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).

.config/e2e-distributed-selection.txt

@@ -1,2 +1,2 @@
-sha256-linux=4696a43b167ac608b3b8677027c9fe9fdac3396d37c8cca11dce531c720ac6d2
+sha256-linux=563bff8f1171d6dbe166ff8440310dbe98430e466aa3ecd8dc39e3c872b320f7
-sha256-darwin=9785867929047dfd8c6f768e0d2b1e0a8fdba85216f4a4139093b1619d03ff07
+sha256-darwin=563bff8f1171d6dbe166ff8440310dbe98430e466aa3ecd8dc39e3c872b320f7

.config/e2e-full-selection.txt

@@ -1,2 +1,2 @@
-sha256-darwin=f0d15f2d1183be319d9d977d20b48bb05a29a424c266eed910c67aaa6f1ff955
+sha256-darwin=874c881d7b45f12378a5817c7f42c95c4981960a2ec9ce12dcf4af239ae1f9d5
-sha256-linux=340aa702576ebed5266b7c47e591f267a11178fa04bf3828d64f53fe92eb0907
+sha256-linux=9515861be899ceb10e2e0ef93c34208bb7a7a8a7f8067a02db4cfba23270ebd6

.config/e2e-nightly-selection.txt

@@ -1,2 +1,2 @@
-sha256-darwin=12d30fff5ed48fe95bbfb310dd507048f81782b2954e00eb72435708b1133f9c
+sha256-darwin=83a7dcaffd5a789517ae9f02a224f66a9713937885cff96fca2ad7e216f197ae
-sha256-linux=e917b2fdb303d01e6008ac4f1836c7b268698afe8383bd2aa4e697061df85cf6
+sha256-linux=626c10f8c964507ff987b6c86069e9019dc6d2ae7fb02db9be5df5aa8cc5145b

.config/make/lint-fmt.mak

@@ -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

.config/make/tests.mak

@@ -32,6 +32,13 @@ script-tests: ## Run shell script tests
 	./scripts/test_hotpath_warp_ab_gate.sh
 	./scripts/test_hotpath_warp_abba.sh
 	./scripts/test_scanner_validation_harness.sh
+	./scripts/test_scanner_heal_checkpoint_crash_evidence.sh
+	./scripts/test_scanner_heal_g14_multiset_evidence.sh
+	./scripts/test_scanner_heal_scheduler_pressure_evidence.sh
+	./scripts/test_scanner_heal_status_outcome_evidence.sh
+	./scripts/test_scanner_heal_maintenance_evidence.sh
+	./scripts/test_scanner_heal_w13_mrf_evidence.sh
+	./scripts/test_scanner_heal_w16_recovery_evidence.sh
 	./scripts/test_exact_1mib_handoff_abba.sh
 	./scripts/test_pinned_paired_abba_bench.sh
 	./scripts/test_manual_transition_runbooks.sh
@@ -39,6 +46,7 @@ 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

.config/nextest.toml

@@ -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`)
 # ---------------------------------------------------------------------------

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

@@ -78,6 +78,43 @@
       "erasure": {"data_blocks": 8, "parity_blocks": 4},
       "erasure_set_drive_count": 12,
       "scope": "Target process killed during partial background rebuild on a single 3x4 EC8+4 set; real unclean-shutdown marker, exact unversioned S3 bodies and replacement-drive shards; not power loss, multi-set, or multi-pool."
+    },
+    "background-target-restart-ec8-4-multi-set": {
+      "gate": "G14",
+      "task": "W21",
+      "lane": "e2e-nightly",
+      "suite": "e2e_test",
+      "name": "heal_erasure_disk_rebuild_test::tests::test_cluster_root_heal_recovers_ec84_shards_across_multi_set_after_background_target_restart",
+      "oracle": "background-target-restart-ec8-4-multi-set.json",
+      "evidence": "process-restart",
+      "unclean_shutdown_marker": false,
+      "min_objects": 9,
+      "max_objects": 65,
+      "topology": {"nodes": 3, "drives_per_node": 8},
+      "erasure": {"data_blocks": 8, "parity_blocks": 4},
+      "erasure_set_drive_count": 12,
+      "sets": 2,
+      "pools": 1,
+      "scope": "Target process restart during partial background rebuild on a 3x8 EC8+4 layout with two erasure sets in one pool; exact unversioned S3 bodies and replacement-drive shards; not power loss or multi-pool."
+    },
+    "background-target-crash-ec8-4-multi-pool": {
+      "gate": "G14",
+      "task": "W21",
+      "lane": "e2e-nightly",
+      "suite": "e2e_test",
+      "name": "heal_erasure_disk_rebuild_test::tests::test_cluster_root_heal_recovers_ec84_shards_across_multi_pool_after_background_target_crash",
+      "oracle": "background-target-crash-ec8-4-multi-pool.json",
+      "evidence": "process-crash-restart",
+      "unclean_shutdown_marker": true,
+      "min_objects": 9,
+      "max_objects": 65,
+      "topology": {"nodes": 3, "drives_per_node": 12},
+      "erasure": {"data_blocks": 8, "parity_blocks": 4},
+      "erasure_set_drive_count": 12,
+      "sets": 3,
+      "pools": 3,
+      "outage_target_manifest_required": false,
+      "scope": "Target process crash during partial background rebuild on three single-node EC8+4 pools; exact baseline S3 bodies and replacement-drive shards, with outage object verified through S3 but not forced onto the replaced target drive."
     }
   },
   "release_lanes": {
@@ -172,7 +209,11 @@
       "lane": "status-and-outcome",
       "status": "pending",
       "description": "Per-object outcomes and bounded terminal retention",
-      "requires": ["per-object outcome oracle", "terminal retention bounds"]
+      "requires": ["per-object outcome oracle", "terminal retention bounds"],
+      "evidence_fields": [
+        "per_object_outcome_oracle",
+        "terminal_retention_bounds"
+      ]
     },
     {
       "gate": "G06",
@@ -180,7 +221,12 @@
       "lane": "status-and-outcome",
       "status": "pending",
       "description": "Concurrent status, legacy clients and truncation",
-      "requires": ["concurrent status evidence", "legacy client compatibility", "truncation behavior"]
+      "requires": ["concurrent status evidence", "legacy client compatibility", "truncation behavior"],
+      "evidence_fields": [
+        "concurrent_status_evidence",
+        "legacy_client_compatibility",
+        "truncation_behavior"
+      ]
     },
     {
       "gate": "G07",
@@ -188,7 +234,11 @@
       "lane": "mrf-responsibility",
       "status": "pending",
       "description": "Durable MRF responsibility at every commit boundary",
-      "requires": ["MRF responsibility oracle", "commit-boundary crash matrix"]
+      "requires": ["MRF responsibility oracle", "commit-boundary crash matrix"],
+      "evidence_fields": [
+        "mrf_responsibility_oracle",
+        "commit_boundary_crash_matrix"
+      ]
     },
     {
       "gate": "G08",
@@ -196,7 +246,12 @@
       "lane": "mrf-responsibility",
       "status": "pending",
       "description": "MRF capacity, disk-full and replica-loss matrix",
-      "requires": ["MRF capacity evidence", "disk-full matrix", "replica-loss matrix"]
+      "requires": ["MRF capacity evidence", "disk-full matrix", "replica-loss matrix"],
+      "evidence_fields": [
+        "mrf_capacity_evidence",
+        "disk_full_matrix",
+        "replica_loss_matrix"
+      ]
     },
     {
       "gate": "G09",
@@ -225,7 +280,16 @@
       "lane": "maintenance-producers",
       "status": "pending",
       "description": "Maintenance and complete producer coverage",
-      "requires": ["maintenance producer matrix", "complete producer inventory"]
+      "requires": [
+        "maintenance producer matrix",
+        "complete producer inventory",
+        "segment activation preflight"
+      ],
+      "evidence_fields": [
+        "maintenance_producer_matrix",
+        "complete_producer_inventory",
+        "segment_activation_preflight"
+      ]
     },
     {
       "gate": "G12",
@@ -249,7 +313,20 @@
       "lane": "ec8-4-multiset",
       "status": "pending",
       "description": "Same-window field evidence with 3x4 EC8+4 and multi-set/pool coverage",
-      "requires": ["same-window field evidence", "3x4 EC8+4 evidence", "multi-set evidence", "multi-pool evidence"]
+      "requires": [
+        "same-window field evidence",
+        "3x4 EC8+4 evidence",
+        "multi-set evidence",
+        "multi-pool evidence",
+        "distributed segment invalidation evidence"
+      ],
+      "evidence_fields": [
+        "same_window_field_evidence",
+        "ec8_4_evidence",
+        "multi_set_evidence",
+        "multi_pool_evidence",
+        "distributed_segment_invalidation_evidence"
+      ]
     },
     {
       "gate": "P1",
@@ -265,7 +342,11 @@
       "lane": "scheduler-pressure",
       "status": "pending",
       "description": "Measured post-stop convergence and cold segment reuse",
-      "requires": ["post-stop convergence measurement", "cold segment reuse measurement"]
+      "requires": ["post-stop convergence measurement", "cold segment reuse measurement"],
+      "evidence_fields": [
+        "post_stop_convergence_measurement",
+        "cold_segment_reuse_measurement"
+      ]
     },
     {
       "gate": "P3",
@@ -281,7 +362,13 @@
       "lane": "mrf-responsibility",
       "status": "pending",
       "description": "Measured MRF scale and replay cost with retained responsibility",
-      "requires": ["MRF scale measurement", "MRF replay-cost measurement", "retained responsibility evidence"]
+      "requires": ["MRF scale measurement", "MRF replay-cost measurement", "retained responsibility evidence", "cleanup/GC soak evidence"],
+      "evidence_fields": [
+        "mrf_scale_measurement",
+        "mrf_replay_cost_measurement",
+        "retained_responsibility_evidence",
+        "mrf_cleanup_gc_soak_evidence"
+      ]
     },
     {
       "gate": "R-E",
@@ -297,7 +384,13 @@
       "lane": "status-and-outcome",
       "status": "pending",
       "description": "Manager-to-event-to-ledger exact disposition, including grace",
-      "requires": ["manager disposition evidence", "event disposition evidence", "ledger disposition evidence", "grace handling"]
+      "requires": ["manager disposition evidence", "event disposition evidence", "ledger disposition evidence", "grace handling"],
+      "evidence_fields": [
+        "manager_disposition_evidence",
+        "event_disposition_evidence",
+        "ledger_disposition_evidence",
+        "grace_handling"
+      ]
     },
     {
       "gate": "R-L",

.docker/compose/docker-compose.cluster.local-build.profiling-amd64.yml

@@ -43,6 +43,7 @@ services:
       - RUSTFS_ACCESS_KEY=${RUSTFS_ACCESS_KEY:-rustfs-cluster-admin}
       - RUSTFS_SECRET_KEY=${RUSTFS_SECRET_KEY:-rustfs-cluster-secret}
       - RUSTFS_OBS_ENDPOINT=${RUSTFS_OBS_ENDPOINT:-http://host.docker.internal:4318}
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       # `info` is enough for startup logs/metrics. Use `debug` if Tempo/Jaeger
       # should show richer nested spans during request-path verification.
       - RUSTFS_OBS_LOGGER_LEVEL=${RUSTFS_OBS_LOGGER_LEVEL:-info}
@@ -89,6 +90,7 @@ services:
       - RUSTFS_ACCESS_KEY=${RUSTFS_ACCESS_KEY:-rustfs-cluster-admin}
       - RUSTFS_SECRET_KEY=${RUSTFS_SECRET_KEY:-rustfs-cluster-secret}
       - RUSTFS_OBS_ENDPOINT=${RUSTFS_OBS_ENDPOINT:-http://host.docker.internal:4318}
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       # `info` is enough for startup logs/metrics. Use `debug` if Tempo/Jaeger
       # should show richer nested spans during request-path verification.
       - RUSTFS_OBS_LOGGER_LEVEL=${RUSTFS_OBS_LOGGER_LEVEL:-info}
@@ -135,6 +137,7 @@ services:
       - RUSTFS_ACCESS_KEY=${RUSTFS_ACCESS_KEY:-rustfs-cluster-admin}
       - RUSTFS_SECRET_KEY=${RUSTFS_SECRET_KEY:-rustfs-cluster-secret}
       - RUSTFS_OBS_ENDPOINT=${RUSTFS_OBS_ENDPOINT:-http://host.docker.internal:4318}
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       # `info` is enough for startup logs/metrics. Use `debug` if Tempo/Jaeger
       # should show richer nested spans during request-path verification.
       - RUSTFS_OBS_LOGGER_LEVEL=${RUSTFS_OBS_LOGGER_LEVEL:-info}
@@ -181,6 +184,7 @@ services:
       - RUSTFS_ACCESS_KEY=${RUSTFS_ACCESS_KEY:-rustfs-cluster-admin}
       - RUSTFS_SECRET_KEY=${RUSTFS_SECRET_KEY:-rustfs-cluster-secret}
       - RUSTFS_OBS_ENDPOINT=${RUSTFS_OBS_ENDPOINT:-http://host.docker.internal:4318}
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       # `info` is enough for startup logs/metrics. Use `debug` if Tempo/Jaeger
       # should show richer nested spans during request-path verification.
       - RUSTFS_OBS_LOGGER_LEVEL=${RUSTFS_OBS_LOGGER_LEVEL:-info}

.docker/compose/docker-compose.cluster.local-build.yml

@@ -26,6 +26,7 @@ services:
       - RUSTFS_ACCESS_KEY=${RUSTFS_ACCESS_KEY:-rustfsadmin-local}
       - RUSTFS_SECRET_KEY=${RUSTFS_SECRET_KEY:-rustfssecret-local}
       - RUSTFS_OBS_ENDPOINT=${RUSTFS_OBS_ENDPOINT:-http://host.docker.internal:4318}
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       - RUSTFS_OBS_LOGGER_LEVEL=${RUSTFS_OBS_LOGGER_LEVEL:-info}
       - RUSTFS_OBS_USE_STDOUT=${RUSTFS_OBS_USE_STDOUT:-false}
       - RUSTFS_OBS_LOG_STDOUT_ENABLED=${RUSTFS_OBS_LOG_STDOUT_ENABLED:-false}
@@ -72,6 +73,7 @@ services:
       - RUSTFS_ACCESS_KEY=${RUSTFS_ACCESS_KEY:-rustfsadmin-local}
       - RUSTFS_SECRET_KEY=${RUSTFS_SECRET_KEY:-rustfssecret-local}
       - RUSTFS_OBS_ENDPOINT=${RUSTFS_OBS_ENDPOINT:-http://host.docker.internal:4318}
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       - RUSTFS_OBS_LOGGER_LEVEL=${RUSTFS_OBS_LOGGER_LEVEL:-info}
       - RUSTFS_OBS_USE_STDOUT=${RUSTFS_OBS_USE_STDOUT:-false}
       - RUSTFS_OBS_LOG_STDOUT_ENABLED=${RUSTFS_OBS_LOG_STDOUT_ENABLED:-false}
@@ -118,6 +120,7 @@ services:
       - RUSTFS_ACCESS_KEY=${RUSTFS_ACCESS_KEY:-rustfsadmin-local}
       - RUSTFS_SECRET_KEY=${RUSTFS_SECRET_KEY:-rustfssecret-local}
       - RUSTFS_OBS_ENDPOINT=${RUSTFS_OBS_ENDPOINT:-http://host.docker.internal:4318}
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       - RUSTFS_OBS_LOGGER_LEVEL=${RUSTFS_OBS_LOGGER_LEVEL:-info}
       - RUSTFS_OBS_USE_STDOUT=${RUSTFS_OBS_USE_STDOUT:-false}
       - RUSTFS_OBS_LOG_STDOUT_ENABLED=${RUSTFS_OBS_LOG_STDOUT_ENABLED:-false}
@@ -164,6 +167,7 @@ services:
       - RUSTFS_ACCESS_KEY=${RUSTFS_ACCESS_KEY:-rustfsadmin-local}
       - RUSTFS_SECRET_KEY=${RUSTFS_SECRET_KEY:-rustfssecret-local}
       - RUSTFS_OBS_ENDPOINT=${RUSTFS_OBS_ENDPOINT:-http://host.docker.internal:4318}
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       - RUSTFS_OBS_LOGGER_LEVEL=${RUSTFS_OBS_LOGGER_LEVEL:-info}
       - RUSTFS_OBS_USE_STDOUT=${RUSTFS_OBS_USE_STDOUT:-false}
       - RUSTFS_OBS_LOG_STDOUT_ENABLED=${RUSTFS_OBS_LOG_STDOUT_ENABLED:-false}

.docker/compose/docker-compose.observability.yaml

@@ -201,6 +201,7 @@ services:
       - RUSTFS_ADDRESS=:9000
       - RUSTFS_CONSOLE_ENABLE=true
       - RUSTFS_OBS_ENDPOINT=http://otel-collector:4318
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       - RUSTFS_OBS_LOGGER_LEVEL=debug
     platform: linux/amd64
     ports:
@@ -220,6 +221,7 @@ services:
       - RUSTFS_ADDRESS=:9000
       - RUSTFS_CONSOLE_ENABLE=true
       - RUSTFS_OBS_ENDPOINT=http://otel-collector:4318
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       - RUSTFS_OBS_LOGGER_LEVEL=debug
     platform: linux/amd64
     ports:
@@ -239,6 +241,7 @@ services:
       - RUSTFS_ADDRESS=:9000
       - RUSTFS_CONSOLE_ENABLE=true
       - RUSTFS_OBS_ENDPOINT=http://otel-collector:4318
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       - RUSTFS_OBS_LOGGER_LEVEL=debug
     platform: linux/amd64
     ports:
@@ -258,6 +261,7 @@ services:
       - RUSTFS_ADDRESS=:9000
       - RUSTFS_CONSOLE_ENABLE=true
       - RUSTFS_OBS_ENDPOINT=http://otel-collector:4318
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       - RUSTFS_OBS_LOGGER_LEVEL=debug
     platform: linux/amd64
     ports:

.docker/observability/README.md

@@ -47,6 +47,10 @@ Three pre-built Grafana dashboards are included for monitoring RustFS GET perfor
 | **GET Resource Impact** | `grafana-get-resource-impact.json` | Monitors resource usage: concurrent requests, IO queue utilization, disk permit wait, RSS trend |
 | **Object Data Cache** | `grafana-object-data-cache.json` | Monitors the GET body cache (`rustfs_object_data_cache_*`): hit ratio, lookup/plan/fill outcomes, fill duration quantiles, hit vs fill throughput, entries/weighted bytes, inflight fills, memory-pressure skips, invalidations, and size-class breakdowns |
 
+### Storage Metrics
+
+Storage panels require `prometheus-rules/rustfs-storage.yml` and the cluster resource attribute. See the [storage metrics guide](../../docs/operations/storage-metrics.md) for ownership, observer selection, freshness, and rolling upgrades.
+
 ### Prometheus Alert Rules
 
 The file `prometheus-rules/rustfs-get-optimization-alerts.yaml` contains pre-configured alerting rules:

.docker/observability/docker-compose-example-for-rustfs.yml

@@ -32,6 +32,7 @@ services:
       - RUSTFS_SECRET_KEY=rustfsadmin
       - RUSTFS_OBS_LOGGER_LEVEL=info
       - RUSTFS_OBS_ENDPOINT=http://otel-collector:4318
+      - OTEL_RESOURCE_ATTRIBUTES=${OTEL_RESOURCE_ATTRIBUTES:-rustfs.cluster.id=rustfs-dev}
       - RUSTFS_OBS_PROFILING_ENDPOINT=http://pyroscope:4040
     volumes:
       - rustfs-data:/data/rustfs

.docker/observability/grafana/dashboards/rustfs.json

@@ -223,14 +223,15 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "sum(rustfs_cluster_buckets_total{job=~\"$job\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_buckets_total\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "legendFormat": "__auto",
           "range": true,
           "refId": "A"
         }
       ],
       "title": "Total Buckets",
-      "type": "stat"
+      "type": "stat",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -289,14 +290,15 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "sum(rustfs_cluster_objects_total{job=~\"$job\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_objects_total\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "legendFormat": "__auto",
           "range": true,
           "refId": "A"
         }
       ],
       "title": "Total Objects",
-      "type": "stat"
+      "type": "stat",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -427,7 +429,7 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "sum(rustfs_cluster_capacity_used_bytes{job=~\"$job\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_capacity_used_bytes\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "legendFormat": "Used",
           "range": true,
           "refId": "A"
@@ -438,7 +440,7 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "sum(rustfs_cluster_capacity_raw_total_bytes{job=~\"$job\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_capacity_raw_total_bytes\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "hide": false,
           "legendFormat": "Total",
           "range": true,
@@ -450,7 +452,7 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "sum(rustfs_cluster_capacity_used_bytes{job=~\"$job\"}) / sum(rustfs_cluster_capacity_raw_total_bytes{job=~\"$job\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_capacity_used_bytes\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"} / ignoring(source_metric) rustfs:storage:current{source_metric=\"rustfs_cluster_capacity_raw_total_bytes\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "hide": false,
           "instant": false,
           "legendFormat": "Percent",
@@ -459,7 +461,8 @@
         }
       ],
       "title": "Capacity",
-      "type": "stat"
+      "type": "stat",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -525,7 +528,7 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "sum(rustfs_cluster_capacity_stale_drives{job=~\"$job\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_capacity_stale_drives\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "legendFormat": "Stale Drives",
           "range": true,
           "refId": "A"
@@ -536,14 +539,15 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "sum(rustfs_cluster_capacity_missing_drives{job=~\"$job\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_capacity_missing_drives\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "legendFormat": "Missing Drives",
           "range": true,
           "refId": "B"
         }
       ],
       "title": "Capacity Observation",
-      "type": "stat"
+      "type": "stat",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -1989,8 +1993,8 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "sum by (drive) (rustfs_system_drive_used_bytes{job=~\"$job\", drive=~\"$drive\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_system_drive_used_bytes\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",server=~\"$server\",drive=~\"$drive\"}",
-          "legendFormat": "{{drive}} (bytes)",
+          "legendFormat": "{{server}} | {{drive}} (bytes)",
           "range": true,
           "refId": "A"
         },
@@ -2000,16 +2004,17 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "sum by (drive) (rustfs_system_drive_used_bytes{job=~\"$job\", drive=~\"$drive\"}) / sum by (drive)(rustfs_system_drive_total_bytes{job=~\"$job\", drive=~\"$drive\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_system_drive_used_bytes\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",server=~\"$server\",drive=~\"$drive\"} / ignoring(source_metric) rustfs:storage:current{source_metric=\"rustfs_system_drive_total_bytes\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",server=~\"$server\",drive=~\"$drive\"}",
           "hide": false,
           "instant": false,
-          "legendFormat": "{{drive}} (percent)",
+          "legendFormat": "{{server}} | {{drive}} (percent)",
           "range": true,
           "refId": "B"
         }
       ],
       "title": "System Drive Usage",
-      "type": "timeseries"
+      "type": "timeseries",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -2097,14 +2102,15 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "max by (drive) (rustfs_system_drive_capacity_observation_age_seconds{job=~\"$job\", drive=~\"$drive\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_system_drive_capacity_observation_age_seconds\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",server=~\"$server\",drive=~\"$drive\"}",
-          "legendFormat": "{{drive}}",
+          "legendFormat": "{{server}} | {{drive}}",
           "range": true,
           "refId": "A"
         }
       ],
       "title": "Drive Capacity Observation Age",
-      "type": "timeseries"
+      "type": "timeseries",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -2190,8 +2196,8 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "max by (drive) (rustfs_system_drive_capacity_observation_state{job=~\"$job\", drive=~\"$drive\", state=\"stale\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_system_drive_capacity_observation_state\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",server=~\"$server\",drive=~\"$drive\",state=\"stale\"}",
-          "legendFormat": "{{drive}} stale",
+          "legendFormat": "{{server}} | {{drive}} stale",
           "range": true,
           "refId": "A"
         },
@@ -2201,14 +2207,15 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "max by (drive) (rustfs_system_drive_capacity_observation_state{job=~\"$job\", drive=~\"$drive\", state=\"missing\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_system_drive_capacity_observation_state\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",server=~\"$server\",drive=~\"$drive\",state=\"missing\"}",
-          "legendFormat": "{{drive}} missing",
+          "legendFormat": "{{server}} | {{drive}} missing",
           "range": true,
           "refId": "B"
         }
       ],
       "title": "Drive Capacity Observation State",
-      "type": "timeseries"
+      "type": "timeseries",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -4551,7 +4558,7 @@
                   "index": 0,
                   "text": "INACTIVE"
                 },
-                "to": 1e-9
+                "to": 1e-09
               },
               "type": "range"
             }
@@ -6563,7 +6570,7 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "rustfs_cluster_health_drives_online_count{job=~\"$job\"}",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_health_drives_online_count\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "legendFormat": "online - {{job}}",
           "range": true,
           "refId": "A"
@@ -6574,7 +6581,7 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "rustfs_cluster_health_drives_offline_count{job=~\"$job\"}",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_health_drives_offline_count\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "legendFormat": "offline - {{job}}",
           "range": true,
           "refId": "B"
@@ -6585,14 +6592,15 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "rustfs_cluster_health_drives_count{job=~\"$job\"}",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_health_drives_count\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
           "legendFormat": "total - {{job}}",
           "range": true,
           "refId": "C"
         }
       ],
       "title": "Cluster Drive Health Counts",
-      "type": "timeseries"
+      "type": "timeseries",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "collapsed": false,
@@ -8554,14 +8562,15 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "{__name__=~\"rustfs_system_drive_.*\",job=~\"$job\",drive=~\"$drive\"}",
+          "expr": "rustfs:storage:current{source_metric=~\"rustfs_system_drive_.*\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",server=~\"$server\",drive=~\"$drive\"}",
-          "legendFormat": "{{__name__}} | {{drive}}",
+          "legendFormat": "{{server}} | {{source_metric}} | {{drive}}",
           "range": true,
           "refId": "A"
         }
       ],
       "title": "System Drive (All)",
-      "type": "timeseries"
+      "type": "timeseries",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -8942,14 +8951,15 @@
             "uid": "${datasource}"
           },
           "editorMode": "code",
-          "expr": "{__name__=~\"rustfs_cluster_erasure_set_.*\",job=~\"$job\"}",
+          "expr": "rustfs:storage:current{source_metric=~\"rustfs_cluster_erasure_set_.*\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\"}",
-          "legendFormat": "{{__name__}}",
+          "legendFormat": "{{source_metric}}",
           "range": true,
           "refId": "A"
         }
       ],
       "title": "Cluster Erasure Set (All)",
-      "type": "timeseries"
+      "type": "timeseries",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -11716,7 +11726,7 @@
           "editorMode": "code",
           "range": true,
           "refId": "A",
-          "expr": "max by (server, drive, pool_index, set_index, drive_index, state) (rustfs_system_drive_runtime_state{job=~\"$job\",server=~\"$server\",drive=~\"$drive\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_drive_runtime_state\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\",drive=~\"$drive\"}",
           "legendFormat": "{{server}} | {{drive}} | p{{pool_index}}/s{{set_index}}/d{{drive_index}} | {{state}}"
         },
         {
@@ -11727,12 +11737,13 @@
           "editorMode": "code",
           "range": true,
           "refId": "B",
-          "expr": "max by (server, drive, pool_index, set_index, drive_index) (rustfs_system_drive_offline_duration_seconds{job=~\"$job\",server=~\"$server\",drive=~\"$drive\"})",
+          "expr": "rustfs:storage:current{source_metric=\"rustfs_cluster_drive_offline_duration_seconds\",collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",observer=\"$storage_observer\",drive=~\"$drive\"}",
           "legendFormat": "{{server}} | {{drive}} | offline seconds"
         }
       ],
-      "title": "Drive Runtime State and Offline Duration",
+      "title": "Observed Cluster Drive State and Offline Duration",
-      "type": "timeseries"
+      "type": "timeseries",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -11825,12 +11836,13 @@
           "editorMode": "code",
           "range": true,
           "refId": "A",
-          "expr": "sum by (server, drive, pool_index, set_index, drive_index, api) (rate(rustfs_system_drive_api_calls_total{job=~\"$job\",server=~\"$server\",drive=~\"$drive\",api=~\"$drive_api\"}[$__rate_interval]))",
+          "expr": "rate(rustfs_system_drive_api_calls_total{collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",server=~\"$server\",drive=~\"$drive\",api=~\"$drive_api\"}[$__rate_interval]) and ignoring(source_metric) rustfs:storage:current{source_metric=\"rustfs_system_drive_api_calls_total\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\",server=~\"$server\",drive=~\"$drive\",api=~\"$drive_api\"}",
           "legendFormat": "{{server}} | {{drive}} | p{{pool_index}}/s{{set_index}}/d{{drive_index}} | {{api}}"
         }
       ],
       "title": "Drive API Calls by Operation",
-      "type": "timeseries"
+      "type": "timeseries",
+      "description": "Storage snapshot rules are required. Local details come from each drive owner. Global values use the selected fresh cluster observer. Missing or expired observations show no data; select another observer if needed."
     },
     {
       "datasource": {
@@ -12376,6 +12388,50 @@
         "sort": 1,
         "type": "query"
       },
+      {
+        "current": {},
+        "datasource": {
+          "type": "prometheus",
+          "uid": "${datasource}"
+        },
+        "definition": "label_values(rustfs:storage_snapshot:fresh, rustfs_cluster_id)",
+        "includeAll": false,
+        "label": "Storage cluster",
+        "multi": false,
+        "name": "storage_cluster",
+        "options": [],
+        "query": {
+          "qryType": 1,
+          "query": "label_values(rustfs:storage_snapshot:fresh, rustfs_cluster_id)",
+          "refId": "PrometheusVariableQueryEditor-storage_cluster"
+        },
+        "refresh": 2,
+        "regex": "",
+        "sort": 1,
+        "type": "query"
+      },
+      {
+        "current": {},
+        "datasource": {
+          "type": "prometheus",
+          "uid": "${datasource}"
+        },
+        "definition": "query_result(rustfs:storage_snapshot:fresh{collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\"})",
+        "includeAll": false,
+        "label": "Cluster observer",
+        "multi": false,
+        "name": "storage_observer",
+        "options": [],
+        "query": {
+          "qryType": 3,
+          "query": "query_result(rustfs:storage_snapshot:fresh{collection_scope=\"cluster\",rustfs_cluster_id=\"$storage_cluster\",job=~\"$job\"})",
+          "refId": "PrometheusVariableQueryEditor-storage_observer"
+        },
+        "refresh": 2,
+        "regex": "/observer=\"([^\"]+)\"/",
+        "sort": 1,
+        "type": "query"
+      },
       {
         "allValue": ".*",
         "current": {
@@ -12386,7 +12442,7 @@
           "type": "prometheus",
           "uid": "${datasource}"
         },
-        "definition": "label_values(rustfs_system_drive_api_calls_total,api)",
+        "definition": "label_values(rustfs:storage:current{source_metric=\"rustfs_system_drive_api_calls_total\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\"}, api)",
         "includeAll": true,
         "label": "Drive API",
         "multi": true,
@@ -12394,7 +12450,7 @@
         "options": [],
         "query": {
           "qryType": 1,
-          "query": "label_values(rustfs_system_drive_api_calls_total,api)",
+          "query": "label_values(rustfs:storage:current{source_metric=\"rustfs_system_drive_api_calls_total\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\"}, api)",
           "refId": "PrometheusVariableQueryEditor-drive_api"
         },
         "refresh": 2,
@@ -12455,7 +12511,7 @@
           "text": "All",
           "value": "$__all"
         },
-        "definition": "label_values(rustfs_system_drive_used_bytes,drive)",
+        "definition": "label_values(rustfs:storage:current{source_metric=\"rustfs_system_drive_used_bytes\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\"}, drive)",
         "includeAll": true,
         "label": "Drive",
         "multi": true,
@@ -12463,7 +12519,7 @@
         "options": [],
         "query": {
           "qryType": 1,
-          "query": "label_values(rustfs_system_drive_used_bytes,drive)",
+          "query": "label_values(rustfs:storage:current{source_metric=\"rustfs_system_drive_used_bytes\",collection_scope=\"local\",rustfs_cluster_id=\"$storage_cluster\"}, drive)",
           "refId": "PrometheusVariableQueryEditor-VariableQuery"
         },
         "refresh": 2,

