chore(gateway): merge main into integration

Merge main d88f524d9 (#8973-#9016) into gateway/integration 1f2dbe36e.

Conflicts:
- admin/handlers/table_catalog/config.rs: keep main's table bucket
  metadata reload notification (#8994) and resolve its app context from
  the owned request envelope.
- protocols/client.rs: take main's session key revalidation before the
  session bucket listing (#8998) and keep integration's conversion of the
  s3s listing into the gateway ListBucketsOutput; the WebDAV capacity view
  reuses list_buckets_for_session as on main.

Owned-type follow-ups for main's new code:
- sftp dir tests, the dummy backend's empty-queue fallback and
  protocols policy tests use gateway DTOs and rustfs_s3_types errors, so
  the s3s footprint stays at its 65-file baseline.
- storage_api re-exports rustfs_s3_types::S3ErrorCode for protocol tests.

Baselines: test-count ratchet raised for the crates main added tests to
(merged counts equal integration plus main's delta per crate).

.agents/skills/adversarial-validation/references/performance.md

@@ -6,7 +6,21 @@
 - Route every new sync/flush through the durability-mode and bucket override
   gates; mode `none` must not pay the new fsync.
 - Keep blocking filesystem/CPU work off async runtime threads, but do not split
-  one small operation into many `spawn_blocking` round trips.
+  one small operation into many `spawn_blocking` round trips; coalesce `open`,
+  `read`/`write`, `fsync`, and `close` into one blocking closure per disk. When
+  coalescing steps across execution lanes, route the combined closure to the
+  dedicated fsync lane whenever any step performs file or directory sync so
+  stalled syncs cannot starve foreground metadata reads.
+- Check for per-disk syscall amplification (`disks × parts/files`): pin a
+  directory once (`openat2` on Linux) and reuse the handle across sibling reads
+  or same-directory renames instead of re-walking ancestor components per file
+  or re-opening the same parent directory twice.
+- When the file size is already known from `fstat` metadata, pre-size the
+  destination buffer (`len + 1`) before reading to EOF so small metadata reads
+  complete in a single `read()` + EOF check rather than exponential-growth
+  reallocations and syscalls.
+- Avoid redundant disk re-reads or existence checks under the same exclusive
+  lock when the state or raw bytes were just loaded or persisted.
 - Measure lock hold time across I/O and compare acquisition order for ABBA.
 - Keep cleanup, extra stat/rename, and diagnostics out of the PUT commit critical
   section when they need not be there.

.agents/skills/rustfs-release-publish/SKILL.md

@@ -25,7 +25,7 @@ Read only the current phase's procedure; retain completed evidence across turns.
 | 4–5 | [Run the downloaded binary, Console CRUD, and latest-rc matrix](references/preview-acceptance.md). Every check must pass. |
 | Confirmation | Complete the gate below and end the turn. |
 | 6 | After fresh confirmation, [publish and verify the final tag](references/final-release.md) at `PREVIEW_HASH`. |
-| 7 | [Reconcile Security Advisories, milestones, installation references, and the website](references/post-release-updates.md) only after complete publication. |
+| 7 | [Reconcile Security Advisories, milestones, installation references, and the website; reconcile main and remove the completed release branch](references/post-release-updates.md) only after complete publication. |
 
 ## Human Confirmation
 
@@ -37,6 +37,6 @@ Only a new affirmative user reply to that exact report authorizes Phase 6. The o
 
 - Before the final tag exists, land fixes on main, backport to the candidate, verify target versions, and restart branch CI/acceptance. Use a new preview iteration once a preview was tagged; main may advance independently.
 - Once the final tag exists, retry publication jobs for that exact tag. A source fix requires a newly confirmed target, not retagging or a newer preview. Phase 7 retries recheck artifact availability and superseding releases.
-- If abandoned after the bump merged, report the unpublished version on main and record whether to revert or let the next release overwrite it. Deleting candidate branches or preview tags requires an explicit decision.
+- If abandoned after the bump merged, report the unpublished version on main and record whether to revert or let the next release overwrite it. Deleting unpublished candidate branches or preview tags requires an explicit decision. Completed release branches follow Phase 7 cleanup; keep release tags.
 
 Report completed phase results and incomplete work with evidence: Console previous/latest tags, release need, hash and run/Release URLs; target, branch CI/SHA, previews and hash, confirmation state; preview/final assets, latest-channel state, Console/rc results and cleanup; advisory reconciliation results and pending disclosure decisions; both milestone states, installation PR checks/merge state, and website text/link/PR/deployment state. Keep artifact publication separate from follow-up delivery; do not call an open PR a live update. Report deviations and their authorization.

.agents/skills/rustfs-release-publish/references/post-release-updates.md

@@ -48,3 +48,10 @@ Use the existing announcement configuration from [website PR #112](https://githu
 3. Use the site's current package manager and required checks. For this config-only change, type-check and lint the changed file, build the site, and inspect the generated homepage for the version, visible link text, and destination. Confirm `/download` actually offers the announced version before using that link.
 4. Commit, push, and create or update the website PR when included in the authorized release scope. Include the published RustFS Release URL and checks run. Keep merge and production-deployment authority separate; reuse permission already given, but this reference grants none by itself.
 5. After an authorized merge/deployment, verify the live `https://rustfs.com/` banner and its destination. If only the PR is ready, report that state and URL; do not report the banner as deployed. A website failure does not invalidate or move the published release tag: retry this phase after resolving the failure.
+
+## Reconcile main and remove the completed release branch
+
+1. Compare the exact published tag with current main. Audit every release-side commit since their merge base, including inherited backports. An ahead count is a history difference, not proof of missing changes. Use ancestry and patch equivalence, then inspect current code and regression evidence for squashed, adapted, superseded, or reverted changes. Record each commit as present, superseded, release-only, or missing; a positive `git cherry` result alone does not prove a missing fix.
+2. Bring missing shared changes to main through a reviewed PR and the required checks before closing the follow-up. Preserve main's later changes; decide version bumps and pinned Console configuration separately. If a history merge is needed, use a merge commit and verify the release commit is an ancestor of main; squash does not establish that ancestry. Do not merge main into the published release branch or move its tags. Report unresolved changes instead of claiming reconciliation is complete.
+3. After publication and authorized follow-ups are complete, delete the remote `release/<target>` branch. Immediately before deletion, verify its HEAD still equals the published tag's commit and no open PR targets or originates from it. Preserve a branch with newer work or open PRs and report the blocker. Use a deletion with an expected-old-SHA check to avoid removing a concurrently advanced branch.
+4. Verify the remote branch is absent and the final and preview tags still resolve to their recorded commits. Remove a local release branch only if it is task-owned, unchecked out, and has no unpreserved work. Record cleanup separately from artifact publication; an explicitly authorized deletion before reconciliation does not make unresolved main changes complete.

.config/e2e-distributed-selection.txt

@@ -1,2 +1,2 @@
-sha256-linux=cf31ed27e4e6bf42da5bf81f248d9d2619ac8e8615186cdacb30bd63996d7607
-sha256-darwin=cf31ed27e4e6bf42da5bf81f248d9d2619ac8e8615186cdacb30bd63996d7607
+sha256-linux=eb33067d90ee66c0a97e8f3658e46c152e47f897d2fbfdc6cb5ad93cfa55414b
+sha256-darwin=eb33067d90ee66c0a97e8f3658e46c152e47f897d2fbfdc6cb5ad93cfa55414b

.config/e2e-full-selection.txt

@@ -1,2 +1,2 @@
-sha256-darwin=b53fe6d46cdffedfad6a83da0dfffa9aec10cd7c8ace53469f129587bc9f4d50
-sha256-linux=6be90f5c3886e9852ac27d51f2d47e6d2fa8bb424e0d03ee931268dcd0020db8
+sha256-darwin=596ca4ce6661a1ccd5c8278a7781fd46922ebe5e23377a254471ec5f4fabfeb0
+sha256-linux=1f3d1d167fbd58a72b46a255edb111c7ee0abdb1ae37ac00400d1239cacdd5e1

.config/e2e-nightly-selection.txt

@@ -1,2 +1,2 @@
-sha256-darwin=347a0d4e5e547d5388703062ce66a66167a845b6e57ec591ccb4c732a567c68d
-sha256-linux=4f9efa47194279185bb68249e82c788fa147abe944b6c764050e3d47f31e7926
+sha256-darwin=3814cdde8cacf4187e18188d342ed70931cc32e268e42e5bb438d565bb4d0f81
+sha256-linux=663025a80273579f95ae3f49af02c0afcbfce94df09701c360a13e607b7637db

.config/e2e-repl-nightly-selection.txt

@@ -1 +1 @@
-sha256=e4ee3898b840a8374cb02c64342cc15f8d7ecd1becaf4b4a2a4657290f9e7ff1
+sha256=4eac269f9a59b2f9be4e7a87ea6aa38454f5ebe10ede35c9b8d4218137d2cd35

.config/e2e-smoke-selection.txt

@@ -1 +1 @@
-sha256=002ec5d973f82288e50a096a2c8b3b38749f62a7a3e169775566874d8ee98b4a
+sha256=57ffdd620dec9cba08a411b8ad681f4eb906cf5764ffed3c7c992084f5fc6092

.config/ecstore-required-tests.json

@@ -29,7 +29,7 @@
     {
       "invariant": "list-uncommitted-version",
       "suite": "rustfs-filemeta",
-      "name": "metacache::tests::resolve_with_write_quorum_slack_keeps_partial_latest_hidden_during_merge"
+      "name": "metacache::tests::resolve_merge_keeps_the_newest_version_at_listing_quorum"
     },
     {
       "invariant": "minio-object-fixture",
@@ -229,7 +229,7 @@
     {
       "invariant": "list-objects-shared-corpus-orphan-cleanup-requires-authoritative-evidence",
       "suite": "rustfs-ecstore",
-      "name": "store::list_objects::differential_tests::list_objects_shared_corpus_orphan_cleanup_requires_authoritative_evidence"
+      "name": "store::list_objects::differential_tests::list_objects_shared_corpus_orphan_listing_preserves_inputs"
     },
     {
       "invariant": "list-objects-shared-corpus-consumer-drop-joins-real-disk-producer",
@@ -241,11 +241,6 @@
       "suite": "rustfs-ecstore",
       "name": "store::list_objects::differential_tests::list_objects_shared_corpus_cancelled_quorum_walk_releases_senders_and_disk_lock"
     },
-    {
-      "invariant": "list-objects-shared-corpus-enabled-index-contracts",
-      "suite": "rustfs-ecstore",
-      "name": "store::list_objects::differential_tests::list_objects_shared_corpus_enabled_index_contracts"
-    },
     {
       "invariant": "borrowed-context-boundary",
       "suite": "rustfs-ecstore",

.config/make/lint-fmt.mak

@@ -70,6 +70,11 @@ s3s-footprint-check: ## Check the s3s dependency footprint ratchet stays frozen
 	@echo "📦 Checking s3s footprint ratchet..."
 	./scripts/check_s3s_footprint.sh
 
+.PHONY: test-count-ratchet-check
+test-count-ratchet-check: ## Check no workspace crate lost tests or gained #[ignore] against the committed baseline
+	@echo "🧪 Checking per-crate test-count ratchet..."
+	./scripts/check_test_count_ratchet.sh
+
 .PHONY: fips-wording-check
 fips-wording-check: ## Check docs and crates/kms do not over-claim crypto capabilities
 	@echo "📣 Checking cryptographic capability wording guard..."

.config/make/pre-commit.mak

@@ -20,11 +20,11 @@ planning-docs-check: ## Check that no planning-type documents are committed
 	./scripts/check_no_planning_docs.sh
 
 .PHONY: pre-commit
-pre-commit: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check quick-check ## Run fast pre-commit checks without clippy/full tests
+pre-commit: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check test-count-ratchet-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check quick-check ## Run fast pre-commit checks without clippy/full tests
 	@echo "✅ All pre-commit checks passed!"
 
 .PHONY: pre-pr
-pre-pr: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check log-analyzer-rules-check offline-enrollment-e2e-check clippy-check test ## Run full pre-PR checks with clippy and tests
+pre-pr: fmt-check unsafe-code-check architecture-migration-check logging-guardrails-check error-other-ratchet-check tokio-io-uring-check extension-schema-check body-cache-whitelist-check s3s-footprint-check test-count-ratchet-check fips-wording-check embedded-secrets-check test-wiring-check doc-paths-check planning-docs-check log-analyzer-rules-check offline-enrollment-e2e-check clippy-check test ## Run full pre-PR checks with clippy and tests
 	@echo "✅ All pre-PR checks passed!"
 
 .PHONY: dev-check

.config/make/tests.mak

@@ -50,8 +50,11 @@ script-tests: ## Run shell script tests
 	./scripts/test_fuzz_runner.sh
 	./scripts/test_python_bin.sh
 	./scripts/check_embedded_secrets.sh --self-test
+	./scripts/check_s3s_footprint.sh --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_test_wiring.py --self-test
+	$(RUSTFS_PYTHON_BIN) ./scripts/codemods/s3_error_to_s3_types.py --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_e2e_binary.py
+	$(RUSTFS_PYTHON_BIN) ./scripts/test_inspect_btf_prototypes.py
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_release_build_workflow.py
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_release_glibc.py
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_upgrade_workflow.py

.config/nextest.toml

@@ -68,6 +68,17 @@ command = ['sh', '-c', 'echo RUST_MIN_STACK=33554432 >> "$NEXTEST_ENV"']
 # boundaries. Reserve the run's capacity so unrelated fixtures cannot starve
 # their existing deadlines and typed-failure assertions (rustfs/backlog#2706).
 # Keep each test's internal concurrency and zero-retry policy unchanged.
+[[profile.default.overrides]]
+filter = """
+package(rustfs-ecstore) & (
+    test(/^set_disk::ops::multipart::tests::complete_multipart_(quorum_success_releases_logical_guards_before_physical_tail|waits_for_tail_before_releasing_guards_and_marking_capacity)$/)
+    | test(=store::heal::tests::full_format_heal_lost_after_last_save_does_not_publish_or_renew_and_latches)
+    | test(/^store::init::tests::(decommission_durable_ilm_receipt_pagination_fails_closed_on_second_page|dispatch_manifest_shutdown_(drains_admitted_delete_batch_without_tail_admission|stops_aborting_validation_tail_reads|stops_authorized_tail_reads_before_receipt_or_cas))$/)
+    | (binary(metadata_read_lock_reuse_test) & test(=replica_multipart_completes_while_metadata_writer_waits_for_its_snapshot))
+)
+"""
+threads-required = "num-test-threads"
+
 [[profile.default.overrides]]
 filter = 'package(rustfs) & test(/^connect::diagnostics::schedule::tests::/)'
 threads-required = "num-test-threads"
@@ -334,6 +345,12 @@ threads-required = "num-test-threads"
 filter = 'package(rustfs-ecstore) & test(=store::init::tests::dispatch_manifest_rollback_bounded_concurrency_reaches_tail_behind_slow_member)'
 threads-required = "num-test-threads"
 
+# These short-deadline real-disk progress probes need the test run's capacity
+# for scanner and object-lock handoffs; keep their assertions and deadlines.
+[[profile.default.overrides]]
+filter = '(package(rustfs-scanner) & (test(=scanner::tests::cycle_persist_failure::later_cycle_post_commit_failure_reloads_durable_state) | test(=scanner::tests::running_main_loop_catches_up_pause_cleared_after_startup_observe))) | (package(rustfs-ecstore) & test(=store::init::tests::restored_transitioned_delete_causally_enqueues_free_version))'
+threads-required = "num-test-threads"
+
 # ---------------------------------------------------------------------------
 # ci profile — the strict CI gate (ci.yml `cargo nextest run --profile ci`)
 # ---------------------------------------------------------------------------
@@ -351,6 +368,17 @@ path = "junit.xml"
 
 # Match the local durable-progress reservations. All members, deadlines,
 # assertions, and internal concurrency remain enabled without retries.
+[[profile.ci.overrides]]
+filter = """
+package(rustfs-ecstore) & (
+    test(/^set_disk::ops::multipart::tests::complete_multipart_(quorum_success_releases_logical_guards_before_physical_tail|waits_for_tail_before_releasing_guards_and_marking_capacity)$/)
+    | test(=store::heal::tests::full_format_heal_lost_after_last_save_does_not_publish_or_renew_and_latches)
+    | test(/^store::init::tests::(decommission_durable_ilm_receipt_pagination_fails_closed_on_second_page|dispatch_manifest_shutdown_(drains_admitted_delete_batch_without_tail_admission|stops_aborting_validation_tail_reads|stops_authorized_tail_reads_before_receipt_or_cas))$/)
+    | (binary(metadata_read_lock_reuse_test) & test(=replica_multipart_completes_while_metadata_writer_waits_for_its_snapshot))
+)
+"""
+threads-required = "num-test-threads"
+
 [[profile.ci.overrides]]
 filter = 'package(rustfs) & test(/^connect::diagnostics::schedule::tests::/)'
 threads-required = "num-test-threads"
@@ -391,6 +419,12 @@ threads-required = "num-test-threads"
 filter = 'package(rustfs-scanner) & (test(/^scanner::backlog::tests::native_/) | test(=scanner_folder::tests::mrf_ownership::mrf_ownership_manager_completion_preserves_scanner_pending) | test(/^scanner_io::tests::service_cohort::/))'
 threads-required = "num-test-threads"
 
+# Keep short-deadline real-disk progress probes from competing with unrelated
+# tests for scanner and object-lock capacity. Assertions and deadlines stay on.
+[[profile.ci.overrides]]
+filter = '(package(rustfs-scanner) & (test(=scanner::tests::cycle_persist_failure::later_cycle_post_commit_failure_reloads_durable_state) | test(=scanner::tests::running_main_loop_catches_up_pause_cleared_after_startup_observe))) | (package(rustfs-ecstore) & test(=store::init::tests::restored_transitioned_delete_causally_enqueues_free_version))'
+threads-required = "num-test-threads"
+
 [[profile.ci.scripts]]
 filter = 'package(rustfs-ecstore) & test(/^(bucket::lifecycle::bucket_lifecycle_ops::tests::manual_transition_worker_result_recovery_marks_unknown_for_corrupt_marker|services::rebalance::entry::tests::real_rebalance_run_fence_loss_blocks_multipart_publication|store::init::tests::(batch_transitioned_delete_uses_free_version_per_item|decommission_entry_(allows_free_version_consumed_before_source_lock|rejects_subquorum_free_version_conflict_and_retains_source|skips_cleanup_only_marker_when_free_version_is_present)|dispatched_tier_delete_recovery_(checks_later_pool_then_commits_after_source_removal|finds_directory_source_on_encoded_set|retains_journal_on_source_metadata_error)|force_tier_remove_blocks_on_physical_free_version_hidden_by_other_pool|legacy_unknown_transition_delete_falls_back_for_single_batch_and_blocks_prefix|multi_pool_(recursive_prefix_rejects_legacy_or_hidden_merge_loser_before_delete|same_remote_tuple_(batch|single)_delete_waits_for_all_sources|same_tuple_recursive_prefix_uses_one_journal_owner|transitioned_delete_persists_one_free_version_per_remote_tuple)|recursive_prefix_partial_(pool|set)_failure_keeps_prepared_cleanup_owners|restored_transitioned_delete_uses_free_version_as_cleanup_owner|stable_transitioned_recursive_prefix_delete_uses_journal_owners|suspended_null_transition_delete_uses_free_version_as_sole_owner|tier_mutation_peer_handler_applies_prepare_commit_and_abort_idempotently|transition_response_loss_persists_unknown_outcome_for_provider_recovery|transition_transaction_recovery_(drops_record_after_confirmed_local_commit|keeps_cleanup_pending_local_commit)|transitioned_delete_(free_version_replays_after_store_restart|local_quorum_failure_rolls_back_without_cleanup_owner|uses_free_version_as_cleanup_owner)|versioned_delete_marker_keeps_transitioned_source_and_remote_object|versioned_explicit_transition_delete_preserves_other_version_then_allows_bucket_delete))$/)'
 setup = 'ecstore-large-stack'
@@ -650,6 +684,14 @@ test-group = 'ecstore-serial-flaky'
 # the PR-lane e2e coverage for scanner usage snapshots consumed by quota and
 # admin surfaces. It uses the same single-node, random-port, isolated-temp-dir
 # fixture as the existing smoke modules.
+#
+# Dual-stack differential (rustfs/backlog#2740): `dual_stack_differential::`
+# starts two single-node servers from the same binary, one per
+# RUSTFS_S3_STACK value, on random ports with isolated temp dirs, and runs one
+# SDK script covering every operation of `impl S3 for FS` against both. Its
+# comparator unit tests need no server. It meets the admission criteria
+# unchanged; the module clause admits a new differential test without another
+# filter edit.
 [profile.e2e-smoke]
 default-filter = """
     package(e2e_test) & (
@@ -658,6 +700,7 @@ default-filter = """
         | test(/^reliant::lifecycle::/)
         | test(/^conditional_delete_objects_test::/)
         | test(/^reliant::tiering::/)
+        | test(/^dual_stack_differential::/)
         | test(/^upgrade_compatibility_test::upgrade_write_readiness_tests::/)
         | test(/^on_demand_migration::(get_basic_test::(get_miss_pulls_inline_and_serves_locally_afterwards|head_miss_answers_from_the_source_without_persisting)|interaction_test::test_odm_admin_config_is_redacted_and_status_counts_match_the_source)$/)
     )

.config/s3s-edge-allowlist.txt

@@ -0,0 +1,37 @@
+# Files that may still reference s3s (`s3s::` or `s3s_sigv4`) when Phase 1 of
+# the gateway migration ends (rustfs/backlog#2734 task T0.6, rustfs/backlog#2741).
+# Consumed only by `scripts/check_s3s_footprint.sh --mode allowlist`.
+#
+# One entry per line: a repository-relative file path or a git `:(glob)`
+# pathspec. `**` matches across directories, `*` stays within one path
+# component, and a directory entry matches everything under it. Blank lines
+# and lines starting with `#` are ignored; an entry may not contain whitespace,
+# so a note belongs on its own comment line.
+#
+# Every Rust file listed by `git grep -l '\bs3s\(::\|_sigv4\)' -- '*.rs'` that
+# no entry matches fails the allowlist mode. Task T1.9 switches CI to that
+# mode; until then `--dry-run` prints the files still outside this list, which
+# is the Phase 1 tracking list.
+#
+# Never add crates/** other than crates/e2e_test: the engine crates leave s3s
+# entirely, and the script rejects such an entry. Only the s3s-facing edge of
+# the rustfs binary and the e2e oracle keep it.
+
+# rustfs binary: the s3s-facing edge that survives Phase 1
+rustfs/src/storage/ecfs.rs
+rustfs/src/storage/s3_compat/**
+rustfs/src/storage/access.rs
+rustfs/src/auth.rs
+rustfs/src/error.rs
+rustfs/src/storage_api.rs
+rustfs/src/app/storage_api.rs
+rustfs/src/admin/router.rs
+rustfs/src/app/metadata_route.rs
+rustfs/src/server/http.rs
+rustfs/src/server/layer.rs
+rustfs/src/server/hybrid.rs
+rustfs/src/protocols/client.rs
+
+# e2e oracle: the test target that verifies S3 behaviour against s3s
+crates/e2e_test/src/fake_s3_target/**
+crates/e2e_test/src/**

.config/s3s-footprint-baseline.txt

@@ -0,0 +1,17 @@
+# Exact baseline for the direct s3s footprint, consumed only by
+# scripts/check_s3s_footprint.sh (rustfs/backlog#1677; rustfs/backlog#2734 T0.4).
+#
+# Each value is the exact measurement of the tree, not a ceiling. CI fails when a
+# measured counter differs from its line here in either direction:
+#   - a larger measurement means new code widened the s3s surface: route it
+#     through the gateway abstractions instead; never raise a value here;
+#   - a smaller measurement means the PR shrank the surface: lower the matching
+#     value to the measured number in the same PR.
+#
+# files           Rust files referencing s3s paths, repo-wide minus crates/e2e_test
+# s3_error_lines  lines holding s3_error!, same scope, minus those the script
+#                 header's rule clears (rustfs_s3_types's macro); fails closed
+# ecstore_files   Rust files referencing s3s paths under crates/ecstore/src
+files=65
+s3_error_lines=289
+ecstore_files=0

.config/test-count-baseline.txt

@@ -0,0 +1,64 @@
+# Per-crate test-count baseline consumed by scripts/check_test_count_ratchet.sh
+# (rustfs/backlog#2734 section 1 rule 3; T0.3 in rustfs/backlog#2738).
+#
+# One row per workspace member: <package>, tests, ignored, tab-separated.
+# tests counts #[test], #[tokio::test], #[tokio::test(...)] and #[test_case(...)]
+# attribute lines under the package directory; ignored counts #[ignore] and
+# #[ignore = "..."]. Both columns are exact: CI fails when a crate's tests fall
+# below or rise above its row, or when its ignored count changes. Raise a row in
+# the PR that adds tests; lowering one is a reviewed statement that tests were
+# intentionally removed. Regenerate with:
+#   scripts/check_test_count_ratchet.sh --update
+e2e_test	991	20
+rustfs	5988	64
+rustfs-audit	70	0
+rustfs-checksums	26	0
+rustfs-common	36	0
+rustfs-concurrency	54	0
+rustfs-config	43	0
+rustfs-connect-symbol-catalog	6	0
+rustfs-credentials	24	0
+rustfs-crypto	67	0
+rustfs-data-usage	55	0
+rustfs-ecstore	7331	2
+rustfs-extension-schema	23	0
+rustfs-filemeta	352	0
+rustfs-heal	1018	3
+rustfs-heal-contracts	8	0
+rustfs-iam	394	0
+rustfs-io-core	49	0
+rustfs-io-metrics	151	0
+rustfs-keystone	29	0
+rustfs-kms	715	25
+rustfs-license	2	0
+rustfs-lifecycle	155	0
+rustfs-lock	143	0
+rustfs-log-analyzer	90	0
+rustfs-madmin	210	0
+rustfs-notify	137	0
+rustfs-object-capacity	92	0
+rustfs-object-data-cache	140	0
+rustfs-obs	381	0
+rustfs-policy	733	0
+rustfs-protocols	748	1
+rustfs-protos	73	0
+rustfs-replication	209	0
+rustfs-rio	206	0
+rustfs-rio-v2	34	6
+rustfs-s3-client	69	0
+rustfs-s3-ops	8	0
+rustfs-s3-types	103	0
+rustfs-s3select-api	207	0
+rustfs-s3select-query	189	2
+rustfs-scanner	1229	20
+rustfs-scanner-metrics	90	0
+rustfs-security-governance	20	0
+rustfs-signer	31	0
+rustfs-storage-api	34	0
+rustfs-targets	603	26
+rustfs-test-utils	1	0
+rustfs-tls-runtime	43	0
+rustfs-trusted-proxies	68	0
+rustfs-utils	247	0
+rustfs-zip	30	0
+zstd	11	0

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

@@ -86,6 +86,10 @@ runs:
       shell: bash
       run: ./scripts/check_s3s_footprint.sh
 
+    - name: Check per-crate test-count ratchet
+      shell: bash
+      run: ./scripts/check_test_count_ratchet.sh
+
     - name: Check cryptographic capability wording
       shell: bash
       run: ./scripts/check_fips_wording.sh

.github/actions/schedule-failure-issue/README.md

@@ -74,7 +74,6 @@ Notes:
 - `.github/workflows/e2e-s3tests.yml` (weekly full s3-tests sweep)
 - `.github/workflows/mint.yml` (weekly multi-SDK mint run)
 - `.github/workflows/fuzz.yml` (nightly fuzz corpus jobs)
-- `.github/workflows/performance-ab.yml` (nightly warp A/B gate)
 - **Pending:** the ci-7 e2e nightly-full workflow does not exist yet
   (backlog#1149 ci-7). When it lands, wire it up with the snippet above.
 

.github/scheduled-validations.json

@@ -18,7 +18,6 @@
     "never_ran_grace_until": "2026-09-08T00:00:00Z"
   },
   { "workflow": ".github/workflows/nightly-gnu.yml", "max_age_hours": 36 },
-  { "workflow": ".github/workflows/performance-ab.yml", "max_age_hours": 36 },
   {
     "workflow": ".github/workflows/runner-hygiene.yml",
     "max_age_hours": 792,

.github/workflows/README.md

@@ -7,7 +7,7 @@
 
 This repository is public. For a public repository, GitHub's standard hosted runners (`ubuntu-latest`, `windows-latest`, `macos-latest`) cost nothing and have no minute cap, whatever plan the organisation is on. GitHub bills only two things: larger hosted runners, which have custom labels and are not used here, and standard runners in private repositories.
 
-Every other label in this repository is capacity RustFS pays for by the hour whether or not a job is running well:
+Every other label in this repository is capacity RustFS pays for. ARC pods and lab machines accrue cost while the machine exists. A Ubicloud label is metered for the minutes a job holds its VM:
 
 | Label | Paid by | What it is | Use it for |
 |---|---|---|---|
@@ -15,7 +15,7 @@ Every other label in this repository is capacity RustFS pays for by the hour whe
 | `windows-latest`, `macos-latest` | nobody | GitHub-hosted standard runners | Windows and macOS targets |
 | `sm-standard-2` | RustFS | ARC pod, Ubuntu 24.04, minimal image: no `gh`, no C toolchain; Docker only on the `dind-` label | only a job that a hosted runner cannot run (see the placement rule) |
 | `sm-standard-4` | RustFS | ARC pod, Ubuntu 24.04, 14 CPUs and 28 GB (the label is not the core count), large disk | workspace-scale Cargo builds |
-| `sm-standard-4-arm` | RustFS | ARC pod, aarch64; reached through the `build.yml` matrix, so `actionlint.yaml` does not list it | native aarch64 release builds |
+| `ubicloud-standard-16-arm` | RustFS | Ubicloud VM, alias of `ubicloud-standard-16-arm-ubuntu-2404`: Ubuntu 24.04 arm64, 16 vCPU, 48 GB, 300 GB disk. Reached through the `build.yml` matrix, so `actionlint.yaml` does not list it | native aarch64 release builds (`linux-aarch64-musl`, `linux-aarch64-gnu`) |
 | `dind-sm-standard-2` | RustFS | ARC pod with Docker-in-Docker | image builds that need `buildx` and a registry push |
 | `smoke-testing`, `pf-testing` | RustFS | long-lived lab machines | the packaged functional chain and performance suites |
 
@@ -40,7 +40,7 @@ A hosted runner is also the only choice for a few things the pods cannot do: `mo
 
 - Every `shared-key` has exactly one writer. `cache-warm.yml` owns `ci-dev`, `ci-feat-rio`, `ci-feat-proto` and `ci-uring`; every other `./.github/actions/setup` call states `cache-save-if` explicitly, and `scripts/security/check_cache_save_if.sh` fails the build when one does not.
 - All readers of a key run on the same image as its writer. Keys that live on hosted runners say so in their name (`ci-uring` is warmed on `ubuntu-latest`; `ci-e2e-distributed-hosted`, `e2e-upgrade-server-hosted`, `e2e-upgrade-tests-hosted`).
-- Moving a job between pools therefore means renaming its key. The new key is empty until a run from `main` (a scheduled run, or a manual dispatch of that workflow from `main`) saves it; until then that lane does a cold build.
+- Moving a job between pools therefore means renaming its key. The new key is empty until a run from `main` (a scheduled run, or a manual dispatch of that workflow from `main`) saves it; until then that lane does a cold build. The `build.yml` aarch64 release legs append `-ubicloud-arm` for this reason. `cache-warm.yml` does not write those keys.
 - `CARGO_BUILD_JOBS` and every other `CARGO*`, `CC*`, `CFLAGS*`, `CXX*`, `CMAKE*` and `RUST*` variable is hashed into the key. `ci.yml` sets `CARGO_BUILD_JOBS` at step level for that reason; promoting it to job level rotates the key for every lane that shares it.
 
 ## Where self-hosted minutes leak
@@ -62,7 +62,7 @@ gh api "repos/rustfs/rustfs/actions/runs/<run_id>/jobs?per_page=100" \
   --jq '.jobs[] | "\(.conclusion)\t\(.labels | join(","))\t\(.started_at)\t\(.completed_at)\t\(.name)"'
 ```
 
-Only rows whose label starts with `sm-`, `dind-`, `smoke-` or `pf-` cost anything. At the October 2026 review, a full pull-request run of `ci.yml` spent about 150 self-hosted minutes, a third of it in jobs that never compiled; treat that as the scale of what a placement mistake costs, not as a current figure.
+Only rows whose label starts with `sm-`, `dind-`, `ubicloud-`, `smoke-` or `pf-` cost anything. At the October 2026 review, a full pull-request run of `ci.yml` spent about 150 self-hosted minutes, a third of it in jobs that never compiled; treat that as the scale of what a placement mistake costs, not as a current figure.
 
 ## Checklist for a new or moved job
 

.github/workflows/audit.yml

@@ -29,7 +29,6 @@ on:
       - 'scripts/test_package_service_scripts.sh'
       - 'scripts/test_package_versions.sh'
       - 'scripts/test_docker_workflow.py'
-      - 'scripts/security/check_performance_ab_workflow.sh'
       - 'scripts/security/check_preview_release_workflow.sh'
       - 'scripts/security/check_tier_artifact_workflow.sh'
       - 'scripts/security/check_workflow_pins.sh'
@@ -48,7 +47,6 @@ on:
       - 'scripts/test_package_service_scripts.sh'
       - 'scripts/test_package_versions.sh'
       - 'scripts/test_docker_workflow.py'
-      - 'scripts/security/check_performance_ab_workflow.sh'
       - 'scripts/security/check_preview_release_workflow.sh'
       - 'scripts/security/check_tier_artifact_workflow.sh'
       - 'scripts/security/check_workflow_pins.sh'
@@ -155,9 +153,6 @@ jobs:
       - name: Check preview release workflow policy
         run: ./scripts/security/check_preview_release_workflow.sh
 
-      - name: Check performance A/B workflow trust boundary
-        run: ./scripts/security/check_performance_ab_workflow.sh
-
       - name: Check tier evidence workflow isolation
         run: ./scripts/security/check_tier_artifact_workflow.sh
 

.github/workflows/btf-prototype-reference.yml

@@ -0,0 +1,75 @@
+name: BTF prototype reference
+
+on:
+  pull_request:
+    paths:
+      - "scripts/inspect_btf_prototypes.py"
+      - "scripts/test_inspect_btf_prototypes.py"
+      - "scripts/fixtures/btf_prototypes.c"
+      - ".github/workflows/btf-prototype-reference.yml"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: btf-prototype-reference-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  reference:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
+        with:
+          persist-credentials: false
+      - uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
+        with:
+          python-version: "3.12"
+      - name: Check prototype and CLI contracts
+        run: python3 -B scripts/test_inspect_btf_prototypes.py
+      - name: Install only missing native reference tools
+        run: |
+          select_bpftool() {
+            local candidate
+            local candidates=()
+            if command -v bpftool >/dev/null; then
+              candidates+=("$(command -v bpftool)")
+            fi
+            for candidate in /usr/lib/linux-tools-*/bpftool /usr/lib/linux-tools/*/bpftool; do
+              if [[ -x "$candidate" ]]; then
+                candidates+=("$candidate")
+              fi
+            done
+            while IFS= read -r candidate; do
+              if [[ -x "$candidate" ]] && file -Lb "$candidate" | grep -q 'ELF' \
+                && "$candidate" version >/dev/null 2>&1; then
+                printf '%s\n' "$candidate"
+                return 0
+              fi
+            done < <(printf '%s\n' "${candidates[@]}" | sort -Vr)
+            return 1
+          }
+          packages=()
+          command -v clang >/dev/null || packages+=(clang)
+          command -v llvm-objcopy >/dev/null || packages+=(llvm)
+          command -v gcc >/dev/null || packages+=(gcc)
+          select_bpftool >/dev/null || packages+=(linux-tools-generic)
+          if (( ${#packages[@]} )); then
+            sudo apt-get update
+            sudo apt-get install --no-install-recommends -y "${packages[@]}"
+          fi
+          # File parsing needs a working binary, not a host-kernel wrapper.
+          selected_bpftool="$(select_bpftool)"
+          "$selected_bpftool" version
+          printf 'BPFTOOL=%s\n' "$selected_bpftool" >> "$GITHUB_ENV"
+      - name: Compare declarations and layouts with independent references
+        run: |
+          clang -target bpf -O2 -g -c scripts/fixtures/btf_prototypes.c -o "$RUNNER_TEMP/prototypes.o"
+          llvm-objcopy --dump-section ".BTF=$RUNNER_TEMP/prototypes.btf" "$RUNNER_TEMP/prototypes.o"
+          gcc -std=c11 -Wall -Wextra -Werror -DRUSTFS_BTF_LAYOUT_REFERENCE \
+            scripts/fixtures/btf_prototypes.c -o "$RUNNER_TEMP/layout-reference"
+          "$RUNNER_TEMP/layout-reference" > "$RUNNER_TEMP/layout-reference.json"
+          python3 -B scripts/test_inspect_btf_prototypes.py --native-reference "$RUNNER_TEMP/prototypes.btf" \
+            --layout-reference "$RUNNER_TEMP/layout-reference.json"

.github/workflows/build.yml

@@ -181,12 +181,18 @@ jobs:
             selected="all"
           fi
 
-
+          # linux-aarch64-* need a real aarch64 host (placement rule: hardware)
+          # so the packaged binary and console smoke execute the artifact.
+          # ubicloud-standard-16-arm is that host: Ubuntu 24.04 arm64, alias of
+          # ubicloud-standard-16-arm-ubuntu-2404 (16 vCPU, 48 GB). linux-x86_64-* stay on the
+          # ARC sm-standard-4 pool. GNU rows keep cross=true so cargo-zigbuild
+          # pins the glibc 2.31 floor on the same architecture; musl stays a
+          # native cargo build.
           all='{"include":[
             {"target_id":"linux-x86_64-musl","os":"sm-standard-4","target":"x86_64-unknown-linux-musl","cross":false,"platform":"linux","rustflags":"","features":"","glibc":""},
-            {"target_id":"linux-aarch64-musl","os":"sm-standard-4-arm","target":"aarch64-unknown-linux-musl","cross":false,"platform":"linux","rustflags":"","features":"","glibc":""},
+            {"target_id":"linux-aarch64-musl","os":"ubicloud-standard-16-arm","target":"aarch64-unknown-linux-musl","cross":false,"platform":"linux","rustflags":"","features":"","glibc":""},
             {"target_id":"linux-x86_64-gnu","os":"sm-standard-4","target":"x86_64-unknown-linux-gnu","cross":true,"platform":"linux","rustflags":"","features":"pyroscope","glibc":"2.31"},
-            {"target_id":"linux-aarch64-gnu","os":"sm-standard-4-arm","target":"aarch64-unknown-linux-gnu","cross":true,"platform":"linux","rustflags":"","features":"pyroscope","glibc":"2.31"},
+            {"target_id":"linux-aarch64-gnu","os":"ubicloud-standard-16-arm","target":"aarch64-unknown-linux-gnu","cross":true,"platform":"linux","rustflags":"","features":"pyroscope","glibc":"2.31"},
             {"target_id":"macos-aarch64","os":"macos-latest","target":"aarch64-apple-darwin","cross":false,"platform":"macos","rustflags":"","features":"pyroscope","glibc":""},
             {"target_id":"windows-x86_64","os":"windows-latest","target":"x86_64-pc-windows-msvc","cross":false,"platform":"windows","rustflags":"","features":"","glibc":""}
           ]}'
@@ -244,7 +250,11 @@ jobs:
         with:
           rust-version: stable
           target: ${{ matrix.target }}
-          cache-shared-key: build-${{ matrix.target }}${{ matrix.glibc && format('-glibc-{0}', matrix.glibc) || '' }}
+          # rust-cache keys include OS and architecture but not the runner image.
+          # The Ubicloud aarch64 legs therefore use their own suffix so they
+          # cannot restore a cache written on a different arm64 image.
+          # x86_64, macOS, and Windows keys stay unsuffixed.
+          cache-shared-key: build-${{ matrix.target }}${{ matrix.glibc && format('-glibc-{0}', matrix.glibc) || '' }}${{ matrix.os == 'ubicloud-standard-16-arm' && '-ubicloud-arm' || '' }}
           # main only. A cache saved on refs/tags/X is scoped to that tag: no
           # other tag, no main run and no PR can restore it, so every release
           # cycle wrote up to 12 entries of 1-2GB (preview tag plus final tag,
@@ -764,9 +774,9 @@ jobs:
             exit 0
           fi
 
-          # The self-hosted Linux runners do not ship the aws CLI (GitHub-hosted
-          # images did). Install the official CLI bundle on demand so R2 uploads
-          # survive a fresh runner without relying on a distro package.
+          # ARC pods (the x86_64 legs) do not ship the aws CLI. Ubicloud arm64
+          # images follow the GitHub runner image and often do. Install the
+          # official CLI bundle when it is missing so R2 uploads work on either.
           if ! command -v aws >/dev/null 2>&1; then
             echo "aws CLI not found on runner; installing..."
             if command -v apt-get >/dev/null 2>&1; then

.github/workflows/ci.yml

@@ -16,7 +16,9 @@ name: Continuous Integration
 
 on:
   push:
-    branches: [ main ]
+    # gateway/integration receives the daily merge of main as a direct push
+    # (rustfs/backlog#2734), so its pushes must prove the result like main's.
+    branches: [ main, gateway/integration ]
     paths-ignore:
       - "**.md"
       - "docs/**"
@@ -737,6 +739,8 @@ jobs:
     needs: [ build-rustfs-debug-binary ]
     runs-on: sm-standard-2
     timeout-minutes: 30
+    env:
+      RUSTFS_S3_STACK: legacy
     steps:
       - name: Checkout repository
         uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
@@ -855,18 +859,105 @@ jobs:
           path: ${{ runner.temp }}/rustfs-e2e-*/rustfs.log
           retention-days: 3
 
+  # The same e2e-smoke suite against the same debug binary with the gateway
+  # S3 stack selected (rustfs/backlog#2734). Report-only: it is outside the
+  # Test and Lint aggregate and becomes required at T2.10. The archive and run
+  # steps must stay identical to e2e-tests; scripts/ci_gate.py enforces that.
+  e2e-tests-gateway:
+    name: e2e-smoke (gateway, report-only until T2.10)
+    needs: [ build-rustfs-debug-binary ]
+    runs-on: sm-standard-2
+    timeout-minutes: 30
+    continue-on-error: true
+    env:
+      RUSTFS_S3_STACK: gateway
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
+        with:
+          persist-credentials: false
+
+      - name: Setup Rust environment
+        uses: ./.github/actions/setup
+        with:
+          rust-version: stable
+          cache-shared-key: ci-dev
+          cache-save-if: 'false'
+          install-build-packaging-tools: 'false'
+
+      - name: Set up Python
+        uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
+        with:
+          python-version: "3.12"
+
+      - name: Install awscurl
+        run: |
+          python3 -m pip install --user --upgrade pip "awscurl==0.44"
+          echo "AWSCURL_PATH=$HOME/.local/bin/awscurl" >> "$GITHUB_ENV"
+
+      - name: Verify awscurl
+        run: test -x "$AWSCURL_PATH"
+
+      - name: Download debug binary
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: rustfs-debug-binary
+          path: target/debug
+
+      - name: Make binary executable
+        run: chmod +x ./target/debug/rustfs
+
+      - name: Archive e2e smoke test binaries
+        env:
+          NEXTEST_ARCHIVE: ${{ runner.temp }}/rustfs-e2e-smoke.tar.zst
+          NEXTEST_LISTING: ${{ runner.temp }}/rustfs-e2e-smoke-list.json
+        run: |
+          cargo nextest archive --profile e2e-smoke -p e2e_test --archive-file "${NEXTEST_ARCHIVE}"
+          cargo nextest list --profile e2e-smoke --archive-file "${NEXTEST_ARCHIVE}" --message-format json > "${NEXTEST_LISTING}"
+          python3 ./scripts/check_test_wiring.py --check-profile e2e-smoke "${NEXTEST_LISTING}"
+          ./scripts/check_security_smoke_count.sh check "${NEXTEST_LISTING}"
+
+      - name: Run e2e smoke suite
+        env:
+          NEXTEST_ARCHIVE: ${{ runner.temp }}/rustfs-e2e-smoke.tar.zst
+          RUSTFS_E2E_LOG_DIR: ${{ runner.temp }}/rustfs-e2e-smoke-logs
+          SAMPLER_INTERVAL_SECS: "15"
+        run: |
+          ./scripts/ci/resource_sampler.sh start e2e-smoke
+          trap './scripts/ci/resource_sampler.sh stop' EXIT
+          python3 scripts/e2e_binary.py run --features e2e-test-hooks -- cargo nextest run --profile e2e-smoke --archive-file "${NEXTEST_ARCHIVE}" \
+            --status-level all --final-status-level all --failure-output final
+
+      - name: Upload e2e smoke diagnostics
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: e2e-smoke-gateway-diagnostics-${{ github.run_number }}
+          path: |
+            ${{ runner.temp }}/rustfs-e2e-smoke-logs/
+            ${{ runner.temp }}/rustfs-e2e-smoke-list.json
+          if-no-files-found: warn
+
+      - name: Upload e2e smoke JUnit report
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: e2e-smoke-gateway-junit-${{ github.run_number }}
+          path: target/nextest/e2e-smoke/junit.xml
+          if-no-files-found: warn
+
   e2e-full:
     name: End-to-End Tests (full merge gate)
     # Merge gate only (backlog#1149 ci-5): the never-automated user-visible
     # suites — KMS, object_lock, multipart_auth, quota, checksum, encryption,
     # security-boundary, ... — via the e2e-full nextest profile. Too heavy for
-    # every PR, so it is gated to main pushes, the merge queue, and manual
-    # dispatch. protocols / the 7 cluster suites / replication / #[ignore] are
-    # owned by other lanes (see .config/nextest.toml profile.e2e-full).
+    # every PR, so it is gated to main and gateway/integration pushes, the merge
+    # queue, and manual dispatch. protocols / the 7 cluster suites / replication
+    # / #[ignore] are owned by other lanes (see .config/nextest.toml profile.e2e-full).
     if: >-
       github.event_name == 'workflow_dispatch' ||
       github.event_name == 'merge_group' ||
-      (github.event_name == 'push' && github.ref == 'refs/heads/main')
+      (github.event_name == 'push' && (github.ref == 'refs/heads/main' || github.ref == 'refs/heads/gateway/integration'))
     needs: [ build-rustfs-debug-binary ]
     runs-on: sm-standard-2
     timeout-minutes: 55

.github/workflows/coverage.yml

@@ -45,7 +45,7 @@ on:
     # 07:00 UTC Sunday — staggered clear of the other Sunday crons: ci (00:00),
     # build (01:00), e2e-s3tests (02:00), audit (03:00), nix-flake-update
     # (05:00), mint (06:00), and the daily fuzz (02:00), minio-interop (03:17),
-    # e2e-replication-nightly (04:00) and performance-ab (06:00) lanes.
+    # and e2e-replication-nightly (04:00) lanes.
     - cron: "43 7 * * 0"
 
 concurrency:

.github/workflows/e2e-replication-nightly.yml

@@ -38,8 +38,8 @@ name: e2e-nightly
 on:
   workflow_dispatch:
   schedule:
-    # 04:00 UTC nightly — staggered clear of fuzz/e2e-s3tests (02:00),
-    # stale (01:30) and performance-ab (06:00).
+    # 04:00 UTC nightly — staggered clear of fuzz/e2e-s3tests (02:00)
+    # and stale (01:30).
     - cron: "29 4 * * *"
 
 # Only alert-on-failure needs more than read access; it declares its own

.github/workflows/e2e-upgrade.yml

@@ -19,9 +19,11 @@ on:
     paths:
       - ".github/workflows/e2e-upgrade.yml"
       - "crates/e2e_test/src/common.rs"
+      - "crates/e2e_test/src/distributed/**"
       - "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/rpc_fault_proxy.rs"
       - "crates/e2e_test/src/upgrade_compatibility_test.rs"
       - "crates/ecstore/**"
       - "crates/filemeta/**"
@@ -128,6 +130,7 @@ jobs:
         run: chmod +x target/debug/rustfs
 
       - name: Verify and unpack pinned previous release
+        id: verify_source
         env:
           SOURCE_DIR: ${{ runner.temp }}/rustfs-upgrade-source
         run: |
@@ -185,6 +188,28 @@ jobs:
           done
           exit "$status"
 
+      - name: Run mixed-version MPU recovery
+        if: ${{ !cancelled() && steps.verify_source.outcome == 'success' }}
+        timeout-minutes: 15
+        env:
+          NO_PROXY: 127.0.0.1,localhost,::1
+          HTTP_PROXY: ""
+          HTTPS_PROXY: ""
+          ALL_PROXY: ""
+          RUSTFS_E2E_LOG_FILTER: "warn,rustfs_ecstore::cluster::rpc::client=debug"
+        run: |
+          set -euo pipefail
+          export no_proxy="$NO_PROXY"
+          export RUSTFS_E2E_LOG_DIR="$RUNNER_TEMP/rustfs-upgrade-logs/mpu-mixed-recovery"
+          mkdir -p "$RUSTFS_E2E_LOG_DIR"
+          mpu_test='distributed::multipart_fence_acceptance_test::mixed_version_failed_complete_requires_current_coordinator_for_abort'
+          python3 scripts/e2e_binary.py run -- cargo test --locked -p e2e_test --lib \
+            "$mpu_test" -- --list --exact | tee "$RUSTFS_E2E_LOG_DIR/list.log"
+          test "$(grep -Fxc "${mpu_test}: test" "$RUSTFS_E2E_LOG_DIR/list.log")" -eq 1
+          python3 scripts/e2e_binary.py run -- cargo test --locked -p e2e_test --lib \
+            "$mpu_test" -- --exact --nocapture --test-threads=1 \
+            2>&1 | tee "$RUSTFS_E2E_LOG_DIR/test.log"
+
       - name: Upload scanner/heal G09 evidence
         if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1

.github/workflows/performance-ab.yml

@@ -1,384 +0,0 @@
-# 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.
-
-# Warp A/B budget gate for the hotpath series (rustfs/backlog#935 HP-14, item 4).
-#
-# Two entry points, honestly scoped:
-#   * schedule (nightly, on main): post-merge detection — catches a regression
-#     within 24h of landing, not before merge.
-#   * workflow_dispatch: an explicitly selected trusted ref.
-# The dispatch input can run the gate with --allow-regression so a deliberate
-# correctness cost (e.g. the #4221 fsync durability fix) is recorded, not
-# blocked (rustfs/backlog#935 correction 1).
-
-name: Performance A/B
-
-on:
-  schedule:
-    - cron: "31 6 * * *" # 06:31 UTC nightly, against main
-  workflow_dispatch:
-    inputs:
-      duration:
-        description: "warp duration per round"
-        required: false
-        default: "12s"
-        type: string
-      allow_regression:
-        description: "Pass the gate despite a FAIL (deliberate tradeoff)"
-        required: false
-        default: false
-        type: boolean
-permissions:
-  actions: read
-  contents: read
-
-env:
-  CARGO_TERM_COLOR: always
-  RUST_BACKTRACE: 1
-
-jobs:
-  warp-ab:
-    if: github.repository == 'rustfs/rustfs'
-    name: Warp A/B budget gate
-    runs-on: sm-standard-2
-    # A normal nightly restores the last successful binary and builds only the
-    # candidate; daily access keeps that cache warm. A cache miss may build both
-    # and needs room for the A/B run plus artifact and cache publication.
-    timeout-minutes: 180
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
-        with:
-          persist-credentials: false
-          fetch-depth: 0 # baseline may be an earlier successful scheduled head
-
-      - name: Setup Rust environment
-        uses: ./.github/actions/setup
-        with:
-          rust-version: stable
-          cache-shared-key: warp-ab-${{ hashFiles('**/Cargo.lock') }}
-          cache-save-if: ${{ github.ref == 'refs/heads/main' }}
-
-      - name: Install warp
-        run: |
-          set -euo pipefail
-          WARP_VERSION="1.0.0"
-          curl -fsSL "https://github.com/minio/warp/releases/download/v${WARP_VERSION}/warp_Linux_x86_64.tar.gz" \
-            | sudo tar -xz -C /usr/local/bin warp
-          warp --version
-
-      - name: Decide exemption
-        id: exempt
-        env:
-          INPUT_ALLOW_REGRESSION: ${{ github.event.inputs.allow_regression }}
-        run: |
-          allow="false"
-          if [[ "$INPUT_ALLOW_REGRESSION" == "true" ]]; then
-            allow="true"
-          fi
-          echo "allow_regression=$allow" >> "$GITHUB_OUTPUT"
-
-      # A failed regression run must keep comparing against the last known-good
-      # scheduled head. Otherwise the next nightly would absorb the regression
-      # into its baseline and turn green without a fix.
-      - name: Find last successful scheduled baseline
-        id: scheduled_baseline
-        if: github.event_name == 'schedule'
-        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
-        with:
-          result-encoding: string
-          script: |
-            const { data } = await github.rest.actions.listWorkflowRuns({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              workflow_id: "performance-ab.yml",
-              event: "schedule",
-              status: "success",
-              per_page: 1,
-            });
-            return data.workflow_runs[0]?.head_sha ?? "";
-
-      # Manual runs compare a selected ref with current main. Scheduled runs
-      # compare current main with the last successful scheduled head. With no
-      # history, the first run measures the candidate against itself and seeds
-      # that head only if the complete rig succeeds.
-      - name: Resolve baseline / candidate commits
-        id: commits
-        env:
-          SCHEDULED_BASELINE_SHA: ${{ steps.scheduled_baseline.outputs.result }}
-        run: |
-          set -euo pipefail
-          candidate_sha="$(git rev-parse HEAD)"
-          if [[ "${{ github.event_name }}" == "schedule" ]]; then
-            baseline_sha="${SCHEDULED_BASELINE_SHA:-$candidate_sha}"
-          else
-            baseline_sha="$(git rev-parse origin/main)"
-          fi
-          git cat-file -e "${baseline_sha}^{commit}"
-          if ! git merge-base --is-ancestor "$baseline_sha" "$candidate_sha"; then
-            echo "::error::baseline $baseline_sha is not an ancestor of candidate $candidate_sha; update the selected ref before comparing" >&2
-            exit 1
-          fi
-          echo "baseline_sha=$baseline_sha" >> "$GITHUB_OUTPUT"
-          echo "candidate_sha=$candidate_sha" >> "$GITHUB_OUTPUT"
-          echo "baseline commit: $baseline_sha"
-          echo "candidate commit: $candidate_sha"
-
-      # Exact-key restore of the candidate binary saved by its successful
-      # scheduled run. A miss leaves cache-hit unset and falls back to a source
-      # build of that known-good head.
-      - name: Restore cached baseline binary
-        id: baseline_cache
-        uses: actions/cache/restore@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6
-        with:
-          path: baseline-bin/rustfs
-          key: rustfs-baseline-${{ steps.commits.outputs.baseline_sha }}
-
-      # Self-heal: on a nightly/dispatch run where the candidate commit IS the
-      # baseline commit, a cache miss would make the rig build the same commit
-      # twice (~65min per side with the post-#4806 LTO profile — no job budget
-      # fits that). Build it once here, reuse it for both phases, and save it
-      # back to the cache so the next run hits.
-      - name: Build baseline on cache miss (same-commit self-heal)
-        id: selfheal
-        if: >-
-          steps.baseline_cache.outputs.cache-hit != 'true' &&
-          steps.commits.outputs.baseline_sha == steps.commits.outputs.candidate_sha
-        run: |
-          set -euo pipefail
-          cargo build --release --bin rustfs
-          mkdir -p baseline-bin
-          cp target/release/rustfs baseline-bin/rustfs
-          echo "built=true" >> "$GITHUB_OUTPUT"
-
-      - name: Build baseline on cache miss (different candidate)
-        id: baseline_build
-        if: >-
-          steps.baseline_cache.outputs.cache-hit != 'true' &&
-          steps.commits.outputs.baseline_sha != steps.commits.outputs.candidate_sha
-        run: |
-          set -euo pipefail
-          baseline_root="$RUNNER_TEMP/rustfs-baseline-${{ github.run_id }}"
-          baseline_target="$RUNNER_TEMP/rustfs-baseline-target-${{ github.run_id }}"
-          git worktree add --detach "$baseline_root" "${{ steps.commits.outputs.baseline_sha }}"
-          cargo build --release --manifest-path "$baseline_root/Cargo.toml" --bin rustfs --target-dir "$baseline_target"
-          mkdir -p baseline-bin
-          cp "$baseline_target/release/rustfs" baseline-bin/rustfs
-          git worktree remove --force "$baseline_root"
-          echo "built=true" >> "$GITHUB_OUTPUT"
-
-      - name: Build candidate binary
-        id: candidate_build
-        if: steps.commits.outputs.baseline_sha != steps.commits.outputs.candidate_sha
-        run: |
-          set -euo pipefail
-          cargo build --release --bin rustfs
-          mkdir -p candidate-bin
-          cp target/release/rustfs candidate-bin/rustfs
-          echo "built=true" >> "$GITHUB_OUTPUT"
-
-      - name: Save self-healed baseline to cache
-        if: steps.selfheal.outputs.built == 'true' || steps.baseline_build.outputs.built == 'true'
-        uses: actions/cache/save@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6
-        with:
-          path: baseline-bin/rustfs
-          key: rustfs-baseline-${{ steps.commits.outputs.baseline_sha }}
-
-      - name: Run warp A/B and gate
-        id: ab
-        env:
-          INPUT_DURATION: ${{ github.event.inputs.duration }}
-        run: |
-          set -euo pipefail
-          # The formal runner executes A1 baseline -> B1 candidate -> B2 candidate
-          # -> A2 baseline for each workload and drive-sync cell. It requires three
-          # rounds per leg to emit tail latency and error-rate evidence.
-          # --health-timeout 180 outlasts the server's own 120s startup-readiness
-          # budget, which the rig's previous 60s health poll undershot (the first
-          # two nightly failures). perf-6 recalibrates these once the noise study
-          # lands.
-          duration="${INPUT_DURATION:-12s}"
-          baseline_sha="${{ steps.commits.outputs.baseline_sha }}"
-          candidate_sha="${{ steps.commits.outputs.candidate_sha }}"
-          baseline_hit="${{ steps.baseline_cache.outputs.cache-hit }}"
-          selfheal_built="${{ steps.selfheal.outputs.built }}"
-          baseline_built="${{ steps.baseline_build.outputs.built }}"
-          candidate_built="${{ steps.candidate_build.outputs.built }}"
-
-          args=(--duration "$duration" --rounds 3 --cooldown 5 --health-timeout 180 --baseline-revision "$baseline_sha" --candidate-revision "$candidate_sha")
-
-          if [[ "$baseline_hit" == "true" || "$selfheal_built" == "true" || "$baseline_built" == "true" ]]; then
-            chmod +x baseline-bin/rustfs
-            base_bin="$PWD/baseline-bin/rustfs"
-            args+=(--baseline-bin "$base_bin")
-            if [[ "$baseline_hit" == "true" ]]; then
-              base_src="actions-cache (rustfs-baseline-$baseline_sha)"
-            elif [[ "$selfheal_built" == "true" ]]; then
-              base_src="source build (cache self-heal, saved as rustfs-baseline-$baseline_sha)"
-            else
-              base_src="isolated baseline source build (saved as rustfs-baseline-$baseline_sha)"
-            fi
-            if [[ "$candidate_sha" == "$baseline_sha" ]]; then
-              # No commits landed since the last successful baseline, so reuse
-              # the one binary for both phases and measure only rig drift.
-              args+=(--candidate-bin "$base_bin")
-              cand_src="same binary as baseline (same commit)"
-            elif [[ "$candidate_built" == "true" ]]; then
-              chmod +x candidate-bin/rustfs
-              args+=(--candidate-bin "$PWD/candidate-bin/rustfs")
-              cand_src="source build of the checked-out ref"
-            else
-              echo "::error::candidate binary was not built" >&2
-              exit 2
-            fi
-          else
-            echo "::error::baseline binary was not restored or built" >&2
-            exit 2
-          fi
-
-          echo "baseline binary: $base_src"
-          echo "candidate binary: $cand_src"
-
-          if [[ "${{ steps.exempt.outputs.allow_regression }}" == "true" ]]; then
-            args+=(--allow-regression --exemption-reason "workflow dispatch override")
-          fi
-          # Do not let a gate FAIL abort the job here; capture status and surface
-          # it after the step summary is written.
-          set +e
-          bash scripts/run_hotpath_warp_abba.sh "${args[@]}"
-          echo "status=$?" >> "$GITHUB_OUTPUT"
-          set -e
-          # Locate the newest run dir + gate.md for the summary/comment/artifact
-          # steps. On a startup failure there is no gate.md, but the run dir still
-          # holds server-logs/ for diagnosis.
-          # Run dirs are UTC-timestamp names (no special chars); ls is safe here.
-          # shellcheck disable=SC2012
-          run_dir="$(ls -td target/hotpath-abba/*/ 2>/dev/null | head -n1 || true)"
-          echo "run_dir=${run_dir%/}" >> "$GITHUB_OUTPUT"
-          # shellcheck disable=SC2012
-          gate_md="$(ls -t target/hotpath-abba/*/candidate_gate.md 2>/dev/null | head -n1 || true)"
-          echo "gate_md=$gate_md" >> "$GITHUB_OUTPUT"
-
-      - name: Upload A/B results
-        if: always()
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
-        with:
-          name: hotpath-warp-ab-${{ github.run_number }}
-          # Includes per-cell median_summary.csv / baseline_compare.csv, both gates,
-          # and server-logs/ (rustfs.log + startup env per phase) so a failed run
-          # is diagnosable. Short retention: this is churny nightly debug data.
-          path: target/hotpath-abba/
-          if-no-files-found: warn
-          retention-days: 14
-
-      - name: Write gate summary
-        if: always()
-        run: |
-          set -euo pipefail
-          status="${{ steps.ab.outputs.status }}"
-          gate_md="${{ steps.ab.outputs.gate_md }}"
-          run_dir="${{ steps.ab.outputs.run_dir }}"
-          {
-            echo "## Hotpath warp A/B — run ${{ github.run_number }}"
-            echo
-            if [[ "$status" == "0" ]]; then
-              echo "Rig/gate exit: \`0\` (pass or warn)."
-            else
-              echo "Rig/gate exit: \`${status:-unknown}\` — **FAILED**."
-            fi
-            echo
-            if [[ -n "$gate_md" && -f "$gate_md" ]]; then
-              cat "$gate_md"
-            else
-              echo "No \`gate.md\` produced — the rig failed **before** the gate"
-              echo "(most likely server startup / health). Failing phase(s) below;"
-              echo "full logs in the \`hotpath-warp-ab-${{ github.run_number }}\` artifact."
-              if [[ -n "$run_dir" && -d "$run_dir/server-logs" ]]; then
-                for f in "$run_dir"/server-logs/*.log; do
-                  [[ -f "$f" ]] || continue
-                  echo
-                  echo "<details><summary>$(basename "$f")</summary>"
-                  echo
-                  echo '```'
-                  tail -n 30 "$f"
-                  echo '```'
-                  echo "</details>"
-                done
-              fi
-            fi
-          } >> "$GITHUB_STEP_SUMMARY"
-
-      - name: Stage successful candidate baseline
-        if: >-
-          steps.ab.outputs.status == '0' &&
-          steps.commits.outputs.baseline_sha != steps.commits.outputs.candidate_sha
-        run: |
-          set -euo pipefail
-          cp candidate-bin/rustfs baseline-bin/rustfs
-
-      - name: Cache successful candidate baseline
-        if: >-
-          steps.ab.outputs.status == '0' &&
-          steps.commits.outputs.baseline_sha != steps.commits.outputs.candidate_sha
-        uses: actions/cache/save@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6
-        with:
-          path: baseline-bin/rustfs
-          key: rustfs-baseline-${{ steps.commits.outputs.candidate_sha }}
-
-      # Scheduled failure alerting is handled by the alert-on-failure job below
-      # (perf-2 consuming ci-8's schedule-failure-issue composite action).
-
-      - name: Enforce gate
-        if: always()
-        run: |
-          status="${{ steps.ab.outputs.status }}"
-          if [[ -z "$status" ]]; then
-            echo "::error::warp A/B setup failed before the rig ran. Check the first failed workflow step." >&2
-            exit 1
-          fi
-          if [[ "$status" != "0" ]]; then
-            echo "::error::warp A/B budget gate failed (exit $status). See the step summary / gate.md artifact." >&2
-            exit "$status"
-          fi
-          echo "warp A/B budget gate passed."
-
-  alert-on-failure:
-    name: Alert on scheduled failure
-    needs: [warp-ab]
-    # `always()` is required: without it this job is skipped when a needed
-    # job fails. Alerts only for scheduled (nightly) runs (backlog#1149
-    # ci-8); manual dispatch failures are already watched by a human.
-    # `cancelled` is included alongside `failure` on purpose: a job that hits
-    # timeout-minutes ends as `cancelled`, and the 2026-07-11..07-14 nightly
-    # timeouts went silent precisely because the guard was failure-only. The
-    # composite action already reports cancelled/timed-out jobs in the issue
-    # body.
-    if: >-
-      github.repository == 'rustfs/rustfs' && (
-        always() && github.event_name == 'schedule' &&
-        (contains(needs.*.result, 'failure') || contains(needs.*.result, 'cancelled'))
-      )
-    runs-on: ubuntu-latest
-    timeout-minutes: 10
-    permissions:
-      contents: read
-      issues: write
-    steps:
-      - uses: actions/checkout@f548e57e544e1ff5a4c46bf1e1b8685f8e4a348a # v7
-        with:
-          persist-credentials: false
-      - name: Open or update failure-tracking issue
-        uses: ./.github/actions/schedule-failure-issue
-        with:
-          github-token: ${{ secrets.BACKLOG_ISSUE_TOKEN }}

.github/workflows/schedule-failure-alert-drill.yml

@@ -16,8 +16,8 @@
 #
 # Forces a job failure and then runs .github/actions/schedule-failure-issue
 # through the exact `needs` + `always() && contains(needs.*.result,
-# 'failure')` wiring used by the real consumers (e2e-s3tests, mint, fuzz,
-# performance-ab). Dispatch it twice to verify both the issue-creation and
+# 'failure')` wiring used by the real consumers (e2e-s3tests, mint, fuzz).
+# Dispatch it twice to verify both the issue-creation and
 # the dedupe-comment paths, then close the resulting
 # "[scheduled-failure] Schedule Failure Alert Drill" issue in rustfs/backlog
 # (the default target repository). Requires the BACKLOG_ISSUE_TOKEN secret

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

@@ -28,7 +28,6 @@ on:
       - "mint"
       - "minio-interop"
       - "Nightly GNU Build"
-      - "Performance A/B"
       - "Runner Hygiene"
     types: [completed]
 

.github/workflows/windows-filesystem.yml

@@ -103,6 +103,10 @@ jobs:
         shell: pwsh
         run: cargo test -p rustfs-ecstore --lib rename_all_ -- --nocapture
 
+      - name: Test multipart parent durability barriers
+        shell: pwsh
+        run: cargo test -p rustfs-ecstore --lib multipart_owned_record_parent_barriers_ -- --nocapture
+
       - name: Test Windows path conversion
         shell: pwsh
         run: "cargo test -p rustfs-utils --no-default-features --features path --lib dunce:: -- --nocapture"

ARCHITECTURE.md

@@ -155,7 +155,7 @@ default build (lifecycle:
    - ⚠️ PARTIALLY VIOLATED: serving-side `s3s` references remain in ecstore
      (bucket metadata/replication/lifecycle DTOs and error mapping). The
      count is ratcheted shrink-only by `scripts/check_s3s_footprint.sh`
-     (`S3S_ECSTORE_FILES_BASELINE`; the `object_lock` module was converted to
+     (the `ecstore_files` line of `.config/s3s-footprint-baseline.txt`; the `object_lock` module was converted to
      storage-level types as the first ratchet step). Target state: the
      baseline reaches zero and ecstore's `Cargo.toml` drops `s3s`.
 

Cargo.lock

@@ -365,9 +365,9 @@ checksum = "d3fb67a6e08acf24fdeccbac2cb6ac4305825bd1f117462e0e6f2f193345ad56"
 
 [[package]]
 name = "arrow"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4c25317ebddb8f2e6f006759336c95e5ded847a446f3524c27853bfe7d77a69b"
+checksum = "3f04e0a89cc175eaf3740282537851e6fd9d7276f97730b9d0ea2859c11a85df"
 dependencies = [
  "arrow-arith",
  "arrow-array",
@@ -386,9 +386,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-arith"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2ab555c0ffe9cb817041c0cee80baa9663a22225d0a035693579771f2693cadc"
+checksum = "157943c720919e8f5b6731407ce5cd4c9a5ef47a75bf10edaf24b7081b5763ef"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -400,9 +400,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-array"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2dae1d91737421e92c6dd7c7f9d0f2d25198fd6fc539738b628f4dad10adf533"
+checksum = "72b5b1682924f11ecf6929689c16cbc69efbb6dd141b06ad98ee1cb10d858c43"
 dependencies = [
  "ahash",
  "arrow-buffer",
@@ -419,9 +419,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-buffer"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1fe9cd0064f2d7cf8ecfaf77cd70e67a980bc71aea92b63102bd37a7d3379a09"
+checksum = "b5d687afc6c927965a8a759f3b68de91933d9dfa2e1bb6e156bac84b85665e92"
 dependencies = [
  "bytes",
  "half",
@@ -431,9 +431,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-cast"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ee510bc252781d700400979a96bd1eadebeae23901132bbabd92118da1b376c5"
+checksum = "3f23c3379feede42135ef27ca60e67f6e643cd4343ae74e4c8b16c83f55afd85"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -453,9 +453,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-csv"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ace83d5935421b950fea3a0a3f9972c0837c297ef98ade57127a5661b681112c"
+checksum = "85b1984628b091b8319b3c305fce21d04f9e9b2578bb733b0d0108462ae1d698"
 dependencies = [
  "arrow-array",
  "arrow-cast",
@@ -468,9 +468,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-data"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "54f128c73bb7fe9f7137218f910edd5b436f390e0f040b1956f5028390787897"
+checksum = "bb0a031c7ff5e2bdb443117054fb466af1d4df5b7f0ecbe197713dc8b113b1ba"
 dependencies = [
  "arrow-buffer",
  "arrow-schema",
@@ -481,9 +481,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-ipc"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8e244f39b82fa7a823e8b5984a0bd988234f9c92c1df79378fb85090e6e4c821"
+checksum = "f5dd1afadda2eb821aaf851bb4c0adef60c30cc0f1a29bd02fb99dd07fb9f58d"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -497,9 +497,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-json"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "727b21ebf8894dcd9504d6ab66bffdac338391e1f85cde594526a0210299ea40"
+checksum = "e2bc4d56a0dad84cc51c5812ccff48b642047a1f3be80349cd84318983e79edb"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -522,9 +522,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-ord"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "70d376b9b883239ae0f9963cdeeb59126ccaac08d60adf208a6ad3e2dd4048c3"
+checksum = "561b8209faa2cd489764b7fd07d7cb9e8d274d78048c38de59f77c41140cb834"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -535,9 +535,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-row"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "48096811f92b5c3a0c59d33cfecd7d7d2ed037d81e54a27a42d232e7700e6de2"
+checksum = "0d8c68b87a569f92b90c46cb0a09d86bd9609ddc9b1bf1fc5a503c8c6907c14e"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -548,9 +548,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-schema"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "37b6c8a6b12cf3869d299ce36ed8e1cd6fe9836264b68fb6c6538473a36124a1"
+checksum = "e87acb0c6bf6237fca7a50d7b51e801c98a18a5fe6f3bb34f7d77b85ca165388"
 dependencies = [
  "serde_core",
  "serde_json",
@@ -558,9 +558,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-select"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e523e85fa9fb3181a391bf0edfa6a1ae0dc46df8cbf7f6b96034dcb1e2a7a04a"
+checksum = "c844141958d805b47d1208e0d12aac53fcf388ce8011639fe82aa492adde0bb4"
 dependencies = [
  "ahash",
  "arrow-array",
@@ -572,9 +572,9 @@ dependencies = [
 
 [[package]]
 name = "arrow-string"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b931a905a0e4b8832d9f54fb369cf6e545f9dc03cb3d9611216a7bd2653df60a"
+checksum = "bc130892b1a9552da77971924b390e06cf375d1f7e4344a7b7b40ca4426fa470"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -1514,6 +1514,23 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "bindgen"
+version = "0.73.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "787ef8ef523575546b106a58213d6e6b06198a05c2f757258c68a74273670cfa"
+dependencies = [
+ "bitflags 2.13.2",
+ "cexpr",
+ "clang-sys",
+ "proc-macro2",
+ "quote",
+ "regex",
+ "rustc-hash",
+ "shlex",
+ "syn 3.0.9",
+]
+
 [[package]]
 name = "binrw"
 version = "0.15.2"
@@ -1726,13 +1743,13 @@ dependencies = [
 
 [[package]]
 name = "brotli"
-version = "9.0.0"
+version = "9.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f8b851b75c23ca7873623d612fe49bd1989aeb03d08fb9432187eb253d3d4c6b"
+checksum = "6737ce8b81460bf9407909f73ff50bf13f7cb8577e1e76715f57ed0eba58ba44"
 dependencies = [
  "alloc-no-stdlib 3.0.0",
  "alloc-stdlib 0.3.0",
- "brotli-decompressor 6.0.1",
+ "brotli-decompressor 6.2.0",
 ]
 
 [[package]]
@@ -1747,9 +1764,9 @@ dependencies = [
 
 [[package]]
 name = "brotli-decompressor"
-version = "6.0.1"
+version = "6.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "941cd9bd4ddab83cb46fa5a2d428f1c857b24ac78cb876cf7beb710840934bd7"
+checksum = "34c6783f42c21c952ad7112b354f0ba2f10760e9decea5b1b60523d6e270faaf"
 dependencies = [
  "alloc-no-stdlib 3.0.0",
  "alloc-stdlib 0.3.0",
@@ -1967,6 +1984,15 @@ dependencies = [
  "shlex",
 ]
 
+[[package]]
+name = "cexpr"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6fac387a98bb7c37292057cffc56d62ecb629900026402633ae9160df93a8766"
+dependencies = [
+ "nom 7.1.3",
+]
+
 [[package]]
 name = "cfg-if"
 version = "1.0.5"
@@ -2083,6 +2109,17 @@ dependencies = [
  "zeroize",
 ]
 
+[[package]]
+name = "clang-sys"
+version = "1.9.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "157a8ba7b480713b56f4c09fd13fc3e0a22a5dfab8097ba61cbc5feef950788a"
+dependencies = [
+ "glob",
+ "libc",
+ "libloading",
+]
+
 [[package]]
 name = "clap"
 version = "4.6.7"
@@ -2199,7 +2236,7 @@ version = "0.4.46"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "cae36683101b2899f48bd31f8db9839437356188e3b1ff23096e73dab09ba8fe"
 dependencies = [
- "brotli 9.0.0",
+ "brotli 9.1.0",
  "bzip2",
  "compression-core",
  "flate2",
@@ -2207,7 +2244,7 @@ dependencies = [
  "lz4",
  "memchr",
  "zstd 0.14.3",
- "zstd-safe 8.2.0",
+ "zstd-safe",
 ]
 
 [[package]]
@@ -2542,9 +2579,9 @@ checksum = "460fbee9c2c2f33933d720630a6a0bac33ba7053db5344fac858d4b8952d77d5"
 
 [[package]]
 name = "crypto-bigint"
-version = "0.7.7"
+version = "0.7.10"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "cc907c45bb7e9ef8fc3778defd0f3cd8c81288f1b4e748eba064948f20c47122"
+checksum = "9c035cb64cba03dad2fb921f3e218785e8cf9a41790da5eb3a1400f62fa2be11"
 dependencies = [
  "cpubits",
  "ctutils",
@@ -2620,9 +2657,9 @@ dependencies = [
 
 [[package]]
 name = "ctutils"
-version = "0.4.5"
+version = "0.4.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "38d28fb6aa7dd680b3ffd1a3e64e115c177bdba63c0ae7f9cd48129ff05c4fa0"
+checksum = "fe3e9d21d39e6f39ff46e03d5b621424f88ee23d8a1672c6bfc9ccc7f2da4f6a"
 dependencies = [
  "cmov",
  "subtle",
@@ -3632,9 +3669,9 @@ dependencies = [
 
 [[package]]
 name = "der"
-version = "0.8.2"
+version = "0.8.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a878c850e9e421b20262e9b41f9c860e4785fa07541c266b62ff9d1ef998a80a"
+checksum = "6a996a108feb8b427a6e12fcf6af4aa83409b79d72c15337d7741368131ee004"
 dependencies = [
  "const-oid 0.10.2",
  "pem-rfc7468 1.0.0",
@@ -4040,6 +4077,7 @@ dependencies = [
  "rustfs-utils",
  "rustls",
  "s3s",
+ "same-file",
  "serde",
  "serde_json",
  "serde_urlencoded",
@@ -4066,7 +4104,7 @@ version = "0.17.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c0681a4fc24c767085329728d8dfba959af91228aa4610cca4f8ce317ba46ae0"
 dependencies = [
- "der 0.8.2",
+ "der 0.8.3",
  "digest 0.11.3",
  "elliptic-curve",
  "rfc6979",
@@ -4131,9 +4169,9 @@ checksum = "0e9c71c2167ca323c882b99918929403426e2373ea17242ff5653e0d5e1058be"
 
 [[package]]
 name = "elliptic-curve"
-version = "0.14.2"
+version = "0.14.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0ad4496221fcfc21015fd8c0e26fcfab00504832dc0604f52acdd7bad2a8dc81"
+checksum = "900f37e7e18d74a2347389af6cd69e3a30d5cc9e697f40004865679aa75e9c7c"
 dependencies = [
  "base16ct",
  "crypto-bigint",
@@ -5235,9 +5273,9 @@ dependencies = [
 
 [[package]]
 name = "hotpath"
-version = "0.29.1"
+version = "0.29.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "62f2351902fa259b758c7303505273393f6b08172ab5bdd5739a2d2bc6d554e8"
+checksum = "5b4af2b5ecf2650278fe33b0e49b68c8c484d2306475c127628316dea740cf5c"
 dependencies = [
  "arc-swap",
  "async-channel",
@@ -5270,15 +5308,15 @@ dependencies = [
 
 [[package]]
 name = "hotpath-drain"
-version = "0.29.1"
+version = "0.29.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "83dab2340134d9976d62f9f5172f9c6584c7f521b092f258e0e58e412e160b53"
+checksum = "a5cceb6c15fbaaeb00b51acc7be7fad533d0b56cb422b1cfc16cf9ca1624d33a"
 
 [[package]]
 name = "hotpath-macros"
-version = "0.29.1"
+version = "0.29.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a04bbe5701b097f224a08187a78b75bb9d9b4bbaff6a279c2a50ba73edbb7aac"
+checksum = "08c0a42eb299ebb7ec6b9161489e5bf132603a34ca12cecadb8490937aa0492a"
 dependencies = [
  "proc-macro2",
  "quote",
@@ -6398,7 +6436,7 @@ dependencies = [
  "hashbrown 0.16.1",
  "indexmap 2.14.2",
  "metrics",
- "ordered-float 5.5.0",
+ "ordered-float 5.5.1",
  "quanta",
  "radix_trie",
  "rand 0.9.5",
@@ -7332,9 +7370,9 @@ dependencies = [
 
 [[package]]
 name = "ordered-float"
-version = "5.5.0"
+version = "5.5.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8c7c9e0d9b23589f26070720bac724174bfec1083e82f7854cdd0267518343c0"
+checksum = "7f59006ce1e204c30956638535b74d317bca99477b9356dfaec54c37f6f600ab"
 dependencies = [
  "num-traits",
 ]
@@ -7451,9 +7489,9 @@ dependencies = [
 
 [[package]]
 name = "pageant"
-version = "0.2.4"
+version = "0.2.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f0786409b5aaa884655c9b4065681db0835db0cc185a6fb19fb17a5d38656866"
+checksum = "e0bc922d9aa8cfcf598de27b3c6122cb2eeda9d6fe71527a22ebcb7f633540ff"
 dependencies = [
  "base16ct",
  "byteorder",
@@ -7522,9 +7560,9 @@ dependencies = [
 
 [[package]]
 name = "parquet"
-version = "59.3.1"
+version = "59.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8b3d2ef0fb136e9341570f219ecb168902e76079c6d81ad555f0a4840b1e08a3"
+checksum = "8eb260d566f2fbdabe356d26aaa5d13403aac29b5b4e44463b8f9bd7b2b12680"
 dependencies = [
  "ahash",
  "arrow-array",
@@ -7809,7 +7847,7 @@ version = "0.8.0-rc.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "986d2e952779af96ea048f160fd9194e1751b4faea78bcf3ceb456efe008088e"
 dependencies = [
- "der 0.8.2",
+ "der 0.8.3",
  "spki 0.8.1",
 ]
 
@@ -7852,7 +7890,7 @@ dependencies = [
  "aes 0.9.3",
  "aes-gcm",
  "cbc 0.2.1",
- "der 0.8.2",
+ "der 0.8.3",
  "pbkdf2 0.13.0",
  "rand_core 0.10.1",
  "scrypt 0.12.0",
@@ -7876,7 +7914,7 @@ version = "0.11.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "451913da69c775a56034ea8d9003d27ee8948e12443eae7c038ba100a4f21cb7"
 dependencies = [
- "der 0.8.2",
+ "der 0.8.3",
  "pkcs5 0.8.1",
  "rand_core 0.10.1",
  "spki 0.8.1",
@@ -8507,6 +8545,15 @@ version = "1.2.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a1d01941d82fa2ab50be1e79e6714289dd7cde78eba4c074bc5a4374f650dfe0"
 
+[[package]]
+name = "quick-xml"
+version = "0.41.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e660451e55124f798a69a5af3f49ccfbefbd41910eefd25caf2393e1f3473ec1"
+dependencies = [
+ "memchr",
+]
+
 [[package]]
 name = "quick-xml"
 version = "0.42.0"
@@ -9138,9 +9185,9 @@ dependencies = [
 
 [[package]]
 name = "russh"
-version = "0.64.1"
+version = "0.65.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ba61e87b9ec9a39a59a6bbed4c0b8ff7fb07073405b24a767a1f2c8560ac8f58"
+checksum = "3bb77a4b259312b2abbc541bca1c13cb4e9b61c560a7fd894068c3b92e0c9dc3"
 dependencies = [
  "aes 0.9.3",
  "aws-lc-rs",
@@ -9155,7 +9202,7 @@ dependencies = [
  "curve25519-dalek 5.0.0",
  "data-encoding",
  "delegate",
- "der 0.8.2",
+ "der 0.8.3",
  "digest 0.11.3",
  "ecdsa",
  "ed25519-dalek 3.0.0",
@@ -9371,7 +9418,7 @@ dependencies = [
  "proptest",
  "prost 0.14.4",
  "pyroscope",
- "quick-xml",
+ "quick-xml 0.42.0",
  "quinn",
  "rand 0.10.3",
  "rcgen",
@@ -9390,6 +9437,7 @@ dependencies = [
  "rustfs-ecstore",
  "rustfs-extension-schema",
  "rustfs-filemeta",
+ "rustfs-gateway-types",
  "rustfs-heal",
  "rustfs-heal-contracts",
  "rustfs-iam",
@@ -9668,6 +9716,7 @@ dependencies = [
  "rustfs-data-usage",
  "rustfs-erasure-codec",
  "rustfs-filemeta",
+ "rustfs-gateway-types",
  "rustfs-heal-contracts",
  "rustfs-io-metrics",
  "rustfs-lifecycle",
@@ -9773,6 +9822,48 @@ dependencies = [
  "xxhash-rust",
 ]
 
+[[package]]
+name = "rustfs-gateway-stream"
+version = "0.2.0"
+source = "git+https://github.com/rustfs/gateway.git?rev=21283e874c2381bab611e9347acbc5b92e9e2b3c#21283e874c2381bab611e9347acbc5b92e9e2b3c"
+dependencies = [
+ "bitflags 2.13.2",
+ "bytes",
+ "http 1.5.0",
+ "http-body",
+]
+
+[[package]]
+name = "rustfs-gateway-types"
+version = "0.32.0+aws.2026-10-02"
+source = "git+https://github.com/rustfs/gateway.git?rev=21283e874c2381bab611e9347acbc5b92e9e2b3c#21283e874c2381bab611e9347acbc5b92e9e2b3c"
+dependencies = [
+ "bytes",
+ "crc-fast",
+ "futures-core",
+ "hex",
+ "http 1.5.0",
+ "md-5 0.11.0",
+ "percent-encoding",
+ "rustfs-gateway-stream",
+ "rustfs-gateway-xml",
+ "s3s",
+ "sha1 0.11.0",
+ "sha2 0.11.1",
+ "time",
+ "twox-hash",
+ "unicode-normalization",
+ "zeroize",
+]
+
+[[package]]
+name = "rustfs-gateway-xml"
+version = "0.4.0"
+source = "git+https://github.com/rustfs/gateway.git?rev=21283e874c2381bab611e9347acbc5b92e9e2b3c#21283e874c2381bab611e9347acbc5b92e9e2b3c"
+dependencies = [
+ "quick-xml 0.41.0",
+]
+
 [[package]]
 name = "rustfs-heal"
 version = "1.0.2"
@@ -10043,10 +10134,10 @@ dependencies = [
  "metrics-util",
  "proptest",
  "rustfs-config",
+ "rustfs-gateway-types",
  "rustfs-replication",
  "rustfs-scanner-metrics",
  "rustfs-storage-api",
- "s3s",
  "serial_test",
  "temp-env",
  "time",
@@ -10109,7 +10200,6 @@ dependencies = [
  "reqwest",
  "rmp-serde",
  "rustfs-signer",
- "s3s",
  "serde",
  "serde_json",
  "sysinfo",
@@ -10149,7 +10239,7 @@ dependencies = [
  "jiff",
  "metrics",
  "percent-encoding",
- "quick-xml",
+ "quick-xml 0.42.0",
  "rayon",
  "rustc-hash",
  "rustfs-config",
@@ -10330,10 +10420,13 @@ dependencies = [
  "rustfs-config",
  "rustfs-credentials",
  "rustfs-ecstore",
+ "rustfs-gateway-stream",
+ "rustfs-gateway-types",
  "rustfs-iam",
  "rustfs-keystone",
  "rustfs-policy",
  "rustfs-rio",
+ "rustfs-s3-types",
  "rustfs-storage-api",
  "rustfs-test-utils",
  "rustfs-tls-runtime",
@@ -10341,13 +10434,13 @@ dependencies = [
  "rustfs-trusted-proxies",
  "rustfs-utils",
  "rustls",
- "s3s",
  "serde",
  "serde_json",
  "sha1 0.11.0",
  "sha2 0.11.1",
  "socket2",
  "subtle",
+ "sync_wrapper",
  "tempfile",
  "thiserror 2.0.21",
  "time",
@@ -10397,7 +10490,7 @@ dependencies = [
  "regex",
  "rmp",
  "rmp-serde",
- "s3s",
+ "rustfs-gateway-types",
  "serde",
  "time",
  "tokio",
@@ -10492,11 +10585,13 @@ dependencies = [
  "hyper-util",
  "lazy_static",
  "md-5 0.11.0",
- "quick-xml",
+ "quick-xml 0.42.0",
  "rand 0.10.3",
  "rustfs-checksums",
  "rustfs-config",
+ "rustfs-gateway-types",
  "rustfs-rio",
+ "rustfs-s3-types",
  "rustfs-signer",
  "rustfs-storage-api",
  "rustfs-tls-runtime",
@@ -10530,9 +10625,19 @@ dependencies = [
 name = "rustfs-s3-types"
 version = "1.0.2"
 dependencies = [
+ "bytes",
+ "futures-core",
  "hotpath",
+ "http 1.5.0",
+ "http-body",
+ "http-body-util",
+ "hyper",
+ "hyper-util",
+ "pin-project-lite",
+ "s3s",
  "serde",
  "serde_json",
+ "tokio",
 ]
 
 [[package]]
@@ -10556,9 +10661,10 @@ dependencies = [
  "parking_lot 0.13.0",
  "rustfs-common",
  "rustfs-ecstore",
+ "rustfs-gateway-types",
+ "rustfs-s3-types",
  "rustfs-storage-api",
  "rustfs-test-utils",
- "s3s",
  "serde_json",
  "serial_test",
  "thiserror 2.0.21",
@@ -10584,7 +10690,6 @@ dependencies = [
  "parking_lot 0.13.0",
  "rustfs-s3select-api",
  "rustfs-test-utils",
- "s3s",
  "serde",
  "serde_json",
  "tokio",
@@ -10613,6 +10718,7 @@ dependencies = [
  "rustfs-data-usage",
  "rustfs-ecstore",
  "rustfs-filemeta",
+ "rustfs-gateway-types",
  "rustfs-heal",
  "rustfs-heal-contracts",
  "rustfs-lifecycle",
@@ -10669,9 +10775,10 @@ dependencies = [
  "bytes",
  "hotpath",
  "http 1.5.0",
+ "http-body",
  "hyper",
+ "rustfs-s3-types",
  "rustfs-utils",
- "s3s",
  "serde_urlencoded",
  "thiserror 2.0.21",
  "time",
@@ -10729,7 +10836,6 @@ dependencies = [
  "rustfs-utils",
  "rustls",
  "rustls-native-certs",
- "s3s",
  "serde",
  "serde_json",
  "sha2 0.11.1",
@@ -10846,7 +10952,7 @@ version = "1.0.2"
 dependencies = [
  "base64-simd",
  "blake2",
- "brotli 9.0.0",
+ "brotli 9.1.0",
  "bytes",
  "convert_case 0.12.0",
  "crc-fast",
@@ -11121,7 +11227,7 @@ dependencies = [
  "numeric_cast",
  "phf 0.14.0",
  "pin-project-lite",
- "quick-xml",
+ "quick-xml 0.42.0",
  "regex",
  "s3s-multipart",
  "s3s-rfc2047",
@@ -11307,7 +11413,7 @@ checksum = "d56d437c2f19203ce5f7122e507831de96f3d2d4d3be5af44a0b0a09d8a80e4d"
 dependencies = [
  "base16ct",
  "ctutils",
- "der 0.8.2",
+ "der 0.8.3",
  "hybrid-array",
  "subtle",
  "zeroize",
@@ -11473,6 +11579,15 @@ dependencies = [
  "syn 3.0.9",
 ]
 
+[[package]]
+name = "serde_spanned"
+version = "1.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4e7523beb55eece201a2356bee0bbca0d1ab466c14c07703b2e0ee6d42cb0c2c"
+dependencies = [
+ "serde_core",
+]
+
 [[package]]
 name = "serde_urlencoded"
 version = "0.7.1"
@@ -11752,9 +11867,9 @@ checksum = "33f4fe9184a62d842c9ef383018f3306d8ba224fd9d836f56d7288308847c256"
 
 [[package]]
 name = "sketches-ddsketch"
-version = "0.3.1"
+version = "0.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0c6f73aeb92d671e0cc4dca167e59b2deb6387c375391bc99ee743f326994a2b"
+checksum = "45f82fa313d625dd72aaa65ec0c9f8c95407add0510c1bef3c18ea7257ec34ce"
 
 [[package]]
 name = "slab"
@@ -11883,7 +11998,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7ef958a98b9d5da290cfc78946e9f3e61e1e62a18db0d92cac0b83cc161491a9"
 dependencies = [
  "base64ct",
- "der 0.8.2",
+ "der 0.8.3",
 ]
 
 [[package]]
@@ -12632,6 +12747,30 @@ dependencies = [
  "webpki-roots 0.26.11",
 ]
 
+[[package]]
+name = "toml"
+version = "0.9.12+spec-1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cf92845e79fc2e2def6a5d828f0801e29a2f8acc037becc5ab08595c7d5e9863"
+dependencies = [
+ "indexmap 2.14.2",
+ "serde_core",
+ "serde_spanned",
+ "toml_datetime 0.7.5+spec-1.1.0",
+ "toml_parser",
+ "toml_writer",
+ "winnow 0.7.15",
+]
+
+[[package]]
+name = "toml_datetime"
+version = "0.7.5+spec-1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "92e1cfed4a3038bc5a127e35a2d360f145e1f4b971b551a2ba5fd7aedf7e1347"
+dependencies = [
+ "serde_core",
+]
+
 [[package]]
 name = "toml_datetime"
 version = "1.1.2+spec-1.1.0"
@@ -12648,9 +12787,9 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e3641d5bbb5349a79e1020a242d251efbc546ad8048d133958323ce9c40a9c9c"
 dependencies = [
  "indexmap 2.14.2",
- "toml_datetime",
+ "toml_datetime 1.1.2+spec-1.1.0",
  "toml_parser",
- "winnow",
+ "winnow 1.0.4",
 ]
 
 [[package]]
@@ -12659,9 +12798,15 @@ version = "1.1.5+spec-1.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "baa693a8032d7e1cada7d0041e96126df243179ff061456783ac7f12bda4744c"
 dependencies = [
- "winnow",
+ "winnow 1.0.4",
 ]
 
+[[package]]
+name = "toml_writer"
+version = "1.1.3+spec-1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "06bdbd8cfc056b8d2e2e85f29b56a3bdbecb527cef81eb39e3e7b98af4652770"
+
 [[package]]
 name = "tonic"
 version = "0.14.6"
@@ -13688,6 +13833,12 @@ version = "0.53.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d6bbff5f0aada427a1e5a6da5f1f98158182f26556f345ac9e04d36d0ebed650"
 
+[[package]]
+name = "winnow"
+version = "0.7.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "df79d97927682d2fd8adb29682d1140b343be4ac0f08fd68b7765d9c059d3945"
+
 [[package]]
 name = "winnow"
 version = "1.0.4"
@@ -13954,9 +14105,9 @@ dependencies = [
 
 [[package]]
 name = "zip"
-version = "9.1.0"
+version = "9.1.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2886b00003a72c4ccee3f42cfc34908d33b2508a74d827dc2a4244706475c704"
+checksum = "ec2e3161de27091f18a831d9494e9c53378a34833af5c1dda35ad85849da18d6"
 dependencies = [
  "aes 0.9.3",
  "bzip2",
@@ -14008,10 +14159,18 @@ dependencies = [
 [[package]]
 name = "zstd"
 version = "0.13.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e91ee311a569c327171651566e07972200e76fcfe2242a4fa446149a3881c08a"
 dependencies = [
- "zstd-safe 7.3.0",
+ "apache-avro",
+ "arrow-array",
+ "arrow-ipc",
+ "arrow-schema",
+ "bytes",
+ "kafka-protocol",
+ "parquet",
+ "semver",
+ "toml",
+ "zstd 0.14.3",
+ "zstd-safe",
 ]
 
 [[package]]
@@ -14020,16 +14179,7 @@ version = "0.14.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "db3ae9bf90af6e40197bef767eeefc46ad6397420fe873830bd85e5fbbfb3c70"
 dependencies = [
- "zstd-safe 8.2.0",
-]
-
-[[package]]
-name = "zstd-safe"
-version = "7.3.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "64d80649ab6db9d9f6f9c80a40becd948eda4714a0a5ac8c4d157a32231c7882"
-dependencies = [
- "zstd-sys",
+ "zstd-safe",
 ]
 
 [[package]]
@@ -14047,6 +14197,7 @@ version = "2.1.2+zstd.1.5.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "481ada9c543b2323cbb7bbdbb5249177fbf0188ed959ed2c15e4b22421c84ed3"
 dependencies = [
+ "bindgen",
  "cc",
  "pkg-config",
 ]

Cargo.toml

@@ -48,7 +48,7 @@ members = [
     "crates/replication", # Replication contracts and wire formats
     "crates/concurrency", # Concurrency management for RustFS - timeout, locking, backpressure, and I/O scheduling
     "crates/s3-client", # S3 client for engine-side consumption of remote S3 endpoints (tiering, transition targets)
-    "crates/s3-types", # S3 event type definitions
+    "crates/s3-types", # S3 event and error type definitions
     "crates/s3-ops", # S3 operation definitions and mapping
     "crates/s3select-api", # S3 Select API interface
     "crates/s3select-query", # S3 Select query engine
@@ -66,6 +66,7 @@ members = [
     "crates/io-metrics", # I/O metrics collection for performance analysis
     "crates/io-core", # Buffered Bytes, mmap-then-copy, and aligned pread I/O helpers
     "crates/zip", # ZIP file handling and compression
+    "crates/zstd-compat", # Fixed codec for dependencies requiring zstd 0.13
 ]
 resolver = "3"
 
@@ -142,6 +143,13 @@ rustfs-tls-runtime = { path = "crates/tls-runtime", version = "1.0.2" }
 rustfs-utils = { path = "crates/utils", version = "1.0.2" }
 rustfs-zip = { path = "./crates/zip", version = "1.0.2" }
 
+# The gateway's S3 types: the DTOs and persisted bucket-configuration codecs the
+# engine crates move onto from s3s (rustfs/backlog#2745). The one pin of the
+# gateway in this workspace; no crate may enable a `compat-s3s*` feature except
+# the s3s-facing edge of the rustfs binary.
+rustfs-gateway-types = { git = "https://github.com/rustfs/gateway.git", rev = "21283e874c2381bab611e9347acbc5b92e9e2b3c", version = "0.32.0", default-features = false }
+rustfs-gateway-stream = { git = "https://github.com/rustfs/gateway.git", rev = "21283e874c2381bab611e9347acbc5b92e9e2b3c", version = "0.2.0", default-features = false }
+
 # Async Runtime and Networking
 async-channel = "2.5.0"
 async_zip = { default-features = false, version = "0.0.19" }
@@ -260,7 +268,7 @@ aws-smithy-http-client = { default-features = false, version = "1.5.1" }
 aws-smithy-runtime-api = { version = "1.19.1" }
 aws-smithy-types = { version = "1.9.0" }
 base64-simd = "0.8.0"
-brotli = "9.0.0"
+brotli = "9.1.0"
 clap = { version = "4.6.7" }
 const-str = { version = "1.1.0" }
 convert_case = "0.12.0"
@@ -302,6 +310,7 @@ num_cpus = { version = "1.17.0" }
 nvml-wrapper = "0.13.0"
 object = { version = "=0.40.0", default-features = false }
 parking_lot = "0.13.0"
+same-file = "1.0.6"
 path-absolutize = "4.0.1"
 percent-encoding = "2.3.2"
 # Server-side QR rendering for TOTP enrollment, so neither the console nor the
@@ -324,6 +333,10 @@ rust-embed = { version = "8.13.0" }
 rustc-hash = { version = "2.1.3" }
 s3s = { git = "https://github.com/s3s-project/s3s", rev = "5761ddfe4c509fcac87e3b99fd5bd9e176b442c0", features = ["minio"] }
 s3s-sigv4 = { git = "https://github.com/s3s-project/s3s", rev = "5761ddfe4c509fcac87e3b99fd5bd9e176b442c0" }
+# The same s3s pin under the name the legacy edge of the gateway migration uses
+# (crates/s3-types, feature `compat-s3s`). Keep `rev` identical to `s3s` above.
+# DELETE BY T4.2 (rustfs/backlog#2784).
+legacy-s3s = { package = "s3s", git = "https://github.com/s3s-project/s3s", rev = "5761ddfe4c509fcac87e3b99fd5bd9e176b442c0", features = ["minio"] }
 serial_test = "4.0.1"
 shadow-rs = { default-features = false, version = "2.0.0" }
 siphasher = "1.0.4"
@@ -332,6 +345,7 @@ compact_str = "0.10.0"
 snap = "1.1.2"
 starshard = { version = "2.4.0" }
 strum = { version = "0.28.0" }
+sync_wrapper = "1.0.2"
 sysinfo = "0.39.6"
 temp-env = "0.3.6"
 tempfile = "3.27.0"
@@ -354,7 +368,7 @@ winapi-util = "0.1.11"
 windows = { version = "0.62.2" }
 windows-sys = "0.61.2"
 xxhash-rust = { version = "0.8.19" }
-zip = "9.1.0"
+zip = "9.1.2"
 zstd = "0.14.3"
 
 # Observability and Metrics
@@ -375,7 +389,7 @@ libunftp = { version = "0.23.1" }
 unftp-core = "0.1.0"
 suppaftp = { version = "12.2.1" }
 rcgen = { version = "0.14.10", default-features = false, features = ["aws_lc_rs", "crypto", "pem"] }
-russh = { version = "0.64.1" }
+russh = { version = "0.65.0" }
 russh-sftp = "3.0.1"
 
 # WebDAV
@@ -384,7 +398,7 @@ dav-server = "0.11.0"
 # Performance Analysis and Memory Profiling
 rustfs-mimalloc = { version = "0.7.2" }
 # Preserve Unicode focus filters until rustfs/backlog#2302 is resolved.
-hotpath = { version = "0.29.1", default-features = false }
+hotpath = { version = "0.29.2", default-features = false }
 # Snapshot testing for output format regression detection
 insta = { version = "1.49" }
 
@@ -424,6 +438,7 @@ strip = "none"
 
 # Pin upstream fixes until compatible crates.io releases are available.
 [patch.crates-io]
+zstd = { path = "crates/zstd-compat" }
 # Preserve RustFS OIDC flows while migrating the old elliptic-curve dependency chain.
 # Upstream: openidconnect-rs#244 and oauth2-rs#352.
 openidconnect = { git = "https://github.com/ramosbugs/openidconnect-rs", rev = "2f77ee73bc4554e1cb6279adb386005429fb4e41" }

README.md

@@ -141,7 +141,7 @@ chown -R 10001:10001 data logs
 docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:latest
 
 # Using specific version
-docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.3
+docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.4
 ```
 
