feat(nightly): publish packages as assets of the rolling 'nightly' release (#7592 )

Replace the assets-branch scheme with a proper GitHub Release on rustfs/auto-testing: a single 'nightly' release whose deb/rpm assets are replaced in place on every build. This is the standard channel — visible on the repo's Releases page, stable download URLs, no git history growth (release assets live outside the repository). - New scripts/release/publish_nightly_assets.sh: resolves-or-creates the 'nightly' release via the REST API, deletes same-name assets, uploads rustfs-nightly-latest.{deb,rpm}, then PATCHes the release body with the build provenance (ref@sha, run link, sizes, SHA256). Plain curl + python3, no gh CLI (the build fleet has none — #7586). - The workflow step shrinks to invoking the script; full flow exercised end-to-end against the real release with probe files (create / upload / overwrite / download round-trip / body update).
fix(kms): prevent transient health failures from latching status (#7578 )
2026-09-09 13:46:05 +00:00 · 2026-09-09 21:33:58 +08:00 · 2026-09-09 11:05:26 +00:00 · 2026-09-09 16:39:59 +08:00 · 2026-09-09 16:39:25 +08:00 · 2026-09-09 14:50:28 +08:00
52 changed files with 3620 additions and 454 deletions
@@ -89,6 +89,7 @@ offline-enrollment-e2e-check: core-deps ## Build and exercise the dedicated offl
 test-wiring-check: ## Check tests stay registered and selected by their intended runners
 	@echo "🧪 Checking test wiring..."
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_test_wiring.py
+	$(RUSTFS_PYTHON_BIN) ./scripts/ci_gate.py --check-workflow

 .PHONY: log-analyzer-rules-check
 log-analyzer-rules-check: core-deps ## Check log-analyzer rule anchors still exist verbatim in source
@@ -39,6 +39,7 @@ script-tests: ## Run shell script tests
 	./scripts/test_python_bin.sh
 	./scripts/check_embedded_secrets.sh --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_test_wiring.py --self-test
+	$(RUSTFS_PYTHON_BIN) ./scripts/ci_gate.py --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_security_coverage.py --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/check_scheduled_validation_freshness.py --self-test
 	$(RUSTFS_PYTHON_BIN) ./scripts/test_security_workflow.py
@@ -197,6 +197,12 @@ test-group = 'e2e-cluster-nightly'
 filter = 'package(e2e_test) & (test(/^kms::kms_vault_test::/) | test(/^kms::kms_rekey_sweep_test::/) | test(/^kms::configured_roundtrip_test::test_configured_vault_kms_admin_and_versioned_cleanup$/))'
 test-group = 'e2e-vault'

+# This four-disk, 65-member rollback probe already drives up to 32 concurrent
+# durable deletions. Reserve this nextest run's capacity for its progress oracle.
+[[profile.default.overrides]]
+filter = 'package(rustfs-ecstore) & test(=store::init::tests::dispatch_manifest_rollback_bounded_concurrency_reaches_tail_behind_slow_member)'
+threads-required = "num-test-threads"
+
 # ---------------------------------------------------------------------------
 # ci profile — the strict CI gate (ci.yml `cargo nextest run --profile ci`)
 # ---------------------------------------------------------------------------
@@ -111,10 +111,6 @@ runs:
      shell: bash
      run: ./scripts/check_no_planning_docs.sh

-    - name: Check CI paths stay in sync
-      shell: bash
-      run: ./scripts/check_ci_paths_sync.sh
-
    - name: Check io_uring lane --lib precondition
      shell: bash
      run: ./scripts/check_uring_lane_lib_only.sh
@@ -1,79 +0,0 @@
-# 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.
-
-# Reports the existing required checks for paths excluded by ci.yml.
-# Mixed PRs can trigger both workflows; their Quick Checks jobs use one shared
-# action to keep validation coverage aligned. Keep this paths list in sync with
-# ci.yml's pull_request.paths-ignore via scripts/check_ci_paths_sync.sh.
-
-name: Continuous Integration (docs only)
-
-on:
-  pull_request:
-    types: [ opened, synchronize, reopened ]
-    branches: [ main ]
-    paths:
-      - "**.md"
-      - "docs/**"
-      - "deploy/**"
-      - "scripts/dev_*.sh"
-      - "scripts/probe.sh"
-      - "LICENSE*"
-      - ".gitignore"
-      - ".dockerignore"
-      - "README*"
-      - "**/*.png"
-      - "**/*.jpg"
-      - "**/*.svg"
-      - ".github/workflows/build.yml"
-      - ".github/workflows/docker.yml"
-      - ".github/workflows/audit.yml"
-      - "flake.lock"
-
-permissions:
-  contents: read
-
-jobs:
-  quick-checks:
-    name: Quick Checks
-    runs-on: ubuntu-latest
-    timeout-minutes: 10
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
-        with:
-          persist-credentials: false
-
-      - name: Run shared quick checks
-        uses: ./.github/actions/quick-checks
-
-  test-and-lint:
-    name: Test and Lint
-    runs-on: ubuntu-latest
-    timeout-minutes: 10
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
-        with:
-          persist-credentials: false
-
-      # Docs-only PRs skip the full code CI, but they are exactly where a
-      # planning-type document could be slipped in (git add -f bypasses
-      # .gitignore). Run the guard here so the required "Test and Lint" check
-      # stays meaningful for docs-only changes.
-      - name: Check no planning docs committed
-        run: ./scripts/check_no_planning_docs.sh
-
-      - name: Satisfy required check for docs-only changes
-        run: echo "Docs-only change — code CI is skipped by paths-ignore; planning-docs guard passed, reporting success for the required 'Test and Lint' check."
@@ -37,25 +37,6 @@ on:
  pull_request:
    types: [ opened, synchronize, reopened, closed ]
    branches: [ main ]
-    # Keep this list in sync with the `paths` list in ci-docs-only.yml, which
-    # reports the required "Test and Lint" check for PRs skipped here.
-    paths-ignore:
-      - "**.md"
-      - "docs/**"
-      - "deploy/**"
-      - "scripts/dev_*.sh"
-      - "scripts/probe.sh"
-      - "LICENSE*"
-      - ".gitignore"
-      - ".dockerignore"
-      - "README*"
-      - "**/*.png"
-      - "**/*.jpg"
-      - "**/*.svg"
-      - ".github/workflows/build.yml"
-      - ".github/workflows/docker.yml"
-      - ".github/workflows/audit.yml"
-      - "flake.lock"
  merge_group:
    types: [ checks_requested ]
  schedule:
@@ -88,6 +69,32 @@ jobs:
      - name: Explain cancellation run
        run: echo "PR closed; this run only cancels older runs in the same concurrency group."

+  classify-changes:
+    name: Select CI scope
+    if: github.event_name != 'pull_request' || github.event.action != 'closed'
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    outputs:
+      mode: ${{ steps.scope.outputs.mode }}
+    steps:
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          fetch-depth: 2
+          persist-credentials: false
+      - name: Select scope using the base revision's policy
+        id: scope
+        env:
+          CI_BASE_SHA: ${{ github.event.pull_request.base.sha }}
+        run: |
+          if [[ "$GITHUB_EVENT_NAME" != "pull_request" ]]; then
+            printf '%s\n' 'mode=full' >> "$GITHUB_OUTPUT"
+          elif [[ "$CI_BASE_SHA" =~ ^[0-9a-f]{40}$ ]] && git show "$CI_BASE_SHA:scripts/ci_gate.py" > "$RUNNER_TEMP/ci-gate-base.py"; then
+            python3 -I "$RUNNER_TEMP/ci-gate-base.py" select
+          else
+            printf '%s\n' 'mode=full' >> "$GITHUB_OUTPUT"
+            echo "Base CI policy unavailable; running the full matrix."
+          fi
+
  typos:
    name: Typos
    if: github.event_name != 'pull_request' || github.event.action != 'closed'
@@ -100,7 +107,7 @@ jobs:
      - name: Typos check with custom config file
        uses: crate-ci/typos@37bb98842b0d8c4ffebdb75301a13db0267cef89 # master

-  # Fail early with compile-free checks shared with docs-only CI.
+  # Fail early with compile-free checks for every pull request.
  quick-checks:
    name: Quick Checks
    if: github.event_name != 'pull_request' || github.event.action != 'closed'
@@ -116,9 +123,9 @@ jobs:
        uses: ./.github/actions/quick-checks

  test-and-lint:
-    name: Test and Lint
-    if: github.event_name != 'pull_request' || github.event.action != 'closed'
-    needs: [ quick-checks ]
+    name: Workspace Test and Lint
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    needs: [ quick-checks, classify-changes ]
    runs-on: sm-standard-4
    timeout-minutes: 90
    env:
@@ -289,45 +296,6 @@ jobs:
      - name: Run rebalance/decommission migration proofs
        run: ./scripts/check_migration_gate_count.sh

-      # Record the reason before this job completes as FAILURE. A separate
-      # dependent job cancels sibling lanes only after GitHub has preserved this
-      # required check's failure verdict.
-      - name: Annotate early-stop reason
-        if: >-
-          failure() && github.event_name == 'pull_request'
-          && github.event.pull_request.head.repo.full_name == github.repository
-        run: |
-          {
-            echo "## CI early-stop"
-            echo "Job \`${GITHUB_JOB}\` (Test and Lint) failed; a follow-up job will cancel sibling lanes to free runners."
-            echo "Sibling jobs showing **cancelled** were stopped by the early-stop follow-up, not by their own failure."
-          } >> "$GITHUB_STEP_SUMMARY"
-
-  # Preserve the required Test and Lint FAILURE verdict before stopping sibling
-  # lanes. Cancelling from inside test-and-lint changed its own conclusion to
-  # CANCELLED and hid the actionable failure in the PR checks UI.
-  cancel-after-test-and-lint-failure:
-    name: Cancel siblings after Test and Lint failure
-    if: >-
-      failure() && needs.test-and-lint.result == 'failure'
-      && github.event_name == 'pull_request'
-      && github.event.pull_request.head.repo.full_name == github.repository
-    needs: [ test-and-lint ]
-    runs-on: ubuntu-latest
-    timeout-minutes: 5
-    permissions:
-      actions: write
-    steps:
-      - name: Cancel remaining jobs
-        env:
-          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          curl -fsS -X POST \
-            -H "Authorization: Bearer ${GH_TOKEN}" \
-            -H "Accept: application/vnd.github+json" \
-            -H "X-GitHub-Api-Version: 2022-11-28" \
-            "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/actions/runs/${GITHUB_RUN_ID}/cancel"
-
  # Dedicated serial lane for the ILM / lifecycle integration tests. These tests
  # drive the object layer through process-global singletons (the GLOBAL_ENV
  # ECStore, the global tier-config manager, background-expiry workers) and bind
@@ -340,8 +308,8 @@ jobs:
  # See rustfs/backlog#1148 (ilm-1) and #1155.
  test-ilm-integration-serial:
    name: ILM Integration (serial)
-    if: github.event_name != 'pull_request' || github.event.action != 'closed'
-    needs: [ quick-checks ]
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    needs: [ quick-checks, classify-changes ]
    runs-on: sm-standard-4
    timeout-minutes: 90
    env:
@@ -408,8 +376,8 @@ jobs:

  test-and-lint-rio-v2:
    name: Test and Lint (rio-v2)
-    if: github.event_name != 'pull_request' || github.event.action != 'closed'
-    needs: [ quick-checks ]
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    needs: [ quick-checks, classify-changes ]
    runs-on: sm-standard-4
    timeout-minutes: 90
    env:
@@ -449,8 +417,8 @@ jobs:

  connect-short-credential-boundary:
    name: Connect Short Credential Boundary
-    if: github.event_name != 'pull_request' || github.event.action != 'closed'
-    needs: [ quick-checks ]
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    needs: [ quick-checks, classify-changes ]
    runs-on: sm-standard-4
    timeout-minutes: 60
    env:
@@ -507,8 +475,8 @@ jobs:

  test-and-lint-protocols:
    name: "Test and Lint (${{ matrix.features.name }})"
-    if: github.event_name != 'pull_request' || github.event.action != 'closed'
-    needs: [ quick-checks ]
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    needs: [ quick-checks, classify-changes ]
    runs-on: sm-standard-4
    timeout-minutes: 90
    strategy:
@@ -561,8 +529,8 @@ jobs:

  build-rustfs-debug-binary:
    name: Build RustFS Debug Binary
-    if: github.event_name != 'pull_request' || github.event.action != 'closed'
-    needs: [ quick-checks ]
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    needs: [ quick-checks, classify-changes ]
    runs-on: sm-standard-4
    timeout-minutes: 30
    env:
@@ -684,8 +652,8 @@ jobs:
    # job had neither, so each closed/merged PR really ran the whole io_uring
    # suite (measured 4m17s / 7m19s / 7m31s on runs 30678272341 / 30678117601 /
    # 30662728539) and kept the cancellation run in progress for minutes.
-    if: github.event_name != 'pull_request' || github.event.action != 'closed'
-    needs: [ quick-checks ]
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    needs: [ quick-checks, classify-changes ]
    # GitHub-hosted ubuntu-latest runs a recent kernel with io_uring and, unlike
    # a container, applies no seccomp filter that would block io_uring_setup — so
    # the probe succeeds and the tests exercise the real UringBackend/FdCache/
@@ -995,11 +963,6 @@ jobs:
          cargo nextest list --profile e2e-full -p e2e_test --message-format json > "${NEXTEST_LISTING}"
          python3 ./scripts/check_test_wiring.py --check-profile e2e-full "${NEXTEST_LISTING}"

-      - name: Prepare Heal logs
-        run: |
-          heal_log_root=$(mktemp -d "${RUNNER_TEMP}/rustfs-heal-logs.XXXXXX")
-          echo "RUSTFS_HEAL_CHAOS_LOG_DIR=$heal_log_root" >> "$GITHUB_ENV"
-
      # Full single-node e2e lane (backlog#1149 ci-5). The e2e-full
      # default-filter in .config/nextest.toml is the single wiring mechanism —
      # extend that filter, never add ad-hoc e2e jobs here. Reuses the downloaded
@@ -1009,18 +972,8 @@ jobs:
          RUSTFS_E2E_STARTUP_CAS_BINARY: ${{ runner.temp }}/rustfs-startup-cas-input/rustfs
          RUSTFS_E2E_STARTUP_CAS_BUILD_MANIFEST: ${{ runner.temp }}/rustfs-startup-cas-input/rustfs.e2e-startup-cas-build.json
          RUSTFS_E2E_STARTUP_CAS_ARTIFACT_DIR: ${{ runner.temp }}/rustfs-startup-cas-evidence
-          RUSTFS_HEAL_CHAOS_SERVER_RUST_LOG: error,rustfs::heal::task=info,rustfs::app::object_usecase=warn,rustfs_ecstore::set_disk::ops::object=warn,rustfs_lock::distributed_lock=debug,rustfs_ecstore::cluster::rpc::remote_locker=warn
        run: cargo nextest run --profile e2e-full -p e2e_test

-      - name: Upload coordinator restart logs
-        if: always() && env.RUSTFS_HEAL_CHAOS_LOG_DIR != ''
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
-        with:
-          name: heal-coordinator-restart-logs-${{ github.run_number }}-${{ github.run_attempt }}
-          path: ${{ env.RUSTFS_HEAL_CHAOS_LOG_DIR }}/coordinator_restart/**/node*.log
-          if-no-files-found: warn
-          retention-days: 7
-
      - name: Upload junit
        if: always()
        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
@@ -1227,9 +1180,44 @@ jobs:
          if-no-files-found: ignore
          retention-days: 3

+  required-checks:
+    name: Test and Lint
+    if: always() && (github.event_name != 'pull_request' || github.event.action != 'closed')
+    needs:
+      - classify-changes
+      - typos
+      - quick-checks
+      - test-and-lint
+      - test-ilm-integration-serial
+      - test-and-lint-rio-v2
+      - connect-short-credential-boundary
+      - test-and-lint-protocols
+      - build-rustfs-debug-binary
+      - uring-integration
+      - e2e-tests
+      - s3-implemented-tests
+      - s3-lifecycle-behavior-tests
+      - build-rustfs-debug-binary-rio-v2
+      - e2e-tests-rio-v2
+      - e2e-full
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
+        with:
+          persist-credentials: false
+      - name: Require the expected result of every CI lane
+        env:
+          CI_NEEDS: ${{ toJSON(needs) }}
+        shell: bash
+        run: python3 scripts/ci_gate.py verify
+
  alert-on-failure:
    name: Alert on scheduled failure
    needs:
+      - classify-changes
+      - connect-short-credential-boundary
+      - required-checks
      - typos
      - quick-checks
      - test-and-lint
@@ -19,17 +19,27 @@ on:
    - cron: "7 0 * * *"
      timezone: "Asia/Shanghai"
  workflow_dispatch:
+    inputs:
+      branch:
+        description: 'Branch/ref to build and publish as the nightly (empty = scheduled source, see NIGHTLY_BUILD_REF)'
+        required: false
+        default: ''

 permissions:
  contents: read

+# Scheduled builds follow the NIGHTLY_BRANCH repo variable so the channel can
+# be pointed at e.g. `release` for the GA cycle and back to `main` afterwards
+# without touching this file. Manual runs take the `branch` input, falling
+# back to the branch the run was dispatched from.
 concurrency:
-  group: nightly-gnu-build-main-${{ github.event_name }}
+  group: nightly-gnu-build-${{ github.event_name }}-${{ github.event_name == 'schedule' && (vars.NIGHTLY_BRANCH || 'main') || (inputs.branch || github.ref_name) }}
  cancel-in-progress: ${{ github.event_name == 'workflow_dispatch' }}

 env:
  CARGO_TERM_COLOR: always
  RUST_BACKTRACE: 1
+  NIGHTLY_BUILD_REF: ${{ github.event_name == 'schedule' && (vars.NIGHTLY_BRANCH || 'main') || (inputs.branch || github.ref_name) }}

 jobs:
  build:
@@ -43,6 +53,7 @@ jobs:
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
        with:
          persist-credentials: false
+          ref: ${{ env.NIGHTLY_BUILD_REF }}

      - name: Setup Rust environment
        uses: ./.github/actions/setup
@@ -152,13 +163,104 @@ jobs:

          fakeroot dpkg-deb --build "${PKG_DIR}"
          ls -lh "${DEB_FILE}"
+          echo "deb_date=${DEB_DATE}" >> "${GITHUB_OUTPUT}"
          echo "deb_file=${DEB_FILE}" >> "${GITHUB_OUTPUT}"

+      # Same packaging scheme as .github/workflows/package.yml (fpm), but from
+      # the locally built nightly binary instead of a release artifact, with a
+      # date-based version that mirrors the DEB.
+      - name: Build RPM package
+        id: rpm
+        shell: bash
+        env:
+          DEB_DATE: ${{ steps.deb.outputs.deb_date }}
+        run: |
+          set -euo pipefail
+
+          if ! command -v fpm >/dev/null 2>&1; then
+            SUDO=""; [ "$(id -u)" -ne 0 ] && SUDO="sudo -n"
+            ${SUDO} apt-get update -qq && ${SUDO} apt-get install -y -qq ruby ruby-dev build-essential rpm >/dev/null
+            ${SUDO} gem install fpm --no-document >/dev/null
+          fi
+
+          RPM_FILE="rustfs-nightly-${DEB_DATE}.rpm"
+          RPM_VERSION="0"
+          RPM_RELEASE="0.nightly.${DEB_DATE//-/.}"
+
+          echo "Building RPM: ${RPM_FILE} (version ${RPM_VERSION}-${RPM_RELEASE})"
+
+          # fpm wants the config file to exist before packaging.
+          mkdir -p ./tmp-pkg/etc/default
+          cat > ./tmp-pkg/etc/default/rustfs << 'ENVEOF'
+          # RustFS Environment Configuration
+          # See https://rustfs.com/docs/ for more information
+          # RUSTFS_VOLUMES=""
+          # RUSTFS_ROOT_USER=""
+          # RUSTFS_ROOT_PASSWORD=""
+          ENVEOF
+
+          fpm -s dir -t rpm \
+            --name rustfs \
+            --version "$RPM_VERSION" \
+            --iteration "$RPM_RELEASE" \
+            --architecture x86_64 \
+            --package "$RPM_FILE" \
+            --depends "glibc >= 2.31" \
+            --maintainer "RustFS Team <support@rustfs.com>" \
+            --description "High-performance distributed object storage" \
+            --url "https://rustfs.com" \
+            --license "Apache-2.0" \
+            --after-install <(cat << 'POSTINST'
+          #!/bin/bash
+          set -e
+          if ! getent passwd rustfs > /dev/null 2>&1; then
+              useradd -r -s /bin/false -d /opt/rustfs rustfs
+          fi
+          mkdir -p /opt/rustfs /data/rustfs /var/log/rustfs
+          chown rustfs:rustfs /opt/rustfs /data/rustfs /var/log/rustfs
+          if [ -d /run/systemd/system ]; then
+              systemctl daemon-reload
+          fi
+          POSTINST
+            ) \
+            --before-remove <(cat << 'PRERM'
+          #!/bin/bash
+          set -e
+          if [ -d /run/systemd/system ] && systemctl is-active --quiet rustfs; then
+              systemctl stop rustfs
+          fi
+          PRERM
+            ) \
+            --after-remove <(cat << 'POSTRM'
+          #!/bin/bash
+          set -e
+          if [ -d /run/systemd/system ]; then
+              systemctl daemon-reload
+          fi
+          POSTRM
+            ) \
+            --config-files /etc/default/rustfs \
+            "rustfs-nightly-${DEB_DATE}/usr/bin/rustfs=/usr/bin/rustfs" \
+            ./tmp-pkg/etc/default/rustfs=/etc/default/rustfs \
+            deploy/build/rustfs.service=/lib/systemd/system/rustfs.service \
+            LICENSE=/usr/share/doc/rustfs/LICENSE \
+            README.md=/usr/share/doc/rustfs/README.md
+
+          [[ -f "$RPM_FILE" ]] || { echo "RPM build failed"; exit 1; }
+          rpm -qpl "$RPM_FILE" | grep -Fx '/usr/bin/rustfs' >/dev/null
+          stat --printf='%n %s bytes\n' "$RPM_FILE"
+          echo "rpm_file=$RPM_FILE" >> "$GITHUB_OUTPUT"
+
      - name: Upload DEB artifact
        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
        with:
          name: ${{ steps.deb.outputs.deb_file }}
          path: ${{ steps.deb.outputs.deb_file }}
+      - name: Upload RPM artifact
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: ${{ steps.rpm.outputs.rpm_file }}
+          path: ${{ steps.rpm.outputs.rpm_file }}
          if-no-files-found: error

      # Persist the nightly deb on Cloudflare R2 (same channel as package.yml)
@@ -187,11 +289,10 @@ jobs:
          export AWS_SECRET_ACCESS_KEY="$R2_SECRET_ACCESS_KEY"
          export AWS_DEFAULT_REGION="auto"

+          # The candidate manifest must describe the tree that was actually
+          # built. With a ref override (NIGHTLY_BRANCH / dispatch input) that
+          # is not necessarily GITHUB_SHA, so always advertise HEAD.
          SOURCE_SHA="$(git rev-parse HEAD)"
-          if [[ "${SOURCE_SHA}" != "${GITHUB_SHA}" ]]; then
-            echo "Checkout SHA does not match the nightly build run" >&2
-            exit 1
-          fi
          DEB_SHA256="$(sha256sum "${DEB_FILE}" | cut -d ' ' -f 1)"
          CANDIDATE_KEY="artifacts/rustfs/packages/nightly/runs/${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}/${DEB_SHA256}/rustfs.deb"
          CANDIDATE_URL="https://dl.rustfs.com/${CANDIDATE_KEY}"
@@ -247,6 +348,20 @@ jobs:
          path: ${{ steps.publish.outputs.candidate_file }}
          if-no-files-found: error

+      # Publish the deb/rpm pair to the auto-testing repo's `assets` branch so
+      # engineers can download and install the nightly directly. The branch is
+      # a single-commit orphan rewritten on every build, which keeps the repo
+      # small while the latest files stay reachable at stable raw URLs.
+      # Publish the deb/rpm pair as assets of the rolling `nightly` release on
+      # rustfs/auto-testing (see scripts/release/publish_nightly_assets.sh).
+      - name: Publish packages to auto-testing release assets
+        env:
+          ASSETS_TOKEN: ${{ secrets.PF_TESTING_GH_TOKEN }}
+          DEB_FILE: ${{ steps.deb.outputs.deb_file }}
+          RPM_FILE: ${{ steps.rpm.outputs.rpm_file }}
+          DEB_DATE: ${{ steps.deb.outputs.deb_date }}
+          BUILD_REF: ${{ env.NIGHTLY_BUILD_REF }}
+        run: bash scripts/release/publish_nightly_assets.sh
  # Live-Vault lane for the rustfs-kms suite (rustfs/backlog#1774).
  #
  # RUSTFS_KMS_VAULT_TOKEN is the single switch that adds the Vault KV2 and
@@ -284,6 +399,7 @@ jobs:
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
        with:
          persist-credentials: false
+          ref: ${{ env.NIGHTLY_BUILD_REF }}

      - name: Setup Rust environment
        uses: ./.github/actions/setup
@@ -372,6 +488,7 @@ jobs:
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
        with:
          persist-credentials: false
+          ref: ${{ env.NIGHTLY_BUILD_REF }}

      - name: Setup Rust environment
        uses: ./.github/actions/setup
@@ -82,6 +82,12 @@ jobs:
  performance-test:
    runs-on: pf-testing
    timeout-minutes: 900
+    env:
+      RUSTFS_BENCH_SCRIPT: ${{ github.workspace }}/auto-testing/rustfs_performance_testing.sh
+      RUSTFS_WARP_METHODS: ${{ inputs.test_method }}
+      RUSTFS_WARP_SIZES: ${{ inputs.object_size }}
+      RUSTFS_WARP_DURATION: ${{ inputs.warp_duration || '5m' }}
+      RUSTFS_WARP_CONCURRENCY: ${{ inputs.warp_concurrency || '64' }}
    # Run on manual dispatch, or when the nightly build completed successfully.
    # Skipped when nightly failed.
    if: ${{ github.event_name == 'workflow_dispatch' || github.event_name == 'repository_dispatch' }}
@@ -158,19 +164,15 @@ jobs:
      - name: Run benchmark (GET/PUT/MIXED)
        id: benchmark
        run: |
-          # Empty on automatic (workflow_run) runs -> full 30 rounds.
-          # Manual dispatch can restrict method(s)/size(s).
-          export WARP_METHODS="${{ inputs.test_method }}"
-          export WARP_SIZES="${{ inputs.object_size }}"
          ./auto-testing/rustfs_performance_test.sh \
            --step 5 -y \
-            --warp-duration "${{ inputs.warp_duration || '5m' }}" \
-            --warp-concurrency "${{ inputs.warp_concurrency || '64' }}" \
            --log-file "${LOG_FILE}"

      - name: Analyze results
        if: ${{ steps.benchmark.conclusion == 'success' }}
        run: |
+          export WARP_METHODS="${RUSTFS_WARP_METHODS}" WARP_SIZES="${RUSTFS_WARP_SIZES}"
+          export WARP_DURATION="${RUSTFS_WARP_DURATION}" WARP_CONCURRENCY="${RUSTFS_WARP_CONCURRENCY}"
          ./auto-testing/rustfs_performance_test.sh --step 6 -y --log-file "${LOG_FILE:-/dev/null}"

      - name: Collect RustFS version info
@@ -52,6 +52,7 @@ docs
 __pycache__/
 !docs/
 docs/*
+!docs/README.md
 !docs/architecture/
 !docs/architecture/**
 !docs/operations/
@@ -7,6 +7,16 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0

 ## [Unreleased]

+### Replication
+
+- Object Lock replication PUTs now carry a required integrity header, fixing target rejection introduced by the plain-payload default ([#7097](https://github.com/rustfs/rustfs/pull/7097)). This changes the default outbound request for locked objects but adds no persisted format.
+- Multipart source objects stay on the multipart transport even when their checksum record is a whole-object checksum, so objects above the single-PUT limit remain replicable ([#7047](https://github.com/rustfs/rustfs/pull/7047)).
+- Targets that mint their own version IDs now use a per-target version ledger for tag, retention, legal-hold, and permanent-delete mutations; ambiguous pre-ledger matches fail with backoff instead of guessing ([#7368](https://github.com/rustfs/rustfs/pull/7368)). This adds dual-prefixed internal metadata keys that older readers ignore.
+- Single-part source checksums are forwarded as `x-amz-checksum-*` headers instead of user metadata, so the replica preserves checksum responses ([#7313](https://github.com/rustfs/rustfs/pull/7313)). This changes the default outbound headers for checksummed objects.
+- Site-replication outage recovery now uses a bounded 30-second retry drain plus the 600-second full reconciliation pass, persists destructive liabilities before local deletion, and fences replay settlement and peer edits ([#7148](https://github.com/rustfs/rustfs/pull/7148)). Persisted additions are optional and ignored by older readers.
+- IAM snapshot/deletion replay, target-assigned delete-marker purges, timestamp ordering, and best-effort peer broadcast now close the control-plane gaps found by the R6 review ([#7195](https://github.com/rustfs/rustfs/pull/7195)).
+- Upgrade and rollback: upgrade every node in one site consecutively and verify reconciliation before moving to the next site; do not intentionally run a site mixed-version. Target-version ledger keys are harmless on rollback, although old code cannot use their routing. Before rolling back past [#7307](https://github.com/rustfs/rustfs/pull/7307), drain or repair every pending version purge: older code can free a retained version's data directory before its remote purge is acknowledged. See `docs/operations/site-replication-operations.md`.
+
 ### Security
 - **Presigned URLs honour only signed headers** (GHSA-g8w9-qw9q-fghr): a SigV4 presigned request that carries an `x-amz-*` request header not listed in `X-Amz-SignedHeaders` is now rejected with `403 AccessDenied` ("There were headers present in the request which were not signed"), matching AWS S3. Previously the holder of a presigned `PutObject` URL could add unsigned `x-amz-tagging`, `x-amz-storage-class`, `x-amz-website-redirect-location`, ACL, metadata, Object Lock or SSE headers and have them applied. Presigners that intend a property must set it before signing so the SDK lists the header in `SignedHeaders`; `x-amz-cf-id` (CloudFront) remains tolerated unsigned. Header-signed SigV4 and SigV2 requests are unchanged.

