fix(odm): decode encoded Azure blob names exactly once (#7251)

* fix(odm): decode encoded Azure blob names exactly once

* test(odm): cover Azure encoded name transport matrix

* test(odm): keep Azure cursor continuation query stable

.config/e2e-distributed-selection.txt

@@ -0,0 +1,2 @@
+sha256-linux=9785867929047dfd8c6f768e0d2b1e0a8fdba85216f4a4139093b1619d03ff07
+sha256-darwin=9785867929047dfd8c6f768e0d2b1e0a8fdba85216f4a4139093b1619d03ff07

.config/e2e-full-selection.txt

@@ -1,2 +1,2 @@
 sha256-darwin=a881fd7d3f5cb94654221ca85b8b30cce1b95e608824a55a15339cbc294e6d34
-sha256-linux=e9a8d64e73f627c4d26c236dbbba690c9ee03a9e26d42a4244515b4439365535
+sha256-linux=a2933d83dfe74ffa03410a0959333a1c48288b8469ca9f17273d449d7510c24b

.config/ecstore-required-tests.json

@@ -0,0 +1,87 @@
+{
+  "lane": "ci/test-and-lint",
+  "tests": [
+    {
+      "invariant": "write-quorum",
+      "suite": "rustfs-ecstore",
+      "name": "set_disk::ops::object::inline_put_commit_path_tests::inline_put_direct_commit_accepts_exact_quorum_and_rejects_quorum_minus_one"
+    },
+    {
+      "invariant": "metadata-rollback",
+      "suite": "rustfs-ecstore",
+      "name": "set_disk::core::io_primitives::tests::write_unique_file_info_reverts_metadata_when_write_quorum_fails"
+    },
+    {
+      "invariant": "stale-writer",
+      "suite": "rustfs-ecstore",
+      "name": "set_disk::ops::object::put_object_tmp_cleanup_tests::put_object_no_lock_aborts_after_outer_namespace_lock_loss"
+    },
+    {
+      "invariant": "range-body",
+      "suite": "rustfs-ecstore",
+      "name": "set_disk::ops::object::transition_upload_integrity_tests::transitioned_compressed_object_range_get_returns_plaintext_slice"
+    },
+    {
+      "invariant": "multipart-cancellation",
+      "suite": "rustfs-ecstore",
+      "name": "set_disk::ops::multipart::tests::cancelled_complete_keeps_upload_lock_through_tail_cleanup"
+    },
+    {
+      "invariant": "list-uncommitted-version",
+      "suite": "rustfs-filemeta",
+      "name": "metacache::tests::resolve_with_write_quorum_slack_keeps_partial_latest_hidden_during_merge"
+    },
+    {
+      "invariant": "minio-object-fixture",
+      "suite": "rustfs-filemeta",
+      "name": "filemeta::test::parses_real_minio_object_xlmeta"
+    },
+    {
+      "invariant": "corrupt-part-arrays",
+      "suite": "rustfs-filemeta",
+      "name": "filemeta::test::crc_valid_but_part_arrays_corrupt_into_fileinfo_errors_not_panics"
+    },
+    {
+      "invariant": "odm-source-contract-s3",
+      "suite": "rustfs",
+      "name": "on_demand_migration::source_client::tests::s3_backend_satisfies_the_shared_backend_contract"
+    },
+    {
+      "invariant": "odm-source-contract-azure",
+      "suite": "rustfs",
+      "name": "on_demand_migration::azure::tests::azure_backend_satisfies_the_shared_backend_contract"
+    },
+    {
+      "invariant": "odm-source-contract-gcs",
+      "suite": "rustfs",
+      "name": "on_demand_migration::gcs::tests::gcs_native_backend_satisfies_the_shared_backend_contract"
+    }
+  ],
+  "fixtures": [
+    {
+      "path": "crates/filemeta/tests/fixtures/minio/object_large_bin.xlmeta.hex",
+      "sha256": "e8093767806d701e639b48d023190e858fbc4cde69bcfd83c22af8cba8452ce5",
+      "source": "MinIO RELEASE.2025-07-23T15-54-02Z; crates/ecstore/tests/fixtures/minio/README.md"
+    },
+    {
+      "path": "crates/filemeta/tests/fixtures/minio/object_small_txt.xlmeta.hex",
+      "sha256": "2a415ad3a3be5a9440035d4026ff880e0e8c1ec1701be9f4e077734e8dce03da",
+      "source": "MinIO RELEASE.2025-07-23T15-54-02Z; crates/ecstore/tests/fixtures/minio/README.md"
+    },
+    {
+      "path": "crates/filemeta/tests/fixtures/minio/object_versioned_txt.xlmeta.hex",
+      "sha256": "7f21f50c326dd8b0228deb6dbdb7052b3d0a3f8ee6c85d43486f0e6bb7a97261",
+      "source": "MinIO RELEASE.2025-07-23T15-54-02Z; crates/ecstore/tests/fixtures/minio/README.md"
+    },
+    {
+      "path": "crates/ecstore/tests/fixtures/minio/bucket_metadata.blob.hex",
+      "sha256": "f2b6e260aff106adf6039feb1c645686e84e75404ff725491fb18668be5db203",
+      "source": "MinIO RELEASE.2025-07-23T15-54-02Z; crates/ecstore/tests/fixtures/minio/README.md"
+    },
+    {
+      "path": "crates/ecstore/tests/fixtures/minio/bucket_metadata_full.xlmeta.hex",
+      "sha256": "3b6de589519c08a1614c8bd409bb8199c17d42043861b07bce513075e6fbfc12",
+      "source": "MinIO RELEASE.2025-07-23T15-54-02Z; crates/ecstore/tests/fixtures/minio/README.md"
+    }
+  ]
+}

.config/make/pre-commit.mak

@@ -3,9 +3,10 @@
 .NOTPARALLEL: pre-commit pre-pr dev-check
 
 .PHONY: setup-hooks
-setup-hooks: ## Set up git hooks
+setup-hooks: ## Install the configured pre-commit hooks
 	@echo "🔧 Setting up git hooks..."
-	chmod +x .git/hooks/pre-commit
+	pre-commit validate-config
+	pre-commit install
 	@echo "✅ Git hooks setup complete!"
 
 .PHONY: doc-paths-check

.config/make/tests.mak

@@ -31,6 +31,7 @@ script-tests: ## Run shell script tests
 	./scripts/test_object_batch_bench_enhanced.sh
 	./scripts/test_hotpath_warp_ab_gate.sh
 	./scripts/test_hotpath_warp_abba.sh
+	./scripts/test_scanner_validation_harness.sh
 	./scripts/test_exact_1mib_handoff_abba.sh
 	./scripts/test_pinned_paired_abba_bench.sh
 	./scripts/test_manual_transition_runbooks.sh
@@ -40,6 +41,8 @@ script-tests: ## Run shell script tests
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_test_wiring.py --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_security_coverage.py --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_scheduled_validation_freshness.py --self-test
+	$(RUSTFS_PYTHON_BIN) ./scripts/test_security_workflow.py
+	$(RUSTFS_PYTHON_BIN) ./scripts/test_nightly_candidate.py
 	$(RUSTFS_PYTHON_BIN) ./scripts/s3-tests/test_report_compat.py
 	bash -n ./scripts/validate_object_data_cache_cold_stampede.sh
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_object_data_cache_follower_samples.py --self-test

.config/migration-gate-floor.txt

@@ -1,10 +1,11 @@
 # Committed floor for the number of tests selected by the migration-critical
 # CI gate (see scripts/check_migration_gate_count.sh, backlog#1153 infra-12).
 #
-# The floor equals the exact count of rustfs-ecstore --lib tests matching the
+# The floor equals the exact count of rustfs-ecstore --lib tests, with the
-# gate filter (name substrings: data_movement, rebalance, decommission,
+# test-util feature enabled, matching the gate filter (name substrings:
-# source_cleanup, delete_marker) at the time this file was last updated.
+# data_movement, rebalance, decommission, source_cleanup, delete_marker) at
+# the time this file was last updated.
 # CI fails if the selected count drops below this number, so renames or
 # removals that thin the gate must update this file in the same PR.
 # Adding tests does not require a bump, but bumping keeps the guard tight.
-571
+946

.config/nextest.toml

@@ -183,6 +183,13 @@ test-group = 'e2e-reliability'
 filter = 'package(e2e_test) & test(/^inline_fast_path_cluster_test::/)'
 test-group = 'e2e-inline-boundaries'
 
+# 4-node 4-drive distributed Actions suite: each case starts four rustfs
+# processes and up to sixteen data directories. Serialize across nextest's
+# process boundary so several 4x4 clusters never overlap.
+[[profile.default.overrides]]
+filter = 'package(e2e_test) & test(/^distributed::/)'
+test-group = 'e2e-cluster-nightly'
+
 # Vault KMS tests share the fixed dev-server port 8200. serial_test's #[serial]
 # does not cross nextest process boundaries, so keep every Vault-backed test in
 # one group.
@@ -526,6 +533,27 @@ path = "junit.xml"
 filter = 'package(e2e_test)'
 test-group = 'e2e-cluster-nightly'
 
+# ---------------------------------------------------------------------------
+# e2e-distributed profile — 4-node 4-disk Actions suite
+# ---------------------------------------------------------------------------
+# Storage-sensitive PR / nightly / dispatch lane owned by
+# .github/workflows/e2e-distributed.yml.
+# Each case starts four rustfs processes (and for site replication, two
+# clusters). Upgrade cases also require RUSTFS_UPGRADE_SOURCE_BINARY.
+# Serialized via e2e-cluster-nightly with no retries.
+[profile.e2e-distributed]
+default-filter = 'package(e2e_test) & test(/^distributed::/)'
+fail-fast = false
+# Decommission / rebalance cases poll for up to 180s with little stdout.
+slow-timeout = { period = "120s", terminate-after = 6 }
+
+[profile.e2e-distributed.junit]
+path = "junit.xml"
+
+[[profile.e2e-distributed.overrides]]
+filter = 'package(e2e_test)'
+test-group = 'e2e-cluster-nightly'
+
 # ---------------------------------------------------------------------------
 # e2e-odm-interop profile — on-demand migration provider interop lane (ODM-20)
 # ---------------------------------------------------------------------------
@@ -586,6 +614,10 @@ path = "junit.xml"
 #     cluster-fault lane. heal_erasure_disk_rebuild is intentionally not
 #     excluded here because backlog#2213 promotes core heal rebuild coverage to
 #     this merge/main lane while retaining nightly coverage.
+#   * distributed:: — 4-node 4-disk Actions suite (S3, lock, versioning,
+#     replication, quota, observability, expand/decommission/rebalance, site
+#     replication, chaos, upgrade history/IAM). Owns [profile.e2e-distributed] and
+#     .github/workflows/e2e-distributed.yml.
 #   * on_demand_migration::interop_test — the ODM-20 provider interoperability
 #     cases, which are meaningless without a source: they run in the dedicated
 #     [profile.e2e-odm-interop] lane below, where the workflow points them at a
@@ -607,6 +639,7 @@ default-filter = """
     package(e2e_test)
     & !test(/^protocols::/)
     & !test(/^(admin_timeout_regression_test|cluster_concurrency_test|cluster_multidrive_pool_test|degraded_listing_availability_test|namespace_lock_quorum_test|object_lambda_test|stale_multipart_cleanup_cluster_test)::/)
+    & !test(/^distributed::/)
     & !test(/^replication_extension_test::/)
     & !test(/^replication_target_matrix_test::/)
     & !test(/^on_demand_migration::(concurrency_test|fault_test|interop_test|real_source_test)::/)

.config/scanner-heal-required-tests.json

@@ -0,0 +1,40 @@
+{
+  "schema": 1,
+  "cases": {
+    "background-target-restart": {
+      "gate": "G14",
+      "task": "W21",
+      "lane": "e2e-nightly",
+      "suite": "e2e_test",
+      "name": "heal_erasure_disk_rebuild_test::tests::test_cluster_root_heal_recovers_remote_shards_after_background_target_restart",
+      "oracle": "background-target-restart.json",
+      "min_objects": 9,
+      "max_objects": 65,
+      "topology": {"nodes": 4, "drives_per_node": 1},
+      "scope": "Target process restart, exact unversioned S3 bodies and replacement-disk shards; not power loss or EC8+4."
+    }
+  },
+  "release_pending": {
+    "G01": "W02/W04 complete root and quota authority coverage",
+    "G02": "W03 bounded checkpoint progress and independent version inventory",
+    "G03": "W17/W18 exact scoped ACK with durable publication and mixed peers",
+    "G04": "W03/W15/W16 crash at every cache/root/floor/intent boundary",
+    "G05": "W06/W07 per-object outcomes and bounded terminal retention",
+    "G06": "W06/W08/W23 concurrent status, legacy clients and truncation",
+    "G07": "W12/W13/W14 durable MRF responsibility at every commit boundary",
+    "G08": "W12/W13/W14 MRF capacity, disk-full and replica-loss matrix",
+    "G09": "W13/W18/W23 actual mixed-version reader/writer and rollback payloads",
+    "G10": "W05/W09/W10/W11 bounded scheduling and pressure recovery",
+    "G11": "W04/W19/W24 maintenance and complete producer coverage",
+    "G12": "W02/W15/W16 both quota paths during reset and settlement",
+    "G13": "W07/W14 quorum-minus-one, unknown disks, remount, Object Lock, dry-run, grace and commit tail",
+    "G14": "W20/W21 same-window field evidence; 3x4 EC8+4 and multi-set/pool coverage",
+    "P1": "W20 measured cold-walk share and foreground latency/throughput",
+    "P2": "W20/W24 measured post-stop convergence and cold segment reuse",
+    "P3": "W20 measured two-hour pressure/heal capacity and recovery window",
+    "P4": "W20 measured MRF scale and replay cost with retained responsibility",
+    "R-E": "W03/W05 fixed-budget real process restart through enumeration and classification",
+    "R-D": "W07/W14 manager-to-event-to-ledger exact disposition, including grace",
+    "R-L": "W13/W14 legacy source conflicts, migration gaps and crash-safe source retirement"
+  }
+}

.github/actions/quick-checks/action.yml

@@ -0,0 +1,120 @@
+# Copyright 2024 RustFS Team
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+name: Quick Checks
+description: Run the shared compile-free RustFS quality checks.
+
+runs:
+  using: composite
+  steps:
+    - name: Install quality tools
+      uses: taiki-e/install-action@bffeee26d4db9be238a4ea78d8826604ebcb594d # v2
+      with:
+        tool: |
+          ripgrep@15.2.0
+          shellcheck@0.11.0
+
+    - name: Install actionlint
+      shell: bash
+      run: |
+        actionlint_dir="$(mktemp -d "${RUNNER_TEMP}/actionlint.XXXXXX")"
+        curl --fail --location --silent --show-error \
+          --output "$actionlint_dir/actionlint.tar.gz" \
+          https://github.com/rhysd/actionlint/releases/download/v1.7.12/actionlint_1.7.12_linux_amd64.tar.gz
+        echo "8aca8db96f1b94770f1b0d72b6dddcb1ebb8123cb3712530b08cc387b349a3d8  $actionlint_dir/actionlint.tar.gz" | sha256sum --check --status
+        tar -xzf "$actionlint_dir/actionlint.tar.gz" -C "$actionlint_dir" actionlint
+        rm "$actionlint_dir/actionlint.tar.gz"
+        echo "$actionlint_dir" >> "$GITHUB_PATH"
+
+    - name: Install Rust toolchain
+      uses: dtolnay/rust-toolchain@29eef336d9b2848a0b548edc03f92a220660cdb8 # stable
+      with:
+        components: rustfmt
+
+    - name: Check workflow syntax and shell scripts
+      shell: bash
+      run: shellcheck --version && actionlint
+
+    - name: Check code formatting
+      shell: bash
+      run: cargo fmt --all --check
+
+    - name: Check unsafe code allowances
+      shell: bash
+      run: ./scripts/check_unsafe_code_allowances.sh
+
+    - name: Check layered dependencies
+      shell: bash
+      run: ./scripts/check_layer_dependencies.sh
+
+    - name: Check architecture migration rules
+      shell: bash
+      run: ./scripts/check_architecture_migration_rules.sh
+
+    - name: Check logging guardrails
+      shell: bash
+      run: ./scripts/check_logging_guardrails.sh
+
+    - name: Check error other(format!) ratchet
+      shell: bash
+      run: ./scripts/check_error_other_format_ratchet.sh
+
+    - name: Check tokio io-uring feature guard
+      shell: bash
+      run: ./scripts/check_no_tokio_io_uring.sh
+
+    - name: Check extension schema boundaries
+      shell: bash
+      run: ./scripts/check_extension_schema_boundaries.sh
+
+    - name: Check body-cache whitelist guard
+      shell: bash
+      run: ./scripts/check_body_cache_whitelist.sh
+
+    - name: Check s3s footprint ratchet
+      shell: bash
+      run: ./scripts/check_s3s_footprint.sh
+
+    - name: Check cryptographic capability wording
+      shell: bash
+      run: ./scripts/check_fips_wording.sh
+
+    - name: Check no embedded secret material
+      shell: bash
+      run: ./scripts/check_embedded_secrets.sh
+
+    - name: Run script contract tests
+      shell: bash
+      run: make script-tests
+
+    - name: Check test wiring
+      shell: bash
+      run: |
+        python3 ./scripts/check_test_wiring.py --self-test
+        python3 ./scripts/check_scheduled_validation_freshness.py --self-test
+        python3 ./scripts/test_security_workflow.py
+        python3 ./scripts/test_nightly_candidate.py
+        python3 ./scripts/check_test_wiring.py
+
+    - name: Check no planning docs committed
+      shell: bash
+      run: ./scripts/check_no_planning_docs.sh
+
+    - name: Check CI paths stay in sync
+      shell: bash
+      run: ./scripts/check_ci_paths_sync.sh
+
+    - name: Check io_uring lane --lib precondition
+      shell: bash
+      run: ./scripts/check_uring_lane_lib_only.sh

.github/pull_request_template.md

@@ -10,16 +10,16 @@ Use N/A when there is no related issue.
 
 ## Summary of Changes
 <!--
-Briefly explain what changed and why reviewers should accept it.
+Describe the concrete problem and resulting behavior. For a behavior change, name the input or state that triggers it and the expected outcome. Explain any new dependency or abstraction that the change needs.
-Focus on behavior, compatibility, and review-relevant context.
 -->
 
 ## Verification
 <!--
-List the commands or checks you ran, for example:
+Give 1–3 concrete pieces of evidence for the changed behavior: the test or command, its observed result, and the regression it catches. For a bug fix, record a failing-before/passing-after check or explain why it was unavailable.
-- `make pre-commit`
 
-Use N/A only when verification is not applicable.
+Identify the tested commit and any local changes. When testing a prebuilt binary or external service, include its source/version and artifact identity; a successful run against a different build is not evidence for this change.
+
+List relevant checks not run and the remaining risk. Use the validation tier in AGENTS.md; do not run broader checks solely to fill this section. For documentation-only changes, list the applicable documentation checks. Use N/A only when verification is not applicable.
 -->
 
 ## Impact

.github/scheduled-validations.json

@@ -4,6 +4,11 @@
   { "workflow": ".github/workflows/ci.yml", "max_age_hours": 192 },
   { "workflow": ".github/workflows/coverage.yml", "max_age_hours": 192 },
   { "workflow": ".github/workflows/e2e-replication-nightly.yml", "max_age_hours": 36 },
+  {
+    "workflow": ".github/workflows/e2e-distributed.yml",
+    "max_age_hours": 36,
+    "never_ran_grace_until": "2026-09-18T00:00:00Z"
+  },
   { "workflow": ".github/workflows/e2e-s3tests.yml", "max_age_hours": 192 },
   { "workflow": ".github/workflows/fuzz.yml", "max_age_hours": 36 },
   { "workflow": ".github/workflows/mint.yml", "max_age_hours": 192 },

.github/workflows/ci-docs-only.yml

@@ -12,24 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# Companion to ci.yml for required status checks.
+# Reports the existing required checks for paths excluded by ci.yml.
-#
+# Mixed PRs can trigger both workflows; their Quick Checks jobs use one shared
-# ci.yml skips docs-only pull requests via paths-ignore, but the branch ruleset
+# action to keep validation coverage aligned. Keep this paths list in sync with
-# requires a check named "Test and Lint" — without this workflow a docs-only PR
+# ci.yml's pull_request.paths-ignore via scripts/check_ci_paths_sync.sh.
-# would wait on it forever. This workflow triggers on exactly the paths ci.yml
-# ignores and reports success under the same job name. Mixed PRs trigger both
-# workflows and the real check still gates: a required check with any failing
-# run blocks the merge.
-# https://docs.github.com/en/repositories/configuring-branches-and-merges-in-your-repository/defining-the-mergeability-of-pull-requests/troubleshooting-required-status-checks#handling-skipped-but-required-checks
-#
-# "Quick Checks" is mirrored here ahead of the ruleset change that will make it
-# required too (rustfs/backlog#1599). Until that change lands this job is
-# inert; mirroring it first is what lets the ruleset change happen without
-# stranding docs-only PRs on a check nobody reports.
-#
-# Keep the paths list below in sync with the pull_request paths-ignore list
-# in ci.yml, and keep the quick-checks steps below byte-identical to the
-# quick-checks job in ci.yml.
 
 name: Continuous Integration (docs only)
 
@@ -59,19 +45,6 @@ permissions:
   contents: read
 
 jobs:
-  # Deliberately NOT a bare `echo`. Once "Quick Checks" becomes a required
-  # check, ci.yml gates every expensive job behind it, so a mixed PR reports
-  # two check runs with this name: the real one (45-51s) and this companion.
-  # GitHub has no written contract for how it picks between same-named
-  # required check runs ("latest wins" vs "any failure blocks"), so instead of
-  # relying on ordering we make both runs execute the same commands against
-  # the same merge ref — their conclusions are then necessarily identical and
-  # the choice does not matter. Keep these steps byte-identical to the
-  # quick-checks job in ci.yml (a guard script that asserts this, and the paths
-  # sync below, is tracked in rustfs/backlog#1603).
-  #
-  # For a genuinely docs-only PR this adds no strictness (no code changed, so
-  # fmt and the guards always pass) and costs ~50s of ubuntu-latest.
   quick-checks:
     name: Quick Checks
     runs-on: ubuntu-latest
@@ -82,63 +55,8 @@ jobs:
         with:
           persist-credentials: false
 
-      - name: Install ripgrep
+      - name: Run shared quick checks
-        uses: taiki-e/install-action@bffeee26d4db9be238a4ea78d8826604ebcb594d # v2
+        uses: ./.github/actions/quick-checks
-        with:
-          tool: ripgrep@15.2.0
-
-      - name: Install Rust toolchain
-        uses: dtolnay/rust-toolchain@29eef336d9b2848a0b548edc03f92a220660cdb8 # stable
-        with:
-          components: rustfmt
-
-      - name: Check code formatting
-        run: cargo fmt --all --check
-
-      - name: Check unsafe code allowances
-        run: ./scripts/check_unsafe_code_allowances.sh
-
-      - name: Check layered dependencies
-        run: ./scripts/check_layer_dependencies.sh
-
-      - name: Check architecture migration rules
-        run: ./scripts/check_architecture_migration_rules.sh
-
-      - name: Check logging guardrails
-        run: ./scripts/check_logging_guardrails.sh
-
-      - name: Check tokio io-uring feature guard
-        run: ./scripts/check_no_tokio_io_uring.sh
-
-      - name: Check extension schema boundaries
-        run: ./scripts/check_extension_schema_boundaries.sh
-
-      - name: Check body-cache whitelist guard
-        run: ./scripts/check_body_cache_whitelist.sh
-
-      - name: Check s3s footprint ratchet
-        run: ./scripts/check_s3s_footprint.sh
-
-      - name: Check cryptographic capability wording
-        run: ./scripts/check_fips_wording.sh
-
-      - name: Check no embedded secret material
-        run: ./scripts/check_embedded_secrets.sh
-
-      - name: Check test wiring
-        run: |
-          python3 ./scripts/check_test_wiring.py --self-test
-          python3 ./scripts/check_scheduled_validation_freshness.py --self-test
-          python3 ./scripts/check_test_wiring.py
-
-      - name: Check no planning docs committed
-        run: ./scripts/check_no_planning_docs.sh
-
-      - name: Check CI paths stay in sync
-        run: ./scripts/check_ci_paths_sync.sh
-
-      - name: Check io_uring lane --lib precondition
-        run: ./scripts/check_uring_lane_lib_only.sh
 
   test-and-lint:
     name: Test and Lint

.github/workflows/ci.yml

@@ -100,12 +100,7 @@ jobs:
       - name: Typos check with custom config file
         uses: crate-ci/typos@37bb98842b0d8c4ffebdb75301a13db0267cef89 # master
 
-  # Fast, compile-free checks that fail early so contributors get feedback in
+  # Fail early with compile-free checks shared with docs-only CI.
-  # ~1 minute instead of waiting for the full test job.
-  #
-  # These steps are mirrored byte-for-byte in ci-docs-only.yml so that a mixed
-  # PR, which reports two check runs named "Quick Checks", cannot get one red
-  # and one green. Edit both jobs together.
   quick-checks:
     name: Quick Checks
     if: github.event_name != 'pull_request' || github.event.action != 'closed'
@@ -117,66 +112,8 @@ jobs:
         with:
           persist-credentials: false
 
-      - name: Install ripgrep
+      - name: Run shared quick checks
-        uses: taiki-e/install-action@bffeee26d4db9be238a4ea78d8826604ebcb594d # v2
+        uses: ./.github/actions/quick-checks
-        with:
-          tool: ripgrep@15.2.0
-
-      - name: Install Rust toolchain
-        uses: dtolnay/rust-toolchain@29eef336d9b2848a0b548edc03f92a220660cdb8 # stable
-        with:
-          components: rustfmt
-
-      - name: Check code formatting
-        run: cargo fmt --all --check
-
-      - name: Check unsafe code allowances
-        run: ./scripts/check_unsafe_code_allowances.sh
-
-      - name: Check layered dependencies
-        run: ./scripts/check_layer_dependencies.sh
-
-      - name: Check architecture migration rules
-        run: ./scripts/check_architecture_migration_rules.sh
-
-      - name: Check logging guardrails
-        run: ./scripts/check_logging_guardrails.sh
-
-      - name: Check error other(format!) ratchet
-        run: ./scripts/check_error_other_format_ratchet.sh
-
-      - name: Check tokio io-uring feature guard
-        run: ./scripts/check_no_tokio_io_uring.sh
-
-      - name: Check extension schema boundaries
-        run: ./scripts/check_extension_schema_boundaries.sh
-
-      - name: Check body-cache whitelist guard
-        run: ./scripts/check_body_cache_whitelist.sh
-
-      - name: Check s3s footprint ratchet
-        run: ./scripts/check_s3s_footprint.sh
-
-      - name: Check cryptographic capability wording
-        run: ./scripts/check_fips_wording.sh
-
-      - name: Check no embedded secret material
-        run: ./scripts/check_embedded_secrets.sh
-
-      - name: Check test wiring
-        run: |
-          python3 ./scripts/check_test_wiring.py --self-test
-          python3 ./scripts/check_scheduled_validation_freshness.py --self-test
-          python3 ./scripts/check_test_wiring.py
-
-      - name: Check no planning docs committed
-        run: ./scripts/check_no_planning_docs.sh
-
-      - name: Check CI paths stay in sync
-        run: ./scripts/check_ci_paths_sync.sh
-
-      - name: Check io_uring lane --lib precondition
-        run: ./scripts/check_uring_lane_lib_only.sh
 
   test-and-lint:
     name: Test and Lint
@@ -269,6 +206,7 @@ jobs:
           CARGO_BUILD_JOBS: ${{ (github.event_name == 'push' || github.event_name == 'workflow_dispatch') && '3' || '2' }}
         run: |
           mkdir -p artifacts/test-and-lint
+          rm -f target/nextest/ci/junit.xml
           ./scripts/ci/resource_sampler.sh start nextest
           trap './scripts/ci/resource_sampler.sh stop' EXIT
           set +e
@@ -277,6 +215,12 @@ jobs:
             --status-level all --final-status-level all \
             2>&1 | tee artifacts/test-and-lint/nextest.log
           status=${PIPESTATUS[0]}
+          if [[ "${status}" -eq 0 ]]; then
+            cargo nextest list --profile ci --all --exclude e2e_test --message-format json \
+              > artifacts/test-and-lint/core-test-listing.json \
+              && python3 scripts/check_test_wiring.py --check-core artifacts/test-and-lint/core-test-listing.json \
+              && test -s target/nextest/ci/junit.xml || status=$?
+          fi
           {
             echo "command=cargo nextest run --profile ci --all --exclude e2e_test"
             echo "exit_status=${status}"
@@ -906,6 +850,11 @@ jobs:
           cache-save-if: 'false'
           install-build-packaging-tools: 'false'
 
+      - name: Install network fault-injection tools
+        run: |
+          sudo apt-get install -y iptables
+          sudo -n iptables --version
+
       - name: Set up Python
         uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
         with:

.github/workflows/e2e-distributed.yml

@@ -0,0 +1,206 @@
+# Copyright 2024 RustFS Team
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# 4-node 4-disk distributed e2e lane.
+#
+# Each selected test starts a real localhost cluster via
+# `RustFSTestClusterEnvironment` (4 processes; 4 drives per node unless the
+# case is a two-site 4-node 1-drive pair or a 4-node upgrade). Membership is
+# `[profile.e2e-distributed]` in `.config/nextest.toml`. Storage-sensitive PRs,
+# nightly runs, and manual dispatches all execute the same fail-closed suite.
+# Upgrade cases download the same pinned previous release as e2e-upgrade.yml.
+#
+# Isolated pool filesystems: expand/decommission/rebalance cases require
+# independent `statfs` capacity. `sm-standard-4` is an ARC pod
+# (`scripts/ci/check_runner_ephemerality.sh`) and usually has no
+# `/dev/loop-control`, so `mount -o loop` fails with ENOENT ("mount failed:
+# No such file or directory"). The prepare step therefore mounts four 1 GiB
+# tmpfs instances and exports them as `RUSTFS_E2E_POOL_ROOTS`.
+
+name: e2e-distributed
+
+on:
+  pull_request:
+    paths:
+      - "Cargo.lock"
+      - "Cargo.toml"
+      - ".config/nextest.toml"
+      - ".github/workflows/e2e-distributed.yml"
+      - "crates/audit/**"
+      - "crates/common/**"
+      - "crates/config/**"
+      - "crates/e2e_test/**"
+      - "crates/ecstore/**"
+      - "crates/filemeta/**"
+      - "crates/heal/**"
+      - "crates/iam/**"
+      - "crates/lock/**"
+      - "crates/madmin/**"
+      - "crates/notify/**"
+      - "crates/replication/**"
+      - "crates/s3-client/**"
+      - "crates/s3-ops/**"
+      - "crates/s3-types/**"
+      - "crates/scanner/**"
+      - "crates/storage-api/**"
+      - "crates/utils/**"
+      - "rustfs/**"
+  workflow_dispatch:
+    inputs:
+      filter:
+        description: "Optional nextest -E filter (default: the whole e2e-distributed profile)"
+        required: false
+        default: ""
+  schedule:
+    # 05:53 UTC nightly — clear of e2e-nightly (04:29) and ODM interop (05:23).
+    - cron: "53 5 * * *"
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.event_name != 'schedule' }}
+
+jobs:
+  distributed:
+    name: Distributed 4-node 4-disk e2e
+    runs-on: sm-standard-4
+    timeout-minutes: 180
+    env:
+      FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+      NO_PROXY: 127.0.0.1,localhost
+      HTTP_PROXY: ""
+      HTTPS_PROXY: ""
+      # Pinned previous release used by distributed::upgrade_test (same pin as e2e-upgrade.yml).
+      UPGRADE_SOURCE_VERSION: 1.0.0-rc.2
+      UPGRADE_SOURCE_ASSET: rustfs-linux-x86_64-gnu-v1.0.0-rc.2.zip
+      UPGRADE_SOURCE_SHA256: 7c789386bf85278f865b8e0d359bf4edb84d5aa408cc3fa54a18c25ca74cd6e7
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Setup Rust environment
+        uses: ./.github/actions/setup
+        with:
+          rust-version: stable
+          cache-shared-key: ci-e2e-distributed
+          cache-save-if: ${{ github.ref == 'refs/heads/main' }}
+          install-build-packaging-tools: 'false'
+
+      - name: Prepare isolated filesystems for pool movement
+        run: |
+          set -euo pipefail
+          mount_base="${RUNNER_TEMP}/rustfs-e2e-pools"
+          mkdir -p "${mount_base}"
+          roots=()
+          for pool in 0 1 2 3; do
+            mountpoint="${mount_base}/pool-${pool}"
+            mkdir -p "${mountpoint}"
+            # sm-standard-4 is an ARC pod without usable loop devices, so
+            # `mount -o loop` fails with ENOENT. Sized tmpfs still reports a
+            # distinct st_dev and independent 1G statfs capacity.
+            sudo mount -t tmpfs -o size=1G,nosuid,nodev,mode=1777 tmpfs "${mountpoint}"
+            sudo chmod 1777 "${mountpoint}"
+            roots+=("${mountpoint}")
+          done
+          printf -v joined_roots '%s:' "${roots[@]}"
+          echo "RUSTFS_E2E_POOL_ROOTS=${joined_roots%:}" >> "${GITHUB_ENV}"
+          findmnt --noheadings --output TARGET,SOURCE,FSTYPE,SIZE --target "${roots[0]}"
+          findmnt --noheadings --output TARGET,SOURCE,FSTYPE,SIZE --target "${roots[1]}"
+          findmnt --noheadings --output TARGET,SOURCE,FSTYPE,SIZE --target "${roots[2]}"
+          findmnt --noheadings --output TARGET,SOURCE,FSTYPE,SIZE --target "${roots[3]}"
+
+      - name: Download pinned previous release
+        env:
+          SOURCE_DIR: ${{ runner.temp }}/rustfs-upgrade-source
+        run: |
+          set -euo pipefail
+          mkdir -p "$SOURCE_DIR"
+          archive="$SOURCE_DIR/$UPGRADE_SOURCE_ASSET"
+          curl --fail --location --retry 3 --output "$archive" \
+            "https://github.com/${GITHUB_REPOSITORY}/releases/download/${UPGRADE_SOURCE_VERSION}/${UPGRADE_SOURCE_ASSET}"
+          echo "$UPGRADE_SOURCE_SHA256  $archive" | sha256sum --check --strict
+          unzip -q "$archive" -d "$SOURCE_DIR"
+          chmod +x "$SOURCE_DIR/rustfs"
+          test -x "$SOURCE_DIR/rustfs"
+          echo "RUSTFS_UPGRADE_SOURCE_BINARY=$SOURCE_DIR/rustfs" >> "$GITHUB_ENV"
+
+      - name: Build rustfs binary
+        run: |
+          cargo build -p rustfs --bins
+          : > target/debug/rustfs.features
+
+      - name: Verify distributed e2e membership
+        env:
+          NEXTEST_LISTING: ${{ runner.temp }}/rustfs-e2e-distributed-list.json
+        run: |
+          cargo nextest list --profile e2e-distributed -p e2e_test --message-format json > "${NEXTEST_LISTING}"
+          python3 ./scripts/check_test_wiring.py --check-profile e2e-distributed "${NEXTEST_LISTING}"
+
+      - name: Run distributed 4-node e2e suite
+        env:
+          RUSTFS_E2E_LOG_DIR: ${{ runner.temp }}/rustfs-e2e-distributed-logs
+          FILTER: ${{ inputs.filter }}
+        run: |
+          set -euo pipefail
+          if [ -n "${FILTER}" ]; then
+            cargo nextest run --profile e2e-distributed -p e2e_test -E "${FILTER}"
+          else
+            cargo nextest run --profile e2e-distributed -p e2e_test --no-tests=fail
+          fi
+
+      - name: Upload distributed e2e diagnostics
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: e2e-distributed-${{ github.run_number }}
+          path: |
+            target/nextest/e2e-distributed/junit.xml
+            ${{ runner.temp }}/rustfs-e2e-distributed-list.json
+            ${{ runner.temp }}/rustfs-e2e-distributed-logs/
+          retention-days: 7
+          if-no-files-found: warn
+
+      - name: Unmount isolated pool filesystems
+        if: always()
+        run: |
+          set -euo pipefail
+          mount_base="${RUNNER_TEMP}/rustfs-e2e-pools"
+          for pool in 0 1 2 3; do
+            mountpoint="${mount_base}/pool-${pool}"
+            if mountpoint --quiet "${mountpoint}"; then
+              sudo umount "${mountpoint}"
+            fi
+          done
+
+  alert-on-failure:
+    name: Alert on scheduled failure
+    needs: [distributed]
+    if: always() && github.event_name == 'schedule' && contains(needs.*.result, 'failure')
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    permissions:
+      contents: read
+      issues: write
+    steps:
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+      - name: Open or update failure-tracking issue
+        uses: ./.github/actions/schedule-failure-issue
+        with:
+          github-token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/e2e-upgrade.yml

@@ -19,7 +19,9 @@ on:
     paths:
       - ".github/workflows/e2e-upgrade.yml"
       - "crates/e2e_test/src/common.rs"
+      - "crates/e2e_test/src/fake_s3_target/**"
       - "crates/e2e_test/src/lib.rs"
+      - "crates/e2e_test/src/replication_extension_test.rs"
       - "crates/e2e_test/src/upgrade_compatibility_test.rs"
       - "crates/ecstore/**"
       - "crates/filemeta/**"
@@ -44,9 +46,9 @@ concurrency:
 env:
   CARGO_TERM_COLOR: always
   RUST_BACKTRACE: 1
-  UPGRADE_SOURCE_VERSION: 1.0.0-rc.2
+  UPGRADE_SOURCE_VERSION: 1.0.0-rc.5
-  UPGRADE_SOURCE_ASSET: rustfs-linux-x86_64-gnu-v1.0.0-rc.2.zip
+  UPGRADE_SOURCE_ASSET: rustfs-linux-x86_64-gnu-v1.0.0-rc.5.zip
-  UPGRADE_SOURCE_SHA256: 7c789386bf85278f865b8e0d359bf4edb84d5aa408cc3fa54a18c25ca74cd6e7
+  UPGRADE_SOURCE_SHA256: 3ee8df71e8edcfada533be452c4135868f697bc515460ae97b027313eade7a3d
 
 jobs:
   upgrade:
@@ -55,14 +57,31 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - name: Direct upgrade from rc.2
+          # The two `_from_rc2_` tests keep their names: they assert
+          # release-independent object contracts and pass unchanged against the
+          # newer pinned source, so renaming them would only churn history and
+          # the CI required-check names. UPGRADE_SOURCE_VERSION above is the
+          # single source of truth for which release they actually run against.
+          - name: Direct upgrade from the previous release
             cache_key: e2e-direct-upgrade
             test: direct_upgrade_from_rc2_preserves_object_contracts
             artifact: direct-upgrade
-          - name: Mixed-version rolling upgrade from rc.2
+          - name: Mixed-version rolling upgrade from the previous release
             cache_key: e2e-mixed-version-upgrade
             test: rolling_upgrade_from_rc2_preserves_mixed_version_contracts
             artifact: mixed-version-upgrade
+          - name: Bucket configuration survives the upgrade
+            cache_key: e2e-bucket-config-upgrade
+            test: direct_upgrade_from_previous_release_preserves_bucket_configuration
+            artifact: bucket-config-upgrade
+          - name: Rollback reads current bucket metadata
+            cache_key: e2e-bucket-config-rollback
+            test: rollback_to_previous_release_reads_current_bucket_metadata
+            artifact: bucket-config-rollback
+          - name: ODM configuration recovery after rc.5 rollback
+            cache_key: e2e-odm-config-rollback
+            test: rc5_rollback_requires_restoring_odm_configuration
+            artifact: odm-config-rollback
     runs-on: ubuntu-latest
     timeout-minutes: 60
     env:

.github/workflows/nightly-gnu.yml

@@ -166,8 +166,9 @@ jobs:
       # e.g. https://dl.rustfs.com/artifacts/rustfs/packages/nightly/... .
       # Skipped when the R2 secrets are not configured (artifact-only mode).
       - name: Upload DEB to Cloudflare R2
-        if: env.R2_ACCESS_KEY_ID != ''
+        id: publish
         env:
+          DEB_FILE: ${{ steps.deb.outputs.deb_file }}
           R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
           R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
           R2_ENDPOINT: ${{ secrets.R2_ENDPOINT }}
@@ -182,28 +183,70 @@ jobs:
             exit 0
           fi
 
-          if ! command -v aws >/dev/null 2>&1; then
-            sudo apt-get update && sudo apt-get install -y -qq awscli
-          fi
-
           export AWS_ACCESS_KEY_ID="$R2_ACCESS_KEY_ID"
           export AWS_SECRET_ACCESS_KEY="$R2_SECRET_ACCESS_KEY"
           export AWS_DEFAULT_REGION="auto"
 
-          DEB_FILE="${{ steps.deb.outputs.deb_file }}"
+          SOURCE_SHA="$(git rev-parse HEAD)"
+          if [[ "${SOURCE_SHA}" != "${GITHUB_SHA}" ]]; then
+            echo "Checkout SHA does not match the nightly build run" >&2
+            exit 1
+          fi
+          DEB_SHA256="$(sha256sum "${DEB_FILE}" | cut -d ' ' -f 1)"
+          CANDIDATE_KEY="artifacts/rustfs/packages/nightly/runs/${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}/${DEB_SHA256}/rustfs.deb"
+          CANDIDATE_URL="https://dl.rustfs.com/${CANDIDATE_KEY}"
+
+          # Old AWS CLI models lack conditional PutObject support. Never fall
+          # back to an overwriting upload for a candidate.
+          AWS_CLI=aws
+          if ! "${AWS_CLI}" s3api put-object --generate-cli-skeleton input | jq -e 'has("IfNoneMatch")' >/dev/null; then
+            sudo apt-get update
+            sudo apt-get install -y -qq python3-venv
+            AWS_CLI_DIR="$(mktemp -d "${RUNNER_TEMP}/nightly-awscli.XXXXXX")"
+            trap 'rm -rf "${AWS_CLI_DIR}"' EXIT
+            python3 -m venv "${AWS_CLI_DIR}"
+            "${AWS_CLI_DIR}/bin/python" -m pip install --disable-pip-version-check 'awscli==1.44.79'
+            AWS_CLI="${AWS_CLI_DIR}/bin/aws"
+          fi
+          "${AWS_CLI}" s3api put-object --generate-cli-skeleton input | jq -e 'has("IfNoneMatch")' >/dev/null
+          "${AWS_CLI}" --version
+          "${AWS_CLI}" s3api put-object --bucket "${R2_BUCKET}" --key "${CANDIDATE_KEY}" \
+            --body "${DEB_FILE}" --if-none-match '*' --endpoint-url "${R2_ENDPOINT}"
+          PUBLISHED_SHA256="$(curl -fsSL --retry 3 --connect-timeout 15 --max-time 300 "${CANDIDATE_URL}" | sha256sum | cut -d ' ' -f 1)"
+          if [[ "${PUBLISHED_SHA256}" != "${DEB_SHA256}" ]]; then
+            echo "Published candidate checksum does not match the built package" >&2
+            exit 1
+          fi
+
           R2_PREFIX="s3://${R2_BUCKET}/artifacts/rustfs/packages/nightly/"
 
           echo "📤 Uploading ${DEB_FILE} to ${R2_PREFIX}"
-          aws s3 cp "${DEB_FILE}" "${R2_PREFIX}" --endpoint-url "$R2_ENDPOINT" --only-show-errors
+          "${AWS_CLI}" s3 cp "${DEB_FILE}" "${R2_PREFIX}" --endpoint-url "$R2_ENDPOINT" --only-show-errors
 
           # Stable "latest" alias so tests can fetch the newest nightly
           # without knowing today's date.
           echo "📤 Uploading latest alias"
-          aws s3 cp "${DEB_FILE}" "${R2_PREFIX}rustfs-nightly-latest.deb" \
+          "${AWS_CLI}" s3 cp "${DEB_FILE}" "${R2_PREFIX}rustfs-nightly-latest.deb" \
             --endpoint-url "$R2_ENDPOINT" --only-show-errors
 
           echo "✅ R2 upload complete"
 
+          CANDIDATE_FILE="${RUNNER_TEMP}/nightly-candidate-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}.json"
+          jq -n --arg source_sha "${SOURCE_SHA}" \
+            --argjson build_run_id "${GITHUB_RUN_ID}" --argjson build_run_attempt "${GITHUB_RUN_ATTEMPT}" \
+            --arg package_url "${CANDIDATE_URL}" --arg package_sha256 "${DEB_SHA256}" \
+            '{schema: 1, source_sha: $source_sha, build_run_id: $build_run_id, build_run_attempt: $build_run_attempt, package_url: $package_url, package_sha256: $package_sha256}' \
+            > "${CANDIDATE_FILE}"
+          echo "candidate_file=${CANDIDATE_FILE}" >> "${GITHUB_OUTPUT}"
+
+      - name: Upload nightly candidate manifest
+        if: ${{ steps.publish.outputs.candidate_file != '' }}
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: nightly-candidate-${{ github.run_id }}-${{ github.run_attempt }}
+          path: ${{ steps.publish.outputs.candidate_file }}
+          if-no-files-found: error
+
   # Live-Vault lane for the rustfs-kms suite (rustfs/backlog#1774).
   #
   # RUSTFS_KMS_VAULT_TOKEN is the single switch that adds the Vault KV2 and

.github/workflows/rustfs-functional-chain.yml

@@ -14,8 +14,8 @@
 
 # Functional chain driver: runs the ten functional suites in a fixed order
 # (upgrade -> s3 -> kms -> tier -> storage -> heal -> pool -> security ->
-# replication, with performance on its own runner in parallel) and guarantees
+# replication -> performance). Each suite attempts the next handoff even
-# the chain keeps moving even when individual suites fail.
+# when its tests fail.
 #
 # Each suite workflow can still be dispatched standalone (workflow_dispatch);
 # only chain-triggered runs forward to the next suite via repository_dispatch,
@@ -59,16 +59,3 @@ jobs:
           gh api --method POST repos/rustfs/rustfs/dispatches \
             -f event_type='rustfs-chain-upgrade' \
             -F 'client_payload[from_suite]=nightly-build'
-
-      - name: Dispatch performance suite (parallel, own runner)
-        env:
-          GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
-        run: |
-          set -euo pipefail
-          if [ -z "${GH_TOKEN:-}" ]; then
-            echo "PF_TESTING_GH_TOKEN is not configured; cannot dispatch performance" >&2
-            exit 1
-          fi
-          gh api --method POST repos/rustfs/rustfs/dispatches \
-            -f event_type='rustfs-chain-performance' \
-            -F 'client_payload[from_suite]=nightly-build'

.github/workflows/rustfs-heal-test.yml

@@ -54,14 +54,26 @@ env:
 jobs:
   heal-test:
     runs-on: smoke-testing
-    # Requirement: a failing suite must not fail the workflow; failures
-    # are filed to rustfs/backlog and the chain continues.
-    continue-on-error: true
     timeout-minutes: 480
     # Standalone manual run, or one link of the nightly functional chain
     # (storage -> heal -> pool). Pool expansion no longer re-runs heal.
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     steps:
+      - name: Initialize functional evidence
+        id: evidence
+        run: |
+          set -euo pipefail
+          umask 077
+          FUNCTIONAL_ARTIFACTS_DIR="${RUNNER_TEMP}/rustfs-heal-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+          mkdir -- "${FUNCTIONAL_ARTIFACTS_DIR}" "${FUNCTIONAL_ARTIFACTS_DIR}-scratch"
+          {
+            printf 'FUNCTIONAL_ARTIFACTS_DIR=%s\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'LOG_FILE=%s/suite.log\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'RUSTFS_WARP_LOG_FILE=%s/warp.log\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'REPORT_FILE=%s/report.md\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'TMPDIR=%s-scratch\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+          } >> "${GITHUB_ENV}"
+
       # auto-testing is private: clone it with the dedicated PF token (not
       # GITHUB_TOKEN) and retry transient GitHub/network failures.
       - name: Checkout auto-testing scripts (with retry)
@@ -117,7 +129,7 @@ jobs:
           else
             ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
           fi
-          ./auto-testing/rustfs_heal_test.sh "${ARGS[@]}"
+          ./auto-testing/rustfs_heal_test.sh "${ARGS[@]}" --log-file "${LOG_FILE}"
 
       - name: Preflight checks
         run: |
@@ -127,7 +139,7 @@ jobs:
           else
             ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
           fi
-          ./auto-testing/rustfs_heal_test.sh "${ARGS[@]}"
+          ./auto-testing/rustfs_heal_test.sh "${ARGS[@]}" --log-file "${LOG_FILE}"
 
       - name: Run heal test (write -> outage -> heal -> verify)
         id: test
@@ -137,13 +149,10 @@ jobs:
             --endpoint "${{ env.RUSTFS_API_ENDPOINT }}" \
             --stop-node-gb "${{ inputs.stop_node_gb || '15' }}" \
             --warp-stop-gb "${{ inputs.warp_stop_gb || '40' }}" \
-            --log-file /tmp/rustfs-heal-test.log
+            --log-file "${LOG_FILE}"
 
       - name: Generate report
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
-        env:
-          LOG_FILE: /tmp/rustfs-heal-test.log
-          REPORT_FILE: /tmp/rustfs-heal-report.md
         run: |
           set -euo pipefail
           PACKAGE_URL='${{ inputs.package_url }}'
@@ -152,8 +161,9 @@ jobs:
           else
             PACKAGE_SOURCE="${RUSTFS_NIGHTLY_PACKAGE_URL}"
           fi
-          STEPS_TABLE="/tmp/rustfs-heal-steps.md"
+          STEPS_TABLE="${FUNCTIONAL_ARTIFACTS_DIR}/steps.md"
-          python3 - "${LOG_FILE}" "${STEPS_TABLE}" <<'PY'
+          CASE_RESULT=success
+          python3 - "${LOG_FILE}" "${STEPS_TABLE}" <<'PY' || CASE_RESULT=failure
           import re
           import sys
 
@@ -165,6 +175,7 @@ jobs:
 
           steps = {}
           order = []
+          status_rank = {'SKIP': 0, 'PASS': 1, 'FAIL': 2}
           version = None
           version_node = None
           verdict = None
@@ -178,14 +189,15 @@ jobs:
                           n, desc, status = m.group(1), m.group(2), m.group(3)
                           if n not in steps:
                               order.append(n)
-                          steps[n] = (desc, status)  # later lines win (fail after pass)
+                          if n not in steps or status_rank[status] > status_rank[steps[n][1]]:
+                              steps[n] = (desc, status)
                           continue
                       m = ver_re.match(line)
                       if m:
                           version, version_node = m.group(1), m.group(2)
                           continue
                       m = result_re.match(line)
-                      if m:
+                      if m and verdict != 'FAIL':
                           verdict, verdict_detail = m.group(1), m.group(2)
           except FileNotFoundError:
               pass
@@ -205,30 +217,43 @@ jobs:
                   out.write(f'| {n} | {desc} | {status} |\n')
               if not order:
                   out.write('| - | - | NOT RUN (no step result lines found) |\n')
+          complete = set(steps) == {str(n) for n in range(1, 8)}
+          sys.exit(0 if complete and verdict != 'FAIL' and all(status == 'PASS' for _, status in steps.values()) else 1)
           PY
+          RESULT=failure
+          if [ '${{ steps.test.outcome }}' = 'success' ] && [ "${CASE_RESULT}" = 'success' ]; then
+            RESULT=success
+          fi
           {
             echo "# RustFS heal test report"
             echo ""
             echo "- Run: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${{ github.event_name }}"
             echo "- Package: ${PACKAGE_SOURCE}"
-            echo "- Test Step Outcome: ${{ steps.test.outcome }}"
+            echo "- Test Step Outcome: ${RESULT}"
+            echo "- Suite Step Outcome: ${{ steps.test.outcome }}"
             echo ""
-            cat "${STEPS_TABLE}" || true
+            if [ "${RESULT}" = "success" ]; then
+              cat "${STEPS_TABLE}"
               echo ""
               echo "## Log tail"
               echo '```text'
-            tail -n 200 "${LOG_FILE}" || true
+              tail -n 200 "${LOG_FILE}"
               echo '```'
+            else
+              echo "The suite or evidence validation failed. See this run's artifact for partial step results and suite.log."
+            fi
           } | tee "${REPORT_FILE}"
           cat "${REPORT_FILE}" >> "${GITHUB_STEP_SUMMARY}"
+          [ "${RESULT}" = "success" ]
 
       - name: Upload functional report to dashboard
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ env.PF_TESTING_GH_TOKEN }}
-          REPORT_FILE: /tmp/rustfs-heal-report.md
           SUITE: heal
         run: |
           set -euo pipefail
@@ -238,28 +263,32 @@ jobs:
           fi
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="functional-reports/${SUITE}/${DATE}.md"
-          CONTENT="$(python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
+          # Base64-encode the report into a temp file and feed it to jq via
+          # --rawfile: large reports (e.g. pool) exceed the OS argv limit and
+          # make `jq --arg content "${CONTENT}"` fail with "Argument list too long".
+          B64_FILE="$(mktemp)"
+          python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}" > "${B64_FILE}"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" --arg sha "${SHA}" \
-              '{message:$msg, content:$content, sha:$sha}' \
+              '{message:$msg, content:($content|rtrimstr("\n")), sha:$sha}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           else
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" \
-              '{message:$msg, content:$content}' \
+              '{message:$msg, content:($content|rtrimstr("\n"))}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           fi
+          rm -f "${B64_FILE}"
 
       - name: File failure issue in rustfs/backlog
         if: ${{ always() && (failure() || steps.test.outcome == 'failure' || steps.test.outcome == 'cancelled') }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
+          EVIDENCE_OUTCOME: ${{ steps.evidence.outcome }}
           SUITE: 'heal'
           SUITE_LABEL: 'Heal'
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-          REPORT_FILE: '/tmp/rustfs-heal-report.md'
-          LOG_FILE: '/tmp/rustfs-heal-test.log'
         run: |
           set -euo pipefail
           if [ -z "${GH_TOKEN:-}" ]; then
@@ -287,14 +316,16 @@ jobs:
             echo ""
             echo "- Suite: \`${SUITE}\`"
             echo "- Run: ${RUN_URL}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${GITHUB_EVENT_NAME}"
             echo "- Date: $(date -u +%Y-%m-%d)"
             echo ""
             echo "## Report (errors and symptoms)"
             echo ""
-            if [ -s "${REPORT_FILE}" ]; then
+            if [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${REPORT_FILE}" ]; then
               redact < "${REPORT_FILE}"
-            elif [ -s "${LOG_FILE:-}" ]; then
+            elif [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${LOG_FILE:-}" ]; then
               echo "(report file missing; log tail below)"
               echo ""
               tail -n 200 "${LOG_FILE}" | redact
@@ -310,14 +341,16 @@ jobs:
           echo "filed backlog issue for suite ${SUITE}"
 
       - name: Upload test logs
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
         with:
-          name: rustfs-heal-test-${{ github.run_id }}
+          name: rustfs-heal-test-${{ github.run_id }}-${{ github.run_attempt }}
           path: |
-            /tmp/rustfs-heal-test*.log
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/report.md
-            /tmp/rustfs-warp.*.log
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/suite.log
-          if-no-files-found: warn
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/warp.log
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/steps.md
+          if-no-files-found: error
 
       - name: Cleanup environment (after)
         if: ${{ always() && inputs.cleanup_after != 'false' }}

.github/workflows/rustfs-kms-test.yml

@@ -49,10 +49,28 @@ env:
 jobs:
   kms-test:
     runs-on: smoke-testing
-    continue-on-error: true
     timeout-minutes: 420
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     steps:
+      - name: Checkout repository (for report parser)
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Initialize functional evidence
+        id: evidence
+        run: |
+          set -euo pipefail
+          umask 077
+          FUNCTIONAL_ARTIFACTS_DIR="${RUNNER_TEMP}/rustfs-kms-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+          mkdir -- "${FUNCTIONAL_ARTIFACTS_DIR}" "${FUNCTIONAL_ARTIFACTS_DIR}-scratch"
+          {
+            printf 'FUNCTIONAL_ARTIFACTS_DIR=%s\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'LOG_FILE=%s/suite.log\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'REPORT_FILE=%s/report.md\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'TMPDIR=%s-scratch\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+          } >> "${GITHUB_ENV}"
+
       # auto-testing is private: clone it with the dedicated PF token (not
       # GITHUB_TOKEN) and retry transient GitHub/network failures.
       - name: Checkout auto-testing scripts (with retry)
@@ -109,9 +127,6 @@ jobs:
 
       - name: Run KMS suite
         id: test
-        continue-on-error: true
-        env:
-          LOG_FILE: /tmp/rustfs-kms.log
         run: |
           set -euo pipefail
           chmod +x auto-testing/rustfs-kms-test.sh
@@ -141,10 +156,7 @@ jobs:
           ./auto-testing/rustfs-kms-test.sh "${ARGS[@]}"
 
       - name: Generate report
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
-        env:
-          LOG_FILE: /tmp/rustfs-kms.log
-          REPORT_FILE: /tmp/rustfs-kms-report.md
         run: |
           set -euo pipefail
           PACKAGE_URL='${{ inputs.package_url }}'
@@ -156,79 +168,43 @@ jobs:
           else
             PACKAGE_SOURCE="${RUSTFS_NIGHTLY_PACKAGE_URL}"
           fi
-          CASE_TABLE="/tmp/rustfs-kms-cases.md"
+          CASE_TABLE="${FUNCTIONAL_ARTIFACTS_DIR}/cases.md"
-          python3 - "${LOG_FILE}" "${CASE_TABLE}" <<'PY'
+          CASE_RESULT=success
-          import re
+          python3 scripts/functional_case_report.py "${LOG_FILE}" "${CASE_TABLE}" || CASE_RESULT=failure
-          import sys
+          RESULT=failure
-
+          if [ '${{ steps.test.outcome }}' = 'success' ] && [ "${CASE_RESULT}" = 'success' ]; then
-          log_file, out_file = sys.argv[1], sys.argv[2]
+            RESULT=success
-          ansi = re.compile(r'\x1b\[[0-9;]*m')
+          fi
-          start_re = re.compile(r'^---\s+([A-Z]+-[0-9]+)\s+(.+?)\s+---$')
-          done_re = re.compile(r'^\[(PASS|FAIL|UNSUPPORTED)\]\s+([A-Z]+-[0-9]+)\b')
-
-          rows = []
-          index = {}
-          try:
-            with open(log_file, 'r', encoding='utf-8', errors='replace') as fh:
-              for raw in fh:
-                line = ansi.sub('', raw).strip()
-                m = start_re.match(line)
-                if m:
-                  case_id, name = m.group(1), m.group(2)
-                  if case_id not in index:
-                    index[case_id] = len(rows)
-                    rows.append([case_id, name, 'RUNNING'])
-                  continue
-                m = done_re.match(line)
-                if m:
-                  status, case_id = m.group(1), m.group(2)
-                  if case_id in index:
-                    rows[index[case_id]][2] = status
-                  else:
-                    rows.append([case_id, case_id, status])
-                    index[case_id] = len(rows) - 1
-          except FileNotFoundError:
-            rows = []
-
-          counts = {'PASS': 0, 'FAIL': 0, 'UNSUPPORTED': 0, 'RUNNING': 0}
-          for _, _, status in rows:
-            counts[status] = counts.get(status, 0) + 1
-
-          with open(out_file, 'w', encoding='utf-8') as out:
-            out.write('## Case Summary\n\n')
-            out.write(f"- Total: {len(rows)}\\n")
-            out.write(f"- PASS: {counts.get('PASS', 0)}\\n")
-            out.write(f"- FAIL: {counts.get('FAIL', 0)}\\n")
-            out.write(f"- UNSUPPORTED: {counts.get('UNSUPPORTED', 0)}\\n")
-            out.write('\\n')
-            out.write('| Case | Name | Status |\\n')
-            out.write('| --- | --- | --- |\\n')
-            for case_id, name, status in rows:
-              out.write(f'| {case_id} | {name} | {status} |\\n')
-          PY
           {
             echo "# RustFS KMS test report"
             echo ""
             echo "- Run: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${{ github.event_name }}"
             echo "- Package: ${PACKAGE_SOURCE}"
-            echo "- Test Step Outcome: ${{ steps.test.outcome }}"
+            echo "- Test Step Outcome: ${RESULT}"
+            echo "- Suite Step Outcome: ${{ steps.test.outcome }}"
             echo ""
-            cat "${CASE_TABLE}" || true
+            if [ "${RESULT}" = "success" ]; then
+              cat "${CASE_TABLE}"
               echo ""
               echo "## Log tail"
               echo '```text'
-            tail -n 200 "${LOG_FILE}" || true
+              tail -n 200 "${LOG_FILE}"
               echo '```'
+            else
+              echo "The suite or evidence validation failed. See this run's artifact for partial case results and suite.log."
+            fi
           } | tee "${REPORT_FILE}"
           cat "${REPORT_FILE}" >> "${GITHUB_STEP_SUMMARY}"
+          [ "${RESULT}" = "success" ]
 
       - name: Upload functional report to dashboard
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ env.PF_TESTING_GH_TOKEN }}
-          REPORT_FILE: /tmp/rustfs-kms-report.md
           SUITE: kms
         run: |
           set -euo pipefail
@@ -238,28 +214,32 @@ jobs:
           fi
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="functional-reports/${SUITE}/${DATE}.md"
-          CONTENT="$(python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
+          # Base64-encode the report into a temp file and feed it to jq via
+          # --rawfile: large reports (e.g. pool) exceed the OS argv limit and
+          # make `jq --arg content "${CONTENT}"` fail with "Argument list too long".
+          B64_FILE="$(mktemp)"
+          python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}" > "${B64_FILE}"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" --arg sha "${SHA}" \
-              '{message:$msg, content:$content, sha:$sha}' \
+              '{message:$msg, content:($content|rtrimstr("\n")), sha:$sha}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           else
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" \
-              '{message:$msg, content:$content}' \
+              '{message:$msg, content:($content|rtrimstr("\n"))}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           fi
+          rm -f "${B64_FILE}"
 
       - name: File failure issue in rustfs/backlog
         if: ${{ always() && (failure() || steps.test.outcome == 'failure' || steps.test.outcome == 'cancelled') }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
+          EVIDENCE_OUTCOME: ${{ steps.evidence.outcome }}
           SUITE: 'kms'
           SUITE_LABEL: 'KMS'
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-          REPORT_FILE: '/tmp/rustfs-kms-report.md'
-          LOG_FILE: '/tmp/rustfs-kms.log'
         run: |
           set -euo pipefail
           if [ -z "${GH_TOKEN:-}" ]; then
@@ -287,14 +267,16 @@ jobs:
             echo ""
             echo "- Suite: \`${SUITE}\`"
             echo "- Run: ${RUN_URL}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${GITHUB_EVENT_NAME}"
             echo "- Date: $(date -u +%Y-%m-%d)"
             echo ""
             echo "## Report (errors and symptoms)"
             echo ""
-            if [ -s "${REPORT_FILE}" ]; then
+            if [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${REPORT_FILE}" ]; then
               redact < "${REPORT_FILE}"
-            elif [ -s "${LOG_FILE:-}" ]; then
+            elif [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${LOG_FILE:-}" ]; then
               echo "(report file missing; log tail below)"
               echo ""
               tail -n 200 "${LOG_FILE}" | redact
@@ -310,14 +292,15 @@ jobs:
           echo "filed backlog issue for suite ${SUITE}"
 
       - name: Upload report and logs
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
         with:
-          name: rustfs-kms-test-${{ github.run_id }}
+          name: rustfs-kms-test-${{ github.run_id }}-${{ github.run_attempt }}
           path: |
-            /tmp/rustfs-kms.log
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/report.md
-            /tmp/rustfs-kms-report.md
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/suite.log
-          if-no-files-found: warn
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/cases.md
+          if-no-files-found: error
 
       - name: Cleanup environment (after)
         if: always()

.github/workflows/rustfs-performance-test.yml

@@ -49,17 +49,16 @@ on:
         type: boolean
         default: true
   repository_dispatch:
-    # Chain entry: dispatched by rustfs-functional-chain.yml (runs on its own
+    # Chain handoff: dispatched when the replication suite finishes.
-    # pf-testing runner, in parallel with the shared-VM chain).
     types: [rustfs-chain-performance]
 
 permissions:
   contents: read
 
-# Dedicated pf-testing runner/environment: own concurrency group so perf runs
+# The default performance nodes overlap the other suites' remote VMs, even
-# never block (or are blocked by) the pool-expansion / heal tests.
+# though the runner differs. Hold the shared lock through cleanup as well.
 concurrency:
-  group: rustfs-performance-test
+  group: rustfs-shared-functional-tests
   cancel-in-progress: false
 
 defaults:
@@ -76,22 +75,33 @@ env:
   # Package used by the nightly run (workflow_dispatch inputs are empty for
   # workflow_run events), i.e. the latest nightly deb published by nightly-gnu.yml.
   RUSTFS_NIGHTLY_PACKAGE_URL: ${{ vars.RUSTFS_NIGHTLY_PACKAGE_URL || 'https://dl.rustfs.com/artifacts/rustfs/packages/nightly/rustfs-nightly-latest.deb' }}
-  # Fixed benchmark result directory so later steps can read summary.md
-  RUSTFS_RESULT_DIR: /tmp/rustfs-perf-results
   # Cross-repo token for uploading reports to rustfs/dashboard (set in repo settings)
   PF_TESTING_GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
 
 jobs:
   performance-test:
     runs-on: pf-testing
-    # Requirement: a failing benchmark must not fail the workflow;
-    # failures are filed to rustfs/backlog.
-    continue-on-error: true
     timeout-minutes: 900
     # Run on manual dispatch, or when the nightly build completed successfully.
     # Skipped when nightly failed.
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     steps:
+      - name: Initialize functional evidence
+        id: evidence
+        run: |
+          set -euo pipefail
+          umask 077
+          FUNCTIONAL_ARTIFACTS_DIR="${RUNNER_TEMP}/rustfs-performance-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+          mkdir -- "${FUNCTIONAL_ARTIFACTS_DIR}" "${FUNCTIONAL_ARTIFACTS_DIR}-scratch"
+          {
+            printf 'FUNCTIONAL_ARTIFACTS_DIR=%s\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'LOG_FILE=%s/suite.log\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'REPORT_FILE=%s/report.md\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'TMPDIR=%s-scratch\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'RUSTFS_RESULT_DIR=%s/results\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'VERSION_FILE=%s/version.txt\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+          } >> "${GITHUB_ENV}"
+
       # auto-testing is private: clone it with the dedicated PF token (not
       # GITHUB_TOKEN) and retry transient GitHub/network failures.
       - name: Checkout auto-testing scripts (with retry)
@@ -123,7 +133,7 @@ jobs:
         if: ${{ inputs.cleanup_before != 'false' }}
         run: |
           chmod +x auto-testing/rustfs_performance_test.sh
-          ./auto-testing/rustfs_performance_test.sh --step 1 -y
+          ./auto-testing/rustfs_performance_test.sh --step 1 -y --log-file "${LOG_FILE:-/dev/null}"
 
       - name: Install RustFS package & start cluster (4x4)
         run: |
@@ -133,7 +143,7 @@ jobs:
           else
             ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
           fi
-          ./auto-testing/rustfs_performance_test.sh "${ARGS[@]}"
+          ./auto-testing/rustfs_performance_test.sh "${ARGS[@]}" --log-file "${LOG_FILE}"
 
       - name: Preflight checks
         run: |
@@ -143,7 +153,7 @@ jobs:
           else
             ARGS+=(--package-url "${{ env.RUSTFS_NIGHTLY_PACKAGE_URL }}")
           fi
-          ./auto-testing/rustfs_performance_test.sh "${ARGS[@]}"
+          ./auto-testing/rustfs_performance_test.sh "${ARGS[@]}" --log-file "${LOG_FILE}"
 
       - name: Run benchmark (GET/PUT/MIXED)
         id: benchmark
@@ -156,17 +166,15 @@ jobs:
             --step 5 -y \
             --warp-duration "${{ inputs.warp_duration || '5m' }}" \
             --warp-concurrency "${{ inputs.warp_concurrency || '64' }}" \
-            --log-file /tmp/rustfs-perf-test.log
+            --log-file "${LOG_FILE}"
 
       - name: Analyze results
         if: ${{ steps.benchmark.conclusion == 'success' }}
         run: |
-          ./auto-testing/rustfs_performance_test.sh --step 6 -y
+          ./auto-testing/rustfs_performance_test.sh --step 6 -y --log-file "${LOG_FILE:-/dev/null}"
 
       - name: Collect RustFS version info
         if: ${{ steps.benchmark.conclusion == 'success' }}
-        env:
-          VERSION_FILE: /tmp/rustfs-version.txt
         run: |
           set -euo pipefail
           read -r -a NODES <<< "${RUSTFS_NODES}"
@@ -186,7 +194,6 @@ jobs:
         env:
           GH_TOKEN: ${{ env.PF_TESTING_GH_TOKEN }}
           RESULT_DIR: ${{ env.RUSTFS_RESULT_DIR }}
-          VERSION_FILE: /tmp/rustfs-version.txt
         run: |
           set -euo pipefail
           if [ -z "${GH_TOKEN:-}" ]; then
@@ -194,7 +201,7 @@ jobs:
             exit 0
           fi
           SUMMARY="${RESULT_DIR}/summary.md"
-          [ -f "${SUMMARY}" ] || { echo "summary.md not found at ${SUMMARY}"; exit 1; }
+          [ -s "${SUMMARY}" ] || { echo "summary.md not found at ${SUMMARY}"; exit 1; }
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="reports/${DATE}.md"
           {
@@ -202,6 +209,8 @@ jobs:
             echo ""
             echo "- **Date**: ${DATE}"
             echo "- **Run**: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+            echo "- **Attempt**: ${GITHUB_RUN_ATTEMPT}"
+            echo "- **Workflow Commit**: ${GITHUB_SHA}"
             echo "- **Trigger**: ${{ github.event_name }}"
             echo "- **Package**: ${{ inputs.package_url || 'nightly (R2 latest)' }}"
             echo ""
@@ -211,8 +220,8 @@ jobs:
             echo '```text'
             cat "${VERSION_FILE}"
             echo '```'
-          } > /tmp/rustfs-perf-report.md
+          } > "${REPORT_FILE}"
-          CONTENT="$(python3 -c 'import base64; print(base64.b64encode(open("/tmp/rustfs-perf-report.md","rb").read()).decode())')"
+          CONTENT="$(python3 -c 'import base64,sys; print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
             jq -n --arg msg "report: ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
@@ -231,11 +240,10 @@ jobs:
         continue-on-error: true
         env:
           GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
+          EVIDENCE_OUTCOME: ${{ steps.evidence.outcome }}
           SUITE: 'performance'
           SUITE_LABEL: 'Performance'
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-          REPORT_FILE: '/tmp/rustfs-perf-report.md'
-          LOG_FILE: '/tmp/rustfs-perf-test.log'
         run: |
           set -euo pipefail
           if [ -z "${GH_TOKEN:-}" ]; then
@@ -263,14 +271,16 @@ jobs:
             echo ""
             echo "- Suite: \`${SUITE}\`"
             echo "- Run: ${RUN_URL}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${GITHUB_EVENT_NAME}"
             echo "- Date: $(date -u +%Y-%m-%d)"
             echo ""
             echo "## Report (errors and symptoms)"
             echo ""
-            if [ -s "${REPORT_FILE}" ]; then
+            if [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${REPORT_FILE}" ]; then
               redact < "${REPORT_FILE}"
-            elif [ -s "${LOG_FILE:-}" ]; then
+            elif [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${LOG_FILE:-}" ]; then
               echo "(report file missing; log tail below)"
               echo ""
               tail -n 200 "${LOG_FILE}" | redact
@@ -286,20 +296,26 @@ jobs:
           echo "filed backlog issue for suite ${SUITE}"
 
       - name: Upload test logs & results
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
         with:
-          name: rustfs-perf-test-${{ github.run_id }}
+          name: rustfs-perf-test-${{ github.run_id }}-${{ github.run_attempt }}
           path: |
-            /tmp/rustfs-perf-test*.log
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/report.md
-            /tmp/rustfs-perf-results/**
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/suite.log
-            /tmp/rustfs-version.txt
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/version.txt
-          if-no-files-found: warn
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/results/master.log
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/results/summary.md
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/results/summary.tsv
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/results/get_*.txt
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/results/put_*.txt
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/results/mixed_*.txt
+          if-no-files-found: error
 
       - name: Reset test environment (after)
         if: ${{ always() && inputs.cleanup_after != 'false' }}
         run: |
-          ./auto-testing/rustfs_performance_test.sh --step 7 -y
+          ./auto-testing/rustfs_performance_test.sh --step 7 -y --log-file "${LOG_FILE:-/dev/null}"
 
       - name: Notify on failure
         if: failure()

.github/workflows/rustfs-pool-expand-test.yml

@@ -76,9 +76,6 @@ jobs:
   pool-expansion-test:
     name: Pool expansion / decommission test
     runs-on: smoke-testing
-    # Requirement: a failing suite must not fail the workflow; failures
-    # are filed to rustfs/backlog and the chain continues.
-    continue-on-error: true
     timeout-minutes: 360
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     env:
@@ -542,17 +539,22 @@ jobs:
           fi
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="functional-reports/${SUITE}/${DATE}.md"
-          CONTENT="$(python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
+          # Base64-encode the report into a temp file and feed it to jq via
+          # --rawfile: large reports (e.g. pool) exceed the OS argv limit and
+          # make `jq --arg content "${CONTENT}"` fail with "Argument list too long".
+          B64_FILE="$(mktemp)"
+          python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}" > "${B64_FILE}"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" --arg sha "${SHA}" \
-              '{message:$msg, content:$content, sha:$sha}' \
+              '{message:$msg, content:($content|rtrimstr("\n")), sha:$sha}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           else
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" \
-              '{message:$msg, content:$content}' \
+              '{message:$msg, content:($content|rtrimstr("\n"))}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           fi
+          rm -f "${B64_FILE}"
 
       - name: File failure issue in rustfs/backlog
         if: ${{ always() && (failure() || steps.pool_test.outcome == 'failure' || steps.pool_test.outcome == 'cancelled') }}

.github/workflows/rustfs-replication-test.yml

@@ -34,8 +34,7 @@ on:
           - site
         default: all
   repository_dispatch:
-    # Chain handoff: dispatched when the security suite finishes. This is the
+    # Chain handoff: dispatched when the security suite finishes.
-    # last link of the functional chain.
     types: [rustfs-chain-replication]
 
 permissions:
@@ -62,12 +61,28 @@ env:
 jobs:
   replication-test:
     runs-on: smoke-testing
-    # A failed replication run must not break the chain or the workflow: the
-    # failure is reported to rustfs/backlog instead (see the issue step).
-    continue-on-error: true
     timeout-minutes: 360
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     steps:
+      - name: Checkout repository (for report parser)
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Initialize functional evidence
+        id: evidence
+        run: |
+          set -euo pipefail
+          umask 077
+          FUNCTIONAL_ARTIFACTS_DIR="${RUNNER_TEMP}/rustfs-replication-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+          mkdir -- "${FUNCTIONAL_ARTIFACTS_DIR}" "${FUNCTIONAL_ARTIFACTS_DIR}-scratch"
+          {
+            printf 'FUNCTIONAL_ARTIFACTS_DIR=%s\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'LOG_FILE=%s/suite.log\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'REPORT_FILE=%s/report.md\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'TMPDIR=%s-scratch\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+          } >> "${GITHUB_ENV}"
+
       # auto-testing is private: clone it with the dedicated PF token (not
       # GITHUB_TOKEN) and retry transient GitHub/network failures.
       - name: Checkout auto-testing scripts (with retry)
@@ -116,9 +131,6 @@ jobs:
 
       - name: Run replication suite
         id: test
-        continue-on-error: true
-        env:
-          LOG_FILE: /tmp/rustfs-replication.log
         run: |
           set -euo pipefail
           chmod +x auto-testing/rustfs-replication-test.sh
@@ -141,10 +153,7 @@ jobs:
           ./auto-testing/rustfs-replication-test.sh "${ARGS[@]}"
 
       - name: Generate report
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
-        env:
-          LOG_FILE: /tmp/rustfs-replication.log
-          REPORT_FILE: /tmp/rustfs-replication-report.md
         run: |
           set -euo pipefail
           PACKAGE_URL='${{ inputs.package_url }}'
@@ -166,80 +175,44 @@ jobs:
               RUSTFS_VERSION_INFO="${DETECTED_VERSION}"
             fi
           fi
-          CASE_TABLE="/tmp/rustfs-replication-cases.md"
+          CASE_TABLE="${FUNCTIONAL_ARTIFACTS_DIR}/cases.md"
-          python3 - "${LOG_FILE}" "${CASE_TABLE}" <<'PY'
+          CASE_RESULT=success
-          import re
+          python3 scripts/functional_case_report.py "${LOG_FILE}" "${CASE_TABLE}" || CASE_RESULT=failure
-          import sys
+          RESULT=failure
-
+          if [ '${{ steps.test.outcome }}' = 'success' ] && [ "${CASE_RESULT}" = 'success' ]; then
-          log_file, out_file = sys.argv[1], sys.argv[2]
+            RESULT=success
-          ansi = re.compile(r'\x1b\[[0-9;]*m')
+          fi
-          start_re = re.compile(r'^---\s+([A-Z0-9]+-[0-9]+)\s+(.+?)\s+---$')
-          done_re = re.compile(r'^\[(PASS|FAIL|UNSUPPORTED)\]\s+([A-Z0-9]+-[0-9]+)\b')
-
-          rows = []
-          index = {}
-          try:
-              with open(log_file, 'r', encoding='utf-8', errors='replace') as fh:
-                  for raw in fh:
-                      line = ansi.sub('', raw).strip()
-                      m = start_re.match(line)
-                      if m:
-                          case_id, name = m.group(1), m.group(2)
-                          if case_id not in index:
-                              index[case_id] = len(rows)
-                              rows.append([case_id, name, 'RUNNING'])
-                          continue
-                      m = done_re.match(line)
-                      if m:
-                          status, case_id = m.group(1), m.group(2)
-                          if case_id in index:
-                              rows[index[case_id]][2] = status
-                          else:
-                              rows.append([case_id, case_id, status])
-                              index[case_id] = len(rows) - 1
-          except FileNotFoundError:
-              rows = []
-
-          counts = {'PASS': 0, 'FAIL': 0, 'UNSUPPORTED': 0, 'RUNNING': 0}
-          for _, _, status in rows:
-              counts[status] = counts.get(status, 0) + 1
-
-          with open(out_file, 'w', encoding='utf-8') as out:
-              out.write('## Case Summary\n\n')
-              out.write(f"- Total: {len(rows)}\\n")
-              out.write(f"- PASS: {counts.get('PASS', 0)}\\n")
-              out.write(f"- FAIL: {counts.get('FAIL', 0)}\\n")
-              out.write(f"- UNSUPPORTED: {counts.get('UNSUPPORTED', 0)}\\n")
-              out.write('\\n')
-              out.write('| Case | Name | Status |\\n')
-              out.write('| --- | --- | --- |\\n')
-              for case_id, name, status in rows:
-                  out.write(f'| {case_id} | {name} | {status} |\\n')
-          PY
           {
             echo "# RustFS replication test report"
             echo ""
             echo "- Run: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${{ github.event_name }}"
             echo "- Package: ${PACKAGE_SOURCE}"
             echo "- RustFS Version: ${RUSTFS_VERSION_INFO}"
-            echo "- Test Step Outcome: ${{ steps.test.outcome }}"
+            echo "- Test Step Outcome: ${RESULT}"
+            echo "- Suite Step Outcome: ${{ steps.test.outcome }}"
             echo ""
-            cat "${CASE_TABLE}" || true
+            if [ "${RESULT}" = "success" ]; then
+              cat "${CASE_TABLE}"
               echo ""
               echo "## Log tail"
               echo '```text'
-            tail -n 200 "${LOG_FILE}" || true
+              tail -n 200 "${LOG_FILE}"
               echo '```'
+            else
+              echo "The suite or evidence validation failed. See this run's artifact for partial case results and suite.log."
+            fi
           } | tee "${REPORT_FILE}"
           cat "${REPORT_FILE}" >> "${GITHUB_STEP_SUMMARY}"
+          [ "${RESULT}" = "success" ]
 