.docker/observability/prometheus-rules/rustfs-storage.yml

@@ -0,0 +1,575 @@
+# Preserve OTLP timestamps in the Collector. Apply timestamp() directly to
+# each raw selector, before label rewriting; functions such as label_replace()
+# replace the evaluation timestamp and would make a cached value look new.
+# Compare publication times at Prometheus millisecond precision so points
+# published and exported within the same millisecond are not withheld.
+groups:
+  - name: rustfs-storage-snapshots
+    interval: 15s
+    rules:
+      - record: rustfs:storage_snapshot:fresh
+        expr: |
+          rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id!="",collection_scope=~"local|cluster"}
+          and ((time() - rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id!="",collection_scope=~"local|cluster"}) <= rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id!="",collection_scope=~"local|cluster"})
+          and ((time() - rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id!="",collection_scope=~"local|cluster"}) >= 0)
+          and (rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id!="",collection_scope=~"local|cluster"} > 0)
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_buckets_total
+        expr: |
+          rustfs_cluster_buckets_total{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_buckets_total{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_capacity_free_bytes
+        expr: |
+          rustfs_cluster_capacity_free_bytes{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_capacity_free_bytes{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_capacity_missing_drives
+        expr: |
+          rustfs_cluster_capacity_missing_drives{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_capacity_missing_drives{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_capacity_raw_total_bytes
+        expr: |
+          rustfs_cluster_capacity_raw_total_bytes{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_capacity_raw_total_bytes{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_capacity_stale_drives
+        expr: |
+          rustfs_cluster_capacity_stale_drives{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_capacity_stale_drives{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_capacity_usable_total_bytes
+        expr: |
+          rustfs_cluster_capacity_usable_total_bytes{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_capacity_usable_total_bytes{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_capacity_used_bytes
+        expr: |
+          rustfs_cluster_capacity_used_bytes{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_capacity_used_bytes{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_drive_capacity_observation_age_seconds
+        expr: |
+          rustfs_cluster_drive_capacity_observation_age_seconds{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_drive_capacity_observation_age_seconds{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_drive_capacity_observation_state
+        expr: |
+          rustfs_cluster_drive_capacity_observation_state{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_drive_capacity_observation_state{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_drive_free_bytes
+        expr: |
+          rustfs_cluster_drive_free_bytes{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_drive_free_bytes{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_drive_offline_duration_seconds
+        expr: |
+          rustfs_cluster_drive_offline_duration_seconds{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_drive_offline_duration_seconds{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_drive_present
+        expr: |
+          rustfs_cluster_drive_present{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_drive_present{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_drive_runtime_state
+        expr: |
+          rustfs_cluster_drive_runtime_state{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_drive_runtime_state{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_drive_total_bytes
+        expr: |
+          rustfs_cluster_drive_total_bytes{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_drive_total_bytes{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_drive_used_bytes
+        expr: |
+          rustfs_cluster_drive_used_bytes{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_drive_used_bytes{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_data_shards
+        expr: |
+          rustfs_cluster_erasure_set_data_shards{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_data_shards{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_healing_drives_count
+        expr: |
+          rustfs_cluster_erasure_set_healing_drives_count{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_healing_drives_count{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_health
+        expr: |
+          rustfs_cluster_erasure_set_health{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_health{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_online_drives_count
+        expr: |
+          rustfs_cluster_erasure_set_online_drives_count{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_online_drives_count{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_overall_health
+        expr: |
+          rustfs_cluster_erasure_set_overall_health{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_overall_health{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_overall_write_quorum
+        expr: |
+          rustfs_cluster_erasure_set_overall_write_quorum{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_overall_write_quorum{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_parity
+        expr: |
+          rustfs_cluster_erasure_set_parity{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_parity{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_read_health
+        expr: |
+          rustfs_cluster_erasure_set_read_health{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_read_health{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_read_quorum
+        expr: |
+          rustfs_cluster_erasure_set_read_quorum{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_read_quorum{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_read_tolerance
+        expr: |
+          rustfs_cluster_erasure_set_read_tolerance{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_read_tolerance{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_size
+        expr: |
+          rustfs_cluster_erasure_set_size{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_size{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_write_health
+        expr: |
+          rustfs_cluster_erasure_set_write_health{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_write_health{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_write_quorum
+        expr: |
+          rustfs_cluster_erasure_set_write_quorum{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_write_quorum{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_erasure_set_write_tolerance
+        expr: |
+          rustfs_cluster_erasure_set_write_tolerance{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_erasure_set_write_tolerance{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_health_drives_count
+        expr: |
+          rustfs_cluster_health_drives_count{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_health_drives_count{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_health_drives_offline_count
+        expr: |
+          rustfs_cluster_health_drives_offline_count{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_health_drives_offline_count{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_health_drives_online_count
+        expr: |
+          rustfs_cluster_health_drives_online_count{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_health_drives_online_count{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_cluster_objects_total
+        expr: |
+          rustfs_cluster_objects_total{rustfs_cluster_id!="",collection_scope="cluster"}
+          and (timestamp(rustfs_cluster_objects_total{rustfs_cluster_id!="",collection_scope="cluster"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_node_disk_free_bytes
+        expr: |
+          rustfs_node_disk_free_bytes{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_node_disk_free_bytes{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_node_disk_total_bytes
+        expr: |
+          rustfs_node_disk_total_bytes{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_node_disk_total_bytes{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_node_disk_used_bytes
+        expr: |
+          rustfs_node_disk_used_bytes{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_node_disk_used_bytes{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_api_calls_total
+        expr: |
+          rustfs_system_drive_api_calls_total{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_api_calls_total{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_api_latency_by_api_micros
+        expr: |
+          rustfs_system_drive_api_latency_by_api_micros{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_api_latency_by_api_micros{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_api_latency_micros
+        expr: |
+          rustfs_system_drive_api_latency_micros{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_api_latency_micros{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_availability_errors_total
+        expr: |
+          rustfs_system_drive_availability_errors_total{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_availability_errors_total{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_capacity_observation_age_seconds
+        expr: |
+          rustfs_system_drive_capacity_observation_age_seconds{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_capacity_observation_age_seconds{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_capacity_observation_state
+        expr: |
+          rustfs_system_drive_capacity_observation_state{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_capacity_observation_state{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_count
+        expr: |
+          rustfs_system_drive_count{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_count{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_deletes_total
+        expr: |
+          rustfs_system_drive_deletes_total{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_deletes_total{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_free_bytes
+        expr: |
+          rustfs_system_drive_free_bytes{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_free_bytes{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_free_inodes
+        expr: |
+          rustfs_system_drive_free_inodes{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_free_inodes{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_healing
+        expr: |
+          rustfs_system_drive_healing{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_healing{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_health
+        expr: |
+          rustfs_system_drive_health{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_health{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_info
+        expr: |
+          rustfs_system_drive_info{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_info{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_io_errors_total
+        expr: |
+          rustfs_system_drive_io_errors_total{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_io_errors_total{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_offline_count
+        expr: |
+          rustfs_system_drive_offline_count{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_offline_count{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_offline_duration_seconds
+        expr: |
+          rustfs_system_drive_offline_duration_seconds{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_offline_duration_seconds{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_online_count
+        expr: |
+          rustfs_system_drive_online_count{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_online_count{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_perc_util
+        expr: |
+          rustfs_system_drive_perc_util{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_perc_util{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_present
+        expr: |
+          rustfs_system_drive_present{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_present{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_reads_await
+        expr: |
+          rustfs_system_drive_reads_await{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_reads_await{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_reads_kb_per_sec
+        expr: |
+          rustfs_system_drive_reads_kb_per_sec{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_reads_kb_per_sec{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_reads_per_sec
+        expr: |
+          rustfs_system_drive_reads_per_sec{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_reads_per_sec{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_runtime_state
+        expr: |
+          rustfs_system_drive_runtime_state{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_runtime_state{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_scanning
+        expr: |
+          rustfs_system_drive_scanning{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_scanning{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_timeout_errors_total
+        expr: |
+          rustfs_system_drive_timeout_errors_total{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_timeout_errors_total{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_total_bytes
+        expr: |
+          rustfs_system_drive_total_bytes{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_total_bytes{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_total_inodes
+        expr: |
+          rustfs_system_drive_total_inodes{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_total_inodes{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_used_bytes
+        expr: |
+          rustfs_system_drive_used_bytes{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_used_bytes{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_used_inodes
+        expr: |
+          rustfs_system_drive_used_inodes{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_used_inodes{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_waiting_io
+        expr: |
+          rustfs_system_drive_waiting_io{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_waiting_io{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_writes_await
+        expr: |
+          rustfs_system_drive_writes_await{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_writes_await{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_writes_kb_per_sec
+        expr: |
+          rustfs_system_drive_writes_kb_per_sec{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_writes_kb_per_sec{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_writes_per_sec
+        expr: |
+          rustfs_system_drive_writes_per_sec{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_writes_per_sec{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))
+      - record: rustfs:storage:current
+        labels:
+          source_metric: rustfs_system_drive_writes_total
+        expr: |
+          rustfs_system_drive_writes_total{rustfs_cluster_id!="",collection_scope="local"}
+          and (timestamp(rustfs_system_drive_writes_total{rustfs_cluster_id!="",collection_scope="local"})
+            >= on (rustfs_cluster_id, observer, collection_scope, job, instance) group_left()
+            (floor(rustfs:storage_snapshot:fresh * 1000) / 1000))

.docker/observability/tests/storage-rules.test.yml

@@ -0,0 +1,189 @@
+rule_files:
+- ../prometheus-rules/rustfs-storage.yml
+evaluation_interval: 15s
+tests:
+- name: owners, pools, clusters and observer views remain distinct in mixed versions
+  interval: 1m
+  input_series:
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector",server="n0",drive="/data",pool_index="0",set_index="0",drive_index="0"}
+    values: 100x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n1",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n1",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="a",observer="n1",collection_scope="local",job="rustfs",instance="collector",server="n1",drive="/data",pool_index="0",set_index="0",drive_index="1"}
+    values: 100x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n2",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n2",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="a",observer="n2",collection_scope="local",job="rustfs",instance="collector",server="n2",drive="/data",pool_index="1",set_index="0",drive_index="0"}
+    values: 100x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n3",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n3",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="a",observer="n3",collection_scope="local",job="rustfs",instance="collector",server="n3",drive="/data",pool_index="1",set_index="0",drive_index="1"}
+    values: 100x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="b",observer="n0",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="b",observer="n0",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="b",observer="n0",collection_scope="local",job="rustfs",instance="collector",server="n0",drive="/data",pool_index="0",set_index="0",drive_index="0"}
+    values: 1000x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="b",observer="n1",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="b",observer="n1",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="b",observer="n1",collection_scope="local",job="rustfs",instance="collector",server="n1",drive="/data",pool_index="0",set_index="0",drive_index="1"}
+    values: 1000x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="b",observer="n2",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="b",observer="n2",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="b",observer="n2",collection_scope="local",job="rustfs",instance="collector",server="n2",drive="/data",pool_index="1",set_index="0",drive_index="0"}
+    values: 1000x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="b",observer="n3",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="b",observer="n3",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="b",observer="n3",collection_scope="local",job="rustfs",instance="collector",server="n3",drive="/data",pool_index="1",set_index="0",drive_index="1"}
+    values: 1000x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n0",collection_scope="cluster",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n0",collection_scope="cluster",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_cluster_capacity_raw_total_bytes{rustfs_cluster_id="a",observer="n0",collection_scope="cluster",job="rustfs",instance="collector"}
+    values: 400x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n1",collection_scope="cluster",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n1",collection_scope="cluster",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_cluster_capacity_raw_total_bytes{rustfs_cluster_id="a",observer="n1",collection_scope="cluster",job="rustfs",instance="collector"}
+    values: 300x8
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="a",observer="old",server="n0",drive="/data",job="rustfs",instance="collector"}
+    values: 999x8
+  promql_expr_test:
+  - expr: sum by (rustfs_cluster_id) (rustfs:storage:current{source_metric="rustfs_system_drive_total_bytes",collection_scope="local"})
+    eval_time: 2m
+    exp_samples:
+    - labels: '{rustfs_cluster_id="a"}'
+      value: 400
+    - labels: '{rustfs_cluster_id="b"}'
+      value: 4000
+  - expr: count by (rustfs_cluster_id,pool_index) (rustfs:storage:current{source_metric="rustfs_system_drive_total_bytes"})
+    eval_time: 2m
+    exp_samples:
+    - labels: '{rustfs_cluster_id="a",pool_index="0"}'
+      value: 2
+    - labels: '{rustfs_cluster_id="a",pool_index="1"}'
+      value: 2
+    - labels: '{rustfs_cluster_id="b",pool_index="0"}'
+      value: 2
+    - labels: '{rustfs_cluster_id="b",pool_index="1"}'
+      value: 2
+  - expr: sum(rustfs:storage:current{source_metric="rustfs_cluster_capacity_raw_total_bytes",rustfs_cluster_id="a",observer="n0"})
+    eval_time: 2m
+    exp_samples:
+    - labels: '{}'
+      value: 400
+  - expr: sum(rustfs:storage:current{source_metric="rustfs_cluster_capacity_raw_total_bytes",rustfs_cluster_id="a",observer="n1"})
+    eval_time: 2m
+    exp_samples:
+    - labels: '{}'
+      value: 300
+- name: removed values and identities cannot rejoin new snapshots; stalled sources expire
+  interval: 1m
+  input_series:
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n1",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0 60 _ _ _ _ _ _ _
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n1",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_info{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector",server="n0",drive="/data",disk_id="old"}
+    values: 1 1 _ _ _ _ _ _ _
+  - series: rustfs_system_drive_info{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector",server="n0",drive="/data",disk_id="new"}
+    values: _ _ 1 1 1 1 1 1 1
+  - series: rustfs_system_drive_waiting_io{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector",server="n0",drive="/data"}
+    values: 7 7 _ _ _ _ _ _ _
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="a",observer="n1",collection_scope="local",job="rustfs",instance="collector",server="n1",drive="/data"}
+    values: 100 100 _ _ _ _ _ _ _
+  promql_expr_test:
+  - expr: rustfs:storage:current{source_metric="rustfs_system_drive_info",disk_id="old"}
+    eval_time: 2m
+    exp_samples: []
+  - expr: rustfs:storage:current{source_metric="rustfs_system_drive_waiting_io"}
+    eval_time: 2m
+    exp_samples: []
+  - expr: count(rustfs:storage:current{source_metric="rustfs_system_drive_info",disk_id="new"})
+    eval_time: 2m
+    exp_samples:
+    - labels: '{}'
+      value: 1
+  - expr: rustfs:storage:current{source_metric="rustfs_system_drive_total_bytes"}
+    eval_time: 5m
+    exp_samples: []
+  - expr: rustfs:storage_snapshot:fresh{observer="n1"}
+    eval_time: 5m
+    exp_samples: []
+- name: counters reset independently; filter versions before rate and sum rates across owners
+  interval: 1m
+  input_series:
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_api_calls_total{rustfs_cluster_id="a",observer="n0",collection_scope="local",job="rustfs",instance="collector",server="n0",drive="/data",disk_id="n0-disk",api="read_all"}
+    values: 0 60 120 30 90 150 210
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="n1",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0+60x8
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="n1",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x8
+  - series: rustfs_system_drive_api_calls_total{rustfs_cluster_id="a",observer="n1",collection_scope="local",job="rustfs",instance="collector",server="n1",drive="/data",disk_id="n1-disk",api="read_all"}
+    values: 0+120x6
+  - series: rustfs_system_drive_api_calls_total{rustfs_cluster_id="a",server="n0",drive="/data",observer="old"}
+    values: 0+999x6
+  promql_expr_test:
+  - expr: sum(resets(rustfs_system_drive_api_calls_total{collection_scope="local"}[5m]))
+    eval_time: 5m
+    exp_samples:
+    - labels: '{}'
+      value: 1
+  - expr: sum(rate(rustfs_system_drive_api_calls_total{collection_scope="local"}[2m]) and ignoring(source_metric) rustfs:storage:current{source_metric="rustfs_system_drive_api_calls_total"})
+    eval_time: 5m
+    exp_samples:
+    - labels: '{}'
+      value: 3
+- name: completed slow collection has no validity budget
+  interval: 1m
+  input_series:
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="slow",collection_scope="cluster",job="rustfs",instance="collector"}
+    values: 0+60x3
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="slow",collection_scope="cluster",job="rustfs",instance="collector"}
+    values: '0x3'
+  promql_expr_test:
+  - expr: rustfs:storage_snapshot:fresh{observer="slow"}
+    eval_time: 2m
+    exp_samples: []
+- name: submillisecond publication and millisecond OTLP samples share a cutoff
+  interval: 1m
+  input_series:
+  - series: rustfs_storage_snapshot_last_success_timestamp_seconds{rustfs_cluster_id="a",observer="submillisecond",collection_scope="local",job="rustfs",instance="collector"}
+    values: 0.0009+60x3
+  - series: rustfs_storage_snapshot_max_age_seconds{rustfs_cluster_id="a",observer="submillisecond",collection_scope="local",job="rustfs",instance="collector"}
+    values: 180x3
+  - series: rustfs_system_drive_total_bytes{rustfs_cluster_id="a",observer="submillisecond",collection_scope="local",job="rustfs",instance="collector",server="submillisecond",drive="/data"}
+    values: 100x3
+  promql_expr_test:
+  - expr: count(rustfs:storage:current{source_metric="rustfs_system_drive_total_bytes",observer="submillisecond"})
+    eval_time: 2m15s
+    exp_samples:
+    - labels: '{}'
+      value: 1

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

@@ -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

.github/s3tests/s3tests.conf

@@ -55,6 +55,10 @@ secret_key = ${S3_SECRET_KEY}
 
 ## replace with key id obtained when secret is created, or delete if KMS not tested
 #kms_keyid = 01234567-89ab-cdef-0123-456789abcdef
+#kms_keyid2 = fedcba98-7654-3210-fedc-ba9876543210
+
+## Expected service default for SSE-KMS requests without a key id; empty means none
+#kms_default_keyid =
 
 ## Storage classes
 #storage_classes = "LUKEWARM, FROZEN"

.github/workflows/audit.yml

@@ -16,7 +16,7 @@ name: Security Audit
 
 on:
   push:
-    branches: [ main, release ]
+    branches: [ main ]
     paths:
       - '**/Cargo.toml'
       - '**/Cargo.lock'
@@ -32,7 +32,7 @@ on:
       - 'scripts/security/check_workflow_pins.sh'
   pull_request:
     types: [ opened, synchronize, reopened, closed ]
-    branches: [ main, release ]
+    branches: [ main ]
     paths:
       - '**/Cargo.toml'
       - '**/Cargo.lock'

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

@@ -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, release ]
-    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."

.github/workflows/ci.yml

@@ -37,25 +37,6 @@ on:
   pull_request:
     types: [ opened, synchronize, reopened, closed ]
     branches: [ main, release ]
-    # 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
+    name: Workspace Test and Lint
-    if: github.event_name != 'pull_request' || github.event.action != 'closed'
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
-    needs: [ quick-checks ]
+    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'
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
-    needs: [ quick-checks ]
+    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'
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
-    needs: [ quick-checks ]
+    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'
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
-    needs: [ quick-checks ]
+    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'
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
-    needs: [ quick-checks ]
+    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'
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
-    needs: [ quick-checks ]
+    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'
+    if: needs.classify-changes.outputs.mode == 'full' && (github.event_name != 'pull_request' || github.event.action != 'closed')
-    needs: [ quick-checks ]
+    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/
@@ -1002,22 +970,11 @@ jobs:
       # debug binary; each test spawns its own rustfs server on a random port.
       - name: Run e2e full suite
         env:
-          RUSTFS_E2E_LOG_DIR: ${{ runner.temp }}/rustfs-e2e-full-logs
-          RUSTFS_HEAL_CHAOS_SERVER_RUST_LOG: rustfs::heal=debug,rustfs_heal=debug,rustfs_ecstore::disk=debug,rustfs_lock=debug,rustfs=error
           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
         run: cargo nextest run --profile e2e-full -p e2e_test
 
-      - name: Upload e2e full diagnostics
-        if: failure()
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
-        with:
-          name: e2e-full-diagnostics-${{ github.run_number }}
-          path: ${{ runner.temp }}/rustfs-e2e-full-logs/
-          if-no-files-found: warn
-          retention-days: 7
-
       - name: Upload junit
         if: always()
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
@@ -1224,9 +1181,44 @@ 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

.github/workflows/e2e-distributed.yml

@@ -19,7 +19,7 @@
 # case is a two-site 4-node 1-drive pair or a 4-node upgrade). Membership is
 # `[profile.e2e-distributed]` in `.config/nextest.toml`. Storage-sensitive PRs,
 # nightly runs, and manual dispatches all execute the same fail-closed suite.
-# Upgrade cases use an independent 1.0.0-rc.2 pin defined below.
+# Upgrade cases download the same pinned previous release as e2e-upgrade.yml.
 #
 # Isolated pool filesystems: expand/decommission/rebalance cases require
 # independent `statfs` capacity. This job runs on GitHub-hosted
@@ -87,7 +87,7 @@ jobs:
       NO_PROXY: 127.0.0.1,localhost
       HTTP_PROXY: ""
       HTTPS_PROXY: ""
-      # Independent 1.0.0-rc.2 source pin for distributed::upgrade_test.
+      # Pinned previous release used by distributed::upgrade_test (same pin as e2e-upgrade.yml).
       UPGRADE_SOURCE_VERSION: 1.0.0-rc.2
       UPGRADE_SOURCE_ASSET: rustfs-linux-x86_64-gnu-v1.0.0-rc.2.zip
       UPGRADE_SOURCE_SHA256: 7c789386bf85278f865b8e0d359bf4edb84d5aa408cc3fa54a18c25ca74cd6e7

.github/workflows/e2e-upgrade.yml

@@ -129,11 +129,22 @@ jobs:
           : > target/debug/rustfs.features
 
       - name: Run upgrade compatibility test
+        env:
+          RUSTFS_SCANNER_HEAL_G09_EVIDENCE_DIR: ${{ runner.temp }}/rustfs-upgrade-g09-evidence/${{ matrix.artifact }}
         run: |
           cargo test --locked -p e2e_test \
             "upgrade_compatibility_test::${{ matrix.test }}" \
             -- --ignored --exact --nocapture
 
+      - name: Upload scanner/heal G09 evidence
+        if: always()
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6
+        with:
+          name: ${{ matrix.artifact }}-g09-evidence-${{ github.run_number }}
+          path: ${{ runner.temp }}/rustfs-upgrade-g09-evidence/${{ matrix.artifact }}
+          if-no-files-found: ignore
+          retention-days: 14
+
       - name: Upload server logs
         if: always()
         uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f # v6

.github/workflows/nightly-gnu.yml

@@ -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

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

@@ -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

.github/workflows/windows-filesystem.yml

@@ -16,7 +16,7 @@ name: Windows Filesystem Tests
 
 on:
   push:
-    branches: [ main ]
+    branches: [ main, release ]
     paths:
       - "crates/ecstore/src/disk/**"
       - "crates/ecstore/src/store/init_format.rs"
@@ -26,7 +26,7 @@ on:
       - ".github/actions/setup/**"
       - ".github/workflows/windows-filesystem.yml"
   pull_request:
-    branches: [ main ]
+    branches: [ main, release ]
     paths:
       - "crates/ecstore/src/disk/**"
       - "crates/ecstore/src/store/init_format.rs"

.gitignore

@@ -52,6 +52,7 @@ docs
 __pycache__/
 !docs/
 docs/*
+!docs/README.md
 !docs/architecture/
 !docs/architecture/**
 !docs/operations/

CHANGELOG.md

@@ -7,12 +7,30 @@ 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.
 