 If you use [podman](https://github.com/containers/podman) instead of docker, you can install the RustFS with the below command
@@ -220,7 +220,7 @@ For developers who want to build RustFS Docker images from source with multi-arc
 ./docker-buildx.sh --push
 
 # Build specific version
-./docker-buildx.sh --release 1.0.3 --push
+./docker-buildx.sh --release 1.0.4 --push
 
 # Build for custom registry
 ./docker-buildx.sh --registry your-registry.com --namespace yourname --push
@@ -238,7 +238,7 @@ You can also use Make targets for convenience:
 ```bash
 make docker-buildx                    # Build locally
 make docker-buildx-push               # Build and push
-make docker-buildx-version VERSION=1.0.3  # Build specific version
+make docker-buildx-version VERSION=1.0.4  # Build specific version
 make help-docker                      # Show all Docker-related commands
 ```
 

README_ZH.md

@@ -138,7 +138,7 @@ chown -R 10001:10001 data logs
 docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:latest
 
 # 使用指定版本运行
-docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.3
+docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.4
 ```
 
 如果您通过绑定挂载启用 TLS 证书目录，也请用同样方式准备该目录：
@@ -176,7 +176,7 @@ docker compose -f docker-compose-simple.yml up -d
 ./docker-buildx.sh --push
 
 # 构建指定版本
-./docker-buildx.sh --release 1.0.3 --push
+./docker-buildx.sh --release 1.0.4 --push
 
 # 构建并推送到自定义仓库
 ./docker-buildx.sh --registry your-registry.com --namespace yourname --push
@@ -193,7 +193,7 @@ docker compose -f docker-compose-simple.yml up -d
 ```bash
 make docker-buildx                    # 本地构建
 make docker-buildx-push               # 构建并推送
-make docker-buildx-version VERSION=1.0.3  # 构建指定版本
+make docker-buildx-version VERSION=1.0.4  # 构建指定版本
 make help-docker                      # 显示所有 Docker 相关命令
 ```
 

crates/config/src/constants/heal.rs

@@ -197,6 +197,10 @@ pub const ENV_HEAL_MRF_REPLAY_BATCH: &str = "RUSTFS_HEAL_MRF_REPLAY_BATCH";
 /// Default behavior keeps the MRF intent pipeline enabled.
 pub const DEFAULT_HEAL_MRF_ENABLE: bool = true;
 
+/// Opt in to lifecycle V2 transitions after qualifying every reader and the rollback binary.
+pub const ENV_MRF_LIFECYCLE_V2_ENABLE: &str = "RUSTFS_MRF_LIFECYCLE_V2_ENABLE";
+pub const DEFAULT_MRF_LIFECYCLE_V2_ENABLE: bool = false;
+
 /// Default MRF queue capacity (matches MinIO's 100k MRF list ceiling).
 pub const DEFAULT_HEAL_MRF_QUEUE_SIZE: usize = 100_000;
 

crates/e2e_test/Cargo.toml

@@ -146,6 +146,7 @@ zip.workspace = true
 clap = { workspace = true, features = ["derive", "env"] }
 
 [dev-dependencies]
+same-file.workspace = true
 aws-smithy-http-client = { workspace = true, default-features = false, features = ["test-util"] }
 rustfs-heal.workspace = true
 tempfile.workspace = true

crates/e2e_test/src/admin_auth_test.rs

@@ -29,6 +29,10 @@
 //! 3. Starting with the default `rustfsadmin` credentials emits the
 //!    default-credentials startup warning (observable operator signal).
 
+// Keep inventory acceptance in the admin-auth PR smoke selection.
+#[cfg(test)]
+mod target_inventory_test;
+
 #[cfg(test)]
 mod tests {
     use crate::common::{RustFSTestEnvironment, init_logging, rustfs_binary_path};

crates/e2e_test/src/admin_auth_test/target_inventory_test.rs

@@ -0,0 +1,205 @@
+// 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.
+
+//! Exercise the target inventory through HTTP authentication and authorization.
+//! Handler-level tests separately control reconciliation locks and config faults.
+
+use crate::common::{
+    RustFSTestEnvironment, admin_create_user, admin_request, admin_request_with_session_token, build_test_sts_client,
+    init_logging, local_http_client,
+};
+use http::header::{AUTHORIZATION, HOST};
+use reqwest::StatusCode;
+use rustfs_signer::{constants::UNSIGNED_PAYLOAD, sign_v4};
+use s3s::Body;
+use serde_json::Value;
+use std::{error::Error, time::Duration};
+
+type TestResult = Result<(), Box<dyn Error + Send + Sync>>;
+
+const SECRET: &str = "inventory-acceptance-token";
+const LIMITED_KEY: &str = "inventory-limited";
+const LIMITED_SECRET: &str = "inventory-limited-secret";
+const PATHS: [&str; 5] = [
+    "/rustfs/admin/v3/target/list",
+    "/rustfs/admin/v3/audit/target/list",
+    "/rustfs/admin/v4/plugins/instances",
+    "/rustfs/admin/v4/plugins/instances/builtin:webhook:notify:acceptance",
+    "/rustfs/admin/v4/extensions/instances",
+];
+
+async fn duplicate_authorization(
+    env: &RustFSTestEnvironment,
+    path: &str,
+) -> Result<reqwest::Response, Box<dyn Error + Send + Sync>> {
+    let url = format!("{}{path}", env.url);
+    let uri: http::Uri = url.parse()?;
+    let authority = uri.authority().ok_or("inventory URL missing authority")?.to_string();
+    let request = http::Request::builder()
+        .method(http::Method::GET)
+        .uri(uri)
+        .header(HOST, authority)
+        .header("x-amz-content-sha256", UNSIGNED_PAYLOAD)
+        .body(Body::empty())?;
+    let signed = sign_v4(request, 0, &env.access_key, &env.secret_key, "", "us-east-1");
+    let mut headers = signed.headers().clone();
+    let authorization = headers.get(AUTHORIZATION).ok_or("signer omitted authorization")?.clone();
+    headers.append(AUTHORIZATION, authorization);
+    assert_eq!(headers.get_all(AUTHORIZATION).iter().count(), 2);
+    Ok(local_http_client().get(url).headers(headers).send().await?)
+}
+
+fn assert_unloaded_target(entry: &Value, path: &str) {
+    assert_eq!(entry["service"], "webhook", "{path}: unexpected target service");
+    assert!(
+        entry["account_id"]
+            .as_str()
+            .is_some_and(|id| id.eq_ignore_ascii_case("acceptance")),
+        "{path}: fixture target is missing"
+    );
+    assert_eq!(entry["status"], "offline", "{path}: unloaded targets must remain visible as offline");
+    if !path.contains("/extensions/") {
+        assert_eq!(entry["health_reason"], "not_loaded_in_runtime", "{path}: unexpected health reason");
+    }
+}
+
+#[tokio::test(flavor = "multi_thread")]
+async fn target_inventory_preserves_unloaded_targets_and_http_auth_boundaries() -> TestResult {
+    init_logging();
+    let mut env = RustFSTestEnvironment::new().await?;
+    // Enabled destinations remain configured while both delivery modules are
+    // disabled, so inventory reads must not require a loaded target runtime.
+    env.start_rustfs_server_with_env(
+        vec![],
+        &[
+            ("RUSTFS_NOTIFY_ENABLE", "false"),
+            ("RUSTFS_AUDIT_ENABLE", "false"),
+            ("RUSTFS_NOTIFY_WEBHOOK_ENABLE_ACCEPTANCE", "on"),
+            ("RUSTFS_NOTIFY_WEBHOOK_ENDPOINT_ACCEPTANCE", "http://127.0.0.1:9/hook"),
+            ("RUSTFS_NOTIFY_WEBHOOK_AUTH_TOKEN_ACCEPTANCE", SECRET),
+            ("RUSTFS_AUDIT_WEBHOOK_ENABLE_ACCEPTANCE", "on"),
+            ("RUSTFS_AUDIT_WEBHOOK_ENDPOINT_ACCEPTANCE", "http://127.0.0.1:9/hook"),
+            ("RUSTFS_AUDIT_WEBHOOK_AUTH_TOKEN_ACCEPTANCE", SECRET),
+        ],
+    )
+    .await?;
+    admin_create_user(&env, LIMITED_KEY, LIMITED_SECRET).await?;
+
+    tokio::time::timeout(Duration::from_secs(60), async {
+        let assumed = build_test_sts_client(&env.url, &env.access_key, &env.secret_key, None, "e2e-target-inventory")
+            .assume_role()
+            .role_arn("arn:aws:iam::123456789012:role/target-inventory")
+            .role_session_name("target-inventory-console-session")
+            .send()
+            .await?;
+        let temporary = assumed
+            .credentials()
+            .ok_or("AssumeRole should return temporary inventory credentials")?;
+        assert_ne!(
+            temporary.access_key_id(),
+            env.access_key,
+            "inventory must exercise a distinct session credential"
+        );
+        assert!(
+            !temporary.session_token().is_empty(),
+            "temporary inventory credentials must include a token"
+        );
+
+        for path in PATHS {
+            for (principal, access_key, secret_key, session_token) in [
+                ("root", env.access_key.as_str(), env.secret_key.as_str(), None),
+                (
+                    "temporary admin",
+                    temporary.access_key_id(),
+                    temporary.secret_access_key(),
+                    Some(temporary.session_token()),
+                ),
+            ] {
+                let (status, body) = admin_request_with_session_token(
+                    &env.url,
+                    http::Method::GET,
+                    path,
+                    None,
+                    access_key,
+                    secret_key,
+                    session_token,
+                )
+                .await?;
+                assert_eq!(status, StatusCode::OK, "{path}: signed {principal} inventory failed");
+                assert!(!body.contains(SECRET), "{path}: inventory exposed a target secret");
+                let value: Value = serde_json::from_str(&body)?;
+                let entries: Vec<&Value> = if path.ends_with("/target/list") {
+                    let field = if path.contains("/audit/") {
+                        "audit_endpoints"
+                    } else {
+                        "notification_endpoints"
+                    };
+                    value[field].as_array().ok_or("missing endpoint inventory")?.iter().collect()
+                } else if path.ends_with("/instances") {
+                    value["instances"]
+                        .as_array()
+                        .ok_or("missing instance inventory")?
+                        .iter()
+                        .collect()
+                } else {
+                    vec![&value]
+                };
+                let targets: Vec<_> = entries
+                    .into_iter()
+                    .filter(|entry| {
+                        entry["account_id"]
+                            .as_str()
+                            .is_some_and(|id| id.eq_ignore_ascii_case("acceptance"))
+                    })
+                    .collect();
+                let expected = if path.ends_with("/instances") { 2 } else { 1 };
+                assert_eq!(targets.len(), expected, "{path}: configured targets disappeared");
+                for entry in &targets {
+                    assert_unloaded_target(entry, path);
+                    if path.contains("/v4/") {
+                        assert_eq!(entry["config"]["auth_token"], "***redacted***", "{path}: missing secret redaction");
+                    }
+                }
+                if expected == 2 {
+                    for domain in ["notify", "audit"] {
+                        assert!(targets.iter().any(|entry| entry["domain"] == domain), "{path}: missing {domain} domain");
+                    }
+                }
+            }
+
+            let unsigned = local_http_client().get(format!("{}{path}", env.url)).send().await?;
+            assert_eq!(unsigned.status(), StatusCode::FORBIDDEN, "{path}: unsigned request was not denied");
+            let body = unsigned.text().await?;
+            assert!(body.contains("<Code>AccessDenied</Code>") && body.contains("Signature is required"));
+
+            let ambiguous = duplicate_authorization(&env, path).await?;
+            assert_eq!(
+                ambiguous.status(),
+                StatusCode::FORBIDDEN,
+                "{path}: duplicate authorization was not rejected"
+            );
+            let body = ambiguous.text().await?;
+            assert!(body.contains("<Code>AccessDenied</Code>") && body.contains("Signature is required"));
+
+            let (status, body) = admin_request(&env.url, http::Method::GET, path, None, LIMITED_KEY, LIMITED_SECRET).await?;
+            assert_eq!(status, StatusCode::FORBIDDEN, "{path}: principal without admin policy was not denied");
+            assert!(body.contains("<Code>AccessDenied</Code>"), "{path}: unexpected authorization error");
+            assert!(!body.contains("Signature is required") && !body.contains("SignatureDoesNotMatch"));
+        }
+        Ok::<(), Box<dyn Error + Send + Sync>>(())
+    })
+    .await??;
+    env.stop_server();
+    Ok(())
+}

crates/e2e_test/src/anonymous_access_test.rs

@@ -17,7 +17,8 @@
 //! when PublicAccessBlock configuration is missing or explicitly set.
 
 use crate::common::{RustFSTestEnvironment, init_logging, local_http_client};
-use aws_sdk_s3::types::PublicAccessBlockConfiguration;
+use crate::sts_query_compat_test::{OPA_AUTH_TOKEN, OpaMock};
+use aws_sdk_s3::types::{PublicAccessBlockConfiguration, Tag, Tagging};
 use tracing::info;
 
 async fn setup_public_bucket(
@@ -69,6 +70,98 @@ async fn anonymous_get_object(
     local_http_client().get(&url).send().await
 }
 
+#[tokio::test]
+async fn test_plain_http_matches_secure_transport_false_bucket_policy() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+
+    let mut env = RustFSTestEnvironment::new().await?;
+    env.start_rustfs_server(vec![]).await?;
+    assert!(env.url.starts_with("http://"), "fixture must exercise plaintext HTTP: {}", env.url);
+
+    let bucket = "secure-transport-policy";
+    let key = "plain-http.txt";
+    let admin = env.create_s3_client();
+    admin.create_bucket().bucket(bucket).send().await?;
+    admin
+        .put_object()
+        .bucket(bucket)
+        .key(key)
+        .body(aws_sdk_s3::primitives::ByteStream::from_static(b"plain http"))
+        .send()
+        .await?;
+    admin
+        .put_bucket_policy()
+        .bucket(bucket)
+        .policy(
+            serde_json::json!({
+                "Version": "2012-10-17",
+                "Statement": [{
+                    "Effect": "Allow",
+                    "Principal": "*",
+                    "Action": "s3:GetObject",
+                    "Resource": format!("arn:aws:s3:::{bucket}/*"),
+                    "Condition": {"Bool": {"aws:SecureTransport": "false"}}
+                }]
+            })
+            .to_string(),
+        )
+        .send()
+        .await?;
+
+    let response = anonymous_get_object(&env, bucket, key).await?;
+    assert_eq!(response.status().as_u16(), 200);
+    assert_eq!(response.bytes().await?.as_ref(), b"plain http");
+    Ok(())
+}
+
+/// Root credentials have no attached IAM policy, so these requests depend on
+/// the owner flag reaching policy evaluation unchanged.
+#[tokio::test]
+async fn test_root_requests_keep_policy_owner_flag() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+
+    let opa = OpaMock::start().await?;
+    let mut env = RustFSTestEnvironment::new().await?;
+    env.start_rustfs_server_with_env(
+        vec![],
+        &[
+            ("RUSTFS_POLICY_PLUGIN_URL", opa.url.as_str()),
+            ("RUSTFS_POLICY_PLUGIN_AUTH_TOKEN", OPA_AUTH_TOKEN),
+        ],
+    )
+    .await?;
+
+    let bucket = "root-owner-flag";
+    let key = "owner.txt";
+    let client = env.create_s3_client();
+    client.create_bucket().bucket(bucket).send().await?;
+    client
+        .put_object()
+        .bucket(bucket)
+        .key(key)
+        .body(aws_sdk_s3::primitives::ByteStream::from_static(b"owner access"))
+        .send()
+        .await?;
+
+    // Duplicate stored tags are ambiguous. An owner request must not enter the
+    // OPA tag-loading path, which rejects this metadata before evaluation.
+    client
+        .put_bucket_tagging()
+        .bucket(bucket)
+        .tagging(
+            Tagging::builder()
+                .tag_set(Tag::builder().key("department").value("finance").build()?)
+                .tag_set(Tag::builder().key("department").value("engineering").build()?)
+                .build()?,
+        )
+        .send()
+        .await?;
+    let object = client.get_object().bucket(bucket).key(key).send().await?;
+
+    assert_eq!(object.body.collect().await?.into_bytes().as_ref(), b"owner access");
+    Ok(())
+}
+
 /// Issue #2036: Anonymous GetObject should succeed when bucket policy allows it
 /// and no PublicAccessBlock configuration exists (ConfigNotFound).
 #[tokio::test]

crates/e2e_test/src/degraded_listing_availability_test.rs

@@ -20,15 +20,14 @@
 //! offline before heal converges, a strict latest-listing quorum of 3 can only
 //! ever observe 2 copies, so ListObjectsV2 silently dropped the object even
 //! though GetObject (read quorum 2) still succeeded. Exposed by the flaky
-//! "Mixed-version rolling upgrade from rc.2" CI lane (run 33478999853); the
-//! product fix relaxes the listing's required object quorum by the number of
-//! set drives the listing could not consult (see
-//! `latest_listing_required_object_quorum` in
-//! `crates/ecstore/src/store/list_objects.rs`).
+//! "Mixed-version rolling upgrade from rc.2" CI lane (run 33478999853). The
+//! listing now follows MinIO's rule: an entry needs `ceil(asked/2)` copies
+//! among the online drives it asked, so three online drives need two copies
+//! (see `SetDisks::list_path` in `crates/ecstore/src/store/list_objects.rs`).
 
 #[cfg(test)]
 mod tests {
-    use crate::common::{RustFSTestClusterEnvironment, init_logging};
+    use crate::common::{ClusterTopology, RustFSTestClusterEnvironment, init_logging};
     use aws_sdk_s3::Client;
     use bytes::Bytes;
     use std::collections::HashSet;
@@ -51,13 +50,13 @@ mod tests {
         format!("degraded-object-{idx:02}")
     }
 
-    async fn list_all_keys(client: &Client) -> Result<HashSet<String>, Box<dyn Error + Send + Sync>> {
+    async fn list_all_keys(client: &Client, bucket: &str) -> Result<HashSet<String>, Box<dyn Error + Send + Sync>> {
         let mut keys = HashSet::new();
         let mut continuation_token: Option<String> = None;
         loop {
             let response = client
                 .list_objects_v2()
-                .bucket(BUCKET)
+                .bucket(bucket)
                 .set_continuation_token(continuation_token.clone())
                 .send()
                 .await?;
@@ -147,7 +146,7 @@ mod tests {
         let expected: HashSet<String> = (0..OBJECT_COUNT).map(object_key).collect();
         let deadline = Instant::now() + LISTING_DEADLINE;
         let listed = loop {
-            let listed = match list_all_keys(&client).await {
+            let listed = match list_all_keys(&client, BUCKET).await {
                 Ok(keys) => keys,
                 Err(error) if Instant::now() < deadline => {
                     info!("retrying degraded listing: {error}");
@@ -171,4 +170,97 @@ mod tests {
 
         Ok(())
     }
+
+    /// 4-node x 4-drive cluster (one 16-drive set), #8828 shape:
+    /// 1. PUT objects while the storage class is EC:8, so each is 8+8.
+    /// 2. Restart with the EC:4 default, as after a storage-class change.
+    /// 3. Stop two nodes (8 drives). Each object keeps eight data shards, so
+    ///    GET succeeds and ListObjectsV2 must list every key: the listing
+    ///    quorum is half of the eight drives it can ask.
+    /// 4. Stop a third node. Four online drives are below the set's read
+    ///    quorum, so LIST must report 503 instead of a partial page.
+    #[tokio::test]
+    async fn ec8_objects_stay_listable_with_half_of_an_ec4_set_offline() -> TestResult {
+        const EC8_BUCKET: &str = "degraded-listing-ec8";
+        init_logging();
+
+        let mut cluster = RustFSTestClusterEnvironment::with_topology(ClusterTopology::single_pool_multidrive(4, 4)).await?;
+        cluster.set_env("RUSTFS_HEAL_ENABLED", "false");
+        cluster.set_env("RUSTFS_SCANNER_ENABLED", "false");
+        cluster.set_env("RUSTFS_STORAGE_CLASS_STANDARD", "EC:8");
+        cluster.start().await?;
+        cluster.create_test_bucket(EC8_BUCKET).await?;
+        let client = cluster.create_s3_client(0)?;
+        for idx in 0..OBJECT_COUNT {
+            let key = object_key(idx);
+            client
+                .put_object()
+                .bucket(EC8_BUCKET)
+                .key(&key)
+                .body(Bytes::from(format!("ec8 listing payload {idx}")).into())
+                .send()
+                .await
+                .map_err(|error| format!("EC:8 PUT for {key} failed: {error}"))?;
+        }
+
+        info!("restarting with the EC:4 default storage class");
+        cluster.stop();
+        // A later entry for the same variable wins when the node process starts.
+        cluster.set_env("RUSTFS_STORAGE_CLASS_STANDARD", "EC:4");
+        cluster.start().await?;
+        let client = cluster.create_s3_client(0)?;
+
+        info!("stopping nodes 2 and 3 (eight of sixteen drives)");
+        cluster.stop_node(2)?;
+        cluster.stop_node(3)?;
+        for idx in 0..OBJECT_COUNT {
+            let key = object_key(idx);
+            let deadline = Instant::now() + GET_RETRY_DEADLINE;
+            loop {
+                match client.get_object().bucket(EC8_BUCKET).key(&key).send().await {
+                    Ok(response) => {
+                        let body = response.body.collect().await?.into_bytes();
+                        assert_eq!(body.as_ref(), format!("ec8 listing payload {idx}").as_bytes());
+                        break;
+                    }
+                    Err(error) if Instant::now() < deadline => {
+                        info!("retrying EC:8 GET for {key}: {error}");
+                        tokio::time::sleep(Duration::from_millis(500)).await;
+                    }
+                    Err(error) => return Err(format!("EC:8 object {key} failed GET with eight drives: {error}").into()),
+                }
+            }
+        }
+
+        let expected: HashSet<String> = (0..OBJECT_COUNT).map(object_key).collect();
+        let deadline = Instant::now() + LISTING_DEADLINE;
+        loop {
+            match list_all_keys(&client, EC8_BUCKET).await {
+                Ok(listed) if listed == expected => break,
+                Ok(listed) => assert!(
+                    Instant::now() < deadline,
+                    "EC:8 objects readable with eight drives stayed missing from ListObjectsV2: listed={listed:?}"
+                ),
+                Err(error) => assert!(
+                    Instant::now() < deadline,
+                    "ListObjectsV2 kept failing with eight of sixteen drives online: {error}"
+                ),
+            }
+            tokio::time::sleep(Duration::from_millis(500)).await;
+        }
+
+        info!("stopping node 1 (four of sixteen drives remain)");
+        cluster.stop_node(1)?;
+        let deadline = Instant::now() + LISTING_DEADLINE;
+        loop {
+            match client.list_objects_v2().bucket(EC8_BUCKET).send().await {
+                Err(error) if error.raw_response().is_some_and(|response| response.status().as_u16() == 503) => break,
+                Err(error) => assert!(Instant::now() < deadline, "below the set read quorum LIST must report 503, got {error}"),
+                Ok(response) => panic!("below the set read quorum LIST must fail, but listed {} keys", response.contents().len()),
+            }
+            tokio::time::sleep(Duration::from_millis(500)).await;
+        }
+
+        Ok(())
+    }
 }

crates/e2e_test/src/distributed/chaos_test.rs

@@ -13,18 +13,20 @@
 // 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,
+    DistCluster, DistLayout, TestResult, assert_object_bytes, cluster_admin_ok, get_object_bytes, payload_for, put_object,
+    retrying_get_equals, retrying_put, s3_probe_client, s3_probe_error_code, 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 aws_sdk_s3::error::SdkError;
+use aws_sdk_s3::operation::{get_object::GetObjectError, put_object::PutObjectError};
 use std::path::PathBuf;
 use std::sync::Arc;
 use std::time::Duration;
 use tokio::sync::{Barrier, mpsc};
-use tokio::time::timeout;
+use tokio::time::{Instant, timeout};
 
 #[tokio::test]
 async fn kill_and_restart_node_preserves_objects() -> TestResult {
@@ -44,6 +46,133 @@ async fn kill_and_restart_node_preserves_objects() -> TestResult {
     Ok(())
 }
 
+#[tokio::test]
+async fn ec4_partial_node_return_restores_read_write_quorum() -> TestResult {
+    init_logging();
+    let mut dist = DistCluster::start_with_env(
+        DistLayout::FourByFour,
+        &[
+            ("RUSTFS_ERASURE_SET_DRIVE_COUNT", "16"),
+            ("RUSTFS_STORAGE_CLASS_STANDARD", "EC:4"),
+        ],
+    )
+    .await?;
+    let storage = cluster_admin_ok(&dist.cluster, http::Method::GET, "/rustfs/admin/v3/storageinfo", None).await?;
+    let storage: serde_json::Value = serde_json::from_str(&storage)?;
+    let storage: rustfs_madmin::StorageInfo = serde_json::from_value(storage["info"].clone())?;
+    assert_eq!(storage.disks.len(), 16);
+    assert_eq!(storage.backend.total_sets, vec![1]);
+    assert_eq!(storage.backend.drives_per_set, vec![16]);
+    assert_eq!(storage.backend.standard_sc_parity, Some(4));
+
+    let bucket = unique_bucket("partial-ec4");
+    dist.create_bucket(&bucket).await?;
+    let client = s3_probe_client(&dist.client(0)?);
+    let baseline = payload_for("partial-recovery/baseline", 8 * 1024 * 1024);
+    put_object(&client, &bucket, "baseline.bin", baseline.clone()).await?;
+    wait_until(
+        Duration::from_secs(30),
+        || async {
+            for node in &dist.cluster.nodes {
+                for directory in &node.data_dirs {
+                    let census = census_object_version_on_disk(std::path::Path::new(directory), &bucket, "baseline.bin", None)?;
+                    if !census.is_complete() {
+                        return Ok(false);
+                    }
+                    assert_eq!(census.data_blocks, Some(12), "the object must require twelve data shards");
+                    assert_eq!(census.parity_blocks, Some(4), "the object must have four parity shards");
+                    assert!(!census.expected_part_numbers.is_empty(), "the baseline must use physical shards");
+                }
+            }
+            Ok(true)
+        },
+        "the baseline is complete on all sixteen EC:4 drives",
+    )
+    .await?;
+
+    dist.cluster.stop_node(2)?;
+    dist.cluster.stop_node(3)?;
+    let read_error = timeout(Duration::from_secs(60), get_object_bytes(&client, &bucket, "baseline.bin"))
+        .await?
+        .expect_err("eight unavailable drives must prevent the baseline GET");
+    assert!(
+        matches!(
+            s3_probe_error_code(read_error.as_ref()),
+            Some("SlowDown" | "SlowDownRead" | "ServiceUnavailable")
+        ),
+        "quorum loss must produce an availability error: {read_error:?}"
+    );
+    assert_eq!(
+        read_error
+            .downcast_ref::<SdkError<GetObjectError>>()
+            .and_then(SdkError::raw_response)
+            .map(|response| response.status().as_u16()),
+        Some(503),
+        "read quorum loss must remain a service availability failure"
+    );
+    let write_error = timeout(Duration::from_secs(60), put_object(&client, &bucket, "unavailable.bin", baseline.clone()))
+        .await?
+        .expect_err("eight unavailable drives must prevent a successful PUT");
+    assert!(
+        matches!(s3_probe_error_code(write_error.as_ref()), Some("SlowDown" | "ServiceUnavailable")),
+        "quorum loss must produce an availability error: {write_error:?}"
+    );
+    assert_eq!(
+        write_error
+            .downcast_ref::<SdkError<PutObjectError>>()
+            .and_then(SdkError::raw_response)
+            .map(|response| response.status().as_u16()),
+        Some(503),
+        "write quorum loss must remain a service availability failure"
+    );
+
+    // Recover only four drives. The remaining node must stay stopped while
+    // every live coordinator serves old data and acknowledges new writes.
+    dist.cluster.start_node_process(2).await?;
+    let recovery_deadline = Instant::now() + Duration::from_secs(90);
+    let mut recovered_objects = Vec::new();
+    for node in 0..3 {
+        assert!(dist.cluster.nodes[3].process.is_none(), "the fourth node must remain unavailable");
+        let client = s3_probe_client(&dist.client(node)?);
+        retrying_get_equals(
+            &client,
+            &bucket,
+            "baseline.bin",
+            &baseline,
+            recovery_deadline.saturating_duration_since(Instant::now()),
+        )
+        .await?;
+        let key = format!("partial-{node}.bin");
+        let body = payload_for(&key, 1024 * 1024);
+        retrying_put(
+            &client,
+            &bucket,
+            &key,
+            body.clone(),
+            recovery_deadline.saturating_duration_since(Instant::now()),
+        )
+        .await?;
+        recovered_objects.push((key, body));
+    }
+    for node in 0..3 {
+        assert!(dist.cluster.nodes[3].process.is_none(), "reads must complete before all drives return");
+        let client = s3_probe_client(&dist.client(node)?);
+        assert_object_bytes(&client, &bucket, "baseline.bin", &baseline).await?;
+        for (key, body) in &recovered_objects {
+            assert_object_bytes(&client, &bucket, key, body).await?;
+        }
+    }
+
+    dist.cluster.start_node(3).await?;
+    wait_for_ready(&dist.cluster).await?;
+    let client = s3_probe_client(&dist.client(3)?);
+    assert_object_bytes(&client, &bucket, "baseline.bin", &baseline).await?;
+    for (key, body) in &recovered_objects {
+        assert_object_bytes(&client, &bucket, key, body).await?;
+    }
+    Ok(())
+}
+
 #[tokio::test]
 async fn full_cluster_restart_preserves_objects() -> TestResult {
     init_logging();

crates/e2e_test/src/distributed/harness.rs

@@ -347,6 +347,9 @@ fn relocate_pool_storage(cluster: &mut RustFSTestClusterEnvironment, node_idx: u
 }
 
 pub(super) fn configure_node_logs(cluster: &mut RustFSTestClusterEnvironment, capture_dir: Option<&Path>) -> TestResult {
+    if let Ok(filter) = std::env::var("RUSTFS_E2E_LOG_FILTER") {
+        cluster.set_env("RUST_LOG", filter);
+    }
     if let Some(directory) = capture_dir {
         std::fs::create_dir_all(directory)?;
         for node_idx in 0..cluster.nodes.len() {
@@ -1178,11 +1181,16 @@ pub(crate) async fn retrying_get_equals(
                 )
                 .into()),
                 Err(error) => {
-                    if matches!(s3_probe_error_code(error.as_ref()), Some("SlowDown" | "ServiceUnavailable")) {
-                        Ok(false)
-                    } else {
-                        Err(error)
-                    }
+                    let retryable = match s3_probe_error_code(error.as_ref()) {
+                        Some("SlowDown" | "ServiceUnavailable") => true,
+                        // GET preserves this compatibility code for a lost data quorum.
+                        Some("SlowDownRead") => error
+                            .downcast_ref::<SdkError<GetObjectError>>()
+                            .and_then(SdkError::raw_response)
+                            .is_some_and(|response| response.status().as_u16() == 503),
+                        _ => false,
+                    };
+                    if retryable { Ok(false) } else { Err(error) }
                 }
             }
         },
@@ -1321,6 +1329,62 @@ mod retry_tests {
         Ok(())
     }
 
+    #[tokio::test(start_paused = true)]
+    async fn retrying_get_allows_read_quorum_recovery() -> TestResult {
+        use aws_sdk_s3::config::Region;
+        use aws_smithy_http_client::test_util::{ReplayEvent, StaticReplayClient};
+        use aws_smithy_types::body::SdkBody;
+
+        let endpoint = "http://read-quorum.example.invalid";
+        let event = |status: u16, body: &'static str| {
+            ReplayEvent::new(
+                http::Request::builder()
+                    .method(Method::GET)
+                    .uri(format!("{endpoint}/retry-probe/key?x-id=GetObject"))
+                    .body(SdkBody::empty())
+                    .expect("build replay request"),
+                http::Response::builder()
+                    .status(status)
+                    .header("content-type", "application/xml")
+                    .header("content-length", body.len())
+                    .body(SdkBody::from(body))
+                    .expect("build replay response"),
+            )
+        };
+        for status in [503, 403, 500] {
+            let transport = StaticReplayClient::new(vec![
+                event(status, "<Error><Code>SlowDownRead</Code></Error>"),
+                event(200, "expected bytes"),
+            ]);
+            let client = Client::from_conf(
+                aws_sdk_s3::Config::builder()
+                    .behavior_version_latest()
+                    .credentials_provider(Credentials::new(FAKE_ACCESS_KEY, FAKE_SECRET_KEY, None, None, "read-quorum-test"))
+                    .region(Region::new("us-east-1"))
+                    .endpoint_url(endpoint)
+                    .force_path_style(true)
+                    .retry_config(RetryConfig::disabled())
+                    .http_client(transport.clone())
+                    .build(),
+            );
+            let result = retrying_get_equals(&client, "retry-probe", "key", b"expected bytes", Duration::from_secs(5)).await;
+            if status == 503 {
+                result?;
+                assert_eq!(
+                    transport.actual_requests().count(),
+                    2,
+                    "the probe must retry the read-quorum error exactly once"
+                );
+                transport.assert_requests_match(&[]);
+            } else {
+                let error = result.expect_err("the compatibility code must not hide a non-503 failure");
+                assert_eq!(s3_probe_error_code(error.as_ref()), Some("SlowDownRead"));
+                assert_eq!(transport.actual_requests().count(), 1, "non-availability errors must fail immediately");
+            }
+        }
+        Ok(())
+    }
+
     #[tokio::test]
     async fn object_polls_distinguish_committed_and_replicated_keys() -> TestResult {
         let target = FakeS3Target::start().await?;

crates/e2e_test/src/distributed/internode_tls_test.rs

@@ -104,7 +104,12 @@ async fn verified_get(
         stage = "response body";
         let mut digest = Sha256::new();
         while let Some(chunk) = output.body.next().await {
-            let chunk = chunk?;
+            let chunk = chunk.inspect_err(|error| {
+                eprintln!(
+                    "TLS acceptance: GET body failed: node={node}, key={}, range={:?}, concurrent={concurrent}, received={received}/{}, elapsed={:?}, error={error}",
+                    object.key, expected.range, expected.length, started.elapsed()
+                );
+            })?;
             received += u64::try_from(chunk.len())?;
             assert!(received <= expected.length, "node {node}: GET exceeded the requested length");
             digest.update(&chunk);

crates/e2e_test/src/distributed/mod.rs

@@ -28,14 +28,18 @@ mod harness;
 mod heal_test;
 mod internode_tls_test;
 mod multipart_fence_acceptance_test;
+mod multipart_rpc_delay_test;
 mod object_lock_test;
 mod observability_test;
+mod publication_log;
 mod replication_delete_marker_test;
 mod replication_quota_test;
 mod s3_basic_test;
 mod s3_during_data_movement_test;
 mod scanner_heal_topology_fixture;
 mod scanner_heal_topology_test;
+mod scanner_heal_topology_tier_test;
 mod site_replication_test;
+mod table_catalog_test;
 mod upgrade_test;
 mod versioning_test;

crates/e2e_test/src/distributed/publication_log.rs

@@ -0,0 +1,413 @@
+// Copyright 2026 RustFS Team
+// SPDX-License-Identifier: Apache-2.0
+
+//! Follow the same append-only log files without rescanning previous JSON rows.
+
+use same_file::Handle;
+use serde::{Deserialize, Deserializer};
+use serde_json::Value;
+use std::fs::File;
+use std::io::{self, Read};
+use std::path::PathBuf;
+
+// Captured RPC spans already produce valid rows above 2 MiB. Keep complete
+// rows bounded independently from the much smaller selected-event history.
+const MAX_LINE_BYTES: usize = 4 << 20;
+const MAX_EVENTS: usize = 128;
+const MAX_EVENT_BYTES: usize = 1 << 20;
+
+fn invalid(reason: &'static str) -> io::Error {
+    io::Error::new(io::ErrorKind::InvalidData, reason)
+}
+
+#[derive(Default, Deserialize)]
+struct Selector {
+    #[serde(default)]
+    event: Value,
+    #[serde(default)]
+    op: Value,
+    #[serde(default)]
+    bucket: Value,
+    #[serde(default)]
+    object: Value,
+}
+
+impl Selector {
+    fn matches(&self, bucket: &str, object: &str) -> bool {
+        self.event.as_str() == Some("object_namespace_lock")
+            && self.op.as_str() == Some("data_movement_publication")
+            && self.bucket.as_str() == Some(bucket)
+            && self.object.as_str() == Some(object)
+    }
+}
+
+#[derive(Deserialize)]
+struct Probe {
+    #[serde(default)]
+    event: Value,
+    #[serde(default)]
+    op: Value,
+    #[serde(default)]
+    bucket: Value,
+    #[serde(default)]
+    object: Value,
+    // A present null is not a missing fields object. The old reader uses the
+    // root only when this key is absent, including for non-object JSON values.
+    #[serde(default, deserialize_with = "present_fields")]
+    fields: Option<Option<Selector>>,
+}
+
+fn present_fields<'de, D: Deserializer<'de>>(d: D) -> Result<Option<Option<Selector>>, D::Error> {
+    Option::<Selector>::deserialize(d).map(Some)
+}
+
+impl Probe {
+    fn matches(&self, bucket: &str, object: &str) -> bool {
+        match &self.fields {
+            Some(Some(fields)) => fields.matches(bucket, object),
+            Some(None) => false,
+            None => {
+                self.event.as_str() == Some("object_namespace_lock")
+                    && self.op.as_str() == Some("data_movement_publication")
+                    && self.bucket.as_str() == Some(bucket)
+                    && self.object.as_str() == Some(object)
+            }
+        }
+    }
+}
+
+fn select(line: &[u8], bucket: &str, object: &str, full_decodes: &mut u64) -> io::Result<Option<Value>> {
+    let text = std::str::from_utf8(line).map_err(|_| invalid("invalid UTF-8 in a completed log row"))?;
+    // Skip unknown fields without allocating a Value tree for every RPC trace.
+    // A shape/duplicate-field error falls back to the original Value parser;
+    // it must not discard valid escaped strings or last-key-wins JSON.
+    if let Ok(probe) = serde_json::from_str::<Probe>(text)
+        && !probe.matches(bucket, object)
+    {
+        return Ok(None);
+    }
+    *full_decodes += 1;
+    let Ok(row) = serde_json::from_str::<Value>(text) else {
+        return Ok(None);
+    };
+    let fields = row.get("fields").unwrap_or(&row);
+    if fields.get("event").and_then(Value::as_str) != Some("object_namespace_lock")
+        || fields.get("op").and_then(Value::as_str) != Some("data_movement_publication")
+        || fields.get("bucket").and_then(Value::as_str) != Some(bucket)
+        || fields.get("object").and_then(Value::as_str) != Some(object)
+    {
+        return Ok(None);
+    }
+    let mut event = fields.clone();
+    event["timestamp"] = row["timestamp"].clone();
+    Ok(Some(event))
+}
+
+struct Tail {
+    path: PathBuf,
+    handle: Handle,
+    offset: u64,
+    pending: Vec<u8>,
+}
+
+impl Tail {
+    fn open(path: PathBuf) -> io::Result<Self> {
+        let file = File::open(&path)?;
+        if !file.metadata()?.is_file() {
+            return Err(invalid("publication log is not a regular file"));
+        }
+        Ok(Self {
+            path,
+            handle: Handle::from_file(file)?,
+            offset: 0,
+            pending: Vec::new(),
+        })
+    }
+
+    fn check_identity(&self) -> io::Result<()> {
+        if Handle::from_path(&self.path)? != self.handle {
+            return Err(invalid("publication log file was replaced"));
+        }
+        Ok(())
+    }
+
+    fn poll(&mut self, bytes_read: &mut u64, mut accept: impl FnMut(&[u8]) -> io::Result<()>) -> io::Result<()> {
+        self.check_identity()?;
+        let end = self.handle.as_file().metadata()?.len();
+        if end < self.offset {
+            return Err(invalid("publication log file was truncated"));
+        }
+        // A fixed high-water snapshot bounds each poll even if writers append
+        // faster than the reader. Bytes after it remain for the next poll.
+        let mut chunk = [0_u8; 64 * 1024];
+        while self.offset < end {
+            let size = usize::try_from((end - self.offset).min(64 * 1024)).expect("a log chunk fits usize");
+            self.handle.as_file_mut().read_exact(&mut chunk[..size])?;
+            self.offset += u64::try_from(size).expect("a bounded log read fits u64");
+            *bytes_read += u64::try_from(size).expect("a bounded log read fits u64");
+            let appended_at = self.pending.len();
+            self.pending.extend_from_slice(&chunk[..size]);
+            let mut start = 0;
+            for end in self.pending[appended_at..]
+                .iter()
+                .enumerate()
+                .filter_map(|(i, byte)| (*byte == b'\n').then_some(appended_at + i + 1))
+            {
+                if end - start > MAX_LINE_BYTES {
+                    return Err(invalid("publication log row exceeds its fixed bound"));
+                }
+                accept(&self.pending[start..end])?;
+                start = end;
+            }
+            self.pending.drain(..start);
+            if self.pending.len() > MAX_LINE_BYTES {
+                return Err(invalid("unfinished publication log row exceeds its fixed bound"));
+            }
+        }
+        self.check_identity()
+    }
+}
+
+pub(super) struct PublicationEvents {
+    tails: Vec<Tail>,
+    bucket: String,
+    object: String,
+    events: Vec<Value>,
+    per_file: Vec<Vec<Value>>,
+    event_count: usize,
+    event_bytes: usize,
+    bytes_read: u64,
+    full_decodes: u64,
+    failed: bool,
+}
+
+impl PublicationEvents {
+    pub(super) fn new(paths: Vec<PathBuf>, bucket: &str, object: &str) -> io::Result<Self> {
+        if paths.is_empty() || paths.len() > 4 {
+            return Err(invalid("publication capture requires one to four log files"));
+        }
+        let tails: Vec<Tail> = paths.into_iter().map(Tail::open).collect::<io::Result<_>>()?;
+        let mut reader = Self {
+            per_file: vec![Vec::new(); tails.len()],
+            tails,
+            bucket: bucket.to_owned(),
+            object: object.to_owned(),
+            events: Vec::new(),
+            event_bytes: 0,
+            event_count: 0,
+            bytes_read: 0,
+            full_decodes: 0,
+            failed: false,
+        };
+        // Prime before holding remote DELETE or starting a movement attempt.
+        // Retain earlier matching events too: the caller still applies its
+        // original timestamp boundary, and evidence keeps the complete history.
+        reader.poll()?;
+        Ok(reader)
+    }
+
+    pub(super) fn poll(&mut self) -> io::Result<&[Value]> {
+        if self.failed {
+            return Err(invalid("publication log reader was already invalidated"));
+        }
+        let result = (|| {
+            for (tail, history) in self.tails.iter_mut().zip(&mut self.per_file) {
+                tail.poll(&mut self.bytes_read, |line| {
+                    if let Some(event) = select(line, &self.bucket, &self.object, &mut self.full_decodes)? {
+                        let bytes = serde_json::to_vec(&event).map_err(io::Error::other)?.len();
+                        if self.event_count >= MAX_EVENTS || bytes > MAX_EVENT_BYTES - self.event_bytes {
+                            return Err(invalid("publication event history exceeds its fixed bound"));
+                        }
+                        self.event_bytes += bytes;
+                        self.event_count += 1;
+                        history.push(event);
+                    }
+                    Ok(())
+                })?;
+            }
+            Ok(())
+        })();
+        // Preserve the old file-then-row order, even when a later poll appends
+        // to an earlier file. Only bounded selected history is copied here.
+        self.events = self.per_file.iter().flatten().cloned().collect();
+        if let Err(error) = result {
+            self.failed = true;
+            return Err(error);
+        }
+        Ok(&self.events)
+    }
+
+    /// Diagnostics only: after a read error this history can be incomplete.
+    pub(super) fn cached(&self) -> &[Value] {
+        &self.events
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use serde_json::json;
+    use std::io::Write;
+
+    fn event(id: &str) -> Value {
+        json!({"timestamp":"2026-10-26T00:00:00Z","event":"object_namespace_lock","op":"data_movement_publication","bucket":"bucket","object":"object","lock_attempt_id":id,"state":"acquiring","extra":{"kept":[1,2]}})
+    }
+    fn append(path: &std::path::Path, raw: &[u8]) {
+        std::fs::OpenOptions::new()
+            .append(true)
+            .open(path)
+            .expect("open append log")
+            .write_all(raw)
+            .expect("append log bytes");
+    }
+    fn legacy(raw: &[u8]) -> Vec<Value> {
+        std::str::from_utf8(raw)
+            .expect("fixture UTF-8")
+            .lines()
+            .filter_map(|line| {
+                let row: Value = serde_json::from_str(line).ok()?;
+                let fields = row.get("fields").unwrap_or(&row);
+                if fields.get("event").and_then(Value::as_str) == Some("object_namespace_lock")
+                    && fields.get("op").and_then(Value::as_str) == Some("data_movement_publication")
+                    && fields.get("bucket").and_then(Value::as_str) == Some("bucket")
+                    && fields.get("object").and_then(Value::as_str) == Some("object")
+                {
+                    let mut result = fields.clone();
+                    result["timestamp"] = row["timestamp"].clone();
+                    Some(result)
+                } else {
+                    None
+                }
+            })
+            .collect()
+    }
+
+    #[test]
+    fn appended_events_keep_full_fields_without_history_rescans() {
+        let file = tempfile::NamedTempFile::new().expect("log file");
+        let mut raw = (json!({"message":"unrelated trace","payload":"x".repeat(4096)}).to_string() + "\n")
+            .repeat(64)
+            .into_bytes();
+        raw.extend_from_slice(
+            format!("{}\n", json!({"event":"object_namespace_lock","op":"get_object","bucket":"bucket","object":"object","spans":"x".repeat(2401832)})).as_bytes(),
+        );
+        raw.extend_from_slice(format!("{}\n", event("first")).as_bytes());
+        std::fs::write(file.path(), &raw).expect("seed log");
+        let mut reader = PublicationEvents::new(vec![file.path().to_owned()], "bucket", "object").expect("prime log");
+        assert_eq!(reader.cached(), legacy(&raw));
+        assert_eq!(reader.full_decodes, 1);
+        let bytes = reader.bytes_read;
+        assert_eq!(reader.poll().expect("unchanged poll"), legacy(&raw));
+        assert_eq!(reader.bytes_read, bytes);
+        let next = format!("{}\n", event("second"));
+        append(file.path(), next.as_bytes());
+        raw.extend_from_slice(next.as_bytes());
+        assert_eq!(reader.poll().expect("new rows"), legacy(&raw));
+        assert_eq!(reader.bytes_read, u64::try_from(raw.len()).expect("fixture length"));
+        assert_eq!(reader.full_decodes, 2);
+    }
+
+    #[test]
+    fn later_appends_preserve_the_old_file_then_row_order() {
+        let first = tempfile::NamedTempFile::new().expect("first log");
+        let second = tempfile::NamedTempFile::new().expect("second log");
+        std::fs::write(first.path(), format!("{}\n", event("a1"))).expect("seed first");
+        std::fs::write(second.path(), format!("{}\n", event("b1"))).expect("seed second");
+        let mut reader = PublicationEvents::new(vec![first.path().to_owned(), second.path().to_owned()], "bucket", "object")
+            .expect("prime both");
+        append(first.path(), format!("{}\n", event("a2")).as_bytes());
+        let mut expected = legacy(&std::fs::read(first.path()).expect("first bytes"));
+        expected.extend(legacy(&std::fs::read(second.path()).expect("second bytes")));
+        assert_eq!(reader.poll().expect("append first after second was read"), expected);
+    }
+
+    #[test]
+    fn root_nested_escaped_and_duplicate_fields_match_the_old_reader() {
+        let file = tempfile::NamedTempFile::new().expect("log file");
+        let root = event("root");
+        let mut nested = json!({"timestamp":root["timestamp"],"fields":root});
+        nested["fields"]["extra"] = json!({"unicode":"é"});
+        let null = json!({"fields":null,"event":"object_namespace_lock","op":"data_movement_publication","bucket":"bucket","object":"object"});
+        let escaped = root
+            .to_string()
+            .replace("object_namespace_lock", "\\u006fbject_namespace_lock");
+        let duplicate = root.to_string().replacen("{", "{\"event\":\"earlier-value\",", 1);
+        let raw = format!("{root}\n{nested}\n{null}\n{escaped}\n{duplicate}\nnot json\n{{\"fields\":[]}}\n");
+        std::fs::write(file.path(), &raw).expect("mixed log");
+        let reader = PublicationEvents::new(vec![file.path().to_owned()], "bucket", "object").expect("probe and fallback");
+        assert_eq!(reader.cached(), legacy(raw.as_bytes()));
+        assert_eq!(reader.cached().len(), 4);
+    }
+
+    #[test]
+    fn incomplete_rows_and_split_utf8_wait_for_the_newline() {
+        let file = tempfile::NamedTempFile::new().expect("log file");
+        let mut row = event("partial");
+        row["extra"] = json!("é");
+        let raw = format!("{row}\n").into_bytes();
+        let cut = raw.iter().position(|b| *b == 0xc3).expect("UTF-8 prefix") + 1;
+        std::fs::write(file.path(), &raw[..cut]).expect("partial code point");
+        let mut reader =
+            PublicationEvents::new(vec![file.path().to_owned()], "bucket", "object").expect("incomplete row is pending");
+        assert!(reader.cached().is_empty());
+        append(file.path(), &raw[cut..raw.len() - 1]);
+        assert!(reader.poll().expect("complete JSON without delimiter").is_empty());
+        append(file.path(), b"\n");
+        assert_eq!(reader.poll().expect("published row"), legacy(&raw));
+    }
+
+    #[test]
+    fn truncation_is_sticky_even_after_the_file_grows_again() {
+        let file = tempfile::NamedTempFile::new().expect("log file");
+        std::fs::write(file.path(), format!("{}\n", event("first"))).expect("seed log");
+        let mut reader = PublicationEvents::new(vec![file.path().to_owned()], "bucket", "object").expect("prime log");
+        file.as_file().set_len(0).expect("truncate");
+        assert!(reader.poll().is_err());
+        append(file.path(), format!("{}\n", event("longer replacement")).as_bytes());
+        assert!(reader.poll().is_err());
+        assert_eq!(reader.cached().len(), 1);
+    }
+
+    #[cfg(unix)]
+    #[test]
+    fn a_replaced_path_does_not_switch_to_the_new_log() {
+        let dir = tempfile::tempdir().expect("log dir");
+        let path = dir.path().join("node.log");
+        std::fs::write(&path, b"").expect("log file");
+        let mut reader = PublicationEvents::new(vec![path.clone()], "bucket", "object").expect("open log");
+        std::fs::rename(&path, dir.path().join("retired.log")).expect("replace log");
+        std::fs::write(&path, format!("{}\n", event("other-generation"))).expect("replacement");
+        assert!(reader.poll().is_err());
+        assert!(reader.cached().is_empty());
+    }
+
+    #[test]
+    fn invalid_utf8_and_oversized_or_excessive_records_fail_closed() {
+        let file = tempfile::NamedTempFile::new().expect("log file");
+        let mut reader = PublicationEvents::new(vec![file.path().to_owned()], "bucket", "object").expect("empty log");
+        append(file.path(), b"\xff\n");
+        assert!(reader.poll().is_err());
+        let file = tempfile::NamedTempFile::new().expect("log file");
+        let mut reader = PublicationEvents::new(vec![file.path().to_owned()], "bucket", "object").expect("empty log");
+        append(file.path(), &vec![b'x'; MAX_LINE_BYTES + 1]);
+        assert!(reader.poll().is_err());
+        let file = tempfile::NamedTempFile::new().expect("log file");
+        let mut reader = PublicationEvents::new(vec![file.path().to_owned()], "bucket", "object").expect("empty log");
+        for _ in 0..=MAX_EVENTS {
+            append(file.path(), format!("{}\n", event("attempt")).as_bytes());
+        }
+        assert!(reader.poll().is_err());
+        assert_eq!(reader.cached().len(), MAX_EVENTS);
+        assert!(reader.poll().is_err());
+        let file = tempfile::NamedTempFile::new().expect("log file");
+        let mut reader = PublicationEvents::new(vec![file.path().to_owned()], "bucket", "object").expect("empty log");
+        let mut large = event("large");
+        large["extra"] = json!("x".repeat(600 * 1024));
+        for _ in 0..2 {
+            append(file.path(), format!("{large}\n").as_bytes());
+        }
+        assert!(reader.poll().is_err());
+        assert_eq!(reader.cached().len(), 1);
+        assert!(reader.poll().is_err());
+    }
+}

crates/e2e_test/src/distributed/scanner_heal_topology_fixture.rs

@@ -216,6 +216,10 @@ pub(crate) struct TopologyRun {
 
 impl TopologyRun {
     pub(crate) async fn start(label: &str) -> TestResult<Self> {
+        Self::start_with_env(label, &[]).await
+    }
+
+    pub(crate) async fn start_with_env(label: &str, extra_env: &[(&str, &str)]) -> TestResult<Self> {
         let run_name = format!("scanner-heal-topology-{label}-{}", Uuid::new_v4().simple());
         // CI uploads RUSTFS_E2E_LOG_DIR; keep node logs, barrier observations
         // and evidence there so a failed lane can be diagnosed.
@@ -236,6 +240,7 @@ impl TopologyRun {
         let mut env: Vec<(&str, &str)> = TOPOLOGY_ENV.to_vec();
         env.extend_from_slice(FAST_DATA_USAGE_SCANNER_ENV);
         env.push((MOVEMENT_BARRIER_DIR_ENV, movement_dir_value.as_str()));
+        env.extend_from_slice(extra_env);
         let mut dist = DistCluster::new_stopped_with_env(DistLayout::SingleNodeFourDrive, &env).await?;
         dist.capture_logs_under(&log_dir)?;
         dist.cluster.start().await?;
@@ -329,6 +334,11 @@ impl FileBarrier {
         &self.nonce
     }
 
+    /// Keep the previous phase's release handle while arming a fresh nonce.
+    pub(crate) fn next_phase(&self) -> Self {
+        Self::new(self.dir.clone())
+    }
+
     fn write_atomic(&self, name: &str, body: &[u8]) -> TestResult {
         let tmp = self.dir.join(format!(".{name}.{}.tmp", Uuid::new_v4().simple()));
         std::fs::write(&tmp, body)?;
@@ -360,6 +370,10 @@ impl FileBarrier {
         self.write_atomic(&format!("release.{}", self.nonce), b"io_error")
     }
 
+    pub(crate) fn release_movement(&self) -> TestResult {
+        self.write_atomic(&format!("release.{}", self.nonce), b"continue")
+    }
+
     pub(crate) fn reached_observations(&self) -> TestResult<Vec<Value>> {
         let prefix = format!("reached.{}.", self.nonce);
         let mut observations = Vec::new();

crates/e2e_test/src/distributed/scanner_heal_topology_test.rs

@@ -528,7 +528,7 @@ async fn scanner_heal_topology_decommission_kill_before_source_cleanup_resumes_w
     run.movement_barrier.disarm()?;
     let restarted_pid = run.restart_killed_node(SEED_POOL).await?;
     let label = "after the killed leader resumed decommission";
-    let outcome = settle_decommission(&run, &manifest, label, DECOMMISSION_RESUME_COMPLETE_TIMEOUT).await;
+    let outcome = settle_decommission(&run, &manifest, label, MOVEMENT_COMPLETE_TIMEOUT).await;
     let strict_nodes = assert_process_durability_strict(&run.node_log_paths());
     record(
         &run,

crates/e2e_test/src/distributed/scanner_heal_topology_tier_test.rs

@@ -0,0 +1,599 @@
+// 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.
+
+//! G-TIER: real mover processes contend with a held remote DELETE while a
+//! free version owns the old remote tuple beside a live transitioned version.
+//! The remote is the HTTP fake S3 target, not a provider interoperability test.
+
+use super::harness::{
+    DECOMMISSION_POOL_ID, DRIVES_PER_NODE, TestResult, cluster_admin_ok, decommission_status_json, enable_versioning,
+    rebalance_status_json, sha256_hex, start_decommission, start_rebalance, unique_bucket, wait_for_decommission_complete,
+    wait_for_rebalance_complete,
+};
+use super::publication_log::PublicationEvents;
+use super::scanner_heal_topology_fixture::{
+    MOVEMENT_COMPLETE_TIMEOUT, MovementOperation, SEED_POOL, STAGE_AFTER_RESPONSIBILITY_PERSISTED, STAGE_BEFORE_SOURCE_CLEANUP,
+    TopologyRun, assert_bucket_durability_strict, assert_process_durability_strict, require_test_hooks,
+};
+use crate::common::init_logging;
+use crate::fake_s3_target::{FAKE_ACCESS_KEY, FAKE_SECRET_KEY, FakeS3Target, Operation, RequestRecord};
+use crate::storage_api::OBJECT_LOCK_TRACE_FILTER;
+use aws_sdk_s3::primitives::ByteStream;
+use aws_sdk_s3::types::{
+    BucketLifecycleConfiguration, ExpirationStatus, LifecycleRule, LifecycleRuleFilter, Transition, TransitionStorageClass,
+};
+use http::Method;
+use rustfs_filemeta::{FileMeta, TransitionVersionState};
+use rustfs_utils::http::{SUFFIX_TRANSITION_TIER_DESTINATION_ID, get_consistent_str};
+use serde::Serialize;
+use serde_json::{Value, json};
+use std::collections::BTreeSet;
+use std::path::Path;
+use std::time::Duration;
+use time::{OffsetDateTime, format_description::well_known::Rfc3339};
+use tokio::time::{Instant, sleep, timeout, timeout_at};
+
+const TIER: &str = "TOPOLOGYCOLD";
+const KEY: &str = "000-tier-owner.bin";
+const TRANSITION_TIMEOUT: Duration = Duration::from_secs(90);
+// Free-version cleanup holds its locks for at most 30 s during remote DELETE.
+// Fail before that deadline instead of confusing a retry with the held attempt.
+const DELETE_OVERLAP_TIMEOUT: Duration = Duration::from_secs(20);
+
+#[derive(Clone, Debug, PartialEq, Eq, Serialize)]
+struct RemoteTuple {
+    bucket: String,
+    key: String,
+    version: String,
+    tier: String,
+    destination: String,
+}
+
+#[derive(Debug, PartialEq, Eq, Serialize)]
+struct Owner {
+    pool: usize,
+    drive: usize,
+    version: String,
+    free: bool,
+    remote: RemoteTuple,
+}
+
+fn owners(run: &TopologyRun, bucket: &str, remote_bucket: &str) -> TestResult<Vec<Owner>> {
+    let mut owners = Vec::new();
+    for node in &run.dist.cluster.nodes {
+        for (drive, root) in node.data_dirs.iter().enumerate() {
+            let bytes = match std::fs::read(Path::new(root).join(bucket).join(KEY).join("xl.meta")) {
+                Ok(bytes) => bytes,
+                Err(error) if error.kind() == std::io::ErrorKind::NotFound => continue,
+                Err(error) => return Err(error.into()),
+            };
+            for version in FileMeta::load(&bytes)?.get_file_info_versions(bucket, KEY, true)?.versions {
+                if version.transition_status != "complete" && !version.tier_free_version() {
+                    continue;
+                }
+                assert_eq!(version.transition_version_state, TransitionVersionState::Exact);
+                owners.push(Owner {
+                    pool: node.pool_idx,
+                    drive,
+                    version: version.version_id.ok_or("tier owner omitted its local version")?.to_string(),
+                    free: version.tier_free_version(),
+                    remote: RemoteTuple {
+                        bucket: remote_bucket.to_string(),
+                        key: version.transitioned_objname.clone(),
+                        version: version
+                            .transition_version
+                            .clone()
+                            .ok_or("tier owner omitted its remote version")?,
+                        tier: version.transition_tier.clone(),
+                        destination: get_consistent_str(&version.metadata, SUFFIX_TRANSITION_TIER_DESTINATION_ID)
+                            .ok_or("tier owner omitted its destination identity")?
+                            .to_string(),
+                    },
+                });
+            }
+        }
+    }
+    Ok(owners)
+}
+
+fn assert_remote(target: &FakeS3Target, remote: &RemoteTuple, expected: &[u8]) -> TestResult {
+    assert_eq!(target.stored_versions(&remote.bucket, &remote.key), vec![(remote.version.clone(), false)]);
+    let (body, metadata) = target
+        .stored_object(&remote.bucket, &remote.key)
+        .ok_or("remote bytes are missing")?;
+    assert_eq!(body.as_ref(), expected, "remote bytes changed for {remote:?}");
+    assert_eq!(
+        get_consistent_str(&metadata, SUFFIX_TRANSITION_TIER_DESTINATION_ID),
+        Some(remote.destination.as_str())
+    );
+    assert_eq!(remote.tier, TIER);
+    Ok(())
+}
+
+async fn transition_version(
+    run: &TopologyRun,
+    target: &FakeS3Target,
+    bucket: &str,
+    remote_bucket: &str,
+    body: &[u8],
+) -> TestResult<(String, RemoteTuple)> {
+    let client = run.dist.client(0)?;
+    let put = client
+        .put_object()
+        .bucket(bucket)
+        .key(KEY)
+        .body(ByteStream::from(body.to_vec()))
+        .send()
+        .await?;
+    let version = put.version_id().ok_or("versioned PUT omitted its version")?.to_string();
+    cluster_admin_ok(
+        &run.dist.cluster,
+        Method::POST,
+        &format!("/rustfs/admin/v3/ilm/transition/run?bucket={bucket}&prefix={KEY}&tier={TIER}&dryRun=false"),
+        None,
+    )
+    .await?;
+    let deadline = Instant::now() + TRANSITION_TIMEOUT;
+    loop {
+        let current = owners(run, bucket, remote_bucket)?;
+        let shards: Vec<_> = current
+            .iter()
+            .filter(|owner| owner.version == version && !owner.free)
+            .collect();
+        if let Some(first) = shards.first()
+            && shards.len() == DRIVES_PER_NODE
+            && shards
+                .iter()
+                .all(|owner| owner.pool == SEED_POOL && owner.remote == first.remote)
+        {
+            let head = client
+                .head_object()
+                .bucket(bucket)
+                .key(KEY)
+                .version_id(&version)
+                .send()
+                .await?;
+            assert_eq!(head.storage_class().map(|class| class.as_str()), Some(TIER));
+            assert_remote(target, &first.remote, body)?;
+            return Ok((version, first.remote.clone()));
+        }
+        if Instant::now() >= deadline {
+            return Err(format!("transition of {bucket}/{KEY}@{version} did not reach every seed drive: {current:?}").into());
+        }
+        sleep(Duration::from_millis(200)).await;
+    }
+}
+
+fn event_time(event: &Value) -> TestResult<OffsetDateTime> {
+    Ok(OffsetDateTime::parse(
+        event["timestamp"].as_str().ok_or("lock event omitted its timestamp")?,
+        &Rfc3339,
+    )?)
+}
+
+async fn pending_publication(events: &mut PublicationEvents, started: OffsetDateTime) -> TestResult<Value> {
+    loop {
+        let observed = events.poll()?;
+        for event in observed {
+            if event_time(event)? < started {
+                continue;
+            }
+            assert_ne!(
+                event["state"], "acquired",
+                "mover acquired its publication lock while remote DELETE was held"
+            );
+            if event["state"] == "acquiring"
+                && !observed
+                    .iter()
+                    .any(|later| later["lock_attempt_id"] == event["lock_attempt_id"] && later["state"] != "acquiring")
+            {
+                assert!(event["lock_attempt_id"].as_str().is_some_and(|id| !id.is_empty()));
+                return Ok(event.clone());
+            }
+        }
+        sleep(Duration::from_millis(50)).await;
+    }
+}
+
+async fn acquired_publication(events: &mut PublicationEvents, pending: &Value) -> TestResult<Value> {
+    loop {
+        for event in events.poll()? {
+            if event["lock_attempt_id"] != pending["lock_attempt_id"] {
+                continue;
+            }
+            match event["state"].as_str() {
+                Some("acquired") => return Ok(event.clone()),
+                Some("failed" | "cancelled") => return Err(format!("held publication attempt did not acquire: {event}").into()),
+                _ => {}
+            }
+        }
+        sleep(Duration::from_millis(50)).await;
+    }
+}
+
+fn deletes(target: &FakeS3Target, remote: &RemoteTuple) -> Vec<RequestRecord> {
+    target
+        .requests()
+        .into_iter()
+        .filter(|request| {
+            request.operation == Operation::DeleteObject
+                && request.bucket == remote.bucket
+                && request.key.as_deref() == Some(remote.key.as_str())
+        })
+        .collect()
+}
+
+async fn verify_live(run: &TopologyRun, bucket: &str, live_version: &str, old_version: &str, body: &[u8]) -> TestResult {
+    for client in run.dist.clients()? {
+        let got = client
+            .get_object()
+            .bucket(bucket)
+            .key(KEY)
+            .version_id(live_version)
+            .send()
+            .await?;
+        assert_eq!(got.body.collect().await?.into_bytes().as_ref(), body);
+        let latest = client.head_object().bucket(bucket).key(KEY).send().await?;
+        assert_eq!(latest.version_id(), Some(live_version));
+        assert_eq!(latest.storage_class().map(|class| class.as_str()), Some(TIER));
+        let deleted = client
+            .get_object()
+            .bucket(bucket)
+            .key(KEY)
+            .version_id(old_version)
+            .send()
+            .await;
+        let error = deleted.expect_err("deleted source version must not be visible");
+        assert_eq!(error.raw_response().map(|response| response.status().as_u16()), Some(404));
+        let listed = client.list_object_versions().bucket(bucket).prefix(KEY).send().await?;
+        assert_eq!(listed.versions().len(), 1, "free owners must not appear as object versions");