@@ -211,7 +211,10 @@ For developers who want to build RustFS Docker images from source with multi-arc

 ```bash
 # Build multi-architecture images locally
-./docker-buildx.sh --build-arg RELEASE=latest
+./docker-buildx.sh
+
+# Build a single-platform image locally
+./docker-buildx.sh -p linux/amd64

 # Build and push to registry
 ./docker-buildx.sh --push
@@ -150,7 +150,10 @@ docker compose -f docker-compose-simple.yml up -d

 ```bash
 # 在本地构建多架构镜像
-./docker-buildx.sh --build-arg RELEASE=latest
+./docker-buildx.sh
+
+# 在本地构建单平台镜像
+./docker-buildx.sh -p linux/amd64

 # 构建并推送到仓库
 ./docker-buildx.sh --push
@@ -923,7 +923,7 @@ mod tests {
                latest_cycle_end = latest_cycle_end.max(cycle_end);
                versions_observed |= versions_scanned > 0;
                observations.push(format!(
-                    "node{node_index}: end={cycle_end}, versions={versions_scanned}, cycle={}, active={}, leader={}, result={}, status={status}",
+                    "node{node_index}: end={cycle_end}, versions={versions_scanned}, cycle={}, active={}, leader={}, result={}",
                    metrics["current_cycle"],
                    metrics["current_cycle_active"],
                    metrics["leader_lock_state"],
@@ -1071,20 +1071,10 @@ mod tests {
        let server_rust_log = std::env::var("RUSTFS_HEAL_CHAOS_SERVER_RUST_LOG")
            .unwrap_or_else(|_| "rustfs::heal::task=info,rustfs=error".to_string());
        cluster.set_env("RUST_LOG", server_rust_log);
-        let log_dir = match std::env::var("RUSTFS_HEAL_CHAOS_LOG_DIR") {
-            Ok(root) => PathBuf::from(root).join(interruption_kind).join(
-                Path::new(&cluster.temp_dir)
-                    .file_name()
-                    .ok_or("cluster temp directory has no basename")?,
-            ),
-            Err(_) => PathBuf::from(&cluster.temp_dir).join("logs"),
-        };
+        let log_dir = std::env::var("RUSTFS_HEAL_CHAOS_LOG_DIR").unwrap_or_else(|_| format!("{}/logs", cluster.temp_dir));
        std::fs::create_dir_all(&log_dir)?;
        for node_index in 0..cluster.nodes.len() {
-            cluster.set_node_capture_log_path(
-                node_index,
-                log_dir.join(format!("node{node_index}.log")).to_string_lossy().into_owned(),
-            )?;
+            cluster.set_node_capture_log_path(node_index, format!("{log_dir}/node{node_index}.log"))?;
        }
        cluster.start_with_binary(&server_binary).await?;
        let clients = cluster.create_all_clients()?;
@@ -1338,7 +1328,7 @@ mod tests {
        let pre_interrupt_status: serde_json::Value = serde_json::from_str(&pre_interrupt_status_body)
            .map_err(|err| format!("pre-interrupt background heal status is not JSON ({err}): {pre_interrupt_status_body}"))?;
        let pre_interrupt_replacement = replacement_recovery_status(&cluster).await?;
-        let coordinator_log = std::fs::read_to_string(log_dir.join("node0.log"))?;
+        let coordinator_log = std::fs::read_to_string(format!("{log_dir}/node0.log"))?;
        assert!(
            coordinator_log
                .lines()
@@ -25,6 +25,190 @@ const KEY: &str = "thumb/79/concurrent-overwrite.jpg";

 type TestResult = Result<(), Box<dyn std::error::Error + Send + Sync>>;

+async fn assert_quorum_object_body(client: &Client, bucket: &str, key: &str, expected: &[u8]) -> TestResult {
+    let body = client
+        .get_object()
+        .bucket(bucket)
+        .key(key)
+        .send()
+        .await?
+        .body
+        .collect()
+        .await?
+        .into_bytes();
+    assert_eq!(body.as_ref(), expected, "quorum read returned incorrect contents for {key}");
+    Ok(())
+}
+
+async fn wait_for_quorum_read_admission(clients: &[Client], bucket: &str) -> TestResult {
+    // SIGKILL can orphan a granted lease. Wait for shared metadata-lock
+    // admission before asserting the stable quorum boundary; cold bodies
+    // remain unread throughout this readiness probe.
+    let deadline =
+        tokio::time::Instant::now() + rustfs_lock::fast_lock::DEFAULT_LOCK_TIMEOUT + std::time::Duration::from_secs(15);
+    loop {
+        let mut ready = true;
+        for client in clients {
+            for key in ["warm-small", "warm-large"] {
+                match client.head_object().bucket(bucket).key(key).send().await {
+                    Ok(_) => {}
+                    Err(error) if error.raw_response().is_some_and(|response| response.status().as_u16() == 503) => {
+                        ready = false;
+                        break;
+                    }
+                    Err(error) => return Err(error.into()),
+                }
+            }
+            if !ready {
+                break;
+            }
+        }
+        if ready {
+            return Ok(());
+        }
+        if tokio::time::Instant::now() >= deadline {
+            return Err(format!("read quorum did not become available after lease convergence for {bucket}").into());
+        }
+        tokio::time::sleep(std::time::Duration::from_millis(100)).await;
+    }
+}
+
+#[tokio::test]
+async fn test_degraded_cluster_read_quorum_follows_erasure_layout() -> TestResult {
+    crate::common::init_logging();
+
+    for (node_count, parity) in [(4, 2), (6, 3), (6, 2)] {
+        let read_quorum = node_count - parity;
+        let write_quorum = read_quorum + usize::from(read_quorum == parity);
+        let mut cluster = RustFSTestClusterEnvironment::new(node_count).await?;
+        cluster.set_env("RUSTFS_STORAGE_CLASS_STANDARD", format!("EC:{parity}"));
+        // Wait for every seed fanout before removing any physical shard.
+        cluster.set_env("RUSTFS_PUT_RENAME_EARLY_ACK_ENABLE", "false");
+        cluster.set_env("RUSTFS_OBS_METRICS_EXPORT_ENABLED", "false");
+        cluster.set_env("RUST_LOG", "warn,rustfs_lock=debug");
+        cluster.start().await?;
+
+        let clients = cluster
+            .create_all_clients()?
+            .into_iter()
+            .map(|client| {
+                Client::from_conf(
+                    client
+                        .config()
+                        .to_builder()
+                        .retry_config(aws_sdk_s3::config::retry::RetryConfig::standard().with_max_attempts(1))
+                        .build(),
+                )
+            })
+            .collect::<Vec<_>>();
+        let bucket = format!("read-quorum-{node_count}-{parity}");
+        clients[0].create_bucket().bucket(&bucket).send().await?;
+        let small = b"read quorum is derived from the erasure layout".to_vec();
+        let large = (0..1_048_576)
+            .map(|index| u8::try_from(index % 251).expect("bounded payload byte"))
+            .collect::<Vec<_>>();
+        for (key, body) in [
+            ("warm-small", &small),
+            ("warm-large", &large),
+            ("cold-small", &small),
+            ("cold-large", &large),
+            ("below-quorum", &large),
+        ] {
+            clients[node_count - 1]
+                .put_object()
+                .bucket(&bucket)
+                .key(key)
+                .body(Bytes::copy_from_slice(body).into())
+                .send()
+                .await?;
+        }
+        for node in &cluster.nodes {
+            for key in ["warm-small", "warm-large", "cold-small", "cold-large", "below-quorum"] {
+                let census =
+                    crate::chaos::census_object_version_on_disk(std::path::Path::new(&node.data_dir), &bucket, key, None)?;
+                assert!(census.is_complete(), "seed shard must be complete before fault injection: {census:?}");
+                assert_eq!(census.data_blocks, Some(read_quorum));
+                assert_eq!(census.parity_blocks, Some(parity));
+            }
+        }
+        for client in &clients {
+            assert_quorum_object_body(client, &bucket, "warm-small", &small).await?;
+            assert_quorum_object_body(client, &bucket, "warm-large", &large).await?;
+        }
+
+        for offline_node in (read_quorum..node_count).rev() {
+            cluster.stop_node(offline_node)?;
+            wait_for_quorum_read_admission(&clients[..offline_node], &bucket).await?;
+            for client in clients.iter().take(offline_node) {
+                client.head_bucket().bucket(&bucket).send().await?;
+                assert_quorum_object_body(client, &bucket, "warm-large", &large).await?;
+            }
+        }
+
+        // Exercise more than the five-second positive bucket-validation TTL.
+        // Every sample must succeed; polling must not hide a transient failure.
+        let validation_deadline = tokio::time::Instant::now() + std::time::Duration::from_secs(6);
+        loop {
+            for client in clients.iter().take(read_quorum) {
+                assert_quorum_object_body(client, &bucket, "warm-small", &small).await?;
+                assert_quorum_object_body(client, &bucket, "warm-large", &large).await?;
+                let listing = client.list_objects_v2().bucket(&bucket).send().await?;
+                for key in ["warm-small", "warm-large", "cold-small", "cold-large", "below-quorum"] {
+                    assert!(listing.contents().iter().any(|entry| entry.key() == Some(key)), "listing omitted {key}");
+                }
+            }
+            if tokio::time::Instant::now() >= validation_deadline {
+                break;
+            }
+            tokio::time::sleep(std::time::Duration::from_millis(250)).await;
+        }
+        for client in clients.iter().take(read_quorum) {
+            assert_quorum_object_body(client, &bucket, "cold-small", &small).await?;
+            assert_quorum_object_body(client, &bucket, "cold-large", &large).await?;
+        }
+
+        let write = clients[0]
+            .put_object()
+            .bucket(&bucket)
+            .key("quorum-write")
+            .body(Bytes::copy_from_slice(&small).into())
+            .send()
+            .await;
+        if read_quorum >= write_quorum {
+            write?;
+        } else {
+            let error = write.expect_err("a read quorum must not authorize a write that needs more votes");
+            assert_eq!(error.as_service_error().and_then(|error| error.meta().code()), Some("ServiceUnavailable"));
+        }
+
+        cluster.stop_node(read_quorum - 1)?;
+        for client in clients.iter().take(read_quorum - 1) {
+            match client.get_object().bucket(&bucket).key("below-quorum").send().await {
+                Ok(response) => assert!(
+                    response.body.collect().await.is_err(),
+                    "fewer than {read_quorum} valid fragments must not reconstruct an uncached object"
+                ),
+                Err(error) => assert_eq!(
+                    error.as_service_error().and_then(|error| error.meta().code()),
+                    Some("ServiceUnavailable"),
+                    "a quorum loss must not be mistaken for a missing object"
+                ),
+            }
+        }
+
+        for node in 0..read_quorum - 1 {
+            cluster.stop_node(node)?;
+        }
+        cluster.start().await?;
+        for client in &clients {
+            assert_quorum_object_body(client, &bucket, "warm-large", &large).await?;
+            assert_quorum_object_body(client, &bucket, "below-quorum", &large).await?;
+        }
+    }
+
+    Ok(())
+}
+
 async fn put_object(client: Client, payload: Vec<u8>, writer_id: usize) -> Result<(), String> {
    client
        .put_object()
@@ -1137,7 +1137,12 @@ async fn test_odm_admin_config_is_redacted_and_status_counts_match_the_source()
        let miss = env.raw_get(bucket, miss_key).await?;
        assert_eq!(miss.status, 404, "{}", String::from_utf8_lossy(&miss.body));
    }
-    assert!(env.wait_local_listed(bucket, hit_key, SETTLE).await?);
+    let (listed, _, _) = tokio::try_join!(
+        env.wait_local_listed(bucket, hit_key, SETTLE),
+        env.wait_for_status_counter(bucket, "/counters/pulled_objects_total/inline", 1, SETTLE),
+        env.wait_for_status_counter(bucket, "/counters/pulled_bytes_total", body.len() as u64, SETTLE),
+    )?;
+    assert!(listed);

    let status = env.status_json(bucket).await?;
    assert_eq!(status.pointer("/configured").and_then(Value::as_bool), Some(true), "{status}");
@@ -4235,6 +4235,16 @@ async fn test_bucket_replication_acceptance_matrix_local_dual_targets() -> TestR
    <ExistingObjectReplication><Status>Enabled</Status></ExistingObjectReplication>
    <Destination><Bucket>{target_b_arn}</Bucket></Destination>
  </Rule>
+  <Rule>
+    <ID>matrix-and-tags</ID>
+    <Priority>135</Priority>
+    <Status>Enabled</Status>
+    <Filter><And><Prefix>and-tags/</Prefix><Tag><Key>env</Key><Value>prod</Value></Tag><Tag><Key>tier</Key><Value>gold</Value></Tag></And></Filter>
+    <DeleteMarkerReplication><Status>Disabled</Status></DeleteMarkerReplication>
+    <DeleteReplication><Status>Enabled</Status></DeleteReplication>
+    <ExistingObjectReplication><Status>Enabled</Status></ExistingObjectReplication>
+    <Destination><Bucket>{target_b_arn}</Bucket></Destination>
+  </Rule>
  <Rule>
    <ID>matrix-disabled</ID>
    <Priority>140</Priority>
@@ -4289,6 +4299,7 @@ async fn test_bucket_replication_acceptance_matrix_local_dual_targets() -> TestR
        "matrix-prefix",
        "matrix-tag",
        "matrix-disabled",
+        "matrix-and-tags",
        "matrix-priority-high",
        "Priority>200",
        "<Status>Disabled</Status>",
@@ -4409,6 +4420,30 @@ async fn test_bucket_replication_acceptance_matrix_local_dual_targets() -> TestR
    put_single_tag_current(&source_client, source_bucket, "tagged/no-match.txt", "route", "tagged").await?;
    assert_replication_key_absent(&target_client_b, target_bucket_b, "tagged/no-match.txt", Duration::from_secs(3)).await?;

+    // S3 and MinIO both read `And.Tags` as AND: an object carrying only one of
+    // the required tags is not admitted. Matching any single tag would push
+    // data to a destination the rule never selected (backlog#2366 P1-1), and
+    // the two-tag rule is the shape `mc replicate add --tags "k1=v1&k2=v2"`
+    // writes, so a single-tag rule passing is not evidence for this.
+    source_client
+        .put_object()
+        .bucket(source_bucket)
+        .key("and-tags/partial.txt")
+        .tagging("env=prod")
+        .body(ByteStream::from_static(b"one of two tags"))
+        .send()
+        .await?;
+    assert_replication_key_absent(&target_client_b, target_bucket_b, "and-tags/partial.txt", Duration::from_secs(3)).await?;
+    source_client
+        .put_object()
+        .bucket(source_bucket)
+        .key("and-tags/full.txt")
+        .tagging("env=prod&tier=gold")
+        .body(ByteStream::from_static(b"both tags"))
+        .send()
+        .await?;
+    wait_for_user_get_object(&target_client_b, target_bucket_b, "and-tags/full.txt").await?;
+
    source_client
        .put_object()
        .bucket(source_bucket)
@@ -3463,6 +3463,11 @@ impl PoolRebalanceActivationFence {
    }
 }

+#[cfg(test)]
+tokio::task_local! {
+    pub(crate) static REBALANCE_ACTIVATION_LOCK_ATTEMPT: Arc<tokio::sync::Notify>;
+}
+
 pub(crate) async fn acquire_pool_rebalance_activation_locks<S>(
    pool: Arc<S>,
    fleet_proof: Option<crate::services::notification_sys::CrossPoolFenceFleetProofToken>,
@@ -3473,17 +3478,21 @@ where
            NamespaceLock = rustfs_lock::NamespaceLockWrapper,
        >,
 {
-    // Activation lock order is always pool.bin -> rebalance.bin.
+    // Match entry admission: rebalance.bin -> pool.bin. An entry retains its
+    // run read fence while target mutations acquire the pool metadata fence;
+    // activation must not hold pool.bin while waiting for that entry to drain.
+    let rebalance_meta_lock = pool.new_ns_lock(RUSTFS_META_BUCKET, REBAL_META_NAME).await?;
+    #[cfg(test)]
+    let _ = REBALANCE_ACTIVATION_LOCK_ATTEMPT.try_with(|attempted| attempted.notify_one());
+    let rebalance_meta_guard = rebalance_meta_lock
+        .get_write_lock(get_lock_acquire_timeout())
+        .await
+        .map_err(activation_rebalance_meta_lock_error)?;
    let pool_meta_lock = pool.new_ns_lock(RUSTFS_META_BUCKET, POOL_META_NAME).await?;
    let pool_meta_guard = pool_meta_lock
        .get_write_lock(get_lock_acquire_timeout())
        .await
        .map_err(activation_pool_meta_lock_error)?;
-    let rebalance_meta_lock = pool.new_ns_lock(RUSTFS_META_BUCKET, REBAL_META_NAME).await?;
-    let rebalance_meta_guard = rebalance_meta_lock
-        .get_write_lock(get_lock_acquire_timeout())
-        .await
-        .map_err(activation_rebalance_meta_lock_error)?;

    Ok(PoolRebalanceActivationFence {
        pool_meta_guard,
@@ -22094,7 +22103,7 @@ mod pools_tests {
                .resources
                .lock()
                .expect("activation lock recorder should not be poisoned"),
-            vec![POOL_META_NAME.to_string(), REBAL_META_NAME.to_string()]
+            vec![REBAL_META_NAME.to_string(), POOL_META_NAME.to_string()]
        );

        let mut second_acquire = Box::pin(acquire_pool_rebalance_activation_locks(second.clone(), None));
@@ -22110,10 +22119,50 @@ mod pools_tests {
                .resources
                .lock()
                .expect("activation lock recorder should not be poisoned"),
-            vec![POOL_META_NAME.to_string(), REBAL_META_NAME.to_string()]
+            vec![REBAL_META_NAME.to_string(), POOL_META_NAME.to_string()]
        );
    }

+    #[tokio::test]
+    async fn test_activation_cancellation_releases_rebalance_fence_while_pool_fence_is_contended() {
+        use crate::storage_api_contracts::namespace::NamespaceLocking as _;
+        let pool = Arc::new(ActivationLockRecorder {
+            lock_manager: Arc::new(rustfs_lock::GlobalLockManager::new()),
+            owner: "activation-cancellation",
+            resources: StdMutex::new(Vec::new()),
+        });
+        let pool_lock = pool
+            .new_ns_lock(crate::disk::RUSTFS_META_BUCKET, POOL_META_NAME)
+            .await
+            .expect("pool lock should be created");
+        let pool_reader = pool_lock
+            .get_read_lock(std::time::Duration::from_secs(5))
+            .await
+            .expect("ordinary mutation should hold the pool read fence");
+        pool.resources.lock().expect("recorder should not be poisoned").clear();
+        let mut activation = Box::pin(acquire_pool_rebalance_activation_locks(Arc::clone(&pool), None));
+        assert!(matches!(futures::poll!(&mut activation), Poll::Pending));
+        assert_eq!(
+            *pool.resources.lock().expect("recorder should not be poisoned"),
+            vec![REBAL_META_NAME.to_string(), POOL_META_NAME.to_string()],
+            "activation must hold the run fence before waiting for the pool fence",
+        );
+        drop(activation);
+        let rebalance_lock = pool
+            .new_ns_lock(crate::disk::RUSTFS_META_BUCKET, REBAL_META_NAME)
+            .await
+            .expect("run lock should be created");
+        let run_writer = rebalance_lock
+            .get_write_lock(std::time::Duration::from_secs(5))
+            .await
+            .expect("cancelling activation must release its already-acquired run fence");
+        assert!(
+            !pool_reader.is_released(),
+            "cancelling activation must not release another caller's pool fence"
+        );
+        assert!(!run_writer.is_lock_lost());
+    }
+
    #[test]
    fn decommission_receipt_run_token_changes_with_persisted_start_time() {
        let first = OffsetDateTime::from_unix_timestamp(1_000).expect("first run timestamp should be valid");
@@ -627,7 +627,13 @@ impl From<tokio::task::JoinError> for DiskError {
 impl Clone for DiskError {
    fn clone(&self) -> Self {
        match self {
-            DiskError::Io(io_error) => DiskError::Io(std::io::Error::new(io_error.kind(), io_error.to_string())),
+            DiskError::Io(io_error) => DiskError::Io(
+                rustfs_rio::clone_internode_http_io_error(io_error)
+                    .and_then(std::io::Error::into_inner)
+                    // The helper derives a kind from the source; Clone must retain the original outer kind.
+                    .map(|source| std::io::Error::new(io_error.kind(), source))
+                    .unwrap_or_else(|| std::io::Error::new(io_error.kind(), io_error.to_string())),
+            ),
            DiskError::MaxVersionsExceeded => DiskError::MaxVersionsExceeded,
            DiskError::Unexpected => DiskError::Unexpected,
            DiskError::CorruptedFormat => DiskError::CorruptedFormat,
@@ -1265,6 +1271,49 @@ mod tests {
        assert!(!bad_request.is_retryable_internode_write_failure());
    }

+    #[test]
+    fn test_internode_http_clone_preserves_retryability_status_and_context() {
+        use http::StatusCode;
+        use rustfs_rio::InternodeHttpErrorKind::{ConnectionRefused, ConnectionReset, HttpStatus, Unknown};
+
+        for (kind, retryable) in [
+            (ConnectionRefused, true),
+            (ConnectionReset, true),
+            (HttpStatus(StatusCode::TOO_MANY_REQUESTS), true),
+            (HttpStatus(StatusCode::SERVICE_UNAVAILABLE), true),
+            (HttpStatus(StatusCode::CONFLICT), true),
+            (Unknown, false),
+            (HttpStatus(StatusCode::BAD_REQUEST), false),
+            (HttpStatus(StatusCode::INTERNAL_SERVER_ERROR), false),
+        ] {
+            let original = DiskError::from(rustfs_rio::new_test_internode_http_io_error(kind));
+            assert_eq!(original.internode_http_error_kind(), Some(kind));
+            assert_eq!(original.is_retryable_internode_write_failure(), retryable);
+
+            let cloned = original.clone();
+            assert_eq!(cloned, original, "clone must preserve the error bucket for {kind:?}");
+            assert_eq!(
+                cloned.is_retryable_internode_write_failure(),
+                retryable,
+                "clone changed retryability for {kind:?}"
+            );
+            assert_eq!(cloned.internode_http_error_kind(), Some(kind));
+            if let HttpStatus(status) = kind {
+                assert!(cloned.is_internode_http_status(status.as_u16()));
+            }
+            let DiskError::Io(io_error) = &cloned else {
+                panic!("unmarked internode error must remain Io: {cloned:?}");
+            };
+            let source = io_error
+                .get_ref()
+                .and_then(|source| source.downcast_ref::<InternodeHttpError>())
+                .expect("clone must retain the structured internode error");
+            assert_eq!(source.context().method(), "PUT");
+            assert_eq!(source.context().target(), "/rustfs/rpc/put_file_stream");
+            assert_eq!(source.context().operation(), Some(INTERNODE_OPERATION_PUT_FILE_STREAM));
+        }
+    }
+
    #[tokio::test]
    async fn read_stream_conflict_is_not_a_retryable_put_file_failure() {
        use tokio::io::{AsyncReadExt, AsyncWriteExt};
@@ -1309,11 +1358,57 @@ mod tests {
                !error.is_retryable_internode_write_failure(),
                "read-operation 409 must not trigger put-file retry"
            );
+            let cloned = error.clone();
+            let reduced = crate::disk::error_reduce::reduce_write_quorum_errs(&[Some(error)], &[], 1)
+                .expect("the read conflict must remain the dominant error");
+            for preserved in [&cloned, &reduced] {
+                assert!(
+                    !preserved.is_retryable_internode_write_failure(),
+                    "cloning or reducing a read conflict must not turn it into a PUT retry"
+                );
+                assert!(preserved.is_internode_http_status(409));
+                let DiskError::Io(io_error) = preserved else {
+                    panic!("read conflict must remain Io: {preserved:?}");
+                };
+                let source = io_error
+                    .get_ref()
+                    .and_then(|source| source.downcast_ref::<InternodeHttpError>())
+                    .expect("read conflict must retain its request context");
+                assert_eq!(source.context().method(), "GET");
+                assert_eq!(source.context().target(), "/rustfs/rpc/read_file_stream");
+                assert_eq!(
+                    source.context().operation(),
+                    Some(rustfs_io_metrics::internode_metrics::INTERNODE_OPERATION_READ_FILE_STREAM)
+                );
+            }
        })
        .await
        .expect("isolated read-conflict test must finish within its budget");
    }

+    #[test]
+    fn test_internode_http_clone_preserves_outer_io_kind_and_message() {
+        let source = rustfs_rio::new_test_internode_http_io_error(InternodeHttpErrorKind::ConnectionReset)
+            .into_inner()
+            .expect("the internode helper must provide a typed source");
+        let original_io = io::Error::new(io::ErrorKind::InvalidData, source);
+        let message = original_io.to_string();
+        let original = DiskError::from(original_io);
+        assert_eq!(original.internode_http_error_kind(), Some(InternodeHttpErrorKind::ConnectionReset));
+        assert!(original.is_retryable_internode_write_failure());
+
+        let cloned = original.clone();
+        let reduced = crate::disk::error_reduce::reduce_write_quorum_errs(&[Some(original)], &[], 1)
+            .expect("the wrapped internode error must remain the dominant error");
+        for preserved in [&cloned, &reduced] {
+            let DiskError::Io(io_error) = preserved else {
+                panic!("the wrapped error must remain Io: {preserved:?}");
+            };
+            assert_eq!(io_error.kind(), io::ErrorKind::InvalidData);
+            assert_eq!(io_error.to_string(), message);
+        }
+    }
+
    #[test]
    fn test_internode_missing_errors_preserve_disk_error_types() {
        let file_missing = DiskError::from(rustfs_rio::new_test_remote_file_not_found_http_io_error());
@@ -1325,6 +1420,17 @@ mod tests {
        assert_eq!(file_missing, DiskError::FileNotFound);
        assert_eq!(volume_missing, DiskError::VolumeNotFound);
        assert!(matches!(unmarked_server_error, DiskError::Io(_)));
+        for missing in [file_missing, volume_missing] {
+            assert_eq!(missing.clone(), missing);
+            assert_eq!(
+                crate::disk::error_reduce::reduce_write_quorum_errs(
+                    &[Some(missing.clone()), Some(missing.clone()), None],
+                    &[],
+                    2
+                ),
+                Some(missing)
+            );
+        }
    }