       - name: Upload functional report to dashboard
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ env.PF_TESTING_GH_TOKEN }}
-          REPORT_FILE: /tmp/rustfs-replication-report.md
           SUITE: replication
         run: |
           set -euo pipefail
@@ -249,28 +222,32 @@ jobs:
           fi
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="functional-reports/${SUITE}/${DATE}.md"
-          CONTENT="$(python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
+          # Base64-encode the report into a temp file and feed it to jq via
+          # --rawfile: large reports (e.g. pool) exceed the OS argv limit and
+          # make `jq --arg content "${CONTENT}"` fail with "Argument list too long".
+          B64_FILE="$(mktemp)"
+          python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}" > "${B64_FILE}"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" --arg sha "${SHA}" \
-              '{message:$msg, content:$content, sha:$sha}' \
+              '{message:$msg, content:($content|rtrimstr("\n")), sha:$sha}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           else
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" \
-              '{message:$msg, content:$content}' \
+              '{message:$msg, content:($content|rtrimstr("\n"))}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           fi
+          rm -f "${B64_FILE}"
 
       - name: File failure issue in rustfs/backlog
         if: ${{ always() && (failure() || steps.test.outcome == 'failure' || steps.test.outcome == 'cancelled') }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
+          EVIDENCE_OUTCOME: ${{ steps.evidence.outcome }}
           SUITE: 'replication'
           SUITE_LABEL: 'Replication (bucket + site)'
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-          REPORT_FILE: '/tmp/rustfs-replication-report.md'
-          LOG_FILE: '/tmp/rustfs-replication.log'
         run: |
           set -euo pipefail
           if [ -z "${GH_TOKEN:-}" ]; then
@@ -298,14 +275,16 @@ jobs:
             echo ""
             echo "- Suite: \`${SUITE}\`"
             echo "- Run: ${RUN_URL}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${GITHUB_EVENT_NAME}"
             echo "- Date: $(date -u +%Y-%m-%d)"
             echo ""
             echo "## Report (errors and symptoms)"
             echo ""
-            if [ -s "${REPORT_FILE}" ]; then
+            if [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${REPORT_FILE}" ]; then
               redact < "${REPORT_FILE}"
-            elif [ -s "${LOG_FILE:-}" ]; then
+            elif [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${LOG_FILE:-}" ]; then
               echo "(report file missing; log tail below)"
               echo ""
               tail -n 200 "${LOG_FILE}" | redact
@@ -321,14 +300,15 @@ jobs:
           echo "filed backlog issue for suite ${SUITE}"
 
       - name: Upload report and logs
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
         with:
-          name: rustfs-replication-${{ github.run_id }}
+          name: rustfs-replication-${{ github.run_id }}-${{ github.run_attempt }}
           path: |
-            /tmp/rustfs-replication.log
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/report.md
-            /tmp/rustfs-replication-report.md
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/suite.log
-          if-no-files-found: warn
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/cases.md
+          if-no-files-found: error
 
       - name: Cleanup environment (after)
         if: always()
@@ -349,13 +329,50 @@ jobs:
             '
           done
 
-      - name: Chain complete
+      - name: "Continue functional chain (next: Performance)"
-        # Replication is the last link of the functional chain: nothing to
-        # dispatch after it. This step just records that the chain finished.
         if: ${{ always() && github.event_name == 'repository_dispatch' }}
+        env:
+          GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
         run: |
-          echo "Functional chain complete: replication (final suite) finished."
+          set -uo pipefail
-          echo "from_suite=security trigger=${{ github.event_name }} outcome=${{ steps.test.outcome }}"
+          if [ -z "${GH_TOKEN:-}" ]; then
+            echo "PF_TESTING_GH_TOKEN is not configured; cannot dispatch the next suite" >&2
+            exit 1
+          fi
+          DISPATCHED=0
+          for attempt in 1 2 3; do
+            if gh api --method POST repos/rustfs/rustfs/dispatches \
+              -f event_type='rustfs-chain-performance' \
+              -F 'client_payload[from_suite]=replication'; then
+              echo "dispatched next suite Performance (attempt ${attempt})"
+              DISPATCHED=1
+              break
+            fi
+            echo "dispatch attempt ${attempt} failed; retrying in ${attempt}0s" >&2
+            sleep "${attempt}0"
+          done
+          if [ "${DISPATCHED:-0}" -ne 1 ]; then
+            echo "ERROR: functional chain stalled: could not dispatch Performance after 3 attempts" >&2
+            TITLE="[functional][chain] stalled after replication (run ${GITHUB_RUN_ID})"
+            BODY_FILE="$(mktemp)"
+            trap 'rm -f "${BODY_FILE}"' EXIT
+            {
+              echo "The functional chain could not hand off from **replication** to **Performance** after 3 attempts."
+              echo ""
+              echo "- Failed suite job: ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/actions/runs/${GITHUB_RUN_ID}"
+              echo "- Expected next event: 'rustfs-chain-performance'"
+              echo "- Likely cause: PF_TESTING_GH_TOKEN lacks contents:write on rustfs/rustfs, or the GitHub API was unavailable."
+              echo "- Recovery: re-dispatch manually with"
+              FENCE="$(printf "\x60\x60\x60")"; echo "  ${FENCE}"
+              echo "  gh api --method POST repos/rustfs/rustfs/dispatches -f event_type='rustfs-chain-performance'"
+              FENCE="$(printf "\x60\x60\x60")"; echo "  ${FENCE}"
+            } > "${BODY_FILE}"
+            gh issue create -R rustfs/backlog --title "${TITLE}" \
+              --body-file "${BODY_FILE}" --label functional-test \
+              || gh issue create -R rustfs/backlog --title "${TITLE}" --body-file "${BODY_FILE}" \
+              || echo "could not file the stall alert issue either; check the token" >&2
+            exit 1
+          fi
 
       - name: Notify on failure
         if: failure()

.github/workflows/rustfs-s3-compat-test.yml

@@ -37,10 +37,28 @@ env:
 jobs:
   s3-compat-test:
     runs-on: smoke-testing
-    continue-on-error: true
     timeout-minutes: 360
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     steps:
+      - name: Checkout repository (for report parser)
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Initialize functional evidence
+        id: evidence
+        run: |
+          set -euo pipefail
+          umask 077
+          FUNCTIONAL_ARTIFACTS_DIR="${RUNNER_TEMP}/rustfs-s3-compat-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+          mkdir -- "${FUNCTIONAL_ARTIFACTS_DIR}" "${FUNCTIONAL_ARTIFACTS_DIR}-scratch"
+          {
+            printf 'FUNCTIONAL_ARTIFACTS_DIR=%s\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'LOG_FILE=%s/suite.log\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'REPORT_FILE=%s/report.md\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'TMPDIR=%s-scratch\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+          } >> "${GITHUB_ENV}"
+
       # auto-testing is private: clone it with the dedicated PF token (not
       # GITHUB_TOKEN) and retry transient GitHub/network failures.
       - name: Checkout auto-testing scripts (with retry)
@@ -88,9 +106,6 @@ jobs:
 
       - name: Run S3 compatibility suite
         id: test
-        continue-on-error: true
-        env:
-          LOG_FILE: /tmp/rustfs-s3-compat.log
         run: |
           set -euo pipefail
           chmod +x auto-testing/rustfs-s3-compat-test.sh
@@ -107,10 +122,7 @@ jobs:
           ./auto-testing/rustfs-s3-compat-test.sh "${ARGS[@]}"
 
       - name: Generate report
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
-        env:
-          LOG_FILE: /tmp/rustfs-s3-compat.log
-          REPORT_FILE: /tmp/rustfs-s3-compat-report.md
         run: |
           set -euo pipefail
           PACKAGE_URL='${{ inputs.package_url }}'
@@ -132,83 +144,44 @@ jobs:
               RUSTFS_VERSION_INFO="${DETECTED_VERSION}"
             fi
           fi
-          CASE_TABLE="/tmp/rustfs-s3-compat-cases.md"
+          CASE_TABLE="${FUNCTIONAL_ARTIFACTS_DIR}/cases.md"
-          python3 - "${LOG_FILE}" "${CASE_TABLE}" <<'PY'
+          CASE_RESULT=success
-          import re
+          python3 scripts/functional_case_report.py "${LOG_FILE}" "${CASE_TABLE}" || CASE_RESULT=failure
-          import sys
+          RESULT=failure
-
+          if [ '${{ steps.test.outcome }}' = 'success' ] && [ "${CASE_RESULT}" = 'success' ]; then
-          log_file, out_file = sys.argv[1], sys.argv[2]
+            RESULT=success
-          ansi = re.compile(r'\x1b\[[0-9;]*m')
+          fi
-          start_re = re.compile(r'^---\s+([A-Z0-9]+-[0-9]+)\s+(.+?)\s+---$')
-          done_re = re.compile(r'^\[(PASS|FAIL|UNSUPPORTED)\]\s+([A-Z0-9]+-[0-9]+)\b')
-
-          rows = []
-          index = {}
-          current = None
-          try:
-            with open(log_file, 'r', encoding='utf-8', errors='replace') as fh:
-              for raw in fh:
-                line = ansi.sub('', raw).strip()
-                m = start_re.match(line)
-                if m:
-                  case_id, name = m.group(1), m.group(2)
-                  current = case_id
-                  if case_id not in index:
-                    index[case_id] = len(rows)
-                    rows.append([case_id, name, 'RUNNING'])
-                  continue
-                m = done_re.match(line)
-                if m:
-                  status, case_id = m.group(1), m.group(2)
-                  if case_id in index:
-                    rows[index[case_id]][2] = status
-                  else:
-                    rows.append([case_id, case_id, status])
-                    index[case_id] = len(rows) - 1
-                  current = None
-          except FileNotFoundError:
-            rows = []
-
-          counts = {'PASS': 0, 'FAIL': 0, 'UNSUPPORTED': 0, 'RUNNING': 0}
-          for _, _, status in rows:
-            counts[status] = counts.get(status, 0) + 1
-
-          with open(out_file, 'w', encoding='utf-8') as out:
-            out.write('## Case Summary\n\n')
-            out.write(f"- Total: {len(rows)}\\n")
-            out.write(f"- PASS: {counts.get('PASS', 0)}\\n")
-            out.write(f"- FAIL: {counts.get('FAIL', 0)}\\n")
-            out.write(f"- UNSUPPORTED: {counts.get('UNSUPPORTED', 0)}\\n")
-            out.write('\\n')
-            out.write('| Case | Name | Status |\\n')
-            out.write('| --- | --- | --- |\\n')
-            for case_id, name, status in rows:
-              out.write(f'| {case_id} | {name} | {status} |\\n')
-          PY
           {
             echo "# RustFS S3 compatibility test report"
             echo ""
             echo "- Run: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${{ github.event_name }}"
             echo "- Package: ${PACKAGE_SOURCE}"
             echo "- RustFS Version: ${RUSTFS_VERSION_INFO}"
-            echo "- Test Step Outcome: ${{ steps.test.outcome }}"
+            echo "- Test Step Outcome: ${RESULT}"
+            echo "- Suite Step Outcome: ${{ steps.test.outcome }}"
             echo ""
-            cat "${CASE_TABLE}" || true
+            if [ "${RESULT}" = "success" ]; then
+              cat "${CASE_TABLE}"
               echo ""
               echo "## Log tail"
               echo '```text'
-            tail -n 200 "${LOG_FILE}" || true
+              tail -n 200 "${LOG_FILE}"
               echo '```'
+            else
+              echo "The suite or evidence validation failed. See this run's artifact for partial case results and suite.log."
+            fi
           } | tee "${REPORT_FILE}"
           cat "${REPORT_FILE}" >> "${GITHUB_STEP_SUMMARY}"
+          [ "${RESULT}" = "success" ]
 
       - name: Upload functional report to dashboard
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ env.PF_TESTING_GH_TOKEN }}
-          REPORT_FILE: /tmp/rustfs-s3-compat-report.md
           SUITE: s3
         run: |
           set -euo pipefail
@@ -218,28 +191,32 @@ jobs:
           fi
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="functional-reports/${SUITE}/${DATE}.md"
-          CONTENT="$(python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
+          # Base64-encode the report into a temp file and feed it to jq via
+          # --rawfile: large reports (e.g. pool) exceed the OS argv limit and
+          # make `jq --arg content "${CONTENT}"` fail with "Argument list too long".
+          B64_FILE="$(mktemp)"
+          python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}" > "${B64_FILE}"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" --arg sha "${SHA}" \
-              '{message:$msg, content:$content, sha:$sha}' \
+              '{message:$msg, content:($content|rtrimstr("\n")), sha:$sha}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           else
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" \
-              '{message:$msg, content:$content}' \
+              '{message:$msg, content:($content|rtrimstr("\n"))}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           fi
+          rm -f "${B64_FILE}"
 
       - name: File failure issue in rustfs/backlog
         if: ${{ always() && (failure() || steps.test.outcome == 'failure' || steps.test.outcome == 'cancelled') }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
+          EVIDENCE_OUTCOME: ${{ steps.evidence.outcome }}
           SUITE: 's3'
           SUITE_LABEL: 'S3 compatibility'
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-          REPORT_FILE: '/tmp/rustfs-s3-compat-report.md'
-          LOG_FILE: '/tmp/rustfs-s3-compat.log'
         run: |
           set -euo pipefail
           if [ -z "${GH_TOKEN:-}" ]; then
@@ -267,14 +244,16 @@ jobs:
             echo ""
             echo "- Suite: \`${SUITE}\`"
             echo "- Run: ${RUN_URL}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${GITHUB_EVENT_NAME}"
             echo "- Date: $(date -u +%Y-%m-%d)"
             echo ""
             echo "## Report (errors and symptoms)"
             echo ""
-            if [ -s "${REPORT_FILE}" ]; then
+            if [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${REPORT_FILE}" ]; then
               redact < "${REPORT_FILE}"
-            elif [ -s "${LOG_FILE:-}" ]; then
+            elif [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${LOG_FILE:-}" ]; then
               echo "(report file missing; log tail below)"
               echo ""
               tail -n 200 "${LOG_FILE}" | redact
@@ -290,14 +269,15 @@ jobs:
           echo "filed backlog issue for suite ${SUITE}"
 
       - name: Upload report and logs
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
         with:
-          name: rustfs-s3-compat-${{ github.run_id }}
+          name: rustfs-s3-compat-${{ github.run_id }}-${{ github.run_attempt }}
           path: |
-            /tmp/rustfs-s3-compat.log
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/report.md
-            /tmp/rustfs-s3-compat-report.md
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/suite.log
-          if-no-files-found: warn
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/cases.md
+          if-no-files-found: error
 
       - name: Cleanup environment (after)
         if: always()

.github/workflows/rustfs-security-test.yml

@@ -74,10 +74,27 @@ env:
 jobs:
   security-test:
     runs-on: smoke-testing
-    continue-on-error: true
     timeout-minutes: 360
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     steps:
+      # Checkout the repository into its own subdirectory. Checking out at
+      # the workspace root would wipe the auto-testing clone above (that is
+      # exactly how run 33934141181 lost rustfs-security-test.sh).
+      - name: Checkout repository (for the OIDC live gate script)
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+          path: rustfs-repo
+
+      - name: Initialize security evidence
+        id: evidence
+        run: |
+          set -euo pipefail
+          umask 077
+          SECURITY_ARTIFACTS_DIR="${RUNNER_TEMP}/rustfs-security-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+          mkdir -- "${SECURITY_ARTIFACTS_DIR}" "${SECURITY_ARTIFACTS_DIR}-scratch"
+          printf 'SECURITY_ARTIFACTS_DIR=%s\n' "${SECURITY_ARTIFACTS_DIR}" >> "${GITHUB_ENV}"
+
       # auto-testing is private: clone it with the dedicated PF token (not
       # GITHUB_TOKEN) and retry transient GitHub/network failures.
       - name: Checkout auto-testing scripts (with retry)
@@ -98,11 +115,6 @@ jobs:
           echo "ERROR: unable to clone rustfs/auto-testing after 5 attempts" >&2
           exit 1
 
-      - name: Checkout repository (for the OIDC live gate script)
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
-        with:
-          persist-credentials: false
-
       - name: Show environment
         run: |
           uname -a
@@ -135,8 +147,9 @@ jobs:
         id: test
         continue-on-error: true
         env:
-          REPORT_FILE: /tmp/rustfs-security-report.md
+          REPORT_FILE: ${{ env.SECURITY_ARTIFACTS_DIR }}/suite-report.md
-          RUSTFS_SECURITY_OIDC_LIVE_SCRIPT: ${{ github.workspace }}/scripts/test/oidc_keycloak_live.sh
+          TMPDIR: ${{ env.SECURITY_ARTIFACTS_DIR }}-scratch
+          RUSTFS_SECURITY_OIDC_LIVE_SCRIPT: ${{ github.workspace }}/rustfs-repo/scripts/test/oidc_keycloak_live.sh
         run: |
           set -euo pipefail
           chmod +x auto-testing/rustfs-security-test.sh
@@ -159,29 +172,48 @@ jobs:
           else
             ARGS+=(--package-url "${RUSTFS_NIGHTLY_PACKAGE_URL}")
           fi
-          ./auto-testing/rustfs-security-test.sh "${ARGS[@]}"
+          GITHUB_STEP_SUMMARY=/dev/null ./auto-testing/rustfs-security-test.sh "${ARGS[@]}" 2>&1 | tee "${SECURITY_ARTIFACTS_DIR}/suite.log"
 
       - name: Generate report
-        if: always()
+        id: report
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
+        env:
+          TEST_OUTCOME: ${{ steps.test.outcome }}
         run: |
           set -euo pipefail
-          if [ ! -f /tmp/rustfs-security-report.md ]; then
+          RESULT=failure
+          if [ "${TEST_OUTCOME}" = "success" ] && [ -s "${SECURITY_ARTIFACTS_DIR}/suite-report.md" ]; then
+            RESULT=success
+          fi
           {
             echo "# RustFS security test report"
             echo ""
-              echo "- Run: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+            echo "- Run: ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/actions/runs/${GITHUB_RUN_ID}"
-              echo "- Trigger: ${{ github.event_name }}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
-              echo "- Test Step Outcome: failure (suite did not produce a report)"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
-            } > /tmp/rustfs-security-report.md
+            echo "- Trigger: ${GITHUB_EVENT_NAME}"
+            echo "- Test Step Outcome: ${RESULT}"
+            echo "- Suite Step Outcome: ${TEST_OUTCOME}"
+            echo ""
+            # The dashboard prioritizes case rows over the step outcome.
+            # Keep partial case results in the artifact when the suite fails.
+            if [ "${RESULT}" = "success" ]; then
+              cat "${SECURITY_ARTIFACTS_DIR}/suite-report.md"
+            elif [ -s "${SECURITY_ARTIFACTS_DIR}/suite-report.md" ]; then
+              echo "The suite did not complete successfully. See suite-report.md in this run's artifact for diagnostics."
+            else
+              echo "The suite did not produce a non-empty report."
             fi
-          cat /tmp/rustfs-security-report.md >> "${GITHUB_STEP_SUMMARY}"
+          } > "${SECURITY_ARTIFACTS_DIR}/report.md"
+          cat "${SECURITY_ARTIFACTS_DIR}/report.md" >> "${GITHUB_STEP_SUMMARY}"
+          [ "${RESULT}" = "success" ]
 
       - name: Upload functional report to dashboard
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ env.PF_TESTING_GH_TOKEN }}
-          REPORT_FILE: /tmp/rustfs-security-report.md
+          REPORT_FILE: ${{ env.SECURITY_ARTIFACTS_DIR }}/report.md
           SUITE: security
         run: |
           set -euo pipefail
@@ -191,17 +223,22 @@ jobs:
           fi
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="functional-reports/${SUITE}/${DATE}.md"
-          CONTENT="$(python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
+          # Base64-encode the report into a temp file and feed it to jq via
+          # --rawfile: large reports (e.g. pool) exceed the OS argv limit and
+          # make `jq --arg content "${CONTENT}"` fail with "Argument list too long".
+          B64_FILE="$(mktemp)"
+          python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}" > "${B64_FILE}"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" --arg sha "${SHA}" \
-              '{message:$msg, content:$content, sha:$sha}' \
+              '{message:$msg, content:($content|rtrimstr("\n")), sha:$sha}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           else
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" \
-              '{message:$msg, content:$content}' \
+              '{message:$msg, content:($content|rtrimstr("\n"))}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           fi
+          rm -f "${B64_FILE}"
 
       - name: File failure issue in rustfs/backlog
         if: ${{ always() && (failure() || steps.test.outcome == 'failure' || steps.test.outcome == 'cancelled') }}
@@ -210,8 +247,9 @@ jobs:
           GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
           SUITE: 'security'
           SUITE_LABEL: 'Security'
+          EVIDENCE_OUTCOME: ${{ steps.evidence.outcome }}
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-          REPORT_FILE: '/tmp/rustfs-security-report.md'
+          REPORT_FILE: ${{ env.SECURITY_ARTIFACTS_DIR }}/report.md
           LOG_FILE: ''
         run: |
           set -euo pipefail
@@ -245,7 +283,7 @@ jobs:
             echo ""
             echo "## Report (errors and symptoms)"
             echo ""
-            if [ -s "${REPORT_FILE}" ]; then
+            if [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${REPORT_FILE}" ]; then
               redact < "${REPORT_FILE}"
             elif [ -s "${LOG_FILE:-}" ]; then
               echo "(report file missing; log tail below)"
@@ -263,14 +301,15 @@ jobs:
           echo "filed backlog issue for suite ${SUITE}"
 
       - name: Upload report and logs
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
         with:
-          name: rustfs-security-test-${{ github.run_id }}
+          name: rustfs-security-test-${{ github.run_id }}-${{ github.run_attempt }}
           path: |
-            /tmp/rustfs-security-report.md
+            ${{ env.SECURITY_ARTIFACTS_DIR }}/report.md
-            /tmp/rustfs-security.*/*
+            ${{ env.SECURITY_ARTIFACTS_DIR }}/suite.log
-          if-no-files-found: ignore
+            ${{ env.SECURITY_ARTIFACTS_DIR }}/suite-report.md
+          if-no-files-found: error
           retention-days: 3
 
       - name: Cleanup environment (after)

.github/workflows/rustfs-storage-test.yml

@@ -46,10 +46,28 @@ env:
 jobs:
   storage-test:
     runs-on: smoke-testing
-    continue-on-error: true
     timeout-minutes: 360
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     steps:
+      - name: Checkout repository (for report parser)
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Initialize functional evidence
+        id: evidence
+        run: |
+          set -euo pipefail
+          umask 077
+          FUNCTIONAL_ARTIFACTS_DIR="${RUNNER_TEMP}/rustfs-storage-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+          mkdir -- "${FUNCTIONAL_ARTIFACTS_DIR}" "${FUNCTIONAL_ARTIFACTS_DIR}-scratch"
+          {
+            printf 'FUNCTIONAL_ARTIFACTS_DIR=%s\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'LOG_FILE=%s/suite.log\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'REPORT_FILE=%s/report.md\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'TMPDIR=%s-scratch\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+          } >> "${GITHUB_ENV}"
+
       # auto-testing is private: clone it with the dedicated PF token (not
       # GITHUB_TOKEN) and retry transient GitHub/network failures.
       - name: Checkout auto-testing scripts (with retry)
@@ -97,9 +115,6 @@ jobs:
 
       - name: Run storage engine suite
         id: test
-        continue-on-error: true
-        env:
-          LOG_FILE: /tmp/rustfs-storage.log
         run: |
           set -euo pipefail
           chmod +x auto-testing/rustfs-storage-test.sh
@@ -122,10 +137,7 @@ jobs:
           ./auto-testing/rustfs-storage-test.sh "${ARGS[@]}"
 
       - name: Generate report
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
-        env:
-          LOG_FILE: /tmp/rustfs-storage.log
-          REPORT_FILE: /tmp/rustfs-storage-report.md
         run: |
           set -euo pipefail
           PACKAGE_URL='${{ inputs.package_url }}'
@@ -147,83 +159,44 @@ jobs:
               RUSTFS_VERSION_INFO="${DETECTED_VERSION}"
             fi
           fi
-          CASE_TABLE="/tmp/rustfs-storage-cases.md"
+          CASE_TABLE="${FUNCTIONAL_ARTIFACTS_DIR}/cases.md"
-          python3 - "${LOG_FILE}" "${CASE_TABLE}" <<'PY'
+          CASE_RESULT=success
-          import re
+          python3 scripts/functional_case_report.py "${LOG_FILE}" "${CASE_TABLE}" || CASE_RESULT=failure
-          import sys
+          RESULT=failure
-
+          if [ '${{ steps.test.outcome }}' = 'success' ] && [ "${CASE_RESULT}" = 'success' ]; then
-          log_file, out_file = sys.argv[1], sys.argv[2]
+            RESULT=success
-          ansi = re.compile(r'\x1b\[[0-9;]*m')
+          fi
-          start_re = re.compile(r'^---\s+([A-Z0-9]+-[0-9]+)\s+(.+?)\s+---$')
-          done_re = re.compile(r'^\[(PASS|FAIL|UNSUPPORTED)\]\s+([A-Z0-9]+-[0-9]+)\b')
-
-          rows = []
-          index = {}
-          current = None
-          try:
-            with open(log_file, 'r', encoding='utf-8', errors='replace') as fh:
-              for raw in fh:
-                line = ansi.sub('', raw).strip()
-                m = start_re.match(line)
-                if m:
-                  case_id, name = m.group(1), m.group(2)
-                  current = case_id
-                  if case_id not in index:
-                    index[case_id] = len(rows)
-                    rows.append([case_id, name, 'RUNNING'])
-                  continue
-                m = done_re.match(line)
-                if m:
-                  status, case_id = m.group(1), m.group(2)
-                  if case_id in index:
-                    rows[index[case_id]][2] = status
-                  else:
-                    rows.append([case_id, case_id, status])
-                    index[case_id] = len(rows) - 1
-                  current = None
-          except FileNotFoundError:
-            rows = []
-
-          counts = {'PASS': 0, 'FAIL': 0, 'UNSUPPORTED': 0, 'RUNNING': 0}
-          for _, _, status in rows:
-            counts[status] = counts.get(status, 0) + 1
-
-          with open(out_file, 'w', encoding='utf-8') as out:
-            out.write('## Case Summary\n\n')
-            out.write(f"- Total: {len(rows)}\\n")
-            out.write(f"- PASS: {counts.get('PASS', 0)}\\n")
-            out.write(f"- FAIL: {counts.get('FAIL', 0)}\\n")
-            out.write(f"- UNSUPPORTED: {counts.get('UNSUPPORTED', 0)}\\n")
-            out.write('\\n')
-            out.write('| Case | Name | Status |\\n')
-            out.write('| --- | --- | --- |\\n')
-            for case_id, name, status in rows:
-              out.write(f'| {case_id} | {name} | {status} |\\n')
-          PY
           {
             echo "# RustFS storage engine test report"
             echo ""
             echo "- Run: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${{ github.event_name }}"
             echo "- Package: ${PACKAGE_SOURCE}"
             echo "- RustFS Version: ${RUSTFS_VERSION_INFO}"
-            echo "- Test Step Outcome: ${{ steps.test.outcome }}"
+            echo "- Test Step Outcome: ${RESULT}"
+            echo "- Suite Step Outcome: ${{ steps.test.outcome }}"
             echo ""
-            cat "${CASE_TABLE}" || true
+            if [ "${RESULT}" = "success" ]; then
+              cat "${CASE_TABLE}"
               echo ""
               echo "## Log tail"
               echo '```text'
-            tail -n 200 "${LOG_FILE}" || true
+              tail -n 200 "${LOG_FILE}"
               echo '```'
+            else
+              echo "The suite or evidence validation failed. See this run's artifact for partial case results and suite.log."
+            fi
           } | tee "${REPORT_FILE}"
           cat "${REPORT_FILE}" >> "${GITHUB_STEP_SUMMARY}"
+          [ "${RESULT}" = "success" ]
 
       - name: Upload functional report to dashboard
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ env.PF_TESTING_GH_TOKEN }}
-          REPORT_FILE: /tmp/rustfs-storage-report.md
           SUITE: storage
         run: |
           set -euo pipefail
@@ -233,28 +206,32 @@ jobs:
           fi
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="functional-reports/${SUITE}/${DATE}.md"
-          CONTENT="$(python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
+          # Base64-encode the report into a temp file and feed it to jq via
+          # --rawfile: large reports (e.g. pool) exceed the OS argv limit and
+          # make `jq --arg content "${CONTENT}"` fail with "Argument list too long".
+          B64_FILE="$(mktemp)"
+          python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}" > "${B64_FILE}"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" --arg sha "${SHA}" \
-              '{message:$msg, content:$content, sha:$sha}' \
+              '{message:$msg, content:($content|rtrimstr("\n")), sha:$sha}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           else
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" \
-              '{message:$msg, content:$content}' \
+              '{message:$msg, content:($content|rtrimstr("\n"))}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           fi
+          rm -f "${B64_FILE}"
 
       - name: File failure issue in rustfs/backlog
         if: ${{ always() && (failure() || steps.test.outcome == 'failure' || steps.test.outcome == 'cancelled') }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
+          EVIDENCE_OUTCOME: ${{ steps.evidence.outcome }}
           SUITE: 'storage'
           SUITE_LABEL: 'Storage engine'
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-          REPORT_FILE: '/tmp/rustfs-storage-report.md'
-          LOG_FILE: '/tmp/rustfs-storage.log'
         run: |
           set -euo pipefail
           if [ -z "${GH_TOKEN:-}" ]; then
@@ -282,14 +259,16 @@ jobs:
             echo ""
             echo "- Suite: \`${SUITE}\`"
             echo "- Run: ${RUN_URL}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${GITHUB_EVENT_NAME}"
             echo "- Date: $(date -u +%Y-%m-%d)"
             echo ""
             echo "## Report (errors and symptoms)"
             echo ""
-            if [ -s "${REPORT_FILE}" ]; then
+            if [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${REPORT_FILE}" ]; then
               redact < "${REPORT_FILE}"
-            elif [ -s "${LOG_FILE:-}" ]; then
+            elif [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${LOG_FILE:-}" ]; then
               echo "(report file missing; log tail below)"
               echo ""
               tail -n 200 "${LOG_FILE}" | redact
@@ -305,14 +284,15 @@ jobs:
           echo "filed backlog issue for suite ${SUITE}"
 
       - name: Upload report and logs
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
         with:
-          name: rustfs-storage-${{ github.run_id }}
+          name: rustfs-storage-${{ github.run_id }}-${{ github.run_attempt }}
           path: |
-            /tmp/rustfs-storage.log
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/report.md
-            /tmp/rustfs-storage-report.md
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/suite.log
-          if-no-files-found: warn
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/cases.md
+          if-no-files-found: error
 
       - name: Cleanup environment (after)
         if: always()

.github/workflows/rustfs-tier-test.yml

@@ -61,9 +61,6 @@ env:
 jobs:
   tier-test:
     runs-on: smoke-testing
-    # Requirement: a failing suite must not fail the workflow; failures
-    # are filed to rustfs/backlog and the chain continues.
-    continue-on-error: true
     timeout-minutes: 420
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     steps:
@@ -380,17 +377,22 @@ jobs:
           fi
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="functional-reports/${SUITE}/${DATE}.md"
-          CONTENT="$(python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
+          # Base64-encode the report into a temp file and feed it to jq via
+          # --rawfile: large reports (e.g. pool) exceed the OS argv limit and
+          # make `jq --arg content "${CONTENT}"` fail with "Argument list too long".
+          B64_FILE="$(mktemp)"
+          python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}" > "${B64_FILE}"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" --arg sha "${SHA}" \
-              '{message:$msg, content:$content, sha:$sha}' \
+              '{message:$msg, content:($content|rtrimstr("\n")), sha:$sha}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           else
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" \
-              '{message:$msg, content:$content}' \
+              '{message:$msg, content:($content|rtrimstr("\n"))}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           fi
+          rm -f "${B64_FILE}"
 
       - name: Verify required tier evidence
         id: evidence_verify

.github/workflows/rustfs-upgrade-test.yml

@@ -18,7 +18,7 @@ on:
   workflow_dispatch:
     inputs:
       from_version:
-        description: 'OLD RustFS release tag (must ship a .deb asset, e.g. 1.0.0-rc.3)'
+        description: 'OLD RustFS release tag, e.g. 1.0.0-rc.3 (its release must ship a .deb asset). Leave empty for the default.'
         required: false
         default: '1.0.0-rc.3'
       from_url:
@@ -26,7 +26,7 @@ on:
         required: false
         type: string
       to_version:
-        description: 'NEW RustFS release tag (leave empty for latest nightly)'
+        description: 'NEW RustFS release tag, e.g. 1.0.0-rc.5 (any version with a .deb asset). Leave empty for latest nightly.'
         required: false
       to_url:
         description: 'NEW .deb URL. Overrides to_version / nightly default.'
@@ -79,10 +79,28 @@ env:
 jobs:
   upgrade-test:
     runs-on: smoke-testing
-    continue-on-error: true
     timeout-minutes: 420
     if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
     steps:
+      - name: Checkout repository (for report parser)
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+
+      - name: Initialize functional evidence
+        id: evidence
+        run: |
+          set -euo pipefail
+          umask 077
+          FUNCTIONAL_ARTIFACTS_DIR="${RUNNER_TEMP}/rustfs-upgrade-${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+          mkdir -- "${FUNCTIONAL_ARTIFACTS_DIR}" "${FUNCTIONAL_ARTIFACTS_DIR}-scratch"
+          {
+            printf 'FUNCTIONAL_ARTIFACTS_DIR=%s\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'LOG_FILE=%s/suite.log\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'REPORT_FILE=%s/report.md\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+            printf 'TMPDIR=%s-scratch\n' "${FUNCTIONAL_ARTIFACTS_DIR}"
+          } >> "${GITHUB_ENV}"
+
       # auto-testing is private: clone it with the dedicated PF token (not
       # GITHUB_TOKEN) and retry transient GitHub/network failures.
       - name: Checkout auto-testing scripts (with retry)
@@ -142,9 +160,8 @@ jobs:
 
       - name: Run upgrade compatibility suite
         id: test
-        continue-on-error: true
         env:
-          LOG_FILE: /tmp/rustfs-upgrade.log
+          GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
         run: |
           set -euo pipefail
           chmod +x auto-testing/rustfs-upgrade-test.sh
@@ -175,13 +192,33 @@ jobs:
           else
             ARGS+=(--to-url "${RUSTFS_NIGHTLY_PACKAGE_URL}")
           fi
+          # Fail fast with a clear message when a requested release tag has
+          # no .deb asset (e.g. 1.0.0-rc.4 ships only zips), instead of
+          # letting the suite die mid-run on a 404.
+          check_release_asset() {
+            local version="$1" tag asset url
+            [ -n "${version}" ] && [ "${version}" != "null" ] || return 0
+            tag="${version#v}"
+            asset="rustfs_${tag//-/.}_amd64.deb"
+            url="https://github.com/rustfs/rustfs/releases/download/${tag}/${asset}"
+            if ! gh api "repos/rustfs/rustfs/releases/tags/${tag}" --jq '.assets[].name' 2>/dev/null | grep -qxF "${asset}"; then
+              echo "ERROR: release ${tag} has no downloadable asset ${asset}:" >&2
+              echo "       ${url}" >&2
+              echo "Pick a tag whose release ships a .deb (check its release assets)." >&2
+              exit 1
+            fi
+            echo "resolved ${tag} -> ${url}"
+          }
+          if [ -z "${FROM_URL}" ]; then
+            check_release_asset "${FROM_VERSION}"
+          fi
+          if [ -z "${TO_URL}" ]; then
+            check_release_asset "${TO_VERSION}"
+          fi
           ./auto-testing/rustfs-upgrade-test.sh "${ARGS[@]}"
 
       - name: Generate report
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
-        env:
-          LOG_FILE: /tmp/rustfs-upgrade.log
-          REPORT_FILE: /tmp/rustfs-upgrade-report.md
         run: |
           set -euo pipefail
           FROM_URL='${{ inputs.from_url }}'
@@ -202,103 +239,47 @@ jobs:
           else
             TO_SOURCE="${RUSTFS_NIGHTLY_PACKAGE_URL}"
           fi
-          CASE_TABLE="/tmp/rustfs-upgrade-cases.md"
+          CASE_TABLE="${FUNCTIONAL_ARTIFACTS_DIR}/cases.md"
-          MATRIX_TABLE="/tmp/rustfs-upgrade-matrix.md"
+          MATRIX_TABLE="${FUNCTIONAL_ARTIFACTS_DIR}/matrix.md"
-          python3 - "${LOG_FILE}" "${CASE_TABLE}" "${MATRIX_TABLE}" <<'PY'
+          CASE_RESULT=success
-          import re
+          python3 scripts/functional_case_report.py "${LOG_FILE}" "${CASE_TABLE}" "${MATRIX_TABLE}" || CASE_RESULT=failure
-          import sys
+          RESULT=failure
-
+          if [ '${{ steps.test.outcome }}' = 'success' ] && [ "${CASE_RESULT}" = 'success' ]; then
-          log_file, out_file, matrix_file = sys.argv[1], sys.argv[2], sys.argv[3]
+            RESULT=success
-          ansi = re.compile(r'\x1b\[[0-9;]*m')
+          fi
-          start_re = re.compile(r'^---\s+([A-Z]+-[0-9]+)\s+(.+?)\s+---$')
-          done_re = re.compile(r'^\[(PASS|FAIL|UNSUPPORTED)\]\s+([A-Z]+-[0-9]+)\b')
-          topo_re = re.compile(
-              r'^\[UPG-TOPO\]\s+(\S+)\s+(\S+)\s+(\S+)\s+(\S+)\s+PASS=(\d+)\s+FAIL=(\d+)\s*$')
-
-          rows = []
-          index = {}
-          topo_rows = []
-          try:
-              with open(log_file, 'r', encoding='utf-8', errors='replace') as fh:
-                  for raw in fh:
-                      line = ansi.sub('', raw).strip()
-                      m = topo_re.match(line)
-                      if m:
-                          topo_rows.append(m.groups())
-                          continue
-                      m = start_re.match(line)
-                      if m:
-                          case_id, name = m.group(1), m.group(2)
-                          if case_id not in index:
-                              index[case_id] = len(rows)
-                              rows.append([case_id, name, 'RUNNING'])
-                          continue
-                      m = done_re.match(line)
-                      if m:
-                          status, case_id = m.group(1), m.group(2)
-                          if case_id in index:
-                              rows[index[case_id]][2] = status
-                          else:
-                              rows.append([case_id, case_id, status])
-                              index[case_id] = len(rows) - 1
-          except FileNotFoundError:
-              rows = []
-
-          counts = {'PASS': 0, 'FAIL': 0, 'UNSUPPORTED': 0, 'RUNNING': 0}
-          for _, _, status in rows:
-              counts[status] = counts.get(status, 0) + 1
-
-          with open(out_file, 'w', encoding='utf-8') as out:
-              out.write('## Case Summary\n\n')
-              out.write(f"- Total: {len(rows)}\\n")
-              out.write(f"- PASS: {counts.get('PASS', 0)}\\n")
-              out.write(f"- FAIL: {counts.get('FAIL', 0)}\\n")
-              out.write(f"- UNSUPPORTED: {counts.get('UNSUPPORTED', 0)}\\n")
-              out.write('\\n')
-              out.write('| Case | Name | Status |\\n')
-              out.write('| --- | --- | --- |\\n')
-              for case_id, name, status in rows:
-                  out.write(f'| {case_id} | {name} | {status} |\\n')
-
-          # Upgrade matrix: one row per topology/backend with the versions
-          # captured on the nodes (rustfs --version) and the aggregated
-          # result. The dashboard renders this table directly.
-          with open(matrix_file, 'w', encoding='utf-8') as out:
-              out.write('## Upgrade Matrix\n\n')
-              out.write('| Topology | KMS Backend | From Version | To Version | Result |\n')
-              out.write('| --- | --- | --- | --- | --- |\n')
-              for topo, backend, old_v, new_v, npass, nfail in topo_rows:
-                  result = 'PASS' if nfail == '0' else 'FAIL'
-                  out.write(f'| {topo} | {backend} | {old_v} | {new_v} | {result} (PASS={npass} FAIL={nfail}) |\n')
-              if not topo_rows:
-                  out.write('| - | - | - | - | NOT RUN (suite failed before upgrade) |\n')
-          PY
           {
             echo "# RustFS upgrade compatibility report"
             echo ""
             echo "- Run: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${{ github.event_name }}"
             echo "- From: ${FROM_SOURCE}"
             echo "- To: ${TO_SOURCE}"
-            echo "- Test Step Outcome: ${{ steps.test.outcome }}"
+            echo "- Test Step Outcome: ${RESULT}"
+            echo "- Suite Step Outcome: ${{ steps.test.outcome }}"
             echo ""
-            cat "${MATRIX_TABLE}" || true
+            if [ "${RESULT}" = "success" ]; then
+              cat "${MATRIX_TABLE}"
               echo ""
-            cat "${CASE_TABLE}" || true
+              cat "${CASE_TABLE}"
               echo ""
               echo "## Log tail"
               echo '```text'
-            tail -n 200 "${LOG_FILE}" || true
+              tail -n 200 "${LOG_FILE}"
               echo '```'
+            else
+              echo "The suite or evidence validation failed. See this run's artifact for partial case results and suite.log."
+            fi
           } | tee "${REPORT_FILE}"
           cat "${REPORT_FILE}" >> "${GITHUB_STEP_SUMMARY}"
+          [ "${RESULT}" = "success" ]
 
       - name: Upload functional report to dashboard
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ env.PF_TESTING_GH_TOKEN }}
-          REPORT_FILE: /tmp/rustfs-upgrade-report.md
           SUITE: upgrade
         run: |
           set -euo pipefail
@@ -308,28 +289,32 @@ jobs:
           fi
           DATE="$(date -u +%Y-%m-%d)"
           REPORT_PATH="functional-reports/${SUITE}/${DATE}.md"
-          CONTENT="$(python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}")"
+          # Base64-encode the report into a temp file and feed it to jq via
+          # --rawfile: large reports (e.g. pool) exceed the OS argv limit and
+          # make `jq --arg content "${CONTENT}"` fail with "Argument list too long".
+          B64_FILE="$(mktemp)"
+          python3 -c 'import base64,sys;print(base64.b64encode(open(sys.argv[1],"rb").read()).decode())' "${REPORT_FILE}" > "${B64_FILE}"
           SHA="$(gh api "repos/rustfs/dashboard/contents/${REPORT_PATH}" -q '.sha' 2>/dev/null || true)"
           if [ -n "${SHA}" ]; then
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" --arg sha "${SHA}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" --arg sha "${SHA}" \
-              '{message:$msg, content:$content, sha:$sha}' \
+              '{message:$msg, content:($content|rtrimstr("\n")), sha:$sha}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           else
-            jq -n --arg msg "report(${SUITE}): ${DATE}" --arg content "${CONTENT}" \
+            jq -n --arg msg "report(${SUITE}): ${DATE}" --rawfile content "${B64_FILE}" \
-              '{message:$msg, content:$content}' \
+              '{message:$msg, content:($content|rtrimstr("\n"))}' \
               | gh api --method PUT "repos/rustfs/dashboard/contents/${REPORT_PATH}" --input - >/dev/null
           fi
+          rm -f "${B64_FILE}"
 
       - name: File failure issue in rustfs/backlog
         if: ${{ always() && (failure() || steps.test.outcome == 'failure' || steps.test.outcome == 'cancelled') }}
         continue-on-error: true
         env:
           GH_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
+          EVIDENCE_OUTCOME: ${{ steps.evidence.outcome }}
           SUITE: 'upgrade'
           SUITE_LABEL: 'Upgrade compatibility'
           RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-          REPORT_FILE: '/tmp/rustfs-upgrade-report.md'
-          LOG_FILE: '/tmp/rustfs-upgrade.log'
         run: |
           set -euo pipefail
           if [ -z "${GH_TOKEN:-}" ]; then
@@ -357,14 +342,16 @@ jobs:
             echo ""
             echo "- Suite: \`${SUITE}\`"
             echo "- Run: ${RUN_URL}"
+            echo "- Attempt: ${GITHUB_RUN_ATTEMPT}"
+            echo "- Workflow Commit: ${GITHUB_SHA}"
             echo "- Trigger: ${GITHUB_EVENT_NAME}"
             echo "- Date: $(date -u +%Y-%m-%d)"
             echo ""
             echo "## Report (errors and symptoms)"
             echo ""
-            if [ -s "${REPORT_FILE}" ]; then
+            if [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${REPORT_FILE}" ]; then
               redact < "${REPORT_FILE}"
-            elif [ -s "${LOG_FILE:-}" ]; then
+            elif [ "${EVIDENCE_OUTCOME}" = "success" ] && [ -s "${LOG_FILE:-}" ]; then
               echo "(report file missing; log tail below)"
               echo ""
               tail -n 200 "${LOG_FILE}" | redact
@@ -380,14 +367,16 @@ jobs:
           echo "filed backlog issue for suite ${SUITE}"
 
       - name: Upload report and logs
-        if: always()
+        if: ${{ always() && steps.evidence.outcome == 'success' }}
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
         with:
-          name: rustfs-upgrade-test-${{ github.run_id }}
+          name: rustfs-upgrade-test-${{ github.run_id }}-${{ github.run_attempt }}
           path: |
-            /tmp/rustfs-upgrade-report.md
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/report.md
-            /tmp/rustfs-upgrade.*/*
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/suite.log
-          if-no-files-found: ignore
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/cases.md
+            ${{ env.FUNCTIONAL_ARTIFACTS_DIR }}/matrix.md
+          if-no-files-found: error
           retention-days: 3
 
       - name: Cleanup environment (after)

.github/workflows/scheduled-validation-freshness.yml

@@ -42,6 +42,7 @@ jobs:
       - name: Check latest scheduled runs
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          RUSTFS_DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
         run: |
           set +e
           python3 scripts/check_scheduled_validation_freshness.py \

.github/workflows/scheduled-validation-watchdog.yml

@@ -22,6 +22,7 @@ on:
       - "Continuous Integration"
       - "coverage"
       - "e2e-nightly"
+      - "e2e-distributed"
       - "e2e-s3tests"
       - "Fuzz"
       - "mint"

.gitignore

@@ -33,6 +33,7 @@ profile.json
 *.zst
 .secrets
 *.go
+!crates/zip/tests/fixtures/snowball/**/generate/*.go
 *.pb
 *.svg
 deploy/logs/*.log.*

.pre-commit-config.yaml

@@ -3,9 +3,9 @@
 repos:
   - repo: local
     hooks:
-      - id: rustfs-dev-check
+      - id: rustfs-fmt-check
-        name: rustfs dev-check
+        name: Rust formatting
-        entry: make dev-check
+        entry: cargo fmt --all --check
         language: system
         types: [rust]
         pass_filenames: false

CHANGELOG.md

@@ -18,7 +18,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   - Read paths: an object at or below `policy.inline_max_bytes` (16 MiB by default) is teed to the client and to the local store in a single source read; a larger object or a Range read streams through and a background pull stores the whole object. A HEAD miss is proxied to the source and stores nothing (`policy.head = local_only` disables it). Every source-backed response carries `x-rustfs-on-demand-migration: source`
   - Protections: a per-source circuit breaker, a per-key negative cache, singleflight per key, a concurrency limit and a bounded pull queue shared by the inline and background paths, an optional bandwidth limit, an anti-loop request marker, and the shared outbound-endpoint (SSRF) policy
   - Metrics under `rustfs_on_demand_migration_*` (`requests_total`, `pulled_bytes_total`, `pulled_objects_total`, `pull_failures_total`, `inflight_pulls`, `queue_depth`, `source_latency_seconds_*`, `breaker_state`), mirrored per node by the admin status route
-  - Limitations: listings show only local objects (the source is not merged into `ListObjectsV2`); PUT and DELETE never reach the source; a source object updated after it was pulled is not re-fetched; SSE-C source objects are unsupported and answer 424; `Last-Modified` on a pulled object is the local write time, with the source timestamp kept in metadata
+  - Listings: `ListObjects` v1 remains local with ordinary key markers. `ListObjectsV2` can merge source objects when `policy.list_through = true`; this is off by default
+  - Upgrade and rollback: finish upgrading every node before enabling ODM. An rc.5 node that writes bucket configuration drops the ODM fields from metadata; neither a later restart nor moving the service out of ECStore recovers them. Before rollback, disable ODM and securely retain the original full configuration and credentials. After every node returns to a compatible version, restore and validate that configuration. Redacted exports cannot replace the credential backup; source-only objects are unavailable through RustFS while ODM is disabled. See the upgrade and rollback section of `docs/operations/on-demand-migration.md`
+  - Optional Google dependencies: default and `full` server builds retain native GCS support. `cargo build -p rustfs --no-default-features --features ftps,webdav` excludes Google SDKs while preserving configuration decoding and redaction; native GCS ODM and tier operations require the `gcs` feature. Do not use that build with existing GCS-tiered data
+  - Limitations: PUT and DELETE never reach the source; a source object updated after it was pulled is not re-fetched; SSE-C source objects are unsupported and answer 424; `Last-Modified` on a pulled object is the local write time, with the source timestamp kept in metadata
 - **NATS JetStream Publish Path**: Opt-in at-least-once delivery for the NATS notify and audit targets. A NATS Core publish flushes to the connection without awaiting a broker acknowledgement, so an event can be lost across a broker restart or a reconnect after the send queue has already cleared it. A queued event now clears only after the JetStream `PublishAck`, so bucket notifications survive those interruptions. Off by default and byte-identical to the NATS Core path when disabled.
   - Three configuration keys per target: `JETSTREAM_ENABLE`, `JETSTREAM_STREAM_NAME`, and `JETSTREAM_ACK_TIMEOUT_SECS`, under the `RUSTFS_NOTIFY_NATS_` and `RUSTFS_AUDIT_NATS_` prefixes
   - Durable store-and-forward with a stable dedup id sent as the `Nats-Msg-Id` header, so a replay after a crash is collapsed by the server duplicate window

CONTRIBUTING.md

@@ -109,24 +109,17 @@ affected boundaries and risks. CI still runs its configured repository gates.
 
 ### 🔒 Git Pre-commit Hooks (optional)
 
-Git hooks are **not** versioned in this repository, so a fresh clone has no
+The optional hook uses the checked-in `.pre-commit-config.yaml`. Install [pre-commit](https://pre-commit.com/#installation), then run this from the checkout or a linked worktree:
-active pre-commit hook. If you add your own `.git/hooks/pre-commit` (a good
-choice is a one-liner that runs `make pre-commit`), you can mark it executable
-with:
 
 ```bash
 make setup-hooks
 ```
 
-Or manually:
+The hook runs `cargo fmt --all --check` when staged files include Rust source. It does not compile the workspace or run tests. Fix formatting with `cargo fmt --all`, inspect and stage the result, then commit again.
 
-```bash
+`pre-commit install` resolves Git's hook directory for linked worktrees and preserves an existing hook in migration mode. If you use `core.hooksPath`, keep that hook manager and integrate `pre-commit run` there; the installer refuses to silently replace that configuration.
-chmod +x .git/hooks/pre-commit
-```
 
-With or without a hook, follow the verification tiers in `AGENTS.md`. Run the
+A local hook provides early formatting feedback. With or without it, follow the verification tiers in `AGENTS.md`, run relevant behavioral tests, and satisfy the CI merge gates. `make pre-commit` and `make dev-check` remain explicit broader commands.
-applicable scoped checks, and reserve `make pre-pr` for broad cross-module
-changes whose impact cannot be bounded by those checks.
 
 ### 📝 Formatting Configuration
 
@@ -138,31 +131,11 @@ fn_call_width = 90
 single_line_let_else_max_width = 100
 ```
 
-### 🚫 Commit Prevention
-
-If you set up a pre-commit hook and your code doesn't meet the formatting requirements, the hook will:
-
-1. **Block the commit** and show clear error messages
-2. **Provide exact commands** to fix the issues
-3. **Guide you through** the resolution process
-
-Example output when formatting fails:
-
-```
-❌ Code formatting check failed!
-💡 Please run 'cargo fmt --all' to format your code before committing.
-
-🔧 Quick fix:
-   cargo fmt --all
-   git add .
-   git commit
-```
-
 ### 🔄 Development Workflow
 
 1. **Make your changes**
 2. **Format your code**: `make fmt` or `cargo fmt --all`
-3. **Run the fast gate**: `make pre-commit` (no clippy, no tests)
+3. **Select relevant checks** using the validation tier in `AGENTS.md`; use `make pre-commit` when its broader fast gate adds useful coverage
 4. **Commit your changes**: `git commit -m "your message"`
 5. **Complete the applicable multi-role adversarial review** for non-exempt changes (see `AGENTS.md`)
 6. **Run applicable scoped checks before opening/updating a PR**; consider
@@ -206,11 +179,12 @@ Configure your IDE to:
 #### Pre-commit hook not running?
 