+        assert_eq!(listed.versions()[0].version_id(), Some(live_version));
+        assert!(
+            listed.delete_markers().is_empty(),
+            "moving the free owner must not create a delete marker"
+        );
+    }
+    Ok(())
+}
+
+async fn tier_delete_during_movement(decommission: bool) -> TestResult {
+    init_logging();
+    require_test_hooks();
+    let operation = if decommission { "decommission" } else { "rebalance" };
+    let target = FakeS3Target::start().await?;
+    let remote_bucket = unique_bucket("sht-cold");
+    target.create_bucket(&remote_bucket);
+    let mut run = TopologyRun::start_with_env(
+        &format!("tier-{operation}"),
+        &[
+            ("RUSTFS_TIER_RUSTFS_ALLOW_LOOPBACK_ENDPOINT", "true"),
+            ("RUSTFS_OBJECT_LOCK_DIAG_ENABLE", "true"),
+            ("RUST_LOG", OBJECT_LOCK_TRACE_FILTER),
+        ],
+    )
+    .await?;
+    let tier = json!({
+        "type": "rustfs",
+        "rustfs": {
+            "name": TIER, "endpoint": target.endpoint(), "accessKey": FAKE_ACCESS_KEY,
+            "secretKey": FAKE_SECRET_KEY, "bucket": remote_bucket, "prefix": "",
+            "region": "us-east-1", "storageClass": ""
+        }
+    });
+    let deadline = Instant::now() + Duration::from_secs(30);
+    loop {
+        match cluster_admin_ok(&run.dist.cluster, Method::PUT, "/rustfs/admin/v3/tier", Some(tier.to_string())).await {
+            Ok(_) => break,
+            Err(_) if Instant::now() < deadline => sleep(Duration::from_millis(250)).await,
+            Err(error) => return Err(error),
+        }
+    }
+    let bucket = unique_bucket("sht-tier");
+    run.dist.create_bucket(&bucket).await?;
+    let client = run.dist.client(0)?;
+    enable_versioning(&client, &bucket).await?;
+    let rule = LifecycleRule::builder()
+        .id("tier-owner")
+        .status(ExpirationStatus::Enabled)
+        .filter(LifecycleRuleFilter::builder().prefix(KEY).build())
+        .transitions(
+            Transition::builder()
+                .days(0)
+                .storage_class(TransitionStorageClass::from(TIER))
+                .build(),
+        )
+        .build()?;
+    client
+        .put_bucket_lifecycle_configuration()
+        .bucket(&bucket)
+        .lifecycle_configuration(BucketLifecycleConfiguration::builder().rules(rule).build()?)
+        .send()
+        .await?;
+    let old_body = vec![0x31; 4096];
+    let live_body = vec![0x92; 8192];
+    let (old_version, old_remote) = transition_version(&run, &target, &bucket, &remote_bucket, &old_body).await?;
+    let (live_version, live_remote) = transition_version(&run, &target, &bucket, &remote_bucket, &live_body).await?;
+    assert_ne!(
+        old_remote.key, live_remote.key,
+        "the obsolete and live versions need distinct remote tuples"
+    );
+    // Keep the seed pool above the rebalance goal until the first key is visited.
+    // These local objects are outside the transition rule's prefix.
+    for index in 0..16 {
+        client
+            .put_object()
+            .bucket(&bucket)
+            .key(format!("zz-load/{index:02}.bin"))
+            .body(ByteStream::from(vec![index as u8; 1024 * 1024]))
+            .send()
+            .await?;
+    }
+    let bucket_durability = assert_bucket_durability_strict(&run.dist.cluster, &[&bucket]).await?;
+    run.dist.expand_to_four_pools().await?;
+    assert_bucket_durability_strict(&run.dist.cluster, &[&bucket]).await?;
+    let seeded = owners(&run, &bucket, &remote_bucket)?;
+    assert_eq!(seeded.len(), 2 * DRIVES_PER_NODE);
+    assert!(seeded.iter().all(|owner| owner.pool == SEED_POOL && !owner.free));
+
+    let mut decommission_start = Box::pin(start_decommission(&run.dist.cluster, DECOMMISSION_POOL_ID));
+    let mut startup = Value::Null;
+    if decommission {
+        run.movement_barrier.arm_movement(
+            "decommission_start",
+            ".rustfs.sys",
+            "pool.bin",
+            STAGE_AFTER_RESPONSIBILITY_PERSISTED,
+        )?;
+        let requested_at = OffsetDateTime::now_utc();
+        let preparation_started = Instant::now();
+        let reached = tokio::select! {
+            reached = run.movement_barrier.wait_reached(MOVEMENT_COMPLETE_TIMEOUT) => {
+                reached.map_err(|error| format!(
+                    "G-TIER decommission start preparation failed before DELETE: {error}"
+                ))?
+            }
+            result = &mut decommission_start => {
+                return Err(format!(
+                    "G-TIER decommission start preparation returned before durable-start barrier {}: {result:?}",
+                    run.movement_barrier.nonce()
+                ).into());
+            }
+        };
+        startup = json!({
+            "requested_at": requested_at.format(&Rfc3339)?,
+            "parked_at": OffsetDateTime::now_utc().format(&Rfc3339)?,
+            "preparation_elapsed_ms": preparation_started.elapsed().as_millis(),
+            "barrier": reached,
+        });
+    }
+    // The durable-start barrier has released the movement writer but has not
+    // started a worker. Install a separate release nonce before holding DELETE.
+    let cleanup_barrier = run.movement_barrier.next_phase();
+    cleanup_barrier.arm_movement(operation, &bucket, KEY, STAGE_BEFORE_SOURCE_CLEANUP)?;
+    let mut publication_events = PublicationEvents::new(run.node_log_paths(), &bucket, KEY)?;
+    let mut delete_gate = target.hold_delete_object(&remote_bucket, &old_remote.key);
+    // The public DELETE persists a free owner and enqueues its remote cleanup.
+    let client = run.dist.client(0)?;
+    client
+        .delete_object()
+        .bucket(&bucket)
+        .key(KEY)
+        .version_id(&old_version)
+        .send()
+        .await
+        .map_err(|error| format!("G-TIER {operation} public DELETE before remote gate failed: {error}; startup={startup}"))?;
+    timeout(TRANSITION_TIMEOUT, delete_gate.wait_until_entered())
+        .await
+        .map_err(|error| format!("G-TIER {operation} DELETE gate entry timed out: {error}; startup={startup}"))?
+        .map_err(|error| format!("G-TIER {operation} DELETE gate entry failed: {error}; startup={startup}"))?;
+    let delete_gate_entered_at = OffsetDateTime::now_utc();
+    let overlap_started = Instant::now();
+    let overlap_deadline = overlap_started + DELETE_OVERLAP_TIMEOUT;
+    let held = owners(&run, &bucket, &remote_bucket)?;
+    assert_eq!(held.len(), 2 * DRIVES_PER_NODE);
+    let old_owners: Vec<_> = held.iter().filter(|owner| owner.remote == old_remote).collect();
+    assert_eq!(old_owners.len(), DRIVES_PER_NODE);
+    assert!(
+        old_owners
+            .iter()
+            .all(|owner| owner.pool == SEED_POOL && owner.free && owner.version != old_version)
+    );
+    assert_eq!(old_owners.iter().map(|owner| &owner.version).collect::<BTreeSet<_>>().len(), 1);
+    assert_eq!(
+        held.iter()
+            .filter(|owner| {
+                owner.pool == SEED_POOL && !owner.free && owner.version == live_version && owner.remote == live_remote
+            })
+            .count(),
+        DRIVES_PER_NODE
+    );
+    assert_remote(&target, &old_remote, &old_body)?;
+    assert_remote(&target, &live_remote, &live_body)?;
+    let held_deletes = deletes(&target, &old_remote);
+    assert!(!held_deletes.is_empty());
+    assert!(held_deletes.iter().all(|request| {
+        request.prepared_response.is_none() && request.version_id.as_deref() == Some(old_remote.version.as_str())
+    }));
+    assert!(deletes(&target, &live_remote).is_empty());
+
+    let movement_started = OffsetDateTime::now_utc();
+    if decommission {
+        startup["released_at"] = json!(movement_started.format(&Rfc3339)?);
+        run.movement_barrier.release_movement()?;
+    }
+    let rebalance_id = timeout_at(overlap_deadline, async {
+        if decommission {
+            decommission_start.as_mut().await?;
+            Ok::<_, Box<dyn std::error::Error + Send + Sync>>(None)
+        } else {
+            Ok(Some(start_rebalance(&run.dist.cluster).await?))
+        }
+    })
+    .await
+    .map_err(|error| {
+        format!("G-TIER {operation} start acknowledgment timed out within DELETE overlap: {error}; startup={startup}")
+    })?
+    .map_err(|error| {
+        format!("G-TIER {operation} start acknowledgment failed within DELETE overlap: {error}; startup={startup}")
+    })?;
+    drop(decommission_start);
+    let movement_acknowledged_at = OffsetDateTime::now_utc();
+    let pending = timeout_at(overlap_deadline, pending_publication(&mut publication_events, movement_started))
+        .await
+        .map_err(|error| format!(
+            "G-TIER {operation} publication pending timed out within DELETE overlap: {error}; startup={startup}; lock_events={:?}; cleanup_barrier={:?}",
+            publication_events.cached(), cleanup_barrier.reached_observations()
+        ))?
+        .map_err(|error| format!("G-TIER {operation} publication pending failed within DELETE overlap: {error}"))?;
+    // Positive gate entry, exact physical ownership and the pending lock attempt
+    // establish the overlap; an absent log alone is not an ordering assertion.
+    assert_eq!(owners(&run, &bucket, &remote_bucket)?, held);
+    assert_remote(&target, &old_remote, &old_body)?;
+    let released_at = OffsetDateTime::now_utc();
+    let delete_overlap_elapsed_ms = overlap_started.elapsed().as_millis();
+    drop(delete_gate);
+    let acquired = timeout(DELETE_OVERLAP_TIMEOUT, acquired_publication(&mut publication_events, &pending))
+        .await
+        .map_err(|error| format!(
+            "G-TIER {operation} publication acquired after DELETE release timed out: {error}; pending={pending}; lock_events={:?}; cleanup_barrier={:?}",
+            publication_events.cached(), cleanup_barrier.reached_observations()
+        ))?
+        .map_err(|error| format!("G-TIER {operation} publication acquired after DELETE release failed: {error}"))?;
+    assert!(event_time(&pending)? >= movement_started && event_time(&pending)? < released_at);
+    assert!(
+        event_time(&acquired)? >= released_at,
+        "the same publication attempt must acquire after DELETE is released"
+    );
+    drop(publication_events);
+    let reached = cleanup_barrier
+        .wait_reached_during(
+            &run.dist.cluster,
+            match rebalance_id.as_deref() {
+                Some(id) => MovementOperation::Rebalance(id),
+                None => MovementOperation::Decommission(DECOMMISSION_POOL_ID),
+            },
+            MOVEMENT_COMPLETE_TIMEOUT,
+        )
+        .await
+        .map_err(|error| format!("G-TIER {operation} before-source-cleanup barrier failed: {error}"))?;
+    let published = owners(&run, &bucket, &remote_bucket)?;
+    assert!(
+        published
+            .iter()
+            .all(|owner| !owner.free && owner.version == live_version && owner.remote == live_remote)
+    );
+    let publication_pools: BTreeSet<_> = published.iter().map(|owner| owner.pool).collect();
+    assert!(publication_pools.contains(&SEED_POOL) && publication_pools.iter().any(|pool| *pool != SEED_POOL));
+    assert_eq!(published.len(), publication_pools.len() * DRIVES_PER_NODE);
+    assert!(target.stored_versions(&remote_bucket, &old_remote.key).is_empty());
+    assert_remote(&target, &live_remote, &live_body)?;
+    cleanup_barrier.release_movement()?;
+    cleanup_barrier.disarm()?;
+    let completion = match rebalance_id.as_deref() {
+        Some(id) => {
+            wait_for_rebalance_complete(&run.dist.cluster, id, MOVEMENT_COMPLETE_TIMEOUT).await?;
+            rebalance_status_json(&run.dist.cluster).await?
+        }
+        None => {
+            wait_for_decommission_complete(&run.dist.cluster, DECOMMISSION_POOL_ID, MOVEMENT_COMPLETE_TIMEOUT).await?;
+            decommission_status_json(&run.dist.cluster).await?
+        }
+    };
+    let settled = owners(&run, &bucket, &remote_bucket)?;
+    assert_eq!(settled.len(), DRIVES_PER_NODE);
+    assert_eq!(settled.iter().map(|owner| owner.pool).collect::<BTreeSet<_>>().len(), 1);
+    assert!(
+        settled
+            .iter()
+            .all(|owner| owner.pool != SEED_POOL && !owner.free && owner.remote == live_remote)
+    );
+    verify_live(&run, &bucket, &live_version, &old_version, &live_body).await?;
+    if decommission {
+        run.dist.remove_retired_pools_and_restart(&[SEED_POOL]).await?;
+    } else {
+        run.dist.restart_current_binary_gracefully().await?;
+    }
+    verify_live(&run, &bucket, &live_version, &old_version, &live_body).await?;
+    let restarted = owners(&run, &bucket, &remote_bucket)?;
+    assert_eq!(restarted.len(), DRIVES_PER_NODE);
+    assert_eq!(restarted.iter().map(|owner| owner.pool).collect::<BTreeSet<_>>().len(), 1);
+    assert!(
+        restarted
+            .iter()
+            .all(|owner| !owner.free && owner.version == live_version && owner.remote == live_remote)
+    );
+    assert_remote(&target, &live_remote, &live_body)?;
+    assert!(target.stored_versions(&remote_bucket, &old_remote.key).is_empty());
+    assert!(
+        deletes(&target, &live_remote).is_empty(),
+        "the live transitioned tuple must never receive DELETE"
+    );
+    let remote_deletes = deletes(&target, &old_remote);
+    assert!(!remote_deletes.is_empty());
+    assert!(
+        remote_deletes
+            .iter()
+            .all(|request| request.version_id.as_deref() == Some(old_remote.version.as_str()))
+    );
+    let successful = remote_deletes
+        .iter()
+        .filter_map(|request| request.prepared_response.as_ref())
+        .find(|response| response.status == 204)
+        .ok_or("old remote tuple never completed DELETE")?;
+    let deleted_at_ms = i128::from(
+        successful
+            .prepared_at_unix_millis
+            .ok_or("DELETE response omitted its timestamp")?,
+    );
+    assert!(deleted_at_ms >= released_at.unix_timestamp_nanos() / 1_000_000);
+    assert!(event_time(&acquired)?.unix_timestamp_nanos() / 1_000_000 >= deleted_at_ms);
+    assert_bucket_durability_strict(&run.dist.cluster, &[&bucket]).await?;
+    let strict_nodes = assert_process_durability_strict(&run.node_log_paths())?;
+    run.write_evidence(
+        &format!("g-tier-{operation}"),
+        &json!({
+            "gate": "G-TIER", "operation": operation, "rebalance_id": rebalance_id,
+            "startup": startup,
+            "delete_gate_entered_at": delete_gate_entered_at.format(&Rfc3339)?,
+            "movement_overlap_started_at": movement_started.format(&Rfc3339)?,
+            "movement_acknowledged_at": movement_acknowledged_at.format(&Rfc3339)?,
+            "delete_overlap_timeout_ms": DELETE_OVERLAP_TIMEOUT.as_millis(),
+            "delete_overlap_elapsed_ms": delete_overlap_elapsed_ms,
+            "remote_backend": "HTTP fake_s3_target with versioning enabled",
+            "ordering": "remote DELETE held, mover publication waiting, DELETE released, same publication acquired",
+            "bucket": bucket, "key": KEY, "bucket_durability": bucket_durability,
+            "process_durability_strict_nodes": strict_nodes,
+            "old_version": old_version, "old_remote": old_remote, "old_sha256": sha256_hex(&old_body),
+            "live_version": live_version, "live_remote": live_remote, "live_sha256": sha256_hex(&live_body),
+            "owners_while_delete_held": held, "publication_waiting": pending, "publication_acquired": acquired,
+            "held_delete_sequences": held_deletes.iter().map(|request| request.sequence).collect::<Vec<_>>(),
+            "delete_released_at": released_at.format(&Rfc3339)?, "remote_delete_completed_at_unix_millis": deleted_at_ms,
+            "barrier": reached, "owners_before_source_cleanup": published, "owners_after_movement": settled,
+            "owners_after_restart": restarted, "completion": completion,
+            "old_remote_delete_attempts": remote_deletes.len(), "live_remote_delete_attempts": 0,
+            "source_readback_nodes": run.dist.cluster.nodes.len(),
+        }),
+    )?;
+    Ok(())
+}
+
+#[tokio::test]
+async fn scanner_heal_topology_tier_free_owner_delete_overlaps_rebalance() -> TestResult {
+    tier_delete_during_movement(false).await
+}
+
+#[tokio::test]
+async fn scanner_heal_topology_tier_free_owner_delete_overlaps_decommission() -> TestResult {
+    tier_delete_during_movement(true).await
+}

crates/e2e_test/src/distributed/table_catalog_test.rs

@@ -0,0 +1,140 @@
+// 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, get_object_bytes, put_object, unique_bucket};
+use crate::common::{init_logging, signed_request};
+use http::{Method, StatusCode};
+use serde_json::{Value, json};
+
+async fn catalog_request(
+    dist: &DistCluster,
+    node: usize,
+    method: Method,
+    path: &str,
+    body: Option<Value>,
+    expected: StatusCode,
+) -> TestResult<Value> {
+    let response = signed_request(
+        method.clone(),
+        &format!("{}{path}", dist.cluster.nodes[node].url),
+        &dist.cluster.access_key,
+        &dist.cluster.secret_key,
+        body.map(|value| serde_json::to_vec(&value)).transpose()?,
+        Some("application/json"),
+    )
+    .await?;
+    let status = response.status();
+    let text = response.text().await?;
+    assert_eq!(status, expected, "node {node}: {method} {path}: {text}");
+    if text.is_empty() {
+        Ok(Value::Null)
+    } else {
+        Ok(serde_json::from_str(&text)?)
+    }
+}
+
+async fn assert_bucket_state(dist: &DistCluster, bucket: &str, enabled: bool) -> TestResult {
+    for node in 0..dist.cluster.nodes.len() {
+        let response =
+            catalog_request(dist, node, Method::GET, &format!("/iceberg/v1/buckets/{bucket}"), None, StatusCode::OK).await?;
+        assert_eq!(response["enabled"], enabled, "node {node}: {response}");
+        assert_eq!(response["table-bucket"], bucket);
+        let expected = if enabled { StatusCode::OK } else { StatusCode::BAD_REQUEST };
+        catalog_request(dist, node, Method::GET, &format!("/iceberg/v1/{bucket}/namespaces"), None, expected).await?;
+        assert_eq!(
+            get_object_bytes(&dist.client(node)?, bucket, "ordinary.txt").await?,
+            b"ordinary bucket object",
+            "node {node}: toggling table support changed an ordinary object"
+        );
+    }
+    Ok(())
+}
+
+async fn assert_table_state(dist: &DistCluster, table_path: &str, created: &Value, pointer: &Value) -> TestResult {
+    for node in 0..dist.cluster.nodes.len() {
+        let loaded = catalog_request(dist, node, Method::GET, table_path, None, StatusCode::OK).await?;
+        assert_eq!(loaded["metadata-location"], created["metadata-location"], "node {node}");
+        assert_eq!(loaded["metadata"], created["metadata"], "node {node}");
+        let actual_pointer =
+            catalog_request(dist, node, Method::GET, &format!("{table_path}/metadata-location"), None, StatusCode::OK).await?;
+        assert_eq!(&actual_pointer, pointer, "node {node}: catalog pointer changed");
+    }
+    Ok(())
+}
+
+#[tokio::test]
+async fn table_bucket_toggles_reload_cached_metadata_on_every_peer() -> TestResult {
+    init_logging();
+    let dist = DistCluster::start_with_env(DistLayout::FourByFour, &[("RUSTFS_TABLE_CATALOG_BACKING", "object")]).await?;
+    let bucket = unique_bucket("table-peer-reload");
+    dist.create_bucket(&bucket).await?;
+    put_object(&dist.client(0)?, &bucket, "ordinary.txt", b"ordinary bucket object".to_vec()).await?;
+    let bucket_path = format!("/iceberg/v1/buckets/{bucket}");
+
+    // Warm each node's false cache before changing the marker. Do not poll:
+    // successful toggles must notify peers before their response is returned.
+    assert_bucket_state(&dist, &bucket, false).await?;
+    for (node, method, enabled) in [(0, Method::PUT, true), (1, Method::DELETE, false), (2, Method::PUT, true)] {
+        let response = catalog_request(&dist, node, method, &bucket_path, None, StatusCode::OK).await?;
+        assert_eq!(response["enabled"], enabled);
+        assert_bucket_state(&dist, &bucket, enabled).await?;
+    }
+
+    let namespaces = format!("/iceberg/v1/{bucket}/namespaces");
+    catalog_request(&dist, 3, Method::POST, &namespaces, Some(json!({"namespace": ["lab"]})), StatusCode::OK).await?;
+    let tables = format!("{namespaces}/lab/tables");
+    let created = catalog_request(
+        &dist,
+        0,
+        Method::POST,
+        &tables,
+        Some(json!({
+            "name": "events",
+            "schema": {
+                "type": "struct",
+                "fields": [{"id": 1, "name": "id", "required": false, "type": "long"}]
+            },
+            "stage-create": false
+        })),
+        StatusCode::OK,
+    )
+    .await?;
+    let uuid = created["metadata"]["table-uuid"]
+        .as_str()
+        .ok_or("create response omitted table UUID")?;
+    assert!(!uuid::Uuid::parse_str(uuid)?.is_nil());
+    assert!(
+        created["metadata-location"]
+            .as_str()
+            .is_some_and(|location| location.starts_with(&format!("s3://{bucket}/")))
+    );
+    let table_path = format!("{tables}/events");
+    let pointer =
+        catalog_request(&dist, 3, Method::GET, &format!("{table_path}/metadata-location"), None, StatusCode::OK).await?;
+    assert_eq!(pointer["metadata-location"], created["metadata-location"]);
+    assert!(pointer["version-token"].as_str().is_some_and(|token| !token.is_empty()));
+    assert!(pointer["generation"].as_u64().is_some());
+    assert_eq!(pointer["warehouse-location"], created["metadata"]["location"]);
+    assert_table_state(&dist, &table_path, &created, &pointer).await?;
+
+    // A rejected toggle must preserve both peer protections and catalog state.
+    catalog_request(&dist, 1, Method::DELETE, &bucket_path, None, StatusCode::CONFLICT).await?;
+    assert_bucket_state(&dist, &bucket, true).await?;
+    assert_table_state(&dist, &table_path, &created, &pointer).await?;
+
+    catalog_request(&dist, 2, Method::DELETE, &table_path, None, StatusCode::NO_CONTENT).await?;
+    catalog_request(&dist, 3, Method::DELETE, &format!("{namespaces}/lab"), None, StatusCode::NO_CONTENT).await?;
+    // The cluster fixture owns all remaining catalog objects and data roots.
+    Ok(())
+}

crates/e2e_test/src/distributed/upgrade_test.rs

@@ -60,7 +60,7 @@ struct UpgradeSeed {
     iam_secret: &'static str,
 }
 
-fn source_binary() -> TestResult<PathBuf> {
+pub(super) 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)"

crates/e2e_test/src/dual_stack_differential/compare.rs

@@ -0,0 +1,512 @@
+// 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.
+
+//! What one stack answered, and how two answers are compared.
+//!
+//! An [`Observation`] keeps only what a client acts on: the HTTP status, the S3 error code, the
+//! allowlisted response headers and a few typed fields decoded from the response. The error
+//! message is kept for diagnostics and never compared: message wording is judged by the wire
+//! corpus, not here. The comparator is pure so it can be tested without a server.
+
+use serde::Serialize;
+use std::collections::{BTreeMap, BTreeSet};
+use std::path::{Path, PathBuf};
+
+/// Response headers compared exactly (after [`normalized_header_value`]).
+const HEADER_ALLOWLIST: [&str; 7] = [
+    "content-type",
+    "x-amz-version-id",
+    "etag",
+    "x-amz-delete-marker",
+    "content-range",
+    "accept-ranges",
+    "x-amz-request-charged",
+];
+
+/// Every response header with this prefix is compared as well.
+const HEADER_ALLOWLIST_PREFIX: &str = "x-amz-server-side-encryption";
+
+/// Placeholder for a field one side did not send.
+const ABSENT: &str = "<absent>";
+
+/// The lowercase header name if the comparator looks at this header.
+pub(crate) fn compared_header(name: &str) -> Option<String> {
+    let name = name.to_ascii_lowercase();
+    (HEADER_ALLOWLIST.contains(&name.as_str()) || name.starts_with(HEADER_ALLOWLIST_PREFIX)).then_some(name)
+}
+
+/// A version ID is minted per process, so its value never matches across two servers; only
+/// whether it is the literal `null` (an unversioned write) carries meaning.
+fn normalized_header_value(name: &str, value: &str) -> String {
+    if name == "x-amz-version-id" && value != "null" {
+        "<version-id>".to_owned()
+    } else {
+        value.to_owned()
+    }
+}
+
+/// One request's outcome as a client sees it.
+#[derive(Clone, Debug, Serialize)]
+pub(crate) struct Observation {
+    /// The S3 operation, named as in the `s3s::S3` trait (`GetObject`).
+    pub op: &'static str,
+    /// Which request of the script this was (`normal`, `negative`, or a setup label).
+    pub case: &'static str,
+    /// HTTP status; 0 when no response arrived.
+    pub status: u16,
+    /// The S3 error code, or the SDK's error kind when there was no service error.
+    pub error_code: Option<String>,
+    /// Allowlisted headers only, lowercase names, normalized values.
+    pub headers: BTreeMap<String, String>,
+    /// Typed values decoded from the response body, never free text.
+    pub fields: BTreeMap<String, String>,
+    /// Diagnostic only; never compared.
+    #[serde(skip)]
+    pub message: Option<String>,
+}
+
+impl Observation {
+    /// Builds an observation, keeping only the allowlisted headers.
+    pub(crate) fn new<'h>(
+        op: &'static str,
+        case: &'static str,
+        status: u16,
+        error_code: Option<String>,
+        headers: impl IntoIterator<Item = (&'h str, &'h str)>,
+    ) -> Self {
+        let headers = headers
+            .into_iter()
+            .filter_map(|(name, value)| compared_header(name).map(|name| (name, value)))
+            .map(|(name, value)| {
+                let value = normalized_header_value(&name, value);
+                (name, value)
+            })
+            .collect();
+        Self {
+            op,
+            case,
+            status,
+            error_code,
+            headers,
+            fields: BTreeMap::new(),
+            message: None,
+        }
+    }
+
+    pub(crate) fn with_field(mut self, name: &str, value: impl Into<String>) -> Self {
+        self.fields.insert(name.to_owned(), value.into());
+        self
+    }
+
+    pub(crate) fn is_success(&self) -> bool {
+        (200..300).contains(&self.status)
+    }
+}
+
+/// One field on which the two stacks answered differently.
+#[derive(Clone, Debug, PartialEq, Eq, Serialize)]
+pub(crate) struct Difference {
+    pub op: String,
+    pub case: String,
+    pub field: String,
+    pub a: String,
+    pub b: String,
+}
+
+/// Every difference between two transcripts of the same script, matched by `(op, case)`.
+pub(crate) fn compare(a: &[Observation], b: &[Observation]) -> Vec<Difference> {
+    let mut differences = Vec::new();
+    let (index_a, duplicates_a) = index(a);
+    let (index_b, duplicates_b) = index(b);
+    for (op, case) in duplicates_a.union(&duplicates_b) {
+        let count = |side: &[Observation]| {
+            side.iter()
+                .filter(|observation| (observation.op, observation.case) == (*op, *case))
+                .count()
+                .to_string()
+        };
+        differences.push(difference(op, case, "duplicate_case", count(a), count(b)));
+    }
+    let keys: BTreeSet<_> = index_a.keys().chain(index_b.keys()).copied().collect();
+    for key @ (op, case) in keys {
+        match (index_a.get(&key), index_b.get(&key)) {
+            (Some(a), Some(b)) => compare_one(a, b, &mut differences),
+            (a, b) => {
+                let presence = |side: Option<&&Observation>| if side.is_some() { "present" } else { ABSENT }.to_owned();
+                differences.push(difference(op, case, "observation", presence(a), presence(b)));
+            }
+        }
+    }
+    differences
+}
+
+type Key = (&'static str, &'static str);
+
+fn index(side: &[Observation]) -> (BTreeMap<Key, &Observation>, BTreeSet<Key>) {
+    let mut index = BTreeMap::new();
+    let mut duplicates = BTreeSet::new();
+    for observation in side {
+        let key = (observation.op, observation.case);
+        if index.insert(key, observation).is_some() {
+            duplicates.insert(key);
+        }
+    }
+    (index, duplicates)
+}
+
+fn difference(op: &str, case: &str, field: &str, a: String, b: String) -> Difference {
+    Difference {
+        op: op.to_owned(),
+        case: case.to_owned(),
+        field: field.to_owned(),
+        a,
+        b,
+    }
+}
+
+fn compare_one(a: &Observation, b: &Observation, differences: &mut Vec<Difference>) {
+    let mut field = |name: String, value_a: Option<&str>, value_b: Option<&str>| {
+        if value_a != value_b {
+            let show = |value: Option<&str>| value.unwrap_or(ABSENT).to_owned();
+            differences.push(difference(a.op, a.case, &name, show(value_a), show(value_b)));
+        }
+    };
+    field(
+        "status".to_owned(),
+        Some(a.status.to_string().as_str()),
+        Some(b.status.to_string().as_str()),
+    );
+    field("error_code".to_owned(), a.error_code.as_deref(), b.error_code.as_deref());
+    let headers: BTreeSet<_> = a.headers.keys().chain(b.headers.keys()).collect();
+    for name in headers {
+        field(
+            format!("header.{name}"),
+            a.headers.get(name).map(String::as_str),
+            b.headers.get(name).map(String::as_str),
+        );
+    }
+    let fields: BTreeSet<_> = a.fields.keys().chain(b.fields.keys()).collect();
+    for name in fields {
+        field(
+            format!("body.{name}"),
+            a.fields.get(name).map(String::as_str),
+            b.fields.get(name).map(String::as_str),
+        );
+    }
+}
+
+/// Operations observed on both sides.
+pub(crate) fn compared_ops(a: &[Observation], b: &[Observation]) -> BTreeSet<&'static str> {
+    let ops_b: BTreeSet<_> = b.iter().map(|observation| observation.op).collect();
+    a.iter()
+        .map(|observation| observation.op)
+        .filter(|op| ops_b.contains(op))
+        .collect()
+}
+
+/// `(op, case)` pairs observed on both sides.
+pub(crate) fn compared_cases(a: &[Observation], b: &[Observation]) -> usize {
+    let cases_b: BTreeSet<_> = b.iter().map(|observation| (observation.op, observation.case)).collect();
+    a.iter()
+        .map(|observation| (observation.op, observation.case))
+        .collect::<BTreeSet<_>>()
+        .intersection(&cases_b)
+        .count()
+}
+
+/// The JSON report written for every run.
+#[derive(Debug, Serialize)]
+pub(crate) struct Report {
+    pub schema: u32,
+    pub run: String,
+    pub stack_a: String,
+    pub stack_b: String,
+    pub region_a: Option<String>,
+    pub region_b: Option<String>,
+    pub compared_ops: usize,
+    pub compared_cases: usize,
+    pub differences: Vec<Difference>,
+}
+
+impl Report {
+    /// Writes the report as `<directory>/<run>.json` and returns its path.
+    pub(crate) fn write_to(&self, directory: &Path) -> std::io::Result<PathBuf> {
+        std::fs::create_dir_all(directory)?;
+        let path = directory.join(format!("{}.json", self.run));
+        let json = serde_json::to_vec_pretty(self).map_err(std::io::Error::other)?;
+        std::fs::write(&path, json)?;
+        Ok(path)
+    }
+}
+
+/// One line per difference, for a failing assertion.
+pub(crate) fn render(differences: &[Difference]) -> String {
+    differences
+        .iter()
+        .map(|d| format!("({}, {}, {}, a={:?}, b={:?})", d.op, d.case, d.field, d.a, d.b))
+        .collect::<Vec<_>>()
+        .join("\n")
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn ok(op: &'static str) -> Observation {
+        Observation::new(op, "normal", 200, None, [("content-type", "application/xml")])
+    }
+
+    fn fields(differences: &[Difference]) -> Vec<&str> {
+        differences.iter().map(|d| d.field.as_str()).collect()
+    }
+
+    #[test]
+    fn identical_transcripts_have_no_difference() {
+        let a = vec![
+            ok("HeadBucket"),
+            ok("GetBucketLocation").with_field("location_constraint", "us-east-1"),
+        ];
+        let b = a.clone();
+        assert_eq!(compare(&a, &b), Vec::new());
+        assert_eq!(compared_ops(&a, &b).len(), 2);
+    }
+
+    #[test]
+    fn a_status_difference_is_reported() {
+        let a = vec![ok("PutBucketAcl")];
+        let b = vec![Observation::new(
+            "PutBucketAcl",
+            "normal",
+            403,
+            None,
+            [("content-type", "application/xml")],
+        )];
+        assert_eq!(
+            compare(&a, &b),
+            vec![Difference {
+                op: "PutBucketAcl".into(),
+                case: "normal".into(),
+                field: "status".into(),
+                a: "200".into(),
+                b: "403".into(),
+            }]
+        );
+    }
+
+    #[test]
+    fn an_error_code_difference_is_reported() {
+        let a = vec![Observation::new("GetObject", "negative", 404, Some("NoSuchKey".into()), [])];
+        let b = vec![Observation::new(
+            "GetObject",
+            "negative",
+            404,
+            Some("NoSuchBucket".into()),
+            [],
+        )];
+        let differences = compare(&a, &b);
+        assert_eq!(fields(&differences), ["error_code"]);
+        assert_eq!((differences[0].a.as_str(), differences[0].b.as_str()), ("NoSuchKey", "NoSuchBucket"));
+    }
+
+    #[test]
+    fn an_error_code_on_one_side_only_is_reported() {
+        let a = vec![Observation::new("HeadBucket", "negative", 404, Some("NotFound".into()), [])];
+        let b = vec![Observation::new("HeadBucket", "negative", 404, None, [])];
+        let differences = compare(&a, &b);
+        assert_eq!(fields(&differences), ["error_code"]);
+        assert_eq!(differences[0].b, ABSENT);
+    }
+
+    #[test]
+    fn an_allowlisted_header_difference_is_reported() {
+        let a = vec![Observation::new("HeadObject", "normal", 200, None, [("ETag", "\"aa\"")])];
+        let b = vec![Observation::new("HeadObject", "normal", 200, None, [("etag", "\"bb\"")])];
+        assert_eq!(fields(&compare(&a, &b)), ["header.etag"]);
+    }
+
+    #[test]
+    fn a_header_sent_by_one_side_only_is_reported() {
+        let a = vec![Observation::new(
+            "DeleteObject",
+            "normal",
+            204,
+            None,
+            [("x-amz-delete-marker", "true")],
+        )];
+        let b = vec![Observation::new("DeleteObject", "normal", 204, None, [])];
+        let differences = compare(&a, &b);
+        assert_eq!(fields(&differences), ["header.x-amz-delete-marker"]);
+        assert_eq!((differences[0].a.as_str(), differences[0].b.as_str()), ("true", ABSENT));
+    }
+
+    #[test]
+    fn every_server_side_encryption_header_is_compared() {
+        let header = "x-amz-server-side-encryption-customer-key-md5";
+        let a = vec![Observation::new("PutObject", "sse-c", 200, None, [(header, "one")])];
+        let b = vec![Observation::new("PutObject", "sse-c", 200, None, [(header, "two")])];
+        assert_eq!(fields(&compare(&a, &b)), [format!("header.{header}")]);
+    }
+
+    #[test]
+    fn headers_outside_the_allowlist_are_not_compared() {
+        let a = vec![Observation::new(
+            "ListBuckets",
+            "normal",
+            200,
+            None,
+            [
+                ("date", "Mon"),
+                ("x-amz-request-id", "1"),
+                ("server", "a"),
+                ("content-length", "10"),
+            ],
+        )];
+        let b = vec![Observation::new(
+            "ListBuckets",
+            "normal",
+            200,
+            None,
+            [
+                ("date", "Tue"),
+                ("x-amz-request-id", "2"),
+                ("server", "b"),
+                ("content-length", "11"),
+            ],
+        )];
+        assert_eq!(compare(&a, &b), Vec::new());
+        assert!(a[0].headers.is_empty(), "non-allowlisted headers were kept: {:?}", a[0].headers);
+    }
+
+    #[test]
+    fn distinct_version_ids_are_not_a_difference() {
+        let a = vec![Observation::new(
+            "PutObject",
+            "normal",
+            200,
+            None,
+            [("x-amz-version-id", "0b1c")],
+        )];
+        let b = vec![Observation::new(
+            "PutObject",
+            "normal",
+            200,
+            None,
+            [("x-amz-version-id", "9f8e")],
+        )];
+        assert_eq!(compare(&a, &b), Vec::new());
+    }
+
+    #[test]
+    fn a_null_version_id_against_a_real_one_is_reported() {
+        let a = vec![Observation::new(
+            "PutObject",
+            "normal",
+            200,
+            None,
+            [("x-amz-version-id", "null")],
+        )];
+        let b = vec![Observation::new(
+            "PutObject",
+            "normal",
+            200,
+            None,
+            [("x-amz-version-id", "9f8e")],
+        )];
+        assert_eq!(fields(&compare(&a, &b)), ["header.x-amz-version-id"]);
+    }
+
+    #[test]
+    fn a_body_field_difference_is_reported() {
+        let a = vec![ok("GetBucketLocation").with_field("location_constraint", "us-east-1")];
+        let b = vec![ok("GetBucketLocation").with_field("location_constraint", "eu-west-1")];
+        let differences = compare(&a, &b);
+        assert_eq!(fields(&differences), ["body.location_constraint"]);
+        assert_eq!((differences[0].a.as_str(), differences[0].b.as_str()), ("us-east-1", "eu-west-1"));
+    }
+
+    #[test]
+    fn a_body_field_on_one_side_only_is_reported() {
+        let a = vec![ok("ListObjectsV2").with_field("keys", "[]")];
+        let b = vec![ok("ListObjectsV2")];
+        assert_eq!(fields(&compare(&a, &b)), ["body.keys"]);
+    }
+
+    #[test]
+    fn error_messages_are_never_compared() {
+        let mut a = Observation::new("GetObject", "negative", 404, Some("NoSuchKey".into()), []);
+        let mut b = a.clone();
+        a.message = Some("The specified key does not exist.".into());
+        b.message = Some("Object not found".into());
+        assert_eq!(compare(&[a], &[b]), Vec::new());
+    }
+
+    #[test]
+    fn a_case_observed_on_one_side_only_is_reported() {
+        let a = vec![ok("HeadBucket"), ok("ListBuckets")];
+        let b = vec![ok("HeadBucket")];
+        let differences = compare(&a, &b);
+        assert_eq!(fields(&differences), ["observation"]);
+        assert_eq!(differences[0].op, "ListBuckets");
+        assert_eq!(compared_ops(&a, &b).into_iter().collect::<Vec<_>>(), ["HeadBucket"]);
+    }
+
+    #[test]
+    fn a_case_recorded_twice_is_reported() {
+        let a = vec![ok("HeadBucket"), ok("HeadBucket")];
+        let b = vec![ok("HeadBucket")];
+        assert_eq!(fields(&compare(&a, &b)), ["duplicate_case"]);
+    }
+
+    #[test]
+    fn compared_ops_counts_operations_not_cases() {
+        let negative = Observation::new("HeadBucket", "negative", 404, Some("NotFound".into()), []);
+        let a = vec![ok("HeadBucket"), negative, ok("ListBuckets")];
+        let b = a.clone();
+        assert_eq!(compared_ops(&a, &b).len(), 2);
+    }
+
+    #[test]
+    fn the_report_lists_each_difference_as_op_field_a_b() {
+        let directory = std::env::temp_dir().join(format!("dual-stack-report-{}", uuid::Uuid::new_v4()));
+        let report = Report {
+            schema: 1,
+            run: "unit".into(),
+            stack_a: "legacy".into(),
+            stack_b: "legacy".into(),
+            region_a: None,
+            region_b: Some("eu-west-1".into()),
+            compared_ops: 1,
+            compared_cases: 1,
+            differences: compare(
+                &[ok("GetBucketLocation").with_field("location_constraint", "us-east-1")],
+                &[ok("GetBucketLocation").with_field("location_constraint", "eu-west-1")],
+            ),
+        };
+        let path = report.write_to(&directory).expect("write the report");
+        let json: serde_json::Value = serde_json::from_slice(&std::fs::read(&path).expect("read the report")).expect("parse");
+        let _ = std::fs::remove_dir_all(&directory);
+        assert_eq!(path.file_name().and_then(|name| name.to_str()), Some("unit.json"));
+        assert_eq!(json["compared_ops"], 1);
+        assert_eq!(
+            json["differences"],
+            serde_json::json!([{
+                "op": "GetBucketLocation",
+                "case": "normal",
+                "field": "body.location_constraint",
+                "a": "us-east-1",
+                "b": "eu-west-1",
+            }])
+        );
+    }
+}

crates/e2e_test/src/dual_stack_differential/mod.rs

@@ -0,0 +1,199 @@
+// 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.
+
+//! Dual-stack differential: one `rustfs` binary, two processes, one SDK script (rustfs/backlog#2740).
+//!
+//! Stack A always runs `RUSTFS_S3_STACK=legacy`. Stack B runs the value of `RUSTFS_E2E_STACK_B`
+//! (default `legacy` until the gateway stack is assembled), and `RUSTFS_E2E_STACK_B_REGION`, when
+//! set, starts B in another region so that a difference can be injected by hand. Both servers
+//! answer the script in `scenarios`, which covers every operation of `impl S3 for FS`; `compare`
+//! reports each `(op, field, a, b)` on which the answers differ, and every run writes its report
+//! to `<target>/dual-stack/<run>.json`. Two processes, not two embedded servers: bucket metadata
+//! is process-global, so two stacks in one process would read each other's buckets.
+
+mod compare;
+mod scenarios;
+
+use crate::common::{RustFSTestEnvironment, init_logging, workspace_root};
+use compare::{Difference, Observation, Report, compare, compared_cases, compared_ops, render};
+use std::path::PathBuf;
+use std::time::{SystemTime, UNIX_EPOCH};
+
+type BoxError = Box<dyn std::error::Error + Send + Sync>;
+
+const STACK_A: &str = "legacy";
+const ENV_STACK_B: &str = "RUSTFS_E2E_STACK_B";
+const ENV_STACK_B_REGION: &str = "RUSTFS_E2E_STACK_B_REGION";
+/// Becomes `gateway` once rustfs/backlog#2734 T2.5 assembles the gateway stack.
+const DEFAULT_STACK_B: &str = "legacy";
+/// The region the negative control gives stack A explicitly, so an inherited `RUSTFS_REGION`
+/// cannot erase the injected difference. Not `us-east-1`: that region answers GetBucketLocation
+/// with an empty LocationConstraint and re-creating an owned bucket with 200, so it would not
+/// report its own name and would add CreateBucket differences to the control.
+const CONTROL_REGION_A: &str = "us-west-2";
+const CONTROL_REGION_B: &str = "eu-west-1";
+/// `normal` cases the legacy stack must answer with a 2xx. The comparison alone cannot tell a
+/// script that exercises real behavior from one in which both stacks fail the same way. Five of
+/// the 73 fail on legacy by design of this script: PutBucketReplication (no registered remote
+/// target), GetBucketReplication (so no configuration), PutBucketNotificationConfiguration (no
+/// notification target configured), RestoreObject (the object is not archived) and
+/// GetObjectTorrent (legacy answers `NoSuchKey`).
+const MIN_NORMAL_SUCCESSES: usize = 68;
+
+struct PairConfig {
+    stack_b: String,
+    region_a: Option<String>,
+    region_b: Option<String>,
+}
+
+impl PairConfig {
+    fn from_env() -> Self {
+        Self {
+            stack_b: std::env::var(ENV_STACK_B).unwrap_or_else(|_| DEFAULT_STACK_B.to_owned()),
+            region_a: None,
+            region_b: std::env::var(ENV_STACK_B_REGION).ok(),
+        }
+    }
+}
+
+/// Starts one server on `stack`, overriding its region only when asked.
+async fn start_stack(stack: &str, region: Option<&str>) -> Result<RustFSTestEnvironment, BoxError> {
+    let mut env = RustFSTestEnvironment::new().await?;
+    let mut vars = vec![("RUSTFS_S3_STACK", stack)];
+    if let Some(region) = region {
+        vars.push(("RUSTFS_REGION", region));
+    }
+    env.start_rustfs_server_without_cleanup_with_env(&vars).await?;
+    Ok(env)
+}
+
+/// Starts stack A and stack B side by side, each with its own port and data directory.
+async fn spawn_pair(config: &PairConfig) -> Result<(RustFSTestEnvironment, RustFSTestEnvironment), BoxError> {
+    tokio::try_join!(
+        start_stack(STACK_A, config.region_a.as_deref()),
+        start_stack(&config.stack_b, config.region_b.as_deref())
+    )
+}
+
+fn report_directory() -> PathBuf {
+    let target = std::env::var_os("CARGO_TARGET_DIR").map_or_else(|| PathBuf::from("target"), PathBuf::from);
+    let target = if target.is_absolute() {
+        target
+    } else {
+        workspace_root().join(target)
+    };
+    target.join("dual-stack")
+}
+
+struct Run {
+    report: Report,
+    path: PathBuf,
+    a: Vec<Observation>,
+}
+
+/// Runs the script against both stacks concurrently and writes the report.
+async fn run_differential(label: &str, config: &PairConfig) -> Run {
+    let (server_a, server_b) = spawn_pair(config).await.expect("start stack A and stack B");
+    let (a, b) = tokio::join!(Box::pin(scenarios::run(&server_a)), Box::pin(scenarios::run(&server_b)));
+    let millis = SystemTime::now()
+        .duration_since(UNIX_EPOCH)
+        .map_or(0, |elapsed| elapsed.as_millis());
+    let report = Report {
+        schema: 1,
+        run: format!("{label}-{millis}-{}", std::process::id()),
+        stack_a: STACK_A.to_owned(),
+        stack_b: config.stack_b.clone(),
+        region_a: config.region_a.clone(),
+        region_b: config.region_b.clone(),
+        compared_ops: compared_ops(&a, &b).len(),
+        compared_cases: compared_cases(&a, &b),
+        differences: compare(&a, &b),
+    };
+    let path = report.write_to(&report_directory()).expect("write the dual-stack report");
+    Run { report, path, a }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[tokio::test]
+    async fn both_stacks_answer_every_operation_identically() {
+        init_logging();
+        let config = PairConfig::from_env();
+        let run = run_differential("self-comparison", &config).await;
+        let report = &run.report;
+
+        let mut observed: Vec<_> = run.a.iter().map(|observation| observation.op).collect();
+        observed.sort_unstable();
+        observed.dedup();
+        assert_eq!(
+            observed,
+            scenarios::OPERATIONS,
+            "the script must drive exactly the operations of impl S3 for FS"
+        );
+        assert_eq!(report.compared_ops, scenarios::OPERATIONS.len(), "report: {}", run.path.display());
+        assert!(
+            report.differences.is_empty(),
+            "stack {} and stack {} differ on {} of {} cases (report: {}):\n{}",
+            report.stack_a,
+            report.stack_b,
+            report.differences.len(),
+            report.compared_cases,
+            run.path.display(),
+            render(&report.differences)
+        );
+
+        let not_4xx: Vec<_> = run
+            .a
+            .iter()
+            .filter(|observation| observation.case == "negative" && !(400..500).contains(&observation.status))
+            .collect();
+        assert!(not_4xx.is_empty(), "negative cases must fail with a 4xx on {STACK_A}: {not_4xx:#?}");
+        let successes = run
+            .a
+            .iter()
+            .filter(|observation| observation.case == "normal" && observation.is_success())
+            .count();
+        assert!(
+            successes >= MIN_NORMAL_SUCCESSES,
+            "only {successes} normal cases succeeded on {STACK_A}: {:#?}",
+            run.a
+        );
+    }
+
+    #[tokio::test]
+    async fn differential_detects_an_injected_difference() {
+        init_logging();
+        let config = PairConfig {
+            stack_b: STACK_A.to_owned(),
+            region_a: Some(CONTROL_REGION_A.to_owned()),
+            region_b: Some(CONTROL_REGION_B.to_owned()),
+        };
+        let run = run_differential("injected-region", &config).await;
+        let expected = Difference {
+            op: "GetBucketLocation".to_owned(),
+            case: "normal".to_owned(),
+            field: "body.location_constraint".to_owned(),
+            a: CONTROL_REGION_A.to_owned(),
+            b: CONTROL_REGION_B.to_owned(),
+        };
+        assert!(
+            run.report.differences.contains(&expected),
+            "B started in {CONTROL_REGION_B} but GetBucketLocation was not reported (report: {}):\n{}",
+            run.path.display(),
+            render(&run.report.differences)
+        );
+    }
+}

crates/e2e_test/src/fake_s3_target/mod.rs

@@ -487,36 +487,43 @@ impl JournaledHeaders {
 struct ControlState {
     scripts: HashMap<Operation, VecDeque<FaultAction>>,
     keyed_scripts: HashMap<(Operation, String), VecDeque<FaultAction>>,
-    held_get: Option<HeldGetObject>,
+    held_get: Option<HeldObject>,
+    held_delete: Option<HeldObject>,
     requests: VecDeque<RequestRecord>,
     next_sequence: u64,
 }
 
 #[derive(Clone)]
-struct HeldGetObject {
+struct HeldObject {
     bucket: String,
     key: String,
     entered: watch::Sender<usize>,
     released: watch::Receiver<bool>,
 }
 
-/// Holds every GET of one object, including retries, until this guard is dropped.
-#[must_use = "dropping the guard releases the held GET requests"]
-pub struct GetObjectGate {
+/// Holds every matching operation, including retries, until this guard is dropped.
+#[must_use = "dropping the guard releases the held requests"]
+pub struct ObjectGate {
     control: Arc<Mutex<ControlState>>,
+    operation: Operation,
     entered: watch::Receiver<usize>,
     released: watch::Sender<bool>,
 }
 
-impl GetObjectGate {
+impl ObjectGate {
     pub async fn wait_until_entered(&mut self) -> Result<(), watch::error::RecvError> {
         self.entered.wait_for(|count| *count > 0).await.map(|_| ())
     }
 }
 
-impl Drop for GetObjectGate {
+impl Drop for ObjectGate {
     fn drop(&mut self) {
-        lock(&self.control).held_get = None;
+        let mut state = lock(&self.control);
+        match self.operation {
+            Operation::GetObject => state.held_get = None,
+            Operation::DeleteObject => state.held_delete = None,
+            _ => unreachable!("only GET and DELETE have object gates"),
+        }
         self.released.send_replace(true);
     }
 }
@@ -1108,24 +1115,40 @@ impl FakeS3Target {
     }
 
     /// Hold one exact bucket/key before any GET response can reach the client.
-    /// The fixture supports one live gate; request and connection deadlines still apply.
-    pub fn hold_get_object(&self, bucket: &str, key: &str) -> GetObjectGate {
+    /// The fixture supports one live GET gate; request and connection deadlines still apply.
+    pub fn hold_get_object(&self, bucket: &str, key: &str) -> ObjectGate {
+        self.hold_object_operation(Operation::GetObject, bucket, key)
+    }
+
+    /// Hold one exact bucket/key before DELETE changes any stored version.
+    /// The fixture supports one live DELETE gate; request deadlines still apply.
+    pub fn hold_delete_object(&self, bucket: &str, key: &str) -> ObjectGate {
+        self.hold_object_operation(Operation::DeleteObject, bucket, key)
+    }
+
+    fn hold_object_operation(&self, operation: Operation, bucket: &str, key: &str) -> ObjectGate {
         assert!(
             bucket.len() <= MAX_RETAINED_IDENTIFIER_BYTES && key.len() <= MAX_RETAINED_IDENTIFIER_BYTES,
-            "held GET identifiers exceed the fixture limit"
+            "held object identifiers exceed the fixture limit"
         );
         let mut state = lock(&self.control);
-        assert!(state.held_get.is_none(), "fake target already holds a GET gate");
+        let slot = match operation {
+            Operation::GetObject => &mut state.held_get,
+            Operation::DeleteObject => &mut state.held_delete,
+            _ => unreachable!("only GET and DELETE have object gates"),
+        };
+        assert!(slot.is_none(), "fake target already holds a {operation:?} gate");
         let (entered, entered_rx) = watch::channel(0);
         let (released, released_rx) = watch::channel(false);
-        state.held_get = Some(HeldGetObject {
+        *slot = Some(HeldObject {
             bucket: bucket.to_string(),
             key: key.to_string(),
             entered,
             released: released_rx,
         });
-        GetObjectGate {
+        ObjectGate {
             control: Arc::clone(&self.control),
+            operation,
             entered: entered_rx,
             released,
         }
@@ -2795,6 +2818,15 @@ impl S3 for FakeBackend {
         apply_non_body_fault(fault.as_ref(), &self.control).await?;
         let headers = req.headers;
         let input = req.input;
+        let held_delete = lock(&self.control)
+            .held_delete
+            .as_ref()
+            .filter(|held| held.bucket == input.bucket && held.key == input.key)
+            .cloned();
+        if let Some(mut held) = held_delete {
+            held.entered.send_modify(|count| *count += 1);
+            let _ = held.released.wait_for(|released| *released).await;
+        }
         let mut state = lock(&self.store);
         let Some(versioned) = state.buckets.get(&input.bucket).map(|bucket| bucket.versioned) else {
             return Err(s3s::s3_error!(NoSuchBucket, "bucket does not exist"));
@@ -3436,6 +3468,68 @@ mod tests {
         Ok(())
     }
 
+    #[tokio::test]
+    async fn delete_object_gate_retains_exact_version_until_release() -> Result<(), BoxError> {
+        let target = FakeS3Target::start().await?;
+        let bucket = "delete-gated-target";
+        target.create_bucket(bucket);
+        for key in ["held", "unrelated"] {
+            target.put_seed_object(bucket, key, Bytes::from_static(b"payload"), &SeedMetadata::default());
+        }
+        let versions = target.stored_versions(bucket, "held");
+        let mut gate = target.hold_delete_object(bucket, "held");
+        let client = client(&target);
+        let held_client = client.clone();
+        let version = versions[0].0.clone();
+        let mut requests = tokio::task::JoinSet::new();
+        requests.spawn(async move {
+            held_client
+                .delete_object()
+                .bucket(bucket)
+                .key("held")
+                .version_id(version)
+                .send()
+                .await
+        });
+        timeout(Duration::from_secs(2), gate.wait_until_entered()).await??;
+        assert_eq!(
+            target.stored_versions(bucket, "held"),
+            versions,
+            "the gate must precede the storage mutation"
+        );
+        assert!(requests.try_join_next().is_none());
+        let held = target
+            .requests()
+            .into_iter()
+            .find(|record| record.operation == Operation::DeleteObject && record.key.as_deref() == Some("held"))
+            .expect("the held DELETE must be journaled");
+        assert_eq!(held.version_id.as_deref(), Some(versions[0].0.as_str()));
+        assert_eq!(held.prepared_response, None);
+        client
+            .delete_object()
+            .bucket(bucket)
+            .key("unrelated")
+            .version_id(&target.stored_versions(bucket, "unrelated")[0].0)
+            .send()
+            .await?;
+        assert!(target.stored_versions(bucket, "unrelated").is_empty());
+        assert_eq!(target.stored_versions(bucket, "held"), versions);
+        drop(gate);
+        timeout(Duration::from_secs(2), requests.join_next())
+            .await?
+            .expect("held DELETE task")??;
+        assert!(target.stored_versions(bucket, "held").is_empty());
+        let released = target
+            .requests()
+            .into_iter()
+            .find(|record| record.sequence == held.sequence)
+            .and_then(|record| record.prepared_response)
+            .expect("released DELETE response");
+        assert_eq!(released.status, 204);
+        assert_eq!(released.version_id.as_deref(), Some(versions[0].0.as_str()));
+        Ok(())
+    }
+
     #[tokio::test]
     async fn object_lock_target_requires_a_checksum_on_locked_puts() -> Result<(), BoxError> {
         use aws_sdk_s3::error::ProvideErrorMetadata;

crates/e2e_test/src/heal_erasure_disk_rebuild_test.rs

@@ -21,7 +21,7 @@ mod tests {
     };
     use crate::common::{
         ClusterTopology, FAST_DATA_USAGE_SCANNER_ENV, RustFSTestClusterEnvironment, RustFSTestEnvironment, admin_request,
-        init_logging, rustfs_binary_path,
+        init_logging, local_http_client, rustfs_binary_path,
     };
     use crate::storage_api::RUSTFS_META_BUCKET;
     use aws_sdk_s3::{
@@ -53,6 +53,196 @@ mod tests {
         expected_pool_metadata: Option<VersionShardCensus>,
     }
 
+    fn pool_metadata_matches_republished_manifest(actual: &VersionShardCensus, expected: &VersionShardCensus) -> bool {
+        for data_dir in [&actual.data_dir, &expected.data_dir] {
+            if data_dir
+                .as_deref()
+                .and_then(|value| uuid::Uuid::parse_str(value).ok())
+                .is_none_or(|value| value.is_nil())
+            {
+                return false;
+            }
+        }
+        // The owning CAS republishes the same canonical payload with a fresh
+        // physical identity. Every shard and layout field must still match.
+        let mut republished = actual.clone();
+        republished.data_dir = expected.data_dir.clone();
+        republished.matches_manifest(expected)
+    }
+
+    fn pool_metadata_node_health_is_ready(status: u16, payload: &serde_json::Value) -> bool {
+        (200..300).contains(&status)
+            && payload["ready"] == true
+            && payload["details"]["poolMetadata"]["ready"] == true
+            && payload["details"]["storage"]["readQuorum"] == true
+            && payload["details"]["storage"]["writeQuorum"] == true
+    }
+
+    async fn all_nodes_pool_metadata_readiness(
+        cluster: &RustFSTestClusterEnvironment,
+        deadline: Instant,
+    ) -> (bool, serde_json::Value) {
+        let http = local_http_client();
+        let mut ready = !cluster.nodes.is_empty();
+        let mut observations = Vec::with_capacity(cluster.nodes.len());
+        for (node, environment) in cluster.nodes.iter().enumerate() {
+            let response = tokio::time::timeout_at(deadline, async {
+                let response = http
+                    .get(format!("{}/health/ready", environment.url))
+                    .timeout(Duration::from_secs(3))
+                    .send()
+                    .await?;
+                let status = response.status().as_u16();
+                let payload = response.json::<serde_json::Value>().await?;
+                Ok::<_, reqwest::Error>((status, payload))
+            })
+            .await;
+            let observation = match response {
+                Ok(Ok((status, payload))) => {
+                    let node_ready = pool_metadata_node_health_is_ready(status, &payload);
+                    ready &= node_ready;
+                    serde_json::json!({"node": node, "status": status, "ready": node_ready, "payload": payload})
+                }
+                Ok(Err(error)) => {
+                    ready = false;
+                    serde_json::json!({"node": node, "error": error.to_string()})
+                }
+                Err(_) => {
+                    ready = false;
+                    serde_json::json!({"node": node, "error": "pool metadata health deadline elapsed"})
+                }
+            };
+            observations.push(observation);
+        }
+        let deadline_elapsed = Instant::now() >= deadline;
+        (
+            ready && !deadline_elapsed,
+            serde_json::json!({"deadline_elapsed": deadline_elapsed, "nodes": observations}),
+        )
+    }
+
+    fn pool_metadata_manifest_fixture() -> VersionShardCensus {
+        VersionShardCensus {
+            version_id: None,
+            has_xl_meta: true,
+            data_dir: Some(uuid::Uuid::from_u128(1).to_string()),
+            erasure_index: Some(1),
+            erasure_distribution: Some(vec![1, 2, 3, 4]),
+            data_blocks: Some(2),
+            parity_blocks: Some(2),
+            expected_part_numbers: Default::default(),
+            present_part_fingerprints: Default::default(),
+            inline_data_fingerprint: Some(crate::chaos::PartShardFingerprint {
+                size: 7,
+                sha256: sha256_hex(b"payload"),
+            }),
+        }
+    }
+
+    #[test]
+    fn pool_metadata_republication_allows_only_valid_data_dir_change() {
+        let expected = pool_metadata_manifest_fixture();
+        let mut actual = expected.clone();
+        actual.data_dir = Some(uuid::Uuid::from_u128(2).to_string());
+        assert!(pool_metadata_matches_republished_manifest(&actual, &expected));
+        assert!(
+            !actual.matches_manifest(&expected),
+            "ordinary object manifests must retain their original identity"
+        );
+        for invalid in [
+            None,
+            Some(String::new()),
+            Some("invalid".to_string()),
+            Some(uuid::Uuid::nil().to_string()),
+        ] {
+            let mut changed = actual.clone();
+            changed.data_dir = invalid.clone();
+            assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+            let mut changed_expected = expected.clone();
+            changed_expected.data_dir = invalid;
+            assert!(!pool_metadata_matches_republished_manifest(&actual, &changed_expected));
+        }
+        let mut changed = actual.clone();
+        changed
+            .inline_data_fingerprint
+            .as_mut()
+            .expect("fixture has inline payload")
+            .sha256 = sha256_hex(b"corrupt");
+        assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+        let mut changed = actual.clone();
+        changed.erasure_index = Some(2);
+        assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+        let mut changed = actual.clone();
+        changed.erasure_distribution = Some(vec![2, 1, 3, 4]);
+        assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+        let mut changed = actual.clone();
+        changed.data_blocks = Some(1);
+        assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+        let mut changed = actual.clone();
+        changed.parity_blocks = Some(3);
+        assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+        let mut changed = actual.clone();
+        changed.version_id = Some(uuid::Uuid::from_u128(3).to_string());
+        assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+        let mut changed = actual;
+        changed.has_xl_meta = false;
+        assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+    }
+
+    #[test]
+    fn pool_metadata_republication_requires_exact_complete_part_fingerprints() {
+        let mut expected = pool_metadata_manifest_fixture();
+        expected.inline_data_fingerprint = None;
+        expected.expected_part_numbers.insert(1);
+        expected.present_part_fingerprints.insert(
+            1,
+            crate::chaos::PartShardFingerprint {
+                size: 7,
+                sha256: sha256_hex(b"payload"),
+            },
+        );
+        let mut actual = expected.clone();
+        actual.data_dir = Some(uuid::Uuid::from_u128(2).to_string());
+        assert!(pool_metadata_matches_republished_manifest(&actual, &expected));
+        let mut changed = actual.clone();
+        changed.present_part_fingerprints.remove(&1);
+        assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+        let mut changed = actual;
+        changed
+            .present_part_fingerprints
+            .get_mut(&1)
+            .expect("fixture has part 1")
+            .sha256 = sha256_hex(b"corrupt");
+        assert!(!pool_metadata_matches_republished_manifest(&changed, &expected));
+    }
+
+    #[test]
+    fn pool_metadata_node_health_requires_explicit_metadata_and_both_quorums() {
+        let expected = serde_json::json!({"ready": true, "details": {
+            "poolMetadata": {"ready": true},
+            "storage": {"readQuorum": true, "writeQuorum": true}
+        }});
+        assert!(pool_metadata_node_health_is_ready(200, &expected));
+        assert!(!pool_metadata_node_health_is_ready(503, &expected));
+        assert!(!pool_metadata_node_health_is_ready(200, &serde_json::json!({"ready": true})));
+        for path in [
+            "/ready",
+            "/details/poolMetadata/ready",
+            "/details/storage/readQuorum",
+            "/details/storage/writeQuorum",
+        ] {
+            for value in [
+                serde_json::Value::Bool(false),
+                serde_json::Value::Null,
+                serde_json::json!("true"),
+            ] {
+                let mut changed = expected.clone();
+                *changed.pointer_mut(path).expect("fixture contains required health field") = value;
+                assert!(!pool_metadata_node_health_is_ready(200, &changed), "required health field {path}");
+            }
+        }
+    }
+
     #[derive(serde::Deserialize)]
     struct EvidenceBuild {
         sha256: String,
@@ -2232,6 +2422,7 @@ mod tests {
             .and_then(|value| value.parse::<u64>().ok())
             .unwrap_or(180);
         let heal_deadline = Instant::now() + Duration::from_secs(heal_timeout_secs);
+        let mut last_pool_metadata_readiness = serde_json::Value::Null;
         loop {
             if restart_recovery_admin_after_failure && !recovery_admin_task_restarted {
                 let task_state = timeout(
@@ -2273,14 +2464,23 @@ mod tests {
                 let outage_census = census_object_version_on_disk(&replaced_disk, bucket, &outage_key, None)?;
                 let pool_metadata_matches = match &expected_pool_metadata {
                     Some(expected) => {
-                        census_object_version_on_disk(&replaced_disk, RUSTFS_META_BUCKET, POOL_METADATA_OBJECT, None)?
-                            .matches_manifest(expected)