    #[test]
@@ -226,6 +226,78 @@ mod tests {
        assert_eq!(res, Some(quorum_err));
    }

+    #[test]
+    fn test_write_quorum_reduction_preserves_internode_http_identity() {
+        use http::StatusCode;
+        use rustfs_rio::InternodeHttpErrorKind::{ConnectionRefused, HttpStatus, Unknown};
+
+        for (kind, retryable) in [
+            (ConnectionRefused, true),
+            (HttpStatus(StatusCode::SERVICE_UNAVAILABLE), true),
+            (HttpStatus(StatusCode::CONFLICT), true),
+            (Unknown, false),
+            (HttpStatus(StatusCode::BAD_REQUEST), false),
+        ] {
+            // Construct both producer errors independently: the reducer owns the first clone.
+            let first = Error::from(rustfs_rio::new_test_internode_http_io_error(kind));
+            let second = Error::from(rustfs_rio::new_test_internode_http_io_error(kind));
+            assert_eq!(first.internode_http_error_kind(), Some(kind));
+            assert_eq!(second.internode_http_error_kind(), Some(kind));
+            assert_eq!(first.is_retryable_internode_write_failure(), retryable);
+            let errors = [Some(first), Some(second), None];
+            let reduced = reduce_write_quorum_errs(&errors, OBJECT_OP_IGNORED_ERRS, 2)
+                .expect("two equal producer errors must dominate one successful write");
+
+            assert_eq!(Some(&reduced), errors[0].as_ref());
+            assert_eq!(
+                reduced.is_retryable_internode_write_failure(),
+                retryable,
+                "quorum reduction changed retryability for {kind:?}"
+            );
+            assert_eq!(reduced.internode_http_error_kind(), Some(kind));
+            if let HttpStatus(status) = kind {
+                assert!(reduced.is_internode_http_status(status.as_u16()));
+            }
+            let Error::Io(io_error) = &reduced else {
+                panic!("the dominant error must remain Io: {reduced:?}");
+            };
+            let source = io_error
+                .get_ref()
+                .and_then(|source| source.downcast_ref::<rustfs_rio::InternodeHttpError>())
+                .expect("quorum reduction must retain the structured internode error");
+            assert_eq!(source.context().method(), "PUT");
+            assert_eq!(source.context().target(), "/rustfs/rpc/put_file_stream");
+            assert_eq!(
+                source.context().operation(),
+                Some(rustfs_io_metrics::internode_metrics::INTERNODE_OPERATION_PUT_FILE_STREAM)
+            );
+        }
+    }
+
+    #[test]
+    fn test_clone_and_write_quorum_do_not_promote_non_retryable_errors() {
+        use http::StatusCode;
+        use rustfs_rio::InternodeHttpErrorKind::{HttpStatus, Unknown};
+
+        for original in [
+            Error::from(rustfs_rio::new_test_internode_http_io_error(Unknown)),
+            Error::from(rustfs_rio::new_test_internode_http_io_error(HttpStatus(StatusCode::BAD_REQUEST))),
+            Error::from(rustfs_rio::new_test_internode_http_io_error(HttpStatus(StatusCode::FORBIDDEN))),
+            Error::from(rustfs_rio::new_test_internode_http_io_error(HttpStatus(StatusCode::NOT_FOUND))),
+            Error::from(rustfs_rio::new_test_internode_http_io_error(HttpStatus(
+                StatusCode::INTERNAL_SERVER_ERROR,
+            ))),
+            err_io("internode connection reset: PUT /rustfs/rpc/put_file_stream"),
+        ] {
+            assert!(!original.is_retryable_internode_write_failure());
+            let cloned = original.clone();
+            let reduced =
+                reduce_write_quorum_errs(&[Some(original)], &[], 1).expect("a non-retryable error must remain an error");
+            assert!(!cloned.is_retryable_internode_write_failure());
+            assert!(!reduced.is_retryable_internode_write_failure());
+        }
+    }
+
    #[test]
    fn test_count_errs() {
        let e1 = err_io("a");
@@ -572,7 +572,7 @@ impl ECStore {
    where
        S: EcstoreObjectIO + StorageNamespaceLocking<Error = Error, NamespaceLock = rustfs_lock::NamespaceLockWrapper>,
    {
-        // Lock order: pool_meta_save_gate -> pool.bin -> rebalance.bin.
+        // Lock order: pool_meta_save_gate -> rebalance.bin -> pool.bin.
        let mut pool_meta_guard = self.pool_meta_save_gate.lock().await;
        pool_meta_guard.ensure_write_safe("rebalance worker activation")?;
        // Classify the durable rebalance record while holding both namespace
@@ -50,6 +50,11 @@ fn ensure_rebalance_entry_active(cancel: &CancellationToken) -> Result<()> {
    Ok(())
 }

+#[cfg(test)]
+tokio::task_local! {
+    static REBALANCE_ENTRY_RUN_FENCE_BARRIER: (Arc<tokio::sync::Notify>, Arc<tokio::sync::Notify>);
+}
+
 #[derive(Debug)]
 struct RebalanceEntryTarget {
    bucket: String,
@@ -256,9 +261,15 @@ impl ECStore {
            .sort_by_key(|v| (v.mod_time.is_none(), std::cmp::Reverse(v.mod_time)));

        // Entry lock order is bucket incarnation -> activation_gate -> rebalance.bin -> movement gate.
+        // Target capacity admission can then acquire pool.bin under the run fence.
        // Stop waits for in-flight entries through cleanup, but not for entries admitted later.
        ensure_rebalance_entry_active(&cancel)?;
        let run_guard = self.rebalance_run_guard(rebalance_id.as_ref(), "rebalance entry").await?;
+        #[cfg(test)]
+        if let Ok((arrived, release)) = REBALANCE_ENTRY_RUN_FENCE_BARRIER.try_with(Clone::clone) {
+            arrived.notify_one();
+            release.notified().await;
+        }
        let lock_lost_signal = run_guard.lock_lost_signal();
        #[cfg(test)]
        let _run_signal_test_fence = lock_lost_signal
@@ -1237,6 +1248,130 @@ mod tests {
        assert_eq!(pool_stats.cleanup_warnings.count, 1, "deferred cleanup must not add a permanent warning");
    }

+    #[tokio::test]
+    #[serial_test::serial]
+    async fn real_rebalance_entry_progresses_while_peer_activation_waits_for_run_fence() {
+        const REBALANCE_ID: &str = "rebalance-peer-activation-lock-order";
+        let (_temp_dirs, store, peer) = crate::services::rebalance::test_two_pool_stores_with_isolated_node_contexts(Some(
+            active_rebalance_meta(REBALANCE_ID),
+        ))
+        .await;
+        assert!(!Arc::ptr_eq(&store.ctx, &peer.ctx), "node-local movement gates must be independent");
+        {
+            let mut meta = peer.rebalance_meta.write().await;
+            let meta = meta.as_mut().expect("peer should know the durable run");
+            meta.activation_gate = Arc::default();
+            meta.cancel = None;
+        }
+        let bucket = crate::disk::RUSTFS_META_BUCKET;
+        let object = "rebalance-peer-activation-object";
+        let version_id = uuid::Uuid::new_v4();
+        let payload = b"entry must drain before peer activation takes the pool fence".repeat(1024);
+        let source_set = store.pools[0].get_disks_by_key(object);
+        let target_set = store.pools[1].get_disks_by_key(object);
+        let opts = ObjectOptions {
+            versioned: true,
+            version_id: Some(version_id.to_string()),
+            ..Default::default()
+        };
+        let mut writer = PutObjReader::from_vec(payload.clone());
+        let source_before = source_set
+            .put_object(bucket, object, &mut writer, &opts)
+            .await
+            .expect("source version should be written");
+        let entry = metacache_entry_from_source(&source_set, bucket, object).await;
+        let arrived = Arc::new(tokio::sync::Notify::new());
+        let release = Arc::new(tokio::sync::Notify::new());
+        // JoinSet aborts both scoped tasks if an assertion or timeout fails.
+        let mut tasks = tokio::task::JoinSet::new();
+        let entry_store = Arc::clone(&store);
+        tasks.spawn(
+            REBALANCE_ENTRY_RUN_FENCE_BARRIER.scope((Arc::clone(&arrived), Arc::clone(&release)), async move {
+                entry_store
+                    .rebalance_entry(
+                        RebalanceEntryTarget {
+                            bucket: bucket.to_string(),
+                            pool_index: 0,
+                        },
+                        entry,
+                        source_set,
+                        Arc::new(RebalanceBucketConfigs::default()),
+                        Arc::from(REBALANCE_ID),
+                        CancellationToken::new(),
+                    )
+                    .await
+            }),
+        );
+        tokio::time::timeout(StdDuration::from_secs(30), arrived.notified())
+            .await
+            .expect("real entry must acquire its persisted run read fence");
+
+        let attempted = Arc::new(tokio::sync::Notify::new());
+        let peer_pool = Arc::clone(&peer.pools[0]);
+        let (activation_done, activation_result) = tokio::sync::oneshot::channel();
+        tasks.spawn(
+            crate::core::pools::REBALANCE_ACTIVATION_LOCK_ATTEMPT.scope(Arc::clone(&attempted), async move {
+                let result = peer.fence_rebalance_worker_activation(peer_pool, REBALANCE_ID).await;
+                let result = result.map(|fence| match fence {
+                    super::super::control::RebalanceWorkerActivationFence::Ready(fence) => {
+                        fence.ensure_held().expect("peer activation must retain both fences");
+                    }
+                    super::super::control::RebalanceWorkerActivationFence::NotStartedTerminal => {
+                        panic!("the paused entry's run must still require activation");
+                    }
+                });
+                activation_done.send(result).expect("activation receiver should remain alive");
+                Ok(RebalanceEntryOutcome::Completed)
+            }),
+        );
+        tokio::time::timeout(StdDuration::from_secs(30), attempted.notified())
+            .await
+            .expect("peer activation must attempt the persisted rebalance write fence");
+        release.notify_one();
+
+        tokio::time::timeout(StdDuration::from_secs(30), async {
+            while let Some(result) = tasks.join_next().await {
+                assert!(matches!(
+                    result
+                        .expect("scoped task must not panic")
+                        .expect("entry must not fail or defer"),
+                    RebalanceEntryOutcome::Completed
+                ));
+            }
+        })
+        .await
+        .expect("entry and peer activation must both make progress");
+        activation_result
+            .await
+            .expect("peer activation result should be sent")
+            .expect("peer activation must not time out behind the entry it blocks");
+
+        let mut reader = target_set
+            .get_object_reader(bucket, object, None, HeaderMap::new(), &opts)
+            .await
+            .expect("the exact target version must be readable");
+        let mut actual = Vec::new();
+        reader
+            .stream
+            .read_to_end(&mut actual)
+            .await
+            .expect("target body should drain completely");
+        assert_eq!(actual, payload);
+        assert_eq!(reader.object_info.version_id, source_before.version_id);
+        assert_eq!(reader.object_info.etag, source_before.etag);
+        assert_eq!(reader.object_info.mod_time, source_before.mod_time);
+        let source_error = store.pools[0]
+            .get_object_info(bucket, object, &opts)
+            .await
+            .expect_err("completed entry must clean up the source version");
+        assert!(crate::error::is_err_object_not_found(&source_error) || crate::error::is_err_version_not_found(&source_error));
+        let meta = store.rebalance_meta.read().await;
+        let stats = &meta.as_ref().expect("local run must remain installed").pool_stats[0];
+        assert_eq!(stats.num_objects, 1);
+        assert_eq!(stats.num_versions, 1);
+        assert_eq!(stats.cleanup_warnings.count, 0);
+    }
+
    #[tokio::test]
    #[serial_test::serial]
    async fn real_rebalance_run_fence_loss_before_target_commit_preserves_target_and_source() {
@@ -1907,6 +1907,124 @@ fn test_is_transient_rebalance_error_accepts_wrapped_disk_timeout() {
    assert!(is_transient_rebalance_error(&Error::Io(std::io::Error::other(DiskError::Timeout))));
 }

+#[test]
+fn test_rebalance_stage_wrapped_transient_errors_remain_retryable() {
+    let cases = [
+        Error::Lock(rustfs_lock::LockError::timeout(".rustfs.sys/pool.bin@latest", Duration::from_secs(5))),
+        Error::Lock(rustfs_lock::LockError::network(
+            "peer unavailable",
+            std::io::Error::from(std::io::ErrorKind::ConnectionReset),
+        )),
+        Error::SlowDown,
+        Error::ErasureReadQuorum,
+        Error::ErasureWriteQuorum,
+        Error::Io(std::io::Error::other(DiskError::Timeout)),
+        Error::Io(std::io::Error::from(std::io::ErrorKind::TimedOut)),
+    ];
+    for mut error in cases {
+        for depth in 0..=3 {
+            assert!(is_transient_rebalance_error(&error), "transient source lost at depth {depth}: {error:?}");
+            assert!(
+                should_defer_rebalance_entry_failure(&error),
+                "exhausted transient entries must be deferred"
+            );
+            assert!(should_retry_rebalance_listing(&error, 0, 3));
+            assert!(
+                !should_retry_rebalance_listing(&error, 2, 3),
+                "wrapping must not bypass the attempt limit"
+            );
+            error = data_movement::data_movement_stage_error_for_test(
+                "rebalance_object",
+                "put_object",
+                "bucket",
+                "baseline/00042.bin",
+                error,
+            );
+        }
+    }
+}
+
+#[test]
+fn test_rebalance_stage_wrapped_terminal_errors_remain_terminal() {
+    let cases = [
+        Error::FileAccessDenied,
+        Error::FileCorrupt,
+        Error::OperationCanceled,
+        Error::DataMovementOverwriteErr("bucket".to_string(), "object".to_string(), "version".to_string()),
+        Error::Lock(rustfs_lock::LockError::already_locked("bucket/object", "owner")),
+        Error::other("permission denied"),
+    ];
+    for mut error in cases {
+        for depth in 0..=3 {
+            assert!(
+                !is_transient_rebalance_error(&error),
+                "terminal source must survive depth {depth}: {error:?}"
+            );
+            assert!(!should_defer_rebalance_entry_failure(&error));
+            // Object names are untrusted context, not evidence of a transient failure.
+            error = data_movement::data_movement_stage_error_for_test(
+                "rebalance_object",
+                "put_object",
+                "bucket",
+                "remote lock rpc timed out",
+                error,
+            );
+        }
+    }
+}
+
+#[tokio::test]
+async fn test_rebalance_stage_wrapped_lock_timeout_retries_real_migration_loop() {
+    for succeeds_on_retry in [true, false] {
+        let backend = MigrationBackendSpy::new(None, None);
+        let attempts = AtomicUsize::new(0);
+        let waits = AtomicUsize::new(0);
+        let mut transfer = |_, _, _| {
+            let attempt = attempts.fetch_add(1, Ordering::SeqCst);
+            async move {
+                if succeeds_on_retry && attempt > 0 {
+                    return Ok(());
+                }
+                Err(data_movement::data_movement_stage_error_for_test(
+                    "rebalance_object",
+                    "put_object",
+                    "bucket",
+                    "baseline/00042.bin",
+                    Error::Lock(rustfs_lock::LockError::timeout(".rustfs.sys/pool.bin@latest", Duration::from_secs(5))),
+                ))
+            }
+        };
+        let version = version_normal();
+        let result = migrate_entry_version_with_retry_wait(
+            &backend,
+            "bucket".to_string(),
+            0,
+            &version,
+            None,
+            3,
+            false,
+            &mut transfer,
+            |_: String, _: String, _: ObjectOptions| async { Ok::<_, Error>(ObjectInfo::default()) },
+            |_| {
+                waits.fetch_add(1, Ordering::SeqCst);
+                std::future::ready(())
+            },
+        )
+        .await;
+        assert_eq!(result.moved, succeeds_on_retry);
+        assert_eq!(result.failed, !succeeds_on_retry);
+        assert_eq!(attempts.load(Ordering::SeqCst), if succeeds_on_retry { 2 } else { 3 });
+        assert_eq!(backend.get_calls(), attempts.load(Ordering::SeqCst));
+        assert_eq!(waits.load(Ordering::SeqCst), attempts.load(Ordering::SeqCst) - 1);
+        if !succeeds_on_retry {
+            assert_eq!(result.stage, Some("write_target"));
+            assert!(should_defer_rebalance_entry_failure(
+                result.error.as_ref().expect("exhaustion must retain its source error")
+            ));
+        }
+    }
+}
+
 #[test]
 fn test_is_transient_rebalance_error_accepts_io_timeout_message() {
    assert!(is_transient_rebalance_error(&Error::Io(std::io::Error::other("timeout"))));
@@ -244,6 +244,7 @@ pub(super) fn resolve_rebalance_bucket_result(
 }

 pub(super) fn is_transient_rebalance_error(err: &Error) -> bool {
+    let err = rebalance_error_source(err);
    match err {
        Error::SlowDown
        | Error::ErasureReadQuorum
@@ -256,6 +257,15 @@ pub(super) fn is_transient_rebalance_error(err: &Error) -> bool {
    }
 }

+fn rebalance_error_source(mut err: &Error) -> &Error {
+    // Stage context contains object names, so classify the preserved source,
+    // not timeout-like text supplied by an object name. Iterate nested stages.
+    while let Some(source) = crate::data_movement::data_movement_stage_source(err) {
+        err = source;
+    }
+    err
+}
+
 fn is_rebalance_transient_lock_error(err: &rustfs_lock::LockError) -> bool {
    match err {
        rustfs_lock::LockError::Timeout { .. } | rustfs_lock::LockError::Network { .. } => true,
@@ -309,6 +319,7 @@ pub(super) fn rebalance_listing_retry_delay(attempt: usize) -> Duration {
 }

 fn is_rebalance_lock_or_rpc_timeout(err: &Error) -> bool {
+    let err = rebalance_error_source(err);
    match err {
        Error::Lock(rustfs_lock::LockError::Timeout { .. }) | Error::Lock(rustfs_lock::LockError::Network { .. }) => true,
        Error::Io(io_err) => is_rebalance_lock_or_rpc_timeout_message(&io_err.to_string()),
@@ -585,3 +596,48 @@ impl SetDisks {
        Ok(())
    }
 }
+
+#[cfg(test)]
+mod error_source_tests {
+    use super::*;
+
+    #[test]
+    fn stage_wrapped_errors_select_the_source_backoff_policy() {
+        let cases = [
+            (
+                Error::Lock(rustfs_lock::LockError::timeout(".rustfs.sys/pool.bin@latest", Duration::from_secs(5))),
+                true,
+            ),
+            (
+                Error::Lock(rustfs_lock::LockError::network(
+                    "peer unavailable",
+                    std::io::Error::from(std::io::ErrorKind::ConnectionReset),
+                )),
+                true,
+            ),
+            (Error::other("remote lock rpc timed out"), true),
+            (Error::SlowDown, false),
+            (Error::Io(std::io::Error::other(DiskError::Timeout)), false),
+            (Error::FileAccessDenied, false),
+        ];
+        for (mut error, lock_backoff) in cases {
+            for depth in 0..=3 {
+                assert_eq!(
+                    is_rebalance_lock_or_rpc_timeout(&error),
+                    lock_backoff,
+                    "wrong backoff at depth {depth}: {error:?}"
+                );
+                if !lock_backoff {
+                    assert_eq!(rebalance_migration_retry_delay(1, &error), REBALANCE_MIGRATION_RETRY_BASE_DELAY * 2);
+                }
+                error = crate::data_movement::data_movement_stage_error_for_test(
+                    "rebalance_object",
+                    "put_object",
+                    "bucket",
+                    "remote lock rpc timed out",
+                    error,
+                );
+            }
+        }
+    }
+}
@@ -326,14 +326,15 @@ impl SetDisks {
        let parity_blocks = Self::common_parity(&parities, default_parity_count as i32);

        if parity_blocks < 0 {
-            // No parity value reached read quorum. Distinguish two cases:
-            // enough disks answered with valid-looking metadata that simply
-            // cannot be reconciled (corrupt/foreign entries — retrying cannot
-            // help, and heal should see Corrupt, rustfs#5801) versus too few
-            // healthy answers (a genuine quorum condition where retry may
-            // succeed once disks recover).
+            // A consistent layout can require more replies than the initial
+            // half-set probe. Reaching that probe alone is not corruption;
+            // only invalid or conflicting healthy replies establish that.
            let healthy_replies = errs.iter().filter(|err| err.is_none()).count();
-            if healthy_replies >= expected_rquorum {
+            let consistent_parity = parities
+                .iter()
+                .find(|&&parity| parity >= 0)
+                .filter(|&&parity| parities.iter().filter(|&&candidate| candidate == parity).count() == healthy_replies);
+            if healthy_replies >= expected_rquorum && consistent_parity.is_none() {
                error!(
                    "object_quorum_from_meta: irreconcilable parity across {healthy_replies} healthy replies (corrupt metadata), errs={errs:?}"
                );
@@ -1652,6 +1653,40 @@ mod tests {
        assert_eq!(err, DiskError::FileCorrupt);
    }

+    #[test]
+    fn consistent_parity_below_its_data_shard_quorum_is_not_corruption() {
+        for (drive_count, parity) in [(6, 2), (8, 2), (12, 4)] {
+            let data = drive_count - parity;
+            let mut metas = (1..=drive_count)
+                .map(|index| {
+                    let mut info = FileInfo::new("bucket/object", data, parity);
+                    info.size = 1024;
+                    info.erasure.index = index;
+                    info
+                })
+                .collect::<Vec<_>>();
+            let mut errs = vec![Some(DiskError::DiskNotFound); drive_count];
+            errs[..data].fill(None);
+            assert_eq!(
+                SetDisks::object_quorum_from_meta(&metas, &errs, parity).expect("exact data quorum should resolve"),
+                (data as i32, data as i32)
+            );
+
+            errs[data - 1] = Some(DiskError::DiskNotFound);
+            assert_eq!(
+                SetDisks::object_quorum_from_meta(&metas, &errs, parity).expect_err("one fewer shard cannot resolve"),
+                DiskError::ErasureReadQuorum,
+                "layout {drive_count}/{parity} has consistent metadata but insufficient shards"
+            );
+
+            metas[0].erasure.parity_blocks = usize::MAX;
+            assert_eq!(
+                SetDisks::object_quorum_from_meta(&metas, &errs, parity).expect_err("corrupt healthy replies must be rejected"),
+                DiskError::FileCorrupt
+            );
+        }
+    }
+
    /// Too few healthy replies remains a genuine quorum condition where a
    /// retry may succeed once disks recover.
    #[test]
@@ -865,6 +865,7 @@ pub(crate) use core::io_primitives::{ENV_RUSTFS_PUT_RENAME_EARLY_ACK_ENABLE, ren
 mod ctx;
 mod metadata;
 mod ops;
+pub(crate) use ops::bucket::BucketInfoQuorum;

 #[cfg(test)]
 pub(crate) use ops::hermetic_set_disks_isolated;
@@ -21,12 +21,72 @@

 use super::super::{
    BUCKET_OP_IGNORED_ERRS, BucketInfo, BucketOperations, BucketOptions, DeleteBucketOptions, DiskError, Error, HashMap,
-    MakeBucketOptions, Result, SetDisks, is_reserved_or_invalid_bucket, join_all, reduce_write_quorum_errs,
+    MakeBucketOptions, Result, SetDisks, is_reserved_or_invalid_bucket, join_all, reduce_read_quorum_errs,
+    reduce_write_quorum_errs,
 };
 use crate::api::bucket::metadata_sys;
 use crate::disk::DiskAPI;

+#[derive(Clone, Copy)]
+pub(crate) enum BucketInfoQuorum {
+    Read,
+    Write,
+}
+
 impl SetDisks {
+    pub(crate) async fn stat_bucket_with_quorum(&self, bucket: &str, quorum: BucketInfoQuorum) -> Result<BucketInfo> {
+        let disks = self.disk_inventory().await;
+        let disk_count = disks.len();
+        let mut futures = Vec::with_capacity(disk_count);
+        for disk in disks {
+            let bucket = bucket.to_string();
+            futures.push(async move {
+                match disk {
+                    Some(disk) => disk.stat_volume(&bucket).await,
+                    None => Err(DiskError::DiskNotFound),
+                }
+            });
+        }
+
+        let results = join_all(futures).await;
+        let mut infos = Vec::with_capacity(results.len());
+        let mut errs = Vec::with_capacity(results.len());
+        for result in results {
+            match result {
+                Ok(info) => {
+                    infos.push(Some(info));
+                    errs.push(None);
+                }
+                Err(err) => {
+                    infos.push(None);
+                    errs.push(Some(err));
+                }
+            }
+        }
+
+        let error = match quorum {
+            // Bucket mutations use a majority regardless of object storage
+            // class. A namespace read must intersect that majority; object
+            // readers still enforce the persisted layout's data-shard quorum.
+            BucketInfoQuorum::Read => reduce_read_quorum_errs(&errs, BUCKET_OP_IGNORED_ERRS, disk_count.div_ceil(2).max(1)),
+            BucketInfoQuorum::Write => reduce_write_quorum_errs(&errs, BUCKET_OP_IGNORED_ERRS, disk_count / 2 + 1),
+        };
+        if let Some(err) = error {
+            return Err(err.into());
+        }
+
+        infos
+            .into_iter()
+            .flatten()
+            .next()
+            .map(|info| BucketInfo {
+                name: info.name,
+                created: info.created,
+                ..Default::default()
+            })
+            .ok_or(Error::VolumeNotFound)
+    }
+
    pub(crate) async fn list_bucket_for_scanner(&self, _opts: &BucketOptions) -> Result<(Vec<BucketInfo>, bool)> {
        let disks = self.disk_inventory().await;
        let write_quorum = (disks.len() / 2) + 1;
@@ -131,59 +191,12 @@ impl BucketOperations for SetDisks {

    #[tracing::instrument(skip(self))]
    async fn get_bucket_info(&self, bucket: &str, _opts: &BucketOptions) -> Result<BucketInfo> {
-        let disks = self.disk_inventory().await;
-        let write_quorum = (disks.len() / 2) + 1;
-
-        let mut futures = Vec::with_capacity(disks.len());
-        for disk in disks {
-            let bucket = bucket.to_string();
-            futures.push(async move {
-                match disk {
-                    Some(disk) => disk.stat_volume(&bucket).await,
-                    None => Err(DiskError::DiskNotFound),
-                }
-            });
-        }
-
-        let results = join_all(futures).await;
-        let mut infos = Vec::with_capacity(results.len());
-        let mut errs = Vec::with_capacity(results.len());
-        for result in results {
-            match result {
-                Ok(info) => {
-                    infos.push(Some(info));
-                    errs.push(None);
-                }
-                Err(err) => {
-                    infos.push(None);
-                    errs.push(Some(err));
-                }
-            }
-        }
-
-        if let Some(err) = reduce_write_quorum_errs(&errs, BUCKET_OP_IGNORED_ERRS, write_quorum) {
-            return Err(err.into());
-        }
-
-        let mut versioning = false;
-        let mut object_locking = false;
+        let mut info = self.stat_bucket_with_quorum(bucket, BucketInfoQuorum::Write).await?;
        if let Ok(sys) = metadata_sys::get(bucket).await {
-            versioning = sys.versioning();
-            object_locking = sys.object_locking();
+            info.versioning = sys.versioning();
+            info.object_locking = sys.object_locking();
        }
-
-        infos
-            .into_iter()
-            .flatten()
-            .next()
-            .map(|info| BucketInfo {
-                name: info.name,
-                created: info.created,
-                versioning,
-                object_locking,
-                ..Default::default()
-            })
-            .ok_or(Error::VolumeNotFound)
+        Ok(info)
    }