 ```bash
-# Check if hook is executable
+pre-commit validate-config
-ls -la .git/hooks/pre-commit
+pre-commit run --all-files
-
+# Inspect any configured hook manager; do not overwrite it.
-# Make it executable if needed
+git config --get core.hooksPath
-chmod +x .git/hooks/pre-commit
+# Install if no separate hook manager is configured.
+make setup-hooks
 ```
 
 #### Formatting issues?

Cargo.toml

@@ -168,7 +168,7 @@ reqwest = "0.13.4"
 rustfs-kafka-async = { version = "1.3.1" }
 socket2 = { version = "0.6.5" }
 tokio = { version = "1.53.1" }
-tokio-rustls = { default-features = false, version = "0.26.4" }
+tokio-rustls = { default-features = false, version = "0.26.5" }
 tokio-stream = { version = "0.1.19" }
 tokio-test = "0.4.5"
 tokio-util = { version = "0.7.19" }
@@ -199,10 +199,10 @@ serde_urlencoded = "0.7.1"
 # matching stable releases are not available yet, while previous stable lines
 # have incompatible APIs. Keep them exact-pinned and monitor upstream for stable
 # releases.
-aes-gcm = { version = "=0.11.1" }
+aes-gcm = { version = "0.11.1" }
-argon2 = { version = "=0.6.0" }
+argon2 = { version = "0.6.0" }
-blake2 = "=0.11.0"
+blake2 = "0.11.0"
-chacha20poly1305 = { version = "=0.11.0" }
+chacha20poly1305 = { version = "0.11.0" }
 crc-fast = "1.10.0"
 hmac = { version = "0.13.0" }
 jsonwebtoken = { version = "11.0.0" }
@@ -234,15 +234,19 @@ tokio-postgres-rustls = "0.14.0"
 # Utilities and Tools
 anyhow = "1.0.104"
 arc-swap = "1.9.2"
-# RUSTFS_COMPAT_TODO(tokio-tar-extension-limits): keep the fork pin until every parser hardening used by Snowball is released upstream. Remove after astral-sh/tokio-tar#118 is merged and a published release includes extension, physical-entry, and sparse limits, cancellation-safe sparse parsing, and error-fused entry streams.
+# RUSTFS_COMPAT_TODO(tokio-tar-extension-limits): keep the fork pin while Snowball and Swift still depend on it. Remove after Snowball uses a released tar-codec/tar-framing API that exposes precedence-resolved MinIO vendor records, RustFS preserves cancellation-safe ownership of large streamed members, footerless minio-go input is accepted only at an authenticated complete request boundary, the existing resource-limit, cancellation, and error-fuse regressions pass, and Swift no longer needs this fork.
 astral-tokio-tar = { git = "https://github.com/cxymds/tokio-tar.git", rev = "603756478b7668436e464519c77ccac22a99ba96" }
+# Candidate Snowball parser versions exercised by rustfs-zip compatibility fixtures.
+tar-codec = "0.0.14"
+tar-framing = "0.0.14"
 atoi = "3.1.0"
 atomic_enum = "0.3.0"
-aws-config = { version = "1.11.0" }
+aws-config = { version = "1.12.0" }
 aws-credential-types = { version = "1.3.0" }
-aws-sdk-kms = { default-features = false, version = "1.117.0" }
+aws-sdk-kms = { default-features = false, version = "1.118.0" }
-aws-sdk-s3 = { default-features = false, version = "1.144.0" }
+aws-sdk-s3 = { default-features = false, version = "1.145.0" }
-aws-sdk-sts = { default-features = false, version = "1.113.0" }
+aws-sdk-sts = { default-features = false, version = "1.114.0" }
+aws-smithy-async = { version = "1.3.0" }
 aws-smithy-http-client = { default-features = false, version = "1.4.0" }
 aws-smithy-runtime-api = { version = "1.16.0" }
 aws-smithy-types = { version = "1.6.3" }
@@ -339,7 +343,7 @@ windows = { version = "0.62.2" }
 windows-sys = "0.61.2"
 xxhash-rust = { version = "0.8.18" }
 zip = "8.6.0"
-zstd = "0.13.3"
+zstd = "0.14.0"
 
 # Observability and Metrics
 metrics = "0.24.6"
@@ -367,7 +371,8 @@ dav-server = "0.11.0"
 
 # Performance Analysis and Memory Profiling
 rustfs-mimalloc = { version = "0.5.3" }
-hotpath = { version = "0.25.0", default-features = false }
+# Preserve Unicode focus filters until rustfs/backlog#2302 is resolved.
+hotpath = { version = "=0.25.0", default-features = false }
 # Snapshot testing for output format regression detection
 insta = { version = "1.48" }
 

crates/config/README.md

@@ -130,6 +130,21 @@ Scanner cycle budget controls:
   - timeout returns S3 `SlowDown`, so clients should use normal SDK retry handling.
   - this is not a fdatasync or group-commit switch. Track fdatasync batching separately with `rustfs_s3_put_object_rename_fdatasync_batch_files`.
 
+## Remote tier timeout environment variables
+
+- `RUSTFS_TIER_REMOTE_CONNECT_TIMEOUT_SECS`
+  - remote tier TCP connect timeout.
+  - default is `10`.
+  - must be positive; zero fails tier client initialization, while an invalid integer is logged and falls back to the default.
+- `RUSTFS_TIER_REMOTE_REQUEST_TIMEOUT_SECS`
+  - remote tier request timeout through response headers.
+  - default is `86400` so large transition uploads keep a production-safe budget.
+  - must be positive; zero fails tier client initialization, while an invalid integer is logged and falls back to the default. Very large values are accepted and act as a correspondingly long budget.
+- `RUSTFS_TIER_REMOTE_RESPONSE_BODY_IDLE_TIMEOUT_SECS`
+  - maximum idle time between remote tier response-body chunks.
+  - default is `60`; the timer resets only when non-empty body data keeps progressing.
+  - must be positive; zero fails tier client initialization, while an invalid integer is logged and falls back to the default.
+
 ## Drive timeout environment variables
 
 - `RUSTFS_DRIVE_METADATA_TIMEOUT_SECS`

crates/config/src/constants/object.rs

@@ -137,6 +137,28 @@ pub const DEFAULT_TIER_REMOTE_VERSION_STATE_FLEET_CONFIRMED: bool = false;
 const _: () = assert!(!DEFAULT_TIER_REMOTE_VERSION_STATE_WRITE);
 const _: () = assert!(!DEFAULT_TIER_REMOTE_VERSION_STATE_FLEET_CONFIRMED);
 
+/// Environment variable for remote tier TCP connect timeout in seconds.
+pub const ENV_TIER_REMOTE_CONNECT_TIMEOUT_SECS: &str = "RUSTFS_TIER_REMOTE_CONNECT_TIMEOUT_SECS";
+/// Default remote tier TCP connect timeout in seconds.
+pub const DEFAULT_TIER_REMOTE_CONNECT_TIMEOUT_SECS: u64 = 10;
+
+/// Environment variable for the remote tier request timeout in seconds.
+///
+/// This bounds upload/download request progress through response headers. The
+/// default is intentionally large so multi-TiB transition uploads keep their
+/// previous production budget while black-hole remotes no longer wait forever.
+pub const ENV_TIER_REMOTE_REQUEST_TIMEOUT_SECS: &str = "RUSTFS_TIER_REMOTE_REQUEST_TIMEOUT_SECS";
+/// Default remote tier request timeout in seconds.
+pub const DEFAULT_TIER_REMOTE_REQUEST_TIMEOUT_SECS: u64 = 24 * 60 * 60;
+
+/// Environment variable for remote tier response-body idle timeout in seconds.
+///
+/// The timer is re-armed on every non-empty response-body chunk, so slow but
+/// progressing remotes can continue while silent response bodies are cancelled.
+pub const ENV_TIER_REMOTE_RESPONSE_BODY_IDLE_TIMEOUT_SECS: &str = "RUSTFS_TIER_REMOTE_RESPONSE_BODY_IDLE_TIMEOUT_SECS";
+/// Default remote tier response-body idle timeout in seconds.
+pub const DEFAULT_TIER_REMOTE_RESPONSE_BODY_IDLE_TIMEOUT_SECS: u64 = 60;
+
 /// Request the object-transaction fencing contract used by storage-owned
 /// cleanup receipts and lock-window optimizations.
 ///
@@ -812,6 +834,16 @@ mod remote_version_state_tests {
         );
     }
 
+    #[test]
+    fn remote_tier_timeout_env_names_are_stable() {
+        assert_eq!(super::ENV_TIER_REMOTE_CONNECT_TIMEOUT_SECS, "RUSTFS_TIER_REMOTE_CONNECT_TIMEOUT_SECS");
+        assert_eq!(super::ENV_TIER_REMOTE_REQUEST_TIMEOUT_SECS, "RUSTFS_TIER_REMOTE_REQUEST_TIMEOUT_SECS");
+        assert_eq!(
+            super::ENV_TIER_REMOTE_RESPONSE_BODY_IDLE_TIMEOUT_SECS,
+            "RUSTFS_TIER_REMOTE_RESPONSE_BODY_IDLE_TIMEOUT_SECS"
+        );
+    }
+
     #[test]
     fn data_movement_part_checksum_gate_uses_stable_environment_names() {
         assert_eq!(super::ENV_DATA_MOVEMENT_PART_CHECKSUMS_WRITE, "RUSTFS_DATA_MOVEMENT_PART_CHECKSUMS_WRITE");

crates/e2e_test/README.md

@@ -26,6 +26,7 @@ Registered in [`src/lib.rs`](src/lib.rs). Grouped by concern:
 | **protocols** | [`src/protocols/`](src/protocols) | FTPS, WebDAV, SFTP compliance. Fixed ports, own guide: [`src/protocols/README.md`](src/protocols/README.md) |
 | **reliant** | [`src/reliant/`](src/reliant) | Tests that reuse an **externally started** server (SQL/select, conditional writes, lifecycle, deleted-object reads, node-interact). Run via [`scripts/run_e2e_tests.sh`](../../scripts/run_e2e_tests.sh); see [`src/reliant/README.md`](src/reliant/README.md) |
 | **cluster** | `cluster_concurrency_test`, `stale_multipart_cleanup_cluster_test`, `namespace_lock_quorum_test`, `admin_timeout_regression_test`, `object_lambda_test`, `replication_extension_test`, `tier_stats_cluster_test` | Multi-node scenarios via `RustFSTestClusterEnvironment` |
+| **distributed 4×4** | [`src/distributed/`](src/distributed) | Storage-sensitive PR and nightly `e2e-distributed` lane: S3, object lock/WORM, versioning, bucket/site replication, quota, expand/decommission/rebalance, concurrency, chaos, 4-node upgrade of historical data and IAM AK/SK. Map: [`docs/testing/distributed-e2e.md`](../../docs/testing/distributed-e2e.md) |
 | **chaos / reliability** | [`src/chaos.rs`](src/chaos.rs), `reliability_disk_fault_test`, `heal_erasure_disk_rebuild_test`, `server_startup_failfast_test` | Disk offline/replace/corrupt, EC rebuild, heal, fail-fast startup |
 | **upgrade compatibility** | `upgrade_compatibility_test` | Pinned previous-release writes followed by current-build reads on the same data directory |
 
@@ -171,6 +172,7 @@ the same profile for membership and execution with one nightly worker.
 | KMS suite | `e2e-full` job, merge queue + main | **Active** |
 | Direct and mixed-version rolling upgrades from pinned previous release | `e2e-upgrade.yml`, storage-sensitive PRs + release tags + weekly | **Active** |
 | Cluster faults (`e2e-nightly` profile) | consolidated nightly workflow | **Active** (backlog#1149 ci-7) |
+| Distributed 4-node 4-disk (`e2e-distributed` profile) | `.github/workflows/e2e-distributed.yml` | **Active** (storage-sensitive PR / nightly / dispatch) |
 | Protocols (FTPS/WebDAV/SFTP) | consolidated nightly workflow, serial | **Active** (backlog#1149 ci-7) |
 | Replication (fast subset) | `e2e-smoke` profile, `e2e-tests` job, every PR | **Active** (backlog#1147 repl-1) |
 | Replication (slow + multi-node) | `e2e-repl-nightly` profile, consolidated nightly workflow | **Active** (backlog#1147 repl-1) |
@@ -191,6 +193,9 @@ cargo nextest run --profile e2e-smoke -p e2e_test
 cargo nextest run --profile e2e-full -p e2e_test
 # Cluster fault nightly lane
 cargo nextest run --profile e2e-nightly -p e2e_test
+# 4-node 4-disk distributed lane (S3 / lock / versioning / replication / decommission / chaos / upgrade)
+# Upgrade cases need RUSTFS_UPGRADE_SOURCE_BINARY; without it they fail closed.
+cargo nextest run --profile e2e-distributed -p e2e_test
 # Replication nightly lane; awscurl is required for STS paths
 cargo nextest run --profile e2e-repl-nightly -p e2e_test
 # Fixed-port protocol nightly lane

crates/e2e_test/build.rs

@@ -0,0 +1,74 @@
+// Copyright 2024 RustFS Team
+// Licensed under the Apache License, Version 2.0.
+
+use std::path::Path;
+use std::process::Command;
+
+fn git(root: &Path, args: &[&str]) -> Option<String> {
+    let output = Command::new("git").args(args).current_dir(root).output().ok()?;
+    output
+        .status
+        .success()
+        .then(|| String::from_utf8_lossy(&output.stdout).trim().to_owned())
+}
+
+fn emit(name: &str, value: &str) {
+    let value = if value.contains(['\n', '\r']) { "unknown" } else { value };
+    println!("cargo:rustc-env=RUSTFS_E2E_BUILD_{name}={value}");
+}
+
+fn main() {
+    let manifest = std::env::var_os("CARGO_MANIFEST_DIR").unwrap_or_default();
+    let root = Path::new(&manifest).join("../..");
+    // Cover dependency/common sources as well as this crate. HEAD/ref/index
+    // changes must refresh identity even when no Rust source mtime changes.
+    for path in [
+        "crates",
+        "rustfs",
+        "Cargo.toml",
+        "Cargo.lock",
+        "rust-toolchain.toml",
+        ".cargo",
+        ".config",
+    ] {
+        println!("cargo:rerun-if-changed={}", root.join(path).display());
+    }
+    let mut git_paths = vec!["HEAD".to_owned(), "index".to_owned(), "packed-refs".to_owned()];
+    if let Some(reference) = git(&root, &["symbolic-ref", "-q", "HEAD"]) {
+        git_paths.push(reference);
+    }
+    for path in git_paths {
+        if let Some(path) = git(&root, &["rev-parse", "--git-path", &path]) {
+            let path = Path::new(&path);
+            let path = if path.is_absolute() {
+                path.to_owned()
+            } else {
+                root.join(path)
+            };
+            if path.exists() {
+                println!("cargo:rerun-if-changed={}", path.display());
+            }
+        }
+    }
+    let revision = git(&root, &["rev-parse", "HEAD"]).unwrap_or_else(|| "unknown".to_owned());
+    let dirty = git(&root, &["status", "--porcelain", "--untracked-files=normal"]).is_none_or(|status| !status.is_empty());
+    let lock = git(&root, &["hash-object", "Cargo.lock"]).unwrap_or_else(|| "unknown".to_owned());
+    let mut features = std::env::vars()
+        .filter_map(|(key, _)| {
+            key.strip_prefix("CARGO_FEATURE_")
+                .map(|name| name.to_ascii_lowercase().replace('_', "-"))
+        })
+        .collect::<Vec<_>>();
+    features.sort();
+    emit("COMMIT", &revision);
+    emit("DIRTY", if dirty { "true" } else { "false" });
+    emit("LOCK", &lock);
+    emit("FEATURES", &features.join(","));
+    for name in ["TARGET", "PROFILE"] {
+        emit(name, &std::env::var(name).unwrap_or_else(|_| "unknown".to_owned()));
+    }
+    println!("cargo:rerun-if-env-changed=CARGO_ENCODED_RUSTFLAGS");
+    let flags = std::env::var("CARGO_ENCODED_RUSTFLAGS").unwrap_or_default();
+    let flags: String = flags.as_bytes().iter().map(|byte| format!("{byte:02x}")).collect();
+    emit("RUSTFLAGS_HEX", &flags);
+}

crates/e2e_test/src/chaos.rs

@@ -55,18 +55,20 @@ type ChaosResult<T> = Result<T, Box<dyn Error + Send + Sync>>;
 /// A successful S3 GET only proves that a quorum can serve an object. Replacement
 /// tests need this lower-level record to prove that the rebuilt target holds the
 /// `xl.meta` selected for a specific version and every `part.N` it declares.
-#[derive(Clone, Debug, Eq, PartialEq)]
+#[derive(Clone, Debug, Eq, PartialEq, serde::Serialize)]
 pub(crate) struct VersionShardCensus {
     pub version_id: Option<String>,
     pub has_xl_meta: bool,
     pub data_dir: Option<String>,
     pub erasure_index: Option<usize>,
+    pub data_blocks: Option<usize>,
+    pub parity_blocks: Option<usize>,
     pub expected_part_numbers: BTreeSet<usize>,
     pub present_part_fingerprints: BTreeMap<usize, PartShardFingerprint>,
     pub inline_data_fingerprint: Option<PartShardFingerprint>,
 }
 
-#[derive(Clone, Debug, Eq, PartialEq)]
+#[derive(Clone, Debug, Eq, PartialEq, serde::Serialize)]
 pub(crate) struct PartShardFingerprint {
     pub size: u64,
     pub sha256: String,
@@ -88,13 +90,15 @@ impl VersionShardCensus {
             && manifest.is_complete()
             && self.data_dir == manifest.data_dir
             && self.erasure_index == manifest.erasure_index
+            && self.data_blocks == manifest.data_blocks
+            && self.parity_blocks == manifest.parity_blocks
             && self.expected_part_numbers == manifest.expected_part_numbers
             && self.present_part_fingerprints == manifest.present_part_fingerprints
             && self.inline_data_fingerprint == manifest.inline_data_fingerprint
     }
 }
 
-fn sha256_hex(data: &[u8]) -> String {
+pub(crate) fn sha256_hex(data: &[u8]) -> String {
     let digest = Sha256::digest(data);
     digest.iter().map(|byte| format!("{byte:02x}")).collect()
 }
@@ -313,6 +317,8 @@ pub(crate) fn census_object_version_on_disk(
             has_xl_meta: false,
             data_dir: None,
             erasure_index: None,
+            data_blocks: None,
+            parity_blocks: None,
             expected_part_numbers: BTreeSet::new(),
             present_part_fingerprints: BTreeMap::new(),
             inline_data_fingerprint: None,
@@ -360,6 +366,8 @@ pub(crate) fn census_object_version_on_disk(
         has_xl_meta: true,
         data_dir,
         erasure_index,
+        data_blocks: Some(file_info.erasure.data_blocks),
+        parity_blocks: Some(file_info.erasure.parity_blocks),
         expected_part_numbers,
         present_part_fingerprints,
         inline_data_fingerprint,
@@ -413,6 +421,8 @@ mod tests {
             has_xl_meta: true,
             data_dir: Some("data-dir".to_string()),
             erasure_index: Some(3),
+            data_blocks: Some(2),
+            parity_blocks: Some(2),
             expected_part_numbers: BTreeSet::from([1]),
             present_part_fingerprints: BTreeMap::from([(1, shard_fingerprint(b"part").unwrap())]),
             inline_data_fingerprint: None,

crates/e2e_test/src/common.rs

@@ -1700,6 +1700,69 @@ impl RustFSTestClusterEnvironment {
         Ok(())
     }
 
+    /// Append a new single-node erasure pool to a stopped multi-pool cluster.
+    ///
+    /// Used to simulate pool expansion on localhost: every pool already owns
+    /// exactly one node with `drives_per_node >= 2` (the only multi-pool layout
+    /// the single-host `RUSTFS_VOLUMES` syntax can express). The new node is
+    /// allocated a fresh port and empty drive directories; callers must
+    /// [`Self::start`] afterwards so every process picks up the extended
+    /// volumes argument. Existing data directories are left untouched.
+    pub async fn append_single_node_pool(&mut self) -> Result<usize, Box<dyn std::error::Error + Send + Sync>> {
+        if self.nodes.iter().any(|node| node.process.is_some()) {
+            return Err("stop the cluster before appending a pool".into());
+        }
+        if self.topology.drives_per_node < 2 {
+            return Err(
+                "append_single_node_pool requires drives_per_node >= 2 (the server parser rejects a single-drive ellipses pool)"
+                    .into(),
+            );
+        }
+
+        let mut pools = self.topology.normalized_pools();
+        for (pool_idx, nodes) in pools.iter().enumerate() {
+            if nodes.len() != 1 {
+                return Err(format!(
+                    "pool {pool_idx} spans {} nodes; append_single_node_pool requires one node per pool",
+                    nodes.len()
+                )
+                .into());
+            }
+        }
+
+        let new_idx = self.nodes.len();
+        let port = RustFSTestEnvironment::find_available_port().await?;
+        let address = format!("127.0.0.1:{port}");
+        let data_dirs: Vec<String> = (0..self.topology.drives_per_node)
+            .map(|drive| format!("{}/node{}/drive{}", self.temp_dir, new_idx, drive))
+            .collect();
+        for dir in &data_dirs {
+            fs::create_dir_all(dir).await?;
+        }
+
+        self.nodes.push(ClusterNode {
+            url: format!("http://{address}"),
+            address,
+            data_dir: data_dirs[0].clone(),
+            data_dirs,
+            pool_idx: pools.len(),
+            process: None,
+        });
+        pools.push(vec![new_idx]);
+        self.topology.node_count = self.nodes.len();
+        self.topology.pools = pools;
+        self.node_extra_env.push(Vec::new());
+        self.node_capture_log_paths.push(None);
+        self.volume_proxy_addresses.push(None);
+
+        if !self.extra_env.iter().any(|(key, _)| key == "RUSTFS_UNSAFE_BYPASS_DISK_CHECK") {
+            self.extra_env
+                .push(("RUSTFS_UNSAFE_BYPASS_DISK_CHECK".to_string(), "true".to_string()));
+        }
+
+        Ok(new_idx)
+    }
+
     /// Gracefully stop one cluster node and wait for its process to exit.
     ///
     /// This is intentionally separate from [`Self::stop_node`]: the latter is

crates/e2e_test/src/data_usage_test.rs

@@ -35,11 +35,15 @@ where
 {
     let mut last_usage = DataUsageInfo::default();
     let mut last_query_error = None;
-    for _ in 0..45 {
+    for _ in 0..90 {
         match get_data_usage_info(env).await {
             Ok(usage) => {
                 last_query_error = None;
-                if usage.buckets_usage.contains_key(bucket) && predicate(&usage) {
+                if usage.is_complete_bucket_usage_snapshot()
+                    && usage.usage_snapshot_converged != Some(false)
+                    && usage.buckets_usage.contains_key(bucket)
+                    && predicate(&usage)
+                {
                     return Ok(usage);
                 }
                 last_usage = usage;

crates/e2e_test/src/distributed/chaos_test.rs

@@ -0,0 +1,222 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{
+    DistCluster, DistLayout, TestResult, assert_object_bytes, payload_for, put_object, retrying_get_equals, unique_bucket,
+    wait_for_ready, wait_until,
+};
+use crate::chaos::{census_object_version_on_disk, signed_admin_post};
+use crate::common::{build_test_s3_config, init_logging};
+use crate::fault_proxy::FaultMode;
+use aws_sdk_s3::Client;
+use std::path::PathBuf;
+use std::sync::Arc;
+use std::time::Duration;
+use tokio::sync::{Barrier, mpsc};
+use tokio::time::timeout;
+
+#[tokio::test]
+async fn kill_and_restart_node_preserves_objects() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("killnode");
+    dist.create_bucket(&bucket).await?;
+    let body = vec![0x11u8; 128 * 1024];
+    put_object(&dist.client(0)?, &bucket, "keep.bin", body.clone()).await?;
+
+    dist.cluster.stop_node(3)?;
+    retrying_get_equals(&dist.client(0)?, &bucket, "keep.bin", &body, Duration::from_secs(20)).await?;
+
+    dist.cluster.start_node(3).await?;
+    wait_for_ready(&dist.cluster).await?;
+    assert_object_bytes(&dist.client(3)?, &bucket, "keep.bin", &body).await?;
+    Ok(())
+}
+
+#[tokio::test]
+async fn full_cluster_restart_preserves_objects() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("pwr");
+    dist.create_bucket(&bucket).await?;
+    let body = vec![0x44u8; 64 * 1024];
+    put_object(&dist.client(1)?, &bucket, "survive.bin", body.clone()).await?;
+
+    dist.cluster.stop();
+    dist.cluster.start().await?;
+    wait_for_ready(&dist.cluster).await?;
+    for node_idx in 0..dist.cluster.nodes.len() {
+        assert_object_bytes(&dist.client(node_idx)?, &bucket, "survive.bin", &body).await?;
+    }
+    Ok(())
+}
+
+#[tokio::test]
+async fn fresh_drive_replacement_is_physically_healed_without_data_change() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start_with_env(DistLayout::FourByFour, &[("RUSTFS_HEAL_ENABLED", "true")]).await?;
+    let bucket = unique_bucket("baddrive");
+    dist.create_bucket(&bucket).await?;
+    let body = payload_for("fresh-drive/durable.bin", 8 * 1024 * 1024);
+    put_object(&dist.client(1)?, &bucket, "durable.bin", body.clone()).await?;
+
+    let replaced_drive = PathBuf::from(&dist.cluster.nodes[0].data_dirs[0]);
+    let baseline = census_object_version_on_disk(&replaced_drive, &bucket, "durable.bin", None)?;
+    assert!(
+        baseline.is_complete(),
+        "replacement target did not hold a complete baseline shard: {baseline:?}"
+    );
+    assert!(
+        !baseline.expected_part_numbers.is_empty(),
+        "replacement witness must use physical part shards: {baseline:?}"
+    );
+
+    dist.cluster.stop_node(0)?;
+    let format_path = replaced_drive.join(".rustfs.sys/format.json");
+    let format = std::fs::read(&format_path)?;
+    let retired_drive = PathBuf::from(format!("{}.retired", replaced_drive.display()));
+    std::fs::rename(&replaced_drive, &retired_drive)?;
+    std::fs::create_dir_all(format_path.parent().ok_or("replacement format path omitted parent")?)?;
+    std::fs::write(&format_path, format)?;
+    let empty = census_object_version_on_disk(&replaced_drive, &bucket, "durable.bin", None)?;
+    assert!(!empty.has_xl_meta, "fresh replacement unexpectedly retained object metadata: {empty:?}");
+
+    dist.cluster.start_node(0).await?;
+    wait_for_ready(&dist.cluster).await?;
+    let heal_body =
+        r#"{"recursive":true,"dryRun":false,"remove":false,"recreate":true,"scanMode":2,"updateParity":false,"nolock":false}"#;
+    let heal_url = format!("{}/rustfs/admin/v3/heal/{bucket}?forceStart=true", dist.cluster.nodes[1].url);
+    signed_admin_post(&heal_url, Some(heal_body), &dist.cluster.access_key, &dist.cluster.secret_key).await?;
+    wait_until(
+        Duration::from_secs(90),
+        || async {
+            let healed = census_object_version_on_disk(&replaced_drive, &bucket, "durable.bin", None)?;
+            Ok(healed.matches_manifest(&baseline))
+        },
+        "fresh replacement contains the original complete shard manifest",
+    )
+    .await?;
+
+    for node_idx in 0..dist.cluster.nodes.len() {
+        assert_object_bytes(&dist.client(node_idx)?, &bucket, "durable.bin", &body).await?;
+    }
+    Ok(())
+}
+
+#[tokio::test]
+async fn concurrent_gets_survive_peer_node_kill() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("getkill");
+    dist.create_bucket(&bucket).await?;
+    let body = payload_for("inflight/steady.bin", 8 * 1024 * 1024);
+    put_object(&dist.client(0)?, &bucket, "steady.bin", body.clone()).await?;
+
+    let live: Vec<_> = (0..3).map(|idx| dist.client(idx)).collect::<Result<Vec<_>, _>>()?;
+    let worker_count = 12;
+    let release = Arc::new(Barrier::new(worker_count + 1));
+    let (started_tx, mut started_rx) = mpsc::unbounded_channel();
+    let mut handles = Vec::new();
+    for idx in 0..worker_count {
+        let client = live[idx % live.len()].clone();
+        let bucket = bucket.clone();
+        let body = body.clone();
+        let release = release.clone();
+        let started_tx = started_tx.clone();
+        handles.push(tokio::spawn(async move {
+            let response = client.get_object().bucket(&bucket).key("steady.bin").send().await?;
+            if response.content_length() != Some(body.len() as i64) {
+                return Err::<(), Box<dyn std::error::Error + Send + Sync>>(
+                    format!("worker {idx} received a wrong content length").into(),
+                );
+            }
+            started_tx.send(idx)?;
+            release.wait().await;
+            let actual = response.body.collect().await?.into_bytes();
+            if actual.as_ref() != body.as_slice() {
+                return Err(format!("worker {idx} received corrupted bytes after peer kill").into());
+            }
+            Ok(())
+        }));
+    }
+    drop(started_tx);
+    for _ in 0..worker_count {
+        timeout(Duration::from_secs(30), started_rx.recv())
+            .await?
+            .ok_or("a streaming GET exited before reaching the kill barrier")?;
+    }
+
+    dist.cluster.stop_node(3)?;
+    release.wait().await;
+    for handle in handles {
+        handle.await??;
+    }
+
+    dist.cluster.start_node(3).await?;
+    wait_for_ready(&dist.cluster).await?;
+    assert_object_bytes(&dist.client(3)?, &bucket, "steady.bin", &body).await?;
+    Ok(())
+}
+
+#[tokio::test]
+async fn blackholed_node_client_network_preserves_cluster_availability_and_recovers() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let proxy = crate::fault_proxy::FaultProxy::start(dist.cluster.nodes[3].address.parse()?).await?;
+    let proxied_url = format!("http://{}", proxy.local_addr());
+    let proxied_client = Client::from_conf(build_test_s3_config(
+        &proxied_url,
+        &dist.cluster.access_key,
+        &dist.cluster.secret_key,
+        None,
+        "distributed-network-chaos",
+    ));
+
+    let result: TestResult = async {
+        let bucket = unique_bucket("netfault");
+        dist.create_bucket(&bucket).await?;
+        let baseline = payload_for("network/baseline.bin", 1024 * 1024);
+        put_object(&dist.client(0)?, &bucket, "baseline.bin", baseline.clone()).await?;
+        assert_object_bytes(&proxied_client, &bucket, "baseline.bin", &baseline).await?;
+
+        proxy.set_mode(FaultMode::Blackhole);
+        assert_eq!(proxy.mode(), FaultMode::Blackhole);
+        if let Ok(Ok(_)) = timeout(
+            Duration::from_secs(5),
+            proxied_client.get_object().bucket(&bucket).key("baseline.bin").send(),
+        )
+        .await
+        {
+            return Err("blackholed node endpoint unexpectedly completed a GET".into());
+        }
+
+        let during = payload_for("network/during.bin", 1024 * 1024);
+        timeout(Duration::from_secs(30), async {
+            put_object(&dist.client(1)?, &bucket, "during-blackhole.bin", during.clone()).await?;
+            assert_object_bytes(&dist.client(2)?, &bucket, "baseline.bin", &baseline).await?;
+            assert_object_bytes(&dist.client(0)?, &bucket, "during-blackhole.bin", &during).await?;
+            Ok::<_, Box<dyn std::error::Error + Send + Sync>>(())
+        })
+        .await??;
+
+        proxy.set_mode(FaultMode::Pass);
+        retrying_get_equals(&proxied_client, &bucket, "during-blackhole.bin", &during, Duration::from_secs(30)).await?;
+        Ok(())
+    }
+    .await;
+
+    proxy.set_mode(FaultMode::Pass);
+    proxy.shutdown().await;
+    result
+}

crates/e2e_test/src/distributed/concurrency_stability_test.rs

@@ -0,0 +1,98 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{DistCluster, DistLayout, TestResult, assert_object_bytes, payload_for, put_object, unique_bucket};
+use crate::common::init_logging;
+use std::collections::BTreeSet;
+use std::sync::Arc;
+use tokio::sync::Barrier;
+
+#[tokio::test]
+async fn four_node_high_concurrency_mixed_workload_is_consistent_on_every_node() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("conc");
+    dist.create_bucket(&bucket).await?;
+    let clients = Arc::new(dist.clients()?);
+    let worker_count = 24;
+    let rounds = 4;
+    let barrier = Arc::new(Barrier::new(worker_count));
+
+    let mut handles = Vec::new();
+    for idx in 0..worker_count {
+        let clients = clients.clone();
+        let barrier = barrier.clone();
+        let bucket = bucket.clone();
+        handles.push(tokio::spawn(async move {
+            barrier.wait().await;
+            let writer = &clients[idx % clients.len()];
+            let reader = &clients[(idx + 1) % clients.len()];
+            let copier = &clients[(idx + 2) % clients.len()];
+            let mut retained = Vec::with_capacity(rounds);
+            for round in 0..rounds {
+                let key = format!("source/worker-{idx:02}-round-{round}.bin");
+                let copy_key = format!("retained/worker-{idx:02}-round-{round}.bin");
+                let body = payload_for(&key, 64 * 1024);
+                put_object(writer, &bucket, &key, body.clone()).await?;
+
+                let head = reader.head_object().bucket(&bucket).key(&key).send().await?;
+                if head.content_length() != Some(body.len() as i64) {
+                    return Err(format!("HEAD returned the wrong size for {key}: {head:?}").into());
+                }
+                assert_object_bytes(reader, &bucket, &key, &body).await?;
+
+                copier
+                    .copy_object()
+                    .bucket(&bucket)
+                    .key(&copy_key)
+                    .copy_source(format!("{bucket}/{key}"))
+                    .send()
+                    .await?;
+                assert_object_bytes(writer, &bucket, &copy_key, &body).await?;
+
+                writer.delete_object().bucket(&bucket).key(&key).send().await?;
+                let missing = reader
+                    .head_object()
+                    .bucket(&bucket)
+                    .key(&key)
+                    .send()
+                    .await
+                    .expect_err("deleted source key must not remain visible");
+                if missing.raw_response().map(|response| response.status().as_u16()) != Some(404) {
+                    return Err(format!("deleted source {key} returned an unexpected result: {missing:?}").into());
+                }
+                retained.push((copy_key, body));
+            }
+            Ok::<_, Box<dyn std::error::Error + Send + Sync>>(retained)
+        }));
+    }
+
+    let mut inventory = Vec::new();
+    for handle in handles {
+        inventory.extend(handle.await??);
+    }
+
+    let expected_keys: BTreeSet<_> = inventory.iter().map(|(key, _)| key.as_str()).collect();
+    for (node_idx, client) in clients.iter().enumerate() {
+        let listed = client.list_objects_v2().bucket(&bucket).prefix("retained/").send().await?;
+        let listed_keys: BTreeSet<_> = listed.contents().iter().filter_map(|object| object.key()).collect();
+        assert_eq!(listed_keys, expected_keys, "node {node_idx} returned a divergent retained-key listing");
+        for (key, body) in &inventory {
+            assert_object_bytes(client, &bucket, key, body)
+                .await
+                .map_err(|error| format!("node {node_idx} failed to read {key}: {error}"))?;
+        }
+    }
+    Ok(())
+}

crates/e2e_test/src/distributed/concurrent_data_movement_test.rs

@@ -0,0 +1,74 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{
+    DECOMMISSION_POOL_ID, DistCluster, DistLayout, TestResult, assert_inventory, decommission_running_with_progress,
+    decommission_status_json, payload_for, put_inventory_retrying, retrying_get_equals, retrying_put, start_decommission,
+    unique_bucket, wait_for_decommission_complete, wait_for_decommission_running_with_progress,
+};
+use crate::common::init_logging;
+use std::sync::Arc;
+use std::time::Duration;
+use tokio::sync::Barrier;
+
+#[tokio::test]
+async fn concurrent_puts_during_decommission_do_not_lose_baseline_or_new_objects() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start(DistLayout::SingleNodeFourDrive).await?;
+    let bucket = unique_bucket("concdecom");
+    dist.create_bucket(&bucket).await?;
+    let baseline_client = dist.client(0)?;
+    let inventory = put_inventory_retrying(&baseline_client, &bucket, 96, 256 * 1024, Duration::from_secs(30)).await?;
+    dist.expand_to_four_pools().await?;
+
+    start_decommission(&dist.cluster, DECOMMISSION_POOL_ID).await?;
+
+    let clients = Arc::new(dist.clients()?);
+    let barrier = Arc::new(Barrier::new(17));
+    let mut handles = Vec::new();
+    for idx in 0..16 {
+        let clients = clients.clone();
+        let barrier = barrier.clone();
+        let bucket = bucket.clone();
+        handles.push(tokio::spawn(async move {
+            barrier.wait().await;
+            let client = &clients[idx % clients.len()];
+            let key = format!("live/{idx:02}.bin");
+            let body = payload_for(&key, 8 * 1024);
+            retrying_put(client, &bucket, &key, body.clone(), Duration::from_secs(45)).await?;
+            Ok::<_, Box<dyn std::error::Error + Send + Sync>>((key, body))
+        }));
+    }
+
+    wait_for_decommission_running_with_progress(&dist.cluster, DECOMMISSION_POOL_ID, Duration::from_secs(30)).await?;
+    barrier.wait().await;
+
+    let mut live_objects = Vec::new();
+    for handle in handles {
+        live_objects.push(handle.await??);
+    }
+    let status = decommission_status_json(&dist.cluster).await?;
+    if !decommission_running_with_progress(&status, DECOMMISSION_POOL_ID)? {
+        return Err(format!("decommission did not remain active across concurrent PUTs: {status}").into());
+    }
+
+    wait_for_decommission_complete(&dist.cluster, DECOMMISSION_POOL_ID, Duration::from_secs(180)).await?;
+
+    let checker = dist.client(2)?;
+    assert_inventory(&checker, &bucket, &inventory).await?;
+    for (key, body) in live_objects {
+        retrying_get_equals(&checker, &bucket, &key, &body, Duration::from_secs(30)).await?;
+    }
+    Ok(())
+}

crates/e2e_test/src/distributed/data_integrity_movement_test.rs

@@ -0,0 +1,156 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{
+    DECOMMISSION_POOL_ID, DistCluster, DistLayout, TestResult, assert_inventory, enable_versioning, put_inventory_retrying,
+    sha256_hex, start_decommission, unique_bucket, wait_for_decommission_active, wait_for_decommission_complete,
+};
+use crate::common::init_logging;
+use aws_sdk_s3::primitives::ByteStream;
+use aws_sdk_s3::types::{CompletedMultipartUpload, CompletedPart};
+use std::time::Duration;
+
+#[tokio::test]
+async fn decommission_does_not_alter_object_sha256_across_pools() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start(DistLayout::SingleNodeFourDrive).await?;
+    let bucket = unique_bucket("integrity");
+    dist.create_bucket(&bucket).await?;
+    let client = dist.client(0)?;
+    enable_versioning(&client, &bucket).await?;
+    let inventory = put_inventory_retrying(&client, &bucket, 96, 256 * 1024, Duration::from_secs(30)).await?;
+    let before: Vec<(String, String)> = inventory.iter().map(|(key, body)| (key.clone(), sha256_hex(body))).collect();
+
+    let versioned_key = "history/versioned.bin";
+    let version_one = b"historical bytes before data movement".to_vec();
+    let version_two = b"current bytes before data movement".to_vec();
+    let version_one_id = client
+        .put_object()
+        .bucket(&bucket)
+        .key(versioned_key)
+        .body(ByteStream::from(version_one.clone()))
+        .send()
+        .await?
+        .version_id()
+        .ok_or("historical PUT omitted version ID")?
+        .to_string();
+    let version_two_id = client
+        .put_object()
+        .bucket(&bucket)
+        .key(versioned_key)
+        .body(ByteStream::from(version_two.clone()))
+        .send()
+        .await?
+        .version_id()
+        .ok_or("current PUT omitted version ID")?
+        .to_string();
+
+    let multipart_key = "multipart/moved.bin";
+    let first_part = vec![0x31; 5 * 1024 * 1024];
+    let second_part = vec![0x72; 1024 * 1024];
+    let upload = client
+        .create_multipart_upload()
+        .bucket(&bucket)
+        .key(multipart_key)
+        .send()
+        .await?;
+    let upload_id = upload.upload_id().ok_or("movement multipart upload omitted upload ID")?;
+    let uploaded_one = client
+        .upload_part()
+        .bucket(&bucket)
+        .key(multipart_key)
+        .upload_id(upload_id)
+        .part_number(1)
+        .body(ByteStream::from(first_part.clone()))
+        .send()
+        .await?;
+    let uploaded_two = client
+        .upload_part()
+        .bucket(&bucket)
+        .key(multipart_key)
+        .upload_id(upload_id)
+        .part_number(2)
+        .body(ByteStream::from(second_part.clone()))
+        .send()
+        .await?;
+    client
+        .complete_multipart_upload()
+        .bucket(&bucket)
+        .key(multipart_key)
+        .upload_id(upload_id)
+        .multipart_upload(
+            CompletedMultipartUpload::builder()
+                .parts(
+                    CompletedPart::builder()
+                        .part_number(1)
+                        .e_tag(uploaded_one.e_tag().ok_or("movement part 1 omitted ETag")?)
+                        .build(),
+                )
+                .parts(
+                    CompletedPart::builder()
+                        .part_number(2)
+                        .e_tag(uploaded_two.e_tag().ok_or("movement part 2 omitted ETag")?)
+                        .build(),
+                )
+                .build(),
+        )
+        .send()
+        .await?;
+
+    dist.expand_to_four_pools().await?;
+
+    start_decommission(&dist.cluster, DECOMMISSION_POOL_ID).await?;
+    wait_for_decommission_active(&dist.cluster, DECOMMISSION_POOL_ID, Duration::from_secs(30)).await?;
+    wait_for_decommission_complete(&dist.cluster, DECOMMISSION_POOL_ID, Duration::from_secs(180)).await?;
+
+    let after_client = dist.client(2)?;
+    assert_inventory(&after_client, &bucket, &inventory).await?;
+    for (key, expected_hash) in before {
+        let got = after_client.get_object().bucket(&bucket).key(&key).send().await?;
+        let body = got.body.collect().await?.into_bytes();
+        assert_eq!(sha256_hex(body.as_ref()), expected_hash, "checksum changed for {key} after decommission");
+    }
+    for (version_id, expected) in [(&version_one_id, &version_one), (&version_two_id, &version_two)] {
+        let got = after_client
+            .get_object()
+            .bucket(&bucket)
+            .key(versioned_key)
+            .version_id(version_id)
+            .send()
+            .await?
+            .body
+            .collect()
+            .await?
+            .into_bytes();
+        assert_eq!(got.as_ref(), expected.as_slice(), "version {version_id} changed after decommission");
+    }
+    let mut expected_multipart = first_part;
+    expected_multipart.extend_from_slice(&second_part);
+    let got_multipart = after_client
+        .get_object()
+        .bucket(&bucket)
+        .key(multipart_key)
+        .send()
+        .await?
+        .body
+        .collect()
+        .await?
+        .into_bytes();
+    assert_eq!(
+        sha256_hex(got_multipart.as_ref()),
+        sha256_hex(&expected_multipart),
+        "multipart checksum changed after decommission"
+    );
+    Ok(())
+}

crates/e2e_test/src/distributed/expand_decommission_rebalance_test.rs

@@ -0,0 +1,81 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{
+    DECOMMISSION_POOL_ID, DistCluster, DistLayout, TestResult, assert_inventory, list_pools_json, put_inventory,
+    put_inventory_retrying, start_decommission, start_rebalance, unique_bucket, wait_for_decommission_active,
+    wait_for_decommission_complete, wait_for_rebalance_active, wait_for_rebalance_complete,
+};
+use crate::common::init_logging;
+use std::time::Duration;
+
+#[tokio::test]
+async fn four_node_pool_expand_preserves_objects_then_rebalance() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start(DistLayout::SingleNodeFourDrive).await?;
+    let bucket = unique_bucket("expand");
+    dist.create_bucket(&bucket).await?;
+    let client = dist.client(0)?;
+    let inventory = put_inventory(&client, &bucket, 64, 256 * 1024).await?;
+    assert_inventory(&client, &bucket, &inventory).await?;
+
+    for expected_nodes in 2..=4 {
+        let new_node = dist.append_pool_and_restart().await?;
+        assert_eq!(new_node + 1, expected_nodes);
+        assert_inventory(&dist.client(new_node)?, &bucket, &inventory).await?;
+    }
+    assert_eq!(dist.cluster.nodes.len(), 4);
+
+    // Prove that the expanded pool map is durable, and clear any recovery
+    // latch raised while the newly-added pool replicas converged.
+    dist.restart_current_binary_gracefully().await?;
+
+    let after_expand = dist.client(0)?;
+    assert_inventory(&after_expand, &bucket, &inventory).await?;
+    let peer = dist.client(3)?;
+    assert_inventory(&peer, &bucket, &inventory).await?;
+
+    let rebalance_id = start_rebalance(&dist.cluster).await?;
+    wait_for_rebalance_active(&dist.cluster, &rebalance_id, Duration::from_secs(30)).await?;
+    wait_for_rebalance_complete(&dist.cluster, &rebalance_id, Duration::from_secs(180)).await?;
+    assert_inventory(&peer, &bucket, &inventory).await?;
+    Ok(())
+}
+
+#[tokio::test]
+async fn four_pool_decommission_moves_objects_without_loss() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start(DistLayout::SingleNodeFourDrive).await?;
+    let bucket = unique_bucket("decom");
+    dist.create_bucket(&bucket).await?;
+    let client = dist.client(0)?;
+    let inventory = put_inventory_retrying(&client, &bucket, 96, 128 * 1024, Duration::from_secs(30)).await?;
+    dist.expand_to_four_pools().await?;
+
+    let pools_before = list_pools_json(&dist.cluster).await?;
+    let pool_count = pools_before
+        .as_array()
+        .map(Vec::len)
+        .or_else(|| pools_before.get("pools").and_then(serde_json::Value::as_array).map(Vec::len))
+        .ok_or_else(|| format!("pool list omitted an array: {pools_before}"))?;
+    assert_eq!(pool_count, 4, "expected exactly four pools before decommission: {pools_before}");
+
+    start_decommission(&dist.cluster, DECOMMISSION_POOL_ID).await?;
+    wait_for_decommission_active(&dist.cluster, DECOMMISSION_POOL_ID, Duration::from_secs(30)).await?;
+    wait_for_decommission_complete(&dist.cluster, DECOMMISSION_POOL_ID, Duration::from_secs(180)).await?;
+
+    let after = dist.client(2)?;
+    assert_inventory(&after, &bucket, &inventory).await?;
+    Ok(())
+}

crates/e2e_test/src/distributed/extra_test.rs

@@ -0,0 +1,149 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{
+    DistCluster, DistLayout, TestResult, assert_object_bytes, get_object_bytes, put_object, unique_bucket, wait_until,
+};
+use crate::common::init_logging;
+use aws_sdk_s3::primitives::ByteStream;
+use aws_sdk_s3::types::{CompletedMultipartUpload, CompletedPart};
+use std::time::Duration;
+
+#[tokio::test]
+async fn four_node_four_drive_multipart_and_cross_node_listing_agree() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("extra");
+    dist.create_bucket(&bucket).await?;
+    let client = dist.client(0)?;
+
+    let key = "multipart.bin";
+    let part1 = vec![0x41u8; 5 * 1024 * 1024];
+    let part2 = vec![0x42u8; 5 * 1024 * 1024];
+    let upload = client.create_multipart_upload().bucket(&bucket).key(key).send().await?;
+    let upload_id = upload.upload_id().ok_or("missing upload id")?.to_string();
+
+    let uploaded1 = client
+        .upload_part()
+        .bucket(&bucket)
+        .key(key)
+        .upload_id(&upload_id)
+        .part_number(1)
+        .body(ByteStream::from(part1.clone()))
+        .send()
+        .await?;
+    let uploaded2 = client
+        .upload_part()
+        .bucket(&bucket)
+        .key(key)
+        .upload_id(&upload_id)
+        .part_number(2)
+        .body(ByteStream::from(part2.clone()))
+        .send()
+        .await?;
+
+    client
+        .complete_multipart_upload()
+        .bucket(&bucket)
+        .key(key)
+        .upload_id(&upload_id)
+        .multipart_upload(
+            CompletedMultipartUpload::builder()
+                .parts(
+                    CompletedPart::builder()
+                        .part_number(1)
+                        .e_tag(uploaded1.e_tag().unwrap_or_default())
+                        .build(),
+                )
+                .parts(
+                    CompletedPart::builder()
+                        .part_number(2)
+                        .e_tag(uploaded2.e_tag().unwrap_or_default())
+                        .build(),
+                )
+                .build(),
+        )
+        .send()
+        .await?;
+
+    let mut expected = part1;
+    expected.extend_from_slice(&part2);
+    for node_idx in 0..dist.cluster.nodes.len() {
+        assert_object_bytes(&dist.client(node_idx)?, &bucket, key, &expected).await?;
+    }
+
+    put_object(&client, &bucket, "list/a", b"a".to_vec()).await?;
+    put_object(&dist.client(2)?, &bucket, "list/b", b"b".to_vec()).await?;
+    let mut seen = Vec::new();
+    for node_idx in 0..dist.cluster.nodes.len() {
+        let listed = dist
+            .client(node_idx)?
+            .list_objects_v2()
+            .bucket(&bucket)
+            .prefix("list/")
+            .send()
+            .await?;
+        let keys: Vec<String> = listed
+            .contents()
+            .iter()
+            .filter_map(|object| object.key().map(str::to_string))
+            .collect();
+        seen.push(keys);
+    }
+    for keys in &seen[1..] {
+        assert_eq!(&seen[0], keys, "list results diverged across nodes: {seen:?}");
+    }
+
+    let got = get_object_bytes(&dist.client(3)?, &bucket, "list/a").await?;
+    assert_eq!(got, b"a");
+    Ok(())
+}
+
+#[tokio::test]
+async fn four_node_list_buckets_agree_across_all_nodes() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("listed");
+    dist.create_bucket(&bucket).await?;
+    put_object(&dist.client(0)?, &bucket, "seed.bin", b"seed".to_vec()).await?;
+
+    for node_idx in 0..dist.cluster.nodes.len() {
+        let client = dist.client(node_idx)?;
+        let name = bucket.clone();
+        wait_until(
+            Duration::from_secs(20),
+            || {
+                let client = client.clone();
+                let name = name.clone();
+                async move {
+                    let listed = client.list_buckets().send().await?;
+                    Ok(listed.buckets().iter().any(|entry| entry.name() == Some(name.as_str())))
+                }
+            },
+            &format!("node {node_idx} lists {bucket}"),
+        )
+        .await?;
+        wait_until(
+            Duration::from_secs(20),
+            || {
+                let client = dist.client(node_idx).expect("client");
+                let name = bucket.clone();
+                async move { Ok(get_object_bytes(&client, &name, "seed.bin").await.ok() == Some(b"seed".to_vec())) }
+            },
+            &format!("node {node_idx} reads seed.bin"),
+        )
+        .await?;
+    }
+    Ok(())
+}

crates/e2e_test/src/distributed/mod.rs

@@ -0,0 +1,35 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! 4-node 4-drive distributed e2e coverage.
+//!
+//! Selected by `[profile.e2e-distributed]` and run from
+//! `.github/workflows/e2e-distributed.yml`. Excluded from `e2e-full` because
+//! each case starts four real `rustfs` processes.
+
+mod chaos_test;
+mod concurrency_stability_test;
+mod concurrent_data_movement_test;
+mod data_integrity_movement_test;
+mod expand_decommission_rebalance_test;
+mod extra_test;
+mod harness;
+mod object_lock_test;
+mod observability_test;
+mod replication_quota_test;
+mod s3_basic_test;
+mod s3_during_data_movement_test;
+mod site_replication_test;
+mod upgrade_test;
+mod versioning_test;

crates/e2e_test/src/distributed/object_lock_test.rs

@@ -0,0 +1,219 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{DistCluster, DistLayout, TestResult, unique_bucket};
+use crate::common::init_logging;
+use crate::object_lock::common::{
+    delete_object_with_bypass, put_object_lock_configuration, put_object_with_legal_hold, put_object_with_retention,
+};
+use aws_sdk_s3::Client;
+use aws_sdk_s3::error::ProvideErrorMetadata;
+use aws_sdk_s3::error::SdkError;
+use aws_sdk_s3::operation::delete_object::DeleteObjectError;
+use aws_sdk_s3::primitives::ByteStream;
+use aws_sdk_s3::types::{
+    DefaultRetention, ObjectLockConfiguration, ObjectLockEnabled, ObjectLockLegalHoldStatus, ObjectLockRetentionMode,
+    ObjectLockRule,
+};
+use chrono::{Duration as ChronoDuration, Utc};
+
+fn delete_denied(error: &SdkError<DeleteObjectError>, context: &str) -> TestResult {
+    let code = error.as_service_error().and_then(ProvideErrorMetadata::code);
+    if code == Some("AccessDenied") {
+        Ok(())
+    } else {
+        Err(format!("{context}: expected AccessDenied, got {error:?}").into())
+    }
+}
+
+async fn expect_versioned_delete_denied(
+    client: &Client,
+    bucket: &str,
+    key: &str,
+    version_id: &str,
+    bypass: bool,
+    context: &str,
+) -> TestResult {
+    match delete_object_with_bypass(client, bucket, key, Some(version_id), bypass).await {
+        Ok(_) => Err(format!("{context}: DeleteObject of retained version must be denied").into()),
+        Err(error) => delete_denied(error.as_ref(), context),
+    }
+}
+
+#[tokio::test]
+async fn four_node_four_drive_object_lock_worm_blocks_delete() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let client = dist.client(0)?;
+    let peer = dist.client(2)?;
+    let bucket = unique_bucket("objlock");
+
+    client
+        .create_bucket()
+        .bucket(&bucket)
+        .object_lock_enabled_for_bucket(true)
+        .send()
+        .await?;
+
+    let retain_until = Utc::now() + ChronoDuration::days(1);
+
+    let compliance_key = "compliance.bin";
+    let compliance_version = put_object_with_retention(
+        &client,
+        &bucket,
+        compliance_key,
+        b"locked-compliance",
+        ObjectLockRetentionMode::Compliance,
+        retain_until,
+    )
+    .await?;
+
+    // Unversioned DELETE is allowed: it only creates a delete marker. WORM
+    // applies to a specific version id.
+    let marker = peer.delete_object().bucket(&bucket).key(compliance_key).send().await?;
+    assert_eq!(
+        marker.delete_marker(),
+        Some(true),
+        "unversioned DELETE on a locked object must create a delete marker"
+    );
+
+    expect_versioned_delete_denied(&peer, &bucket, compliance_key, &compliance_version, false, "COMPLIANCE without bypass")
+        .await?;
+    expect_versioned_delete_denied(&peer, &bucket, compliance_key, &compliance_version, true, "COMPLIANCE with bypass").await?;
+
+    let governance_key = "governance.bin";
+    let governance_version = put_object_with_retention(
+        &client,
+        &bucket,
+        governance_key,
+        b"locked-governance",
+        ObjectLockRetentionMode::Governance,
+        retain_until,
+    )
+    .await?;
+
+    expect_versioned_delete_denied(&peer, &bucket, governance_key, &governance_version, false, "GOVERNANCE without bypass")
+        .await?;
+    delete_object_with_bypass(&peer, &bucket, governance_key, Some(&governance_version), true).await?;
+    let deleted_governance = peer
+        .head_object()
+        .bucket(&bucket)
+        .key(governance_key)
+        .version_id(&governance_version)
+        .send()
+        .await
+        .expect_err("GOVERNANCE bypass must remove the retained version");
+    assert_eq!(
+        deleted_governance.raw_response().map(|response| response.status().as_u16()),
+        Some(404),
+        "deleted GOVERNANCE version returned an unexpected HEAD result: {deleted_governance:?}"
+    );
+
+    let hold_key = "legal-hold.bin";
+    let hold_version =
+        put_object_with_legal_hold(&client, &bucket, hold_key, b"legal-hold", ObjectLockLegalHoldStatus::On).await?;
+    expect_versioned_delete_denied(&peer, &bucket, hold_key, &hold_version, false, "legal hold without bypass").await?;
+    expect_versioned_delete_denied(&peer, &bucket, hold_key, &hold_version, true, "legal hold with bypass").await?;
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn four_node_default_retention_is_visible_and_non_lock_bucket_rejects_configuration() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let writer = dist.client(0)?;
+    let reader = dist.client(3)?;
+    let bucket = unique_bucket("default-lock");
+
+    writer
+        .create_bucket()
+        .bucket(&bucket)
+        .object_lock_enabled_for_bucket(true)
+        .send()
+        .await?;
+    put_object_lock_configuration(&writer, &bucket, ObjectLockRetentionMode::Governance, Some(1), None).await?;
+
+    let key = "default-governance.bin";
+    let put = writer
+        .put_object()
+        .bucket(&bucket)
+        .key(key)
+        .body(ByteStream::from_static(b"default retention payload"))
+        .send()
+        .await?;
+    let version_id = put.version_id().ok_or("default-retained PUT omitted version ID")?;
+
+    let config = reader.get_object_lock_configuration().bucket(&bucket).send().await?;
+    let default_retention = config
+        .object_lock_configuration()
+        .and_then(|configuration| configuration.rule())
+        .and_then(|rule| rule.default_retention())
+        .ok_or("GetObjectLockConfiguration omitted default retention")?;
+    assert_eq!(default_retention.mode().map(|mode| mode.as_str()), Some("GOVERNANCE"));
+    assert_eq!(default_retention.days(), Some(1));
+
+    let retention = reader
+        .get_object_retention()
+        .bucket(&bucket)
+        .key(key)
+        .version_id(version_id)
+        .send()
+        .await?;
+    let retention = retention.retention().ok_or("GetObjectRetention omitted applied retention")?;
+    assert_eq!(retention.mode().map(|mode| mode.as_str()), Some("GOVERNANCE"));
+    let retain_until = retention
+        .retain_until_date()
+        .ok_or("default retention omitted retain-until date")?;
+    assert!(retain_until.secs() > Utc::now().timestamp(), "default retention is not in the future");
+
+    let versioning = reader.get_bucket_versioning().bucket(&bucket).send().await?;
+    assert_eq!(versioning.status().map(|status| status.as_str()), Some("Enabled"));
+    expect_versioned_delete_denied(&reader, &bucket, key, version_id, false, "default GOVERNANCE retention without bypass")
+        .await?;
+
+    let plain_bucket = unique_bucket("no-lock");
+    dist.create_bucket(&plain_bucket).await?;
+    let configuration = ObjectLockConfiguration::builder()
+        .object_lock_enabled(ObjectLockEnabled::Enabled)
+        .rule(
+            ObjectLockRule::builder()
+                .default_retention(
+                    DefaultRetention::builder()
+                        .mode(ObjectLockRetentionMode::Governance)
+                        .days(1)
+                        .build(),
+                )
+                .build(),
+        )
+        .build();
+    let error = writer
+        .put_object_lock_configuration()
+        .bucket(&plain_bucket)
+        .object_lock_configuration(configuration)
+        .send()
+        .await
+        .expect_err("an unversioned bucket must reject Object Lock enablement");
+    let service_error = error
+        .as_service_error()
+        .ok_or("non-lock bucket rejection was not an S3 service error")?;
+    assert_eq!(service_error.code(), Some("InvalidBucketState"), "unexpected error: {error:?}");
+    assert_eq!(
+        service_error.message(),
+        Some("Object Lock configuration cannot be enabled on existing buckets"),
+        "unexpected error: {error:?}"
+    );
+
+    Ok(())
+}

crates/e2e_test/src/distributed/observability_test.rs

@@ -0,0 +1,236 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{DistCluster, DistLayout, TestResult, cluster_admin_ok, unique_bucket, wait_for_ready};
+use crate::common::{admin_request, init_logging, local_http_client};
+use aws_sdk_s3::operation::RequestId;
+use aws_sdk_s3::primitives::ByteStream;
+use bytes::Bytes;
+use http::Method;
+use http_body_util::{BodyExt, Empty};
+use hyper::body::Incoming;
+use hyper::service::service_fn;
+use hyper::{Request, Response};
+use hyper_util::rt::TokioIo;
+use local_ip_address::local_ip;
+use rustfs_madmin::metrics::RealtimeMetrics;
+use rustfs_utils::egress::ENV_OUTBOUND_ALLOW_ORIGINS;
+use serde_json::Value;
+use std::convert::Infallible;
+use std::time::Duration;
+use tokio::net::TcpListener;
+use tokio::sync::mpsc;
+use tokio::task::JoinHandle;
+use tokio::time::{Instant, timeout};
+
+async fn spawn_audit_collector() -> TestResult<(String, mpsc::UnboundedReceiver<Value>, JoinHandle<()>)> {
+    let listener = TcpListener::bind("0.0.0.0:0").await?;
+    let endpoint = format!("http://{}/audit", std::net::SocketAddr::new(local_ip()?, listener.local_addr()?.port()));
+    let (tx, rx) = mpsc::unbounded_channel();
+    let handle = tokio::spawn(async move {
+        loop {
+            let Ok((stream, _)) = listener.accept().await else {
+                return;
+            };
+            let tx = tx.clone();
+            tokio::spawn(async move {
+                let service = service_fn(move |request: Request<Incoming>| {
+                    let tx = tx.clone();
+                    async move {
+                        let method = request.method().clone();
+                        if let Ok(body) = request.into_body().collect().await
+                            && method == Method::POST
+                            && let Ok(payload) = serde_json::from_slice::<Value>(&body.to_bytes())
+                        {
+                            if let Some(records) = payload["Records"].as_array() {
+                                for entry in records {
+                                    let _ = tx.send(entry.clone());
+                                }
+                            } else {
+                                let _ = tx.send(payload);
+                            }
+                        }
+                        Ok::<_, Infallible>(Response::new(Empty::<Bytes>::new()))
+                    }
+                });
+                let _ = hyper::server::conn::http1::Builder::new()
+                    .serve_connection(TokioIo::new(stream), service)
+                    .await;
+            });
+        }
+    });
+    Ok((endpoint, rx, handle))
+}
+
+async fn wait_for_audit_entry(
+    rx: &mut mpsc::UnboundedReceiver<Value>,
+    bucket: &str,
+    key: &str,
+    request_id: &str,
+) -> TestResult<Value> {
+    let deadline = Instant::now() + Duration::from_secs(30);
+    let mut seen = Vec::new();
+    loop {
+        let remaining = deadline.saturating_duration_since(Instant::now());
+        if remaining.is_zero() {
+            return Err(format!(
+                "audit webhook did not receive PutObject for {bucket}/{key}; received {} other records: {seen:?}",
+                seen.len()
+            )
+            .into());
+        }
+        let entry = match timeout(remaining, rx.recv()).await {
+            Ok(Some(entry)) => entry,
+            Ok(None) => return Err("audit collector stopped before the expected entry arrived".into()),
+            Err(_) => {
+                return Err(format!(
+                    "audit webhook did not receive PutObject for {bucket}/{key}; received {} other records: {seen:?}",
+                    seen.len()
+                )
+                .into());
+            }
+        };
+        if entry["api"]["name"].as_str() == Some("s3:PutObject")
+            && entry["api"]["bucket"].as_str() == Some(bucket)
+            && entry["api"]["object"].as_str() == Some(key)
+            && entry["requestID"].as_str() == Some(request_id)
+        {
+            return Ok(entry);
+        }
+        if seen.len() < 8 {
+            seen.push(format!(
+                "api={:?} bucket={:?} object={:?} requestID={:?}",
+                entry["api"]["name"].as_str(),
+                entry["api"]["bucket"].as_str(),
+                entry["api"]["object"].as_str(),
+                entry["requestID"].as_str()
+            ));
+        }
+    }
+}
+
+#[tokio::test]
+async fn four_node_health_inventory_metrics_and_audit_delivery_are_consistent() -> TestResult {
+    init_logging();
+    let (audit_endpoint, mut audit_entries, collector) = spawn_audit_collector().await?;
+    let audit_origin = reqwest::Url::parse(&audit_endpoint)?.origin().ascii_serialization();
+    let audit_env = [
+        ("RUSTFS_AUDIT_ENABLE", "true"),
+        ("RUSTFS_AUDIT_WEBHOOK_ENABLE_DISTRIBUTED", "on"),
+        ("RUSTFS_AUDIT_WEBHOOK_ENDPOINT_DISTRIBUTED", audit_endpoint.as_str()),
+        (ENV_OUTBOUND_ALLOW_ORIGINS, audit_origin.as_str()),
+    ];
+    let mut dist = DistCluster::new_stopped_with_env(DistLayout::FourByFour, &audit_env).await?;
+    for node_idx in 0..dist.cluster.nodes.len() {
+        let queue_dir = format!("{}/audit-queue-node-{node_idx}", dist.cluster.temp_dir);
+        tokio::fs::create_dir_all(&queue_dir).await?;
+        dist.cluster
+            .set_node_env(node_idx, "RUSTFS_AUDIT_WEBHOOK_QUEUE_DIR_DISTRIBUTED", queue_dir)?;
+    }
+    dist.cluster.start().await?;
+    wait_for_ready(&dist.cluster).await?;
+
+    let http = local_http_client();
+    for node in &dist.cluster.nodes {
+        for probe in ["ready", "live"] {
+            let response = http.get(format!("{}/health/{probe}", node.url)).send().await?;
+            assert!(
+                response.status().is_success(),
+                "node {} {probe} probe failed: {}",
+                node.address,
+                response.status()
+            );
+        }
+    }
+
+    let info_body = cluster_admin_ok(&dist.cluster, Method::GET, "/rustfs/admin/v3/info", None).await?;
+    let info: Value = serde_json::from_str(&info_body)?;
+    let servers = info["info"]["servers"]
+        .as_array()
+        .ok_or_else(|| format!("admin info omitted servers: {info}"))?;
+    assert_eq!(servers.len(), 4, "admin info did not report all four nodes: {info}");
+
+    let storage_body = cluster_admin_ok(&dist.cluster, Method::GET, "/rustfs/admin/v3/storageinfo", None).await?;
+    let storage: Value = serde_json::from_str(&storage_body)?;
+    let disks = storage["info"]["disks"]
+        .as_array()
+        .ok_or_else(|| format!("storageinfo omitted disks: {storage}"))?;
+    assert_eq!(disks.len(), 16, "storageinfo did not report all sixteen drives: {storage}");
+    assert!(
+        disks.iter().all(|disk| {
+            disk["state"].as_str().is_some_and(|state| state.eq_ignore_ascii_case("ok"))
+                && disk["runtimeState"]
+                    .as_str()
+                    .is_some_and(|state| state.eq_ignore_ascii_case("online"))
+        }),
+        "storageinfo reported a drive that was not healthy and online: {storage}"
+    );
+
+    for (node_idx, node) in dist.cluster.nodes.iter().enumerate() {
+        let (status, metrics_body) = admin_request(
+            &node.url,
+            Method::GET,
+            "/rustfs/admin/v3/metrics?n=1&by-host=true&by-disk=true",
+            None,
+            &dist.cluster.access_key,
+            &dist.cluster.secret_key,
+        )
+        .await?;
+        assert!(status.is_success(), "node {node_idx} metrics failed: {status} {metrics_body}");
+        let sample: RealtimeMetrics = serde_json::from_str(
+            metrics_body
+                .lines()
+                .next()
+                .ok_or_else(|| format!("node {node_idx} returned empty metrics"))?,
+        )?;
+        assert!(sample.finally, "node {node_idx} metrics sample was not terminal");
+        assert!(sample.errors.is_empty(), "node {node_idx} metrics reported errors: {:?}", sample.errors);
+        assert!(!sample.hosts.is_empty(), "node {node_idx} metrics omitted hosts");
+    }
+
+    let targets_body = cluster_admin_ok(&dist.cluster, Method::GET, "/rustfs/admin/v3/audit/target/list", None).await?;
+    let targets: Value = serde_json::from_str(&targets_body)?;
+    let configured = targets["audit_endpoints"]
+        .as_array()
+        .ok_or_else(|| format!("audit target list omitted audit_endpoints: {targets}"))?
+        .iter()
+        .any(|target| target["account_id"].as_str() == Some("distributed") && target["service"].as_str() == Some("webhook"));
+    assert!(configured, "configured audit webhook was missing: {targets}");
+
+    let bucket = unique_bucket("audit");
+    dist.create_bucket(&bucket).await?;
+    let key = "correlated/audit-object.bin";
+    let put = dist
+        .client(2)?
+        .put_object()
+        .bucket(&bucket)
+        .key(key)
+        .body(ByteStream::from_static(b"distributed audit payload"))
+        .send()
+        .await?;
+    let request_id = put.request_id().ok_or("PutObject response omitted request ID")?;
+    let audit = wait_for_audit_entry(&mut audit_entries, &bucket, key, request_id).await?;
+    assert_eq!(
+        audit["api"]["status_code"].as_i64(),
+        Some(200),
+        "audit entry did not report success: {audit}"
+    );
+    assert!(
+        !audit.to_string().contains(&dist.cluster.secret_key),
+        "audit entry leaked the root secret key"
+    );
+
+    collector.abort();
+    Ok(())
+}

crates/e2e_test/src/distributed/replication_quota_test.rs

@@ -0,0 +1,191 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{
+    DistCluster, DistLayout, TestResult, enable_versioning, put_bucket_replication, put_object, retrying_put, set_bucket_quota,
+    set_remote_target, unique_bucket, wait_for_ready, wait_for_replicated_bytes, wait_until,
+};
+use crate::common::{FAST_DATA_USAGE_SCANNER_ENV, init_logging};
+use aws_sdk_s3::error::ProvideErrorMetadata;
+use aws_sdk_s3::primitives::ByteStream;
+use http::Method;
+use std::time::Duration;
+
+async fn wait_for_replication_status(
+    client: &aws_sdk_s3::Client,
+    bucket: &str,
+    key: &str,
+    expected: &[&str],
+    timeout: Duration,
+) -> TestResult {
+    wait_until(
+        timeout,
+        || async {
+            let head = client.head_object().bucket(bucket).key(key).send().await?;
+            Ok(head
+                .replication_status()
+                .is_some_and(|status| expected.contains(&status.as_str())))
+        },
+        &format!("replication status for {bucket}/{key} in {expected:?}"),
+    )
+    .await
+}
+
+#[tokio::test]
+async fn four_node_bucket_replication_converges_to_peer_cluster() -> TestResult {
+    init_logging();
+    let (source, mut target) = DistCluster::start_replication_pair().await?;
+    let source_bucket = unique_bucket("replsrc");
+    let target_bucket = unique_bucket("repldst");
+    source.create_bucket(&source_bucket).await?;
+    target.create_bucket(&target_bucket).await?;
+
+    let source_client = source.client(0)?;
+    let target_client = target.client(0)?;
+    enable_versioning(&source_client, &source_bucket).await?;
+    enable_versioning(&target_client, &target_bucket).await?;
+
+    let arn = set_remote_target(&source.cluster, &source_bucket, &target.cluster, &target_bucket).await?;
+    put_bucket_replication(&source.cluster, &source_bucket, &arn).await?;
+
+    let key = "replicated/metadata-and-tags.bin";
+    let body = b"distributed-bucket-replication".to_vec();
+    source_client
+        .put_object()
+        .bucket(&source_bucket)
+        .key(key)
+        .metadata("origin", "four-node-source")
+        .tagging("suite=distributed&shape=metadata")
+        .body(ByteStream::from(body.clone()))
+        .send()
+        .await?;
+    wait_for_replicated_bytes(&target_client, &target_bucket, key, &body, Duration::from_secs(45)).await?;
+    wait_for_replication_status(&source_client, &source_bucket, key, &["COMPLETED"], Duration::from_secs(30)).await?;
+
+    let peer_read = target.client(3)?;
+    wait_for_replicated_bytes(&peer_read, &target_bucket, key, &body, Duration::from_secs(15)).await?;
+    let replica_head = peer_read.head_object().bucket(&target_bucket).key(key).send().await?;
+    assert_eq!(
+        replica_head
+            .metadata()
+            .and_then(|metadata| metadata.get("origin"))
+            .map(String::as_str),
+        Some("four-node-source")
+    );
+    assert_eq!(replica_head.replication_status().map(|status| status.as_str()), Some("REPLICA"));
+    let replica_tags = peer_read.get_object_tagging().bucket(&target_bucket).key(key).send().await?;
+    let tags: std::collections::BTreeMap<_, _> = replica_tags.tag_set().iter().map(|tag| (tag.key(), tag.value())).collect();
+    assert_eq!(tags.get("suite"), Some(&"distributed"));
+    assert_eq!(tags.get("shape"), Some(&"metadata"));
+
+    target.cluster.stop();
+    let outage_key = "replicated/queued-during-target-outage.bin";
+    let outage_body = b"retry-after-target-restart".to_vec();
+    put_object(&source_client, &source_bucket, outage_key, outage_body.clone()).await?;
+    wait_for_replication_status(
+        &source_client,
+        &source_bucket,
+        outage_key,
+        &["PENDING", "FAILED"],
+        Duration::from_secs(30),
+    )
+    .await?;
+
+    target.cluster.start().await?;
+    wait_for_ready(&target.cluster).await?;
+    wait_for_replicated_bytes(&target.client(2)?, &target_bucket, outage_key, &outage_body, Duration::from_secs(90)).await?;
+    wait_for_replication_status(&source_client, &source_bucket, outage_key, &["COMPLETED"], Duration::from_secs(45)).await?;
+    Ok(())
+}
+
+#[tokio::test]
+async fn four_node_four_drive_hard_quota_rejects_over_limit_put() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start_with_env(DistLayout::FourByFour, FAST_DATA_USAGE_SCANNER_ENV).await?;
+    let bucket = unique_bucket("quota");
+    dist.create_bucket(&bucket).await?;
+    set_bucket_quota(&dist.cluster, &bucket, 8 * 1024).await?;
+
+    let client = dist.client(1)?;
+    retrying_put(&client, &bucket, "small.bin", vec![0u8; 1024], Duration::from_secs(30)).await?;
+    wait_until(
+        Duration::from_secs(30),
+        || async {
+            let (status, body) = super::harness::cluster_admin(
+                &dist.cluster,
+                Method::GET,
+                &format!("/rustfs/admin/v3/quota-stats/{bucket}"),
+                None,
+            )
+            .await?;
+            if !status.is_success() {
+                return Ok(false);
+            }
+            let stats: serde_json::Value =
+                serde_json::from_str(&body).map_err(|error| format!("quota stats returned invalid JSON: {error}: {body}"))?;
+            let usage = stats
+                .get("current_usage")
+                .and_then(serde_json::Value::as_u64)
+                .ok_or_else(|| format!("quota stats omitted current_usage: {stats}"))?;
+            Ok(usage >= 1024)
+        },
+        "quota stats observe small object",
+    )
+    .await?;
+
+    let oversized_key = "too-big.bin";
+    let error = client
+        .put_object()
+        .bucket(&bucket)
+        .key(oversized_key)
+        .body(vec![0u8; 16 * 1024].into())
+        .send()
+        .await
+        .expect_err("hard quota must reject the oversized PUT");
+    let service_error = error
+        .as_service_error()
+        .ok_or("quota rejection was not an S3 service error")?;
+    assert_eq!(
+        error.raw_response().map(|response| response.status().as_u16()),
+        Some(400),
+        "quota rejection must be HTTP 400: {error:?}"
+    );
+    assert_eq!(service_error.code(), Some("InvalidRequest"), "unexpected quota error: {error:?}");
+    assert!(
+        service_error
+            .message()
+            .is_some_and(|message| message.starts_with("Bucket quota exceeded")),
+        "PUT must fail specifically at quota admission: {error:?}"
+    );
+
+    let missing = client
+        .head_object()
+        .bucket(&bucket)
+        .key(oversized_key)
+        .send()
+        .await
+        .expect_err("an object rejected by quota must not become visible");
+    assert_eq!(
+        missing.raw_response().map(|response| response.status().as_u16()),
+        Some(404),
+        "quota-rejected object returned an unexpected HEAD result: {missing:?}"
+    );
+
+    let listed = client.list_objects_v2().bucket(&bucket).send().await?;
+    assert!(
+        listed.contents().iter().all(|object| object.key() != Some(oversized_key)),
+        "quota-rejected key leaked into ListObjectsV2"
+    );
+    Ok(())
+}

crates/e2e_test/src/distributed/s3_basic_test.rs

@@ -0,0 +1,258 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{DistCluster, DistLayout, TestResult, assert_object_bytes, get_object_bytes, put_object, unique_bucket};
+use crate::common::{init_logging, local_http_client};
+use aws_sdk_s3::error::ProvideErrorMetadata;
+use aws_sdk_s3::presigning::PresigningConfig;
+use aws_sdk_s3::primitives::ByteStream;
+use aws_sdk_s3::types::{Delete, MetadataDirective, ObjectIdentifier};
+use std::time::Duration;
+
+#[tokio::test]
+async fn four_node_four_drive_s3_put_get_head_list_copy_rename_delete_and_presign() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("s3basic");
+    dist.create_bucket(&bucket).await?;
+
+    let writer = dist.client(0)?;
+    let reader = dist.client(3)?;
+    let key = "dir/object.bin";
+    let body = vec![0xA5u8; 256 * 1024];
+    put_object(&writer, &bucket, key, body.clone()).await?;
+
+    let head = reader.head_object().bucket(&bucket).key(key).send().await?;
+    assert_eq!(head.content_length(), Some(body.len() as i64));
+    assert_object_bytes(&reader, &bucket, key, &body).await?;
+
+    let ranged = reader
+        .get_object()
+        .bucket(&bucket)
+        .key(key)
+        .range("bytes=0-15")
+        .send()
+        .await?;
+    let ranged_body = ranged.body.collect().await?.into_bytes();
+    assert_eq!(ranged_body.as_ref(), &body[..16]);
+
+    let listed = reader.list_objects_v2().bucket(&bucket).prefix("dir/").send().await?;
+    let keys: Vec<_> = listed.contents().iter().filter_map(|object| object.key()).collect();
+    assert_eq!(keys, vec![key]);
+
+    let copy_key = "dir/object-copy.bin";
+    reader
+        .copy_object()
+        .bucket(&bucket)
+        .key(copy_key)
+        .copy_source(format!("{bucket}/{key}"))
+        .metadata_directive(MetadataDirective::Copy)
+        .send()
+        .await?;
+    assert_object_bytes(&writer, &bucket, copy_key, &body).await?;
+
+    let moved_key = "dir/object-moved.bin";
+    writer
+        .copy_object()
+        .bucket(&bucket)
+        .key(moved_key)
+        .copy_source(format!("{bucket}/{copy_key}"))
+        .send()
+        .await?;
+    writer.delete_object().bucket(&bucket).key(copy_key).send().await?;
+    match writer.head_object().bucket(&bucket).key(copy_key).send().await {
+        Ok(_) => return Err("copied source still present after rename delete".into()),
+        Err(error) if error.as_service_error().is_some_and(|err| err.is_not_found()) => {}
+        Err(error) => return Err(error.into()),
+    }
+    assert_object_bytes(&reader, &bucket, moved_key, &body).await?;
+
+    let presigned = writer
+        .get_object()
+        .bucket(&bucket)
+        .key(key)
+        .presigned(PresigningConfig::expires_in(Duration::from_secs(120))?)
+        .await?;
+    let response = local_http_client().get(presigned.uri().to_string()).send().await?;
+    assert!(response.status().is_success(), "presigned GET failed: {}", response.status());
+    let presigned_body = response.bytes().await?;
+    assert_eq!(presigned_body.as_ref(), body.as_slice());
+
+    let empty_key = "empty";
+    put_object(&writer, &bucket, empty_key, Vec::new()).await?;
+    let empty = get_object_bytes(&reader, &bucket, empty_key).await?;
+    assert!(empty.is_empty());
+
+    let deleted = writer
+        .delete_objects()
+        .bucket(&bucket)
+        .delete(
+            Delete::builder()
+                .objects(ObjectIdentifier::builder().key(key).build()?)
+                .objects(ObjectIdentifier::builder().key(moved_key).build()?)
+                .objects(ObjectIdentifier::builder().key(empty_key).build()?)
+                .build()?,
+        )
+        .send()
+        .await?;
+    assert!(deleted.errors().is_empty(), "DeleteObjects reported failures: {deleted:?}");
+    assert_eq!(deleted.deleted().len(), 3, "DeleteObjects did not acknowledge every key");
+
+    let remaining = reader.list_objects_v2().bucket(&bucket).send().await?;
+    assert!(remaining.contents().is_empty(), "bucket still has objects after delete");
+    Ok(())
+}
+
+#[tokio::test]
+async fn four_node_s3_metadata_tags_special_keys_pagination_and_multipart_abort() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("s3matrix");
+    dist.create_bucket(&bucket).await?;
+    let writer = dist.client(0)?;
+    let reader = dist.client(3)?;
+
+    let special_key = "unicode/测试 space+percent%25.txt";
+    let special_body = b"metadata and tagging survive distributed routing".to_vec();
+    let put = writer
+        .put_object()
+        .bucket(&bucket)
+        .key(special_key)
+        .metadata("test-meta", "distributed")
+        .tagging("purpose=compatibility&scope=four-by-four")
+        .body(ByteStream::from(special_body.clone()))
+        .send()
+        .await?;
+    let etag = put.e_tag().ok_or("PutObject omitted ETag")?.to_string();
+
+    let head = reader.head_object().bucket(&bucket).key(special_key).send().await?;
+    assert_eq!(
+        head.metadata()
+            .and_then(|metadata| metadata.get("test-meta"))
+            .map(String::as_str),
+        Some("distributed")
+    );
+    assert_eq!(head.e_tag(), Some(etag.as_str()));
+    let tags = reader.get_object_tagging().bucket(&bucket).key(special_key).send().await?;
+    let actual_tags: std::collections::BTreeMap<_, _> = tags
+        .tag_set()
+        .iter()
+        .map(|tag| (tag.key().to_string(), tag.value().to_string()))
+        .collect();
+    assert_eq!(actual_tags.get("purpose").map(String::as_str), Some("compatibility"));
+    assert_eq!(actual_tags.get("scope").map(String::as_str), Some("four-by-four"));
+
+    let conditional = reader
+        .get_object()
+        .bucket(&bucket)
+        .key(special_key)
+        .if_match(&etag)
+        .send()
+        .await?;
+    assert_eq!(conditional.body.collect().await?.into_bytes().as_ref(), special_body.as_slice());
+    let invalid_range = reader
+        .get_object()
+        .bucket(&bucket)
+        .key(special_key)
+        .range("bytes=999999-1000000")
+        .send()
+        .await
+        .expect_err("an unsatisfiable range must fail");
+    assert_eq!(
+        invalid_range.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("InvalidRange"),
+        "unexpected invalid-range error: {invalid_range:?}"
+    );
+
+    let upload_key = "multipart/aborted.bin";
+    let upload = writer
+        .create_multipart_upload()
+        .bucket(&bucket)
+        .key(upload_key)
+        .send()
+        .await?;
+    let upload_id = upload.upload_id().ok_or("CreateMultipartUpload omitted upload ID")?;
+    writer
+        .upload_part()
+        .bucket(&bucket)
+        .key(upload_key)
+        .upload_id(upload_id)
+        .part_number(1)
+        .body(ByteStream::from(vec![0x5Au8; 5 * 1024 * 1024]))
+        .send()
+        .await?;
+    let pending = reader
+        .list_multipart_uploads()
+        .bucket(&bucket)
+        .prefix("multipart/")
+        .send()
+        .await?;
+    assert!(pending.uploads().iter().any(|entry| entry.upload_id() == Some(upload_id)));
+    writer
+        .abort_multipart_upload()
+        .bucket(&bucket)
+        .key(upload_key)
+        .upload_id(upload_id)
+        .send()
+        .await?;
+    let after_abort = reader
+        .list_multipart_uploads()
+        .bucket(&bucket)
+        .prefix("multipart/")
+        .send()
+        .await?;
+    assert!(after_abort.uploads().iter().all(|entry| entry.upload_id() != Some(upload_id)));
+    let aborted_head = reader
+        .head_object()
+        .bucket(&bucket)
+        .key(upload_key)
+        .send()
+        .await
+        .expect_err("aborted multipart upload must not create an object");
+    assert_eq!(
+        aborted_head.raw_response().map(|response| response.status().as_u16()),
+        Some(404),
+        "aborted multipart object returned an unexpected HEAD result: {aborted_head:?}"
+    );
+
+    for index in 0..113 {
+        let key = format!("page/{index:04}.txt");
+        put_object(&writer, &bucket, &key, format!("page-{index}").into_bytes()).await?;
+    }
+    let mut token = None;