+                        let actual =
+                            census_object_version_on_disk(&replaced_disk, RUSTFS_META_BUCKET, POOL_METADATA_OBJECT, None)?;
+                        pool_metadata_matches_republished_manifest(&actual, expected)
                     }
                     None => true,
                 };
+                let pool_metadata_ready = if expected_pool_metadata.is_some() && pool_metadata_matches {
+                    let (ready, observation) = all_nodes_pool_metadata_readiness(&cluster, heal_deadline).await;
+                    last_pool_metadata_readiness = observation;
+                    ready
+                } else {
+                    expected_pool_metadata.is_none()
+                };
                 if matching == expected_manifests.len()
                     && (!outage_target_manifest_required || outage_census.is_complete())
                     && pool_metadata_matches
+                    && pool_metadata_ready
                 {
                     break;
                 }
@@ -2290,6 +2490,15 @@ mod tests {
                 let outage_census = census_object_version_on_disk(&replaced_disk, bucket, &outage_key, None)?;
                 let pool_metadata =
                     census_object_version_on_disk(&replaced_disk, RUSTFS_META_BUCKET, POOL_METADATA_OBJECT, None)?;
+                // This post-deadline observation is diagnostic only. It cannot
+                // make an expired recovery attempt satisfy the original budget.
+                let pool_metadata_readiness = if expected_pool_metadata.is_some() {
+                    let (_, observation) =
+                        all_nodes_pool_metadata_readiness(&cluster, Instant::now() + Duration::from_secs(3)).await;
+                    serde_json::json!({"last_eligible_check": last_pool_metadata_readiness, "timeout_check": observation})
+                } else {
+                    serde_json::Value::Null
+                };
                 let final_status = signed_admin_post(&status_url, None, &cluster.access_key, &cluster.secret_key)
                     .await
                     .unwrap_or_else(|err| format!("status request failed: {err}"));
@@ -2313,7 +2522,7 @@ mod tests {
                     Err(_) => "replacement status request exceeded 5s diagnostic budget".to_string(),
                 };
                 return Err(format!(
-                    "root heal did not recover after {interruption_kind} within {heal_timeout_secs}s: baseline={matching}/{}, outage={outage_census:?}, pool_metadata={pool_metadata:?}, status={final_status}, task_status={task_status}, pre_interrupt_status={pre_interrupt_status}, pre_heal_replacement={pre_heal_replacement}, pre_interrupt_replacement={pre_interrupt_replacement}, replacement_status={replacement_status}",
+                    "root heal did not recover after {interruption_kind} within {heal_timeout_secs}s: baseline={matching}/{}, outage={outage_census:?}, pool_metadata_expected={expected_pool_metadata:?}, pool_metadata={pool_metadata:?}, pool_metadata_readiness={pool_metadata_readiness}, status={final_status}, task_status={task_status}, pre_interrupt_status={pre_interrupt_status}, pre_heal_replacement={pre_heal_replacement}, pre_interrupt_replacement={pre_interrupt_replacement}, replacement_status={replacement_status}",
                     expected_manifests.len()
                 )
                 .into());

crates/e2e_test/src/lib.rs

@@ -31,6 +31,12 @@ pub mod chaos;
 #[cfg(test)]
 pub mod fake_s3_target;
 
+// Dual-stack differential (rustfs/backlog#2740): one binary, two processes on
+// different RUSTFS_S3_STACK values, the same SDK script for every S3 operation,
+// compared on status, error code, allowlisted headers and typed body fields.
+#[cfg(test)]
+mod dual_stack_differential;
+
 // On-demand migration (backlog#2147): shared two-server environment, admin
 // wrappers, and the harness self-test (backlog#2151). Behavior scenarios are
 // added by later ODM tasks.

crates/e2e_test/src/list_objects_v2_pagination_test.rs

@@ -1129,12 +1129,14 @@ mod tests {
     /// and the marker plus the cursor tag.
     ///
     /// Bug Reference: backlog#1047
-    /// The V2 continuation token appends a `[rustfs_cache:...]` cursor tag to the
-    /// last returned key. The orchestration layer passed that raw string to
-    /// `forward_past`, so any key whose byte after the shared stem sorts below
-    /// `[` (0x5B) — `.`, `/`, `-`, digits, uppercase — was silently dropped on
-    /// the next page: with keys `a`, `a.txt`, `zz` and max_keys=1, page 2
-    /// returned `zz` and `a.txt` was never listed.
+    /// Earlier releases appended a `[rustfs_cache:...]` cursor tag to the last
+    /// returned key inside the V2 continuation token. The orchestration layer
+    /// passed that raw string to `forward_past`, so any key whose byte after the
+    /// shared stem sorts below `[` (0x5B) — `.`, `/`, `-`, digits, uppercase —
+    /// was silently dropped on the next page: with keys `a`, `a.txt`, `zz` and
+    /// max_keys=1, page 2 returned `zz` and `a.txt` was never listed. The token
+    /// is now base64 of the plain key, and this test keeps guarding the
+    /// continuation boundary.
     #[tokio::test]
     async fn test_list_objects_v2_continuation_keeps_keys_after_marker_stem() {
         init_logging();

crates/e2e_test/src/protocols/webdav_core.rs

@@ -32,7 +32,9 @@
 //! Advisory: <https://github.com/rustfs/rustfs/security/advisories/GHSA-3p3x-734c-h5vx>
 
 use crate::common::rustfs_binary_path_with_features;
-use crate::common::{AdminTransport, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user_via};
+use crate::common::{
+    AdminTransport, admin_add_canned_policy_via, admin_attach_user_policy_via, admin_create_user_via, signed_s3_request,
+};
 use crate::common::{FAST_DATA_USAGE_SCANNER_ENV, admin_request};
 use crate::protocols::test_env::{DEFAULT_ACCESS_KEY, DEFAULT_SECRET_KEY, ProtocolTestEnvironment};
 use anyhow::Result;
@@ -774,6 +776,81 @@ async fn test_webdav_core_operations_direct() -> Result<()> {
     test_webdav_core_operations().await
 }
 
+#[tokio::test]
+async fn test_webdav_content_type_directory_marker() -> Result<()> {
+    let env = ProtocolTestEnvironment::new().map_err(|e| anyhow::anyhow!("{}", e))?;
+    let admin_base_url = format!("http://{}", S3_TEST_ADDRESS);
+    let binary_path = rustfs_binary_path_with_features(Some("webdav"));
+    let mut server_process = Command::new(&binary_path)
+        .arg("--address")
+        .arg(S3_TEST_ADDRESS)
+        .env("RUSTFS_CONSOLE_ENABLE", "false")
+        .env("RUSTFS_WEBDAV_ENABLE", "true")
+        .env("RUSTFS_WEBDAV_ADDRESS", WEBDAV_ADDRESS)
+        .env("RUSTFS_WEBDAV_TLS_ENABLED", "false")
+        .arg(&env.temp_dir)
+        .spawn()?;
+
+    let result = async {
+        ProtocolTestEnvironment::wait_for_port_ready(WEBDAV_PORT, 30)
+            .await
+            .map_err(|e| anyhow::anyhow!("{}", e))?;
+
+        let client = create_client();
+        let bucket_name = "webdav-content-type-bucket";
+        let directory_name = "content-type-directory";
+        let webdav_base_url = format!("http://{}", WEBDAV_ADDRESS);
+        let mkcol_bucket = client
+            .request(
+                reqwest::Method::from_bytes(b"MKCOL").expect("MKCOL is a valid HTTP method"),
+                format!("{}/{}", webdav_base_url, bucket_name),
+            )
+            .header("Authorization", basic_auth_header())
+            .send()
+            .await?;
+        anyhow::ensure!(
+            mkcol_bucket.status().is_success() || mkcol_bucket.status().as_u16() == 201,
+            "MKCOL bucket should succeed, got: {}",
+            mkcol_bucket.status()
+        );
+
+        let mkcol_directory = client
+            .request(
+                reqwest::Method::from_bytes(b"MKCOL").expect("MKCOL is a valid HTTP method"),
+                format!("{}/{}/{}", webdav_base_url, bucket_name, directory_name),
+            )
+            .header("Authorization", basic_auth_header())
+            .send()
+            .await?;
+        anyhow::ensure!(
+            mkcol_directory.status().is_success() || mkcol_directory.status().as_u16() == 201,
+            "MKCOL directory should succeed, got: {}",
+            mkcol_directory.status()
+        );
+
+        let head_url = format!("{}/{}/{}/", admin_base_url, bucket_name, directory_name);
+        let head = signed_s3_request(http::Method::HEAD, &head_url, None, None, DEFAULT_ACCESS_KEY, DEFAULT_SECRET_KEY)
+            .await
+            .map_err(|error| anyhow::anyhow!("{error}"))?;
+        anyhow::ensure!(head.status().is_success(), "HEAD directory marker should succeed, got: {}", head.status());
+        let content_type = head
+            .headers()
+            .get("content-type")
+            .and_then(|value| value.to_str().ok())
+            .unwrap_or_default();
+        anyhow::ensure!(
+            content_type == "application/x-directory",
+            "WebDAV directory marker must preserve its content type, got: {content_type}"
+        );
+        Ok(())
+    }
+    .await;
+
+    let _ = server_process.kill().await;
+    let _ = server_process.wait().await;
+    result
+}
+
 /// Extract the numeric value of the first occurrence of a DAV property in a
 /// PROPFIND multistatus body. Namespace-prefix agnostic; a self-closing
 /// (unreported) property yields None.

crates/e2e_test/src/quota_test.rs

@@ -19,6 +19,19 @@ use http::{Method, StatusCode};
 use tokio::time::{Duration, sleep, timeout};
 use tracing::{debug, info};
 
+/// A successful reporting response can still have unknown scanner usage.
+fn reported_quota_usage(response: &serde_json::Value) -> std::io::Result<Option<u64>> {
+    let value = response
+        .get("current_usage")
+        .ok_or_else(|| std::io::Error::new(std::io::ErrorKind::InvalidData, "quota response is missing current_usage"))?;
+    if value.is_null() {
+        return Ok(None);
+    }
+    value.as_u64().map(Some).ok_or_else(|| {
+        std::io::Error::new(std::io::ErrorKind::InvalidData, "quota current_usage is not an unsigned integer or null")
+    })
+}
+
 /// Test environment setup for quota tests
 pub struct QuotaTestEnv {
     pub env: RustFSTestEnvironment,
@@ -87,35 +100,69 @@ impl QuotaTestEnv {
         }
     }
 
-    pub async fn get_bucket_quota_stats(&self) -> Result<serde_json::Value, Box<dyn std::error::Error + Send + Sync>> {
-        let url = format!("{}/rustfs/admin/v3/quota-stats/{}", self.env.url, self.bucket_name);
-        let response = awscurl_get(&url, &self.env.access_key, &self.env.secret_key).await?;
-
-        if response.contains("error") {
-            Err(format!("Failed to get quota stats: {}", response).into())
-        } else {
-            Ok(serde_json::from_str(&response)?)
+    async fn wait_for_usage_response(
+        &self,
+        method: Method,
+        path: &str,
+        body: Option<String>,
+        expected_usage: u64,
+    ) -> Result<serde_json::Value, Box<dyn std::error::Error + Send + Sync>> {
+        let convergence = async {
+            loop {
+                let (status, response) = admin_request(
+                    &self.env.url,
+                    method.clone(),
+                    path,
+                    body.clone(),
+                    &self.env.access_key,
+                    &self.env.secret_key,
+                )
+                .await?;
+                if status.is_success() {
+                    let response: serde_json::Value = serde_json::from_str(&response)?;
+                    if let Some(usage) = reported_quota_usage(&response)?
+                        && usage == expected_usage
+                    {
+                        return Ok::<_, Box<dyn std::error::Error + Send + Sync>>(response);
+                    }
+                } else if status != StatusCode::SERVICE_UNAVAILABLE {
+                    return Err(format!("quota usage request failed for {path}: {status} {response}").into());
+                }
+                sleep(Duration::from_millis(100)).await;
+            }
+        };
+        match timeout(Duration::from_secs(30), convergence).await {
+            Ok(result) => result,
+            Err(_) => Err(format!("quota usage did not converge to {expected_usage} for {path} within 30 seconds").into()),
         }
     }
 
-    pub async fn check_bucket_quota(
+    async fn wait_for_bucket_quota_stats(
+        &self,
+        bucket: &str,
+        expected_usage: u64,
+    ) -> Result<serde_json::Value, Box<dyn std::error::Error + Send + Sync>> {
+        self.wait_for_usage_response(Method::GET, &format!("/rustfs/admin/v3/quota-stats/{bucket}"), None, expected_usage)
+            .await
+    }
+
+    async fn wait_for_bucket_quota_check(
         &self,
         operation_type: &str,
         operation_size: u64,
+        expected_usage: u64,
     ) -> Result<serde_json::Value, Box<dyn std::error::Error + Send + Sync>> {
-        let url = format!("{}/rustfs/admin/v3/quota-check/{}", self.env.url, self.bucket_name);
         let check_request = serde_json::json!({
             "operation_type": operation_type,
             "operation_size": operation_size
         });
-
-        let response = awscurl_post(&url, &check_request.to_string(), &self.env.access_key, &self.env.secret_key).await?;
-
-        if response.contains("error") {
-            Err(format!("Failed to check quota: {}", response).into())
-        } else {
-            Ok(serde_json::from_str(&response)?)
-        }
+        self.wait_for_usage_response(
+            Method::POST,
+            &format!("/rustfs/admin/v3/quota-check/{}", self.bucket_name),
+            Some(check_request.to_string()),
+            expected_usage,
+        )
+        .await
     }
 
     pub async fn upload_object(&self, key: &str, size_bytes: usize) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
@@ -145,25 +192,9 @@ impl QuotaTestEnv {
         }
     }
 
-    pub async fn get_bucket_usage(&self) -> Result<u64, Box<dyn std::error::Error + Send + Sync>> {
-        let stats = self.get_bucket_quota_stats().await?;
-        Ok(stats.get("current_usage").and_then(|v| v.as_u64()).unwrap_or(0))
-    }
-
     async fn wait_for_bucket_usage(&self, expected: u64) -> Result<u64, Box<dyn std::error::Error + Send + Sync>> {
-        let convergence = async {
-            loop {
-                let usage = self.get_bucket_usage().await?;
-                if usage == expected {
-                    return Ok::<u64, Box<dyn std::error::Error + Send + Sync>>(usage);
-                }
-                sleep(Duration::from_millis(100)).await;
-            }
-        };
-        match timeout(Duration::from_secs(30), convergence).await {
-            Ok(result) => result,
-            Err(_) => Err(format!("bucket usage did not converge to {expected} bytes within 30 seconds").into()),
-        }
+        let stats = self.wait_for_bucket_quota_stats(&self.bucket_name, expected).await?;
+        reported_quota_usage(&stats)?.ok_or_else(|| "converged quota usage is missing".into())
     }
 
     pub async fn set_bucket_quota_for(
@@ -171,8 +202,6 @@ impl QuotaTestEnv {
         bucket: &str,
         quota_bytes: u64,
     ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-        self.wait_for_quota_usage_for(bucket).await?;
-
         let quota_path = format!("/rustfs/admin/v3/quota/{bucket}");
         let quota_config = serde_json::json!({
             "quota": quota_bytes,
@@ -206,46 +235,6 @@ impl QuotaTestEnv {
         }
     }
 
-    pub async fn wait_for_quota_usage_for(&self, bucket: &str) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
-        let stats_path = format!("/rustfs/admin/v3/quota-stats/{bucket}");
-        let readiness = async {
-            loop {
-                let (status, response) =
-                    admin_request(&self.env.url, Method::GET, &stats_path, None, &self.env.access_key, &self.env.secret_key)
-                        .await?;
-                if status.is_success() {
-                    return Ok::<(), Box<dyn std::error::Error + Send + Sync>>(());
-                }
-                if status != StatusCode::SERVICE_UNAVAILABLE {
-                    return Err(format!("quota usage readiness failed for {bucket}: {status} {response}").into());
-                }
-                sleep(Duration::from_secs(1)).await;
-            }
-        };
-        match timeout(Duration::from_secs(30), readiness).await {
-            Ok(result) => result,
-            Err(_) => Err(format!("quota usage did not become authoritative for {bucket} within 30 seconds").into()),
-        }
-    }
-
-    /// Get bucket quota statistics for specific bucket
-    pub async fn get_bucket_quota_stats_for(
-        &self,
-        bucket: &str,
-    ) -> Result<serde_json::Value, Box<dyn std::error::Error + Send + Sync>> {
-        debug!("Getting quota stats for bucket: {}", bucket);
-
-        let url = format!("{}/rustfs/admin/v3/quota-stats/{}", self.env.url, bucket);
-        let response = awscurl_get(&url, &self.env.access_key, &self.env.secret_key).await?;
-
-        if response.contains("error") {
-            Err(format!("Failed to get quota stats: {}", response).into())
-        } else {
-            let stats: serde_json::Value = serde_json::from_str(&response)?;
-            Ok(stats)
-        }
-    }
-
     /// Upload an object to specific bucket
     pub async fn upload_object_to_bucket(
         &self,
@@ -270,6 +259,132 @@ impl QuotaTestEnv {
     }
 }
 
+#[cfg(test)]
+mod usage_response_tests {
+    use super::*;
+    use tokio::io::{AsyncReadExt, AsyncWriteExt};
+    use tokio::net::TcpListener;
+    use tokio::task::JoinSet;
+
+    #[test]
+    fn unknown_usage_is_not_zero() {
+        assert_eq!(
+            reported_quota_usage(&serde_json::json!({"current_usage": null})).expect("valid unknown usage"),
+            None
+        );
+        assert_eq!(
+            reported_quota_usage(&serde_json::json!({"current_usage": 0})).expect("known zero usage"),
+            Some(0)
+        );
+    }
+
+    #[test]
+    fn missing_or_invalid_usage_is_not_readiness() {
+        for response in [
+            serde_json::json!({}),
+            serde_json::json!({"current_usage": -1}),
+            serde_json::json!({"current_usage": "0"}),
+            serde_json::json!({"current_usage": 1.5}),
+            serde_json::json!({"current_usage": false}),
+        ] {
+            let error = reported_quota_usage(&response).expect_err("malformed usage must fail");
+            assert_eq!(error.kind(), std::io::ErrorKind::InvalidData);
+        }
+    }
+
+    async fn mock_environment(
+        responses: Vec<(StatusCode, &'static str)>,
+    ) -> Result<(QuotaTestEnv, JoinSet<()>), Box<dyn std::error::Error + Send + Sync>> {
+        let listener = TcpListener::bind((std::net::Ipv4Addr::LOCALHOST, 0)).await?;
+        let address = listener.local_addr()?;
+        let mut tasks = JoinSet::new();
+        tasks.spawn(async move {
+            for (status, body) in responses {
+                let (mut stream, _) = listener.accept().await.expect("accept quota reporting request");
+                let mut request = Vec::new();
+                while !request.windows(4).any(|part| part == b"\r\n\r\n") {
+                    let mut buffer = [0_u8; 1024];
+                    let read = stream.read(&mut buffer).await.expect("read reporting request");
+                    assert!(read > 0 && request.len() + read <= 8192, "bounded HTTP request headers");
+                    request.extend_from_slice(&buffer[..read]);
+                }
+                let response = format!(
+                    "HTTP/1.1 {} Test\r\nContent-Type: application/json\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{}",
+                    status.as_u16(),
+                    body.len(),
+                    body
+                );
+                stream
+                    .write_all(response.as_bytes())
+                    .await
+                    .expect("write quota reporting response");
+                stream.shutdown().await.expect("close reporting response");
+            }
+        });
+        let mut env = RustFSTestEnvironment::new().await?;
+        env.url = format!("http://{address}");
+        let client = env.create_s3_client();
+        Ok((
+            QuotaTestEnv {
+                env,
+                client,
+                bucket_name: "quota-reporting-fixture".to_owned(),
+            },
+            tasks,
+        ))
+    }
+
+    #[tokio::test]
+    async fn reporting_waits_for_known_matching_usage_without_selecting_allowed()
+    -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+        let (env, mut tasks) = mock_environment(vec![
+            (StatusCode::SERVICE_UNAVAILABLE, "unavailable"),
+            (StatusCode::OK, r#"{"current_usage":null}"#),
+            (StatusCode::OK, r#"{"current_usage":1}"#),
+            (StatusCode::OK, r#"{"current_usage":0,"allowed":false}"#),
+        ])
+        .await?;
+        let response = env.wait_for_usage_response(Method::GET, "/reporting", None, 0).await?;
+        assert_eq!(response["current_usage"], 0);
+        assert_eq!(response["allowed"], false, "readiness must not select a desired admission outcome");
+        timeout(Duration::from_secs(1), tasks.join_next())
+            .await?
+            .expect("reporting mock task")
+            .expect("all responses consumed");
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn reporting_rejects_malformed_success_instead_of_retrying() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+        let (env, mut tasks) = mock_environment(vec![(StatusCode::OK, r#"{"current_usage":"unknown"}"#)]).await?;
+        let error = env
+            .wait_for_usage_response(Method::GET, "/reporting", None, 0)
+            .await
+            .expect_err("malformed usage must fail");
+        assert!(error.to_string().contains("not an unsigned integer or null"));
+        timeout(Duration::from_secs(1), tasks.join_next())
+            .await?
+            .expect("reporting mock task")
+            .expect("response consumed");
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn reporting_does_not_retry_authorization_failure() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+        let (env, mut tasks) = mock_environment(vec![(StatusCode::FORBIDDEN, "denied")]).await?;
+        let error = env
+            .wait_for_usage_response(Method::GET, "/reporting", None, 0)
+            .await
+            .expect_err("authorization error must fail");
+        assert!(error.to_string().contains("403"));
+        timeout(Duration::from_secs(1), tasks.join_next())
+            .await?
+            .expect("reporting mock task")
+            .expect("response consumed");
+        Ok(())
+    }
+}
+
 #[cfg(test)]
 mod integration_tests {
     use super::*;
@@ -482,7 +597,7 @@ mod integration_tests {
         env.upload_object("file2.txt", 256 * 1024).await?;
 
         // Check usage
-        let usage = env.get_bucket_usage().await?;
+        let usage = env.wait_for_bucket_usage((512 + 256) * 1024).await?;
         assert_eq!(usage, (512 + 256) * 1024);
 
         // Delete a file
@@ -517,7 +632,7 @@ mod integration_tests {
         env.upload_object("file2.txt", 512 * 1024).await?;
 
         // Get detailed quota statistics
-        let stats = env.get_bucket_quota_stats().await?;
+        let stats = env.wait_for_bucket_quota_stats(&env.bucket_name, (1024 + 512) * 1024).await?;
 
         assert_eq!(stats.get("bucket").unwrap().as_str().unwrap(), env.bucket_name);
         assert_eq!(stats.get("quota_limit").unwrap().as_u64().unwrap(), 2 * 1024 * 1024);
@@ -546,22 +661,22 @@ mod integration_tests {
         env.upload_object("existing_file.txt", 512 * 1024).await?;
 
         // Check if we can upload another 512KB (should succeed, exactly fill quota)
-        let check_result = env.check_bucket_quota("PUT", 512 * 1024).await?;
+        let check_result = env.wait_for_bucket_quota_check("PUT", 512 * 1024, 512 * 1024).await?;
         assert!(check_result.get("allowed").unwrap().as_bool().unwrap());
         assert_eq!(check_result.get("remaining_quota").unwrap().as_u64().unwrap(), 0);
 
         // Note: we haven't actually uploaded the second file yet, so current_usage is still 512KB
         // Check if we can upload 1KB (should succeed - we haven't used the full quota yet)
-        let check_result = env.check_bucket_quota("PUT", 1024).await?;
+        let check_result = env.wait_for_bucket_quota_check("PUT", 1024, 512 * 1024).await?;
         assert!(check_result.get("allowed").unwrap().as_bool().unwrap());
         assert_eq!(check_result.get("remaining_quota").unwrap().as_u64().unwrap(), 523264); // 511 * 1024
 
         // Check if we can upload 600KB (should fail - would exceed quota)
-        let check_result = env.check_bucket_quota("PUT", 600 * 1024).await?;
+        let check_result = env.wait_for_bucket_quota_check("PUT", 600 * 1024, 512 * 1024).await?;
         assert!(!check_result.get("allowed").unwrap().as_bool().unwrap());
 
         // Check delete operation (should always be allowed)
-        let check_result = env.check_bucket_quota("DELETE", 512 * 1024).await?;
+        let check_result = env.wait_for_bucket_quota_check("DELETE", 512 * 1024, 512 * 1024).await?;
         assert!(check_result.get("allowed").unwrap().as_bool().unwrap());
 
         env.cleanup_bucket().await?;
@@ -593,8 +708,8 @@ mod integration_tests {
         env.upload_object_to_bucket(&bucket2, "another_file.txt", 512 * 1024).await?;
 
         // Verify statistics are independent
-        let stats1 = env.get_bucket_quota_stats_for(&bucket1).await?;
-        let stats2 = env.get_bucket_quota_stats_for(&bucket2).await?;
+        let stats1 = env.wait_for_bucket_quota_stats(&bucket1, 1024 * 1024).await?;
+        let stats2 = env.wait_for_bucket_quota_stats(&bucket2, (1024 + 512) * 1024).await?;
 
         assert_eq!(stats1.get("current_usage").unwrap().as_u64().unwrap(), 1024 * 1024);
         assert_eq!(stats2.get("current_usage").unwrap().as_u64().unwrap(), (1024 + 512) * 1024);
@@ -655,7 +770,6 @@ mod integration_tests {
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
-        env.wait_for_quota_usage_for(&env.bucket_name).await?;
 
         // Test 1: GET quota for bucket without quota config
         let url = format!("{}/rustfs/admin/v3/quota/{}", env.env.url, env.bucket_name);

crates/e2e_test/src/reliant/grpc_lock_server.rs

@@ -548,6 +548,20 @@ impl NodeService for MinimalLockNodeService {
         Err(Status::unimplemented("lock-only test server"))
     }
 
+    async fn rename_multipart_data(
+        &self,
+        _request: Request<rustfs_protos::proto_gen::node_service::RenameDataRequest>,
+    ) -> Result<Response<rustfs_protos::proto_gen::node_service::RenameDataResponse>, Status> {
+        Err(Status::unimplemented("lock-only test server"))
+    }
+
+    async fn recover_multipart_source(
+        &self,
+        _request: Request<rustfs_protos::proto_gen::node_service::RecoverMultipartSourceRequest>,
+    ) -> Result<Response<rustfs_protos::proto_gen::node_service::RecoverMultipartSourceResponse>, Status> {
+        Err(Status::unimplemented("lock-only test server"))
+    }
+
     async fn rename_data_at_incarnation(
         &self,
         _request: Request<rustfs_protos::proto_gen::node_service::RenameDataRequest>,
@@ -562,6 +576,25 @@ impl NodeService for MinimalLockNodeService {
         Err(Status::unimplemented("lock-only test server"))
     }
 
+    async fn rename_data_with_pool_meta_cas(
+        &self,
+        _request: Request<rustfs_protos::proto_gen::node_service::RenameDataRequest>,
+    ) -> Result<Response<rustfs_protos::proto_gen::node_service::RenameDataResponse>, Status> {
+        Err(Status::unimplemented("lock-only test server"))
+    }
+    async fn undo_pool_meta_write(
+        &self,
+        _request: Request<rustfs_protos::proto_gen::node_service::DeleteVersionRequest>,
+    ) -> Result<Response<rustfs_protos::proto_gen::node_service::DeleteVersionResponse>, Status> {
+        Err(Status::unimplemented("lock-only test server"))
+    }
+    async fn pool_meta_native_recovery(
+        &self,
+        _request: Request<rustfs_protos::proto_gen::node_service::PoolMetaNativeRecoveryRequest>,
+    ) -> Result<Response<rustfs_protos::proto_gen::node_service::PoolMetaNativeRecoveryResponse>, Status> {
+        Err(Status::unimplemented("lock-only test server"))
+    }
+
     async fn make_volumes(
         &self,
         _request: Request<rustfs_protos::proto_gen::node_service::MakeVolumesRequest>,

crates/e2e_test/src/storage_api.rs

@@ -15,7 +15,11 @@
 #[cfg(test)]
 pub(crate) use rustfs_ecstore::api::bucket::bucket_target_sys::BucketTargetSys;
 #[cfg(test)]
-pub(crate) use rustfs_ecstore::api::disk::{DiskOption, Endpoint, RUSTFS_META_BUCKET, VolumeInfo, WalkDirOptions, new_disk};
+pub(crate) const OBJECT_LOCK_TRACE_FILTER: &str = "info,rustfs_ecstore::object_lock_diag=trace";
+#[cfg(test)]
+pub(crate) use rustfs_ecstore::api::disk::{
+    DiskOption, Endpoint, RUSTFS_META_BUCKET, RUSTFS_META_MULTIPART_BUCKET, VolumeInfo, WalkDirOptions, new_disk,
+};
 pub(crate) use rustfs_ecstore::api::rpc::{AuthenticatedChannel, TonicInterceptor, node_service_time_out_client_no_auth};
 #[cfg(test)]
 pub(crate) use rustfs_ecstore::api::rpc::{
@@ -55,5 +59,24 @@ pub(crate) mod replication_extension {
 
 #[cfg(test)]
 pub(crate) mod upgrade_compatibility {
-    pub(crate) use super::{DiskOption, Endpoint, new_disk};
+    pub(crate) use super::{DiskOption, Endpoint, RUSTFS_META_MULTIPART_BUCKET, new_disk};
+}
+
+#[cfg(test)]
+pub(crate) mod multipart_rpc_delay {
+    pub(crate) use super::RUSTFS_META_MULTIPART_BUCKET;
+}
+
+#[cfg(test)]
+pub(crate) mod multipart_fence_acceptance {
+    pub(crate) use super::RUSTFS_META_MULTIPART_BUCKET;
+}
+
+#[cfg(test)]
+pub(crate) mod multipart_source_rpc {
+    pub(crate) use rustfs_ecstore::api::disk::{
+        MULTIPART_SOURCE_PROTOCOL_VERSION, MultipartAbortCertificate, MultipartAbortProof, MultipartAbortProofKind,
+        MultipartAbortRound, MultipartSourceOperation, MultipartSourcePhase, MultipartSourceRequest, MultipartSourceSnapshot,
+        PublicationIdentity, SourceGeneration, SourceGeometry, SourceLineage,
+    };
 }

crates/e2e_test/src/upgrade_compatibility_test.rs

@@ -21,6 +21,7 @@ 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 crate::storage_api::upgrade_compatibility::RUSTFS_META_MULTIPART_BUCKET;
 use aws_sdk_s3::Client;
 use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::primitives::ByteStream;
@@ -52,6 +53,8 @@ const SSE_MASTER_KEY: &str = "QkJCQkJCQkJCQkJCQkJCQkJCQkJCQkJCQkJCQkJCQkI=";
 const PLAIN_BUCKET: &str = "upgrade-plain-data";
 const VERSIONED_BUCKET: &str = "upgrade-versioned-data";
 const MIXED_BUCKET: &str = "upgrade-mixed-version-data";
+const PENDING_MULTIPART_BUCKET: &str = "upgrade-mpu-continuity";
+const PENDING_MULTIPART_KEY: &str = "pending/legacy-multipart.bin";
 const MIXED_NODE_COUNT: usize = 4;
 const UPGRADE_READINESS_BODY: &[u8] = b"upgrade write readiness";
 // Written by the "previous-seed" readiness probe through node 0 before any
@@ -301,18 +304,27 @@ async fn read_object(
     Ok((encryption, body))
 }
 
-async fn write_multipart(client: &Client, bucket: &str, key: &str, parts: &[Vec<u8>]) -> TestResult {
-    let created = client.create_multipart_upload().bucket(bucket).key(key).send().await?;
-    let upload_id = created.upload_id().ok_or("CreateMultipartUpload omitted upload ID")?;
-    let mut completed_parts = Vec::with_capacity(parts.len());
+struct PendingMultipart {
+    upload_id: String,
+    completed_parts: Vec<CompletedPart>,
+    part_sizes: Vec<i64>,
+}
 
+async fn stage_multipart(client: &Client, bucket: &str, key: &str, parts: &[Vec<u8>]) -> Result<PendingMultipart, BoxError> {
+    let created = client.create_multipart_upload().bucket(bucket).key(key).send().await?;
+    let upload_id = created
+        .upload_id()
+        .ok_or("CreateMultipartUpload omitted upload ID")?
+        .to_owned();
+    let mut completed_parts = Vec::with_capacity(parts.len());
+    let mut part_sizes = Vec::with_capacity(parts.len());
     for (index, part) in parts.iter().enumerate() {
         let part_number = i32::try_from(index + 1)?;
         let uploaded = client
             .upload_part()
             .bucket(bucket)
             .key(key)
-            .upload_id(upload_id)
+            .upload_id(&upload_id)
             .part_number(part_number)
             .body(ByteStream::from(part.clone()))
             .send()
@@ -323,16 +335,38 @@ async fn write_multipart(client: &Client, bucket: &str, key: &str, parts: &[Vec<
                 .e_tag(uploaded.e_tag().ok_or("UploadPart omitted ETag")?)
                 .build(),
         );
+        part_sizes.push(i64::try_from(part.len())?);
     }
+    Ok(PendingMultipart {
+        upload_id,
+        completed_parts,
+        part_sizes,
+    })
+}
 
-    client
+async fn complete_pending_multipart(
+    client: &Client,
+    bucket: &str,
+    key: &str,
+    pending: &PendingMultipart,
+) -> Result<aws_sdk_s3::operation::complete_multipart_upload::CompleteMultipartUploadOutput, BoxError> {
+    Ok(client
         .complete_multipart_upload()
         .bucket(bucket)
         .key(key)
-        .upload_id(upload_id)
-        .multipart_upload(CompletedMultipartUpload::builder().set_parts(Some(completed_parts)).build())
+        .upload_id(&pending.upload_id)
+        .multipart_upload(
+            CompletedMultipartUpload::builder()
+                .set_parts(Some(pending.completed_parts.clone()))
+                .build(),
+        )
         .send()
-        .await?;
+        .await?)
+}
+
+async fn write_multipart(client: &Client, bucket: &str, key: &str, parts: &[Vec<u8>]) -> TestResult {
+    let pending = stage_multipart(client, bucket, key, parts).await?;
+    complete_pending_multipart(client, bucket, key, &pending).await?;
     Ok(())
 }
 
@@ -379,6 +413,330 @@ async fn write_multipart_load(clients: &[Client], phase: &str) -> Result<Vec<Str
     Ok(keys)
 }
 
+async fn write_mixed_put_load(clients: &[Client], phase: &str) -> Result<Vec<String>, BoxError> {
+    let mut tasks = JoinSet::new();
+    for worker in 0..MULTIPART_WORKERS {
+        let node = worker % clients.len();
+        let client = clients[node].clone();
+        let phase = phase.to_owned();
+        tasks.spawn(async move {
+            let mut keys = Vec::with_capacity(MULTIPART_UPLOADS_PER_WORKER);
+            for upload in 0..MULTIPART_UPLOADS_PER_WORKER {
+                let key = format!("{phase}/put/{worker:02}/{upload:02}");
+                tokio::time::timeout(
+                    MIXED_MULTIPART_UPLOAD_TIMEOUT,
+                    client
+                        .put_object()
+                        .bucket(MIXED_BUCKET)
+                        .key(&key)
+                        .body(ByteStream::from(vec![u8::try_from(worker)?; 64 * 1024]))
+                        .send(),
+                )
+                .await
+                .map_err(|_| format!("node {node}: PUT {key} acknowledgement deadline exceeded"))??;
+                keys.push(key);
+            }
+            Ok::<_, BoxError>(keys)
+        });
+    }
+    let mut keys = Vec::with_capacity(MULTIPART_WORKERS * MULTIPART_UPLOADS_PER_WORKER);
+    while let Some(result) = tasks.join_next().await {
+        keys.extend(result??);
+    }
+    Ok(keys)
+}
+
+async fn verify_load_bodies(clients: &[Client], phase: &str, operation: &str) -> TestResult {
+    for worker in 0..MULTIPART_WORKERS {
+        let reader = &clients[(worker + 1) % clients.len()];
+        for upload in 0..MULTIPART_UPLOADS_PER_WORKER {
+            let key = format!("{phase}/{operation}/{worker:02}/{upload:02}");
+            assert_eq!(
+                read_object(reader, MIXED_BUCKET, &key, None).await?.1,
+                vec![u8::try_from(worker)?; 64 * 1024],
+                "{phase}: acknowledged {key} body"
+            );
+        }
+    }
+    Ok(())
+}
+
+async fn exercise_homogeneous_multipart(cluster: &RustFSTestClusterEnvironment, phase: &str) -> TestResult {
+    let clients = cluster.create_all_clients()?;
+    wait_for_upgrade_write_readiness(&clients, MIXED_BUCKET, phase, LISTING_CONVERGENCE_TIMEOUT).await?;
+    wait_for_drive_readmission(&clients, MIXED_BUCKET, phase, DRIVE_READMISSION_TIMEOUT).await?;
+    let keys = match write_multipart_load(&clients, phase).await {
+        Ok(keys) => keys,
+        Err(error) => {
+            let log_dir = std::env::var_os("RUSTFS_E2E_LOG_DIR");
+            return observe_mixed_multipart_failure(cluster, phase, error, log_dir.as_deref().map(Path::new)).await;
+        }
+    };
+    assert_eq!(keys.len(), MULTIPART_WORKERS * MULTIPART_UPLOADS_PER_WORKER);
+    verify_load_bodies(&clients, phase, "multipart").await?;
+    for (reader, client) in clients.iter().enumerate() {
+        wait_for_phase_listing(client, phase, keys.len(), &format!("homogeneous reader {reader}")).await?;
+    }
+    Ok(())
+}
+
+struct PendingSourceBytes {
+    root: PathBuf,
+    files: Vec<(PathBuf, Vec<u8>)>,
+}
+
+fn pending_source_bytes(
+    cluster: &RustFSTestClusterEnvironment,
+    pending: &PendingMultipart,
+) -> Result<Vec<PendingSourceBytes>, BoxError> {
+    let source_hash =
+        rustfs_utils::hex_sha256(format!("{PENDING_MULTIPART_BUCKET}/{PENDING_MULTIPART_KEY}").as_bytes(), str::to_owned);
+    let mut sources = Vec::new();
+    for volume in cluster.nodes.iter().flat_map(|node| &node.data_dirs) {
+        let hash_root = Path::new(volume).join(RUSTFS_META_MULTIPART_BUCKET).join(&source_hash);
+        let roots = std::fs::read_dir(&hash_root)?
+            .map(|entry| {
+                let entry = entry?;
+                Ok(entry.file_type()?.is_dir().then(|| entry.path()))
+            })
+            .collect::<Result<Vec<_>, std::io::Error>>()?
+            .into_iter()
+            .flatten()
+            .collect::<Vec<_>>();
+        if roots.is_empty() {
+            return Err(std::io::Error::new(std::io::ErrorKind::NotFound, "pending upload source has not settled").into());
+        }
+        assert_eq!(roots.len(), 1, "the dedicated key has exactly one real pending upload");
+        let root = roots.into_iter().next().ok_or("pending upload source missing")?;
+        let mut files = vec![(PathBuf::from("xl.meta"), std::fs::read(root.join("xl.meta"))?)];
+        let data_dirs = std::fs::read_dir(&root)?
+            .map(|entry| {
+                let entry = entry?;
+                Ok((entry.file_type()?.is_dir() && entry.path().join("part.1").is_file()).then(|| entry.path()))
+            })
+            .collect::<Result<Vec<_>, std::io::Error>>()?
+            .into_iter()
+            .flatten()
+            .collect::<Vec<_>>();
+        if data_dirs.is_empty() {
+            return Err(std::io::Error::new(std::io::ErrorKind::NotFound, "selected part publication has not settled").into());
+        }
+        assert_eq!(data_dirs.len(), 1, "the upload has one actual selected-part directory");
+        let data_dir = data_dirs.into_iter().next().ok_or("pending upload parts missing")?;
+        for part in &pending.completed_parts {
+            let number = part.part_number().ok_or("acknowledged part omitted its number")?;
+            for name in [format!("part.{number}"), format!("part.{number}.meta")] {
+                let path = data_dir.join(name);
+                files.push((path.strip_prefix(&root)?.to_owned(), std::fs::read(path)?));
+            }
+        }
+        sources.push(PendingSourceBytes { root, files });
+    }
+    assert_eq!(
+        sources.len(),
+        MIXED_NODE_COUNT,
+        "the controlled legacy fixture requires every physical source"
+    );
+    Ok(sources)
+}
+
+async fn wait_for_pending_source_bytes(
+    cluster: &RustFSTestClusterEnvironment,
+    pending: &PendingMultipart,
+) -> Result<Vec<PendingSourceBytes>, BoxError> {
+    let deadline = Instant::now() + DRIVE_READMISSION_TIMEOUT;
+    loop {
+        match pending_source_bytes(cluster, pending) {
+            Ok(sources) => return Ok(sources),
+            Err(error)
+                if error
+                    .downcast_ref::<std::io::Error>()
+                    .is_some_and(|error| error.kind() == std::io::ErrorKind::NotFound)
+                    && Instant::now() < deadline =>
+            {
+                sleep(Duration::from_millis(500)).await
+            }
+            Err(error) => return Err(error),
+        }
+    }
+}
+
+async fn assert_pending_parts(client: &Client, pending: &PendingMultipart) -> TestResult {
+    let listed = client
+        .list_parts()
+        .bucket(PENDING_MULTIPART_BUCKET)
+        .key(PENDING_MULTIPART_KEY)
+        .upload_id(&pending.upload_id)
+        .send()
+        .await?;
+    assert_eq!(listed.upload_id(), Some(pending.upload_id.as_str()));
+    assert_ne!(listed.is_truncated(), Some(true));
+    assert_eq!(listed.parts().len(), pending.completed_parts.len());
+    for ((actual, expected), expected_size) in listed.parts().iter().zip(&pending.completed_parts).zip(&pending.part_sizes) {
+        assert_eq!(actual.part_number(), expected.part_number());
+        assert_eq!(actual.e_tag(), expected.e_tag());
+        assert_eq!(actual.size(), Some(*expected_size));
+    }
+    Ok(())
+}
+
+async fn assert_mixed_pending_refusal(
+    cluster: &RustFSTestClusterEnvironment,
+    pending: &PendingMultipart,
+    sources: &[PendingSourceBytes],
+) -> TestResult {
+    let clients = cluster.create_all_clients()?;
+    assert_pending_parts(&clients[0], pending).await?;
+    let error = upgrade_probe_client(&clients[0])
+        .complete_multipart_upload()
+        .bucket(PENDING_MULTIPART_BUCKET)
+        .key(PENDING_MULTIPART_KEY)
+        .upload_id(&pending.upload_id)
+        .multipart_upload(
+            CompletedMultipartUpload::builder()
+                .set_parts(Some(pending.completed_parts.clone()))
+                .build(),
+        )
+        .send()
+        .await
+        .expect_err("one native participant cannot authorize a legacy MPU completion");
+    assert_eq!(error.raw_response().map(|response| response.status().as_u16()), Some(503), "{error:?}");
+    assert_eq!(error.as_service_error().and_then(ProvideErrorMetadata::code), Some("ServiceUnavailable"));
+    for client in &clients {
+        assert_pending_parts(client, pending).await?;
+        let error = client
+            .get_object()
+            .bucket(PENDING_MULTIPART_BUCKET)
+            .key(PENDING_MULTIPART_KEY)
+            .send()
+            .await
+            .expect_err("the rejected Complete must not publish a target");
+        assert_eq!(error.as_service_error().and_then(ProvideErrorMetadata::code), Some("NoSuchKey"));
+    }
+    for source in sources {
+        for entry in std::fs::read_dir(&source.root)? {
+            assert!(
+                !entry?.file_name().to_string_lossy().starts_with(".mpu-"),
+                "the failed preflight must not upgrade one physical legacy source"
+            );
+        }
+        for (relative, expected) in &source.files {
+            let actual = std::fs::read(source.root.join(relative))?;
+            assert!(actual == *expected, "rejected Complete changed actual source {}", relative.display());
+        }
+    }
+    Ok(())
+}
+
+struct CompletedPendingMultipart {
+    version_id: String,
+    etag: String,
+    body: Vec<u8>,
+}
+
+async fn assert_resumed_multipart(clients: &[Client], completed: &CompletedPendingMultipart) -> TestResult {
+    for (node, client) in clients.iter().enumerate() {
+        let response = client
+            .get_object()
+            .bucket(PENDING_MULTIPART_BUCKET)
+            .key(PENDING_MULTIPART_KEY)
+            .version_id(&completed.version_id)
+            .send()
+            .await?;
+        assert_eq!(response.version_id(), Some(completed.version_id.as_str()), "reader {node}");
+        assert_eq!(response.e_tag(), Some(completed.etag.as_str()), "reader {node}");
+        assert_eq!(
+            response.body.collect().await?.into_bytes().as_ref(),
+            completed.body.as_slice(),
+            "reader {node}"
+        );
+        assert_eq!(
+            read_object(client, PENDING_MULTIPART_BUCKET, PENDING_MULTIPART_KEY, None)
+                .await?
+                .1,
+            completed.body
+        );
+        let versions = client
+            .list_object_versions()
+            .bucket(PENDING_MULTIPART_BUCKET)
+            .prefix(PENDING_MULTIPART_KEY)
+            .send()
+            .await?;
+        assert_eq!(versions.versions().len(), 1, "same-upload replay must not add a version");
+        assert_eq!(versions.versions()[0].version_id(), Some(completed.version_id.as_str()));
+        assert!(versions.delete_markers().is_empty());
+    }
+    Ok(())
+}
+
+async fn finish_pending_multipart(
+    clients: &[Client],
+    pending: &PendingMultipart,
+    body: Vec<u8>,
+) -> Result<CompletedPendingMultipart, BoxError> {
+    let complete = complete_pending_multipart(&clients[0], PENDING_MULTIPART_BUCKET, PENDING_MULTIPART_KEY, pending).await?;
+    let version_id = complete
+        .version_id()
+        .filter(|version| !version.is_empty() && *version != "null")
+        .ok_or("versioned legacy upload completion omitted a concrete version")?
+        .to_owned();
+    let etag = complete.e_tag().ok_or("CompleteMultipartUpload omitted ETag")?.to_owned();
+    let replay = complete_pending_multipart(&clients[0], PENDING_MULTIPART_BUCKET, PENDING_MULTIPART_KEY, pending).await?;
+    assert_eq!(replay.version_id(), Some(version_id.as_str()));
+    assert_eq!(replay.e_tag(), Some(etag.as_str()));
+    let completed = CompletedPendingMultipart { version_id, etag, body };
+    assert_resumed_multipart(clients, &completed).await?;
+    Ok(completed)
+}
+
+async fn wait_for_multipart_quiescence(cluster: &RustFSTestClusterEnvironment, sources: &[PendingSourceBytes]) -> TestResult {
+    let deadline = Instant::now() + LISTING_CONVERGENCE_TIMEOUT;
+    loop {
+        let pending_roots = sources.iter().filter(|source| source.root.exists()).count();
+        let mut native_controls = 0;
+        let mut remaining_upload_roots = 0;
+        for volume in cluster.nodes.iter().flat_map(|node| &node.data_dirs) {
+            let root = Path::new(volume).join(RUSTFS_META_MULTIPART_BUCKET);
+            if !root.exists() {
+                continue;
+            }
+            for entry in walkdir::WalkDir::new(root) {
+                let entry = match entry {
+                    Ok(entry) => entry,
+                    Err(error)
+                        if error
+                            .io_error()
+                            .is_some_and(|error| error.kind() == std::io::ErrorKind::NotFound) =>
+                    {
+                        native_controls += 1;
+                        continue;
+                    }
+                    Err(error) => return Err(error.into()),
+                };
+                // Every fixture upload lives at MPI/<SHA>/<upload>. A born-only
+                // source can contain just xl.meta and its data directory, with
+                // no .mpu-* record. Only empty SHA parents may survive rollback.
+                if entry.depth() == 2 || (entry.depth() == 1 && !entry.file_type().is_dir()) {
+                    remaining_upload_roots += 1;
+                }
+                if entry.file_name().to_string_lossy().starts_with(".mpu-") {
+                    native_controls += 1;
+                }
+            }
+        }
+        if pending_roots == 0 && remaining_upload_roots == 0 && native_controls == 0 {
+            return Ok(());
+        }
+        if Instant::now() >= deadline {
+            return Err(format!(
+                "downgrade requires settled native MPU state: {pending_roots} fixed pending roots, {remaining_upload_roots} inventoried upload roots, {native_controls} native controls"
+            )
+            .into());
+        }
+        sleep(Duration::from_millis(500)).await;
+    }
+}
+
 async fn mixed_multipart_before(
     deadline: Instant,
     node: usize,
@@ -625,14 +983,11 @@ async fn exercise_mixed_cluster(
         .await?;
     assert_eq!(read_object(current_client, MIXED_BUCKET, &previous_key, None).await?.1, previous_body);
 
-    let multipart_keys = match write_multipart_load(&clients, phase).await {
-        Ok(keys) => keys,
-        Err(error) => {
-            let log_dir = std::env::var_os("RUSTFS_E2E_LOG_DIR");
-            return observe_mixed_multipart_failure(cluster, phase, error, log_dir.as_deref().map(Path::new)).await;
-        }
-    };
-    let expected_count = multipart_keys.len() + 2;
+    // Native MPU mutations are operationally paused while RPC generations
+    // differ. Keep the same mixed traffic volume with ordinary PUT instead.
+    let written_keys = write_mixed_put_load(&clients, phase).await?;
+    let expected_count = written_keys.len() + 2;
+    verify_load_bodies(&clients, phase, "put").await?;
     for (label, client) in [("current", current_client), ("previous", previous_client)] {
         assert_eq!(
             read_object(client, MIXED_BUCKET, PREVIOUS_RELEASE_SEED_KEY, None).await?.1,
@@ -648,9 +1003,9 @@ async fn exercise_mixed_cluster(
         .await?;
     }
 
-    let last_multipart_key = format!("{phase}/multipart/{:02}/{:02}", MULTIPART_WORKERS - 1, MULTIPART_UPLOADS_PER_WORKER - 1);
+    let last_put_key = format!("{phase}/put/{:02}/{:02}", MULTIPART_WORKERS - 1, MULTIPART_UPLOADS_PER_WORKER - 1);
     assert_eq!(
-        read_object(previous_client, MIXED_BUCKET, &last_multipart_key, None).await?.1,
+        read_object(previous_client, MIXED_BUCKET, &last_put_key, None).await?.1,
         vec![u8::try_from(MULTIPART_WORKERS - 1)?; 64 * 1024]
     );
 
@@ -790,12 +1145,20 @@ fn previous_baseline_writers(clients: &[Client], restarted_node: usize) -> &[Cli
 }
 
 async fn prepare_previous_release_baseline(cluster: &mut RustFSTestClusterEnvironment, previous_binary: &Path) -> TestResult {
+    prepare_release_baseline(cluster, previous_binary, "previous").await
+}
+
+async fn prepare_release_baseline(
+    cluster: &mut RustFSTestClusterEnvironment,
+    previous_binary: &Path,
+    prefix: &str,
+) -> TestResult {
     let clients: Vec<_> = cluster.create_all_clients()?.iter().map(upgrade_probe_client).collect();
     // rc.5 can latch a non-elected node's write fence while its first startup
     // waits for pool metadata (#7473). First prove the elected writer can
     // persist data, then restart each old process once with three peers still
-    // readable. This preparation ends before any current binary is started.
-    wait_for_upgrade_write_readiness(&clients[..1], MIXED_BUCKET, "previous-seed", LISTING_CONVERGENCE_TIMEOUT).await?;
+    // readable. The same preparation is reused only after a quiescent homogeneous rollback.
+    wait_for_upgrade_write_readiness(&clients[..1], MIXED_BUCKET, &format!("{prefix}-seed"), LISTING_CONVERGENCE_TIMEOUT).await?;
     // A node that just came back is still replaying its startup traffic against
     // every peer. Killing the next node during that burst gives the fresh node
     // two failures in a row and it drops that peer's drive to Offline, where
@@ -805,7 +1168,7 @@ async fn prepare_previous_release_baseline(cluster: &mut RustFSTestClusterEnviro
     for node in [1, 2, 3, 0] {
         cluster.stop_node(node)?;
         cluster.start_node_from_binary(node, previous_binary).await?;
-        let context = format!("previous-restart-{node}");
+        let context = format!("{prefix}-restart-{node}");
         wait_for_upgrade_write_readiness(
             std::slice::from_ref(&clients[node]),
             MIXED_BUCKET,
@@ -822,15 +1185,17 @@ async fn prepare_previous_release_baseline(cluster: &mut RustFSTestClusterEnviro
         .await?;
     }
 
-    wait_for_upgrade_write_readiness(&clients, MIXED_BUCKET, "previous-baseline", LISTING_CONVERGENCE_TIMEOUT).await?;
+    wait_for_upgrade_write_readiness(&clients, MIXED_BUCKET, &format!("{prefix}-baseline"), LISTING_CONVERGENCE_TIMEOUT).await?;
     for (reader, client) in clients.iter().enumerate() {
         assert_eq!(
-            read_object(client, MIXED_BUCKET, PREVIOUS_RELEASE_SEED_KEY, None).await?.1,
+            read_object(client, MIXED_BUCKET, &format!(".upgrade-readiness/{prefix}-seed/node-0"), None)
+                .await?
+                .1,
             UPGRADE_READINESS_BODY,
             "previous-release node {reader} must retain the seed across its rolling restart"
         );
         for writer in 0..clients.len() {
-            let key = format!(".upgrade-readiness/previous-baseline/node-{writer}");
+            let key = format!(".upgrade-readiness/{prefix}-baseline/node-{writer}");
             assert_eq!(
                 read_object(client, MIXED_BUCKET, &key, None).await?.1,
                 UPGRADE_READINESS_BODY,
@@ -1883,10 +2248,21 @@ async fn rolling_upgrade_from_previous_release_preserves_mixed_version_contracts
     cluster.start_with_binary(&previous_binary).await?;
     cluster.create_test_bucket(MIXED_BUCKET).await?;
     prepare_previous_release_baseline(&mut cluster, &previous_binary).await?;
+    exercise_homogeneous_multipart(&cluster, "homogeneous-previous").await?;
+    cluster.create_test_bucket(PENDING_MULTIPART_BUCKET).await?;
+    let previous_clients = cluster.create_all_clients()?;
+    enable_versioning(&previous_clients[0], PENDING_MULTIPART_BUCKET).await?;
+    let pending_parts = vec![vec![0x63; 5 * 1024 * 1024], vec![0x7d; 4096]];
+    let pending = stage_multipart(&previous_clients[0], PENDING_MULTIPART_BUCKET, PENDING_MULTIPART_KEY, &pending_parts).await?;
+    let sources = wait_for_pending_source_bytes(&cluster, &pending).await?;
+    assert_pending_parts(&previous_clients[0], &pending).await?;
 
     cluster.stop_node(0)?;
     cluster.start_node_from_binary(0, &current_binary).await?;
     exercise_mixed_cluster(&cluster, "one-current-node", 0, 1).await?;
+    // One controlled out-of-policy attempt proves fail-closed behavior below
+    // native write quorum; positive MPU writers remain paused while mixed.
+    assert_mixed_pending_refusal(&cluster, &pending, &sources).await?;
 
     for (node_idx, phase) in [(1, "two-current-nodes"), (2, "one-previous-node")] {
         cluster.stop_node(node_idx)?;
@@ -1898,6 +2274,9 @@ async fn rolling_upgrade_from_previous_release_preserves_mixed_version_contracts
     cluster.start_node_from_binary(3, &current_binary).await?;
     let clients = cluster.create_all_clients()?;
     wait_for_drive_readmission(&clients, MIXED_BUCKET, "homogeneous-current", DRIVE_READMISSION_TIMEOUT).await?;
+    let completed_pending = finish_pending_multipart(&clients, &pending, pending_parts.concat()).await?;
+    exercise_homogeneous_multipart(&cluster, "homogeneous-current").await?;
+    verify_load_bodies(&clients, "homogeneous-previous", "multipart").await?;
 
     for (node_idx, client) in clients.iter().enumerate() {
         assert_eq!(
@@ -1916,47 +2295,6 @@ async fn rolling_upgrade_from_previous_release_preserves_mixed_version_contracts
         }
     }
 
-    if let Some(context) = evidence_context.as_ref() {
-        let phases = ["one-current-node", "two-current-nodes", "one-previous-node"];
-        let objects_per_phase = MULTIPART_WORKERS * MULTIPART_UPLOADS_PER_WORKER + 2;
-        write_g09_evidence(
-            context,
-            "mixed_version_reader_evidence",
-            "mixed-version-reader",
-            &["old-writer-new-reader", "new-writer-old-reader"],
-            "upgrade_compatibility_test::rolling_upgrade_from_previous_release_preserves_mixed_version_contracts",
-            serde_json::json!({
-                "bucket": MIXED_BUCKET,
-                "phases": phases,
-                "objects_per_phase": objects_per_phase,
-                "assertions": [
-                    "current node reads objects written through previous-release client",
-                    "previous-release node reads objects written through current client",
-                    "all nodes list every mixed-version object after homogeneous-current convergence",
-                    "the previous-release seed survives the baseline restart and every upgrade phase"
-                ],
-            }),
-        )?;
-        write_g09_evidence(
-            context,
-            "mixed_version_writer_evidence",
-            "mixed-version-writer",
-            &["old-reader-new-writer", "new-reader-old-writer"],
-            "upgrade_compatibility_test::rolling_upgrade_from_previous_release_preserves_mixed_version_contracts",
-            serde_json::json!({
-                "bucket": MIXED_BUCKET,
-                "phases": phases,
-                "objects_per_phase": objects_per_phase,
-                "assertions": [
-                    "current writer publishes objects readable by previous-release node",
-                    "previous-release writer publishes objects readable by current node",
-                    "multipart writers continue at every individual node upgrade",
-                    "modern DELETE fails closed and preserves every byte while old peers remain"
-                ],
-            }),
-        )?;
-    }
-
     let clients = cluster.create_all_clients()?;
     wait_for_upgrade_write_readiness(&clients, MIXED_BUCKET, "all-current", LISTING_CONVERGENCE_TIMEOUT).await?;
     clients[0]
@@ -1980,6 +2318,115 @@ async fn rolling_upgrade_from_previous_release_preserves_mixed_version_contracts
         );
     }
 
+    wait_for_multipart_quiescence(&cluster, &sources).await?;
+    // No MPU mutation runs during a heterogeneous downgrade. All current
+    // callbacks and source controls settle before old binaries reopen volumes.
+    cluster.stop();
+    cluster.start_with_binary(&previous_binary).await?;
+    prepare_release_baseline(&mut cluster, &previous_binary, "rollback").await?;
+    let old_clients = cluster.create_all_clients()?;
+    assert_resumed_multipart(&old_clients, &completed_pending).await?;
+    for phase in ["homogeneous-previous", "homogeneous-current"] {
+        verify_load_bodies(&old_clients, phase, "multipart").await?;
+        for client in &old_clients {
+            wait_for_phase_listing(
+                client,
+                phase,
+                MULTIPART_WORKERS * MULTIPART_UPLOADS_PER_WORKER,
+                "previous reader after rollback",
+            )
+            .await?;
+        }
+    }
+    for phase in ["one-current-node", "two-current-nodes", "one-previous-node"] {
+        verify_load_bodies(&old_clients, phase, "put").await?;
+        for client in &old_clients {
+            wait_for_phase_listing(
+                client,
+                phase,
+                MULTIPART_WORKERS * MULTIPART_UPLOADS_PER_WORKER + 2,
+                "previous reader after rollback",
+            )
+            .await?;
+            for (key, body) in [
+                ("written-by-current", format!("{phase}: current RustFS build")),
+                ("written-by-previous", format!("{phase}: previous RustFS release")),
+            ] {
+                assert_eq!(
+                    read_object(client, MIXED_BUCKET, &format!("{phase}/{key}"), None).await?.1,
+                    body.as_bytes()
+                );
+            }
+        }
+    }
+    for (node, client) in old_clients.iter().enumerate() {
+        let error = client
+            .get_object()
+            .bucket(MIXED_BUCKET)
+            .key(PREVIOUS_RELEASE_SEED_KEY)
+            .send()
+            .await
+            .expect_err("current DELETE must remain visible to previous readers after rollback");
+        assert_eq!(
+            error.as_service_error().and_then(ProvideErrorMetadata::code),
+            Some("NoSuchKey"),
+            "reader {node}"
+        );
+    }
+    exercise_homogeneous_multipart(&cluster, "homogeneous-rollback").await?;
+
+    if let Some(context) = evidence_context.as_ref() {
+        let phases = ["one-current-node", "two-current-nodes", "one-previous-node"];
+        let objects_per_phase = MULTIPART_WORKERS * MULTIPART_UPLOADS_PER_WORKER + 2;
+        write_g09_evidence(
+            context,
+            "mixed_version_reader_evidence",
+            "mixed-version-reader",
+            &["old-writer-new-reader", "new-writer-old-reader"],
+            "upgrade_compatibility_test::rolling_upgrade_from_previous_release_preserves_mixed_version_contracts",
+            serde_json::json!({
+                "bucket": MIXED_BUCKET,
+                "phases": phases,
+                "objects_per_phase": objects_per_phase,
+                "mixed_load": "plain-put",
+                "homogeneous_multipart_phases": ["homogeneous-previous", "homogeneous-current", "homogeneous-rollback"],
+                "multipart_objects_per_homogeneous_phase": MULTIPART_WORKERS * MULTIPART_UPLOADS_PER_WORKER,
+                "assertions": [
+                    "current node reads objects written through previous-release client",
+                    "previous-release node reads objects written through current client",
+                    "all nodes list every mixed-version object after homogeneous-current convergence",
+                    "the previous-release seed survives the baseline restart and every upgrade phase",
+                    "previous readers read every current-native MPU body after quiescent rollback",
+                    "same legacy upload resumes with identical body, ETag and version after full upgrade"
+                ],
+            }),
+        )?;
+        write_g09_evidence(
+            context,
+            "mixed_version_writer_evidence",
+            "mixed-version-writer",
+            &["old-reader-new-writer", "new-reader-old-writer"],
+            "upgrade_compatibility_test::rolling_upgrade_from_previous_release_preserves_mixed_version_contracts",
+            serde_json::json!({
+                "bucket": MIXED_BUCKET,
+                "phases": phases,
+                "objects_per_phase": objects_per_phase,
+                "mixed_load": "plain-put",
+                "homogeneous_multipart_phases": ["homogeneous-previous", "homogeneous-current", "homogeneous-rollback"],
+                "multipart_objects_per_homogeneous_phase": MULTIPART_WORKERS * MULTIPART_UPLOADS_PER_WORKER,
+                "assertions": [
+                    "current writer publishes objects readable by previous-release node",
+                    "previous-release writer publishes objects readable by current node",
+                    "mixed phases retain 16x16 plain PUT writers while positive multipart writers are paused",
+                    "16x16 multipart load runs in homogeneous previous, current and rollback phases",
+                    "one-current-node Complete returns typed 503 and preserves actual source authority and parts",
+                    "the same legacy upload completes and replays once fully upgraded without adding a version",
+                    "modern DELETE fails closed and preserves every byte while old peers remain"
+                ],
+            }),
+        )?;
+    }
+
     Ok(())
 }
 

crates/ecstore/AGENTS.md

@@ -12,6 +12,9 @@ Applies to `crates/ecstore/`.
 
 - Be careful with allocations and locking in hot paths.
 - Keep network and disk operations async-friendly; avoid introducing unnecessary blocking.
+- Coalesce related `open`, `read`/`write`, `fsync`, and `close` filesystem steps into a single per-disk blocking closure rather than chaining multiple async or blocking dispatches, and route any closure that performs file or directory sync to the disk's dedicated fsync lane so stalled syncs cannot starve foreground metadata reads.
+- Pin a parent directory once (`openat2(BENEATH | NO_SYMLINKS)` on Linux) and reuse the pinned handle across sibling reads or same-directory renames instead of re-walking ancestor components per file.
+- When the file size is already known from `fstat` metadata, pre-size the destination buffer (`len + 1`) before reading to EOF so small metadata reads complete in a single `read()` + EOF check rather than exponential-growth reallocations and syscalls.
 - Benchmark-sensitive changes should include measurable rationale.
 
 ## Allocation Discipline in Hot Paths

crates/ecstore/Cargo.toml

@@ -33,11 +33,6 @@ workspace = true
 default = []
 diagnostic-read-version-caller = ["rustfs-io-metrics/diagnostic-read-version-caller"]
 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
-# state in a stock build (backlog#1832).
-list-chaos = []
 rio-v2 = ["dep:rustfs-rio-v2"]
 hotpath = [
     "hotpath/hotpath",
@@ -142,6 +137,7 @@ rustfs-lifecycle.workspace = true
 rustfs-s3-client = { workspace = true }
 rustfs-s3-types = { workspace = true }
 rustfs-data-usage.workspace = true
+rustfs-gateway-types.workspace = true
 rustfs-object-capacity.workspace = true
 rustfs-object-data-cache = { workspace = true, features = ["runtime-memory"] }
 arc-swap.workspace = true
@@ -160,7 +156,6 @@ serde = { workspace = true, features = ["derive"] }
 time = { workspace = true, features = ["parsing", "formatting", "macros", "serde"] }
 bytesize.workspace = true
 serde_json = { workspace = true, features = ["raw_value"] }
-s3s = { workspace = true, features = ["minio"] }
 http.workspace = true
 opentelemetry.workspace = true
 http-body = { workspace = true }
@@ -249,6 +244,7 @@ windows-sys = { workspace = true, features = [
 windows-sys = { workspace = true, features = ["Win32_System_Ioctl"] }
 
 [dev-dependencies]
+s3s = { workspace = true, features = ["minio"] }
 aws-smithy-async.workspace = true
 rustfs-s3-ops.workspace = true
 tokio = { workspace = true, features = ["rt-multi-thread", "macros", "test-util", "fs"] }

crates/ecstore/src/api/mod.rs