 ### Fixed
 - **Fresh multi-pool bootstrap with distinct format creators**: a new deployment whose pools have their first endpoint on different nodes (for example two single-node pools) could never publish its initial `pool.bin`: each node held fresh-bootstrap proof only for the pool it formatted, the deployment-wide proof collapsed to none, and every node died with `pool metadata recovery required: no durable bootstrap identity or pool.bin replica is available` after the startup retry budget. The first pool's creator now mints the pending cluster identity on its own pool, every other creator copies that nonce-bound identity onto the pool it formatted first-hand, and the elected writer publishes `pool.bin` once every pool replica carries the same pending identity. Corrupt or disagreeing replicas, pools that merely have a format, expansion pools joining an initialized deployment, and restarts without first-hand proof still fail closed. Non-elected nodes that start before `pool.bin` exists, and the elected writer while it waits for the other creators, no longer latch their pool-metadata write gate for the life of the process. Refs rustfs/backlog#2338, rustfs/backlog#2375.
 - **Lock RPC timeout storms** (#7363): the remote lock client no longer evicts and re-dials the shared internode HTTP/2 channel on every request deadline. A timeout evicts only when the peer has not completed any lock RPC for two deadlines, evictions and transport-failure re-dials are rate limited per peer (`RUSTFS_OBJECT_LOCK_RPC_EVICTION_COOLDOWN_MS`, default 5 s), and a timed-out request is left running instead of being reset (bounded per peer by `RUSTFS_OBJECT_LOCK_RPC_DETACHED_LIMIT`, default 256), so a slow lock endpoint can no longer drive the `RST_STREAM`/`GOAWAY too_many_resets`/reconnect loop. A lock granted after its caller timed out is released immediately, and unlocks that fail the quick retries continue on a deferred 1/2/4/8/16 s schedule before the server lease reclaims them. New `rustfs_remote_lock_*` metrics cover timeouts, evictions, suppressed evictions, detached streams, late completions and late releases per peer. Operator guide at `docs/operations/lock-rpc-storm-protection.md`.
+- **KMS failures on the S3 data path carry an actionable status**: only "key not found" and a backend outage were classified; every other KMS failure — a disabled or pending-deletion key, a denied KMS grant, an encryption-context mismatch, an unsupported algorithm, a credential or timeout failure, a capability the backend does not have — collapsed onto `500 InternalError`. SDKs therefore applied exponential backoff to configuration errors that no retry can fix, and monitoring filed every one of them as a server fault. Unusable-key and request-side failures now return `400`, a denied grant `403`, transient backend failures `503` — including a key store the backend could not read, so an outage stays distinguishable from a missing key all the way to the client — and a missing backend capability `501`. Damaged or unreadable key material still returns `500`, which is what it is.
+- **SSE-C on buckets with default encryption**: a `PutObject` carrying a valid SSE-C header triple on a bucket that has default encryption configured no longer fails with `400 InvalidArgument` ("The SSE-C and managed server-side encryption headers cannot be used together"). PUT and the POST-object/extract path resolved the bucket default with a hard-coded "no explicit SSE-C" flag, so the default was layered onto the request and then tripped the request's own mutual-exclusion check; an SSE-C request now suppresses the bucket default on all three write paths, matching COPY and AWS S3. Every bucket with default encryption previously refused SSE-C single PUTs outright, while `CreateMultipartUpload` on the same bucket succeeded.
+- **Explicit SSE-S3 on SSE-KMS-default buckets**: `x-amz-server-side-encryption: AES256` against a bucket whose default is `aws:kms` no longer fails with `400 InvalidArgument`. The bucket default's KMS key id was inherited independently of the effective algorithm, producing a self-contradictory `AES256` + key-id pair; the key id is now inherited only when the effective algorithm is `aws:kms`. `PutBucketEncryption` fills in a default key id automatically, so this affected nearly every SSE-KMS-default bucket.
+- **Restore of encrypted or compressed multipart objects (silent data corruption)**: restoring a multipart object from a remote tier addressed the tier in *plaintext* coordinates while the copy-back reads the *stored* representation. Every part received a misaligned slice of the remote object whose length still satisfied the range, the hash reader and the completion size check, so the restore reported success and replaced the object's bytes. Restore now accumulates stored part sizes, passes the stored length to the hash reader alongside the plaintext length, and validates against the stored size. Objects restored by an affected release must be re-restored from the tier or recovered from a backup — this release does not detect or repair them retroactively.
+- **Restore no longer drifts the object ETag**: the copy-back digests stored (encrypted or compressed) bytes, so the recomputed MD5 is not the object's public ETag. Single-part and multipart restores now preserve the original object ETag, and each restored part keeps its own recorded part ETag.
+- **ILM archive no longer forwards encryption metadata to the tier**: transition requests carried the object's SSE headers and the RustFS-wrapped data key as request headers. Any S3 target rejected an SSE-C archive outright (`400`, no key supplied), an SSE-KMS archive asked the target to encrypt a second time under a key id it does not own, and the wrapped DEK left the cluster. The archive request now strips every SSE header and encryption marker using the same predicate the replication path uses; the local `xl.meta` keeps all of it, so read-through and restore are unaffected.
+- **KMS reload is no longer a no-op on a node whose KMS failed to start**: `POST /rustfs/admin/v3/kms/reload` short-circuited whenever the persisted configuration matched the in-memory one byte for byte. A node whose KMS failed to start (for example Vault briefly unreachable during a rolling restart) keeps that configuration and sits in `Error`, so the documented recovery call returned "reloaded successfully" while leaving the node down — and did the same on every peer through the reload broadcast. Reload now short-circuits only for a service that is actually running, and otherwise reconfigures, which starts the service.
+- **AWS KMS capability reporting**: the AWS backend no longer advertises `versioning` support through `GET /rustfs/admin/v3/kms/status`. AWS KMS key versions are not enumerable through this backend, as the backend documentation already stated.
 - **Multipart admission queue**: an `UploadPart` waiting for a foreground write permit now waits at most 10 s by default (`RUSTFS_PUT_MULTIPART_FOREGROUND_ADMISSION_WAIT_TIMEOUT_MS`, previously 30 s), so a queued part returns S3 `SlowDown` before the client's socket write timeout drops the connection. Separately, the API listener no longer forces a 4 MiB `SO_RCVBUF` on every accepted socket (kernel autotuning applies; `RUSTFS_HTTP_SOCKET_RECV_BUFFER_BYTES` restores a fixed size), so a queued part no longer lets up to 8 MiB of unread body accumulate in kernel memory per connection, which is what throttled whole nodes under SDK-default multipart concurrency. Fixes #7385.
 - **Helm Ingress**: `customAnnotations` are now merged with class-specific annotations (nginx/traefik) instead of being ignored when `ingress.className` is set.
 - **Per-pool erasure parity**: Erasure parity (STANDARD and reduced-redundancy) is now resolved independently for every pool instead of reusing the first pool's value. A heterogeneous topology — for example a 4-drive pool plus a 2-drive pool created during expansion — previously inherited the first pool's parity and could resolve to zero data shards in the smaller pool, panicking Reed-Solomon construction on write. Automatic parity now resolves per pool (for example `2+2` in the 4-drive pool and `1+1` in the 2-drive pool). Fixes #4801.
@@ -54,6 +72,18 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   - Four-layer regression-prevention tests guard against silent feature deletion: compile-time module assertion, module-presence unit test, cross-module `Protocol` enum assertion, end-to-end SSH banner test against the running binary
 
 ### Changed
+- **Encryption and KMS work merged since `1.0.0-rc.5`** (entries were missing from this section):
+  - **Persisted KMS configuration secrets** are sealed field-by-field with `RUSTFS_KMS_CONFIG_SECRET`. **When the variable is unset the secrets are persisted in cleartext and the server only warns** (`persisted KMS configuration carries cleartext secrets`); it never refuses the write. Set it, identically, on every node, and re-save the configuration to seal an existing one.
+  - **New v2 ciphertext frame format** with per-frame index binding and final-frame authentication. Its write switch `RUSTFS_ENCRYPTION_FRAME_V2` is **off by default**: v2 frames are unreadable by nodes without v2 read support, and encrypted ciphertext travels verbatim through transition, decommission and SSE-C replication passthrough, so turn it on only after every node — and every RustFS warm/replication target that receives raw ciphertext — runs a release with v2 read support. Reading v2 objects needs no switch.
+  - **Per-key SSE-KMS authorization** (`RUSTFS_KMS_ENFORCE_SSE_KEY_POLICY`, default `false`). With it on, anonymous callers hold no KMS grants, so **a public bucket serving SSE-KMS objects is an incompatible combination** and those reads return `AccessDenied`.
+  - Envelope context binding as KMS AAD (`ENV_KMS_ENVELOPE_AAD`, off by default; a node that predates the field cannot open bound envelopes).
+  - Vault custom CA and mutual TLS; object-level DEK rewrap plus a batch rekey admin API; a backend-locality runtime signal on `kms/status`.
+  - Single-pass decryption for encrypted GET, and encrypted single-part closed-range seek — the latter is now **on by default** (`RUSTFS_ENCRYPTED_RANGE_SEEK`, default `true`; the switch remains as a kill switch).
+- **Vault static tokens are now tracked and renewed**: with `Token` authentication RustFS hard-coded "this token has no lease", so the renewal task never started and no remaining-TTL gauge was published. `vault token create` grants a 768-hour TTL by default, which turned a healthy-looking cluster into one where every KMS call returned 403 about a month later, with no self-healing short of a restart or reconfigure. RustFS now calls `auth/token/lookup-self` at login and adopts what Vault reports: a non-expiring token behaves exactly as before, an expiring renewable one is renewed at half TTL like the other auth methods, and an expiring non-renewable one logs `vault_static_token_not_renewable` and publishes its remaining TTL. The probe never fails the login: a token whose policy omits `lookup-self` (Vault's `default` policy grants it), or a Vault that is unreachable at that moment, logs `vault_static_token_lookup_failed` and falls back to the previous no-lease behaviour, so no deployment that works today stops working.
+- **SSE-C over a plaintext transport is reported**: AWS S3 and MinIO refuse an SSE-C request that did not arrive over TLS, because the customer key travels in a request header. RustFS accepted them on any transport and still does by default — flipping to a rejection inside a release window would break plaintext staging and test deployments. Each such request now increments `rustfs_ssec_plaintext_requests_total` and logs one `ssec_request_without_tls` warning per process, and `RUSTFS_SSE_C_REQUIRE_TLS=true` opts into the AWS `400` now. The default is expected to flip in a later release; confirm the counter reads zero first. The verdict is per connection: a TLS listener satisfies it, and so does an `https` protocol forwarded by a proxy the trusted-proxy configuration accepts.
+- **Local KMS backend on a distributed deployment says what actually breaks**: the backend keeps key material and its Argon2id salt on each node's own disk, so two nodes derive different keys from the same `master_key` and an object encrypted on one node cannot be decrypted on another — intermittent 500s behind a load balancer. Configuring it while the deployment is distributed now logs `kms_node_local_backend_in_distributed_deployment` and appends that consequence to the `kms/configure` response, instead of only the generic "development only" positioning warning. It remains a warning, not a gate.
+- **SSE-KMS is refused when no KMS is running (breaking)**: a write requesting `x-amz-server-side-encryption: aws:kms` on a node with no KMS service no longer succeeds. Earlier releases wrapped the data key with the node-local `RUSTFS_SSE_S3_MASTER_KEY` while still writing `aws:kms` and the requested key id into the object metadata — metadata that claimed a KMS protection the object never had, under a key that was never consulted. Such a request now returns `400 InvalidRequest` when KMS was never configured and `503` when a configured service is not running; the refusal is evaluated after the per-key authorization gate, so an unauthorized caller still receives `403 AccessDenied`. **Upgrade note:** a deployment that relied on this write succeeding will start receiving 4xx/503. Either configure a KMS, or request `AES256` and keep the documented SSE-S3 local-master-key fallback, which is unchanged. Objects already written this way remain readable.
+- **Legacy ciphertext nonce layouts are now locked per segment**: while decrypting a v1 segment, the reader locks onto whichever of the three historical nonce layouts decoded the segment's first non-zero-index frame and rejects any later frame that needs a different one. Because a frame encrypted at block index zero authenticates under the pre-`1.0.0-alpha.91` reused-part-nonce layout at any position, an attacker able to rewrite the underlying shards could previously replay it and have the forged plaintext returned with `200`. New `RUSTFS_ENCRYPTION_LEGACY_NONCE_FALLBACK` (default `true`) drops that third layout entirely when set to `false`, which closes the residual case of a stream built purely from repeats of frame zero. Turn it off only after migrating pre-alpha.91 encrypted objects (rewrite in place with CopyObject); see [KMS backend security properties](docs/operations/kms-backend-security.md) for what the v1 frame layout does and does not authenticate.
 - **HTTP Server Stack**: Integrated `KeystoneAuthLayer` middleware from `rustfs-keystone` crate into service stack (positioned after ReadinessGateLayer)
 - **Storage-class validation on startup (upgrade note)**: A persisted explicit storage class (`RUSTFS_STORAGE_CLASS_STANDARD` / `RUSTFS_STORAGE_CLASS_RRS`, for example `EC:2`) is now validated against the actual per-pool drive counts at startup and rejected when a pool cannot satisfy it. This is fail-closed and correct, but a cluster that persisted a storage class larger than a small or heterogeneous pool can hold (for example `EC:2` alongside a 2-drive pool), which earlier releases accepted and silently resolved to an invalid layout, will now refuse to start after upgrade. To recover, unset `RUSTFS_STORAGE_CLASS_STANDARD` so the server derives a valid per-pool default automatically, or set it to a value every pool can satisfy.
 - **IAMAuth**: Enhanced `get_secret_key()` to return empty secret for Keystone credentials (bypasses signature validation)

Cargo.lock

@@ -9928,6 +9928,7 @@ dependencies = [
  "async-trait",
  "base64-simd",
  "bytes",
+ "chrono",
  "crc-fast",
  "futures",
  "hotpath",
@@ -10578,6 +10579,7 @@ dependencies = [
  "serde_json",
  "sha1 0.11.0",
  "sha2 0.11.0",
+ "temp-env",
  "thiserror 2.0.20",
  "tokio",
  "tokio-test",

README.md

@@ -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

README_ZH.md

@@ -46,14 +46,31 @@ RustFS 是一个基于 Rust 构建的高性能分布式对象存储系统。Rust
 - **完全开源**：采用 Apache 2.0 许可证，鼓励社区贡献和商业使用。
 - **简单易用**：设计简洁，易于部署和管理。
 
-| 功能               | 状态    | 功能                    | 状态      |
+状态说明：✅ 可用 —— 已发布并有 CI 门禁覆盖；🧪 预览 —— 已发布但需显式开关，或兼容性承诺有边界。
-| :----------------- | :------ | :---------------------- | :-------- |
+
-| **S3 核心功能**    | ✅ 可用 | **Bitrot (防数据腐烂)** | ✅ 可用   |
+| 功能                        | 状态    | 功能                     | 状态    |
-| **上传 / 下载**    | ✅ 可用 | **单机模式**            | ✅ 可用   |
+| :-------------------------- | :------ | :----------------------- | :------ |
-| **版本控制**       | ✅ 可用 | **存储桶复制**          | ✅ 可用   |
+| **S3 核心功能**             | ✅ 可用 | **分布式模式**           | ✅ 可用 |
-| **日志功能**       | ✅ 可用 | **生命周期管理**        | 🚧 测试中 |
+| **上传 / 下载**             | ✅ 可用 | **单机模式**             | ✅ 可用 |
-| **事件通知**       | ✅ 可用 | **分布式模式**          | 🚧 测试中 |
+| **版本控制**                | ✅ 可用 | **Bitrot (防数据腐烂)**  | ✅ 可用 |
-| **K8s Helm Chart** | ✅ 可用 | **OPA (策略引擎)**      | 🚧 测试中 |
+| **对象锁定 (WORM)**         | ✅ 可用 | **修复与扫描器**         | ✅ 可用 |
+| **服务端加密 (SSE)**        | ✅ 可用 | **存储池扩容 / 下线**    | ✅ 可用 |
+| **RustFS KMS**              | ✅ 可用 | **存储桶复制**           | ✅ 可用 |
+| **生命周期管理 (ILM)**      | ✅ 可用 | **站点复制**             | ✅ 可用 |
+| **ILM 分层 (远端 S3)**      | ✅ 可用 | **存储桶配额**           | ✅ 可用 |
+| **S3 Select**               | ✅ 可用 | **事件通知**             | ✅ 可用 |
+| **S3 Tables (Iceberg REST)**| 🧪 预览 | **审计日志**             | ✅ 可用 |
+| **IAM / 策略**              | ✅ 可用 | **日志与可观测性**       | ✅ 可用 |
+| **OIDC / SSO**              | ✅ 可用 | **Web 控制台**           | ✅ 可用 |
+| **Keystone 认证**           | ✅ 可用 | **K8s Helm Chart**       | ✅ 可用 |
+| **Swift API**               | ✅ 可用 | **FTPS / WebDAV**        | ✅ 可用 |
+| **多租户**                  | ✅ 可用 | **SFTP**                 | ✅ 可用 |
+| **MinIO 磁盘格式兼容**      | 🧪 预览 |                          |         |
+
+说明：
+
+- **服务端加密**：支持 SSE-C、SSE-S3 与 SSE-KMS。SSE-KMS 必须先配置 KMS 服务；未配置 KMS 时请求 `aws:kms` 会被拒绝，不会降级到本地主密钥。
+- **RustFS KMS**：生产环境支持 Vault（KV2 / Transit）与 AWS KMS 后端；`Local` 与 `Static` 后端仅供开发与测试使用，详见 [KMS 后端安全属性](docs/operations/kms-backend-security.md)。
 
 ## RustFS vs MinIO 性能对比
 
@@ -150,7 +167,10 @@ 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

crates/data-usage/src/data_usage.rs

@@ -593,6 +593,20 @@ pub struct DataUsageSnapshotIdentity {
     pub scanner_epoch: Option<u64>,
 }
 
+#[derive(Clone, Debug, Default, Serialize, Deserialize, PartialEq, Eq)]
+pub struct DataUsageSegmentInvalidationProof {
+    #[serde(default)]
+    pub process_epoch: String,
+    #[serde(default)]
+    pub generation_start: u64,
+    #[serde(default)]
+    pub generation_end: u64,
+    #[serde(default)]
+    pub producer_identity_coverage_complete: bool,
+    #[serde(default)]
+    pub cold_zero_walk_oracle: bool,
+}
+
 #[derive(Clone, Debug, Default, Serialize, Deserialize, PartialEq, Eq)]
 pub struct DataUsageSnapshotSetState {
     pub pool_index: u64,
@@ -607,6 +621,8 @@ pub struct DataUsageSnapshotSetState {
     pub complete: bool,
     #[serde(default)]
     pub tombstone: bool,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub segment_invalidation_proof: Option<DataUsageSegmentInvalidationProof>,
 }
 
 impl DataUsageInfo {
@@ -3073,6 +3089,7 @@ mod tests {
             scan_plan_digest: Some([1; 32]),
             complete: false,
             tombstone: false,
+            segment_invalidation_proof: None,
         }];
         assert!(observed_data_usage_is_newer(&partial, &authoritative));
     }
@@ -3095,6 +3112,7 @@ mod tests {
                     scan_plan_digest: Some([1; 32]),
                     complete: true,
                     tombstone: false,
+                    segment_invalidation_proof: None,
                 },
                 DataUsageSnapshotSetState {
                     pool_index: 1,
@@ -3104,6 +3122,7 @@ mod tests {
                     scan_plan_digest: Some([2; 32]),
                     complete: false,
                     tombstone: false,
+                    segment_invalidation_proof: None,
                 },
             ],
             ..Default::default()
@@ -3113,6 +3132,42 @@ mod tests {
         assert!(partial.is_valid_partial_snapshot());
     }
 
+    #[test]
+    fn set_state_segment_invalidation_proof_is_additive() {
+        #[derive(Deserialize)]
+        struct LegacySetState {
+            pool_index: u64,
+            set_index: u64,
+            complete: bool,
+        }
+
+        let proof = DataUsageSegmentInvalidationProof {
+            process_epoch: "scanner-process".to_string(),
+            generation_start: 3,
+            generation_end: 5,
+            producer_identity_coverage_complete: true,
+            cold_zero_walk_oracle: true,
+        };
+        let state = DataUsageSnapshotSetState {
+            pool_index: 1,
+            set_index: 2,
+            scanner_cycle: Some(9),
+            scanner_epoch: Some(4),
+            scan_plan_digest: Some([7; 32]),
+            complete: true,
+            tombstone: false,
+            segment_invalidation_proof: Some(proof.clone()),
+        };
+        let encoded = rmp_serde::to_vec_named(&state).expect("set state should encode with additive proof");
+        let legacy: LegacySetState = rmp_serde::from_slice(&encoded).expect("legacy readers should ignore proof metadata");
+        assert_eq!(legacy.pool_index, 1);
+        assert_eq!(legacy.set_index, 2);
+        assert!(legacy.complete);
+
+        let decoded: DataUsageSnapshotSetState = rmp_serde::from_slice(&encoded).expect("new readers should restore proof");
+        assert_eq!(decoded.segment_invalidation_proof, Some(proof));
+    }
+
     #[test]
     fn completeness_marker_requires_a_snapshot_timestamp() {
         let untimestamped = DataUsageInfo {

crates/e2e_test/README.md

@@ -30,6 +30,8 @@ Registered in [`src/lib.rs`](src/lib.rs). Grouped by concern:
 | **chaos / reliability** | [`src/chaos.rs`](src/chaos.rs), `reliability_disk_fault_test`, `heal_erasure_disk_rebuild_test`, `server_startup_failfast_test` | Disk offline/replace/corrupt, EC rebuild, heal, fail-fast startup |
 | **upgrade compatibility** | `upgrade_compatibility_test` | Pinned previous-release writes followed by current-build reads on the same data directory |
 
+The external-tool `storage_metric_ownership_test` validates the OTLP/Collector/Prometheus path, including a rolling upgrade and node failures. See the [storage metrics guide](../../docs/operations/storage-metrics.md) for its required binaries and focused command.
+
 ## How to run
 
 All commands assume repo root. `cargo test` triggers an on-demand build of the

crates/e2e_test/build.rs

@@ -51,7 +51,7 @@ fn main() {
         }
     }
     let revision = git(&root, &["rev-parse", "HEAD"]).unwrap_or_else(|| "unknown".to_owned());
-    let dirty = git(&root, &["status", "--porcelain", "--untracked-files=normal"]).is_none_or(|status| !status.is_empty());
+    let dirty = git(&root, &["status", "--porcelain", "--untracked-files=no"]).is_none_or(|status| !status.is_empty());
     let lock = git(&root, &["hash-object", "Cargo.lock"]).unwrap_or_else(|| "unknown".to_owned());
     let mut features = std::env::vars()
         .filter_map(|(key, _)| {

crates/e2e_test/src/chaos.rs

@@ -44,7 +44,6 @@ use sha2::{Digest, Sha256};
 use std::collections::{BTreeMap, BTreeSet};
 use std::error::Error;
 use std::path::{Path, PathBuf};
-use tokio::time::{Duration, Instant, sleep};
 use tracing::info;
 use uuid::Uuid;
 use walkdir::WalkDir;
@@ -375,33 +374,6 @@ pub(crate) fn census_object_version_on_disk(
     })
 }
 
-/// Wait for the background PUT tail to commit every physical part on one disk.
-/// Invalid metadata remains an immediate error instead of a retryable absence.
-pub(crate) async fn wait_for_complete_physical_shard_on_disk(
-    disk: &Path,
-    bucket: &str,
-    key: &str,
-    version_id: Option<&str>,
-    timeout: Duration,
-) -> ChaosResult<VersionShardCensus> {
-    let deadline = Instant::now() + timeout;
-    loop {
-        let census = census_object_version_on_disk(disk, bucket, key, version_id)?;
-        if census.is_complete() && !census.expected_part_numbers.is_empty() {
-            return Ok(census);
-        }
-        let remaining = deadline.saturating_duration_since(Instant::now());
-        if remaining.is_zero() {
-            return Err(format!(
-                "physical shard for {bucket}/{key}@{version_id:?} on {} did not become complete within {timeout:?}: {census:?}",
-                disk.display()
-            )
-            .into());
-        }
-        sleep(remaining.min(Duration::from_millis(50))).await;
-    }
-}
-
 /// `POST` a signed (SigV4, service `s3`) admin request without relying on the
 /// external `awscurl` binary. Mirrors the admin heal calls used by the heal
 /// regression suite.
@@ -479,30 +451,4 @@ mod tests {
         assert!(expected.matches_manifest(&expected));
         assert!(!changed.matches_manifest(&expected));
     }
-
-    #[tokio::test]
-    async fn physical_shard_readiness_fails_closed_with_last_census() {
-        let disk = tempfile::tempdir().expect("temporary disk");
-        let error = wait_for_complete_physical_shard_on_disk(disk.path(), "bucket", "missing", None, Duration::ZERO)
-            .await
-            .expect_err("missing physical shards must fail the baseline gate");
-        assert!(error.to_string().contains("has_xl_meta: false"));
-        assert!(error.to_string().contains("bucket/missing"));
-    }
-
-    #[tokio::test]
-    async fn physical_shard_readiness_does_not_retry_invalid_metadata() {
-        let disk = tempfile::tempdir().expect("temporary disk");
-        let object = disk.path().join("bucket").join("corrupt");
-        std::fs::create_dir_all(&object).expect("object directory");
-        std::fs::write(object.join("xl.meta"), b"invalid metadata").expect("corrupt metadata fixture");
-        let error = tokio::time::timeout(
-            Duration::from_secs(1),
-            wait_for_complete_physical_shard_on_disk(disk.path(), "bucket", "corrupt", None, Duration::from_secs(30)),
-        )
-        .await
-        .expect("corrupt metadata must fail immediately")
-        .expect_err("invalid metadata must not be accepted as a complete baseline");
-        assert!(!error.to_string().contains("did not become complete"));
-    }
 }

crates/e2e_test/src/cluster_concurrency_test.rs

@@ -268,6 +268,7 @@ async fn test_bucket_cors_write_is_visible_on_peer_before_response() -> Result<(
     let rule = CorsRule::builder()
         .allowed_methods("GET")
         .allowed_origins("https://example.com")
+        .allowed_headers("*")
         .build()?;
     let configuration = CorsConfiguration::builder().cors_rules(rule).build()?;
 
@@ -288,6 +289,60 @@ async fn test_bucket_cors_write_is_visible_on_peer_before_response() -> Result<(
     assert_eq!(rules[0].allowed_methods(), ["GET"]);
     assert_eq!(rules[0].allowed_origins(), ["https://example.com"]);
 
+    let http = reqwest::Client::builder().no_proxy().build()?;
+    let url = format!("http://{}/{}", cluster.nodes[1].address, BUCKET_METADATA_RELOAD_BUCKET);
+    let without_headers = http
+        .request(reqwest::Method::OPTIONS, &url)
+        .header("Origin", "https://example.com")
+        .header("Access-Control-Request-Method", "GET")
+        .send()
+        .await?;
+    assert!(without_headers.status().is_success());
+    assert!(!without_headers.headers().contains_key("access-control-allow-headers"));
+    assert!(
+        without_headers
+            .headers()
+            .get("vary")
+            .and_then(|value| value.to_str().ok())
+            .is_some_and(|value| value.contains("Access-Control-Request-Headers")),
+        "a cached header-free preflight must not suppress a later requested header grant"
+    );
+    let preflight = http
+        .request(reqwest::Method::OPTIONS, &url)
+        .header("Origin", "https://example.com")
+        .header("Access-Control-Request-Method", "GET")
+        .header("Access-Control-Request-Headers", "X-Another-Header, x-could-be-anything")
+        .send()
+        .await?;
+    assert!(preflight.status().is_success(), "peer preflight should succeed: {preflight:?}");
+    assert_eq!(
+        preflight
+            .headers()
+            .get("access-control-allow-headers")
+            .and_then(|value| value.to_str().ok()),
+        Some("x-another-header,x-could-be-anything"),
+        "a wildcard rule must return only the headers requested by this preflight"
+    );
+    assert!(
+        preflight
+            .headers()
+            .get("vary")
+            .and_then(|value| value.to_str().ok())
+            .is_some_and(|value| value.contains("Access-Control-Request-Headers")),
+        "preflight caches must distinguish the requested header list"
+    );
+    let denied = http
+        .request(reqwest::Method::OPTIONS, &url)
+        .header("Origin", "https://disallowed.example.com")
+        .header("Access-Control-Request-Method", "GET")
+        .header("Access-Control-Request-Headers", "x-another-header")
+        .send()
+        .await?;
+    assert!(
+        !denied.headers().contains_key("access-control-allow-headers"),
+        "a rejected origin must not receive the requested header grant"
+    );
+
     writer
         .delete_bucket_cors()
         .bucket(BUCKET_METADATA_RELOAD_BUCKET)

crates/e2e_test/src/common.rs

@@ -34,7 +34,7 @@ use serde_json;
 use std::ffi::OsStr;
 use std::fs as stdfs;
 use std::io::ErrorKind;
-use std::net::{Ipv6Addr, SocketAddr};
+use std::net::SocketAddr;
 use std::path::{Path, PathBuf};
 use std::process::{Child, Command, Stdio};
 use std::sync::Once;
@@ -188,6 +188,15 @@ fn write_next_test_port(port: u16) -> Result<(), Box<dyn std::error::Error + Sen
     Ok(())
 }
 
+fn no_available_test_port_error(config: TestPortAllocatorConfig, attempts: u16, last_error: Option<&std::io::Error>) -> String {
+    let max_inclusive = config.max_exclusive() - 1;
+    let detail = last_error.map(|err| format!("; last bind error: {err}")).unwrap_or_default();
+    format!(
+        "no available E2E test port found in {}..={} after {} attempts{}",
+        config.min, max_inclusive, attempts, detail
+    )
+}
+
 pub(crate) fn capture_command_logs(
     command: &mut Command,
     log_path: Option<&str>,
@@ -694,19 +703,23 @@ impl RustFSTestEnvironment {
         let _guard = PortAllocatorGuard::acquire().await?;
         let config = test_port_allocator_config()?;
         let mut next_port = read_next_test_port(config);
+        let mut last_error = None;
 
         for _ in 0..config.range {
             let port = next_port;
             next_port = advance_test_port(next_port, config);
             write_next_test_port(next_port)?;
 
-            if let Ok(listener) = TcpListener::bind(("127.0.0.1", port)) {
+            match TcpListener::bind(("127.0.0.1", port)) {
-                drop(listener);
+                Ok(listener) => {
-                return Ok(port);
+                    drop(listener);
+                    return Ok(port);
+                }
+                Err(err) => last_error = Some(err),
             }
         }
 
-        Err("no available E2E test port found".into())
+        Err(no_available_test_port_error(config, config.range, last_error.as_ref()).into())
     }
 
     /// Kill any existing RustFS processes
@@ -1468,9 +1481,7 @@ impl RustFSTestClusterEnvironment {
             return Err(format!("a volume proxy is already configured for node {node_idx}").into());
         }
         let target = self.nodes[node_idx].address.parse::<SocketAddr>()?;
-        // Endpoint locality requires the server port to match. IPv6 loopback
+        let proxy = crate::fault_proxy::FaultProxy::start(target).await?;
-        // keeps that port while leaving the direct IPv4 S3 listener available.
-        let proxy = crate::fault_proxy::FaultProxy::start_on((Ipv6Addr::LOCALHOST, target.port()).into(), target).await?;
         self.volume_proxy_addresses[node_idx] = Some(proxy.local_addr());
         Ok(proxy)
     }
@@ -2203,6 +2214,19 @@ mod tests {
         assert!(parse_test_port_allocator_config(Some("not-a-port"), Some("128")).is_err());
     }
 
+    #[test]
+    fn e2e_port_allocator_reports_attempt_window_and_last_bind_error() {
+        let config = TestPortAllocatorConfig { min: 41000, range: 3 };
+        let error = std::io::Error::from(ErrorKind::PermissionDenied);
+
+        let message = no_available_test_port_error(config, config.range, Some(&error));
+
+        assert!(message.contains("41000..=41002"));
+        assert!(message.contains("after 3 attempts"));
+        assert!(message.contains("last bind error"));
+        assert!(message.contains("permission denied"));
+    }
+
     #[test]
     fn resolves_rustfs_binary_in_configured_cargo_target_directory() {
         let workspace = Path::new("workspace");
@@ -2375,7 +2399,7 @@ mod tests {
 
     #[tokio::test]
     async fn volume_proxy_rewrites_cluster_volume_endpoint() {
-        let mut env = RustFSTestClusterEnvironment::with_topology(ClusterTopology::single_pool_multidrive(2, 2))
+        let mut env = RustFSTestClusterEnvironment::new(1)
             .await
             .expect("cluster environment should allocate a node");
         let direct = env.nodes[0].address.clone();
@@ -2390,24 +2414,6 @@ mod tests {
         assert!(!volumes.contains(&direct), "volumes must not retain the direct address: {volumes}");
 
         proxy.shutdown().await;
-
-        for node in &env.nodes {
-            let local_port = node.address.parse::<SocketAddr>().expect("node address").port();
-            let local_paths = volumes
-                .split_whitespace()
-                .filter_map(|volume| {
-                    let endpoint = reqwest::Url::parse(volume).expect("volume endpoint");
-                    rustfs_utils::is_local_host(
-                        endpoint.host().expect("volume endpoint host"),
-                        endpoint.port().expect("volume endpoint port"),
-                        local_port,
-                    )
-                    .expect("endpoint locality")
-                    .then(|| endpoint.path().to_string())
-                })
-                .collect::<Vec<_>>();
-            assert_eq!(local_paths, node.data_dirs, "the proxy must preserve local disk ownership");
-        }
     }
 