+    let mut paged_keys = Vec::new();
+    loop {
+        let page = reader
+            .list_objects_v2()
+            .bucket(&bucket)
+            .prefix("page/")
+            .max_keys(37)
+            .set_continuation_token(token.take())
+            .send()
+            .await?;
+        paged_keys.extend(page.contents().iter().filter_map(|object| object.key().map(str::to_string)));
+        if page.is_truncated() != Some(true) {
+            break;
+        }
+        token = Some(
+            page.next_continuation_token()
+                .ok_or("truncated ListObjectsV2 page omitted next continuation token")?
+                .to_string(),
+        );
+    }
+    assert_eq!(paged_keys.len(), 113);
+    let expected: Vec<_> = (0..113).map(|index| format!("page/{index:04}.txt")).collect();
+    assert_eq!(paged_keys, expected, "pagination lost, duplicated, or reordered keys");
+    Ok(())
+}

crates/e2e_test/src/distributed/s3_during_data_movement_test.rs

@@ -0,0 +1,94 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{
+    DECOMMISSION_POOL_ID, DistCluster, DistLayout, TestResult, assert_inventory, decommission_running_with_progress,
+    decommission_status_json, put_inventory_retrying, rebalance_running_with_progress, rebalance_status_json,
+    retrying_get_equals, retrying_put, start_decommission, start_rebalance, unique_bucket, wait_for_decommission_complete,
+    wait_for_decommission_running_with_progress, wait_for_rebalance_complete, wait_for_rebalance_running_with_progress,
+};
+use crate::common::init_logging;
+use std::time::Duration;
+
+#[tokio::test]
+async fn s3_put_get_list_succeed_during_decommission_and_rebalance() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start(DistLayout::SingleNodeFourDrive).await?;
+    let bucket = unique_bucket("s3move");
+    dist.create_bucket(&bucket).await?;
+    let client = dist.client(0)?;
+    let inventory = put_inventory_retrying(&client, &bucket, 96, 256 * 1024, Duration::from_secs(30)).await?;
+    dist.expand_to_four_pools().await?;
+
+    start_decommission(&dist.cluster, DECOMMISSION_POOL_ID).await?;
+    wait_for_decommission_running_with_progress(&dist.cluster, DECOMMISSION_POOL_ID, Duration::from_secs(30)).await?;
+    let live = dist.client(2)?;
+    retrying_put(
+        &live,
+        &bucket,
+        "during-decommission.bin",
+        b"written-while-decommissioning".to_vec(),
+        Duration::from_secs(30),
+    )
+    .await?;
+    retrying_get_equals(
+        &live,
+        &bucket,
+        "during-decommission.bin",
+        b"written-while-decommissioning",
+        Duration::from_secs(30),
+    )
+    .await?;
+    let listed = live.list_objects_v2().bucket(&bucket).send().await?;
+    assert!(
+        listed
+            .contents()
+            .iter()
+            .any(|object| object.key() == Some("during-decommission.bin")),
+        "list during decommission missed the newly written key"
+    );
+    let status = decommission_status_json(&dist.cluster).await?;
+    if !decommission_running_with_progress(&status, DECOMMISSION_POOL_ID)? {
+        return Err(format!("decommission did not remain active across the S3 operations: {status}").into());
+    }
+
+    wait_for_decommission_complete(&dist.cluster, DECOMMISSION_POOL_ID, Duration::from_secs(180)).await?;
+    assert_inventory(&live, &bucket, &inventory).await?;
+
+    let rebalance_id = start_rebalance(&dist.cluster).await?;
+    wait_for_rebalance_running_with_progress(&dist.cluster, &rebalance_id, Duration::from_secs(30)).await?;
+    retrying_put(
+        &live,
+        &bucket,
+        "during-rebalance.bin",
+        b"written-while-rebalancing".to_vec(),
+        Duration::from_secs(30),
+    )
+    .await?;
+    retrying_get_equals(
+        &live,
+        &bucket,
+        "during-rebalance.bin",
+        b"written-while-rebalancing",
+        Duration::from_secs(30),
+    )
+    .await?;
+    let status = rebalance_status_json(&dist.cluster).await?;
+    if !rebalance_running_with_progress(&status, &rebalance_id)? {
+        return Err(format!("rebalance did not remain active across the S3 operations: {status}").into());
+    }
+    wait_for_rebalance_complete(&dist.cluster, &rebalance_id, Duration::from_secs(180)).await?;
+    assert_inventory(&dist.client(1)?, &bucket, &inventory).await?;
+    Ok(())
+}

crates/e2e_test/src/distributed/site_replication_test.rs

@@ -0,0 +1,128 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{
+    DistCluster, TestResult, cluster_admin_ok, enable_versioning, put_object, unique_bucket, wait_for_replicated_bytes,
+    wait_until,
+};
+use crate::common::{init_logging, signed_request};
+use http::{Method, StatusCode};
+use rustfs_madmin::{PeerSite, ReplicateAddStatus, SiteReplicationInfo, SyncStatus};
+use std::time::Duration;
+
+async fn site_replication_add(
+    cluster: &crate::common::RustFSTestClusterEnvironment,
+    sites: &[PeerSite],
+) -> TestResult<ReplicateAddStatus> {
+    let url = format!("{}/rustfs/admin/v3/site-replication/add?replicateILMExpiry=false", cluster.nodes[0].url);
+    let response = signed_request(
+        Method::PUT,
+        &url,
+        &cluster.access_key,
+        &cluster.secret_key,
+        Some(serde_json::to_vec(sites)?),
+        Some("application/json"),
+    )
+    .await?;
+    if response.status() != StatusCode::OK {
+        let status = response.status();
+        let body = response.text().await.unwrap_or_default();
+        return Err(format!("site replication add failed: {status} {body}").into());
+    }
+    Ok(serde_json::from_slice(&response.bytes().await?)?)
+}
+
+async fn site_replication_info(cluster: &crate::common::RustFSTestClusterEnvironment) -> TestResult<SiteReplicationInfo> {
+    let body = cluster_admin_ok(cluster, Method::GET, "/rustfs/admin/v3/site-replication/info", None).await?;
+    Ok(serde_json::from_str(&body)?)
+}
+
+async fn wait_for_site_replication_enabled(cluster: &crate::common::RustFSTestClusterEnvironment) -> TestResult {
+    wait_until(
+        Duration::from_secs(30),
+        || async {
+            let info = site_replication_info(cluster).await?;
+            Ok(info.enabled && info.sites.len() == 2 && info.sites.iter().all(|site| site.sync_state == SyncStatus::Enable))
+        },
+        "site replication enabled with two synchronized sites",
+    )
+    .await
+}
+
+#[tokio::test]
+async fn four_node_site_replication_replicates_object_to_peer_site() -> TestResult {
+    init_logging();
+    let (site_a, site_b) = DistCluster::start_replication_pair().await?;
+    let bucket = unique_bucket("siterepl");
+    site_a.create_bucket(&bucket).await?;
+    site_b.create_bucket(&bucket).await?;
+
+    let client_a = site_a.client(0)?;
+    let client_b = site_b.client(0)?;
+    enable_versioning(&client_a, &bucket).await?;
+    enable_versioning(&client_b, &bucket).await?;
+
+    let sites = vec![
+        PeerSite {
+            name: "site-a".to_string(),
+            endpoint: site_a.cluster.nodes[0].url.clone(),
+            access_key: site_a.cluster.access_key.clone(),
+            secret_key: site_a.cluster.secret_key.clone(),
+            ..Default::default()
+        },
+        PeerSite {
+            name: "site-b".to_string(),
+            endpoint: site_b.cluster.nodes[0].url.clone(),
+            access_key: site_b.cluster.access_key.clone(),
+            secret_key: site_b.cluster.secret_key.clone(),
+            ..Default::default()
+        },
+    ];
+    let add_status = site_replication_add(&site_a.cluster, &sites).await?;
+    assert!(
+        add_status.success && add_status.err_detail.is_empty() && add_status.initial_sync_error_message.is_empty(),
+        "site replication add reported failure: {add_status:?}"
+    );
+    wait_for_site_replication_enabled(&site_a.cluster).await?;
+    wait_for_site_replication_enabled(&site_b.cluster).await?;
+
+    let info_a = site_replication_info(&site_a.cluster).await?;
+    let remote = info_a
+        .sites
+        .iter()
+        .find(|site| site.name == "site-b")
+        .ok_or_else(|| format!("site A info omitted the configured site-b peer: {info_a:?}"))?;
+    assert_eq!(remote.endpoint, site_b.cluster.nodes[0].url);
+    let deployment_ids: std::collections::BTreeSet<_> = info_a.sites.iter().map(|site| site.deployment_id.as_str()).collect();
+    assert!(
+        deployment_ids.iter().all(|deployment_id| !deployment_id.is_empty()) && deployment_ids.len() == 2,
+        "site peers must have two distinct non-empty deployment IDs: {info_a:?}"
+    );
+    assert!(info_a.retry_stats.is_none(), "site A has pending replication retries: {info_a:?}");
+    assert!(info_a.pending_operation.is_none(), "site A has a pending operation: {info_a:?}");
+
+    let key = "site-object.bin";
+    let body = b"four-node-site-replication".to_vec();
+    put_object(&client_a, &bucket, key, body.clone()).await?;
+    wait_for_replicated_bytes(&client_b, &bucket, key, &body, Duration::from_secs(60)).await?;
+
+    let peer_b = site_b.client(3)?;
+    wait_for_replicated_bytes(&peer_b, &bucket, key, &body, Duration::from_secs(20)).await?;
+
+    let reverse_key = "reverse/site-object.bin";
+    let reverse_body = b"site-b-to-site-a".to_vec();
+    put_object(&site_b.client(2)?, &bucket, reverse_key, reverse_body.clone()).await?;
+    wait_for_replicated_bytes(&site_a.client(3)?, &bucket, reverse_key, &reverse_body, Duration::from_secs(60)).await?;
+    Ok(())
+}

crates/e2e_test/src/distributed/upgrade_test.rs

@@ -0,0 +1,345 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! 4-node upgrade coverage for historical objects and IAM AK/SK.
+//!
+//! Complements `upgrade_compatibility_test` (single-node SSE/multipart and
+//! mixed-version listing). This module pins the distributed contract the
+//! hardware upgrade chain is meant to catch: after a 4-node upgrade, objects
+//! written on the previous release still read back, and IAM user credentials
+//! created before the upgrade still authenticate.
+//!
+//! Requires `RUSTFS_UPGRADE_SOURCE_BINARY` pointing at the pinned previous
+//! release. The `e2e-distributed` workflow downloads that binary; a local run
+//! without it fails closed rather than skipping.
+
+use super::harness::{
+    DistCluster, DistLayout, TestResult, assert_object_bytes, cluster_admin_ok, enable_versioning, get_object_bytes, put_object,
+    unique_bucket, wait_until,
+};
+use crate::common::{
+    AdminTransport, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user_via, init_logging,
+};
+use aws_sdk_s3::Client;
+use aws_sdk_s3::error::ProvideErrorMetadata;
+use std::path::{Path, PathBuf};
+use std::time::Duration;
+use uuid::Uuid;
+
+const SOURCE_BINARY_ENV: &str = "RUSTFS_UPGRADE_SOURCE_BINARY";
+const IAM_SECRET: &str = "UpgradeTestSecretKey1";
+const WRONG_SECRET: &str = "WrongSecretKey000000";
+const CREDENTIAL_TIMEOUT: Duration = Duration::from_secs(30);
+
+struct UpgradeSeed {
+    history_bucket: String,
+    history_key: &'static str,
+    history_body: Vec<u8>,
+    versioned_bucket: String,
+    versioned_key: &'static str,
+    version1: String,
+    version1_body: Vec<u8>,
+    version2: String,
+    version2_body: Vec<u8>,
+    iam_bucket: String,
+    iam_key: &'static str,
+    iam_body: Vec<u8>,
+    iam_user: String,
+    iam_secret: &'static str,
+}
+
+fn source_binary() -> TestResult<PathBuf> {
+    let path = std::env::var_os(SOURCE_BINARY_ENV).map(PathBuf::from).ok_or_else(|| {
+        format!(
+            "{SOURCE_BINARY_ENV} must point to the pinned previous release binary (the e2e-distributed workflow downloads it)"
+        )
+    })?;
+    if !path.is_file() {
+        return Err(format!("upgrade source binary does not exist: {}", path.display()).into());
+    }
+    Ok(path)
+}
+
+fn capture_upgrade_logs(cluster: &mut DistCluster, label: &str) -> TestResult {
+    let Some(log_dir) = std::env::var_os("RUSTFS_E2E_LOG_DIR") else {
+        return Ok(());
+    };
+    std::fs::create_dir_all(&log_dir)?;
+    for node_idx in 0..cluster.cluster.nodes.len() {
+        let path = Path::new(&log_dir).join(format!("{label}-node-{node_idx}.log"));
+        cluster
+            .cluster
+            .set_node_capture_log_path(node_idx, path.to_string_lossy().into_owned())?;
+    }
+    Ok(())
+}
+
+fn iam_rw_policy(bucket: &str) -> String {
+    serde_json::json!({
+        "Version": "2012-10-17",
+        "Statement": [{
+            "Effect": "Allow",
+            "Action": ["s3:*"],
+            "Resource": [
+                format!("arn:aws:s3:::{bucket}"),
+                format!("arn:aws:s3:::{bucket}/*")
+            ]
+        }]
+    })
+    .to_string()
+}
+
+async fn create_iam_user(dist: &DistCluster, user: &str, secret: &str, policy_name: &str, bucket: &str) -> TestResult {
+    let url = &dist.cluster.nodes[0].url;
+    let access = &dist.cluster.access_key;
+    let admin_secret = &dist.cluster.secret_key;
+    admin_create_user_via(AdminTransport::Signed, url, access, admin_secret, user, secret).await?;
+    admin_add_canned_policy_via(AdminTransport::Signed, url, access, admin_secret, policy_name, &iam_rw_policy(bucket)).await?;
+    admin_attach_user_policy_via(AdminTransport::Signed, url, access, admin_secret, policy_name, user).await?;
+    Ok(())
+}
+
+async fn wait_for_put(client: &Client, bucket: &str, key: &str, body: Vec<u8>, label: &str) -> TestResult {
+    wait_until(
+        CREDENTIAL_TIMEOUT,
+        || {
+            let client = client.clone();
+            let bucket = bucket.to_string();
+            let key = key.to_string();
+            let body = body.clone();
+            async move {
+                put_object(&client, &bucket, &key, body).await?;
+                Ok(true)
+            }
+        },
+        label,
+    )
+    .await
+}
+
+async fn wait_for_bytes(client: &Client, bucket: &str, key: &str, expected: &[u8], label: &str) -> TestResult {
+    wait_until(
+        CREDENTIAL_TIMEOUT,
+        || {
+            let client = client.clone();
+            let bucket = bucket.to_string();
+            let key = key.to_string();
+            let expected = expected.to_vec();
+            async move {
+                let got = get_object_bytes(&client, &bucket, &key).await?;
+                Ok(got == expected)
+            }
+        },
+        label,
+    )
+    .await
+}
+
+async fn seed_history_and_iam(dist: &DistCluster) -> TestResult<UpgradeSeed> {
+    let history_bucket = unique_bucket("upg-hist");
+    let versioned_bucket = unique_bucket("upg-ver");
+    let iam_bucket = unique_bucket("upg-iam");
+    dist.create_bucket(&history_bucket).await?;
+    dist.create_bucket(&versioned_bucket).await?;
+    dist.create_bucket(&iam_bucket).await?;
+
+    let root = dist.client(0)?;
+    enable_versioning(&root, &versioned_bucket).await?;
+
+    let history_key = "plain-history.bin";
+    let history_body = b"written by the previous 4-node release".to_vec();
+    put_object(&root, &history_bucket, history_key, history_body.clone()).await?;
+
+    let versioned_key = "versioned-history.txt";
+    let version1_body = b"version-one-before-upgrade".to_vec();
+    let version1 = root
+        .put_object()
+        .bucket(&versioned_bucket)
+        .key(versioned_key)
+        .body(aws_sdk_s3::primitives::ByteStream::from(version1_body.clone()))
+        .send()
+        .await?
+        .version_id()
+        .ok_or("first versioned PUT omitted version ID")?
+        .to_string();
+    let version2_body = b"version-two-before-upgrade".to_vec();
+    let version2 = root
+        .put_object()
+        .bucket(&versioned_bucket)
+        .key(versioned_key)
+        .body(aws_sdk_s3::primitives::ByteStream::from(version2_body.clone()))
+        .send()
+        .await?
+        .version_id()
+        .ok_or("second versioned PUT omitted version ID")?
+        .to_string();
+
+    let iam_user = format!("upg{}", &Uuid::new_v4().simple().to_string()[..8]);
+    let policy_name = format!("upgpol{}", &Uuid::new_v4().simple().to_string()[..8]);
+    create_iam_user(dist, &iam_user, IAM_SECRET, &policy_name, &iam_bucket).await?;
+
+    let iam_key = "iam-history.bin";
+    let iam_body = b"written with pre-upgrade IAM AK/SK".to_vec();
+    let iam_client = dist.client_with_credentials(1, &iam_user, IAM_SECRET)?;
+    wait_for_put(&iam_client, &iam_bucket, iam_key, iam_body.clone(), "IAM user PUT before upgrade").await?;
+
+    Ok(UpgradeSeed {
+        history_bucket,
+        history_key,
+        history_body,
+        versioned_bucket,
+        versioned_key,
+        version1,
+        version1_body,
+        version2,
+        version2_body,
+        iam_bucket,
+        iam_key,
+        iam_body,
+        iam_user,
+        iam_secret: IAM_SECRET,
+    })
+}
+
+async fn assert_history_and_iam(dist: &DistCluster, seed: &UpgradeSeed, context: &str) -> TestResult {
+    let root_a = dist.client(0)?;
+    let root_b = dist.client(3)?;
+    wait_for_bytes(
+        &root_b,
+        &seed.history_bucket,
+        seed.history_key,
+        &seed.history_body,
+        &format!("{context}: root GET historical object"),
+    )
+    .await?;
+    assert_object_bytes(&root_a, &seed.history_bucket, seed.history_key, &seed.history_body).await?;
+
+    let v1 = root_b
+        .get_object()
+        .bucket(&seed.versioned_bucket)
+        .key(seed.versioned_key)
+        .version_id(&seed.version1)
+        .send()
+        .await?;
+    let v1_body = v1.body.collect().await?.into_bytes();
+    if v1_body.as_ref() != seed.version1_body.as_slice() {
+        return Err(format!("{context}: version 1 bytes changed after upgrade").into());
+    }
+    let v2 = root_a
+        .get_object()
+        .bucket(&seed.versioned_bucket)
+        .key(seed.versioned_key)
+        .version_id(&seed.version2)
+        .send()
+        .await?;
+    let v2_body = v2.body.collect().await?.into_bytes();
+    if v2_body.as_ref() != seed.version2_body.as_slice() {
+        return Err(format!("{context}: version 2 bytes changed after upgrade").into());
+    }
+
+    let users = cluster_admin_ok(&dist.cluster, http::Method::GET, "/rustfs/admin/v3/list-users", None).await?;
+    if !users.contains(&seed.iam_user) {
+        return Err(format!("{context}: list-users lost IAM user {}: {users}", seed.iam_user).into());
+    }
+
+    let iam_on_upgraded = dist.client_with_credentials(0, &seed.iam_user, seed.iam_secret)?;
+    let iam_on_peer = dist.client_with_credentials(3, &seed.iam_user, seed.iam_secret)?;
+    wait_for_bytes(
+        &iam_on_upgraded,
+        &seed.iam_bucket,
+        seed.iam_key,
+        &seed.iam_body,
+        &format!("{context}: IAM GET historical object on node 0"),
+    )
+    .await?;
+    wait_for_bytes(
+        &iam_on_peer,
+        &seed.iam_bucket,
+        seed.iam_key,
+        &seed.iam_body,
+        &format!("{context}: IAM GET historical object on node 3"),
+    )
+    .await?;
+
+    let post_key = format!("after-upgrade-{context}.txt");
+    let post_body = format!("{context}: written with the same IAM AK/SK after upgrade").into_bytes();
+    wait_for_put(
+        &iam_on_peer,
+        &seed.iam_bucket,
+        &post_key,
+        post_body.clone(),
+        &format!("{context}: IAM PUT after upgrade"),
+    )
+    .await?;
+    assert_object_bytes(&iam_on_upgraded, &seed.iam_bucket, &post_key, &post_body).await?;
+
+    let bad = dist.client_with_credentials(1, &seed.iam_user, WRONG_SECRET)?;
+    match bad.get_object().bucket(&seed.iam_bucket).key(seed.iam_key).send().await {
+        Ok(_) => return Err(format!("{context}: wrong secret must not read the IAM object").into()),
+        Err(error) => {
+            let code = error.as_service_error().and_then(ProvideErrorMetadata::code);
+            let rejected = code == Some("SignatureDoesNotMatch")
+                || code == Some("InvalidAccessKeyId")
+                || code == Some("AccessDenied")
+                || code == Some("InvalidArgument")
+                || error.raw_response().is_some_and(|response| response.status().as_u16() == 403);
+            if !rejected {
+                return Err(format!("{context}: wrong secret failed with unexpected error {error:?}").into());
+            }
+        }
+    }
+
+    let post_root_key = format!("root-after-{context}.bin");
+    let post_root_body = format!("{context}: root write after upgrade").into_bytes();
+    put_object(&root_a, &seed.history_bucket, &post_root_key, post_root_body.clone()).await?;
+    assert_object_bytes(&root_b, &seed.history_bucket, &post_root_key, &post_root_body).await?;
+    Ok(())
+}
+
+#[tokio::test]
+async fn four_node_direct_upgrade_preserves_history_and_iam_credentials() -> TestResult {
+    init_logging();
+    let previous = source_binary()?;
+    let mut dist = DistCluster::new_stopped(DistLayout::FourNodeFourDisk).await?;
+    capture_upgrade_logs(&mut dist, "direct-upgrade")?;
+    dist.start_from_binary(&previous).await?;
+
+    let seed = seed_history_and_iam(&dist).await?;
+    dist.restart_with_current_binary().await?;
+    assert_history_and_iam(&dist, &seed, "direct").await?;
+    Ok(())
+}
+
+#[tokio::test]
+async fn four_node_rolling_upgrade_preserves_history_and_iam_credentials() -> TestResult {
+    init_logging();
+    let previous = source_binary()?;
+    let mut dist = DistCluster::new_stopped(DistLayout::FourNodeFourDisk).await?;
+    capture_upgrade_logs(&mut dist, "rolling-upgrade")?;
+    dist.start_from_binary(&previous).await?;
+
+    let seed = seed_history_and_iam(&dist).await?;
+
+    dist.replace_node_with_current_binary(0).await?;
+    assert_history_and_iam(&dist, &seed, "one-current-node").await?;
+
+    for node_idx in [1, 2] {
+        dist.replace_node_with_current_binary(node_idx).await?;
+    }
+    assert_history_and_iam(&dist, &seed, "one-previous-node").await?;
+
+    dist.replace_node_with_current_binary(3).await?;
+    assert_history_and_iam(&dist, &seed, "homogeneous-current").await?;
+    Ok(())
+}

crates/e2e_test/src/distributed/versioning_test.rs

@@ -0,0 +1,188 @@
+// Copyright 2026 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use super::harness::{DistCluster, DistLayout, TestResult, enable_versioning, get_object_bytes, put_object, unique_bucket};
+use crate::common::init_logging;
+use aws_sdk_s3::error::ProvideErrorMetadata;
+use aws_sdk_s3::types::{BucketVersioningStatus, VersioningConfiguration};
+
+#[tokio::test]
+async fn four_node_four_drive_versioning_put_list_get_delete_marker() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("version");
+    dist.create_bucket(&bucket).await?;
+    let writer = dist.client(0)?;
+    let reader = dist.client(3)?;
+    enable_versioning(&writer, &bucket).await?;
+
+    let key = "versioned.txt";
+    let v1_id = writer
+        .put_object()
+        .bucket(&bucket)
+        .key(key)
+        .body(b"v1".to_vec().into())
+        .send()
+        .await?
+        .version_id()
+        .ok_or("v1 PUT omitted version ID")?
+        .to_string();
+    let v2_id = writer
+        .put_object()
+        .bucket(&bucket)
+        .key(key)
+        .body(b"v2".to_vec().into())
+        .send()
+        .await?
+        .version_id()
+        .ok_or("v2 PUT omitted version ID")?
+        .to_string();
+
+    let versions = reader.list_object_versions().bucket(&bucket).prefix(key).send().await?;
+    let matching_versions: Vec<_> = versions
+        .versions()
+        .iter()
+        .filter(|version| version.key() == Some(key))
+        .collect();
+    assert_eq!(matching_versions.len(), 2, "fresh key must have exactly two versions: {versions:?}");
+    assert!(versions.delete_markers().is_empty(), "fresh key unexpectedly has a delete marker");
+    assert!(
+        matching_versions
+            .iter()
+            .any(|version| version.version_id() == Some(v1_id.as_str()) && version.is_latest() != Some(true)),
+        "v1 was not the historical version: {versions:?}"
+    );
+    assert!(
+        matching_versions
+            .iter()
+            .any(|version| version.version_id() == Some(v2_id.as_str()) && version.is_latest() == Some(true)),
+        "v2 was not the latest version: {versions:?}"
+    );
+
+    let latest = get_object_bytes(&reader, &bucket, key).await?;
+    assert_eq!(latest, b"v2");
+
+    let older = reader.get_object().bucket(&bucket).key(key).version_id(&v1_id).send().await?;
+    let older_body = older.body.collect().await?.into_bytes();
+    assert_eq!(older_body.as_ref(), b"v1");
+
+    let deleted = writer.delete_object().bucket(&bucket).key(key).send().await?;
+    assert_eq!(deleted.delete_marker(), Some(true));
+    let marker_id = deleted.version_id().ok_or("DeleteObject omitted delete-marker version ID")?;
+    let after_delete = reader.list_object_versions().bucket(&bucket).prefix(key).send().await?;
+    let matching_markers: Vec<_> = after_delete
+        .delete_markers()
+        .iter()
+        .filter(|marker| marker.key() == Some(key))
+        .collect();
+    assert_eq!(
+        matching_markers.len(),
+        1,
+        "delete marker missing or duplicated after current-version delete: {after_delete:?}"
+    );
+    assert!(
+        matching_markers[0].version_id() == Some(marker_id) && matching_markers[0].is_latest() == Some(true),
+        "DeleteObject response and ListObjectVersions disagree about the marker: {after_delete:?}"
+    );
+
+    let latest_after_delete = reader.get_object().bucket(&bucket).key(key).send().await;
+    match latest_after_delete {
+        Ok(_) => return Err("current version should be a delete marker".into()),
+        Err(error)
+            if error
+                .as_service_error()
+                .and_then(ProvideErrorMetadata::code)
+                .is_some_and(|code| code == "NoSuchKey" || code == "NotFound") => {}
+        Err(error) => return Err(error.into()),
+    }
+
+    let restored = reader.get_object().bucket(&bucket).key(key).version_id(&v1_id).send().await?;
+    let restored_body = restored.body.collect().await?.into_bytes();
+    assert_eq!(restored_body.as_ref(), b"v1");
+
+    writer
+        .delete_object()
+        .bucket(&bucket)
+        .key(key)
+        .version_id(marker_id)
+        .send()
+        .await?;
+    assert_eq!(get_object_bytes(&reader, &bucket, key).await?, b"v2");
+    Ok(())
+}
+
+#[tokio::test]
+async fn four_node_versioning_suspension_keeps_one_null_version_and_history() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start(DistLayout::FourByFour).await?;
+    let bucket = unique_bucket("suspend");
+    dist.create_bucket(&bucket).await?;
+    let writer = dist.client(0)?;
+    let reader = dist.client(3)?;
+    enable_versioning(&writer, &bucket).await?;
+
+    let key = "suspended.txt";
+    let original = writer
+        .put_object()
+        .bucket(&bucket)
+        .key(key)
+        .body(b"enabled-history".to_vec().into())
+        .send()
+        .await?
+        .version_id()
+        .ok_or("enabled PUT omitted version ID")?
+        .to_string();
+    writer
+        .put_bucket_versioning()
+        .bucket(&bucket)
+        .versioning_configuration(
+            VersioningConfiguration::builder()
+                .status(BucketVersioningStatus::Suspended)
+                .build(),
+        )
+        .send()
+        .await?;
+
+    put_object(&writer, &bucket, key, b"null-one".to_vec()).await?;
+    put_object(&writer, &bucket, key, b"null-two".to_vec()).await?;
+    assert_eq!(get_object_bytes(&reader, &bucket, key).await?, b"null-two");
+
+    let versions = reader.list_object_versions().bucket(&bucket).prefix(key).send().await?;
+    let matching: Vec<_> = versions
+        .versions()
+        .iter()
+        .filter(|version| version.key() == Some(key))
+        .collect();
+    assert!(matching.iter().any(|version| version.version_id() == Some(original.as_str())));
+    let null_version_count = matching
+        .iter()
+        .filter(|version| {
+            matches!(
+                version.version_id(),
+                None | Some("") | Some("null") | Some("00000000-0000-0000-0000-000000000000")
+            )
+        })
+        .count();
+    assert_eq!(null_version_count, 1, "suspended overwrites must keep one null version: {versions:?}");
+
+    let historical = reader
+        .get_object()
+        .bucket(&bucket)
+        .key(key)
+        .version_id(&original)
+        .send()
+        .await?;
+    assert_eq!(historical.body.collect().await?.into_bytes().as_ref(), b"enabled-history");
+    Ok(())
+}

crates/e2e_test/src/heal_erasure_disk_rebuild_test.rs

@@ -16,15 +16,18 @@
 
 #[cfg(test)]
 mod tests {
-    use crate::chaos::{VersionShardCensus, census_object_version_on_disk, signed_admin_post};
+    use crate::chaos::{VersionShardCensus, census_object_version_on_disk, sha256_hex, signed_admin_post};
     use crate::common::{
         FAST_DATA_USAGE_SCANNER_ENV, RustFSTestClusterEnvironment, RustFSTestEnvironment, admin_request, init_logging,
+        rustfs_binary_path,
     };
     use crate::storage_api::RUSTFS_META_BUCKET;
     use aws_sdk_s3::primitives::ByteStream;
     use http::Method;
+    use sha2::{Digest, Sha256};
     use std::collections::HashSet;
     use std::error::Error;
+    use std::io::{Read, Write};
     use std::net::SocketAddr;
     use std::path::{Path, PathBuf};
     use std::process::Command;
@@ -34,6 +37,76 @@ mod tests {
 
     const POOL_METADATA_OBJECT: &str = "pool.bin";
 
+    #[derive(serde::Deserialize)]
+    struct EvidenceBuild {
+        sha256: String,
+    }
+
+    #[derive(serde::Deserialize)]
+    struct RestartEvidenceRun {
+        schema: u32,
+        run_id: String,
+        source_revision: String,
+        test_build: serde_json::Value,
+        binary: EvidenceBuild,
+        test_binary: EvidenceBuild,
+    }
+
+    fn file_sha256(path: &Path) -> Result<String, Box<dyn Error + Send + Sync>> {
+        let mut file = std::fs::File::open(path)?;
+        let mut digest = Sha256::new();
+        let mut buffer = [0_u8; 64 * 1024];
+        loop {
+            let read = file.read(&mut buffer)?;
+            if read == 0 {
+                break;
+            }
+            digest.update(&buffer[..read]);
+        }
+        Ok(digest.finalize().iter().map(|byte| format!("{byte:02x}")).collect())
+    }
+
+    fn restart_evidence_run(binary: &Path) -> Result<Option<(PathBuf, RestartEvidenceRun)>, Box<dyn Error + Send + Sync>> {
+        let Some(directory) = std::env::var_os("RUSTFS_SCANNER_HEAL_RUN_DIR") else {
+            return Ok(None);
+        };
+        let directory = PathBuf::from(directory);
+        let receipt = directory.join("run.json");
+        if receipt.metadata()?.len() > 1024 * 1024 {
+            return Err("oversized scanner/heal execution receipt".into());
+        }
+        let run: RestartEvidenceRun = serde_json::from_slice(&std::fs::read(receipt)?)?;
+        if run.schema != 1 || run.run_id.len() != 32 || run.source_revision.len() != 40 {
+            return Err("invalid scanner/heal execution identity".into());
+        }
+        let built = compiled_test_identity();
+        for key in ["source_revision", "dirty", "lock_blob", "features"] {
+            assert_eq!(built[key], run.test_build[key], "compiled test identity differs for {key}");
+        }
+        assert_eq!(file_sha256(binary)?, run.binary.sha256, "server binary must match the run receipt");
+        assert_eq!(
+            file_sha256(&std::env::current_exe()?)?,
+            run.test_binary.sha256,
+            "test executable must match the run receipt"
+        );
+        if directory.join("background-target-restart.json").exists() {
+            return Err("scanner/heal oracle already exists; create a new execution receipt".into());
+        }
+        Ok(Some((directory, run)))
+    }
+
+    fn compiled_test_identity() -> serde_json::Value {
+        serde_json::json!({
+            "source_revision": env!("RUSTFS_E2E_BUILD_COMMIT"),
+            "dirty": env!("RUSTFS_E2E_BUILD_DIRTY") != "false",
+            "lock_blob": env!("RUSTFS_E2E_BUILD_LOCK"),
+            "features": env!("RUSTFS_E2E_BUILD_FEATURES"),
+            "target": env!("RUSTFS_E2E_BUILD_TARGET"),
+            "profile": env!("RUSTFS_E2E_BUILD_PROFILE"),
+            "rustflags_hex": env!("RUSTFS_E2E_BUILD_RUSTFLAGS_HEX"),
+        })
+    }
+
     struct TcpPortBlackhole {
         port: u16,
         comment: String,
@@ -195,8 +268,9 @@ mod tests {
         clients: &[aws_sdk_s3::Client],
         bucket: &str,
         expected_keys: &HashSet<String>,
-    ) -> Result<(), Box<dyn Error + Send + Sync>> {
+    ) -> Result<Vec<Vec<String>>, Box<dyn Error + Send + Sync>> {
         const PAGE_SIZE: i32 = 10;
+        let mut node_listings = Vec::with_capacity(clients.len());
         for (node_index, client) in clients.iter().enumerate() {
             let mut listed_keys = Vec::new();
             let mut continuation_token = None;
@@ -243,8 +317,10 @@ mod tests {
                 &listed_key_set, expected_keys,
                 "node {node_index} did not expose the complete recovered namespace"
             );
+            listed_keys.sort();
+            node_listings.push(listed_keys);
         }
-        Ok(())
+        Ok(node_listings)
     }
 
     fn heal_task_status_diagnostic(body: &str) -> String {
@@ -808,6 +884,13 @@ mod tests {
     }
 
     async fn run_cluster_root_heal_interruption(scenario: InterruptionScenario) -> Result<(), Box<dyn Error + Send + Sync>> {
+        let server_binary = rustfs_binary_path();
+        let evidence_run = if scenario == InterruptionScenario::BackgroundTargetRestart {
+            restart_evidence_run(&server_binary)?
+        } else {
+            None
+        };
+        let mut evidence_objects = Vec::new();
         let (background_enabled, interruption_node, interruption_kind) = match scenario {
             InterruptionScenario::IsolatedTargetRestart => (false, 1, "target_restart"),
             InterruptionScenario::BackgroundTargetRestart => (true, 1, "background_target_restart"),
@@ -855,7 +938,7 @@ mod tests {
         for node_index in 0..cluster.nodes.len() {
             cluster.set_node_capture_log_path(node_index, format!("{log_dir}/node{node_index}.log"))?;
         }
-        cluster.start().await?;
+        cluster.start_with_binary(&server_binary).await?;
         let clients = cluster.create_all_clients()?;
 
         let bucket = "heal-restart-during-rebuild";
@@ -996,7 +1079,7 @@ mod tests {
             }
         }
 
-        cluster.start_node(1).await?;
+        cluster.start_node_from_binary(1, &server_binary).await?;
 
         let status_url = format!("{}/rustfs/admin/v3/background-heal/status", cluster.nodes[0].url);
         let recovery_deadline = Instant::now() + Duration::from_secs(60);
@@ -1274,7 +1357,7 @@ mod tests {
                     }
                 }
             }
-            cluster.start_node(interruption_node).await?;
+            cluster.start_node_from_binary(interruption_node, &server_binary).await?;
             if interruption_node == 0 {
                 let target = cluster.nodes[1]
                     .process
@@ -1373,7 +1456,7 @@ mod tests {
             .map(|manifest| manifest.key.clone())
             .collect::<HashSet<_>>();
         assert!(expected_keys.insert(outage_key.to_string()));
-        assert_all_nodes_list_exact_keys(&clients, bucket, &expected_keys).await?;
+        let node_listings = assert_all_nodes_list_exact_keys(&clients, bucket, &expected_keys).await?;
 
         let target_client = cluster.create_s3_client(1)?;
         for expected in &expected_manifests {
@@ -1381,11 +1464,31 @@ mod tests {
             let actual = response.body.collect().await?.into_bytes();
             let expected_body = deterministic_object_body(object_size_bytes, expected.payload_seed);
             assert_eq!(actual.as_ref(), expected_body.as_slice(), "object body changed for {}", expected.key);
+            if evidence_run.is_some() {
+                evidence_objects.push(serde_json::json!({
+                    "key": expected.key, "version_id": expected.shard_census.version_id,
+                    "expected_bytes": expected_body.len(), "actual_bytes": actual.len(),
+                    "expected_sha256": sha256_hex(&expected_body),
+                    "actual_sha256": sha256_hex(&actual),
+                    "expected_physical": expected.shard_census,
+                    "physical": census_object_version_on_disk(&replaced_disk, bucket, &expected.key, None)?,
+                }));
+            }
         }
         let response = target_client.get_object().bucket(bucket).key(outage_key).send().await?;
         let actual = response.body.collect().await?.into_bytes();
         let expected_outage_body = deterministic_object_body(object_size_bytes, outage_payload_seed);
         assert_eq!(actual.as_ref(), expected_outage_body.as_slice(), "object body changed for {outage_key}");
+        if evidence_run.is_some() {
+            evidence_objects.push(serde_json::json!({
+                "key": outage_key, "version_id": null,
+                "expected_bytes": expected_outage_body.len(), "actual_bytes": actual.len(),
+                "expected_sha256": sha256_hex(&expected_outage_body),
+                "actual_sha256": sha256_hex(&actual),
+                "expected_physical": null,
+                "physical": census_object_version_on_disk(&replaced_disk, bucket, outage_key, None)?,
+            }));
+        }
 
         let terminal_deadline = Instant::now() + Duration::from_secs(30);
         loop {
@@ -1432,6 +1535,31 @@ mod tests {
             return Err(format!("heal data rebuilt but task did not finish successfully: {task_status}").into());
         }
 
+        if let Some((directory, run)) = evidence_run {
+            let restarted_pid = cluster.nodes[1].process.as_ref().ok_or("restarted target is absent")?.id();
+            assert_ne!(target_pid, restarted_pid, "target must be a new process");
+            assert_eq!(file_sha256(&server_binary)?, run.binary.sha256, "server build changed during restart");
+            let evidence = serde_json::json!({
+                "schema": 1, "case": "background-target-restart", "evidence": "process-restart",
+                "run_id": run.run_id, "source_revision": run.source_revision,
+                "test_build": compiled_test_identity(),
+                "binary_sha256": run.binary.sha256, "test_binary_sha256": run.test_binary.sha256,
+                "topology": {"nodes": cluster.nodes.len(), "drives_per_node": cluster.nodes[0].data_dirs.len()},
+                "pid_before": target_pid, "pid_after": restarted_pid,
+                "objects": evidence_objects, "node_listings": node_listings,
+            });
+            let data = serde_json::to_vec(&evidence)?;
+            if data.len() > 1024 * 1024 {
+                return Err("scanner/heal oracle exceeds the 1 MiB artifact budget".into());
+            }
+            let mut output = std::fs::OpenOptions::new()
+                .write(true)
+                .create_new(true)
+                .open(directory.join("background-target-restart.json"))?;
+            output.write_all(&data)?;
+            output.sync_all()?;
+        }
+
         Ok(())
     }
 

crates/e2e_test/src/lib.rs

@@ -378,6 +378,11 @@ mod bucket_stats_regression_test;
 #[cfg(test)]
 mod distributed_startup_regression_test;
 
+// 4-node / 4-disk distributed Actions suite (S3, lock, versioning, replication,
+// quota, observability, expand/decommission/rebalance, site replication, chaos).
+#[cfg(test)]
+mod distributed;
+
 // P1 regression: tier/ILM transition (rustfs#5218, #5130, #5011, #4826, #5024)
 #[cfg(test)]
 mod tier_transition_regression_test;

crates/e2e_test/src/on_demand_migration/concurrency_test.rs

@@ -20,9 +20,10 @@
 //! journal (`count_requests`) carries the assertion in every one of them.
 
 use super::common::{BoxError, OdmTestEnv, RawResponse, SeedObject, start_configured_env};
-use crate::fake_s3_target::Operation;
+use crate::fake_s3_target::{FaultAction, Operation};
 use aws_sdk_s3::types::{BucketVersioningStatus, VersioningConfiguration};
 use bytes::Bytes;
+use futures::{StreamExt, TryStreamExt};
 use std::time::Duration;
 
 type TestResult = Result<(), BoxError>;
@@ -145,14 +146,38 @@ async fn test_odm_range_burst_overflows_the_pull_queue_without_failing_clients()
     .await?;
 
     let body = payload(128 * 1024);
+    let blocker = "queue/blocker.bin";
+    env.seed_source(SOURCE_BUCKET, &[SeedObject::new(blocker, body.clone())]);
+    // The one-chunk range completes immediately; its full background pull
+    // occupies the only slot while the remaining requests fill the queue.
+    env.source.inject_for_key(
+        Operation::GetObject,
+        blocker,
+        FaultAction::SlowSendBody {
+            chunk_bytes: 1024,
+            delay: Duration::from_millis(100),
+        },
+        2,
+    );
+    let response = env
+        .raw_object_request(http::Method::GET, bucket, blocker, &[("range", "bytes=0-1023")])
+        .await?;
+    assert_eq!(response.status, 206);
+    assert_eq!(response.body, body.slice(0..1024));
+    env.wait_for_status_counter(bucket, "/inflight_pulls", 1, SETTLE).await?;
+
     let keys: Vec<String> = (0..REQUESTS).map(|index| format!("queue/object-{index:03}.bin")).collect();
     let seeds: Vec<SeedObject> = keys.iter().map(|key| SeedObject::new(key.clone(), body.clone())).collect();
     env.seed_source(SOURCE_BUCKET, &seeds);
 
-    let responses: Vec<RawResponse> = futures::future::try_join_all(
+    // Bound source connections below the fixture's limit while still
+    // submitting all 100 requests to the eight-slot background queue.
+    let responses: Vec<RawResponse> = futures::stream::iter(
         keys.iter()
             .map(|key| env.raw_object_request(http::Method::GET, bucket, key, &[("range", "bytes=0-1023")])),
     )
+    .buffered(16)
+    .try_collect()
     .await?;
     for (key, response) in keys.iter().zip(&responses) {
         assert_eq!(response.status, 206, "{key}: {}", String::from_utf8_lossy(&response.body));
@@ -168,6 +193,15 @@ async fn test_odm_range_burst_overflows_the_pull_queue_without_failing_clients()
         .wait_for_status_counter(bucket, "/counters/pull_failures_total/queue_full", 1, SETTLE)
         .await?;
     assert!(queue_full > 0, "a 100-deep burst must overflow an 8-slot queue");
+    let queue_full = usize::try_from(queue_full)?;
+    assert!(queue_full <= REQUESTS);
+    env.wait_for_status_counter(
+        bucket,
+        "/counters/pulled_objects_total/background",
+        u64::try_from(REQUESTS + 1 - queue_full)?,
+        SETTLE,
+    )
+    .await?;
 
     let ranged_reads: usize = keys.iter().map(|key| source_get_count(&env, key)).sum();
     assert!(
@@ -175,9 +209,6 @@ async fn test_odm_range_burst_overflows_the_pull_queue_without_failing_clients()
         "every reader is served from the source: {ranged_reads} GETs for {REQUESTS} readers"
     );
     let dropped = keys.iter().filter(|key| source_get_count(&env, key) == 1).count();
-    assert!(
+    assert_eq!(dropped, queue_full, "only overflowed keys remain without a background GET");
-        dropped > 0,
-        "the overflowed keys are the ones with no backfill GET, but every key got one"
-    );
     Ok(())
 }

crates/e2e_test/src/on_demand_migration/list_through_test.rs

@@ -265,16 +265,13 @@ async fn list_through_rejects_a_tampered_continuation_token() -> TestResult {
     let decoded = String::from_utf8(base64_simd::STANDARD.decode_to_vec(token.as_bytes())?)?;
     assert!(decoded.contains("\"t\":\"odm-list\""), "the merged token is an envelope: {decoded}");
 
-    let tampered = base64_simd::STANDARD.encode_to_string(decoded.replace("\"v\":1", "\"v\":2").as_bytes());
+    let tampered = base64_simd::STANDARD.encode_to_string(decoded.replace("\"v\":1", "\"v\":3").as_bytes());
-    let rejected = env
+    assert_ne!(tampered, token, "the test must change the token version");
-        .raw_list_objects_v2(bucket, &format!("continuation-token={tampered}"))
+    let query = serde_urlencoded::to_string([("continuation-token", tampered.as_str())])?;
-        .await?;
+    let rejected = env.raw_list_objects_v2(bucket, &query).await?;
-    assert_eq!(
+    let error_body = String::from_utf8_lossy(&rejected.body);
-        rejected.status,
+    assert_eq!(rejected.status, 400, "a bumped token version is a client error: {}", error_body);
-        400,
+    assert!(error_body.contains("<Code>InvalidArgument</Code>"), "{error_body}");
-        "a bumped token version is a client error: {}",
-        String::from_utf8_lossy(&rejected.body)
-    );
     Ok(())
 }
 

crates/e2e_test/src/replication_extension_test.rs

@@ -6743,6 +6743,99 @@ async fn test_site_replication_replicates_object_with_bucket_versioning_real_dua
     Ok(())
 }
 
+#[tokio::test]
+async fn test_site_replication_replays_bucket_created_during_peer_outage_real_dual_node() -> TestResult {
+    init_logging();
+
+    // Keep compilation outside the scenario timeout. Recovery itself waits
+    // for the production 30-second lightweight retry tick.
+    let _rustfs_binary = rustfs_binary_path();
+
+    match timeout(Duration::from_secs(150), async {
+        let mut site_env = replication_fast_env();
+        site_env.extend_from_slice(LOOPBACK_REPLICATION_TARGET_ENV);
+
+        let mut site_a_env = RustFSTestEnvironment::new().await?;
+        site_a_env.start_rustfs_server_with_env(vec![], &site_env).await?;
+
+        let mut site_b_env = RustFSTestEnvironment::new().await?;
+        site_b_env.start_rustfs_server_without_cleanup_with_env(&site_env).await?;
+
+        let site_a_client = site_a_env.create_s3_client();
+        let site_b_client = site_b_env.create_s3_client();
+        let bucket = "site-repl-peer-outage";
+        let key = "after-recovery.txt";
+        let payload = b"site replication recovered the missed bucket".to_vec();
+
+        let add_status = site_replication_add(
+            &site_a_env,
+            &[
+                PeerSite {
+                    name: "outage-site-a".to_string(),
+                    endpoint: site_a_env.url.clone(),
+                    access_key: site_a_env.access_key.clone(),
+                    secret_key: site_a_env.secret_key.clone(),
+                    ..Default::default()
+                },
+                PeerSite {
+                    name: "outage-site-b".to_string(),
+                    endpoint: site_b_env.url.clone(),
+                    access_key: site_b_env.access_key.clone(),
+                    secret_key: site_b_env.secret_key.clone(),
+                    ..Default::default()
+                },
+            ],
+        )
+        .await?;
+        assert!(add_status.success, "unexpected site add result: {add_status:?}");
+        wait_for_site_replication_enabled(&site_a_env, 2).await?;
+        wait_for_site_replication_enabled(&site_b_env, 2).await?;
+
+        site_b_env.stop_server();
+        site_a_client.create_bucket().bucket(bucket).send().await?;
+        site_a_client.head_bucket().bucket(bucket).send().await?;
+
+        let queued = site_replication_info(&site_a_env)
+            .await?
+            .retry_stats
+            .ok_or("peer outage did not persist a site replication retry event")?;
+        assert!(queued.pending + queued.failed > 0, "peer outage retry queue was unexpectedly empty");
+
+        site_b_env.restart_server_preserving_data(vec![], &site_env).await?;
+        let recovery_deadline = tokio::time::Instant::now() + Duration::from_secs(75);
+        loop {
+            let bucket_recovered = site_b_client.head_bucket().bucket(bucket).send().await.is_ok();
+            let queue_empty = site_replication_info(&site_a_env).await?.retry_stats.is_none();
+            if bucket_recovered && queue_empty {
+                break;
+            }
+            if tokio::time::Instant::now() >= recovery_deadline {
+                return Err(format!(
+                    "site replication retry did not settle after peer recovery; bucket_recovered={bucket_recovered}, queue_empty={queue_empty}"
+                )
+                .into());
+            }
+            sleep(Duration::from_millis(250)).await;
+        }
+
+        site_a_client
+            .put_object()
+            .bucket(bucket)
+            .key(key)
+            .body(ByteStream::from(payload.clone()))
+            .send()
+            .await?;
+        assert_eq!(wait_for_object_on_target(&site_b_client, bucket, key).await?, payload);
+
+        Ok(())
+    })
+    .await
+    {
+        Ok(result) => result,
+        Err(_) => Err("site replication peer-outage recovery timed out after 150 seconds".into()),
+    }
+}
+
 /// Re-applying a site's own replication config must not disable the peer's reverse direction.
 ///
 /// `PutBucketReplication` broadcasts the config to every peer — the console's replication

crates/e2e_test/src/upgrade_compatibility_test.rs

@@ -12,21 +12,34 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use crate::common::{RustFSTestClusterEnvironment, RustFSTestEnvironment, init_logging, rustfs_binary_path};
+use crate::common::{
+    RustFSTestClusterEnvironment, RustFSTestEnvironment, admin_request, init_logging, replication_fast_env, rustfs_binary_path,
+};
+use crate::fake_s3_target::{BucketMode, FAKE_ACCESS_KEY, FAKE_SECRET_KEY, FakeS3Target};
+use crate::on_demand_migration::common::{ODM_SERVER_ENV, OdmTestEnv, SeedObject};
+use crate::replication_extension_test::{
+    LOOPBACK_REPLICATION_TARGET_ENV, ReplicationTargetOptions, put_bucket_replication, set_replication_target_with_options,
+};
 use aws_sdk_s3::Client;
 use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{
-    BucketVersioningStatus, CompletedMultipartUpload, CompletedPart, ServerSideEncryption, VersioningConfiguration,
+    BucketLifecycleConfiguration, BucketVersioningStatus, CompletedMultipartUpload, CompletedPart, DefaultRetention,
+    ExpirationStatus, LifecycleExpiration, LifecycleRule, LifecycleRuleFilter, ObjectLockConfiguration, ObjectLockEnabled,
+    ObjectLockRetentionMode, ObjectLockRule, PublicAccessBlockConfiguration, ServerSideEncryption, ServerSideEncryptionByDefault,
+    ServerSideEncryptionConfiguration, ServerSideEncryptionRule, Tag, Tagging, VersioningConfiguration,
 };
+use http::{Method, StatusCode};
 use std::path::{Path, PathBuf};
 use std::time::Duration;
 use tokio::task::JoinSet;
 use tokio::time::{Instant, sleep};
 
 type TestResult = Result<(), Box<dyn std::error::Error + Send + Sync>>;
+type BoxError = Box<dyn std::error::Error + Send + Sync>;
 
 const SOURCE_BINARY_ENV: &str = "RUSTFS_UPGRADE_SOURCE_BINARY";
+const RC5_COMMIT: &str = "40a2470feb567201165a5b809b7598bb4b1f68f5";
 const SSE_MASTER_KEY_ENV: &str = "RUSTFS_SSE_S3_MASTER_KEY";
 const SSE_MASTER_KEY: &str = "QkJCQkJCQkJCQkJCQkJCQkJCQkJCQkJCQkJCQkJCQkI=";
 const PLAIN_BUCKET: &str = "upgrade-plain-data";
@@ -40,6 +53,32 @@ const MULTIPART_UPLOADS_PER_WORKER: usize = 16;
 // comfortably covers that window plus CI scheduling jitter.
 const LISTING_CONVERGENCE_TIMEOUT: Duration = Duration::from_secs(30);
 
+// Bucket-configuration upgrade/rollback scenarios (rustfs#7172, #7183, #7089).
+const CONFIG_PLAIN_BUCKET: &str = "upgrade-config-plain";
+const CONFIG_ENCRYPTED_BUCKET: &str = "upgrade-config-encrypted";
+const CONFIG_REPLICATED_BUCKET: &str = "upgrade-config-replicated";
+const CONFIG_LOCKED_BUCKET: &str = "upgrade-config-locked";
+const CONFIG_REPLICA_BUCKET: &str = "upgrade-config-replica";
+const ROLLBACK_BUCKET: &str = "rollback-config-data";
+const ROLLBACK_REPLICA_BUCKET: &str = "rollback-config-replica";
+const BUCKET_QUOTA_BYTES: u64 = 64 * 1024 * 1024;
+const LIFECYCLE_RULE_ID: &str = "upgrade-expire-logs";
+const LIFECYCLE_PREFIX: &str = "logs/";
+const LIFECYCLE_DAYS: i32 = 30;
+const BUCKET_TAG_KEY: &str = "owner";
+const BUCKET_TAG_VALUE: &str = "upgrade-compatibility";
+const OBJECT_LOCK_DAYS: i32 = 1;
+// `set-bucket-quota` answers 503 until the scanner has made the bucket's usage
+// authoritative; the quota test uses the same 30s budget.
+const QUOTA_READINESS_TIMEOUT: Duration = Duration::from_secs(30);
+// Quota admission fails closed while a freshly started server has neither
+// authoritative usage nor a persisted degraded baseline for the bucket
+// (rustfs#5716), so a write to a quota-enabled bucket is retryable-503 for that
+// window. It is a restart property, not an upgrade property — the same window
+// opens on the very first start — so the write assertions ride it out instead
+// of treating it as an upgrade failure.
+const QUOTA_ADMISSION_WARMUP_TIMEOUT: Duration = Duration::from_secs(90);
+
 fn source_binary() -> Result<PathBuf, Box<dyn std::error::Error + Send + Sync>> {
     let path = std::env::var_os(SOURCE_BINARY_ENV)
         .map(PathBuf::from)
@@ -240,6 +279,93 @@ async fn exercise_mixed_cluster(
     Ok(())
 }
 
+/// Pins the published old writer's limitation and the supported recovery
+/// procedure. This is not a promise that mixed-version ODM is supported.
+/// Replace the loss assertion when ODM gains independent persistence;
+/// preserving configuration across rc.5 writes is then an improvement.
+#[tokio::test]
+#[ignore = "requires the pinned 1.0.0-rc.5 release binary"]
+async fn rc5_rollback_requires_restoring_odm_configuration() -> TestResult {
+    init_logging();
+    let previous_binary = source_binary()?;
+    let version = tokio::process::Command::new(&previous_binary)
+        .arg("--version")
+        .output()
+        .await?;
+    assert!(version.status.success(), "previous binary must report its version");
+    assert!(
+        String::from_utf8(version.stdout)?.contains(RC5_COMMIT),
+        "this compatibility scenario requires the published rc.5 writer"
+    );
+    let mut env = OdmTestEnv::start().await?;
+    let bucket = "odm-rc5-rollback";
+    let source_bucket = "odm-rc5-source";
+    env.source.create_bucket_with_mode(source_bucket, BucketMode::Unversioned);
+    env.seed_source(
+        source_bucket,
+        &[SeedObject::new(
+            "source-only",
+            bytes::Bytes::from_static(b"source read after recovery"),
+        )],
+    );
+    env.rustfs.create_test_bucket(bucket).await?;
+    let saved_config = env.fake_source_spec(source_bucket);
+    assert_eq!(env.configure_source(bucket, &saved_config).await?.status, 200);
+    let before = env.get_config(bucket).await?;
+    assert_eq!(before.status, 200);
+    let expected_config = before
+        .json()?
+        .get("config")
+        .cloned()
+        .ok_or("configuration response omitted config")?;
+    env.client
+        .put_object()
+        .bucket(bucket)
+        .key("local")
+        .body(ByteStream::from_static(b"local data survives rollback"))
+        .send()
+        .await?;
+    env.rustfs.restart_server_preserving_data(vec![], ODM_SERVER_ENV).await?;
+    let restarted = env.get_config(bucket).await?;
+    assert_eq!(restarted.status, 200, "a current writer preserves ODM across restart");
+    assert_eq!(restarted.json()?.get("config"), Some(&expected_config));
+
+    restart_from_binary(&mut env.rustfs, &previous_binary, &[]).await?;
+    env.client
+        .put_bucket_tagging()
+        .bucket(bucket)
+        .tagging(
+            Tagging::builder()
+                .tag_set(Tag::builder().key("writer").value("rc5").build()?)
+                .build()?,
+        )
+        .send()
+        .await?;
+    env.rustfs.restart_server_preserving_data(vec![], ODM_SERVER_ENV).await?;
+    let missing = env.get_config(bucket).await?;
+    assert_eq!(missing.status, 404, "rc.5 rewrites metadata without ODM keys");
+    assert!(missing.body.contains("NoSuchConfiguration"));
+    assert_eq!(read_object(&env.client, bucket, "local", None).await?.1, b"local data survives rollback");
+    let tags = env.client.get_bucket_tagging().bucket(bucket).send().await?;
+    assert!(tags.tag_set().iter().any(|tag| tag.key() == "writer" && tag.value() == "rc5"));
+
+    assert_eq!(
+        env.configure_source(bucket, &saved_config).await?.status,
+        200,
+        "restore from saved full configuration"
+    );
+    env.rustfs.restart_server_preserving_data(vec![], ODM_SERVER_ENV).await?;
+    let restored = env.get_config(bucket).await?;
+    assert_eq!(restored.status, 200, "restored ODM configuration persists");
+    assert_eq!(restored.json()?.get("config"), Some(&expected_config));
+    env.wait_until_source_consulted(bucket).await?;
+    assert_eq!(
+        read_object(&env.client, bucket, "source-only", None).await?.1,
+        b"source read after recovery"
+    );
+    Ok(())
+}
+
 #[tokio::test]
 #[ignore = "requires a pinned previous RustFS release binary"]
 async fn direct_upgrade_from_rc2_preserves_object_contracts() -> TestResult {
@@ -429,3 +555,653 @@ async fn rolling_upgrade_from_rc2_preserves_mixed_version_contracts() -> TestRes
 
     Ok(())
 }
+
+/// Child-process environment shared by both bucket-configuration scenarios.
+///
+/// The replication target is an in-process fake bound to `127.0.0.1`, which
+/// `set-remote-target` rejects as an SSRF risk without the loopback opt-in, and
+/// the proxy bypass keeps a developer's `HTTP_PROXY` from intercepting the
+/// server's outbound health check.
+fn bucket_config_server_env() -> Vec<(&'static str, &'static str)> {
+    let mut env = vec![
+        (SSE_MASTER_KEY_ENV, SSE_MASTER_KEY),
+        ("NO_PROXY", "127.0.0.1,localhost"),
+        ("HTTP_PROXY", ""),
+        ("HTTPS_PROXY", ""),
+        // Shorten the scanner cycle so the bucket's usage becomes authoritative
+        // in seconds; both `set-bucket-quota` and quota admission block on it.
+        ("RUSTFS_SCANNER_CYCLE", "1"),
+        ("RUSTFS_SCANNER_START_DELAY_SECS", "0"),
+    ];
+    env.extend_from_slice(LOOPBACK_REPLICATION_TARGET_ENV);
+    env.extend(replication_fast_env());
+    env
+}
+
+/// Restart `env` in place on the same data directory using an explicit binary.
+///
+/// [`RustFSTestEnvironment::restart_server_preserving_data`] always relaunches
+/// the workspace build, which is the upgrade direction only. The rollback
+/// scenario needs the reverse: stop the current build and bring the pinned
+/// previous release up on the metadata that build just wrote.
+async fn restart_from_binary(env: &mut RustFSTestEnvironment, binary: &Path, server_env: &[(&str, &str)]) -> TestResult {
+    env.stop_server();
+    env.start_rustfs_server_from_binary(binary, vec![], server_env).await
+}
+
+async fn set_bucket_quota(env: &RustFSTestEnvironment, bucket: &str, quota_bytes: u64) -> TestResult {
+    let path = format!("/rustfs/admin/v3/quota/{bucket}");
+    let body = serde_json::json!({ "quota": quota_bytes, "quota_type": "HARD" }).to_string();
+    let deadline = Instant::now() + QUOTA_READINESS_TIMEOUT;
+    loop {
+        let (status, response) =
+            admin_request(&env.url, Method::PUT, &path, Some(body.clone()), &env.access_key, &env.secret_key).await?;
+        if status.is_success() {
+            return Ok(());
+        }
+        if status != StatusCode::SERVICE_UNAVAILABLE || Instant::now() >= deadline {
+            return Err(format!("setting the quota of {bucket} failed: {status} {response}").into());
+        }
+        sleep(Duration::from_millis(500)).await;
+    }
+}
+
+/// PUT into a quota-enabled bucket, riding out the post-start quota-admission
+/// warm-up described on [`QUOTA_ADMISSION_WARMUP_TIMEOUT`].
+///
+/// Only `ServiceUnavailable` is retried: any other failure, and a warm-up that
+/// never ends, is a genuine regression and surfaces as an error.
+async fn put_object_through_quota_warmup(client: &Client, bucket: &str, key: &str, body: &'static [u8]) -> TestResult {
+    let deadline = Instant::now() + QUOTA_ADMISSION_WARMUP_TIMEOUT;
+    loop {
+        let result = client
+            .put_object()
+            .bucket(bucket)
+            .key(key)
+            .body(ByteStream::from_static(body))
+            .send()
+            .await;
+        let error = match result {
+            Ok(_) => return Ok(()),
+            Err(error) => error,
+        };
+        let retryable = error.as_service_error().and_then(ProvideErrorMetadata::code) == Some("ServiceUnavailable");
+        if !retryable || Instant::now() >= deadline {
+            return Err(format!("PUT {bucket}/{key} failed after the quota warm-up window: {error}").into());
+        }
+        sleep(Duration::from_millis(500)).await;
+    }
+}
+
+async fn get_bucket_quota(env: &RustFSTestEnvironment, bucket: &str) -> Result<Option<u64>, BoxError> {
+    let path = format!("/rustfs/admin/v3/quota/{bucket}");
+    let (status, response) = admin_request(&env.url, Method::GET, &path, None, &env.access_key, &env.secret_key).await?;
+    if status != StatusCode::OK {
+        return Err(format!("reading the quota of {bucket} failed: {status} {response}").into());
+    }
+    let quota: serde_json::Value = serde_json::from_str(&response)?;
+    Ok(quota.get("quota").and_then(serde_json::Value::as_u64))
+}
+
+/// `GET /rustfs/admin/v3/list-remote-targets?bucket=...`.
+///
+/// Returns an error for any non-200, because rustfs#7172 made this endpoint
+/// fail closed on a `bucket-targets.json` blob the running build cannot parse.
+/// An upgrade that misreads a blob written by the previous release therefore
+/// shows up here as an error, and a silently dropped target shows up as an
+/// empty list — the caller must distinguish the two.
+async fn list_remote_targets(env: &RustFSTestEnvironment, bucket: &str) -> Result<Vec<serde_json::Value>, BoxError> {
+    let path = format!("/rustfs/admin/v3/list-remote-targets?bucket={}", urlencoding::encode(bucket));
+    let (status, response) = admin_request(&env.url, Method::GET, &path, None, &env.access_key, &env.secret_key).await?;
+    if status != StatusCode::OK {
+        return Err(format!("list-remote-targets for {bucket} failed: {status} {response}").into());
+    }
+    Ok(serde_json::from_str(&response)?)
+}
+
+/// Assert that `bucket` still carries exactly the replication target `arn`.
+async fn assert_remote_target_preserved(env: &RustFSTestEnvironment, bucket: &str, arn: &str, context: &str) -> TestResult {
+    let targets = list_remote_targets(env, bucket).await?;
+    assert_eq!(
+        targets.len(),
+        1,
+        "{context}: list-remote-targets must still report the single configured target, got {targets:?}"
+    );
+    assert_eq!(
+        targets[0].get("arn").and_then(serde_json::Value::as_str),
+        Some(arn),
+        "{context}: the target ARN changed across the restart: {targets:?}"
+    );
+    Ok(())
+}
+
+/// Configure a replication target on `bucket` pointing at the in-process fake,
+/// then attach an enabled replication rule for it. Returns the target ARN.
+async fn configure_replication(
+    env: &RustFSTestEnvironment,
+    bucket: &str,
+    target: &FakeS3Target,
+    target_bucket: &str,
+) -> Result<String, BoxError> {
+    let arn = set_replication_target_with_options(
+        env,
+        bucket,
+        ReplicationTargetOptions {
+            endpoint: &target.address(),
+            access_key: FAKE_ACCESS_KEY,
+            secret_key: FAKE_SECRET_KEY,
+            target_bucket,
+            secure: false,
+            skip_tls_verify: false,
+            ca_cert_pem: None,
+        },
+    )
+    .await?;
+    put_bucket_replication(env, bucket, &arn).await?;
+    Ok(arn)
+}
+
+async fn put_default_sse_s3_encryption(client: &Client, bucket: &str) -> TestResult {
+    let configuration = ServerSideEncryptionConfiguration::builder()
+        .rules(
+            ServerSideEncryptionRule::builder()
+                .apply_server_side_encryption_by_default(
+                    ServerSideEncryptionByDefault::builder()
+                        .sse_algorithm(ServerSideEncryption::Aes256)
+                        .build()?,
+                )
+                .build(),
+        )
+        .build()?;
+    client
+        .put_bucket_encryption()
+        .bucket(bucket)
+        .server_side_encryption_configuration(configuration)
+        .send()
+        .await?;
+    Ok(())
+}
+
+async fn assert_default_sse_s3_encryption(client: &Client, bucket: &str, context: &str) -> TestResult {
+    let response = client.get_bucket_encryption().bucket(bucket).send().await?;
+    let rules = response
+        .server_side_encryption_configuration()
+        .ok_or("GetBucketEncryption omitted the configuration")?
+        .rules();
+    assert_eq!(rules.len(), 1, "{context}: expected exactly one encryption rule, got {rules:?}");
+    assert_eq!(
+        rules[0]
+            .apply_server_side_encryption_by_default()
+            .map(ServerSideEncryptionByDefault::sse_algorithm),
+        Some(&ServerSideEncryption::Aes256),
+        "{context}: the default encryption algorithm changed"
+    );
+    Ok(())
+}
+
+async fn put_bucket_tag(client: &Client, bucket: &str) -> TestResult {
+    let tagging = Tagging::builder()
+        .tag_set(Tag::builder().key(BUCKET_TAG_KEY).value(BUCKET_TAG_VALUE).build()?)
+        .build()?;
+    client.put_bucket_tagging().bucket(bucket).tagging(tagging).send().await?;
+    Ok(())
+}
+
+async fn assert_bucket_tag(client: &Client, bucket: &str, context: &str) -> TestResult {
+    let tags = client.get_bucket_tagging().bucket(bucket).send().await?;
+    let tag_set = tags.tag_set();
+    assert_eq!(tag_set.len(), 1, "{context}: expected exactly one bucket tag, got {tag_set:?}");
+    assert_eq!(tag_set[0].key(), BUCKET_TAG_KEY, "{context}: bucket tag key changed");
+    assert_eq!(tag_set[0].value(), BUCKET_TAG_VALUE, "{context}: bucket tag value changed");
+    Ok(())
+}
+
+async fn assert_versioning_enabled(client: &Client, bucket: &str, context: &str) -> TestResult {
+    let versioning = client.get_bucket_versioning().bucket(bucket).send().await?;
+    assert_eq!(
+        versioning.status(),
+        Some(&BucketVersioningStatus::Enabled),
+        "{context}: versioning is no longer Enabled on {bucket}"
+    );
+    Ok(())
+}
+
+fn bucket_policy_document(bucket: &str) -> serde_json::Value {
+    serde_json::json!({
+        "Version": "2012-10-17",
+        "Statement": [{
+            "Sid": "UpgradePublicRead",
+            "Effect": "Allow",
+            "Principal": { "AWS": ["*"] },
+            "Action": ["s3:GetObject"],
+            "Resource": [format!("arn:aws:s3:::{bucket}/public/*")]
+        }]
+    })
+}
+
+/// `GET .../on-demand-migration/{bucket}/status`.
+///
+/// The migration module defaults on from rustfs#7089, so a bucket that never
+/// configured a source must still answer `configured: false` rather than
+/// engaging the migration path.
+async fn assert_migration_not_configured(env: &RustFSTestEnvironment, bucket: &str) -> TestResult {
+    let path = format!("/rustfs/admin/v3/on-demand-migration/{bucket}/status");
+    let (status, response) = admin_request(&env.url, Method::GET, &path, None, &env.access_key, &env.secret_key).await?;
+    assert_eq!(
+        status,
+        StatusCode::OK,
+        "the migration status endpoint must answer for an unconfigured bucket: {status} {response}"
+    );
+    let body: serde_json::Value = serde_json::from_str(&response)?;
+    assert_eq!(
+        body.get("configured"),
+        Some(&serde_json::Value::Bool(false)),
+        "a bucket upgraded from the previous release must not look migration-configured: {body}"
+    );
+    Ok(())
+}
+
+/// A GET for a key that was never written must be a plain `NoSuchKey`.
+///
+/// With the migration module on by default this is the cheap proof that an
+/// unconfigured bucket never consults a source: any migration engagement would
+/// surface as a different status or error code here.
+async fn assert_missing_key_is_no_such_key(client: &Client, bucket: &str, key: &str) -> TestResult {
+    let error = client
+        .get_object()
+        .bucket(bucket)
+        .key(key)
+        .send()
+        .await
+        .expect_err("a key that was never written must not be readable");
+    assert_eq!(
+        error.raw_response().map(|response| response.status().as_u16()),
+        Some(404),
+        "a missing key must stay a 404 on a bucket with no migration configuration"
+    );
+    assert_eq!(
+        error.as_service_error().and_then(ProvideErrorMetadata::code),
+        Some("NoSuchKey"),
+        "a missing key must stay NoSuchKey on a bucket with no migration configuration"
+    );
+    Ok(())
+}
+
+/// Bucket configuration written by the pinned previous release must survive an
+/// upgrade to the current build unchanged, and must keep working.
+///
+/// This pins the three on-disk surfaces the on-demand-migration series moved:
+///
+/// * `BucketMetadata` grew two msgpack keys (encoded map length 44 -> 46), so
+///   every configuration read below decodes a 44-key blob on 46-key code.
+/// * rustfs#7172 made an unreadable `bucket-targets.json` / encryption /
+///   public-access-block / quota blob "present but unreadable" instead of
+///   silently defaulting, and made `list-remote-targets` fail closed on it. A
+///   replication target configured by the old release must therefore still be
+///   *listed*, not dropped and not an error.
+/// * rustfs#7183 made the object write path refuse a PUT when the bucket's
+///   encryption configuration cannot be read, so a misparsed SSE config would
+///   turn every PUT to that bucket into a 500.
+///
+/// Not covered on purpose: on-demand-migration configuration itself, which the
+/// previous release has no public API for — the reverse direction is asserted
+/// instead (an upgraded bucket reports `configured: false`).
+#[tokio::test]
+#[ignore = "requires a pinned previous RustFS release binary"]
+async fn direct_upgrade_from_previous_release_preserves_bucket_configuration() -> TestResult {
+    init_logging();
+    let previous_binary = source_binary()?;
+
+    // In-process: the fake target outlives both server processes, so the
+    // replication target stays reachable across the upgrade.
+    let replication_target = FakeS3Target::start().await?;
+    replication_target.create_bucket(CONFIG_REPLICA_BUCKET);
+
+    let mut env = RustFSTestEnvironment::new().await?;
+    let server_env = bucket_config_server_env();
+    env.start_rustfs_server_from_binary(&previous_binary, vec![], &server_env)
+        .await?;
+    let old_client = env.create_s3_client();
+
+    env.create_test_bucket(CONFIG_PLAIN_BUCKET).await?;
+    env.create_test_bucket(CONFIG_ENCRYPTED_BUCKET).await?;
+    env.create_test_bucket(CONFIG_REPLICATED_BUCKET).await?;
+    old_client
+        .create_bucket()
+        .bucket(CONFIG_LOCKED_BUCKET)
+        .object_lock_enabled_for_bucket(true)
+        .send()
+        .await?;
+
+    // Plain bucket: policy, tags, lifecycle, quota.
+    let policy = bucket_policy_document(CONFIG_PLAIN_BUCKET);
+    old_client
+        .put_bucket_policy()
+        .bucket(CONFIG_PLAIN_BUCKET)
+        .policy(policy.to_string())
+        .send()
+        .await?;
+    put_bucket_tag(&old_client, CONFIG_PLAIN_BUCKET).await?;
+    old_client
+        .put_bucket_lifecycle_configuration()
+        .bucket(CONFIG_PLAIN_BUCKET)
+        .lifecycle_configuration(
+            BucketLifecycleConfiguration::builder()
+                .rules(
+                    LifecycleRule::builder()
+                        .id(LIFECYCLE_RULE_ID)
+                        .status(ExpirationStatus::Enabled)
+                        .filter(LifecycleRuleFilter::builder().prefix(LIFECYCLE_PREFIX).build())
+                        .expiration(LifecycleExpiration::builder().days(LIFECYCLE_DAYS).build())
+                        .build()?,
+                )
+                .build()?,
+        )
+        .send()
+        .await?;
+    set_bucket_quota(&env, CONFIG_PLAIN_BUCKET, BUCKET_QUOTA_BYTES).await?;
+
+    // Encrypted bucket: SSE-S3 default encryption plus a fully restrictive
+    // public access block, both of which rustfs#7172 now fails closed on.
+    put_default_sse_s3_encryption(&old_client, CONFIG_ENCRYPTED_BUCKET).await?;
+    old_client
+        .put_public_access_block()
+        .bucket(CONFIG_ENCRYPTED_BUCKET)
+        .public_access_block_configuration(
+            PublicAccessBlockConfiguration::builder()
+                .block_public_acls(true)
+                .ignore_public_acls(true)
+                .block_public_policy(true)
+                .restrict_public_buckets(true)
+                .build(),
+        )
+        .send()
+        .await?;
+
+    // Replicated bucket: versioning, a validated remote target, a rule.
+    enable_versioning(&old_client, CONFIG_REPLICATED_BUCKET).await?;
+    let target_arn = configure_replication(&env, CONFIG_REPLICATED_BUCKET, &replication_target, CONFIG_REPLICA_BUCKET).await?;
+    assert_remote_target_preserved(&env, CONFIG_REPLICATED_BUCKET, &target_arn, "before the upgrade").await?;
+
+    // Object-lock bucket: a default GOVERNANCE retention on a fresh bucket.
+    old_client
+        .put_object_lock_configuration()
+        .bucket(CONFIG_LOCKED_BUCKET)
+        .object_lock_configuration(
+            ObjectLockConfiguration::builder()
+                .object_lock_enabled(ObjectLockEnabled::Enabled)
+                .rule(
+                    ObjectLockRule::builder()
+                        .default_retention(
+                            DefaultRetention::builder()
+                                .mode(ObjectLockRetentionMode::Governance)
+                                .days(OBJECT_LOCK_DAYS)
+                                .build(),
+                        )
+                        .build(),
+                )
+                .build(),
+        )
+        .send()
+        .await?;
+
+    let plain_key = "plain/written-by-previous";
+    let plain_bytes = b"plain object written by the previous RustFS release";
+    put_object_through_quota_warmup(&old_client, CONFIG_PLAIN_BUCKET, plain_key, plain_bytes).await?;
+
+    let encrypted_key = "encrypted/written-by-previous";
+    let encrypted_bytes = b"default-encrypted object written by the previous RustFS release";
+    old_client
+        .put_object()
+        .bucket(CONFIG_ENCRYPTED_BUCKET)
+        .key(encrypted_key)
+        .body(ByteStream::from_static(encrypted_bytes))
+        .send()
+        .await?;
+    assert_eq!(
+        read_object(&old_client, CONFIG_ENCRYPTED_BUCKET, encrypted_key, None)
+            .await?
+            .0,
+        Some(ServerSideEncryption::Aes256),
+        "the previous release must apply the bucket default encryption it just accepted"
+    );
+
+    // The multipart object lives in the default-encrypted bucket so the
+    // upgraded build has to reassemble parts *and* re-derive the object key.
+    let multipart_key = "encrypted/multipart-written-by-previous";
+    let multipart_parts = vec![vec![b'm'; 5 * 1024 * 1024], b"final multipart bytes".to_vec()];
+    let multipart_bytes = multipart_parts.concat();
+    write_multipart(&old_client, CONFIG_ENCRYPTED_BUCKET, multipart_key, &multipart_parts).await?;
+
+    let versioned_key = "versioned/written-by-previous";
+    let versioned_bytes = b"versioned object written by the previous RustFS release";
+    let versioned_id = old_client
+        .put_object()
+        .bucket(CONFIG_REPLICATED_BUCKET)
+        .key(versioned_key)
+        .body(ByteStream::from_static(versioned_bytes))
+        .send()
+        .await?
+        .version_id()
+        .ok_or("versioned PUT omitted version ID")?
+        .to_string();
+
+    env.restart_server_preserving_data(vec![], &server_env).await?;
+    let new_client = env.create_s3_client();
+
+    // Every configuration must read back unchanged on the upgraded build.
+    let upgraded_policy = new_client.get_bucket_policy().bucket(CONFIG_PLAIN_BUCKET).send().await?;
+    let upgraded_policy: serde_json::Value =
+        serde_json::from_str(upgraded_policy.policy().ok_or("GetBucketPolicy omitted the document")?)?;
+    assert_eq!(upgraded_policy, policy, "the bucket policy changed across the upgrade");
+    assert_bucket_tag(&new_client, CONFIG_PLAIN_BUCKET, "after the upgrade").await?;
+
+    let lifecycle = new_client
+        .get_bucket_lifecycle_configuration()
+        .bucket(CONFIG_PLAIN_BUCKET)
+        .send()
+        .await?;
+    let rules = lifecycle.rules();
+    assert_eq!(rules.len(), 1, "the lifecycle rule count changed across the upgrade: {rules:?}");
+    assert_eq!(rules[0].id(), Some(LIFECYCLE_RULE_ID));
+    assert_eq!(rules[0].status(), &ExpirationStatus::Enabled);
+    assert_eq!(
+        rules[0].expiration().and_then(LifecycleExpiration::days),
+        Some(LIFECYCLE_DAYS),
+        "the lifecycle expiration changed across the upgrade"
+    );
+
+    assert_eq!(
+        get_bucket_quota(&env, CONFIG_PLAIN_BUCKET).await?,
+        Some(BUCKET_QUOTA_BYTES),
+        "the bucket quota changed across the upgrade"
+    );
+
+    assert_default_sse_s3_encryption(&new_client, CONFIG_ENCRYPTED_BUCKET, "after the upgrade").await?;
+    let public_access_block = new_client
+        .get_public_access_block()
+        .bucket(CONFIG_ENCRYPTED_BUCKET)
+        .send()
+        .await?;
+    let public_access_block = public_access_block
+        .public_access_block_configuration()
+        .ok_or("GetPublicAccessBlock omitted the configuration")?;
+    assert_eq!(public_access_block.block_public_acls(), Some(true));
+    assert_eq!(public_access_block.ignore_public_acls(), Some(true));
+    assert_eq!(public_access_block.block_public_policy(), Some(true));
+    assert_eq!(public_access_block.restrict_public_buckets(), Some(true));
+
+    assert_versioning_enabled(&new_client, CONFIG_REPLICATED_BUCKET, "after the upgrade").await?;
+    // rustfs#7172: neither an empty list nor an error is acceptable here.
+    assert_remote_target_preserved(&env, CONFIG_REPLICATED_BUCKET, &target_arn, "after the upgrade").await?;
+    let replication = new_client
+        .get_bucket_replication()
+        .bucket(CONFIG_REPLICATED_BUCKET)
+        .send()
+        .await?;
+    let replication_rules = replication
+        .replication_configuration()
+        .ok_or("GetBucketReplication omitted the configuration")?
+        .rules();
+    assert_eq!(
+        replication_rules.len(),
+        1,
+        "the replication rule count changed across the upgrade: {replication_rules:?}"
+    );
+    assert_eq!(
+        replication_rules[0].destination().map(|destination| destination.bucket()),
+        Some(target_arn.as_str()),
+        "the replication rule no longer points at the configured target"
+    );
+
+    let object_lock = new_client
+        .get_object_lock_configuration()
+        .bucket(CONFIG_LOCKED_BUCKET)
+        .send()
+        .await?;
+    let object_lock = object_lock
+        .object_lock_configuration()
+        .ok_or("GetObjectLockConfiguration omitted the configuration")?;
+    assert_eq!(object_lock.object_lock_enabled(), Some(&ObjectLockEnabled::Enabled));
+    let retention = object_lock
+        .rule()
+        .and_then(ObjectLockRule::default_retention)
+        .ok_or("the object lock configuration lost its default retention")?;
+    assert_eq!(retention.mode(), Some(&ObjectLockRetentionMode::Governance));
+    assert_eq!(retention.days(), Some(OBJECT_LOCK_DAYS));
+
+    // rustfs#7183: a PUT into the default-encrypted bucket must still succeed
+    // and still come back encrypted.
+    let post_upgrade_encrypted_key = "encrypted/written-after-upgrade";
+    let post_upgrade_encrypted_bytes = b"default-encrypted object written by the current RustFS build";
+    new_client
+        .put_object()
+        .bucket(CONFIG_ENCRYPTED_BUCKET)
+        .key(post_upgrade_encrypted_key)
+        .body(ByteStream::from_static(post_upgrade_encrypted_bytes))
+        .send()
+        .await?;
+    let (encryption, body) = read_object(&new_client, CONFIG_ENCRYPTED_BUCKET, post_upgrade_encrypted_key, None).await?;
+    assert_eq!(
+        encryption,
+        Some(ServerSideEncryption::Aes256),
+        "a PUT after the upgrade lost the bucket default encryption"
+    );
+    assert_eq!(body, post_upgrade_encrypted_bytes);
+
+    let post_upgrade_plain_key = "plain/written-after-upgrade";
+    let post_upgrade_plain_bytes = b"plain object written by the current RustFS build";
+    put_object_through_quota_warmup(&new_client, CONFIG_PLAIN_BUCKET, post_upgrade_plain_key, post_upgrade_plain_bytes).await?;
+    let (encryption, body) = read_object(&new_client, CONFIG_PLAIN_BUCKET, post_upgrade_plain_key, None).await?;
+    assert_eq!(encryption, None, "a bucket without default encryption must not encrypt a PUT");
+    assert_eq!(body, post_upgrade_plain_bytes);
+
+    // Every object written by the previous release reads back byte-identical.
+    assert_eq!(read_object(&new_client, CONFIG_PLAIN_BUCKET, plain_key, None).await?.1, plain_bytes);
+    let (encryption, body) = read_object(&new_client, CONFIG_ENCRYPTED_BUCKET, encrypted_key, None).await?;
+    assert_eq!(encryption, Some(ServerSideEncryption::Aes256));
+    assert_eq!(body, encrypted_bytes);
+    let (encryption, body) = read_object(&new_client, CONFIG_ENCRYPTED_BUCKET, multipart_key, None).await?;
+    assert_eq!(encryption, Some(ServerSideEncryption::Aes256));
+    assert_eq!(body, multipart_bytes, "the multipart object did not survive the upgrade");
+    assert_eq!(
+        read_object(&new_client, CONFIG_REPLICATED_BUCKET, versioned_key, Some(&versioned_id))
+            .await?
+            .1,
+        versioned_bytes
+    );
+
+    // rustfs#7089: the migration module is on by default, but a bucket that
+    // never configured a source behaves exactly as before.
+    assert_migration_not_configured(&env, CONFIG_PLAIN_BUCKET).await?;
+    assert_missing_key_is_no_such_key(&new_client, CONFIG_PLAIN_BUCKET, "plain/never-written").await?;
+
+    replication_target.shutdown().await;
+    Ok(())
+}
+
+/// Rolling back to the pinned previous release must still read the bucket
+/// metadata the current build wrote.
+///
+/// This is the other half of the `BucketMetadata` 44 -> 46 key change: the
+/// current build writes a 46-key msgpack map with `OnDemandMigrationConfigJSON`
+/// and `OnDemandMigrationConfigUpdatedAt`, and the previous release's decoder
+/// has to skip those two unknown keys instead of failing the whole blob. If it
+/// did not, every configuration read below would come back empty or error and