    #[tracing::instrument(skip(self))]
@@ -19,7 +19,7 @@ use crate::bucket::{
 };
 use crate::error::is_err_bucket_not_found;
 use crate::runtime::sources as runtime_sources;
-use crate::set_disk::get_lock_acquire_timeout;
+use crate::set_disk::{BucketInfoQuorum, get_lock_acquire_timeout};
 use crate::storage_api_contracts::bucket::{BUCKET_LIFECYCLE_LOCK_OBJECT, SRBucketDeleteOp};
 use crate::storage_api_contracts::namespace::NamespaceLocking as _;
 use futures::stream::{self, StreamExt};
@@ -772,17 +772,30 @@ impl ECStore {

    #[instrument(skip(self))]
    pub(crate) async fn get_bucket_info_from_sets(&self, bucket: &str, opts: &BucketOptions) -> Result<BucketInfo> {
+        self.get_bucket_info_from_sets_with_quorum(bucket, opts, BucketInfoQuorum::Write)
+            .await
+    }
+
+    async fn get_bucket_info_from_sets_with_quorum(
+        &self,
+        bucket: &str,
+        opts: &BucketOptions,
+        quorum: BucketInfoQuorum,
+    ) -> Result<BucketInfo> {
        // One host may participate in several pools after expansion. Resolve the
        // namespace against each erasure set so disks from different pools can
        // never be combined into one bucket quorum.
        // Bucket validation is request-path IO. Keep the previous peer fanout's
        // latency shape by probing every set concurrently; scanner listings use
        // a separate bounded path below because they run continuously.
-        let mut scoped_results =
-            futures::future::join_all(self.bucket_sets().map(|(pool_index, set_index, set)| async move {
-                (pool_index, set_index, set.get_bucket_info(bucket, opts).await)
-            }))
-            .await;
+        let mut scoped_results = futures::future::join_all(self.bucket_sets().map(|(pool_index, set_index, set)| async move {
+            let result = match quorum {
+                BucketInfoQuorum::Read => set.stat_bucket_with_quorum(bucket, quorum).await,
+                BucketInfoQuorum::Write => set.get_bucket_info(bucket, opts).await,
+            };
+            (pool_index, set_index, result)
+        }))
+        .await;
        scoped_results.sort_unstable_by_key(|(pool_index, set_index, _)| (*pool_index, *set_index));

        let mut first_info = None;
@@ -806,7 +819,11 @@ impl ECStore {

    #[instrument(skip(self))]
    pub(super) async fn handle_get_bucket_info(&self, bucket: &str, opts: &BucketOptions) -> Result<BucketInfo> {
-        let mut info = self.get_bucket_info_from_sets(bucket, opts).await?;
+        let mut info = match self.get_bucket_info_from_sets(bucket, opts).await {
+            Ok(info) => info,
+            Err(Error::ErasureWriteQuorum) => return self.get_bucket_info_at_read_quorum(bucket, opts).await,
+            Err(err) => return Err(err),
+        };

        if let Ok(sys) = metadata_sys::get_in(&self.ctx, bucket).await {
            if should_override_created_from_metadata(sys.created) {
@@ -819,6 +836,35 @@ impl ECStore {
        Ok(info)
    }

+    async fn get_bucket_info_at_read_quorum(&self, bucket: &str, opts: &BucketOptions) -> Result<BucketInfo> {
+        // Lock order: bucket lifecycle -> internal metadata object read locks.
+        // Keep create/delete from changing the namespace while a read quorum
+        // confirms both physical presence and persisted bucket metadata.
+        let guard = self.acquire_bucket_lifecycle_read_lock(bucket).await?;
+        await_bucket_namespace_operation(Some(&guard), bucket, "bucket read quorum validation", async {
+            let mut info = self
+                .get_bucket_info_from_sets_with_quorum(bucket, opts, BucketInfoQuorum::Read)
+                .await?;
+            let (metadata, persisted) = metadata_sys::get_config_from_disk_with_presence_in(&self.ctx, bucket).await?;
+            if !persisted {
+                // A minority of directories left by failed creation is not an
+                // authoritative bucket. Never turn fabricated defaults into
+                // permission to serve degraded reads.
+                return Err(Error::ErasureReadQuorum);
+            }
+            if metadata.name != bucket {
+                return Err(Error::FileCorrupt);
+            }
+            if should_override_created_from_metadata(metadata.created) {
+                info.created = Some(metadata.created);
+            }
+            info.versioning = metadata.versioning();
+            info.object_locking = metadata.object_locking();
+            Ok(info)
+        })
+        .await
+    }
+
    #[instrument(skip(self))]
    pub(super) async fn handle_list_bucket(&self, opts: &BucketOptions) -> Result<Vec<BucketInfo>> {
        // TODO(backlog): support cached bucket listing via opts.cached
@@ -1049,7 +1095,7 @@ mod tests {
        run_physical_bucket_deletion, scan_metadata_less_residue, scan_metadata_less_residue_with_budget,
        should_override_created_from_metadata, validate_table_bucket_delete_allowed,
    };
-    use crate::bucket::metadata::table_bucket_catalog_metadata_prefix;
+    use crate::bucket::metadata::{BucketMetadata, table_bucket_catalog_metadata_prefix};
    use crate::bucket::metadata_sys;
    use crate::cluster::rpc::peer_s3_client::install_delete_bucket_empty_scan_barrier;
    use crate::disk::{BUCKET_META_PREFIX, DiskAPI, RUSTFS_META_BUCKET, STORAGE_FORMAT_FILE};
@@ -1076,6 +1122,7 @@ mod tests {
    use std::sync::atomic::{AtomicBool, Ordering};
    use std::time::{Duration, SystemTime};
    use time::OffsetDateTime;
+    use tokio::io::AsyncReadExt;
    use tokio::sync::{Notify, OnceCell};
    use tokio_util::sync::CancellationToken;
    use uuid::Uuid;
@@ -1359,11 +1406,18 @@ mod tests {
    }

    async fn setup_multi_pool_bucket_test_env() -> (tempfile::TempDir, Arc<ECStore>) {
+        setup_bucket_quorum_test_env(&[4, 4], None).await
+    }
+
+    async fn setup_bucket_quorum_test_env(
+        drives_per_pool: &[usize],
+        standard_parity: Option<usize>,
+    ) -> (tempfile::TempDir, Arc<ECStore>) {
        let temp_dir = tempfile::tempdir().expect("multi-pool bucket test directory should be created");
        let mut pools = Vec::new();
-        for pool_index in 0..2 {
+        for (pool_index, &drive_count) in drives_per_pool.iter().enumerate() {
            let mut endpoints = Vec::new();
-            for disk_index in 0..4 {
+            for disk_index in 0..drive_count {
                let disk_path = temp_dir.path().join(format!("pool{pool_index}-disk{disk_index}"));
                tokio::fs::create_dir_all(&disk_path)
                    .await
@@ -1378,7 +1432,7 @@ mod tests {
            pools.push(PoolEndpoints {
                legacy: false,
                set_count: 1,
-                drives_per_set: 4,
+                drives_per_set: drive_count,
                endpoints: Endpoints::from(endpoints),
                cmd_line: format!("bucket-test-pool-{pool_index}"),
                platform: format!("OS: {} | Arch: {}", std::env::consts::OS, std::env::consts::ARCH),
@@ -1399,9 +1453,12 @@ mod tests {
        )
        .await
        .expect("multi-pool ECStore should initialize");
-        let storage_class =
-            crate::config::storageclass::lookup_config_for_pools_without_env(&rustfs_config::server_config::KVS::new(), &[4, 4])
-                .expect("multi-pool storage class should match both four-disk pools");
+        let mut storage_class_kvs = rustfs_config::server_config::KVS::new();
+        if let Some(parity) = standard_parity {
+            storage_class_kvs.insert(crate::config::storageclass::CLASS_STANDARD.to_string(), format!("EC:{parity}"));
+        }
+        let storage_class = crate::config::storageclass::lookup_config_for_pools_without_env(&storage_class_kvs, drives_per_pool)
+            .expect("storage class should match every test erasure set");
        for pool in &ecstore.pools {
            for set in &pool.disk_set {
                set.set_test_storage_class_config(storage_class.clone());
@@ -2067,6 +2124,218 @@ mod tests {
        }
    }

+    #[tokio::test]
+    #[serial]
+    async fn bucket_info_read_quorum_tracks_erasure_layout() {
+        for (drive_count, parity) in [(2, 1), (3, 1), (4, 2), (5, 2), (6, 3), (8, 4), (6, 2), (12, 6)] {
+            let (_temp_dir, store) = setup_bucket_quorum_test_env(&[drive_count], Some(parity)).await;
+            metadata_sys::init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+            let bucket = format!("read-quorum-{drive_count}-{parity}");
+            let object = "uncached-object";
+            let body = b"erasure read quorum must follow the persisted layout".repeat(32_768);
+            store
+                .make_bucket(&bucket, &MakeBucketOptions::default())
+                .await
+                .expect("healthy namespace should accept bucket creation");
+            store
+                .put_object(&bucket, object, &mut PutObjReader::from_vec(body.clone()), &ObjectOptions::default())
+                .await
+                .expect("healthy erasure set should accept the seed object");
+            let set = &store.pools[0].disk_set[0];
+            let lock = set
+                .new_ns_lock(&bucket, object)
+                .await
+                .expect("seed namespace lock should resolve");
+            drop(
+                lock.get_write_lock(Duration::from_secs(30))
+                    .await
+                    .expect("seed physical fanout must finish before taking disks offline"),
+            );
+            if (drive_count, parity) == (6, 3) {
+                let mut kvs = rustfs_config::server_config::KVS::new();
+                kvs.insert(crate::config::storageclass::CLASS_STANDARD.to_string(), "EC:2".to_string());
+                set.set_test_storage_class_config(
+                    crate::config::storageclass::lookup_config_for_pools_without_env(&kvs, &[drive_count])
+                        .expect("a later storage-class change must not raise old objects' read quorum"),
+                );
+            }
+
+            let offline_indexes = (0..parity).collect::<Vec<_>>();
+            let offline = take_set_disks_offline(&store, set, &offline_indexes).await;
+            let info = store
+                .get_bucket_info(&bucket, &BucketOptions::default())
+                .await
+                .expect("bucket validation must admit the object's exact read quorum");
+            assert_eq!(info.name, bucket);
+
+            let mut reader = store
+                .get_object_reader(&bucket, object, None, Default::default(), &ObjectOptions::default())
+                .await
+                .expect("the persisted layout should remain readable at its exact data-shard quorum");
+            let mut restored = Vec::new();
+            reader
+                .stream
+                .read_to_end(&mut restored)
+                .await
+                .expect("quorum read should reconstruct the body");
+            assert_eq!(restored, body, "layout {drive_count}/{parity} must retain exact object contents");
+            drop(reader);
+
+            if drive_count - parity == drive_count / 2 {
+                let error = store
+                    .get_bucket_info_from_sets(&bucket, &BucketOptions::default())
+                    .await
+                    .expect_err("bucket mutations must retain their majority namespace check");
+                assert_eq!(error, StorageError::ErasureWriteQuorum);
+            }
+
+            let below_quorum = take_set_disks_offline(&store, set, &[parity]).await;
+            let read = store
+                .get_object_reader(&bucket, object, None, Default::default(), &ObjectOptions::default())
+                .await;
+            match read {
+                Ok(mut reader) => assert!(
+                    reader.stream.read_to_end(&mut Vec::new()).await.is_err(),
+                    "layout {drive_count}/{parity} must reject fewer than its data-shard quorum"
+                ),
+                Err(error) => assert!(
+                    matches!(error, StorageError::ErasureReadQuorum | StorageError::InsufficientReadQuorum(_, _)),
+                    "a missing shard must report read quorum loss, got {error}"
+                ),
+            }
+            restore_set_disks(&store, set, below_quorum).await;
+            restore_set_disks(&store, set, offline).await;
+        }
+    }
+
+    #[tokio::test]
+    #[serial]
+    async fn bucket_info_read_quorum_is_scoped_to_each_erasure_set() {
+        let (_temp_dir, store) = setup_bucket_quorum_test_env(&[4, 6], None).await;
+        metadata_sys::init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+        let bucket = "read-quorum-mixed-pools";
+        store
+            .make_bucket(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("healthy pools should accept bucket creation");
+
+        let first_set = &store.pools[0].disk_set[0];
+        let second_set = &store.pools[1].disk_set[0];
+        let first_offline = take_set_disks_offline(&store, first_set, &[0, 1]).await;
+        let second_offline = take_set_disks_offline(&store, second_set, &[0, 1, 2]).await;
+        store
+            .get_bucket_info(bucket, &BucketOptions::default())
+            .await
+            .expect("each set independently satisfies its namespace read quorum");
+
+        for (set, extra_disk) in [(first_set, 2), (second_set, 3)] {
+            let extra_offline = take_set_disks_offline(&store, set, &[extra_disk]).await;
+            assert_eq!(
+                store
+                    .get_bucket_info(bucket, &BucketOptions::default())
+                    .await
+                    .expect_err("another pool must not subsidize a set below its read quorum"),
+                StorageError::ErasureReadQuorum
+            );
+            restore_set_disks(&store, set, extra_offline).await;
+        }
+        restore_set_disks(&store, first_set, first_offline).await;
+        restore_set_disks(&store, second_set, second_offline).await;
+    }
+
+    #[tokio::test]
+    #[serial]
+    async fn bucket_info_read_quorum_requires_authoritative_metadata() {
+        for state in ["missing", "corrupt", "foreign", "incarnation"] {
+            let (_temp_dir, store) = setup_bucket_quorum_test_env(&[4], None).await;
+            metadata_sys::init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+            let bucket = format!("read-quorum-{state}-metadata");
+            let mut metadata = if state == "missing" {
+                store
+                    .make_bucket_on_sets(&bucket, &MakeBucketOptions::default())
+                    .await
+                    .expect("simulate directories left before bucket metadata is published");
+                BucketMetadata::new(&bucket)
+            } else {
+                store
+                    .make_bucket(&bucket, &MakeBucketOptions::default())
+                    .await
+                    .expect("healthy bucket should publish metadata");
+                metadata_sys::get_in(&store.ctx, &bucket)
+                    .await
+                    .expect("seed metadata should be cached")
+                    .as_ref()
+                    .clone()
+            };
+            let path = metadata.save_file_path();
+            match state {
+                "corrupt" => crate::config::com::save_config(store.clone(), &path, b"corrupt".to_vec())
+                    .await
+                    .expect("persist corrupt metadata while the cached copy remains valid"),
+                "foreign" => {
+                    metadata.name = "different-bucket".to_string();
+                    let mut encoded = vec![1, 0, 1, 0];
+                    encoded.extend(metadata.marshal_msg().expect("foreign metadata should encode"));
+                    crate::config::com::save_config(store.clone(), &path, encoded)
+                        .await
+                        .expect("persist metadata for a different bucket at the requested path");
+                }
+                "incarnation" => crate::bucket::metadata::save_bucket_incarnation(store.clone(), &bucket, Uuid::new_v4())
+                    .await
+                    .expect("persist a different bucket generation"),
+                _ => {}
+            }
+
+            let set = &store.pools[0].disk_set[0];
+            let offline = take_set_disks_offline(&store, set, &[0, 1]).await;
+            let error = store
+                .get_bucket_info(&bucket, &BucketOptions::default())
+                .await
+                .expect_err("read admission must not trust residual directories or cached metadata");
+            match state {
+                "missing" => assert_eq!(error, StorageError::ErasureReadQuorum),
+                "foreign" => assert_eq!(error, StorageError::FileCorrupt),
+                "incarnation" => assert!(error.to_string().contains("sidecar does not match bucket metadata")),
+                "corrupt" => assert!(error.to_string().contains("format invalid"), "unexpected corruption error: {error}"),
+                _ => unreachable!(),
+            }
+            restore_set_disks(&store, set, offline).await;
+        }
+    }
+
+    #[tokio::test]
+    #[serial]
+    async fn bucket_info_read_quorum_accepts_persisted_legacy_metadata() {
+        let (_temp_dir, store) = setup_bucket_quorum_test_env(&[4], None).await;
+        metadata_sys::init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+        let bucket = "interop";
+        store
+            .make_bucket_on_sets(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("legacy bucket directories should exist");
+        let hex = include_str!("../../tests/fixtures/minio/bucket_metadata.blob.hex")
+            .split_whitespace()
+            .collect::<String>();
+        let body = (0..hex.len())
+            .step_by(2)
+            .map(|index| u8::from_str_radix(&hex[index..index + 2], 16).expect("pinned MinIO metadata fixture"))
+            .collect();
+        crate::config::com::save_config(store.clone(), &BucketMetadata::new(bucket).save_file_path(), body)
+            .await
+            .expect("legacy metadata should be persisted without an incarnation sidecar");
+
+        let set = &store.pools[0].disk_set[0];
+        let offline = take_set_disks_offline(&store, set, &[0, 1]).await;
+        let info = store
+            .get_bucket_info(bucket, &BucketOptions::default())
+            .await
+            .expect("persisted MinIO metadata should authorize reads at the namespace read quorum");
+        assert_eq!(info.name, bucket);
+        assert!(info.versioning);
+        assert!(info.object_locking);
+        restore_set_disks(&store, set, offline).await;
+    }
+
    #[tokio::test]
    #[serial]
    async fn bucket_namespace_reads_report_missing_when_every_set_is_absent() {
@@ -2090,6 +2359,7 @@ mod tests {
    #[serial]
    async fn bucket_namespace_reads_fail_closed_when_any_set_loses_quorum() {
        let (_temp_dir, ecstore) = setup_multi_pool_bucket_test_env().await;
+        metadata_sys::init_bucket_metadata_sys(ecstore.clone(), Vec::new()).await;
        let bucket = format!("degraded-expansion-{}", Uuid::new_v4().simple());
        ecstore.pools[0].disk_set[0]
            .make_bucket(&bucket, &MakeBucketOptions::default())
@@ -2097,6 +2367,7 @@ mod tests {
            .expect("bucket should be created in the original pool only");
        ecstore.pools[1].disk_set[0].disks.write().await[0] = None;
        ecstore.pools[1].disk_set[0].disks.write().await[1] = None;
+        ecstore.pools[1].disk_set[0].disks.write().await[2] = None;

        let list_err = ecstore
            .list_bucket(&BucketOptions::default())
@@ -2108,7 +2379,7 @@ mod tests {
            .get_bucket_info(&bucket, &BucketOptions::default())
            .await
            .expect_err("bucket validation must fail when an expansion pool is unavailable");
-        assert_eq!(info_err, StorageError::ErasureWriteQuorum);
+        assert_eq!(info_err, StorageError::ErasureReadQuorum);
    }

    #[tokio::test]
@@ -577,13 +577,16 @@ impl KmsServiceManager {
        Some(service_version.probe_worker.as_ref()?.status())
    }

-    /// Health check for the KMS service
+    /// Check backend health without changing the service lifecycle state.
+    ///
+    /// A transient backend failure leaves the published service available for
+    /// subsequent checks and operations. Readiness uses the background probe
+    /// to evaluate backend availability independently of lifecycle state.
    pub async fn health_check(&self) -> Result<bool> {
        let checked_state = self.state.load_full();
        match checked_state.current_service.as_ref() {
            Some(service_version) => {
                let manager = service_version.manager.clone();
-                let checked_version = service_version.version;
                // Perform health check on the backend
                match manager.health_check().await {
                    Ok(healthy) => {
@@ -594,8 +597,6 @@ impl KmsServiceManager {
                    }
                    Err(e) => {
                        error!("KMS health check error: {}", e);
-                        let _guard = self.lifecycle_mutex.lock().await;
-                        self.mark_health_error_if_current(checked_version, &e);
                        Err(e)
                    }
                }
@@ -739,17 +740,6 @@ impl KmsServiceManager {
            task: std::sync::Mutex::new(Some(task)),
        }))
    }
-
-    fn mark_health_error_if_current(&self, checked_version: u64, error: &KmsError) {
-        let current = self.state.load_full();
-        if current.current_service.as_ref().map(|version| version.version) == Some(checked_version) {
-            self.state.store(Arc::new(RuntimeState {
-                config: current.config.clone(),
-                status: KmsServiceStatus::Error(format!("Health check failed: {error}")),
-                current_service: current.current_service.clone(),
-            }));
-        }
-    }
 }

 impl Default for KmsServiceManager {
@@ -1004,19 +994,6 @@ mod tests {
        assert!(manager.get_service_version().await.expect("restarted version") > first_version);
    }

-    #[tokio::test]
-    async fn stale_health_failure_cannot_poison_new_service_status() {
-        let manager = KmsServiceManager::new();
-        manager.configure(static_config("key-a", 0x11)).await.expect("configure");
-        manager.start().await.expect("start");
-        let old_version = manager.get_service_version().await.expect("old version");
-        manager.restart().await.expect("restart");
-
-        manager.mark_health_error_if_current(old_version, &KmsError::backend_error("stale failure"));
-
-        assert_eq!(manager.get_status().await, KmsServiceStatus::Running);
-    }
-
    #[tokio::test]
    async fn forbidden_local_master_key_change_preserves_running_config_and_service() {
        use crate::types::{CreateKeyRequest, KeyUsage};
@@ -75,6 +75,44 @@ fn unreachable_vault_config() -> KmsConfig {
    }
 }

+#[tokio::test]
+async fn transient_health_failure_does_not_latch_the_service_status() {
+    let kms = TestKms::local().await;
+    let manager = kms.manager();
+    let service = manager.get_encryption_service().await.expect("running service");
+    let version = manager.get_service_version().await.expect("running version");
+    assert!(manager.health_check().await.expect("initial backend health"));
+
+    // Move only this test's keys out of reach, then restore the same backend.
+    let key_dir = kms.key_dir().expect("local key directory");
+    let outage = tempfile::TempDir::new().expect("temporary outage directory");
+    let hidden_keys = outage.path().join("keys");
+    tokio::fs::rename(&key_dir, &hidden_keys)
+        .await
+        .expect("make backend unavailable");
+    let failure = manager.health_check().await;
+    let outage_status = manager.get_status().await;
+    tokio::fs::rename(&hidden_keys, &key_dir).await.expect("restore backend");
+
+    assert!(failure.is_err(), "the outage must surface as a health-check error");
+    assert!(manager.health_check().await.expect("backend recovers without restart"));
+    assert!(Arc::ptr_eq(
+        &service,
+        &manager.get_encryption_service().await.expect("service survives the outage")
+    ));
+    assert_eq!(manager.get_service_version().await, Some(version));
+    assert_eq!(
+        manager.get_status().await,
+        KmsServiceStatus::Running,
+        "a recovered backend must not leave service-status and readiness latched in Error"
+    );
+    assert_eq!(
+        outage_status,
+        KmsServiceStatus::Running,
+        "backend health does not change the running service's lifecycle state"
+    );
+}
+
 #[tokio::test]
 async fn starting_against_an_unreachable_backend_fails_without_publishing_a_service() {
    let manager = KmsServiceManager::new();
@@ -166,6 +166,24 @@ fn rule_replicates(rule: &ReplicationRule, obj: &ObjectOpts) -> bool {
    }
 }

+fn replication_filter_tags_match(filter: &s3s::dto::ReplicationRuleFilter, object_tags: &HashMap<String, String>) -> bool {
+    let tag_matches = |tag: &s3s::dto::Tag| match (&tag.key, &tag.value) {
+        (None, None) => true,
+        (Some(key), _) if key.is_empty() => true,
+        (Some(key), Some(value)) => object_tags.get(key) == Some(value),
+        _ => false,
+    };
+
+    filter
+        .and
+        .as_ref()
+        .and_then(|and| and.tags.as_deref())
+        .into_iter()
+        .flatten()
+        .chain(filter.tag.iter())
+        .all(tag_matches)
+}
+
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum ReplicationTargetValidationError {
    RoleWithMultipleDestinations,
@@ -704,7 +722,7 @@ impl ReplicationConfigurationExt for ReplicationConfiguration {

            if let Some(filter) = &rule.filter {
                let object_tags = ReplicationTagFilter::decode_tags_to_map(&obj.user_tags);
-                if filter.test_tags(&object_tags) {
+                if replication_filter_tags_match(filter, &object_tags) {
                    rules.push(rule.clone());
                }
            } else {
@@ -1139,6 +1157,47 @@ mod tests {
        assert_eq!(validate_replication_config_structure(&structure_config(vec![rule])), Ok(()));
    }

+    #[test]
+    fn actionable_rules_require_every_and_tag_to_match() {
+        let mut rule = replication_rule("rule-1", "arn:target:a");
+        rule.filter = Some(s3s::dto::ReplicationRuleFilter {
+            and: Some(s3s::dto::ReplicationRuleAndOperator {
+                prefix: None,
+                tags: Some(vec![
+                    s3s::dto::Tag {
+                        key: Some("env".to_string()),
+                        value: Some("prod".to_string()),
+                    },
+                    s3s::dto::Tag {
+                        key: Some("tier".to_string()),
+                        value: Some("gold".to_string()),
+                    },
+                ]),
+            }),
+            ..Default::default()
+        });
+        let config = structure_config(vec![rule]);
+        let object = |user_tags: &str| ObjectOpts {
+            name: "object".to_string(),
+            user_tags: user_tags.to_string(),
+            ..Default::default()
+        };
+
+        assert!(config.filter_target_arns(&object("env=prod")).is_empty());
+        assert_eq!(config.filter_target_arns(&object("env=prod&tier=gold")), vec!["arn:target:a"]);
+        assert!(config.filter_target_arns(&object("")).is_empty());
+
+        let mut malformed = config;
+        malformed.rules[0].filter.as_mut().unwrap().and.as_mut().unwrap().tags = Some(vec![s3s::dto::Tag {
+            key: Some("env".to_string()),
+            value: None,
+        }]);
+        assert!(
+            malformed.filter_target_arns(&object("env=prod")).is_empty(),
+            "a malformed tag filter must fail closed"
+        );
+    }
+
    #[test]
    fn structure_validation_allows_tag_filter_when_delete_marker_replication_disabled() {
        let mut rule = replication_rule("rule-1", "arn:target:a");
@@ -580,6 +580,30 @@ impl FailStats {
        FailedMetric { count, size }
    }

+    /// Both rolling windows from one walk of the samples. `short` must be the
+    /// narrower window; the walk stops at `long`. Callers that need both (the
+    /// per-node site snapshot) would otherwise scan the deque twice while
+    /// holding the bucket-stats read lock, and the deque is only bounded by
+    /// the one-hour window - an unreachable target under load fills it.
+    pub fn recent_windows(&self, short: Duration, long: Duration) -> (FailedMetric, FailedMetric) {
+        let now = Instant::now();
+        let mut short_metric = FailedMetric::default();
+        let mut long_metric = FailedMetric::default();
+        for sample in self.recent.iter().rev() {
+            let age = now.duration_since(sample.observed_at);
+            if age > long {
+                break;
+            }
+            if age <= short {
+                short_metric.count += 1;
+                short_metric.size += sample.size;
+            }
+            long_metric.count += 1;
+            long_metric.size += sample.size;
+        }
+        (short_metric, long_metric)
+    }
+
    pub fn merge(&self, other: &FailStats) -> Self {
        Self {
            count: self.count.saturating_add(other.count),
@@ -912,6 +936,26 @@ mod tests {
        assert_eq!(last_hour.size, 96);
    }