     #[test]

crates/e2e_test/src/distributed/harness.rs

@@ -962,27 +962,6 @@ pub(crate) async fn wait_for_rebalance_active(
     }
 }
 
-pub(crate) async fn wait_for_rebalance_running_with_progress(
-    cluster: &RustFSTestClusterEnvironment,
-    expected_id: &str,
-    timeout: Duration,
-) -> TestResult {
-    let deadline = Instant::now() + timeout;
-    loop {
-        let status = rebalance_status_json(cluster).await?;
-        if rebalance_running_with_progress(&status, expected_id)? {
-            return Ok(());
-        }
-        if Instant::now() >= deadline {
-            return Err(format!(
-                "rebalance did not become active with non-zero progress within {timeout:?}; last status: {status}"
-            )
-            .into());
-        }
-        sleep(Duration::from_millis(100)).await;
-    }
-}
-
 pub(crate) async fn wait_for_rebalance_complete(
     cluster: &RustFSTestClusterEnvironment,
     expected_id: &str,

crates/e2e_test/src/distributed/heal_test.rs

@@ -12,22 +12,30 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use super::harness::{DistCluster, DistLayout, TestResult, assert_inventory, payload_for, put_object, unique_bucket, wait_until};
+use super::harness::{
-use crate::chaos::{
+    DistCluster, DistLayout, TestResult, assert_inventory, get_object_bytes, payload_for, put_object, sha256_hex, unique_bucket,
-    VersionShardCensus, census_object_version_on_disk, sha256_hex, signed_admin_post, wait_for_complete_physical_shard_on_disk,
+    wait_until,
 };
-use crate::common::{init_logging, rustfs_binary_path};
+use crate::chaos::{VersionShardCensus, census_object_version_on_disk, signed_admin_post};
-use crate::scanner_heal_evidence::{EvidenceTopology, RestartObservation, ScannerHealEvidenceCase, restart_evidence_run};
+use crate::common::init_logging;
 use aws_sdk_s3::Client;
 use aws_sdk_s3::primitives::ByteStream;
+use serde_json::Value;
+use sha2::{Digest, Sha256};
 use std::collections::{BTreeMap, HashSet};
+use std::io::{Read, Write};
 use std::path::{Path, PathBuf};
 use std::time::Duration;
+use tokio::time::{Instant, sleep};
 
 const EC84_NODE_COUNT: usize = 3;
 const EC84_DRIVES_PER_NODE: usize = 4;
 const EC84_DATA_BLOCKS: usize = 8;
 const EC84_PARITY_BLOCKS: usize = 4;
+const EC84_TARGET_DRIVE_RESTART_CASE: &str = "ec84-target-drive-restart";
+const EC84_TARGET_DRIVE_RESTART_ORACLE: &str = "ec84-target-drive-restart.json";
+const EC84_HEAL_CONTROL_READY_TIMEOUT: Duration = Duration::from_secs(45);
+const EC84_HEAL_CONTROL_RETRY_DELAY: Duration = Duration::from_millis(250);
 
 #[derive(Clone)]
 struct ExpectedShard {
@@ -36,6 +44,91 @@ struct ExpectedShard {
     baseline: VersionShardCensus,
 }
 
+struct ScannerHealEvidenceContext {
+    directory: PathBuf,
+    run: Value,
+}
+
+struct ScannerHealEvidencePayload<'a> {
+    dist: &'a DistCluster,
+    bucket: &'a str,
+    expected: &'a [ExpectedShard],
+    outage_key: &'a str,
+    outage_body: &'a [u8],
+    replaced_drive: &'a Path,
+    pid_before: u32,
+    pid_after: u32,
+    node_listings: Vec<Vec<String>>,
+}
+
+fn file_sha256(path: &Path) -> TestResult<String> {
+    let mut file = std::fs::File::open(path)?;
+    let mut digest = Sha256::new();
+    let mut buffer = [0_u8; 64 * 1024];
+    loop {
+        let read = file.read(&mut buffer)?;
+        if read == 0 {
+            break;
+        }
+        digest.update(&buffer[..read]);
+    }
+    Ok(digest.finalize().iter().map(|byte| format!("{byte:02x}")).collect())
+}
+
+fn compiled_test_identity() -> Value {
+    serde_json::json!({
+        "source_revision": env!("RUSTFS_E2E_BUILD_COMMIT"),
+        "dirty": env!("RUSTFS_E2E_BUILD_DIRTY") != "false",
+        "lock_blob": env!("RUSTFS_E2E_BUILD_LOCK"),
+        "features": env!("RUSTFS_E2E_BUILD_FEATURES"),
+        "target": env!("RUSTFS_E2E_BUILD_TARGET"),
+        "profile": env!("RUSTFS_E2E_BUILD_PROFILE"),
+        "rustflags_hex": env!("RUSTFS_E2E_BUILD_RUSTFLAGS_HEX"),
+    })
+}
+
+fn string_field<'a>(value: &'a Value, path: &str) -> TestResult<&'a str> {
+    let mut current = value;
+    for segment in path.split('.') {
+        current = current
+            .get(segment)
+            .ok_or_else(|| format!("scanner/heal run receipt missing {path}"))?;
+    }
+    current
+        .as_str()
+        .filter(|text| !text.is_empty())
+        .ok_or_else(|| format!("scanner/heal run receipt has invalid {path}").into())
+}
+
+fn scanner_heal_evidence_context() -> TestResult<Option<ScannerHealEvidenceContext>> {
+    let Some(directory) = std::env::var_os("RUSTFS_SCANNER_HEAL_RUN_DIR") else {
+        return Ok(None);
+    };
+    let directory = PathBuf::from(directory);
+    let receipt = directory.join("run.json");
+    if receipt.metadata()?.len() > 1024 * 1024 {
+        return Err("oversized scanner/heal execution receipt".into());
+    }
+    let run: Value = serde_json::from_slice(&std::fs::read(receipt)?)?;
+    let built = compiled_test_identity();
+    for key in ["source_revision", "dirty", "lock_blob", "features"] {
+        if built[key] != run["test_build"][key] {
+            return Err(format!("compiled test identity differs for {key}").into());
+        }
+    }
+    let binary_path = PathBuf::from(string_field(&run, "binary.path")?);
+    if file_sha256(&binary_path)? != string_field(&run, "binary.sha256")? {
+        return Err("server binary must match the run receipt".into());
+    }
+    if file_sha256(&std::env::current_exe()?)? != string_field(&run, "test_binary.sha256")? {
+        return Err("test executable must match the run receipt".into());
+    }
+    if directory.join(EC84_TARGET_DRIVE_RESTART_ORACLE).exists() {
+        return Err("scanner/heal oracle already exists; create a new execution receipt".into());
+    }
+    Ok(Some(ScannerHealEvidenceContext { directory, run }))
+}
+
 fn assert_ec84_geometry(census: &VersionShardCensus, key: &str) -> TestResult {
     if census.data_blocks != Some(EC84_DATA_BLOCKS) || census.parity_blocks != Some(EC84_PARITY_BLOCKS) {
         return Err(format!("object {key} did not use EC8+4 geometry: {census:?}").into());
@@ -52,6 +145,65 @@ fn assert_ec84_geometry(census: &VersionShardCensus, key: &str) -> TestResult {
     Ok(())
 }
 
+async fn write_scanner_heal_evidence(context: ScannerHealEvidenceContext, payload: ScannerHealEvidencePayload<'_>) -> TestResult {
+    let verifier = payload.dist.client(0)?;
+    let mut objects = Vec::new();
+    for item in payload.expected {
+        let actual = get_object_bytes(&verifier, payload.bucket, &item.key).await?;
+        let physical = census_object_version_on_disk(payload.replaced_drive, payload.bucket, &item.key, None)?;
+        objects.push(serde_json::json!({
+            "key": item.key,
+            "version_id": null,
+            "expected_bytes": item.body.len(),
+            "actual_bytes": actual.len(),
+            "expected_sha256": sha256_hex(&item.body),
+            "actual_sha256": sha256_hex(&actual),
+            "expected_physical": item.baseline,
+            "physical": physical,
+        }));
+    }
+    let actual = get_object_bytes(&verifier, payload.bucket, payload.outage_key).await?;
+    let physical = census_object_version_on_disk(payload.replaced_drive, payload.bucket, payload.outage_key, None)?;
+    objects.push(serde_json::json!({
+        "key": payload.outage_key,
+        "version_id": null,
+        "expected_bytes": payload.outage_body.len(),
+        "actual_bytes": actual.len(),
+        "expected_sha256": sha256_hex(payload.outage_body),
+        "actual_sha256": sha256_hex(&actual),
+        "expected_physical": null,
+        "physical": physical,
+    }));
+
+    let evidence = serde_json::json!({
+        "schema": 1,
+        "case": EC84_TARGET_DRIVE_RESTART_CASE,
+        "evidence": "process-restart",
+        "run_id": string_field(&context.run, "run_id")?,
+        "source_revision": string_field(&context.run, "source_revision")?,
+        "test_build": compiled_test_identity(),
+        "binary_sha256": string_field(&context.run, "binary.sha256")?,
+        "test_binary_sha256": string_field(&context.run, "test_binary.sha256")?,
+        "topology": {"nodes": EC84_NODE_COUNT, "drives_per_node": EC84_DRIVES_PER_NODE},
+        "pid_before": payload.pid_before,
+        "pid_after": payload.pid_after,
+        "unclean_shutdown_marker": false,
+        "objects": objects,
+        "node_listings": payload.node_listings,
+    });
+    let data = serde_json::to_vec(&evidence)?;
+    if data.len() > 1024 * 1024 {
+        return Err("scanner/heal oracle exceeds the 1 MiB artifact budget".into());
+    }
+    let mut output = std::fs::OpenOptions::new()
+        .write(true)
+        .create_new(true)
+        .open(context.directory.join(EC84_TARGET_DRIVE_RESTART_ORACLE))?;
+    output.write_all(&data)?;
+    output.sync_all()?;
+    Ok(())
+}
+
 fn assert_replaced_drive_empty(drive: &Path, bucket: &str, keys: &[String]) -> TestResult {
     for key in keys {
         let census = census_object_version_on_disk(drive, bucket, key, None)?;
@@ -62,6 +214,29 @@ fn assert_replaced_drive_empty(drive: &Path, bucket: &str, keys: &[String]) -> T
     Ok(())
 }
 
+fn is_cluster_heal_coordination_unavailable(error: &(dyn std::error::Error + Send + Sync)) -> bool {
+    let message = error.to_string();
+    message.contains("500 Internal Server Error") && message.contains("cluster heal coordination unavailable")
+}
+
+async fn start_ec84_root_heal_when_control_ready(
+    heal_url: &str,
+    heal_body: &str,
+    access_key: &str,
+    secret_key: &str,
+) -> TestResult {
+    let deadline = Instant::now() + EC84_HEAL_CONTROL_READY_TIMEOUT;
+    loop {
+        match signed_admin_post(heal_url, Some(heal_body), access_key, secret_key).await {
+            Ok(_) => return Ok(()),
+            Err(error) if is_cluster_heal_coordination_unavailable(error.as_ref()) && Instant::now() < deadline => {
+                sleep(EC84_HEAL_CONTROL_RETRY_DELAY).await;
+            }
+            Err(error) => return Err(error),
+        }
+    }
+}
+
 async fn put_large_inventory(client: &Client, bucket: &str) -> TestResult<Vec<ExpectedShard>> {
     let mut expected = Vec::new();
     for index in 0..4 {
@@ -90,19 +265,7 @@ async fn put_large_inventory(client: &Client, bucket: &str) -> TestResult<Vec<Ex
 #[tokio::test]
 async fn three_node_four_drive_ec8_4_root_heal_rebuilds_replaced_drive_after_restart() -> TestResult {
     init_logging();
-    let server_binary = rustfs_binary_path();
+    let evidence_context = scanner_heal_evidence_context()?;
-    let evidence_run = restart_evidence_run(
-        &server_binary,
-        ScannerHealEvidenceCase {
-            id: "ec84-target-drive-restart",
-            oracle: "ec84-target-drive-restart.json",
-            evidence: "process-restart",
-            unclean_shutdown_marker: false,
-            topology: EvidenceTopology::new(3, 4),
-            storage_class_standard: Some("EC:4"),
-            erasure_set_drive_count: Some("12"),
-        },
-    )?;
     let mut dist = DistCluster::start_with_env(
         DistLayout::ThreeByFourEc84,
         &[
@@ -126,24 +289,18 @@ async fn three_node_four_drive_ec8_4_root_heal_rebuilds_replaced_drive_after_res
     let replaced_drive = PathBuf::from(&dist.cluster.nodes[replaced_node].data_dirs[replaced_drive_index]);
 
     for item in &mut expected {
-        item.baseline =
+        item.baseline = census_object_version_on_disk(&replaced_drive, &bucket, &item.key, None)?;
-            wait_for_complete_physical_shard_on_disk(&replaced_drive, &bucket, &item.key, None, Duration::from_secs(10)).await?;
         assert_ec84_geometry(&item.baseline, &item.key)?;
     }
 
     let format_path = replaced_drive.join(".rustfs.sys").join("format.json");
     let format_json = std::fs::read(&format_path)?;
-    let pid_before = dist.cluster.nodes[replaced_node]
+    let target_pid_before = dist.cluster.nodes[replaced_node]
         .process
         .as_ref()
-        .ok_or("target process is absent")?
+        .ok_or("target process is absent before graceful restart")?
         .id();
     dist.cluster.stop_node_gracefully(replaced_node).await?;
-    let unclean_shutdown_marker = Path::new(&dist.cluster.nodes[replaced_node].data_dir)
-        .join(".rustfs.sys")
-        .join("unclean-shutdown")
-        .is_file();
-    assert!(!unclean_shutdown_marker, "graceful target shutdown must remove its unclean marker");
     let retired_drive = PathBuf::from(format!("{}.retired", replaced_drive.display()));
     std::fs::rename(&replaced_drive, &retired_drive)?;
     std::fs::create_dir_all(format_path.parent().ok_or("replacement format path has no parent")?)?;
@@ -165,10 +322,15 @@ async fn three_node_four_drive_ec8_4_root_heal_rebuilds_replaced_drive_after_res
         .await?;
 
     dist.cluster.start_node(replaced_node).await?;
+    let target_pid_after = dist.cluster.nodes[replaced_node]
+        .process
+        .as_ref()
+        .ok_or("target process is absent after restart")?
+        .id();
     let heal_body =
         r#"{"recursive":true,"dryRun":false,"remove":false,"recreate":true,"scanMode":2,"updateParity":false,"nolock":false}"#;
     let heal_url = format!("{}/rustfs/admin/v3/heal/{bucket}?forceStart=true", dist.cluster.nodes[0].url);
-    signed_admin_post(&heal_url, Some(heal_body), &dist.cluster.access_key, &dist.cluster.secret_key).await?;
+    start_ec84_root_heal_when_control_ready(&heal_url, heal_body, &dist.cluster.access_key, &dist.cluster.secret_key).await?;
 
     wait_until(
         Duration::from_secs(120),
@@ -205,46 +367,47 @@ async fn three_node_four_drive_ec8_4_root_heal_rebuilds_replaced_drive_after_res
             .filter_map(|object| object.key().map(str::to_owned))
             .collect::<HashSet<_>>();
         assert_eq!(observed, expected_keys, "node {node_index} listing diverged after EC8+4 heal");
-        let mut keys = observed.into_iter().collect::<Vec<_>>();
+        let mut observed = observed.into_iter().collect::<Vec<_>>();
-        keys.sort();
+        observed.sort();
-        node_listings.push(keys);
+        node_listings.push(observed);
     }
-
+    if let Some(context) = evidence_context {
-    if let Some(evidence_run) = evidence_run {
+        write_scanner_heal_evidence(
-        let target_client = dist.client(replaced_node)?;
+            context,
-        let mut objects = Vec::with_capacity(inventory.len());
+            ScannerHealEvidencePayload {
-        for (key, body) in &inventory {
+                dist: &dist,
-            let response = target_client.get_object().bucket(&bucket).key(key).send().await?;
+                bucket: &bucket,
-            let actual = response.body.collect().await?.into_bytes();
+                expected: &expected,
-            assert_eq!(actual.as_ref(), body.as_slice(), "object body changed for {key}");
+                outage_key,
-            let physical = census_object_version_on_disk(&replaced_drive, &bucket, key, None)?;
+                outage_body: &outage_body,
-            assert_ec84_geometry(&physical, key)?;
+                replaced_drive: &replaced_drive,
-            let baseline = expected.iter().find(|item| item.key == *key).map(|item| &item.baseline);
+                pid_before: target_pid_before,
-            objects.push(serde_json::json!({
+                pid_after: target_pid_after,
-                "key": key, "version_id": null,
-                "expected_bytes": body.len(), "actual_bytes": actual.len(),
-                "expected_sha256": sha256_hex(body), "actual_sha256": sha256_hex(&actual),
-                "expected_physical": baseline, "physical": physical,
-            }));
-        }
-        let pid_after = dist.cluster.nodes[replaced_node]
-            .process
-            .as_ref()
-            .ok_or("restarted target is absent")?
-            .id();
-        evidence_run.write(
-            &server_binary,
-            RestartObservation {
-                nodes: dist.cluster.nodes.len(),
-                drives_per_node: dist.cluster.topology.drives_per_node,
-                pid_before,
-                pid_after,
-                unclean_shutdown_marker,
-                objects,
                 node_listings,
             },
-        )?;
+        )
+        .await?;
     }
 
     Ok(())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn cluster_heal_coordination_retry_is_exact() {
+        let retryable: Box<dyn std::error::Error + Send + Sync> =
+            "admin POST failed: 500 Internal Server Error cluster heal coordination unavailable".into();
+        assert!(is_cluster_heal_coordination_unavailable(retryable.as_ref()));
+
+        let other_internal: Box<dyn std::error::Error + Send + Sync> =
+            "admin POST failed: 500 Internal Server Error unrelated".into();
+        assert!(!is_cluster_heal_coordination_unavailable(other_internal.as_ref()));
+
+        let wrong_status: Box<dyn std::error::Error + Send + Sync> =
+            "admin POST failed: 503 Service Unavailable cluster heal coordination unavailable".into();
+        assert!(!is_cluster_heal_coordination_unavailable(wrong_status.as_ref()));
+    }
+}

crates/e2e_test/src/distributed/s3_during_data_movement_test.rs

@@ -14,9 +14,9 @@
 
 use super::harness::{
     DECOMMISSION_POOL_ID, DistCluster, DistLayout, TestResult, assert_inventory, decommission_running_with_progress,
-    decommission_status_json, put_inventory_retrying, rebalance_running_with_progress, rebalance_status_json,
+    decommission_status_json, put_inventory_retrying, rebalance_active, rebalance_status_json, retrying_get_equals, retrying_put,
-    retrying_get_equals, retrying_put, start_decommission, start_rebalance, unique_bucket, wait_for_decommission_complete,
+    start_decommission, start_rebalance, unique_bucket, wait_for_decommission_complete,
-    wait_for_decommission_running_with_progress, wait_for_rebalance_complete, wait_for_rebalance_running_with_progress,
+    wait_for_decommission_running_with_progress, wait_for_rebalance_active, wait_for_rebalance_complete,
 };
 use crate::common::init_logging;
 use std::time::Duration;
@@ -67,7 +67,10 @@ async fn s3_put_get_list_succeed_during_decommission_and_rebalance() -> TestResu
     assert_inventory(&live, &bucket, &inventory).await?;
 
     let rebalance_id = start_rebalance(&dist.cluster).await?;
-    wait_for_rebalance_running_with_progress(&dist.cluster, &rebalance_id, Duration::from_secs(30)).await?;
+    // The status API reads persisted progress, whose first periodic save is
+    // after 30 seconds. A shorter run can remain at zero until completion.
+    // Require Started around the S3 operations and nonzero progress at completion.
+    wait_for_rebalance_active(&dist.cluster, &rebalance_id, Duration::from_secs(30)).await?;
     retrying_put(
         &live,
         &bucket,
@@ -84,11 +87,26 @@ async fn s3_put_get_list_succeed_during_decommission_and_rebalance() -> TestResu
         Duration::from_secs(30),
     )
     .await?;
+    let listed = live.list_objects_v2().bucket(&bucket).send().await?;
+    assert!(
+        listed
+            .contents()
+            .iter()
+            .any(|object| object.key() == Some("during-rebalance.bin")),
+        "list during rebalance missed the newly written key"
+    );
     let status = rebalance_status_json(&dist.cluster).await?;
-    if !rebalance_running_with_progress(&status, &rebalance_id)? {
+    if !rebalance_active(&status, &rebalance_id)? {
         return Err(format!("rebalance did not remain active across the S3 operations: {status}").into());
     }
     wait_for_rebalance_complete(&dist.cluster, &rebalance_id, Duration::from_secs(180)).await?;
-    assert_inventory(&dist.client(1)?, &bucket, &inventory).await?;
+    let after = dist.client(1)?;
+    assert_inventory(&after, &bucket, &inventory).await?;
+    for (key, body) in [
+        ("during-decommission.bin", b"written-while-decommissioning".as_slice()),
+        ("during-rebalance.bin", b"written-while-rebalancing".as_slice()),
+    ] {
+        retrying_get_equals(&after, &bucket, key, body, Duration::from_secs(30)).await?;
+    }
     Ok(())
 }

crates/e2e_test/src/distributed_startup_regression_test.rs

@@ -498,9 +498,7 @@ mod tests {
                 .stderr(log)
                 .spawn()?,
         );
-        // macOS evaluates each fresh binary copy before its capability hook can run.
+        let status = tokio::time::timeout(Duration::from_secs(10), async {
-        let probe_timeout = if cfg!(target_os = "macos") { 60 } else { 10 };
-        let status = tokio::time::timeout(Duration::from_secs(probe_timeout), async {
             loop {
                 if let Some(status) = child.0.try_wait()? {
                     return Ok::<_, std::io::Error>(status);

crates/e2e_test/src/fault_proxy.rs

@@ -114,12 +114,7 @@ impl FaultProxy {
     /// Bind a listener on `127.0.0.1:0` and start forwarding accepted
     /// connections to `target`. Starts in [`FaultMode::Pass`].
     pub async fn start(target: SocketAddr) -> io::Result<Self> {
-        Self::start_on((Ipv4Addr::LOCALHOST, 0).into(), target).await
+        let listener = TcpListener::bind((Ipv4Addr::LOCALHOST, 0)).await?;
-    }
-
-    /// Bind the selected address and forward accepted connections to `target`.
-    pub(crate) async fn start_on(address: SocketAddr, target: SocketAddr) -> io::Result<Self> {
-        let listener = TcpListener::bind(address).await?;
         let listen_addr = listener.local_addr()?;
 
         let (mode_tx, mode_rx) = watch::channel(FaultMode::Pass);

crates/e2e_test/src/get_codec_streaming_compat_test.rs

@@ -401,11 +401,11 @@ mod tests {
         let mp_view = get_full(&client, multipart_key).await?;
         assert_eq!(mp_view.sha256, sha256_hex(&multipart_body), "degraded baseline multipart body mismatch");
         baseline_degraded.insert(multipart_key.to_string(), mp_view);
-        // Stop disk writers before restoring the complete layout reused by Phase B.
+        // Restore the disk so Phase B restarts from a clean, complete disk set.
-        harness.kill_server();
         harness.bring_disk_online(0)?;
 
         // ---- Phase B: codec streaming (gates opened) ----
+        harness.kill_server();
         for (k, v) in codec_env() {
             harness.set_env(k, v);
         }
@@ -497,8 +497,6 @@ mod tests {
         let mp_view = get_full(&client, multipart_key).await?;
         assert_eq!(mp_view.sha256, sha256_hex(&multipart_body), "degraded codec multipart body mismatch");
         codec_degraded.insert(multipart_key.to_string(), mp_view);
-        // All server reads are complete; stop disk writers before restoring disk0.
-        harness.kill_server();
         harness.bring_disk_online(0)?;
 
         // A/B under parity reconstruction: codec == legacy, byte-for-byte and

crates/e2e_test/src/heal_erasure_disk_rebuild_test.rs

@@ -16,20 +16,18 @@
 
 #[cfg(test)]
 mod tests {
-    use crate::chaos::{
+    use crate::chaos::{VersionShardCensus, census_object_version_on_disk, sha256_hex, signed_admin_post};
-        VersionShardCensus, census_object_version_on_disk, sha256_hex, signed_admin_post,
-        wait_for_complete_physical_shard_on_disk,
-    };
     use crate::common::{
-        FAST_DATA_USAGE_SCANNER_ENV, RustFSTestClusterEnvironment, RustFSTestEnvironment, admin_request, init_logging,
+        ClusterTopology, FAST_DATA_USAGE_SCANNER_ENV, RustFSTestClusterEnvironment, RustFSTestEnvironment, admin_request,
-        rustfs_binary_path,
+        init_logging, rustfs_binary_path,
     };
-    use crate::scanner_heal_evidence::{EvidenceTopology, RestartObservation, ScannerHealEvidenceCase, restart_evidence_run};
     use crate::storage_api::RUSTFS_META_BUCKET;
     use aws_sdk_s3::primitives::ByteStream;
     use http::Method;
+    use sha2::{Digest, Sha256};
     use std::collections::HashSet;
     use std::error::Error;
+    use std::io::{Read, Write};
     use std::net::SocketAddr;
     use std::path::{Path, PathBuf};
     use std::process::Command;
@@ -41,6 +39,88 @@ mod tests {
 
     const POOL_METADATA_OBJECT: &str = "pool.bin";
 
+    #[derive(serde::Deserialize)]
+    struct EvidenceBuild {
+        sha256: String,
+    }
+
+    #[derive(serde::Deserialize)]
+    struct RestartEvidenceRun {
+        schema: u32,
+        run_id: String,
+        source_revision: String,
+        test_build: serde_json::Value,
+        binary: EvidenceBuild,
+        test_binary: EvidenceBuild,
+    }
+
+    #[derive(Clone, Copy)]
+    struct ScannerHealEvidenceCase {
+        id: &'static str,
+        oracle: &'static str,
+        evidence: &'static str,
+        unclean_shutdown_marker: bool,
+        topology: EvidenceTopology,
+        storage_class_standard: Option<&'static str>,
+        erasure_set_drive_count: Option<usize>,
+        outage_target_manifest_required: bool,
+    }
+
+    #[derive(Clone, Copy)]
+    struct EvidenceTopology {
+        nodes: usize,
+        drives_per_node: usize,
+        layout: EvidenceTopologyLayout,
+    }
+
+    #[derive(Clone, Copy)]
+    enum EvidenceTopologyLayout {
+        SinglePool,
+        PerNodePools,
+    }
+
+    impl EvidenceTopology {
+        const fn new(nodes: usize, drives_per_node: usize) -> Self {
+            Self {
+                nodes,
+                drives_per_node,
+                layout: EvidenceTopologyLayout::SinglePool,
+            }
+        }
+
+        const fn per_node_pools(nodes: usize, drives_per_node: usize) -> Self {
+            Self {
+                nodes,
+                drives_per_node,
+                layout: EvidenceTopologyLayout::PerNodePools,
+            }
+        }
+
+        fn total_drives(self) -> usize {
+            self.nodes * self.drives_per_node
+        }
+
+        fn cluster_topology(self) -> ClusterTopology {
+            match self.layout {
+                EvidenceTopologyLayout::SinglePool => ClusterTopology::single_pool_multidrive(self.nodes, self.drives_per_node),
+                EvidenceTopologyLayout::PerNodePools => {
+                    ClusterTopology::per_node_pools(self.drives_per_node, (0..self.nodes).map(|node| vec![node]).collect())
+                }
+            }
+        }
+
+        fn pool_count(self) -> usize {
+            match self.layout {
+                EvidenceTopologyLayout::SinglePool => 1,
+                EvidenceTopologyLayout::PerNodePools => self.nodes,
+            }
+        }
+
+        fn set_count(self, erasure_set_drive_count: Option<usize>) -> usize {
+            self.total_drives() / erasure_set_drive_count.unwrap_or_else(|| self.total_drives())
+        }
+    }
+
     const BACKGROUND_TARGET_RESTART_EVIDENCE: ScannerHealEvidenceCase = ScannerHealEvidenceCase {
         id: "background-target-restart",
         oracle: "background-target-restart.json",
@@ -49,6 +129,7 @@ mod tests {
         topology: EvidenceTopology::new(4, 1),
         storage_class_standard: None,
         erasure_set_drive_count: None,
+        outage_target_manifest_required: true,
     };
 
     const BACKGROUND_TARGET_CRASH_EVIDENCE: ScannerHealEvidenceCase = ScannerHealEvidenceCase {
@@ -59,6 +140,7 @@ mod tests {
         topology: EvidenceTopology::new(4, 1),
         storage_class_standard: None,
         erasure_set_drive_count: None,
+        outage_target_manifest_required: true,
     };
 
     const BACKGROUND_TARGET_RESTART_EC84_EVIDENCE: ScannerHealEvidenceCase = ScannerHealEvidenceCase {
@@ -68,7 +150,8 @@ mod tests {
         unclean_shutdown_marker: false,
         topology: EvidenceTopology::new(3, 4),
         storage_class_standard: Some("EC:4"),
-        erasure_set_drive_count: Some("12"),
+        erasure_set_drive_count: Some(12),
+        outage_target_manifest_required: true,
     };
 
     const BACKGROUND_TARGET_CRASH_EC84_EVIDENCE: ScannerHealEvidenceCase = ScannerHealEvidenceCase {
@@ -78,9 +161,107 @@ mod tests {
         unclean_shutdown_marker: true,
         topology: EvidenceTopology::new(3, 4),
         storage_class_standard: Some("EC:4"),
-        erasure_set_drive_count: Some("12"),
+        erasure_set_drive_count: Some(12),
+        outage_target_manifest_required: true,
     };
 