@@ -62,6 +62,8 @@ pub mod bucket {
         }
 
         pub mod bucket_lifecycle_ops {
+            pub use rustfs_lifecycle::{RestoreOutputLocation, RestoreRequest, RestoreS3Location, RestoreTag, RestoreTagging};
+
             pub use crate::bucket::lifecycle::bucket_lifecycle_ops::{
                 ExpiryState, LifecycleOps, ManualTransitionCancelCheck, ManualTransitionProgressSink,
                 ManualTransitionQueueSnapshot, ManualTransitionRunExecution, ManualTransitionRunOptions,
@@ -133,7 +135,7 @@ pub mod bucket {
             pub use crate::bucket::lifecycle::lifecycle::{
                 Event, ExpirationOptions, IlmAction, LIFECYCLE_MALFORMED_XML_ERROR_KIND, Lifecycle, LifecycleCalculate,
                 ObjectOpts, RuleValidate, TRANSITION_COMPLETE, TRANSITION_PENDING, TransitionOptions, expected_expiry_time,
-                object_opts_from_object_info,
+                format_lifecycle_date, object_opts_from_object_info, parse_lifecycle_date,
             };
         }
 
@@ -225,7 +227,7 @@ pub mod bucket {
     }
 
     pub mod object_lock {
-        pub use crate::bucket::object_lock::{ObjectLockApi, ObjectLockStatusExt};
+        pub use crate::bucket::object_lock::ObjectLockApi;
 
         pub mod types {
             pub use crate::bucket::object_lock::types::{
@@ -320,9 +322,9 @@ pub mod bucket {
 
     pub mod utils {
         pub use crate::bucket::utils::{
-            check_bucket_and_object_names, check_list_objs_args, check_object_name_for_length_and_slash,
-            check_valid_bucket_name_strict, deserialize, has_bad_path_component, is_meta_bucketname, is_valid_object_prefix,
-            serialize,
+            BucketConfigXml, BucketConfigXmlError, check_bucket_and_object_names, check_list_objs_args,
+            check_object_name_for_length_and_slash, check_valid_bucket_name_strict, deserialize, has_bad_path_component,
+            is_meta_bucketname, is_valid_object_prefix, serialize,
         };
     }
 
@@ -341,15 +343,15 @@ pub mod cache {
 
 pub mod capacity {
     pub use crate::core::pools::{
-        DecommissionUnresolvedEntry, PoolDecommissionInfo, PoolMeta, PoolStatus, get_total_usable_capacity,
-        get_total_usable_capacity_free, is_pool_activation_fleet_proof_error, path2_bucket_object,
-        path2_bucket_object_with_base_path,
+        DecommissionUnresolvedEntry, PoolAdminDiagnostic, PoolAvailability, PoolCapacityStatus, PoolDecommissionInfo,
+        PoolDiagnosticStatus, PoolMeta, PoolStatus, get_total_usable_capacity, get_total_usable_capacity_free,
+        is_pool_activation_fleet_proof_error, path2_bucket_object, path2_bucket_object_with_base_path,
     };
     pub use crate::store::utils::is_reserved_or_invalid_bucket;
 }
 
 pub mod object_api_utils {
-    pub use crate::object_api::object_api_utils::{ObjReaderFn, PutObjReader, get_raw_etag, new_getobjectreader, to_s3s_etag};
+    pub use crate::object_api::object_api_utils::{ObjReaderFn, PutObjReader, get_raw_etag, new_getobjectreader};
 }
 
 pub mod cluster {
@@ -429,6 +431,16 @@ pub mod data_usage {
 pub mod disk {
     pub use crate::disk::disk_store::get_object_disk_read_timeout;
     pub use crate::disk::local::{ReplacementExecutionLease, ScanGuard};
+    #[cfg(feature = "test-util")]
+    pub use crate::disk::local::{local_durability_modes, set_local_rename_data_fail_after_metadata_commit};
+    pub use crate::disk::multipart_source::{
+        MULTIPART_SOURCE_PROTOCOL_VERSION, MultipartAbortCertificate, MultipartAbortProof, MultipartAbortProofKind,
+        MultipartAbortRound, MultipartSourceOperation, MultipartSourcePhase, MultipartSourceRequest, MultipartSourceResponse,
+        MultipartSourceSnapshot, PublicationIdentity, SourceGeneration, SourceGeometry, SourceLineage, capture_source_condition,
+        require_source_condition, strip_native_controls,
+    };
+    #[cfg(all(feature = "test-util", unix))]
+    pub use crate::disk::os::set_local_dir_fsync_failure;
     #[cfg(all(feature = "test-util", not(windows)))]
     pub use crate::disk::os::{LocalPublicationPause, LocalPublicationStage};
     pub use crate::disk::{
@@ -438,9 +450,12 @@ pub mod disk {
         DeleteRecoveryInspection, DeleteRecoveryPage, DeleteRecoveryRpcOperation, DeleteRecoveryRpcReply, DeleteRecoveryState,
         Disk, DiskAPI, DiskInfo, DiskInfoOptions, DiskLocation, DiskOption, DiskStore, FileInfoVersions, FileReader, FileWriter,
         GuardedDeleteVersionsPayload, GuardedFileInfoVersions, HEALING_MARKER_PATH, NsScannerOpenRequest, OldCurrentSize,
-        PartTransactionAction, QUOTA_MUTATION_FENCE_METADATA_SUFFIX, RUSTFS_META_BUCKET, ReadMultipleReq, ReadMultipleResp,
-        ReadOptions, RenameDataResp, STORAGE_FORMAT_FILE, SnapshotLeaseToken, UpdateMetadataOpts, VolumeInfo, WalkDirOptions,
-        is_quota_mutation_fence_path, new_disk, validate_batch_read_version_item_count,
+        POOL_META_NATIVE_CAS_METADATA_SUFFIX, PartTransactionAction, PoolMetaNativeBackup, PoolMetaNativeInspection,
+        PoolMetaNativeObject, PoolMetaNativeRecoveryRequest, PoolMetaNativeRecoveryResponse,
+        QUOTA_MUTATION_FENCE_METADATA_SUFFIX, RUSTFS_META_BUCKET, RUSTFS_META_MULTIPART_BUCKET, RUSTFS_META_TMP_BUCKET,
+        ReadMultipleReq, ReadMultipleResp, ReadOptions, RenameDataResp, STORAGE_FORMAT_FILE, SnapshotLeaseToken,
+        UpdateMetadataOpts, VolumeInfo, WalkDirOptions, is_quota_mutation_fence_path, new_disk,
+        validate_batch_read_version_item_count,
     };
     pub use bytes::Bytes;
     pub use endpoint::Endpoint;
@@ -513,13 +528,16 @@ pub mod metrics {
 pub mod notification {
     pub use crate::services::notification_sys::{
         ClusterTierDailyStats, CrossPoolFenceFleetProofToken, IlmRecoveryExportFleetProofToken,
-        LegacyTransitionStateReconcileFleetProofToken, NotificationPeerErr, NotificationSys, ScannerPublicationLeaseGrant,
-        acquire_cross_pool_fence_fleet_proof, acquire_ilm_recovery_export_fleet_proof,
-        acquire_legacy_transition_state_reconcile_fleet_proof, configured_fleet_topology_fingerprint,
-        cross_pool_fence_fleet_proof_matches, get_global_notification_sys, ilm_recovery_export_fleet_proof_matches,
-        ilm_recovery_export_local_process_epoch, ilm_recovery_export_member_epochs_sha256,
-        ilm_recovery_export_topology_generation, legacy_transition_state_reconcile_fleet_proof_matches,
-        new_global_notification_sys, scanner_peer_transport_error_message_is_retryable, start_remote_version_state_fleet_probe,
+        LegacyTransitionStateReconcileFleetProofToken, MrfLifecycleWriterFleetProofToken, NotificationPeerErr, NotificationSys,
+        ScannerPublicationLeaseGrant, acquire_cross_pool_fence_fleet_proof, acquire_ilm_recovery_export_fleet_proof,
+        acquire_legacy_transition_state_reconcile_fleet_proof, acquire_mrf_lifecycle_writer_fleet_proof,
+        configured_fleet_topology_fingerprint, cross_pool_fence_fleet_proof_matches, get_global_notification_sys,
+        ilm_recovery_export_fleet_proof_matches, ilm_recovery_export_local_process_epoch,
+        ilm_recovery_export_member_epochs_sha256, ilm_recovery_export_topology_generation,
+        legacy_transition_state_reconcile_fleet_proof_matches, mrf_lifecycle_reader_local_process_epoch,
+        mrf_lifecycle_writer_fleet_proof_current, mrf_lifecycle_writer_fleet_proof_matches, new_global_notification_sys,
+        scanner_peer_transport_error_message_is_retryable, start_remote_version_state_fleet_probe,
+        with_mrf_lifecycle_writer_fleet_proof,
     };
     #[cfg(any(test, feature = "test-util"))]
     pub use crate::services::notification_sys::{

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

@@ -87,12 +87,14 @@ use rustfs_filemeta::{
     FileInfo, FileInfoOpts, NULL_VERSION_ID, RestoreStatus, RestoreStatusOps, TRANSITION_COMPLETE, get_file_info,
     is_restored_object_on_disk,
 };
+use rustfs_gateway_types::persistence::PersistedLifecycleConfiguration;
+use rustfs_gateway_types::persistence::PersistedObjectLockConfiguration as ObjectLockConfiguration;
+use rustfs_lifecycle::{RestoreRequest, RestoreS3Location};
 use rustfs_lock::{NamespaceLockGuard, error::LockError};
 use rustfs_scanner_metrics::metrics::{
     IlmAction, Metrics, ScannerLifecycleExpiryStateUpdate, ScannerLifecycleTransitionStateUpdate, global_metrics,
 };
 use rustfs_utils::{get_env_i64, get_env_usize, path::encode_dir_object, string::parse_bool};
-use s3s::dto::{BucketLifecycleConfiguration, ExpirationStatus, ObjectLockConfiguration, RestoreRequest, RestoreRequestType};
 use sha2::{Digest, Sha256};
 use std::any::Any;
 use std::collections::{BTreeMap, HashMap, HashSet};
@@ -298,7 +300,7 @@ impl LifecycleSys {
         Arc::new(Self)
     }
 
-    pub async fn get(&self, bucket: &str) -> Option<BucketLifecycleConfiguration> {
+    pub async fn get(&self, bucket: &str) -> Option<PersistedLifecycleConfiguration> {
         match metadata_boundary::get_lifecycle_config(bucket).await {
             Ok((lc, _)) => Some(lc),
             Err(Error::ConfigNotFound) => None,
@@ -3934,7 +3936,7 @@ pub fn init_background_stale_multipart_upload_cleanup(api: Arc<ECStore>) {
     });
 }
 
-pub async fn validate_transition_tier(lc: &BucketLifecycleConfiguration) -> Result<(), std::io::Error> {
+pub async fn validate_transition_tier(lc: &PersistedLifecycleConfiguration) -> Result<(), std::io::Error> {
     for rule in &lc.rules {
         if let Some(transitions) = &rule.transitions {
             for transition in transitions {
@@ -4051,8 +4053,9 @@ pub(crate) async fn enqueue_immediate_expiry(api: Arc<ECStore>, oi: &ObjectInfo,
         .map(lifecycle::object_opts_from_object_info)
         .collect::<Vec<ObjectOpts>>();
     let lock_config = configs.object_lock;
+    let evaluator_lock_config = lock_config.clone();
     let events = match Evaluator::new(lifecycle)
-        .with_lock_retention(lock_config.clone())
+        .with_lock_retention(evaluator_lock_config)
         .eval(&object_opts)
         .await
     {
@@ -4532,19 +4535,26 @@ pub async fn enqueue_transition_for_existing_objects_scoped(
     }
 }
 
-fn lifecycle_rule_has_date_expiration(lc: &BucketLifecycleConfiguration, rule_id: &str) -> bool {
+fn lifecycle_rule_has_date_expiration(lc: &PersistedLifecycleConfiguration, rule_id: &str) -> bool {
     lc.rules.iter().any(|rule| {
-        rule.status == ExpirationStatus::from_static(ExpirationStatus::ENABLED)
+        rule.status == "Enabled"
             && rule.id.as_deref() == Some(rule_id)
             && rule.expiration.as_ref().is_some_and(|expiration| expiration.date.is_some())
     })
 }
 
-fn should_defer_date_expiry_for_recent_config_update(lc: &BucketLifecycleConfiguration, now: OffsetDateTime) -> bool {
-    lc.expiry_updated_at.as_ref().is_some_and(|updated_at| {
-        let updated_at = OffsetDateTime::from(updated_at.clone());
-        now.unix_timestamp().saturating_sub(updated_at.unix_timestamp()) < DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS
-    })
+fn should_defer_date_expiry_for_recent_config_update(lc: &PersistedLifecycleConfiguration, now: OffsetDateTime) -> bool {
+    lc.expiry_updated_at
+        .as_ref()
+        .is_some_and(|updated_at| match rustfs_lifecycle::parse_lifecycle_date(updated_at) {
+            Ok(updated_at) => {
+                now.unix_timestamp().saturating_sub(updated_at.unix_timestamp()) < DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS
+            }
+            Err(error) => {
+                warn!(error = %error, "Deferring date expiry with an unreadable update timestamp");
+                true
+            }
+        })
 }
 
 async fn apply_existing_object_expiry(
@@ -4564,7 +4574,7 @@ async fn apply_existing_object_expiry(
 struct ExistingObjectExpiryContext<'a> {
     api: Arc<ECStore>,
     bucket: &'a str,
-    lc: Arc<BucketLifecycleConfiguration>,
+    lc: Arc<PersistedLifecycleConfiguration>,
     lock_config: Option<Arc<ObjectLockConfiguration>>,
     bucket_incarnation_id: Uuid,
     src: &'a LcEventSrc,
@@ -4584,8 +4594,9 @@ async fn enqueue_expiry_for_existing_object_group(
         .iter()
         .map(lifecycle::object_opts_from_object_info)
         .collect::<Vec<ObjectOpts>>();
+    let evaluator_lock_config = context.lock_config.clone();
     let events = match Evaluator::new(context.lc.clone())
-        .with_lock_retention(context.lock_config.clone())
+        .with_lock_retention(evaluator_lock_config)
         .eval(&object_opts)
         .await
     {
@@ -4779,7 +4790,7 @@ fn manual_transition_version_marker(oi: &ObjectInfo) -> String {
 async fn enqueue_transition_with_lifecycle_report(
     api: Option<Arc<ECStore>>,
     oi: &ObjectInfo,
-    lc: &BucketLifecycleConfiguration,
+    lc: &PersistedLifecycleConfiguration,
     src: &LcEventSrc,
     options: &ManualTransitionRunOptions,
     report: &mut ManualTransitionRunReport,
@@ -4845,7 +4856,7 @@ async fn enqueue_transition_with_lifecycle_report(
     false
 }
 
-async fn enqueue_transition_with_lifecycle(oi: &ObjectInfo, lc: &BucketLifecycleConfiguration, src: &LcEventSrc) -> bool {
+async fn enqueue_transition_with_lifecycle(oi: &ObjectInfo, lc: &PersistedLifecycleConfiguration, src: &LcEventSrc) -> bool {
     let event = lc.eval(&oi.to_lifecycle_opts()).await;
     match event.action {
         IlmAction::TransitionAction | IlmAction::TransitionVersionAction => {
@@ -5359,12 +5370,8 @@ pub async fn post_restore_opts(version_id: &str, bucket: &str, object: &str) ->
     })
 }
 
-fn select_restore_s3_location(rreq: &RestoreRequest) -> Result<Option<&s3s::dto::S3Location>, std::io::Error> {
-    if rreq
-        .type_
-        .as_ref()
-        .is_none_or(|type_| type_.as_str() != RestoreRequestType::SELECT)
-    {
+fn select_restore_s3_location(rreq: &RestoreRequest) -> Result<Option<&RestoreS3Location>, std::io::Error> {
+    if rreq.type_.as_ref().is_none_or(|type_| type_.as_str() != "SELECT") {
         return Ok(None);
     }
     let output_location = rreq
@@ -5406,27 +5413,25 @@ pub async fn put_restore_opts(
     // the OutputLocation metadata instead of a write to `OutputLocation.S3`
     // (backlog#1341). RestoreObject rejects SELECT at the API boundary; this
     // is the fail-closed backstop for any other caller.
-    if rreq
-        .type_
-        .as_ref()
-        .is_some_and(|type_| type_.as_str() == RestoreRequestType::SELECT)
-    {
+    if rreq.type_.as_ref().is_some_and(|type_| type_.as_str() == "SELECT") {
         return Err(std::io::Error::other("SELECT restore requests are not supported"));
     }
     for (k, v) in oi.user_defined.iter() {
         meta.insert(k.to_string(), v.clone());
     }
+    // The fleet capability gate guarantees physical copy-back support. Keep
+    // the completed transition identity for orphan recovery and remote cleanup.
     rustfs_utils::http::metadata_compat::remove_str(&mut meta, rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_OPERATION_ID);
-    rustfs_utils::http::metadata_compat::remove_str(&mut meta, rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_WORKER_LOCK);
     if !oi.user_tags.is_empty() {
         meta.insert(AMZ_OBJECT_TAGGING.to_string(), (*oi.user_tags).clone());
     }
-    let restore_expiry = lifecycle::expected_expiry_time(OffsetDateTime::now_utc(), rreq.days.unwrap_or(1));
+    // Finalization publishes completion and its expiry under the object lock.
+    // The local copy-back commit must not make unfinished work eligible for expiry.
     meta.insert(
         metadata_keys::RESTORE.to_string(),
         RestoreStatus {
-            is_restore_in_progress: Some(false),
-            restore_expiry_date: Some(restore_expiry),
+            is_restore_in_progress: Some(true),
+            restore_expiry_date: None,
         }
         .to_string(),
     );
@@ -5440,6 +5445,7 @@ pub async fn put_restore_opts(
         // computed MD5 is not the object's public plaintext ETag.
         preserve_etag: oi.etag.clone(),
         shard_integrity_write_mode: Some(oi.shard_integrity_write_mode()),
+        restore_copyback: true,
         //expires:           oi.expires,
         ..Default::default()
     })
@@ -5459,6 +5465,18 @@ impl LifecycleOps for ObjectInfo {
         if self.transitioned_object.status != lifecycle::TRANSITION_COMPLETE {
             return false;
         }
+        if rustfs_utils::http::metadata_compat::get_consistent_str(
+            &self.user_defined,
+            rustfs_utils::http::metadata_compat::SUFFIX_RESTORE_COPYBACK,
+        ) == Some(rustfs_utils::http::metadata_compat::RESTORE_COPYBACK_PROTOCOL_V1)
+            && self
+                .user_defined
+                .get(metadata_keys::RESTORE)
+                .and_then(|value| rustfs_filemeta::parse_restore_obj_status(value).ok())
+                .is_some_and(|status| status.is_restore_in_progress == Some(true))
+        {
+            return false;
+        }
         !is_restored_object_on_disk(&self.user_defined)
     }
 }
@@ -5470,25 +5488,25 @@ pub trait RestoreRequestOps {
 impl RestoreRequestOps for RestoreRequest {
     fn validate(&self, _api: Arc<ECStore>) -> Result<(), std::io::Error> {
         // SELECT type requires select_parameters, and vice versa
-        if self.type_.as_ref().is_none_or(|t| t.as_str() != RestoreRequestType::SELECT) && self.select_parameters.is_some() {
+        if self.type_.as_ref().is_none_or(|t| t.as_str() != "SELECT") && self.select_parameters.is_some() {
             return Err(std::io::Error::other("Select parameters can only be specified with SELECT request type"));
         }
         if let Some(type_) = &self.type_
-            && type_.as_str() == RestoreRequestType::SELECT
+            && type_.as_str() == "SELECT"
             && self.select_parameters.is_none()
         {
             return Err(std::io::Error::other("SELECT restore request requires select parameters to be specified"));
         }
 
         // OutputLocation is only valid for SELECT requests
-        if self.type_.as_ref().is_none_or(|t| t.as_str() != RestoreRequestType::SELECT) && self.output_location.is_some() {
+        if self.type_.as_ref().is_none_or(|t| t.as_str() != "SELECT") && self.output_location.is_some() {
             return Err(std::io::Error::other("OutputLocation can only be specified with SELECT request type"));
         }
         select_restore_s3_location(self)?;
 
         // Days must not be specified with SELECT requests
         if let Some(type_) = &self.type_
-            && type_.as_str() == RestoreRequestType::SELECT
+            && type_.as_str() == "SELECT"
             && self.days.is_some_and(|d| d > 0)
         {
             return Err(std::io::Error::other("Days cannot be specified with SELECT restore request"));
@@ -5530,7 +5548,7 @@ pub fn lifecycle_version_delete_target(oi: &ObjectInfo) -> Option<ObjectToDelete
 }
 
 pub async fn eval_action_from_lifecycle(
-    lc: &BucketLifecycleConfiguration,
+    lc: &PersistedLifecycleConfiguration,
     lock_config: Option<&ObjectLockConfiguration>,
     oi: &ObjectInfo,
 ) -> lifecycle::Event {
@@ -6235,14 +6253,19 @@ mod tests {
     use rustfs_data_usage::TierStats;
     use rustfs_filemeta::metadata_keys;
     use rustfs_filemeta::{FileInfo, FileMeta};
+    use rustfs_gateway_types::dto::MetadataEntry;
+    use rustfs_gateway_types::persistence::{
+        PersistedDefaultRetention as DefaultRetention, PersistedObjectLockConfiguration as ObjectLockConfiguration,
+        PersistedObjectLockRule as ObjectLockRule,
+    };
+    use rustfs_gateway_types::persistence::{
+        PersistedLifecycleConfiguration, PersistedLifecycleExpiration, PersistedLifecycleRule,
+        PersistedNoncurrentVersionExpiration, PersistedTransition,
+    };
+    use rustfs_lifecycle::{RestoreOutputLocation as OutputLocation, RestoreRequest, RestoreS3Location as S3Location};
     #[cfg(feature = "test-util")]
     use rustfs_s3_client::transition_api::ReaderImpl;
     use rustfs_scanner_metrics::metrics::{IlmAction, global_metrics};
-    use s3s::dto::{
-        BucketLifecycleConfiguration, DefaultRetention, ExpirationStatus, LifecycleExpiration, LifecycleRule, MetadataEntry,
-        NoncurrentVersionExpiration, ObjectLockConfiguration, ObjectLockEnabled, ObjectLockRetentionMode, ObjectLockRule,
-        OutputLocation, RestoreRequest, RestoreRequestType, S3Location, Timestamp, Transition, TransitionStorageClass,
-    };
     use serial_test::serial;
     use sha2::{Digest, Sha256};
     use std::collections::HashMap;
@@ -9249,7 +9272,7 @@ mod tests {
             output_location,
             select_parameters: None,
             tier: None,
-            type_: Some(RestoreRequestType::from_static(RestoreRequestType::SELECT)),
+            type_: Some("SELECT".to_owned()),
         }
     }
 
@@ -9566,12 +9589,12 @@ mod tests {
 
     #[test]
     fn lifecycle_rule_has_date_expiration_detects_enabled_date_rule() {
-        let lc = BucketLifecycleConfiguration {
+        let lc = PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
-                    date: Some(Timestamp::from(OffsetDateTime::now_utc())),
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
+                    date: Some(lifecycle::format_lifecycle_date(OffsetDateTime::now_utc()).unwrap()),
                     ..Default::default()
                 }),
                 id: Some("rule-date".to_string()),
@@ -10088,14 +10111,15 @@ mod tests {
     #[test]
     fn should_defer_date_expiry_for_recent_config_update_respects_grace_window() {
         let now = OffsetDateTime::now_utc();
-        let recent = BucketLifecycleConfiguration {
-            expiry_updated_at: Some(Timestamp::from(now - time::Duration::seconds(1))),
+        let recent = PersistedLifecycleConfiguration {
+            expiry_updated_at: Some(lifecycle::format_lifecycle_date(now - time::Duration::seconds(1)).unwrap()),
             rules: Vec::new(),
         };
-        let stale = BucketLifecycleConfiguration {
-            expiry_updated_at: Some(Timestamp::from(
-                now - time::Duration::seconds(DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS + 1),
-            )),
+        let stale = PersistedLifecycleConfiguration {
+            expiry_updated_at: Some(
+                lifecycle::format_lifecycle_date(now - time::Duration::seconds(DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS + 1))
+                    .unwrap(),
+            ),
             rules: Vec::new(),
         };
 
@@ -10223,12 +10247,12 @@ mod tests {
         assert_eq!(deleted.version_id, None);
     }
 
-    fn expired_delete_marker_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn expired_delete_marker_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     expired_object_delete_marker: Some(true),
                     ..Default::default()
                 }),
@@ -10244,12 +10268,12 @@ mod tests {
         }
     }
 
-    fn latest_expiration_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn latest_expiration_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     days: Some(1),
                     ..Default::default()
                 }),
@@ -10265,12 +10289,12 @@ mod tests {
         }
     }
 
-    fn current_and_noncurrent_expiration_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn current_and_noncurrent_expiration_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     days: Some(30),
                     ..Default::default()
                 }),
@@ -10278,7 +10302,7 @@ mod tests {
                 del_marker_expiration: None,
                 filter: None,
                 id: Some("expire-current-and-noncurrent".to_string()),
-                noncurrent_version_expiration: Some(NoncurrentVersionExpiration {
+                noncurrent_version_expiration: Some(PersistedNoncurrentVersionExpiration {
                     noncurrent_days: Some(1),
                     newer_noncurrent_versions: None,
                 }),
@@ -10289,12 +10313,12 @@ mod tests {
         }
     }
 
-    fn all_versions_expiration_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn all_versions_expiration_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     days: Some(1),
                     expired_object_all_versions: Some(true),
                     ..Default::default()
@@ -10313,29 +10337,29 @@ mod tests {
 
     fn lock_enabled_without_default_retention() -> ObjectLockConfiguration {
         ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: None,
         }
     }
 
     fn lock_enabled_with_default_retention() -> ObjectLockConfiguration {
         ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: Some(ObjectLockRule {
                 default_retention: Some(DefaultRetention {
                     days: Some(30),
-                    mode: Some(ObjectLockRetentionMode::from_static(ObjectLockRetentionMode::COMPLIANCE)),
+                    mode: Some("COMPLIANCE".to_owned()),
                     years: None,
                 }),
             }),
         }
     }
 
-    fn latest_transition_lifecycle() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn latest_transition_lifecycle() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
                 expiration: None,
                 abort_incomplete_multipart_upload: None,
                 del_marker_expiration: None,
@@ -10344,10 +10368,10 @@ mod tests {
                 noncurrent_version_expiration: None,
                 noncurrent_version_transitions: None,
                 prefix: None,
-                transitions: Some(vec![Transition {
+                transitions: Some(vec![PersistedTransition {
                     days: Some(1),
                     date: None,
-                    storage_class: Some(TransitionStorageClass::from_static("WARM")),
+                    storage_class: Some("WARM".to_owned()),
                 }]),
             }],
         }
@@ -11314,13 +11338,7 @@ mod tests {
         let marker = marker
             .as_deref()
             .expect("production checkpoint should persist an object marker");
-        let marker_suffix = marker
-            .strip_prefix(&expected_marker)
-            .expect("production checkpoint should stop at the first page boundary");
-        assert!(
-            marker_suffix.is_empty() || marker_suffix.starts_with("[rustfs_cache:v2,"),
-            "unexpected production checkpoint marker suffix: {marker_suffix}"
-        );
+        assert_eq!(marker, expected_marker, "production checkpoint should stop at the first page boundary");
         assert_eq!(version_marker.as_deref(), Some("null"));
     }
 
@@ -12890,8 +12908,9 @@ mod tests {
 
         for (name, object, lock_config, expected) in cases {
             let object_opts = lifecycle::object_opts_from_object_info(&object);
+            let evaluator_lock_config = lock_config.clone().map(Arc::new);
             let batch_event = Evaluator::new(Arc::new(lifecycle.clone()))
-                .with_lock_retention(lock_config.clone().map(Arc::new))
+                .with_lock_retention(evaluator_lock_config)
                 .eval(&[object_opts])
                 .await
                 .expect("batch lifecycle evaluation should succeed")
@@ -14148,10 +14167,7 @@ mod tests {
         let object = current_object_with_metadata(
             ReplicationStatusType::Completed,
             HashMap::from([
-                (
-                    metadata_keys::OBJECT_LOCK_MODE.to_string(),
-                    s3s::dto::ObjectLockRetentionMode::COMPLIANCE.to_string(),
-                ),
+                (metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string()),
                 (metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(), retain_until),
             ]),
         );
@@ -14163,7 +14179,7 @@ mod tests {
 
     #[tokio::test]
     async fn restored_copy_expiry_is_not_blocked_by_retention() {
-        let lifecycle = BucketLifecycleConfiguration {
+        let lifecycle = PersistedLifecycleConfiguration {
             expiry_updated_at: None,
             rules: Vec::new(),
         };
@@ -14173,10 +14189,7 @@ mod tests {
         let mut object = current_object_with_metadata(
             ReplicationStatusType::Completed,
             HashMap::from([
-                (
-                    metadata_keys::OBJECT_LOCK_MODE.to_string(),
-                    ObjectLockRetentionMode::COMPLIANCE.to_string(),
-                ),
+                (metadata_keys::OBJECT_LOCK_MODE.to_string(), "COMPLIANCE".to_string()),
                 (metadata_keys::OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(), retain_until),
             ]),
         );
@@ -14393,7 +14406,7 @@ mod tests {
                 "removed" => {
                     metadata_sys::delete(&bucket, BUCKET_LIFECYCLE_CONFIG).await.unwrap();
                 }
-                "disabled" => changed.rules[0].status = ExpirationStatus::from_static(ExpirationStatus::DISABLED),
+                "disabled" => changed.rules[0].status = "Disabled".to_owned(),
                 "longer-days" => {
                     changed.rules[0]
                         .noncurrent_version_expiration
@@ -14509,7 +14522,7 @@ mod tests {
         );
     }
 
-    async fn install_test_expiry_policy(ecstore: &Arc<ECStore>, bucket: &str) -> Arc<BucketLifecycleConfiguration> {
+    async fn install_test_expiry_policy(ecstore: &Arc<ECStore>, bucket: &str) -> Arc<PersistedLifecycleConfiguration> {
         let xml = br#"<LifecycleConfiguration><Rule><ID>expire</ID><Status>Enabled</Status><Filter><Prefix></Prefix></Filter><Expiration><Days>0</Days></Expiration><NoncurrentVersionExpiration><NoncurrentDays>0</NoncurrentDays></NoncurrentVersionExpiration></Rule></LifecycleConfiguration>"#;
         metadata_sys::update_in(&ecstore.ctx, bucket, BUCKET_LIFECYCLE_CONFIG, xml.to_vec())
             .await
@@ -14982,7 +14995,11 @@ mod tests {
     #[tokio::test]
     #[serial]
     async fn tier_free_version_recovery_production_entrypoint_enqueues_seeded_item() {
-        let (disk_paths, ecstore) = setup_test_env().await;
+        let (disk_paths, ecstore) =
+            temp_env::async_with_vars([(super::ENV_TIER_FREE_VERSION_RECOVERY_ENABLED, Some("false"))], async {
+                setup_test_env().await
+            })
+            .await;
         let tier = format!("RECOVERY{}", Uuid::new_v4().simple()).to_uppercase();
         crate::services::tier::test_util::register_mock_tier(&ecstore.tier_config_mgr(), &tier).await;
         let runtime_state = install_unconsumed_runtime_expiry_worker(&ecstore, 1).await;

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

@@ -17,7 +17,7 @@ use crate::object_api::ObjectInfo;
 pub use rustfs_lifecycle::{
     Event, ExpirationOptions, IlmAction, LIFECYCLE_MALFORMED_XML_ERROR_KIND, Lifecycle, LifecycleCalculate, ObjectOpts,
     RuleValidate, TRANSITION_COMPLETE, TRANSITION_PENDING, TransitionOptions, abort_incomplete_multipart_upload_due,
-    expected_expiry_time, expiration_action_has_valid_target,
+    expected_expiry_time, expiration_action_has_valid_target, format_lifecycle_date, parse_lifecycle_date,
 };
 
 pub fn object_opts_from_object_info(oi: &ObjectInfo) -> ObjectOpts {

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

@@ -14,7 +14,8 @@
 
 use std::sync::Arc;
 
-use s3s::dto::{BucketLifecycleConfiguration, ObjectLockConfiguration};
+use rustfs_gateway_types::persistence::PersistedLifecycleConfiguration;
+use rustfs_gateway_types::persistence::PersistedObjectLockConfiguration as ObjectLockConfiguration;
 use time::OffsetDateTime;
 use uuid::Uuid;
 
@@ -24,7 +25,7 @@ use crate::error::{Error, Result};
 
 #[derive(Debug)]
 pub(crate) struct LifecycleExpiryConfigs {
-    pub(crate) lifecycle: Option<Arc<BucketLifecycleConfiguration>>,
+    pub(crate) lifecycle: Option<Arc<PersistedLifecycleConfiguration>>,
     pub(crate) object_lock: Option<Arc<ObjectLockConfiguration>>,
     pub(crate) bucket_incarnation_id: Uuid,
     pub(crate) table_bucket_enabled: bool,
@@ -107,7 +108,7 @@ pub(crate) async fn get_expiry_configs_for_options(
     })
 }
 
-pub(crate) async fn get_lifecycle_config(bucket: &str) -> Result<(BucketLifecycleConfiguration, OffsetDateTime)> {
+pub(crate) async fn get_lifecycle_config(bucket: &str) -> Result<(PersistedLifecycleConfiguration, OffsetDateTime)> {
     metadata_sys::get_lifecycle_config(bucket).await
 }
 
@@ -117,15 +118,15 @@ mod tests {
     use crate::bucket::metadata::BucketMetadata;
     use crate::bucket::metadata_sys::{self, test_support::isolated_store_over_temp_disks};
     use crate::storage_api_contracts::bucket::MakeBucketOptions;
-    use s3s::dto::{ExpirationStatus, LifecycleExpiration, LifecycleRule};
+    use rustfs_gateway_types::persistence::{PersistedLifecycleExpiration, PersistedLifecycleRule};
     use serial_test::serial;
 
-    fn lifecycle_config() -> BucketLifecycleConfiguration {
-        BucketLifecycleConfiguration {
+    fn lifecycle_config() -> PersistedLifecycleConfiguration {
+        PersistedLifecycleConfiguration {
             expiry_updated_at: None,
-            rules: vec![LifecycleRule {
-                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
-                expiration: Some(LifecycleExpiration {
+            rules: vec![PersistedLifecycleRule {
+                status: "Enabled".to_owned(),
+                expiration: Some(PersistedLifecycleExpiration {
                     days: Some(1),
                     ..Default::default()
                 }),

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

@@ -22,7 +22,7 @@ pub(crate) fn is_object_locked_by_metadata(user_defined: &HashMap<String, String
 }
 
 pub(crate) fn check_object_lock_for_deletion_with_config(
-    config: Option<&s3s::dto::ObjectLockConfiguration>,
+    config: Option<&rustfs_gateway_types::persistence::PersistedObjectLockConfiguration>,
     obj_info: &ObjectInfo,
     bypass_governance: bool,
 ) -> crate::error::Result<Option<ObjectLockBlockReason>> {

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

@@ -82,12 +82,13 @@ mod tests {
     use crate::bucket::replication::{DeleteReplicationConfigSnapshot, ReplicationObjectBridge};
     use crate::object_api::{ObjectInfo, ObjectOptions};
     use crate::storage_api_contracts::object::ObjectToDelete;
-    use rustfs_scanner_metrics::metrics::IlmAction;
-    use s3s::dto::{
-        BucketVersioningStatus, DeleteMarkerReplication, DeleteMarkerReplicationStatus, DeleteReplication,
-        DeleteReplicationStatus, Destination, ReplicationConfiguration, ReplicationRule, ReplicationRuleStatus,
-        VersioningConfiguration,
+    use rustfs_gateway_types::dto::Status;
+    use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
+    use rustfs_gateway_types::persistence::{
+        PersistedOptionalReplicationStatus, PersistedReplicationConfiguration, PersistedReplicationDestination,
+        PersistedReplicationRule, PersistedReplicationStatus,
     };
+    use rustfs_scanner_metrics::metrics::IlmAction;
     use uuid::Uuid;
 
     use super::*;
@@ -160,27 +161,27 @@ mod tests {
             for purge_enabled in [false, true] {
                 let snapshot = DeleteReplicationConfigSnapshot::from_configs_for_test(
                     VersioningConfiguration {
-                        status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                        status: Some("Enabled".to_owned()),
                         ..Default::default()
                     },
-                    Some(ReplicationConfiguration {
+                    Some(PersistedReplicationConfiguration {
                         role: String::new(),
-                        rules: vec![ReplicationRule {
-                            delete_marker_replication: Some(DeleteMarkerReplication {
+                        rules: vec![PersistedReplicationRule {
+                            delete_marker_replication: Some(PersistedOptionalReplicationStatus {
                                 status: Some(if marker_enabled {
-                                    DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::ENABLED)
+                                    Status::ENABLED.to_string()
                                 } else {
-                                    DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::DISABLED)
+                                    Status::DISABLED.to_string()
                                 }),
                             }),
-                            delete_replication: Some(DeleteReplication {
+                            delete_replication: Some(PersistedReplicationStatus {
                                 status: if purge_enabled {
-                                    DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED)
+                                    Status::ENABLED.to_string()
                                 } else {
-                                    DeleteReplicationStatus::from_static(DeleteReplicationStatus::DISABLED)
+                                    Status::DISABLED.to_string()
                                 },
                             }),
-                            destination: Destination {
+                            destination: PersistedReplicationDestination {
                                 bucket: "arn:rustfs:replication:target".to_string(),
                                 ..Default::default()
                             },
@@ -190,7 +191,7 @@ mod tests {
                             prefix: Some(String::new()),
                             priority: Some(1),
                             source_selection_criteria: None,
-                            status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                            status: Status::ENABLED.to_string(),
                         }],
                     }),
                 );

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

@@ -14,7 +14,7 @@
 
 use std::sync::Arc;
 
-use s3s::dto::BucketLifecycleConfiguration;
+use rustfs_gateway_types::persistence::PersistedLifecycleConfiguration;
 use tokio::sync::RwLock;
 use tokio_util::sync::CancellationToken;
 
@@ -52,7 +52,7 @@ pub(crate) fn deployment_id() -> Option<String> {
     sources::deployment_id()
 }
 
-pub(crate) async fn bucket_lifecycle_config(bucket: &str) -> Option<BucketLifecycleConfiguration> {
+pub(crate) async fn bucket_lifecycle_config(bucket: &str) -> Option<PersistedLifecycleConfiguration> {
     sources::bucket_lifecycle_config(bucket).await
 }
 

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

@@ -23,6 +23,7 @@ use std::{
 };
 
 use futures::StreamExt;
+use rand::RngExt as _;
 use serde::{Deserialize, Serialize};
 use sha2::{Digest, Sha256};
 use tokio_util::sync::CancellationToken;
@@ -70,6 +71,7 @@ pub(crate) const EVENT_LIFECYCLE_TIER_DELETE_JOURNAL: &str = "lifecycle_tier_del
 // successful pages while its wall-clock budget remains.
 pub const DEFAULT_TIER_DELETE_JOURNAL_RECOVERY_LIMIT: usize = 8;
 const TIER_DELETE_JOURNAL_RECOVERY_INTERVAL: Duration = Duration::from_secs(60);
+const TIER_DELETE_JOURNAL_RECOVERY_MAX_IDLE_INTERVAL: Duration = Duration::from_secs(5 * 60);
 const TIER_DELETE_JOURNAL_RECOVERY_TIMEOUT: Duration = Duration::from_secs(300);
 const TIER_DELETE_JOURNAL_RECOVERY_PASS_BUDGET: Duration = Duration::from_secs(240);
 const TIER_DELETE_REMOTE_DEADLINE: Duration = Duration::from_secs(30);
@@ -6104,57 +6106,114 @@ pub(crate) async fn recover_test_tier_delete_journal_pass_with_budget(
     recover_tier_delete_journal_pass(api, cancel_token, manifest_marker, journal_marker, budget).await
 }
 
-pub async fn run_tier_delete_journal_recovery_loop(api: Arc<ECStore>, cancel_token: CancellationToken) {
+fn tier_delete_journal_recovery_pass_is_idle(stats: &TierDeleteJournalRecoveryPassStats) -> bool {
+    stats.manifest_pages > 0
+        && stats.journal_pages > 0
+        && stats.manifest_errors == 0
+        && stats.journal_errors == 0
+        && !stats.deadline_exhausted
+        && !stats.canceled
+        && stats.manifests == TierDeleteDispatchManifestRecoveryStats::default()
+        && stats.journals == TierDeleteJournalRecoveryStats::default()
+}
+
+fn bounded_tier_delete_journal_recovery_delay(delay: Duration, percent: u32) -> Duration {
+    (delay.saturating_mul(percent) / 100)
+        .clamp(TIER_DELETE_JOURNAL_RECOVERY_INTERVAL, TIER_DELETE_JOURNAL_RECOVERY_MAX_IDLE_INTERVAL)
+}
+
+fn jitter_tier_delete_journal_recovery_delay(delay: Duration) -> Duration {
+    bounded_tier_delete_journal_recovery_delay(delay, rand::rng().random_range(90..=110))
+}
+
+async fn drive_tier_delete_journal_recovery(
+    cancel_token: &CancellationToken,
+    wait_for_wakeup: impl AsyncFn(),
+    mut recover: impl AsyncFnMut() -> TierDeleteJournalRecoveryPassStats,
+    jitter_delay: fn(Duration) -> Duration,
+) {
     let mut interval = tokio::time::interval(TIER_DELETE_JOURNAL_RECOVERY_INTERVAL);
     interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
-    let mut marker: Option<String> = None;
-    let mut manifest_marker: Option<String> = None;
+    let mut idle_interval = TIER_DELETE_JOURNAL_RECOVERY_INTERVAL;
 
     loop {
         tokio::select! {
             biased;
             _ = cancel_token.cancelled() => return,
+            _ = wait_for_wakeup() => {
+                idle_interval = TIER_DELETE_JOURNAL_RECOVERY_INTERVAL;
+                interval.reset();
+            },
             _ = interval.tick() => {},
-            _ = api.ctx.wait_for_tier_delete_journal_recovery() => {},
         }
 
-        let stats = recover_tier_delete_journal_pass(
-            api.clone(),
-            &cancel_token,
-            &mut manifest_marker,
-            &mut marker,
-            TIER_DELETE_JOURNAL_RECOVERY_PASS_BUDGET,
-        )
-        .await;
+        let stats = recover().await;
         if stats.canceled {
             return;
         }
-        debug!(
-            event = EVENT_LIFECYCLE_TIER_DELETE_JOURNAL,
-            component = LOG_COMPONENT_ECSTORE,
-            subsystem = LOG_SUBSYSTEM_LIFECYCLE,
-            manifest_pages = stats.manifest_pages,
-            manifest_scanned = stats.manifests.scanned,
-            manifest_advanced = stats.manifests.advanced,
-            manifest_deleted = stats.manifests.deleted,
-            manifest_retained = stats.manifests.retained,
-            manifest_failed = stats.manifests.failed,
-            manifest_errors = stats.manifest_errors,
-            manifest_truncated = stats.manifests.truncated,
-            manifest_next_marker = ?manifest_marker,
-            journal_pages = stats.journal_pages,
-            journal_scanned = stats.journals.scanned,
-            journal_deleted = stats.journals.deleted,
-            journal_failed = stats.journals.failed,
-            journal_errors = stats.journal_errors,
-            journal_truncated = stats.journals.truncated,
-            journal_next_marker = ?marker,
-            deadline_exhausted = stats.deadline_exhausted,
-            "Recovered tier delete journal pass"
-        );
+        if tier_delete_journal_recovery_pass_is_idle(&stats) {
+            // Only a completed empty sweep may defer discovery. Retained,
+            // failed, or paginated work keeps the original retry tick budget.
+            interval.reset_after(jitter_delay(idle_interval));
+            idle_interval = idle_interval
+                .saturating_mul(2)
+                .min(TIER_DELETE_JOURNAL_RECOVERY_MAX_IDLE_INTERVAL);
+        } else {
+            idle_interval = TIER_DELETE_JOURNAL_RECOVERY_INTERVAL;
+        }
     }
 }
 
+pub async fn run_tier_delete_journal_recovery_loop(api: Arc<ECStore>, cancel_token: CancellationToken) {
+    let mut marker: Option<String> = None;
+    let mut manifest_marker: Option<String> = None;
+    let wake_context = api.ctx.clone();
+    let recovery_cancel_token = cancel_token.clone();
+    drive_tier_delete_journal_recovery(
+        &cancel_token,
+        async move || wake_context.wait_for_tier_delete_journal_recovery().await,
+        async move || {
+            let stats = recover_tier_delete_journal_pass(
+                api.clone(),
+                &recovery_cancel_token,
+                &mut manifest_marker,
+                &mut marker,
+                TIER_DELETE_JOURNAL_RECOVERY_PASS_BUDGET,
+            )
+            .await;
+            if stats.canceled {
+                return stats;
+            }
+            debug!(
+                event = EVENT_LIFECYCLE_TIER_DELETE_JOURNAL,
+                component = LOG_COMPONENT_ECSTORE,
+                subsystem = LOG_SUBSYSTEM_LIFECYCLE,
+                manifest_pages = stats.manifest_pages,
+                manifest_scanned = stats.manifests.scanned,
+                manifest_advanced = stats.manifests.advanced,
+                manifest_deleted = stats.manifests.deleted,
+                manifest_retained = stats.manifests.retained,
+                manifest_failed = stats.manifests.failed,
+                manifest_errors = stats.manifest_errors,
+                manifest_truncated = stats.manifests.truncated,
+                manifest_next_marker = ?manifest_marker,
+                journal_pages = stats.journal_pages,
+                journal_scanned = stats.journals.scanned,
+                journal_deleted = stats.journals.deleted,
+                journal_failed = stats.journals.failed,
+                journal_errors = stats.journal_errors,
+                journal_truncated = stats.journals.truncated,
+                journal_next_marker = ?marker,
+                deadline_exhausted = stats.deadline_exhausted,
+                "Recovered tier delete journal pass"
+            );
+            stats
+        },
+        jitter_tier_delete_journal_recovery_delay,
+    )
+    .await;
+}
+
 async fn await_tier_delete_journal_recovery<T, F>(
     cancel_token: &CancellationToken,
     timeout: Duration,
@@ -6180,16 +6239,20 @@ mod tests {
     use super::{
         PersistedTierDeleteJournalEntry, TIER_DELETE_DISPATCH_MANIFEST_VERSION, TIER_DELETE_DISPATCH_PARENT_RECORD_TYPE,
         TIER_DELETE_DISPATCH_PARENT_VERSION, TIER_DELETE_JOURNAL_EXACT_VERSION, TIER_DELETE_JOURNAL_LEGACY_PREFIX,
+        TIER_DELETE_JOURNAL_RECOVERY_INTERVAL, TIER_DELETE_JOURNAL_RECOVERY_MAX_IDLE_INTERVAL,
         TIER_DELETE_JOURNAL_SOLE_OWNER_VERSION, TIER_DELETE_JOURNAL_STATE_VERSION, TIER_DELETE_JOURNAL_TRANSACTION_VERSION,
         TIER_DELETE_JOURNAL_V1_RECOVERY_SCHEMA, TIER_DELETE_JOURNAL_V2_RECOVERY_SCHEMA, TIER_DELETE_JOURNAL_V6_PREFIX,
-        TIER_DELETE_JOURNAL_VERSION, TierDeleteDispatchChunkBinding, TierDeleteDispatchManifest, TierDeleteDispatchManifestState,
-        TierDeleteDispatchParent, TierDeleteDispatchParentState, TierDeleteDispatchRecord, await_tier_delete_journal_recovery,
-        decode_tier_delete_dispatch_record, decode_tier_delete_journal_entry, encode_tier_delete_dispatch_manifest,
+        TIER_DELETE_JOURNAL_VERSION, TierDeleteDispatchChunkBinding, TierDeleteDispatchManifest,
+        TierDeleteDispatchManifestRecoveryStats, TierDeleteDispatchManifestState, TierDeleteDispatchParent,
+        TierDeleteDispatchParentState, TierDeleteDispatchRecord, TierDeleteJournalRecoveryPassStats,
+        await_tier_delete_journal_recovery, bounded_tier_delete_journal_recovery_delay, decode_tier_delete_dispatch_record,
+        decode_tier_delete_journal_entry, drive_tier_delete_journal_recovery, encode_tier_delete_dispatch_manifest,
         encode_tier_delete_dispatch_parent, encode_tier_delete_journal_entry, object_info_references_tier_delete,
         record_tier_delete_journal_backend_identity, same_tier_delete_authorization_identity, same_tier_delete_journal_identity,
         tier_delete_dispatch_child_matches_parent, tier_delete_dispatch_chunk_manifest_object_name,
         tier_delete_dispatch_journal_set_digest, tier_delete_dispatch_manifest_object_name, tier_delete_journal_object_name,
-        tier_delete_source_matches_dispatch_scope, validate_legacy_tier_delete_recovery_source,
+        tier_delete_journal_recovery_pass_is_idle, tier_delete_source_matches_dispatch_scope,
+        validate_legacy_tier_delete_recovery_source,
     };
     use crate::bucket::lifecycle::tier_sweeper::{
         Jentry, TierDeleteDispatchBinding, TierDeleteJournalState, TierDeleteSourceIdentity,
@@ -6833,6 +6896,213 @@ mod tests {
         assert!(err.to_string().contains("decode tier delete journal failed"));
     }
 
+    fn empty_recovery_pass() -> TierDeleteJournalRecoveryPassStats {
+        TierDeleteJournalRecoveryPassStats {
+            manifest_pages: 1,
+            journal_pages: 1,
+            ..Default::default()
+        }
+    }
+
+    #[test]
+    fn tier_delete_journal_recovery_requires_both_completed_empty_queues_for_idle() {
+        assert!(tier_delete_journal_recovery_pass_is_idle(&empty_recovery_pass()));
+        let mut cases = Vec::new();
+        cases.push(TierDeleteJournalRecoveryPassStats::default());
+        for change in [
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.manifest_pages = 0,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.journal_pages = 0,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.manifest_errors = 1,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.journal_errors = 1,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.deadline_exhausted = true,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.canceled = true,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.manifests.scanned = 1,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.manifests.retained = 1,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.manifests.failed = 1,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.manifests.truncated = true,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.manifests.next_marker = Some("next".to_owned()),
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.journals.scanned = 1,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.journals.failed = 1,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.journals.truncated = true,
+            |stats: &mut TierDeleteJournalRecoveryPassStats| stats.journals.next_marker = Some("next".to_owned()),
+        ] {
+            let mut stats = empty_recovery_pass();
+            change(&mut stats);
+            cases.push(stats);
+        }
+        for stats in cases {
+            assert!(
+                !tier_delete_journal_recovery_pass_is_idle(&stats),
+                "unfinished recovery was considered idle: {stats:?}"
+            );
+        }
+    }
+
+    #[test]
+    fn tier_delete_journal_recovery_jitter_preserves_discovery_bounds() {
+        for delay in [
+            Duration::ZERO,
+            Duration::from_secs(60),
+            Duration::from_secs(120),
+            Duration::from_secs(300),
+            Duration::MAX,
+        ] {
+            for percent in [90, 100, 110] {
+                let jittered = bounded_tier_delete_journal_recovery_delay(delay, percent);
+                assert!(jittered >= TIER_DELETE_JOURNAL_RECOVERY_INTERVAL);
+                assert!(jittered <= TIER_DELETE_JOURNAL_RECOVERY_MAX_IDLE_INTERVAL);
+                if delay == Duration::from_secs(120) {
+                    assert!((Duration::from_secs(108)..=Duration::from_secs(132)).contains(&jittered));
+                }
+            }
+        }
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn tier_delete_journal_recovery_loop_starts_immediately_backs_off_and_wakes() {
+        let cancel = CancellationToken::new();
+        let wake = tokio::sync::Notify::new();
+        let (tx, mut rx) = tokio::sync::mpsc::unbounded_channel();
+        let started = tokio::time::Instant::now();
+        let recovery = drive_tier_delete_journal_recovery(
+            &cancel,
+            async || wake.notified().await,
+            async || {
+                tx.send(tokio::time::Instant::now()).expect("test must observe each pass");
+                empty_recovery_pass()
+            },
+            |delay| bounded_tier_delete_journal_recovery_delay(delay, 110),
+        );
+        let exercise = async {
+            assert_eq!(rx.recv().await.expect("startup recovery must run"), started);
+            let mut elapsed = Duration::ZERO;
+            for delay in [66, 132, 264, 300, 300] {
+                tokio::time::advance(Duration::from_secs(delay - 1)).await;
+                tokio::task::yield_now().await;
+                assert!(rx.try_recv().is_err(), "idle pass ran before its backoff");
+                tokio::time::advance(Duration::from_secs(1)).await;
+                elapsed += Duration::from_secs(delay);
+                assert_eq!(rx.recv().await.expect("bounded idle recovery must run"), started + elapsed);
+            }
+            wake.notify_one();
+            assert_eq!(rx.recv().await.expect("notification must bypass idle delay"), started + elapsed);
+            let first_idle_delay = Duration::from_secs(66);
+            tokio::time::advance(first_idle_delay).await;
+            assert_eq!(
+                rx.recv().await.expect("wakeup must reset idle backoff"),
+                started + elapsed + first_idle_delay
+            );
+            cancel.cancel();
+        };
+        tokio::join!(recovery, exercise);
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn tier_delete_journal_recovery_loop_preserves_known_work_tick_deadline() {
+        let cancel = CancellationToken::new();
+        let (tx, mut rx) = tokio::sync::mpsc::unbounded_channel();
+        let started = tokio::time::Instant::now();
+        let mut pass = 0;
+        let recovery = drive_tier_delete_journal_recovery(
+            &cancel,
+            async || std::future::pending::<()>().await,
+            async || {
+                tx.send(tokio::time::Instant::now()).expect("test must observe each pass");
+                pass += 1;
+                if pass <= 3 {
+                    return empty_recovery_pass();
+                }
+                if pass == 4 {
+                    tokio::time::sleep(Duration::from_secs(40)).await;
+                }
+                let mut stats = empty_recovery_pass();
+                match pass {
+                    4 => {
+                        stats.manifests = TierDeleteDispatchManifestRecoveryStats {
+                            retained: 1,
+                            scanned: 1,
+                            ..Default::default()
+                        }
+                    }
+                    5 => stats.manifest_errors = 1,
+                    6 => stats.journal_errors = 1,
+                    7 => stats.manifests.truncated = true,
+                    8 => stats.journals.failed = 1,
+                    9 => stats.deadline_exhausted = true,
+                    _ => stats.journals.next_marker = Some("next-page".to_owned()),
+                }
+                stats
+            },
+            std::convert::identity,
+        );
+        let exercise = async {
+            for expected in [0, 60, 180, 420] {
+                assert_eq!(
+                    rx.recv().await.expect("idle and work passes must run"),
+                    started + Duration::from_secs(expected)
+                );
+            }
+            tokio::time::advance(Duration::from_secs(59)).await;
+            tokio::task::yield_now().await;
+            assert!(rx.try_recv().is_err(), "known work ran before the next tick");
+            tokio::time::advance(Duration::from_secs(1)).await;
+            for expected in [480, 540, 600, 660, 720, 780, 840] {
+                assert_eq!(
+                    rx.recv().await.expect("unfinished work must retain the base tick deadline"),
+                    started + Duration::from_secs(expected)
+                );
+            }
+            cancel.cancel();
+        };
+        tokio::join!(recovery, exercise);
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn tier_delete_journal_recovery_loop_keeps_wakeup_received_during_a_pass() {
+        let cancel = CancellationToken::new();
+        let wake = tokio::sync::Notify::new();
+        let (tx, mut rx) = tokio::sync::mpsc::unbounded_channel();
+        let started = tokio::time::Instant::now();
+        let mut first = true;
+        let recovery = drive_tier_delete_journal_recovery(
+            &cancel,
+            async || wake.notified().await,
+            async || {
+                tx.send(tokio::time::Instant::now()).expect("test must observe each pass");
+                if first {
+                    first = false;
+                    tokio::time::sleep(Duration::from_secs(40)).await;
+                }
+                empty_recovery_pass()
+            },
+            std::convert::identity,
+        );
+        let exercise = async {
+            assert_eq!(rx.recv().await.expect("startup recovery must run"), started);
+            wake.notify_one();
+            tokio::time::advance(Duration::from_secs(40)).await;
+            assert_eq!(
+                rx.recv().await.expect("a notification during recovery must remain pending"),
+                started + Duration::from_secs(40)
+            );
+            cancel.cancel();
+        };
+        tokio::join!(recovery, exercise);
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn tier_delete_journal_recovery_loop_cancellation_wins_over_ready_wakeup() {
+        let cancel = CancellationToken::new();
+        cancel.cancel();
+        drive_tier_delete_journal_recovery(
+            &cancel,
+            async || {},
+            async || panic!("shutdown must prevent admission even when a wakeup and startup tick are ready"),
+            std::convert::identity,
+        )
+        .await;
+    }
+
     #[tokio::test]
     async fn tier_delete_journal_recovery_has_a_hard_outer_timeout() {
         let result = await_tier_delete_journal_recovery(

crates/ecstore/src/bucket/metadata.rs

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 use super::msgp_decode::{read_msgp_ext8_time, skip_msgp_value, write_msgp_time};
-use super::object_lock::{ObjectLockApi, ObjectLockStatusExt};
+use super::object_lock::ObjectLockApi;
 use super::versioning::VersioningApi;
 use super::{quota::BucketQuota, target::BucketTargets};
 use crate::bucket::replication::invalid_replication_config_status_field;
@@ -25,12 +25,17 @@ use crate::object_api::WriteCompletion;
 use crate::runtime::sources as runtime_sources;
 use crate::store::ECStore;
 use byteorder::{BigEndian, ByteOrder, LittleEndian};
-use rustfs_policy::policy::BucketPolicy;
-use s3s::dto::{
-    AccelerateConfiguration, BucketLifecycleConfiguration, BucketLoggingStatus, BucketVersioningStatus, CORSConfiguration,
-    NotificationConfiguration, ObjectLockConfiguration, PublicAccessBlockConfiguration, ReplicationConfiguration,
-    RequestPaymentConfiguration, ServerSideEncryptionConfiguration, Tagging, VersioningConfiguration, WebsiteConfiguration,
+use rustfs_gateway_types::cors_tagging::{PersistedCorsConfiguration as CORSConfiguration, PersistedTagging as Tagging};
+use rustfs_gateway_types::persistence::{
+    PersistedAccelerateConfiguration as AccelerateConfiguration,
+    PersistedBucketEncryptionConfiguration as ServerSideEncryptionConfiguration,
+    PersistedBucketLoggingStatus as BucketLoggingStatus, PersistedLifecycleConfiguration,
+    PersistedNotificationConfiguration as NotificationConfiguration, PersistedObjectLockConfiguration as ObjectLockConfiguration,
+    PersistedPublicAccessBlockConfiguration as PublicAccessBlockConfiguration, PersistedReplicationConfiguration,
+    PersistedRequestPaymentConfiguration as RequestPaymentConfiguration,
+    PersistedVersioningConfiguration as VersioningConfiguration, PersistedWebsiteConfiguration as WebsiteConfiguration,
 };
+use rustfs_policy::policy::BucketPolicy;
 use serde::Serializer;
 use sha2::{Digest, Sha256};
 use std::collections::HashMap;
@@ -40,23 +45,11 @@ use time::{Date, OffsetDateTime, PrimitiveDateTime, Time as CivilTime, UtcOffset
 use tracing::error;
 use uuid::Uuid;
 
-// The serving-layer DTO impls for the storage-level Object Lock traits live
-// here because this module owns the persisted `ObjectLockConfiguration`
-// during the s3s ratchet migration (rustfs/backlog#1842).
+// The persisted Object Lock configuration answers the storage-level trait
+// here because this module owns the persisted `ObjectLockConfiguration`.
 impl ObjectLockApi for ObjectLockConfiguration {
     fn enabled(&self) -> bool {
-        self.object_lock_enabled
-            .as_ref()
-            .is_some_and(|v| v.as_str() == s3s::dto::ObjectLockEnabled::ENABLED)
-    }
-}
-
-impl ObjectLockStatusExt for s3s::dto::ObjectLockLegalHoldStatus {
-    fn valid(&self) -> bool {
-        matches!(
-            self.as_str(),
-            s3s::dto::ObjectLockLegalHoldStatus::ON | s3s::dto::ObjectLockLegalHoldStatus::OFF
-        )
+        self.object_lock_enabled.as_deref() == Some("Enabled")
     }
 }
 
@@ -439,6 +432,8 @@ pub struct BucketMetadata {
     pub bucket_incarnation_id: Uuid,
     pub(crate) bucket_incarnation_sidecar: bool,
     pub(crate) bucket_creation_committed: bool,
+    // A legacy default is not proof of an interrupted creation.
+    pub(crate) bucket_creation_commit_record_present: bool,
     pub policy_config_json: Vec<u8>,
     pub notification_config_xml: Vec<u8>,
     pub lifecycle_config_xml: Vec<u8>,
@@ -487,13 +482,13 @@ pub struct BucketMetadata {
 
     pub policy_config: Option<BucketPolicy>,
     pub notification_config: Option<NotificationConfiguration>,
-    pub lifecycle_config: Option<BucketLifecycleConfiguration>,
+    pub lifecycle_config: Option<PersistedLifecycleConfiguration>,
     pub object_lock_config: Option<ObjectLockConfiguration>,
     pub versioning_config: Option<VersioningConfiguration>,
     pub sse_config: Option<ServerSideEncryptionConfiguration>,
     pub tagging_config: Option<Tagging>,
     pub quota_config: Option<BucketQuota>,
-    pub replication_config: Option<ReplicationConfiguration>,
+    pub replication_config: Option<PersistedReplicationConfiguration>,
     pub bucket_target_config: Option<BucketTargets>,
     pub bucket_target_config_meta: Option<HashMap<String, String>>,
     pub cors_config: Option<CORSConfiguration>,
@@ -514,6 +509,7 @@ impl Default for BucketMetadata {
             bucket_incarnation_id: Uuid::nil(),
             bucket_incarnation_sidecar: false,
             bucket_creation_committed: false,
+            bucket_creation_commit_record_present: false,
             policy_config_json: Default::default(),
             notification_config_xml: Default::default(),
             lifecycle_config_xml: Default::default(),
@@ -752,7 +748,10 @@ impl BucketMetadata {
                 "TableBucketConfigUpdatedAt" => self.table_bucket_config_updated_at = read_msgp_time_value(rd)?,
                 "DurabilityConfigUpdatedAt" => self.durability_config_updated_at = read_msgp_time_value(rd)?,
                 "OnDemandMigrationConfigUpdatedAt" => self.on_demand_migration_config_updated_at = read_msgp_time_value(rd)?,
-                "BucketCreationCommitted" => self.bucket_creation_committed = read_msgp_bool(rd)?,
+                "BucketCreationCommitted" => {
+                    self.bucket_creation_committed = read_msgp_bool(rd)?;
+                    self.bucket_creation_commit_record_present = true;
+                }
                 other => {
                     tracing::debug!(field = %other, "BucketMetadata decode_from: skipping unknown field");
                     skip_msgp_value(rd)?;
@@ -765,8 +764,10 @@ impl BucketMetadata {
 
     /// Encode to msgp bytes. Field order follows MinIO BucketMetadata for compatibility.
     pub fn encode_to<W: Write>(&self, wr: &mut W) -> Result<()> {
-        // Map size: MinIO fields (25) + RustFS extensions (22)
-        let map_len: u32 = 47;
+        // Preserve the absence of a legacy commit record across config rewrites.
+        let commit_record_present = self.bucket_creation_commit_record_present || self.bucket_creation_committed;
+        // Map size: MinIO fields (25) + RustFS extensions (21) + optional commit record.
+        let map_len: u32 = 46 + u32::from(commit_record_present);
         rmp::encode::write_map_len(wr, map_len)?;
 
         // MinIO field order (same as Go struct)
@@ -847,8 +848,10 @@ impl BucketMetadata {
         write_msgp_time(wr, self.durability_config_updated_at)?;
         rmp::encode::write_str(wr, "OnDemandMigrationConfigUpdatedAt")?;
         write_msgp_time(wr, self.on_demand_migration_config_updated_at)?;
-        rmp::encode::write_str(wr, "BucketCreationCommitted")?;
-        rmp::encode::write_bool(wr, self.bucket_creation_committed)?;
+        if commit_record_present {
+            rmp::encode::write_str(wr, "BucketCreationCommitted")?;
+            rmp::encode::write_bool(wr, self.bucket_creation_committed)?;
+        }
 
         Ok(())
     }
@@ -1018,9 +1021,11 @@ impl BucketMetadata {
                     None
                 } else {
                     let config = deserialize::<VersioningConfiguration>(&data)?;
-                    if config.status.as_ref().is_some_and(|status| {
-                        !matches!(status.as_str(), BucketVersioningStatus::ENABLED | BucketVersioningStatus::SUSPENDED)
-                    }) {
+                    if config
+                        .status
+                        .as_ref()
+                        .is_some_and(|status| !matches!(status.as_str(), "Enabled" | "Suspended"))
+                    {
                         return Err(Error::other("bucket versioning configuration has an invalid status"));
                     }
                     Some(config)
@@ -1033,7 +1038,7 @@ impl BucketMetadata {
                 let config = if data.is_empty() {
                     None
                 } else {
-                    let config = deserialize::<ReplicationConfiguration>(&data)?;
+                    let config = deserialize::<PersistedReplicationConfiguration>(&data)?;
                     if let Some(field) = invalid_replication_config_status_field(&config) {