+    #[test]
+    fn fail_stats_recent_windows_matches_two_separate_scans() {
+        let mut stats = FailStats::default();
+        stats.add_size(64, None::<&()>);
+        stats.add_size(32, None::<&()>);
+
+        let (minute, hour) = stats.recent_windows(Duration::from_secs(60), Duration::from_secs(60 * 60));
+        let expected_minute = stats.recent_since(Duration::from_secs(60));
+        let expected_hour = stats.recent_since(Duration::from_secs(60 * 60));
+
+        assert_eq!((minute.count, minute.size), (expected_minute.count, expected_minute.size));
+        assert_eq!((hour.count, hour.size), (expected_hour.count, expected_hour.size));
+        assert_eq!(minute.count, 2);
+        assert_eq!(hour.size, 96);
+
+        let empty = FailStats::default();
+        let (minute, hour) = empty.recent_windows(Duration::from_secs(60), Duration::from_secs(60 * 60));
+        assert_eq!((minute.count, minute.size, hour.count, hour.size), (0, 0, 0, 0));
+    }
+
    #[test]
    fn fail_stats_saturate_instead_of_wrapping() {
        let mut stats = FailStats {
@@ -919,6 +919,22 @@ where
            }
        }

+        // The filename's item count is untrusted. Reject a payload that contains
+        // more items than advertised instead of returning success and allowing the
+        // caller to delete the entry with trailing events still in the file.
+        match deserializer.next() {
+            None => {}
+            Some(Ok(_)) => {
+                return Err(StoreError::Deserialization(format!(
+                    "Batch for key {key} contains more than {} items",
+                    key.item_count
+                )));
+            }
+            Some(Err(e)) => {
+                return Err(StoreError::Deserialization(format!("Failed to deserialize trailing batch item: {e}")));
+            }
+        }
+
        if items.is_empty() && key.item_count > 0 {
            return Err(StoreError::Deserialization("No items found".to_string()));
        }
@@ -1381,6 +1397,39 @@ mod tests {
        let _ = store.delete();
    }

+    #[test]
+    fn get_multiple_errors_on_batch_with_trailing_items_instead_of_partial_success() {
+        let dir = temp_store_dir("trailing-batch-items");
+        let store = QueueStore::<String>::new_with_compression(&dir, 8, ".test", false);
+        store.open().unwrap();
+
+        let items = vec!["aa".to_string(), "bb".to_string(), "cc".to_string()];
+        let original_key = store.put_multiple(items).unwrap();
+        assert_eq!(original_key.item_count, 3);
+
+        // Keep the three-item payload but make its filename claim that it contains
+        // only two items, simulating a corrupt or otherwise untrusted queue key.
+        let original_path = store.file_path(&original_key);
+        let advertised_key = Key {
+            item_count: 2,
+            ..original_key
+        };
+        let advertised_path = store.file_path(&advertised_key);
+        std::fs::rename(&original_path, &advertised_path).unwrap();
+
+        let err = store.get_multiple(&advertised_key).unwrap_err();
+        assert!(
+            matches!(err, StoreError::Deserialization(_)),
+            "expected Deserialization error, got {err:?}"
+        );
+
+        // Because get_multiple failed, the batch entry remains available for
+        // inspection or recovery instead of being silently discarded.
+        assert!(advertised_path.exists());
+
+        let _ = store.delete();
+    }
+
    #[test]
    fn concurrent_put_raw_respects_entry_limit() {
        let dir = temp_store_dir("concurrent-limit");
@@ -0,0 +1,23 @@
+# Documentation
+
+Use the focused indexes rather than treating this directory as an unordered
+collection:
+
+- [Architecture knowledge base](architecture/README.md)
+- [Testing references](testing/README.md)
+
+## Operations
+
+Operational runbooks live under [`operations/`](operations/). Replication
+operators should start with:
+
+| Runbook | Use it for |
+|---|---|
+| [Site replication operations](operations/site-replication-operations.md) | Health fields, pending operations, outage recovery, re-pair admission, IAM/SSE boundaries, and upgrades. |
+| [Replication target check](operations/replication-check.md) | Validating an S3 destination and version fidelity before enabling replication. |
+| [Replication object size limits](operations/replication-object-size-limits.md) | Multipart routing, large-object limits, and retry characteristics. |
+| [Replication outbound transport](operations/replication-outbound-transport.md) | Integrity headers, generic target behavior, and transport knobs. |
+
+Other runbooks remain grouped by filename in [`operations/`](operations/);
+architecture pages link to the relevant runbook where a cross-boundary
+procedure is required.
@@ -60,6 +60,8 @@ Required headings and strings in these files are asserted by `scripts/check_arch
 | [minio-rustfs-router-compatibility.md](minio-rustfs-router-compatibility.md) | a client or `mc` call that works against MinIO fails against RustFS and you need to know whether the endpoint is missing, stubbed, or deliberately different |
 | [minio-file-format-compat.md](minio-file-format-compat.md) | deciding whether a MinIO drive set, bucket-metadata blob, or SSE object can be read or imported by a given RustFS build, or before touching a listed version anchor |

-Operations runbooks live in [../operations/](../operations/) and testing references in [../testing/README.md](../testing/README.md).
+Operations runbooks are registered in the [documentation operations index](../README.md#operations), and testing references live in [../testing/README.md](../testing/README.md).
+
+For replication operations, start with [site replication operations](../operations/site-replication-operations.md), [replication target check](../operations/replication-check.md), [replication object size limits](../operations/replication-object-size-limits.md), and [replication outbound transport](../operations/replication-outbound-transport.md).

 For per-node HTTP failure ratios and cached storage probe provenance, see [S3 write failure diagnostics](../operations/s3-write-failure-diagnostics.md).
@@ -38,6 +38,42 @@ Counts ignore blank lines and comments; compute them from the files. The lifecyc

 "Supported" for the SSE row means RustFS encrypts and decrypts its own objects. MinIO SSE objects (SSE-S3, SSE-KMS, SSE-C) are not readable in default builds; see [minio-file-format-compat.md Part C](minio-file-format-compat.md#part-c--server-side-encryption-sse) for the `rio-v2` migration build.

+## Replication Support Boundary
+
+Site replication and bucket replication are not the same compatibility claim.
+Site replication requires RustFS-compatible peer admin APIs and coordinates
+IAM, topology, buckets, and metadata. A generic S3-compatible service can only
+be a bucket-replication data target.
+
+For a generic S3 target, RustFS supports object PUT/HEAD/DELETE, multipart
+uploads, tags, version deletes, and Object Lock mutations when the target
+implements the corresponding S3 APIs and has versioning enabled. Targets that
+mint their own version IDs are supported through a per-target version ledger;
+pre-ledger replicas are adopted only when exact key and ETag identify one
+unambiguous target version. `NoSuchVersion` for an already absent addressed
+replica is treated as converged.
+
+The following are capability boundaries, not universal S3 claims:
+
+- `GET /BUCKET?replication-check` must pass the phases required by the intended
+  workload. `VersionFidelity` may report a minting target as mismatched even
+  though ledger-addressed delete and Object Lock phases succeed.
+- A target that rejects standard multipart constraints, required Object Lock
+  integrity headers, or the configured checksum framing is unsupported until
+  its transport settings are made compatible.
+- SSE-S3 and SSE-KMS are decrypted at the source and re-encrypted by the
+  destination's KMS. SSE-C uses ciphertext passthrough and requires target
+  evidence. Unsupported or ambiguous encryption metadata fails closed.
+- ACL authorization is intentionally unsupported, and generic targets never
+  receive RustFS IAM/site-control-plane state.
+- RustFS does not guess between multiple target versions with the same key and
+  ETag. The mutation remains failed and retryable until repair establishes an
+  unambiguous mapping.
+
+See [site replication operations](../operations/site-replication-operations.md)
+for health, recovery, and upgrade rules and [replication outbound transport](../operations/replication-outbound-transport.md)
+for the tested target classes and knobs.
+
 ## Not Yet Passing

 Standard S3 areas that must not be described as complete:
@@ -0,0 +1,258 @@
+# Site Replication Operations
+
+**Use this when:** operating a site-replication deployment, diagnosing a peer
+outage or incomplete topology change, pairing sites that already contain data,
+or planning an upgrade.
+
+**Source of truth:** `rustfs/src/admin/handlers/site_replication.rs`,
+`rustfs/src/site_replication/`, and the bucket-replication worker under
+`crates/ecstore/src/bucket/replication/`.
+
+Site replication combines two different convergence paths:
+
+- the control plane replicates buckets, bucket metadata, IAM, and topology;
+- ordinary bucket replication moves object versions and delete operations.
+
+An `enabled: true` response only says that a site has more than one configured
+peer. It does not prove that every peer is reachable or caught up. Always read
+`pendingOperation`, `retryStats`, `PeerErrors`, and `Metrics` as well.
+
+## Routine checks
+
+Run these commands from an admin workstation with one alias per site:
+
+```console
+mc admin replicate info site-a
+mc admin replicate status site-a
+```
+
+Check more than one site. A partition can leave each side with a different but
+locally valid view.
+
+`replicate info` is the compact control-plane view:
+
+| Field | Interpretation |
+|---|---|
+| `enabled` | More than one site is configured; this is not a health verdict. |
+| `sites` | The locally persisted topology. Compare deployment IDs and endpoints on every site. |
+| `retryStats.pending` | Collapsed peer deliveries waiting to be retried. |
+| `retryStats.failed` | Deliveries that crossed the escalation threshold and require attention. |
+| `retryStats.lastError` | A redacted summary of the most recent delivery failure. |
+| `pendingOperation` | A durable multi-step topology operation described below. Absence is the healthy steady state. |
+
+`replicate status` adds detailed convergence state:
+
+| Field | Interpretation |
+|---|---|
+| `Sites` / `PeerStates` | Configured peers and derived reachability/configuration state. |
+| `PeerErrors` | A peer could not be queried. Its detailed counters may be absent; do not read zeros as success. |
+| `BucketStats` | Per-bucket presence and versioning, replication, lifecycle, Object Lock, and metadata mismatches. |
+| `PolicyStats`, `UserStats`, `GroupStats` | IAM inventory mismatches. |
+| `RetryStats` | Durable control-plane retry backlog and escalation count. |
+| `Metrics.replMetrics` | Per-destination online state, downtime, replicated counts/bytes, and `failed` totals/windows. |
+| `Metrics.queued` / `Metrics.inProgress` | Object work waiting or active on the responding node. |
+| `Metrics.errors` | Node-level object-replication failures. When only queue statistics are available, RustFS synthesizes a node entry and preserves this counter rather than reporting zero. |
+| `Metrics.retries` | Redeliveries. Always zero today: a failed object is not retried by an event, it waits for the scanner pass described below. Read `errors` instead. |
+
+Healthy means: the same topology is visible on all sites, no pending operation,
+no peer error, no failed retry escalation, required bucket/IAM state is in sync,
+and queue/error counters are stable or falling. Counters are cumulative; alert on
+their rate and on a backlog that does not drain, not merely on a non-zero total.
+
+## Pending operations and recovery
+
+`pendingOperation` contains `operation`, an opaque `id`, `pendingPeers`, and
+`ackedPeers`. Do not edit the site-replication state object by hand. The marker
+is the crash-recovery journal and removing it can make a partially applied
+operation look complete.
+
+The heavyweight reconciler runs once at startup and every 600 seconds. The
+lightweight retry drain runs every 30 seconds. A restart is therefore a valid
+way to cause an immediate heavyweight pass after the underlying fault has been
+fixed, but it is not a substitute for fixing connectivity, credentials, TLS,
+or the remote endpoint.
+
+### `remove`
+
+The original topology and each peer acknowledgement are persisted before the
+operation finalizes. While peers remain in `pendingPeers`, restore access to
+them and wait for reconciliation. If a peer is permanently gone, a new remove
+request may remove all currently active unacknowledged peers; RustFS permits
+that request and then finalizes against the remaining topology. Removing the
+local site or all sites is also an explicit completion path.
+
+Do not re-add a site merely to hide this marker. First compare the topology on
+all reachable peers. If the same operation ID makes no progress for more than
+one heavyweight interval, collect `PeerErrors`, `RetryStats`, and the
+site-replication logs before retrying the remove.
+
+### `rotate-svc-acct`
+
+Service-account rotation keeps the candidate secrets and peer acknowledgements
+until every current remote peer accepts the rotation. Restore the failing peer
+and allow the reconciler to resume it. Do not manually delete either candidate
+credential during this window: doing so can remove the only credential that a
+not-yet-acknowledged peer accepts.
+
+After the marker clears, verify `replicate status` from every site, then retire
+any separately retained old credential material according to local policy.
+
+### `endpoint-refresh`
+
+An endpoint, CA, or TLS-verification edit first refreshes the replication
+target on every active peer and records acknowledgements. On startup and every
+heavyweight pass, RustFS probes peer capability, uses the endpoint-refresh API
+when supported (or the legacy peer-edit fallback), refreshes local bucket
+targets, and commits the edit only after every still-active peer acknowledges.
+
+If this marker is stuck:
+
+1. Confirm that the proposed endpoint and CA are correct and reachable from
+   every site, not only from the admin workstation.
+2. Restore the site-replication service account and TLS trust path.
+3. Wait for one 600-second pass or restart one healthy node to trigger the
+   startup pass.
+4. Re-run the identical edit only if the operation remains visible; a different
+   endpoint edit is rejected while the existing refresh is pending. The journal
+   pins the edit's payload, so a re-run without `--replicate-ilm-expiry` keeps
+   the value the first attempt recorded, and a re-run asking for a different
+   value is rejected. Finish or remove the pending refresh before changing it.
+
+A peer removed from the topology no longer blocks completion. A remove request
+is accepted when it removes every active unacknowledged peer.
+
+While this marker is present, control-plane retry replay to the other peers
+keeps running, but bucket wiring reconciliation waits: it rewrites the same
+targets the refresh is changing. Expect bucket-level drift on this site to
+persist until the refresh settles.
+
+## Outage recovery and convergence time
+
+Control-plane retry begins on the 30-second drain, while heavyweight snapshots,
+pending topology operations, and bucket wiring are revisited on the 600-second
+pass. Object MRF entries are persisted every 10 seconds by default and target
+health is probed every 5 seconds. These are scheduling bounds, not delivery
+SLAs: network timeouts and the amount of queued work add to them.
+
+Objects that must be rediscovered by the scanner have this conservative upper
+bound before discovery:
+
+```text
+RUSTFS_DATA_USAGE_UPDATE_DIR_CYCLES
+    × max(RUSTFS_SCANNER_CYCLE, actual duration of one scanner cycle)
+```
+
+The defaults re-descend a compacted directory every 16 cycles. A practical
+production starting point for a tighter recovery objective is
+`RUSTFS_DATA_USAGE_UPDATE_DIR_CYCLES=4`; `1` forces re-descent every cycle.
+Measure the additional disk and metadata load before lowering it further or
+tuning the scanner cadence. For an immediate operator-driven recovery, start a
+site resync with `mc admin replicate resync start` and monitor its status.
+Transfer time after discovery remains proportional to backlog size, bandwidth,
+worker capacity, and target latency. Use queue depth and the rate of
+`Metrics.errors` rather than the formula alone to decide whether convergence is
+progressing.
+
+## Pairing sites that already contain data
+
+When more than one requested site is non-empty, preflight considers each bucket
+name held by more than one site:
+
+- versioning must be `Enabled` on every site holding the shared bucket;
+- Object Lock enablement must be identical on every holder.
+
+A bucket present on only one site is safe: post-add backfill creates it on the
+other peers. A shared unversioned bucket is rejected because merging can
+overwrite the only copy of an object. An Object Lock mismatch is rejected
+because lock enablement cannot be changed after bucket creation and convergence
+could otherwise strip a WORM guarantee.
+
+If preflight rejects the pair, keep the authoritative copy, delete the
+conflicting bucket (or its contents) from all other sites, run `replicate add`
+again, and then start `replicate resync` from the surviving site. Back up and
+validate the authoritative data before deleting anything.
+
+## IAM convergence and repair boundary
+
+Ordinary IAM changes are delivered to each peer. A successful bulk IAM import
+also schedules one collapsed full-IAM snapshot per remote peer. A failed IAM
+deletion is replayed before that snapshot so the snapshot cannot re-create a
+principal or grant that was already revoked.
+
+The safety state has two bounds:
+
+- deletion high-water marks are retained for 30 days;
+- deletion replay bodies are capped at 256 distinct entities per peer.
+
+Repeated deletion of the same entity replaces its saved body. When the per-peer
+cap is exceeded or the body cannot be serialized, the retry entry remains
+escalated rather than pretending the deletion is replayable. An item from an
+older sender without a source timestamp cannot install the 30-day high-water
+mark, so verify it explicitly after a prolonged split. A successful drain
+clears replay bodies; removing the peer prunes its bodies. For an escalated IAM
+retry, use the site-replication repair workflow for the affected peer and IAM
+family, then verify users, service accounts, groups, policies, and mappings on
+both sides. Repair is the operator's explicit accountability transfer and
+clears the saved deletion bodies only after the IAM repair succeeds.
+
+A group's status converges in one direction. An explicit disable is applied
+everywhere, including through a snapshot, but a membership change never
+carries an enable - it would otherwise re-enable a group frozen on the
+receiving site. If a group ended up disabled on one site only, re-enable it
+there explicitly with `mc admin group enable`; a snapshot or repair will not
+do it.
+
+Treat IAM divergence as a security incident: a user deleted on one site can
+remain usable on an unreachable peer until replay or repair completes. A peer
+whose IAM entry is escalated does not receive scheduled snapshots either -
+including the one a bulk import schedules - until the repair settles it.
+
+## Encrypted objects
+
+| Source form | Replication behavior | Fail-closed condition |
+|---|---|---|
+| SSE-S3 | The source decrypts the object; the request sends only `AES256` intent; the destination encrypts with its own KMS. Source envelope material never leaves the site. | The destination cannot satisfy the encryption request, or the source metadata is incomplete/unsupported. The replica is `FAILED`; plaintext is not silently stored. |
+| SSE-KMS | The source decrypts the object; the request sends `aws:kms` intent without the source-local key ID; the destination selects its own configured KMS key. | Either side cannot decrypt/encrypt, or the metadata mixes incompatible encryption evidence. |
+| SSE-C | Stored ciphertext and the required SSE-C replication transport metadata pass through. RustFS verifies target evidence before accepting the replica. | The target does not echo the customer-algorithm evidence, required material/layout is absent, or the metadata is ambiguous. |
+
+Unknown MinIO/RustFS encryption markers are never forwarded as ordinary user
+metadata. They fail replication so an operator must migrate or repair the
+object with a supported format.
+
+## Rolling upgrades and rollback
+
+Keep every node in one site on the same version whenever possible. Upgrade all
+nodes of one site consecutively, verify its startup reconciliation and status,
+then move to the next site. Do not intentionally leave a site mixed-version:
+admin requests can land on different nodes, and an older node may not resume a
+new pending-operation shape or expose its health fields.
+
+Current state additions are optional and defaulted, so older readers ignore
+them. The target-version ledger is stored as dual-prefixed internal object
+metadata and is also ignored by older readers; rollback does not corrupt the
+object format, but older code loses the assigned-version routing improvement.
+
+Before rolling back across the fix that retains the data directory of a version
+awaiting purge replication (rustfs/rustfs#7307), ensure no version purge is
+pending. Older code can free that retained version's data directory before the
+remote purge is acknowledged, leaving unreadable metadata and blocking bucket
+deletion. Drain or repair replication and take a metadata/data backup first.
+
+## Runtime knobs
+
+These values are read when the owning background task starts. Restart the
+server after changing them. The millisecond intervals have a 10 ms floor;
+invalid values fall back to the default with a warning.
+
+| Variable | Default | Effect |
+|---|---:|---|
+| `RUSTFS_REPL_HEALTH_CHECK_INTERVAL_MS` | `5000` | Remote-target health probe interval. Lowering it increases outbound probes. |
+| `RUSTFS_REPL_MRF_FLUSH_INTERVAL_MS` | `10000` | Maximum periodic interval between MRF persistence flushes; 1,000 new entries also trigger a flush. |
+| `RUSTFS_REPL_RESYNC_POLL_MAX_MS` | `60000` | Upper bound for randomized resync retry-poll sleep. |
+| `RUSTFS_REPL_RESYNC_MAX_JOBS` | `2` | Concurrent resync jobs; values are bounded to `1..=32`. |
+
+Transport-specific controls and target behavior are documented in
+[Replication outbound transport](replication-outbound-transport.md). Validate a
+new destination with [Replication target check](replication-check.md), and read
+[Replication object size limits](replication-object-size-limits.md) before
+moving large objects.
@@ -3,7 +3,7 @@
 **Use this when:** a check is red and you need to know whether it blocks the merge, which workflow and job produced it, and how to reproduce it locally.
 **Source of truth:** the live `main` ruleset (command below) for required status; `.github/workflows/<file>.yml` for triggers, `paths`, `timeout-minutes`, and cron; `.config/nextest.toml` for e2e profile filters; `.github/scheduled-validations.json` for the freshness-watchdog list.

-A job blocks a merge only when its exact check name is in the live `main` ruleset. A workflow name, a `merge_group` trigger, or a red PR check does not make a job required by itself.
+A job blocks a merge when its exact check name is required by the live `main` ruleset, or when its result is required by the `Test and Lint` aggregate. A workflow name, a `merge_group` trigger, or an unrelated red PR check does not make a job required by itself.

 ## Required merge checks

@@ -13,9 +13,11 @@ The `main` ruleset (`6436880`) requires exactly these contexts, with `strict_req
 |---|---|---|
 | `CLA Check` | `cla.yml` | Contributor agreement |
 | `Quick Checks` | `ci.yml` job `quick-checks` | Formatting and repository guard scripts |
-| `Test and Lint` | `ci.yml` job `test-and-lint` | Clippy, workspace nextest (`ci` profile, excluding `e2e_test`), doctests, migration-gate count (`scripts/check_migration_gate_count.sh`) |
+| `Test and Lint` | `ci.yml` job `required-checks` | Exact expected results for every CI validation job, including workspace checks, critical E2E, feature lanes, and event-specific full suites |

-For PRs limited to the `paths-ignore` list in `ci.yml`, `ci-docs-only.yml` reports `Quick Checks` and `Test and Lint` under the same names; it runs the quick checks and `scripts/check_no_planning_docs.sh`, not a Rust build or tests. `scripts/check_ci_paths_sync.sh` keeps the two path lists aligned.
+Every PR enters `ci.yml`. The `classify-changes` job uses the base revision of `scripts/ci_gate.py` to select a conservative documentation-only path: root Markdown/licenses, `AGENTS.md`, Markdown under `docs/` or `.agents/skills/`, and documentation images. Unknown paths, unavailable Git history, an empty diff, or a missing base policy select the full matrix. Renames include their deleted source path. Documentation-only PRs still run Quick Checks and Typos; the aggregate requires the expensive jobs to be skipped exactly as selected.
+
+`required-checks` runs even after failed or skipped dependencies. `scripts/ci_gate.py verify` rejects missing jobs, unexpected jobs, failure, cancellation, and unexpected skips; optional lanes are required only on their declared events. `Workspace Test and Lint` is the ordinary Rust job, while `Test and Lint` uniquely names the aggregate. New validation jobs must update both its direct dependencies and the script contract. Test this wiring and its failure cases with `python3 scripts/ci_gate.py --self-test`.

 Verify the live rule before changing merge policy:

@@ -24,25 +26,25 @@ gh api repos/rustfs/rustfs/rulesets/6436880 \
  --jq '.rules[] | select(.type == "required_status_checks") | .parameters'
 ```

-Promotion rule: never promote a report-only lane to required from one green run. Require at least 14 days and 30 representative PRs with at least 99% complete execution, then update the ruleset and this file together.
+The aggregate requires the validation lanes already selected by `ci.yml`; this closes the gap where a failing critical lane left the required workspace check green. Independent workflows remain report-only unless separately required. Before adding a new expensive lane or moving existing PR coverage to a schedule, collect representative execution and regression evidence, establish ownership and a working scheduled replacement, and update this reference with the resulting policy.

 ## Pull request and merge matrix

-"Report-only" means visible and actionable but not in the required list. Budgets are each job's `timeout-minutes` in the named workflow and are not copied here.
+"Via aggregate" means a wrong result fails the required `Test and Lint` check. "Report-only" means visible and actionable but outside both the required list and aggregate. Budgets are each job's `timeout-minutes` in the named workflow and are not copied here.

 | Event | Check name | Workflow / job | Merge status | Reproduce |
 |---|---|---|---|---|
 | PR, non-doc change | `Quick Checks` | `ci.yml` `quick-checks` | Required | `make pre-commit` |
-| PR, non-doc change | `Test and Lint` | `ci.yml` `test-and-lint` | Required | `cargo clippy --all-targets -- -D warnings`; `cargo nextest run --profile ci --all --exclude e2e_test`; `cargo test --all --doc`; `scripts/check_migration_gate_count.sh` |
-| PR, non-doc change | `Typos` | `ci.yml` `typos` | Report-only | `typos` |
-| PR, non-doc change | `ILM Integration (serial)` | `ci.yml` `test-ilm-integration-serial` | Report-only | exact command in the job |
-| PR, non-doc change | `Test and Lint (rio-v2)`, `Test and Lint (swift)`, `Test and Lint (sftp)` | `ci.yml` `test-and-lint-rio-v2`, `test-and-lint-protocols` | Report-only | `cargo nextest run` with the job's `--features` |
-| PR, non-doc change | `Connect Short Credential Boundary` | `ci.yml` `connect-short-credential-boundary` | Report-only | `cargo test -p rustfs --test connect_registration --features connect-e2e-short-credentials`; `cargo check -p rustfs --release --features connect-e2e-short-credentials` must fail |
-| PR, non-doc change | `Build RustFS Debug Binary` | `ci.yml` `build-rustfs-debug-binary` | Report-only; prerequisite for the black-box jobs | `cargo build -p rustfs --bins` |
-| PR, non-doc change | `io_uring Integration (real)` | `ci.yml` `uring-integration` | Report-only | `cargo test -p rustfs-ecstore --lib uring_ -- --test-threads=1 --nocapture` |
-| PR, non-doc change | `End-to-End Tests` | `ci.yml` `e2e-tests` | Report-only | `cargo nextest run --profile e2e-smoke -p e2e_test`, then `./scripts/e2e-run.sh ./target/debug/rustfs <data-dir>`; membership guards `scripts/check_test_wiring.py --check-profile e2e-smoke <listing.json>` and `scripts/check_security_smoke_count.sh check <listing.json>` |
-| PR, non-doc change | `S3 Implemented Tests` | `ci.yml` `s3-implemented-tests` | Report-only | build `rustfs`, then `scripts/s3-tests/run.sh` with the job's `DEPLOY_MODE` / `TEST_MODE` / `MAXFAIL` env |
-| PR, non-doc change | `S3 Lifecycle Behavior Tests` | `ci.yml` `s3-lifecycle-behavior-tests` | Report-only | `scripts/s3-tests/run.sh` with the job's accelerated-scanner env |
+| PR, non-doc change | `Workspace Test and Lint` | `ci.yml` `test-and-lint` | Via aggregate | `cargo clippy --all-targets -- -D warnings`; `cargo nextest run --profile ci --all --exclude e2e_test`; `cargo test --all --doc`; `scripts/check_migration_gate_count.sh` |
+| PR, non-doc change | `Typos` | `ci.yml` `typos` | Via aggregate | `typos` |
+| PR, non-doc change | `ILM Integration (serial)` | `ci.yml` `test-ilm-integration-serial` | Via aggregate | exact command in the job |
+| PR, non-doc change | `Test and Lint (rio-v2)`, `Test and Lint (swift)`, `Test and Lint (sftp)` | `ci.yml` `test-and-lint-rio-v2`, `test-and-lint-protocols` | Via aggregate | `cargo nextest run` with the job's `--features` |
+| PR, non-doc change | `Connect Short Credential Boundary` | `ci.yml` `connect-short-credential-boundary` | Via aggregate | `cargo test -p rustfs --test connect_registration --features connect-e2e-short-credentials`; `cargo check -p rustfs --release --features connect-e2e-short-credentials` must fail |
+| PR, non-doc change | `Build RustFS Debug Binary` | `ci.yml` `build-rustfs-debug-binary` | Via aggregate; prerequisite for black-box jobs | `cargo build -p rustfs --bins --features e2e-test-hooks` |
+| PR, non-doc change | `io_uring Integration (real)` | `ci.yml` `uring-integration` | Via aggregate | `cargo test -p rustfs-ecstore --lib uring_ -- --test-threads=1 --nocapture` |
+| PR, non-doc change | `End-to-End Tests` | `ci.yml` `e2e-tests` | Via aggregate | `cargo nextest run --profile e2e-smoke -p e2e_test`, then `./scripts/e2e-run.sh ./target/debug/rustfs <data-dir>`; membership guards `scripts/check_test_wiring.py --check-profile e2e-smoke <listing.json>` and `scripts/check_security_smoke_count.sh check <listing.json>` |
+| PR, non-doc change | `S3 Implemented Tests` | `ci.yml` `s3-implemented-tests` | Via aggregate | build `rustfs`, then `scripts/s3-tests/run.sh` with the job's `DEPLOY_MODE` / `TEST_MODE` / `MAXFAIL` env |
+| PR, non-doc change | `S3 Lifecycle Behavior Tests` | `ci.yml` `s3-lifecycle-behavior-tests` | Via aggregate | `scripts/s3-tests/run.sh` with the job's accelerated-scanner env |
 | PR touching `paths` in `audit.yml` | `Cargo Deny`, `Workflow Pin Report`, `Dependency Review` | `audit.yml` `cargo-deny`, `workflow-pin-report`, `dependency-review` | Report-only | `cargo deny check`; `scripts/security/check_workflow_pins.sh` |
 | PR touching `paths` in `architecture-migration-rules.yml` | `Architecture Migration Rules` | `architecture-migration-rules.yml` `architecture-migration-rules` | Report-only | `scripts/check_architecture_migration_rules.sh` |
 | PR touching `paths` in `nix.yml` | `Nix Build & Check` | `nix.yml` `nix-validation` | Report-only | `nix flake check` |
@@ -52,8 +54,8 @@ Promotion rule: never promote a report-only lane to required from one green run.
 | PR touching `paths` in `e2e-upgrade.yml` | `Direct upgrade from the previous release`, `Mixed-version rolling upgrade from the previous release`, `Bucket configuration survives the upgrade`, `Rollback reads current bucket metadata` | `e2e-upgrade.yml` `upgrade` matrix | Report-only | the `cargo test --locked -p e2e_test` command in the job with `RUSTFS_UPGRADE_SOURCE_BINARY` pointing at the pinned previous release (`UPGRADE_SOURCE_VERSION`) |
 | PR touching `paths` in `oidc-keycloak.yml` | `OIDC Keycloak live gate` | `oidc-keycloak.yml` `oidc-keycloak-live` | Report-only | `cargo build --locked -p rustfs --bin rustfs`, then `bash scripts/test/oidc_keycloak_live.sh ./target/debug/rustfs` |
 | PR touching `paths` in `targets-integration.yml` | `PostgreSQL, MySQL, AMQP, and NATS` | `targets-integration.yml` `targets-live` | Report-only | start the containers as in the job, export the `RUSTFS_TEST_*` DSNs, then the job's `cargo test --locked -p rustfs-targets --test <name> -- --ignored --test-threads=1` commands |
-| PR limited to main-CI-excluded paths | `Quick Checks`, `Test and Lint` | `ci-docs-only.yml` `quick-checks`, `test-and-lint` | Required | `git diff --check`; `make doc-paths-check`; `scripts/check_no_planning_docs.sh` |
-| `merge_group`; push to `main` | `End-to-End Tests (full merge gate)` | `ci.yml` `e2e-full` | Report-only | `cargo nextest run --profile e2e-full -p e2e_test` |
+| PR, documentation-only selection | `Quick Checks`, `Typos`, `Test and Lint` | `ci.yml` `quick-checks`, `typos`, `required-checks` | Required directly or via aggregate | Quick Checks commands; `python3 scripts/ci_gate.py --self-test` |
+| `merge_group`; push to `main` | `End-to-End Tests (full merge gate)` | `ci.yml` `e2e-full` | Via aggregate on these events | `cargo nextest run --profile e2e-full -p e2e_test` |

 e2e filters live in `.config/nextest.toml`; extend a profile instead of adding a second selector. Before a profile runs, `scripts/check_test_wiring.py` compares its listing to the committed digest in `.config/e2e-<profile>-selection.txt`, so a silent test drop fails closed.

@@ -67,11 +69,11 @@ the serialized cluster fault-domain suites for scheduled soak signal.

 ## Scheduled validation

-Scheduled lanes never block a PR. Their workflow-local gate fails the run, scheduled failures route to the shared failure-issue action, and `scheduled-validation-freshness.yml` fails when a workflow listed in `.github/scheduled-validations.json` has not run within its `max_age_hours` (a `never_ran_grace_until` entry covers the window before a newly enabled cron's first slot). Cadence is qualitative here; the cron lives in each workflow's `on.schedule`.
+Scheduled lanes never block a PR. Their workflow-local gate fails the run, scheduled failures route to the shared failure-issue action, and `scheduled-validation-freshness.yml` fails when a workflow listed in `.github/scheduled-validations.json` has no recent attempt or completed successful scheduled run within its `max_age_hours` (a `never_ran_grace_until` entry covers the window before a newly enabled cron's first slot). Cadence is qualitative here; the cron lives in each workflow's `on.schedule`.

 | Workflow (cadence) | Jobs | Verdict and artifacts | In freshness list | Reproduce |
 |---|---|---|---|---|
-| `ci.yml` (weekly) | full matrix, including the schedule/dispatch-only rio-v2 jobs `build-rustfs-debug-binary-rio-v2` and `e2e-tests-rio-v2` | per-job | yes | dispatch `ci.yml` |
+| `ci.yml` (weekly) | full matrix, including the schedule/dispatch-only rio-v2 jobs `build-rustfs-debug-binary-rio-v2` and `e2e-tests-rio-v2` | strict aggregate; the full E2E lane runs on dispatch, merge groups, and main pushes | yes | dispatch `ci.yml` |
 | `build.yml` (weekly) | `build-rustfs` over the six-target platform matrix in `prepare-platform-matrix` (four Linux, macOS aarch64, Windows x86_64) | build/package integrity | yes | dispatch `build.yml` with an exact platform set |
 | `e2e-replication-nightly.yml` (nightly) | `repl-nightly`, `cluster-nightly`, `protocols-nightly` | three independent gates; JUnit, membership listing, server logs | yes | `cargo nextest run --profile e2e-repl-nightly -p e2e_test`; `--profile e2e-nightly`; `-j 1 --profile e2e-protocols` |
 | `e2e-distributed.yml` (storage-sensitive PRs + nightly) | `distributed` | fail-closed 4-node 4-disk S3, durability, replication, movement, fault, and direct/rolling upgrade gate; JUnit, membership listing, per-node server logs | yes, with `never_ran_grace_until` | download the pinned previous release as in the workflow, export `RUSTFS_UPGRADE_SOURCE_BINARY`, then `cargo nextest run --profile e2e-distributed -p e2e_test` |
@@ -88,7 +90,7 @@ Scheduled lanes never block a PR. Their workflow-local gate fails the run, sched
 | `e2e-upgrade.yml` (weekly) | `upgrade` (4-case matrix) | upgrade and rollback gate; server logs | no | see the PR row |
 | `oidc-keycloak.yml` (weekly) | `oidc-keycloak-live` | live OIDC gate | no | see the PR row |
 | `targets-integration.yml` (nightly) | `targets-live` | live target gate; container logs | no | see the PR row |
-| `scheduled-validation-freshness.yml` (nightly) | `check-freshness` | fails on a never-created or stale schedule | n/a | dispatch |
+| `scheduled-validation-freshness.yml` (nightly) | `check-freshness` | fails on missing or stale attempts or completed successes | n/a | dispatch |

 Manual `workflow_dispatch` runs are debugging evidence and do not open scheduled-failure issues. A manual performance run may explicitly allow a known regression; that override is not a passing baseline.

@@ -627,6 +627,32 @@ pub(crate) async fn cluster_replication_stats(bucket: &str, context: Option<Arc<
        .await
 }

+/// Reload the bucket's metadata on every peer so a follow-up
+/// `put-bucket-replication` on another node does not read a stale target.
+///
+/// Best effort, like every S3 bucket-config write path
+/// (`app::bucket_usecase::notify_bucket_metadata_reload`): the target is
+/// already persisted and live on this node, and the 15-minute refresh closes
+/// the gap, so a peer that cannot be reached must not turn a completed write
+/// into a failed request.
+async fn notify_remote_target_metadata_reload(bucket: &str, context: Option<Arc<AppContext>>, action: &'static str) {
+    let Some(notification_system) = current_notification_system_for_context(context.as_deref()) else {
+        return;
+    };
+    if let Err(err) = notification_system.load_bucket_metadata(bucket).await {
+        warn!(
+            event = EVENT_ADMIN_REMOTE_TARGET_STATE,
+            component = LOG_COMPONENT_ADMIN,
+            subsystem = LOG_SUBSYSTEM_REPLICATION,
+            action = action,
+            result = "peer_metadata_reload_failed",
+            bucket = %bucket,
+            error = ?err,
+            "admin remote target state"
+        );
+    }
+}
+
 fn unique_replication_peers(peer_clients: &[Option<PeerRestClient>]) -> (Vec<&PeerRestClient>, u32) {
    let mut seen_grid_hosts = HashSet::new();
    let peers: Vec<_> = peer_clients
@@ -699,6 +725,7 @@ pub struct SetRemoteTargetHandler {}
 impl Operation for SetRemoteTargetHandler {
    async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
        let cred = validate_replication_admin_request(&req, AdminAction::SetBucketTargetAction).await?;
+        let app_context = app_context_from_req(&req);

        let queries = extract_query_params(&req.uri);

@@ -926,6 +953,8 @@ impl Operation for SetRemoteTargetHandler {
            .map_err(map_bucket_target_error)?;
        let _targets_guard = lock_bucket_targets_metadata(bucket).await;
        let arn = persist_remote_target_write(bucket, remote_target, incarnation, mode).await?;
+        drop(_targets_guard);
+        notify_remote_target_metadata_reload(bucket, app_context, "set_remote_target").await;
        let arn_str = serde_json::to_string(&arn)
            .map_err(|_| S3Error::with_message(S3ErrorCode::InternalError, "Failed to serialize target ARN"))?;

@@ -1006,6 +1035,7 @@ pub struct RemoveRemoteTargetHandler {}
 impl Operation for RemoveRemoteTargetHandler {
    async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
        validate_replication_admin_request(&req, AdminAction::SetBucketTargetAction).await?;
+        let app_context = app_context_from_req(&req);

        debug!("remove remote target called");
        let queries = extract_query_params(&req.uri);
@@ -1081,6 +1111,7 @@ impl Operation for RemoveRemoteTargetHandler {
        }
        let json_targets = serde_json::to_vec(&targets)
            .map_err(|_| S3Error::with_message(S3ErrorCode::InternalError, "Failed to serialize targets"))?;
+        let notification_bucket = bucket.clone();
        let bucket = bucket.clone();
        let arn = arn_str.clone();
        // The pool cancellation owns a detached task. Both outer guards must
@@ -1101,6 +1132,8 @@ impl Operation for RemoveRemoteTargetHandler {
            S3Error::with_message(S3ErrorCode::InternalError, format!("remote target removal task failed: {error}"))
        })??;

+        notify_remote_target_metadata_reload(&notification_bucket, app_context, "remove_remote_target").await;
+
        Ok(S3Response::new((StatusCode::NO_CONTENT, Body::from("".to_string()))))
    }
 }
@@ -1787,6 +1820,25 @@ mod tests {
        pairs.iter().map(|(k, v)| (k.to_string(), v.to_string())).collect()
    }

+    #[test]
+    fn remote_target_writes_notify_peer_metadata_caches() {
+        let source = include_str!("replication.rs");
+        for (start, end) in [
+            ("impl Operation for SetRemoteTargetHandler", "pub struct ListRemoteTargetHandler"),
+            ("impl Operation for RemoveRemoteTargetHandler", "async fn cancel_active_resync_intent"),
+        ] {
+            let body = source
+                .split(start)
+                .nth(1)
+                .and_then(|rest| rest.split(end).next())
+                .expect(start);
+            assert!(
+                body.contains("notify_remote_target_metadata_reload"),
+                "{start} must notify every node before returning success"
+            );
+        }
+    }
+
    #[test]
    fn update_ops_parse_minio_query_contract() {
        let ops = parse_remote_target_update_ops(&query_map(&[
@@ -1318,6 +1318,24 @@ impl Operation for ImportIam {
            failed,
        };

+        // The entities are already imported locally. A snapshot that cannot be
+        // scheduled is a convergence delay the reconcile pass still closes, so
+        // it must not turn a completed import into a failed request - the same
+        // best-effort contract every other site-replication hook here follows.
+        if let Err(err) =
+            crate::site_replication::enqueue_site_replication_iam_snapshot("iam import scheduled a full snapshot").await
+        {
+            warn!(
+                component = LOG_COMPONENT_ADMIN,
+                subsystem = LOG_SUBSYSTEM_USER,
+                event = EVENT_ADMIN_USER_STATE,
+                action = "import_iam",
+                result = "site_replication_snapshot_not_scheduled",
+                error = ?err,
+                "admin user state"
+            );
+        }
+
        let body = serde_json::to_vec(&ret).map_err(|e| S3Error::with_message(S3ErrorCode::InternalError, e.to_string()))?;

        let mut header = HeaderMap::new();
@@ -1424,6 +1442,16 @@ mod tests {
        assert!(include_str!("user.rs").contains(mapper_call));
    }

+    #[test]
+    fn import_iam_enqueues_a_site_replication_snapshot() {
+        let body = source_block(include_str!("user.rs"), "impl Operation for ImportIam");
+
+        assert!(
+            body.contains("enqueue_site_replication_iam_snapshot"),
+            "a successful IAM import must schedule a full IAM snapshot for every remote site"
+        );
+    }
+
    #[test]
    fn test_should_check_deny_only_for_regular_self_request() {
        let cred = Credentials {
@@ -409,6 +409,26 @@ fn transfer_summaries(stats: &InternalReplicationStats) -> (XferSummaryWire, Tar
    (summary, per_target)
 }

+/// Node-level failure counters for `errors`. The sibling `retries` field
+/// stays zero on purpose: it means redeliveries in the minio-go shape, and a
+/// failed object is not retried by an event today (it waits for the scanner's
+/// heal pass), so reporting failures there would claim a redelivery that
+/// never happened.
+fn failure_counters(stats: &InternalReplicationStats) -> CounterSummaryWire {
+    let (total, last1m, last1hr) = stats.stats.values().fold((0i64, 0i64, 0i64), |acc, stat| {
+        (
+            acc.0.saturating_add(stat.fail_stats.count),
+            acc.1.saturating_add(stat.fail_stats.last_minute.count),
+            acc.2.saturating_add(stat.fail_stats.last_hour.count),
+        )
+    });
+    CounterSummaryWire {
+        total: u64::try_from(total.max(0)).unwrap_or_default(),
+        last1m: u64::try_from(last1m.max(0)).unwrap_or_default(),
+        last1hr: u64::try_from(last1hr.max(0)).unwrap_or_default(),
+    }
+}
+
 impl MetricsV2Wire {
    /// Project the aggregated internal stats onto the `MetricsV2` shape.
    ///
@@ -418,6 +438,7 @@ impl MetricsV2Wire {
    /// `queueStats.nodes` and treats an empty list as "no data".
    pub(crate) fn from_stats(bucket_stats: &BucketStats, node_name: &str) -> Self {
        let (xfer_stats, tgt_xfer_stats) = transfer_summaries(&bucket_stats.replication_stats);
+        let failed = failure_counters(&bucket_stats.replication_stats);
        let mut nodes: Vec<ReplQNodeStatsWire> = bucket_stats
            .queue_stats
            .nodes
@@ -436,6 +457,7 @@ impl MetricsV2Wire {
                q_stats: InQueueMetricWire::from(&bucket_stats.replication_stats.q_stat),
                xfer_stats: xfer_stats.clone(),
                tgt_xfer_stats: tgt_xfer_stats.clone(),
+                errors: failed,
                ..Default::default()
            });
        } else {
@@ -444,6 +466,7 @@ impl MetricsV2Wire {
            if let Some(first) = nodes.first_mut() {
                first.xfer_stats = xfer_stats.clone();
                first.tgt_xfer_stats = tgt_xfer_stats.clone();
+                first.errors = failed;
            }
        }

@@ -478,6 +501,12 @@ mod tests {
        target.replicated_size = 4096;
        target.failed.count = 3;
        target.failed.size = 900;
+        target.fail_stats.count = 3;
+        target.fail_stats.size = 900;
+        target.fail_stats.last_minute.count = 2;
+        target.fail_stats.last_minute.size = 600;
+        target.fail_stats.last_hour.count = 3;
+        target.fail_stats.last_hour.size = 900;
        target.bandwidth_limit_bytes_per_sec = 1024;
        target.current_bandwidth_bytes_per_sec = 512.5;
        stats
@@ -537,6 +566,10 @@ mod tests {
        assert_eq!(node["queueStats"]["peak"], node["queueStats"]["max"]);
        assert!(node["activeWorkers"].get("curr").is_some());
        assert!(node["transferSummary"].get("Total").is_some());
+        assert_eq!(node["errors"]["total"], 3);
+        assert_eq!(node["errors"]["last1m"], 2);
+        assert_eq!(node["errors"]["last1hr"], 3);
+        assert_eq!(node["retries"]["total"], 0, "failures are not redeliveries; retries must not claim one");
        assert_eq!(json["downtimeInfo"], serde_json::json!({}));
    }

@@ -217,6 +217,26 @@ pub(crate) fn settle_observed_site_replication_retry_event(
    before.saturating_sub(queue.len())
 }

+/// Make sure `peer` has a collapsed entry for `path` without counting the
+/// call as a delivery failure. A bulk local mutation (`import-iam`) needs the
+/// entry to exist so the next drain sends the snapshot; routing it through
+/// [`upsert_site_replication_retry_event`] would raise `retry_count` on every
+/// import and escalate a healthy peer to `failed` after
+/// [`SITE_REPLICATION_RETRY_FAILED_AFTER`] of them, with the scheduling note
+/// shown to operators as `lastError`.
+pub(crate) fn ensure_site_replication_retry_event(
+    queue: &mut Vec<SiteReplicationRetryEvent>,
+    peer: &PeerInfo,
+    path: &str,
+    reason: &str,
+) -> S3Result<Vec<SiteReplicationRetryEvent>> {
+    let path = collapsed_retry_queue_path(path).unwrap_or(path);
+    if queue.iter().any(|event| retry_event_matches(event, peer, path)) {
+        return Ok(Vec::new());
+    }
+    push_site_replication_retry_event(queue, peer, path, summarize_peer_error_detail(reason), false, None)
+}
+
 pub(crate) fn upsert_site_replication_retry_event(
    queue: &mut Vec<SiteReplicationRetryEvent>,
    peer: &PeerInfo,
@@ -244,6 +264,17 @@ pub(crate) fn upsert_site_replication_retry_event(
        return Ok(Vec::new());
    }

+    push_site_replication_retry_event(queue, peer, path, detail, peer_unreachable, generation)
+}
+
+fn push_site_replication_retry_event(
+    queue: &mut Vec<SiteReplicationRetryEvent>,
+    peer: &PeerInfo,
+    path: &str,
+    detail: String,
+    peer_unreachable: bool,
+    generation: Option<u64>,
+) -> S3Result<Vec<SiteReplicationRetryEvent>> {
    let slots_needed = queue
        .len()
        .saturating_add(1)
@@ -274,7 +305,7 @@ pub(crate) fn upsert_site_replication_retry_event(
        retry_count: 1,
        failed: false,
        last_error: detail,
-        updated_at: Some(now),
+        updated_at: Some(OffsetDateTime::now_utc()),
        edit_generation: generation,
        peer_unreachable,
        deletions_recorded: false,
@@ -365,6 +396,60 @@ pub(crate) async fn enqueue_site_replication_retry_event_for_generation(
    }
 }

+/// Returns the number of peers whose snapshot entry is escalated and therefore
+/// will not carry this scheduling: the marker records a deletion that a
+/// snapshot cannot replay, and only a repair settles it, so clearing it to make
+/// the entry drainable again would drop that liability.
+pub(crate) fn record_iam_snapshot_retries(
+    state: &mut SiteReplicationState,
+    local_peer: &PeerInfo,
+    reason: &str,
+) -> S3Result<usize> {
+    let peers = state
+        .peers
+        .values()
+        .filter(|peer| {
+            peer.deployment_id != local_peer.deployment_id && !same_identity_endpoint(&peer.endpoint, &local_peer.endpoint)
+        })
+        .cloned()
+        .collect::<Vec<_>>();
+    let mut escalated = 0usize;
+    for peer in peers {
+        if state.retry_queue.iter().any(|event| {
+            retry_event_matches(event, &peer, SITE_REPLICATION_RETRY_IAM_SNAPSHOT_PATH)
+                && event.last_error == SITE_REPLICATION_RETRY_SNAPSHOT_REPLAYED_MARKER
+        }) {
+            escalated += 1;
+            continue;
+        }
+        ensure_site_replication_retry_event(&mut state.retry_queue, &peer, SITE_REPLICATION_RETRY_IAM_SNAPSHOT_PATH, reason)?;
+    }
+    Ok(escalated)
+}
+
+/// Schedule one collapsed full-IAM snapshot per remote peer after a bulk
+/// local mutation such as `import-iam`.
+pub(crate) async fn enqueue_site_replication_iam_snapshot(reason: &str) -> S3Result<()> {
+    let state = load_site_replication_state().await?;
+    if !state.enabled() {
+        return Ok(());
+    }
+    let local_peer = current_local_runtime_peer(&state);
+    let reason = reason.to_string();
+    let escalated = update_site_replication_state(move |state| record_iam_snapshot_retries(state, &local_peer, &reason)).await?;
+    if escalated > 0 {
+        warn!(
+            component = LOG_COMPONENT_ADMIN,
+            subsystem = LOG_SUBSYSTEM_SITE_REPLICATION,
+            event = EVENT_ADMIN_SITE_REPLICATION_STATE,
+            escalated,
+            result = "iam_snapshot_not_scheduled_for_escalated_peer",
+            "site replication peers hold an escalated IAM entry; the snapshot waits for a repair"
+        );
+    }
+    Ok(())
+}
+
 pub(crate) const SITE_REPLICATION_PEER_IAM_ITEM_WIRE_PATH: &str = "/rustfs/admin/v3/site-replication/peer/iam-item";

 /// Per-peer cap on recorded deletion bodies. Beyond it the peer's collapsed
@@ -168,6 +168,8 @@ mod rfc3339_map {
 pub(crate) struct PendingEndpointRefresh {
    pub(crate) id: String,
    pub(crate) peer: PeerInfo,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub(crate) ilm_expiry_override: Option<bool>,
    #[serde(default, skip_serializing_if = "BTreeMap::is_empty")]
    pub(crate) remote_peers: BTreeMap<String, PeerInfo>,
    #[serde(default, skip_serializing_if = "BTreeSet::is_empty")]
@@ -693,6 +693,122 @@ fn test_record_iam_deletion_marks_newest_wins_and_expires_by_age_only() {
    );
 }

+/// Scheduling a snapshot is not a delivery failure. Repeated imports - the
+/// normal way a bulk IAM migration is done, one archive at a time - must not
+/// walk the peer's entry up to the escalation threshold and report a healthy
+/// site as `retryStats.failed` with the scheduling note as its `lastError`.
+#[test]
+fn repeated_iam_import_snapshots_do_not_escalate_a_healthy_peer() {
+    let local = PeerInfo {
+        deployment_id: "local-dep".to_string(),
+        ..peer("local", "https://local.example.com")
+    };
+    let remote = PeerInfo {
+        deployment_id: "remote-a".to_string(),
+        ..peer("remote-a", "https://a.example.com")
+    };
+    let mut state = SiteReplicationState {
+        peers: BTreeMap::from([
+            (local.deployment_id.clone(), local.clone()),
+            (remote.deployment_id.clone(), remote),
+        ]),
+        ..Default::default()
+    };
+
+    for _ in 0..(SITE_REPLICATION_RETRY_FAILED_AFTER + 2) {
+        record_iam_snapshot_retries(&mut state, &local, "iam import scheduled a full snapshot").expect("record snapshot");
+    }
+
+    assert_eq!(state.retry_queue.len(), 1);
+    let event = &state.retry_queue[0];
+    assert_eq!(event.retry_count, 1, "a schedule must not count as a delivery attempt");
+    assert!(!event.failed, "a scheduled snapshot must not report as an escalated failure");
+}
+
+/// An escalated entry records a deletion a snapshot cannot replay: only a
+/// repair settles it. Scheduling an import snapshot must not clear that
+/// marker to make the entry drainable again, and the peer it skips has to be
+/// reported rather than silently left behind.
+#[test]
+fn an_escalated_peer_keeps_its_marker_and_is_reported() {
+    let local = PeerInfo {
+        deployment_id: "local-dep".to_string(),
+        ..peer("local", "https://local.example.com")
+    };
+    let remote = PeerInfo {
+        deployment_id: "remote-a".to_string(),
+        ..peer("remote-a", "https://a.example.com")
+    };