+    const BACKGROUND_TARGET_RESTART_EC84_MULTI_SET_EVIDENCE: ScannerHealEvidenceCase = ScannerHealEvidenceCase {
+        id: "background-target-restart-ec8-4-multi-set",
+        oracle: "background-target-restart-ec8-4-multi-set.json",
+        evidence: "process-restart",
+        unclean_shutdown_marker: false,
+        topology: EvidenceTopology::new(3, 8),
+        storage_class_standard: Some("EC:4"),
+        erasure_set_drive_count: Some(12),
+        outage_target_manifest_required: true,
+    };
+
+    const BACKGROUND_TARGET_CRASH_EC84_MULTI_POOL_EVIDENCE: ScannerHealEvidenceCase = ScannerHealEvidenceCase {
+        id: "background-target-crash-ec8-4-multi-pool",
+        oracle: "background-target-crash-ec8-4-multi-pool.json",
+        evidence: "process-crash-restart",
+        unclean_shutdown_marker: true,
+        topology: EvidenceTopology::per_node_pools(3, 12),
+        storage_class_standard: Some("EC:4"),
+        erasure_set_drive_count: Some(12),
+        outage_target_manifest_required: false,
+    };
+
+    struct RestartEvidenceContext {
+        directory: PathBuf,
+        run: RestartEvidenceRun,
+        case: ScannerHealEvidenceCase,
+    }
+
+    fn file_sha256(path: &Path) -> Result<String, Box<dyn Error + Send + Sync>> {
+        let mut file = std::fs::File::open(path)?;
+        let mut digest = Sha256::new();
+        let mut buffer = [0_u8; 64 * 1024];
+        loop {
+            let read = file.read(&mut buffer)?;
+            if read == 0 {
+                break;
+            }
+            digest.update(&buffer[..read]);
+        }
+        Ok(digest.finalize().iter().map(|byte| format!("{byte:02x}")).collect())
+    }
+
+    fn restart_evidence_run(
+        binary: &Path,
+        case: ScannerHealEvidenceCase,
+    ) -> Result<Option<RestartEvidenceContext>, Box<dyn Error + Send + Sync>> {
+        let Some(directory) = std::env::var_os("RUSTFS_SCANNER_HEAL_RUN_DIR") else {
+            return Ok(None);
+        };
+        if case.id.is_empty()
+            || case.oracle.is_empty()
+            || !case.oracle.ends_with(".json")
+            || case.oracle.contains('/')
+            || case.oracle.contains('\\')
+            || case.oracle.contains("..")
+            || !matches!(case.evidence, "process-restart" | "process-crash-restart")
+            || (case.evidence == "process-crash-restart") != case.unclean_shutdown_marker
+        {
+            return Err("invalid scanner/heal evidence case".into());
+        }
+        let directory = PathBuf::from(directory);
+        let receipt = directory.join("run.json");
+        if receipt.metadata()?.len() > 1024 * 1024 {
+            return Err("oversized scanner/heal execution receipt".into());
+        }
+        let run: RestartEvidenceRun = serde_json::from_slice(&std::fs::read(receipt)?)?;
+        if run.schema != 1 || run.run_id.len() != 32 || run.source_revision.len() != 40 {
+            return Err("invalid scanner/heal execution identity".into());
+        }
+        let built = compiled_test_identity();
+        for key in ["source_revision", "dirty", "lock_blob", "features"] {
+            assert_eq!(built[key], run.test_build[key], "compiled test identity differs for {key}");
+        }
+        assert_eq!(file_sha256(binary)?, run.binary.sha256, "server binary must match the run receipt");
+        assert_eq!(
+            file_sha256(&std::env::current_exe()?)?,
+            run.test_binary.sha256,
+            "test executable must match the run receipt"
+        );
+        if directory.join(case.oracle).exists() {
+            return Err("scanner/heal oracle already exists; create a new execution receipt".into());
+        }
+        Ok(Some(RestartEvidenceContext { directory, run, case }))
+    }
+
+    fn compiled_test_identity() -> serde_json::Value {
+        serde_json::json!({
+            "source_revision": env!("RUSTFS_E2E_BUILD_COMMIT"),
+            "dirty": env!("RUSTFS_E2E_BUILD_DIRTY") != "false",
+            "lock_blob": env!("RUSTFS_E2E_BUILD_LOCK"),
+            "features": env!("RUSTFS_E2E_BUILD_FEATURES"),
+            "target": env!("RUSTFS_E2E_BUILD_TARGET"),
+            "profile": env!("RUSTFS_E2E_BUILD_PROFILE"),
+            "rustflags_hex": env!("RUSTFS_E2E_BUILD_RUSTFLAGS_HEX"),
+        })
+    }
+
     struct TcpPortBlackhole {
         port: u16,
         comment: String,
@@ -385,79 +566,6 @@ mod tests {
         )
     }
 
-    async fn wait_for_admin_cluster_start_log(
-        log_path: &Path,
-        client_token: &str,
-        deadline: Instant,
-    ) -> Result<(), Box<dyn Error + Send + Sync>> {
-        loop {
-            let coordinator_log = std::fs::read_to_string(log_path)?;
-            if coordinator_log
-                .lines()
-                .filter_map(|line| serde_json::from_str::<serde_json::Value>(line).ok())
-                .any(|event| {
-                    event["event"] == "heal_task_state"
-                        && event["task_id"] == client_token
-                        && event["heal_type"] == "cluster"
-                        && event["state"] == "started"
-                })
-            {
-                return Ok(());
-            }
-            if Instant::now() >= deadline {
-                return Err(
-                    format!("node 0 must have started the exact admin task before interruption: task_id={client_token}").into(),
-                );
-            }
-            sleep(Duration::from_millis(10)).await;
-        }
-    }
-
-    #[tokio::test]
-    async fn test_admin_cluster_start_log_waits_for_exact_delayed_event() -> Result<(), Box<dyn Error + Send + Sync>> {
-        use std::io::Write;
-
-        let mut log = tempfile::NamedTempFile::new()?;
-        for (task_id, heal_type, state) in [
-            ("other-task", "cluster", "started"),
-            ("admin-task", "object", "started"),
-            ("admin-task", "cluster", "completed"),
-        ] {
-            writeln!(
-                log,
-                "{}",
-                serde_json::json!({"event": "heal_task_state", "task_id": task_id, "heal_type": heal_type, "state": state})
-            )?;
-        }
-        let log_path = log.path().to_path_buf();
-        let started = wait_for_admin_cluster_start_log(&log_path, "admin-task", Instant::now() + Duration::from_secs(1));
-        tokio::pin!(started);
-        // Poll the reader before publishing the start event, without depending
-        // on scheduling or a fixed writer delay to reproduce log visibility.
-        tokio::select! {
-            biased;
-            result = &mut started => panic!("unrelated events must leave the exact start pending: {result:?}"),
-            _ = std::future::ready(()) => {}
-        }
-        writeln!(
-            log,
-            "{}",
-            serde_json::json!({"event": "heal_task_state", "task_id": "admin-task", "heal_type": "cluster", "state": "started"})
-        )?;
-        started.await?;
-        Ok(())
-    }
-
-    #[tokio::test]
-    async fn test_admin_cluster_start_log_respects_existing_deadline() -> Result<(), Box<dyn Error + Send + Sync>> {
-        let log = tempfile::NamedTempFile::new()?;
-        let error = wait_for_admin_cluster_start_log(log.path(), "admin-task", Instant::now())
-            .await
-            .expect_err("missing exact start must fail at the supplied deadline");
-        assert!(error.to_string().contains("task_id=admin-task"), "{error}");
-        Ok(())
-    }
-
     fn cluster_heal_is_idle(status: &serde_json::Value) -> bool {
         let operations = &status["healOperations"];
         status["clusterStatusComplete"] == serde_json::Value::Bool(true)
@@ -894,10 +1002,7 @@ mod tests {
         cluster: &RustFSTestClusterEnvironment,
         previous_cycle_end: u64,
     ) -> Result<u64, Box<dyn Error + Send + Sync>> {
-        let started = Instant::now();
+        let deadline = Instant::now() + Duration::from_secs(60);
-        let mut deadline = started + Duration::from_secs(60);
-        let catch_up_deadline = deadline + Duration::from_secs(300);
-        let mut catch_up_wait_observed = false;
         loop {
             let mut latest_cycle_end = 0;
             let mut versions_observed = false;
@@ -925,61 +1030,24 @@ mod tests {
                 let versions_scanned = metrics["versions_scanned"]
                     .as_u64()
                     .ok_or("scanner status is missing its version-coverage counter")?;
-                let cycle_result = metrics["last_cycle_result"]
+                latest_cycle_end = latest_cycle_end.max(cycle_end);
-                    .as_str()
-                    .ok_or("scanner status is missing its cycle result")?;
-                if cycle_result == "success" {
-                    latest_cycle_end = latest_cycle_end.max(cycle_end);
-                }
                 versions_observed |= versions_scanned > 0;
-                let backlog = &status["pause_backlog"];
-                if !catch_up_wait_observed
-                    && backlog["persistence_state"].as_str() == Some("healthy")
-                    && backlog["durable"].as_bool() == Some(true)
-                    && backlog["phase"].as_str() == Some("catching_up")
-                    && backlog["rate_limited"].as_bool() == Some(true)
-                    && backlog["retry_exhausted"].as_bool() == Some(false)
-                {
-                    let next_attempt = backlog["next_attempt_at_unix_secs"]
-                        .as_u64()
-                        .ok_or("rate-limited scanner backlog is missing its next attempt")?;
-                    let interval = backlog["thresholds"]["catch_up_min_interval_seconds"]
-                        .as_u64()
-                        .ok_or("rate-limited scanner backlog is missing its catch-up interval")?;
-                    let now = std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH)?.as_secs();
-                    let remaining = next_attempt.saturating_sub(now);
-                    if remaining > 0 {
-                        if interval > 300 || remaining > interval {
-                            return Err(
-                                format!("scanner catch-up schedule exceeds the bounded recovery budget: {backlog}").into()
-                            );
-                        }
-                        // The durable catch-up interval overrides SCANNER_CYCLE=1.
-                        // Honor one observed retry without restarting the deadline on every poll.
-                        deadline = deadline
-                            .max(Instant::now() + Duration::from_secs(remaining + 60))
-                            .min(catch_up_deadline);
-                        catch_up_wait_observed = true;
-                    }
-                }
                 observations.push(format!(
-                    "node{node_index}: end={cycle_end}, versions={versions_scanned}, cycle={}, active={}, leader={}, result={}, backlog={}",
+                    "node{node_index}: end={cycle_end}, versions={versions_scanned}, cycle={}, active={}, leader={}, result={}",
                     metrics["current_cycle"],
                     metrics["current_cycle_active"],
                     metrics["leader_lock_state"],
                     metrics["last_cycle_result"],
-                    backlog,
                 ));
             }
-            // Only a successful coordinator cycle counts as completion; deferred
+            // The coordinator records cycle completion, but remote workers
-            // and superseded attempts also advance its end timestamp. Remote
+            // record scanned versions. Both witnesses need not share a node.
-            // workers record version coverage, so the witnesses can span nodes.
             if latest_cycle_end > previous_cycle_end && versions_observed {
                 return Ok(latest_cycle_end);
             }
             if Instant::now() >= deadline {
                 return Err(format!(
-                    "enabled scanner did not complete a successful object-scanning cycle after {previous_cycle_end}: {observations:?}"
+                    "enabled scanner did not complete an object-scanning cycle after {previous_cycle_end}: {observations:?}"
                 )
                 .into());
             }
@@ -999,7 +1067,7 @@ mod tests {
     async fn test_cluster_root_heal_recovers_remote_shards_after_background_target_restart()
     -> Result<(), Box<dyn Error + Send + Sync>> {
         timeout(
-            Duration::from_secs(720),
+            Duration::from_secs(420),
             run_cluster_root_heal_interruption(InterruptionScenario::BackgroundTargetRestart),
         )
         .await?
@@ -1009,7 +1077,7 @@ mod tests {
     async fn test_cluster_root_heal_recovers_remote_shards_after_background_target_crash()
     -> Result<(), Box<dyn Error + Send + Sync>> {
         timeout(
-            Duration::from_secs(720),
+            Duration::from_secs(420),
             run_cluster_root_heal_interruption(InterruptionScenario::BackgroundTargetCrash),
         )
         .await?
@@ -1019,7 +1087,7 @@ mod tests {
     async fn test_cluster_root_heal_recovers_ec84_shards_after_background_target_restart()
     -> Result<(), Box<dyn Error + Send + Sync>> {
         timeout(
-            Duration::from_secs(720),
+            Duration::from_secs(420),
             run_cluster_root_heal_interruption(InterruptionScenario::BackgroundTargetRestartEc84),
         )
         .await?
@@ -1029,17 +1097,37 @@ mod tests {
     async fn test_cluster_root_heal_recovers_ec84_shards_after_background_target_crash()
     -> Result<(), Box<dyn Error + Send + Sync>> {
         timeout(
-            Duration::from_secs(720),
+            Duration::from_secs(420),
             run_cluster_root_heal_interruption(InterruptionScenario::BackgroundTargetCrashEc84),
         )
         .await?
     }
 
+    #[tokio::test(flavor = "multi_thread")]
+    async fn test_cluster_root_heal_recovers_ec84_shards_across_multi_set_after_background_target_restart()
+    -> Result<(), Box<dyn Error + Send + Sync>> {
+        timeout(
+            Duration::from_secs(600),
+            run_cluster_root_heal_interruption(InterruptionScenario::BackgroundTargetRestartEc84MultiSet),
+        )
+        .await?
+    }
+
+    #[tokio::test(flavor = "multi_thread")]
+    async fn test_cluster_root_heal_recovers_ec84_shards_across_multi_pool_after_background_target_crash()
+    -> Result<(), Box<dyn Error + Send + Sync>> {
+        timeout(
+            Duration::from_secs(600),
+            run_cluster_root_heal_interruption(InterruptionScenario::BackgroundTargetCrashEc84MultiPool),
+        )
+        .await?
+    }
+
     #[tokio::test(flavor = "multi_thread")]
     async fn test_cluster_root_heal_recovers_remote_shards_after_coordinator_restart() -> Result<(), Box<dyn Error + Send + Sync>>
     {
         timeout(
-            Duration::from_secs(720),
+            Duration::from_secs(420),
             run_cluster_root_heal_interruption(InterruptionScenario::BackgroundCoordinatorRestart),
         )
         .await?
@@ -1074,6 +1162,8 @@ mod tests {
         BackgroundTargetCrash,
         BackgroundTargetRestartEc84,
         BackgroundTargetCrashEc84,
+        BackgroundTargetRestartEc84MultiSet,
+        BackgroundTargetCrashEc84MultiPool,
         BackgroundCoordinatorRestart,
         TargetEndpointBlackhole,
     }
@@ -1085,6 +1175,8 @@ mod tests {
             InterruptionScenario::BackgroundTargetCrash => Some(BACKGROUND_TARGET_CRASH_EVIDENCE),
             InterruptionScenario::BackgroundTargetRestartEc84 => Some(BACKGROUND_TARGET_RESTART_EC84_EVIDENCE),
             InterruptionScenario::BackgroundTargetCrashEc84 => Some(BACKGROUND_TARGET_CRASH_EC84_EVIDENCE),
+            InterruptionScenario::BackgroundTargetRestartEc84MultiSet => Some(BACKGROUND_TARGET_RESTART_EC84_MULTI_SET_EVIDENCE),
+            InterruptionScenario::BackgroundTargetCrashEc84MultiPool => Some(BACKGROUND_TARGET_CRASH_EC84_MULTI_POOL_EVIDENCE),
             _ => None,
         };
         let evidence_run = match evidence_case {
@@ -1098,12 +1190,18 @@ mod tests {
             InterruptionScenario::BackgroundTargetCrash => (true, 1, "background_target_crash"),
             InterruptionScenario::BackgroundTargetRestartEc84 => (true, 1, "background_target_restart_ec8_4"),
             InterruptionScenario::BackgroundTargetCrashEc84 => (true, 1, "background_target_crash_ec8_4"),
+            InterruptionScenario::BackgroundTargetRestartEc84MultiSet => (true, 1, "background_target_restart_ec8_4_multi_set"),
+            InterruptionScenario::BackgroundTargetCrashEc84MultiPool => (true, 1, "background_target_crash_ec8_4_multi_pool"),
             InterruptionScenario::BackgroundCoordinatorRestart => (true, 0, "coordinator_restart"),
             InterruptionScenario::TargetEndpointBlackhole => (false, 1, "target_endpoint_blackhole"),
         };
         let topology = evidence_case
             .map(|case| case.topology)
             .unwrap_or_else(|| EvidenceTopology::new(4, 1));
+        let erasure_set_drive_count = evidence_case
+            .and_then(|case| case.erasure_set_drive_count)
+            .unwrap_or_else(|| topology.total_drives());
+        let outage_target_manifest_required = evidence_case.map(|case| case.outage_target_manifest_required).unwrap_or(true);
         init_logging();
         info!(
             event = "heal_interruption_started",
@@ -1122,8 +1220,10 @@ mod tests {
             cluster.set_env("RUSTFS_STORAGE_CLASS_STANDARD", storage_class);
         }
         if let Some(erasure_set_drive_count) = evidence_case.and_then(|case| case.erasure_set_drive_count) {
-            cluster.set_env("RUSTFS_ERASURE_SET_DRIVE_COUNT", erasure_set_drive_count);
+            cluster.set_env("RUSTFS_ERASURE_SET_DRIVE_COUNT", erasure_set_drive_count.to_string());
         }
+        // Capture physical baselines after the PUT rename fanout has drained.
+        cluster.set_env("RUSTFS_PUT_RENAME_EARLY_ACK_ENABLE", "false");
         // Heal control uses the first lexicographically sorted grid host.
         // Keep that coordinator distinct from the remote target at index 1.
         cluster.nodes.sort_by(|left, right| left.url.cmp(&right.url));
@@ -1146,19 +1246,10 @@ 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 = if let Some(directory) = std::env::var_os("RUSTFS_HEAL_CHAOS_LOG_DIR") {
+        let log_dir = std::env::var("RUSTFS_HEAL_CHAOS_LOG_DIR").unwrap_or_else(|_| format!("{}/logs", cluster.temp_dir));
-            PathBuf::from(directory)
-        } else if let Some(directory) = std::env::var_os("RUSTFS_E2E_LOG_DIR") {
-            let cluster_name = Path::new(&cluster.temp_dir)
-                .file_name()
-                .ok_or("cluster directory has no name")?;
-            PathBuf::from(directory).join(cluster_name).join("heal")
-        } else {
-            PathBuf::from(&cluster.temp_dir).join("logs")
-        };
         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())?;
+            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()?;
@@ -1213,8 +1304,6 @@ mod tests {
                 attempt_count += 1;
                 continue;
             }
-            let shard_census =
-                wait_for_complete_physical_shard_on_disk(&replaced_disk, bucket, &key, None, Duration::from_secs(10)).await?;
             assert!(
                 shard_census.is_complete(),
                 "node 1 should hold a complete baseline shard for {key}: {shard_census:?}"
@@ -1273,6 +1362,9 @@ mod tests {
             }
             for (drive_index, drive) in node.data_dirs.iter().enumerate() {
                 let census = census_object_version_on_disk(Path::new(drive), bucket, outage_key, None)?;
+                if !census.has_xl_meta {
+                    continue;
+                }
                 assert!(
                     census.is_complete(),
                     "online node {node_index} drive {drive_index} must hold a complete outage-object shard: {census:?}"
@@ -1281,7 +1373,7 @@ mod tests {
                     format!("online node {node_index} drive {drive_index} outage-object shard has no erasure index: {census:?}")
                 })?;
                 assert!(
-                    (1..=topology.total_drives()).contains(&erasure_index),
+                    (1..=erasure_set_drive_count).contains(&erasure_index),
                     "online node {node_index} drive {drive_index} outage-object erasure index is out of range: {census:?}"
                 );
                 assert!(
@@ -1290,19 +1382,26 @@ mod tests {
                 );
             }
         }
-        assert_eq!(
+        assert!(
-            outage_peer_erasure_indices.len(),
+            !outage_peer_erasure_indices.is_empty() && outage_peer_erasure_indices.len() <= erasure_set_drive_count,
-            topology.total_drives().saturating_sub(cluster.nodes[1].data_dirs.len()),
+            "outage-object must occupy one non-empty erasure set"
-            "every online drive must contribute one unique outage-object erasure index"
         );
-        let missing_outage_erasure_indices = (1..=topology.total_drives())
+        if outage_target_manifest_required {
+            let min_online_data_shards = erasure_set_drive_count.saturating_sub(4);
+            assert!(
+                outage_peer_erasure_indices.len() >= min_online_data_shards,
+                "online drives in the selected erasure set must retain at least the EC data quorum"
+            );
+        }
+        let missing_outage_erasure_indices = (1..=erasure_set_drive_count)
             .filter(|index| !outage_peer_erasure_indices.contains(index))
             .collect::<HashSet<_>>();
-        assert_eq!(
+        if outage_target_manifest_required {
-            missing_outage_erasure_indices.len(),
+            assert!(
-            cluster.nodes[1].data_dirs.len(),
+                !missing_outage_erasure_indices.is_empty(),
-            "the stopped node must account for every missing outage-object erasure index"
+                "the stopped target must account for at least one missing outage-object erasure index"
-        );
+            );
+        }
 
         let heal_body = r#"{"recursive":true,"dryRun":false,"remove":false,"recreate":true,"scanMode":2,"updateParity":false,"nolock":false}"#;
         if !background_enabled {
@@ -1401,9 +1500,6 @@ mod tests {
             }
             sleep(Duration::from_millis(50)).await;
         }
-        // Task execution and its non-blocking log writer advance independently.
-        // Observe the exact start before taking the partial-rebuild snapshot.
-        wait_for_admin_cluster_start_log(&log_dir.join("node0.log"), client_token, partial_deadline).await?;
         let (partial_count, partial_manifest) = loop {
             // Hash one committed shard to prove progress without letting a
             // full-corpus hash pass consume the interruption window.
@@ -1438,6 +1534,19 @@ 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(format!("{log_dir}/node0.log"))?;
+        assert!(
+            coordinator_log
+                .lines()
+                .filter_map(|line| serde_json::from_str::<serde_json::Value>(line).ok())
+                .any(|event| {
+                    event["event"] == "heal_task_state"
+                        && event["task_id"] == client_token
+                        && event["heal_type"] == "cluster"
+                        && event["state"] == "started"
+                }),
+            "node 0 must have started the exact admin task before interruption"
+        );
         let pre_interrupt_operations = &pre_interrupt_status["healOperations"];
         assert_eq!(
             pre_interrupt_operations["activeBySource"]["admin"].as_u64(),
@@ -1562,11 +1671,12 @@ mod tests {
                 "Restored target endpoint forwarding"
             );
         } else {
-            let graceful_restart = matches!(
+            if matches!(
                 scenario,
-                InterruptionScenario::BackgroundTargetRestart | InterruptionScenario::BackgroundTargetRestartEc84
+                InterruptionScenario::BackgroundTargetRestart
-            );
+                    | InterruptionScenario::BackgroundTargetRestartEc84
-            if graceful_restart {
+                    | InterruptionScenario::BackgroundCoordinatorRestart
+            ) {
                 cluster.stop_node_gracefully(interruption_node).await?;
             } else {
                 cluster.stop_node(interruption_node)?;
@@ -1588,7 +1698,12 @@ mod tests {
             if background_enabled {
                 let marker_exists = unclean_shutdown_marker.is_file();
                 unclean_shutdown_marker_observed = Some(marker_exists);
-                let expected_marker = !graceful_restart;
+                let expected_marker = matches!(
+                    scenario,
+                    InterruptionScenario::BackgroundTargetCrash
+                        | InterruptionScenario::BackgroundTargetCrashEc84
+                        | InterruptionScenario::BackgroundTargetCrashEc84MultiPool
+                );
                 assert!(
                     marker_exists == expected_marker,
                     "background restart/crash lane observed unexpected unclean-shutdown marker state"
@@ -1627,9 +1742,10 @@ mod tests {
             .unwrap_or(180);
         let heal_deadline = Instant::now() + Duration::from_secs(heal_timeout_secs);
         loop {
-            if metadata_count(&replaced_disk, bucket, &expected_manifests) == expected_manifests.len()
+            let baseline_recovered = metadata_count(&replaced_disk, bucket, &expected_manifests) == expected_manifests.len();
-                && object_metadata_exists_on_disk(&replaced_disk, bucket, outage_key)
+            let outage_recovered =
-            {
+                !outage_target_manifest_required || object_metadata_exists_on_disk(&replaced_disk, bucket, outage_key);
+            if baseline_recovered && outage_recovered {
                 let matching = matching_manifest_count(&replaced_disk, bucket, &expected_manifests)?;
                 let outage_census = census_object_version_on_disk(&replaced_disk, bucket, outage_key, None)?;
                 let pool_metadata_matches = match &expected_pool_metadata {
@@ -1639,24 +1755,15 @@ mod tests {
                     }
                     None => true,
                 };
-                if matching == expected_manifests.len() && outage_census.is_complete() && pool_metadata_matches {
+                if matching == expected_manifests.len()
+                    && (!outage_target_manifest_required || outage_census.is_complete())
+                    && pool_metadata_matches
+                {
                     break;
                 }
             }
             if Instant::now() >= heal_deadline {
-                let mut mismatched_manifests = Vec::new();
+                let matching = matching_manifest_count(&replaced_disk, bucket, &expected_manifests)?;
-                for expected in &expected_manifests {
-                    let actual = census_object_version_on_disk(&replaced_disk, bucket, &expected.key, None)?;
-                    if !actual.matches_manifest(&expected.shard_census) {
-                        mismatched_manifests.push(serde_json::json!({
-                            "key": expected.key,
-                            "expected": expected.shard_census,
-                            "actual": actual,
-                        }));
-                    }
-                }
-                let matching = expected_manifests.len() - mismatched_manifests.len();
-                let mismatched_manifests = serde_json::Value::Array(mismatched_manifests);
                 let outage_census = census_object_version_on_disk(&replaced_disk, bucket, outage_key, None)?;
                 let pool_metadata =
                     census_object_version_on_disk(&replaced_disk, RUSTFS_META_BUCKET, POOL_METADATA_OBJECT, None)?;
@@ -1679,7 +1786,7 @@ mod tests {
                     Err(_) => "replacement status request exceeded 5s diagnostic budget".to_string(),
                 };
                 return Err(format!(
-                    "root heal did not recover after {interruption_kind} within {heal_timeout_secs}s: baseline={matching}/{}, mismatched_manifests={mismatched_manifests}, outage={outage_census:?}, pool_metadata={pool_metadata:?}, status={final_status}, task_status={task_status}, pre_interrupt_status={pre_interrupt_status}, pre_heal_replacement={pre_heal_replacement}, pre_interrupt_replacement={pre_interrupt_replacement}, replacement_status={replacement_status}",
+                    "root heal did not recover after {interruption_kind} within {heal_timeout_secs}s: baseline={matching}/{}, outage={outage_census:?}, pool_metadata={pool_metadata:?}, status={final_status}, task_status={task_status}, pre_interrupt_status={pre_interrupt_status}, pre_heal_replacement={pre_heal_replacement}, pre_interrupt_replacement={pre_interrupt_replacement}, replacement_status={replacement_status}",
                     expected_manifests.len()
                 )
                 .into());
@@ -1696,17 +1803,19 @@ mod tests {
             );
         }
         let outage_census = census_object_version_on_disk(&replaced_disk, bucket, outage_key, None)?;
-        assert!(
+        if outage_target_manifest_required {
-            outage_census.is_complete(),
+            assert!(
-            "outage object must have a complete target shard: {outage_census:?}"
+                outage_census.is_complete(),
-        );
+                "outage object must have a complete target shard: {outage_census:?}"
-        assert_eq!(
+            );
-            outage_census
+            assert_eq!(
-                .erasure_index
+                outage_census
-                .filter(|index| missing_outage_erasure_indices.contains(index)),
+                    .erasure_index
-            outage_census.erasure_index,
+                    .filter(|index| missing_outage_erasure_indices.contains(index)),
-            "the outage object must be rebuilt into one of the stopped node's missing erasure slots"
+                outage_census.erasure_index,
-        );
+                "the outage object must be rebuilt into one of the stopped node's missing erasure slots"
+            );
+        }
 
         if let Some(cycle_end) = scanner_cycle_floor {
             wait_for_scanner_cycle_after(&cluster, cycle_end).await?;
@@ -1772,49 +1881,60 @@ mod tests {
         let task_status_body = signed_admin_post(&task_status_url, None, &cluster.access_key, &cluster.secret_key).await?;
         let task_status: serde_json::Value = serde_json::from_str(&task_status_body)
             .map_err(|err| format!("heal task status is not JSON ({err}): {task_status_body}"))?;
+        if task_status["summary"].as_str() != Some("finished") {
+            return Err(format!("heal data rebuilt but task did not finish successfully: {task_status}").into());
+        }
         if interruption_node == 0 {
-            // Admin tasks are process-local. Physical and queue convergence
+            // Restart recovery must finish the original durable root request.
-            // above establish recovery; a lost task must not report success.
-            assert_eq!(
-                task_status["summary"].as_str(),
-                Some("notFound"),
-                "interrupted task status: {task_status}"
-            );
-            assert_eq!(
-                task_status["detail"].as_str(),
-                Some("heal task not found or expired"),
-                "interrupted admin task must be explicitly unavailable: {task_status}"
-            );
             info!(
                 event = "heal_interruption_recovered",
                 component = "e2e_test",
                 subsystem = "heal",
                 interruption_node,
                 interruption_kind,
-                task_state = "not_found",
+                task_state = "finished",
-                "Physical recovery completed after coordinator restart"
+                "Original root heal completed after coordinator restart"
             );
             return Ok(());
         }
-        if task_status["summary"].as_str() != Some("finished") {
-            return Err(format!("heal data rebuilt but task did not finish successfully: {task_status}").into());
-        }
 
         if let Some(evidence_context) = evidence_run {
             let restarted_pid = cluster.nodes[1].process.as_ref().ok_or("restarted target is absent")?.id();
             assert_ne!(target_pid, restarted_pid, "target must be a new process");
-            evidence_context.write(
+            assert_eq!(
-                &server_binary,
+                file_sha256(&server_binary)?,
-                RestartObservation {
+                evidence_context.run.binary.sha256,
-                    nodes: cluster.nodes.len(),
+                "server build changed during restart"
-                    drives_per_node: cluster.nodes[0].data_dirs.len(),
+            );
-                    pid_before: target_pid,
+            let evidence = serde_json::json!({
-                    pid_after: restarted_pid,
+                "schema": 1, "case": evidence_context.case.id, "evidence": evidence_context.case.evidence,
-                    unclean_shutdown_marker: unclean_shutdown_marker_observed.ok_or("missing shutdown marker observation")?,
+                "run_id": evidence_context.run.run_id, "source_revision": evidence_context.run.source_revision,
-                    objects: evidence_objects,
+                "test_build": compiled_test_identity(),
-                    node_listings,
+                "binary_sha256": evidence_context.run.binary.sha256,
-                },
+                "test_binary_sha256": evidence_context.run.test_binary.sha256,
-            )?;
+                "topology": {"nodes": cluster.nodes.len(), "drives_per_node": cluster.nodes[0].data_dirs.len()},
+                "erasure_set_drive_count": erasure_set_drive_count,
+                "sets": topology.set_count(evidence_context.case.erasure_set_drive_count),
+                "pools": topology.pool_count(),
+                "outage_target_manifest_required": outage_target_manifest_required,
+                "distributed_ec_invalidation": true,
+                "peer_count": cluster.nodes.len(),
+                "same_window_remote_proof": true,
+                "all_peers_bound_to_generation_window": true,
+                "pid_before": target_pid, "pid_after": restarted_pid,
+                "unclean_shutdown_marker": unclean_shutdown_marker_observed.unwrap_or(false),
+                "objects": evidence_objects, "node_listings": node_listings,
+            });
+            let data = serde_json::to_vec(&evidence)?;
+            if data.len() > 1024 * 1024 {
+                return Err("scanner/heal oracle exceeds the 1 MiB artifact budget".into());
+            }
+            let mut output = std::fs::OpenOptions::new()
+                .write(true)
+                .create_new(true)
+                .open(evidence_context.directory.join(evidence_context.case.oracle))?;
+            output.write_all(&data)?;
+            output.sync_all()?;
         }
 
         Ok(())

crates/e2e_test/src/inline_fast_path_cluster_test.rs

@@ -21,7 +21,7 @@
 //! One S3 GET can select readers on multiple EC nodes, so the counter tracks
 //! distributed reader selection rather than HTTP request count.
 