+/// the rollback would silently discard the bucket's configuration.
+#[tokio::test]
+#[ignore = "requires a pinned previous RustFS release binary"]
+async fn rollback_to_previous_release_reads_current_bucket_metadata() -> TestResult {
+    init_logging();
+    let previous_binary = source_binary()?;
+
+    let replication_target = FakeS3Target::start().await?;
+    replication_target.create_bucket(ROLLBACK_REPLICA_BUCKET);
+
+    let mut env = RustFSTestEnvironment::new().await?;
+    let server_env = bucket_config_server_env();
+    env.start_rustfs_server_with_env(vec![], &server_env).await?;
+    let new_client = env.create_s3_client();
+
+    env.create_test_bucket(ROLLBACK_BUCKET).await?;
+    enable_versioning(&new_client, ROLLBACK_BUCKET).await?;
+    put_default_sse_s3_encryption(&new_client, ROLLBACK_BUCKET).await?;
+    put_bucket_tag(&new_client, ROLLBACK_BUCKET).await?;
+    let target_arn = configure_replication(&env, ROLLBACK_BUCKET, &replication_target, ROLLBACK_REPLICA_BUCKET).await?;
+    assert_remote_target_preserved(&env, ROLLBACK_BUCKET, &target_arn, "before the rollback").await?;
+
+    let single_key = "rollback/single";
+    let single_bytes = b"single-part object written by the current RustFS build";
+    let single_version = new_client
+        .put_object()
+        .bucket(ROLLBACK_BUCKET)
+        .key(single_key)
+        .body(ByteStream::from_static(single_bytes))
+        .send()
+        .await?
+        .version_id()
+        .ok_or("versioned PUT omitted version ID")?
+        .to_string();
+
+    let multipart_key = "rollback/multipart";
+    let multipart_parts = vec![vec![b'r'; 5 * 1024 * 1024], b"final rollback bytes".to_vec()];
+    let multipart_bytes = multipart_parts.concat();
+    write_multipart(&new_client, ROLLBACK_BUCKET, multipart_key, &multipart_parts).await?;
+
+    restart_from_binary(&mut env, &previous_binary, &server_env).await?;
+    let old_client = env.create_s3_client();
+
+    assert_versioning_enabled(&old_client, ROLLBACK_BUCKET, "after the rollback").await?;
+    assert_default_sse_s3_encryption(&old_client, ROLLBACK_BUCKET, "after the rollback").await?;
+    assert_bucket_tag(&old_client, ROLLBACK_BUCKET, "after the rollback").await?;
+    assert_remote_target_preserved(&env, ROLLBACK_BUCKET, &target_arn, "after the rollback").await?;
+
+    let (encryption, body) = read_object(&old_client, ROLLBACK_BUCKET, single_key, Some(&single_version)).await?;
+    assert_eq!(encryption, Some(ServerSideEncryption::Aes256));
+    assert_eq!(body, single_bytes);
+    let (encryption, body) = read_object(&old_client, ROLLBACK_BUCKET, multipart_key, None).await?;
+    assert_eq!(encryption, Some(ServerSideEncryption::Aes256));
+    assert_eq!(body, multipart_bytes, "the multipart object did not survive the rollback");
+
+    // A PUT on the rolled-back release must still honour the encryption
+    // configuration it decoded out of the current build's metadata blob.
+    let post_rollback_key = "rollback/written-after-rollback";
+    let post_rollback_bytes = b"object written by the previous RustFS release after the rollback";
+    old_client
+        .put_object()
+        .bucket(ROLLBACK_BUCKET)
+        .key(post_rollback_key)
+        .body(ByteStream::from_static(post_rollback_bytes))
+        .send()
+        .await?;
+    let (encryption, body) = read_object(&old_client, ROLLBACK_BUCKET, post_rollback_key, None).await?;
+    assert_eq!(
+        encryption,
+        Some(ServerSideEncryption::Aes256),
+        "the rolled-back release lost the bucket default encryption"
+    );
+    assert_eq!(body, post_rollback_bytes);
+
+    replication_target.shutdown().await;
+    Ok(())
+}

crates/ecstore/Cargo.toml

@@ -31,6 +31,7 @@ workspace = true
 
 [features]
 default = []
+gcs = ["dep:google-cloud-storage", "dep:google-cloud-auth"]
 # Compiles the controlled list-objects namespace-journal chaos injector into a
 # production binary (it is always available to tests). Off by default so the
 # RUSTFS_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_* env vars cannot rewrite journal
@@ -212,8 +213,8 @@ aws-smithy-runtime-api = { workspace = true, features = ["http-1x"] }
 parking_lot = { workspace = true }
 base64-simd.workspace = true
 serde_urlencoded.workspace = true
-google-cloud-storage = { workspace = true }
+google-cloud-storage = { workspace = true, optional = true }
-google-cloud-auth = { workspace = true }
+google-cloud-auth = { workspace = true, optional = true }
 faster-hex = { workspace = true }
 ratelimit = { workspace = true }
 aws-smithy-http-client = { workspace = true, default-features = false, features = ["rustls-aws-lc"] }
@@ -244,6 +245,7 @@ windows-sys = { workspace = true, features = [
 windows-sys = { workspace = true, features = ["Win32_System_Ioctl"] }
 
 [dev-dependencies]
+aws-smithy-async.workspace = true
 tokio = { workspace = true, features = ["rt-multi-thread", "macros", "test-util", "fs"] }
 criterion = { workspace = true, features = ["html_reports"] }
 temp-env = { workspace = true, features = ["async_closure"] }

crates/ecstore/src/api/mod.rs

@@ -32,7 +32,7 @@ pub mod bucket {
     pub mod bucket_target_sys {
         pub use crate::bucket::bucket_target_sys::{
             AdvancedPutOptions, BucketTargetError, BucketTargetSys, PutObjectOptions, RemoveObjectOptions, S3ClientError,
-            SsecPassthroughCapability, TargetClient, append_version_id_query,
+            SsecPassthroughCapability, TargetClient, UnreadableTargetsPolicy, append_version_id_query,
         };
     }
 
@@ -69,6 +69,13 @@ pub mod bucket {
             };
         }
 
+        pub mod recovery_control {
+            pub use crate::bucket::lifecycle::recovery_control::{
+                IlmRecoveryClassification, IlmRecoveryControlPage, IlmRecoveryControlView, IlmRecoveryProtocol,
+                inspect_recovery_control, list_recovery_controls,
+            };
+        }
+
         pub mod transition_transaction {
             pub use crate::bucket::lifecycle::transition_transaction::{
                 TransitionOperatorDeleteResult, TransitionOperatorError, TransitionOperatorProbe, TransitionOperatorStatus,
@@ -89,8 +96,9 @@ pub mod bucket {
         #[allow(clippy::module_inception)]
         pub mod lifecycle {
             pub use crate::bucket::lifecycle::lifecycle::{
-                Event, ExpirationOptions, IlmAction, Lifecycle, LifecycleCalculate, ObjectOpts, RuleValidate,
+                Event, ExpirationOptions, IlmAction, LIFECYCLE_MALFORMED_XML_ERROR_KIND, Lifecycle, LifecycleCalculate,
-                TRANSITION_COMPLETE, TRANSITION_PENDING, TransitionOptions, expected_expiry_time, object_opts_from_object_info,
+                ObjectOpts, RuleValidate, TRANSITION_COMPLETE, TRANSITION_PENDING, TransitionOptions, expected_expiry_time,
+                object_opts_from_object_info,
             };
         }
 
@@ -146,66 +154,23 @@ pub mod bucket {
         };
     }
 
-    pub mod on_demand_migration {
-        pub use crate::bucket::on_demand_migration::{
-            ApplyOutcome, BREAKER_FAILURE_THRESHOLD, BREAKER_FAILURE_WINDOW, BREAKER_HALF_OPEN_MAX_PROBES, BREAKER_OPEN_DURATION,
-            Breaker, BreakerState, BreakerTransition, BreakerVerdict, BucketOdmState, GLOBAL_ON_DEMAND_MIGRATION_SYS, GaugeGuard,
-            LastSourceError, LatencyBucketSnapshot, NEGATIVE_CACHE_MAX_ENTRIES, NegativeCache, OdmBucketSnapshot, OdmLookup,
-            OdmOp, OdmOutcome, OdmStateError, OdmStats, OdmStatsSnapshot, OnDemandMigrationSys, PullError, PullFailureReason,
-            PullFollower, PullLeader, PullOutcome, PullPath, PullResult, PullSlot, SOURCE_LATENCY_BUCKET_BOUNDS_MS,
-            SourceLatencySnapshot, source_client_spec,
-        };
-        pub use crate::bucket::on_demand_migration::{
-            ConfigPublishHook, FilterConfig, HeadPolicy, ON_DEMAND_MIGRATION_CONFIG_HOOK, ON_DEMAND_MIGRATION_CONFIG_VERSION,
-            OnDemandMigrationConfig, OnDemandMigrationConfigError, PathStyle, PolicyConfig, Provider, RangeGetPolicy,
-            SourceConfig, SourceCredentials, SourceErrorPolicy, SourceTimeout, TlsConfig, ValidationContext,
-        };
-        pub use crate::bucket::on_demand_migration::{
-            EnqueueOutcome, LocalObject, MAX_MULTIPART_PARTS, OdmWriteBack, PULL_MAX_RETRIES, PULL_RETRY_BASE_DELAYS,
-            PullCompletion, PullQueue, PullReason, PullSource, QueuedPullOutcome, SourceBody, SourceIdleGuard, WriteBackBody,
-            WriteBackError, WriteBackOutcome, WriteBackPart, WriteBackRequest, commit_inline, commit_inline_with,
-            idle_guarded_body,
-        };
-        pub use crate::bucket::on_demand_migration::{
-            FetchRequest, LIST_THROUGH_TOKEN_VERSION, ListEntryKey, ListThroughCursor, ListThroughMerger, ListThroughToken,
-            ListThroughTokenError, MAX_LIST_FETCHES_PER_SIDE, MergeOutcome, MergePick, MergeSide, SOURCE_LIST_MAX_RATE_WAIT,
-            SOURCE_LIST_RATE_PER_SEC, SourceListPlan, SourceListRateLimiter, decode_continuation_token, source_list_plan,
-        };
-        pub mod backfill {
-            pub use crate::bucket::on_demand_migration::backfill::{
-                BACKFILL_CHECKPOINT_FILE, BACKFILL_CHECKPOINT_FORMAT_VERSION, BACKFILL_FAILED_KEYS_CAPACITY, BACKFILL_LEASE,
-                BACKFILL_LEASE_LOCK_PREFIX, BACKFILL_LIST_PAGE_SIZE, BACKFILL_RECOVERY_INTERVAL, BACKFILL_SAVE_EVERY_KEYS,
-                BACKFILL_SAVE_INTERVAL, BackfillCheckpoint, BackfillContext, BackfillContextFactory, BackfillError,
-                BackfillLastError, BackfillOwner, BackfillRecoveryStats, BackfillRequest, BackfillRunner, BackfillState,
-                BucketBackfillContext, LocalBackfillObject, PriorityPullPermits, PullPermit, PullPriority, SkipExisting,
-                StoredCheckpoint, SysBackfillContexts, global_backfill_runner, install_global_backfill_runner, key_hash,
-                read_checkpoint, run_backfill_recovery_loop, spawn_backfill_recovery_loop,
-            };
-        }
-        pub mod source_client {
-            pub use crate::bucket::on_demand_migration::source_client::{
-                SourceClient, SourceClientSpec, SourceError, SourceGet, SourceHead, SourceListRequest, SourceObject, SourcePage,
-                SourceProbe, SourceProvider, SourceSse, SourceTimeouts, USER_AGENT_SUFFIX, is_multipart_etag, range_header_value,
-                resolve_path_style,
-            };
-        }
-    }
-
     pub mod metadata_sys {
-        #[cfg(feature = "test-util")]
-        pub use crate::bucket::metadata_sys::ConfigWriteLockProbe;
         pub use crate::bucket::metadata_sys::{
-            BucketMetadataMutationGuard, BucketMetadataSys, ObjectLockConfigState, acquire_bucket_metadata_transaction_lock,
+            BUCKET_CONFIG_PUBLISH_HOOK, BucketConfigPublishHook, BucketMetadataMutationGuard, BucketMetadataSys,
-            acquire_bucket_metadata_transaction_lock_for_incarnation, capture_bucket_metadata_incarnation, delete,
+            ObjectLockConfigState, acquire_bucket_metadata_transaction_lock,
-            delete_if_incarnation, delete_under_transaction_lock, get, get_accelerate_config, get_bucket_policy,
+            acquire_bucket_metadata_transaction_lock_for_incarnation, acquire_scanner_bucket_incarnation_fence,
-            get_bucket_policy_raw, get_bucket_targets_config, get_config_from_disk, get_cors_config, get_durability_config,
+            capture_bucket_metadata_incarnation, delete, delete_if_incarnation, delete_under_transaction_lock, get,
-            get_global_bucket_metadata_sys, get_lifecycle_config, get_logging_config, get_notification_config,
+            get_accelerate_config, get_bucket_policy, get_bucket_policy_raw, get_bucket_targets_config, get_config_from_disk,
-            get_object_lock_config, get_object_lock_config_state, get_on_demand_migration_config, get_public_access_block_config,
+            get_cors_config, get_durability_config, get_global_bucket_metadata_sys, get_lifecycle_config, get_logging_config,
-            get_quota_config, get_replication_config, get_request_payment_config, get_sse_config, get_tagging_config,
+            get_notification_config, get_object_lock_config, get_object_lock_config_state, get_on_demand_migration_config,
-            get_versioning_config, get_website_config, init_bucket_metadata_sys, list_bucket_targets, reload_bucket_metadata,
+            get_on_demand_migration_config_in, get_public_access_block_config, get_quota_config, get_replication_config,
-            remove_bucket_metadata, set_bucket_metadata, update, update_bucket_targets_under_transaction_lock,
+            get_request_payment_config, get_sse_config, get_tagging_config, get_versioning_config, get_website_config,
-            update_config_with, update_if_incarnation, update_quota_if_incarnation, update_under_transaction_lock,
+            init_bucket_metadata_sys, list_bucket_targets, reload_bucket_metadata, remove_bucket_metadata, set_bucket_metadata,
+            update, update_bucket_targets_under_transaction_lock, update_config_with, update_if_incarnation,
+            update_quota_if_incarnation, update_under_transaction_lock,
         };
+        #[cfg(feature = "test-util")]
+        pub use crate::bucket::metadata_sys::{ConfigWriteLockProbe, test_support};
     }
 
     pub mod migration {
@@ -248,7 +213,7 @@ pub mod bucket {
     pub mod remote_s3_client {
         pub use crate::bucket::remote_s3_client::{
             PathStyle, RemoteCredentials, RemoteS3ClientError, RemoteS3EndpointSpec, RemoteS3RetryPolicy, build_remote_s3_client,
-            validate_remote_endpoint,
+            build_remote_s3_config, validate_remote_endpoint, validate_target_ca_pem,
         };
     }
 
@@ -479,9 +444,11 @@ pub mod notification {
     #[cfg(any(test, feature = "test-util"))]
     pub use crate::services::notification_sys::rotate_cross_pool_fence_fleet_proof_for_test;
     pub use crate::services::notification_sys::{
-        ClusterTierDailyStats, CrossPoolFenceFleetProofToken, NotificationPeerErr, NotificationSys, ScannerPublicationLeaseGrant,
+        ClusterTierDailyStats, CrossPoolFenceFleetProofToken, LegacyTransitionStateReconcileFleetProofToken, NotificationPeerErr,
-        acquire_cross_pool_fence_fleet_proof, cross_pool_fence_fleet_proof_matches, get_global_notification_sys,
+        NotificationSys, ScannerPublicationLeaseGrant, acquire_cross_pool_fence_fleet_proof,
-        new_global_notification_sys, scanner_peer_transport_error_message_is_retryable, start_remote_version_state_fleet_probe,
+        acquire_legacy_transition_state_reconcile_fleet_proof, cross_pool_fence_fleet_proof_matches, get_global_notification_sys,
+        legacy_transition_state_reconcile_fleet_proof_matches, new_global_notification_sys,
+        scanner_peer_transport_error_message_is_retryable, start_remote_version_state_fleet_probe,
     };
 }
 
@@ -492,9 +459,9 @@ pub mod object {
         ObjectInfo, ObjectLockConfigSnapshot, ObjectMutationHook, ObjectOptions, PutObjReader, QuotaAdmission,
         RangedDecompressReader, ReadEncryptionMaterial, ReadEncryptionMode, ReadEncryptionRequest,
         SCANNER_PUBLICATION_LEASE_FENCE_METADATA_KEY, ScannerPublicationCommitScope, ScannerPublicationCommitStartError,
-        ScannerPublicationCommitState, StreamConsumer, get_object_body_cache_plaintext_len, lookup_get_object_body_cache_hook,
+        ScannerPublicationCommitState, StreamConsumer, WriteCompletion, get_object_body_cache_plaintext_len,
-        register_get_object_body_cache_hook, register_object_mutation_hook, unregister_get_object_body_cache_hook,
+        lookup_get_object_body_cache_hook, register_get_object_body_cache_hook, register_object_mutation_hook,
-        unregister_object_mutation_hook,
+        unregister_get_object_body_cache_hook, unregister_object_mutation_hook,
     };
     pub use crate::store::{
         PrepareSelectObjectSnapshotError, PreparedGetObjectReader, SelectObjectSnapshot, SelectObjectSnapshotReadError,
@@ -558,6 +525,12 @@ pub mod set_disk {
     pub mod test_util {
         pub use crate::bucket::quota::reservation::fail_next_quota_ledger_save_for_test;
         pub use crate::set_disk::{MultipartCommitBarrier, MultipartCommitPause, PutObjectCommitBarrier, PutObjectCommitPause};
+
+        /// Keep a namespace commit pending until the returned owner is dropped.
+        #[must_use]
+        pub fn hold_namespace_commit(store: &crate::store::ECStore) -> impl Send + Sync {
+            store.ctx.begin_namespace_commit()
+        }
     }
 }
 

crates/ecstore/src/bucket/bucket_target_sys.rs

@@ -59,7 +59,7 @@ use rustfs_utils::http::{
     insert_header,
 };
 use serde::{Deserialize, Serialize};
-use std::collections::HashMap;
+use std::collections::{HashMap, HashSet};
 use std::error::Error;
 use std::fmt;
 use std::str::FromStr as _;
@@ -369,6 +369,26 @@ struct SsecPassthroughRecord {
     recorded_at: Instant,
 }
 
+/// What a target write does when the bucket's persisted target set exists but
+/// cannot be decoded.
+///
+/// `docs/architecture/remote-credential-sealing-adr.md` forbids rewriting a
+/// configuration that could not be fully read, because re-serializing a
+/// partial in-memory view is the one mechanism by which a configured target
+/// really disappears. That rule guards against an *unintentional* overwrite,
+/// so an operator who names the hazard keeps a repair path
+/// (rustfs/backlog#2309); everything that does not name it stays refused.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
+pub enum UnreadableTargetsPolicy {
+    /// Refuse the write with [`BucketTargetError::BucketRemoteTargetsUnreadable`].
+    #[default]
+    FailClosed,
+    /// Discard the unreadable set; the target being written becomes the whole
+    /// configuration. Reachable only from an admin request that asked for it
+    /// explicitly, and audited by the caller.
+    Replace,
+}
+
 #[derive(Debug, Default)]
 pub struct BucketTargetSys {
     pub arn_remotes_map: Arc<RwLock<HashMap<String, ArnTarget>>>,
@@ -376,6 +396,11 @@ pub struct BucketTargetSys {
     /// [`SsecPassthroughCapability`]; reset alongside `arn_remotes_map`.
     ssec_passthrough_map: Arc<RwLock<HashMap<String, SsecPassthroughRecord>>>,
     pub targets_map: Arc<RwLock<HashMap<String, Vec<BucketTarget>>>>,
+    /// Buckets whose persisted `bucket-targets.json` exists but cannot be
+    /// decoded (rustfs/backlog#2282). Written under the bucket's update mutex
+    /// alongside `targets_map`, and read before it so an unreadable
+    /// configuration surfaces as a typed error instead of an empty target set.
+    unreadable_targets: Arc<RwLock<HashSet<String>>>,
     pub h_mutex: Arc<RwLock<HashMap<String, EpHealth>>>,
     target_h_mutex: Arc<RwLock<HashMap<String, EpHealth>>>,
     pub hc_client: Arc<HttpClient>,
@@ -419,6 +444,7 @@ impl BucketTargetSys {
             arn_remotes_map: Arc::new(RwLock::new(HashMap::new())),
             ssec_passthrough_map: Arc::new(RwLock::new(HashMap::new())),
             targets_map: Arc::new(RwLock::new(HashMap::new())),
+            unreadable_targets: Arc::new(RwLock::new(HashSet::new())),
             h_mutex: Arc::new(RwLock::new(HashMap::new())),
             target_h_mutex: Arc::new(RwLock::new(HashMap::new())),
             hc_client: Arc::new(build_health_check_client()),
@@ -628,12 +654,19 @@ impl BucketTargetSys {
         health_map.clone()
     }
 
-    pub async fn list_targets(&self, bucket: &str, arn_type: &str) -> Vec<BucketTarget> {
+    /// Targets of one bucket, or of every bucket when `bucket` is empty.
+    ///
+    /// A bucket that simply has no targets yields an empty list; a bucket
+    /// whose persisted configuration cannot be decoded is an error, so an
+    /// admin listing reports the fault instead of an empty list that reads as
+    /// "replication is not configured" (rustfs/backlog#2282).
+    pub async fn list_targets(&self, bucket: &str, arn_type: &str) -> Result<Vec<BucketTarget>, BucketTargetError> {
         let health_stats = self.target_health_stats().await;
         let mut targets = Vec::new();
 
         if !bucket.is_empty() {
-            if let Ok(bucket_targets) = self.list_bucket_targets(bucket).await {
+            match self.list_bucket_targets(bucket).await {
+                Ok(bucket_targets) => {
                     for mut target in bucket_targets.targets {
                         if arn_type.is_empty() || target.target_type.to_string() == arn_type {
                             if let Some(health) = health_stats.get(&target.arn) {
@@ -651,7 +684,10 @@ impl BucketTargetSys {
                         }
                     }
                 }
-            return targets;
+                Err(BucketTargetError::BucketRemoteTargetNotFound { .. }) => {}
+                Err(err) => return Err(err),
+            }
+            return Ok(targets);
         }
 
         let targets_map = self.targets_map.read().await;
@@ -674,10 +710,16 @@ impl BucketTargetSys {
             }
         }
 
-        targets
+        Ok(targets)
     }
 
     pub async fn list_bucket_targets(&self, bucket: &str) -> Result<BucketTargets, BucketTargetError> {
+        if self.unreadable_targets.read().await.contains(bucket) {
+            return Err(BucketTargetError::BucketRemoteTargetsUnreadable {
+                bucket: bucket.to_string(),
+            });
+        }
+
         let targets_map = self.targets_map.read().await;
         if let Some(targets) = targets_map.get(bucket) {
             Ok(BucketTargets {
@@ -690,13 +732,30 @@ impl BucketTargetSys {
         }
     }
 
+    /// Record that this bucket's persisted targets configuration exists but
+    /// cannot be decoded (rustfs/backlog#2282).
+    ///
+    /// Any snapshot published from an earlier readable load is deliberately
+    /// left in place: withdrawing it would produce exactly the silent "no
+    /// targets configured" state this marker exists to prevent. The marker is
+    /// cleared by the next successful publish, which is what makes a repaired
+    /// configuration take effect without a restart.
+    pub async fn mark_targets_unreadable(&self, bucket: &str) {
+        let update_mutex = self.target_update_mutex(bucket).await;
+        let _update_guard = update_mutex.lock().await;
+
+        self.unreadable_targets.write().await.insert(bucket.to_string());
+    }
+
     pub async fn delete(&self, bucket: &str) {
         let update_mutex = self.target_update_mutex(bucket).await;
         let _update_guard = update_mutex.lock().await;
 
-        // Lock order: targets_map, then arn_remotes_map, then target_h_mutex,
+        // Lock order: unreadable_targets, then targets_map, then
-        // then ssec_passthrough_map (always last; also taken standalone by the
+        // arn_remotes_map, then target_h_mutex, then ssec_passthrough_map
-        // capability accessors).
+        // (always last; also taken standalone by the capability accessors).
+        self.unreadable_targets.write().await.remove(bucket);
+
         let mut targets_map = self.targets_map.write().await;
         let mut arn_remotes_map = self.arn_remotes_map.write().await;
         let mut health_map = self.target_h_mutex.write().await;
@@ -752,20 +811,45 @@ impl BucketTargetSys {
         bucket: &str,
         target: &BucketTarget,
         update: bool,
+        unreadable_policy: UnreadableTargetsPolicy,
     ) -> Result<BucketTargets, BucketTargetError> {
         self.validate_target(bucket, target).await?;
 
-        let mut bucket_targets = match self.list_bucket_targets(bucket).await {
+        let mut bucket_targets = self.targets_base_for_write(bucket, unreadable_policy).await?;
-            Ok(targets) => targets,
-            Err(BucketTargetError::BucketRemoteTargetNotFound { .. }) => BucketTargets::default(),
-            Err(err) => return Err(err),
-        };
 
         Self::upsert_target_entry(&mut bucket_targets.targets, target, update)?;
 
         Ok(bucket_targets)
     }
 
+    /// The persisted target set a write merges into.
+    ///
+    /// An absent configuration starts from the empty set. An unreadable one is
+    /// refused, because re-serializing a partial view of a set this node could
+    /// not decode is how a configured target disappears for good — unless the
+    /// caller carries the operator's explicit
+    /// [`UnreadableTargetsPolicy::Replace`] opt-in, which discards it
+    /// deliberately (rustfs/backlog#2309).
+    async fn targets_base_for_write(
+        &self,
+        bucket: &str,
+        unreadable_policy: UnreadableTargetsPolicy,
+    ) -> Result<BucketTargets, BucketTargetError> {
+        match self.list_bucket_targets(bucket).await {
+            Ok(targets) => Ok(targets),
+            Err(BucketTargetError::BucketRemoteTargetNotFound { .. }) => Ok(BucketTargets::default()),
+            // The opt-in discards only a set this node genuinely cannot read.
+            // A readable set still merges through the arm above, so the policy
+            // can never drop a target that was visible here.
+            Err(BucketTargetError::BucketRemoteTargetsUnreadable { .. })
+                if unreadable_policy == UnreadableTargetsPolicy::Replace =>
+            {
+                Ok(BucketTargets::default())
+            }
+            Err(err) => Err(err),
+        }
+    }
+
     pub async fn validate_target(&self, bucket: &str, target: &BucketTarget) -> Result<(), BucketTargetError> {
         if !target.target_type.is_valid() {
             return Err(BucketTargetError::BucketRemoteArnTypeInvalid {
@@ -1093,6 +1177,11 @@ impl BucketTargetSys {
     /// Keeping persisted-config reads under the same mutex prevents a stale
     /// reload from overwriting a concurrent credential rotation.
     async fn update_all_targets_locked(&self, bucket: &str, targets: Option<&BucketTargets>) {
+        // Reaching here means the persisted configuration decoded, so the
+        // unreadable marker (if any) is stale. Cleared before the maps below
+        // so `unreadable_targets` stays the outermost of this module's locks.
+        self.unreadable_targets.write().await.remove(bucket);
+
         let mut clients = Vec::new();
         if let Some(new_targets) = targets {
             for target in &new_targets.targets {
@@ -1100,9 +1189,9 @@ impl BucketTargetSys {
             }
         }
 
-        // Lock order: targets_map, then arn_remotes_map, then target_h_mutex,
+        // Lock order: unreadable_targets (above), then targets_map, then
-        // then ssec_passthrough_map (always last; also taken standalone by the
+        // arn_remotes_map, then target_h_mutex, then ssec_passthrough_map
-        // capability accessors).
+        // (always last; also taken standalone by the capability accessors).
         let mut targets_map = self.targets_map.write().await;
         let mut arn_remotes_map = self.arn_remotes_map.write().await;
         let mut health_map = self.target_h_mutex.write().await;
@@ -1161,6 +1250,11 @@ impl BucketTargetSys {
     }
 
     pub async fn set(&self, bucket: &str, meta: &BucketMetadata) {
+        if meta.bucket_targets_unreadable() {
+            self.mark_targets_unreadable(bucket).await;
+            return;
+        }
+
         let Some(config) = &meta.bucket_target_config else {
             return;
         };
@@ -2276,6 +2370,13 @@ pub enum BucketTargetError {
     BucketRemoteTargetNotFound {
         bucket: String,
     },
+    /// The bucket's persisted targets configuration exists but cannot be
+    /// decoded. Distinct from `BucketRemoteTargetNotFound`, which means the
+    /// bucket genuinely has no targets: callers must not degrade this one to
+    /// an empty target set (rustfs/backlog#2282).
+    BucketRemoteTargetsUnreadable {
+        bucket: String,
+    },
     BucketRemoteArnTypeInvalid {
         bucket: String,
     },
@@ -2309,6 +2410,9 @@ impl fmt::Display for BucketTargetError {
             BucketTargetError::BucketRemoteTargetNotFound { bucket } => {
                 write!(f, "Remote target not found for bucket: {bucket}")
             }
+            BucketTargetError::BucketRemoteTargetsUnreadable { bucket } => {
+                write!(f, "Persisted replication target configuration is unreadable for bucket: {bucket}")
+            }
             BucketTargetError::BucketRemoteArnTypeInvalid { bucket } => {
                 write!(f, "Invalid ARN type for bucket: {bucket}")
             }
@@ -3256,7 +3360,7 @@ mod tests {
             }],
         );
 
-        let targets = sys.list_targets("", "").await;
+        let targets = sys.list_targets("", "").await.expect("listing every bucket's targets");
 
         assert_eq!(targets.len(), 1);
         assert!(!targets[0].online);
@@ -4254,4 +4358,75 @@ mod tests {
         let window = LastMinuteLatency::new();
         assert_eq!(window.get_total().avg, Duration::from_secs(0));
     }
+
+    fn repair_target(bucket: &str, id: &str) -> BucketTarget {
+        BucketTarget {
+            source_bucket: bucket.to_string(),
+            endpoint: "remote.example.com".to_string(),
+            target_bucket: "remote".to_string(),
+            arn: format!("arn:rustfs:replication:us-east-1:{bucket}:{id}"),
+            target_type: BucketTargetType::ReplicationService,
+            region: "us-east-1".to_string(),
+            ..Default::default()
+        }
+    }
+
+    /// rustfs/backlog#2309: after rustfs/rustfs#7172 an undecodable
+    /// `bucket-targets.json` left the bucket with no API repair path at all.
+    /// The refusal is the default and stays the default; the operator's
+    /// explicit opt-in is the only thing that discards the set, and it starts
+    /// the replacement from empty rather than from a partial view of bytes
+    /// this node never decoded.
+    #[tokio::test]
+    async fn an_unreadable_target_set_is_replaced_only_with_the_explicit_opt_in() {
+        let sys = BucketTargetSys::default();
+        let bucket = "targets-repair-opt-in";
+        sys.mark_targets_unreadable(bucket).await;
+
+        assert!(
+            matches!(
+                sys.targets_base_for_write(bucket, UnreadableTargetsPolicy::FailClosed).await,
+                Err(BucketTargetError::BucketRemoteTargetsUnreadable { .. })
+            ),
+            "without the opt-in an unreadable target set must still refuse the write"
+        );
+        assert_eq!(
+            UnreadableTargetsPolicy::default(),
+            UnreadableTargetsPolicy::FailClosed,
+            "a caller that says nothing must get the refusal"
+        );
+
+        let base = sys
+            .targets_base_for_write(bucket, UnreadableTargetsPolicy::Replace)
+            .await
+            .expect("the explicit opt-in must let an operator replace an unreadable set");
+        assert!(
+            base.is_empty(),
+            "the replacement must start from an empty set, never from a partial decode"
+        );
+    }
+
+    /// The opt-in is not a wipe switch. On a set this node can read, both
+    /// policies take the same merge path, so a stray `replace-unreadable=true`
+    /// cannot drop a visible target — which is what makes the flag safe to
+    /// repeat in an operator's repair script.
+    #[tokio::test]
+    async fn the_opt_in_never_discards_a_readable_target_set() {
+        let sys = BucketTargetSys::default();
+        let bucket = "targets-repair-readable";
+        let existing = repair_target(bucket, "keep");
+        sys.targets_map
+            .write()
+            .await
+            .insert(bucket.to_string(), vec![existing.clone()]);
+
+        for policy in [UnreadableTargetsPolicy::FailClosed, UnreadableTargetsPolicy::Replace] {
+            let base = sys
+                .targets_base_for_write(bucket, policy)
+                .await
+                .expect("a readable target set must be readable under either policy");
+            assert_eq!(base.targets.len(), 1, "{policy:?} must keep the persisted target");
+            assert_eq!(base.targets[0].arn, existing.arn, "{policy:?} must not rewrite the persisted target");
+        }
+    }
 }

crates/ecstore/src/bucket/lifecycle/core.rs

@@ -15,9 +15,9 @@
 use crate::object_api::ObjectInfo;
 
 pub use rustfs_lifecycle::{
-    Event, ExpirationOptions, IlmAction, Lifecycle, LifecycleCalculate, ObjectOpts, RuleValidate, TRANSITION_COMPLETE,
+    Event, ExpirationOptions, IlmAction, LIFECYCLE_MALFORMED_XML_ERROR_KIND, Lifecycle, LifecycleCalculate, ObjectOpts,
-    TRANSITION_PENDING, TransitionOptions, abort_incomplete_multipart_upload_due, expected_expiry_time,
+    RuleValidate, TRANSITION_COMPLETE, TRANSITION_PENDING, TransitionOptions, abort_incomplete_multipart_upload_due,
-    expiration_action_has_valid_target,
+    expected_expiry_time, expiration_action_has_valid_target,
 };
 
 pub fn object_opts_from_object_info(oi: &ObjectInfo) -> ObjectOpts {

crates/ecstore/src/bucket/lifecycle/durable_namespace.rs

@@ -22,9 +22,10 @@ use super::{
     bucket_lifecycle_ops::{
         ManualTransitionQueueSnapshot, ManualTransitionRunReport, decode_manual_transition_continuation_token,
     },
-    manual_transition_job, tier_delete_journal, transition_transaction,
+    manual_transition_job, recovery_control, tier_delete_journal, transition_transaction,
 };
 use crate::error::{Error, Result};
+use crate::services::tier::tier_probe_intent;
 
 pub(crate) const ILM_META_PREFIX: &str = "ilm";
 const ILM_META_OBJECT_PREFIX: &str = "ilm/";
@@ -35,10 +36,12 @@ pub(crate) enum DurableIlmRecordKind {
     TierDeleteJournal,
     TierDeleteDispatchManifest,
     TransitionTransaction,
+    TierProbeIntent,
     ManualTransitionJob,
     ManualTransitionScope,
     ManualTransitionTask,
     ManualTransitionWorkerResult,
+    RecoveryControl,
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
@@ -73,6 +76,12 @@ pub(crate) const TRANSITION_TRANSACTION_NAMESPACE: DurableIlmNamespace = Durable
     max_record_size: transition_transaction::MAX_TRANSITION_TRANSACTION_SIZE,
     kind: DurableIlmRecordKind::TransitionTransaction,
 };
+pub(crate) const TIER_PROBE_INTENT_NAMESPACE: DurableIlmNamespace = DurableIlmNamespace {
+    name: "tier-probe-intent",
+    prefix: tier_probe_intent::TIER_PROBE_INTENT_RECORD_PREFIX,
+    max_record_size: tier_probe_intent::MAX_TIER_PROBE_INTENT_SIZE,
+    kind: DurableIlmRecordKind::TierProbeIntent,
+};
 pub(crate) const MANUAL_TRANSITION_JOB_NAMESPACE: DurableIlmNamespace = DurableIlmNamespace {
     name: "manual-transition-job",
     prefix: "ilm/manual-transition/jobs",
@@ -97,16 +106,24 @@ pub(crate) const MANUAL_TRANSITION_WORKER_RESULT_NAMESPACE: DurableIlmNamespace
     max_record_size: manual_transition_job::MAX_MANUAL_TRANSITION_WORKER_RESULT_RECORD_SIZE,
     kind: DurableIlmRecordKind::ManualTransitionWorkerResult,
 };
+pub(crate) const RECOVERY_CONTROL_NAMESPACE: DurableIlmNamespace = DurableIlmNamespace {
+    name: "recovery-control",
+    prefix: recovery_control::ILM_RECOVERY_CONTROL_PREFIX,
+    max_record_size: recovery_control::MAX_ILM_RECOVERY_CONTROL_SIZE,
+    kind: DurableIlmRecordKind::RecoveryControl,
+};
 
-pub(crate) const DURABLE_ILM_NAMESPACES: [DurableIlmNamespace; 8] = [
+pub(crate) const DURABLE_ILM_NAMESPACES: [DurableIlmNamespace; 10] = [
     TIER_DELETE_JOURNAL_NAMESPACE,
     TIER_DELETE_JOURNAL_V6_NAMESPACE,
     TIER_DELETE_DISPATCH_MANIFEST_NAMESPACE,
     TRANSITION_TRANSACTION_NAMESPACE,
+    TIER_PROBE_INTENT_NAMESPACE,
     MANUAL_TRANSITION_JOB_NAMESPACE,
     MANUAL_TRANSITION_SCOPE_NAMESPACE,
     MANUAL_TRANSITION_TASK_NAMESPACE,
     MANUAL_TRANSITION_WORKER_RESULT_NAMESPACE,
+    RECOVERY_CONTROL_NAMESPACE,
 ];
 
 #[derive(Debug, Clone, PartialEq, Eq)]
@@ -200,6 +217,15 @@ pub(crate) enum DurableIlmRecordCheckpoint {
         revision: u64,
         state: transition_transaction::TransitionTransactionState,
     },
+    TierProbeIntent {
+        content_sha256: String,
+        identity_sha256: String,
+        remote_version_sha256: String,
+        remote_version_known: bool,
+        owner_fence_sha256: String,
+        revision: u64,
+        state: tier_probe_intent::TierProbeIntentState,
+    },
     ManualTransitionJob {
         content_sha256: String,
         identity_sha256: String,
@@ -223,6 +249,18 @@ pub(crate) enum DurableIlmRecordCheckpoint {
     ManualTransitionWorkerResult {
         content_sha256: String,
     },
+    RecoveryControl {
+        content_sha256: String,
+        identity_sha256: String,
+        source_generation_sha256: String,
+        first_seen_at_unix_nanos: i64,
+        revision: u64,
+        classification: recovery_control::IlmRecoveryClassification,
+        attempt_count: u64,
+        consecutive_failure_count: u32,
+        #[serde(default, skip_serializing_if = "Option::is_none")]
+        owner_fence_sha256: Option<String>,
+    },
 }
 
 impl DurableIlmRecordCheckpoint {
@@ -232,10 +270,12 @@ impl DurableIlmRecordCheckpoint {
             | Self::TierDeleteDispatchManifest { content_sha256, .. }
             | Self::TierDeleteDispatchParent { content_sha256, .. }
             | Self::TransitionTransaction { content_sha256, .. }
+            | Self::TierProbeIntent { content_sha256, .. }
             | Self::ManualTransitionJob { content_sha256, .. }
             | Self::ManualTransitionScope { content_sha256, .. }
             | Self::ManualTransitionTask { content_sha256 }
-            | Self::ManualTransitionWorkerResult { content_sha256 } => content_sha256,
+            | Self::ManualTransitionWorkerResult { content_sha256 }
+            | Self::RecoveryControl { content_sha256, .. } => content_sha256,
         }
     }
 
@@ -421,6 +461,32 @@ impl DurableIlmRecordCheckpoint {
                         .is_some_and(|expected_revision| *next_revision == expected_revision)
                     && (!previous_remote_version_known || previous_remote_version == next_remote_version)
             }
+            (
+                Self::TierProbeIntent {
+                    identity_sha256: previous_identity,
+                    remote_version_sha256: previous_remote_version,
+                    remote_version_known: previous_remote_version_known,
+                    owner_fence_sha256: previous_owner_fence,
+                    revision: previous_revision,
+                    state: previous_state,
+                    ..
+                },
+                Self::TierProbeIntent {
+                    identity_sha256: next_identity,
+                    remote_version_sha256: next_remote_version,
+                    owner_fence_sha256: next_owner_fence,
+                    revision: next_revision,
+                    state: next_state,
+                    ..
+                },
+            ) => {
+                previous_identity == next_identity
+                    && previous_owner_fence == next_owner_fence
+                    && next_revision
+                        .checked_sub(*previous_revision)
+                        .is_some_and(|distance| distance == 1 && tier_probe_state_reaches(*previous_state, *next_state, distance))
+                    && (!previous_remote_version_known || previous_remote_version == next_remote_version)
+            }
             (
                 Self::ManualTransitionJob {
                     content_sha256: previous_content,
@@ -483,6 +549,51 @@ impl DurableIlmRecordCheckpoint {
                     ..
                 },
             ) => previous_identity == next_identity && next_updated_at > previous_updated_at,
+            (
+                Self::RecoveryControl {
+                    identity_sha256: previous_identity,
+                    source_generation_sha256: previous_generation,
+                    first_seen_at_unix_nanos: previous_first_seen,
+                    revision: previous_revision,
+                    classification: previous_classification,
+                    attempt_count: previous_attempts,
+                    consecutive_failure_count: previous_failures,
+                    owner_fence_sha256: previous_owner,
+                    ..
+                },
+                Self::RecoveryControl {
+                    identity_sha256: next_identity,
+                    source_generation_sha256: next_generation,
+                    first_seen_at_unix_nanos: next_first_seen,
+                    revision: next_revision,
+                    classification: next_classification,
+                    attempt_count: next_attempts,
+                    consecutive_failure_count: next_failures,
+                    owner_fence_sha256: next_owner,
+                    ..
+                },
+            ) => {
+                let adjacent = previous_identity == next_identity
+                    && previous_first_seen == next_first_seen
+                    && previous_revision.checked_add(1) == Some(*next_revision);
+                let claim = next_owner.is_some()
+                    && *previous_classification == recovery_control::IlmRecoveryClassification::Retrying
+                    && *next_classification == recovery_control::IlmRecoveryClassification::Retrying
+                    && previous_attempts.checked_add(1) == Some(*next_attempts)
+                    && previous_failures == next_failures;
+                let source_refresh = previous_owner.is_some()
+                    && previous_owner == next_owner
+                    && *previous_classification == recovery_control::IlmRecoveryClassification::Retrying
+                    && *next_classification == recovery_control::IlmRecoveryClassification::Retrying
+                    && previous_attempts == next_attempts
+                    && previous_failures == next_failures
+                    && previous_generation != next_generation;
+                let completion = previous_owner.is_some()
+                    && next_owner.is_none()
+                    && previous_generation == next_generation
+                    && previous_attempts == next_attempts;
+                adjacent && (claim || source_refresh || completion)
+            }
             _ => false,
         };
 
@@ -500,6 +611,22 @@ impl DurableIlmRecordCheckpoint {
     /// after the exact terminal ETag and terminal receipt were committed, to
     /// purge older object versions exposed by that deletion.
     pub(crate) fn is_predecessor_of_terminal(&self, terminal: &Self) -> bool {
+        if let Self::TierProbeIntent { state, .. } = terminal
+            && !matches!(
+                state,
+                tier_probe_intent::TierProbeIntentState::AbortedNoRemote | tier_probe_intent::TierProbeIntentState::Completed
+            )
+        {
+            return false;
+        }
+        if let Self::RecoveryControl { classification, .. } = terminal
+            && !matches!(
+                classification,
+                recovery_control::IlmRecoveryClassification::Terminal | recovery_control::IlmRecoveryClassification::Abandoned
+            )
+        {
+            return false;
+        }
         if self == terminal || self.validate_successor(terminal).is_ok() {
             return true;
         }
@@ -568,6 +695,63 @@ impl DurableIlmRecordCheckpoint {
                         }
                     })
             }
+            (
+                Self::TierProbeIntent {
+                    identity_sha256: previous_identity,
+                    remote_version_sha256: previous_remote_version,
+                    remote_version_known: previous_remote_version_known,
+                    owner_fence_sha256: previous_owner_fence,
+                    revision: previous_revision,
+                    state: previous_state,
+                    ..
+                },
+                Self::TierProbeIntent {
+                    identity_sha256: terminal_identity,
+                    remote_version_sha256: terminal_remote_version,
+                    owner_fence_sha256: terminal_owner_fence,
+                    revision: terminal_revision,
+                    state: terminal_state,
+                    ..
+                },
+            ) => {
+                previous_identity == terminal_identity
+                    && previous_owner_fence == terminal_owner_fence
+                    && matches!(
+                        terminal_state,
+                        tier_probe_intent::TierProbeIntentState::AbortedNoRemote
+                            | tier_probe_intent::TierProbeIntentState::Completed
+                    )
+                    && terminal_revision
+                        .checked_sub(*previous_revision)
+                        .is_some_and(|distance| tier_probe_state_reaches(*previous_state, *terminal_state, distance))
+                    && (!previous_remote_version_known || previous_remote_version == terminal_remote_version)
+            }
+            (
+                Self::RecoveryControl {
+                    identity_sha256: previous_identity,
+                    source_generation_sha256: previous_generation,
+                    first_seen_at_unix_nanos: previous_first_seen,
+                    revision: previous_revision,
+                    attempt_count: previous_attempts,
+                    ..
+                },
+                Self::RecoveryControl {
+                    identity_sha256: terminal_identity,
+                    source_generation_sha256: terminal_generation,
+                    first_seen_at_unix_nanos: terminal_first_seen,
+                    revision: terminal_revision,
+                    attempt_count: terminal_attempts,
+                    classification:
+                        recovery_control::IlmRecoveryClassification::Terminal | recovery_control::IlmRecoveryClassification::Abandoned,
+                    ..
+                },
+            ) => {
+                previous_identity == terminal_identity
+                    && (previous_generation == terminal_generation || terminal_attempts > previous_attempts)
+                    && previous_first_seen == terminal_first_seen
+                    && terminal_revision > previous_revision
+                    && terminal_attempts >= previous_attempts
+            }
             _ => false,
         }
     }
@@ -606,6 +790,23 @@ fn transition_state_distance(
     }
 }
 
+fn tier_probe_state_reaches(
+    from: tier_probe_intent::TierProbeIntentState,
+    to: tier_probe_intent::TierProbeIntentState,
+    revision_distance: u64,
+) -> bool {
+    use tier_probe_intent::TierProbeIntentState::{AbortedNoRemote, CleanupPending, Completed, UploadOutcomeUnknown, Uploaded};
+
+    match (from, to) {
+        (UploadOutcomeUnknown, Uploaded | CleanupPending | AbortedNoRemote) => revision_distance == 1,
+        (UploadOutcomeUnknown, Completed) => matches!(revision_distance, 2 | 3),
+        (Uploaded, CleanupPending) => revision_distance == 1,
+        (Uploaded, Completed) => revision_distance == 2,
+        (CleanupPending, Completed) => revision_distance == 1,
+        _ => false,
+    }
+}
+
 fn manual_job_state_reaches(
     from: manual_transition_job::ManualTransitionJobState,
     to: manual_transition_job::ManualTransitionJobState,
@@ -1082,6 +1283,71 @@ pub(crate) fn validate_durable_ilm_record(path: &str, data: &[u8]) -> Result<Val
                 },
             )
         }
+        DurableIlmRecordKind::TierProbeIntent => {
+            let probe_id = tier_probe_intent::tier_probe_intent_id_from_record_object_name(path)
+                .map_err(|err| Error::other(err.to_string()))?;
+            let intent =
+                tier_probe_intent::TierProbeIntent::decode(probe_id, data).map_err(|err| Error::other(err.to_string()))?;
+            let canonical =
+                tier_probe_intent::tier_probe_intent_record_object_name(probe_id).map_err(|err| Error::other(err.to_string()))?;
+            if canonical != path {
+                return Err(Error::other("tier probe intent path is not canonical"));
+            }
+            let identity_sha256 = checkpoint_hash(&(
+                intent.probe_id,
+                &intent.operation,
+                &intent.tier_name,
+                intent.destination_id,
+                &intent.probe_object,
+                &intent.creator_id,
+                intent.creator_epoch,
+                intent.created_at_unix_nanos,
+            ))?;
+            let remote_version_sha256 = checkpoint_hash(&intent.remote_version)?;
+            let owner_fence_sha256 = checkpoint_hash(&intent.owner)?;
+            (
+                "probe_id",
+                probe_id.to_string(),
+                DurableIlmRecordCheckpoint::TierProbeIntent {
+                    content_sha256,
+                    identity_sha256,
+                    remote_version_sha256,
+                    remote_version_known: !intent.remote_version.is_unknown(),
+                    owner_fence_sha256,
+                    revision: intent.revision,
+                    state: intent.state,
+                },
+            )
+        }
+        DurableIlmRecordKind::RecoveryControl => {
+            let (protocol, control_id) = recovery_control::recovery_control_id_from_record_object_name(path)
+                .map_err(|err| Error::other(err.to_string()))?;
+            let control =
+                recovery_control::IlmRecoveryControl::decode(&control_id, data).map_err(|err| Error::other(err.to_string()))?;
+            let canonical = recovery_control::recovery_control_record_object_name(protocol, &control_id)
+                .map_err(|err| Error::other(err.to_string()))?;
+            if canonical != path || control.identity.protocol != protocol {
+                return Err(Error::other("ILM recovery control path is not canonical"));
+            }
+            let identity_sha256 = checkpoint_hash(&control.identity)?;
+            let source_generation_sha256 = checkpoint_hash(&control.observed_source_generation)?;
+            let owner_fence_sha256 = control.owner.as_ref().map(checkpoint_hash).transpose()?;
+            (
+                "control_id",
+                control_id,
+                DurableIlmRecordCheckpoint::RecoveryControl {
+                    content_sha256,
+                    identity_sha256,
+                    source_generation_sha256,
+                    first_seen_at_unix_nanos: control.first_seen_at_unix_nanos,
+                    revision: control.revision,
+                    classification: control.classification,
+                    attempt_count: control.attempt_count,
+                    consecutive_failure_count: control.consecutive_failure_count,
+                    owner_fence_sha256,
+                },
+            )
+        }
         DurableIlmRecordKind::ManualTransitionJob => {
             let job_id = manual_transition_job::manual_transition_job_id_from_record_object_name(path)
                 .map_err(|err| Error::other(err.to_string()))?;
@@ -1237,6 +1503,190 @@ mod tests {
         }
     }
 
+    fn tier_probe_intent_fixture() -> tier_probe_intent::TierProbeIntent {
+        let probe_id = Uuid::parse_str("36e2220e-9ad2-495b-b3bc-c4d2caf70a31").expect("fixture uuid should parse");
+        tier_probe_intent::TierProbeIntent {
+            probe_id,
+            revision: 1,
+            state: tier_probe_intent::TierProbeIntentState::UploadOutcomeUnknown,
+            operation: tier_probe_intent::TierProbeOperationIdentity::Verify {
+                config_etag: "config-etag".to_string(),
+                backend_identity: [1; 32],
+            },
+            tier_name: "COLD-A".to_string(),
+            destination_id: [1; 32],
+            probe_object: tier_probe_intent::tier_probe_object_name(probe_id),
+            creator_id: "node-a".to_string(),
+            creator_epoch: Uuid::parse_str("76746062-c05a-40b7-9e38-d2722d7e0332").expect("fixture creator epoch should parse"),
+            created_at_unix_nanos: 1_780_000_000_000_000_000,
+            owner: tier_probe_intent::TierProbeOwnerFence {
+                owner_id: "node-a".to_string(),
+                owner_epoch: Uuid::parse_str("76746062-c05a-40b7-9e38-d2722d7e0332").expect("fixture owner epoch should parse"),
+                not_after_unix_nanos: 1_780_000_900_000_000_000,
+            },
+            remote_version: tier_probe_intent::TierProbeRemoteVersion::default(),
+        }
+    }
+
+    fn tier_probe_checkpoint(intent: &tier_probe_intent::TierProbeIntent) -> DurableIlmRecordCheckpoint {
+        let path =
+            tier_probe_intent::tier_probe_intent_record_object_name(intent.probe_id).expect("tier probe path should build");
+        let encoded = intent.encode().expect("tier probe intent should encode");
+        let namespace = classify_durable_ilm_record(&path)
+            .expect("tier probe namespace should classify")
+            .expect("tier probe intent should be durable");
+        assert_eq!(namespace, &TIER_PROBE_INTENT_NAMESPACE);
+        validate_durable_ilm_record(&path, &encoded)
+            .expect("tier probe intent should validate")
+            .checkpoint
+    }
+
+    fn recovery_control_fixture() -> recovery_control::IlmRecoveryControl {
+        let source_path = "ilm/transition-transactions/records/12/34/1234567890abcdef1234567890abcdef.json";
+        let generation = recovery_control::IlmRecoverySourceGeneration::new(
+            transition_transaction::TRANSITION_TRANSACTION_SCHEMA,
+            "source-etag",
+            "a".repeat(64),
+            vec![recovery_control::IlmRecoverySourceCopy {
+                authority: "pool-0/set-0".to_string(),
+                canonical_path: source_path.to_string(),
+                etag: "source-etag".to_string(),
+                encoded_len: 128,
+                content_sha256: "a".repeat(64),
+            }],
+        )
+        .expect("source generation should build");
+        recovery_control::IlmRecoveryControl::new(
+            recovery_control::IlmRecoveryControlIdentity {
+                protocol: recovery_control::IlmRecoveryProtocol::TransitionTransaction,
+                canonical_source_path: source_path.to_string(),
+                stable_operation_identity: "12345678-90ab-cdef-1234-567890abcdef".to_string(),
+                record_class: "transition_transaction_v1".to_string(),
+            },
+            generation,
+            recovery_control::IlmRecoveryClassification::Retrying,
+            1_000_000_000,
+            recovery_control::IlmRecoveryErrorCode::None,
+        )
+        .expect("recovery control should build")
+    }
+
+    fn recovery_control_checkpoint(control: &recovery_control::IlmRecoveryControl) -> DurableIlmRecordCheckpoint {
+        let control_id = control.identity.source_operation_digest().expect("control id should derive");
+        let path = recovery_control::recovery_control_record_object_name(control.identity.protocol, &control_id)
+            .expect("control path should build");
+        let encoded = control.encode().expect("control should encode");
+        let namespace = classify_durable_ilm_record(&path)
+            .expect("recovery control namespace should classify")
+            .expect("recovery control should be durable");
+        assert_eq!(namespace, &RECOVERY_CONTROL_NAMESPACE);
+        validate_durable_ilm_record(&path, &encoded)
+            .expect("recovery control should validate")
+            .checkpoint
+    }
+
+    #[test]
+    fn recovery_control_checkpoint_tracks_claim_retry_and_terminal_generations() {
+        let initial_control = recovery_control_fixture();
+        let initial = recovery_control_checkpoint(&initial_control);
+
+        let mut claimed_control = initial_control;
+        let mut advanced_generation = claimed_control.observed_source_generation.clone();
+        advanced_generation.source_schema = "rustfs-transition-transaction-v2".to_string();
+        claimed_control
+            .claim_for_source_generation("node-a", Uuid::new_v4(), 2_000_000_000, 300_000_000_000, advanced_generation)
+            .expect("control should claim");
+        let claimed = recovery_control_checkpoint(&claimed_control);
+        initial.validate_successor(&claimed).expect("claim should advance receipt");
+
+        let mut retry_control = claimed_control;
+        retry_control
+            .record_retryable_failure(3_000_000_000, recovery_control::IlmRecoveryErrorCode::BackendTimeout)
+            .expect("retry should persist");
+        let retry = recovery_control_checkpoint(&retry_control);
+        claimed.validate_successor(&retry).expect("retry should advance receipt");
+
+        let ready_at = retry_control
+            .next_attempt_at_unix_nanos
+            .expect("retry deadline should persist");
+        let mut terminal_control = retry_control;
+        terminal_control
+            .claim("node-b", Uuid::new_v4(), ready_at, 300_000_000_000)
+            .expect("retry should claim");
+        let reclaimed = recovery_control_checkpoint(&terminal_control);
+        retry.validate_successor(&reclaimed).expect("reclaim should advance receipt");
+        terminal_control
+            .finish_attempt(
+                recovery_control::IlmRecoveryClassification::Terminal,
+                recovery_control::IlmRecoveryErrorCode::None,
+            )
+            .expect("control should terminate");
+        let terminal = recovery_control_checkpoint(&terminal_control);
+        reclaimed
+            .validate_successor(&terminal)
+            .expect("terminal state should advance receipt");
+        assert!(initial.is_predecessor_of_terminal(&terminal));
+        assert!(!initial.is_predecessor_of_terminal(&retry));
+    }
+
+    #[test]
+    fn tier_probe_intent_checkpoint_tracks_exact_monotonic_generations() {
+        let initial_intent = tier_probe_intent_fixture();
+        let initial = tier_probe_checkpoint(&initial_intent);
+
+        let mut uploaded_intent = initial_intent;
+        uploaded_intent
+            .advance(
+                tier_probe_intent::TierProbeIntentState::Uploaded,
+                tier_probe_intent::TierProbeRemoteVersion::versioned("opaque-v1"),
+            )
+            .expect("uploaded state should advance");
+        let uploaded = tier_probe_checkpoint(&uploaded_intent);
+        initial
+            .validate_successor(&uploaded)
+            .expect("durable receipt may adopt the exact uploaded generation");
+
+        let mut cleanup_intent = uploaded_intent.clone();
+        cleanup_intent
+            .advance(
+                tier_probe_intent::TierProbeIntentState::CleanupPending,
+                uploaded_intent.remote_version.clone(),
+            )
+            .expect("cleanup state should advance");
+        let cleanup = tier_probe_checkpoint(&cleanup_intent);
+        uploaded
+            .validate_successor(&cleanup)
+            .expect("durable receipt may adopt the exact cleanup generation");
+
+        let mut completed_intent = cleanup_intent.clone();
+        completed_intent
+            .advance(tier_probe_intent::TierProbeIntentState::Completed, cleanup_intent.remote_version.clone())
+            .expect("completed state should advance");
+        let completed = tier_probe_checkpoint(&completed_intent);
+        cleanup
+            .validate_successor(&completed)
+            .expect("durable receipt may adopt the exact terminal generation");
+        assert!(
+            initial.is_predecessor_of_terminal(&completed),
+            "terminal cleanup must recognize the full acknowledged-PUT path"
+        );
+        assert!(
+            initial.validate_successor(&completed).is_err(),
+            "ordinary receipt advancement must not skip intermediate generations"
+        );
+        assert!(
+            !initial.is_predecessor_of_terminal(&uploaded),
+            "a nonterminal generation must not be accepted as terminal proof"
+        );
+
+        let mut rebound = uploaded_intent;
+        rebound.owner.owner_epoch = Uuid::new_v4();
+        assert!(
+            rebound.encode().is_err(),
+            "dormant v1 must reject owner takeover before producing a checkpoint"
+        );
+    }
+
     #[test]
     fn tier_delete_dispatch_manifest_namespace_validates_monotonic_branches() {
         use tier_delete_journal::TierDeleteDispatchManifestState::{Aborted, Aborting, Completed, DispatchAuthorized, Preparing};

crates/ecstore/src/bucket/lifecycle/manual_transition_job.rs

@@ -1170,6 +1170,7 @@ pub async fn save_manual_transition_job_record_if_current(
         data.clone(),
         &ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             http_preconditions: Some(HTTPPreconditions {
                 if_match: Some(current_etag.to_string()),
                 ..Default::default()
@@ -1242,6 +1243,7 @@ pub(crate) async fn save_manual_transition_worker_result_if_absent(
         data,
         &ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             http_preconditions: Some(HTTPPreconditions {
                 if_none_match: Some("*".to_string()),
                 ..Default::default()
@@ -1270,6 +1272,7 @@ pub(crate) async fn save_manual_transition_task_if_absent(
         data,
         &ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             http_preconditions: Some(HTTPPreconditions {
                 if_none_match: Some("*".to_string()),
                 ..Default::default()
@@ -1621,6 +1624,7 @@ pub async fn save_manual_transition_scope_admission_if_absent(
         data.clone(),
         &ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             http_preconditions: Some(HTTPPreconditions {
                 if_none_match: Some("*".to_string()),
                 ..Default::default()
@@ -1672,6 +1676,7 @@ pub async fn save_manual_transition_scope_admission_if_current(
         data.clone(),
         &ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             http_preconditions: Some(HTTPPreconditions {
                 if_match: Some(current_etag.to_string()),
                 ..Default::default()

crates/ecstore/src/bucket/lifecycle/mod.rs

@@ -24,6 +24,7 @@ pub(crate) use metadata_boundary::{LifecycleExpiryConfigs, get_expiry_configs, g
 mod object_handlers_common;
 mod object_lock_boundary;
 pub use self::core as lifecycle;
+pub mod recovery_control;
 mod replication_sink;
 pub mod rule;
 mod runtime_boundary;

crates/ecstore/src/bucket/lifecycle/tier_delete_journal.rs

@@ -1733,6 +1733,7 @@ async fn save_config_if_none_fenced(
         data.clone(),
         &ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             http_preconditions: Some(HTTPPreconditions {
                 if_none_match: Some("*".to_string()),
                 ..Default::default()
@@ -1832,6 +1833,7 @@ async fn save_decommission_manifest_checkpoint_if_match(
 
         let mut opts = ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             no_lock: true,
             http_preconditions: Some(HTTPPreconditions {
                 if_match: Some(observed_etag),
@@ -1960,6 +1962,7 @@ async fn save_config_if_match_fenced(
         data.clone(),
         &ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             http_preconditions: Some(HTTPPreconditions {
                 if_match: Some(etag.to_string()),
                 ..Default::default()
@@ -3780,6 +3783,7 @@ where
                 data.clone(),
                 &ObjectOptions {
                     max_parity: true,
+                    write_completion: crate::object_api::WriteCompletion::TailDrained,
                     http_preconditions: Some(HTTPPreconditions {
                         if_none_match: Some("*".to_string()),
                         ..Default::default()
@@ -3869,6 +3873,7 @@ where
                 data.clone(),
                 &ObjectOptions {
                     max_parity: true,
+                    write_completion: crate::object_api::WriteCompletion::TailDrained,
                     http_preconditions: Some(HTTPPreconditions {
                         if_match: Some(etag),
                         ..Default::default()
@@ -3893,6 +3898,7 @@ where
                 data.clone(),
                 &ObjectOptions {
                     max_parity: true,
+                    write_completion: crate::object_api::WriteCompletion::TailDrained,
                     http_preconditions: Some(HTTPPreconditions {
                         if_none_match: Some("*".to_string()),
                         ..Default::default()

crates/ecstore/src/bucket/lifecycle/tier_sweeper.rs

@@ -15,8 +15,6 @@
 #![allow(unused_variables)]
 #![allow(unused_mut)]
 #![allow(unused_assignments)]
-#![allow(unused_must_use)]
-#![allow(clippy::all)]
 
 use super::runtime_boundary as runtime_sources;
 use crate::bucket::lifecycle::bucket_lifecycle_ops::ExpiryOp;
@@ -72,9 +70,11 @@ static REMOTE_DELETE_BREAKER: LazyLock<Mutex<RemoteDeleteBreaker>> = LazyLock::n
 });
 
 #[cfg(test)]
-static REMOTE_TIER_DELETE_TEST_HOOK: std::sync::LazyLock<
+type RemoteTierDeleteTestHook = Box<dyn Fn(&str, &str, &str) -> std::io::Result<()> + Send + Sync>;
-    std::sync::Mutex<Option<Box<dyn Fn(&str, &str, &str) -> std::io::Result<()> + Send + Sync>>>,
+
-> = std::sync::LazyLock::new(|| std::sync::Mutex::new(None));
+#[cfg(test)]
+static REMOTE_TIER_DELETE_TEST_HOOK: std::sync::LazyLock<std::sync::Mutex<Option<RemoteTierDeleteTestHook>>> =
+    std::sync::LazyLock::new(|| std::sync::Mutex::new(None));
 
 #[derive(Debug)]
 struct RemoteDeleteBreaker {
@@ -107,7 +107,7 @@ impl RemoteDeleteBreaker {
     fn prune(&mut self, now: Instant) {
         while let Some(ts) = self.failures.front().copied() {
             if now.duration_since(ts) > self.window {
-                self.failures.pop_front();
+                let _ = self.failures.pop_front();
             } else {
                 break;
             }
@@ -137,11 +137,11 @@ fn is_signer_header_error(err: &std::io::Error) -> bool {
         return false;
     }
 
-    if let Some(source) = err.get_ref() {
+    if let Some(source) = err.get_ref()
-        if error_chain_contains_signer_header_marker(source) {
+        && error_chain_contains_signer_header_marker(source)
+    {
         return true;
     }
-    }
 
     let message = err.to_string().to_ascii_lowercase();
     message.contains("invalid utf-8 header value")
@@ -205,7 +205,7 @@ impl ObjSweeper {
 
     #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     pub fn with_version(&mut self, vid: Option<Uuid>) -> &Self {
-        self.version_id = vid.clone();
+        self.version_id = vid;
         self
     }
 
@@ -219,7 +219,7 @@ impl ObjSweeper {
     #[allow(dead_code, reason = "MinIO-parity surface with no caller in this port (backlog#1823)")]
     pub fn get_opts(&self) -> lifecycle::ObjectOpts {
         let mut opts = ObjectOpts {
-            version_id: self.version_id.clone(),
+            version_id: self.version_id,
             versioned: self.versioned,
             version_suspended: self.suspended,
             ..Default::default()
@@ -388,8 +388,8 @@ impl Jentry {
 impl ExpiryOp for Jentry {
     fn op_hash(&self) -> u64 {
         let mut hasher = Sha256::new();
-        hasher.update(format!("{}", self.tier_name).as_bytes());
+        hasher.update(self.tier_name.as_bytes());
-        hasher.update(format!("{}", self.obj_name).as_bytes());
+        hasher.update(self.obj_name.as_bytes());
         xxh64::xxh64(hasher.finalize().as_slice(), XXHASH_SEED)
     }
 
@@ -436,7 +436,7 @@ async fn delete_object_from_remote_tier_raw_with_manager(
     tier_name: &str,
     tier_config_mgr: &Arc<tokio::sync::RwLock<TierConfigMgr>>,
 ) -> Result<(), std::io::Error> {
-    let lease = TierConfigMgr::acquire_operation_lease(&tier_config_mgr, tier_name)
+    let lease = TierConfigMgr::acquire_operation_lease(tier_config_mgr, tier_name)
         .await
         .map_err(std::io::Error::other)?;
     delete_object_from_remote_tier_raw_with_lease(obj_name, rv_id, &lease, false, true).await
@@ -575,7 +575,7 @@ pub(crate) async fn delete_confirmed_transition_candidate_exact_with_lease_idemp
 #[cfg(test)]
 static CONFIRMED_TRANSITION_EMPTY_GUARD_DISPATCHES: std::sync::atomic::AtomicUsize = std::sync::atomic::AtomicUsize::new(0);
 
-#[cfg(test)]
+#[cfg(all(test, feature = "test-util"))]
 pub(crate) async fn delete_confirmed_transition_candidate_exact_with_manager_and_identity(
     obj_name: &str,
     rv_id: &str,
@@ -706,15 +706,16 @@ pub(crate) fn transitioned_delete_journal_entry_for_source(
 
 #[cfg(test)]
 mod test {
+    #[cfg(feature = "test-util")]
+    use super::delete_confirmed_transition_candidate_exact_with_manager_and_identity;
     use rustfs_s3_client::signer_error::invalid_utf8_header_error;
 
     use super::{
         CONFIRMED_TRANSITION_EMPTY_GUARD_DISPATCHES, ERR_REMOTE_DELETE_BREAKER_OPEN, ERR_REMOTE_DELETE_LIMITER_CLOSED, Jentry,
         RemoteDeleteBreaker, RemoteTierDeleteOutcome, TierDeleteJournalState, TierDeleteSourceIdentity,
-        delete_confirmed_transition_candidate_exact_with_manager_and_identity, delete_object_from_remote_tier_idempotent,
+        delete_object_from_remote_tier_idempotent, delete_object_from_remote_tier_idempotent_with_manager_and_identity,
-        delete_object_from_remote_tier_idempotent_with_manager_and_identity, is_remote_tier_not_found_error,
+        is_remote_tier_not_found_error, is_signer_header_error, lifecycle, set_remote_tier_delete_test_hook,
-        is_signer_header_error, lifecycle, set_remote_tier_delete_test_hook, should_record_remote_delete_failure,
+        should_record_remote_delete_failure, transitioned_delete_journal_entry, transitioned_force_delete_journal_entry,
-        transitioned_delete_journal_entry, transitioned_force_delete_journal_entry,
     };
     use crate::storage_api_contracts::lifecycle::TransitionedObject;
     use rustfs_filemeta::TransitionVersionState;

crates/ecstore/src/bucket/lifecycle/transition_transaction.rs

@@ -23,6 +23,11 @@ use uuid::Uuid;
 use crate::bucket::lifecycle::config_boundary;
 use crate::bucket::lifecycle::durable_namespace::TRANSITION_TRANSACTION_NAMESPACE;
 use crate::bucket::lifecycle::lifecycle::TRANSITION_COMPLETE;
+use crate::bucket::lifecycle::recovery_control::{
+    IlmRecoveryClassification, IlmRecoveryControl, IlmRecoveryControlIdentity, IlmRecoveryErrorCode, IlmRecoveryProtocol,
+    ObservedIlmRecoveryControl, load_recovery_control, observe_recovery_source, recovery_control_record_object_name,
+    save_recovery_control_if_absent, save_recovery_control_if_current,
+};
 use crate::bucket::lifecycle::tier_sweeper::{
     delete_confirmed_transition_candidate_exact_with_lease_idempotent,
     delete_object_from_remote_tier_idempotent_with_manager_and_identity,
@@ -44,6 +49,7 @@ const EVENT_LIFECYCLE_TRANSITION_TRANSACTION_RECOVERY: &str = "lifecycle_transit
 pub const DEFAULT_TRANSITION_TRANSACTION_RECOVERY_LIMIT: usize = 1_000;
 const TRANSITION_TRANSACTION_RECOVERY_INTERVAL: Duration = Duration::from_secs(60);
 const TRANSITION_TRANSACTION_RECOVERY_TIMEOUT: Duration = Duration::from_secs(300);
+const TRANSITION_RECOVERY_CONTROL_LEASE_NANOS: i64 = 15 * 60 * 1_000_000_000;
 pub const TRANSITION_TRANSACTION_SCHEMA: &str = "rustfs-transition-transaction-v1";
 pub const TRANSITION_TRANSACTION_PREFIX: &str = "ilm/transition-transactions";
 pub const TRANSITION_TRANSACTION_RECORD_PREFIX: &str = TRANSITION_TRANSACTION_NAMESPACE.prefix;
@@ -612,6 +618,7 @@ pub(crate) async fn save_transition_transaction_record(
         data.clone(),
         &ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             http_preconditions: Some(HTTPPreconditions {
                 if_none_match: Some("*".to_string()),
                 ..Default::default()
@@ -658,6 +665,7 @@ pub(crate) async fn save_transition_transaction_record_if_current(
         data.clone(),
         &ObjectOptions {
             max_parity: true,
+            write_completion: crate::object_api::WriteCompletion::TailDrained,
             http_preconditions: Some(HTTPPreconditions {
                 if_match: Some(etag),
                 ..Default::default()
@@ -735,9 +743,11 @@ pub enum TransitionTransactionRecoveryOutcome {
     RemoteCandidateDeleted,
     RecordDeleted,
     Retained,
+    RetainedAmbiguous(IlmRecoveryErrorCode),
+    OperatorRequired(IlmRecoveryErrorCode),
 }
 
-#[cfg(test)]
+#[cfg(all(test, feature = "test-util"))]
 #[derive(Default)]
 struct TransitionRecoveryClaimBarrierState {
     transaction_id: Uuid,
@@ -745,17 +755,17 @@ struct TransitionRecoveryClaimBarrierState {
     release: tokio::sync::Notify,
 }
 
-#[cfg(test)]
+#[cfg(all(test, feature = "test-util"))]
 pub(crate) struct TransitionRecoveryClaimBarrier {
     state: Arc<TransitionRecoveryClaimBarrierState>,
 }
 
-#[cfg(test)]
+#[cfg(all(test, feature = "test-util"))]
 static TRANSITION_RECOVERY_CLAIM_BARRIER: std::sync::OnceLock<
     std::sync::Mutex<Option<Arc<TransitionRecoveryClaimBarrierState>>>,
 > = std::sync::OnceLock::new();
 
-#[cfg(test)]
+#[cfg(all(test, feature = "test-util"))]
 impl TransitionRecoveryClaimBarrier {
     pub(crate) fn install(transaction_id: Uuid) -> Self {
         let state = Arc::new(TransitionRecoveryClaimBarrierState {
@@ -786,7 +796,7 @@ impl TransitionRecoveryClaimBarrier {
     }
 }
 
-#[cfg(test)]
+#[cfg(all(test, feature = "test-util"))]
 impl Drop for TransitionRecoveryClaimBarrier {
     fn drop(&mut self) {
         self.state.release.notify_one();
@@ -800,7 +810,7 @@ impl Drop for TransitionRecoveryClaimBarrier {
     }
 }
 
-#[cfg(test)]
+#[cfg(all(test, feature = "test-util"))]
 async fn pause_before_transition_recovery_claim(transaction_id: Uuid) {
     let barrier = TRANSITION_RECOVERY_CLAIM_BARRIER
         .get_or_init(|| std::sync::Mutex::new(None))
@@ -815,6 +825,80 @@ async fn pause_before_transition_recovery_claim(transaction_id: Uuid) {
     }
 }
 
+#[cfg(all(test, feature = "test-util"))]
+#[derive(Default)]
+struct TransitionRecoveryTerminalBarrierState {
+    transaction_id: Uuid,
+    arrived: tokio::sync::Notify,
+    release: tokio::sync::Notify,
+}
+
+#[cfg(all(test, feature = "test-util"))]
+pub(crate) struct TransitionRecoveryTerminalBarrier {
+    state: Arc<TransitionRecoveryTerminalBarrierState>,
+}
+
+#[cfg(all(test, feature = "test-util"))]
+static TRANSITION_RECOVERY_TERMINAL_BARRIER: std::sync::OnceLock<
+    std::sync::Mutex<Option<Arc<TransitionRecoveryTerminalBarrierState>>>,
+> = std::sync::OnceLock::new();
+
+#[cfg(all(test, feature = "test-util"))]
+impl TransitionRecoveryTerminalBarrier {
+    pub(crate) fn install(transaction_id: Uuid) -> Self {
+        let state = Arc::new(TransitionRecoveryTerminalBarrierState {
+            transaction_id,
+            ..Default::default()
+        });
+        let mut slot = TRANSITION_RECOVERY_TERMINAL_BARRIER
+            .get_or_init(|| std::sync::Mutex::new(None))
+            .lock()
+            .expect("transition recovery terminal barrier mutex should not poison");
+        assert!(
+            slot.is_none(),
+            "transition recovery terminal barrier must be installed by one test at a time"
+        );
+        *slot = Some(Arc::clone(&state));
+        drop(slot);
+        Self { state }
+    }
+
+    pub(crate) async fn wait_until_paused(&self) {
+        tokio::time::timeout(Duration::from_secs(30), self.state.arrived.notified())
+            .await
+            .expect("transition recovery should persist terminal control before source cleanup");
+    }
+}
+
+#[cfg(all(test, feature = "test-util"))]
+impl Drop for TransitionRecoveryTerminalBarrier {
+    fn drop(&mut self) {
+        self.state.release.notify_one();
+        let mut slot = TRANSITION_RECOVERY_TERMINAL_BARRIER
+            .get_or_init(|| std::sync::Mutex::new(None))
+            .lock()
+            .expect("transition recovery terminal barrier mutex should not poison");
+        if slot.as_ref().is_some_and(|state| Arc::ptr_eq(state, &self.state)) {
+            *slot = None;
+        }
+    }
+}
+
+#[cfg(all(test, feature = "test-util"))]
+async fn pause_after_transition_recovery_terminal(transaction_id: Uuid) {
+    let barrier = TRANSITION_RECOVERY_TERMINAL_BARRIER
+        .get_or_init(|| std::sync::Mutex::new(None))
+        .lock()
+        .expect("transition recovery terminal barrier mutex should not poison")
+        .as_ref()
+        .filter(|barrier| barrier.transaction_id == transaction_id)
+        .cloned();
+    if let Some(barrier) = barrier {
+        barrier.arrived.notify_one();
+        barrier.release.notified().await;
+    }
+}
+
 #[derive(Debug, Clone, PartialEq, Eq, Serialize)]
 #[serde(rename_all = "snake_case")]
 pub enum TransitionOperatorProbe {
@@ -1018,17 +1102,35 @@ fn transition_transaction_id_from_record_object_name(object: &str) -> Result<Uui
     let suffix = object
         .strip_prefix(&prefix)
         .ok_or(TransitionTransactionError::Corrupt("transaction record path has wrong prefix"))?;
-    let file_name = suffix
+    let mut parts = suffix.split('/');
-        .rsplit('/')
+    let shard_a = parts
         .next()
         .ok_or(TransitionTransactionError::Corrupt("transaction record path is incomplete"))?;
+    let shard_b = parts
+        .next()
+        .ok_or(TransitionTransactionError::Corrupt("transaction record path is incomplete"))?;
+    let file_name = parts
+        .next()
+        .ok_or(TransitionTransactionError::Corrupt("transaction record path is incomplete"))?;
+    if parts.next().is_some() {
+        return Err(TransitionTransactionError::Corrupt("transaction record path is not canonical"));
+    }
     let transaction_key = file_name
         .strip_suffix(".json")
         .ok_or(TransitionTransactionError::Corrupt("transaction record path has wrong suffix"))?;
-    if transaction_key.len() != 32 || !transaction_key.bytes().all(|byte| byte.is_ascii_hexdigit()) {
+    if transaction_key.len() != 32
+        || !transaction_key
+            .bytes()
+            .all(|byte| byte.is_ascii_hexdigit() && !byte.is_ascii_uppercase())
+        || shard_a != &transaction_key[..2]
+        || shard_b != &transaction_key[2..4]
+    {
         return Err(TransitionTransactionError::Corrupt("transaction record path has invalid transaction id"));
     }
-    Uuid::parse_str(transaction_key).map_err(|_| TransitionTransactionError::Corrupt("transaction record path has invalid uuid"))
+    Uuid::parse_str(transaction_key)
+        .ok()
+        .filter(|transaction_id| !transaction_id.is_nil())
+        .ok_or(TransitionTransactionError::Corrupt("transaction record path has invalid uuid"))
 }
 
 pub async fn process_transition_transaction_record(
@@ -1053,6 +1155,27 @@ async fn process_transition_transaction_record_at(
 ) -> EcstoreResult<TransitionTransactionRecoveryOutcome> {
     let record_name =
         transition_transaction_record_object_name(observed.transaction_id).map_err(transition_transaction_store_error)?;
+    let now_unix_nanos =
+        i64::try_from(now_unix_nanos).map_err(|_| Error::other("transition transaction recovery timestamp does not fit i64"))?;
+    let recovery_control_identity = transition_recovery_control_identity(observed, &record_name);
+    let recovery_control_id = recovery_control_identity
+        .source_operation_digest()
+        .map_err(|err| Error::other(err.to_string()))?;
+    let control_record_name =
+        recovery_control_record_object_name(IlmRecoveryProtocol::TransitionTransaction, &recovery_control_id)
+            .map_err(|err| Error::other(err.to_string()))?;
+    let control_lock = if transition_state_needs_recovery_control(observed, now_unix_nanos) {
+        Some(
+            api.new_ns_lock(RUSTFS_META_BUCKET, &format!("{control_record_name}.recovery-lock"))
+                .await?,
+        )
+    } else {
+        None
+    };
+    let _control_guard = match &control_lock {
+        Some(lock) => Some(lock.get_write_lock(crate::set_disk::get_lock_acquire_timeout()).await?),
+        None => None,
+    };
     // The synthetic key avoids nesting the recovery lock with the config
     // object's own I/O lock. Holding it across the bounded source proof and
     // remote DELETE elects one destructive recovery worker across nodes.
@@ -1071,11 +1194,40 @@ async fn process_transition_transaction_record_at(
         return Ok(TransitionTransactionRecoveryOutcome::Retained);
     }
 
-    match current.state {
+    let mut recovery_control = if transition_state_needs_recovery_control(&current, now_unix_nanos) {
-        TransitionTransactionState::Uploaded => {
+        if cleanup_terminal_transition_recovery_control(
-            if transition_transaction_ownership_is_active(&current, now_unix_nanos) {
+            api.clone(),
-                return Ok(TransitionTransactionRecoveryOutcome::Retained);
+            &current,
+            &record_name,
+            &recovery_control_identity,
+            &recovery_control_id,
+        )
+        .await?
+        {
+            return Ok(TransitionTransactionRecoveryOutcome::RecordDeleted);
         }
+        match claim_transition_recovery_control(
+            api.clone(),
+            &current,
+            &record_name,
+            recovery_control_identity,
+            &recovery_control_id,
+            now_unix_nanos,
+        )
+        .await?
+        {
+            Some(control) => Some(control),
+            None => return Ok(TransitionTransactionRecoveryOutcome::Retained),
+        }
+    } else {
+        None
+    };
+
+    let recovery = match current.state {
+        TransitionTransactionState::Uploaded => {
+            if transition_transaction_ownership_is_active(&current, i128::from(now_unix_nanos)) {
+                Ok(TransitionTransactionRecoveryOutcome::Retained)
+            } else {
                 let mut cleanup = current.clone();
                 cleanup
                     .mark_cleanup_pending(
@@ -1090,36 +1242,352 @@ async fn process_transition_transaction_record_at(
                         },
                     )
                     .map_err(transition_transaction_store_error)?;
-            #[cfg(test)]
+                #[cfg(all(test, feature = "test-util"))]
                 pause_before_transition_recovery_claim(current.transaction_id).await;
                 match save_transition_transaction_record_if_current(api.clone(), &current, &cleanup).await {
-                Ok(()) => recover_cleanup_pending(api, &cleanup).await,
+                    Ok(()) => recover_cleanup_pending(api.clone(), &cleanup).await,
-                Err(Error::PreconditionFailed) | Err(Error::ConfigNotFound) => Ok(TransitionTransactionRecoveryOutcome::Retained),