                         return Err(Error::other(format!("replication field {field} has an invalid status")));
                     }
@@ -1227,8 +1232,8 @@ impl BucketMetadata {
         }
         self.lifecycle_config = None;
         if !self.lifecycle_config_xml.is_empty()
-            && let Err(e) =
-                deserialize::<BucketLifecycleConfiguration>(&self.lifecycle_config_xml).map(|c| self.lifecycle_config = Some(c))
+            && let Err(e) = deserialize::<PersistedLifecycleConfiguration>(&self.lifecycle_config_xml)
+                .map(|c| self.lifecycle_config = Some(c))
         {
             tracing::warn!(
                 event = "bucket_metadata_parse_failed",
@@ -1314,8 +1319,8 @@ impl BucketMetadata {
         }
         self.replication_config = None;
         if !self.replication_config_xml.is_empty()
-            && let Err(e) =
-                deserialize::<ReplicationConfiguration>(&self.replication_config_xml).map(|c| self.replication_config = Some(c))
+            && let Err(e) = deserialize::<PersistedReplicationConfiguration>(&self.replication_config_xml)
+                .map(|c| self.replication_config = Some(c))
         {
             tracing::warn!(
                 event = "bucket_metadata_parse_failed",
@@ -1691,6 +1696,13 @@ mod test {
         let mut bm = BucketMetadata::unmarshal(&body[4..]).expect("unmarshal recovered blob");
         assert_eq!(bm.name, "interop");
         assert!(!bm.bucket_creation_committed, "legacy metadata must default to uncommitted");
+        assert!(!bm.bucket_creation_commit_record_present, "legacy metadata has no creation-commit record");
+        let rewritten = BucketMetadata::unmarshal(&bm.marshal_msg().expect("rewrite legacy metadata"))
+            .expect("read rewritten legacy metadata");
+        assert!(
+            !rewritten.bucket_creation_commit_record_present,
+            "rewriting legacy metadata must not fabricate a creation intent"
+        );
         bm.parse_all_configs().expect("parse recovered configs");
         assert!(bm.lifecycle_config.is_some());
     }
@@ -1708,6 +1720,7 @@ mod test {
 
         assert_eq!(bm.name, new.name);
         assert!(new.bucket_creation_committed);
+        assert!(new.bucket_creation_commit_record_present);
         assert!(!bm.bucket_incarnation_id.is_nil());
         assert_eq!(bm.bucket_incarnation_id, new.bucket_incarnation_id);
     }
@@ -1812,17 +1825,15 @@ mod test {
 
     #[test]
     fn object_locking_requires_lock_metadata_not_plain_versioning() {
-        use s3s::dto::ObjectLockEnabled;
-
         let mut bm = BucketMetadata::new("test-bucket");
         bm.versioning_config = Some(VersioningConfiguration {
-            status: Some(s3s::dto::BucketVersioningStatus::from_static("Enabled")),
+            status: Some("Enabled".to_owned()),
             ..Default::default()
         });
         assert!(!bm.object_locking());
 
         bm.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             ..Default::default()
         });
         assert!(bm.object_locking());

crates/ecstore/src/bucket/metadata_sys.rs

@@ -30,14 +30,18 @@ use crate::storage_api_contracts::heal::HealOperations as _;
 use crate::storage_api_contracts::namespace::NamespaceLocking as _;
 use crate::store::{ECStore, await_bucket_namespace_operation};
 use futures::future::join_all;
+use rustfs_gateway_types::cors_tagging::{PersistedCorsConfiguration as CORSConfiguration, PersistedTagging as Tagging};
+use rustfs_gateway_types::persistence::{
+    PersistedAccelerateConfiguration as AccelerateConfiguration,
+    PersistedBucketEncryptionConfiguration as ServerSideEncryptionConfiguration,
+    PersistedBucketLoggingStatus as BucketLoggingStatus, PersistedLifecycleConfiguration,
+    PersistedNotificationConfiguration as NotificationConfiguration, PersistedObjectLockConfiguration as ObjectLockConfiguration,
+    PersistedPublicAccessBlockConfiguration as PublicAccessBlockConfiguration, PersistedReplicationConfiguration,
+    PersistedRequestPaymentConfiguration as RequestPaymentConfiguration,
+    PersistedVersioningConfiguration as VersioningConfiguration, PersistedWebsiteConfiguration as WebsiteConfiguration,
+};
 use rustfs_heal_contracts::heal_channel::HealOpts;
 use rustfs_policy::policy::BucketPolicy;
-use s3s::dto::ReplicationConfiguration;
-use s3s::dto::{
-    AccelerateConfiguration, BucketLifecycleConfiguration, BucketLoggingStatus, CORSConfiguration, NotificationConfiguration,
-    ObjectLockConfiguration, ObjectLockEnabled, ObjectLockRetentionMode, PublicAccessBlockConfiguration,
-    RequestPaymentConfiguration, ServerSideEncryptionConfiguration, Tagging, VersioningConfiguration, WebsiteConfiguration,
-};
 use std::collections::HashSet;
 use std::time::Duration;
 use std::{
@@ -168,7 +172,7 @@ pub(crate) fn object_lock_config_state_from_authoritative_metadata(bm: &BucketMe
     if bm.lock_enabled {
         return Ok(ObjectLockConfigState::Configured {
             config: ObjectLockConfiguration {
-                object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+                object_lock_enabled: Some("Enabled".to_owned()),
                 rule: None,
             },
             updated_at: bm.object_lock_config_updated_at,
@@ -212,17 +216,10 @@ pub(crate) fn configured_object_lock_state_for_tests(
 ) -> ObjectLockConfigState {
     ObjectLockConfigState::Configured {
         config: ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
-            rule: Some(s3s::dto::ObjectLockRule {
-                default_retention: Some(s3s::dto::DefaultRetention {
-                    mode: Some(s3s::dto::ObjectLockRetentionMode::from_static(match mode {
-                        crate::bucket::object_lock::types::RetentionMode::Governance => {
-                            s3s::dto::ObjectLockRetentionMode::GOVERNANCE
-                        }
-                        crate::bucket::object_lock::types::RetentionMode::Compliance => {
-                            s3s::dto::ObjectLockRetentionMode::COMPLIANCE
-                        }
-                    })),
+            object_lock_enabled: Some("Enabled".to_owned()),
+            rule: Some(rustfs_gateway_types::persistence::PersistedObjectLockRule {
+                default_retention: Some(rustfs_gateway_types::persistence::PersistedDefaultRetention {
+                    mode: Some(mode.as_str().to_owned()),
                     days: Some(days),
                     years: None,
                 }),
@@ -233,7 +230,7 @@ pub(crate) fn configured_object_lock_state_for_tests(
 }
 
 fn validate_authoritative_object_lock_config(config: &ObjectLockConfiguration) -> Result<()> {
-    if config.object_lock_enabled.as_ref().map(ObjectLockEnabled::as_str) != Some(ObjectLockEnabled::ENABLED) {
+    if config.object_lock_enabled.as_deref() != Some("Enabled") {
         return Err(Error::other("persisted bucket Object Lock enabled state is invalid"));
     }
 
@@ -246,7 +243,7 @@ fn validate_authoritative_object_lock_config(config: &ObjectLockConfiguration) -
     if !retention
         .mode
         .as_ref()
-        .is_some_and(|mode| matches!(mode.as_str(), ObjectLockRetentionMode::COMPLIANCE | ObjectLockRetentionMode::GOVERNANCE))
+        .is_some_and(|mode| matches!(mode.as_str(), "COMPLIANCE" | "GOVERNANCE"))
     {
         return Err(Error::other("persisted bucket Object Lock retention mode is invalid"));
     }
@@ -1368,7 +1365,7 @@ pub async fn get_public_access_block_config(bucket: &str) -> Result<(PublicAcces
     bucket_meta_sys.get_public_access_block_config(bucket).await
 }
 
-pub async fn get_lifecycle_config(bucket: &str) -> Result<(BucketLifecycleConfiguration, OffsetDateTime)> {
+pub async fn get_lifecycle_config(bucket: &str) -> Result<(PersistedLifecycleConfiguration, OffsetDateTime)> {
     let bucket_meta_sys_lock = get_bucket_metadata_sys()?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
@@ -1587,7 +1584,7 @@ pub(crate) async fn get_cached_quota_config_and_incarnation_in(
     bucket_meta_sys.cached_quota_config_and_incarnation(bucket).await
 }
 
-pub async fn get_replication_config(bucket: &str) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
+pub async fn get_replication_config(bucket: &str) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
     let bucket_meta_sys_lock = get_bucket_metadata_sys()?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
@@ -1597,7 +1594,7 @@ pub async fn get_replication_config(bucket: &str) -> Result<(ReplicationConfigur
 pub(crate) async fn get_replication_config_in(
     ctx: &crate::runtime::instance::InstanceContext,
     bucket: &str,
-) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
+) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
     let bucket_meta_sys_lock = bucket_metadata_sys_of(ctx)?;
     let bucket_meta_sys = bucket_meta_sys_lock.read().await;
 
@@ -1723,9 +1720,27 @@ struct MetadataPublishGuard {
     _guard: tokio::sync::OwnedMutexGuard<MetadataPublishLockState>,
 }
 
+#[derive(Debug, Clone)]
+pub(crate) struct PendingBucketCreationProof {
+    pub(crate) generation: Uuid,
+    pub(crate) intent: Vec<u8>,
+    pub(crate) versioning_requested: bool,
+    pub(crate) witnesses: Vec<crate::disk::os::NativeBucketCreationWitness>,
+}
+
+impl PendingBucketCreationProof {
+    pub(crate) fn matches_metadata(&self, metadata: &BucketMetadata) -> Result<bool> {
+        Ok(!metadata.bucket_creation_committed
+            && metadata.bucket_incarnation_sidecar
+            && metadata.bucket_incarnation_id == self.generation
+            && metadata.marshal_msg()? == self.intent)
+    }
+}
+
 #[derive(Debug, Clone)]
 pub struct BucketMetadataSys {
     metadata_map: Arc<RwLock<HashMap<String, Arc<BucketMetadata>>>>,
+    pending_bucket_creations: Arc<RwLock<HashMap<String, PendingBucketCreationProof>>>,
     /// Serializes metadata-map commits and their derived cache updates for one
     /// bucket. Namespace locks, when present, are acquired before this lock.
     metadata_publish_locks: Arc<MetadataPublishLockRegistry>,
@@ -1769,6 +1784,7 @@ impl BucketMetadataSys {
     pub fn new(api: Arc<ECStore>) -> Self {
         Self {
             metadata_map: Arc::new(RwLock::new(HashMap::new())),
+            pending_bucket_creations: Arc::new(RwLock::new(HashMap::new())),
             metadata_publish_locks: Arc::new(MetadataPublishLockRegistry {
                 locks: StdMutex::new(HashMap::new()),
             }),
@@ -1799,6 +1815,81 @@ impl BucketMetadataSys {
             .expect("bucket metadata object store should still be live")
     }
 
+    pub(crate) async fn pending_bucket_creation(&self, bucket: &str) -> Option<PendingBucketCreationProof> {
+        self.pending_bucket_creations.read().await.get(bucket).cloned()
+    }
+
+    pub(crate) async fn grant_pending_bucket_creation(&self, bucket: &str, proof: PendingBucketCreationProof) {
+        self.pending_bucket_creations.write().await.insert(bucket.to_owned(), proof);
+    }
+
+    pub(crate) async fn revoke_pending_bucket_creation(&self, bucket: &str) {
+        self.pending_bucket_creations.write().await.remove(bucket);
+    }
+
+    pub(crate) async fn consume_pending_bucket_creation(
+        &self,
+        bucket: &str,
+        metadata: &BucketMetadata,
+        witnesses: &[crate::disk::os::NativeBucketCreationWitness],
+    ) -> Result<bool> {
+        let mut pending = self.pending_bucket_creations.write().await;
+        let Some(proof) = pending.get(bucket) else {
+            return Ok(false);
+        };
+        if !proof.matches_metadata(metadata)?
+            || proof.witnesses.len() != witnesses.len()
+            || !proof
+                .witnesses
+                .iter()
+                .zip(witnesses)
+                .all(|(old, current)| old.matches(current))
+            || !crate::disk::os::consume_native_creation_witnesses(witnesses)
+        {
+            return Ok(false);
+        }
+        pending.remove(bucket);
+        Ok(true)
+    }
+
+    #[cfg(test)]
+    pub(crate) async fn refresh_bucket_for_test(&self, bucket: &str) -> Result<()> {
+        let expected = self.metadata_map.read().await.get(bucket).cloned();
+        let lock = self.object_store().new_ns_lock(bucket, bucket).await?;
+        let guard = lock.get_read_lock(crate::set_disk::get_lock_acquire_timeout()).await?;
+        self.load_bucket_under_namespace(bucket, MetadataLoadMode::Refresh, expected.as_ref(), &guard)
+            .await
+    }
+
+    #[cfg(test)]
+    pub(crate) async fn publish_persisted_metadata_for_test(&self, bucket: &str) -> Result<()> {
+        let expected = self.metadata_map.read().await.get(bucket).cloned();
+        let lock = self.object_store().new_ns_lock(bucket, bucket).await?;
+        let guard = lock.get_read_lock(crate::set_disk::get_lock_acquire_timeout()).await?;
+        let (metadata, persisted) = await_bucket_namespace_operation(
+            Some(&guard),
+            bucket,
+            "test cached metadata load",
+            load_bucket_metadata_parse_with_presence(self.object_store(), bucket, true),
+        )
+        .await?;
+        self.publish_if_unchanged(bucket, expected.as_ref(), metadata, persisted, &guard)
+            .await
+    }
+
+    #[cfg(test)]
+    pub(crate) async fn migrate_bucket_creation_commit_for_test(&self, bucket: &str) -> Result<()> {
+        self.migrate_bucket_creation_commit(bucket).await.map(|_| ())
+    }
+
+    #[cfg(test)]
+    pub(crate) async fn legacy_migration_reports_missing_bucket_for_test(&self, bucket: &str) -> Result<bool> {
+        Ok(matches!(
+            self.migrate_legacy_metadata(bucket).await?,
+            BucketMetadataAuthority::MissingBucket
+        ))
+    }
+
     fn object_store_if_live(&self) -> Option<Arc<ECStore>> {
         self.api.upgrade()
     }
@@ -1989,6 +2080,7 @@ impl BucketMetadataSys {
         expected: Option<&Arc<BucketMetadata>>,
         namespace_guard: &rustfs_lock::NamespaceLockGuard,
     ) -> Result<()> {
+        self.revoke_pending_bucket_creation(bucket).await;
         if !await_bucket_namespace_operation(
             Some(namespace_guard),
             bucket,
@@ -2072,6 +2164,7 @@ impl BucketMetadataSys {
         persisted: bool,
         namespace_guard: &rustfs_lock::NamespaceLockGuard,
     ) -> Result<()> {
+        self.revoke_pending_bucket_creation(bucket).await;
         if !persisted {
             let _publish_guard = self
                 .lock_metadata_publish(bucket, namespace_guard, "refreshed bucket metadata absence publish")
@@ -2133,6 +2226,7 @@ impl BucketMetadataSys {
     }
 
     pub async fn set(&self, bucket: String, bm: Arc<BucketMetadata>) {
+        self.revoke_pending_bucket_creation(&bucket).await;
         if !is_meta_bucketname(&bucket) {
             let publish_lock = self.metadata_publish_lock(&bucket);
             let _publish_guard = publish_lock.lock().await;
@@ -2154,6 +2248,7 @@ impl BucketMetadataSys {
         if is_meta_bucketname(bucket) {
             return false;
         }
+        self.revoke_pending_bucket_creation(bucket).await;
         let publish_lock = self.metadata_publish_lock(bucket);
         let _publish_guard = publish_lock.lock().await;
         let mut map = self.metadata_map.write().await;
@@ -2171,6 +2266,7 @@ impl BucketMetadataSys {
     }
 
     async fn _reset(&mut self) {
+        self.pending_bucket_creations.write().await.clear();
         let mut map = self.metadata_map.write().await;
         map.clear();
         drop(map);
@@ -2309,6 +2405,7 @@ impl BucketMetadataSys {
     /// server's metadata never leaks into the ambient (first) instance.
     pub(crate) async fn persist_and_set(&self, bm: BucketMetadata) -> Result<()> {
         let mut bm = bm;
+        self.revoke_pending_bucket_creation(&bm.name).await;
         bm.save_with_store(self.object_store()).await?;
 
         self.set(bm.name.clone(), Arc::new(bm)).await;
@@ -2317,6 +2414,7 @@ impl BucketMetadataSys {
     }
 
     async fn persist_new_and_set(&self, mut bm: BucketMetadata) -> Result<()> {
+        self.revoke_pending_bucket_creation(&bm.name).await;
         bm.bucket_creation_committed = true;
         bm.save_with_store_committed(self.object_store()).await?;
         save_bucket_incarnation(self.object_store(), &bm.name, bm.bucket_incarnation_id).await?;
@@ -2326,7 +2424,9 @@ impl BucketMetadataSys {
     }
 
     async fn persist_new_bucket_metadata_intent(&self, mut bm: BucketMetadata) -> Result<()> {
+        self.revoke_pending_bucket_creation(&bm.name).await;
         bm.bucket_creation_committed = false;
+        bm.bucket_creation_commit_record_present = true;
         bm.save_with_store(self.object_store()).await?;
         save_bucket_incarnation(self.object_store(), &bm.name, bm.bucket_incarnation_id).await?;
         bm.bucket_incarnation_sidecar = true;
@@ -2336,6 +2436,7 @@ impl BucketMetadataSys {
     }
 
     async fn commit_bucket_metadata(&self, mut bm: BucketMetadata) -> Result<()> {
+        self.revoke_pending_bucket_creation(&bm.name).await;
         bm.parse_all_configs()?;
         bm.bucket_incarnation_sidecar = true;
         // The caller holds the full bucket creation fence (lifecycle, metadata
@@ -2384,6 +2485,7 @@ impl BucketMetadataSys {
         bucket: &str,
         namespace_guard: &rustfs_lock::NamespaceLockGuard,
     ) -> Result<()> {
+        self.revoke_pending_bucket_creation(bucket).await;
         let expected = self.metadata_map.read().await.get(bucket).cloned();
         if !self
             .bucket_exists(bucket, namespace_guard, "peer bucket metadata existence check")
@@ -2440,6 +2542,7 @@ impl BucketMetadataSys {
             #[cfg(test)]
             self.lazy_disk_loads.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
 
+            self.revoke_pending_bucket_creation(bucket).await;
             let lock = self.object_store().new_ns_lock(bucket, bucket).await?;
             let guard = lock.get_read_lock(crate::set_disk::get_lock_acquire_timeout()).await?;
             #[cfg(test)]
@@ -2765,6 +2868,7 @@ impl BucketMetadataSys {
     /// revalidated at write quorum, so a stale snapshot can neither revert an
     /// acknowledged configuration update nor outlive a delete/recreate.
     async fn migrate_bucket_creation_commit(&self, bucket: &str) -> Result<BucketMetadataAuthority> {
+        self.revoke_pending_bucket_creation(bucket).await;
         let transaction_lock = self
             .object_store()
             .new_ns_lock(RUSTFS_META_BUCKET, &bucket_metadata_transaction_lock_key(bucket))
@@ -2842,6 +2946,7 @@ impl BucketMetadataSys {
         &self,
         bucket: &str,
     ) -> Result<BucketMetadataAuthority> {
+        self.revoke_pending_bucket_creation(bucket).await;
         #[cfg(test)]
         if self.object_lock_disk_read_errors.write().await.remove(bucket) {
             return Err(Error::other(format!("injected Object Lock metadata disk read failure: {bucket}")));
@@ -3017,7 +3122,7 @@ impl BucketMetadataSys {
         }
     }
 
-    pub async fn get_lifecycle_config(&self, bucket: &str) -> Result<(BucketLifecycleConfiguration, OffsetDateTime)> {
+    pub async fn get_lifecycle_config(&self, bucket: &str) -> Result<(PersistedLifecycleConfiguration, OffsetDateTime)> {
         let (bm, _) = self.get_config(bucket).await?;
 
         match ConfigState::of(&bm.lifecycle_config_xml, &bm.lifecycle_config).require(bucket, BUCKET_LIFECYCLE_CONFIG)? {
@@ -3140,7 +3245,7 @@ impl BucketMetadataSys {
         }))
     }
 
-    pub async fn get_replication_config(&self, bucket: &str) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
+    pub async fn get_replication_config(&self, bucket: &str) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
         let (bm, _) = self.get_config(bucket).await?;
 
         if !bm.replication_config_xml.is_empty() && bm.replication_config.is_none() {
@@ -3234,6 +3339,7 @@ mod tests {
     };
     use crate::bucket::target::{BucketTarget, BucketTargetType, Credentials};
     use crate::config::com::read_config;
+    use crate::disk::{DiskAPI as _, RUSTFS_META_TMP_BUCKET};
     use crate::storage_api_contracts::bucket::{BucketOperations as _, DeleteBucketOptions, MakeBucketOptions};
     use byteorder::{ByteOrder as _, LittleEndian};
     use serial_test::serial;
@@ -3288,6 +3394,180 @@ mod tests {
         ),
     ];
 
+    #[tokio::test]
+    #[serial(storage_class_env)]
+    async fn persistence_entry_failures_revoke_pending_creation_before_cache_publish() {
+        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("instance metadata system");
+        let sys = sys.read().await.clone();
+        let options = MakeBucketOptions {
+            lock_enabled: true,
+            ..Default::default()
+        };
+        for entry in ["save", "persist", "persist-new", "commit"] {
+            let bucket = format!("revoke-{entry}-{}", Uuid::new_v4());
+            store.fail_next_bucket_creation_after_intent_for_test(&bucket);
+            store
+                .make_bucket(&bucket, &options)
+                .await
+                .expect_err("stop the real producer before physical publication");
+            sys.pending_bucket_creation(&bucket)
+                .await
+                .expect("the producer must leave a real live creation permission");
+            let (intent, persisted) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
+                .await
+                .expect("read the original producer intent");
+            assert!(persisted, "{entry}: the complete original intent must be persisted");
+            let original_intent = intent.marshal_msg().expect("encode the original intent");
+
+            // Keep the creation fence order used by the real producer.
+            let lifecycle_guard = store
+                .acquire_bucket_lifecycle_write_lock(&bucket)
+                .await
+                .expect("hold the bucket lifecycle fence");
+            let transaction_guard = acquire_bucket_metadata_transaction_lock_in(&store.ctx, &bucket)
+                .await
+                .expect("hold the metadata transaction fence");
+            let namespace_lock = store
+                .new_ns_lock(&bucket, &bucket)
+                .await
+                .expect("create the exact bucket namespace lock");
+            let namespace_guard = namespace_lock
+                .get_write_lock(crate::set_disk::get_lock_acquire_timeout())
+                .await
+                .expect("hold the exact bucket namespace fence");
+
+            // A file at each native staging root causes actual IO to fail before
+            // metadata can reach the cache-publishing set call. Preserve the
+            // original staging directories so no fixture data is discarded.
+            for dir in &dirs {
+                let staging = dir.path().join(RUSTFS_META_TMP_BUCKET);
+                let saved = dir.path().join(".rustfs.sys/revocation-saved-tmp");
+                std::fs::rename(&staging, &saved).expect("preserve the native staging directory");
+                std::fs::write(&staging, b"block metadata staging with a non-directory")
+                    .expect("install an actual non-directory IO obstruction");
+            }
+            let result = match entry {
+                "save" => sys.save(intent).await,
+                "persist" => sys.persist_and_set(intent).await,
+                "persist-new" => sys.persist_new_and_set(intent).await,
+                "commit" => sys.commit_bucket_metadata(intent).await,
+                _ => unreachable!(),
+            };
+            for dir in &dirs {
+                let staging = dir.path().join(RUSTFS_META_TMP_BUCKET);
+                let saved = dir.path().join(".rustfs.sys/revocation-saved-tmp");
+                std::fs::remove_file(&staging).expect("remove only the fixture's staging obstruction");
+                std::fs::rename(&saved, &staging).expect("restore the complete original staging directory");
+            }
+            drop((namespace_guard, transaction_guard, lifecycle_guard));
+            result.expect_err("native persistence must fail before its downstream cache set");
+            assert!(
+                sys.pending_bucket_creation(&bucket).await.is_none(),
+                "{entry}: the entry revoker must retire live permission even when persistence fails"
+            );
+            let (after_failure, present) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
+                .await
+                .expect("read the intent after the failed persistence entry");
+            assert!(present, "{entry}: a failed persistence entry must retain the original metadata");
+            assert_eq!(
+                after_failure
+                    .marshal_msg()
+                    .expect("encode the original intent after persistence failure"),
+                original_intent,
+                "{entry}: native IO failure must not change the original generation or complete intent"
+            );
+            let error = store
+                .make_bucket(&bucket, &options)
+                .await
+                .expect_err("a failed ordinary persistence entry cannot preserve creation permission");
+            assert!(error.to_string().contains("administrator recovery"), "{entry}: {error}");
+            let (after_retry, _) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
+                .await
+                .expect("read the original intent after the rejected retry");
+            assert_eq!(
+                after_retry.marshal_msg().expect("encode the intent after the rejected retry"),
+                original_intent,
+                "{entry}: the rejected retry must preserve the complete original intent"
+            );
+            for disk in store.pools[0].disk_set[0].disks.read().await.iter() {
+                assert!(
+                    matches!(
+                        disk.as_ref().expect("native disk").stat_volume(&bucket).await,
+                        Err(crate::disk::error::DiskError::VolumeNotFound)
+                    ),
+                    "{entry}: failed persistence and the rejected retry must not publish a physical volume"
+                );
+            }
+        }
+    }
+
+    #[tokio::test]
+    #[serial(storage_class_env)]
+    async fn metadata_reset_revokes_real_pending_creation_without_changing_persisted_intent() {
+        let (_dirs, store) = isolated_store_over_temp_disks().await;
+        init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+        let bucket = format!("revoke-reset-{}", Uuid::new_v4());
+        let options = MakeBucketOptions {
+            lock_enabled: true,
+            ..Default::default()
+        };
+        store.fail_next_bucket_creation_after_intent_for_test(&bucket);
+        store
+            .make_bucket(&bucket, &options)
+            .await
+            .expect_err("stop the real producer before physical publication");
+        let sys = bucket_metadata_sys_of(&store.ctx).expect("instance metadata system");
+        let mut sys = sys.read().await.clone();
+        sys.pending_bucket_creation(&bucket)
+            .await
+            .expect("the real producer must leave live creation permission before reset");
+        let (intent, persisted) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
+            .await
+            .expect("read the original intent before cache reset");
+        assert!(persisted, "cache reset starts from a complete real persisted intent");
+        let original_intent = intent.marshal_msg().expect("encode the original reset intent");
+        sys._reset().await;
+        assert!(
+            sys.pending_bucket_creation(&bucket).await.is_none(),
+            "reset must revoke the producer-issued permission shared with the instance metadata system"
+        );
+        let (after_reset, present) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
+            .await
+            .expect("read the original intent after cache reset");
+        assert!(present, "cache reset must retain the persisted intent");
+        assert_eq!(
+            after_reset.marshal_msg().expect("encode the intent after cache reset"),
+            original_intent,
+            "cache reset must leave the original generation and complete persisted intent unchanged"
+        );
+        let error = store
+            .make_bucket(&bucket, &options)
+            .await
+            .expect_err("reset cannot regrant the producer's old creation permission");
+        assert!(error.to_string().contains("administrator recovery"), "{error}");
+        let (after_retry, _) = get_config_from_disk_with_presence_in(&store.ctx, &bucket)
+            .await
+            .expect("read the intent after the reset-induced retry rejection");
+        assert_eq!(
+            after_retry
+                .marshal_msg()
+                .expect("encode the intent after the rejected reset retry"),
+            original_intent,
+            "the reset-induced rejection must preserve the original complete intent"
+        );
+        for disk in store.pools[0].disk_set[0].disks.read().await.iter() {
+            assert!(
+                matches!(
+                    disk.as_ref().expect("native disk").stat_volume(&bucket).await,
+                    Err(crate::disk::error::DiskError::VolumeNotFound)
+                ),
+                "cache reset and the rejected retry must not publish a physical volume"
+            );
+        }
+    }
+
     #[tokio::test]
     async fn g_d3_003_new_writer_replication_loads_without_fail_closed_state() {
         let (dirs, store) = isolated_store_over_temp_disks().await;
@@ -3476,7 +3756,7 @@ mod tests {
         let mut versioned = BucketMetadata::new("authoritative-versioned");
         versioned.versioning_config_xml = b"<VersioningConfiguration><Status>Enabled</Status></VersioningConfiguration>".to_vec();
         versioned.versioning_config = Some(VersioningConfiguration {
-            status: Some(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::ENABLED)),
+            status: Some("Enabled".to_owned()),
             ..Default::default()
         });
         sys.set("authoritative-versioned".to_string(), Arc::new(versioned)).await;
@@ -3599,25 +3879,24 @@ mod tests {
         else {
             panic!("legacy lock-enabled metadata must remain configured");
         };
-        assert_eq!(
-            config.object_lock_enabled.as_ref().map(|value| value.as_str()),
-            Some(ObjectLockEnabled::ENABLED)
-        );
+        assert_eq!(config.object_lock_enabled.as_deref(), Some("Enabled"));
         assert!(config.rule.is_none());
     }
 
     #[test]
     fn authoritative_object_lock_state_rejects_invalid_retention_periods() {
-        use s3s::dto::{DefaultRetention, ObjectLockRule};
+        use rustfs_gateway_types::persistence::{
+            PersistedDefaultRetention as DefaultRetention, PersistedObjectLockRule as ObjectLockRule,
+        };
 
         for (days, years) in [(Some(0), None), (Some(-1), None), (Some(1), Some(1)), (None, None)] {
             let mut metadata = BucketMetadata::new("invalid-object-lock");
             metadata.object_lock_config = Some(ObjectLockConfiguration {
-                object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+                object_lock_enabled: Some("Enabled".to_owned()),
                 rule: Some(ObjectLockRule {
                     default_retention: Some(DefaultRetention {
                         days,
-                        mode: Some(ObjectLockRetentionMode::from_static(ObjectLockRetentionMode::COMPLIANCE)),
+                        mode: Some("COMPLIANCE".to_owned()),
                         years,
                     }),
                 }),
@@ -3632,9 +3911,11 @@ mod tests {
 
     #[test]
     fn authoritative_object_lock_state_rejects_invalid_enabled_mode_and_rule() {
-        use s3s::dto::{DefaultRetention, ObjectLockRule};
+        use rustfs_gateway_types::persistence::{
+            PersistedDefaultRetention as DefaultRetention, PersistedObjectLockRule as ObjectLockRule,
+        };
 
-        for enabled in [None, Some(ObjectLockEnabled::from("Disabled".to_string()))] {
+        for enabled in [None, Some("Disabled".to_string())] {
             let mut metadata = BucketMetadata::new("invalid-object-lock-enabled");
             metadata.lock_enabled = true;
             metadata.object_lock_config = Some(ObjectLockConfiguration {
@@ -3646,11 +3927,11 @@ mod tests {
 
         let mut invalid_mode = BucketMetadata::new("invalid-object-lock-mode");
         invalid_mode.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: Some(ObjectLockRule {
                 default_retention: Some(DefaultRetention {
                     days: Some(1),
-                    mode: Some(ObjectLockRetentionMode::from("invalid".to_string())),
+                    mode: Some("invalid".to_string()),
                     years: None,
                 }),
             }),
@@ -3659,7 +3940,7 @@ mod tests {
 
         let mut missing_retention = BucketMetadata::new("missing-object-lock-retention");
         missing_retention.object_lock_config = Some(ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: Some(ObjectLockRule { default_retention: None }),
         });
         assert!(object_lock_config_state_from_authoritative_metadata(&missing_retention).is_err());
@@ -3669,7 +3950,7 @@ mod tests {
     fn object_lock_ever_enabled_marker_survives_config_removal() {
         let mut metadata = BucketMetadata::new("lock-config-removed");
         let config = ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: None,
         };
         metadata
@@ -3687,10 +3968,7 @@ mod tests {
         else {
             panic!("an ever-enabled bucket must remain Object Lock configured");
         };
-        assert_eq!(
-            config.object_lock_enabled.as_ref().map(ObjectLockEnabled::as_str),
-            Some(ObjectLockEnabled::ENABLED)
-        );
+        assert_eq!(config.object_lock_enabled.as_deref(), Some("Enabled"));
     }
 
     #[tokio::test]
@@ -3706,7 +3984,7 @@ mod tests {
 
         let node = Arc::new(RwLock::new(BucketMetadataSys::new(ecstore.clone())));
         let config = ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: None,
         };
         update_with_sys(
@@ -4687,7 +4965,7 @@ mod tests {
     #[tokio::test]
     async fn update_config_with_persists_tagging_rewrite_across_disk_reload() {
         use crate::bucket::metadata::BUCKET_TAGGING_CONFIG;
-        use s3s::dto::Tag;
+        use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
 
         let (dirs, ecstore) = isolated_store_over_temp_disks().await;
 
@@ -4851,7 +5129,7 @@ mod tests {
     #[tokio::test]
     async fn concurrent_update_config_with_calls_do_not_lose_writes() {
         use crate::bucket::metadata::BUCKET_TAGGING_CONFIG;
-        use s3s::dto::Tag;
+        use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
 
         let (dirs, ecstore) = isolated_store_over_temp_disks().await;
         let sys = Arc::new(RwLock::new(BucketMetadataSys::new(ecstore)));
@@ -4951,7 +5229,7 @@ mod tests {
         let mut persisted = BucketMetadata::new("reload-bucket");
         persisted.policy_config_json = b"persisted-marker".to_vec();
         let lock_config = ObjectLockConfiguration {
-            object_lock_enabled: Some(ObjectLockEnabled::from_static(ObjectLockEnabled::ENABLED)),
+            object_lock_enabled: Some("Enabled".to_owned()),
             rule: None,
         };
         persisted
@@ -5230,7 +5508,7 @@ mod tests {
     #[serial]
     async fn bucket_delete_waits_for_config_mutation_fence() {
         use crate::bucket::metadata::BUCKET_TAGGING_CONFIG;
-        use s3s::dto::Tag;
+        use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
 
         let (_dirs, store) = isolated_store_over_temp_disks().await;
         init_bucket_metadata_sys(store.clone(), Vec::new()).await;

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

@@ -17,15 +17,8 @@ pub mod objectlock_sys;
 pub mod types;
 
 /// Whether a bucket Object Lock configuration has locking enabled. The
-/// serving-layer `ObjectLockConfiguration` DTO implements this in
-/// the bucket-metadata module, which owns the persisted configuration type
-/// during the s3s ratchet migration (rustfs/backlog#1842).
+/// persisted `ObjectLockConfiguration` implements this in the bucket-metadata
+/// module, which owns the persisted configuration type.
 pub trait ObjectLockApi {
     fn enabled(&self) -> bool;
 }
-
-/// Whether a legal-hold status value is one of the two valid wire values.
-/// Implemented for the serving-layer DTO in the bucket-metadata module.
-pub trait ObjectLockStatusExt {
-    fn valid(&self) -> bool;
-}

crates/ecstore/src/bucket/quota/reservation.rs

@@ -707,13 +707,6 @@ struct ShardedReservationData {
 }
 
 impl ReservationLedgerData {
-    fn store(&self) -> &Arc<ECStore> {
-        match self {
-            Self::Single(data) => &data.store,
-            Self::Sharded(data) => &data.store,
-        }
-    }
-
     fn bucket(&self) -> &str {
         match self {
             Self::Single(data) => &data.bucket,
@@ -765,7 +758,7 @@ impl QuotaReservation {
         let Some(ledger) = self.ledger.as_ref() else {
             return;
         };
-        crate::store::list_objects::observe_list_objects_mutation(ledger.store(), ledger.bucket()).await;
+        crate::store::list_objects::observe_scanner_namespace_mutations(ledger.bucket(), 1);
         match settle(ledger, true).await {
             Ok(()) => self.ledger = None,
             Err(err) => log_deferred_settlement(ledger, "commit_deferred", &err),
@@ -1290,7 +1283,7 @@ async fn reconcile_exact(store: &Arc<ECStore>, bucket: &str, ledger: &mut QuotaL
 }
 
 async fn exact_bucket_usage(store: &Arc<ECStore>, bucket: &str) -> Result<u64> {
-    crate::store::list_objects::observe_list_objects_mutation(store, bucket).await;
+    crate::store::list_objects::observe_scanner_namespace_mutations(bucket, 1);
     Ok(compute_bucket_usage(Arc::clone(store), bucket).await?.size)
 }
 

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

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+pub use rustfs_replication::config_cache_serde;
 pub use rustfs_replication::{
     ObjectOpts, OperatorRuleContract, REMOTE_TARGET_CAPABILITY_CONTRACT_VERSION, REMOTE_TARGET_READ_ONLY_HISTORICAL_FIELDS,
     REMOTE_TARGET_UNSUPPORTED_FIELDS, REMOTE_TARGET_WRITABLE_FIELDS, REPLICATION_CAPABILITY_CONTRACT_VERSION,

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

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use s3s::dto::ReplicationConfiguration;
+use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
 
 use super::replication_config_boundary::ReplicationConfigurationExt as _;
 use super::replication_filemeta_boundary::{
@@ -42,7 +42,7 @@ impl ReplicationLifecycleBridge {
         dead_code,
         reason = "declared boundary surface for the ECStore replication split plan; no caller in this port (backlog#1823)"
     )]
-    pub(crate) fn new_config(config: ReplicationConfiguration) -> ReplicationLifecycleConfig {
+    pub(crate) fn new_config(config: PersistedReplicationConfiguration) -> ReplicationLifecycleConfig {
         ReplicationConfig::new(Some(config), None)
     }
 
@@ -118,16 +118,17 @@ impl ReplicationLifecycleBridge {
 
 #[cfg(test)]
 mod tests {
-    use s3s::dto::{Destination, ReplicationRule, ReplicationRuleStatus};
+    use rustfs_gateway_types::dto::Status;
+    use rustfs_gateway_types::persistence::{PersistedReplicationDestination, PersistedReplicationRule};
 
     use super::super::replication_filemeta_boundary::{ReplicateTargetDecision, VersionPurgeStatusType};
     use super::*;
 
-    fn replication_rule() -> ReplicationRule {
-        ReplicationRule {
+    fn replication_rule() -> PersistedReplicationRule {
+        PersistedReplicationRule {
             delete_marker_replication: None,
             delete_replication: None,
-            destination: Destination {
+            destination: PersistedReplicationDestination {
                 bucket: "arn:aws:s3:::target-bucket".to_string(),
                 ..Default::default()
             },
@@ -137,13 +138,13 @@ mod tests {
             prefix: Some(String::new()),
             priority: Some(1),
             source_selection_criteria: None,
-            status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+            status: Status::ENABLED.to_string(),
         }
     }
 
     #[test]
     fn has_pending_version_purge_preserves_replication_active_rule_behavior() {
-        let config = ReplicationLifecycleBridge::new_config(ReplicationConfiguration {
+        let config = ReplicationLifecycleBridge::new_config(PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![replication_rule()],
         });

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

@@ -17,8 +17,8 @@ use std::sync::Arc;
 use crate::bucket::{metadata::BucketMetadata, metadata_sys};
 use crate::disk::{BUCKET_META_PREFIX, RUSTFS_META_BUCKET};
 use crate::runtime::instance::InstanceContext;
+use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
 use rustfs_utils::path::path_join_buf;
-use s3s::dto::ReplicationConfiguration;
 use time::OffsetDateTime;
 
 use super::replication_error_boundary::{Error, Result};
@@ -36,11 +36,11 @@ impl ReplicationMetadataStore {
     pub(crate) const FORCE_DELETE_REPLICATION_FILE: &'static str = "config/replication/force-delete.bin";
     pub(crate) const FORCE_DELETE_REPLICATION_TRANSACTION_LOCK: &'static str = "config/replication/force-delete.bin.transaction";
 
-    pub(crate) async fn replication_config(bucket: &str) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
+    pub(crate) async fn replication_config(bucket: &str) -> Result<(PersistedReplicationConfiguration, OffsetDateTime)> {
         metadata_sys::get_replication_config(bucket).await
     }
 
-    pub(crate) async fn optional_replication_config(bucket: &str) -> Result<Option<ReplicationConfiguration>> {
+    pub(crate) async fn optional_replication_config(bucket: &str) -> Result<Option<PersistedReplicationConfiguration>> {
         let config = match Self::replication_config(bucket).await {
             Ok((config, _)) => Some(config),
             Err(err) => {

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

@@ -16,7 +16,9 @@ use std::{collections::HashMap, fmt, sync::Arc};
 
 use super::replication_filemeta_boundary::metadata_keys;
 use crate::bucket::metadata::BucketMetadata;
-use s3s::dto::{BucketVersioningStatus, ReplicationConfiguration, ReplicationRuleStatus, VersioningConfiguration};
+use rustfs_gateway_types::dto::Status;
+use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
+use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
 use serde::{Deserialize, Serialize};
 use time::OffsetDateTime;
 use tracing::error;
@@ -39,7 +41,7 @@ use super::replication_target_boundary::{BucketTargets, ReplicationTargetStore};
 use super::replication_versioning_boundary::ReplicationVersioningStore;
 use super::runtime_boundary as runtime_sources;
 
-pub(crate) async fn get_replication_config(bucket: &str) -> Result<Option<ReplicationConfiguration>> {
+pub(crate) async fn get_replication_config(bucket: &str) -> Result<Option<PersistedReplicationConfiguration>> {
     let config = ReplicationMetadataStore::optional_replication_config(bucket).await?;
     validate_delete_replication_config(&VersioningConfiguration::default(), config.as_ref())?;
     Ok(config)
@@ -64,7 +66,7 @@ impl DeleteReplicationConfigSnapshot {
     #[cfg(test)]
     pub(crate) fn from_configs_for_test(
         versioning: VersioningConfiguration,
-        replication: Option<ReplicationConfiguration>,
+        replication: Option<PersistedReplicationConfiguration>,
     ) -> Self {
         let metadata = replication.map(|config| {
             let mut metadata = BucketMetadata::new("test-bucket");
@@ -78,7 +80,7 @@ impl DeleteReplicationConfigSnapshot {
         &self.versioning
     }
 
-    pub fn replication_config(&self) -> Option<&ReplicationConfiguration> {
+    pub fn replication_config(&self) -> Option<&PersistedReplicationConfiguration> {
         self.metadata
             .as_ref()
             .and_then(|metadata| metadata.replication_config.as_ref())
@@ -101,7 +103,7 @@ impl DeleteReplicationConfigSnapshot {
     pub(crate) fn active_delete_marker_rules_require_tags(&self, object: &str) -> bool {
         self.replication_config().is_some_and(|config| {
             config.rules.iter().any(|rule| {
-                if rule.status == ReplicationRuleStatus::from_static(ReplicationRuleStatus::DISABLED) {
+                if rule.status == Status::DISABLED.as_str() {
                     return false;
                 }
                 if !object.starts_with(rule.prefix()) {
@@ -122,12 +124,12 @@ impl DeleteReplicationConfigSnapshot {
 
 fn validate_delete_replication_config(
     versioning: &VersioningConfiguration,
-    config: Option<&ReplicationConfiguration>,
+    config: Option<&PersistedReplicationConfiguration>,
 ) -> Result<()> {
     if versioning
         .status
         .as_ref()
-        .is_some_and(|status| !matches!(status.as_str(), BucketVersioningStatus::ENABLED | BucketVersioningStatus::SUSPENDED))
+        .is_some_and(|status| !matches!(status.as_str(), "Enabled" | "Suspended"))
     {
         return Err(super::replication_error_boundary::Error::other(
             "bucket versioning configuration has an invalid status",
@@ -144,7 +146,7 @@ fn validate_delete_replication_config(
         let role = config.role.trim();
         let mut role_destination = None;
         for rule in &config.rules {
-            if rule.status.as_str() == ReplicationRuleStatus::ENABLED {
+            if rule.status.as_str() == Status::ENABLED.as_str() {
                 let destination = rule.destination.bucket.trim();
                 if role.is_empty() && destination.is_empty() {
                     return Err(super::replication_error_boundary::Error::other(
@@ -169,7 +171,7 @@ fn validate_delete_replication_config(
     Ok(())
 }
 
-fn replication_config_from_metadata(metadata: &BucketMetadata) -> Result<Option<&ReplicationConfiguration>> {
+fn replication_config_from_metadata(metadata: &BucketMetadata) -> Result<Option<&PersistedReplicationConfiguration>> {
     if !metadata.replication_config_xml.is_empty() && metadata.replication_config.is_none() {
         return Err(super::replication_error_boundary::Error::other(
             "persisted bucket replication configuration is invalid",
@@ -250,12 +252,15 @@ pub(crate) async fn load_delete_replication_config_in(
 
 #[derive(Debug, Clone, Default, Serialize, Deserialize)]
 pub struct ReplicationConfig {
-    pub config: Option<ReplicationConfiguration>,
+    /// Carried in the scanner's persisted data usage cache, so its serde form
+    /// is the one the s3s DTO wrote there (see `config_cache_serde` in the replication crate).
+    #[serde(with = "super::replication_config_boundary::config_cache_serde")]
+    pub config: Option<PersistedReplicationConfiguration>,
     pub remotes: Option<BucketTargets>,
 }
 
 impl ReplicationConfig {
-    pub fn new(config: Option<ReplicationConfiguration>, remotes: Option<BucketTargets>) -> Self {
+    pub fn new(config: Option<PersistedReplicationConfiguration>, remotes: Option<BucketTargets>) -> Self {
         Self { config, remotes }
     }
 
@@ -468,7 +473,7 @@ fn check_replicate_delete_with_config(
     oi: &ObjectInfo,
     del_opts: &ObjectOptions,
     source_error: bool,
-    config: Option<&ReplicationConfiguration>,
+    config: Option<&PersistedReplicationConfiguration>,
     existing_delete_marker: bool,
     trust_persisted_replica_status: bool,
 ) -> ReplicateDecision {
@@ -587,7 +592,7 @@ pub(crate) async fn must_replicate(bucket: &str, object: &str, mopts: MustReplic
 }
 
 fn metadata_target_should_replicate(
-    cfg: &ReplicationConfiguration,
+    cfg: &PersistedReplicationConfiguration,
     opts: &ObjectOpts,
     mopts: &MustReplicateOptions,
     arn: &str,
@@ -597,20 +602,21 @@ fn metadata_target_should_replicate(
 
 #[cfg(test)]
 mod tests {
-    use s3s::dto::{
-        DeleteMarkerReplication, DeleteMarkerReplicationStatus, DeleteReplication, DeleteReplicationStatus, Destination,
-        ReplicaModifications, ReplicationRule, ReplicationRuleFilter, ReplicationRuleStatus, SourceSelectionCriteria, Tag,
+    use rustfs_gateway_types::dto::Status;
+    use rustfs_gateway_types::persistence::{
+        PersistedOptionalReplicationStatus, PersistedReplicationDestination, PersistedReplicationFilter,
+        PersistedReplicationRule, PersistedReplicationStatus, PersistedReplicationTag, PersistedSourceSelectionCriteria,
     };
 
     use super::super::replication_filemeta_boundary::VersionPurgeStatusType;
     use super::super::replication_target_boundary::BucketTarget;
     use super::*;
 
-    fn replication_rule() -> ReplicationRule {
-        ReplicationRule {
+    fn replication_rule() -> PersistedReplicationRule {
+        PersistedReplicationRule {
             delete_marker_replication: None,
             delete_replication: None,
-            destination: Destination {
+            destination: PersistedReplicationDestination {
                 bucket: "arn:aws:s3:::target-bucket".to_string(),
                 ..Default::default()
             },
@@ -620,7 +626,7 @@ mod tests {
             prefix: Some(String::new()),
             priority: Some(1),
             source_selection_criteria: None,
-            status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+            status: Status::ENABLED.to_string(),
         }
     }
 
@@ -629,11 +635,11 @@ mod tests {
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
         let mut rule = replication_rule();
         rule.destination.bucket = arn.to_string();
-        rule.filter = Some(ReplicationRuleFilter {
+        rule.filter = Some(PersistedReplicationFilter {
             prefix: Some("admitted/".to_string()),
             ..Default::default()
         });
-        let cfg = ReplicationConfiguration {
+        let cfg = PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![rule],
         };
@@ -671,7 +677,7 @@ mod tests {
         assert!(!empty.replicate(&ObjectOpts::default()));
 
         let config = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
                 rules: vec![replication_rule()],
             }),
@@ -725,7 +731,7 @@ mod tests {
     fn delete_snapshot_rejects_enabled_rules_without_a_destination() {
         let mut rule = replication_rule();
         rule.destination.bucket.clear();
-        let config = ReplicationConfiguration {
+        let config = PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![rule],
         };
@@ -741,7 +747,7 @@ mod tests {
         let first = replication_rule();
         let mut second = replication_rule();
         second.destination.bucket = "arn:aws:s3:::other-target".to_string();
-        let config = ReplicationConfiguration {
+        let config = PersistedReplicationConfiguration {
             role: "arn:aws:s3:::role-target".to_string(),
             rules: vec![first, second],
         };
@@ -752,47 +758,47 @@ mod tests {
     #[test]
     fn delete_snapshot_rejects_unknown_status_values() {
         let invalid_versioning = VersioningConfiguration {
-            status: Some("Enabld".to_string().into()),
+            status: Some("Enabld".to_string()),
             ..Default::default()
         };
         assert!(validate_delete_replication_config(&invalid_versioning, None).is_err());
 
         let mut invalid_rule = replication_rule();
-        invalid_rule.status = "Enabld".to_string().into();
-        let config = ReplicationConfiguration {
+        invalid_rule.status = "Enabld".to_string();
+        let config = PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![invalid_rule],
         };
         assert!(validate_delete_replication_config(&VersioningConfiguration::default(), Some(&config)).is_err());
 
         let mut invalid_delete = replication_rule();
-        invalid_delete.delete_replication = Some(DeleteReplication {
-            status: "Enabld".to_string().into(),
+        invalid_delete.delete_replication = Some(PersistedReplicationStatus {
+            status: "Enabld".to_string(),
         });
-        let config = ReplicationConfiguration {
+        let config = PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![invalid_delete],
         };
         assert!(validate_delete_replication_config(&VersioningConfiguration::default(), Some(&config)).is_err());
 
         let mut invalid_delete_marker = replication_rule();
-        invalid_delete_marker.delete_marker_replication = Some(DeleteMarkerReplication {
-            status: Some("Enabld".to_string().into()),
+        invalid_delete_marker.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
+            status: Some("Enabld".to_string()),
         });
-        let config = ReplicationConfiguration {
+        let config = PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![invalid_delete_marker],
         };
         assert!(validate_delete_replication_config(&VersioningConfiguration::default(), Some(&config)).is_err());
 
         let mut invalid_replica_modifications = replication_rule();
-        invalid_replica_modifications.source_selection_criteria = Some(SourceSelectionCriteria {
-            replica_modifications: Some(ReplicaModifications {
-                status: "Enabld".to_string().into(),
+        invalid_replica_modifications.source_selection_criteria = Some(PersistedSourceSelectionCriteria {
+            replica_modifications: Some(PersistedReplicationStatus {
+                status: "Enabld".to_string(),
             }),
             sse_kms_encrypted_objects: None,
         });
-        let config = ReplicationConfiguration {
+        let config = PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![invalid_replica_modifications],
         };
@@ -804,11 +810,11 @@ mod tests {
         let mut metadata = BucketMetadata::new("bucket");
         metadata.versioning_config_xml = b"configured".to_vec();
         metadata.versioning_config = Some(VersioningConfiguration {
-            status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+            status: Some("Enabled".to_owned()),
             ..Default::default()
         });
         metadata.replication_config_xml = b"configured".to_vec();
-        metadata.replication_config = Some(ReplicationConfiguration {
+        metadata.replication_config = Some(PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![replication_rule()],
         });
@@ -848,7 +854,7 @@ mod tests {
         );
 
         let mut inconsistent = BucketMetadata::new("bucket");
-        inconsistent.replication_config = Some(ReplicationConfiguration {
+        inconsistent.replication_config = Some(PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![replication_rule()],
         });
@@ -863,11 +869,11 @@ mod tests {
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
         let mut rule = replication_rule();
         rule.destination.bucket = arn.to_string();
-        rule.delete_marker_replication = Some(DeleteMarkerReplication {
-            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::ENABLED)),
+        rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
+            status: Some(Status::ENABLED.to_string()),
         });
         let mut metadata = BucketMetadata::new("bucket");
-        metadata.replication_config = Some(ReplicationConfiguration {
+        metadata.replication_config = Some(PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![rule],
         });
@@ -898,27 +904,27 @@ mod tests {
     fn delete_marker_source_read_is_required_only_for_tag_filtered_rules() {
         let mut prefix_rule = replication_rule();
         prefix_rule.prefix = Some("logs/".to_string());
-        prefix_rule.delete_marker_replication = Some(DeleteMarkerReplication {
-            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::ENABLED)),
+        prefix_rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
+            status: Some(Status::ENABLED.to_string()),
         });
 
         let prefix_snapshot = DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                status: Some("Enabled".to_owned()),
                 ..Default::default()
             },
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
                 rules: vec![prefix_rule],
             }),
         );
 
         let mut tag_rule = replication_rule();
-        tag_rule.delete_marker_replication = Some(DeleteMarkerReplication {
-            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::ENABLED)),
+        tag_rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
+            status: Some(Status::ENABLED.to_string()),
         });
-        tag_rule.filter = Some(ReplicationRuleFilter {
-            tag: Some(Tag {
+        tag_rule.filter = Some(PersistedReplicationFilter {
+            tag: Some(PersistedReplicationTag {
                 key: Some("class".to_string()),
                 value: Some("audit".to_string()),
             }),
@@ -926,10 +932,10 @@ mod tests {
         });
         let tag_snapshot = DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                status: Some("Enabled".to_owned()),
                 ..Default::default()
             },
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
                 rules: vec![tag_rule],
             }),
@@ -944,14 +950,14 @@ mod tests {
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
         let mut rule = replication_rule();
         rule.destination.bucket = arn.to_string();
-        rule.delete_replication = Some(DeleteReplication {
-            status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+        rule.delete_replication = Some(PersistedReplicationStatus {
+            status: Status::ENABLED.to_string(),
         });
-        rule.delete_marker_replication = Some(DeleteMarkerReplication {
-            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::DISABLED)),
+        rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
+            status: Some(Status::DISABLED.to_string()),
         });
         let config = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
                 rules: vec![rule],
             }),
@@ -985,22 +991,22 @@ mod tests {
     #[test]
     fn live_delete_does_not_trust_persisted_replica_status() {
         let mut rule = replication_rule();
-        rule.delete_replication = Some(DeleteReplication {
-            status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+        rule.delete_replication = Some(PersistedReplicationStatus {
+            status: Status::ENABLED.to_string(),
         });
-        rule.source_selection_criteria = Some(SourceSelectionCriteria {
-            replica_modifications: Some(ReplicaModifications {
-                status: s3s::dto::ReplicaModificationsStatus::from_static(s3s::dto::ReplicaModificationsStatus::DISABLED),
+        rule.source_selection_criteria = Some(PersistedSourceSelectionCriteria {
+            replica_modifications: Some(PersistedReplicationStatus {
+                status: Status::DISABLED.to_string(),
             }),
             sse_kms_encrypted_objects: None,
         });
-        let replication = ReplicationConfiguration {
+        let replication = PersistedReplicationConfiguration {
             role: String::new(),
             rules: vec![rule],
         };
         let snapshot = DeleteReplicationConfigSnapshot::from_configs_for_test(
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                status: Some("Enabled".to_owned()),
                 ..Default::default()
             },
             Some(replication.clone()),
@@ -1036,14 +1042,14 @@ mod tests {
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
         let mut rule = replication_rule();
         rule.destination.bucket = arn.to_string();
-        rule.delete_replication = Some(DeleteReplication {
-            status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+        rule.delete_replication = Some(PersistedReplicationStatus {
+            status: Status::ENABLED.to_string(),
         });
-        rule.delete_marker_replication = Some(DeleteMarkerReplication {
-            status: Some(DeleteMarkerReplicationStatus::from_static(DeleteMarkerReplicationStatus::DISABLED)),
+        rule.delete_marker_replication = Some(PersistedOptionalReplicationStatus {
+            status: Some(Status::DISABLED.to_string()),
         });
         let config = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
                 rules: vec![rule],
             }),

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

@@ -4794,9 +4794,10 @@ mod tests {
     #[tokio::test]
     async fn heal_queue_marks_missing_versioning_state_as_missed() {
         use super::super::replication_target_boundary::BucketTargets;
-        use s3s::dto::{
-            DeleteReplication, DeleteReplicationStatus, Destination, ReplicationConfiguration, ReplicationRule,
-            ReplicationRuleStatus,
+        use rustfs_gateway_types::dto::Status;
+        use rustfs_gateway_types::persistence::{
+            PersistedReplicationConfiguration, PersistedReplicationDestination, PersistedReplicationRule,
+            PersistedReplicationStatus,
         };
 
         let arn = "arn:rustfs:replication:us-east-1:target:bucket";
@@ -4811,14 +4812,14 @@ mod tests {
                 ..Default::default()
             },
             ReplicationConfig::new(
-                Some(ReplicationConfiguration {
+                Some(PersistedReplicationConfiguration {
                     role: String::new(),
-                    rules: vec![ReplicationRule {
+                    rules: vec![PersistedReplicationRule {
                         delete_marker_replication: None,
-                        delete_replication: Some(DeleteReplication {
-                            status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+                        delete_replication: Some(PersistedReplicationStatus {
+                            status: Status::ENABLED.to_string(),
                         }),
-                        destination: Destination {
+                        destination: PersistedReplicationDestination {
                             bucket: arn.to_string(),
                             ..Default::default()
                         },
@@ -4828,7 +4829,7 @@ mod tests {
                         prefix: Some(String::new()),
                         priority: Some(1),
                         source_selection_criteria: None,
-                        status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                        status: Status::ENABLED.to_string(),
                     }],
                 }),
                 Some(BucketTargets::default()),

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

@@ -85,6 +85,8 @@ use http_body_util::StreamBody;
 use metrics::counter;
 #[cfg(test)]
 use rmp_serde;
+#[cfg(test)]
+use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
 use rustfs_s3_types::EventName;
 use rustfs_utils::http::{
     AMZ_OBJECT_LOCK_LEGAL_HOLD, AMZ_OBJECT_LOCK_MODE, AMZ_OBJECT_LOCK_RETAIN_UNTIL_DATE, AMZ_TAGGING_DIRECTIVE,
@@ -92,8 +94,6 @@ use rustfs_utils::http::{
     insert_str, replication_target_versions,
 };
 use rustfs_utils::{DEFAULT_SIP_HASH_KEY, get_env_usize, sip_hash};
-#[cfg(test)]
-use s3s::dto::ReplicationConfiguration;
 use std::collections::{HashMap, HashSet};
 use std::fmt::Display;
 use std::pin::Pin;
@@ -6500,9 +6500,10 @@ mod tests {
 
     use super::super::replication_target_boundary::{BucketTarget, BucketTargets};
     use super::*;
-    use s3s::dto::{
-        BucketVersioningStatus, DeleteReplication, DeleteReplicationStatus, Destination, ExcludedPrefix, ReplicationRule,
-        ReplicationRuleStatus, VersioningConfiguration,
+    use rustfs_gateway_types::dto::Status;
+    use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
+    use rustfs_gateway_types::persistence::{
+        PersistedReplicationDestination, PersistedReplicationRule, PersistedReplicationStatus,
     };
     use std::collections::HashMap;
     use time::OffsetDateTime;
@@ -8568,7 +8569,7 @@ mod tests {
             ..Default::default()
         };
         let rcfg = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: role.to_string(),
                 rules: vec![],
             }),
@@ -8594,7 +8595,7 @@ mod tests {
         ReplicationVersioningStore::install_prefix_state_test_config(
             &bucket,
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
+                status: Some("Enabled".to_owned()),
                 ..Default::default()
             },
         );
@@ -8627,22 +8628,20 @@ mod tests {
         ReplicationVersioningStore::install_prefix_state_test_config(
             &bucket,
             VersioningConfiguration {
-                status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
-                excluded_prefixes: Some(vec![ExcludedPrefix {
-                    prefix: Some("archive/".to_string()),
-                }]),
+                status: Some("Enabled".to_owned()),
+                excluded_prefixes: Some(vec![Some("archive/".to_string())]),
                 ..Default::default()
             },
         );
         let rcfg = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
-                rules: vec![ReplicationRule {
+                rules: vec![PersistedReplicationRule {
                     delete_marker_replication: None,
-                    delete_replication: Some(DeleteReplication {
-                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::ENABLED),
+                    delete_replication: Some(PersistedReplicationStatus {
+                        status: Status::ENABLED.to_string(),
                     }),
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: arn.to_string(),
                         ..Default::default()