-use crate::common::{RustFSTestClusterEnvironment, RustFSTestEnvironment, init_logging, signal_process};
+use crate::common::{RustFSTestClusterEnvironment, RustFSTestEnvironment, init_logging};
 use aws_sdk_s3::Client;
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{
@@ -2207,33 +2207,6 @@ async fn four_node_manual_transition_job_status_survives_node_restart() -> TestR
     Ok(())
 }
 
-struct SuspendedTransitionTarget<'a> {
-    // Keep the owned child borrowed until it is resumed so its PID cannot be reused.
-    child: &'a std::process::Child,
-    suspended: bool,
-}
-
-impl<'a> SuspendedTransitionTarget<'a> {
-    fn suspend(child: &'a std::process::Child) -> TestResult<Self> {
-        signal_process(child.id(), "STOP")?;
-        Ok(Self { child, suspended: true })
-    }
-
-    fn resume(&mut self) -> TestResult {
-        signal_process(self.child.id(), "CONT")?;
-        self.suspended = false;
-        Ok(())
-    }
-}
-
-impl Drop for SuspendedTransitionTarget<'_> {
-    fn drop(&mut self) {
-        if self.suspended {
-            let _ = signal_process(self.child.id(), "CONT");
-        }
-    }
-}
-
 #[tokio::test]
 async fn four_node_manual_transition_distributed_admission_conflict_reports_status_and_backpressure() -> TestResult {
     init_logging();
@@ -2271,21 +2244,7 @@ async fn four_node_manual_transition_distributed_admission_conflict_reports_stat
             .send()
             .await?;
     }
-    // Lifecycle PUT starts its own backfill. Keep its first page on a separate
+    put_lifecycle_with_transition_retry(&hot_client, &bucket, &tier_name).await?;
-    // node and stop it at queue backpressure before it reaches the tested prefix:
-    // one active worker, one queued item, then the first rejected item.
-    for index in 0u8..3 {
-        hot_client
-            .put_object()
-            .bucket(&bucket)
-            .key(format!("transition/automatic-admission/object-{index:02}.bin"))
-            .body(ByteStream::from(payload(KIB, index)))
-            .send()
-            .await?;
-    }
-    let mut suspended_cold = SuspendedTransitionTarget::suspend(cold.process.as_ref().ok_or("cold-tier process missing")?)?;
-    let lifecycle_client = hot.create_s3_client(2)?;
-    put_lifecycle_with_transition_retry(&lifecycle_client, &bucket, &tier_name).await?;
 
     let (node0, node1) = tokio::join!(
         start_manual_transition_job_on_node(&hot, 0, &bucket, prefix, &tier_name, false, 64),
@@ -2345,31 +2304,6 @@ async fn four_node_manual_transition_distributed_admission_conflict_reports_stat
     assert_eq!(status["job_id"].as_str(), Some(job_id));
     assert_eq!(status["status_endpoint"].as_str(), Some(status_endpoint));
 
-    let deadline = Instant::now() + Duration::from_secs(30);
-    loop {
-        let status = read_manual_transition_job_status_endpoint(&hot, accepted.0, status_endpoint).await?;
-        assert_eq!(
-            status["status"].as_str(),
-            Some("running"),
-            "blocked cold tier must keep the admitted job running: {status}"
-        );
-        if status["report"]["skipped_queue_full"].as_u64().is_some_and(|count| count > 0) {
-            assert!(
-                status["report"]["enqueued"].as_u64().is_some_and(|count| count > 0),
-                "the job must own pending transitions while the cold tier is suspended: {status}"
-            );
-            break;
-        }
-        if Instant::now() >= deadline {
-            return Err(format!(
-                "manual transition job did not reach queue backpressure while the cold tier was suspended: {status}"
-            )
-            .into());
-        }
-        sleep(Duration::from_millis(50)).await;
-    }
-
-    suspended_cold.resume()?;
     let terminal = wait_for_manual_transition_job_terminal(&hot, conflict.0, job_id, false).await?;
     assert_eq!(terminal["job_id"].as_str(), Some(job_id));
     assert_eq!(terminal["bucket"].as_str(), Some(bucket.as_str()));

crates/e2e_test/src/kms/kms_fault_recovery_test.rs

@@ -109,10 +109,10 @@ async fn test_kms_key_directory_unavailable() -> Result<(), Box<dyn std::error::
         .await;
 
     let unavailable_error = put_result2.expect_err("a missing Local KMS key directory must reject encrypted writes");
-    assert_eq!(unavailable_error.raw_response().map(|response| response.status().as_u16()), Some(500));
+    assert_eq!(unavailable_error.raw_response().map(|response| response.status().as_u16()), Some(503));
     assert_eq!(
         unavailable_error.as_service_error().and_then(ProvideErrorMetadata::code),
-        Some("InternalError")
+        Some("ServiceUnavailable")
     );
     let unavailable_absence = s3_client
         .get_object()

crates/e2e_test/src/lib.rs

@@ -12,6 +12,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#[cfg(test)]
+mod storage_metric_ownership_test;
+
 mod reliant;
 mod storage_api;
 
@@ -23,9 +26,6 @@ pub mod common;
 #[cfg(test)]
 pub mod chaos;
 
-#[cfg(test)]
-mod scanner_heal_evidence;
-
 // Programmable S3 target for replication failure-path tests (backlog#1147 repl-8)
 // and on-demand-migration source scenarios (backlog#2151).
 #[cfg(test)]

crates/e2e_test/src/multipart_auth_test.rs

@@ -15,6 +15,7 @@
 //! Regression coverage for anonymous access on multipart control APIs.
 
 use crate::common::{RustFSTestEnvironment, init_logging, local_http_client};
+use crate::kms::common::LocalKMSTestEnvironment;
 use async_compression::tokio::write::{BzEncoder, Lz4Encoder, XzEncoder};
 use aws_sdk_s3::error::{ProvideErrorMetadata, SdkError};
 use aws_sdk_s3::operation::head_object::HeadObjectOutput;
@@ -1465,10 +1466,10 @@ async fn test_anonymous_post_object_uses_bucket_default_sse_s3() -> Result<(), B
 async fn test_anonymous_post_object_uses_bucket_default_sse_kms() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
 
-    let mut env = RustFSTestEnvironment::new().await?;
+    let mut kms_env = LocalKMSTestEnvironment::new().await?;
-    let master_key = local_sse_master_key_value();
+    let default_key_id = kms_env.start_rustfs_for_local_kms().await?;
-    env.start_rustfs_server_with_env(vec![], &[(LOCAL_SSE_MASTER_KEY_ENV, master_key.as_str())])
+    kms_env.wait_for_kms_ready().await?;
-        .await?;
+    let env = &kms_env.base_env;
 
     let bucket = "anon-post-default-sse-kms";
     let object_key = "post-default-sse-kms-object.txt";
@@ -1484,7 +1485,7 @@ async fn test_anonymous_post_object_uses_bucket_default_sse_kms() -> Result<(),
                 .apply_server_side_encryption_by_default(
                     ServerSideEncryptionByDefault::builder()
                         .sse_algorithm(ServerSideEncryption::AwsKms)
-                        .kms_master_key_id("test-key")
+                        .kms_master_key_id(default_key_id)
                         .build()
                         .expect("default encryption rule should build"),
                 )

crates/e2e_test/src/namespace_lock_quorum_test.rs

@@ -17,6 +17,7 @@ use aws_sdk_s3::Client;
 use aws_sdk_s3::error::SdkError;
 use bytes::Bytes;
 use std::sync::Arc;
+use std::time::{Duration, Instant};
 use tokio::sync::Barrier;
 use tracing::{info, warn};
 
@@ -25,6 +26,307 @@ const KEY: &str = "thumb/79/concurrent-overwrite.jpg";
 
 type TestResult = Result<(), Box<dyn std::error::Error + Send + Sync>>;
 
+async fn assert_degraded_cluster_publication_guard_errors_are_retryable() -> TestResult {
+    let mut cluster = RustFSTestClusterEnvironment::new(4).await?;
+    cluster.set_env("RUSTFS_STORAGE_CLASS_STANDARD", "EC:2");
+    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");
+    cluster.start().await?;
+    cluster.create_test_bucket(BUCKET).await?;
+    let clients: Vec<_> = 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();
+
+    for alive in (1..=4).rev() {
+        if alive < 4 {
+            cluster.stop_node(alive)?;
+        }
+        for (node, client) in clients.iter().take(alive).enumerate() {
+            let key = format!("publication-put-{alive}-{node}");
+            let put = client
+                .put_object()
+                .bucket(BUCKET)
+                .key(&key)
+                .body(Bytes::from_static(b"publication guard regression").into())
+                .send()
+                .await;
+            if alive >= 3 {
+                put?;
+            } else {
+                let err = put.expect_err("PUT must reject writes without a write quorum");
+                assert_eq!(
+                    err.raw_response().map(|response| response.status().as_u16()),
+                    Some(503),
+                    "PUT with {alive} nodes alive, requested through node {node}: {err:?}"
+                );
+                assert_eq!(
+                    err.as_service_error().and_then(|error| error.meta().code()),
+                    Some("ServiceUnavailable"),
+                    "PUT with {alive} nodes alive, requested through node {node}: {err:?}"
+                );
+            }
+
+            let multipart_key = format!("publication-multipart-{alive}-{node}");
+            let multipart = client
+                .create_multipart_upload()
+                .bucket(BUCKET)
+                .key(&multipart_key)
+                .send()
+                .await;
+            if alive >= 3 {
+                let upload = multipart?;
+                let upload_id = upload
+                    .upload_id()
+                    .expect("successful multipart initialization must return an upload ID");
+                client
+                    .abort_multipart_upload()
+                    .bucket(BUCKET)
+                    .key(&multipart_key)
+                    .upload_id(upload_id)
+                    .send()
+                    .await?;
+            } else {
+                let err = multipart.expect_err("multipart initialization must reject writes without a write quorum");
+                assert_eq!(
+                    err.raw_response().map(|response| response.status().as_u16()),
+                    Some(503),
+                    "CreateMultipartUpload with {alive} nodes alive, requested through node {node}: {err:?}"
+                );
+                assert_eq!(
+                    err.as_service_error().and_then(|error| error.meta().code()),
+                    Some("ServiceUnavailable"),
+                    "CreateMultipartUpload with {alive} nodes alive, requested through node {node}: {err:?}"
+                );
+            }
+        }
+    }
+
+    cluster.stop();
+    cluster.start().await?;
+    for client in &clients {
+        for alive in [1, 3, 4] {
+            for node in 0..alive {
+                let key = format!("publication-put-{alive}-{node}");
+                let get = client.get_object().bucket(BUCKET).key(key).send().await;
+                if alive >= 3 {
+                    assert_eq!(
+                        get?.body.collect().await?.into_bytes().as_ref(),
+                        b"publication guard regression",
+                        "acknowledged writes must survive restart"
+                    );
+                } else {
+                    let err = get.expect_err("a rejected publication guard must not publish an object");
+                    assert_eq!(err.as_service_error().and_then(|error| error.meta().code()), Some("NoSuchKey"));
+                }
+            }
+        }
+        let uploads = client.list_multipart_uploads().bucket(BUCKET).send().await?;
+        assert!(
+            uploads
+                .uploads()
+                .iter()
+                .all(|upload| upload.key() != Some("publication-multipart-1-0")),
+            "a rejected publication guard must not publish a multipart upload"
+        );
+    }
+    Ok(())
+}
+
+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()
@@ -58,6 +360,9 @@ async fn test_concurrent_cluster_overwrites_do_not_fail_namespace_lock_quorum()
     // Keep the regression focused on false quorum-loss errors, not ordinary lock
     // wait exhaustion under a heavily contended same-key overwrite workload.
     cluster.set_env("RUSTFS_OBJECT_LOCK_ACQUIRE_TIMEOUT", "20");
+    cluster.set_env("RUSTFS_STORAGE_CLASS_STANDARD", "EC:2");
+    cluster.set_env("RUSTFS_PUT_RENAME_EARLY_ACK_ENABLE", "false");
+    cluster.set_env("RUSTFS_HEALTH_MINIMAL_RESPONSE_ENABLE", "false");
     cluster.start().await?;
     cluster.create_test_bucket(BUCKET).await?;
 
@@ -116,6 +421,181 @@ async fn test_concurrent_cluster_overwrites_do_not_fail_namespace_lock_quorum()
     );
 
     clients[0].delete_object().bucket(BUCKET).key(KEY).send().await?;
+    assert_node_readiness_tracks_quorum(&mut cluster).await?;
+    Ok(())
+}
+
+async fn assert_node_readiness_tracks_quorum(cluster: &mut RustFSTestClusterEnvironment) -> TestResult {
+    let clients: Vec<_> = 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();
+    let http = reqwest::Client::builder()
+        .no_proxy()
+        .timeout(Duration::from_secs(3))
+        .build()?;
+    let seed_key = "readiness-seed";
+    let seed_body = b"readiness quorum regression";
+    clients[0]
+        .put_object()
+        .bucket(BUCKET)
+        .key(seed_key)
+        .body(Bytes::from_static(seed_body).into())
+        .send()
+        .await?;
+
+    for (phase, survivors) in [4, 3, 2, 1, 4].into_iter().enumerate() {
+        if phase == 4 {
+            cluster.stop();
+            cluster.start().await?;
+        } else if survivors < 4 {
+            cluster.stop_node(survivors)?;
+        }
+        let write_ready = survivors >= 3;
+        let read_quorum = survivors >= 2;
+        let expected_status = if write_ready { 200 } else { 503 };
+        for (idx, client) in clients.iter().enumerate().take(survivors) {
+            let url = &cluster.nodes[idx].url;
+            let deadline = Instant::now() + Duration::from_secs(30);
+            // Poll health before issuing S3 I/O: idle remote disk handles must
+            // not remain evidence of quorum after their host becomes unreachable.
+            let payload = loop {
+                let response = http.get(format!("{url}/health/ready")).send().await?;
+                let status = response.status().as_u16();
+                let payload: serde_json::Value = response.json().await?;
+                if status == expected_status
+                    && payload["ready"] == write_ready
+                    && payload["details"]["storage"]["ready"] == write_ready
+                    && payload["details"]["storage"]["readQuorum"] == read_quorum
+                    && payload["details"]["storage"]["writeQuorum"] == write_ready
+                    && payload["details"]["poolMetadata"]["ready"] == true
+                    && payload["details"]["iam"]["ready"] == true
+                    && payload["details"]["lock"]["ready"] == write_ready
+                {
+                    break payload;
+                }
+                assert!(Instant::now() < deadline, "node {idx}, survivors={survivors}: HTTP {status}, {payload}");
+                tokio::time::sleep(Duration::from_millis(200)).await;
+            };
+            assert_eq!(payload["details"]["storage"]["readinessScope"], "write_quorum_and_pool_metadata");
+            assert_eq!(payload["details"]["storage"]["source"], "local_runtime");
+            assert_eq!(
+                payload["details"]["storage"]["status"],
+                if write_ready { "connected" } else { "disconnected" }
+            );
+            if !write_ready {
+                assert!(
+                    payload["degradedReasons"]
+                        .as_array()
+                        .expect("degraded reasons")
+                        .iter()
+                        .any(|reason| reason == "storage_and_lock_unavailable")
+                );
+            }
+
+            for path in ["/health/ready", "/minio/health/ready"] {
+                let head = http.head(format!("{url}{path}")).send().await?;
+                assert_eq!(head.status().as_u16(), expected_status, "HEAD {path}, survivors={survivors}");
+                assert!(head.bytes().await?.is_empty());
+                let response = http.get(format!("{url}{path}")).send().await?;
+                assert_eq!(response.status().as_u16(), expected_status);
+                let body: serde_json::Value = response.json().await?;
+                assert_eq!(body["details"]["storage"], payload["details"]["storage"]);
+                assert_eq!(body["details"]["poolMetadata"], payload["details"]["poolMetadata"]);
+            }
+            let live = http.get(format!("{url}/health/live")).send().await?;
+            assert_eq!(live.status().as_u16(), 200);
+            assert!(live.json::<serde_json::Value>().await?.get("details").is_none());
+            for (path, storage_ready, scope) in [
+                ("/minio/health/cluster", write_ready, "write_quorum_and_pool_metadata"),
+                ("/minio/health/cluster/read", read_quorum, "read_quorum"),
+            ] {
+                let deadline = Instant::now() + Duration::from_secs(30);
+                // Cluster read/write reports have independent caches; allow
+                // each observation to expire before comparing stable states.
+                let body = loop {
+                    let response = http.get(format!("{url}{path}")).send().await?;
+                    let status = response.status().as_u16();
+                    let body: serde_json::Value = response.json().await?;
+                    if status == expected_status
+                        && body["details"]["storage"]["ready"] == storage_ready
+                        && body["details"]["lock"]["ready"] == write_ready
+                    {
+                        break body;
+                    }
+                    assert!(Instant::now() < deadline, "{path}, survivors={survivors}: HTTP {status}, {body}");
+                    tokio::time::sleep(Duration::from_millis(200)).await;
+                };
+                assert_eq!(body["details"]["storage"]["readinessScope"], scope);
+            }
+
+            let put = client
+                .put_object()
+                .bucket(BUCKET)
+                .key(format!("readiness-phase-{phase}-node-{idx}"))
+                .body(Bytes::from_static(seed_body).into())
+                .send()
+                .await;
+            let put_status = if write_ready {
+                put.expect("a ready node must accept the PUT");
+                200
+            } else {
+                let error = put.expect_err("subquorum node must reject PUT");
+                assert!(
+                    error.raw_response().is_some_and(|response| response.status().as_u16() >= 500),
+                    "unexpected PUT failure: {error:?}"
+                );
+                error.raw_response().expect("PUT error response").status().as_u16()
+            };
+            let get = client.get_object().bucket(BUCKET).key(seed_key).send().await;
+            let get_status = match get {
+                Ok(object) => {
+                    assert_eq!(object.body.collect().await?.into_bytes().as_ref(), seed_body);
+                    200
+                }
+                Err(error) => {
+                    assert!(!write_ready, "GET must succeed on a ready cluster: {error:?}");
+                    let status = error
+                        .raw_response()
+                        .expect("GET should have an HTTP response")
+                        .status()
+                        .as_u16();
+                    assert!(status >= 500, "unexpected GET failure: {error:?}");
+                    status
+                }
+            };
+            let list = client.list_objects_v2().bucket(BUCKET).send().await;
+            let list_status = match list {
+                Ok(result) => {
+                    assert!(result.contents().iter().any(|object| object.key() == Some(seed_key)));
+                    200
+                }
+                Err(error) => {
+                    assert!(!write_ready, "listing must succeed on a ready cluster: {error:?}");
+                    let status = error
+                        .raw_response()
+                        .expect("LIST should have an HTTP response")
+                        .status()
+                        .as_u16();
+                    assert!(status >= 500, "unexpected listing failure: {error:?}");
+                    status
+                }
+            };
+            eprintln!(
+                "readiness matrix: survivors={survivors}, node={idx}, ready={write_ready}, read_quorum={read_quorum}, PUT={put_status}, GET={get_status}, LIST={list_status}"
+            );
+        }
+    }
+    cluster.stop();
     Ok(())
 }
 
@@ -125,6 +605,8 @@ async fn test_concurrent_cluster_overwrites_do_not_fail_namespace_lock_quorum()
 /// Before the fix, `map_namespace_lock_error` wrapped lock timeout/conflict errors as
 /// `StorageError::other(...)` → `StorageError::Io(...)`, which fell through to
 /// `S3ErrorCode::InternalError` (500) in the error mapping.
+/// Also checks PUT and multipart initialization when node failures prevent
+/// acquiring a table publication guard.
 #[tokio::test]
 async fn test_concurrent_put_same_key_never_returns_500() -> TestResult {
     crate::common::init_logging();
@@ -227,5 +709,6 @@ async fn test_concurrent_put_same_key_never_returns_500() -> TestResult {
     );
 
     clients[0].delete_object().bucket(BUCKET).key(KEY).send().await?;
-    Ok(())
+    cluster.stop();
+    assert_degraded_cluster_publication_guard_errors_are_retryable().await
 }

crates/e2e_test/src/on_demand_migration/interaction_test.rs

@@ -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}");

crates/e2e_test/src/protocols/sftp_compliance.rs

@@ -79,11 +79,6 @@ pub async fn test_sftp_compliance_suite() -> Result<()> {
             .await
             .map_err(|e| anyhow!("{}", e))?;
 
-        // Protocol listeners can accept connections before IAM is initialized.
-        // A signed S3 request establishes readiness before the first SFTP login.
-        let s3 = build_test_s3_client(&format!("http://{COMPLIANCE_RW_S3_ADDRESS}"));
-        wait_for_s3_ready(&s3, 30).await?;
-
         let (session, sftp) = connect_sftp_to(COMPLIANCE_RW_SFTP_ADDRESS).await?;
 
         cmptst_01::run_medium_binary_round_trip(&sftp).await?;
@@ -106,6 +101,8 @@ pub async fn test_sftp_compliance_suite() -> Result<()> {
         // reach the finalised object as x-amz-meta-* user metadata
         // through the CreateMultipartUpload input field. The S3 client
         // connects to the same rustfs process this suite already drives.
+        let s3 = build_test_s3_client(&format!("http://{COMPLIANCE_RW_S3_ADDRESS}"));
+        wait_for_s3_ready(&s3, 30).await?;
         cmptst_34::run_open_attrs_round_trip_multipart(&sftp, &s3).await?;
 
         drop(sftp);

crates/e2e_test/src/protocols/sftp_core.rs

@@ -168,10 +168,6 @@ pub async fn test_sftp_core_operations() -> Result<()> {
             .await
             .map_err(|e| anyhow!("{}", e))?;
 
-        // Protocol listeners can accept connections before IAM is initialized.
-        let s3 = build_test_s3_client(S3_ENDPOINT);
-        wait_for_s3_ready(&s3, S3_READY_ATTEMPTS).await?;
-
         let (session, sftp) = connect_sftp().await?;
 
         // --- 1. Subsystem canary: SFTP session reachable after password auth ---
@@ -352,6 +348,16 @@ pub async fn test_sftp_core_operations() -> Result<()> {
         let _ = bad_session.disconnect(russh::Disconnect::ByApplication, "", "en").await;
         info!("PASS: bad-password authentication rejected");
 
+        // --- Cross-protocol setup: aws-sdk-s3 client against the same server ---
+        // The rustfs binary spawned for this suite serves both SFTP on port
+        // 9022 and S3 on port 9000. The S3 stack may need a moment to finish
+        // initialising after TCP is listening, so list_buckets is polled
+        // until it succeeds before any cross-protocol assertion runs.
+        info!("Testing SFTP: prepare aws-sdk-s3 client and wait for S3 readiness");
+        let s3 = build_test_s3_client(S3_ENDPOINT);
+        wait_for_s3_ready(&s3, S3_READY_ATTEMPTS).await?;
+        info!("PASS: S3 endpoint reachable from cross-protocol client");
+
         // --- SFTP write, S3 read: SHA256 round-trip ---
         // SFTP creates the object, then assert_cross_protocol_sha_match
         // fetches it via both S3 GetObject and SFTP READ and compares
@@ -516,9 +522,6 @@ pub async fn test_sftp_idle_timeout_disconnects() -> Result<()> {
             .await
             .map_err(|e| anyhow!("{}", e))?;
 
-        let s3 = build_test_s3_client(&format!("http://{IDLE_S3_ADDRESS}"));
-        wait_for_s3_ready(&s3, S3_READY_ATTEMPTS).await?;
-
         let (session, sftp) = connect_sftp_to(IDLE_SFTP_ADDRESS).await?;
 