+    let mut state = SiteReplicationState {
+        peers: BTreeMap::from([
+            (local.deployment_id.clone(), local.clone()),
+            (remote.deployment_id.clone(), remote.clone()),
+        ]),
+        retry_queue: vec![SiteReplicationRetryEvent {
+            id: "escalated".to_string(),
+            peer_deployment_id: remote.deployment_id.clone(),
+            peer_endpoint: remote.endpoint,
+            path: SITE_REPLICATION_RETRY_IAM_SNAPSHOT_PATH.to_string(),
+            retry_count: SITE_REPLICATION_RETRY_FAILED_AFTER,
+            failed: true,
+            last_error: SITE_REPLICATION_RETRY_SNAPSHOT_REPLAYED_MARKER.to_string(),
+            deletions_recorded: true,
+            ..Default::default()
+        }],
+        ..Default::default()
+    };
+
+    let escalated =
+        record_iam_snapshot_retries(&mut state, &local, "iam import scheduled a full snapshot").expect("record snapshot retries");
+
+    assert_eq!(escalated, 1);
+    assert_eq!(state.retry_queue.len(), 1);
+    assert_eq!(
+        state.retry_queue[0].last_error, SITE_REPLICATION_RETRY_SNAPSHOT_REPLAYED_MARKER,
+        "the unreplayable-deletion marker must survive a snapshot schedule"
+    );
+}
+
+#[test]
+fn iam_import_snapshot_retry_is_recorded_once_per_remote_peer() {
+    let local = PeerInfo {
+        deployment_id: "local-dep".to_string(),
+        ..peer("local", "https://local.example.com")
+    };
+    let remote_a = PeerInfo {
+        deployment_id: "remote-a".to_string(),
+        ..peer("remote-a", "https://a.example.com")
+    };
+    let remote_b = PeerInfo {
+        deployment_id: "remote-b".to_string(),
+        ..peer("remote-b", "https://b.example.com")
+    };
+    let mut state = SiteReplicationState {
+        peers: BTreeMap::from([
+            (local.deployment_id.clone(), local.clone()),
+            (remote_a.deployment_id.clone(), remote_a),
+            (remote_b.deployment_id.clone(), remote_b),
+        ]),
+        ..Default::default()
+    };
+
+    record_iam_snapshot_retries(&mut state, &local, "IAM import snapshot pending").expect("record snapshot retries");
+
+    assert_eq!(state.retry_queue.len(), 2);
+    assert!(
+        state
+            .retry_queue
+            .iter()
+            .all(|event| event.path == SITE_REPLICATION_RETRY_IAM_SNAPSHOT_PATH)
+    );
+    assert!(
+        state
+            .retry_queue
+            .iter()
+            .all(|event| event.peer_deployment_id != local.deployment_id)
+    );
+}
+
 /// A failed deletion delivery persists a replay record next to the collapsed
 /// retry entry; a fresh entry is stamped `deletions_recorded` so a later
 /// replay can settle it, and a repeated deletion of the same entity keeps the
@@ -17,6 +17,7 @@
 use std::collections::hash_map::DefaultHasher;
 use std::hash::{Hash, Hasher};
 use std::sync::{Arc, LazyLock};
+use std::time::Duration;

 use rand::RngExt as _;
 use rustfs_storage_api as storage_contracts;
@@ -836,6 +837,39 @@ impl StorageReplicationStatsHandle {

    pub(crate) async fn site_metrics_snapshot(&self) -> ReplicationSiteMetricsSnapshot {
        let metrics = self.inner.get_sr_metrics_for_node().await;
+        // Aggregate under the read lock rather than through `get_all`: that
+        // clones every bucket's stats, and `FailStats.recent` is bounded only
+        // by the one-hour window, so an unreachable target under load - the
+        // very case an operator polls this for - makes the copy large. The
+        // windows come from the live samples; the serialized `last_minute` /
+        // `last_hour` snapshots are stamped onto per-bucket clones elsewhere
+        // and stay zero in this node-local cache.
+        let (
+            failed_count,
+            failed_bytes,
+            failed_last_minute_count,
+            failed_last_minute_bytes,
+            failed_last_hour_count,
+            failed_last_hour_bytes,
+        ) = {
+            let cache = self.inner.cache.read().await;
+            cache
+                .values()
+                .flat_map(|bucket| bucket.stats.values())
+                .fold((0i64, 0i64, 0i64, 0i64, 0i64, 0i64), |totals, stat| {
+                    let (minute, hour) = stat
+                        .fail_stats
+                        .recent_windows(Duration::from_secs(60), Duration::from_secs(3600));
+                    (
+                        totals.0.saturating_add(stat.fail_stats.count),
+                        totals.1.saturating_add(stat.fail_stats.size),
+                        totals.2.saturating_add(minute.count),
+                        totals.3.saturating_add(minute.size),
+                        totals.4.saturating_add(hour.count),
+                        totals.5.saturating_add(hour.size),
+                    )
+                })
+        };
        ReplicationSiteMetricsSnapshot {
            uptime: metrics.uptime,
            queued_curr_count: metrics.queued.curr.count,
@@ -859,6 +893,12 @@ impl StorageReplicationStatsHandle {
            proxy_delete_tag_failed: metrics.proxied.delete_tag_failed,
            replica_size: metrics.replica_size,
            replica_count: metrics.replica_count,
+            failed_count,
+            failed_bytes,
+            failed_last_minute_count,
+            failed_last_minute_bytes,
+            failed_last_hour_count,
+            failed_last_hour_bytes,
        }
    }

@@ -899,6 +939,12 @@ pub(crate) struct ReplicationSiteMetricsSnapshot {
    pub(crate) proxy_delete_tag_failed: i64,
    pub(crate) replica_size: i64,
    pub(crate) replica_count: i64,
+    pub(crate) failed_count: i64,
+    pub(crate) failed_bytes: i64,
+    pub(crate) failed_last_minute_count: i64,
+    pub(crate) failed_last_minute_bytes: i64,
+    pub(crate) failed_last_hour_count: i64,
+    pub(crate) failed_last_hour_bytes: i64,
 }

 pub(crate) async fn get_local_server_property() -> rustfs_madmin::ServerProperties {
@@ -2043,13 +2089,32 @@ pub(crate) async fn init_compression_total_memory_from_backend(store: Arc<ECStor
 #[cfg(test)]
 mod tests {
    use super::{
-        BUCKET_RESYNC_LOCK_RETRY_MAX_MS, apply_active_resync_intents, bucket_resync_transaction_lock_retry_ceiling_ms,
-        bucket_resync_transaction_lock_retry_delay, bucket_resync_transaction_lock_retry_reason,
-        bucket_targets_metadata_lock_shard, ecstore_bucket, lock_bucket_targets_metadata, new_instance_ctx,
-        retry_bucket_resync_transaction_lock, scanner_maintenance_config_file,
+        BUCKET_RESYNC_LOCK_RETRY_MAX_MS, StorageReplicationStatsHandle, apply_active_resync_intents,
+        bucket_resync_transaction_lock_retry_ceiling_ms, bucket_resync_transaction_lock_retry_delay,
+        bucket_resync_transaction_lock_retry_reason, bucket_targets_metadata_lock_shard, ecstore_bucket,
+        lock_bucket_targets_metadata, new_instance_ctx, retry_bucket_resync_transaction_lock, scanner_maintenance_config_file,
    };
    use std::time::Duration;

+    #[tokio::test]
+    async fn site_metrics_snapshot_includes_live_failure_windows() {
+        let stats = StorageReplicationStatsHandle::new();
+        let mut target = ecstore_bucket::replication::BucketReplicationStat::default();
+        target.fail_stats.add_size(2048, None::<&std::io::Error>);
+        let mut bucket = ecstore_bucket::replication::BucketReplicationStats::new();
+        bucket.stats.insert("arn:replication::remote:photos".to_string(), target);
+        stats.inner.cache.write().await.insert("photos".to_string(), bucket);
+
+        let snapshot = stats.site_metrics_snapshot().await;
+
+        assert_eq!(snapshot.failed_count, 1);
+        assert_eq!(snapshot.failed_bytes, 2048);
+        assert_eq!(snapshot.failed_last_minute_count, 1);
+        assert_eq!(snapshot.failed_last_minute_bytes, 2048);
+        assert_eq!(snapshot.failed_last_hour_count, 1);
+        assert_eq!(snapshot.failed_last_hour_bytes, 2048);
+    }
+
    #[tokio::test]
    async fn bucket_target_metadata_locks_serialize_only_matching_shards() {
        let bucket = "bucket-target-lock";
@@ -1,83 +0,0 @@
-#!/usr/bin/env bash
-# ci.yml's pull_request paths-ignore and ci-docs-only.yml's paths must be equal.
-#
-# ci-docs-only.yml exists to report the required checks for pull requests that
-# ci.yml skips. The two lists are the complement of each other, so any drift
-# breaks one of two ways, both silent:
-#
-#   - an entry only in ci.yml's paths-ignore: a PR touching only those files
-#     triggers neither workflow, nobody reports "Test and Lint" or "Quick
-#     Checks", and the PR waits on a required check forever;
-#   - an entry only in ci-docs-only.yml's paths: both workflows run, which is
-#     merely wasteful — but it also means the lists no longer describe the same
-#     intent, and the next edit is made against a wrong assumption.
-#
-# The push paths-ignore in ci.yml is deliberately NOT compared: no required
-# check is reported for push events, so it does not have to pair with anything.
-#
-# Also asserts ci-docs-only.yml still declares both companion job names, since a
-# rename there produces exactly the permanent-pending failure above.
-#
-# Usage: scripts/check_ci_paths_sync.sh
-set -euo pipefail
-
-cd "$(dirname "$0")/.."
-
-CI=".github/workflows/ci.yml"
-DOCS=".github/workflows/ci-docs-only.yml"
-
-# Print the quoted list items that follow $2 within the block introduced by $1.
-# Both files keep these as a flat list of quoted scalars, so no YAML parser is
-# needed and the script stays dependency-free like its check_* siblings.
-extract() {
-    local file="$1" event="$2" key="$3"
-    awk -v event="$event" -v key="$key" '
-        $0 ~ "^  " event ":[[:space:]]*$" { in_event = 1; next }
-        in_event && /^  [a-z_]+:[[:space:]]*$/ { in_event = 0 }
-        in_event && $0 ~ "^    " key ":[[:space:]]*$" { in_list = 1; next }
-        in_list {
-            if ($0 ~ /^      - /) {
-                item = $0
-                sub(/^      - /, "", item)
-                gsub(/^"|"$/, "", item)
-                print item
-                next
-            }
-            if ($0 !~ /^[[:space:]]*#/ && $0 !~ /^[[:space:]]*$/) in_list = 0
-        }
-    ' "$file" | sort
-}
-
-ci_list="$(extract "$CI" "pull_request" "paths-ignore")"
-docs_list="$(extract "$DOCS" "pull_request" "paths")"
-
-if [ -z "$ci_list" ] || [ -z "$docs_list" ]; then
-    echo "ERROR: could not read one of the path lists — did the file structure change?" >&2
-    echo "  $CI pull_request.paths-ignore: $(printf '%s' "$ci_list" | grep -c . || true) entries" >&2
-    echo "  $DOCS pull_request.paths: $(printf '%s' "$docs_list" | grep -c . || true) entries" >&2
-    exit 1
-fi
-
-status=0
-
-if ! diff_out="$(diff <(printf '%s\n' "$ci_list") <(printf '%s\n' "$docs_list"))"; then
-    echo "ERROR: $CI pull_request paths-ignore and $DOCS paths have drifted." >&2
-    echo "  '<' is only in $CI, '>' is only in $DOCS:" >&2
-    printf '%s\n' "$diff_out" | sed 's/^/    /' >&2
-    status=1
-fi
-
-for job_name in "Test and Lint" "Quick Checks"; do
-    if ! grep -q "name: ${job_name}\$" "$DOCS"; then
-        echo "ERROR: $DOCS no longer declares a job named '${job_name}'." >&2
-        echo "  It is a required status check; without a companion job here, a" >&2
-        echo "  docs-only PR waits on it forever." >&2
-        status=1
-    fi
-done
-
-if [ "$status" -ne 0 ]; then
-    exit 1
-fi
-
-echo "OK: ci.yml and ci-docs-only.yml path lists agree ($(printf '%s\n' "$ci_list" | wc -l | tr -d ' ') entries)"
@@ -55,8 +55,11 @@ cd "$(dirname "$0")/.."
 # now reports an unreadable configuration as a plain string instead of raising
 # an S3 error per arm (24 invocation lines removed from
 # rustfs/src/admin/handlers/bucket_meta.rs; measured after merging the two).
+# 1589 -> 1588 on 2026-09-08: the GA blocker set (rustfs/backlog#2366) added
+# three invocation lines to the endpoint-refresh paths and folded the five
+# copies of the concurrent-change error into one constructor, netting -1.
 S3S_IMPORT_FILES_BASELINE=213
-S3_ERROR_LINES_BASELINE=1589
+S3_ERROR_LINES_BASELINE=1588
 # ecstore-scoped ratchet (rustfs/backlog#1842): the storage engine must not
 # know S3 wire/DTO types (ARCHITECTURE.md invariant 4). The S3-*consuming*
 # client was extracted to crates/s3-client, where s3s usage is legitimate;
@@ -542,7 +542,7 @@ def yaml_scalar_continues(lines: list[str], index: int, indent: int) -> bool:
 def check_quick_checks(root: Path) -> list[str]:
    errors: list[str] = []
    bypass_key = r'''(?:if|continue-on-error|needs|"if"|"continue-on-error"|"needs"|'if'|'continue-on-error'|'needs')\s*:'''
-    for name in ("ci.yml", "ci-docs-only.yml"):
+    for name in ("ci.yml",):
        relative = f".github/workflows/{name}"
        path = root / relative
        job = yaml_block(path.read_text().splitlines(), "quick-checks", 2) if path.is_file() else None
@@ -1165,7 +1165,6 @@ class SelfTests(unittest.TestCase):
                ".github/workflows/ci.yml": caller.replace(
                    "    steps:", "    if: github.event_name != 'pull_request' || github.event.action != 'closed'\n    steps:"
                ),
-                ".github/workflows/ci-docs-only.yml": caller,
                ".github/actions/quick-checks/action.yml": action,
            }
            for relative, source in sources.items():
@@ -1173,7 +1172,7 @@ class SelfTests(unittest.TestCase):
                path.parent.mkdir(parents=True, exist_ok=True)
                path.write_text(source)
            self.assertEqual(check_quick_checks(root), [])
-            for relative in (".github/workflows/ci.yml", ".github/workflows/ci-docs-only.yml"):
+            for relative in (".github/workflows/ci.yml",):
                source = sources[relative]
                mutations = {
                    "different action": source.replace("./.github/actions/quick-checks", "./.github/actions/other"),
@@ -0,0 +1,369 @@
+#!/usr/bin/env python3
+"""Select safe documentation-only CI and verify the complete required job set."""
+from __future__ import annotations
+
+import json
+import os
+from pathlib import Path, PurePosixPath
+import re
+import subprocess
+import sys
+import tempfile
+import unittest
+
+ROOT = Path(__file__).resolve().parent.parent
+ALWAYS_JOBS = ("classify-changes", "typos", "quick-checks")
+CODE_JOBS = (
+    "test-and-lint", "test-ilm-integration-serial", "test-and-lint-rio-v2",
+    "connect-short-credential-boundary", "test-and-lint-protocols",
+    "build-rustfs-debug-binary", "uring-integration", "e2e-tests",
+    "s3-implemented-tests", "s3-lifecycle-behavior-tests",
+)
+OPTIONAL_JOBS = ("build-rustfs-debug-binary-rio-v2", "e2e-tests-rio-v2", "e2e-full")
+NON_VALIDATION_JOBS = {"required-checks", "cancel-closed-pr-runs", "alert-on-failure"}
+
+
+def documentation_path(path: str) -> bool:
+    parts = PurePosixPath(path).parts
+    if not parts or path.startswith("/") or any(part in (".", "..") for part in parts) or any(ord(c) < 32 for c in path):
+        return False
+    if parts[-1] == "AGENTS.md":
+        return True
+    if len(parts) == 1 and (path.endswith(".md") or path == "LICENSE" or path.startswith("LICENSE-")):
+        return True
+    if path.startswith(("docs/", ".agents/skills/")) and path.endswith(".md"):
+        return True
+    return path.startswith("docs/") and path.endswith((".png", ".jpg", ".svg"))
+
+
+def select_mode(event: str, base: str, head: str, root: Path) -> str:
+    if event != "pull_request" or not all(re.fullmatch(r"[0-9a-f]{40}", sha) for sha in (base, head)):
+        return "full"
+    try:
+        changed = subprocess.check_output(
+            ["git", "diff", "--no-ext-diff", "--no-textconv", "--no-renames", "--name-only", "-z", base, head, "--"],
+            cwd=root, stderr=subprocess.PIPE,
+        ).decode("utf-8")
+    except (subprocess.CalledProcessError, UnicodeError):
+        return "full"
+    paths = changed.rstrip("\0").split("\0") if changed else []
+    return "docs" if paths and all(documentation_path(path) for path in paths) else "full"
+
+
+def expected_results(mode: str, event: str, ref: str) -> dict[str, str]:
+    if event not in ("pull_request", "push", "merge_group", "schedule", "workflow_dispatch"):
+        raise ValueError(f"unsupported CI event: {event!r}")
+    if mode not in ("docs", "full") or (mode == "docs" and event != "pull_request"):
+        raise ValueError(f"invalid CI selection: {mode!r} for {event!r}")
+    expected = {job: "success" for job in ALWAYS_JOBS}
+    expected.update({job: "success" if mode == "full" else "skipped" for job in CODE_JOBS})
+    rio = mode == "full" and event in ("schedule", "workflow_dispatch")
+    expected.update({job: "success" if rio else "skipped" for job in OPTIONAL_JOBS[:2]})
+    full = mode == "full" and (event in ("merge_group", "workflow_dispatch") or (event == "push" and ref == "refs/heads/main"))
+    expected["e2e-full"] = "success" if full else "skipped"
+    return expected
+
+
+def verify_results(needs: object, event: str, ref: str) -> list[str]:
+    if not isinstance(needs, dict):
+        return ["needs must be a job-result object"]
+    selection = needs.get("classify-changes", {})
+    outputs = selection.get("outputs", {}) if isinstance(selection, dict) else {}
+    mode = outputs.get("mode") if isinstance(outputs, dict) else None
+    try:
+        expected = expected_results(mode, event, ref)
+    except ValueError as error:
+        return [str(error)]
+    errors = []
+    if set(needs) != set(expected):
+        errors.append(f"job set differs: missing={sorted(set(expected) - set(needs))}, unexpected={sorted(set(needs) - set(expected))}")
+    for job, required in expected.items():
+        result = needs.get(job, {})
+        actual = result.get("result") if isinstance(result, dict) else None
+        if actual != required:
+            errors.append(f"{job}: expected {required}, got {actual!r}")
+    return errors
+
+
+def check_workflow(root: Path) -> list[str]:
+    # Reuse the repository's canonical-indentation checker; actionlint validates YAML syntax.
+    from check_test_wiring import yaml_block, yaml_scalar_continues
+
+    errors = []
+    lines = (root / ".github/workflows/ci.yml").read_text().splitlines()
+    jobs = yaml_block(lines, "jobs", 0) or []
+    names = set()
+    for index, line in enumerate(jobs):
+        if not re.match(r"^  \S", line) or line.lstrip().startswith("#"):
+            continue
+        header = re.fullmatch(r'''  (["']?)([A-Za-z_][A-Za-z0-9_-]*)\1\s*:\s*(?:#.*)?''', line)
+        if header is None:
+            errors.append("CI job declarations must use single-line job IDs")
+            continue
+        name = header[2]
+        if name in names:
+            errors.append(f"duplicate CI job ID: {name}")
+        names.add(name)
+        jobs[index] = f"  {name}:"
+    required = set(ALWAYS_JOBS + CODE_JOBS + OPTIONAL_JOBS)
+    if names - NON_VALIDATION_JOBS != required:
+        errors.append("CI verification jobs and the required gate contract differ")
+    for job in required:
+        block = yaml_block(jobs, job, 2) or []
+        if any(re.match(r"\s+(?:- )?[\"']?continue-on-error[\"']?\s*:", line) for line in block):
+            errors.append(f"{job} cannot convert a validation failure into success")
+    gate = yaml_block(jobs, "required-checks", 2) or []
+    def scalar(block, key, indent):
+        prefix = " " * indent + key + ": "
+        matches = [index for index, line in enumerate(block) if line.startswith(prefix)]
+        if len(matches) != 1:
+            return None
+        index = matches[0]
+        if yaml_scalar_continues(block, index, indent):
+            return None
+        return block[index][len(prefix):]
+
+    display_names = {}
+    for job in names:
+        block = [re.sub(r'''^    (?:'name'|"name")\s*:\s*''', "    name: ", line)
+                 for line in yaml_block(jobs, job, 2) or []]
+        value = scalar(block, "name", 4)
+        display = re.fullmatch(r'''(?:"([^"\\]*)"|'([^']*)'|([^'"#][^#]*?))(?:\s+#.*)?\s*''', (value or "").strip())
+        if display is None or (display[3] is not None and display[3].startswith(tuple("|>*&!{[?"))):
+            errors.append(f"{job} must use a verifiable single-line display name")
+            continue
+        name = next(value for value in display.groups() if value is not None)
+        if "${{" in name and (job != "test-and-lint-protocols" or name != "Test and Lint (${{ matrix.features.name }})"):
+            errors.append(f"{job} has an unverifiable dynamic display name")
+        display_names[job] = name
+
+    dependencies = yaml_block(gate, "needs", 4) or []
+    declared = [line.strip().removeprefix("- ") for line in dependencies if line.strip()]
+    if set(declared) != required or len(declared) != len(required):
+        errors.append("required-checks must directly depend on every verification job exactly once")
+    if display_names.get("required-checks") != "Test and Lint" or list(display_names.values()).count("Test and Lint") != 1:
+        errors.append("Test and Lint must uniquely name the aggregate gate")
+    if scalar(gate, "if", 4) != "always() && (github.event_name != 'pull_request' || github.event.action != 'closed')":
+        errors.append("required-checks must run after failed or skipped dependencies")
+    if scalar(gate, "shell", 8) != "bash" or scalar(gate, "run", 8) != "python3 scripts/ci_gate.py verify" or scalar(gate, "CI_NEEDS", 10) != "${{ toJSON(needs) }}":
+        errors.append("required-checks must verify the actual needs results")
+    if any(re.match(r'''\s+(?:- )?(?:["']?continue-on-error["']?\s*:|["']?if["']?\s*:)''', line) and not line.startswith("    if:") for line in gate):
+        errors.append("required-checks cannot ignore failures")
+    pr = yaml_block(lines, "pull_request", 2) or []
+    if any(line.strip().startswith(("paths:", "paths-ignore:")) for line in pr):
+        errors.append("all pull requests must enter the single CI workflow")
+    if (root / ".github/workflows/ci-docs-only.yml").exists():
+        errors.append("the duplicate required-status companion must be removed")
+    return errors
+
+
+class SelfTests(unittest.TestCase):
+    def test_documentation_paths_do_not_hide_build_or_fixture_changes(self):
+        for path in ("README.md", "AGENTS.md", "crates/utils/AGENTS.md", "docs/testing/README.md", "docs/diagram.svg", ".agents/skills/example/SKILL.md"):
+            self.assertTrue(documentation_path(path), path)
+        for path in ("", "src/lib.rs", "crates/foo/tests/fixtures/data.md", "Cargo.lock", "build.rs", "deploy/chart.yaml", ".github/workflows/ci.yml", "scripts/dev_build.sh", "assets/logo.png", "docs/test.rs", "README.md\n", "../README.md"):
+            self.assertFalse(documentation_path(path), path)
+
+    def test_git_range_includes_deleted_source_and_rename_origins(self):
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            def git(*args):
+                return subprocess.check_output(["git", "-c", "core.hooksPath=/dev/null", "-c", "user.name=CI Test", "-c", "user.email=ci@example.invalid", *args], cwd=root, stderr=subprocess.PIPE).decode().strip()
+            git("init", "-q")
+            (root / "server.rs").write_text("fn main() {}\n")
+            (root / "README.md").write_text("old\n")
+            git("add", "."); git("commit", "-qm", "base")
+            base = git("rev-parse", "HEAD")
+            (root / "README.md").write_text("new\n")
+            git("add", "."); git("commit", "-qm", "docs")
+            docs = git("rev-parse", "HEAD")
+            self.assertEqual(select_mode("pull_request", base, docs, root), "docs")
+            (root / "server.rs").rename(root / "server.md")
+            git("add", "-A"); git("commit", "-qm", "rename source")
+            head = git("rev-parse", "HEAD")
+            self.assertEqual(select_mode("pull_request", base, head, root), "full")
+            self.assertEqual(select_mode("pull_request", docs, docs, root), "full")
+            self.assertEqual(select_mode("pull_request", "0" * 40, head, root), "full")
+            self.assertEqual(select_mode("pull_request", "--output=bad", head, root), "full")
+            self.assertEqual(select_mode("merge_group", base, docs, root), "full")
+
+    def test_event_contract_requires_complete_candidate_and_optional_lanes(self):
+        ordinary = expected_results("full", "pull_request", "refs/pull/1/merge")
+        self.assertEqual({job for job, state in ordinary.items() if state == "skipped"}, set(OPTIONAL_JOBS))
+        docs = expected_results("docs", "pull_request", "refs/pull/1/merge")
+        self.assertEqual({job for job, state in docs.items() if state == "success"}, set(ALWAYS_JOBS))
+        for event in ("schedule", "workflow_dispatch", "merge_group", "push"):
+            result = expected_results("full", event, "refs/heads/main")
+            self.assertEqual(result["e2e-full"], "skipped" if event == "schedule" else "success")
+            self.assertEqual(result["e2e-tests-rio-v2"], "success" if event in ("schedule", "workflow_dispatch") else "skipped")
+            with self.assertRaises(ValueError):
+                expected_results("docs", event, "refs/heads/main")
+
+    def test_every_wrong_result_missing_job_or_selection_fails_closed(self):
+        for mode, event in (("full", "pull_request"), ("docs", "pull_request"), ("full", "schedule"), ("full", "workflow_dispatch"), ("full", "merge_group")):
+            good = {job: {"result": value} for job, value in expected_results(mode, event, "refs/heads/main").items()}
+            good["classify-changes"]["outputs"] = {"mode": mode}
+            self.assertEqual(verify_results(good, event, "refs/heads/main"), [])
+            for job in good:
+                for value in ("success", "skipped", "failure", "cancelled", "neutral", "", None):
+                    if value == good[job]["result"]:
+                        continue
+                    with self.subTest(mode=mode, event=event, job=job, result=value):
+                        bad = {**good, job: {**good[job], "result": value}}
+                        self.assertTrue(verify_results(bad, event, "refs/heads/main"))
+                self.assertTrue(verify_results({key: value for key, value in good.items() if key != job}, event, "refs/heads/main"))
+                missing_result = {key: value for key, value in good[job].items() if key != "result"}
+                self.assertTrue(verify_results({**good, job: missing_result}, event, "refs/heads/main"))
+            self.assertTrue(verify_results({**good, "unknown-job": {"result": "success"}}, event, "refs/heads/main"))
+            for selection in ({}, {"mode": ""}, {"mode": True}, []):
+                bad = {**good, "classify-changes": {"result": "success", "outputs": selection}}
+                self.assertTrue(verify_results(bad, event, "refs/heads/main"))
+
+    def test_repository_wiring_and_missing_dependency_regression(self):
+        self.assertEqual(check_workflow(ROOT), [])
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            (root / ".github/workflows").mkdir(parents=True)
+            source = (ROOT / ".github/workflows/ci.yml").read_text()
+            path = root / ".github/workflows/ci.yml"
+            for job in ALWAYS_JOBS + CODE_JOBS + OPTIONAL_JOBS:
+                before, gate = source.split("  required-checks:\n", 1)
+                path.write_text(before + "  required-checks:\n" + gate.replace(f"      - {job}\n", "", 1))
+                self.assertTrue(check_workflow(root), job)
+            for old, new in (
+                ("run: python3 scripts/ci_gate.py verify", "run: python3 scripts/ci_gate.py verify || true"),
+                ("run: python3 scripts/ci_gate.py verify", "run: python3 scripts/ci_gate.py verify\n          || true"),
+                ("CI_NEEDS: ${{ toJSON(needs) }}", "CI_NEEDS: '{}'"),
+                ("name: Test and Lint\n", "name: Unrequired result\n"),
+                ("        shell: bash\n        run: python3 scripts/ci_gate.py verify", "        shell: echo {0}\n        run: python3 scripts/ci_gate.py verify"),
+                ("        shell: bash\n        run: python3 scripts/ci_gate.py verify", "        run: python3 scripts/ci_gate.py verify"),
+                ("        run: python3 scripts/ci_gate.py verify", '        "if": false\n        run: python3 scripts/ci_gate.py verify'),
+            ):
+                path.write_text(source.replace(old, new))
+                self.assertTrue(check_workflow(root), new)
+            for job in ALWAYS_JOBS + CODE_JOBS + OPTIONAL_JOBS:
+                for field in ("continue-on-error", '"continue-on-error"', "'continue-on-error'"):
+                    path.write_text(source.replace(f"  {job}:\n", f"  {job}:\n    {field}: true\n", 1))
+                    self.assertTrue(check_workflow(root), (job, field))