+                    Err(Error::PreconditionFailed) | Err(Error::ConfigNotFound) => {
+                        Ok(TransitionTransactionRecoveryOutcome::Retained)
+                    }
                     Err(err) => Err(err),
                 }
             }
-        TransitionTransactionState::CleanupPending => recover_cleanup_pending(api, &current).await,
-        TransitionTransactionState::LocalCommitStarted => match local_commit_matches_transaction(api.clone(), &current).await {
-            Ok(true) => {
-                delete_transition_transaction_record(api, &current).await?;
-                Ok(TransitionTransactionRecoveryOutcome::RecordDeleted)
         }
-            Ok(false) => Ok(TransitionTransactionRecoveryOutcome::Retained),
+        TransitionTransactionState::CleanupPending => recover_cleanup_pending(api.clone(), &current).await,
-            Err(err) if transition_source_is_missing(&err) => Ok(TransitionTransactionRecoveryOutcome::Retained),
+        TransitionTransactionState::LocalCommitStarted => match local_commit_matches_transaction(api.clone(), &current).await {
+            Ok(true) => Ok(TransitionTransactionRecoveryOutcome::RecordDeleted),
+            Ok(false) => Ok(TransitionTransactionRecoveryOutcome::OperatorRequired(
+                IlmRecoveryErrorCode::LocalCommitAmbiguous,
+            )),
+            Err(err) if transition_source_is_missing(&err) => Ok(TransitionTransactionRecoveryOutcome::OperatorRequired(
+                IlmRecoveryErrorCode::LocalCommitAmbiguous,
+            )),
             Err(err) => Err(err),
         },
         TransitionTransactionState::AbortedNoRemote | TransitionTransactionState::Committed => {
-            delete_transition_transaction_record(api, &current).await?;
             Ok(TransitionTransactionRecoveryOutcome::RecordDeleted)
         }
         TransitionTransactionState::UploadOutcomeUnknown => {
-            if transition_transaction_ownership_is_active(&current, now_unix_nanos) {
+            if transition_transaction_ownership_is_active(&current, i128::from(now_unix_nanos)) {
                 Ok(TransitionTransactionRecoveryOutcome::Retained)
             } else {
-                recover_unknown_upload_outcome(api, &current).await
+                recover_unknown_upload_outcome(api.clone(), &current).await
             }
         }
-        TransitionTransactionState::UploadStarted => Ok(TransitionTransactionRecoveryOutcome::Retained),
+        TransitionTransactionState::UploadStarted => {
+            if transition_transaction_ownership_is_active(&current, i128::from(now_unix_nanos)) {
+                Ok(TransitionTransactionRecoveryOutcome::Retained)
+            } else {
+                Ok(TransitionTransactionRecoveryOutcome::RetainedAmbiguous(
+                    IlmRecoveryErrorCode::RemoteVersionUnknown,
+                ))
+            }
+        }
+    };
+
+    if let Some(mut control) = recovery_control.take() {
+        let source_to_delete = if matches!(
+            recovery,
+            Ok(TransitionTransactionRecoveryOutcome::RemoteCandidateDeleted
+                | TransitionTransactionRecoveryOutcome::RecordDeleted)
+        ) {
+            let refreshed =
+                refresh_transition_recovery_control_source(api.clone(), control, &record_name, current.transaction_id).await?;
+            control = refreshed.0;
+            refreshed.1
+        } else {
+            None
+        };
+        persist_transition_recovery_result(api.clone(), control, &recovery, now_unix_nanos).await?;
+        if let Some(source) = source_to_delete {
+            #[cfg(all(test, feature = "test-util"))]
+            pause_after_transition_recovery_terminal(source.transaction_id).await;
+            delete_transition_transaction_record(api, &source).await?;
+        }
+    } else if matches!(
+        recovery,
+        Ok(TransitionTransactionRecoveryOutcome::RemoteCandidateDeleted | TransitionTransactionRecoveryOutcome::RecordDeleted)
+    ) {
+        delete_transition_transaction_record(api, &current).await?;
+    }
+    recovery
+}
+
+fn transition_recovery_control_identity(transaction: &TransitionTransaction, record_name: &str) -> IlmRecoveryControlIdentity {
+    IlmRecoveryControlIdentity {
+        protocol: IlmRecoveryProtocol::TransitionTransaction,
+        canonical_source_path: record_name.to_string(),
+        stable_operation_identity: transaction.transaction_id.to_string(),
+        record_class: "transition_transaction_v1".to_string(),
+    }
+}
+
+#[cfg(all(test, feature = "test-util"))]
+pub(crate) fn transition_recovery_control_id(transaction: &TransitionTransaction) -> Result<String> {
+    let record_name = transition_transaction_record_object_name(transaction.transaction_id)?;
+    transition_recovery_control_identity(transaction, &record_name)
+        .source_operation_digest()
+        .map_err(|_| TransitionTransactionError::Corrupt("transition recovery control identity is invalid"))
+}
+
+fn transition_state_needs_recovery_control(transaction: &TransitionTransaction, now_unix_nanos: i64) -> bool {
+    now_unix_nanos >= transaction.not_after_unix_nanos
+        && !matches!(
+            transaction.state,
+            TransitionTransactionState::AbortedNoRemote | TransitionTransactionState::Committed
+        )
+}
+
+async fn cleanup_terminal_transition_recovery_control(
+    api: Arc<ECStore>,
+    transaction: &TransitionTransaction,
+    record_name: &str,
+    identity: &IlmRecoveryControlIdentity,
+    control_id: &str,
+) -> EcstoreResult<bool> {
+    let observed = match load_recovery_control(api.clone(), IlmRecoveryProtocol::TransitionTransaction, control_id).await {
+        Ok(observed) => observed,
+        Err(Error::ConfigNotFound) => return Ok(false),
+        Err(err) => return Err(err),
+    };
+    if observed.control.classification != IlmRecoveryClassification::Terminal {
+        return Ok(false);
+    }
+    let source = observe_recovery_source(api.clone(), record_name, TRANSITION_TRANSACTION_SCHEMA).await?;
+    let exact_source = source.is_consistent()
+        && source.generation == observed.control.observed_source_generation
+        && source.canonical_data.as_deref().is_some_and(|data| {
+            TransitionTransaction::decode(transaction.transaction_id, data).is_ok_and(|decoded| decoded == *transaction)
+        });
+    if observed.control.identity != *identity || !exact_source {
+        return Ok(false);
+    }
+    delete_transition_transaction_record(api, transaction).await?;
+    Ok(true)
+}
+
+async fn claim_transition_recovery_control(
+    api: Arc<ECStore>,
+    transaction: &TransitionTransaction,
+    record_name: &str,
+    identity: IlmRecoveryControlIdentity,
+    control_id: &str,
+    now_unix_nanos: i64,
+) -> EcstoreResult<Option<ObservedIlmRecoveryControl>> {
+    let existing = match load_recovery_control(api.clone(), IlmRecoveryProtocol::TransitionTransaction, control_id).await {
+        Ok(control) => Some(control),
+        Err(Error::ConfigNotFound) => None,
+        Err(err) => return Err(err),
+    };
+    if let Some(observed) = existing.as_ref() {
+        if observed.control.identity != identity {
+            return Ok(None);
+        }
+        if observed
+            .control
+            .owner
+            .as_ref()
+            .is_some_and(|owner| owner.lease_expires_at_unix_nanos <= now_unix_nanos)
+        {
+            let mut expired = observed.control.clone();
+            expired
+                .record_expired_attempt(now_unix_nanos)
+                .map_err(|err| Error::other(err.to_string()))?;
+            save_recovery_control_if_current(api, observed, &expired).await?;
+            return Ok(None);
+        }
+        if !observed.control.should_attempt_at(now_unix_nanos) {
+            return Ok(None);
+        }
+    }
+
+    let source = match observe_recovery_source(api.clone(), record_name, TRANSITION_TRANSACTION_SCHEMA).await {
+        Ok(source) => source,
+        Err(err) => {
+            if let Some(observed) = existing {
+                persist_transition_recovery_source_failure(api, observed, now_unix_nanos).await?;
+                return Ok(None);
+            }
+            return Err(err);
+        }
+    };
+    let source_matches = source.is_consistent()
+        && source.canonical_data.as_deref().is_some_and(|data| {
+            TransitionTransaction::decode(transaction.transaction_id, data).is_ok_and(|observed| observed == *transaction)
+        });
+    let source_error = if source_matches {
+        IlmRecoveryErrorCode::None
+    } else if source.canonical_data.is_some() {
+        IlmRecoveryErrorCode::SourceGenerationChanged
+    } else {
+        IlmRecoveryErrorCode::SourceDivergent
+    };
+
+    let mut observed = match existing {
+        Some(control) => control,
+        None => {
+            let candidate = IlmRecoveryControl::new(
+                identity.clone(),
+                source.generation.clone(),
+                if source_matches {
+                    IlmRecoveryClassification::Retrying
+                } else {
+                    IlmRecoveryClassification::Corrupt
+                },
+                now_unix_nanos,
+                source_error,
+            )
+            .map_err(|err| Error::other(err.to_string()))?;
+            match save_recovery_control_if_absent(api.clone(), &candidate).await {
+                Ok(()) | Err(Error::PreconditionFailed) => {}
+                Err(err) => return Err(err),
+            }
+            load_recovery_control(api.clone(), IlmRecoveryProtocol::TransitionTransaction, control_id).await?
+        }
+    };
+    if observed.control.identity != identity || !observed.control.should_attempt_at(now_unix_nanos) {
+        return Ok(None);
+    }
+
+    let mut claimed = observed.control.clone();
+    claimed
+        .claim_for_source_generation(
+            api.id.to_string(),
+            Uuid::new_v4(),
+            now_unix_nanos,
+            TRANSITION_RECOVERY_CONTROL_LEASE_NANOS,
+            source.generation,
+        )
+        .map_err(|err| Error::other(err.to_string()))?;
+    save_recovery_control_if_current(api.clone(), &observed, &claimed).await?;
+    observed = load_recovery_control(api.clone(), IlmRecoveryProtocol::TransitionTransaction, control_id).await?;
+    if observed.control != claimed {
+        return Err(Error::PreconditionFailed);
+    }
+    if !source_matches {
+        let mut corrupt = observed.control.clone();
+        corrupt
+            .finish_attempt(IlmRecoveryClassification::Corrupt, source_error)
+            .map_err(|err| Error::other(err.to_string()))?;
+        save_recovery_control_if_current(api, &observed, &corrupt).await?;
+        return Ok(None);
+    }
+    Ok(Some(observed))
+}
+
+async fn persist_transition_recovery_source_failure(
+    api: Arc<ECStore>,
+    observed: ObservedIlmRecoveryControl,
+    now_unix_nanos: i64,
+) -> EcstoreResult<()> {
+    let mut claimed = observed.control.clone();
+    claimed
+        .claim(
+            api.id.to_string(),
+            Uuid::new_v4(),
+            now_unix_nanos,
+            TRANSITION_RECOVERY_CONTROL_LEASE_NANOS,
+        )
+        .map_err(|err| Error::other(err.to_string()))?;
+    save_recovery_control_if_current(api.clone(), &observed, &claimed).await?;
+    let claimed = load_recovery_control(
+        api.clone(),
+        IlmRecoveryProtocol::TransitionTransaction,
+        &claimed
+            .identity
+            .source_operation_digest()
+            .map_err(|err| Error::other(err.to_string()))?,
+    )
+    .await?;
+    let mut failed = claimed.control.clone();
+    failed
+        .record_retryable_failure(now_unix_nanos, IlmRecoveryErrorCode::SourceUnavailable)
+        .map_err(|err| Error::other(err.to_string()))?;
+    save_recovery_control_if_current(api, &claimed, &failed).await
+}
+
+async fn refresh_transition_recovery_control_source(
+    api: Arc<ECStore>,
+    mut observed: ObservedIlmRecoveryControl,
+    record_name: &str,
+    transaction_id: Uuid,
+) -> EcstoreResult<(ObservedIlmRecoveryControl, Option<TransitionTransaction>)> {
+    let transaction = match load_transition_transaction_record(api.clone(), transaction_id).await {
+        Ok(transaction) => transaction,
+        Err(Error::ConfigNotFound) => return Ok((observed, None)),
+        Err(err) => return Err(err),
+    };
+    let source = observe_recovery_source(api.clone(), record_name, TRANSITION_TRANSACTION_SCHEMA).await?;
+    let exact_source = source.is_consistent()
+        && source
+            .canonical_data
+            .as_deref()
+            .is_some_and(|data| TransitionTransaction::decode(transaction_id, data).is_ok_and(|decoded| decoded == transaction));
+    if !exact_source {
+        return Err(Error::PreconditionFailed);
+    }
+    if observed.control.observed_source_generation != source.generation {
+        let mut refreshed = observed.control.clone();
+        refreshed
+            .refresh_owned_source_generation(source.generation)
+            .map_err(|err| Error::other(err.to_string()))?;
+        save_recovery_control_if_current(api.clone(), &observed, &refreshed).await?;
+        observed = load_recovery_control(
+            api,
+            IlmRecoveryProtocol::TransitionTransaction,
+            &refreshed
+                .identity
+                .source_operation_digest()
+                .map_err(|err| Error::other(err.to_string()))?,
+        )
+        .await?;
+        if observed.control != refreshed {
+            return Err(Error::PreconditionFailed);
+        }
+    }
+    Ok((observed, Some(transaction)))
+}
+
+async fn persist_transition_recovery_result(
+    api: Arc<ECStore>,
+    observed: ObservedIlmRecoveryControl,
+    recovery: &EcstoreResult<TransitionTransactionRecoveryOutcome>,
+    now_unix_nanos: i64,
+) -> EcstoreResult<()> {
+    let mut next = observed.control.clone();
+    match recovery {
+        Ok(
+            TransitionTransactionRecoveryOutcome::RemoteCandidateDeleted | TransitionTransactionRecoveryOutcome::RecordDeleted,
+        ) => next
+            .finish_attempt(IlmRecoveryClassification::Terminal, IlmRecoveryErrorCode::None)
+            .map_err(|err| Error::other(err.to_string()))?,
+        Ok(TransitionTransactionRecoveryOutcome::Retained) => next
+            .record_retryable_failure(now_unix_nanos, IlmRecoveryErrorCode::SourceGenerationChanged)
+            .map_err(|err| Error::other(err.to_string()))?,
+        Ok(TransitionTransactionRecoveryOutcome::RetainedAmbiguous(code)) => next
+            .finish_attempt(IlmRecoveryClassification::RetainedAmbiguous, *code)
+            .map_err(|err| Error::other(err.to_string()))?,
+        Ok(TransitionTransactionRecoveryOutcome::OperatorRequired(code)) => next
+            .finish_attempt(IlmRecoveryClassification::OperatorRequired, *code)
+            .map_err(|err| Error::other(err.to_string()))?,
+        Err(err) => next
+            .record_retryable_failure(now_unix_nanos, transition_recovery_error_code(err))
+            .map_err(|err| Error::other(err.to_string()))?,
+    }
+    save_recovery_control_if_current(api, &observed, &next).await
+}
+
+fn transition_recovery_error_code(err: &Error) -> IlmRecoveryErrorCode {
+    match err {
+        Error::PreconditionFailed => IlmRecoveryErrorCode::CasConflict,
+        Error::ConfigNotFound
+        | Error::FileNotFound
+        | Error::FileVersionNotFound
+        | Error::ObjectNotFound(_, _)
+        | Error::VersionNotFound(_, _, _)
+        | Error::BucketNotFound(_) => IlmRecoveryErrorCode::SourceUnavailable,
+        Error::SlowDown => IlmRecoveryErrorCode::BackendThrottled,
+        _ => IlmRecoveryErrorCode::Unknown,
     }
 }
 
@@ -1132,10 +1600,7 @@ async fn recover_cleanup_pending(
     transaction: &TransitionTransaction,
 ) -> EcstoreResult<TransitionTransactionRecoveryOutcome> {
     match local_commit_matches_transaction(api.clone(), transaction).await {
-        Ok(true) => {
+        Ok(true) => Ok(TransitionTransactionRecoveryOutcome::RecordDeleted),
-            delete_transition_transaction_record(api, transaction).await?;
-            Ok(TransitionTransactionRecoveryOutcome::RecordDeleted)
-        }
         Ok(false) => delete_unreferenced_transition_candidate(api, transaction).await,
         Err(err) if transition_source_is_missing(&err) => delete_unreferenced_transition_candidate(api, transaction).await,
         Err(err) => Err(err),
@@ -1155,7 +1620,6 @@ async fn delete_unreferenced_transition_candidate(
         return Ok(TransitionTransactionRecoveryOutcome::Retained);
     }
     delete_transition_remote_candidate(api.clone(), &current).await?;
-    delete_transition_transaction_record(api, &current).await?;
     Ok(TransitionTransactionRecoveryOutcome::RemoteCandidateDeleted)
 }
 
@@ -1176,24 +1640,26 @@ async fn recover_unknown_upload_outcome(
         .await
         .map_err(Error::other)?
     {
-        TransitionCandidateProbe::Missing => {
+        TransitionCandidateProbe::Missing => Ok(TransitionTransactionRecoveryOutcome::RecordDeleted),
-            delete_transition_transaction_record(api, transaction).await?;
-            Ok(TransitionTransactionRecoveryOutcome::RecordDeleted)
-        }
         TransitionCandidateProbe::UnversionedPresent => {
             cleanup_recovered_unknown_upload_candidate(api, transaction, TransitionRemoteVersion::unversioned()).await
         }
         TransitionCandidateProbe::VersionedPresent(version_id)
             if Uuid::parse_str(&version_id).is_ok_and(|version_id| version_id.is_nil()) =>
         {
-            Ok(TransitionTransactionRecoveryOutcome::Retained)
+            Ok(TransitionTransactionRecoveryOutcome::RetainedAmbiguous(
+                IlmRecoveryErrorCode::RemoteVersionUnknown,
+            ))
         }
         TransitionCandidateProbe::VersionedPresent(version_id) => {
             cleanup_recovered_unknown_upload_candidate(api, transaction, TransitionRemoteVersion::versioned(version_id)).await
         }
-        TransitionCandidateProbe::Ambiguous | TransitionCandidateProbe::Unsupported => {
+        TransitionCandidateProbe::Ambiguous => Ok(TransitionTransactionRecoveryOutcome::RetainedAmbiguous(
-            Ok(TransitionTransactionRecoveryOutcome::Retained)
+            IlmRecoveryErrorCode::RemoteProbeAmbiguous,
-        }
+        )),
+        TransitionCandidateProbe::Unsupported => Ok(TransitionTransactionRecoveryOutcome::RetainedAmbiguous(
+            IlmRecoveryErrorCode::RemoteProbeUnsupported,
+        )),
     }
 }
 
@@ -1287,7 +1753,7 @@ pub async fn recover_transition_transaction_records(
     recover_transition_transaction_records_with_now(api, limit, marker, None).await
 }
 
-#[cfg(any(test, feature = "test-util"))]
+#[cfg(feature = "test-util")]
 pub async fn recover_transition_transaction_records_at(
     api: Arc<ECStore>,
     limit: usize,
@@ -1321,6 +1787,11 @@ async fn recover_transition_transaction_records_with_now(
             false,
         )
         .await?;
+    if list.is_truncated && list.next_continuation_token.is_none() {
+        return Err(Error::other(
+            "transition transaction recovery returned a truncated page without a continuation marker",
+        ));
+    }
 
     let mut stats = TransitionTransactionRecoveryStats {
         scanned: 0,
@@ -1379,7 +1850,11 @@ async fn recover_transition_transaction_records_with_now(
             ) => {
                 stats.recovered += 1;
             }
-            Ok(TransitionTransactionRecoveryOutcome::Retained) => {
+            Ok(
+                TransitionTransactionRecoveryOutcome::Retained
+                | TransitionTransactionRecoveryOutcome::RetainedAmbiguous(_)
+                | TransitionTransactionRecoveryOutcome::OperatorRequired(_),
+            ) => {
                 stats.retained += 1;
                 debug!(
                     event = EVENT_LIFECYCLE_TRANSITION_TRANSACTION_RECOVERY,
@@ -1507,11 +1982,74 @@ fn state_requires_known_remote_version(state: TransitionTransactionState) -> boo
 #[cfg(test)]
 mod tests {
     use std::collections::HashMap;
+    use std::sync::atomic::{AtomicBool, Ordering};
 
     use super::*;
 
     const BACKEND_FINGERPRINT: [u8; 32] = [7; 32];
 
+    struct RecoveryAttemptDropGuard(Arc<AtomicBool>);
+
+    impl Drop for RecoveryAttemptDropGuard {
+        fn drop(&mut self) {
+            self.0.store(true, Ordering::SeqCst);
+        }
+    }
+
+    async fn pending_recovery_attempt(started: Arc<tokio::sync::Notify>, dropped: Arc<AtomicBool>) -> EcstoreResult<()> {
+        let _drop_guard = RecoveryAttemptDropGuard(dropped);
+        started.notify_one();
+        std::future::pending().await
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn transition_recovery_timeout_and_cancellation_drop_inflight_attempts() {
+        let timeout_started = Arc::new(tokio::sync::Notify::new());
+        let timeout_dropped = Arc::new(AtomicBool::new(false));
+        let timeout_task = tokio::spawn({
+            let started = Arc::clone(&timeout_started);
+            let dropped = Arc::clone(&timeout_dropped);
+            async move {
+                await_transition_transaction_recovery(
+                    &CancellationToken::new(),
+                    TRANSITION_TRANSACTION_RECOVERY_TIMEOUT,
+                    pending_recovery_attempt(started, dropped),
+                )
+                .await
+            }
+        });
+        timeout_started.notified().await;
+        tokio::time::advance(TRANSITION_TRANSACTION_RECOVERY_TIMEOUT).await;
+        let timed_out = timeout_task.await.expect("timeout wrapper task should join");
+        assert!(matches!(timed_out, Some(Err(_))), "outer timeout should fail the recovery pass");
+        assert!(timeout_dropped.load(Ordering::SeqCst), "outer timeout must drop its in-flight attempt");
+
+        let cancel_token = CancellationToken::new();
+        let cancel_started = Arc::new(tokio::sync::Notify::new());
+        let cancel_dropped = Arc::new(AtomicBool::new(false));
+        let cancel_task = tokio::spawn({
+            let cancel_token = cancel_token.clone();
+            let started = Arc::clone(&cancel_started);
+            let dropped = Arc::clone(&cancel_dropped);
+            async move {
+                await_transition_transaction_recovery(
+                    &cancel_token,
+                    TRANSITION_TRANSACTION_RECOVERY_TIMEOUT,
+                    pending_recovery_attempt(started, dropped),
+                )
+                .await
+            }
+        });
+        cancel_started.notified().await;
+        cancel_token.cancel();
+        let cancelled = cancel_task.await.expect("cancellation wrapper task should join");
+        assert!(cancelled.is_none(), "outer cancellation should stop the recovery loop");
+        assert!(
+            cancel_dropped.load(Ordering::SeqCst),
+            "outer cancellation must drop its in-flight attempt"
+        );
+    }
+
     #[derive(Default)]
     struct MemoryTransactionStore {
         records: HashMap<Uuid, Vec<u8>>,
@@ -1966,5 +2504,19 @@ mod tests {
             transition_transaction_record_object_name(Uuid::nil()),
             Err(TransitionTransactionError::Corrupt("transaction_id is nil"))
         ));
+        assert_eq!(
+            transition_transaction_id_from_record_object_name(&object).expect("canonical record path should parse"),
+            transaction_id
+        );
+        for malformed in [
+            object.to_ascii_uppercase(),
+            object.replace("/aa/aa/", "/ff/aa/"),
+            object.replace("/aa/aa/", "/aa/aa/extra/"),
+        ] {
+            assert!(matches!(
+                transition_transaction_id_from_record_object_name(&malformed),
+                Err(TransitionTransactionError::Corrupt(_))
+            ));
+        }
     }
 }

crates/ecstore/src/bucket/metadata.rs

@@ -477,28 +477,29 @@ impl BucketMetadata {
         !self.table_bucket_config_json.is_empty()
     }
 
-    /// Parsed per-bucket durability override, if a valid one is stored.
+    /// `bucket-targets.json` is stored for this bucket but this build cannot
+    /// decode it.
     ///
-    /// Absent/empty/unparsable payloads all mean "no override" (the bucket
+    /// Keeps "no replication targets configured" and "the target
-    /// follows the global durability mode); a parse failure is logged so a
+    /// configuration cannot be read" apart, the same distinction the
-    /// corrupted entry cannot silently change fsync behavior.
+    /// `fabricated` marker draws for the bucket metadata as a whole. Only
-    /// Parsed on-demand migration config, if one is stored.
+    /// meaningful after [`Self::parse_all_configs`] has run; readers must fail
-    ///
+    /// closed on `true` instead of serving an empty target set.
-    /// `Ok(None)` means no config (absent or cleared). A stored payload that
+    pub fn bucket_targets_unreadable(&self) -> bool {
-    /// does not parse is an error, never a default: the runtime must not
+        !self.bucket_targets_config_json.is_empty() && self.bucket_target_config.is_none()
-    /// pull from a source it cannot describe.
-    pub fn on_demand_migration_config(
-        &self,
-    ) -> std::result::Result<
-        Option<super::on_demand_migration::OnDemandMigrationConfig>,
-        super::on_demand_migration::OnDemandMigrationConfigError,
-    > {
-        if self.on_demand_migration_config_json.is_empty() {
-            return Ok(None);
-        }
-        super::on_demand_migration::OnDemandMigrationConfig::from_json(&self.on_demand_migration_config_json).map(Some)
     }
 
+    /// Opaque application-owned configuration with its persisted update time.
+    /// Empty bytes mean absent or cleared; decoding belongs to the consumer.
+    pub fn on_demand_migration_config(&self) -> Option<(&[u8], OffsetDateTime)> {
+        (!self.on_demand_migration_config_json.is_empty()).then_some((
+            self.on_demand_migration_config_json.as_slice(),
+            self.on_demand_migration_config_updated_at,
+        ))
+    }
+
+    /// Parsed per-bucket durability override, if a valid one is stored.
+    /// Invalid payloads follow the global mode after logging a parse failure.
     pub fn durability_config(&self) -> Option<super::durability::BucketDurabilityConfig> {
         if self.durability_config_json.is_empty() {
             return None;
@@ -904,13 +905,6 @@ impl BucketMetadata {
                 self.durability_config_updated_at = updated;
             }
             BUCKET_ON_DEMAND_MIGRATION_CONFIG => {
-                // Structural check only (shape, unknown fields); the
-                // deployment-relative rules run in the admin handler with a
-                // `ValidationContext`. A blob this build cannot read must not
-                // be persisted for every later reader to trip over.
-                if !data.is_empty() {
-                    super::on_demand_migration::OnDemandMigrationConfig::from_json(&data).map_err(Error::other)?;
-                }
                 self.on_demand_migration_config_json = data;
                 self.on_demand_migration_config_updated_at = updated;
             }
@@ -964,7 +958,32 @@ impl BucketMetadata {
         Ok(())
     }
 
-    fn parse_all_configs(&mut self) -> Result<()> {
+    /// Decode every stored sub-configuration into its typed field.
+    ///
+    /// A decode failure never fails the whole load: this runs on every bucket
+    /// metadata read, including startup and peer reload, so one bucket's
+    /// corrupt sub-configuration must not make the bucket — or the node —
+    /// unloadable. Instead the failure is *retained*: the raw bytes stay
+    /// untouched and the typed field stays `None`, so `!raw.is_empty() &&
+    /// typed.is_none()` is the durable "exists but cannot be read" signal that
+    /// each accessor keys off. Which accessors must fail closed on it:
+    ///
+    /// | Config | Verdict |
+    /// |---|---|
+    /// | policy | Fails closed: `get_bucket_policy` re-parses the raw JSON and propagates the error; `get_bucket_policy_raw` returns the stored bytes. |
+    /// | object lock | Fails closed in `object_lock_config_state_from_authoritative_metadata`; a retention decision may never be taken on a guess. |
+    /// | versioning | Fails closed in `get_versioning_config`; guessing Unversioned would make delete markers and version ids diverge from what is on disk. |
+    /// | replication | Fails closed in `get_replication_config`. |
+    /// | bucket targets | Fails closed in `get_bucket_targets_config`, and `sync_bucket_target_sys` marks the bucket unreadable in `BucketTargetSys` instead of publishing an empty target set (rustfs/backlog#2282). |
+    /// | encryption | Fails closed in `get_sse_config`: degrading to "no default encryption" stores plaintext objects the operator required to be encrypted. |
+    /// | public access block | Fails closed in `get_public_access_block_config`: degrading grants the anonymous access the operator asked to block. |
+    /// | quota | Fails closed in `get_quota_config`; the enforcement path in `quota::checker` already re-parses the raw JSON and refuses on error. |
+    /// | lifecycle | Safe to degrade: no rules means no expiration and no transition, so nothing is deleted or moved on the strength of an unreadable rule set. The bucket keeps serving reads and writes. |
+    /// | notification | Safe to degrade: events are an outbound side channel; no consumer draws a durability or authorization conclusion from their absence. |
+    /// | tagging | Safe to degrade: bucket tags are cost-allocation labels here; object-level tag conditions come from object metadata, not this blob. |
+    /// | CORS | Safe to degrade: an absent CORS configuration rejects cross-origin browser requests, which is already the restrictive direction. |
+    /// | logging, website, accelerate, request payment, bucket ACL | Safe to degrade: each only shapes an optional response or an optional side channel, and none of them authorizes an action or decides whether data is retained. |
+    pub(super) fn parse_all_configs(&mut self) -> Result<()> {
         if let Err(e) = self.parse_policy_config() {
             tracing::warn!(
                 event = "bucket_metadata_parse_failed",
@@ -1088,20 +1107,26 @@ impl BucketMetadata {
                 "Failed to parse bucket metadata config"
             );
         }
+        // A stored targets blob that cannot be decoded must not collapse into
+        // the empty target set: that is indistinguishable from "no replication
+        // configured", so replication stops and no caller ever sees an error
+        // (rustfs/backlog#2282). Leaving the typed field `None` while the raw
+        // bytes stay non-empty is the retained parse failure every targets
+        // reader keys off; the bytes are preserved so the configuration is
+        // still recoverable.
+        self.bucket_target_config = None;
         if !self.bucket_targets_config_json.is_empty() {
-            if let Err(e) = serde_json::from_slice::<BucketTargets>(&self.bucket_targets_config_json)
+            match serde_json::from_slice::<BucketTargets>(&self.bucket_targets_config_json) {
-                .map(|t| self.bucket_target_config = Some(t))
+                Ok(targets) => self.bucket_target_config = Some(targets),
-            {
+                Err(e) => tracing::error!(
-                tracing::warn!(
                     event = "bucket_metadata_parse_failed",
                     component = "ecstore",
                     subsystem = "bucket_metadata",
                     bucket = %self.name,
                     config = "bucket_targets",
                     error = %e,
-                    "Failed to parse bucket metadata config"
+                    "Bucket replication targets are unreadable; replication for this bucket fails closed"
-                );
+                ),
-                self.bucket_target_config = Some(BucketTargets::default());
             }
         } else {
             self.bucket_target_config = Some(BucketTargets::default());
@@ -1535,6 +1560,145 @@ mod test {
         assert_eq!(bucket_targets.targets[0].target_bucket, "target-bucket");
     }
 
+    /// rustfs/backlog#2282: a stored targets blob this build cannot decode
+    /// must not become the empty target set, and must stay distinguishable
+    /// from a bucket that never configured a target.
+    #[test]
+    fn unreadable_bucket_targets_never_degrade_to_an_empty_target_set() {
+        let truncated = br#"{"targets":[{"endpoint":"s3.example.com","#.to_vec();
+        let mut corrupt = BucketMetadata::new("corrupt-targets");
+        corrupt.bucket_targets_config_json = truncated.clone();
+
+        corrupt
+            .parse_all_configs()
+            .expect("one unreadable sub-config must not fail the whole metadata load");
+
+        assert!(
+            corrupt.bucket_target_config.is_none(),
+            "an undecodable targets blob must not produce a target set at all"
+        );
+        assert!(corrupt.bucket_targets_unreadable());
+        assert_eq!(
+            corrupt.bucket_targets_config_json, truncated,
+            "the raw bytes must survive so the configuration stays recoverable"
+        );
+
+        // The genuinely-absent case is unchanged, and the two now diverge.
+        let mut absent = BucketMetadata::new("no-targets");
+        absent.parse_all_configs().expect("absent targets parse");
+        assert!(
+            absent.bucket_target_config.as_ref().is_some_and(BucketTargets::is_empty),
+            "a bucket that configured no target still reads as an empty target set"
+        );
+        assert!(!absent.bucket_targets_unreadable());
+    }
+
+    /// `Credentials` carries no struct-level `serde(default)`, so one target
+    /// missing `secretKey` is a hard parse error for the whole document. That
+    /// must surface as "unreadable", never as "no targets configured".
+    #[test]
+    fn bucket_targets_missing_secret_key_are_unreadable_not_empty() {
+        let mut bm = BucketMetadata::new("missing-secret-key");
+        bm.bucket_targets_config_json = br#"{"targets":[{"endpoint":"s3.example.com","targetbucket":"remote","arn":"arn:rustfs:replication:us-east-1:src:1","credentials":{"accessKey":"AKIAEXAMPLE"}}]}"#.to_vec();
+
+        bm.parse_all_configs()
+            .expect("a rejected targets document must not fail the whole metadata load");
+
+        assert!(
+            bm.bucket_targets_unreadable(),
+            "a targets document rejected for a missing secretKey is unreadable, not empty"
+        );
+        assert!(bm.bucket_target_config.is_none());
+    }
+
+    /// rustfs/backlog#2309: the MinIO-origin `.metadata.bin` this repository
+    /// already carries as a compatibility fixture stores
+    /// `BucketTargetsConfigJSON` as a bare JSON array, which `BucketTargets`
+    /// (a `{"targets":[…]}` struct with no array fallback) cannot decode. The
+    /// bytes below are the exact payload the fixture in
+    /// `metadata_test.rs::TEST_BUCKET_METADATA_HEX` decodes to, so if RustFS
+    /// ever grows the array-shaped compatibility parse, this test is where the
+    /// upgrade break is pinned and where the decision has to be recorded.
+    #[test]
+    fn minio_array_shaped_bucket_targets_are_unreadable() {
+        let minio_array = br#"[{"endpoint":"http://target.example.com","targetBucket":"tb","region":"us-east-1"}]"#.to_vec();
+        let mut bm = BucketMetadata::new("minio-array-targets");
+        bm.bucket_targets_config_json = minio_array.clone();
+
+        bm.parse_all_configs()
+            .expect("a MinIO-shaped targets blob must not fail the whole metadata load");
+
+        assert!(
+            bm.bucket_targets_unreadable(),
+            "an array-shaped MinIO targets blob is unreadable, not an empty target set"
+        );
+        assert!(bm.bucket_target_config.is_none());
+        assert_eq!(
+            bm.bucket_targets_config_json, minio_array,
+            "the raw MinIO bytes must survive so the configuration stays recoverable"
+        );
+    }
+
+    /// The invariant every branch of `parse_all_configs` shares: a stored but
+    /// undecodable payload keeps its raw bytes and leaves the typed field
+    /// `None`, so no branch fabricates a value. What a reader may then do with
+    /// that state is decided per config; see the table on `parse_all_configs`.
+    #[test]
+    fn every_config_branch_retains_its_parse_failure_instead_of_defaulting() {
+        let malformed_xml = b"<not-a-valid-document".to_vec();
+        let malformed_json = b"{not-json".to_vec();
+
+        let mut bm = BucketMetadata::new("all-configs-malformed");
+        bm.policy_config_json = malformed_json.clone();
+        bm.quota_config_json = malformed_json.clone();
+        bm.bucket_targets_config_json = malformed_json.clone();
+        bm.notification_config_xml = malformed_xml.clone();
+        bm.lifecycle_config_xml = malformed_xml.clone();
+        bm.object_lock_config_xml = malformed_xml.clone();
+        bm.versioning_config_xml = malformed_xml.clone();
+        bm.encryption_config_xml = malformed_xml.clone();
+        bm.tagging_config_xml = malformed_xml.clone();
+        bm.replication_config_xml = malformed_xml.clone();
+        bm.cors_config_xml = malformed_xml.clone();
+        bm.logging_config_xml = malformed_xml.clone();
+        bm.website_config_xml = malformed_xml.clone();
+        bm.accelerate_config_xml = malformed_xml.clone();
+        bm.request_payment_config_xml = malformed_xml.clone();
+        bm.public_access_block_config_xml = malformed_xml.clone();
+        // `bucket_acl_config_json` is only checked for UTF-8, so only invalid
+        // UTF-8 exercises its failure branch.
+        bm.bucket_acl_config_json = vec![0xff, 0xfe];
+
+        bm.parse_all_configs()
+            .expect("a bucket whose every config is corrupt must still load its metadata");
+
+        let cleared: [(&str, bool); 17] = [
+            ("policy", bm.policy_config.is_none()),
+            ("quota", bm.quota_config.is_none()),
+            ("bucket_targets", bm.bucket_target_config.is_none()),
+            ("notification", bm.notification_config.is_none()),
+            ("lifecycle", bm.lifecycle_config.is_none()),
+            ("object_lock", bm.object_lock_config.is_none()),
+            ("versioning", bm.versioning_config.is_none()),
+            ("encryption", bm.sse_config.is_none()),
+            ("tagging", bm.tagging_config.is_none()),
+            ("replication", bm.replication_config.is_none()),
+            ("cors", bm.cors_config.is_none()),
+            ("logging", bm.logging_config.is_none()),
+            ("website", bm.website_config.is_none()),
+            ("accelerate", bm.accelerate_config.is_none()),
+            ("request_payment", bm.request_payment_config.is_none()),
+            ("public_access_block", bm.public_access_block_config.is_none()),
+            ("bucket_acl", bm.bucket_acl_config.is_none()),
+        ];
+        for (config, is_cleared) in cleared {
+            assert!(is_cleared, "{config}: a corrupt payload must not be replaced by a default");
+        }
+
+        assert_eq!(bm.bucket_targets_config_json, malformed_json, "raw bytes are retained");
+        assert_eq!(bm.lifecycle_config_xml, malformed_xml, "raw bytes are retained");
+    }
+
     #[test]
     fn lifecycle_update_config_clears_parsed_config_on_delete() {
         let mut bm = BucketMetadata::new("test-bucket");
@@ -1824,51 +1988,30 @@ mod test {
 
     const ODM_JSON: &[u8] = br#"{"version":1,"enabled":true,"source":{"provider":"minio","endpoint":"https://legacy.example.com:9000","region":"auto","bucket":"legacy-bucket","credentials":{"access_key":"AK","secret_key":"SK"}}}"#;
 
-    /// rustfs/backlog#2148: the on-demand migration config is a RustFS
+    /// The metadata codec preserves application-owned bytes and timestamps.
-    /// extension entry that round-trips through `update_config` and the
-    /// msgpack codec, clears on delete, and never parses corruption into a
-    /// default.
     #[test]
     fn on_demand_migration_config_round_trips_and_tracks_updates() {
-        use crate::bucket::on_demand_migration::{OnDemandMigrationConfig, OnDemandMigrationConfigError};
-
         let mut bm = BucketMetadata::new("odm-bucket");
-        assert_eq!(bm.on_demand_migration_config(), Ok(None), "fresh metadata carries no config");
+        assert_eq!(bm.on_demand_migration_config(), None, "fresh metadata carries no config");
-
-        let expected = OnDemandMigrationConfig::from_json(ODM_JSON).unwrap();
         bm.update_config(BUCKET_ON_DEMAND_MIGRATION_CONFIG, ODM_JSON.to_vec())
-            .expect("valid config is accepted");
+            .expect("opaque config is accepted");
-        assert_ne!(bm.on_demand_migration_config_updated_at, OffsetDateTime::UNIX_EPOCH);
+        let stamped = bm.on_demand_migration_config_updated_at;
-        assert_eq!(bm.on_demand_migration_config(), Ok(Some(expected.clone())));
+        assert_ne!(stamped, OffsetDateTime::UNIX_EPOCH);
-
+        assert_eq!(bm.on_demand_migration_config(), Some((ODM_JSON, stamped)));
         let back = BucketMetadata::unmarshal(&bm.marshal_msg().unwrap()).unwrap();
         assert_eq!(back.on_demand_migration_config_json, bm.on_demand_migration_config_json);
-        assert_eq!(
+        assert_eq!(back.on_demand_migration_config_updated_at.unix_timestamp(), stamped.unix_timestamp());
-            back.on_demand_migration_config_updated_at.unix_timestamp(),
-            bm.on_demand_migration_config_updated_at.unix_timestamp()
-        );
-        assert_eq!(back.on_demand_migration_config(), Ok(Some(expected)));
-
-        // A blob this build cannot read is rejected at the write boundary
-        // rather than persisted for every reader to trip over.
-        let before = bm.on_demand_migration_config_json.clone();
-        assert!(
-            bm.update_config(BUCKET_ON_DEMAND_MIGRATION_CONFIG, br#"{"source":{"provider":"s3"},"bogus":1}"#.to_vec())
-                .is_err()
-        );
-        assert_eq!(bm.on_demand_migration_config_json, before, "a rejected update leaves the blob untouched");
-
-        // Delete clears the entry.
-        let stamped = bm.on_demand_migration_config_updated_at;
         bm.update_config(BUCKET_ON_DEMAND_MIGRATION_CONFIG, Vec::new()).unwrap();
         assert!(bm.on_demand_migration_config_json.is_empty());
-        assert_eq!(bm.on_demand_migration_config(), Ok(None));
+        assert_eq!(bm.on_demand_migration_config(), None);
         assert!(bm.on_demand_migration_config_updated_at >= stamped);
-
+        bm.update_config(BUCKET_ON_DEMAND_MIGRATION_CONFIG, b"not-json".to_vec())
-        // Corruption that bypassed `update_config` (disk, another writer)
+            .unwrap();
-        // is a typed error, never a default.
+        let back = BucketMetadata::unmarshal(&bm.marshal_msg().unwrap()).unwrap();
-        bm.on_demand_migration_config_json = b"not-json".to_vec();
+        assert_eq!(
-        assert!(matches!(bm.on_demand_migration_config(), Err(OnDemandMigrationConfigError::Malformed(_))));
+            back.on_demand_migration_config_json, b"not-json",
+            "metadata must not reinterpret application bytes"
+        );
     }
 
     /// rustfs/backlog#2148: a `.metadata.bin` written before the on-demand
@@ -1880,7 +2023,7 @@ mod test {
         let mut bm = BucketMetadata::unmarshal(&blob[4..]).expect("unmarshal MinIO bucket metadata");
         assert!(bm.on_demand_migration_config_json.is_empty());
         assert_eq!(bm.on_demand_migration_config_updated_at, OffsetDateTime::UNIX_EPOCH);
-        assert_eq!(bm.on_demand_migration_config(), Ok(None));
+        assert_eq!(bm.on_demand_migration_config(), None);
 
         bm.default_timestamps();
         assert_ne!(bm.created, OffsetDateTime::UNIX_EPOCH, "fixture must carry a real creation time");

crates/ecstore/src/bucket/metadata_sys.rs

@@ -19,7 +19,6 @@ use super::quota::BucketQuota;
 use super::target::BucketTargets;
 use crate::bucket::bucket_target_sys::BucketTargetSys;
 use crate::bucket::metadata::{load_bucket_metadata_parse, load_bucket_metadata_parse_with_presence};
-use crate::bucket::on_demand_migration::{ON_DEMAND_MIGRATION_CONFIG_HOOK, OnDemandMigrationConfig};
 use crate::bucket::utils::is_meta_bucketname;
 use crate::disk::RUSTFS_META_BUCKET;
 use crate::error::{Error, Result, is_err_bucket_not_found, is_err_strict_volume_not_found};
@@ -49,6 +48,11 @@ use tokio_util::sync::CancellationToken;
 use tracing::{error, warn};
 use uuid::Uuid;
 
+/// Opaque bucket configuration notifications for application-owned services.
+/// `None` withdraws a configuration; consumers validate nonempty bytes.
+pub type BucketConfigPublishHook = Box<dyn Fn(&str, &str, Option<(&[u8], OffsetDateTime, Uuid)>) + Send + Sync>;
+pub static BUCKET_CONFIG_PUBLISH_HOOK: std::sync::OnceLock<BucketConfigPublishHook> = std::sync::OnceLock::new();
+
 const BUCKET_METADATA_REFRESH_INTERVAL: Duration = Duration::from_secs(15 * 60);
 
 #[cfg(any(test, feature = "test-util"))]
@@ -360,6 +364,16 @@ async fn refresh_buckets_metadata_once(sys: Arc<RwLock<BucketMetadataSys>>) {
 }
 
 async fn sync_bucket_target_sys(bucket: &str, bm: &BucketMetadata) {
+    if bm.bucket_targets_unreadable() {
+        // "The configuration cannot be read" is not "no targets configured".
+        // Publishing an empty snapshot here is what silently stopped
+        // replication (rustfs/backlog#2282): mark the bucket instead, so every
+        // targets reader gets a typed error, and leave any snapshot from an
+        // earlier readable load in place rather than withdrawing it.
+        BucketTargetSys::get().mark_targets_unreadable(bucket).await;
+        return;
+    }
+
     BucketTargetSys::get()
         .update_all_targets(bucket, bm.bucket_target_config.as_ref())
         .await;
@@ -385,39 +399,21 @@ fn clear_bucket_durability(bucket: &str) {
     crate::disk::local::bucket_durability::set(bucket, None);
 }
 
-/// Publish the bucket's on-demand migration config (or its absence) to the
+/// Publish application-owned bytes on every cache install path.
-/// runtime registered in `ON_DEMAND_MIGRATION_CONFIG_HOOK`.
-///
-/// Called from the same five cache-install paths as
-/// [`sync_bucket_durability`]. A stored payload this build cannot parse is
-/// published as `None`: the runtime must stop pulling for that bucket rather
-/// than keep an older config or guess.
 fn sync_on_demand_migration(bucket: &str, bm: &BucketMetadata) {
-    let Some(hook) = ON_DEMAND_MIGRATION_CONFIG_HOOK.get() else {
+    if let Some(hook) = BUCKET_CONFIG_PUBLISH_HOOK.get() {
-        return;
+        hook(
-    };
+            bucket,
-    match bm.on_demand_migration_config() {
+            super::metadata::BUCKET_ON_DEMAND_MIGRATION_CONFIG,
-        Ok(config) => hook(bucket, config.as_ref()),
+            bm.on_demand_migration_config()
-        Err(err) => {
+                .map(|(bytes, stamp)| (bytes, stamp, bm.bucket_incarnation_id)),
-            warn!(
-                event = "bucket_metadata_parse_failed",
-                component = "ecstore",
-                subsystem = "bucket_metadata",
-                bucket = %bucket,
-                config = "on_demand_migration",
-                error = %err,
-                "Failed to parse bucket metadata config"
         );
-            hook(bucket, None);
-        }
     }
 }
 
-/// Withdraw a bucket's on-demand migration config when its metadata leaves
-/// the cache.
 fn clear_on_demand_migration(bucket: &str) {
-    if let Some(hook) = ON_DEMAND_MIGRATION_CONFIG_HOOK.get() {
+    if let Some(hook) = BUCKET_CONFIG_PUBLISH_HOOK.get() {
-        hook(bucket, None);
+        hook(bucket, super::metadata::BUCKET_ON_DEMAND_MIGRATION_CONFIG, None);
     }
 }
 
@@ -645,6 +641,12 @@ pub struct BucketMetadataMutationGuard {
 }
 
 impl BucketMetadataMutationGuard {
+    /// Returns the storage-verified identity while both incarnation fences remain valid.
+    pub fn checked_bucket_incarnation(&self) -> Result<(&str, Uuid)> {
+        self.ensure_valid(&self.bucket)?;
+        Ok((&self.bucket, self.incarnation_id))
+    }
+
     fn ensure_valid(&self, bucket: &str) -> Result<()> {
         if self.bucket != bucket {
             return Err(Error::other("bucket metadata mutation guard does not match bucket"));
@@ -664,6 +666,29 @@ async fn acquire_config_write_guard_for_incarnation(
     sys: Arc<RwLock<BucketMetadataSys>>,
     bucket: &str,
     expected_incarnation_id: Option<Uuid>,
+) -> Result<BucketMetadataMutationGuard> {
+    acquire_config_write_guard_with_migration(sys, bucket, expected_incarnation_id, true).await
+}
+
+/// Scanner probes must not create an incarnation to make a capability available.
+pub async fn acquire_scanner_bucket_incarnation_fence(
+    bucket: &str,
+    expected_incarnation_id: Uuid,
+    expected_owner_id: Uuid,
+) -> Result<BucketMetadataMutationGuard> {
+    super::utils::check_valid_bucket_name(bucket)?;
+    let sys = get_bucket_metadata_sys()?;
+    if expected_owner_id.is_nil() || sys.read().await.api.id != expected_owner_id || expected_incarnation_id.is_nil() {
+        return Err(Error::other("scanner bucket incarnation owner does not match"));
+    }
+    acquire_config_write_guard_with_migration(sys, bucket, Some(expected_incarnation_id), false).await
+}
+
+async fn acquire_config_write_guard_with_migration(
+    sys: Arc<RwLock<BucketMetadataSys>>,
+    bucket: &str,
+    expected_incarnation_id: Option<Uuid>,
+    migrate: bool,
 ) -> Result<BucketMetadataMutationGuard> {
     let metadata_sys = sys.read().await.clone();
     let lifecycle_guard = metadata_sys.api.acquire_bucket_lifecycle_read_lock(bucket).await?;
@@ -671,6 +696,7 @@ async fn acquire_config_write_guard_for_incarnation(
     // Legacy buckets are migrated while the lifecycle fence prevents a
     // same-name replacement. The second read under the write transaction is
     // the CAS source of truth for the actual rewrite.
+    if migrate {
         await_bucket_namespace_operation(
             Some(&lifecycle_guard),
             bucket,
@@ -678,6 +704,7 @@ async fn acquire_config_write_guard_for_incarnation(
             metadata_sys.get_bucket_incarnation_id(bucket),
         )
         .await?;
+    }
     let transaction_guard = await_bucket_namespace_operation(
         Some(&lifecycle_guard),
         bucket,
@@ -1008,15 +1035,21 @@ pub async fn get_durability_config(
 }
 
 /// The bucket's on-demand migration config with its update time, or
-/// `Ok(None)` when the bucket has none. A stored payload that does not parse
+/// `Ok(None)` when the bucket has none. Bytes are opaque to the metadata owner.
-/// is a typed error (`OnDemandMigrationConfigError` inside `Error::Io`).
+pub async fn get_on_demand_migration_config(bucket: &str) -> Result<Option<(Vec<u8>, OffsetDateTime)>> {
-pub async fn get_on_demand_migration_config(bucket: &str) -> Result<Option<(OnDemandMigrationConfig, OffsetDateTime)>> {
     let bucket_meta_sys_lock = get_bucket_metadata_sys()?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
     bucket_meta_sys.get_on_demand_migration_config(bucket).await
 }
 
+/// Resolve opaque configuration from the store's own metadata system.
+pub async fn get_on_demand_migration_config_in(api: &ECStore, bucket: &str) -> Result<Option<(Vec<u8>, OffsetDateTime)>> {
+    let sys = bucket_metadata_sys_of(&api.ctx)?;
+    let lock = sys.read().await;
+    lock.get_on_demand_migration_config(bucket).await
+}
+
 pub async fn get_quota_config(bucket: &str) -> Result<(BucketQuota, OffsetDateTime)> {
     let bucket_meta_sys_lock = get_bucket_metadata_sys()?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
@@ -2118,7 +2151,9 @@ impl BucketMetadataSys {
     pub async fn get_public_access_block_config(&self, bucket: &str) -> Result<(PublicAccessBlockConfiguration, OffsetDateTime)> {
         let (bm, _) = self.get_config(bucket).await?;
 
-        if let Some(config) = &bm.public_access_block_config {
+        if !bm.public_access_block_config_xml.is_empty() && bm.public_access_block_config.is_none() {
+            Err(Error::other("persisted bucket public access block configuration is invalid"))
+        } else if let Some(config) = &bm.public_access_block_config {
             Ok((config.clone(), bm.public_access_block_config_updated_at))
         } else {
             Err(Error::ConfigNotFound)
@@ -2429,7 +2464,9 @@ impl BucketMetadataSys {
     pub async fn get_sse_config(&self, bucket: &str) -> Result<(ServerSideEncryptionConfiguration, OffsetDateTime)> {
         let (bm, _) = self.get_config(bucket).await?;
 
-        if let Some(config) = &bm.sse_config {
+        if !bm.encryption_config_xml.is_empty() && bm.sse_config.is_none() {
+            Err(Error::other("persisted bucket encryption configuration is invalid"))
+        } else if let Some(config) = &bm.sse_config {
             Ok((config.clone(), bm.encryption_config_updated_at))
         } else {
             Err(Error::ConfigNotFound)
@@ -2500,7 +2537,9 @@ impl BucketMetadataSys {
     pub async fn get_quota_config(&self, bucket: &str) -> Result<(BucketQuota, OffsetDateTime)> {
         let (bm, _) = self.get_config(bucket).await?;
 
-        if let Some(config) = &bm.quota_config {
+        if !bm.quota_config_json.is_empty() && bm.quota_config.is_none() {
+            Err(Error::other("persisted bucket quota configuration is invalid"))
+        } else if let Some(config) = &bm.quota_config {
             Ok((config.clone(), bm.quota_config_updated_at))
         } else {
             Err(Error::ConfigNotFound)
@@ -2522,7 +2561,9 @@ impl BucketMetadataSys {
     pub async fn get_bucket_targets_config(&self, bucket: &str) -> Result<BucketTargets> {
         let (bm, _) = self.get_config(bucket).await?;
 
-        if let Some(config) = &bm.bucket_target_config {
+        if bm.bucket_targets_unreadable() {
+            Err(Error::other("persisted bucket replication target configuration is invalid"))
+        } else if let Some(config) = &bm.bucket_target_config {
             Ok(config.clone())
         } else {
             Err(Error::ConfigNotFound)
@@ -2530,29 +2571,27 @@ impl BucketMetadataSys {
     }
 
     /// See [`get_on_demand_migration_config`].
-    pub async fn get_on_demand_migration_config(
+    pub async fn get_on_demand_migration_config(&self, bucket: &str) -> Result<Option<(Vec<u8>, OffsetDateTime)>> {
-        &self,
-        bucket: &str,
-    ) -> Result<Option<(OnDemandMigrationConfig, OffsetDateTime)>> {
         let (bm, _) = self.get_config(bucket).await?;
 
-        let config = bm.on_demand_migration_config().map_err(Error::other)?;
+        Ok(bm
-        Ok(config.map(|config| (config, bm.on_demand_migration_config_updated_at)))
+            .on_demand_migration_config()
+            .map(|(bytes, updated_at)| (bytes.to_vec(), updated_at)))
     }
 }
 
 /// Test-only fixture shared with sibling modules (e.g. the quota checker
 /// tests): a 4-disk `ECStore` on an isolated instance context, so tests
 /// exercising the metadata system never touch ambient process state.
-#[cfg(test)]
+#[cfg(any(test, feature = "test-util"))]
-pub(crate) mod test_support {
+pub mod test_support {
     use super::*;
     use crate::disk::endpoint::Endpoint;
     use crate::layout::endpoints::{EndpointServerPools, Endpoints, PoolEndpoints};
     use crate::runtime::instance::InstanceContext;
     use crate::store::init_local_disks_with_instance_ctx;
 
-    pub(crate) async fn isolated_store_over_temp_disks() -> (Vec<tempfile::TempDir>, Arc<ECStore>) {
+    pub async fn isolated_store_over_temp_disks() -> (Vec<tempfile::TempDir>, Arc<ECStore>) {
         let mut dirs = Vec::with_capacity(4);
         let mut endpoints = Vec::with_capacity(4);
         for disk_idx in 0..4 {
@@ -2593,6 +2632,7 @@ pub(crate) mod test_support {
 mod tests {
     use super::test_support::isolated_store_over_temp_disks;
     use super::*;
+    use crate::bucket::bucket_target_sys::BucketTargetError;
     use crate::bucket::metadata::{
         BUCKET_ACCELERATE_CONFIG, BUCKET_CORS_CONFIG, BUCKET_LIFECYCLE_CONFIG, BUCKET_LOGGING_CONFIG, BUCKET_NOTIFICATION_CONFIG,
         BUCKET_POLICY_CONFIG, BUCKET_PUBLIC_ACCESS_BLOCK_CONFIG, BUCKET_REPLICATION_CONFIG, BUCKET_REQUEST_PAYMENT_CONFIG,
@@ -2788,6 +2828,36 @@ mod tests {
         );
     }
 
+    /// The `parse_all_configs` audit (rustfs/backlog#2282): every accessor
+    /// whose configuration grants something — plaintext storage, anonymous
+    /// access, capacity, replication targets — reports a corrupt payload as
+    /// invalid rather than as absent, because "absent" is what grants it.
+    #[tokio::test]
+    async fn malformed_permissive_configs_are_not_reported_as_absent() {
+        let (_dirs, ecstore) = isolated_store_over_temp_disks().await;
+        let sys = BucketMetadataSys::new(ecstore);
+        let bucket = "malformed-permissive-config";
+        let mut metadata = BucketMetadata::new(bucket);
+        metadata.encryption_config_xml = b"<ServerSideEncryptionConfiguration".to_vec();
+        metadata.public_access_block_config_xml = b"<PublicAccessBlockConfiguration".to_vec();
+        metadata.quota_config_json = b"{not-json".to_vec();
+        metadata.bucket_targets_config_json = b"{not-json".to_vec();
+        metadata
+            .parse_all_configs()
+            .expect("a corrupt sub-config must not fail the load");
+        sys.set(bucket.to_string(), Arc::new(metadata)).await;
+
+        for (config, result) in [
+            ("encryption", sys.get_sse_config(bucket).await.err()),
+            ("public access block", sys.get_public_access_block_config(bucket).await.err()),
+            ("quota", sys.get_quota_config(bucket).await.err()),
+            ("bucket targets", sys.get_bucket_targets_config(bucket).await.err()),
+        ] {
+            let err = result.unwrap_or_else(|| panic!("malformed {config} metadata must not read as a value"));
+            assert_ne!(err, Error::ConfigNotFound, "malformed {config} metadata must not be reported as absent");
+        }
+    }
+
     #[tokio::test]
     async fn config_states_distinguish_authoritative_absence_from_fabricated_metadata() {
         use std::sync::atomic::Ordering;
@@ -3127,6 +3197,82 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn scoped_dirty_usage_incarnation_probe_does_not_migrate_legacy_metadata() {
+        let (dirs, store) = isolated_store_over_temp_disks().await;
+        let sys = Arc::new(RwLock::new(BucketMetadataSys::new(store.clone())));
+        let bucket = "scoped-ack-legacy";
+        for dir in &dirs {
+            std::fs::create_dir_all(dir.path().join(bucket)).expect("create legacy bucket");
+        }
+        let mut metadata = BucketMetadata::new(bucket);
+        metadata.bucket_incarnation_id = Uuid::nil();
+        sys.read()
+            .await
+            .persist_and_set(metadata)
+            .await
+            .expect("persist legacy metadata");
+        assert!(
+            acquire_config_write_guard_with_migration(sys.clone(), bucket, Some(Uuid::new_v4()), false)
+                .await
+                .is_err()
+        );
+        assert!(load_bucket_incarnation(store, bucket).await.expect("read sidecar").is_none());
+        assert!(
+            sys.read()
+                .await
+                .get_config_from_disk(bucket)
+                .await
+                .expect("read metadata")
+                .bucket_incarnation_id
+                .is_nil()
+        );
+    }
+
+    #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+    #[serial]
+    async fn scoped_dirty_usage_incarnation_rejects_deleted_and_recreated_bucket() {
+        let (_dirs, store) = isolated_store_over_temp_disks().await;
+        init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+        let sys = bucket_metadata_sys_of(&store.ctx).expect("metadata owner");
+        let bucket = "scoped-ack-recreated";
+        store
+            .make_bucket(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("create bucket");
+        let old = store.bucket_incarnation_id_from_disk(bucket).await.expect("old incarnation");
+        let guard = acquire_config_write_guard_with_migration(sys.clone(), bucket, Some(old), false)
+            .await
+            .expect("trusted incarnation fence");
+        assert_eq!(guard.checked_bucket_incarnation().expect("valid fences"), (bucket, old));
+        drop(guard);
+        store
+            .delete_bucket(bucket, &DeleteBucketOptions::default())
+            .await
+            .expect("delete bucket");
+        assert!(
+            acquire_config_write_guard_with_migration(sys.clone(), bucket, Some(old), false)
+                .await
+                .is_err()
+        );
+        store
+            .make_bucket(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("recreate bucket");
+        let new = store.bucket_incarnation_id_from_disk(bucket).await.expect("new incarnation");
+        assert_ne!(old, new);
+        assert!(
+            acquire_config_write_guard_with_migration(sys.clone(), bucket, Some(old), false)
+                .await
+                .is_err()
+        );
+        assert!(
+            acquire_config_write_guard_with_migration(sys, bucket, Some(new), false)
+                .await
+                .is_ok()
+        );
+    }
+
     #[tokio::test]
     async fn old_node_metadata_rewrite_cannot_replace_bucket_incarnation_sidecar() {
         let (dirs, ecstore) = isolated_store_over_temp_disks().await;
@@ -4066,6 +4212,114 @@ mod tests {
         target_sys.delete(bucket).await;
     }
 
+    /// rustfs/backlog#2282: an unreadable `bucket-targets.json` reaches every
+    /// targets reader as a typed error; it neither withdraws a snapshot a
+    /// previous readable load published, nor collapses into the "no targets
+    /// configured" state that a bucket with an absent configuration reports.
+    #[tokio::test]
+    #[serial]
+    async fn unreadable_bucket_targets_fail_closed_and_stay_distinct_from_absent() {
+        let (_dirs, ecstore) = isolated_store_over_temp_disks().await;
+        let sys = BucketMetadataSys::new(ecstore);
+        let target_sys = BucketTargetSys::get();
+        let unreadable = "targets-unreadable";
+        let absent = "targets-absent";
+        target_sys.delete(unreadable).await;
+        target_sys.delete(absent).await;
+
+        // A readable load publishes this bucket's targets.
+        let mut readable = BucketMetadata::new(unreadable);
+        readable.bucket_target_config = Some(BucketTargets {
+            targets: vec![target(unreadable, "live")],
+        });
+        sync_bucket_target_sys(unreadable, &readable).await;
+        assert_eq!(
+            target_sys
+                .list_bucket_targets(unreadable)
+                .await
+                .expect("readable targets publish")
+                .targets
+                .len(),
+            1
+        );
+
+        // The same bucket reloaded with a blob that cannot be decoded.
+        let mut corrupt = BucketMetadata::new(unreadable);
+        corrupt.bucket_targets_config_json = br#"{"targets":[{"endpoint":"#.to_vec();
+        corrupt
+            .parse_all_configs()
+            .expect("an unreadable targets blob must not fail the metadata load");
+        sys.set(unreadable.to_string(), Arc::new(corrupt)).await;
+
+        assert!(
+            matches!(
+                target_sys.list_bucket_targets(unreadable).await,
+                Err(BucketTargetError::BucketRemoteTargetsUnreadable { .. })
+            ),
+            "an unreadable configuration must not read as an empty or a missing target set"
+        );
+        assert!(
+            target_sys.list_targets(unreadable, "").await.is_err(),
+            "the admin listing must surface the fault instead of an empty list"
+        );
+        let err = sys
+            .get_bucket_targets_config(unreadable)
+            .await
+            .expect_err("an unreadable targets configuration must not read as a value");
+        assert_ne!(err, Error::ConfigNotFound, "unreadable must not be reported as absent");
+
+        // A bucket that never configured a target keeps its previous behavior.
+        let mut no_targets = BucketMetadata::new(absent);
+        no_targets.parse_all_configs().expect("absent targets parse");
+        sys.set(absent.to_string(), Arc::new(no_targets)).await;
+        assert!(
+            matches!(
+                target_sys.list_bucket_targets(absent).await,
+                Err(BucketTargetError::BucketRemoteTargetNotFound { .. })
+            ),
+            "an absent configuration must still report as a missing target set"
+        );
+        assert!(
+            target_sys
+                .list_targets(absent, "")
+                .await
+                .expect("an absent configuration lists no targets")
+                .is_empty()
+        );
+        assert!(
+            sys.get_bucket_targets_config(absent)
+                .await
+                .expect("an absent targets configuration still reads as an empty set")
+                .is_empty(),
+            "the absent path must keep returning an empty target set, exactly as before"
+        );
+
+        // One bucket's unreadable configuration does not reach another bucket.
+        assert!(!matches!(
+            target_sys.list_bucket_targets(absent).await,
+            Err(BucketTargetError::BucketRemoteTargetsUnreadable { .. })
+        ));
+
+        // A repaired configuration takes effect on the next load, no restart.
+        let mut repaired = BucketMetadata::new(unreadable);
+        repaired.bucket_target_config = Some(BucketTargets {
+            targets: vec![target(unreadable, "repaired")],
+        });
+        sync_bucket_target_sys(unreadable, &repaired).await;
+        assert_eq!(
+            target_sys
+                .list_bucket_targets(unreadable)
+                .await
+                .expect("a repaired configuration clears the unreadable marker")
+                .targets
+                .len(),
+            1
+        );
+
+        target_sys.delete(unreadable).await;
+        target_sys.delete(absent).await;
+    }
+
     #[tokio::test]
     #[serial]
     async fn metadata_reload_clears_stale_bucket_targets_when_config_is_removed() {
@@ -4121,19 +4375,26 @@ mod tests {
 
     const ODM_JSON: &[u8] = br#"{"source":{"provider":"minio","endpoint":"https://legacy.example.com:9000","region":"auto","bucket":"legacy-bucket","credentials":{"access_key":"AK","secret_key":"SK"}}}"#;
 
+    type RecordedOdmConfig = Option<(Vec<u8>, OffsetDateTime, Uuid)>;
+    type RecordedOdmHookCall = (String, RecordedOdmConfig);
+
     /// Every `(bucket, config)` the recording hook has seen. Tests filter by
     /// their own bucket name; the hook is process-wide and set once.
-    static ODM_HOOK_CALLS: std::sync::Mutex<Vec<(String, Option<OnDemandMigrationConfig>)>> = std::sync::Mutex::new(Vec::new());
+    static ODM_HOOK_CALLS: std::sync::Mutex<Vec<RecordedOdmHookCall>> = std::sync::Mutex::new(Vec::new());
 
     fn install_recording_odm_hook() {
-        ON_DEMAND_MIGRATION_CONFIG_HOOK.get_or_init(|| {
+        BUCKET_CONFIG_PUBLISH_HOOK.get_or_init(|| {
-            Box::new(|bucket, config| {
+            Box::new(|bucket, config_file, config| {
-                ODM_HOOK_CALLS.lock().unwrap().push((bucket.to_string(), config.cloned()));
+                assert_eq!(config_file, super::super::metadata::BUCKET_ON_DEMAND_MIGRATION_CONFIG);
+                ODM_HOOK_CALLS.lock().unwrap().push((
+                    bucket.to_string(),
+                    config.map(|(bytes, stamp, incarnation)| (bytes.to_vec(), stamp, incarnation)),
+                ));
             })
         });
     }
 
-    fn odm_hook_calls(bucket: &str) -> Vec<Option<OnDemandMigrationConfig>> {
+    fn odm_hook_calls(bucket: &str) -> Vec<RecordedOdmConfig> {
         ODM_HOOK_CALLS
             .lock()
             .unwrap()
@@ -4143,54 +4404,6 @@ mod tests {
             .collect()
     }
 
-    /// rustfs/backlog#2148: the accessor reports absence as `Ok(None)` and a
-    /// stored payload it cannot parse as a typed error, never as a default
-    /// and never as `ConfigNotFound`.
-    #[tokio::test]
-    async fn get_on_demand_migration_config_distinguishes_absent_from_corrupt() {
-        use crate::bucket::on_demand_migration::OnDemandMigrationConfigError;
-
-        let (_dirs, ecstore) = isolated_store_over_temp_disks().await;
-        let sys = BucketMetadataSys::new(ecstore);
-        let bucket = "odm-accessor";
-
-        sys.set(bucket.to_string(), Arc::new(BucketMetadata::new(bucket))).await;
-        assert_eq!(sys.get_on_demand_migration_config(bucket).await.unwrap(), None);
-
-        let mut corrupt = BucketMetadata::new(bucket);
-        corrupt.on_demand_migration_config_json = br#"{"source":{"provider":"s3"},"bogus":1}"#.to_vec();
-        sys.set(bucket.to_string(), Arc::new(corrupt)).await;
-        let err = sys
-            .get_on_demand_migration_config(bucket)
-            .await
-            .expect_err("corrupt config must not read as a default");
-        assert_ne!(err, Error::ConfigNotFound, "corruption must not be reported as absence");
-        let typed = match &err {
-            Error::Io(io) => io
-                .get_ref()
-                .and_then(|source| source.downcast_ref::<OnDemandMigrationConfigError>()),
-            _ => None,
-        };
-        assert!(
-            matches!(typed, Some(OnDemandMigrationConfigError::Malformed(_))),
-            "typed parse error must survive the Result boundary, got: {err:?}"
-        );
-
-        let mut valid = BucketMetadata::new(bucket);
-        valid
-            .update_config(crate::bucket::metadata::BUCKET_ON_DEMAND_MIGRATION_CONFIG, ODM_JSON.to_vec())
-            .unwrap();
-        let stamped = valid.on_demand_migration_config_updated_at;
-        sys.set(bucket.to_string(), Arc::new(valid)).await;
-        let (config, updated_at) = sys
-            .get_on_demand_migration_config(bucket)
-            .await
-            .unwrap()
-            .expect("stored config is returned");
-        assert_eq!(config, OnDemandMigrationConfig::from_json(ODM_JSON).unwrap());
-        assert_eq!(updated_at, stamped);
-    }
-
     /// rustfs/backlog#2148: the publish hook fires on every path that
     /// installs bucket metadata into the cache (set, initial load, peer
     /// reload, refresh loop, lazy load) and withdraws on removal, mirroring
@@ -4204,15 +4417,22 @@ mod tests {
         for dir in &dirs {
             std::fs::create_dir_all(dir.path().join(bucket)).expect("physical bucket should exist");
         }
-        let expected = OnDemandMigrationConfig::from_json(ODM_JSON).unwrap();
+
+        let incarnation = Uuid::new_v4();
         let expect_publish = |before: usize, label: &str| {
             let calls = odm_hook_calls(bucket);
             assert_eq!(calls.len(), before + 1, "{label} must publish exactly once");
-            assert_eq!(calls.last().unwrap().as_ref(), Some(&expected), "{label} must publish the stored config");
+            assert_eq!(
+                calls.last().unwrap().as_ref().map(|(bytes, _, _)| bytes.as_slice()),
+                Some(ODM_JSON),
+                "{label} must publish the stored bytes"
+            );
+            assert_eq!(calls.last().unwrap().as_ref().map(|(_, _, id)| *id), Some(incarnation));
         };
 
         // set (via persist_new_and_set, which installs through `set`).
         let mut bm = BucketMetadata::new(bucket);
+        bm.bucket_incarnation_id = incarnation;
         bm.update_config(crate::bucket::metadata::BUCKET_ON_DEMAND_MIGRATION_CONFIG, ODM_JSON.to_vec())
             .unwrap();
         let writer = BucketMetadataSys::new(ecstore.clone());
@@ -4254,14 +4474,18 @@ mod tests {
         assert_eq!(calls.len(), before + 1, "remove must withdraw exactly once");
         assert_eq!(calls.last().unwrap(), &None);
 
-        // A corrupt payload is withdrawn, never published as a config.
+        // Opaque bytes reach the application even if they are not valid JSON.
         let mut corrupt = BucketMetadata::new(bucket);
         corrupt.on_demand_migration_config_json = b"not-json".to_vec();
         let before = odm_hook_calls(bucket).len();
         lazy.set(bucket.to_string(), Arc::new(corrupt)).await;
         let calls = odm_hook_calls(bucket);
         assert_eq!(calls.len(), before + 1);
-        assert_eq!(calls.last().unwrap(), &None, "unreadable config must publish absence");
+        assert_eq!(
+            calls.last().unwrap().as_ref().map(|(bytes, _, _)| bytes.as_slice()),
+            Some(b"not-json".as_slice()),
+            "the application validates opaque config bytes"
+        );
     }
 
     #[tokio::test]

crates/ecstore/src/bucket/mod.rs

@@ -26,7 +26,6 @@ mod metadata_test;
 pub mod migration;
 mod msgp_decode;
 pub mod object_lock;
-pub mod on_demand_migration;
 pub mod policy_sys;
 pub mod quota;
 pub mod remote_s3_client;

crates/ecstore/src/bucket/remote_s3_client.rs

@@ -180,6 +180,8 @@ impl RemoteS3EndpointSpec {
 
 #[derive(Debug, thiserror::Error)]
 pub enum RemoteS3ClientError {
+    #[error("the {0} backend is not included in this build")]
+    BackendNotCompiled(&'static str),
     #[error("remote endpoint requires credentials")]
     MissingCredentials,
     #[error("{0}")]
@@ -281,9 +283,7 @@ impl Intercept for UserAgentSuffixInterceptor {
 
 /// Builds the SDK config for `spec` without finalizing it, so callers can add
 /// interceptors or (in tests) swap the HTTP client before `build()`.
-pub(crate) async fn build_remote_s3_config(
+pub async fn build_remote_s3_config(spec: &RemoteS3EndpointSpec) -> Result<aws_sdk_s3::config::Builder, RemoteS3ClientError> {
-    spec: &RemoteS3EndpointSpec,
-) -> Result<aws_sdk_s3::config::Builder, RemoteS3ClientError> {
     let Some(credentials) = &spec.credentials else {
         return Err(RemoteS3ClientError::MissingCredentials);
     };
@@ -523,7 +523,7 @@ fn validate_ca_pem_bundle(ca_cert_pem: &[u8]) -> Result<(), String> {
     Ok(())
 }
 
-pub(crate) fn validate_target_ca_pem(ca_cert_pem: &str) -> Result<(), RemoteS3ClientError> {
+pub fn validate_target_ca_pem(ca_cert_pem: &str) -> Result<(), RemoteS3ClientError> {
     validate_ca_pem_bundle(ca_cert_pem.as_bytes()).map_err(RemoteS3ClientError::InvalidCaPem)
 }
 
@@ -652,9 +652,10 @@ async fn build_aws_s3_http_client_from_tls_path() -> Option<SharedHttpClient> {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use aws_smithy_async::time::TimeSource;
     use aws_smithy_runtime_api::http::StatusCode as SmithyStatusCode;
     use std::sync::Mutex;
-    use std::sync::atomic::{AtomicUsize, Ordering};
+    use std::sync::atomic::{AtomicU64, AtomicUsize, Ordering};
 
     fn spec(endpoint: &str, secure: bool) -> RemoteS3EndpointSpec {
         RemoteS3EndpointSpec {
@@ -824,6 +825,174 @@ mod tests {
         );
     }
 
+    #[derive(Clone, Debug)]
+    struct ClockSkewTimeSource(Arc<AtomicU64>);
+
+    impl TimeSource for ClockSkewTimeSource {
+        fn now(&self) -> SystemTime {
+            SystemTime::UNIX_EPOCH + Duration::from_secs(self.0.load(Ordering::SeqCst))
+        }
+    }
+
+    #[derive(Clone, Debug)]
+    struct ClockSkewConnector {
+        request_headers: RecordedHeaders,
+        error_code: &'static str,
+        skew_seconds: i64,
+        clock: ClockSkewTimeSource,
+    }
+
+    fn recorded_header<'a>(headers: &'a [(String, String)], name: &str) -> &'a str {
+        headers
+            .iter()
+            .find(|(key, _)| key.eq_ignore_ascii_case(name))
+            .map(|(_, value)| value.as_str())
+            .unwrap_or_else(|| panic!("signed request must contain {name}"))
+    }
+
+    fn signing_time(headers: &[(String, String)]) -> chrono::NaiveDateTime {
+        chrono::NaiveDateTime::parse_from_str(recorded_header(headers, "x-amz-date"), "%Y%m%dT%H%M%SZ")
+            .expect("SDK signing timestamp must use the SigV4 format")
+    }
+
+    impl SmithyHttpConnector for ClockSkewConnector {
+        fn call(&self, request: HttpRequest) -> HttpConnectorFuture {
+            let mut headers = self.request_headers.lock().expect("clock skew request capture lock");
+            assert!(headers.len() < 3, "clock skew fixture must not exceed two GET attempts and one HEAD");
+            headers.push(
+                request
+                    .headers()
+                    .iter()
+                    .map(|(key, value)| (key.to_string(), value.to_string()))
+                    .collect(),
+            );
+            let server_time = chrono::DateTime::<chrono::Utc>::from(self.clock.now()).naive_utc()
+                + chrono::Duration::seconds(self.skew_seconds);
+            let (status, body) = if headers.len() == 1 {
+                (
+                    403,
+                    format!("<Error><Code>{}</Code><Message>Clock skew fixture</Message></Error>", self.error_code),
+                )
+            } else {
+                (200, String::new())
+            };
+            let response = http::Response::builder()
+                .status(status)
+                .header("date", server_time.format("%a, %d %b %Y %H:%M:%S GMT").to_string())
+                .header("content-type", "application/xml")
+                .header("content-length", body.len())
+                .body(SdkBody::from(body))
+                .expect("clock skew fixture response");
+            HttpConnectorFuture::ready(Ok(HttpResponse::try_from(response).expect("Smithy fixture response")))
+        }
+    }
+
+    async fn clock_skew_client(
+        error_code: &'static str,
+        skew_seconds: i64,
+        retry: RemoteS3RetryPolicy,
+    ) -> (S3Client, RecordedHeaders, ClockSkewTimeSource) {
+        let headers: RecordedHeaders = Arc::new(Mutex::new(Vec::new()));
+        let clock = ClockSkewTimeSource(Arc::new(AtomicU64::new(1_700_000_000)));
+        let connector = SharedHttpConnector::new(ClockSkewConnector {
+            request_headers: Arc::clone(&headers),
+            error_code,
+            skew_seconds,
+            clock: clock.clone(),
+        });
+        let mut spec = spec("s3.example.com", true);
+        spec.retry = retry;
+        let config = build_remote_s3_config(&spec)
+            .await
+            .expect("clock skew fixture uses the production outbound configuration")
+            .http_client(http_client_fn(move |_settings, _components| connector.clone()))
+            .time_source(clock.clone())
+            .build();
+        (S3Client::from_conf(config), headers, clock)
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn remote_s3_clock_skew_retries_resign_and_seed_next_operation() {
+        for error_code in ["RequestTimeTooSkewed", "SignatureDoesNotMatch"] {
+            for skew_seconds in [-600, 600] {
+                let (client, headers, clock) = clock_skew_client(error_code, skew_seconds, REPLICATION_TARGET_RETRY_POLICY).await;
+                let initial = chrono::DateTime::<chrono::Utc>::from(clock.now()).naive_utc();
+                client
+                    .get_object()
+                    .bucket("bucket")
+                    .key("object")
+                    .send()
+                    .await
+                    .expect("clock skew GET must retry successfully");
+                assert_eq!(
+                    headers.lock().expect("captured requests").len(),
+                    2,
+                    "{error_code}: GET needs exactly one retry"
+                );
+                clock.0.fetch_add(17, Ordering::SeqCst);
+                // SDK signing time is independent of Tokio's retry/scheduler clock.
+                tokio::time::advance(Duration::from_secs(61)).await;
+                client
+                    .head_bucket()
+                    .bucket("bucket")
+                    .send()
+                    .await
+                    .expect("subsequent HEAD must use the client's cached skew");
+                let headers = headers.lock().expect("captured signed requests");
+                assert_eq!(headers.len(), 3, "subsequent operation must succeed on its first attempt");
+                assert_eq!(signing_time(&headers[0]), initial, "the first attempt must use the injected clock");
+                assert_eq!(
+                    signing_time(&headers[1]),
+                    initial + chrono::Duration::seconds(skew_seconds),
+                    "{error_code}: retry must apply the measured offset exactly"
+                );
+                assert_eq!(
+                    signing_time(&headers[2]),
+                    initial + chrono::Duration::seconds(skew_seconds + 17),
+                    "{error_code}: the next operation must apply cached skew to the advanced signing clock"
+                );
+                let signature = |index: usize| {
+                    recorded_header(&headers[index], "authorization")
+                        .rsplit_once("Signature=")
+                        .expect("SigV4 authorization contains a signature")
+                        .1
+                };
+                assert_ne!(
+                    signature(0),
+                    signature(1),
+                    "{error_code}: retry must be signed again after adjusting its date"
+                );
+            }
+        }
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn remote_s3_clock_skew_respects_one_attempt_policy() {
+        use aws_smithy_types::error::metadata::ProvideErrorMetadata;
+
+        for error_code in ["RequestTimeTooSkewed", "SignatureDoesNotMatch"] {
+            for retry in [
+                RemoteS3RetryPolicy::Disabled,
+                RemoteS3RetryPolicy::Standard { max_attempts: 1 },
+            ] {
+                let (client, headers, _clock) = clock_skew_client(error_code, 600, retry).await;
+                let error = client
+                    .get_object()
+                    .bucket("bucket")
+                    .key("object")
+                    .send()
+                    .await
+                    .expect_err("clock skew must not override the caller's one-attempt budget");
+                assert_eq!(error.as_service_error().and_then(ProvideErrorMetadata::code), Some(error_code));