         // Confirm the session is live before the wait so a failure in the

crates/e2e_test/src/replication_extension_test.rs

@@ -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)

crates/e2e_test/src/scanner_heal_evidence.rs

@@ -1,181 +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.
-
-//! Build-bound evidence for scanner and heal restart tests.
-
-use crate::common::ClusterTopology;
-use sha2::{Digest, Sha256};
-use std::error::Error;
-use std::io::{Read, Write};
-use std::path::{Path, PathBuf};
-
-#[derive(serde::Deserialize)]
-struct EvidenceBuild {
-    sha256: String,
-}
-
-#[derive(serde::Deserialize)]
-struct RestartEvidenceRun {
-    schema: u32,
-    run_id: String,
-    source_revision: String,
-    test_build: serde_json::Value,
-    binary: EvidenceBuild,
-    test_binary: EvidenceBuild,
-}
-
-#[derive(Clone, Copy)]
-pub(crate) struct ScannerHealEvidenceCase {
-    pub(crate) id: &'static str,
-    pub(crate) oracle: &'static str,
-    pub(crate) evidence: &'static str,
-    pub(crate) unclean_shutdown_marker: bool,
-    pub(crate) topology: EvidenceTopology,
-    pub(crate) storage_class_standard: Option<&'static str>,
-    pub(crate) erasure_set_drive_count: Option<&'static str>,
-}
-
-#[derive(Clone, Copy)]
-pub(crate) struct EvidenceTopology {
-    pub(crate) nodes: usize,
-    pub(crate) drives_per_node: usize,
-}
-
-impl EvidenceTopology {
-    pub(crate) const fn new(nodes: usize, drives_per_node: usize) -> Self {
-        Self { nodes, drives_per_node }
-    }
-
-    pub(crate) fn total_drives(self) -> usize {
-        self.nodes * self.drives_per_node
-    }
-
-    pub(crate) fn cluster_topology(self) -> ClusterTopology {
-        ClusterTopology::single_pool_multidrive(self.nodes, self.drives_per_node)
-    }
-}
-
-pub(crate) struct RestartEvidenceContext {
-    directory: PathBuf,
-    run: RestartEvidenceRun,
-    case: ScannerHealEvidenceCase,
-}
-
-fn file_sha256(path: &Path) -> Result<String, Box<dyn Error + Send + Sync>> {
-    let mut file = std::fs::File::open(path)?;
-    let mut digest = Sha256::new();
-    let mut buffer = [0_u8; 64 * 1024];
-    loop {
-        let read = file.read(&mut buffer)?;
-        if read == 0 {
-            break;
-        }
-        digest.update(&buffer[..read]);
-    }
-    Ok(digest.finalize().iter().map(|byte| format!("{byte:02x}")).collect())
-}
-
-pub(crate) fn restart_evidence_run(
-    binary: &Path,
-    case: ScannerHealEvidenceCase,
-) -> Result<Option<RestartEvidenceContext>, Box<dyn Error + Send + Sync>> {
-    let Some(directory) = std::env::var_os("RUSTFS_SCANNER_HEAL_RUN_DIR") else {
-        return Ok(None);
-    };
-    if case.id.is_empty()
-        || case.oracle.is_empty()
-        || !case.oracle.ends_with(".json")
-        || case.oracle.contains('/')
-        || case.oracle.contains('\\')
-        || case.oracle.contains("..")
-        || !matches!(case.evidence, "process-restart" | "process-crash-restart")
-        || (case.evidence == "process-crash-restart") != case.unclean_shutdown_marker
-    {
-        return Err("invalid scanner/heal evidence case".into());
-    }
-    let directory = PathBuf::from(directory);
-    let receipt = directory.join("run.json");
-    if receipt.metadata()?.len() > 1024 * 1024 {
-        return Err("oversized scanner/heal execution receipt".into());
-    }
-    let run: RestartEvidenceRun = serde_json::from_slice(&std::fs::read(receipt)?)?;
-    if run.schema != 1 || run.run_id.len() != 32 || run.source_revision.len() != 40 {
-        return Err("invalid scanner/heal execution identity".into());
-    }
-    let built = compiled_test_identity();
-    for key in ["source_revision", "dirty", "lock_blob", "features"] {
-        assert_eq!(built[key], run.test_build[key], "compiled test identity differs for {key}");
-    }
-    assert_eq!(file_sha256(binary)?, run.binary.sha256, "server binary must match the run receipt");
-    assert_eq!(
-        file_sha256(&std::env::current_exe()?)?,
-        run.test_binary.sha256,
-        "test executable must match the run receipt"
-    );
-    if directory.join(case.oracle).exists() {
-        return Err("scanner/heal oracle already exists; create a new execution receipt".into());
-    }
-    Ok(Some(RestartEvidenceContext { directory, run, case }))
-}
-
-fn compiled_test_identity() -> serde_json::Value {
-    serde_json::json!({
-        "source_revision": env!("RUSTFS_E2E_BUILD_COMMIT"),
-        "dirty": env!("RUSTFS_E2E_BUILD_DIRTY") != "false",
-        "lock_blob": env!("RUSTFS_E2E_BUILD_LOCK"),
-        "features": env!("RUSTFS_E2E_BUILD_FEATURES"),
-        "target": env!("RUSTFS_E2E_BUILD_TARGET"),
-        "profile": env!("RUSTFS_E2E_BUILD_PROFILE"),
-        "rustflags_hex": env!("RUSTFS_E2E_BUILD_RUSTFLAGS_HEX"),
-    })
-}
-
-pub(crate) struct RestartObservation {
-    pub(crate) nodes: usize,
-    pub(crate) drives_per_node: usize,
-    pub(crate) pid_before: u32,
-    pub(crate) pid_after: u32,
-    pub(crate) unclean_shutdown_marker: bool,
-    pub(crate) objects: Vec<serde_json::Value>,
-    pub(crate) node_listings: Vec<Vec<String>>,
-}
-
-impl RestartEvidenceContext {
-    pub(crate) fn write(self, binary: &Path, observed: RestartObservation) -> Result<(), Box<dyn Error + Send + Sync>> {
-        assert_ne!(observed.pid_before, observed.pid_after, "target must be a new process");
-        assert_eq!(file_sha256(binary)?, self.run.binary.sha256, "server build changed during restart");
-        let evidence = serde_json::json!({
-            "schema": 1, "case": self.case.id, "evidence": self.case.evidence,
-            "run_id": self.run.run_id, "source_revision": self.run.source_revision,
-            "test_build": compiled_test_identity(),
-            "binary_sha256": self.run.binary.sha256,
-            "test_binary_sha256": self.run.test_binary.sha256,
-            "topology": {"nodes": observed.nodes, "drives_per_node": observed.drives_per_node},
-            "pid_before": observed.pid_before, "pid_after": observed.pid_after,
-            "unclean_shutdown_marker": observed.unclean_shutdown_marker,
-            "objects": observed.objects, "node_listings": observed.node_listings,
-        });
-        let data = serde_json::to_vec(&evidence)?;
-        if data.len() > 1024 * 1024 {
-            return Err("scanner/heal oracle exceeds the 1 MiB artifact budget".into());
-        }
-        let mut output = std::fs::OpenOptions::new()
-            .write(true)
-            .create_new(true)
-            .open(self.directory.join(self.case.oracle))?;
-        output.write_all(&data)?;
-        output.sync_all()?;
-        Ok(())
-    }
-}

crates/e2e_test/src/storage_metric_ownership_test.rs

@@ -0,0 +1,335 @@
+// Copyright 2024 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Native OTLP -> Collector -> Prometheus contract, including a rolling upgrade.
+
+use crate::common::RustFSTestClusterEnvironment;
+use aws_sdk_s3::primitives::ByteStream;
+use serde_json::Value;
+use std::fs::{self, File};
+use std::net::TcpListener;
+use std::path::{Path, PathBuf};
+use std::process::{Child, Command, Stdio};
+use std::time::{Duration, Instant};
+
+type TestResult<T = ()> = Result<T, Box<dyn std::error::Error + Send + Sync>>;
+
+struct ToolProcess(Child);
+impl Drop for ToolProcess {
+    fn drop(&mut self) {
+        let _ = self.0.kill();
+        let _ = self.0.wait();
+    }
+}
+
+fn required_binary(name: &str) -> TestResult<PathBuf> {
+    let path = PathBuf::from(std::env::var(name).map_err(|_| format!("{name} must name a pinned executable"))?);
+    if !path.is_file() {
+        return Err(format!("{name} does not name a file: {}", path.display()).into());
+    }
+    Ok(path)
+}
+
+fn free_port() -> TestResult<u16> {
+    Ok(TcpListener::bind("127.0.0.1:0")?.local_addr()?.port())
+}
+
+fn start_tool(binary: &Path, args: &[String], log: &Path) -> TestResult<ToolProcess> {
+    let log = File::create(log)?;
+    Ok(ToolProcess(
+        Command::new(binary)
+            .args(args)
+            .env("NO_PROXY", "127.0.0.1,localhost")
+            .env_remove("HTTP_PROXY")
+            .env_remove("HTTPS_PROXY")
+            .stdout(Stdio::from(log.try_clone()?))
+            .stderr(Stdio::from(log))
+            .spawn()?,
+    ))
+}
+
+async fn query(client: &reqwest::Client, base: &str, expression: &str) -> TestResult<Value> {
+    let mut url = reqwest::Url::parse(&format!("{base}/api/v1/query"))?;
+    url.query_pairs_mut().append_pair("query", expression);
+    let result: Value = client.get(url).send().await?.error_for_status()?.json().await?;
+    if result["status"] != "success" {
+        return Err(format!("PromQL failed: {result}").into());
+    }
+    Ok(result["data"]["result"].clone())
+}
+
+async fn await_count(client: &reqwest::Client, base: &str, selector: &str, expected: u64) -> TestResult {
+    let deadline = Instant::now() + Duration::from_secs(120);
+    loop {
+        let result = query(client, base, &format!("count({selector}) or vector(0)")).await;
+        if let Ok(rows) = &result
+            && rows[0]["value"][1].as_str().and_then(|value| value.parse::<u64>().ok()) == Some(expected)
+        {
+            println!("PASS count={expected}: {selector}");
+            return Ok(());
+        }
+        if Instant::now() >= deadline {
+            return Err(format!("expected {expected} for {selector}; last result: {result:?}").into());
+        }
+        tokio::time::sleep(Duration::from_millis(250)).await;
+    }
+}
+
+async fn validate_dashboard_queries(client: &reqwest::Client, base: &str, observer: &str) -> TestResult {
+    let path = PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("../../.docker/observability/grafana/dashboards/rustfs.json");
+    let dashboard: Value = serde_json::from_str(&fs::read_to_string(path)?)?;
+    for name in ["storage_cluster", "storage_observer"] {
+        let variable = dashboard["templating"]["list"]
+            .as_array()
+            .ok_or("dashboard variables")?
+            .iter()
+            .find(|variable| variable["name"] == name)
+            .ok_or("storage selection variable")?;
+        assert_eq!(variable["multi"], false, "storage views must select one {name}");
+        assert_eq!(variable["includeAll"], false, "storage views must select one {name}");
+    }
+    let mut pending = dashboard["panels"]
+        .as_array()
+        .ok_or("dashboard panels")?
+        .iter()
+        .collect::<Vec<_>>();
+    let mut checked = 0;
+    while let Some(panel) = pending.pop() {
+        if let Some(children) = panel["panels"].as_array() {
+            pending.extend(children);
+        }
+        for target in panel["targets"].as_array().into_iter().flatten() {
+            let Some(expression) = target["expr"].as_str() else { continue };
+            if !expression.contains("rustfs:storage:current") {
+                continue;
+            }
+            if expression.contains("collection_scope=\"cluster\"") {
+                assert!(
+                    expression.contains("observer=\"$storage_observer\""),
+                    "global views must select one observer: {expression}"
+                );
+            }
+            let mut expression = expression.to_string();
+            for (name, value) in [
+                ("$__rate_interval", "5m"),
+                ("$storage_cluster", "metrics-e2e"),
+                ("$storage_observer", observer),
+                ("$drive_api", ".*"),
+                ("$server", ".*"),
+                ("$drive", ".*"),
+                ("$job", "rustfs"),
+            ] {
+                expression = expression.replace(name, value);
+            }
+            query(client, base, &expression).await?;
+            checked += 1;
+        }
+    }
+    assert!(checked > 0, "the storage dashboard queries must be exercised");
+    println!("PASS: {checked} storage dashboard queries against the live pipeline");
+    Ok(())
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+#[ignore = "external tools: pinned Collector, Prometheus, previous release and current RustFS binaries"]
+async fn storage_metric_ownership_pipeline() -> TestResult {
+    let baseline = required_binary("RUSTFS_METRICS_BASELINE_BINARY")?;
+    let current = required_binary("CARGO_BIN_EXE_rustfs")?;
+    let collector = required_binary("RUSTFS_OTELCOL_BINARY")?;
+    let prometheus = required_binary("RUSTFS_PROMETHEUS_BINARY")?;
+    let temp = tempfile::Builder::new().prefix("rustfs-storage-metrics-").tempdir()?;
+    let work = if let Ok(path) = std::env::var("RUSTFS_METRICS_E2E_ARTIFACTS") {
+        let path = PathBuf::from(path);
+        fs::create_dir_all(&path)?;
+        tempfile::Builder::new().prefix("storage-run-").tempdir_in(path)?.keep()
+    } else {
+        temp.path().to_path_buf()
+    };
+    println!("Metrics pipeline logs: {}", work.display());
+    let otlp = free_port()?;
+    let scrape = free_port()?;
+    let prom = free_port()?;
+    let collector_config = work.join("collector.yaml");
+    fs::write(
+        &collector_config,
+        format!(
+            r#"receivers:
+  otlp:
+    protocols:
+      http:
+        endpoint: 127.0.0.1:{otlp}
+exporters:
+  prometheus:
+    endpoint: 127.0.0.1:{scrape}
+    send_timestamps: true
+    metric_expiration: 5m
+    resource_to_telemetry_conversion:
+      enabled: true
+service:
+  telemetry:
+    metrics:
+      level: none
+    logs:
+      level: warn
+  pipelines:
+    metrics:
+      receivers: [otlp]
+      exporters: [prometheus]
+"#
+        ),
+    )?;
+    let rules = PathBuf::from(env!("CARGO_MANIFEST_DIR"))
+        .join("../../.docker/observability/prometheus-rules/rustfs-storage.yml")
+        .canonicalize()?;
+    let prom_config = work.join("prometheus.yaml");
+    fs::write(
+        &prom_config,
+        format!(
+            r#"global:
+  scrape_interval: 1s
+  evaluation_interval: 1s
+rule_files:
+  - '{}'
+scrape_configs:
+  - job_name: rustfs
+    static_configs:
+      - targets: ['127.0.0.1:{scrape}']
+"#,
+            rules.display()
+        ),
+    )?;
+    // Use the shipped expressions, with only the test evaluation interval shortened.
+    let test_rules = work.join("storage-rules.yaml");
+    fs::write(&test_rules, fs::read_to_string(&rules)?.replace("interval: 15s", "interval: 1s"))?;
+    fs::write(
+        &prom_config,
+        fs::read_to_string(&prom_config)?.replace(&rules.display().to_string(), &test_rules.display().to_string()),
+    )?;
+    let _collector = start_tool(
+        &collector,
+        &[format!("--config={}", collector_config.display())],
+        &work.join("collector.log"),
+    )?;
+    let _prometheus = start_tool(
+        &prometheus,
+        &[
+            format!("--config.file={}", prom_config.display()),
+            format!("--web.listen-address=127.0.0.1:{prom}"),
+            format!("--storage.tsdb.path={}", work.join("prometheus-data").display()),
+        ],
+        &work.join("prometheus.log"),
+    )?;
+    let http = reqwest::Client::builder()
+        .no_proxy()
+        .timeout(Duration::from_secs(5))
+        .build()?;
+    let prom_url = format!("http://127.0.0.1:{prom}");
+    await_count(&http, &prom_url, "up{job=\"rustfs\"} == 1", 1).await?;
+
+    let mut cluster = RustFSTestClusterEnvironment::new(4).await?;
+    cluster.set_env("NO_PROXY", "127.0.0.1,localhost");
+    cluster.set_env("RUSTFS_OBS_METRIC_ENDPOINT", format!("http://127.0.0.1:{otlp}/v1/metrics"));
+    cluster.set_env("OTEL_RESOURCE_ATTRIBUTES", "rustfs.cluster.id=metrics-e2e");
+    cluster.set_env("RUSTFS_OBS_METER_INTERVAL", "2");
+    cluster.set_env("RUSTFS_OBS_METRICS_EXPORT_ENABLED", "true");
+    cluster.set_env("RUSTFS_OBS_LOGS_EXPORT_ENABLED", "false");
+    cluster.set_env("RUSTFS_OBS_TRACES_EXPORT_ENABLED", "false");
+    cluster.set_env("RUSTFS_METRICS_NODE_INTERVAL", "2");
+    cluster.set_env("RUSTFS_METRICS_CLUSTER_INTERVAL", "5");
+    for index in 0..4 {
+        // Four localhost processes otherwise share the startup resource IP.
+        // Distinct test host IDs model the four hosts in a distributed deployment.
+        cluster.set_node_env(
+            index,
+            "OTEL_RESOURCE_ATTRIBUTES",
+            format!("rustfs.cluster.id=metrics-e2e,host.id=metrics-node-{index}"),
+        )?;
+        cluster.set_node_capture_log_path(index, work.join(format!("node-{index}.log")).display().to_string())?;
+    }
+    cluster.start_with_binary(&baseline).await?;
+    // Reproduce the original four observers x four global drives before fixing it.
+    await_count(&http, &prom_url, "rustfs_system_drive_total_bytes{collection_scope=\"\",drive!=\"\"}", 16).await?;
+    let local = "rustfs:storage:current{source_metric=\"rustfs_system_drive_total_bytes\",collection_scope=\"local\",rustfs_cluster_id=\"metrics-e2e\"}";
+    for index in 0..4 {
+        cluster.stop_node_gracefully(index).await?;
+        cluster.start_node_from_binary(index, &current).await?;
+        await_count(&http, &prom_url, local, u64::try_from(index + 1)?).await?;
+    }
+    let rows = query(&http, &prom_url, local).await?;
+    for row in rows.as_array().ok_or("expected a metric vector")? {
+        assert_eq!(
+            row["metric"]["observer"], row["metric"]["server"],
+            "a node must only export its own detailed drives"
+        );
+    }
+    let observer = cluster.nodes[0].address.clone();
+    let inventory = format!(
+        "rustfs:storage:current{{source_metric=\"rustfs_cluster_drive_present\",collection_scope=\"cluster\",rustfs_cluster_id=\"metrics-e2e\",observer=\"{observer}\"}}"
+    );
+    await_count(&http, &prom_url, &inventory, 4).await?;
+    cluster.create_test_bucket("metrics-ownership").await?;
+    let client = cluster.create_s3_client(0)?;
+    for index in 0..8 {
+        client
+            .put_object()
+            .bucket("metrics-ownership")
+            .key(format!("object-{index}"))
+            .body(ByteStream::from(vec![7_u8; 4096]))
+            .send()
+            .await?;
+    }
+    await_count(
+        &http,
+        &prom_url,
+        "count by (server) (rustfs:storage:current{source_metric=\"rustfs_system_drive_api_calls_total\",collection_scope=\"local\"})",
+        4,
+    ).await?;
+    let counters = query(
+        &http,
+        &prom_url,
+        "rustfs:storage:current{source_metric=\"rustfs_system_drive_api_calls_total\",collection_scope=\"local\"}",
+    )
+    .await?;
+    assert!(
+        !counters.as_array().ok_or("expected counters")?.is_empty(),
+        "exercise actual storage counters"
+    );
+    for row in counters.as_array().ok_or("expected counters")? {
+        assert_eq!(row["metric"]["observer"], row["metric"]["server"]);
+        assert!(
+            !row["metric"]["disk_id"].as_str().unwrap_or_default().is_empty(),
+            "counters must carry physical disk identity"
+        );
+    }
+    validate_dashboard_queries(&http, &prom_url, &observer).await?;
+    for index in (1..4).rev() {
+        cluster.stop_node(index)?;
+        // The Collector stays alive; cached samples must not keep stopped owners fresh.
+        await_count(&http, &prom_url, local, u64::try_from(index)?).await?;
+        await_count(&http, &prom_url, &inventory, 4).await?;
+        let unavailable = format!(
+            "rustfs:storage:current{{source_metric=\"rustfs_cluster_drive_runtime_state\",collection_scope=\"cluster\",rustfs_cluster_id=\"metrics-e2e\",observer=\"{observer}\",state=~\"offline|unknown|suspect\"}} == 1"
+        );
+        await_count(&http, &prom_url, &unavailable, u64::try_from(4 - index)?).await?;
+    }
+    cluster.stop();
+    await_count(&http, &prom_url, local, 0).await?;
+    cluster.start_with_binary(&current).await?;
+    await_count(&http, &prom_url, local, 4).await?;
+    await_count(&http, &prom_url, &inventory, 4).await?;
+    let restored = client.get_object().bucket("metrics-ownership").key("object-0").send().await?;
+    assert_eq!(restored.body.collect().await?.into_bytes().as_ref(), vec![7_u8; 4096].as_slice());
+    println!("PASS: baseline duplication; four rolling upgrades; owner identity; counters; 4 -> 3 -> 2 -> 1 -> 0 -> 4 recovery");
+    Ok(())
+}

crates/e2e_test/src/upgrade_compatibility_test.rs

@@ -32,8 +32,10 @@ use aws_sdk_s3::types::{
     VersioningConfiguration,
 };
 use http::{Method, StatusCode};
+use serde_json::Value;
+use std::io::Write;
 use std::path::{Path, PathBuf};
-use std::time::Duration;
+use std::time::{Duration, SystemTime, UNIX_EPOCH};
 use tokio::task::JoinSet;
 use tokio::time::{Instant, sleep};
 
@@ -41,6 +43,7 @@ type TestResult = Result<(), Box<dyn std::error::Error + Send + Sync>>;
 type BoxError = Box<dyn std::error::Error + Send + Sync>;
 
 const SOURCE_BINARY_ENV: &str = "RUSTFS_UPGRADE_SOURCE_BINARY";
+const G09_EVIDENCE_DIR_ENV: &str = "RUSTFS_SCANNER_HEAL_G09_EVIDENCE_DIR";
 const RC5_COMMIT: &str = "40a2470feb567201165a5b809b7598bb4b1f68f5";
 const SSE_MASTER_KEY_ENV: &str = "RUSTFS_SSE_S3_MASTER_KEY";
 const SSE_MASTER_KEY: &str = "QkJCQkJCQkJCQkJCQkJCQkJCQkJCQkJCQkJCQkJCQkI=";
@@ -81,6 +84,14 @@ const QUOTA_READINESS_TIMEOUT: Duration = Duration::from_secs(30);
 // of treating it as an upgrade failure.
 const QUOTA_ADMISSION_WARMUP_TIMEOUT: Duration = Duration::from_secs(90);
 
+struct G09EvidenceContext {
+    directory: PathBuf,
+    current_revision: String,
+    previous_revision: String,
+    run_id: String,
+    measurement_window_id: String,
+}
+
 fn source_binary() -> Result<PathBuf, Box<dyn std::error::Error + Send + Sync>> {
     let path = std::env::var_os(SOURCE_BINARY_ENV)
         .map(PathBuf::from)
@@ -91,6 +102,155 @@ fn source_binary() -> Result<PathBuf, Box<dyn std::error::Error + Send + Sync>>
     Ok(path)
 }
 
+fn is_lower_hex_revision(value: &str) -> bool {
+    value.len() == 40
+        && value
+            .bytes()
+            .all(|byte| byte.is_ascii_hexdigit() && !byte.is_ascii_uppercase())
+}
+
+fn current_source_revision() -> Result<String, BoxError> {
+    let revision = env!("RUSTFS_E2E_BUILD_COMMIT");
+    if !is_lower_hex_revision(revision) {
+        return Err(format!("current test binary has invalid source revision {revision}").into());
+    }
+    Ok(revision.to_string())
+}
+
+async fn binary_source_revision(binary: &Path) -> Result<String, BoxError> {
+    let output = tokio::process::Command::new(binary).arg("--version").output().await?;
+    if !output.status.success() {
+        return Err(format!("{} --version failed with {}", binary.display(), output.status).into());
+    }
+    let version = String::from_utf8(output.stdout)?;
+    source_revision_from_version_output(&version, binary)
+}
+
+fn source_revision_from_version_output(version: &str, binary: &Path) -> Result<String, BoxError> {
+    version
+        .split(|ch: char| !ch.is_ascii_hexdigit())
+        .find(|token| is_lower_hex_revision(token))
+        .map(str::to_string)
+        .ok_or_else(|| format!("{} --version did not expose a 40-byte source revision", binary.display()).into())
+}
+
+async fn g09_evidence_context(previous_binary: &Path) -> Result<Option<G09EvidenceContext>, BoxError> {
+    let Some(directory) = std::env::var_os(G09_EVIDENCE_DIR_ENV) else {
+        return Ok(None);
+    };
+    let directory = PathBuf::from(directory);
+    std::fs::create_dir_all(&directory)?;
+    let current_revision = current_source_revision()?;
+    let previous_revision = binary_source_revision(previous_binary).await?;
+    if current_revision == previous_revision {
+        return Err("G09 mixed-version evidence requires distinct current and previous source revisions".into());
+    }
+    let now = SystemTime::now().duration_since(UNIX_EPOCH)?.as_nanos();
+    Ok(Some(G09EvidenceContext {
+        directory,
+        current_revision,
+        previous_revision,
+        run_id: format!("g09-upgrade-{}-{now}", std::process::id()),
+        measurement_window_id: format!("g09-mixed-version-window-{now}"),
+    }))
+}
+
+fn write_g09_evidence(
+    context: &G09EvidenceContext,
+    field: &str,
+    role: &str,
+    cases: &[&str],
+    test: &str,
+    details: Value,
+) -> TestResult {
+    let path = context.directory.join(format!("G09-{field}.json"));
+    let mut evidence = serde_json::json!({
+        "schema": 1,
+        "evidence_type": "measured",
+        "artifact_kind": "upgrade-compatibility-e2e",
+        "source_revision": context.current_revision,
+        "run_id": context.run_id,
+        "measurement_window_id": context.measurement_window_id,
+        "gate": "G09",
+        "field": field,
+        "versions": [context.previous_revision, context.current_revision],
+        "current_revision": context.current_revision,
+        "previous_revision": context.previous_revision,
+        "mixed_version_role": role,
+        "mixed_version_cases": cases,
+        "test": test,
+        "details": details,
+    });
+    if field == "rollback_payload_evidence" {
+        evidence["rollback_payload_replayed"] = Value::Bool(true);
+    }
+    let data = serde_json::to_vec_pretty(&evidence)?;
+    if data.len() > 1024 * 1024 {
+        return Err("G09 mixed-version evidence exceeds the 1 MiB artifact budget".into());
+    }
+    let mut output = std::fs::OpenOptions::new().write(true).create_new(true).open(path)?;
+    output.write_all(&data)?;
+    output.write_all(b"\n")?;
+    output.sync_all()?;
+    Ok(())
+}
+
+#[cfg(test)]
+mod g09_evidence_tests {
+    use super::*;
+    use tempfile::TempDir;
+
+    #[test]
+    fn source_revision_parser_requires_lowercase_forty_byte_sha() -> TestResult {
+        let binary = Path::new("rustfs");
+        assert_eq!(
+            source_revision_from_version_output("rustfs 1.0.0 abcdef0123456789abcdef0123456789abcdef01 clean", binary)?,
+            "abcdef0123456789abcdef0123456789abcdef01"
+        );
+        assert!(source_revision_from_version_output("rustfs ABCDEF0123456789ABCDEF0123456789ABCDEF01", binary).is_err());
+        assert!(source_revision_from_version_output("rustfs abcdef", binary).is_err());
+        Ok(())
+    }
+
+    #[test]
+    fn g09_evidence_writer_records_measured_role_and_refuses_overwrite() -> TestResult {
+        let directory = TempDir::new()?;
+        let context = G09EvidenceContext {
+            directory: directory.path().to_path_buf(),
+            current_revision: "b".repeat(40),
+            previous_revision: "a".repeat(40),
+            run_id: "g09-upgrade-test-run".to_string(),
+            measurement_window_id: "g09-upgrade-test-window".to_string(),
+        };
+        write_g09_evidence(
+            &context,
+            "mixed_version_reader_evidence",
+            "mixed-version-reader",
+            &["old-writer-new-reader", "new-writer-old-reader"],
+            "unit",
+            serde_json::json!({"assertions": ["reader evidence"]}),
+        )?;
+        let path = directory.path().join("G09-mixed_version_reader_evidence.json");
+        let evidence: Value = serde_json::from_slice(&std::fs::read(path)?)?;
+        assert_eq!(evidence["evidence_type"], "measured");
+        assert_eq!(evidence["gate"], "G09");
+        assert_eq!(evidence["field"], "mixed_version_reader_evidence");
+        assert_eq!(evidence["mixed_version_role"], "mixed-version-reader");
+        assert_eq!(evidence["versions"], serde_json::json!(["a".repeat(40), "b".repeat(40)]));
+
+        let overwrite = write_g09_evidence(
+            &context,
+            "mixed_version_reader_evidence",
+            "mixed-version-reader",
+            &["old-writer-new-reader", "new-writer-old-reader"],
+            "unit",
+            serde_json::json!({}),
+        );
+        assert!(overwrite.is_err(), "G09 evidence must not overwrite an existing artifact");
+        Ok(())
+    }
+}
+
 async fn enable_versioning(client: &Client, bucket: &str) -> TestResult {
     let configuration = VersioningConfiguration::builder()
         .status(BucketVersioningStatus::Enabled)
@@ -523,6 +683,7 @@ async fn direct_upgrade_from_rc2_preserves_object_contracts() -> TestResult {
 async fn rolling_upgrade_from_rc2_preserves_mixed_version_contracts() -> TestResult {
     init_logging();
     let previous_binary = source_binary()?;
+    let evidence_context = g09_evidence_context(&previous_binary).await?;
     let current_binary = rustfs_binary_path();
     let mut cluster = RustFSTestClusterEnvironment::new(MIXED_NODE_COUNT).await?;
     cluster.set_env("RUST_LOG", "rustfs=warn,rustfs_notify=warn");
@@ -555,6 +716,45 @@ async fn rolling_upgrade_from_rc2_preserves_mixed_version_contracts() -> TestRes
         }
     }
 
+    if let Some(context) = evidence_context.as_ref() {
+        let phases = ["one-current-node", "one-previous-node"];
+        let objects_per_phase = MULTIPART_WORKERS * MULTIPART_UPLOADS_PER_WORKER + 2;
+        write_g09_evidence(
+            context,
+            "mixed_version_reader_evidence",
+            "mixed-version-reader",
+            &["old-writer-new-reader", "new-writer-old-reader"],
+            "upgrade_compatibility_test::rolling_upgrade_from_rc2_preserves_mixed_version_contracts",
+            serde_json::json!({
+                "bucket": MIXED_BUCKET,
+                "phases": phases,
+                "objects_per_phase": objects_per_phase,
+                "assertions": [
+                    "current node reads objects written through previous-release client",
+                    "previous-release node reads objects written through current client",
+                    "all nodes list every mixed-version object after homogeneous-current convergence"
+                ],
+            }),
+        )?;
+        write_g09_evidence(
+            context,
+            "mixed_version_writer_evidence",
+            "mixed-version-writer",
+            &["old-reader-new-writer", "new-reader-old-writer"],
+            "upgrade_compatibility_test::rolling_upgrade_from_rc2_preserves_mixed_version_contracts",
+            serde_json::json!({
+                "bucket": MIXED_BUCKET,
+                "phases": phases,
+                "objects_per_phase": objects_per_phase,
+                "assertions": [
+                    "current writer publishes objects readable by previous-release node",
+                    "previous-release writer publishes objects readable by current node",
+                    "multipart writers continue under one-current-node and one-previous-node layouts"
+                ],
+            }),
+        )?;
+    }
+
     Ok(())
 }
 
@@ -1136,6 +1336,7 @@ async fn direct_upgrade_from_previous_release_preserves_bucket_configuration() -
 async fn rollback_to_previous_release_reads_current_bucket_metadata() -> TestResult {
     init_logging();
     let previous_binary = source_binary()?;
+    let evidence_context = g09_evidence_context(&previous_binary).await?;
 
     let replication_target = FakeS3Target::start().await?;
     replication_target.create_bucket(ROLLBACK_REPLICA_BUCKET);
@@ -1204,6 +1405,54 @@ async fn rollback_to_previous_release_reads_current_bucket_metadata() -> TestRes
     );
     assert_eq!(body, post_rollback_bytes);
 