+                    before, block = source.split(f"  {job}:\n", 1)
+                    block = block.replace("      - name:", f"      - {field}: true\n        name:", 1)
+                    path.write_text(before + f"  {job}:\n" + block)
+                    self.assertTrue(check_workflow(root), (job, field, "step"))
+            path.write_text(source + "\n  cancel-after-test-and-lint-failure:\n    runs-on: ubuntu-latest\n")
+            self.assertTrue(check_workflow(root))
+
+    def test_job_ids_and_display_names_cannot_hide_validation(self):
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            (root / ".github/workflows").mkdir(parents=True)
+            source = (ROOT / ".github/workflows/ci.yml").read_text()
+            path = root / ".github/workflows/ci.yml"
+            for header in ("typos", "'typos'", '"typos"'):
+                path.write_text(source.replace("  typos:\n", f"  {header}: # spelling\n"))
+                self.assertEqual(check_workflow(root), [], header)
+            for name in ("Test and Lint # required", "'Test and Lint'", '"Test and Lint" # required'):
+                path.write_text(source.replace("    name: Test and Lint\n", f"    name: {name}\n"))
+                self.assertEqual(check_workflow(root), [], name)
+            for key in ("'name'", '"name"'):
+                path.write_text(source.replace("    name: Typos\n", f"    {key}: Typos\n"))
+                self.assertEqual(check_workflow(root), [], key)
+            for header in ("new_test", "NewTest", "_new_test", "'new_test'", '"new_test"', '"new\\u005ftest"'):
+                path.write_text(source + f"\n  {header}:\n    name: New test\n    runs-on: ubuntu-latest\n    steps:\n      - run: exit 1\n")
+                self.assertTrue(check_workflow(root), header)
+            path.write_text(source + "\n  'typos':\n    name: Duplicate\n    runs-on: ubuntu-latest\n    steps:\n      - run: exit 1\n")
+            self.assertIn("duplicate CI job ID: typos", check_workflow(root))
+            for name in (
+                "Test and Lint", "Test and Lint # duplicate", "'Test and Lint'",
+                '"Test and Lint" # duplicate', '"Test\\u0020and Lint"',
+                ">-\n      Test and Lint", "|-\n      Test and Lint", "Test and\n      Lint",
+                "*required_name", "&required_name Test and Lint", "!!str Test and Lint",
+                "${{ 'Test and Lint' }}", '"${{ github.event.inputs.check_name }}"',
+            ):
+                path.write_text(source.replace("    name: Typos\n", f"    name: {name}\n"))
+                self.assertTrue(check_workflow(root), name)
+            path.write_text(source.replace("    name: Typos\n", ""))
+            self.assertIn("typos must use a verifiable single-line display name", check_workflow(root))
+
+    def test_verify_command_preserves_failures(self):
+        good = {job: {"result": value} for job, value in expected_results("full", "pull_request", "refs/pull/1/merge").items()}
+        good["classify-changes"]["outputs"] = {"mode": "full"}
+        failed = {**good, "e2e-tests": {"result": "failure"}}
+        for needs, code in ((json.dumps(good), 0), (json.dumps(failed), 1), ("{}", 1), ("{", 1)):
+            with self.subTest(needs=needs):
+                env = dict(os.environ, CI_NEEDS=needs, GITHUB_EVENT_NAME="pull_request", GITHUB_REF="refs/pull/1/merge")
+                result = subprocess.run([sys.executable, str(Path(__file__).resolve()), "verify"], env=env, capture_output=True, text=True)
+                self.assertEqual(result.returncode, code, result.stderr)
+                self.assertIn("ERROR:" if code else "CI contract passed", result.stderr if code else result.stdout)
+
+    def test_actual_selector_bootstrap_uses_base_policy_and_fails_closed(self):
+        from check_test_wiring import yaml_block
+        jobs = yaml_block((ROOT / ".github/workflows/ci.yml").read_text().splitlines(), "jobs", 0)
+        selector = yaml_block(jobs, "classify-changes", 2)
+        body = "\n".join(line[10:] for line in selector[selector.index("        run: |") + 1:])
+        for event, changed, base_sha, available, broken, expected in (
+            ("pull_request", "README.md", "b" * 40, True, False, "docs"),
+            ("pull_request", "src/server.rs", "b" * 40, True, False, "full"),
+            ("pull_request", "README.md", "b" * 40, False, False, "full"),
+            ("merge_group", "README.md", "b" * 40, False, False, "full"),
+            ("pull_request", "README.md", "b" * 40, True, True, None),
+            ("pull_request", "README.md", "", True, True, "full"),
+        ):
+            with self.subTest(event=event, changed=changed, available=available, broken=broken), tempfile.TemporaryDirectory() as directory:
+                root = Path(directory)
+                (root / "scripts").mkdir()
+                (root / "scripts/ci_gate.py").write_text("raise SystemExit(71)\n")
+                (root / "python3").symlink_to(sys.executable)
+                base = root / "base-policy.py"
+                base.write_text("raise SystemExit(29)\n" if broken else Path(__file__).read_text())
+                git = root / "git"
+                git.write_text('''#!/bin/sh
+if [ "$1" = show ]; then
+  [ "$2" = "$CI_BASE_SHA:scripts/ci_gate.py" ] || exit 19
+  [ "$BASE_AVAILABLE" = yes ] || exit 128
+  cat "$BASE_POLICY"
+elif [ "$1" = diff ]; then
+  printf '%s\\0' "$CHANGED_PATH"
+else
+  exit 20
+fi
+''')
+                git.chmod(0o755)
+                output = root / "output"
+                output.touch()
+                env = dict(os.environ, GITHUB_EVENT_NAME=event, CI_BASE_SHA=base_sha, GITHUB_SHA="c" * 40,
+                           RUNNER_TEMP=str(root), GITHUB_OUTPUT=str(output), BASE_POLICY=str(base),
+                           BASE_AVAILABLE="yes" if available else "no", CHANGED_PATH=changed,
+                           PATH=f"{root}{os.pathsep}{os.environ['PATH']}")
+                result = subprocess.run(["bash", "--noprofile", "--norc", "-e", "-o", "pipefail", "-c", body], cwd=root, env=env, capture_output=True, text=True)
+                self.assertEqual(result.returncode, 29 if expected is None else 0, result.stderr)
+                self.assertEqual(output.read_text(), "" if expected is None else f"mode={expected}\n")
+
+
+def main() -> int:
+    if sys.argv[1:] == ["--self-test"]:
+        return not unittest.TextTestRunner(verbosity=2).run(unittest.defaultTestLoader.loadTestsFromTestCase(SelfTests)).wasSuccessful()
+    if sys.argv[1:] == ["select"]:
+        mode = select_mode(os.environ.get("GITHUB_EVENT_NAME", ""), os.environ.get("CI_BASE_SHA", ""), os.environ.get("GITHUB_SHA", ""), Path.cwd())
+        with open(os.environ["GITHUB_OUTPUT"], "a") as output:
+            output.write(f"mode={mode}\n")
+        print(f"CI selection: {mode}")
+        return 0
+    if sys.argv[1:] == ["verify"]:
+        try:
+            errors = verify_results(json.loads(os.environ["CI_NEEDS"]), os.environ.get("GITHUB_EVENT_NAME", ""), os.environ.get("GITHUB_REF", ""))
+        except (KeyError, ValueError) as error:
+            errors = [str(error)]
+    elif sys.argv[1:] == ["--check-workflow"]:
+        errors = check_workflow(ROOT)
+    else:
+        print("usage: ci_gate.py {select|verify|--check-workflow|--self-test}", file=sys.stderr)
+        return 2
+    for error in errors:
+        print(f"ERROR: {error}", file=sys.stderr)
+    if not errors:
+        print("CI contract passed")
+    return bool(errors)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
@@ -0,0 +1,87 @@
+#!/usr/bin/env bash
+# Publish the nightly DEB/RPM as assets of the rolling `nightly` release on
+# rustfs/auto-testing, replacing the previous build's files in place.
+#
+# Required environment:
+#   ASSETS_TOKEN  token with contents:write on rustfs/auto-testing
+#   DEB_FILE      path to the built .deb
+#   RPM_FILE      path to the built .rpm
+#   DEB_DATE      build date (YYYY-MM-DD)
+#   BUILD_REF     branch/ref the nightly was built from
+# Optional environment:
+#   GITHUB_SHA / GITHUB_RUN_ID / GITHUB_REPOSITORY / GITHUB_SERVER_URL
+#
+# Plain curl + python3 by design: the nightly build fleet has no gh CLI.
+
+set -euo pipefail
+
+: "${ASSETS_TOKEN:?ASSETS_TOKEN is required}"
+: "${DEB_FILE:?DEB_FILE is required}"
+: "${RPM_FILE:?RPM_FILE is required}"
+: "${DEB_DATE:?DEB_DATE is required}"
+: "${BUILD_REF:?BUILD_REF is required}"
+
+for f in "${DEB_FILE}" "${RPM_FILE}"; do
+  [ -f "$f" ] || { echo "missing package: $f" >&2; exit 1; }
+done
+
+API="https://api.github.com/repos/rustfs/auto-testing"
+UPLOADS="https://uploads.github.com/repos/rustfs/auto-testing/releases"
+AUTH="Authorization: token ${ASSETS_TOKEN}"
+SOURCE_SHA="$(git rev-parse HEAD 2>/dev/null || echo "${GITHUB_SHA:-unknown}")"
+
+release_id="$(curl -fsS --retry 3 -H "${AUTH}" "${API}/releases/tags/nightly" \
+  | python3 -c 'import json,sys; print(json.load(sys.stdin).get("id", ""))' 2>/dev/null || true)"
+if [ -z "${release_id}" ]; then
+  echo "creating the rolling nightly release"
+  release_id="$(curl -fsS --retry 3 -X POST -H "${AUTH}" -H "Content-Type: application/json" \
+    -d '{"tag_name":"nightly","name":"Nightly builds","body":"Rolling nightly builds. Assets are replaced on every build; the release body documents the provenance of the current files."}' \
+    "${API}/releases" | python3 -c 'import json,sys; print(json.load(sys.stdin)["id"])')"
+fi
+[ -n "${release_id}" ] || { echo "could not resolve the nightly release id" >&2; exit 1; }
+
+upload_asset() {
+  local name="$1" file="$2" asset_id
+  asset_id="$(curl -fsS --retry 3 -H "${AUTH}" "${API}/releases/tags/nightly" \
+    | ASSET_NAME="${name}" python3 -c '
+import json, sys, os
+d = json.load(sys.stdin)
+name = os.environ["ASSET_NAME"]
+print(next((a["id"] for a in d.get("assets", []) if a["name"] == name), ""))')"
+  if [ -n "${asset_id}" ]; then
+    curl -fsS --retry 3 -X DELETE -H "${AUTH}" "${API}/releases/assets/${asset_id}" >/dev/null
+  fi
+  curl -fsS --retry 3 --max-time 900 -X POST \
+    -H "${AUTH}" -H "Content-Type: application/octet-stream" \
+    --data-binary "@${file}" \
+    "${UPLOADS}/${release_id}/assets?name=${name}" >/dev/null
+  echo "uploaded ${name}"
+}
+
+upload_asset "rustfs-nightly-latest.deb" "${DEB_FILE}"
+upload_asset "rustfs-nightly-latest.rpm" "${RPM_FILE}"
+
+DEB_SHA="$(sha256sum "${DEB_FILE}" | cut -d ' ' -f 1)"
+RPM_SHA="$(sha256sum "${RPM_FILE}" | cut -d ' ' -f 1)"
+RUN_URL="${GITHUB_SERVER_URL:-https://github.com}/${GITHUB_REPOSITORY:-/rustfs/rustfs}/actions/runs/${GITHUB_RUN_ID:-0}"
+export BUILD_REF SOURCE_SHA DEB_DATE DEB_FILE RPM_FILE DEB_SHA RPM_SHA RUN_URL
+
+python3 - << 'PY' > /tmp/release-body.json
+import json, os
+e = os.environ
+deb_mb = os.path.getsize(e["DEB_FILE"]) // 1048576
+rpm_mb = os.path.getsize(e["RPM_FILE"]) // 1048576
+body = (
+    f"Nightly build from `{e['BUILD_REF']}@{e['SOURCE_SHA'][:12]}`, built {e['DEB_DATE']}.\n\n"
+    f"[Build run]({e['RUN_URL']}). The `latest` assets are replaced in place on every nightly.\n\n"
+    f"| Asset | Size | SHA256 |\n|---|---|---|\n"
+    f"| rustfs-nightly-latest.deb (={e['DEB_FILE']}) | {deb_mb} MB | `{e['DEB_SHA']}` |\n"
+    f"| rustfs-nightly-latest.rpm (={e['RPM_FILE']}) | {rpm_mb} MB | `{e['RPM_SHA']}` |\n"
+)
+print(json.dumps({"body": body}))
+PY
+
+curl -fsS --retry 3 -X PATCH -H "${AUTH}" -H "Content-Type: application/json" \
+  --data-binary @/tmp/release-body.json "${API}/releases/${release_id}" >/dev/null
+
+echo "published ${DEB_FILE} and ${RPM_FILE} to the rustfs/auto-testing 'nightly' release"
@@ -163,12 +163,13 @@ SH
                self.assertFalse(self.store.exists())
                self.assertEqual(list(self.root.glob("nightly-awscli.*")), [])

-    def test_checkout_sha_mismatch_fails_before_upload(self):
+    def test_manifest_advertises_checked_out_head_even_when_github_sha_differs(self):
+        # With a ref override (NIGHTLY_BRANCH variable / dispatch `branch`
+        # input) the checked-out HEAD intentionally differs from GITHUB_SHA;
+        # the candidate manifest must record the tree that was built.
        result = self.run_publish(GITHUB_SHA="f" * 40)
-        self.assertNotEqual(result.returncode, 0)
-        self.assertIn("Checkout SHA", result.stderr)
-        self.assertFalse(self.output.exists())
-        self.assertFalse((self.root / "aws.log").exists())
+        self.assertEqual(result.returncode, 0, result.stderr)
+        self.assertEqual(self.manifest()["source_sha"], self.sha)

    def test_same_date_builds_and_reruns_keep_distinct_candidates(self):
        urls = []
@@ -837,6 +837,55 @@ emit_step_result() {
            self.assertIn(value, contents)
        self.assertNotIn("OLD RUN EVIDENCE", contents)

+    def test_performance_commands_bind_runner_selection_and_preserve_failures(self) -> None:
+        self.prepare("performance")
+        source = self.source.splitlines()
+        job = yaml_block(source, "performance-test", 2)
+        runner = WorkflowSteps()
+        runner.directory = self.directory / "workspace with spaces"
+        scripts = runner.directory / "auto-testing"
+        scripts.mkdir(parents=True)
+        wrapper = scripts / "rustfs_performance_test.sh"
+        wrapper.write_text(f"#!{sys.executable}\nimport json, os, sys\n" +
+                           "print(json.dumps({'args': sys.argv[1:], 'env': {key: os.environ.get(key) for key in " +
+                           "('RUSTFS_BENCH_SCRIPT', 'RUSTFS_WARP_METHODS', 'RUSTFS_WARP_SIZES', " +
+                           "'RUSTFS_WARP_DURATION', 'RUSTFS_WARP_CONCURRENCY', 'WARP_METHODS', " +
+                           "'WARP_SIZES', 'WARP_DURATION', 'WARP_CONCURRENCY')}}))\n" +
+                           "sys.exit(int(os.environ['FAKE_BENCH_EXIT']))\n")
+        wrapper.chmod(0o755)
+        runner.steps = named_steps(job)
+        for methods, sizes, duration, concurrency in (
+            ("get", "1KiB", "1s", "7"), ("all", "all", "5m", "64"), ("", "", "5m", "64")
+        ):
+            runner.context = {"github.workspace": str(runner.directory), "inputs.test_method": methods,
+                              "inputs.object_size": sizes, "inputs.warp_duration || '5m'": duration,
+                              "inputs.warp_concurrency || '64'": concurrency}
+            runner.env = {**self.env, "RUSTFS_BENCH_SCRIPT": "/unverified/home-script.sh",
+                          "RUSTFS_WARP_METHODS": "put", "RUSTFS_WARP_SIZES": "64MiB",
+                          "RUSTFS_WARP_DURATION": "99h", "RUSTFS_WARP_CONCURRENCY": "2",
+                          "WARP_DURATION": "88h", "WARP_CONCURRENCY": "3", "WARP_METHODS": "mixed", "WARP_SIZES": "32MiB",
+                          "LOG_FILE": str(self.directory / "suite.log")}
+            runner.env.update(runner.step_env(job, indent=4))
+            for step, number in (("Run benchmark (GET/PUT/MIXED)", "5"), ("Analyze results", "6")):
+                for code in (0, 42):
+                    with self.subTest(methods=methods, sizes=sizes, step=step, exit=code):
+                        runner.env["FAKE_BENCH_EXIT"] = str(code)
+                        result = runner.run_step(step)
+                        self.assertEqual(result.returncode, code, result.stderr)
+                        invocation = json.loads(result.stdout)
+                        expected = ["--step", number, "-y", "--log-file", runner.env["LOG_FILE"]]
+                        self.assertEqual(invocation["args"], expected)
+                        self.assertEqual(invocation["env"]["RUSTFS_BENCH_SCRIPT"], str(scripts / "rustfs_performance_testing.sh"))
+                        self.assertEqual(invocation["env"]["RUSTFS_WARP_METHODS"], methods)
+                        self.assertEqual(invocation["env"]["RUSTFS_WARP_SIZES"], sizes)
+                        self.assertEqual(invocation["env"]["RUSTFS_WARP_DURATION"], duration)
+                        self.assertEqual(invocation["env"]["RUSTFS_WARP_CONCURRENCY"], concurrency)
+                        if number == "6":
+                            self.assertEqual(invocation["env"]["WARP_METHODS"], methods)
+                            self.assertEqual(invocation["env"]["WARP_SIZES"], sizes)
+                            self.assertEqual(invocation["env"]["WARP_DURATION"], duration)
+                            self.assertEqual(invocation["env"]["WARP_CONCURRENCY"], concurrency)
+

 if __name__ == "__main__":
    unittest.main()
Author	SHA1	Message	Date
hector	8552c9f8aa	feat(nightly): publish packages as assets of the rolling 'nightly' release (#7592 ) Replace the assets-branch scheme with a proper GitHub Release on rustfs/auto-testing: a single 'nightly' release whose deb/rpm assets are replaced in place on every build. This is the standard channel — visible on the repo's Releases page, stable download URLs, no git history growth (release assets live outside the repository). - New scripts/release/publish_nightly_assets.sh: resolves-or-creates the 'nightly' release via the REST API, deletes same-name assets, uploads rustfs-nightly-latest.{deb,rpm}, then PATCHes the release body with the build provenance (ref@sha, run link, sizes, SHA256). Plain curl + python3, no gh CLI (the build fleet has none — #7586). - The workflow step shrinks to invoking the script; full flow exercised end-to-end against the real release with probe files (create / upload / overwrite / download round-trip / body update).	2026-09-09 21:33:58 +08:00
唐小鸭	27d66c159f	fix(kms): prevent transient health failures from latching status (#7578 ) Keep backend health checks from overwriting the running service lifecycle state so subsequent admin checks and probe-based readiness can recover without a restart. Add a regression test that fails on the original implementation after backend recovery and verifies the service instance and version remain unchanged. Validation: 39 focused resilience, lifecycle, concurrency, and service manager tests passed; one existing live AWS test remained ignored. cargo fmt --all --check and git diff --check passed. Thanks to @stevapple for reporting the issue and providing a detailed diagnosis and reproduction. Fixes #7554	2026-09-09 11:05:26 +00:00
hector	e546ae9c62	fix(nightly): push assets with plain git — the build fleet has no gh CLI (#7572 ) The sm-standard-4 runners used by the nightly build have no gh binary (functional-chain workflows run elsewhere, on the jumpbox). The assets publish step died on 'gh: command not found' at the credential-helper setup, and the preceding clone failure had been masked by 2>/dev/null, misleading the step into the orphan path. Swap clone and remote setup to plain git with the token embedded in the URL; push semantics are unchanged.	2026-09-09 16:39:59 +08:00
JaySon	35b5cfcf8d	docs: fix invalid docker-buildx.sh usage example in README and README_ZH (#7570 ) docs: fix invalid docker-buildx.sh usage in README and README_ZH Replace the docker-buildx.sh --build-arg RELEASE=latest example, which the script never accepted as a CLI flag (--build-arg is only used internally for docker buildx build), with the supported invocations: - bare ./docker-buildx.sh for the default local build - ./docker-buildx.sh -p linux/amd64 for a single-platform local build Also update the surrounding comments to reflect single-platform local builds and add the multi-arch example comment accordingly.	2026-09-09 16:39:25 +08:00
cxymds	d7c2fc7587	fix(ecstore): apply read quorum to bucket validation (#7556 ) * fix(ecstore): apply read quorum to bucket validation * fix(e2e): remove needless borrow in quorum test	2026-09-09 14:50:28 +08:00
Zhengchao An	7db206466b	test(ci): reserve runner capacity for bounded rollback probe (#7560 )	2026-09-09 14:24:57 +08:00
hector	e549252ac6	feat(nightly): make the build branch configurable (NIGHTLY_BRANCH var + dispatch input) (#7557 )	2026-09-09 14:24:44 +08:00
Zhengchao An	17807b05fb	fix(ecstore): preserve internode error context when cloning (#7548 )	2026-09-09 14:24:23 +08:00
cxymds	789f1832a4	fix(rebalance): align activation locks and preserve retryable causes (#7551 )	2026-09-09 03:25:05 +00:00
Zhengchao An	a722fa80d5	fix(ci): require every selected validation lane before merge (#7529 )	2026-09-09 05:16:48 +08:00
Zhengchao An	9ecb500cbf	test(odm): wait for pull counters before status assertions (#7533 )	2026-09-08 17:35:43 +00:00
cui fliter	8462b3492b	fix(targets): reject trailing batch items (#7508 )	2026-09-08 15:32:24 +00:00
Zhengchao An	ac44f8968e	fix(ci): bind performance runs to selected inputs (#7512 )	2026-09-08 23:16:54 +08:00
唐小鸭	46907c05cf	fix(replication): close the GA blocker set from backlog#2366 (#7503 ) * fix(replication): close GA blockers from backlog#2366 Implements the P1 set from the pre-GA replication audit: - Replication rule tag filters now require every And.Tag to match, replacing the s3s OR semantics with a local AND matcher that fails closed on a malformed tag. - A replicated group membership change no longer writes the group status, so a membership update carrying the default Enabled status cannot silently re-enable a disabled group on the peer. - A successful IAM import schedules one collapsed full-IAM snapshot per remote peer instead of leaving the imported entities local-only. - A pending endpoint refresh is redriven by the heavyweight reconcile tick, carries its own ilm-expiry override, and no longer blocks a remove that drops every unacknowledged peer. - Site metrics expose local replication failure totals and rolling windows; node-level counters no longer report a constructed zero. - set/remove-remote-target notify peer metadata caches before returning, so a follow-up put-bucket-replication on another node sees the target. - Adds the site-replication operations runbook, a docs index, a replication support boundary section, and the Replication changelog section. * fix(site-replication): resume only a locally driven endpoint refresh The peer-side edit handler journals a pending endpoint refresh with an empty `remote_peers` map and commits it inside the same request through `apply_internal_peer_edit`. The reconcile tick could not tell that journal from the coordinator's own: with no required peers it reads as complete on sight, so the tick committed it with `edit_state` - losing the local-name sync - and cleared it under the request that owned it, whose commit then reported the refresh as changed and denied the coordinator the peer acknowledgement it was waiting for. Resume now runs only for a journal that carries the fan-out topology. A receiver's journal stays for the coordinator to redrive with the same refresh id, which is the path that already recovers it. * fix(site-replication): keep an explicit disabled group status on a snapshot Skipping the group-status write whenever an item carries members stopped a membership change from re-enabling a disabled group, but it also silenced the full-IAM snapshot, which always sends members together with the sender's real status. A peer that did not have the group yet created it through `GroupInfo::new` - enabled - so a bootstrap, a repair, or the snapshot an IAM import now schedules handed every member of a frozen group live access there. The madmin wire maps an unset `groupStatus` to Enabled, so only Enabled can be a default. Disabled is always explicit and is applied again. * fix(site-replication): schedule the import snapshot without recording a failure `import-iam` reused the failure-recording path to queue its full-IAM snapshot. That raises `retry_count` on every call, so three imports - the normal shape of a bulk migration done one archive at a time - escalated a healthy peer to `retryStats.failed` with the scheduling note shown as `lastError`, which is exactly the signal the runbook tells operators to repair. A full retry queue also turned a completed import into a 503. Scheduling now only ensures the collapsed entry exists, and a failure to schedule is logged instead of failing the request: the entities are already imported and the reconcile pass still closes the gap. * fix(admin): stop reporting replication failures as retries `retries` is the minio-go counter for redeliveries, and mc prints it as such. Filling it with the failure count claimed a redelivery that never happens: a failed object is not retried by an event today, it waits for the scanner heal pass. `errors` keeps the failure counters; `retries` stays zero until there is a real redelivery to count, and the runbook now says so. * perf(site-replication): aggregate failure windows without cloning bucket stats `site_metrics_snapshot` went through `get_all`, which clones every bucket's stats, and then scanned each target's sample deque twice. That deque is bounded only by the one-hour window, so an unreachable target under load - the case an operator polls this endpoint for - made every `mc admin replicate status` copy the whole backlog and hold the read lock against the failure path while doing it. It now folds under the read lock and takes both windows in one walk. The `max` against the serialized `last_minute` / `last_hour` snapshots is dropped: those are stamped onto per-bucket clones elsewhere and are always zero in this node-local cache. * fix(site-replication): reject a conflicting ilm-expiry override on a re-run The commit now reads the ilm-expiry override back out of the pending refresh journal, so a second edit that asks for a different value had it dropped while the request still reported success. Re-running without the flag keeps pinning the recorded value - that is the documented way to redrive a stuck refresh - but an explicit different value is now rejected instead of ignored. * fix(admin): do not fail a remote-target write on a peer reload error set/remove-remote-target propagated the peer metadata reload error, so a target that was already persisted and live on this node reported a 5xx to the client whenever one peer could not be reached. Every S3 bucket-config write path treats that reload as best effort and only warns; these two admin handlers now do the same, and the reason is logged with the bucket and action. * fix(site-replication): undo every bucket a cut-short refresh rewrote When a remove accepted on another node clears the refresh journal mid-pass, only the bucket holding the lock at that moment had its restored target undone. The buckets rewritten earlier in the same pass kept a target pointing at the removed peer whenever the remove's own cleanup had already walked past them. The undo now covers every bucket this pass rewrote, attempting all of them so one failure does not strand the rest. * fix(site-replication): keep replay running while an endpoint refresh is pending A pending endpoint refresh took the whole heavyweight pass with it, so a peer that never came back froze IAM and bucket replay to every healthy peer too - the stall this journal's resume path was meant to end. The refresh arm now drains the retry queue before returning; it replays per-peer deliveries against the endpoints currently committed in state, so it is unaffected by the edit in flight. Bucket wiring reconciliation still waits, because it rewrites the very targets the refresh is changing, and the runbook now says so. * test(e2e): cover the AND semantics of a two-tag replication filter The acceptance matrix only had a single-tag rule, which matches under both AND and OR semantics and therefore proved nothing about the filter this fix changed. It now also carries a two-tag `And` rule - the shape `mc replicate add --tags "k1=v1&k2=v2"` writes - and asserts that an object with one of the two tags is not admitted while an object with both is. No new test function, so the nightly selection digest is unchanged. * refactor(site-replication): fold the refresh state-change error into one constructor The endpoint-refresh work added three `s3_error!` invocation lines, which the s3s footprint ratchet is meant to prevent. Five copies of the same concurrent-change error now share one constructor, so the surface nets one line smaller than main; the baseline is retightened to match. * fix(site-replication): report a peer whose IAM snapshot waits for a repair An escalated snapshot entry records a deletion a snapshot cannot replay, so only a repair settles it and the marker must survive. Scheduling an import snapshot therefore leaves that peer's entry alone - and now says so, instead of returning success while nothing was scheduled for it. * docs(operations): state the group-status and escalation convergence limits Two boundaries the fixes in this branch make load-bearing: a membership change never carries an enable, so a group disabled on one site only has to be re-enabled there explicitly; and a peer holding an escalated IAM entry does not receive a scheduled snapshot, including the one a bulk import schedules, until a repair settles it.	2026-09-08 14:58:41 +00:00