+                assert_eq!(
+                    headers.lock().expect("captured requests").len(),
+                    1,
+                    "{error_code}: {retry:?} must send exactly one request"
+                );
+            }
+        }
+    }
+
     #[test]
     fn path_style_auto_and_path_force_path_style() {
         assert!(PathStyle::Auto.force_path_style());

crates/ecstore/src/bucket/replication/replication_pool.rs

@@ -46,7 +46,7 @@ use super::replication_storage_boundary::{
     HTTPPreconditions, ObjectInfo, ObjectOptions, ObjectToDelete, ReplicationDeletedObject, ReplicationObjectIO,
     ReplicationStorage,
 };
-use super::replication_target_boundary::{ReplicationTargetStore, replication_object_is_ssec_encrypted};
+use super::replication_target_boundary::{BucketTargetError, ReplicationTargetStore, replication_object_is_ssec_encrypted};
 use super::replication_versioning_boundary::ReplicationVersioningStore;
 use super::runtime_boundary as runtime_sources;
 use futures_util::stream::{self, StreamExt};
@@ -3084,6 +3084,23 @@ pub async fn queue_replication_heal(bucket: &str, oi: ObjectInfo, retry_count: u
 
     let tgts = match ReplicationTargetStore::list_bucket_targets(bucket).await {
         Ok(targets) => Some(targets),
+        // A bucket whose persisted target configuration cannot be decoded has
+        // an unknown target set, not an empty one: scheduling against `None`
+        // here would drop every heal for it without a trace
+        // (rustfs/backlog#2282). Report it missed so the object is retried
+        // once the configuration is readable again.
+        Err(BucketTargetError::BucketRemoteTargetsUnreadable { .. }) => {
+            warn!(
+                event = EVENT_REPLICATION_CONFIG_LOOKUP_SKIPPED,
+                component = LOG_COMPONENT_ECSTORE,
+                subsystem = LOG_SUBSYSTEM_REPLICATION,
+                bucket,
+                reason = "target_config_unreadable",
+                "Bucket replication targets are unreadable; replication heal queue fails closed"
+            );
+
+            return ReplicationQueueAdmission::Missed;
+        }
         Err(err) => {
             debug!(
                 event = EVENT_REPLICATION_CONFIG_LOOKUP_SKIPPED,

crates/ecstore/src/bucket/replication/replication_target_boundary.rs

@@ -15,7 +15,8 @@
 use std::collections::HashMap;
 use std::sync::Arc;
 
-use crate::bucket::bucket_target_sys::{BucketTargetError, BucketTargetSys};
+pub(crate) use crate::bucket::bucket_target_sys::BucketTargetError;
+use crate::bucket::bucket_target_sys::BucketTargetSys;
 use aws_sdk_s3::operation::head_object::HeadObjectOutput;
 use aws_sdk_s3::types::{ObjectLockLegalHoldStatus, ObjectLockRetentionMode};
 use http::HeaderMap;

crates/ecstore/src/bucket/sealed_credentials.rs

@@ -203,7 +203,7 @@ mod tests {
     use parking_lot::Mutex;
     use std::collections::BTreeMap;
 
-    fn encode_context(context: &HashMap<String, String>) -> String {
+    fn encode_context(context: &BTreeMap<String, String>) -> String {
         let ordered = context.iter().collect::<BTreeMap<_, _>>();
         serde_json::to_string(&ordered).expect("context serializes")
     }

crates/ecstore/src/cluster/rpc/client.rs

@@ -30,6 +30,7 @@ use rustfs_protos::{
     ChannelClass, create_new_channel, get_channel_for_class,
     proto_gen::node_service::{
         heal_control_service_client::HealControlServiceClient, node_service_client::NodeServiceClient,
+        scanner_control_service_client::ScannerControlServiceClient,
         tier_mutation_control_service_client::TierMutationControlServiceClient,
     },
 };
@@ -60,6 +61,24 @@ pub async fn node_service_time_out_client(
     node_service_time_out_client_for_class(addr, interceptor, ChannelClass::Control).await
 }
 
+pub(crate) async fn scanner_control_time_out_client(
+    addr: &str,
+    interceptor: TonicInterceptor,
+) -> crate::error::Result<ScannerControlServiceClient<InterceptedService<AuthenticatedChannel, TonicInterceptor>>> {
+    let interceptor = interceptor.with_rpc_audience(addr)?;
+    let channel = match runtime_sources::cached_node_channel(addr).await {
+        Some(channel) => channel,
+        None => create_new_channel(addr)
+            .await
+            .map_err(|err| crate::error::Error::other(err.to_string()))?,
+    };
+    let channel = ReplayScopeChannel::new(channel, interceptor.replay_scope_audience());
+    let limit = rustfs_protos::scoped_dirty_usage::SCOPED_DIRTY_USAGE_MAX_REQUEST_BYTES as usize;
+    Ok(ScannerControlServiceClient::with_interceptor(channel, interceptor)
+        .max_decoding_message_size(limit)
+        .max_encoding_message_size(limit))
+}
+
 pub async fn heal_control_time_out_client(
     addr: &str,
     interceptor: TonicInterceptor,

crates/ecstore/src/cluster/rpc/peer_rest_client.rs

@@ -2050,6 +2050,53 @@ impl PeerRestClient {
         .await
     }
 
+    /// Probe only: scoped ACK production requires a durable per-bucket proof.
+    pub async fn scanner_scoped_dirty_usage_capability(
+        &self,
+        owner_id: String,
+        instance_id: String,
+        entries: Vec<rustfs_protos::proto_gen::node_service::ScannerScopedDirtyUsageEntry>,
+    ) -> Result<bool> {
+        use rustfs_protos::scoped_dirty_usage::*;
+        let payload = rustfs_protos::proto_gen::node_service::ScannerScopedDirtyUsageAckRequest {
+            challenge: Uuid::new_v4().as_bytes().to_vec().into(),
+            protocol_version: SCOPED_DIRTY_USAGE_PROTOCOL_VERSION,
+            owner_id,
+            instance_id,
+            scope: SCOPED_DIRTY_USAGE_BUCKET_SCOPE,
+            probe_only: true,
+            entries,
+        };
+        let canonical = canonical_scoped_dirty_usage_request(&payload).map_err(|err| Error::other(err.to_string()))?;
+        self.finalize_result(
+            async {
+                let mut client = super::client::scanner_control_time_out_client(
+                    &self.grid_host,
+                    TonicInterceptor::Signature(gen_tonic_signature_interceptor()),
+                )
+                .await?;
+                let mut request = Request::new(payload.clone());
+                set_tonic_canonical_body_digest(&mut request, &canonical)?;
+                let response = client.scanner_scoped_dirty_usage_ack(request).await?.into_inner();
+                let body = canonical_scoped_dirty_usage_response(&canonical, &response)
+                    .map_err(|_| Error::other("scoped dirty usage capability response is too large"))?;
+                verify_tonic_rpc_response_proof(&body, response.response_proof.as_ref())?;
+                if response.protocol_version != SCOPED_DIRTY_USAGE_PROTOCOL_VERSION
+                    || response.owner_id != payload.owner_id
+                    || response.instance_id != payload.instance_id
+                    || response.max_entries != SCOPED_DIRTY_USAGE_MAX_ENTRIES
+                    || response.max_request_bytes != SCOPED_DIRTY_USAGE_MAX_REQUEST_BYTES
+                    || response.cleared != 0
+                {
+                    return Err(Error::other("scoped dirty usage capability response does not match request"));
+                }
+                Ok(response.supported)
+            }
+            .await,
+        )
+        .await
+    }
+
     pub async fn acknowledge_scanner_dirty_usage(&self, instance_id: String, generation: u64) -> Result<ScannerPeerActivity> {
         let result = self
             .scanner_activity_request_with_protocol(instance_id.clone(), generation, SCANNER_ACTIVITY_PROTOCOL_VERSION)

crates/ecstore/src/core/pools.rs

@@ -3013,7 +3013,7 @@ fn parse_decommission_durable_ilm_receipt_path(path: &str) -> Result<Decommissio
         .ok_or_else(|| Error::other(format!("durable ILM receipt path `{path}` is missing its record id")))?;
     let id_kind = parts
         .next()
-        .filter(|id_kind| matches!(*id_kind, "operation_id" | "transaction_id" | "job_id"))
+        .filter(|id_kind| matches!(*id_kind, "operation_id" | "transaction_id" | "job_id" | "control_id"))
         .ok_or_else(|| Error::other(format!("durable ILM receipt path `{path}` has an invalid id kind")))?;
     let source_path = parts
         .next()
@@ -3023,8 +3023,9 @@ fn parse_decommission_durable_ilm_receipt_path(path: &str) -> Result<Decommissio
         return Err(Error::other(format!("durable ILM receipt path `{path}` has an invalid run token")));
     }
     match id_kind {
-        "operation_id" if !is_sha256_checksum(id) => {
+        "operation_id" | "control_id" if !is_sha256_checksum(id) => {
-            return Err(Error::other(format!("durable ILM receipt path `{path}` has an invalid operation id")));
+            let id_label = id_kind.trim_end_matches("_id");
+            return Err(Error::other(format!("durable ILM receipt path `{path}` has an invalid {id_label} id")));
         }
         "transaction_id" | "job_id" if uuid::Uuid::parse_str(id).is_err() => {
             return Err(Error::other(format!("durable ILM receipt path `{path}` has an invalid UUID")));
@@ -5493,6 +5494,7 @@ where
     fence.ensure_held()?;
     let mut opts = ObjectOptions {
         max_parity: true,
+        write_completion: crate::object_api::WriteCompletion::TailDrained,
         no_lock: true,
         http_preconditions: Some(pool_meta_cas_preconditions(token, object)?),
         ..Default::default()
@@ -14412,6 +14414,7 @@ impl ECStore {
                 encoded.clone(),
                 &ObjectOptions {
                     max_parity: true,
+                    write_completion: crate::object_api::WriteCompletion::TailDrained,
                     http_preconditions: Some(HTTPPreconditions {
                         if_none_match: Some("*".to_string()),
                         ..Default::default()
@@ -14566,6 +14569,7 @@ impl ECStore {
                 encoded,
                 &ObjectOptions {
                     max_parity: true,
+                    write_completion: crate::object_api::WriteCompletion::TailDrained,
                     http_preconditions: Some(http_preconditions),
                     ..Default::default()
                 },
@@ -14957,6 +14961,7 @@ impl ECStore {
                 encoded,
                 &ObjectOptions {
                     max_parity: true,
+                    write_completion: crate::object_api::WriteCompletion::TailDrained,
                     http_preconditions: Some(HTTPPreconditions {
                         if_match: Some(etag),
                         ..Default::default()
@@ -19296,8 +19301,8 @@ mod pools_tests {
         load_decommission_entry_versions, local_decommission_queue_prefix, mark_decommission_bucket_done,
         merge_decommission_durable_ilm_receipts, merge_pool_meta_updates_for_save, merge_pool_status_refresh,
         missing_decommission_worker_prefix, next_decommission_capacity_generation, observe_decommission_terminal_reload_result,
-        pool_meta_has_active_decommission, publish_pool_meta_updates, read_pool_meta_replica,
+        parse_decommission_durable_ilm_receipt_path, pool_meta_has_active_decommission, publish_pool_meta_updates,
-        reconcile_decommission_meta_buckets, reconcile_decommission_unresolved_entries_for_completion,
+        read_pool_meta_replica, reconcile_decommission_meta_buckets, reconcile_decommission_unresolved_entries_for_completion,
         record_decommission_unresolved_entry, recover_decommission_capacity_reservations,
         renew_decommission_capacity_reservation, require_decommission_store, reserve_decommission_start_cancelers,
         reserve_decommission_start_target_capacity, resolve_decommission_bucket_state,
@@ -20165,6 +20170,26 @@ mod pools_tests {
         assert!(!old_receipt.starts_with(&decommission_durable_ilm_receipt_run_prefix(&second_token)));
     }
 
+    #[test]
+    fn decommission_recovery_control_receipt_path_round_trips() {
+        let run_token = "b".repeat(64);
+        let control_id = "a".repeat(64);
+        let source_path = format!(
+            "ilm/recovery-controls/transition_transaction/{}/{}/{}.json",
+            &control_id[..2],
+            &control_id[2..4],
+            control_id
+        );
+        let path = decommission_durable_ilm_receipt_path(&run_token, &source_path, "control_id", &control_id);
+
+        let locator = parse_decommission_durable_ilm_receipt_path(&path).expect("recovery control receipt path should parse");
+
+        assert_eq!(locator.run_token, run_token);
+        assert_eq!(locator.source_path, source_path);
+        assert_eq!(locator.id_kind, "control_id");
+        assert_eq!(locator.id, control_id);
+    }
+
     #[test]
     fn decommission_receipt_merge_preserves_terminal_proof() {
         let operation_id = "a".repeat(64);

crates/ecstore/src/diagnostics/get.rs

@@ -99,11 +99,17 @@ pub(crate) const GET_STAGE_READER_OPEN_MMAP_COPY_FALLBACK: &str = "reader_open_m
 pub(crate) const GET_STAGE_READER_OPEN_MMAP_COPY_SUCCESS: &str = "reader_open_mmap_copy_success";
 pub(crate) const GET_STAGE_READER_OPEN_STREAM: &str = "reader_open_stream";
 pub(crate) const GET_STAGE_READER_MMAP_ACCESS_CHECK: &str = "reader_mmap_access_check";
+#[cfg(unix)]
 pub(crate) const GET_STAGE_READER_MMAP_BLOCKING_TASK: &str = "reader_mmap_blocking_task";
+#[cfg(unix)]
 pub(crate) const GET_STAGE_READER_MMAP_BLOCKING_WAIT: &str = "reader_mmap_blocking_wait";
+#[cfg(unix)]
 pub(crate) const GET_STAGE_READER_MMAP_COPY_BUFFER: &str = "reader_mmap_copy_buffer";
+#[cfg(unix)]
 pub(crate) const GET_STAGE_READER_MMAP_DIRECT_READ_COPY: &str = "reader_mmap_direct_read_copy";
+#[cfg(unix)]
 pub(crate) const GET_STAGE_READER_MMAP_FILE_OPEN: &str = "reader_mmap_file_open";
+#[cfg(unix)]
 pub(crate) const GET_STAGE_READER_MMAP_MAP: &str = "reader_mmap_map";
 pub(crate) const GET_STAGE_READER_MMAP_METADATA_LOOKUP: &str = "reader_mmap_metadata_lookup";
 pub(crate) const GET_STAGE_READER_MMAP_METADATA_VALIDATE: &str = "reader_mmap_metadata_validate";

crates/ecstore/src/disk/disk_store.rs

@@ -317,6 +317,74 @@ impl DiskStoreRenameDataExt for LocalDiskWrapper {
         dst_path: &str,
         external_guard: Option<Arc<dyn Send + Sync>>,
     ) -> Result<RenameDataResp> {
+        self.rename_data_observed(src_volume, src_path, fi, dst_volume, dst_path, external_guard)
+            .await
+            .result
+    }
+}
+
+impl LocalDiskWrapper {
+    pub(in crate::disk) async fn undo_write_with_namespace_owner(
+        &self,
+        volume: &str,
+        path: &str,
+        fi: FileInfo,
+        opts: DeleteOptions,
+        namespace_owner: Option<Arc<dyn Send + Sync>>,
+    ) -> Result<()> {
+        self.track_disk_health_mutation(
+            "delete_version",
+            DiskMetricMutation::Delete,
+            || async {
+                // Preserve the old DiskAPI future's boxing boundary.
+                Box::pin(
+                    self.disk
+                        .undo_write_with_namespace_owner(volume, path, fi, opts, namespace_owner),
+                )
+                .await
+            },
+            get_max_timeout_duration(),
+        )
+        .await
+    }
+
+    pub(in crate::disk) async fn rename_data_observed(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+        external_guard: Option<Arc<dyn Send + Sync>>,
+    ) -> super::RenameDataObservation {
+        self.rename_data_observed_with_guards(
+            src_volume,
+            src_path,
+            fi,
+            dst_volume,
+            dst_path,
+            super::RenameDataGuards {
+                external_guard,
+                ..Default::default()
+            },
+        )
+        .await
+    }
+
+    pub(in crate::disk) async fn rename_data_observed_with_guards(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+        guards: super::RenameDataGuards,
+    ) -> super::RenameDataObservation {
+        let super::RenameDataGuards {
+            external_guard,
+            namespace_owner,
+            ..
+        } = guards;
         let operation = self.clone();
         let src_volume = src_volume.to_owned();
         let src_path = src_path.to_owned();
@@ -333,22 +401,37 @@ impl DiskStoreRenameDataExt for LocalDiskWrapper {
         } else {
             get_max_timeout_duration()
         };
-        run_owned_mutation(external_guard, move || async move {
+        let observed = run_owned_mutation(external_guard, move || async move {
-            operation
+            let mut preflight_rejection = None;
+            let result = operation
                 .track_disk_health_mutation(
                     "rename_data",
                     DiskMetricMutation::Write,
                     || async {
-                        operation
+                        // Preserve the former DiskAPI future's single boxing boundary.
-                            .disk
+                        let observed = Box::pin(operation.disk.rename_data_observed(
-                            .rename_data_borrowed(&src_volume, &src_path, &fi, &dst_volume, &dst_path)
+                            &src_volume,
-                            .await
+                            &src_path,
+                            &fi,
+                            &dst_volume,
+                            &dst_path,
+                            namespace_owner,
+                        ))
+                        .await;
+                        preflight_rejection = observed.preflight_rejection;
+                        observed.result
                     },
                     timeout_duration,
                 )
-                .await
+                .await;
+            // Health tracking must observe the real disk error, not an Ok tuple.
+            Ok(super::RenameDataObservation {
+                result,
+                preflight_rejection,
             })
-        .await
+        })
+        .await;
+        observed.unwrap_or_else(|error| super::RenameDataObservation::unknown(Err(error)))
     }
 }
 
@@ -1272,6 +1355,7 @@ impl LocalDiskWrapper {
         self.disk.get_object_path(volume, path)
     }
 
+    #[cfg(unix)]
     pub(crate) fn get_object_path_for_io(&self, volume: &str, path: &str) -> crate::disk::error::Result<std::path::PathBuf> {
         self.disk.get_object_path_for_io(volume, path)
     }
@@ -2588,6 +2672,46 @@ mod tests {
         assert_eq!(wrapper.metrics_snapshot().api_calls.get("unknown"), Some(&1));
     }
 
+    #[tokio::test]
+    async fn rename_preflight_evidence_preserves_health_errors_and_owned_reply() {
+        for source_exists in [false, true] {
+            for guarded in [false, true] {
+                let dir = tempfile::tempdir().expect("temp dir should be created");
+                let endpoint = Endpoint::try_from(dir.path().to_str().expect("temp dir should be valid UTF-8"))
+                    .expect("endpoint should parse");
+                let disk = Arc::new(LocalDisk::new(&endpoint, false).await.expect("local disk should be created"));
+                if source_exists {
+                    disk.make_volume("source").await.expect("source volume should exist");
+                }
+                let wrapper = LocalDiskWrapper::new(disk, false);
+                let drops = Arc::new(std::sync::atomic::AtomicUsize::new(0));
+                let external_guard = guarded.then(|| Arc::new(DropProbe(Arc::clone(&drops))) as Arc<dyn Send + Sync>);
+                let mut file_info = FileInfo::new("object", 1, 0);
+                file_info.mod_time = Some(::time::OffsetDateTime::now_utc());
+                file_info.erasure.index = 1;
+                let observed = wrapper
+                    .rename_data_observed("source", "object", &file_info, "missing-destination", "object", external_guard)
+                    .await;
+                assert!(observed.rejected_before_publication(), "normal access rejection must carry proof");
+                assert!(matches!(observed.result, Err(DiskError::VolumeNotFound)));
+                let snapshot = wrapper.metrics_snapshot();
+                assert_eq!(snapshot.api_calls.get("rename_data"), Some(&1));
+                assert_eq!(snapshot.total_writes, 0, "health tracking must not observe the rejection as Ok");
+                assert_eq!(drops.load(Ordering::SeqCst), usize::from(guarded));
+
+                wrapper.health.force_runtime_state_for_test(RuntimeDriveHealthState::Offline);
+                let observed = wrapper
+                    .rename_data_observed("source", "object", &file_info, "missing-destination", "object", None)
+                    .await;
+                assert!(!observed.rejected_before_publication(), "wrapper errors carry no local preflight proof");
+                assert!(matches!(observed.result, Err(DiskError::FaultyDisk)));
+                let snapshot = wrapper.metrics_snapshot();
+                assert_eq!(snapshot.total_errors_availability, 1);
+                assert_eq!(snapshot.total_writes, 0);
+            }
+        }
+    }
+
     #[tokio::test]
     async fn local_disk_health_wrapper_counts_returned_availability_errors() {
         let dir = tempfile::tempdir().expect("temp dir should be created");

crates/ecstore/src/disk/fs.rs

@@ -218,10 +218,12 @@ pub async fn rename(from: impl AsRef<Path>, to: impl AsRef<Path>) -> io::Result<
     fs::rename(from, to).await
 }
 
+#[cfg(any(not(windows), test))]
 pub fn rename_std(from: impl AsRef<Path>, to: impl AsRef<Path>) -> io::Result<()> {
     std::fs::rename(from, to)
 }
 
+#[cfg(any(not(windows), test))]
 #[tracing::instrument(level = "debug", skip_all)]
 pub async fn read_file(path: impl AsRef<Path>) -> io::Result<Vec<u8>> {
     fs::read(path.as_ref()).await

crates/ecstore/src/disk/mod.rs

@@ -75,6 +75,33 @@ use time::OffsetDateTime;
 use tokio::io::{AsyncRead, AsyncWrite};
 use uuid::Uuid;
 
+/// Independent admission and physical ownership for one disk rename.
+#[derive(Default)]
+pub(crate) struct RenameDataGuards {
+    pub(crate) scanner_publication_lease_token: Option<Uuid>,
+    pub(crate) external_guard: Option<Arc<dyn Send + Sync>>,
+    pub(crate) namespace_owner: Option<Arc<dyn Send + Sync>>,
+}
+
+/// Local preflight evidence stays outside DiskAPI and the RPC response format.
+pub(crate) struct RenameDataObservation {
+    pub(crate) result: Result<RenameDataResp>,
+    preflight_rejection: Option<local::LocalRenamePreflightRejection>,
+}
+
+impl RenameDataObservation {
+    fn unknown(result: Result<RenameDataResp>) -> Self {
+        Self {
+            result,
+            preflight_rejection: None,
+        }
+    }
+
+    pub(crate) fn rejected_before_publication(&self) -> bool {
+        self.result.is_err() && self.preflight_rejection.is_some()
+    }
+}
+
 const QUOTA_MUTATION_FENCE_PREFIX: &str = "tmp/quota-mutation-fences/";
 pub(crate) const QUOTA_MUTATION_FENCE_METADATA_SUFFIX: &str = "quota-mutation-fence-token";
 
@@ -171,11 +198,17 @@ pub struct MmapCopyStageMetrics {
     pub(crate) path_resolve_stage: &'static str,
     pub(crate) metadata_lookup_stage: &'static str,
     pub(crate) metadata_validate_stage: &'static str,
+    #[cfg(unix)]
     pub(crate) blocking_wait_stage: &'static str,
+    #[cfg(unix)]
     pub(crate) blocking_task_stage: &'static str,
+    #[cfg(unix)]
     pub(crate) file_open_stage: &'static str,
+    #[cfg(unix)]
     pub(crate) mmap_map_stage: &'static str,
+    #[cfg(unix)]
     pub(crate) mmap_copy_stage: &'static str,
+    #[cfg(unix)]
     pub(crate) direct_read_copy_stage: &'static str,
 }
 
@@ -699,6 +732,25 @@ impl Disk {
         }
     }
 
+    /// Keep local undo publication owned independently of the wrapper deadline.
+    /// Remote undo retains its existing RPC contract; this is not a remote drain proof.
+    pub(crate) async fn undo_write_with_namespace_owner(
+        &self,
+        volume: &str,
+        path: &str,
+        fi: FileInfo,
+        opts: DeleteOptions,
+        namespace_owner: Option<Arc<dyn Send + Sync>>,
+    ) -> Result<()> {
+        match self {
+            Self::Local(disk) => {
+                disk.undo_write_with_namespace_owner(volume, path, fi, opts, namespace_owner)
+                    .await
+            }
+            Self::Remote(disk) => disk.delete_version(volume, path, fi, false, opts).await,
+        }
+    }
+
     pub(crate) async fn rename_data_borrowed(
         &self,
         src_volume: &str,
@@ -711,6 +763,36 @@ impl Disk {
             .await
     }
 
+    pub(crate) async fn rename_data_borrowed_with_fence_observed(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+        guards: RenameDataGuards,
+    ) -> RenameDataObservation {
+        match self {
+            Disk::Local(local_disk) => {
+                local_disk
+                    .rename_data_observed_with_guards(src_volume, src_path, fi, dst_volume, dst_path, guards)
+                    .await
+            }
+            Disk::Remote(remote_disk) => RenameDataObservation::unknown(
+                remote_disk
+                    .rename_data_borrowed_with_fence(
+                        src_volume,
+                        src_path,
+                        fi,
+                        dst_volume,
+                        dst_path,
+                        guards.scanner_publication_lease_token,
+                    )
+                    .await,
+            ),
+        }
+    }
+
     pub(crate) async fn rename_data_borrowed_with_fence(
         &self,
         src_volume: &str,
@@ -873,6 +955,7 @@ impl Disk {
         }
     }
 
+    #[cfg(unix)]
     pub(crate) fn get_object_path_for_io_if_local(
         &self,
         volume: &str,

crates/ecstore/src/disk/os.rs

@@ -91,6 +91,7 @@ pub(crate) mod fsync_dir_recorder {
     static RECORDED: Mutex<Vec<PathBuf>> = Mutex::new(Vec::new());
     static LIMITED: Mutex<Vec<PathBuf>> = Mutex::new(Vec::new());
     static GROUPED: Mutex<Vec<(PathBuf, usize)>> = Mutex::new(Vec::new());
+    #[cfg(unix)]
     static BEFORE_LIMITED: std::sync::LazyLock<Mutex<HashMap<PathBuf, Hook>>> =
         std::sync::LazyLock::new(|| Mutex::new(HashMap::new()));
     static BEFORE_GROUP_BATCH: std::sync::LazyLock<Mutex<HashMap<PathBuf, Hook>>> =
@@ -150,6 +151,7 @@ pub(crate) mod fsync_dir_recorder {
         contains_path(&RECORDED.lock().expect("fsync dir recorder poisoned"), dir)
     }
 
+    #[cfg(unix)]
     pub(crate) fn record_limited(dir: &Path) {
         record_path(&LIMITED, dir, "limited fsync dir recorder");
         let hook = remove_hook(&BEFORE_LIMITED, dir, "limited fsync hook poisoned");
@@ -162,6 +164,7 @@ pub(crate) mod fsync_dir_recorder {
         contains_path(&LIMITED.lock().expect("limited fsync dir recorder poisoned"), dir)
     }
 
+    #[cfg(unix)]
     pub(crate) fn set_before_limited(dir: &Path, hook: impl FnOnce() + Send + 'static) {
         BEFORE_LIMITED
             .lock()
@@ -237,11 +240,56 @@ pub(crate) mod fsync_dir_recorder {
             .insert(dir.to_path_buf(), kind);
     }
 
+    #[cfg(unix)]
     pub(crate) fn take_grouped_failure(dir: &Path) -> Option<io::ErrorKind> {
         remove_path_keyed(&GROUPED_FAILURES, dir, "grouped fsync failure hook poisoned")
     }
 }
 
+/// Pause a real namespace mutation inside its physical executor.
+#[cfg(all(test, not(windows)))]
+pub(crate) mod prepared_publication_test_hooks {
+    use super::*;
+
+    #[derive(Clone, Copy, Debug, Eq, Hash, PartialEq)]
+    pub(crate) enum Stage {
+        PreparedRename,
+        Rename,
+        Remove,
+        Rollback,
+        DirFsync,
+    }
+
+    type Hook = Box<dyn FnOnce() + Send>;
+    type Key = (Stage, PathBuf);
+    static BEFORE_PUBLICATION: LazyLock<Mutex<HashMap<Key, Hook>>> = LazyLock::new(|| Mutex::new(HashMap::new()));
+
+    pub(crate) struct Guard(Key);
+
+    impl Drop for Guard {
+        fn drop(&mut self) {
+            BEFORE_PUBLICATION.lock().remove(&self.0);
+        }
+    }
+
+    pub(crate) fn install(path: &Path, hook: impl FnOnce() + Send + 'static) -> Guard {
+        install_at(Stage::PreparedRename, path, hook)
+    }
+
+    pub(crate) fn install_at(stage: Stage, path: &Path, hook: impl FnOnce() + Send + 'static) -> Guard {
+        let key = (stage, path.to_path_buf());
+        assert!(BEFORE_PUBLICATION.lock().insert(key.clone(), Box::new(hook)).is_none());
+        Guard(key)
+    }
+
+    pub(crate) fn run(stage: Stage, path: &Path) {
+        let hook = BEFORE_PUBLICATION.lock().remove(&(stage, path.to_path_buf()));
+        if let Some(hook) = hook {
+            hook();
+        }
+    }
+}
+
 #[cfg(all(test, windows))]
 pub(crate) mod windows_rename_test_hooks {
     use super::*;
@@ -576,6 +624,7 @@ impl OpenedDstDirFsyncGroup {
 }
 
 struct DstDirFsyncWaiter {
+    namespace_owner: Option<Arc<dyn Send + Sync>>,
     result_tx: oneshot::Sender<SharedDstDirFsyncResult>,
 }
 
@@ -634,6 +683,7 @@ impl DstDirFsyncGroupCommit {
     fn enqueue_opened(
         &self,
         opened: OpenedDstDirFsyncGroup,
+        namespace_owner: Option<Arc<dyn Send + Sync>>,
     ) -> io::Result<(oneshot::Receiver<SharedDstDirFsyncResult>, Option<Arc<DstDirFsyncGroup>>)> {
         let (result_tx, result_rx) = oneshot::channel();
         let mut registry = self.inner.lock();
@@ -664,7 +714,10 @@ impl DstDirFsyncGroupCommit {
             group
         };
         let mut group_state = group.inner.lock();
-        group_state.pending.push_back(DstDirFsyncWaiter { result_tx });
+        group_state.pending.push_back(DstDirFsyncWaiter {
+            result_tx,
+            namespace_owner,
+        });
         let start_worker = !group_state.worker_running;
         if start_worker {
             group_state.worker_running = true;
@@ -686,7 +739,13 @@ impl DstDirFsyncGroupCommit {
     fn remove_idle_group(&self, group: &Arc<DstDirFsyncGroup>) {
         let mut registry = self.inner.lock();
         let group_state = group.inner.lock();
-        if !group_state.worker_running && group_state.pending.is_empty() {
+        if !group_state.worker_running
+            && group_state.pending.is_empty()
+            && registry
+                .groups
+                .get(&group.key)
+                .is_some_and(|registered| Arc::ptr_eq(registered, group))
+        {
             registry.groups.remove(&group.key);
         }
     }
@@ -709,16 +768,20 @@ impl DstDirFsyncGroupCommit {
         &self,
         dir: &Path,
     ) -> io::Result<(oneshot::Receiver<SharedDstDirFsyncResult>, Option<Arc<DstDirFsyncGroup>>)> {
-        self.enqueue_opened(OpenedDstDirFsyncGroup::open(dir)?)
+        self.enqueue_opened(OpenedDstDirFsyncGroup::open(dir)?, None)
     }
 }
 
 #[cfg(unix)]
-async fn fsync_open_dst_dir_group(group: &DstDirFsyncGroup) -> io::Result<()> {
+async fn fsync_open_dst_dir_group(group: &DstDirFsyncGroup, namespace_owners: Vec<Arc<dyn Send + Sync>>) -> io::Result<()> {
     #[cfg(test)]
     let dir = group.dir.clone();
     let dir_file = group.dir_file.clone();
     fsync_spawn_blocking(move || {
+        // The batch worker may be cancelled while this syscall is still running.
+        let _namespace_owners = namespace_owners;
+        #[cfg(all(test, not(windows)))]
+        prepared_publication_test_hooks::run(prepared_publication_test_hooks::Stage::DirFsync, &dir);
         #[cfg(test)]
         {
             if let Some(kind) = fsync_dir_recorder::take_grouped_failure(&dir) {
@@ -733,38 +796,85 @@ async fn fsync_open_dst_dir_group(group: &DstDirFsyncGroup) -> io::Result<()> {
 }
 
 #[cfg(not(unix))]
-async fn fsync_open_dst_dir_group(group: &DstDirFsyncGroup) -> io::Result<()> {
+async fn fsync_open_dst_dir_group(group: &DstDirFsyncGroup, namespace_owners: Vec<Arc<dyn Send + Sync>>) -> io::Result<()> {
+    let _namespace_owners = namespace_owners;
     fsync_dir(&group.dir).await
 }
 
-async fn run_dst_dir_fsync_group_worker(group: Arc<DstDirFsyncGroup>) {
+struct DstDirFsyncWorkerGuard {
+    group: Arc<DstDirFsyncGroup>,
+    in_flight: usize,
+    armed: bool,
+}
+
+impl Drop for DstDirFsyncWorkerGuard {
+    fn drop(&mut self) {
+        if !self.armed {
+            return;
+        }
+        // Cancellation must release queued owners, but the physical batch keeps
+        // its own owners until its blocking syscall returns.
+        let pending = {
+            let mut registry = DST_DIR_FSYNC_GROUP_COMMIT.inner.lock();
+            let mut group_state = self.group.inner.lock();
+            let pending = std::mem::take(&mut group_state.pending);
+            group_state.worker_running = false;
+            if registry
+                .groups
+                .get(&self.group.key)
+                .is_some_and(|group| Arc::ptr_eq(group, &self.group))
+            {
+                registry.total_waiters = registry.total_waiters.saturating_sub(pending.len() + self.in_flight);
+                registry.groups.remove(&self.group.key);
+            }
+            pending
+        };
+        // Lease and channel destructors must run outside the registry locks.
+        drop(pending);
+    }
+}
+
+fn run_dst_dir_fsync_group_worker(group: Arc<DstDirFsyncGroup>) -> impl std::future::Future<Output = ()> {
+    // Capture before spawning: shutdown may drop the future without polling it.
+    let worker_guard = DstDirFsyncWorkerGuard {
+        group: group.clone(),
+        in_flight: 0,
+        armed: true,
+    };
+    async move {
+        let mut worker_guard = worker_guard;
         loop {
             #[cfg(test)]
             fsync_dir_recorder::run_before_group_batch(&group.dir);
             tokio::task::yield_now().await;
-        let batch: Vec<DstDirFsyncWaiter> = {
+            let mut batch: Vec<DstDirFsyncWaiter> = {
                 let mut group_state = group.inner.lock();
                 group_state.pending.drain(..).collect()
             };
             if batch.is_empty() {
                 let mut group_state = group.inner.lock();
+                worker_guard.armed = false;
                 group_state.worker_running = false;
                 drop(group_state);
                 DST_DIR_FSYNC_GROUP_COMMIT.remove_idle_group(&group);
                 return;
             }
+            worker_guard.in_flight = batch.len();
 
             #[cfg(test)]
             fsync_dir_recorder::record_grouped(&group.dir, batch.len());
-        let result = fsync_open_dst_dir_group(&group)
+            let namespace_owners = batch.iter_mut().filter_map(|waiter| waiter.namespace_owner.take()).collect();
+            let result = fsync_open_dst_dir_group(&group, namespace_owners)
                 .await
                 .map_err(SharedDstDirFsyncError::from_error);
             let batch_len = batch.len();
             DST_DIR_FSYNC_GROUP_COMMIT.complete_batch(batch_len);
+            worker_guard.in_flight = 0;
 
             let should_stop = {
                 let mut group_state = group.inner.lock();
                 if group_state.pending.is_empty() {
+                    worker_guard.armed = false;
                     group_state.worker_running = false;
                     true
                 } else {
@@ -781,18 +891,23 @@ async fn run_dst_dir_fsync_group_worker(group: Arc<DstDirFsyncGroup>) {
                 return;
             }
         }
+    }
 }
 
-async fn fsync_dst_dir_group_commit_with_enabled(dir: impl AsRef<Path>, enabled: bool) -> io::Result<()> {
+async fn fsync_dst_dir_group_commit_with_enabled(
+    dir: impl AsRef<Path>,
+    enabled: bool,
+    namespace_owner: Option<Arc<dyn Send + Sync>>,
+) -> io::Result<()> {
     if !enabled {
-        return fsync_dir(dir).await;
+        return fsync_dir_with_owner(dir.as_ref(), namespace_owner).await;
     }
 
     let dir = dir.as_ref().to_path_buf();
     let opened = tokio::task::spawn_blocking(move || OpenedDstDirFsyncGroup::open(&dir))
         .await
         .map_err(|err| io::Error::other(format!("blocking dst dir group open failed: {err}")))??;
-    let (result_rx, worker) = DST_DIR_FSYNC_GROUP_COMMIT.enqueue_opened(opened)?;
+    let (result_rx, worker) = DST_DIR_FSYNC_GROUP_COMMIT.enqueue_opened(opened, namespace_owner)?;
     if let Some(group) = worker {
         tokio::spawn(run_dst_dir_fsync_group_worker(group));
     }
@@ -804,8 +919,11 @@ async fn fsync_dst_dir_group_commit_with_enabled(dir: impl AsRef<Path>, enabled:
     }
 }
 
-pub(crate) async fn fsync_dst_dir_group_commit(dir: impl AsRef<Path>) -> io::Result<()> {
+pub(crate) async fn fsync_dst_dir_group_commit(
-    fsync_dst_dir_group_commit_with_enabled(dir, dst_dir_fsync_group_commit_enabled()).await
+    dir: impl AsRef<Path>,
+    namespace_owner: Option<Arc<dyn Send + Sync>>,
+) -> io::Result<()> {
+    fsync_dst_dir_group_commit_with_enabled(dir, dst_dir_fsync_group_commit_enabled(), namespace_owner).await
 }
 
 pub(crate) async fn fsync_dst_dir_group_commit_or_namespace_file_sync_limit(
@@ -814,7 +932,7 @@ pub(crate) async fn fsync_dst_dir_group_commit_or_namespace_file_sync_limit(
     admission: &FileSyncAdmission,
 ) -> io::Result<()> {
     if dst_dir_fsync_group_commit_enabled() {
-        fsync_dst_dir_group_commit_with_enabled(dir, true).await
+        fsync_dst_dir_group_commit_with_enabled(dir, true, Some(lease)).await
     } else {
         fsync_dir_with_namespace_file_sync_limit(dir, lease, admission).await
     }
@@ -822,7 +940,7 @@ pub(crate) async fn fsync_dst_dir_group_commit_or_namespace_file_sync_limit(
 
 #[cfg(test)]
 pub(crate) async fn fsync_dst_dir_group_commit_for_test(dir: impl AsRef<Path>, enabled: bool) -> io::Result<()> {
-    fsync_dst_dir_group_commit_with_enabled(dir, enabled).await
+    fsync_dst_dir_group_commit_with_enabled(dir, enabled, None).await
 }
 
 #[cfg(test)]
@@ -1229,6 +1347,8 @@ pub(crate) struct NamespaceMutationLease {
     _namespace_guard: OwnedMutexGuard<()>,
     _volume_guard: Option<OwnedRwLockReadGuard<()>>,
     external_guard: Mutex<Option<Arc<dyn Send + Sync>>>,
+    // Independent of the quota claim; both survive cancellation of the waiter.
+    _namespace_owner: Option<Arc<dyn Send + Sync>>,
 }
 
 impl NamespaceMutationLease {
@@ -1238,10 +1358,18 @@ impl NamespaceMutationLease {
 }
 
 async fn acquire_namespace_mutation_lease(path: &Path) -> Arc<NamespaceMutationLease> {
+    acquire_namespace_mutation_lease_with_owner(path, None).await
+}
+
+async fn acquire_namespace_mutation_lease_with_owner(
+    path: &Path,
+    namespace_owner: Option<Arc<dyn Send + Sync>>,
+) -> Arc<NamespaceMutationLease> {
     Arc::new(NamespaceMutationLease {
         _namespace_guard: disk_namespace_mutation_lock(path).lock_owned().await,
         _volume_guard: None,
         external_guard: Mutex::new(None),
+        _namespace_owner: namespace_owner,
     })
 }
 
@@ -1251,6 +1379,15 @@ pub(crate) async fn acquire_rename_data_mutation_lease(
     root: &Path,
     volume: &str,
     destination_object: &Path,
+) -> Arc<NamespaceMutationLease> {
+    acquire_rename_data_mutation_lease_with_owner(root, volume, destination_object, None).await
+}
+
+pub(crate) async fn acquire_rename_data_mutation_lease_with_owner(
+    root: &Path,
+    volume: &str,
+    destination_object: &Path,
+    namespace_owner: Option<Arc<dyn Send + Sync>>,
 ) -> Arc<NamespaceMutationLease> {
     let namespace_guard = disk_namespace_mutation_lock(destination_object).lock_owned().await;
     let volume_guard = disk_volume_mutation_lock(root, volume).read_owned().await;
@@ -1258,6 +1395,7 @@ pub(crate) async fn acquire_rename_data_mutation_lease(
         _namespace_guard: namespace_guard,
         _volume_guard: Some(volume_guard),
         external_guard: Mutex::new(None),
+        _namespace_owner: namespace_owner,
     })
 }
 
@@ -1747,6 +1885,69 @@ pub async fn rename_all(
     Ok(())
 }
 
+pub(crate) async fn fsync_dir_with_owner(path: &Path, namespace_owner: Option<Arc<dyn Send + Sync>>) -> io::Result<()> {
+    #[cfg(unix)]
+    {
+        if namespace_owner.is_none() {
+            return fsync_dir(path).await;
+        }
+        let path = path.to_path_buf();
+        fsync_spawn_blocking(move || {
+            let _namespace_owner = namespace_owner;
+            fsync_dir_std(path)
+        })
+        .await?
+    }
+    #[cfg(not(unix))]
+    {
+        let _ = namespace_owner;
+        fsync_dir(path).await
+    }
+}
+
+/// Retain namespace ownership in the actual filesystem executor after timeout.
+pub(crate) async fn remove_file_with_owner(
+    path: impl AsRef<Path>,
+    namespace_owner: Option<Arc<dyn Send + Sync>>,
+) -> io::Result<()> {
+    if namespace_owner.is_none() {
+        return tokio::fs::remove_file(path).await;
+    }
+    let path = path.as_ref().to_path_buf();
+    let lease = acquire_namespace_mutation_lease_with_owner(&path, namespace_owner).await;
+    run_blocking_namespace_operation(lease, move || {
+        #[cfg(all(test, not(windows)))]
+        prepared_publication_test_hooks::run(prepared_publication_test_hooks::Stage::Remove, &path);
+        std::fs::remove_file(path)
+    })
+    .await
+}
+
+/// Retain namespace ownership in the actual filesystem executor after timeout.
+pub(crate) async fn remove_dir_with_owner(
+    path: impl AsRef<Path>,
+    namespace_owner: Option<Arc<dyn Send + Sync>>,
+) -> io::Result<()> {
+    if namespace_owner.is_none() {
+        return tokio::fs::remove_dir(path).await;
+    }
+    let path = path.as_ref().to_path_buf();
+    let lease = acquire_namespace_mutation_lease_with_owner(&path, namespace_owner).await;
+    run_blocking_namespace_operation(lease, move || std::fs::remove_dir(path)).await
+}
+
+#[tracing::instrument(name = "rename_all", level = "debug", skip_all)]
+pub(crate) async fn rename_all_with_owner(
+    src_file_path: impl AsRef<Path>,
+    dst_file_path: impl AsRef<Path>,
+    base_dir: impl AsRef<Path>,
+    publication_root: &PublicationRoot,
+    namespace_owner: Option<Arc<dyn Send + Sync>>,
+) -> Result<()> {
+    let lease = acquire_namespace_mutation_lease_with_owner(dst_file_path.as_ref(), namespace_owner).await;
+    rename_all_with_lease(src_file_path, dst_file_path, base_dir, publication_root, lease).await
+}
+
 pub(crate) async fn rename_all_with_lease(
     src_file_path: impl AsRef<Path>,
     dst_file_path: impl AsRef<Path>,
@@ -1939,6 +2140,8 @@ pub(crate) async fn rename_all_with_prepared_source(
         move || {
             validate_prepared_rename_source(&prepared_source, &src_file_path)?;
             let preparation = prepare_rename_with_retry(&src_file_path, &dst_file_path, &base_dir, &publication_root)?;
+            #[cfg(test)]
+            prepared_publication_test_hooks::run(prepared_publication_test_hooks::Stage::PreparedRename, &dst_file_path);
             rename_prepared(&src_file_path, &dst_file_path, &preparation)
         }
     };
@@ -1977,6 +2180,32 @@ pub async fn rename_all_ignore_missing_source(
     }
 }
 
+#[tracing::instrument(name = "rename_all_ignore_missing_source", level = "debug", skip_all)]
+pub(crate) async fn rename_all_ignore_missing_source_with_owner(
+    src_file_path: impl AsRef<Path>,
+    dst_file_path: impl AsRef<Path>,
+    base_dir: impl AsRef<Path>,
+    publication_root: &PublicationRoot,
+    namespace_owner: Option<Arc<dyn Send + Sync>>,
+) -> Result<()> {
+    let src_file_path = src_file_path.as_ref();
+    let lease = acquire_namespace_mutation_lease_with_owner(dst_file_path.as_ref(), namespace_owner).await;
+    match reliable_rename_inner_with_lease(
+        src_file_path.to_path_buf(),
+        dst_file_path.as_ref().to_path_buf(),
+        base_dir.as_ref().to_path_buf(),
+        publication_root.clone(),
+        false,
+        lease,
+    )
+    .await
+    {
+        Ok(()) => Ok(()),
+        Err(err) if err.kind() == io::ErrorKind::NotFound && rename_source_is_missing(src_file_path, publication_root) => Ok(()),
+        Err(err) => Err(to_file_error(err).into()),
+    }
+}
+
 #[cfg(windows)]
 pub(crate) fn rename_source_is_missing(src_file_path: &Path, publication_root: &PublicationRoot) -> bool {
     let Some(source_parent) = src_file_path.parent() else {
@@ -2042,6 +2271,11 @@ async fn reliable_rename_inner_with_lease(
         let base_dir = base_dir.clone();
         move || {
             let preparation = prepare_rename_with_retry(&src_file_path, &dst_file_path, &base_dir, &publication_root)?;
+            #[cfg(all(test, not(windows)))]
+            {
+                prepared_publication_test_hooks::run(prepared_publication_test_hooks::Stage::Rename, &src_file_path);
+                prepared_publication_test_hooks::run(prepared_publication_test_hooks::Stage::Rename, &dst_file_path);
+            }
             rename_prepared(&src_file_path, &dst_file_path, &preparation)
         }
     };
@@ -6136,6 +6370,245 @@ mod tests {
         wait_for_dst_dir_fsync_group_commit_idle().await;
     }
 
+    #[cfg(unix)]
+    #[tokio::test]
+    #[serial_test::serial(dst_dir_fsync_group_commit)]
+    async fn grouped_fsync_physical_batch_keeps_all_owners_after_worker_cancellation() {
+        let temp_dir = tempdir().expect("fixture directory");
+        let dir = temp_dir.path().canonicalize().expect("canonical fsync path");
+        let first_ctx = Arc::new(crate::runtime::instance::InstanceContext::new());
+        let second_ctx = Arc::new(crate::runtime::instance::InstanceContext::new());
+        let first_owner = first_ctx.begin_namespace_commit();
+        let second_owner = second_ctx.begin_namespace_commit();
+        let first_probe = Arc::downgrade(&first_owner);
+        let second_probe = Arc::downgrade(&second_owner);
+        let (first_rx, group) = DST_DIR_FSYNC_GROUP_COMMIT
+            .enqueue_opened(
+                OpenedDstDirFsyncGroup::open(&dir).expect("open first waiter directory"),
+                Some(first_owner),
+            )
+            .expect("queue first real waiter");
+        let group = group.expect("first waiter starts the group");
+        let (second_rx, second_worker) = DST_DIR_FSYNC_GROUP_COMMIT
+            .enqueue_opened(
+                OpenedDstDirFsyncGroup::open(&dir).expect("open second waiter directory"),
+                Some(second_owner),
+            )
+            .expect("queue second real waiter");
+        assert!(second_worker.is_none(), "same directory must join the same batch");
+        assert_eq!(group.inner.lock().pending.len(), 2);
+        let (entered_tx, entered_rx) = tokio::sync::oneshot::channel();
+        let (release_tx, release_rx) = std::sync::mpsc::channel::<()>();
+        let _hook =
+            prepared_publication_test_hooks::install_at(prepared_publication_test_hooks::Stage::DirFsync, &dir, move || {
+                let _ = entered_tx.send(());
+                let _ = release_rx.recv();
+            });
+        let worker = tokio::spawn(run_dst_dir_fsync_group_worker(group.clone()));
+        tokio::time::timeout(Duration::from_secs(5), entered_rx)
+            .await
+            .expect("batch must reach its physical fsync")
+            .expect("physical fsync entry");
+        assert_eq!(fsync_dir_recorder::grouped_batch_sizes(&dir), vec![2]);
+        assert!(
+            group.inner.lock().pending.is_empty(),
+            "both waiters were transferred into the physical batch"
+        );
+        let queued_ctx = Arc::new(crate::runtime::instance::InstanceContext::new());
+        let queued_owner = queued_ctx.begin_namespace_commit();
+        let queued_probe = Arc::downgrade(&queued_owner);
+        let queued_generation = queued_ctx.namespace_commit_generation();
+        let (queued_rx, queued_worker) = DST_DIR_FSYNC_GROUP_COMMIT
+            .enqueue_opened(
+                OpenedDstDirFsyncGroup::open(&dir).expect("open queued waiter directory"),
+                Some(queued_owner),
+            )
+            .expect("queue a waiter after the physical batch was frozen");
+        assert!(queued_worker.is_none());
+        assert_eq!(group.inner.lock().pending.len(), 1);
+        drop((first_rx, second_rx));
+        worker.abort();
+        assert!(worker.await.expect_err("cancel the async batch owner").is_cancelled());
+        assert!(queued_rx.await.is_err(), "an undispatched waiter must observe worker cancellation");
+        assert!(queued_probe.upgrade().is_none());
+        assert!(!queued_ctx.namespace_commits_pending());
+        assert!(queued_ctx.namespace_commit_generation() > queued_generation);
+        assert!(group.inner.lock().pending.is_empty());
+        assert!(!group.inner.lock().worker_running);
+        assert_eq!(DST_DIR_FSYNC_GROUP_COMMIT.counts_for_test(), (0, 0));
+        let first_pending = first_ctx.namespace_commits_pending() && first_probe.upgrade().is_some();
+        let second_pending = second_ctx.namespace_commits_pending() && second_probe.upgrade().is_some();
+        let generations = (first_ctx.namespace_commit_generation(), second_ctx.namespace_commit_generation());
+        drop(release_tx);
+        tokio::time::timeout(Duration::from_secs(5), async {
+            while Arc::strong_count(&group.dir_file) != 1
+                || first_probe.upgrade().is_some()
+                || second_probe.upgrade().is_some()
+                || first_ctx.namespace_commits_pending()
+                || second_ctx.namespace_commits_pending()
+            {
+                tokio::task::yield_now().await;
+            }
+        })
+        .await
+        .expect("physical fsync must release every batch owner");
+        assert!(fsync_dir_recorder::was_fsynced(&dir), "the detached syscall must really execute");
+        assert!(
+            first_pending && second_pending,
+            "one physical batch must preserve both independent namespace owners"
+        );
+        assert!(!first_ctx.namespace_commits_pending());
+        assert!(!second_ctx.namespace_commits_pending());
+        assert!(first_ctx.namespace_commit_generation() > generations.0);
+        assert!(second_ctx.namespace_commit_generation() > generations.1);
+    }
+
+    #[cfg(unix)]
+    #[tokio::test]
+    #[serial_test::serial(dst_dir_fsync_group_commit)]
+    async fn grouped_fsync_unpolled_worker_releases_queued_owner() {
+        let temp_dir = tempdir().expect("fixture directory");
+        let ctx = Arc::new(crate::runtime::instance::InstanceContext::new());
+        let owner = ctx.begin_namespace_commit();
+        let probe = Arc::downgrade(&owner);
+        let generation = ctx.namespace_commit_generation();
+        let (rx, group) = DST_DIR_FSYNC_GROUP_COMMIT
+            .enqueue_opened(
+                OpenedDstDirFsyncGroup::open(temp_dir.path()).expect("open queued waiter directory"),
+                Some(owner),
+            )
+            .expect("queue a real waiter");
+        let group = group.expect("first waiter starts the group");
+        let worker = run_dst_dir_fsync_group_worker(group.clone());
+        assert!(ctx.namespace_commits_pending());
+        drop(worker);
+        assert!(rx.await.is_err(), "shutdown before first poll must release the waiter");
+        assert!(probe.upgrade().is_none());
+        assert!(!ctx.namespace_commits_pending());
+        assert!(ctx.namespace_commit_generation() > generation);
+        assert!(group.inner.lock().pending.is_empty());
+        assert!(!group.inner.lock().worker_running);
+        assert_eq!(DST_DIR_FSYNC_GROUP_COMMIT.counts_for_test(), (0, 0));
+        assert!(
+            fsync_dir_recorder::grouped_batch_sizes(temp_dir.path()).is_empty(),
+            "the dropped future must not dispatch a physical batch"
+        );
+    }
+
+    #[cfg(unix)]
+    #[test]
+    fn stale_idle_group_cleanup_preserves_successor_registration() {
+        let temp_dir = tempdir().expect("fixture directory");
+        let registry = DstDirFsyncGroupCommit::default();
+        let (mut first_rx, first_worker) = registry.enqueue_for_test(temp_dir.path()).expect("enqueue first worker");
+        let old_group = first_worker.expect("first waiter starts a worker");
+        // W1 has completed its batch and marked G idle, but has not cleaned G up.
+        let first_waiter = old_group.inner.lock().pending.pop_front().expect("first batch waiter");
+        registry.complete_batch(1);
+        old_group.inner.lock().worker_running = false;
+
+        let (mut second_rx, second_worker) = registry.enqueue_for_test(temp_dir.path()).expect("enqueue second worker");
+        let reused_group = second_worker.expect("idle G starts another worker");
+        assert!(Arc::ptr_eq(&old_group, &reused_group));
+        let second_waiter = reused_group.inner.lock().pending.pop_front().expect("second batch waiter");
+        registry.complete_batch(1);
+        reused_group.inner.lock().worker_running = false;
+        registry.remove_idle_group(&reused_group);
+        assert_eq!(registry.counts_for_test(), (0, 0), "normal idle cleanup must remove G");
+        assert!(second_waiter.result_tx.send(Ok(())).is_ok());
+        assert!(second_rx.try_recv().expect("second worker reports completion").is_ok());
+
+        let (mut successor_rx, successor_worker) = registry.enqueue_for_test(temp_dir.path()).expect("enqueue successor");
+        let successor = successor_worker.expect("successor starts a new group");
+        assert!(!Arc::ptr_eq(&old_group, &successor));
+        assert_eq!(registry.counts_for_test(), (1, 1));
+        // W1 resumes with its old Arc after W2 removed G and W3 installed G2.
+        registry.remove_idle_group(&old_group);
+        assert!(first_waiter.result_tx.send(Ok(())).is_ok());
+        assert!(first_rx.try_recv().expect("first worker reports completion").is_ok());
+        assert!(
+            registry
+                .inner
+                .lock()
+                .groups
+                .get(&successor.key)
+                .is_some_and(|registered| Arc::ptr_eq(registered, &successor)),
+            "stale cleanup must retain the exact successor Arc"
+        );
+        assert_eq!(registry.counts_for_test(), (1, 1));
+        assert!(successor.inner.lock().worker_running);
+        assert_eq!(successor.inner.lock().pending.len(), 1);
+        assert!(matches!(successor_rx.try_recv(), Err(oneshot::error::TryRecvError::Empty)));
+
+        let (_joined_rx, new_worker) = registry.enqueue_for_test(temp_dir.path()).expect("join successor");
+        assert!(new_worker.is_none(), "a later waiter must join G2 instead of creating G3");
+        assert_eq!(successor.inner.lock().pending.len(), 2);
+        assert_eq!(registry.counts_for_test(), (1, 2));
+    }
+
+    #[cfg(unix)]
+    #[tokio::test]
+    #[serial_test::serial(dst_dir_fsync_group_commit)]
+    async fn stale_idle_cleanup_then_unpolled_worker_drop_releases_waiter_budget() {
+        wait_for_dst_dir_fsync_group_commit_idle().await;
+        let temp_dir = tempdir().expect("fixture directory");
+        let (old_rx, old_worker) = DST_DIR_FSYNC_GROUP_COMMIT
+            .enqueue_for_test(temp_dir.path())
+            .expect("enqueue old group");
+        let old_group = old_worker.expect("old group starts a worker");
+        tokio::time::timeout(Duration::from_secs(5), run_dst_dir_fsync_group_worker(old_group.clone()))
+            .await
+            .expect("old worker must finish its actual fsync");
+        assert!(old_rx.await.expect("old worker reports completion").is_ok());
+        assert!(fsync_dir_recorder::was_fsynced(temp_dir.path()));
+        assert_eq!(fsync_dir_recorder::grouped_batch_sizes(temp_dir.path()), vec![1]);
+        assert_eq!(DST_DIR_FSYNC_GROUP_COMMIT.counts_for_test(), (0, 0));
+
+        let ctx = Arc::new(crate::runtime::instance::InstanceContext::new());
+        let owner = ctx.begin_namespace_commit();
+        let probe = Arc::downgrade(&owner);
+        let generation = ctx.namespace_commit_generation();
+        let (rx, successor_worker) = DST_DIR_FSYNC_GROUP_COMMIT
+            .enqueue_opened(
+                OpenedDstDirFsyncGroup::open(temp_dir.path()).expect("open successor directory"),
+                Some(owner),
+            )
+            .expect("enqueue successor owner");
+        let successor = successor_worker.expect("successor starts a new group");
+        assert!(!Arc::ptr_eq(&old_group, &successor));
+        let worker = run_dst_dir_fsync_group_worker(successor.clone());
+        // The stale Arc represents W1 resuming after another worker removed G.
+        DST_DIR_FSYNC_GROUP_COMMIT.remove_idle_group(&old_group);
+        assert!(ctx.namespace_commits_pending());
+        assert!(probe.upgrade().is_some());
+        drop(worker);
+        let channel_closed = tokio::time::timeout(Duration::from_secs(5), rx)
+            .await
+            .expect("dropping the unpolled worker must release its channel")
+            .is_err();
+        let counts_after_drop = DST_DIR_FSYNC_GROUP_COMMIT.counts_for_test();
+        let owner_released = probe.upgrade().is_none();
+        let namespace_pending = ctx.namespace_commits_pending();
+        let generation_after_drop = ctx.namespace_commit_generation();
+        let successor_pending = successor.inner.lock().pending.len();
+        let worker_running = successor.inner.lock().worker_running;
+        // Preserve the observed result before cleanup, so a RED run cannot leak
+        // its phantom count into unrelated tests in the same process.
+        clear_dst_dir_fsync_group_commit_for_test();
+        assert!(channel_closed);
+        assert!(owner_released);
+        assert!(!namespace_pending);
+        assert!(generation_after_drop > generation);
+        assert_eq!(successor_pending, 0);
+        assert!(!worker_running);
+        assert_eq!(
+            fsync_dir_recorder::grouped_batch_sizes(temp_dir.path()),
+            vec![1],
+            "dropping the successor before its first poll must not dispatch another fsync"
+        );
+        assert_eq!(counts_after_drop, (0, 0), "stale cleanup must not strand a phantom waiter");
+    }
+
     #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
     #[serial_test::serial(dst_dir_fsync_group_commit)]
     async fn dst_dir_fsync_group_commit_cancellation_releases_waiter_state() {

crates/ecstore/src/io_support/bitrot.rs

@@ -13,12 +13,15 @@
 // limitations under the License.
 
 use crate::diagnostics::get::{
-    GET_STAGE_READER_MMAP_ACCESS_CHECK, GET_STAGE_READER_MMAP_BLOCKING_TASK, GET_STAGE_READER_MMAP_BLOCKING_WAIT,
+    GET_STAGE_READER_MMAP_ACCESS_CHECK, GET_STAGE_READER_MMAP_METADATA_LOOKUP, GET_STAGE_READER_MMAP_METADATA_VALIDATE,
-    GET_STAGE_READER_MMAP_COPY_BUFFER, GET_STAGE_READER_MMAP_DIRECT_READ_COPY, GET_STAGE_READER_MMAP_FILE_OPEN,
-    GET_STAGE_READER_MMAP_MAP, GET_STAGE_READER_MMAP_METADATA_LOOKUP, GET_STAGE_READER_MMAP_METADATA_VALIDATE,
     GET_STAGE_READER_MMAP_PATH_RESOLVE, GET_STAGE_READER_OPEN_MMAP_COPY_FALLBACK, GET_STAGE_READER_OPEN_MMAP_COPY_SUCCESS,
     GET_STAGE_READER_OPEN_STREAM, GET_STAGE_READER_STREAM_FIRST_READ, record_get_stage_duration_if_enabled,
 };
+#[cfg(unix)]
+use crate::diagnostics::get::{
+    GET_STAGE_READER_MMAP_BLOCKING_TASK, GET_STAGE_READER_MMAP_BLOCKING_WAIT, GET_STAGE_READER_MMAP_COPY_BUFFER,
+    GET_STAGE_READER_MMAP_DIRECT_READ_COPY, GET_STAGE_READER_MMAP_FILE_OPEN, GET_STAGE_READER_MMAP_MAP,
+};
 #[cfg(feature = "hotpath")]
 use crate::disk::FileWriter;
 use crate::disk::{self, DiskAPI as _, DiskStore, FileReader, MmapCopyStageMetrics, error::DiskError};
@@ -406,11 +409,17 @@ async fn open_disk_reader(
             path_resolve_stage: GET_STAGE_READER_MMAP_PATH_RESOLVE,
             metadata_lookup_stage: GET_STAGE_READER_MMAP_METADATA_LOOKUP,
             metadata_validate_stage: GET_STAGE_READER_MMAP_METADATA_VALIDATE,
+            #[cfg(unix)]
             blocking_wait_stage: GET_STAGE_READER_MMAP_BLOCKING_WAIT,
+            #[cfg(unix)]
             blocking_task_stage: GET_STAGE_READER_MMAP_BLOCKING_TASK,
+            #[cfg(unix)]
             file_open_stage: GET_STAGE_READER_MMAP_FILE_OPEN,
+            #[cfg(unix)]
             mmap_map_stage: GET_STAGE_READER_MMAP_MAP,
+            #[cfg(unix)]
             mmap_copy_stage: GET_STAGE_READER_MMAP_COPY_BUFFER,
+            #[cfg(unix)]
             direct_read_copy_stage: GET_STAGE_READER_MMAP_DIRECT_READ_COPY,
         });
         let mmap_result = {

crates/ecstore/src/object_api/types.rs

@@ -870,6 +870,18 @@ impl TierFreeVersionReceiptSink {
     }
 }
 
+/// Internal PUT completion boundary; this does not change fsync or write quorum.
+#[doc(hidden)]
+#[derive(Debug, Default, Clone, Copy, PartialEq, Eq)]
+pub enum WriteCompletion {
+    /// Return at write quorum when the commit owner can retain its guards.
+    #[default]
+    Quorum,
+    /// Drain the rename fan-out before returning. Minority failures still heal
+    /// after a successful quorum commit; this does not require every disk to succeed.
+    TailDrained,
+}
+
 #[derive(Default, Clone)]
 pub struct ObjectOptions {
     // Use the maximum parity (N/2), used when saving server configuration files
@@ -896,6 +908,10 @@ pub struct ObjectOptions {
     /// Persisted bucket incarnation observed before authorization.
     pub expected_bucket_incarnation_id: Option<Uuid>,
     pub no_lock: bool,
+    /// Control-plane writers that immediately read or CAS the same namespace
+    /// key use TailDrained without changing namespace lock ownership.
+    #[doc(hidden)]
+    pub write_completion: WriteCompletion,
     /// True when an upper layer already holds the object read lock before
     /// forwarding a no_lock read to the set layer.
     pub metadata_cache_safe: bool,
@@ -940,6 +956,9 @@ pub struct ObjectOptions {
     pub preserve_etag: Option<String>,
     pub metadata_chg: bool,
     pub http_preconditions: Option<HTTPPreconditions>,
+    /// Internal create-only writes may also preserve an acknowledged deletion.
+    /// Evaluated with `http_preconditions` under the namespace commit lock.
+    pub preserve_delete_marker: bool,
 
     pub delete_replication: Option<ReplicationState>,
     pub delete_replication_config_snapshot: Option<Arc<DeleteReplicationConfigSnapshot>>,

crates/ecstore/src/runtime/instance.rs

@@ -78,6 +78,21 @@ pub(crate) struct ScannerPublicationLeaseEntry {
     pub(crate) _operation_guard: OwnedRwLockReadGuard<()>,
 }
 
+pub(crate) struct NamespaceCommitGuard {
+    ctx: Arc<InstanceContext>,
+    counted: bool,
+}
+
+impl Drop for NamespaceCommitGuard {
+    fn drop(&mut self) {
+        if self.counted {
+            // Publish the new generation before a zero-pending publication probe.
+            self.ctx.advance_namespace_commit_generation();
+            self.ctx.namespace_commits.fetch_sub(1, Ordering::AcqRel);
+        }
+    }
+}
+
 /// Runtime state owned by a single `ECStore` instance.
 ///
 /// This is intentionally minimal in the first migration slice; subsequent
@@ -209,9 +224,13 @@ pub struct InstanceContext {
     /// Last storage-owned movement snapshot observed under the operation
     /// gate. SetDisks cache writers fail closed until ECStore refreshes it.
     scanner_publication_state: AtomicU8,
+    namespace_commits: AtomicU64,
+    namespace_commit_generation: AtomicU64,
     /// Resolves object-encryption material at the application boundary.
     object_encryption_resolver: OnceLock<Arc<dyn ObjectEncryptionResolver>>,
     tier_delete_journal_recovery_stores: std::sync::Mutex<HashSet<Uuid>>,
+    #[cfg(test)]
+    suppress_tier_delete_journal_recovery: bool,
     transition_transaction_recovery_stores: std::sync::Mutex<HashSet<Uuid>>,
     tier_delete_journal_recovery_wakeup: tokio::sync::Notify,
 }
@@ -256,8 +275,12 @@ impl InstanceContext {
             data_movement_generation_exhausted: AtomicBool::new(false),
             data_movement_generation_notify: Arc::new(Notify::new()),
             scanner_publication_state: AtomicU8::new(SCANNER_PUBLICATION_STATE_UNKNOWN),
+            namespace_commits: AtomicU64::new(0),
+            namespace_commit_generation: AtomicU64::new(0),
             object_encryption_resolver: OnceLock::new(),
             tier_delete_journal_recovery_stores: std::sync::Mutex::new(HashSet::new()),
+            #[cfg(test)]
+            suppress_tier_delete_journal_recovery: false,
             transition_transaction_recovery_stores: std::sync::Mutex::new(HashSet::new()),
             tier_delete_journal_recovery_wakeup: tokio::sync::Notify::new(),
         }
@@ -385,6 +408,36 @@ impl InstanceContext {
             && self.scanner_publication_state.load(Ordering::Acquire) == SCANNER_PUBLICATION_STATE_ALLOWED
     }
 
+    pub(crate) fn begin_namespace_commit(self: &Arc<Self>) -> Arc<NamespaceCommitGuard> {
+        let counted = self
+            .namespace_commits
+            .fetch_update(Ordering::AcqRel, Ordering::Acquire, |count| count.checked_add(1))
+            .is_ok();
+        if counted {
+            self.advance_namespace_commit_generation();
+        } else {
+            self.namespace_commit_generation.store(u64::MAX, Ordering::Release);
+        }
+        Arc::new(NamespaceCommitGuard {
+            ctx: Arc::clone(self),
+            counted,
+        })
+    }
+
+    fn advance_namespace_commit_generation(&self) {
+        let _ = self
+            .namespace_commit_generation
+            .fetch_update(Ordering::AcqRel, Ordering::Acquire, |generation| Some(generation.saturating_add(1)));
+    }
+
+    pub(crate) fn namespace_commit_generation(&self) -> u64 {
+        self.namespace_commit_generation.load(Ordering::Acquire)
+    }
+
+    pub(crate) fn namespace_commits_pending(&self) -> bool {
+        self.namespace_commits.load(Ordering::Acquire) != 0 || self.namespace_commit_generation() == u64::MAX
+    }
+
     pub(crate) fn set_scanner_publication_state(&self, blocked: bool) {
         self.scanner_publication_state.store(
             if blocked {
@@ -640,12 +693,21 @@ impl InstanceContext {
     }
 
     pub(crate) fn mark_tier_delete_journal_recovery_started(&self, store_id: Uuid) -> bool {
+        #[cfg(test)]
+        if self.suppress_tier_delete_journal_recovery {
+            return false;
+        }
         self.tier_delete_journal_recovery_stores
             .lock()
             .unwrap_or_else(std::sync::PoisonError::into_inner)
             .insert(store_id)
     }
 
+    #[cfg(test)]
+    pub(crate) fn suppress_tier_delete_journal_recovery_for_test(&mut self) {
+        self.suppress_tier_delete_journal_recovery = true;
+    }
+
     pub(crate) fn mark_transition_transaction_recovery_started(&self, store_id: Uuid) -> bool {
         self.transition_transaction_recovery_stores
             .lock()
@@ -756,6 +818,50 @@ pub fn bootstrap_ctx() -> Arc<InstanceContext> {
 mod tests {
     use super::*;
 
+    #[test]
+    fn namespace_commit_guards_are_instance_local_and_count_until_last_owner() {
+        let first = Arc::new(InstanceContext::new());
+        let other = Arc::new(InstanceContext::new());
+        first.set_scanner_publication_state(false);
+        other.set_scanner_publication_state(false);
+        assert!(first.scanner_publication_state_allowed());
+        let one = first.begin_namespace_commit();
+        let shared_owner = Arc::clone(&one);
+        let two = first.begin_namespace_commit();
+        assert!(first.namespace_commits_pending());
+        assert!(first.scanner_publication_state_allowed(), "pending writes must not block scan admission");
+        assert_eq!(first.namespace_commit_generation(), 2);
+        assert!(!other.namespace_commits_pending());
+        assert_eq!(other.namespace_commit_generation(), 0);
+        assert!(other.scanner_publication_state_allowed());
+        drop(one);
+        assert_eq!(first.namespace_commit_generation(), 2);
+        drop(shared_owner);
+        assert!(first.namespace_commits_pending());
+        assert_eq!(first.namespace_commit_generation(), 3);
+        drop(two);
+        assert!(!first.namespace_commits_pending());
+        assert_eq!(first.namespace_commit_generation(), 4);
+        assert!(first.scanner_publication_state_allowed());
+    }
+
+    #[test]
+    fn namespace_commit_counter_exhaustion_keeps_publication_blocked() {
+        for (count, generation) in [(0, u64::MAX - 1), (u64::MAX, 0)] {
+            let ctx = Arc::new(InstanceContext::new());
+            ctx.set_scanner_publication_state(false);
+            ctx.namespace_commits.store(count, Ordering::Release);
+            ctx.namespace_commit_generation.store(generation, Ordering::Release);
+            let guard = ctx.begin_namespace_commit();
+            assert!(ctx.namespace_commits_pending());
+            assert_eq!(ctx.namespace_commit_generation(), u64::MAX);
+            drop(guard);
+            assert!(ctx.namespace_commits_pending());
+            assert_eq!(ctx.namespace_commit_generation(), u64::MAX);
+            assert_eq!(ctx.namespace_commits.load(Ordering::Acquire), count);
+        }
+    }
+
     // The SetupType inputs must derive the exact (is_erasure,
     // is_dist_erasure, is_erasure_sd) triples that the original three
     // process-global erasure bools produced via update_erasure_type().
@@ -1073,6 +1179,12 @@ mod tests {
         assert!(!ctx_a.mark_tier_delete_journal_recovery_started(store_a));
         assert!(ctx_a.mark_tier_delete_journal_recovery_started(store_b));
         assert!(ctx_b.mark_tier_delete_journal_recovery_started(store_a));
+
+        let mut manual_ctx = InstanceContext::new();
+        manual_ctx.suppress_tier_delete_journal_recovery_for_test();
+        assert!(!manual_ctx.mark_tier_delete_journal_recovery_started(store_a));
+        assert!(!manual_ctx.mark_tier_delete_journal_recovery_started(store_b));
+        assert!(ctx_b.mark_tier_delete_journal_recovery_started(store_b));
     }
 
     #[test]

crates/ecstore/src/services/notification_sys.rs

@@ -62,12 +62,27 @@ const REMOTE_VERSION_STATE_PROOF_TTL: Duration = Duration::from_secs(30);
 const CROSS_POOL_FENCE_SUPPORTED_VERSION: u32 = 2;
 const TIER_DELETE_JOURNAL_POLICY_SUPPORTED_VERSION: u32 = 3;
 const DECOMMISSION_TARGET_FENCE_POLICY_SUPPORTED_VERSION: u32 = 4;
+// Keep this synchronized with the version served by node_service. Including
+// the local member in the minimum prevents an older coordinator from
+// self-authorizing a policy implemented only by newer remote peers.
+const LOCAL_CROSS_POOL_FENCE_POLICY_SUPPORTED_VERSION: u32 = 4;
+/// Version 5 is reserved for a fleet whose every metadata writer preserves
+/// explicit transition version state and destination identity, and implements
+/// conditional per-generation `xl.meta` writes with strong readback. The node
+/// service must not advertise this version until the conditional writer from
+/// rustfs/backlog#684 is available.
+const LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION: u32 = 5;
 type CrossPoolFencePolicyResult = Result<BTreeMap<String, Uuid>>;
 
 fn cross_pool_fence_policy_results(
     peer_epochs: BTreeMap<String, Uuid>,
     minimum_version: u32,
-) -> (CrossPoolFencePolicyResult, CrossPoolFencePolicyResult, CrossPoolFencePolicyResult) {
+) -> (
+    CrossPoolFencePolicyResult,
+    CrossPoolFencePolicyResult,
+    CrossPoolFencePolicyResult,
+    CrossPoolFencePolicyResult,
+) {
     let journal_result = if minimum_version >= TIER_DELETE_JOURNAL_POLICY_SUPPORTED_VERSION {
         Ok(peer_epochs.clone())
     } else {
@@ -78,7 +93,18 @@ fn cross_pool_fence_policy_results(
     } else {
         Err(Error::other("decommission target fence policy capability version is unsupported"))
     };
-    (Ok(peer_epochs), journal_result, decommission_target_fence_result)
+    let legacy_transition_state_reconcile_result =
+        if minimum_version >= LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION {
+            Ok(peer_epochs.clone())
+        } else {
+            Err(Error::other("legacy transition state reconcile policy capability version is unsupported"))
+        };
+    (
+        Ok(peer_epochs),
+        journal_result,
+        decommission_target_fence_result,
+        legacy_transition_state_reconcile_result,
+    )
 }
 
 #[derive(Clone, Debug)]
@@ -252,10 +278,21 @@ pub(crate) struct TierDeleteJournalFleetProofToken {
     _permit: FleetCapabilityProofPermit,
 }
 
+/// Effect-window authority for one legacy transition-state reconciliation.
+///
+/// The token intentionally cannot be cloned. Its permit keeps the admitted
+/// fleet generation alive until the caller finishes the final strong
+/// readback, while revocation makes every later validation fail immediately.
+pub struct LegacyTransitionStateReconcileFleetProofToken {
+    token: FleetCapabilityProofToken,
+    _permit: FleetCapabilityProofPermit,
+}
+
 static REMOTE_VERSION_STATE_FLEET_PROOF: OnceLock<std::sync::RwLock<FleetCapabilityProofState>> = OnceLock::new();
 static CROSS_POOL_FENCE_FLEET_PROOF: OnceLock<std::sync::RwLock<FleetCapabilityProofState>> = OnceLock::new();
 static TIER_DELETE_JOURNAL_FLEET_PROOF: OnceLock<std::sync::RwLock<FleetCapabilityProofState>> = OnceLock::new();
 static DECOMMISSION_TARGET_FENCE_FLEET_PROOF: OnceLock<std::sync::RwLock<FleetCapabilityProofState>> = OnceLock::new();
+static LEGACY_TRANSITION_STATE_RECONCILE_FLEET_PROOF: OnceLock<std::sync::RwLock<FleetCapabilityProofState>> = OnceLock::new();
 static REMOTE_VERSION_STATE_PROBE_TOPOLOGY: OnceLock<String> = OnceLock::new();
 
 fn cross_pool_fence_fleet_proof_slot() -> &'static std::sync::RwLock<FleetCapabilityProofState> {
@@ -274,6 +311,10 @@ fn decommission_target_fence_fleet_proof_slot() -> &'static std::sync::RwLock<Fl
     DECOMMISSION_TARGET_FENCE_FLEET_PROOF.get_or_init(|| std::sync::RwLock::new(FleetCapabilityProofState::default()))
 }
 
+fn legacy_transition_state_reconcile_fleet_proof_slot() -> &'static std::sync::RwLock<FleetCapabilityProofState> {
+    LEGACY_TRANSITION_STATE_RECONCILE_FLEET_PROOF.get_or_init(|| std::sync::RwLock::new(FleetCapabilityProofState::default()))
+}
+
 fn revoke_fleet_capability_proof_state(state: &mut FleetCapabilityProofState) {
     if let Some(proof) = state.proof.take() {
         proof.generation.revoke();
@@ -444,6 +485,125 @@ pub(crate) fn tier_delete_journal_topology_generation(proof: &TierDeleteJournalF
     stable_tier_delete_journal_topology_generation(&proof.token.topology_fingerprint)
 }
 
+/// Acquire one non-cloneable authority that must span the complete reconcile
+/// effect window, including its final strong readback.
+pub async fn acquire_legacy_transition_state_reconcile_fleet_proof() -> Option<LegacyTransitionStateReconcileFleetProofToken> {
+    let expected_topology = REMOTE_VERSION_STATE_PROBE_TOPOLOGY.get()?;
+    let proof = {
+        let state = legacy_transition_state_reconcile_fleet_proof_slot()
+            .read()
+            .unwrap_or_else(std::sync::PoisonError::into_inner);
+        acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, expected_topology, Instant::now())?
+    };
+    let observed_peer_epochs = observe_legacy_transition_state_reconcile_fleet(expected_topology).await?;
+    let state = legacy_transition_state_reconcile_fleet_proof_slot()
+        .read()
+        .unwrap_or_else(std::sync::PoisonError::into_inner);
+    legacy_transition_state_reconcile_fleet_proof_matches_observation_at(
+        &state,
+        &proof,
+        expected_topology,
+        &observed_peer_epochs,
+        Instant::now(),
+    )
+    .then_some(proof)
+}
+
+fn acquire_legacy_transition_state_reconcile_fleet_proof_from(
+    state: &FleetCapabilityProofState,
+    expected_topology: &str,
+    now: Instant,
+) -> Option<LegacyTransitionStateReconcileFleetProofToken> {
+    let token = acquire_fleet_capability_proof_from(state, expected_topology, now)?;
+    let permit = state.proof.as_ref()?.generation.try_acquire()?;
+    Some(LegacyTransitionStateReconcileFleetProofToken { token, _permit: permit })
+}
+
+async fn observe_legacy_transition_state_reconcile_fleet(expected_topology: &str) -> Option<BTreeMap<String, Uuid>> {
+    let notification_sys = get_global_notification_sys()?;
+    let (peer_epochs, minimum_version) = timeout(
+        REMOTE_VERSION_STATE_PROBE_TIMEOUT,
+        notification_sys.probe_cross_pool_fence_fleet(expected_topology),
+    )
+    .await
+    .ok()?
+    .ok()?;
+    let (_, _, _, reconcile_result) = cross_pool_fence_policy_results(peer_epochs, minimum_version);
+    reconcile_result.ok()
+}
+
+/// Revalidate the exact fleet generation captured by a reconcile token with a
+/// fresh synchronous observation. Callers must await this before each
+/// conditional metadata write and after the final strong readback.
+pub async fn legacy_transition_state_reconcile_fleet_proof_matches(
+    proof: &LegacyTransitionStateReconcileFleetProofToken,
+) -> bool {
+    let Some(expected_topology) = REMOTE_VERSION_STATE_PROBE_TOPOLOGY.get() else {
+        return false;
+    };
+    legacy_transition_state_reconcile_fleet_proof_matches_with_observer(
+        legacy_transition_state_reconcile_fleet_proof_slot(),
+        proof,
+        expected_topology,
+        || observe_legacy_transition_state_reconcile_fleet(expected_topology),
+    )
+    .await
+}
+
+async fn legacy_transition_state_reconcile_fleet_proof_matches_with_observer<F, Fut>(
+    slot: &std::sync::RwLock<FleetCapabilityProofState>,
+    proof: &LegacyTransitionStateReconcileFleetProofToken,
+    expected_topology: &str,
+    observe: F,
+) -> bool
+where
+    F: FnOnce() -> Fut,
+    Fut: Future<Output = Option<BTreeMap<String, Uuid>>>,
+{
+    {
+        let state = slot.read().unwrap_or_else(std::sync::PoisonError::into_inner);
+        if !legacy_transition_state_reconcile_fleet_proof_matches_at(&state, proof, expected_topology, Instant::now()) {
+            return false;
+        }
+    }
+    let Some(observed_peer_epochs) = observe().await else {
+        return false;
+    };
+    let state = slot.read().unwrap_or_else(std::sync::PoisonError::into_inner);
+    legacy_transition_state_reconcile_fleet_proof_matches_observation_at(
+        &state,
+        proof,
+        expected_topology,
+        &observed_peer_epochs,
+        Instant::now(),
+    )
+}
+
+fn legacy_transition_state_reconcile_fleet_proof_matches_at(
+    state: &FleetCapabilityProofState,
+    proof: &LegacyTransitionStateReconcileFleetProofToken,
+    expected_topology: &str,
+    now: Instant,
+) -> bool {
+    proof._permit.generation.is_accepting()
+        && fleet_capability_proof_matches_at(state, &proof.token, expected_topology, now)
+        && state
+            .proof
+            .as_ref()
+            .is_some_and(|current| Arc::ptr_eq(&current.generation, &proof._permit.generation))
+}
+
+fn legacy_transition_state_reconcile_fleet_proof_matches_observation_at(
+    state: &FleetCapabilityProofState,
+    proof: &LegacyTransitionStateReconcileFleetProofToken,
+    expected_topology: &str,
+    observed_peer_epochs: &BTreeMap<String, Uuid>,
+    now: Instant,
+) -> bool {
+    legacy_transition_state_reconcile_fleet_proof_matches_at(state, proof, expected_topology, now)
+        && proof.token.peer_epochs.as_ref() == observed_peer_epochs
+}
+
 #[cfg(all(test, feature = "test-util"))]
 pub(crate) fn tier_delete_journal_fleet_proof_has_inflight_for_test() -> bool {
     let state = tier_delete_journal_fleet_proof_slot()
@@ -766,6 +926,7 @@ pub fn start_remote_version_state_fleet_probe(topology_fingerprint: String) {
                 cross_pool_fence_fleet_proof_slot(),
                 tier_delete_journal_fleet_proof_slot(),
                 decommission_target_fence_fleet_proof_slot(),
+                legacy_transition_state_reconcile_fleet_proof_slot(),
             ] {
                 mark_fleet_capability_topology_conflict(slot);
             }
@@ -798,11 +959,12 @@ pub fn start_remote_version_state_fleet_probe(topology_fingerprint: String) {
                 .unwrap_or_else(|_| Err(Error::other("cross-pool fence fleet capability probe timed out"))),
                 None => Err(Error::other("cross-pool fence fleet capability notification system is unavailable")),
             };
-            let (fence_result, journal_result, decommission_target_fence_result) = match fence_probe {
+            let (fence_result, journal_result, decommission_target_fence_result, reconcile_result) = match fence_probe {
                 Ok((peer_epochs, minimum_version)) => cross_pool_fence_policy_results(peer_epochs, minimum_version),
                 Err(err) => {
                     let message = err.to_string();
                     (
+                        Err(Error::other(message.clone())),
                         Err(Error::other(message.clone())),
                         Err(Error::other(message.clone())),
                         Err(Error::other(message)),
@@ -818,6 +980,7 @@ pub fn start_remote_version_state_fleet_probe(topology_fingerprint: String) {
                 revoke_fleet_capability_proof(cross_pool_fence_fleet_proof_slot());
                 revoke_fleet_capability_proof(tier_delete_journal_fleet_proof_slot());
                 revoke_fleet_capability_proof(decommission_target_fence_fleet_proof_slot());
+                revoke_fleet_capability_proof(legacy_transition_state_reconcile_fleet_proof_slot());
             } else if let Some(err) = publish_fleet_capability_probe_result(
                 remote_version_state_fleet_proof_slot(),
                 &topology_fingerprint,
@@ -880,6 +1043,24 @@ pub fn start_remote_version_state_fleet_probe(topology_fingerprint: String) {
                     "notification capability probe"
                 );
             }
+            if !topology_conflict
+                && let Some(err) = publish_fleet_capability_probe_result(
+                    legacy_transition_state_reconcile_fleet_proof_slot(),
+                    &topology_fingerprint,
+                    reconcile_result,
+                    Instant::now(),
+                )
+            {
+                debug!(
+                    event = EVENT_NOTIFICATION_CAPABILITY_PROBE,
+                    component = LOG_COMPONENT_ECSTORE,
+                    subsystem = LOG_SUBSYSTEM_NOTIFICATION,
+                    capability = "legacy_transition_state_reconcile_v1",
+                    state = "failed_closed",
+                    error = %err,
+                    "notification capability probe"
+                );
+            }
             sleep(REMOTE_VERSION_STATE_PROBE_INTERVAL).await;
         }
     });
@@ -959,7 +1140,7 @@ impl NotificationSys {
             client.probe_cross_pool_fence(topology_fingerprint.to_string()).await
         });
         let mut peer_epochs = BTreeMap::new();
-        let mut minimum_version = u32::MAX;
+        let mut minimum_version = LOCAL_CROSS_POOL_FENCE_POLICY_SUPPORTED_VERSION;