                     },
@@ -8652,7 +8651,7 @@ mod tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                    status: Status::ENABLED.to_string(),
                 }],
             }),
             Some(BucketTargets {
@@ -8688,14 +8687,14 @@ mod tests {
         let admitted_arn = "arn:rustfs:replication:us-east-1:target:admitted";
         let current_role = "arn:rustfs:replication:us-east-1:target:current";
         let rcfg = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: current_role.to_string(),
-                rules: vec![ReplicationRule {
+                rules: vec![PersistedReplicationRule {
                     delete_marker_replication: None,
-                    delete_replication: Some(DeleteReplication {
-                        status: DeleteReplicationStatus::from_static(DeleteReplicationStatus::DISABLED),
+                    delete_replication: Some(PersistedReplicationStatus {
+                        status: Status::DISABLED.to_string(),
                     }),
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: current_role.to_string(),
                         ..Default::default()
                     },
@@ -8705,7 +8704,7 @@ mod tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static(ReplicationRuleStatus::ENABLED),
+                    status: Status::ENABLED.to_string(),
                 }],
             }),
             Some(BucketTargets {
@@ -8753,12 +8752,12 @@ mod tests {
     #[tokio::test]
     async fn heal_rejects_semantically_invalid_replication_config() {
         let rcfg = ReplicationConfig::new(
-            Some(ReplicationConfiguration {
+            Some(PersistedReplicationConfiguration {
                 role: String::new(),
-                rules: vec![ReplicationRule {
+                rules: vec![PersistedReplicationRule {
                     delete_marker_replication: None,
                     delete_replication: None,
-                    destination: Destination {
+                    destination: PersistedReplicationDestination {
                         bucket: "arn:rustfs:replication:us-east-1:target:bucket".to_string(),
                         ..Default::default()
                     },
@@ -8768,7 +8767,7 @@ mod tests {
                     prefix: Some(String::new()),
                     priority: Some(1),
                     source_selection_criteria: None,
-                    status: ReplicationRuleStatus::from_static("Enabld"),
+                    status: "Enabld".to_owned(),
                 }],
             }),
             Some(BucketTargets::default()),

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

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use s3s::dto::ReplicationConfiguration;
+use rustfs_gateway_types::persistence::PersistedReplicationConfiguration;
 
 use super::replication_config_boundary::{ObjectOpts, ReplicationConfigurationExt};
 use super::replication_filemeta_boundary::ReplicationType;
@@ -20,7 +20,7 @@ use super::replication_filemeta_boundary::ReplicationType;
 pub(crate) struct ReplicationTargetConfigBridge;
 
 impl ReplicationTargetConfigBridge {
-    pub(crate) fn target_is_used_by_rules(config: &ReplicationConfiguration, arn: &str) -> bool {
+    pub(crate) fn target_is_used_by_rules(config: &PersistedReplicationConfiguration, arn: &str) -> bool {
         config
             .filter_target_arns(&ObjectOpts {
                 op_type: ReplicationType::All,
@@ -37,7 +37,7 @@ mod tests {
 
     #[test]
     fn target_config_bridge_matches_role_target() {
-        let config = ReplicationConfiguration {
+        let config = PersistedReplicationConfiguration {
             role: " arn:target ".to_string(),
             rules: Vec::new(),
         };

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

@@ -15,7 +15,7 @@
 use super::replication_error_boundary::Result;
 use crate::bucket::{versioning::VersioningApi as _, versioning_sys::BucketVersioningSys};
 #[cfg(test)]
-use s3s::dto::VersioningConfiguration;
+use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
 #[cfg(test)]
 use std::{collections::HashMap, sync::LazyLock, sync::Mutex};
 

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

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use s3s::dto::Tag;
+use rustfs_gateway_types::cors_tagging::PersistedTag as Tag;
 use std::cmp::Ordering;
 use std::collections::HashMap;
 use url::form_urlencoded;

crates/ecstore/src/bucket/utils.rs

@@ -15,9 +15,23 @@
 use crate::disk::{MIGRATING_META_BUCKET, RUSTFS_META_BUCKET};
 use crate::error::{Error, Result, StorageError};
 use regex::Regex;
+use rustfs_gateway_types::cors_tagging::{
+    CorsTaggingCodecError, PersistedCorsConfiguration, PersistedTagging, parse_cors, parse_tagging, serialize_cors,
+    serialize_tagging,
+};
+use rustfs_gateway_types::persistence::{
+    PersistedAccelerateConfiguration, PersistedBucketEncryptionConfiguration, PersistedBucketLoggingStatus,
+    PersistedLifecycleConfiguration, PersistedNotificationConfiguration, PersistedObjectLockConfiguration,
+    PersistedPublicAccessBlockConfiguration, PersistedReplicationConfiguration, PersistedRequestPaymentConfiguration,
+    PersistedVersioningConfiguration, PersistedWebsiteConfiguration, PersistenceCodecError, parse_accelerate,
+    parse_bucket_encryption, parse_bucket_logging, parse_lifecycle_legacy_dates, parse_notification, parse_object_lock,
+    parse_public_access_block, parse_replication, parse_request_payment, parse_versioning, parse_website, serialize_accelerate,
+    serialize_bucket_encryption, serialize_bucket_logging, serialize_lifecycle_legacy_dates, serialize_notification,
+    serialize_object_lock, serialize_public_access_block, serialize_replication, serialize_request_payment, serialize_versioning,
+    serialize_website,
+};
 use rustfs_utils::WindowsFileNameError;
 use rustfs_utils::path::SLASH_SEPARATOR;
-use s3s::xml;
 use tracing::instrument;
 
 pub fn is_meta_bucketname(name: &str) -> bool {
@@ -74,37 +88,173 @@ pub fn check_valid_bucket_name_strict(bucket_name: &str) -> Result<()> {
     check_bucket_name_common(bucket_name, true)
 }
 
-// RUSTFS_COMPAT_TODO(s3gate-metadata-xml): the s3s codec reads persisted XML during migration. Remove after every supported writer uses the gateway codec and every retained metadata object and backup archive is verified or rewritten.
-pub fn deserialize<T>(input: &[u8]) -> xml::DeResult<T>
-where
-    T: for<'xml> xml::Deserialize<'xml>,
-{
-    let mut d = xml::Deserializer::new(input);
-    let ans = T::deserialize(&mut d)?;
-    d.expect_eof()?;
-    Ok(ans)
+/// A bucket configuration persisted as XML in the bucket metadata, together
+/// with the codec that reads and writes its stored bytes.
+///
+/// Every family reads and writes its stored bytes through the gateway
+/// persistence codecs (`rustfs_gateway_types::persistence` and
+/// `rustfs_gateway_types::cors_tagging`, rustfs/backlog#2745). The codecs keep
+/// the persisted bytes the `bucket_config_goldens` pin, so a document the old
+/// s3s codec wrote reads back and re-encodes unchanged.
+pub trait BucketConfigXml: Sized {
+    /// Decodes the stored bytes of one configuration document.
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError>;
+
+    /// Encodes the value as RustFS stores it.
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError>;
 }
 
-#[allow(
-    dead_code,
-    reason = "xml serialize helper with no caller in this port; the live sibling is deserialize (backlog#1823)"
-)]
-pub fn serialize_content<T: xml::SerializeContent>(val: &T) -> xml::SerResult<String> {
-    let mut buf = Vec::with_capacity(256);
-    {
-        let mut ser = xml::Serializer::new(&mut buf);
-        val.serialize_content(&mut ser)?;
-    }
-    Ok(String::from_utf8(buf).unwrap())
+/// Why a persisted bucket configuration could not be decoded or encoded.
+#[derive(Debug, thiserror::Error)]
+pub enum BucketConfigXmlError {
+    /// A gateway persistence codec refused the bytes or the value.
+    #[error(transparent)]
+    Persisted(#[from] PersistenceCodecError),
+    /// The gateway CORS or Tagging codec refused the bytes or the value.
+    #[error(transparent)]
+    CorsTagging(#[from] CorsTaggingCodecError),
 }
 
-pub fn serialize<T: xml::Serialize>(val: &T) -> xml::SerResult<Vec<u8>> {
-    let mut buf = Vec::with_capacity(256);
-    {
-        let mut ser = xml::Serializer::new(&mut buf);
-        val.serialize(&mut ser)?;
+/// Decodes one persisted bucket configuration document.
+pub fn deserialize<T: BucketConfigXml>(input: &[u8]) -> std::result::Result<T, BucketConfigXmlError> {
+    T::decode_config_xml(input)
+}
+
+/// Encodes one bucket configuration as RustFS persists it.
+pub fn serialize<T: BucketConfigXml>(val: &T) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+    val.encode_config_xml()
+}
+
+impl BucketConfigXml for PersistedReplicationConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_replication(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_replication(self)?)
+    }
+}
+
+impl BucketConfigXml for PersistedLifecycleConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_lifecycle_legacy_dates(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_lifecycle_legacy_dates(self)?)
+    }
+}
+
+// The gateway writers of these families are infallible: a value that exists is
+// always encodable, so only the readers can refuse bytes.
+impl BucketConfigXml for PersistedAccelerateConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_accelerate(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_accelerate(self))
+    }
+}
+
+impl BucketConfigXml for PersistedBucketLoggingStatus {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_bucket_logging(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_bucket_logging(self))
+    }
+}
+
+impl BucketConfigXml for PersistedCorsConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_cors(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_cors(self))
+    }
+}
+
+impl BucketConfigXml for PersistedNotificationConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_notification(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_notification(self))
+    }
+}
+
+impl BucketConfigXml for PersistedObjectLockConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_object_lock(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_object_lock(self))
+    }
+}
+
+impl BucketConfigXml for PersistedPublicAccessBlockConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_public_access_block(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_public_access_block(self))
+    }
+}
+
+impl BucketConfigXml for PersistedRequestPaymentConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_request_payment(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_request_payment(self))
+    }
+}
+
+impl BucketConfigXml for PersistedBucketEncryptionConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_bucket_encryption(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_bucket_encryption(self))
+    }
+}
+
+impl BucketConfigXml for PersistedTagging {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_tagging(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_tagging(self))
+    }
+}
+
+impl BucketConfigXml for PersistedVersioningConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_versioning(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_versioning(self))
+    }
+}
+
+impl BucketConfigXml for PersistedWebsiteConfiguration {
+    fn decode_config_xml(input: &[u8]) -> std::result::Result<Self, BucketConfigXmlError> {
+        Ok(parse_website(input)?)
+    }
+
+    fn encode_config_xml(&self) -> std::result::Result<Vec<u8>, BucketConfigXmlError> {
+        Ok(serialize_website(self))
     }
-    Ok(buf)
 }
 
 pub fn has_bad_path_component(path: &str) -> bool {
@@ -434,6 +584,40 @@ mod tests {
     use super::*;
     use proptest::prelude::*;
 
+    /// Every bucket-configuration family reads and writes through its own
+    /// gateway codec (rustfs/backlog#2745): the family's golden document
+    /// decodes and re-encodes to the same bytes, and a document of another
+    /// family is refused instead of decoding into an empty configuration.
+    #[test]
+    fn every_bucket_config_family_uses_its_own_gateway_codec() {
+        fn assert_family<T: BucketConfigXml>(family: &str, own: &[u8], foreign: &[u8]) {
+            let value = deserialize::<T>(own).unwrap_or_else(|err| panic!("{family}: own document must decode: {err}"));
+            let encoded = serialize(&value).unwrap_or_else(|err| panic!("{family}: value must encode: {err}"));
+            assert_eq!(encoded, own, "{family}: re-encoding must keep the persisted bytes");
+            assert!(deserialize::<T>(foreign).is_err(), "{family}: a foreign root element must be refused");
+        }
+
+        macro_rules! golden {
+            ($family:literal) => {
+                include_bytes!(concat!("../../tests/fixtures/bucket-config-goldens/", $family, "/minio.out.xml"))
+            };
+        }
+        let tagging = golden!("tagging");
+        let versioning = golden!("versioning");
+
+        assert_family::<PersistedAccelerateConfiguration>("accelerate", golden!("accelerate"), tagging);
+        assert_family::<PersistedBucketLoggingStatus>("logging", golden!("logging"), tagging);
+        assert_family::<PersistedCorsConfiguration>("cors", golden!("cors"), tagging);
+        assert_family::<PersistedNotificationConfiguration>("notification", golden!("notification"), tagging);
+        assert_family::<PersistedObjectLockConfiguration>("object_lock", golden!("object_lock"), tagging);
+        assert_family::<PersistedPublicAccessBlockConfiguration>("public_access_block", golden!("public_access_block"), tagging);
+        assert_family::<PersistedRequestPaymentConfiguration>("request_payment", golden!("request_payment"), tagging);
+        assert_family::<PersistedBucketEncryptionConfiguration>("encryption", golden!("encryption"), tagging);
+        assert_family::<PersistedTagging>("tagging", tagging, versioning);
+        assert_family::<PersistedVersioningConfiguration>("versioning", versioning, tagging);
+        assert_family::<PersistedWebsiteConfiguration>("website", golden!("website"), tagging);
+    }
+
     /// rustfs#5785: keys whose path segments exceed the on-disk NAME_MAX
     /// budget must be rejected up front as ObjectNameInvalid (4xx), not leak
     /// ENAMETOOLONG as InternalError 500 from the disk layer.

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

@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
 use rustfs_utils::string::match_simple;
-use s3s::dto::{BucketVersioningStatus, VersioningConfiguration};
 
 pub trait VersioningApi {
     fn enabled(&self) -> bool;
@@ -28,10 +28,10 @@ pub trait VersioningApi {
 
 impl VersioningApi for VersioningConfiguration {
     fn enabled(&self) -> bool {
-        self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED))
+        self.status.as_deref() == Some("Enabled")
     }
     fn prefix_enabled(&self, prefix: &str) -> bool {
-        if self.status != Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)) {
+        if self.status.as_deref() != Some("Enabled") {
             return false;
         }
 
@@ -44,12 +44,10 @@ impl VersioningApi for VersioningConfiguration {
         }
 
         if let Some(ref excluded_prefixes) = self.excluded_prefixes {
-            for p in excluded_prefixes.iter() {
-                if let Some(ref sprefix) = p.prefix {
-                    let pattern = format!("{sprefix}*");
-                    if match_simple(&pattern, prefix) {
-                        return false;
-                    }
+            for sprefix in excluded_prefixes.iter().flatten() {
+                let pattern = format!("{sprefix}*");
+                if match_simple(&pattern, prefix) {
+                    return false;
                 }
             }
         }
@@ -58,11 +56,11 @@ impl VersioningApi for VersioningConfiguration {
     }
 
     fn prefix_suspended(&self, prefix: &str) -> bool {
-        if self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::SUSPENDED)) {
+        if self.status.as_deref() == Some("Suspended") {
             return true;
         }
 
-        if self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)) {
+        if self.status.as_deref() == Some("Enabled") {
             if prefix.is_empty() {
                 return false;
             }
@@ -75,12 +73,10 @@ impl VersioningApi for VersioningConfiguration {
             }
 
             if let Some(ref excluded_prefixes) = self.excluded_prefixes {
-                for p in excluded_prefixes.iter() {
-                    if let Some(ref sprefix) = p.prefix {
-                        let pattern = format!("{sprefix}*");
-                        if match_simple(&pattern, prefix) {
-                            return true;
-                        }
+                for sprefix in excluded_prefixes.iter().flatten() {
+                    let pattern = format!("{sprefix}*");
+                    if match_simple(&pattern, prefix) {
+                        return true;
                     }
                 }
             }
@@ -92,14 +88,12 @@ impl VersioningApi for VersioningConfiguration {
         self.prefix_enabled(prefix) || self.prefix_suspended(prefix)
     }
     fn suspended(&self) -> bool {
-        self.status == Some(BucketVersioningStatus::from_static(BucketVersioningStatus::SUSPENDED))
+        self.status.as_deref() == Some("Suspended")
     }
 }
 
 #[cfg(test)]
 mod tests {
-    use s3s::dto::{BucketVersioningStatus, ExcludedPrefix};
-
     use super::*;
 
     struct LegacyVersioning;
@@ -134,10 +128,8 @@ mod tests {
     #[test]
     fn delete_state_treats_excluded_prefixes_as_unversioned() {
         let config = VersioningConfiguration {
-            status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::ENABLED)),
-            excluded_prefixes: Some(vec![ExcludedPrefix {
-                prefix: Some("archive/".to_string()),
-            }]),
+            status: Some("Enabled".to_owned()),
+            excluded_prefixes: Some(vec![Some("archive/".to_string())]),
             ..Default::default()
         };
 
@@ -146,7 +138,7 @@ mod tests {
         assert_eq!(config.delete_state("live/object"), (true, false));
 
         let suspended = VersioningConfiguration {
-            status: Some(BucketVersioningStatus::from_static(BucketVersioningStatus::SUSPENDED)),
+            status: Some("Suspended".to_owned()),
             ..Default::default()
         };
         assert_eq!(suspended.delete_state("archive/object"), (false, true));

crates/ecstore/src/bucket/versioning_sys.rs

@@ -17,7 +17,7 @@ use super::metadata::unreadable_config_refusal;
 use super::{metadata_sys::get_bucket_metadata_sys, versioning::VersioningApi};
 use crate::disk::RUSTFS_META_BUCKET;
 use crate::error::Result;
-use s3s::dto::VersioningConfiguration;
+use rustfs_gateway_types::persistence::PersistedVersioningConfiguration as VersioningConfiguration;
 use tracing::{error, warn};
 
 pub struct BucketVersioningSys {}
@@ -163,7 +163,7 @@ mod tests {
 
     fn enabled() -> VersioningConfiguration {
         VersioningConfiguration {
-            status: Some(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::ENABLED)),
+            status: Some("Enabled".to_owned()),
             ..Default::default()
         }
     }

crates/ecstore/src/cache_value/metacache_set.rs

@@ -43,11 +43,6 @@ const EVENT_METACACHE_LISTING: &str = "metacache_listing";
 pub type AgreedFn = Box<dyn Fn(MetaCacheEntry) -> Pin<Box<dyn Future<Output = Result<(), DiskError>> + Send>> + Send + 'static>;
 pub type PartialFn =
     Box<dyn Fn(MetaCacheEntries, &[Option<DiskError>]) -> Pin<Box<dyn Future<Output = ()> + Send>> + Send + 'static>;
-pub(crate) type PartialResultFn = Box<
-    dyn Fn(MetaCacheEntries, &[Option<DiskError>]) -> Pin<Box<dyn Future<Output = disk::error::Result<()>> + Send>>
-        + Send
-        + 'static,
->;
 type FinishedFn = Box<dyn Fn(&[Option<DiskError>]) -> Pin<Box<dyn Future<Output = ()> + Send>> + Send + 'static>;
 
 #[derive(Clone, Default)]
@@ -60,6 +55,7 @@ impl FallbackClaimTracker {
         self.claimed.lock().await.insert(disk.endpoint().to_string());
     }
 
+    #[cfg(test)]
     pub(crate) async fn claimed_keys(&self) -> HashSet<String> {
         self.claimed.lock().await.clone()
     }
@@ -317,7 +313,17 @@ pub struct ListPathRawOutcome {
 pub async fn list_path_raw(rx: CancellationToken, opts: ListPathRawOptions) -> disk::error::Result<()> {
     let rx = rx.child_token();
     let _cancel_guard = rx.clone().drop_guard();
-    list_path_raw_inner(rx, opts, None, None).await.map(|_| ())
+    list_path_raw_inner(rx, opts, None).await.map(|_| ())
+}
+
+/// Like [`list_path_raw`], but reports how the merged stream ended.
+pub(crate) async fn list_path_raw_with_outcome(
+    rx: CancellationToken,
+    opts: ListPathRawOptions,
+) -> disk::error::Result<ListPathRawOutcome> {
+    let rx = rx.child_token();
+    let _cancel_guard = rx.clone().drop_guard();
+    list_path_raw_inner(rx, opts, None).await
 }
 
 pub(crate) async fn list_path_raw_with_claim_tracker(
@@ -327,25 +333,13 @@ pub(crate) async fn list_path_raw_with_claim_tracker(
 ) -> disk::error::Result<ListPathRawOutcome> {
     let rx = rx.child_token();
     let _cancel_guard = rx.clone().drop_guard();
-    list_path_raw_inner(rx, opts, Some(claim_tracker), None).await
-}
-
-pub(crate) async fn list_path_raw_with_partial_result(
-    rx: CancellationToken,
-    opts: ListPathRawOptions,
-    claim_tracker: FallbackClaimTracker,
-    partial_result: PartialResultFn,
-) -> disk::error::Result<ListPathRawOutcome> {
-    let rx = rx.child_token();
-    let _cancel_guard = rx.clone().drop_guard();
-    list_path_raw_inner(rx, opts, Some(claim_tracker), Some(partial_result)).await
+    list_path_raw_inner(rx, opts, Some(claim_tracker)).await
 }
 
 async fn list_path_raw_inner(
     rx: CancellationToken,
     opts: ListPathRawOptions,
     fallback_claim_tracker: Option<FallbackClaimTracker>,
-    partial_result: Option<PartialResultFn>,
 ) -> disk::error::Result<ListPathRawOutcome> {
     if opts.disks.is_empty() {
         return Err(DiskError::ErasureReadQuorum);
@@ -1020,13 +1014,7 @@ async fn list_path_raw_inner(
                 }
             }
 
-            if let Some(partial_fn) = partial_result.as_ref() {
-                tokio::select! {
-                    biased;
-                    _ = revjob_rx.cancelled() => return Ok(ListPathRawOutcome { limit_reached }),
-                    result = partial_fn(MetaCacheEntries(top_entries), &errs) => result?,
-                }
-            } else if let Some(partial_fn) = opts.partial.as_ref() {
+            if let Some(partial_fn) = opts.partial.as_ref() {
                 tokio::select! {
                     biased;
                     _ = revjob_rx.cancelled() => return Ok(ListPathRawOutcome { limit_reached }),
@@ -1308,7 +1296,6 @@ mod tests {
                 ..Default::default()
             },
             None,
-            None,
         )
         .await
         .expect("listing should complete");
@@ -1333,7 +1320,6 @@ mod tests {
                 ..Default::default()
             },
             None,
-            None,
         )
         .await
         .expect("listing should complete");
@@ -1827,39 +1813,6 @@ mod tests {
         );
     }
 
-    #[tokio::test]
-    async fn list_path_raw_propagates_partial_callback_failure() {
-        let first = MetaCacheEntry {
-            name: "bucket/object-a".to_string(),
-            metadata: vec![1],
-            ..Default::default()
-        };
-        let second = MetaCacheEntry {
-            name: "bucket/object-b".to_string(),
-            metadata: vec![2],
-            ..Default::default()
-        };
-
-        let err = list_path_raw_with_partial_result(
-            CancellationToken::new(),
-            ListPathRawOptions {
-                disks: vec![None, None],
-                min_disks: 2,
-                test_reader_behaviors: vec![
-                    TestReaderBehavior::Entries(vec![first]),
-                    TestReaderBehavior::Entries(vec![second]),
-                ],
-                ..Default::default()
-            },
-            FallbackClaimTracker::default(),
-            Box::new(move |_, _| Box::pin(async { Err(DiskError::ErasureReadQuorum) })),
-        )
-        .await
-        .expect_err("a resolution failure after reader advancement must fail the listing");
-
-        assert_eq!(err, DiskError::ErasureReadQuorum);
-    }
-
     #[tokio::test]
     async fn list_path_raw_cancels_a_blocked_partial_callback() {
         let cancel = CancellationToken::new();

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

@@ -105,7 +105,7 @@ pub async fn tier_mutation_control_time_out_client(
 
 /// Build a `NodeServiceClient` bound to the [`ChannelClass`]-appropriate channel for `addr`.
 ///
-/// Bulk `bytes`-carrying RPCs (ReadAll/WriteAll/CompareAndUpdateFile/ReadMultiple/BatchReadVersion) pass
+/// Bulk `bytes`-carrying RPCs, including native pool metadata recovery inventories, pass
 /// [`ChannelClass::Bulk`] so, when channel isolation is enabled, they are physically isolated
 /// from lock/health RPCs; everything else uses [`ChannelClass::Control`]. When isolation is
 /// disabled the two classes resolve to the same cached channel, i.e. legacy behavior.
@@ -357,6 +357,7 @@ fn rename_data_grpc_stage(path: &str) -> bool {
     matches!(
         path,
         "/node_service.NodeService/RenameData"
+            | "/node_service.NodeService/RenameMultipartData"
             | "/node_service.NodeService/RenameDataAtIncarnation"
             | "/node_service.NodeService/RenameDataWithQuotaFence"
     )
@@ -379,6 +380,7 @@ where
     }
 
     fn call(&mut self, mut request: HttpRequest<ReqBody>) -> Self::Future {
+        let rpc_method = request.extensions().get::<tonic::GrpcMethod<'_>>().cloned();
         let observe_rename_data = rustfs_io_metrics::put_stage_metrics_enabled() && rename_data_grpc_stage(request.uri().path());
         let request_scope_started = if observe_rename_data {
             rustfs_io_metrics::put_stage_timer()
@@ -447,13 +449,21 @@ where
                 let response_state = verify_tonic_peer_replay_capabilities_response(&audience, challenge, response.headers());
                 if let Err(error) = &response_state
                     && (response.headers().contains_key(RPC_REPLAY_CACHE_CAPABILITY_HEADER)
-                        || response.headers().contains_key(RPC_REPLAY_CACHE_CAPABILITY_PROOF_HEADER))
+                        || response.headers().contains_key(RPC_REPLAY_CACHE_CAPABILITY_PROOF_HEADER)
+                        || matches!(sent_state.cache_capability, Some(PeerReplayCapability::Capable { .. })))
                 {
                     debug!(
                         event = "internode_rpc_capability_proof_rejected",
                         component = "ecstore",
                         subsystem = "rpc_client",
                         result = "rejected",
+                        audience = audience.rsplit('@').next().unwrap_or("unknown"),
+                        rpc_service = rpc_method.as_ref().map(|method| method.service()).unwrap_or("unknown"),
+                        rpc_method = rpc_method.as_ref().map(|method| method.method()).unwrap_or("unknown"),
+                        http_status = response.status().as_u16(),
+                        grpc_status = response.headers().get("grpc-status").and_then(|value| value.to_str().ok()).unwrap_or("trailers"),
+                        capability_present = response.headers().contains_key(RPC_REPLAY_CACHE_CAPABILITY_HEADER),
+                        capability_proof_present = response.headers().contains_key(RPC_REPLAY_CACHE_CAPABILITY_PROOF_HEADER),
                         error = %error,
                         "internode RPC capability proof rejected"
                     )

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

@@ -1965,6 +1965,15 @@ impl PeerRestClient {
         Ok((self.topology_member.clone(), epoch))
     }
 
+    pub async fn probe_restore_copyback_capability(&self, topology_fingerprint: String) -> Result<(String, Uuid)> {
+        let probe = rustfs_protos::restore_copyback_capability_probe(Uuid::new_v4().as_bytes());
+        let result = self
+            .heal_control(rustfs_protos::HEAL_CONTROL_PROTOCOL_VERSION, topology_fingerprint, probe)
+            .await?;
+        let epoch = decode_remote_version_state_capability(&self.topology_member, &result)?;
+        Ok((self.topology_member.clone(), epoch))
+    }
+
     pub async fn probe_cross_pool_fence(&self, topology_fingerprint: String) -> Result<(String, u32, Uuid)> {
         let mut probe = rustfs_protos::CROSS_POOL_FENCE_CAPABILITY_PROBE_PREFIX.to_vec();
         probe.extend_from_slice(Uuid::new_v4().as_bytes());

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

@@ -22,13 +22,17 @@ use crate::cluster::rpc::internode_data_transport::{
     WriteStreamRequest,
 };
 use crate::disk::error::{Error, Result};
+use crate::disk::multipart_source::{
+    MULTIPART_SOURCE_PROTOCOL_VERSION, MultipartSourceOperation, MultipartSourceRequest, MultipartSourceResponse,
+};
 use crate::disk::{
     BatchReadVersionReq, BatchReadVersionResp, CheckPartsResp, ConditionalFileUpdate, DELETE_RECOVERY_PROTOCOL_VERSION,
     DeleteOptions, DeleteRecoveryAction, DeleteRecoveryCursor, DeleteRecoveryDigest, DeleteRecoveryInspection,
     DeleteRecoveryPage, DeleteRecoveryRpcOperation, DeleteRecoveryRpcReply, DiskAPI, DiskInfo, DiskInfoOptions, DiskLocation,
     DiskOption, FileInfoVersions, FileReader, FileWriter, GuardedDeleteVersionsPayload, GuardedFileInfoVersions,
-    PartTransactionAction, ReadMultipleReq, ReadMultipleResp, ReadOptions, RenameDataResp, SnapshotLeaseToken,
-    UpdateMetadataOpts, VolumeInfo, WalkDirOptions, batch_read_version_one_by_one,
+    PartTransactionAction, PoolMetaNativeRecoveryRequest, PoolMetaNativeRecoveryResponse, ReadMultipleReq, ReadMultipleResp,
+    ReadOptions, RenameDataResp, SnapshotLeaseToken, UpdateMetadataOpts, VolumeInfo, WalkDirOptions,
+    batch_read_version_one_by_one,
     disk_store::{
         DEFAULT_RUSTFS_DRIVE_ACTIVE_MONITORING, ENV_RUSTFS_DRIVE_ACTIVE_MONITORING, SKIP_IF_SUCCESS_BEFORE,
         get_drive_active_check_interval, get_drive_active_check_timeout, get_drive_disk_info_timeout, get_drive_list_dir_timeout,
@@ -62,11 +66,14 @@ use rustfs_protos::proto_gen::node_service::{
     BatchReadVersionRequest, BatchReadVersionResponse, CheckPartsRequest, CompareAndUpdateFileOutcome,
     CompareAndUpdateFileRequest, DeletePathsRequest, DeleteRequest, DeleteVersionRequest, DeleteVersionsRequest,
     DeleteVersionsResponse, DeleteVolumeRequest, DiskInfoRequest, GuardedDeleteVersionsRequest, ListDirRequest,
-    ListVolumesRequest, MakeVolumeRequest, MakeVolumesRequest, PreparePartTransactionRequest, ReadAllRequest,
-    ReadMetadataRequest, ReadMultipleRequest, ReadMultipleResponse, ReadPartsRequest, ReadVersionRequest, ReadXlRequest,
-    RecoverDeleteTransactionRequest, RenameDataRequest, RenameFileRequest, SettlePartTransactionRequest,
-    SnapshotLeaseReleaseRequest, SnapshotLeaseRenewRequest, SnapshotLeaseRequest, SnapshotLeaseResponse, StatVolumeRequest,
-    UpdateMetadataRequest, VerifyFileRequest, WriteAllRequest, WriteMetadataRequest, node_service_client::NodeServiceClient,
+    ListVolumesRequest, MakeVolumeRequest, MakeVolumesRequest, PoolMetaNativeRecoveryRequest as PoolMetaNativeRecoveryRpcRequest,
+    PreparePartTransactionRequest, ReadAllRequest, ReadMetadataRequest, ReadMultipleRequest, ReadMultipleResponse,
+    ReadPartsRequest, ReadVersionRequest, ReadXlRequest, RecoverDeleteTransactionRequest,
+    RecoverMultipartSourceRequest as RecoverMultipartSourceRpcRequest,
+    RecoverMultipartSourceResponse as RecoverMultipartSourceRpcResponse, RenameDataRequest, RenameFileRequest,
+    SettlePartTransactionRequest, SnapshotLeaseReleaseRequest, SnapshotLeaseRenewRequest, SnapshotLeaseRequest,
+    SnapshotLeaseResponse, StatVolumeRequest, UpdateMetadataRequest, VerifyFileRequest, WriteAllRequest, WriteMetadataRequest,
+    node_service_client::NodeServiceClient,
 };
 use serde::{Serialize, de::DeserializeOwned};
 use std::{
@@ -237,6 +244,36 @@ fn attach_quota_mutation_body_digest<T>(
     crate::cluster::rpc::set_tonic_canonical_body_digest(request, &body).map_err(Error::other)
 }
 
+fn decode_multipart_source_recovery_response(
+    request: &MultipartSourceRequest,
+    response: RecoverMultipartSourceRpcResponse,
+) -> Result<MultipartSourceResponse> {
+    if response.protocol_version != u32::from(MULTIPART_SOURCE_PROTOCOL_VERSION) {
+        return Err(DiskError::UnsupportedDisk);
+    }
+    if response.result.len() > MultipartSourceOperation::max_wire_payload_bytes() {
+        return Err(DiskError::FileCorrupt);
+    }
+    if !response.success {
+        if !response.result.is_empty() {
+            return Err(DiskError::FileCorrupt);
+        }
+        let error = response.error.ok_or(DiskError::FileCorrupt)?;
+        if DiskError::from_u32(error.code).is_none() {
+            return Err(DiskError::FileCorrupt);
+        }
+        return Err(error.into());
+    }
+    if response.error.is_some() {
+        return Err(DiskError::FileCorrupt);
+    }
+    let result: MultipartSourceResponse =
+        rustfs_protos::decode_multipart_source_recovery_payload(&response.result).map_err(|_| DiskError::FileCorrupt)?;
+    result.validate_wire()?;
+    result.validate_for(request)?;
+    Ok(result)
+}
+
 fn decode_volume_infos(volume_infos: Vec<String>) -> Result<Vec<VolumeInfo>> {
     volume_infos
         .into_iter()
@@ -2128,7 +2165,7 @@ impl RemoteDisk {
             .map_err(|err| Error::RemoteClientUnavailable(err.to_string()))
     }
 
-    /// Client for large `bytes`-carrying RPCs (ReadAll/WriteAll/CompareAndUpdateFile/ReadMultiple/BatchReadVersion).
+    /// Client for large `bytes`-carrying RPCs, including native pool metadata recovery inventories.
     /// Routes onto the isolated bulk channel pool so large transfers cannot head-of-line block
     /// lock/health RPCs (grpc-optimization P1). Falls back to the control channel when isolation
     /// is disabled.
@@ -2224,6 +2261,38 @@ fn decode_msgpack_or_json<T: DeserializeOwned>(binary: &[u8], json: &str, value_
     }
 }
 
+fn decode_multipart_publication_response(binary: &[u8]) -> Result<RenameDataResp> {
+    let result = if binary.len() > MultipartSourceOperation::max_wire_payload_bytes() {
+        Err(DiskError::FileCorrupt)
+    } else {
+        rustfs_protos::decode_multipart_source_recovery_payload(binary).map_err(|_| DiskError::FileCorrupt)
+    };
+    if result.is_ok() {
+        crate::cluster::rpc::runtime_sources::record_response_msgpack_decode("RenameDataResp");
+    } else {
+        crate::cluster::rpc::runtime_sources::record_response_msgpack_decode_error("RenameDataResp");
+    }
+    // A codec error must never echo a malformed opaque capability.
+    result
+}
+
+fn validate_multipart_publication_envelope(response: &rustfs_protos::proto_gen::node_service::RenameDataResponse) -> Result<()> {
+    if !response.success {
+        if !response.rename_data_resp.is_empty() || !response.rename_data_resp_bin.is_empty() {
+            return Err(DiskError::FileCorrupt);
+        }
+        let error = response.error.as_ref().ok_or(DiskError::FileCorrupt)?;
+        if DiskError::from_u32(error.code).is_none() {
+            return Err(DiskError::FileCorrupt);
+        }
+        return Err(error.clone().into());
+    }
+    if response.error.is_some() {
+        return Err(DiskError::FileCorrupt);
+    }
+    Ok(())
+}
+
 fn read_version_stage_timer(attribution_enabled: bool) -> Option<Instant> {
     attribution_enabled.then(Instant::now)
 }
@@ -2399,6 +2468,18 @@ impl RemoteDisk {
         dst_path: &str,
         scanner_publication_lease_token: Option<Uuid>,
     ) -> Result<RenameDataResp> {
+        let multipart_rename = src_volume == crate::disk::RUSTFS_META_MULTIPART_BUCKET;
+        let pool_meta_cas = rustfs_utils::http::metadata_compat::contains_key_str(
+            &fi.metadata,
+            crate::disk::POOL_META_NATIVE_CAS_METADATA_SUFFIX,
+        );
+        if pool_meta_cas
+            && (dst_volume != crate::disk::RUSTFS_META_BUCKET
+                || dst_path != "pool.bin"
+                || scanner_publication_lease_token.is_some())
+        {
+            return Err(DiskError::UnsupportedDisk);
+        }
         trace!(
             event = EVENT_REMOTE_DISK_RPC,
             component = LOG_COMPONENT_ECSTORE,
@@ -2441,9 +2522,12 @@ impl RemoteDisk {
                     &fi.metadata,
                     crate::disk::QUOTA_MUTATION_FENCE_METADATA_SUFFIX,
                 );
-                if quota_fenced {
+                if pool_meta_cas {
+                    let body = canonical_body.map_err(|_| Error::other("pool metadata CAS body length cannot be represented"))?;
+                    crate::cluster::rpc::set_tonic_canonical_body_digest(&mut request, &body).map_err(Error::other)?;
+                } else if quota_fenced {
                     attach_quota_mutation_body_digest(&mut request, canonical_body)?;
-                } else if scanner_publication_lease_token.is_some() || incarnation_bound {
+                } else if multipart_rename || scanner_publication_lease_token.is_some() || incarnation_bound {
                     let canonical_body =
                         canonical_body.map_err(|_| Error::other("rename_data request length cannot be represented"))?;
                     crate::cluster::rpc::set_tonic_canonical_body_digest(&mut request, &canonical_body).map_err(Error::other)?;
@@ -2458,7 +2542,12 @@ impl RemoteDisk {
                 let rpc_started = rustfs_io_metrics::put_stage_timer();
                 let response = rustfs_io_metrics::observe_put_stage_future(
                     async {
-                        if quota_fenced {
+                        if multipart_rename {
+                            // Old peers must reject before an upload source can be changed.
+                            client.rename_multipart_data(request).await
+                        } else if pool_meta_cas {
+                            client.rename_data_with_pool_meta_cas(request).await
+                        } else if quota_fenced {
                             // The method itself proves quota-token enforcement; an old peer must fail closed.
                             client.rename_data_with_quota_fence(request).await
                         } else if incarnation_bound {
@@ -2472,7 +2561,14 @@ impl RemoteDisk {
                     rustfs_io_metrics::PUT_STAGE_SET_DISK_RENAME_REMOTE_CLIENT_RPC_AWAIT_POLL_PENDING_COUNT,
                     rustfs_io_metrics::PUT_STAGE_SET_DISK_RENAME_REMOTE_CLIENT_RPC_AWAIT_FIRST_PENDING_TO_READY,
                 )
-                .await?;
+                .await
+                .map_err(|status| {
+                    if (multipart_rename || pool_meta_cas) && status.code() == Code::Unimplemented {
+                        DiskError::UnsupportedDisk
+                    } else {
+                        status.into()
+                    }
+                })?;
                 let response_process_started = rustfs_io_metrics::put_stage_timer();
                 let response_headers_started = rustfs_io_metrics::put_stage_timer();
                 let _ = response.metadata().len();
@@ -2492,7 +2588,9 @@ impl RemoteDisk {
                 );
 
                 let success_check_started = rustfs_io_metrics::put_stage_timer();
-                if !response.success {
+                if multipart_rename {
+                    validate_multipart_publication_envelope(&response)?;
+                } else if !response.success {
                     return Err(response.error.unwrap_or_default().into());
                 }
                 rustfs_io_metrics::record_put_object_stage_duration_from(
@@ -2501,11 +2599,15 @@ impl RemoteDisk {
                 );
 
                 let decode_started = rustfs_io_metrics::put_stage_timer();
-                let rename_data_resp = decode_msgpack_or_json::<RenameDataResp>(
-                    &response.rename_data_resp_bin,
-                    &response.rename_data_resp,
-                    "RenameDataResp",
-                )?;
+                let rename_data_resp = if multipart_rename {
+                    decode_multipart_publication_response(&response.rename_data_resp_bin)?
+                } else {
+                    decode_msgpack_or_json::<RenameDataResp>(
+                        &response.rename_data_resp_bin,
+                        &response.rename_data_resp,
+                        "RenameDataResp",
+                    )?
+                };
                 rustfs_io_metrics::record_put_object_stage_duration_from(
                     rustfs_io_metrics::PUT_STAGE_SET_DISK_RENAME_REMOTE_CLIENT_DECODE,
                     decode_started,
@@ -2515,6 +2617,14 @@ impl RemoteDisk {
                     response_process_started,
                 );
 
+                if multipart_rename {
+                    let publication = rename_data_resp.multipart_source.as_ref().ok_or(DiskError::FileCorrupt)?;
+                    publication.validate_wire()?;
+                    let expected = crate::disk::multipart_source::publication_identity(fi)?.ok_or(DiskError::FileCorrupt)?;
+                    if publication.publication != expected {
+                        return Err(DiskError::FileCorrupt);
+                    }
+                }
                 Ok(rename_data_resp)
             },
             get_max_timeout_duration(),
@@ -2907,6 +3017,20 @@ impl DiskAPI for RemoteDisk {
                 let file_info_bin = encode_file_info_msgpack(&fi)?;
                 let opts_bin = encode_delete_options_msgpack(&opts)?;
                 let conditional_marker = opts.expected_delete_marker.is_some();
+                let pool_meta_undo = rustfs_utils::http::metadata_compat::contains_key_str(
+                    &fi.metadata,
+                    crate::disk::POOL_META_NATIVE_CAS_METADATA_SUFFIX,
+                );
+                if pool_meta_undo
+                    && (volume != crate::disk::RUSTFS_META_BUCKET
+                        || path != "pool.bin"
+                        || !opts.undo_write
+                        || opts.undo_delete
+                        || force_del_marker
+                        || conditional_marker)
+                {
+                    return Err(DiskError::UnsupportedDisk);
+                }
                 let file_info = serde_json::to_string(&fi)?;
                 let opts = serde_json::to_string(&opts)?;
 
@@ -2926,7 +3050,7 @@ impl DiskAPI for RemoteDisk {
                 });
                 let canonical_body = rustfs_protos::canonical_delete_version_request_body(request.get_ref());
                 let incarnation_bound = !request.get_ref().bucket_incarnation_id.is_empty();
-                if incarnation_bound {
+                if incarnation_bound || pool_meta_undo {
                     let body = canonical_body.map_err(|_| Error::other("delete-version body length cannot be represented"))?;
                     crate::cluster::rpc::set_tonic_canonical_body_digest(&mut request, &body).map_err(Error::other)?;
                 } else {
@@ -2935,7 +3059,15 @@ impl DiskAPI for RemoteDisk {
 
                 // The marker-specific method rejects older peers; its body digest
                 // also binds any incarnation, so neither precondition can be lost.
-                let response = if conditional_marker {
+                let response = if pool_meta_undo {
+                    client.undo_pool_meta_write(request).await.map_err(|status| {
+                        if status.code() == Code::Unimplemented {
+                            DiskError::UnsupportedDisk
+                        } else {
+                            status.into()
+                        }
+                    })?
+                } else if conditional_marker {
                     client.delete_retired_marker(request).await?
                 } else if incarnation_bound {
                     client.delete_version_at_incarnation(request).await?
@@ -3617,6 +3749,107 @@ impl DiskAPI for RemoteDisk {
             .await
     }
 
+    async fn rename_data_with_pool_meta_cas(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+    ) -> Result<RenameDataResp> {
+        if !rustfs_utils::http::metadata_compat::contains_key_str(&fi.metadata, crate::disk::POOL_META_NATIVE_CAS_METADATA_SUFFIX)
+        {
+            return Err(DiskError::UnsupportedDisk);
+        }
+        self.rename_data_borrowed(src_volume, src_path, &fi, dst_volume, dst_path)
+            .await
+    }
+
+    async fn undo_pool_meta_write(&self, volume: &str, path: &str, fi: FileInfo, opts: DeleteOptions) -> Result<()> {
+        if !rustfs_utils::http::metadata_compat::contains_key_str(&fi.metadata, crate::disk::POOL_META_NATIVE_CAS_METADATA_SUFFIX)
+        {
+            return Err(DiskError::UnsupportedDisk);
+        }
+        self.delete_version(volume, path, fi, false, opts).await
+    }
+
+    async fn pool_meta_native_recovery(
+        &self,
+        operation: PoolMetaNativeRecoveryRequest,
+    ) -> Result<PoolMetaNativeRecoveryResponse> {
+        self.execute_with_timeout_for_op(
+            "pool_meta_native_recovery",
+            || async {
+                let payload = rmp_serde::to_vec_named(&operation).map_err(Error::other)?;
+                if payload.len() > rustfs_protos::POOL_META_NATIVE_RECOVERY_MAX_PAYLOAD_BYTES {
+                    return Err(DiskError::FileCorrupt);
+                }
+                let mut client = self.get_bulk_client().await?;
+                let mut request = Request::new(PoolMetaNativeRecoveryRpcRequest {
+                    disk: self.endpoint.to_string(),
+                    payload: payload.into(),
+                });
+                let body = rustfs_protos::canonical_pool_meta_native_recovery_request_body(request.get_ref())
+                    .map_err(|_| Error::other("pool metadata recovery body length cannot be represented"))?;
+                crate::cluster::rpc::set_tonic_canonical_body_digest(&mut request, &body).map_err(Error::other)?;
+                let response = client
+                    .pool_meta_native_recovery(request)
+                    .await
+                    .map_err(|status| {
+                        if status.code() == Code::Unimplemented {
+                            DiskError::UnsupportedDisk
+                        } else {
+                            status.into()
+                        }
+                    })?
+                    .into_inner();
+                if !response.success {
+                    return Err(response.error.unwrap_or_default().into());
+                }
+                rustfs_protos::decode_pool_meta_native_recovery_payload(&response.payload).map_err(Error::other)
+            },
+            get_max_timeout_duration(),
+        )
+        .await
+    }
+
+    async fn recover_multipart_source(&self, mut source: MultipartSourceRequest) -> Result<MultipartSourceResponse> {
+        source.prepare_wire()?;
+        let payload = rmp_serde::to_vec_named(&source.operation).map_err(Error::other)?;
+        if payload.len() > MultipartSourceOperation::max_wire_payload_bytes() {
+            return Err(DiskError::FileCorrupt);
+        }
+        self.execute_with_timeout_for_op(
+            "recover_multipart_source",
+            || async {
+                let mut client = self.get_bulk_client().await?;
+                let mut request = Request::new(RecoverMultipartSourceRpcRequest {
+                    disk: self.endpoint.to_string(),
+                    path: source.path.clone(),
+                    protocol_version: u32::from(source.protocol_version),
+                    operation: payload.into(),
+                });
+                let body = rustfs_protos::canonical_recover_multipart_source_request_body(request.get_ref())
+                    .map_err(|_| Error::other("multipart source recovery request length cannot be represented"))?;
+                crate::cluster::rpc::set_tonic_canonical_body_digest(&mut request, &body).map_err(Error::other)?;
+                let response = client
+                    .recover_multipart_source(request)
+                    .await
+                    .map_err(|status| {
+                        if status.code() == Code::Unimplemented {
+                            DiskError::UnsupportedDisk
+                        } else {
+                            status.into()
+                        }
+                    })?
+                    .into_inner();
+                decode_multipart_source_recovery_response(&source, response)
+            },
+            get_max_timeout_duration(),
+        )
+        .await
+    }
+
     #[tracing::instrument(level = "trace", skip_all)]
     async fn list_dir(&self, _origvolume: &str, volume: &str, dir_path: &str, count: i32) -> Result<Vec<String>> {
         trace!(volume, dir_path, "Remote disk list_dir RPC started");
@@ -4553,6 +4786,322 @@ mod tests {
 
     static INIT: Once = Once::new();
 
+    fn multipart_source_claim_request() -> MultipartSourceRequest {
+        use crate::disk::multipart_source::{
+            MultipartRecoveryRound, MultipartSourcePhase, MultipartSourceSnapshot, SourceGeneration, SourceGeometry,
+            SourceLineage,
+        };
+        let original_data_dir = Uuid::new_v4();
+        let lineage = SourceLineage {
+            bucket: "bucket".to_string(),
+            object: "object".to_string(),
+            path: "upload".to_string(),
+            original_data_dir,
+            original_version_id: None,
+            created_unix_nanos: 1,
+            original_identity: [1; 32],
+            geometry: SourceGeometry {
+                data_blocks: 2,
+                parity_blocks: 2,
+                block_size: 1024,
+                distribution: vec![1, 2, 3, 4],
+            },
+        };
+        let generation = SourceGeneration {
+            epoch: 0,
+            nonce: original_data_dir,
+        };
+        MultipartSourceRequest {
+            protocol_version: MULTIPART_SOURCE_PROTOCOL_VERSION,
+            path: lineage.path.clone(),
+            operation: MultipartSourceOperation::Claim {
+                expected: MultipartSourceSnapshot {
+                    disk_id: Uuid::new_v4(),
+                    physical_slot: 0,
+                    lineage: lineage.clone(),
+                    generation,
+                    revision: 0,
+                    phase: MultipartSourcePhase::Active,
+                    state_sha256: [2; 32],
+                    head_sha256: Some([3; 32]),
+                    inventory_sha256: [4; 32],
+                    capability_sha256: None,
+                    publication: None,
+                },
+                round: MultipartRecoveryRound {
+                    lineage,
+                    from_generation: generation,
+                    old_attempt: Uuid::new_v4(),
+                    generation: SourceGeneration {
+                        epoch: 1,
+                        nonce: Uuid::new_v4(),
+                    },
+                    round: Uuid::new_v4(),
+                },
+            },
+        }
+    }
+
+    #[tokio::test]
+    async fn multipart_source_recovery_old_peer_rejects_without_generic_fallback() {
+        runtime_sources::ensure_test_rpc_secret();
+        let peer = TestGrpcPeer::spawn(Bytes::new(), Bytes::new())
+            .await
+            .expect("old peer must bind");
+        let remote = remote_disk_for_addr(&peer.addr).await;
+        for request in [
+            MultipartSourceRequest {
+                protocol_version: MULTIPART_SOURCE_PROTOCOL_VERSION,
+                path: "upload".to_string(),
+                operation: MultipartSourceOperation::Inspect { lineage: None },
+            },
+            multipart_source_claim_request(),
+        ] {
+            assert_eq!(
+                remote
+                    .recover_multipart_source(request)
+                    .await
+                    .expect_err("old peer must reject before recovery work"),
+                DiskError::UnsupportedDisk
+            );
+        }
+        assert_eq!(
+            *peer.peer.unsupported_methods.lock().expect("observed methods lock"),
+            vec!["/node_service.NodeService/RecoverMultipartSource"; 2],
+            "read and mutation requests must never fall back to generic reads, writes or legacy recovery",
+        );
+        peer.stop().await;
+    }
+
+    #[test]
+    fn multipart_publication_response_is_strict_and_redacts_codec_errors() {
+        let encoded = rmp_serde::to_vec_named(&RenameDataResp::default()).expect("publication response");
+        assert!(decode_multipart_publication_response(&encoded).is_ok());
+        let mut trailing = encoded;
+        trailing.push(0xc0);
+        let secret = "private-undo-capability";
+        let malformed =
+            rmp_serde::to_vec_named(&serde_json::json!({ "multipart_source": secret })).expect("malformed capability response");
+        for binary in [Vec::new(), trailing, malformed] {
+            let error = decode_multipart_publication_response(&binary).expect_err("invalid publication response");
+            assert!(matches!(error, DiskError::FileCorrupt));
+            assert!(!error.to_string().contains(secret));
+        }
+        use rustfs_protos::proto_gen::node_service::RenameDataResponse;
+        for response in [
+            RenameDataResponse {
+                success: true,
+                error: Some(DiskError::FileNotFound.into()),
+                ..Default::default()
+            },
+            RenameDataResponse::default(),
+            RenameDataResponse {
+                success: false,
+                error: Some(rustfs_protos::proto_gen::node_service::Error {
+                    code: 0,
+                    error_info: secret.to_string(),
+                }),
+                ..Default::default()
+            },
+        ] {
+            assert!(matches!(validate_multipart_publication_envelope(&response), Err(DiskError::FileCorrupt)));
+        }
+        let response = RenameDataResponse {
+            success: false,
+            error: Some(DiskError::ConditionalDeleteConflict.into()),
+            ..Default::default()
+        };
+        assert!(matches!(
+            validate_multipart_publication_envelope(&response),
+            Err(DiskError::ConditionalDeleteConflict)
+        ));
+    }
+
+    #[test]
+    fn multipart_source_recovery_rejects_malformed_and_unrelated_replies() {
+        use crate::disk::multipart_source::MultipartSourceInspection;
+        let source = MultipartSourceRequest {
+            protocol_version: MULTIPART_SOURCE_PROTOCOL_VERSION,
+            path: "upload".to_string(),
+            operation: MultipartSourceOperation::Inspect { lineage: None },
+        };
+        let result = MultipartSourceResponse::Inspected(MultipartSourceInspection {
+            disk_id: Uuid::new_v4(),
+            snapshot: None,
+            original: None,
+            recovery: None,
+            negative_certificate: None,
+            ready_certificate: None,
+            terminal_certificate: Vec::new(),
+            root_missing: false,
+            abort_round: None,
+            abort_certificate: None,
+        });
+        let valid = RecoverMultipartSourceRpcResponse {
+            success: true,
+            protocol_version: u32::from(MULTIPART_SOURCE_PROTOCOL_VERSION),
+            result: rmp_serde::to_vec_named(&result).expect("inspection reply").into(),
+            error: None,
+        };
+        assert!(decode_multipart_source_recovery_response(&source, valid.clone()).is_ok());
+        let claim = multipart_source_claim_request();
+        let MultipartSourceOperation::Claim { mut expected, .. } = claim.operation else { unreachable!() };
+        expected.lineage.path = "other-upload".to_string();
+        let wrong_scope = MultipartSourceResponse::Inspected(MultipartSourceInspection {
+            disk_id: expected.disk_id,
+            snapshot: Some(expected),
+            original: None,
+            recovery: None,
+            negative_certificate: None,
+            ready_certificate: None,
+            terminal_certificate: Vec::new(),
+            root_missing: false,
+            abort_round: None,
+            abort_certificate: None,
+        });
+        let wrong_scope = RecoverMultipartSourceRpcResponse {
+            result: rmp_serde::to_vec_named(&wrong_scope).expect("unrelated inspection").into(),
+            ..valid.clone()
+        };
+        assert!(decode_multipart_source_recovery_response(&source, wrong_scope).is_err());
+        let mut wrong_version = valid.clone();
+        wrong_version.protocol_version += 1;
+        assert!(matches!(
+            decode_multipart_source_recovery_response(&source, wrong_version),
+            Err(DiskError::UnsupportedDisk)
+        ));
+        for invalid in [
+            RecoverMultipartSourceRpcResponse {
+                error: Some(DiskError::FileNotFound.into()),
+                ..valid.clone()
+            },
+            RecoverMultipartSourceRpcResponse {
+                success: false,
+                error: None,
+                result: Bytes::new(),
+                ..valid.clone()
+            },
+            RecoverMultipartSourceRpcResponse {
+                success: false,
+                error: Some(rustfs_protos::proto_gen::node_service::Error {
+                    code: 0,
+                    error_info: "invalid code".to_string(),
+                }),
+                result: Bytes::new(),
+                ..valid.clone()
+            },
+            RecoverMultipartSourceRpcResponse {
+                success: false,
+                error: Some(DiskError::FileNotFound.into()),
+                ..valid.clone()
+            },
+            RecoverMultipartSourceRpcResponse {
+                result: Bytes::from_static(b"invalid-msgpack"),
+                ..valid.clone()
+            },
+            {
+                let mut trailing = valid;
+                let mut bytes = trailing.result.to_vec();
+                bytes.push(0xc0);
+                trailing.result = bytes.into();
+                trailing
+            },
+        ] {
+            assert!(decode_multipart_source_recovery_response(&source, invalid).is_err());
+        }
+        let native_error = RecoverMultipartSourceRpcResponse {
+            success: false,
+            protocol_version: u32::from(MULTIPART_SOURCE_PROTOCOL_VERSION),
+            result: Bytes::new(),
+            error: Some(DiskError::ConditionalDeleteConflict.into()),
+        };
+        assert!(matches!(
+            decode_multipart_source_recovery_response(&source, native_error),
+            Err(DiskError::ConditionalDeleteConflict)
+        ));
+    }
+
+    #[test]
+    fn multipart_source_recovery_does_not_accept_caller_constructed_claim_proofs() {
+        use crate::disk::multipart_source::{MultipartProofKind, MultipartSourcePhase, MultipartSourceProof};
+        runtime_sources::ensure_test_rpc_secret();
+        let request = multipart_source_claim_request();
+        let MultipartSourceOperation::Claim { expected, round } = &request.operation else {
+            unreachable!()
+        };
+        let mut snapshot = expected.clone();
+        snapshot.generation = round.generation;
+        snapshot.phase = MultipartSourcePhase::Recovering;
+        snapshot.revision += 1;
+        for authentication in [
+            serde_json::json!({ "class": "cluster_hmac", "mac": vec![0_u8; 32] }),
+            serde_json::json!({ "class": "native_local" }),
+        ] {
+            let proof: MultipartSourceProof = serde_json::from_value(serde_json::json!({
+                "kind": MultipartProofKind::Claimed,
+                "challenge": round.round,
+                "recovery": round,
+                "snapshot": snapshot,
+                "authentication": authentication,
+            }))
+            .expect("untrusted wire fields can decode without gaining native authority");
+            let response = RecoverMultipartSourceRpcResponse {
+                success: true,
+                protocol_version: u32::from(MULTIPART_SOURCE_PROTOCOL_VERSION),
+                result: rmp_serde::to_vec_named(&MultipartSourceResponse::Claimed(proof))
+                    .expect("forged claim fields")
+                    .into(),
+                error: None,
+            };
+            assert!(
+                decode_multipart_source_recovery_response(&request, response).is_err(),
+                "matching caller fields cannot mint a native H vote"
+            );
+        }
+    }
+
+    #[tokio::test]
+    async fn multipart_rename_old_peer_rejects_without_legacy_fallback() {
+        runtime_sources::ensure_test_rpc_secret();
+        let peer = TestGrpcPeer::spawn(Bytes::new(), Bytes::new())
+            .await
+            .expect("old peer must bind");
+        let remote = remote_disk_for_addr(&peer.addr).await;
+        let file_info = sample_rename_data_file_info();
+        for scanner_lease in [None, Some(Uuid::new_v4())] {
+            assert_eq!(
+                remote
+                    .rename_data_borrowed_with_fence(
+                        crate::disk::RUSTFS_META_MULTIPART_BUCKET,
+                        "upload",
+                        &file_info,
+                        "bucket",
+                        "object",
+                        scanner_lease,
+                    )
+                    .await
+                    .expect_err("old peer cannot preserve multipart source data"),
+                DiskError::UnsupportedDisk,
+            );
+        }
+        assert!(
+            remote
+                .rename_data("tmp", "source", file_info, "bucket", "object")
+                .await
+                .is_err()
+        );
+        assert_eq!(
+            *peer.peer.unsupported_methods.lock().expect("observed methods lock"),
+            vec![
+                "/node_service.NodeService/RenameMultipartData",
+                "/node_service.NodeService/RenameMultipartData",
+                "/node_service.NodeService/RenameData",
+            ],
+            "each dispatch must use exactly one method; multipart publication must never downgrade",
+        );
+        peer.stop().await;
+    }
+
     #[tokio::test]
     async fn quota_fence_old_peer_rejects_without_legacy_fallback() {
         runtime_sources::ensure_test_rpc_secret();
@@ -4607,6 +5156,153 @@ mod tests {
         peer.stop().await;
     }
 
+    #[tokio::test]
+    async fn pool_meta_native_recovery_uses_bulk_channel_without_replacing_control() {
+        runtime_sources::ensure_test_rpc_secret();
+        temp_env::async_with_vars(
+            [
+                (rustfs_config::ENV_INTERNODE_CHANNEL_ISOLATION, Some("true")),
+                (rustfs_config::ENV_INTERNODE_BULK_CHANNELS, Some("1")),
+            ],
+            async {
+                let peer = TestGrpcPeer::spawn(Bytes::new(), Bytes::new())
+                    .await
+                    .expect("route test peer must bind");
+                let remote = remote_disk_for_addr(&peer.addr).await;
+                let bulk_key = format!("{}\u{0}bulk\u{0}0", peer.addr);
+                drop(remote.get_client().await.expect("control channel must connect"));
+                assert!(rustfs_common::has_cached_connection(&peer.addr).await, "control channel must be warm");
+                assert!(
+                    !rustfs_common::has_cached_connection(&bulk_key).await,
+                    "the unique peer must have no bulk channel yet"
+                );
+
+                assert_eq!(
+                    remote
+                        .pool_meta_native_recovery(PoolMetaNativeRecoveryRequest::Probe {})
+                        .await,
+                    Err(DiskError::UnsupportedDisk),
+                    "the old-peer response must still traverse the real recovery RPC route",
+                );
+                assert!(
+                    rustfs_common::has_cached_connection(&peer.addr).await,
+                    "recovery must retain the control cache"
+                );
+                assert!(
+                    rustfs_common::has_cached_connection(&bulk_key).await,
+                    "recovery must establish its isolated bulk channel"
+                );
+                for request in [
+                    MultipartSourceRequest {
+                        protocol_version: MULTIPART_SOURCE_PROTOCOL_VERSION,
+                        path: "upload".to_owned(),
+                        operation: MultipartSourceOperation::Inspect { lineage: None },
+                    },
+                    multipart_source_claim_request(),
+                ] {
+                    rustfs_common::evict_connection(&bulk_key).await;
+                    assert!(
+                        !rustfs_common::has_cached_connection(&bulk_key).await,
+                        "each multipart recovery must establish its own observed bulk route"
+                    );
+                    assert_eq!(
+                        remote
+                            .recover_multipart_source(request)
+                            .await
+                            .expect_err("old peer must reject before recovery work"),
+                        DiskError::UnsupportedDisk,
+                    );
+                    assert!(
+                        rustfs_common::has_cached_connection(&peer.addr).await,
+                        "multipart recovery must retain the control channel"
+                    );