         for result in join_all(probes).await {
             let (peer, version, epoch) = result?;
             if version < CROSS_POOL_FENCE_SUPPORTED_VERSION {
@@ -968,11 +1149,6 @@ impl NotificationSys {
             minimum_version = minimum_version.min(version);
             insert_remote_version_state_peer(&mut peer_epochs, peer, epoch)?;
         }
-        // A single-node deployment has no remote member to lower the local
-        // policy version advertised by this binary.
-        if minimum_version == u32::MAX {
-            minimum_version = DECOMMISSION_TARGET_FENCE_POLICY_SUPPORTED_VERSION;
-        }
         Ok((peer_epochs, minimum_version))
     }
 }
@@ -3190,20 +3366,36 @@ mod tests {
     #[test]
     fn cross_pool_policy_versions_authorize_only_their_supported_protocols() {
         let peers = BTreeMap::from([("node-b:9000".to_string(), Uuid::new_v4())]);
-        let (generic_v2, journal_v2, decommission_v2) = cross_pool_fence_policy_results(peers.clone(), 2);
+        let (generic_v2, journal_v2, decommission_v2, reconcile_v2) = cross_pool_fence_policy_results(peers.clone(), 2);
         assert!(generic_v2.is_ok(), "v2 remains valid for existing cross-pool fencing");
         assert!(journal_v2.is_err(), "a mixed v2/v3 fleet must fail closed for journal-v6 deletion");
         assert!(decommission_v2.is_err(), "v2 cannot authorize the sticky per-target decommission fence");
+        assert!(reconcile_v2.is_err(), "v2 cannot authorize legacy transition-state reconciliation");
 
-        let (generic_v3, journal_v3, decommission_v3) = cross_pool_fence_policy_results(peers.clone(), 3);
+        let (generic_v3, journal_v3, decommission_v3, reconcile_v3) = cross_pool_fence_policy_results(peers.clone(), 3);
         assert!(generic_v3.is_ok());
         assert!(journal_v3.is_ok(), "an all-v3 fleet may authorize journal-v6 deletion");
         assert!(decommission_v3.is_err(), "v3 members do not understand the per-target decommission fence");
+        assert!(reconcile_v3.is_err());
 
-        let (generic_v4, journal_v4, decommission_v4) = cross_pool_fence_policy_results(peers, 4);
+        let (generic_v4, journal_v4, decommission_v4, reconcile_v4) =
+            cross_pool_fence_policy_results(peers.clone(), LOCAL_CROSS_POOL_FENCE_POLICY_SUPPORTED_VERSION);
         assert!(generic_v4.is_ok());
         assert!(journal_v4.is_ok());
         assert!(decommission_v4.is_ok(), "an all-v4 fleet may create sticky per-target reservations");
+        assert!(
+            reconcile_v4.is_err(),
+            "the current local policy lacks the conditional xl.meta writer required by reconcile"
+        );
+
+        let (generic_v5, journal_v5, decommission_v5, reconcile_v5) = cross_pool_fence_policy_results(peers, 5);
+        assert!(generic_v5.is_ok());
+        assert!(journal_v5.is_ok());
+        assert!(decommission_v5.is_ok());
+        assert!(
+            reconcile_v5.is_ok(),
+            "only an all-v5 fleet preserves destination identity and conditional reconcile writes"
+        );
     }
 
     #[test]
@@ -3458,6 +3650,234 @@ mod tests {
         );
     }
 
+    #[test]
+    fn legacy_transition_state_reconcile_admits_only_compatible_single_and_multi_node_fleets() {
+        let now = Instant::now();
+        for peers in [
+            BTreeMap::new(),
+            BTreeMap::from([("peer-a".to_string(), Uuid::new_v4()), ("peer-b".to_string(), Uuid::new_v4())]),
+        ] {
+            let slot = std::sync::RwLock::new(FleetCapabilityProofState::default());
+            let (_, _, _, result) =
+                cross_pool_fence_policy_results(peers, LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION);
+            assert!(publish_fleet_capability_probe_result(&slot, "topology-a", result, now).is_none());
+
+            let admitted = {
+                let state = slot.read().expect("reconcile proof slot should not poison");
+                acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, "topology-a", now)
+                    .expect("an all-compatible fleet should admit reconciliation")
+            };
+            let state = slot.read().expect("reconcile proof slot should not poison");
+            assert!(legacy_transition_state_reconcile_fleet_proof_matches_at(
+                &state,
+                &admitted,
+                "topology-a",
+                now,
+            ));
+        }
+    }
+
+    #[test]
+    fn legacy_transition_state_reconcile_restart_drains_concurrent_effect_windows() {
+        let slot = std::sync::RwLock::new(FleetCapabilityProofState::default());
+        let now = Instant::now();
+        let original_peers = BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())]);
+        let (_, _, _, original_result) =
+            cross_pool_fence_policy_results(original_peers, LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION);
+        assert!(publish_fleet_capability_probe_result(&slot, "topology-a", original_result, now).is_none());
+
+        let (first, second) = {
+            let state = slot.read().expect("reconcile proof slot should not poison");
+            (
+                acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, "topology-a", now)
+                    .expect("the first reconcile writer should be admitted"),
+                acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, "topology-a", now)
+                    .expect("the second reconcile writer should be admitted"),
+            )
+        };
+
+        let restarted_peers = BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())]);
+        let (_, _, _, restarted_result) =
+            cross_pool_fence_policy_results(restarted_peers.clone(), LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION);
+        let blocked =
+            publish_fleet_capability_probe_result(&slot, "topology-a", restarted_result, now + Duration::from_millis(1))
+                .expect("a restarted member must revoke the old generation and wait for both writers");
+        assert!(blocked.to_string().contains("previous generation to drain"));
+        {
+            let state = slot.read().expect("reconcile proof slot should not poison");
+            assert!(state.proof.is_none());
+            assert!(state.draining_generation.is_some());
+            assert!(!legacy_transition_state_reconcile_fleet_proof_matches_at(
+                &state,
+                &first,
+                "topology-a",
+                now + Duration::from_millis(1),
+            ));
+            assert!(!legacy_transition_state_reconcile_fleet_proof_matches_at(
+                &state,
+                &second,
+                "topology-a",
+                now + Duration::from_millis(1),
+            ));
+        }
+
+        drop(first);
+        let (_, _, _, still_blocked_result) =
+            cross_pool_fence_policy_results(restarted_peers.clone(), LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION);
+        assert!(
+            publish_fleet_capability_probe_result(&slot, "topology-a", still_blocked_result, now + Duration::from_millis(2),)
+                .is_some(),
+            "one remaining writer must keep the successor generation closed"
+        );
+
+        drop(second);
+        let (_, _, _, admitted_result) =
+            cross_pool_fence_policy_results(restarted_peers, LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION);
+        assert!(
+            publish_fleet_capability_probe_result(&slot, "topology-a", admitted_result, now + Duration::from_millis(3),)
+                .is_none(),
+            "the restarted generation may publish only after every old writer drains"
+        );
+    }
+
+    #[test]
+    fn legacy_transition_state_reconcile_fresh_observation_closes_the_polling_window() {
+        let slot = std::sync::RwLock::new(FleetCapabilityProofState::default());
+        let now = Instant::now();
+        let original_peers = BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())]);
+        let (_, _, _, original_result) =
+            cross_pool_fence_policy_results(original_peers.clone(), LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION);
+        assert!(publish_fleet_capability_probe_result(&slot, "topology-a", original_result, now).is_none());
+        let admitted = {
+            let state = slot.read().expect("reconcile proof slot should not poison");
+            acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, "topology-a", now)
+                .expect("the original fleet should admit reconciliation")
+        };
+
+        let restarted_peers = BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())]);
+        let state = slot.read().expect("reconcile proof slot should not poison");
+        assert!(
+            legacy_transition_state_reconcile_fleet_proof_matches_at(&state, &admitted, "topology-a", now),
+            "the periodic cache has not observed the restart yet"
+        );
+        assert!(!legacy_transition_state_reconcile_fleet_proof_matches_observation_at(
+            &state,
+            &admitted,
+            "topology-a",
+            &restarted_peers,
+            now,
+        ));
+
+        let (_, _, _, downgraded) = cross_pool_fence_policy_results(original_peers, 4);
+        assert!(
+            downgraded.is_err(),
+            "a synchronous observation of a downgraded peer must fail before any cached proof can authorize a write"
+        );
+    }
+
+    #[tokio::test]
+    async fn legacy_transition_state_reconcile_invalid_token_skips_fleet_observation() {
+        let slot = std::sync::RwLock::new(FleetCapabilityProofState::default());
+        let now = Instant::now();
+        let peers = BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())]);
+        assert!(publish_fleet_capability_probe_result(&slot, "topology-a", Ok(peers), now).is_none());
+        let admitted = {
+            let state = slot.read().expect("reconcile proof slot should not poison");
+            acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, "topology-a", now)
+                .expect("the original fleet should admit reconciliation")
+        };
+        revoke_fleet_capability_proof(&slot);
+
+        assert!(
+            !legacy_transition_state_reconcile_fleet_proof_matches_with_observer(&slot, &admitted, "topology-a", || async {
+                panic!("an invalid local generation must not trigger a fleet observation");
+            },)
+            .await
+        );
+    }
+
+    #[test]
+    fn legacy_transition_state_reconcile_membership_and_topology_changes_revoke_authority() {
+        let now = Instant::now();
+        for replacement in [
+            BTreeMap::from([("peer-a".to_string(), Uuid::new_v4()), ("peer-b".to_string(), Uuid::new_v4())]),
+            BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())]),
+        ] {
+            let slot = std::sync::RwLock::new(FleetCapabilityProofState::default());
+            let original = BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())]);
+            assert!(publish_fleet_capability_probe_result(&slot, "topology-a", Ok(original), now).is_none());
+            let admitted = {
+                let state = slot.read().expect("reconcile proof slot should not poison");
+                acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, "topology-a", now)
+                    .expect("the original fleet should admit reconciliation")
+            };
+
+            assert!(
+                publish_fleet_capability_probe_result(&slot, "topology-a", Ok(replacement), now + Duration::from_millis(1),)
+                    .is_some(),
+                "membership or process-epoch replacement must wait for the admitted writer"
+            );
+            let state = slot.read().expect("reconcile proof slot should not poison");
+            assert!(!legacy_transition_state_reconcile_fleet_proof_matches_at(
+                &state,
+                &admitted,
+                "topology-a",
+                now + Duration::from_millis(1),
+            ));
+        }
+
+        let slot = std::sync::RwLock::new(FleetCapabilityProofState::default());
+        assert!(publish_fleet_capability_probe_result(&slot, "topology-a", Ok(BTreeMap::new()), now).is_none());
+        let admitted = {
+            let state = slot.read().expect("reconcile proof slot should not poison");
+            acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, "topology-a", now)
+                .expect("the original topology should admit reconciliation")
+        };
+        mark_fleet_capability_topology_conflict(&slot);
+        let state = slot.read().expect("reconcile proof slot should not poison");
+        assert!(state.topology_conflict);
+        assert!(!legacy_transition_state_reconcile_fleet_proof_matches_at(
+            &state,
+            &admitted,
+            "topology-a",
+            now,
+        ));
+    }
+
+    #[test]
+    fn legacy_transition_state_reconcile_capability_downgrade_fails_closed() {
+        let slot = std::sync::RwLock::new(FleetCapabilityProofState::default());
+        let now = Instant::now();
+        let peers = BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())]);
+        let (_, _, _, compatible_result) =
+            cross_pool_fence_policy_results(peers.clone(), LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION);
+        assert!(publish_fleet_capability_probe_result(&slot, "topology-a", compatible_result, now).is_none());
+        let admitted = {
+            let state = slot.read().expect("reconcile proof slot should not poison");
+            acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, "topology-a", now)
+                .expect("v5 should admit reconciliation")
+        };
+
+        let (_, _, _, downgraded_result) =
+            cross_pool_fence_policy_results(peers, LEGACY_TRANSITION_STATE_RECONCILE_POLICY_SUPPORTED_VERSION - 1);
+        let err = publish_fleet_capability_probe_result(&slot, "topology-a", downgraded_result, now + Duration::from_millis(1))
+            .expect("a v4 member must revoke reconcile authority");
+        assert!(err.to_string().contains("reconcile policy capability version is unsupported"));
+        let state = slot.read().expect("reconcile proof slot should not poison");
+        assert!(state.proof.is_none());
+        assert!(!legacy_transition_state_reconcile_fleet_proof_matches_at(
+            &state,
+            &admitted,
+            "topology-a",
+            now + Duration::from_millis(1),
+        ));
+        assert!(
+            acquire_legacy_transition_state_reconcile_fleet_proof_from(&state, "topology-a", now + Duration::from_millis(1),)
+                .is_none(),
+            "a downgraded fleet must remain inspect-only"
+        );
+    }
+
     #[test]
     fn remote_version_state_fleet_proof_conflict_revokes_atomic_snapshot() {
         let now = Instant::now();
@@ -3539,6 +3959,57 @@ mod tests {
         assert!(err.to_string().contains("incomplete"));
     }
 
+    #[tokio::test]
+    async fn legacy_transition_state_reconcile_probe_rejects_missing_or_unreachable_members() {
+        let missing = NotificationSys {
+            peer_clients: Vec::new(),
+            all_peer_clients: vec![None],
+            peer_topology_hosts: vec!["peer-a".to_string()],
+            peer_admin_caches: Vec::new(),
+            tier_config_reload_workers: Default::default(),
+        };
+        let missing_err = missing
+            .probe_cross_pool_fence_fleet("topology-a")
+            .await
+            .expect_err("a missing member slot must prevent reconcile capability proof");
+        assert!(missing_err.to_string().contains("incomplete"));
+
+        let unreachable = NotificationSys {
+            peer_clients: vec![None],
+            all_peer_clients: vec![None, None],
+            peer_topology_hosts: vec!["peer-a".to_string()],
+            peer_admin_caches: vec![Mutex::new(PeerAdminCache::new())],
+            tier_config_reload_workers: Default::default(),
+        };
+        let unreachable_err = unreachable
+            .probe_cross_pool_fence_fleet("topology-a")
+            .await
+            .expect_err("an unreachable member must prevent reconcile capability proof");
+        assert!(unreachable_err.to_string().contains("unreachable"));
+    }
+
+    #[tokio::test]
+    async fn legacy_transition_state_reconcile_single_node_stays_closed_before_local_cas_support() {
+        let notification_sys = NotificationSys {
+            peer_clients: Vec::new(),
+            all_peer_clients: vec![None],
+            peer_topology_hosts: Vec::new(),
+            peer_admin_caches: Vec::new(),
+            tier_config_reload_workers: Default::default(),
+        };
+        let (peers, minimum_version) = notification_sys
+            .probe_cross_pool_fence_fleet("topology-a")
+            .await
+            .expect("a single-node capability probe should complete");
+        assert!(peers.is_empty());
+        assert_eq!(minimum_version, LOCAL_CROSS_POOL_FENCE_POLICY_SUPPORTED_VERSION);
+        let (_, _, _, reconcile_result) = cross_pool_fence_policy_results(peers, minimum_version);
+        assert!(
+            reconcile_result.is_err(),
+            "the current node must not self-authorize reconcile before the conditional writer lands"
+        );
+    }
+
     fn build_props(endpoint: &str) -> ServerProperties {
         ServerProperties {
             endpoint: endpoint.to_string(),

crates/ecstore/src/services/tier/mod.rs

@@ -21,9 +21,11 @@ pub mod tier_gen;
 pub mod tier_handlers;
 pub(crate) mod tier_mutation_intent;
 pub mod tier_mutation_peer;
+pub(crate) mod tier_probe_intent;
 pub mod warm_backend;
 pub mod warm_backend_aliyun;
 pub mod warm_backend_azure;
+#[cfg(feature = "gcs")]
 pub mod warm_backend_gcs;
 pub mod warm_backend_huaweicloud;
 pub mod warm_backend_minio;

crates/ecstore/src/services/tier/test_util.rs

@@ -56,6 +56,7 @@
 
 use std::collections::HashMap;
 use std::io::Cursor;
+#[cfg(feature = "test-util")]
 use std::path::Path;
 use std::sync::{
     Arc,
@@ -68,21 +69,28 @@ use tokio::io::AsyncReadExt;
 use tokio::sync::{Mutex, Notify, RwLock};
 use uuid::Uuid;
 
+#[cfg(feature = "test-util")]
 use crate::disk::endpoint::Endpoint;
+#[cfg(feature = "test-util")]
 use crate::disk::format::FormatV3;
+#[cfg(feature = "test-util")]
 use crate::disk::{DiskAPI, DiskOption, FORMAT_CONFIG_FILE, RUSTFS_META_BUCKET, STORAGE_FORMAT_FILE, new_disk};
 use crate::services::tier::tier::TierConfigMgr;
 use crate::services::tier::tier_config::{TierConfig, TierMinIO, TierType};
 use crate::services::tier::warm_backend::{
     TransitionCandidateProbe, WarmBackend, WarmBackendGetOpts, build_transition_put_options,
 };
+#[cfg(feature = "test-util")]
 use rustfs_filemeta::FileMeta;
 use rustfs_s3_client::transition_api::{ReadCloser, ReaderImpl};
+#[cfg(feature = "test-util")]
 use rustfs_utils::path::path_join_buf;
 
 /// One-shot barrier before rejected transition cleanup resolves its ECStore.
+#[cfg(feature = "test-util")]
 pub struct TransitionCleanupStoreBarrier(crate::set_disk::SetDiskTransitionCleanupStoreBarrier);
 
+#[cfg(feature = "test-util")]
 impl TransitionCleanupStoreBarrier {
     /// Install the barrier for the next rejected transition cleanup.
     pub fn install() -> Self {
@@ -96,6 +104,7 @@ impl TransitionCleanupStoreBarrier {
 }
 
 /// Default polling cadence used by the `wait_for_*` helpers.
+#[cfg(feature = "test-util")]
 const POLL_INTERVAL: Duration = Duration::from_millis(50);
 
 /// A fault to inject into [`MockWarmBackend`] operations.
@@ -208,10 +217,12 @@ impl Drop for MockRemoveOperationGuard {
 }
 
 /// One-shot barrier that pauses a mock tier PUT after storing its remote body.
+#[cfg(feature = "test-util")]
 pub struct MockPutBarrier {
     state: Arc<MockPutBarrierState>,
 }
 
+#[cfg(feature = "test-util")]
 impl MockPutBarrier {
     /// Wait until the remote body is stored and the PUT is paused before returning.
     pub async fn wait_until_paused(&self) {
@@ -226,6 +237,7 @@ impl MockPutBarrier {
     }
 }
 
+#[cfg(feature = "test-util")]
 impl Drop for MockPutBarrier {
     fn drop(&mut self) {
         self.state.release.notify_one();
@@ -258,10 +270,12 @@ impl Drop for MockGetBarrier {
 }
 
 /// One-shot barrier that pauses and then fails a mock tier DELETE.
+#[cfg(feature = "test-util")]
 pub struct MockRemoveBarrier {
     state: Arc<MockRemoveBarrierState>,
 }
 
+#[cfg(feature = "test-util")]
 impl MockRemoveBarrier {
     /// Wait until DELETE reaches the deterministic failure point.
     pub async fn wait_until_paused(&self) {
@@ -283,6 +297,7 @@ impl MockRemoveBarrier {
     }
 }
 
+#[cfg(feature = "test-util")]
 impl Drop for MockRemoveBarrier {
     fn drop(&mut self) {
         self.state.release.notify_one();
@@ -306,6 +321,7 @@ impl MockWarmBackend {
     }
 
     /// Arm a one-shot pause after the next tier PUT stores its remote body.
+    #[cfg(feature = "test-util")]
     pub async fn arm_put_barrier(&self) -> MockPutBarrier {
         let state = Arc::new(MockPutBarrierState::default());
         *self.inner.put_barrier.lock().await = Some(Arc::clone(&state));
@@ -313,6 +329,7 @@ impl MockWarmBackend {
     }
 
     /// Pause and then fail the next DELETE after it reaches the backend.
+    #[cfg(feature = "test-util")]
     pub async fn arm_failing_remove_barrier(&self) -> MockRemoveBarrier {
         let state = Arc::new(MockRemoveBarrierState::default());
         let mut barrier = self.inner.remove_barrier.lock().await;
@@ -323,6 +340,7 @@ impl MockWarmBackend {
 
     /// Arm a one-shot pause before the next tier GET, then return an error
     /// after the test releases it.
+    #[cfg(feature = "test-util")]
     pub async fn arm_failing_get_barrier(&self) -> MockGetBarrier {
         let state = Arc::new(MockGetBarrierState {
             fail_after_release: true,
@@ -343,6 +361,7 @@ impl MockWarmBackend {
     // ---- fault injection -------------------------------------------------
 
     /// Replace the entire fault configuration.
+    #[cfg(feature = "test-util")]
     pub async fn set_faults(&self, faults: FaultConfig) {
         *self.inner.faults.lock().await = faults;
     }
@@ -353,6 +372,7 @@ impl MockWarmBackend {
     }
 
     /// Toggle "HTTP 5xx" server errors on every operation.
+    #[cfg(feature = "test-util")]
     pub async fn set_server_error(&self, server_error: bool) {
         self.inner.faults.lock().await.server_error = server_error;
     }
@@ -363,11 +383,13 @@ impl MockWarmBackend {
     }
 
     /// Set (or clear, with `None`) injected latency applied before each op.
+    #[cfg(feature = "test-util")]
     pub async fn set_latency(&self, latency: Option<Duration>) {
         self.inner.faults.lock().await.latency = latency;
     }
 
     /// Clear all injected faults, restoring healthy behaviour.
+    #[cfg(feature = "test-util")]
     pub async fn clear_faults(&self) {
         *self.inner.faults.lock().await = FaultConfig::default();
     }
@@ -375,6 +397,7 @@ impl MockWarmBackend {
     /// Limit how many body bytes a successful mock PUT consumes. `None` drains
     /// the complete body. This models a backend that incorrectly accepts a
     /// truncated stream while still returning success.
+    #[cfg(feature = "test-util")]
     pub async fn set_put_read_limit(&self, limit: Option<usize>) {
         *self.inner.put_read_limit.lock().await = limit;
     }
@@ -395,12 +418,14 @@ impl MockWarmBackend {
     }
 
     /// Reject non-empty remote versions before transition metadata is committed.
+    #[cfg(feature = "test-util")]
     pub fn set_reject_non_empty_remote_versions(&self, reject: bool) {
         self.inner.reject_non_empty_remote_versions.store(reject, Ordering::Release);
     }
 
     /// Reject the next non-empty remote version validation without changing
     /// subsequent exact-version backend cleanup behavior.
+    #[cfg(feature = "test-util")]
     pub fn reject_next_non_empty_remote_version_validation(&self) {
         self.inner
             .reject_non_empty_remote_version_validations
@@ -438,6 +463,7 @@ impl MockWarmBackend {
     }
 
     /// Clear the operation log without touching stored objects or faults.
+    #[cfg(feature = "test-util")]
     pub async fn clear_op_log(&self) {
         self.inner.op_log.lock().await.clear();
     }
@@ -459,11 +485,13 @@ impl MockWarmBackend {
     }
 
     /// Return the exact object/version pairs produced by successful tier PUTs.
+    #[cfg(feature = "test-util")]
     pub async fn put_versions(&self) -> Vec<(String, String)> {
         self.inner.put_versions.lock().await.clone()
     }
 
     /// Return the exact object/version pairs passed to successful tier removes.
+    #[cfg(feature = "test-util")]
     pub async fn remove_versions(&self) -> Vec<(String, String)> {
         self.inner.remove_versions.lock().await.clone()
     }
@@ -475,6 +503,7 @@ impl MockWarmBackend {
 
     /// Number of `get` calls recorded — useful to assert restore reads hit the
     /// local copy rather than the remote tier.
+    #[cfg(feature = "test-util")]
     pub async fn get_count(&self) -> usize {
         self.inner
             .op_log
@@ -486,6 +515,7 @@ impl MockWarmBackend {
     }
 
     /// Number of `put` calls recorded.
+    #[cfg(feature = "test-util")]
     pub async fn put_count(&self) -> usize {
         self.inner
             .op_log
@@ -499,6 +529,7 @@ impl MockWarmBackend {
     // ---- storage inspection ---------------------------------------------
 
     /// Whether the backend currently stores `object`.
+    #[cfg(feature = "test-util")]
     pub async fn contains(&self, object: &str) -> bool {
         self.inner.objects.lock().await.contains_key(object)
     }
@@ -509,11 +540,13 @@ impl MockWarmBackend {
     }
 
     /// A clone of the stored object, if present.
+    #[cfg(feature = "test-util")]
     pub async fn stored(&self, object: &str) -> Option<MockStoredObject> {
         self.inner.objects.lock().await.get(object).cloned()
     }
 
     /// A clone of the raw bytes stored for `object`, if present.
+    #[cfg(feature = "test-util")]
     pub async fn bytes(&self, object: &str) -> Option<Vec<u8>> {
         self.inner.objects.lock().await.get(object).map(|o| o.bytes.clone())
     }
@@ -538,6 +571,7 @@ impl MockWarmBackend {
 
     /// Poll until `object` is absent from the backend, or `timeout` elapses.
     /// Returns `true` if the object disappeared within the budget.
+    #[cfg(feature = "test-util")]
     pub async fn wait_for_remote_absence(&self, object: &str, timeout: Duration) -> bool {
         let deadline = tokio::time::Instant::now() + timeout;
         loop {
@@ -553,6 +587,7 @@ impl MockWarmBackend {
 
     /// Poll until the backend holds exactly `expected` objects, or `timeout`
     /// elapses. Returns `true` if the count was reached within the budget.
+    #[cfg(feature = "test-util")]
     pub async fn wait_for_object_count(&self, expected: usize, timeout: Duration) -> bool {
         let deadline = tokio::time::Instant::now() + timeout;
         loop {
@@ -701,7 +736,7 @@ impl WarmBackend for MockWarmBackend {
         Ok(version)
     }
 
-    async fn get(&self, object: &str, _rv: &str, opts: WarmBackendGetOpts) -> Result<ReadCloser, std::io::Error> {
+    async fn get(&self, object: &str, rv: &str, opts: WarmBackendGetOpts) -> Result<ReadCloser, std::io::Error> {
         self.precondition().await?;
         let barrier = self.inner.get_barrier.lock().await.take();
         if let Some(barrier) = barrier {
@@ -719,6 +754,9 @@ impl WarmBackend for MockWarmBackend {
         let Some(stored) = objects.get(object) else {
             return Err(std::io::Error::new(std::io::ErrorKind::NotFound, "mock object not found"));
         };
+        if !rv.is_empty() && stored.remote_version_id != rv {
+            return Err(std::io::Error::new(std::io::ErrorKind::NotFound, "NoSuchVersion"));
+        }
         let bytes = &stored.bytes;
 
         let start = opts.start_offset.max(0) as usize;
@@ -844,6 +882,7 @@ pub async fn register_mock_tier_backend(handle: &Arc<RwLock<TierConfigMgr>>, tie
 /// The transition-state tuple read from an on-disk `xl.meta`, plus the object's
 /// free-version count.
 #[derive(Clone, Debug, PartialEq, Eq)]
+#[cfg(feature = "test-util")]
 pub struct TransitionMeta {
     /// `transition_status` (e.g. `"complete"`), empty when not transitioned.
     pub status: String,
@@ -857,6 +896,7 @@ pub struct TransitionMeta {
     pub free_version_count: usize,
 }
 
+#[cfg(feature = "test-util")]
 async fn open_disk(disk_path: &Path) -> Option<crate::disk::DiskStore> {
     // `LocalDisk::new` rejects an endpoint whose (set_idx, disk_idx) disagrees
     // with the position recorded in the disk's own format.json, so derive the
@@ -887,6 +927,7 @@ async fn open_disk(disk_path: &Path) -> Option<crate::disk::DiskStore> {
 /// The free-version metadata removal lands asynchronously after the remote
 /// object disappears, so callers typically poll via
 /// [`wait_for_free_version_absence`] instead of asserting a single read.
+#[cfg(feature = "test-util")]
 pub async fn free_version_count(disk_path: &Path, bucket: &str, object: &str) -> usize {
     let Some(disk) = open_disk(disk_path).await else {
         return 0;
@@ -911,6 +952,7 @@ pub async fn free_version_count(disk_path: &Path, bucket: &str, object: &str) ->
 /// fields are taken from the newest version that carries a transition record;
 /// if no version is transitioned, they are taken from the current version (and
 /// will be empty).
+#[cfg(feature = "test-util")]
 pub async fn read_transition_meta(disk_path: &Path, bucket: &str, object: &str) -> Option<TransitionMeta> {
     let disk = open_disk(disk_path).await?;
     let data = disk
@@ -944,6 +986,7 @@ pub async fn read_transition_meta(disk_path: &Path, bucket: &str, object: &str)
 /// disk is missing the object or disagrees — this is the shard-consistency
 /// check required by ilm-6 (the `(status, tier, remote key, remote version id)`
 /// four-tuple plus free-version count must match across all erasure shards).
+#[cfg(feature = "test-util")]
 pub async fn assert_transition_meta_consistent<P: AsRef<Path>>(disk_paths: &[P], bucket: &str, object: &str) -> TransitionMeta {
     assert!(!disk_paths.is_empty(), "assert_transition_meta_consistent needs at least one disk");
 
@@ -969,6 +1012,7 @@ pub async fn assert_transition_meta_consistent<P: AsRef<Path>>(disk_paths: &[P],
 
 /// Poll until `object` retains no free versions on `disk_path`, or `timeout`
 /// elapses. Returns `true` if the free versions drained within the budget.
+#[cfg(feature = "test-util")]
 pub async fn wait_for_free_version_absence(disk_path: &Path, bucket: &str, object: &str, timeout: Duration) -> bool {
     let deadline = tokio::time::Instant::now() + timeout;
     loop {
@@ -1037,6 +1081,44 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn mock_metadata_survives_put_and_external_delete_is_distinct() {
+        let backend = MockWarmBackend::new();
+        let metadata = HashMap::from([
+            ("content-type".to_string(), "text/plain".to_string()),
+            ("project".to_string(), "archive".to_string()),
+        ]);
+        let version = backend
+            .put_with_meta("object", ReaderImpl::Body(Bytes::from_static(b"body")), 4, metadata.clone())
+            .await
+            .expect("mock PUT should preserve remote metadata");
+        assert_eq!(backend.metadata("object").await, Some(metadata));
+        assert_eq!(
+            backend
+                .probe_transition_candidate_state("object")
+                .await
+                .expect("probe stored object"),
+            TransitionCandidateProbe::VersionedPresent(version)
+        );
+
+        backend.external_remove("object").await;
+        assert_eq!(backend.metadata("object").await, None);
+        assert_eq!(
+            backend
+                .probe_transition_candidate_state("object")
+                .await
+                .expect("probe removed object"),
+            TransitionCandidateProbe::Missing
+        );
+        let operations = backend.op_log().await;
+        assert!(
+            operations
+                .iter()
+                .any(|op| matches!(op, MockWarmOp::ExternalRemove { object } if object == "object"))
+        );
+        assert!(!operations.iter().any(|op| matches!(op, MockWarmOp::Remove { .. })));
+    }
+
     #[tokio::test]
     async fn mock_probe_preserves_fault_fail_closed_behavior() {
         let backend = MockWarmBackend::new();

crates/ecstore/src/services/tier/tier.rs

@@ -143,11 +143,12 @@ struct TierDriverBuildBarrier {
 static TIER_DRIVER_BUILD_BARRIER: LazyLock<Mutex<Option<Arc<TierDriverBuildBarrier>>>> = LazyLock::new(|| Mutex::new(None));
 
 #[cfg(test)]
-type TierDriverTestFactory = Arc<dyn Fn(&TierConfig) -> std::result::Result<WarmBackendImpl, AdminError> + Send + Sync + 'static>;
+pub(crate) type TierDriverTestFactory =
+    Arc<dyn Fn(&TierConfig) -> std::result::Result<WarmBackendImpl, AdminError> + Send + Sync + 'static>;
 
 #[cfg(test)]
 tokio::task_local! {
-    static TIER_DRIVER_TEST_FACTORY: TierDriverTestFactory;
+    pub(crate) static TIER_DRIVER_TEST_FACTORY: TierDriverTestFactory;
 }
 
 #[cfg(test)]
@@ -1377,27 +1378,42 @@ async fn ensure_no_authoritative_persisted_references<S>(
 where
     S: TierReferenceProofStore,
 {
-    ensure_no_authoritative_persisted_references_with(api.clone(), TIER_DELETE_JOURNAL_PREFIX, |_object, data| {
+    ensure_no_authoritative_persisted_references_with(
+        api.clone(),
+        TIER_DELETE_JOURNAL_PREFIX,
+        "tier-delete journal",
+        |_object, data| {
             let journal = decode_tier_delete_journal_entry(data).map_err(io::Error::other)?;
             Ok((
                 journal.tier_name.clone(),
                 tier_persisted_reference_blocks_any_target(&journal.tier_name, journal.backend_identity, targets),
             ))
-    })
+        },
+    )
     .await?;
-    ensure_no_authoritative_persisted_references_with(api, TRANSITION_TRANSACTION_RECORD_PREFIX, |object, data| {
+    ensure_no_authoritative_persisted_references_with(
+        api,
+        TRANSITION_TRANSACTION_RECORD_PREFIX,
+        "transition transaction",
+        |object, data| {
             let transaction = decode_transition_transaction_record(object, data).map_err(io::Error::other)?;
             Ok((
                 transaction.tier_name.clone(),
-            tier_persisted_reference_blocks_any_target(&transaction.tier_name, Some(transaction.backend_fingerprint), targets),
+                tier_persisted_reference_blocks_any_target(
+                    &transaction.tier_name,
+                    Some(transaction.backend_fingerprint),
+                    targets,
+                ),
             ))
-    })
+        },
+    )
     .await
 }
 
 async fn ensure_no_authoritative_persisted_references_with<S, F>(
     api: Arc<S>,
     prefix: &str,
+    reference_kind: &str,
     blocks_target: F,
 ) -> std::result::Result<(), AdminError>
 where
@@ -1426,7 +1442,7 @@ where
                 .map_err(tier_reference_proof_admin_error)?;
             let (tier_name, blocks) = blocks_target(&object.name, &data).map_err(tier_reference_proof_admin_error)?;
             if blocks {
-                return Err(tier_reference_proof_persisted_in_use_error(&tier_name, &object.name));
+                return Err(tier_reference_proof_persisted_in_use_error(&tier_name, reference_kind, &object.name));
             }
         }
         if !page.is_truncated {
@@ -1488,16 +1504,21 @@ fn tier_persisted_reference_blocks_target(
 
 fn tier_reference_proof_in_use_error(tier_name: &str, object: &ObjectInfo) -> AdminError {
     let mut err = ERR_TIER_BACKEND_IN_USE.clone();
+    let reference_kind = if object.transitioned_object.free_version {
+        "free-version ownership"
+    } else {
+        "transitioned-object"
+    };
     err.message = format!(
-        "Remote tier {tier_name} still has object references, for example {}/{}",
+        "Remote tier {tier_name} still has a {reference_kind} reference, for example {}/{}",
         object.bucket, object.name
     );
     err
 }
 
-fn tier_reference_proof_persisted_in_use_error(tier_name: &str, object: &str) -> AdminError {
+fn tier_reference_proof_persisted_in_use_error(tier_name: &str, reference_kind: &str, object: &str) -> AdminError {
     let mut err = ERR_TIER_BACKEND_IN_USE.clone();
-    err.message = format!("Remote tier {tier_name} still has a persisted reference, for example {object}");
+    err.message = format!("Remote tier {tier_name} still has a {reference_kind} reference, for example {object}");
     err
 }
 
@@ -2346,6 +2367,10 @@ impl WarmBackend for SharedWarmBackendProxy {
         self.0.probe_transition_candidate(object).await
     }
 
+    async fn probe_transition_version(&self, object: &str, remote_version_id: &str) -> io::Result<TransitionCandidateProbe> {
+        self.0.probe_transition_version(object, remote_version_id).await
+    }
+
     async fn in_use(&self) -> io::Result<bool> {
         self.0.in_use().await
     }
@@ -2458,6 +2483,15 @@ impl TierOperationLease {
         Ok(())
     }
 
+    pub(crate) async fn probe_transition_version(
+        &self,
+        object: &str,
+        remote_version_id: &str,
+    ) -> io::Result<TransitionCandidateProbe> {
+        self.validate_remote_version_id(remote_version_id)?;
+        self.inner.driver.probe_transition_version(object, remote_version_id).await
+    }
+
     pub(crate) fn is_current_generation(&self) -> bool {
         lock_unpoisoned(&self.runtime)
             .generations
@@ -3528,7 +3562,7 @@ impl TierConfigMgr {
         // Get tier configuration and create new driver
         let tier_config = self.tiers.get(tier_name).ok_or_else(|| ERR_TIER_NOT_FOUND.clone())?;
 
-        let driver = new_warm_backend(tier_config, false).await?;
+        let driver = construct_warm_backend(tier_config).await?;
 
         self.replace_driver(tier_name, driver)?;
         Ok(self
@@ -3672,13 +3706,17 @@ impl TierConfigMgr {
         let manager = handle.read().await;
         let runtime = tier_driver_runtime(handle, &manager);
         let runtime = lock_unpoisoned(&runtime);
-        if !runtime
+        let prepared = runtime
+            .prepared_mutation_blocks
+            .values()
+            .any(|blocked_mutation_id| *blocked_mutation_id == mutation_id);
+        let committed = runtime
             .committed_mutation_blocks
             .values()
-            .any(|mutation_ids| mutation_ids.contains(&mutation_id))
+            .any(|mutation_ids| mutation_ids.contains(&mutation_id));
-        {
+        if !prepared && !committed {
             let mut err = ERR_TIER_INVALID_CONFIG.clone();
-            err.message = "Remote tier committed mutation fence was not installed".to_string();
+            err.message = "Remote tier mutation fence was not installed".to_string();
             return Err(err);
         }
         Ok(MutationBlockAllowance {
@@ -3885,14 +3923,6 @@ impl TierConfigMgr {
         Self::begin_tier_transition_with_destinations(handle, manager, changed, replaced_destinations, mutation_block_allowance)
     }
 
-    fn begin_tier_transition(
-        handle: &Arc<RwLock<Self>>,
-        manager: &mut Self,
-        changed: HashSet<String>,
-    ) -> std::result::Result<TierPublishTransition, AdminError> {
-        Self::begin_tier_transition_with_destinations(handle, manager, changed, HashMap::new(), None)
-    }
-
     fn begin_tier_transition_with_destinations(
         handle: &Arc<RwLock<Self>>,
         manager: &mut Self,
@@ -4163,7 +4193,7 @@ impl TierConfigMgr {
                     let mut config_lock = config_lock;
                     let coordinated_config_update = config_lock.is_some();
                     let mut update = Some(update);
-                    let (mutation_kind, explicit_tier_name, mutation_force, current_for_targets, driver_tier, mut transition) =
+                    let (mutation_kind, explicit_tier_name, mutation_force, current_for_targets, driver_tier, target_tiers) =
                         match mutation {
                             TierCandidateMutation::Prevalidated(prepared) => {
                                 if version != prepared.version {
@@ -4179,9 +4209,8 @@ impl TierConfigMgr {
                                     )));
                                 }
                                 candidate = prepared.candidate;
-                                let validation_deadline = Instant::now() + TIER_REMOTE_VALIDATION_TIMEOUT;
+                                let target_tiers = {
-                                let mut transition = {
+                                    let manager = handle.read().await;
-                                    let mut manager = handle.write().await;
                                     let mut target_tiers = changed_tier_names(&manager, &candidate);
                                     if let Some(tier_name) = prepared.explicit_tier_name.as_ref()
                                         && (manager.tiers.contains_key(tier_name)
@@ -4190,8 +4219,7 @@ impl TierConfigMgr {
                                     {
                                         target_tiers.insert(tier_name.clone());
                                     }
-                                    Self::begin_tier_transition(&handle, &mut manager, target_tiers)
+                                    target_tiers
-                                        .map_err(TierConfigUpdateError::Publish)?
                                 };
                                 (
                                     prepared.kind,
@@ -4199,7 +4227,7 @@ impl TierConfigMgr {
                                     prepared.force,
                                     prepared.current,
                                     prepared.driver_tier,
-                                    transition,
+                                    target_tiers,
                                 )
                             }
                             mutation => {
@@ -4224,11 +4252,9 @@ impl TierConfigMgr {
                                     last_refreshed_at: candidate.last_refreshed_at,
                                 };
                                 let validation_deadline = Instant::now() + TIER_REMOTE_VALIDATION_TIMEOUT;
-                                let mut transition = {
+                                let target_tiers = {
-                                    let mut manager = handle.write().await;
+                                    let manager = handle.read().await;
-                                    let target_tiers = mutation.target_tiers(&manager, &candidate);
+                                    mutation.target_tiers(&manager, &candidate)
-                                    Self::begin_tier_transition(&handle, &mut manager, target_tiers)
-                                        .map_err(TierConfigUpdateError::Publish)?
                                 };
                                 let driver_tier = apply_tier_candidate_mutation(mutation, &mut candidate, validation_deadline)
                                     .await
@@ -4239,7 +4265,7 @@ impl TierConfigMgr {
                                     mutation_force,
                                     current_for_targets,
                                     driver_tier,
-                                    transition,
+                                    target_tiers,
                                 )
                             }
                         };
@@ -4262,28 +4288,83 @@ impl TierConfigMgr {
                     save_coordinator_tier_mutation_intent(api.clone(), coordinator_intent.as_ref())
                         .await
                         .map_err(TierConfigUpdateError::Save)?;
+                    let mut blocked_target_tiers = target_tiers.clone();
                     if let Some(intent) = coordinator_intent.as_ref() {
-                        TierConfigMgr::apply_prepared_mutation_intent_block(&handle, intent)
+                        blocked_target_tiers.extend(intent.affected_targets.iter().map(|target| target.tier_name.clone()));
+                    }
+                    if let Some(intent) = coordinator_intent.as_ref() {
+                        // `target_tiers` may include a stale local-only manager
+                        // entry that is absent from the persisted proof
+                        // snapshot. Fence that local transition under the same
+                        // mutation ID as well; recovery may discard this
+                        // process-local superset, which advances the revision
+                        // and makes the deferred transition fail closed.
+                        TierConfigMgr::apply_prepared_mutation_intent_block_for_tiers(&handle, intent, &blocked_target_tiers)
                             .await
                             .map_err(TierConfigUpdateError::Publish)?;
                     }
+                    let prepared_mutation_block_allowance = match coordinator_intent.as_ref() {
+                        Some(intent) => Some(
+                            TierConfigMgr::mutation_block_allowance_for(&handle, intent.mutation_id)
+                                .await
+                                .map_err(TierConfigUpdateError::Publish)?,
+                        ),
+                        None => None,
+                    };
+                    // A durable coordinator intent supplies the admission
+                    // fence that lets us defer generation revocation. Keep the
+                    // original early transition for no-intent paths (for
+                    // example, reconciling a stale local manager to an
+                    // idempotently removed persisted tier), where there is no
+                    // Prepared record capable of blocking a new lease.
+                    let (mut transition, deferred_target_tiers) = if coordinator_intent.is_some() {
+                        (None, Some(target_tiers))
+                    } else {
+                        let transition = {
+                            let mut manager = handle.write().await;
+                            Self::begin_tier_transition_with_destinations(
+                                &handle,
+                                &mut manager,
+                                target_tiers,
+                                HashMap::new(),
+                                None,
+                            )
+                            .map_err(TierConfigUpdateError::Publish)?
+                        };
+                        (Some(transition), None)
+                    };
                     if coordinated_config_update {
                         drop(update.take());
                         drop(config_lock.take());
                     }
-                    let drain_deadline = Instant::now() + TIER_REMOTE_VALIDATION_TIMEOUT;
+                    // The durable Prepared block closes admission before the
-                    if let Err(drain_error) = transition.wait_for_active_leases_until(drain_deadline).await {
+                    // zero-reference proof, but deliberately leaves already
-                        if !abort_prepared_tier_mutation(&handle, api.clone(), coordinator_intent.as_ref(), Vec::new()).await {
+                    // issued generations current. In particular, an exact
+                    // free-version cleanup that has completed remote DELETE
+                    // must still be able to remove its local ownership marker;
+                    // revoking its generation here would strand that marker
+                    // and make this mutation reject its own interrupted work.
+                    if let Some(intent) = coordinator_intent.as_ref()
+                        && let Err(drain_error) =
+                            TierConfigMgr::wait_for_blocked_tier_operation_leases_for_tiers(&handle, &blocked_target_tiers).await
+                    {
+                        if !abort_prepared_tier_mutation(&handle, api.clone(), Some(intent), Vec::new()).await {
                             warn!(
                                 event = "tier_mutation_abort",
                                 component = LOG_COMPONENT_ECSTORE,
                                 subsystem = LOG_SUBSYSTEM_TIER,
                                 result = "prepared_intent_retained",
-                                coordinator_intent = coordinator_intent.is_some(),
+                                mutation_id = %intent.mutation_id,
                                 "tier mutation lease drain failed and abort was incomplete"
                             );
                         }
                         return Err(TierConfigUpdateError::Publish(drain_error));
+                    } else if let Some(transition) = transition.as_ref() {
+                        let drain_deadline = Instant::now() + TIER_REMOTE_VALIDATION_TIMEOUT;
+                        transition
+                            .wait_for_active_leases_until(drain_deadline)
+                            .await
+                            .map_err(TierConfigUpdateError::Publish)?;
                     }
                     let prepared_peers = if let Some(intent) = coordinator_intent.as_ref() {
                         let peers = match remote_tier_mutation_peers().await {
@@ -4350,6 +4431,71 @@ impl TierConfigMgr {
                         }
                         return Err(TierConfigUpdateError::Publish(proof_error));
                     }
+                    // No affected-tier lease can start after Prepared, and the
+                    // existing set was drained above. It is now safe to revoke
+                    // the generation for publication without invalidating a
+                    // cleanup between its remote and local commit boundaries.
+                    if transition.is_none() {
+                        let target_tiers = deferred_target_tiers.ok_or_else(|| {
+                            let mut err = ERR_TIER_INVALID_CONFIG.clone();
+                            err.message = "Remote tier mutation lost its deferred transition targets".to_string();
+                            TierConfigUpdateError::Publish(err)
+                        })?;
+                        let mut manager = handle.write().await;
+                        transition = Some(
+                            match Self::begin_tier_transition_with_destinations(
+                                &handle,
+                                &mut manager,
+                                target_tiers,
+                                HashMap::new(),
+                                prepared_mutation_block_allowance.as_ref(),
+                            ) {
+                                Ok(transition) => transition,
+                                Err(transition_error) => {
+                                    drop(manager);
+                                    if !abort_prepared_tier_mutation(
+                                        &handle,
+                                        api.clone(),
+                                        coordinator_intent.as_ref(),
+                                        prepared_peers,
+                                    )
+                                    .await
+                                    {
+                                        warn!(
+                                            event = "tier_mutation_abort",
+                                            component = LOG_COMPONENT_ECSTORE,
+                                            subsystem = LOG_SUBSYSTEM_TIER,
+                                            result = "prepared_intent_retained",
+                                            coordinator_intent = coordinator_intent.is_some(),
+                                            "tier mutation publish transition failed and abort was incomplete"
+                                        );
+                                    }
+                                    return Err(TierConfigUpdateError::Publish(transition_error));
+                                }
+                            },
+                        );
+                    }
+                    let mut transition = transition.ok_or_else(|| {
+                        let mut err = ERR_TIER_INVALID_CONFIG.clone();
+                        err.message = "Remote tier mutation lost its publish transition".to_string();
+                        TierConfigUpdateError::Publish(err)
+                    })?;
+                    let drain_deadline = Instant::now() + TIER_REMOTE_VALIDATION_TIMEOUT;
+                    if let Err(drain_error) = transition.wait_for_active_leases_until(drain_deadline).await {
+                        drop(transition);
+                        if !abort_prepared_tier_mutation(&handle, api.clone(), coordinator_intent.as_ref(), prepared_peers).await
+                        {
+                            warn!(
+                                event = "tier_mutation_abort",
+                                component = LOG_COMPONENT_ECSTORE,
+                                subsystem = LOG_SUBSYSTEM_TIER,
+                                result = "prepared_intent_retained",
+                                coordinator_intent = coordinator_intent.is_some(),
+                                "tier mutation publish drain failed and abort was incomplete"
+                            );
+                        }
+                        return Err(TierConfigUpdateError::Publish(drain_error));
+                    }
                     let candidate_digest = tier_config_candidate_digest(&candidate).map_err(TierConfigUpdateError::Save)?;
                     if coordinated_config_update {
                         config_lock = match Self::acquire_tier_config_write_lock(api.clone()).await {
@@ -4473,6 +4619,11 @@ impl TierConfigMgr {
                     let committed_coordinator_intent =
                         committed_tier_mutation_intent(coordinator_intent.as_ref(), &committed_config_etag)
                             .map_err(TierConfigUpdateError::Save)?;
+                    // Persist Committed before notifying refresh; a Prepared disk record
+                    // would restore the prepared block and invalidate our publish allowance.
+                    let coordinator_commit =
+                        commit_coordinator_tier_mutation_intent(api.clone(), coordinator_intent.as_ref(), &committed_config_etag)
+                            .await;
                     if let Some(intent) = committed_coordinator_intent.as_ref() {
                         TierConfigMgr::apply_committed_mutation_intent_block(&handle, intent)
                             .await
@@ -4483,9 +4634,9 @@ impl TierConfigMgr {
                                 .map_err(TierConfigUpdateError::Publish)?,
                         );
                     }
-                    commit_coordinator_tier_mutation_intent(api.clone(), coordinator_intent.as_ref(), &committed_config_etag)
+                    // Config is already saved: retain the committed fence and wake recovery
-                        .await
+                    // even when the coordinator commit failed or its outcome is unknown.
-                        .map_err(TierConfigUpdateError::Save)?;
+                    coordinator_commit.map_err(TierConfigUpdateError::Save)?;
                     if coordinated_config_update {
                         drop(update.take());
                         drop(config_lock.take());
@@ -5396,11 +5547,40 @@ impl TierConfigMgr {
         handle: &Arc<RwLock<Self>>,
         intent: &TierMutationIntent,
     ) -> std::result::Result<(), AdminError> {
+        let target_tiers = intent
+            .affected_targets
+            .iter()
+            .map(|target| target.tier_name.clone())
+            .collect();
+        Self::apply_prepared_mutation_intent_block_for_tiers(handle, intent, &target_tiers).await
+    }
+
+    async fn apply_prepared_mutation_intent_block_for_tiers(
+        handle: &Arc<RwLock<Self>>,
+        intent: &TierMutationIntent,
+        target_tiers: &HashSet<String>,
+    ) -> std::result::Result<(), AdminError> {
+        if intent.state != TierMutationIntentState::Prepared {
+            return Ok(());
+        }
         let manager = handle.read().await;
         let runtime = tier_driver_runtime(handle, &manager);
         let mut runtime = lock_unpoisoned(&runtime);
         let mut prepared_mutation_blocks = runtime.prepared_mutation_blocks.clone();
         Self::collect_prepared_mutation_intent_block(&mut prepared_mutation_blocks, intent)?;
+        for tier_name in target_tiers {
+            match prepared_mutation_blocks.entry(tier_name.clone()) {
+                Entry::Vacant(entry) => {
+                    entry.insert(intent.mutation_id);
+                }
+                Entry::Occupied(entry) if *entry.get() == intent.mutation_id => {}
+                Entry::Occupied(_) => {
+                    let mut err = ERR_TIER_BACKEND_IN_USE.clone();
+                    err.message = format!("Remote tier {tier_name} already has another prepared mutation");
+                    return Err(err);
+                }
+            }
+        }
         if prepared_mutation_blocks == runtime.prepared_mutation_blocks {
             return Ok(());
         }
@@ -5416,6 +5596,18 @@ impl TierConfigMgr {
     pub(crate) async fn wait_for_blocked_tier_operation_leases(
         handle: &Arc<RwLock<Self>>,
         intent: &TierMutationIntent,
+    ) -> std::result::Result<(), AdminError> {
+        let target_tiers = intent
+            .affected_targets
+            .iter()
+            .map(|target| target.tier_name.clone())
+            .collect();
+        Self::wait_for_blocked_tier_operation_leases_for_tiers(handle, &target_tiers).await
+    }
+
+    async fn wait_for_blocked_tier_operation_leases_for_tiers(
+        handle: &Arc<RwLock<Self>>,
+        target_tiers: &HashSet<String>,
     ) -> std::result::Result<(), AdminError> {
         let generations = {
             let manager = handle.read().await;
@@ -5423,10 +5615,9 @@ impl TierConfigMgr {
                 return Ok(());
             };
             let runtime = lock_unpoisoned(&runtime);
-            intent
+            target_tiers
-                .affected_targets
                 .iter()
-                .filter_map(|target| runtime.generations.get(&target.tier_name).cloned())
+                .filter_map(|tier_name| runtime.generations.get(tier_name).cloned())
                 .collect::<Vec<_>>()
         };
         let drain = async {
@@ -10590,6 +10781,11 @@ mod tests {
             .expect_err("coordinator committed-state CAS failure must be observable");
         assert!(matches!(err, TierConfigUpdateError::Save(_)));
         assert!(manager.read().await.tiers.contains_key("COLD-A"));
+        assert!(TierConfigMgr::has_committed_mutation_block(&manager).await);
+        let refresh = TierConfigMgr::mutation_refresh_notifier(&manager).await;
+        tokio::time::timeout(Duration::from_secs(1), refresh.notified())
+            .await
+            .expect("failed coordinator commit must notify recovery after saving config");
         let blocked = match TierConfigMgr::acquire_operation_lease(&manager, "COLD-A").await {
             Ok(_) => panic!("failed coordinator commit CAS must retain the local committed fence"),
             Err(err) => err,
@@ -14316,6 +14512,12 @@ mod tests {
         after_commit: bool,
     }
 
+    #[derive(Debug, Default)]
+    struct CasCoordinatorCommitBarrier {
+        arrived: tokio::sync::Notify,
+        release: tokio::sync::Notify,
+    }
+
     #[derive(Debug)]
     struct CasConfigStore {
         objects: tokio::sync::Mutex<HashMap<String, (Vec<u8>, String)>>,
@@ -14328,6 +14530,7 @@ mod tests {
         fail_delete_prefix: tokio::sync::Mutex<Option<(String, usize)>>,
         delete_log: tokio::sync::Mutex<Vec<String>>,
         list_barrier: tokio::sync::Mutex<Option<Arc<CasListBarrier>>>,
+        coordinator_commit_barrier: tokio::sync::Mutex<Option<Arc<CasCoordinatorCommitBarrier>>>,
         intent_list_calls: AtomicUsize,
         fail_reference_walk: AtomicBool,
         reference_walk_send_count: AtomicUsize,
@@ -14350,6 +14553,7 @@ mod tests {
                 fail_delete_prefix: tokio::sync::Mutex::new(None),
                 delete_log: tokio::sync::Mutex::new(Vec::new()),
                 list_barrier: tokio::sync::Mutex::new(None),
+                coordinator_commit_barrier: tokio::sync::Mutex::new(None),
                 intent_list_calls: AtomicUsize::new(0),
                 fail_reference_walk: AtomicBool::new(false),
                 reference_walk_send_count: AtomicUsize::new(0),
@@ -14383,6 +14587,13 @@ mod tests {
                 .push(object);
         }
 
+        fn remove_listed_version(&self, bucket: &str, object: &str) {
+            self.listed_versions
+                .lock()
+                .expect("tier reference fixture should not poison")
+                .retain(|version| version.bucket != bucket || version.name != object);
+        }
+
         fn add_lifecycle_config(&self, bucket: &str, config: BucketLifecycleConfiguration) {
             self.lifecycle_configs
                 .lock()
@@ -14541,6 +14752,19 @@ mod tests {
             }
             let mut payload = Vec::new();
             tokio::io::AsyncReadExt::read_to_end(&mut data.stream, &mut payload).await?;
+            if object.starts_with(crate::services::tier::tier_mutation_intent::TIER_COORDINATOR_MUTATION_INTENT_RECORD_PREFIX)
+                && opts
+                    .http_preconditions
+                    .as_ref()
+                    .and_then(HTTPPreconditions::if_match_value)
+                    .is_some()
+            {
+                let barrier = self.coordinator_commit_barrier.lock().await.take();
+                if let Some(barrier) = barrier {
+                    barrier.arrived.notify_one();
+                    barrier.release.notified().await;
+                }
+            }
             let race_rewrite = if opts
                 .http_preconditions
                 .as_ref()
@@ -15639,13 +15863,152 @@ mod tests {
     }
 
     #[tokio::test]
-    async fn force_remove_and_save_bypasses_lifecycle_only_reference() {
+    async fn tier_remove_prepared_fence_allows_inflight_free_version_cleanup_to_finish() {
-        // rustfs/rustfs#6832: reproduces the admin RemoveTier path (not just the lower-level
+        let store = Arc::new(CasConfigStore::default());
-        // reference-proof function) for a tier with zero transitioned objects but a lifecycle
+        let tier = build_rustfs_tier("COLD-A");
-        // rule still pointing at it — the exact shape of
+        let identity = tier_backend_identity(&tier).expect("test tier identity should encode");
-        // `test_manual_transition_async_tier_failure_reports_terminal_partial` in e2e_test,
+        let mut persisted = empty_mgr();
-        // which force-removes a tier a lifecycle rule still references to simulate a
+        persisted.tiers.insert("COLD-A".to_string(), tier.clone_with_credentials());
-        // decommissioned backend.
+        persisted
+            .save_tiering_config_if_current(store.clone(), None)
+            .await
+            .expect("free-version drain fixture should persist");
+
+        let manager = TierConfigMgr::new();
+        {
+            let mut guard = manager.write().await;
+            guard.tiers.insert("COLD-A".to_string(), tier);
+            guard.tiers.insert("COLD-B".to_string(), build_rustfs_tier("COLD-B"));
+            guard
+                .replace_driver("COLD-A", Box::new(LeaseTestBackend::ready("cleanup")))
+                .expect("cleanup driver generation should install");
+            guard
+                .replace_driver("COLD-B", Box::new(LeaseTestBackend::ready("stale-local")))
+                .expect("stale local driver generation should install");
+        }
+        let cleanup_lease = TierConfigMgr::acquire_operation_lease(&manager, "COLD-A")
+            .await
+            .expect("in-flight cleanup lease should be available");
+        let stale_local_lease = TierConfigMgr::acquire_operation_lease(&manager, "COLD-B")
+            .await
+            .expect("stale local tier lease should be available");
+        let mut free_version = transitioned_tier_object("photos", "2026/free-version.jpg", "COLD-A", Some(identity));
+        free_version.transitioned_object.status = "pending".to_string();
+        free_version.transitioned_object.free_version = true;
+        store.add_listed_version(free_version);
+
+        let remove_manager = manager.clone();
+        let remove_store = store.clone();
+        let remove = tokio::spawn(async move {
+            TIER_MUTATION_TEST_PEERS
+                .scope(
+                    Vec::new(),
+                    TierConfigMgr::remove_and_save_with(&remove_manager, remove_store, "COLD-A", true),
+                )
+                .await
+        });
+
+        tokio::time::timeout(Duration::from_secs(1), async {
+            loop {
+                let guard = manager.read().await;
+                let runtime = registered_tier_driver_runtime(&guard).expect("runtime should remain registered");
+                let prepared = {
+                    let runtime = lock_unpoisoned(&runtime);
+                    runtime.prepared_mutation_blocks.contains_key("COLD-A")
+                        && runtime.prepared_mutation_blocks.contains_key("COLD-B")
+                };
+                if prepared {
+                    break;
+                }
+                drop(guard);
+                tokio::task::yield_now().await;
+            }
+        })
+        .await
+        .expect("tier remove should install its durable prepared fence");
+
+        assert!(
+            cleanup_lease.is_current(&manager).await,
+            "the prepared fence must let the already leased cleanup finish its exact local marker deletion"
+        );
+        assert!(
+            stale_local_lease.is_current(&manager).await,
+            "the local superset fence must also let an already leased stale-manager operation finish"
+        );
+        let blocked = match TierConfigMgr::acquire_operation_lease(&manager, "COLD-A").await {
+            Ok(_) => panic!("the prepared fence must reject new tier operations"),
+            Err(err) => err,
+        };
+        assert!(TierConfigMgr::operation_lease_blocked_by_mutation(&blocked));
+        let stale_blocked = match TierConfigMgr::acquire_operation_lease(&manager, "COLD-B").await {
+            Ok(_) => panic!("the local superset fence must reject new stale-manager operations"),
+            Err(err) => err,
+        };
+        assert!(TierConfigMgr::operation_lease_blocked_by_mutation(&stale_blocked));
+
+        store.remove_listed_version("photos", "2026/free-version.jpg");
+        drop(cleanup_lease);
+        drop(stale_local_lease);
+
+        tokio::time::timeout(Duration::from_secs(5), remove)
+            .await
+            .expect("tier remove should finish after both in-flight operations release their leases")
+            .expect("tier remove task should join")
+            .expect("tier remove should pass once the in-flight cleanup removes its marker");
+        assert!(!manager.read().await.tiers.contains_key("COLD-A"));
+        assert!(!manager.read().await.tiers.contains_key("COLD-B"));
+        assert!(
+            !load_tier_config_for_update(store)
+                .await
+                .expect("removed tier config should reload")
+                .0
+                .tiers
+                .contains_key("COLD-A")
+        );
+    }
+
+    #[tokio::test]
+    async fn no_intent_stale_manager_removal_keeps_early_generation_drain() {
+        let store = Arc::new(CasConfigStore::default());
+        empty_mgr()
+            .save_tiering_config_if_current(store.clone(), None)
+            .await
+            .expect("empty persisted tier config should exist");
+        let manager = TierConfigMgr::new();
+        {
+            let mut guard = manager.write().await;
+            install_lease_backend(&mut guard, "COLD-A", LeaseTestBackend::ready("stale"));
+        }
+        let old = TierConfigMgr::acquire_operation_lease(&manager, "COLD-A")
+            .await
+            .expect("stale manager lease should be available");
+
+        let remove_manager = manager.clone();
+        let remove_store = store.clone();
+        let remove =
+            tokio::spawn(async move { TierConfigMgr::remove_and_save_with(&remove_manager, remove_store, "COLD-A", true).await });
+        tokio::time::timeout(Duration::from_secs(1), async {
+            while old.is_current(&manager).await {
+                tokio::task::yield_now().await;
+            }
+        })
+        .await
+        .expect("no-intent stale-manager reconciliation should revoke before its proof");
+        let blocked = match TierConfigMgr::acquire_operation_lease(&manager, "COLD-A").await {
+            Ok(_) => panic!("stale-manager reconciliation must not admit a new operation"),
+            Err(err) => err,
+        };
+        assert!(TierConfigMgr::operation_lease_blocked_by_mutation(&blocked));
+
+        drop(old);
+        remove
+            .await
+            .expect("stale-manager removal task should join")
+            .expect("stale-manager removal should converge to the persisted empty config");
+        assert!(!manager.read().await.tiers.contains_key("COLD-A"));
+    }
+
+    async fn assert_lifecycle_only_reference_obeys_force(clear: bool, force: bool) {
         let store = Arc::new(CasConfigStore::default());
         let tier = build_rustfs_tier("COLD-A");
         let mut persisted = empty_mgr();
@@ -15686,22 +16049,55 @@ mod tests {
 
         let manager = TierConfigMgr::new();
         manager.write().await.tiers.insert("COLD-A".to_string(), tier);
-        TierConfigMgr::remove_and_save_with(&manager, store.clone(), "COLD-A", true)
+        let mutation = if clear {
-            .await
+            TierCandidateMutation::Clear(force)
-            .expect("force remove must bypass a lifecycle-config-only reference");
+        } else {
+            TierCandidateMutation::Remove("COLD-A".to_string(), force)
+        };
+        let result = TIER_DRIVER_TEST_FACTORY
+            .scope(
+                healthy_driver_factory(),
+                TierConfigMgr::update_candidate_with_config_lock(&manager, store.clone(), mutation),
+            )
+            .await;
+        if force {
+            result.expect("force mutation must bypass a lifecycle-config-only reference");
+        } else {
+            let err = result.expect_err("non-force mutation must reject a lifecycle-only reference");
+            let TierConfigUpdateError::Publish(err) = err else {
+                panic!("non-force mutation must fail during reference proof: {err:?}");
+            };
+            assert_eq!(err.code, ERR_TIER_BACKEND_IN_USE.code);
+            assert!(err.message.contains("move-current"), "{err}");
+        }
 
-        assert!(!manager.read().await.tiers.contains_key("COLD-A"));
+        assert_eq!(manager.read().await.tiers.contains_key("COLD-A"), !force);
-        assert!(
+        assert_eq!(
-            !load_tier_config_for_update(store)
+            load_tier_config_for_update(store)
                 .await
                 .expect("config should still reload")
                 .0
                 .tiers
                 .contains_key("COLD-A"),
-            "force removal must persist the empty candidate"
+            !force,
+            "persisted state must match the force mutation result"
         );
     }
 
+    #[tokio::test]
+    async fn remove_with_config_lock_obeys_force_for_lifecycle_only_reference() {
+        for force in [false, true] {
+            assert_lifecycle_only_reference_obeys_force(false, force).await;
+        }
+    }
+
+    #[tokio::test]
+    async fn clear_with_config_lock_obeys_force_for_lifecycle_only_reference() {
+        for force in [false, true] {
+            assert_lifecycle_only_reference_obeys_force(true, force).await;
+        }
+    }
+
     #[tokio::test]
     async fn zero_reference_proof_blocks_clear_before_config_save() {
         let store = Arc::new(CasConfigStore::default());
@@ -16515,12 +16911,23 @@ mod tests {
             .await
         });
         tokio::time::timeout(Duration::from_secs(1), async {
-            while old.is_current(&manager).await {
+            loop {
+                let guard = manager.read().await;
+                let runtime = registered_tier_driver_runtime(&guard).expect("runtime should remain registered");
+                let prepared = lock_unpoisoned(&runtime).prepared_mutation_blocks.contains_key("COLD-A");
+                if prepared {
+                    break;
+                }
+                drop(guard);
                 tokio::task::yield_now().await;
             }
         })
         .await
-        .expect("owned update should revoke before caller cancellation");
+        .expect("owned update should install its prepared fence before caller cancellation");
+        assert!(
+            old.is_current(&manager).await,
+            "an already leased operation must remain current until it can finish"
+        );
         caller.abort();
         drop(old);
 
@@ -16565,12 +16972,23 @@ mod tests {
             .await
         });
         tokio::time::timeout(Duration::from_secs(1), async {
-            while old.is_current(&manager).await {
+            loop {
+                let guard = manager.read().await;
+                let runtime = registered_tier_driver_runtime(&guard).expect("runtime should remain registered");
+                let prepared = lock_unpoisoned(&runtime).prepared_mutation_blocks.contains_key("COLD-A");
+                if prepared {
+                    break;
+                }
+                drop(guard);
                 tokio::task::yield_now().await;
             }
         })
         .await
-        .expect("owned update should revoke before caller cancellation");
+        .expect("owned update should install its prepared fence before caller cancellation");
+        assert!(
+            old.is_current(&manager).await,
+            "the prepared fence must not invalidate an already leased operation"
+        );
         caller.abort();
 
         let config_file = tier_config_lock_path();
@@ -16956,6 +17374,71 @@ mod tests {
         assert!(current.tiers.contains_key("COLD-B"));
     }
 
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn reference_proof_rejects_a_changed_prepared_fence_revision_before_publish() {
+        let manager = TierConfigMgr::new();
+        let store = Arc::new(CasConfigStore::default());
+        let mut persisted = empty_mgr();
+        persisted.tiers.insert("COLD-A".to_string(), build_rustfs_tier("COLD-A"));
+        persisted
+            .save_tiering_config_if_current(store.clone(), None)
+            .await
+            .expect("prepared-fence revision fixture should persist");
+        {
+            let mut guard = manager.write().await;
+            install_lease_backend(&mut guard, "COLD-A", LeaseTestBackend::ready("old"));
+        }
+
+        let barrier = tier_reference_proof_test_barrier();
+        let scoped_barrier = barrier.clone();
+        let update_manager = manager.clone();
+        let update_store = store.clone();
+        let update = tokio::spawn(async move {
+            TIER_REFERENCE_PROOF_TEST_BARRIER
+                .scope(
+                    scoped_barrier,
+                    TIER_MUTATION_TEST_PEERS.scope(
+                        Vec::new(),
+                        TierConfigMgr::update_candidate_with_config_lock(
+                            &update_manager,
+                            update_store,
+                            TierCandidateMutation::Remove("COLD-A".to_string(), true),
+                        ),
+                    ),
+                )
+                .await
+        });
+        barrier.arrived.notified().await;
+
+        let unrelated = prepared_remove_intent("COLD-B", uuid::Uuid::from_u128(0x2237));
+        TierConfigMgr::apply_prepared_mutation_intent_block(&manager, &unrelated)
+            .await
+            .expect("an unrelated prepared fence should advance the runtime revision");
+        barrier.release.add_permits(1);
+
+        let err = update
+            .await
+            .expect("tier update task should join")
+            .expect_err("a reference proof cannot authorize publication across a fence revision change");
+        let TierConfigUpdateError::Publish(err) = err else {
+            panic!("the stale prepared-fence allowance should fail publication: {err:?}");
+        };
+        assert!(err.message.contains("changed before replacement"), "{err}");
+        assert!(manager.read().await.tiers.contains_key("COLD-A"));
+        assert!(
+            load_tier_config_for_update(store)
+                .await
+                .expect("rejected tier config should remain readable")
+                .0
+                .tiers
+                .contains_key("COLD-A")
+        );
+        TierConfigMgr::clear_prepared_mutation_intent_block(&manager, unrelated.mutation_id)
+            .await
+            .expect("unrelated test fence should clear");
+    }
+
     #[tokio::test]
     #[serial_test::serial]
     async fn caller_cancellation_after_durable_prepare_does_not_hide_the_mutation() {
@@ -17242,6 +17725,98 @@ mod tests {
         assert_ne!(manager_a.read().await.empty(), manager_b.read().await.empty());
     }
 
+    async fn assert_coordinator_commit_refresh_succeeds(mutation: TierCandidateMutation) {
+        let adding = matches!(mutation, TierCandidateMutation::Add(..));
+        let manager = TierConfigMgr::new();
+        let store = Arc::new(CasConfigStore::default());
+        if !adding {
+            let mut persisted = empty_mgr();
+            persisted.tiers.insert("COLD-A".to_string(), build_rustfs_tier("COLD-A"));
+            persisted
+                .save_tiering_config_if_current(store.clone(), None)
+                .await
+                .expect("existing tier fixture should persist");
+            let mut guard = manager.write().await;
+            install_lease_backend(&mut guard, "COLD-A", LeaseTestBackend::ready("old"));
+        }
+        let barrier = Arc::new(CasCoordinatorCommitBarrier::default());
+        *store.coordinator_commit_barrier.lock().await = Some(barrier.clone());
+        let update_manager = manager.clone();
+        let update_store = store.clone();
+        let update = tokio::spawn(async move {
+            TIER_DRIVER_TEST_FACTORY
+                .scope(
+                    healthy_driver_factory(),
+                    TIER_MUTATION_TEST_PEERS.scope(
+                        Vec::new(),
+                        TierConfigMgr::update_candidate_with_config_lock(&update_manager, update_store, mutation),
+                    ),
+                )
+                .await
+        });
+        tokio::time::timeout(Duration::from_secs(5), barrier.arrived.notified())
+            .await
+            .expect("mutation should reach coordinator commit after saving config");
+        assert_eq!(
+            load_tier_config_for_update(store.clone())
+                .await
+                .expect("saved config should be readable before coordinator commit")
+                .0
+                .tiers
+                .contains_key("COLD-A"),
+            adding
+        );
+        assert_eq!(
+            TierConfigMgr::load_coordinator_mutation_intents(store.clone())
+                .await
+                .expect("coordinator intent should remain readable")[0]
+                .state,
+            TierMutationIntentState::Prepared
+        );
+
+        let lock_requests = lock_unpoisoned(&store.lock_requests).len();
+        // Also exercise an independently scheduled refresh while the durable
+        // coordinator record is still Prepared, before its commit notification.
+        TierConfigMgr::request_committed_mutation_refresh(&manager).await;
+        TIER_MUTATION_TEST_PEERS
+            .scope(Vec::new(), async {
+                let worker = TierConfigMgr::refresh_tier_config_handle_with(manager.clone(), store.clone());
+                tokio::pin!(worker);
+                tokio::time::timeout(Duration::from_secs(5), async {
+                    while lock_unpoisoned(&store.lock_requests).len() == lock_requests {
+                        tokio::select! {
+                            _ = &mut worker => panic!("refresh worker must remain available"),
+                            _ = tokio::task::yield_now() => {}
+                        }
+                    }
+                })
+                .await
+                .expect("refresh should reconcile the Prepared record before waiting for the config lock");
+                barrier.release.notify_one();
+                let result = tokio::time::timeout(Duration::from_secs(5), async {
+                    tokio::select! {
+                        _ = &mut worker => panic!("refresh worker must remain available"),
+                        result = update => result.expect("tier mutation task should join"),
+                    }
+                })
+                .await
+                .expect("tier mutation should finish with refresh running");
+                result.expect("saved tier mutation must publish successfully on the first attempt");
+            })
+            .await;
+        assert_eq!(manager.read().await.tiers.contains_key("COLD-A"), adding);
+    }
+
+    #[tokio::test]
+    async fn tier_add_succeeds_with_refresh_during_coordinator_commit() {
+        assert_coordinator_commit_refresh_succeeds(TierCandidateMutation::Add(build_rustfs_tier("COLD-A"), true)).await;
+    }
+
+    #[tokio::test]
+    async fn tier_remove_succeeds_with_refresh_during_coordinator_commit() {
+        assert_coordinator_commit_refresh_succeeds(TierCandidateMutation::Remove("COLD-A".to_string(), true)).await;
+    }
+
     async fn committed_refresh_fixture(fail_cleanup: bool) -> (Arc<RwLock<TierConfigMgr>>, Arc<CasConfigStore>, uuid::Uuid) {
         let manager = TierConfigMgr::new();
         {