+                    assert!(
+                        rustfs_common::has_cached_connection(&bulk_key).await,
+                        "multipart inspection and mutation must both establish an isolated bulk channel"
+                    );
+                }
+                assert_eq!(
+                    *peer.peer.unsupported_methods.lock().expect("observed methods lock"),
+                    vec![
+                        "/node_service.NodeService/PoolMetaNativeRecovery",
+                        "/node_service.NodeService/RecoverMultipartSource",
+                        "/node_service.NodeService/RecoverMultipartSource",
+                    ],
+                );
+
+                remote.cancel_token.cancel();
+                drop(remote);
+                rustfs_common::evict_connection(&peer.addr).await;
+                rustfs_common::evict_connection(&bulk_key).await;
+                peer.stop().await;
+            },
+        )
+        .await;
+    }
+
+    #[tokio::test]
+    async fn pool_meta_native_old_peer_rejects_without_legacy_fallback() {
+        runtime_sources::ensure_test_rpc_secret();
+        let peer = TestGrpcPeer::spawn(Bytes::new(), Bytes::new())
+            .await
+            .expect("old peer must bind");
+        let remote = remote_disk_for_addr(&peer.addr).await;
+        let mut file_info = sample_rename_data_file_info();
+        crate::disk::PoolMetaNativeCasCondition {
+            attempt_id: Uuid::new_v4(),
+            parent: None,
+            parent_raw_sha256: None,
+        }
+        .insert_into(&mut file_info.metadata)
+        .expect("native condition must encode");
+        assert_eq!(
+            remote
+                .rename_data_borrowed_with_fence(
+                    crate::disk::RUSTFS_META_TMP_BUCKET,
+                    "staged",
+                    &file_info,
+                    crate::disk::RUSTFS_META_BUCKET,
+                    "pool.bin",
+                    None
+                )
+                .await
+                .expect_err("old peer must reject native CAS"),
+            DiskError::UnsupportedDisk,
+        );
+        assert_eq!(
+            remote
+                .delete_version(
+                    crate::disk::RUSTFS_META_BUCKET,
+                    "pool.bin",
+                    file_info,
+                    false,
+                    DeleteOptions {
+                        undo_write: true,
+                        old_data_dir: Some(Uuid::new_v4()),
+                        ..Default::default()
+                    }
+                )
+                .await,
+            Err(DiskError::UnsupportedDisk),
+        );
+        assert_eq!(
+            remote
+                .pool_meta_native_recovery(PoolMetaNativeRecoveryRequest::Probe {})
+                .await,
+            Err(DiskError::UnsupportedDisk),
+        );
+        assert_eq!(
+            *peer.peer.unsupported_methods.lock().expect("observed methods lock"),
+            vec![
+                "/node_service.NodeService/RenameDataWithPoolMetaCas",
+                "/node_service.NodeService/UndoPoolMetaWrite",
+                "/node_service.NodeService/PoolMetaNativeRecovery"
+            ],
+            "protected mutations and capability probes must never fall back to legacy RPCs",
+        );
+        peer.stop().await;
+    }
+
     #[tokio::test]
     async fn guarded_delete_old_peer_rejects_without_legacy_fallback() {
         runtime_sources::ensure_test_rpc_secret();
@@ -5494,6 +6190,7 @@ mod tests {
     fn rename_data_response_accepts_legacy_json_without_decode_error() {
         crate::cluster::rpc::runtime_sources::reset_internode_metrics_for_test();
         let response = RenameDataResp {
+            multipart_source: None,
             old_data_dir: Some(Uuid::new_v4()),
             rollback_data_dir: Some(Uuid::new_v4()),
             cleanup_data_dir: Some(Uuid::new_v4()),
@@ -5898,6 +6595,7 @@ mod tests {
     #[test]
     fn rename_data_resp_named_msgpack_is_smaller_than_json() {
         let response = RenameDataResp {
+            multipart_source: None,
             old_data_dir: Some(Uuid::new_v4()),
             rollback_data_dir: Some(Uuid::new_v4()),
             cleanup_data_dir: Some(Uuid::new_v4()),
@@ -8050,6 +8748,89 @@ mod tests {
         peer.stop().await;
     }
 
+    #[tokio::test]
+    #[serial(remote_disk_recovery_probe)]
+    async fn endpoint_reconnect_admits_ready_drive_while_another_format_probe_is_blocked() {
+        runtime_sources::ensure_test_rpc_secret();
+        let reference = crate::layout::format::FormatV3::new(1, 2);
+        let mut peers = Vec::new();
+        let mut endpoints = Vec::new();
+        for index in 0..2 {
+            let mut format = reference.clone();
+            format.erasure.this = reference.erasure.sets[0][index];
+            let peer = TestGrpcPeer::spawn(
+                Bytes::from(format.to_json().expect("format should serialize")),
+                Bytes::from_static(b"recovered-peer-data"),
+            )
+            .await
+            .expect("authenticated regression peer is required");
+            endpoints.push(Endpoint {
+                url: url::Url::parse(&format!("{}/data/rustfs{index}", peer.addr)).expect("endpoint should parse"),
+                is_local: false,
+                pool_idx: 0,
+                set_idx: 0,
+                disk_idx: i32::try_from(index).expect("fixture index fits"),
+            });
+            peers.push(peer);
+        }
+        let set = crate::set_disk::SetDisks::new(
+            "independent-reconnect-test".to_owned(),
+            Arc::new(tokio::sync::RwLock::new(vec![None; 2])),
+            2,
+            1,
+            0,
+            0,
+            endpoints,
+            reference.clone(),
+            Vec::new(),
+        )
+        .await;
+        let gate = peers[0].peer.pause_next_format_read();
+        let pending_set = Arc::clone(&set);
+        let pending = tokio::spawn(async move { pending_set.connect_disks().await });
+        time::timeout(Duration::from_secs(10), gate.entered.notified())
+            .await
+            .expect("the first endpoint must reach the format barrier");
+
+        let ready = time::timeout(Duration::from_secs(10), async {
+            loop {
+                if let Some(disk) = set.disks.read().await[1].clone() {
+                    return disk;
+                }
+                tokio::task::yield_now().await;
+            }
+        })
+        .await;
+        if let Ok(disk) = &ready {
+            assert!(set.disks.read().await[0].is_none(), "the blocked endpoint must remain unpublished");
+            assert!(!pending.is_finished(), "the sweep still owns its blocked endpoint probe");
+            assert_eq!(
+                disk.read_all("bucket", "object")
+                    .await
+                    .expect("the ready endpoint must serve real I/O"),
+                Bytes::from_static(b"recovered-peer-data")
+            );
+            assert_eq!(
+                peers[1].peer.object_read_all_disks(),
+                vec![reference.erasure.sets[0][1].to_string()],
+                "the usable handle must carry its validated slot identity"
+            );
+        }
+        gate.release.cancel();
+        pending.await.expect("all endpoint probes must finish after release");
+        for disk in set.disks.read().await.iter().flatten() {
+            disk.close().await.expect("stop each published handle's monitoring");
+        }
+        for peer in peers {
+            peer.stop().await;
+        }
+        ready.expect("a blocked earlier endpoint must not delay a ready drive in the same set");
+        assert!(
+            set.disks.read().await.iter().all(Option::is_some),
+            "both endpoints must eventually be admitted"
+        );
+    }
+
     #[tokio::test]
     #[serial(remote_disk_recovery_probe)]
     async fn missing_read_disks_preserve_concurrently_published_handles() {

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

@@ -41,6 +41,10 @@ use tonic::service::interceptor::InterceptedService;
 use tonic::{Request, Response};
 use tracing::{debug, info, warn};
 
+const EVENT_NAMESPACE_LOCK_RELEASE: &str = "namespace_lock_release";
+const LOG_COMPONENT_ECSTORE: &str = "ecstore";
+const LOG_SUBSYSTEM_REMOTE_LOCK: &str = "remote_lock";
+
 fn attach_lock_mutation_body_digest<T: rustfs_protos::CanonicalMutationBody>(request: &mut Request<T>) -> std::io::Result<()> {
     set_tonic_rolling_mutation_body_digest(request)
 }
@@ -233,6 +237,20 @@ pub struct RemoteClient {
 impl RemoteClient {
     const ONLINE_CHECK_RESOURCE: &'static str = "health-lock-online";
 
+    fn record_release_ack(&self, lock_id: &LockId) {
+        tracing::trace!(
+            target: "rustfs_ecstore::object_lock_diag",
+            event = EVENT_NAMESPACE_LOCK_RELEASE,
+            component = LOG_COMPONENT_ECSTORE,
+            subsystem = LOG_SUBSYSTEM_REMOTE_LOCK,
+            state = "acknowledged",
+            peer_kind = "remote",
+            lock_id = %lock_id,
+            peer = %self.addr,
+            "Remote namespace lock release acknowledged"
+        );
+    }
+
     pub fn new(endpoint: String) -> Self {
         Self { addr: endpoint }
     }
@@ -1204,8 +1222,6 @@ impl LockClient for RemoteClient {
     }
 
     async fn release(&self, lock_id: &LockId) -> Result<bool> {
-        info!("remote release for {}", lock_id);
-
         let unlock_request = Self::create_unlock_request(lock_id);
         let request_string = serde_json::to_string(&unlock_request)
             .map_err(|e| LockError::internal(format!("Failed to serialize request: {e}")))?;
@@ -1226,6 +1242,9 @@ impl LockClient for RemoteClient {
         if let Some(error_info) = resp.error_info {
             return Err(LockError::internal(error_info));
         }
+        if resp.success {
+            self.record_release_ack(lock_id);
+        }
         Ok(resp.success)
     }
 
@@ -1261,7 +1280,13 @@ impl LockClient for RemoteClient {
         Ok(lock_ids
             .iter()
             .enumerate()
-            .map(|(idx, _)| resp.results.get(idx).map(|result| result.success).unwrap_or(false))
+            .map(|(idx, lock_id)| {
+                let result = resp.results.get(idx);
+                if result.is_some_and(|result| result.success && result.error_info.is_none()) {
+                    self.record_release_ack(lock_id);
+                }
+                result.map(|result| result.success).unwrap_or(false)
+            })
             .collect())
     }
 

crates/ecstore/src/core/pool_meta_native_integration_tests.rs

@@ -0,0 +1,630 @@
+// 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::*;
+use crate::disk::error::DiskError;
+use crate::storage_api_contracts::namespace::NamespaceLocking as _;
+use codec::{fixture_inline_pool_meta_replicas as replicas, fixture_split_inline_pool_meta_replicas as split};
+use std::path::PathBuf;
+use tokio::fs;
+
+struct PhysicalFixture {
+    _directories: Vec<tempfile::TempDir>,
+    store: Arc<ECStore>,
+    canonical: Vec<u8>,
+    pending: Vec<u8>,
+    identity: Vec<u8>,
+    revision: PoolMetaRevision,
+}
+
+fn object_path(disk: &DiskStore) -> PathBuf {
+    disk.get_object_path_if_local(RUSTFS_META_BUCKET, POOL_META_NAME)
+        .expect("physical fixture must use local disks")
+        .expect("pool metadata path should resolve")
+}
+
+async fn disks(store: &ECStore) -> Vec<Vec<DiskStore>> {
+    let mut all = Vec::with_capacity(store.pools.len());
+    for pool in &store.pools {
+        let set = pool.get_disks_by_key(POOL_META_NAME);
+        let disks = set
+            .disks
+            .read()
+            .await
+            .iter()
+            .cloned()
+            .map(|disk| disk.expect("physical fixture must keep every disk online"))
+            .collect::<Vec<_>>();
+        assert_eq!(disks.len(), 4, "fixture requires an EC 2+2 set");
+        all.push(disks);
+    }
+    all
+}
+
+async fn owner(store: &ECStore) -> Arc<rustfs_lock::NamespaceLockGuard> {
+    let lock = store.pools[0]
+        .new_ns_lock(RUSTFS_META_BUCKET, POOL_META_NAME)
+        .await
+        .expect("real namespace lock should be created");
+    Arc::new(
+        lock.get_write_lock(get_lock_acquire_timeout())
+            .await
+            .expect("physical recovery must hold the namespace writer"),
+    )
+}
+
+fn clean_state(fixture: &PhysicalFixture) -> PoolMetaWriteState {
+    PoolMetaWriteState::for_startup(fixture.revision.cluster_id.expect("fixture V3 revision requires identity"), false)
+}
+
+async fn snapshots(store: &ECStore) -> Vec<Vec<PoolMetaNativeInspection>> {
+    let mut all = Vec::new();
+    for pool in disks(store).await {
+        let mut current = Vec::new();
+        for disk in pool {
+            match disk
+                .pool_meta_native_recovery(Request::Inspect {
+                    object: PoolMetaNativeObject::PoolBin,
+                })
+                .await
+                .expect("physical native inspection should succeed")
+            {
+                Response::Inspected(inspection) => current.push(inspection),
+                _ => panic!("inspection must return a disk snapshot"),
+            }
+        }
+        all.push(current);
+    }
+    all
+}
+
+async fn fixture(split_pools: usize) -> PhysicalFixture {
+    let (directories, store, _other_store) =
+        crate::services::rebalance::test_three_pool_stores_with_four_disk_sets_with_isolated_node_contexts(None).await;
+    persist_v3_pool_meta_for_test(&store).await;
+    let previous = load_pool_meta_replicas(store.pools.clone(), true)
+        .await
+        .expect("baseline V3 pool metadata must be ordinary-readable");
+    let revision = PoolMetaRevision {
+        generation: previous
+            .revision
+            .generation
+            .checked_add(1)
+            .expect("test generation should increment"),
+        transaction_id: Some(uuid::Uuid::new_v4()),
+        ..previous.revision
+    };
+    let candidate = PoolMetaCommittedCandidate {
+        canonical: previous.canonical.expect("baseline must have a canonical V3 COMMIT"),
+        meta: previous.meta.clone(),
+        revision: previous.revision,
+    };
+    let canonical = encode_pool_meta_v3_envelope(&previous.meta, revision, true, None).expect("next COMMIT should encode");
+    let pending =
+        encode_pool_meta_v3_envelope(&previous.meta, revision, false, Some(&candidate)).expect("next PREPARE should encode");
+    let identity = load_pool_meta_identity(
+        store.pools.clone(),
+        revision
+            .cluster_id
+            .expect("baseline V3 metadata should have a cluster identity"),
+    )
+    .await
+    .expect("real initialized identities must agree")
+    .identity
+    .expect("real fixture identity must be present");
+    let identity = encode_pool_meta_identity(identity).expect("real fixture identity should encode canonically");
+    let modified = OffsetDateTime::now_utc();
+    let split = split(&pending, &canonical, modified).await;
+    let commit = replicas(&canonical, modified + Duration::seconds(1)).await;
+    let prepared = replicas(&pending, modified).await;
+    let directory = uuid::Uuid::new_v4();
+    for (pool_index, pool) in disks(&store).await.into_iter().enumerate() {
+        let heads = if pool_index < split_pools { &split } else { &commit };
+        for (disk_index, disk) in pool.into_iter().enumerate() {
+            let object = object_path(&disk);
+            fs::remove_dir_all(&object)
+                .await
+                .expect("fault injection should remove only its own pool.bin fixture");
+            fs::create_dir_all(object.join(directory.to_string()))
+                .await
+                .expect("fixture rollback directory should be created");
+            fs::write(object.join("xl.meta"), &heads[disk_index])
+                .await
+                .expect("physical split head should be installed");
+            fs::write(object.join(directory.to_string()).join("xl.meta.bkp"), &prepared[disk_index])
+                .await
+                .expect("physical PREPARE backup should be installed");
+        }
+    }
+    PhysicalFixture {
+        _directories: directories,
+        store,
+        canonical,
+        pending,
+        identity,
+        revision,
+    }
+}
+
+async fn assert_committed(fixture: &PhysicalFixture) {
+    let confirmed = load_pool_meta_replicas(fixture.store.pools.clone(), true)
+        .await
+        .expect("ordinary full-pool read must succeed after physical recovery");
+    assert_eq!(confirmed.canonical.as_deref(), Some(fixture.canonical.as_slice()));
+    assert_eq!(confirmed.revision, fixture.revision);
+    assert!(!confirmed.replica_state.needs_repair);
+    assert!(confirmed.replica_state.repair_write_safe);
+    for pool in snapshots(&fixture.store).await {
+        let raw = pool
+            .iter()
+            .map(|disk| disk.current.clone().expect("every recovered physical head must exist"))
+            .collect::<Vec<_>>();
+        assert_eq!(
+            codec::decode_inline_pool_meta_replicas(&raw)
+                .await
+                .expect("every recovered shard must verify"),
+            vec![Bytes::copy_from_slice(&fixture.canonical)]
+        );
+    }
+}
+
+#[tokio::test]
+#[serial_test::serial(pool_meta_version_env)]
+async fn pool_meta_native_physical_split_repair_completes_normal_cas_and_quorum_readback() {
+    let fixture = fixture(1).await;
+    let before = read_pool_meta_replicas(fixture.store.pools.clone(), true).await;
+    assert!(matches!(before[0].replica, PoolMetaReplica::Unreadable(_)));
+    assert!(matches!(before[1].replica, PoolMetaReplica::Valid { committed: true, .. }));
+    let mut state = clean_state(&fixture);
+    let fence = PoolMetaPersistenceFence::Owned(owner(&fixture.store).await);
+    assert!(
+        repair(&fixture.store, &mut state, &fence)
+            .await
+            .expect("healthy ordinary COMMIT must authorize physical repair")
+    );
+    state
+        .ensure_write_safe("physical recovery readback")
+        .expect("completed recovery must disarm its transaction");
+    assert_committed(&fixture).await;
+}
+
+#[tokio::test]
+#[serial_test::serial(pool_meta_version_env)]
+async fn pool_meta_native_physical_split_without_ordinary_commit_performs_no_mutation() {
+    let fixture = fixture(3).await;
+    let before = snapshots(&fixture.store).await;
+    let mut state = clean_state(&fixture);
+    let fence = PoolMetaPersistenceFence::Owned(owner(&fixture.store).await);
+    assert!(
+        repair(&fixture.store, &mut state, &fence).await.is_err(),
+        "reconstruction alone must never authorize a COMMIT"
+    );
+    assert_eq!(
+        snapshots(&fixture.store).await,
+        before,
+        "all current heads, identities and complete backups must remain unchanged"
+    );
+    assert_eq!(state.active_transactions.load(Ordering::SeqCst), 0);
+}
+
+struct IdentityEpochFork {
+    path: PathBuf,
+    original: Bytes,
+    replacement: Bytes,
+    original_identity: PersistedPoolMetaIdentity,
+}
+
+async fn identity_epoch_fork(fixture: &PhysicalFixture) -> IdentityEpochFork {
+    let identity_set = fixture.store.pools[0].get_disks_by_key(POOL_META_IDENTITY_NAME);
+    assert!(
+        !Arc::ptr_eq(&identity_set, &fixture.store.pools[0].get_disks_by_key(POOL_META_NAME)),
+        "the identity evidence must come from its independently placed physical set"
+    );
+    let identity_disks = identity_set
+        .disks
+        .read()
+        .await
+        .iter()
+        .cloned()
+        .map(|disk| disk.expect("actual identity set must retain all four disks"))
+        .collect::<Vec<_>>();
+    let mut raw = Vec::with_capacity(identity_disks.len());
+    for disk in &identity_disks {
+        match disk
+            .pool_meta_native_recovery(Request::InspectIdentity {})
+            .await
+            .expect("actual identity set must support native inspection")
+        {
+            Response::IdentityInspected { disk_id, raw: current } => {
+                assert!(!disk_id.is_nil());
+                raw.push(current);
+            }
+            _ => panic!("identity inspection must return its physical raw shard"),
+        }
+    }
+    assert_eq!(raw.len(), 4);
+    let original_identity = match decode_pool_meta_identity(&fixture.identity) {
+        PoolMetaIdentityReplica::Valid(identity) => identity,
+        _ => panic!("fixture must provide a valid initialized identity"),
+    };
+    let forked_identity = PersistedPoolMetaIdentity {
+        epoch: original_identity
+            .epoch
+            .checked_add(1)
+            .expect("test identity epoch should increment"),
+        ..original_identity
+    };
+    let forked = encode_pool_meta_identity(forked_identity).expect("foreign epoch identity should remain validly encoded");
+    assert!(matches!(decode_pool_meta_identity(&forked), PoolMetaIdentityReplica::Valid(_)));
+    let modified = raw
+        .iter()
+        .map(|raw| {
+            codec::parse_inline_info(raw)
+                .expect("actual identity shard layout should parse")
+                .mod_time
+        })
+        .collect::<Option<Vec<_>>>()
+        .expect("every actual identity shard should have a modification time")
+        .into_iter()
+        .max()
+        .expect("actual identity set should contain four shards")
+        + Duration::nanoseconds(1);
+    let replacements = codec::encode_inline_pool_meta_replicas(&raw, &forked, modified)
+        .await
+        .expect("epoch fork must retain the actual set's valid erasure layout");
+    assert_eq!(
+        codec::parse_inline_info(&raw[0]).expect("original layout").erasure,
+        codec::parse_inline_info(&replacements[0])
+            .expect("replacement layout")
+            .erasure
+    );
+    codec::validate_inline_pool_meta_shard(&replacements[0], 0, &forked)
+        .await
+        .expect("forked shard must pass the physical integrity checks");
+    let identity_path = identity_disks[0]
+        .get_object_path_if_local(RUSTFS_META_BUCKET, POOL_META_IDENTITY_NAME)
+        .expect("identity fixture must use local storage")
+        .expect("actual identity metadata path should resolve");
+    IdentityEpochFork {
+        path: identity_path.join("xl.meta"),
+        original: raw[0].clone(),
+        replacement: replacements[0].clone(),
+        original_identity,
+    }
+}
+
+#[tokio::test]
+#[serial_test::serial(pool_meta_version_env)]
+async fn pool_meta_native_physical_identity_actual_set_minority_epoch_fork_performs_no_apply() {
+    let fixture = fixture(1).await;
+    let fork = identity_epoch_fork(&fixture).await;
+    let heads = |pools: Vec<Vec<PoolMetaNativeInspection>>| {
+        pools
+            .into_iter()
+            .map(|pool| pool.into_iter().map(|disk| (disk.current, disk.backups)).collect::<Vec<_>>())
+            .collect::<Vec<_>>()
+    };
+    let before = heads(snapshots(&fixture.store).await);
+    fs::write(&fork.path, &fork.replacement)
+        .await
+        .expect("only one actual identity-set disk should receive the epoch fork");
+    let selected = load_pool_meta_identity(fixture.store.pools.clone(), fork.original_identity.cluster_id)
+        .await
+        .expect("three matching identity shards must remain ordinary-readable");
+    assert_eq!(selected.identity, Some(fork.original_identity));
+    let reads = read_pool_meta_replicas(fixture.store.pools.clone(), true).await;
+    assert!(matches!(reads[0].replica, PoolMetaReplica::Unreadable(_)));
+    assert!(matches!(reads[1].replica, PoolMetaReplica::Valid { committed: true, .. }));
+    let mut state = clean_state(&fixture);
+    let fence = PoolMetaPersistenceFence::Owned(owner(&fixture.store).await);
+    let error = repair(&fixture.store, &mut state, &fence)
+        .await
+        .expect_err("ordinary identity quorum must not hide a physical minority epoch fork");
+    assert_eq!(
+        error
+            .pool_metadata_failure()
+            .expect("known identity fork must retain typed failure")
+            .kind,
+        crate::error::PoolMetadataFailure::RecoveryRequired
+    );
+    assert!(state.write_blocked, "known minority fork must latch the write gate");
+    assert_eq!(
+        heads(snapshots(&fixture.store).await),
+        before,
+        "identity rejection must precede every pool.bin Apply"
+    );
+    assert_eq!(state.active_transactions.load(Ordering::SeqCst), 0);
+    fs::write(&fork.path, &fork.original)
+        .await
+        .expect("restoring the input identity must not clear the observed fork latch");
+    let error = repair(&fixture.store, &mut state, &fence)
+        .await
+        .expect_err("the same state must retain operator recovery after identity becomes consistent");
+    assert_eq!(
+        error
+            .pool_metadata_failure()
+            .expect("sticky failure must retain its classification")
+            .kind,
+        crate::error::PoolMetadataFailure::RecoveryRequired
+    );
+    assert!(state.write_blocked);
+    assert_eq!(heads(snapshots(&fixture.store).await), before);
+}
+
+#[tokio::test]
+#[serial_test::serial(pool_meta_version_env)]
+async fn pool_meta_native_physical_runtime_fork_after_apply_latches_original_gate() {
+    let fixture = fixture(1).await;
+    let fork = identity_epoch_fork(&fixture).await;
+    let since = OffsetDateTime::from_unix_timestamp(1_700_000_000).expect("fixed original failure timestamp should be valid");
+    {
+        let mut state = fixture.store.pool_meta_save_gate.lock().await;
+        assert_eq!(state.active_transactions.load(Ordering::SeqCst), 0);
+        state.write_blocked = false;
+        state.block_context = None;
+        state.aborted_transaction.store(true, Ordering::SeqCst);
+        let mut failure = pool_metadata_error(crate::error::PoolMetadataFailure::TransactionUnknown, "commit_cas", None);
+        failure.since = since;
+        *state
+            .transaction_failure
+            .lock()
+            .expect("original transaction failure mutex should lock") = Some(failure);
+        *state
+            .block_started_at
+            .lock()
+            .expect("original block timestamp mutex should lock") = Some(since);
+    }
+    let pause = Arc::new(NativeApplyPause {
+        reached: tokio::sync::Notify::new(),
+        resume: tokio::sync::Notify::new(),
+    });
+    let reached = pause.reached.notified();
+    tokio::pin!(reached);
+    reached.as_mut().enable();
+    let recovery = tokio::spawn({
+        let store = Arc::clone(&fixture.store);
+        let pause = Arc::clone(&pause);
+        async move {
+            AFTER_NATIVE_APPLY
+                .scope(pause, async move { store.recover_pool_meta_transaction().await })
+                .await
+        }
+    });
+    tokio::time::timeout(std::time::Duration::from_secs(10), reached)
+        .await
+        .expect("real runtime recovery should pause after completing physical Apply");
+    let readable = load_pool_meta_replicas(fixture.store.pools.clone(), true)
+        .await
+        .expect("physical Apply must already make every pool ordinary-readable");
+    assert_eq!(readable.canonical.as_deref(), Some(fixture.canonical.as_slice()));
+    fs::write(&fork.path, &fork.replacement)
+        .await
+        .expect("a valid minority identity fork should arrive after physical repair");
+    pause.resume.notify_one();
+    let error = tokio::time::timeout(std::time::Duration::from_secs(10), recovery)
+        .await
+        .expect("resumed runtime recovery should finish its validation")
+        .expect("runtime recovery task should not panic")
+        .expect_err("post-Apply identity fork must reject runtime recovery");
+    assert_eq!(
+        error
+            .pool_metadata_failure()
+            .expect("runtime fork failure must remain typed")
+            .kind,
+        crate::error::PoolMetadataFailure::RecoveryRequired
+    );
+    assert_committed(&fixture).await;
+    {
+        let state = fixture.store.pool_meta_save_gate.lock().await;
+        assert!(
+            state.write_blocked,
+            "candidate validation must propagate its permanent latch to the original state"
+        );
+        let context = state
+            .block_context
+            .as_ref()
+            .expect("original state must retain the known fork context");
+        assert_eq!(context.kind, crate::error::PoolMetadataFailure::RecoveryRequired);
+        assert_eq!(context.since, since, "candidate failure must preserve the original outage timestamp");
+        assert_eq!(*state.block_started_at.lock().expect("original block timestamp should lock"), Some(since));
+    }
+    assert!(
+        !fixture
+            .store
+            .recover_pool_meta_transaction()
+            .await
+            .expect("permanently blocked recovery should return without mutation"),
+        "ordinary-readable repaired heads must not clear the original known-fork latch"
+    );
+    let state = fixture.store.pool_meta_save_gate.lock().await;
+    assert!(state.write_blocked);
+    assert_eq!(
+        state
+            .block_context
+            .as_ref()
+            .expect("sticky fork context must remain present")
+            .since,
+        since
+    );
+}
+
+#[tokio::test]
+#[serial_test::serial(pool_meta_version_env)]
+async fn pool_meta_native_physical_partial_apply_resumes_after_conditional_rejection() {
+    let fixture = fixture(1).await;
+    let fence = PoolMetaPersistenceFence::Owned(owner(&fixture.store).await);
+    let inspected = inspect(&fixture.store.pools, &fence)
+        .await
+        .expect("real native snapshots should be obtained");
+    let replacements = plan(&inspected, &fixture.canonical, fixture.revision, &fixture.identity)
+        .await
+        .expect("physical recovery plan should validate");
+    let snapshot = &inspected[0].snapshots[0];
+    assert_eq!(
+        inspected[0].disks[0]
+            .pool_meta_native_recovery(Request::Apply {
+                object: PoolMetaNativeObject::PoolBin,
+                expected_disk_id: snapshot.disk_id,
+                expected_current_sha256: snapshot.current.as_ref().map(|raw| Sha256::digest(raw).into()),
+                expected_backup_inventory_sha256: snapshot.inventory_sha256().expect("snapshot inventory should hash"),
+                replacement: replacements[0][0].clone(),
+            })
+            .await
+            .expect("first real Apply must atomically publish and sync"),
+        Response::Applied
+    );
+    let before_rejected = snapshots(&fixture.store).await;
+    let snapshot = &inspected[0].snapshots[1];
+    let mut stale_inventory = snapshot.inventory_sha256().expect("snapshot inventory should hash");
+    stale_inventory[0] ^= 1;
+    assert!(matches!(
+        inspected[0].disks[1]
+            .pool_meta_native_recovery(Request::Apply {
+                object: PoolMetaNativeObject::PoolBin,
+                expected_disk_id: snapshot.disk_id,
+                expected_current_sha256: snapshot.current.as_ref().map(|raw| Sha256::digest(raw).into()),
+                expected_backup_inventory_sha256: stale_inventory,
+                replacement: replacements[0][1].clone(),
+            })
+            .await,
+        Err(DiskError::ConditionalDeleteConflict)
+    ));
+    assert_eq!(
+        snapshots(&fixture.store).await,
+        before_rejected,
+        "rejected Apply must preserve the partial repair evidence"
+    );
+    drop(inspected);
+    drop(fence);
+    let fence = PoolMetaPersistenceFence::Owned(owner(&fixture.store).await);
+    let mut state = clean_state(&fixture);
+    assert!(
+        repair(&fixture.store, &mut state, &fence)
+            .await
+            .expect("a fresh owner should resume the partial physical repair")
+    );
+    assert_committed(&fixture).await;
+}
+
+#[tokio::test]
+#[serial_test::serial(pool_meta_version_env)]
+async fn pool_meta_native_physical_noninline_conditional_io_round_trip() {
+    let fixture = fixture(0).await;
+    let pool = Arc::clone(&fixture.store.pools[0]);
+    let fence = PoolMetaPersistenceFence::Owned(owner(&fixture.store).await);
+    let (original, info) = read_config_no_lock_preserve_empty_with_metadata(Arc::clone(&pool), POOL_META_NAME)
+        .await
+        .expect("baseline conditional object must be readable");
+    let token = pool_meta_cas_token_from_info(info);
+    // This smoke test targets the physical conditional I/O boundary. The large
+    // opaque body deliberately exceeds the inline writer's one-block limit.
+    let payload = vec![0x5a; rustfs_filemeta::BLOCK_SIZE_V2 + 137];
+    let state = clean_state(&fixture);
+    let mut arm = state.arm_transaction();
+    save_pool_meta_object_cas(
+        Arc::clone(&pool),
+        POOL_META_NAME,
+        payload.clone(),
+        &token,
+        &fence,
+        "noninline_smoke",
+        &mut arm,
+    )
+    .await
+    .expect("normal native CAS must support noninline pool metadata storage");
+    let (readback, info) = read_config_no_lock_preserve_empty_with_metadata(Arc::clone(&pool), POOL_META_NAME)
+        .await
+        .expect("noninline conditional write must be ordinary-readable");
+    assert_eq!(readback, payload);
+    for disk in &snapshots(&fixture.store).await[0] {
+        let raw = disk
+            .current
+            .as_ref()
+            .expect("noninline head should be installed on every disk");
+        let info = rustfs_filemeta::get_file_info(
+            raw,
+            RUSTFS_META_BUCKET,
+            POOL_META_NAME,
+            "",
+            rustfs_filemeta::FileInfoOpts {
+                data: true,
+                include_free_versions: false,
+                include_part_checksums: true,
+            },
+        )
+        .expect("physical noninline metadata should decode");
+        assert!(!info.inline_data());
+        assert!(info.data.is_none());
+    }
+    save_pool_meta_object_cas(
+        pool,
+        POOL_META_NAME,
+        original.clone(),
+        &pool_meta_cas_token_from_info(info),
+        &fence,
+        "noninline_to_inline_smoke",
+        &mut arm,
+    )
+    .await
+    .expect("normal CAS must also replace a noninline parent with inline metadata");
+    arm.disarm();
+    assert_committed(&fixture).await;
+}
+
+#[tokio::test]
+#[serial_test::serial(pool_meta_version_env)]
+async fn pool_meta_native_physical_three_plus_one_checkpoint_converges_all_shards() {
+    let fixture = fixture(0).await;
+    let pending = replicas(&fixture.pending, OffsetDateTime::now_utc() - Duration::seconds(1)).await;
+    let all_disks = disks(&fixture.store).await;
+    fs::write(object_path(&all_disks[0][0]).join("xl.meta"), &pending[0])
+        .await
+        .expect("one stale physical PREPARE head should be installed");
+    let selection = load_pool_meta_replicas(fixture.store.pools.clone(), true)
+        .await
+        .expect("three matching COMMIT shards must remain ordinary-readable");
+    assert_eq!(selection.canonical.as_deref(), Some(fixture.canonical.as_slice()));
+    let guard = owner(&fixture.store).await;
+    let fence = PoolMetaPersistenceFence::Owned(Arc::clone(&guard));
+    let mut state = clean_state(&fixture);
+    assert!(
+        !repair(&fixture.store, &mut state, &fence)
+            .await
+            .expect("ordinary-readable pools should use normal checkpoint recovery")
+    );
+    selection
+        .meta
+        .save_for_startup_guarded(fixture.store.pools.clone(), &mut state, guard)
+        .await
+        .expect("normal checkpoint must converge a readable 3+1 pool without an unsafe native overwrite");
+    let checkpoint = load_pool_meta_replicas(fixture.store.pools.clone(), true)
+        .await
+        .expect("checkpoint must remain ordinary-readable across all pools");
+    assert!(!checkpoint.replica_state.needs_repair);
+    assert!(checkpoint.replica_state.repair_write_safe);
+    assert!(checkpoint.revision.generation > fixture.revision.generation);
+    let canonical = checkpoint.canonical.expect("checkpoint should expose a canonical COMMIT");
+    for pool in snapshots(&fixture.store).await {
+        let raw = pool
+            .into_iter()
+            .map(|disk| disk.current.expect("checkpoint head must exist"))
+            .collect::<Vec<_>>();
+        assert_eq!(
+            codec::decode_inline_pool_meta_replicas(&raw)
+                .await
+                .expect("all four checkpoint shards must verify"),
+            vec![Bytes::copy_from_slice(&canonical)]
+        );
+    }
+}

crates/ecstore/src/core/pool_meta_native_recovery.rs

@@ -0,0 +1,736 @@
+// 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::pool_meta_recovery_codec as codec;
+use super::*;
+use crate::disk::{
+    DiskAPI, DiskStore, PoolMetaNativeInspection, PoolMetaNativeObject, PoolMetaNativeRecoveryRequest as Request,
+    PoolMetaNativeRecoveryResponse as Response,
+};
+use bytes::Bytes;
+
+#[cfg(test)]
+#[derive(Default)]
+pub(super) struct NativeApplyPause {
+    pub(super) reached: tokio::sync::Notify,
+    pub(super) resume: tokio::sync::Notify,
+}
+
+#[cfg(test)]
+tokio::task_local! {
+    pub(super) static AFTER_NATIVE_APPLY: Arc<NativeApplyPause>;
+}
+
+struct InspectedPool {
+    disks: Vec<DiskStore>,
+    snapshots: Vec<PoolMetaNativeInspection>,
+}
+
+fn invalid() -> Error {
+    Error::other("native pool metadata recovery requires one proven V3 transaction and complete matching disk evidence")
+}
+
+fn known_validation(source: Error, phase: &'static str) -> Error {
+    Error::other(pool_metadata_error(
+        crate::error::PoolMetadataFailure::RecoveryRequired,
+        phase,
+        Some(Arc::new(source)),
+    ))
+}
+
+fn codec_validation(source: codec::PoolMetaCodecError, phase: &'static str) -> Error {
+    // A supported transport may carry a valid layout outside this narrow
+    // recovery codec. That is not evidence of corruption or a transaction fork.
+    if matches!(source, codec::PoolMetaCodecError::UnsupportedLayout) {
+        return Error::other(source);
+    }
+    known_validation(Error::other(source), phase)
+}
+
+fn latch_validation(state: &mut PoolMetaWriteState, error: Error) -> Error {
+    if let Some(context) = error.pool_metadata_failure()
+        && context.kind == crate::error::PoolMetadataFailure::RecoveryRequired
+    {
+        state.block_with_context(context.clone());
+    }
+    error
+}
+
+fn current(pool: &InspectedPool) -> Result<Vec<Bytes>> {
+    pool.snapshots
+        .iter()
+        .map(|snapshot| snapshot.current.clone().ok_or_else(invalid))
+        .collect()
+}
+
+async fn inspect(pools: &[Arc<Sets>], fence: &PoolMetaPersistenceFence<'_>) -> Result<Vec<InspectedPool>> {
+    let mut inspected = Vec::with_capacity(pools.len());
+    // Pool -> disk order is stable; all capability probes precede every mutation.
+    for pool in pools {
+        fence.ensure_held()?;
+        let set = pool.get_disks_by_key(POOL_META_NAME);
+        let disks = set
+            .disks
+            .read()
+            .await
+            .iter()
+            .cloned()
+            .map(|disk| disk.ok_or_else(invalid))
+            .collect::<Result<Vec<_>>>()?;
+        let probes = futures::stream::iter(disks.iter().cloned().map(|disk| async move {
+            match disk.pool_meta_native_recovery(Request::Probe {}).await? {
+                Response::Probed => Ok(()),
+                _ => Err(invalid()),
+            }
+        }))
+        .buffered(4)
+        .collect::<Vec<Result<()>>>()
+        .await;
+        for result in probes {
+            result?;
+        }
+        inspected.push(InspectedPool {
+            disks,
+            snapshots: Vec::new(),
+        });
+    }
+    for pool in &mut inspected {
+        fence.ensure_held()?;
+        let results = futures::stream::iter(pool.disks.iter().cloned().map(|disk| async move {
+            match disk
+                .pool_meta_native_recovery(Request::Inspect {
+                    object: PoolMetaNativeObject::PoolBin,
+                })
+                .await?
+            {
+                Response::Inspected(snapshot) if !snapshot.disk_id.is_nil() => Ok(snapshot),
+                _ => Err(invalid()),
+            }
+        }))
+        .buffered(4)
+        .collect::<Vec<Result<PoolMetaNativeInspection>>>()
+        .await;
+        for result in results {
+            pool.snapshots.push(result?);
+        }
+    }
+    fence.ensure_held()?;
+    Ok(inspected)
+}
+
+async fn validate_identities(pools: &[Arc<Sets>], canonical: &[u8], fence: &PoolMetaPersistenceFence<'_>) -> Result<()> {
+    for pool in pools {
+        fence.ensure_held()?;
+        // Identity and pool.bin have independent placement hashes. Read every
+        // member of the identity's actual set instead of assuming co-location.
+        let set = pool.get_disks_by_key(POOL_META_IDENTITY_NAME);
+        let disks = set
+            .disks
+            .read()
+            .await
+            .iter()
+            .cloned()
+            .map(|disk| disk.ok_or_else(invalid))
+            .collect::<Result<Vec<_>>>()?;
+        let results = futures::stream::iter(disks.into_iter().enumerate().map(|(index, disk)| async move {
+            match disk.pool_meta_native_recovery(Request::InspectIdentity {}).await? {
+                Response::IdentityInspected { disk_id, raw } if !disk_id.is_nil() => {
+                    codec::validate_inline_pool_meta_shard(&raw, index, canonical)
+                        .await
+                        .map_err(|error| codec_validation(error, "native_identity_validation"))
+                }
+                _ => Err(invalid()),
+            }
+        }))
+        .buffered(4)
+        .collect::<Vec<Result<()>>>()
+        .await;
+        for result in results {
+            result?;
+        }
+    }
+    fence.ensure_held()
+}
+
+async fn plan(
+    inspected: &[InspectedPool],
+    canonical: &[u8],
+    revision: PoolMetaRevision,
+    identity: &[u8],
+) -> Result<Vec<Vec<Bytes>>> {
+    let mut pending = Vec::<Vec<u8>>::new();
+    let mut predecessors = Vec::<Vec<u8>>::new();
+    let mut decoded = Vec::<Vec<u8>>::new();
+    let mut newest = OffsetDateTime::now_utc();
+    for pool in inspected {
+        let current = current(pool)?;
+        for (index, snapshot) in pool.snapshots.iter().enumerate() {
+            if let Some(local_identity) = snapshot.identity.as_ref() {
+                codec::validate_inline_pool_meta_shard(local_identity, index, identity)
+                    .await
+                    .map_err(|error| codec_validation(error, "native_metadata_validation"))?;
+            }
+            for raw in std::iter::once(current[index].clone()).chain(snapshot.backups.iter().map(|backup| backup.data.clone())) {
+                let info =
+                    codec::parse_inline_info(&raw).map_err(|error| codec_validation(error, "native_metadata_validation"))?;
+                newest = newest.max(info.mod_time.ok_or_else(invalid)?);
+            }
+        }
+        let mut inventories = vec![current.clone()];
+        let directories = pool
+            .snapshots
+            .iter()
+            .flat_map(|snapshot| snapshot.backups.iter().map(|backup| backup.directory))
+            .collect::<HashSet<_>>();
+        for directory in directories {
+            // A missing backup contributes its actual current shard only to
+            // integrity reconstruction. It never invents backup evidence.
+            inventories.push(
+                pool.snapshots
+                    .iter()
+                    .enumerate()
+                    .map(|(index, snapshot)| {
+                        snapshot
+                            .backups
+                            .iter()
+                            .find(|backup| backup.directory == directory)
+                            .map(|backup| backup.data.clone())
+                            .unwrap_or_else(|| current[index].clone())
+                    })
+                    .collect(),
+            );
+        }
+        for raw in inventories {
+            for generation in codec::decode_inline_pool_meta_replicas(&raw)
+                .await
+                .map_err(|error| codec_validation(error, "native_metadata_validation"))?
+            {
+                let payload = generation.to_vec();
+                if !decoded.contains(&payload) {
+                    decoded.push(payload);
+                }
+            }
+        }
+    }
+    let mut obsolete = Vec::<Vec<u8>>::new();
+    for payload in &decoded {
+        match decode_pool_meta_replica(payload.clone()) {
+            PoolMetaReplica::Valid {
+                canonical: candidate,
+                revision: candidate_revision,
+                committed: false,
+                previous: Some(previous),
+                ..
+            } if candidate_revision == revision && candidate == canonical => {
+                if !pending.contains(payload) {
+                    pending.push(payload.clone());
+                }
+                if !predecessors.contains(&previous.canonical) {
+                    predecessors.push(previous.canonical);
+                }
+            }
+            PoolMetaReplica::Valid { revision: older, .. }
+                if older.is_generation_protocol()
+                    && older.cluster_id == revision.cluster_id
+                    && older.epoch == revision.epoch
+                    && older.generation < revision.generation =>
+            {
+                // Retained evidence from an earlier successful recovery is
+                // obsolete, not a competing commit authority. It remains
+                // fully reconstructed and bound into the native CAS.
+                obsolete.push(payload.clone());
+            }
+            PoolMetaReplica::Valid {
+                canonical: candidate,
+                revision: candidate_revision,
+                committed: true,
+                ..
+            } if candidate_revision == revision && candidate == canonical => {}
+            PoolMetaReplica::Valid { revision: conflict, .. } if conflict.is_generation_protocol() => {
+                return Err(known_validation(
+                    Error::other("native V3 transaction conflicts with committed authority"),
+                    "native_transaction_validation",
+                ));
+            }
+            PoolMetaReplica::Corrupt(_) | PoolMetaReplica::Incompatible(_) => {
+                return Err(known_validation(
+                    Error::other("native V3 payload is corrupt or incompatible"),
+                    "native_transaction_validation",
+                ));
+            }
+            _ => {}
+        }
+    }
+    if pending.len() > 1 {
+        return Err(known_validation(
+            Error::other("native V3 prepare records disagree on the predecessor"),
+            "native_transaction_validation",
+        ));
+    }
+    if pending.is_empty() {
+        return Err(invalid());
+    }
+    for payload in &decoded {
+        if payload.as_slice() != canonical
+            && !pending.contains(payload)
+            && !predecessors.contains(payload)
+            && !obsolete.contains(payload)
+        {
+            return Err(invalid());
+        }
+    }
+    for pool in inspected {
+        for (index, snapshot) in pool.snapshots.iter().enumerate() {
+            let raw = snapshot.current.as_ref().ok_or_else(invalid)?;
+            let mut matched = false;
+            for payload in std::iter::once(canonical).chain(pending.iter().map(Vec::as_slice)) {
+                matched |= codec::validate_inline_pool_meta_shard(raw, index, payload).await.is_ok();
+            }
+            if !matched {
+                return Err(invalid());
+            }
+            for backup in &snapshot.backups {
+                let mut matched = false;
+                for payload in std::iter::once(canonical).chain(
+                    pending
+                        .iter()
+                        .chain(predecessors.iter())
+                        .chain(obsolete.iter())
+                        .map(Vec::as_slice),
+                ) {
+                    matched |= codec::validate_inline_pool_meta_shard(&backup.data, index, payload)
+                        .await
+                        .is_ok();
+                }
+                if !matched {
+                    return Err(invalid());
+                }
+            }
+        }
+    }
+    let modified = newest.checked_add(Duration::nanoseconds(1)).ok_or_else(invalid)?;
+    let mut replacements = Vec::with_capacity(inspected.len());
+    for pool in inspected {
+        replacements.push(
+            codec::encode_inline_pool_meta_replicas(&current(pool)?, canonical, modified)
+                .await
+                .map_err(|error| codec_validation(error, "native_metadata_validation"))?,
+        );
+    }
+    Ok(replacements)
+}
+
+pub(super) async fn repair(
+    store: &ECStore,
+    state: &mut PoolMetaWriteState,
+    fence: &PoolMetaPersistenceFence<'_>,
+) -> Result<bool> {
+    fence.ensure_held()?;
+    let reads = read_pool_meta_replicas(store.pools.clone(), true).await;
+    if !reads
+        .iter()
+        .any(|read| matches!(read.replica, PoolMetaReplica::Unreadable(_)))
+    {
+        return Ok(false);
+    }
+    state.ensure_write_safe("native pool metadata recovery")?;
+    if reads
+        .iter()
+        .any(|read| matches!(read.replica, PoolMetaReplica::Corrupt(_) | PoolMetaReplica::Incompatible(_)))
+    {
+        return Err(latch_validation(
+            state,
+            known_validation(Error::other("pool metadata is corrupt or incompatible"), "native_metadata_validation"),
+        ));
+    }
+    if reads
+        .iter()
+        .any(|read| !matches!(read.replica, PoolMetaReplica::Valid { .. } | PoolMetaReplica::Unreadable(_)))
+    {
+        return Err(invalid());
+    }
+    // Only an ordinary, successfully read COMMIT grants authority. Rebuilding
+    // two erasure shards below proves integrity, never a commit decision.
+    let ordinary_commits = reads
+        .iter()
+        .filter_map(|read| match &read.replica {
+            PoolMetaReplica::Valid {
+                raw,
+                canonical,
+                revision,
+                committed: true,
+                ..
+            } if revision.is_generation_protocol() && raw == canonical => Some((canonical.clone(), *revision)),
+            _ => None,
+        })
+        .collect::<Vec<_>>();
+    if ordinary_commits.is_empty() {
+        return Err(invalid());
+    }
+    let selection = select_pool_meta_replica_reads(reads)
+        .map_err(|error| latch_validation(state, known_validation(error, "native_metadata_validation")))?;
+    state
+        .validate_selection(&selection)
+        .map_err(|error| latch_validation(state, known_validation(error, "native_metadata_validation")))?;
+    if selection.meta.pools.len() != store.pools.len()
+        || selection
+            .meta
+            .pools
+            .iter()
+            .zip(&store.pools)
+            .enumerate()
+            .any(|(index, (meta, pool))| meta.id != index || meta.cmd_line != pool.endpoints.cmd_line)
+    {
+        return Err(invalid());
+    }
+    let canonical = selection.canonical.as_ref().ok_or_else(invalid)?;
+    if !selection.revision.is_generation_protocol()
+        || !ordinary_commits
+            .iter()
+            .any(|(raw, revision)| raw == canonical && *revision == selection.revision)
+    {
+        return Err(invalid());
+    }
+    let cluster_id = selection.revision.cluster_id.ok_or_else(invalid)?;
+    if state.expected_cluster_id != Some(cluster_id) {
+        return Err(invalid());
+    }
+    let identities = join_all(store.pools.iter().cloned().map(read_pool_meta_identity_replica)).await;
+    if identities.iter().any(|read| {
+        matches!(
+            read.replica,
+            PoolMetaIdentityReplica::Corrupt(_) | PoolMetaIdentityReplica::Incompatible(_)
+        )
+    }) {
+        return Err(latch_validation(
+            state,
+            known_validation(Error::other("pool identity is corrupt or incompatible"), "native_identity_validation"),
+        ));
+    }
+    if identities
+        .iter()
+        .any(|read| !matches!(read.replica, PoolMetaIdentityReplica::Valid(_)))
+    {
+        return Err(invalid());
+    }
+    let identity = select_pool_meta_identity(identities, cluster_id)
+        .map_err(|error| latch_validation(state, known_validation(error, "native_identity_validation")))?;
+    state
+        .validate_identity(&identity)
+        .map_err(|error| latch_validation(state, known_validation(error, "native_identity_validation")))?;
+    let identity = identity.identity.ok_or_else(invalid)?;
+    if !identity.initialized {
+        return Err(invalid());
+    }
+    if identity.epoch != selection.revision.epoch {
+        return Err(latch_validation(
+            state,
+            known_validation(Error::other("pool identity epoch conflicts with metadata"), "native_identity_validation"),
+        ));
+    }
+    let identity_bytes = encode_pool_meta_identity(identity)?;
+    let repair_started = std::time::Instant::now();
+    validate_identities(&store.pools, &identity_bytes, fence)
+        .await
+        .map_err(|error| latch_validation(state, error))?;
+    let inspected = inspect(&store.pools, fence).await?;
+    let replacements = plan(&inspected, canonical, selection.revision, &identity_bytes)
+        .await
+        .map_err(|error| latch_validation(state, error))?;
+    fence.ensure_held()?;
+    let mut arm = state.arm_transaction();
+    arm.phase = Some("native_repair_apply");
+    for (pool, replacements) in inspected.iter().zip(replacements) {
+        for ((disk, snapshot), replacement) in pool.disks.iter().zip(&pool.snapshots).zip(replacements) {
+            fence.ensure_held()?;
+            let request = Request::Apply {
+                object: PoolMetaNativeObject::PoolBin,
+                expected_disk_id: snapshot.disk_id,
+                expected_current_sha256: snapshot.current.as_ref().map(|raw| Sha256::digest(raw).into()),
+                expected_backup_inventory_sha256: snapshot.inventory_sha256()?,
+                replacement,
+            };
+            match disk.pool_meta_native_recovery(request).await? {
+                Response::Applied => {}
+                _ => return Err(invalid()),
+            }
+        }
+    }
+    #[cfg(test)]
+    if let Ok(pause) = AFTER_NATIVE_APPLY.try_with(Arc::clone) {
+        pause.reached.notify_one();
+        pause.resume.notified().await;
+    }
+    fence.ensure_held()?;
+    let confirmed = load_pool_meta_replicas(store.pools.clone(), true).await?;
+    if confirmed.canonical.as_ref() != Some(canonical)
+        || confirmed.revision != selection.revision
+        || !confirmed.replica_state.repair_write_safe
+        || confirmed.replica_state.needs_repair
+    {
+        return Err(invalid());
+    }
+    // Re-enter the normal V3 conditional-write path, including native CAS,
+    // full tail completion, and the final ordinary quorum readback.
+    for (pool, token) in store.pools.iter().cloned().zip(&confirmed.cas_tokens) {
+        save_pool_meta_object_cas(pool, POOL_META_NAME, canonical.clone(), token, fence, "native_repair_confirm", &mut arm)
+            .await?;
+    }
+    let confirmed = load_pool_meta_replicas(store.pools.clone(), true).await?;
+    if confirmed.canonical.as_ref() != Some(canonical)
+        || confirmed.revision != selection.revision
+        || confirmed.replica_state.needs_repair
+        || !confirmed.replica_state.repair_write_safe
+    {
+        return Err(invalid());
+    }
+    validate_identities(&store.pools, &identity_bytes, fence)
+        .await
+        .map_err(|error| latch_validation(state, error))?;
+    let identities = join_all(store.pools.iter().cloned().map(read_pool_meta_identity_replica)).await;
+    if identities.iter().any(|read| {
+        matches!(
+            read.replica,
+            PoolMetaIdentityReplica::Corrupt(_) | PoolMetaIdentityReplica::Incompatible(_)
+        )
+    }) {
+        return Err(latch_validation(
+            state,
+            known_validation(Error::other("pool identity is corrupt or incompatible"), "native_identity_validation"),
+        ));
+    }
+    if identities
+        .iter()
+        .any(|read| !matches!(read.replica, PoolMetaIdentityReplica::Valid(_)))
+    {
+        return Err(invalid());
+    }
+    let identity = select_pool_meta_identity(identities, cluster_id)
+        .map_err(|error| latch_validation(state, known_validation(error, "native_identity_validation")))?;
+    state
+        .validate_identity(&identity)
+        .map_err(|error| latch_validation(state, known_validation(error, "native_identity_validation")))?;
+    if identity.identity.map(encode_pool_meta_identity).transpose()?.as_deref() != Some(identity_bytes.as_slice()) {
+        return Err(latch_validation(
+            state,
+            known_validation(Error::other("native identity changed after repair"), "native_identity_validation"),
+        ));
+    }
+    fence.ensure_held()?;
+    arm.disarm();
+    info!(
+        event = EVENT_DECOMMISSION_STATE,
+        component = LOG_COMPONENT_ECSTORE,
+        subsystem = LOG_SUBSYSTEM_POOLS,
+        state = "pool_metadata_native_repaired",
+        generation = selection.revision.generation,
+        pools = inspected.len(),
+        disks = inspected.iter().map(|pool| pool.disks.len()).sum::<usize>(),
+        elapsed_ms = repair_started.elapsed().as_secs_f64() * 1000.0,
+        "Pool metadata native replicas repaired"
+    );
+    Ok(true)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use codec::{fixture_inline_pool_meta_replicas as replicas, fixture_split_inline_pool_meta_replicas as split};
+
+    struct Fixture {
+        pools: Vec<InspectedPool>,
+        commit: Vec<u8>,
+        pending: Vec<u8>,
+        identity: Vec<u8>,
+        revision: PoolMetaRevision,
+        modified: OffsetDateTime,
+    }
+
+    async fn fixture() -> Fixture {
+        let cluster_id = uuid::Uuid::new_v4();
+        let revision = PoolMetaRevision {
+            version: POOL_META_GENERATION_VERSION,
+            cluster_id: Some(cluster_id),
+            epoch: 1,
+            generation: 1,
+            transaction_id: Some(uuid::Uuid::new_v4()),
+        };
+        let meta = PoolMeta {
+            version: POOL_META_GENERATION_VERSION,
+            ..Default::default()
+        };
+        let previous_meta = PoolMeta {
+            version: POOL_META_V1_VERSION,
+            ..Default::default()
+        };
+        let previous = PoolMetaCommittedCandidate {
+            canonical: previous_meta.encode_config_data_for_v2_gate(true).unwrap(),
+            meta: previous_meta,
+            revision: PoolMetaRevision::legacy(POOL_META_V1_VERSION),
+        };
+        let pending = encode_pool_meta_v3_envelope(&meta, revision, false, Some(&previous)).unwrap();
+        let commit = encode_pool_meta_v3_envelope(&meta, revision, true, None).unwrap();
+        let identity = encode_pool_meta_identity(PersistedPoolMetaIdentity {
+            version: POOL_META_IDENTITY_VERSION,
+            cluster_id,
+            epoch: 1,
+            initialized: true,
+            fresh_bootstrap_nonce: None,
+        })
+        .unwrap();
+        let modified = OffsetDateTime::from_unix_timestamp(1_700_000_000).unwrap();
+        let current_split = split(&pending, &commit, modified).await;
+        let current_commit = replicas(&commit, modified + Duration::seconds(1)).await;
+        let backups = replicas(&pending, modified).await;
+        let identities = replicas(&identity, modified).await;
+        let directory = uuid::Uuid::new_v4();
+        let pools = [current_split.clone(), current_split, current_commit.clone(), current_commit]
+            .into_iter()
+            .map(|current| InspectedPool {
+                disks: Vec::new(),
+                snapshots: current
+                    .into_iter()
+                    .enumerate()
+                    .map(|(index, current)| PoolMetaNativeInspection {
+                        disk_id: uuid::Uuid::new_v4(),
+                        identity: Some(identities[index].clone()),
+                        current: Some(current),
+                        backups: vec![crate::disk::PoolMetaNativeBackup {
+                            directory,
+                            data: backups[index].clone(),
+                        }],
+                    })
+                    .collect(),
+            })
+            .collect();
+        Fixture {
+            pools,
+            commit,
+            pending,
+            identity,
+            revision,
+            modified,
+        }
+    }
+
+    #[tokio::test]
+    async fn pool_meta_native_plan_restores_physical_split_with_fresh_common_identity() {
+        let f = fixture().await;
+        let repaired = plan(&f.pools, &f.commit, f.revision, &f.identity).await.unwrap();
+        assert_eq!(repaired.len(), 4);
+        for pool in repaired {
+            assert_eq!(
+                codec::decode_inline_pool_meta_replicas(&pool).await.unwrap(),
+                vec![Bytes::copy_from_slice(&f.commit)]
+            );
+            let times = pool
+                .iter()
+                .map(|raw| codec::parse_inline_info(raw).unwrap().mod_time.unwrap())
+                .collect::<HashSet<_>>();
+            assert_eq!(times.len(), 1);
+            assert!(*times.iter().next().unwrap() > f.modified + Duration::seconds(1));
+        }
+    }
+
+    #[tokio::test]
+    async fn pool_meta_native_plan_rejects_single_hidden_identity_fork() {
+        let mut f = fixture().await;
+        let wrong = replicas(b"different identity", f.modified).await;
+        f.pools[0].snapshots[0].identity = Some(wrong[0].clone());
+        assert!(plan(&f.pools, &f.commit, f.revision, &f.identity).await.is_err());
+    }
+
+    #[tokio::test]
+    async fn pool_meta_native_plan_rejects_hidden_higher_generation_and_unknown_backup() {
+        let mut f = fixture().await;
+        let higher_revision = PoolMetaRevision {
+            generation: 2,
+            transaction_id: Some(uuid::Uuid::new_v4()),
+            ..f.revision
+        };
+        let higher = encode_pool_meta_v3_envelope(
+            &PoolMeta {
+                version: POOL_META_GENERATION_VERSION,
+                ..Default::default()
+            },
+            higher_revision,
+            true,
+            None,
+        )
+        .unwrap();
+        let wrong = replicas(&higher, f.modified + Duration::seconds(2)).await;
+        f.pools[0].snapshots[0].current = Some(wrong[0].clone());
+        assert!(plan(&f.pools, &f.commit, f.revision, &f.identity).await.is_err());
+        f.pools[0].snapshots[0].current = Some(replicas(&f.pending, f.modified).await[0].clone());
+        f.pools[0].snapshots[0].backups[0].data = wrong[0].clone();
+        assert!(plan(&f.pools, &f.commit, f.revision, &f.identity).await.is_err());
+    }
+
+    #[tokio::test]
+    async fn pool_meta_native_plan_accepts_retained_older_recovery_backups() {
+        let mut f = fixture().await;
+        let older_commit = f.commit.clone();
+        let older_pending = f.pending.clone();
+        let next = PoolMetaRevision {
+            generation: 2,
+            transaction_id: Some(uuid::Uuid::new_v4()),
+            ..f.revision
+        };
+        let PoolMetaReplica::Valid { meta, .. } = decode_pool_meta_replica(older_commit.clone()) else {
+            panic!("valid fixture")
+        };
+        let previous = PoolMetaCommittedCandidate {
+            canonical: older_commit.clone(),
+            meta: meta.clone(),
+            revision: f.revision,
+        };
+        f.commit = encode_pool_meta_v3_envelope(&meta, next, true, None).unwrap();
+        f.pending = encode_pool_meta_v3_envelope(&meta, next, false, Some(&previous)).unwrap();
+        f.revision = next;
+        let current = split(&f.pending, &f.commit, f.modified + Duration::seconds(2)).await;
+        let old_commit = replicas(&older_commit, f.modified + Duration::seconds(1)).await;
+        let current_pending = replicas(&f.pending, f.modified + Duration::seconds(2)).await;
+        let new_directory = uuid::Uuid::new_v4();
+        let old_directory = uuid::Uuid::new_v4();
+        for pool in &mut f.pools {
+            for (index, snapshot) in pool.snapshots.iter_mut().enumerate() {
+                snapshot.current = Some(current[index].clone());
+                snapshot.backups.push(crate::disk::PoolMetaNativeBackup {
+                    directory: new_directory,
+                    data: current_pending[index].clone(),
+                });
+                snapshot.backups.push(crate::disk::PoolMetaNativeBackup {
+                    directory: old_directory,
+                    data: old_commit[index].clone(),
+                });
+            }
+        }
+        assert!(older_pending != f.pending);
+        assert!(plan(&f.pools, &f.commit, f.revision, &f.identity).await.is_ok());
+    }
+
+    #[tokio::test]
+    async fn pool_meta_native_plan_rejects_missing_or_corrupt_evidence() {
+        let mut f = fixture().await;
+        let before = f.pools[1].snapshots[2].current.take();
+        assert!(plan(&f.pools, &f.commit, f.revision, &f.identity).await.is_err());
+        f.pools[1].snapshots[2].current = before;
+        f.pools[1].snapshots[2].backups[0].data = Bytes::from_static(b"invalid metadata");
+        assert!(plan(&f.pools, &f.commit, f.revision, &f.identity).await.is_err());
+    }
+}
+
+#[cfg(all(test, feature = "test-util"))]
+#[path = "pool_meta_native_integration_tests.rs"]
+mod integration_tests;