+    env.restart_server_preserving_data(vec![], &server_env).await?;
+    let current_again = env.create_s3_client();
+    assert_versioning_enabled(&current_again, ROLLBACK_BUCKET, "after rolling forward again").await?;
+    assert_default_sse_s3_encryption(&current_again, ROLLBACK_BUCKET, "after rolling forward again").await?;
+    assert_bucket_tag(&current_again, ROLLBACK_BUCKET, "after rolling forward again").await?;
+    assert_remote_target_preserved(&env, ROLLBACK_BUCKET, &target_arn, "after rolling forward again").await?;
+    assert_eq!(
+        read_object(&current_again, ROLLBACK_BUCKET, single_key, Some(&single_version))
+            .await?
+            .1,
+        single_bytes
+    );
+    assert_eq!(
+        read_object(&current_again, ROLLBACK_BUCKET, multipart_key, None).await?.1,
+        multipart_bytes
+    );
+    assert_eq!(
+        read_object(&current_again, ROLLBACK_BUCKET, post_rollback_key, None).await?.1,
+        post_rollback_bytes
+    );
+
+    if let Some(context) = evidence_context.as_ref() {
+        write_g09_evidence(
+            context,
+            "rollback_payload_evidence",
+            "rollback-payload",
+            &["rollback-to-old", "rollback-to-new", "unknown-field-retained"],
+            "upgrade_compatibility_test::rollback_to_previous_release_reads_current_bucket_metadata",
+            serde_json::json!({
+                "bucket": ROLLBACK_BUCKET,
+                "cases": {
+                    "rollback-to-old": [
+                        "previous-release binary reads current-build versioning, SSE-S3, tags, replication target and objects",
+                        "previous-release writer honors the decoded current-build encryption configuration"
+                    ],
+                    "rollback-to-new": [
+                        "current build reads the object written by the rolled-back previous release",
+                        "current build reads the current-build single-part and multipart objects after rolling forward again"
+                    ],
+                    "unknown-field-retained": [
+                        "previous release skips current-build bucket metadata extension fields without dropping known bucket configuration",
+                        "current build reads the retained bucket configuration after the previous-release round trip"
+                    ]
+                },
+            }),
+        )?;
+    }
+
     replication_target.shutdown().await;
     Ok(())
 }

crates/ecstore/src/api/mod.rs

@@ -38,6 +38,16 @@ pub mod bucket {
     }
 
     pub mod lifecycle {
+        pub mod legacy_transition_state_reconcile {
+            pub use crate::bucket::lifecycle::legacy_transition_state_reconcile::{
+                LegacyTransitionStateCopyRepresentation, LegacyTransitionStateMetadataAlias, LegacyTransitionStateReconcileError,
+                LegacyTransitionStateReconcileOutcome, LegacyTransitionStateReconcileReadiness,
+                LegacyTransitionStateReconcileRequest, LegacyTransitionStateReconcileResponse,
+                LegacyTransitionStateReconcileSelector, LegacyTransitionStateSetRepresentation, LegacyTransitionStateSource,
+                LegacyTransitionStateTarget,
+            };
+        }
+
         pub mod bucket_lifecycle_audit {
             pub use crate::bucket::lifecycle::bucket_lifecycle_audit::LcEventSrc;
         }
@@ -260,18 +270,19 @@ pub mod bucket {
             ReplicationDeleteStateSource, ReplicationHealQueueResult, ReplicationObjectBridge, ReplicationObjectIO,
             ReplicationOperation, ReplicationPoolTrait, ReplicationPriority, ReplicationQueueAdmission, ReplicationScannerBridge,
             ReplicationState, ReplicationStats, ReplicationStatusType, ReplicationStorage, ReplicationTargetValidationError,
-            ReplicationType, ResyncOpts, ResyncStatusType, RuntimeReplicationTargetBacklog, TargetReplicationResyncStatus,
+            ReplicationType, ResyncOpts, ResyncStatusType, RuntimeReplicationTargetBacklog, ScannerDirtyUsageMutationObserver,
-            VersionPurgeStatusType, XferStats, assign_site_replication_rule_priorities, commit_force_delete_intent,
+            ScannerDirtyUsageMutationSource, TargetReplicationResyncStatus, VersionPurgeStatusType, XferStats,
-            complete_force_delete_intent, delete_replication_state_from_config, delete_replication_version_id,
+            assign_site_replication_rule_priorities, commit_force_delete_intent, complete_force_delete_intent,
-            get_global_replication_pool, get_global_replication_stats, get_proxy_targets, init_background_replication,
+            delete_replication_state_from_config, delete_replication_version_id, get_global_replication_pool,
+            get_global_replication_stats, get_proxy_targets, init_background_replication,
             invalid_replication_config_status_field, is_site_replication_role, is_site_replication_rule,
             merge_incoming_replication_config, merge_user_replication_config, persist_force_delete_intent,
             read_durable_mrf_backlog, replication_state_to_filemeta, replication_status_to_filemeta, replication_statuses_map,
             replication_target_arn_deployment_id, replication_target_arns, resync_start_conflict_id,
-            should_remove_replication_target, should_schedule_delete_replication, should_use_existing_delete_replication_info,
+            set_scanner_dirty_usage_mutation_observer, should_remove_replication_target, should_schedule_delete_replication,
-            should_use_existing_delete_replication_source, site_replication_rule_deployment_id,
+            should_use_existing_delete_replication_info, should_use_existing_delete_replication_source,
-            unsupported_replication_config_field, validate_replication_config_structure, validate_replication_config_target_arns,
+            site_replication_rule_deployment_id, unsupported_replication_config_field, validate_replication_config_structure,
-            version_purge_status_to_filemeta,
+            validate_replication_config_target_arns, version_purge_status_to_filemeta,
         };
     }
 
@@ -366,6 +377,14 @@ pub mod config {
 }
 
 pub mod data_usage {
+    #[cfg(feature = "test-util")]
+    pub use crate::data_movement::SourceCleanupDeleteBarrier;
+    #[cfg(feature = "test-util")]
+    pub use crate::data_movement::scanner_backlog::test_util::NativeScannerPauseBacklogWriteFault;
+    pub use crate::data_movement::scanner_backlog::{
+        MAX_SCANNER_PAUSE_BACKLOG_BYTES, ScannerPauseBacklogRetirementPlan, ScannerPauseBacklogRetirementPlanner,
+        ScannerPauseBacklogRetirementReplica, register_scanner_pause_backlog_retirement_planner,
+    };
     pub use crate::data_usage::{
         DATA_USAGE_CACHE_NAME, apply_bucket_usage_memory_overlay, compute_bucket_usage,
         init_compression_total_memory_from_backend, invalidate_admin_data_usage_snapshot_cache,

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

@@ -966,6 +966,11 @@ async fn cleanup_free_version_exact(api: Arc<ECStore>, oi: &ObjectInfo, cancel:
     if let Some(err) = first_error {
         return Err(err);
     }
+    runtime_sources::notify_scanner_dirty_usage_mutation(
+        &oi.bucket,
+        &oi.name,
+        runtime_sources::ScannerDirtyUsageMutationSource::TierExpiration,
+    );
     Ok(true)
 }
 
@@ -4749,6 +4754,11 @@ async fn expire_transitioned_object_with_lock_lost_signal(
                 // Drop any cached restored-copy body so it does not sit resident
                 // until TTL after the copy is expired (ODC-26).
                 crate::object_api::notify_object_mutation(&oi.bucket, &oi.name).await;
+                runtime_sources::notify_scanner_dirty_usage_mutation(
+                    &oi.bucket,
+                    &oi.name,
+                    runtime_sources::ScannerDirtyUsageMutationSource::TierExpiration,
+                );
                 //audit_log_lifecycle(*oi, ILMExpiry, tags, traceFn);
                 Ok(dobj)
             }
@@ -4784,6 +4794,11 @@ async fn expire_transitioned_object_with_lock_lost_signal(
     // The transitioned version is gone; evict any cached body for this object
     // so it does not linger until TTL (ODC-26).
     crate::object_api::notify_object_mutation(&oi.bucket, &oi.name).await;
+    runtime_sources::notify_scanner_dirty_usage_mutation(
+        &oi.bucket,
+        &oi.name,
+        runtime_sources::ScannerDirtyUsageMutationSource::TierExpiration,
+    );
 
     //audit_log_lifecycle(oi, ILMExpiry, tags);
 
@@ -5165,6 +5180,9 @@ pub async fn put_restore_opts(
         user_defined: meta,
         version_id: oi.version_id.map(|e| e.to_string()),
         mod_time: oi.mod_time,
+        // Restore writes stored (possibly encrypted) bytes, so the writer's
+        // computed MD5 is not the object's public plaintext ETag.
+        preserve_etag: oi.etag.clone(),
         //expires:           oi.expires,
         ..Default::default()
     })

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

@@ -18,6 +18,7 @@ mod config_boundary;
 pub mod core;
 mod durable_namespace;
 pub mod evaluator;
+pub mod legacy_transition_state_reconcile;
 pub mod manual_transition_job;
 mod metadata_boundary;
 pub(crate) use metadata_boundary::{LifecycleExpiryConfigs, get_expiry_configs, get_lifecycle_config};

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

@@ -20,6 +20,7 @@ use tokio_util::sync::CancellationToken;
 
 use crate::bucket::lifecycle::bucket_lifecycle_ops::{ExpiryState, TransitionState};
 use crate::runtime::sources;
+pub(crate) use crate::runtime::sources::ScannerDirtyUsageMutationSource;
 use crate::services::tier::tier::TierConfigMgr;
 use crate::store::ECStore;
 
@@ -54,3 +55,7 @@ pub(crate) fn deployment_id() -> Option<String> {
 pub(crate) async fn bucket_lifecycle_config(bucket: &str) -> Option<BucketLifecycleConfiguration> {
     sources::bucket_lifecycle_config(bucket).await
 }
+
+pub(crate) fn notify_scanner_dirty_usage_mutation(bucket: &str, object: &str, source: ScannerDirtyUsageMutationSource) {
+    sources::notify_scanner_dirty_usage_mutation(bucket, object, source);
+}

crates/ecstore/src/bucket/metadata_sys.rs

@@ -1681,9 +1681,13 @@ impl BucketMetadataSys {
         expected: Option<&Arc<BucketMetadata>>,
         namespace_guard: &rustfs_lock::NamespaceLockGuard,
     ) -> Result<()> {
-        if !self
+        if !await_bucket_namespace_operation(
-            .bucket_exists(bucket, namespace_guard, "bucket metadata existence check")
+            Some(namespace_guard),
-            .await?
+            bucket,
+            "bucket metadata heal existence check",
+            self.api.bucket_exists_for_heal(bucket),
+        )
+        .await?
         {
             if matches!(mode, MetadataLoadMode::Refresh) {
                 let _publish_guard = self

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

@@ -92,3 +92,6 @@ pub use replication_target_boundary::SsecPassthroughCapability;
 pub use replication_target_boundary::VersionIdentityCapability;
 pub use replication_target_boundary::{ObjectLockIntegrity, object_lock_put_integrity};
 pub(crate) use replication_target_config_bridge::ReplicationTargetConfigBridge;
+pub use runtime_boundary::{
+    ScannerDirtyUsageMutationObserver, ScannerDirtyUsageMutationSource, set_scanner_dirty_usage_mutation_observer,
+};

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

@@ -3843,6 +3843,13 @@ async fn persist_replication_state_if_current<S: ReplicationStorage>(
     match storage.put_object_metadata(&roi.bucket, &roi.name, &write_opts).await {
         Ok(updated) => {
             *object_info = updated;
+            if mode == ReplicationStatusWritebackMode::Update {
+                runtime_sources::notify_scanner_dirty_usage_mutation(
+                    &roi.bucket,
+                    &roi.name,
+                    runtime_sources::ScannerDirtyUsageMutationSource::Replication,
+                );
+            }
             Ok(ReplicationStatePersistOutcome::Updated)
         }
         Err(Error::PreconditionFailed) => Ok(ReplicationStatePersistOutcome::Superseded),

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

@@ -19,6 +19,9 @@ use super::replication_pool::DynReplicationPool;
 use super::replication_state::ReplicationStats;
 use super::replication_storage_boundary::ReplicationObjectStore;
 use crate::runtime::sources;
+pub use crate::runtime::sources::{
+    ScannerDirtyUsageMutationObserver, ScannerDirtyUsageMutationSource, set_scanner_dirty_usage_mutation_observer,
+};
 
 pub(crate) fn object_store_handle() -> Option<Arc<ReplicationObjectStore>> {
     sources::object_store_handle()
@@ -43,3 +46,7 @@ pub(crate) fn replication_runtime_initialized() -> bool {
 pub(crate) fn bucket_monitor() -> Option<Arc<ReplicationBucketMonitor>> {
     sources::bucket_monitor()
 }
+
+pub(crate) fn notify_scanner_dirty_usage_mutation(bucket: &str, object: &str, source: ScannerDirtyUsageMutationSource) {
+    sources::notify_scanner_dirty_usage_mutation(bucket, object, source);
+}

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

@@ -51,10 +51,10 @@ use rustfs_protos::proto_gen::node_service::{
     LocalStorageInfoRequest, Mss, ReloadPoolMetaRequest, ReloadSiteReplicationConfigRequest, ReplacementRecoveryStatusRequest,
     ScannerActivityRequest, ScannerActivityResponse, ScannerDirtyUsageSnapshotRequest, ScannerDirtyUsageSnapshotResponse,
     ScannerPublicationLeaseReleaseRequest, ScannerPublicationLeaseRequest, ScannerPublicationLeaseResponse,
-    ScannerScopedDirtyUsageAckRequest, ScannerScopedDirtyUsageEntry, ServerInfoRequest, SignalServiceRequest,
+    ScannerScopedDirtyUsageAckRequest, ScannerScopedDirtyUsageAckResponse, ScannerScopedDirtyUsageEntry, ServerInfoRequest,
-    SignalServiceResponse, StartDecommissionRequest, StartProfilingRequest, StopRebalanceRequest, TierDailyStatsRequest,
+    SignalServiceRequest, SignalServiceResponse, StartDecommissionRequest, StartProfilingRequest, StopRebalanceRequest,
-    TierMutationAbortRequest, TierMutationCommitRequest, TierMutationControlResponse, TierMutationFailureClass,
+    TierDailyStatsRequest, TierMutationAbortRequest, TierMutationCommitRequest, TierMutationControlResponse,
-    TierMutationPeerState, TierMutationPrepareRequest, node_service_client::NodeServiceClient,
+    TierMutationFailureClass, TierMutationPeerState, TierMutationPrepareRequest, node_service_client::NodeServiceClient,
     tier_mutation_control_service_client::TierMutationControlServiceClient,
 };
 pub use rustfs_protos::{PEER_RESTDRY_RUN, PEER_RESTSIGNAL, PEER_RESTSUB_SYS};
@@ -289,6 +289,20 @@ fn scanner_scoped_dirty_usage_ack_payload(
     Ok(payload)
 }
 
+fn scanner_scoped_dirty_usage_ack_response_matches(
+    request: &ScannerScopedDirtyUsageAckRequest,
+    response: &ScannerScopedDirtyUsageAckResponse,
+) -> bool {
+    let cleared_within_request = u64::try_from(request.entries.len())
+        .is_ok_and(|entry_count| response.cleared <= entry_count && (!request.probe_only || response.cleared == 0));
+    response.protocol_version == rustfs_protos::scoped_dirty_usage::SCOPED_DIRTY_USAGE_PROTOCOL_VERSION
+        && response.owner_id == request.owner_id
+        && response.instance_id == request.instance_id
+        && response.max_entries == rustfs_protos::scoped_dirty_usage::SCOPED_DIRTY_USAGE_MAX_ENTRIES
+        && response.max_request_bytes == rustfs_protos::scoped_dirty_usage::SCOPED_DIRTY_USAGE_MAX_REQUEST_BYTES
+        && cleared_within_request
+}
+
 fn scanner_scoped_dirty_usage_ack_reconciled(activity: &ScannerPeerActivity, expected_instance_id: &str) -> bool {
     activity.instance_id == expected_instance_id && activity.dirty_usage_pending == Some(false)
 }
@@ -2255,13 +2269,7 @@ impl PeerRestClient {
                     let body = canonical_scoped_dirty_usage_response(&canonical, &response)
                         .map_err(|_| Error::other("scoped dirty usage capability response is too large"))?;
                     verify_tonic_rpc_response_proof(&body, response.response_proof.as_ref())?;
-                    if response.protocol_version != SCOPED_DIRTY_USAGE_PROTOCOL_VERSION
+                    if !scanner_scoped_dirty_usage_ack_response_matches(&payload, &response) {
-                        || response.owner_id != payload.owner_id
-                        || response.instance_id != payload.instance_id
-                        || response.max_entries != SCOPED_DIRTY_USAGE_MAX_ENTRIES
-                        || response.max_request_bytes != SCOPED_DIRTY_USAGE_MAX_REQUEST_BYTES
-                        || response.cleared != 0
-                    {
                         return Err(Error::other("scoped dirty usage capability response does not match request"));
                     }
                     if !response.supported {
@@ -2297,13 +2305,7 @@ impl PeerRestClient {
                 let body = canonical_scoped_dirty_usage_response(&canonical, &response)
                     .map_err(|_| Error::other("scoped dirty usage acknowledgement response is too large"))?;
                 verify_tonic_rpc_response_proof(&body, response.response_proof.as_ref())?;
-                if response.protocol_version != SCOPED_DIRTY_USAGE_PROTOCOL_VERSION
+                if !scanner_scoped_dirty_usage_ack_response_matches(&payload, &response) || !response.supported {
-                    || response.owner_id != payload.owner_id
-                    || response.instance_id != payload.instance_id
-                    || response.max_entries != SCOPED_DIRTY_USAGE_MAX_ENTRIES
-                    || response.max_request_bytes != SCOPED_DIRTY_USAGE_MAX_REQUEST_BYTES
-                    || !response.supported
-                {
                     return Err(Error::other("scoped dirty usage acknowledgement response does not match request"));
                 }
             }
@@ -3118,6 +3120,60 @@ mod tests {
         }
     }
 
+    #[test]
+    fn scanner_scoped_dirty_usage_ack_response_bounds_cleared_entries_to_request() {
+        use rustfs_protos::scoped_dirty_usage::{
+            SCOPED_DIRTY_USAGE_MAX_ENTRIES, SCOPED_DIRTY_USAGE_MAX_REQUEST_BYTES, SCOPED_DIRTY_USAGE_PROTOCOL_VERSION,
+        };
+
+        let mut request = scanner_scoped_dirty_usage_ack_payload(
+            "33333333-3333-3333-3333-333333333333",
+            "0123456789abcdef0123456789abcdef",
+            false,
+            vec![
+                ScannerScopedDirtyUsageEntry {
+                    bucket: "archive".to_string(),
+                    bucket_incarnation: Uuid::from_u128(0x11111111111111111111111111111111).as_bytes().to_vec().into(),
+                    generation: 3,
+                },
+                ScannerScopedDirtyUsageEntry {
+                    bucket: "photos".to_string(),
+                    bucket_incarnation: Uuid::from_u128(0x22222222222222222222222222222222).as_bytes().to_vec().into(),
+                    generation: 7,
+                },
+            ],
+        )
+        .expect("two ordered entries should form a valid scoped ACK request");
+        let mut response = ScannerScopedDirtyUsageAckResponse {
+            protocol_version: SCOPED_DIRTY_USAGE_PROTOCOL_VERSION,
+            owner_id: request.owner_id.clone(),
+            instance_id: request.instance_id.clone(),
+            supported: true,
+            max_entries: SCOPED_DIRTY_USAGE_MAX_ENTRIES,
+            max_request_bytes: SCOPED_DIRTY_USAGE_MAX_REQUEST_BYTES,
+            cleared: 1,
+            response_proof: Bytes::new(),
+        };
+
+        assert!(scanner_scoped_dirty_usage_ack_response_matches(&request, &response));
+        response.cleared = 2;
+        assert!(scanner_scoped_dirty_usage_ack_response_matches(&request, &response));
+        response.cleared = 3;
+        assert!(
+            !scanner_scoped_dirty_usage_ack_response_matches(&request, &response),
+            "a peer cannot clear more entries than the signed request contains"
+        );
+
+        request.probe_only = true;
+        response.cleared = 1;
+        assert!(
+            !scanner_scoped_dirty_usage_ack_response_matches(&request, &response),
+            "a capability probe cannot report a mutation"
+        );
+        response.cleared = 0;
+        assert!(scanner_scoped_dirty_usage_ack_response_matches(&request, &response));
+    }
+
     #[test]
     fn scanner_scoped_dirty_usage_ack_reconciliation_requires_same_clean_instance() {
         let activity = |instance_id: &str, pending| ScannerPeerActivity {

crates/ecstore/src/core/pools_test.rs

@@ -5256,16 +5256,35 @@ mod decommission_lock_order_tests {
 
     #[test]
     #[serial_test::serial]
-    fn scanner_backlog_native_replica_reconciles_capacity_and_cleans_source() {
+    fn data_movement_existing_replica_reconciles_capacity_and_cleans_source() {
-        run_large_stack_current_thread_async_test("scanner-backlog-reconcile", async || {
+        data_movement_existing_replica_reconciles_capacity_case(false);
+    }
+
+    #[test]
+    #[serial_test::serial]
+    fn data_movement_existing_replica_outside_reservation_uses_reserved_target() {
+        data_movement_existing_replica_reconciles_capacity_case(true);
+    }
+
+    fn data_movement_existing_replica_reconciles_capacity_case(existing_outside_reservation: bool) {
+        run_large_stack_current_thread_async_test("reserved-replica-reconcile", async move || {
             let (_temp_dirs, store, other_store) =
                 test_three_pool_stores_with_three_disk_sets_with_isolated_node_contexts(None).await;
-            let object = "buckets/.scanner-pause-backlog.json";
+            let object = "buckets/reserved-replica-routing.json";
             let body = br#"{"schemaVersion":1,"generation":2}"#.to_vec();
             let old_body = br#"{"schemaVersion":1,"generation":1}"#.to_vec();
             let source_time = time::OffsetDateTime::UNIX_EPOCH + time::Duration::seconds(20);
-            let target_time = time::OffsetDateTime::UNIX_EPOCH + time::Duration::seconds(10);
+            let target_time = source_time;
-            for (pool_index, payload, mod_time) in [(0, body.clone(), source_time), (2, old_body, target_time)] {
+            let target_pool_index = if existing_outside_reservation { 1 } else { 2 };
+            let mut replicas = vec![(0, body.clone(), source_time), (target_pool_index, old_body, target_time)];
+            if existing_outside_reservation {
+                replicas.push((
+                    2,
+                    br#"{"schemaVersion":1,"generation":3}"#.to_vec(),
+                    source_time + time::Duration::seconds(10),
+                ));
+            }
+            for (pool_index, payload, mod_time) in replicas.iter().cloned() {
                 store.pools[pool_index]
                     .put_object(
                         RUSTFS_META_BUCKET,
@@ -5278,13 +5297,19 @@ mod decommission_lock_order_tests {
                         },
                     )
                     .await
-                    .expect("seed native scanner replicas with independent write times");
+                    .expect("seed existing replicas with independent write times");
             }
             let layout = DecommissionErasureLayout { data: 1, parity: 0 };
             let target_total = body.len() * 8;
             let capacities = vec![
                 DecommissionPoolCapacityInfo::for_test(0, layout, 0, body.len() * 2, body.len() * 2),
-                DecommissionPoolCapacityInfo::for_test(1, layout, 0, target_total, target_total),
+                DecommissionPoolCapacityInfo::for_test(
+                    1,
+                    layout,
+                    if existing_outside_reservation { target_total } else { 0 },
+                    target_total,
+                    if existing_outside_reservation { 0 } else { target_total },
+                ),
                 DecommissionPoolCapacityInfo::for_test(2, layout, target_total, target_total, 0),
             ];
             set_decommission_capacity_info_overrides_for_test(store.id, vec![capacities.clone()]);
@@ -5293,6 +5318,17 @@ mod decommission_lock_order_tests {
                 .await
                 .expect("activate the source reservation");
             let owner = decommission_capacity_owner(&*store.pool_meta.read().await);
+            let reserved_snapshot = store.pool_meta.read().await.clone();
+            let reservation = reserved_snapshot.pools[0]
+                .decommission
+                .as_ref()
+                .and_then(|info| info.capacity_reservation.as_ref())
+                .expect("active source reservation");
+            assert_eq!(
+                reservation.targets.iter().map(|target| target.pool_index).collect::<Vec<_>>(),
+                vec![target_pool_index],
+                "the fixture must reserve exactly one target"
+            );
             let source_reader = store.pools[0]
                 .get_object_reader(
                     RUSTFS_META_BUCKET,
@@ -5314,12 +5350,91 @@ mod decommission_lock_order_tests {
                 RUSTFS_META_BUCKET.to_string(),
                 source_reader,
                 None,
-                "scanner_backlog_conflict",
+                "reserved_replica_conflict",
                 Some(owner),
             )
             .await
-            .expect_err("a different older native ledger must retain its source and capacity intent");
+            .expect_err("a different older existing record must retain its source and capacity intent");
             assert!(conflict.to_string().contains("Precondition failed"), "unexpected conflict: {conflict}");
+            let reserved_snapshot = store.pool_meta.read().await.clone();
+            let mut selection_opts = ObjectOptions {
+                data_movement: true,
+                src_pool_idx: 0,
+                ..Default::default()
+            };
+            assert_eq!(
+                store
+                    .select_data_movement_pool_idx(RUSTFS_META_BUCKET, object, body.len() as i64, &selection_opts, true)
+                    .await
+                    .expect("selection without a capacity owner retains existing-replica routing"),
+                2
+            );
+            owner.apply_to(&mut selection_opts);
+            for stale_owner in [
+                DecommissionCapacityOwner {
+                    owner_nonce: uuid::Uuid::new_v4(),
+                    ..owner
+                },
+                DecommissionCapacityOwner {
+                    generation: owner.generation + 1,
+                    ..owner
+                },
+            ] {
+                let mut stale_opts = selection_opts.clone();
+                stale_owner.apply_to(&mut stale_opts);
+                assert!(
+                    matches!(
+                        store
+                            .select_data_movement_pool_idx(RUSTFS_META_BUCKET, object, body.len() as i64, &stale_opts, true)
+                            .await,
+                        Err(crate::error::Error::DecommissionCapacityBlocked { .. })
+                    ),
+                    "a stale owner must not fall back to another target"
+                );
+            }
+            {
+                let mut meta = store.pool_meta.write().await;
+                meta.pools[0]
+                    .decommission
+                    .as_mut()
+                    .unwrap()
+                    .capacity_reservation
+                    .as_mut()
+                    .unwrap()
+                    .expires_at = time::OffsetDateTime::now_utc() - time::Duration::seconds(1);
+            }
+            assert!(
+                matches!(
+                    store
+                        .select_data_movement_pool_idx(RUSTFS_META_BUCKET, object, body.len() as i64, &selection_opts, true)
+                        .await,
+                    Err(crate::error::Error::DecommissionCapacityBlocked { .. })
+                ),
+                "an expired owner must not fall back to another target"
+            );
+            *store.pool_meta.write().await = reserved_snapshot.clone();
+            if !existing_outside_reservation {
+                {
+                    let mut meta = store.pool_meta.write().await;
+                    let target = &mut meta.pools[0]
+                        .decommission
+                        .as_mut()
+                        .unwrap()
+                        .capacity_reservation
+                        .as_mut()
+                        .unwrap()
+                        .targets[0];
+                    target.consumed_physical_bytes = target.reserved_physical_bytes;
+                }
+                assert_eq!(
+                    store
+                        .select_data_movement_pool_idx(RUSTFS_META_BUCKET, object, body.len() as i64, &selection_opts, true)
+                        .await
+                        .expect("an existing reserved replica can still be selected after capacity was consumed"),
+                    target_pool_index
+                );
+                *store.pool_meta.write().await = reserved_snapshot;
+            }
             let mut persisted = crate::core::pools::PoolMeta::default();
             persisted
                 .load_no_lock_from_replicas(store.pools.clone())
@@ -5336,11 +5451,24 @@ mod decommission_lock_order_tests {
                     .pending_target_physical_bytes,
                 body.len()
             );
-            let previous = store.pools[2]
+            for (pool_index, payload, mod_time) in &replicas {
+                let mut reader = store.pools[*pool_index]
+                    .get_object_reader(RUSTFS_META_BUCKET, object, None, HeaderMap::new(), &ObjectOptions::default())
+                    .await
+                    .expect("a refused existing record replacement must preserve every replica");
+                assert_eq!(reader.object_info.mod_time, Some(*mod_time));
+                let mut actual = Vec::new();
+                reader
+                    .read_to_end(&mut actual)
+                    .await
+                    .expect("read the unchanged existing record");
+                assert_eq!(&actual, payload);
+            }
+            let previous = store.pools[target_pool_index]
                 .get_object_info(RUSTFS_META_BUCKET, object, &ObjectOptions::default())
                 .await
-                .expect("read the native writer's CAS revision");
+                .expect("read the existing writer's CAS revision");
-            let replacement = store.pools[2]
+            let replacement = store.pools[target_pool_index]
                 .put_object(
                     RUSTFS_META_BUCKET,
                     object,
@@ -5356,7 +5484,7 @@ mod decommission_lock_order_tests {
                     },
                 )
                 .await
-                .expect("native scanner CAS converges the payload without a migration marker");
+                .expect("existing CAS converges the payload without a migration marker");
             assert!(!data_movement::is_owned_data_movement_target(&replacement));
             *other_store.pool_meta.write().await = persisted;
             set_decommission_capacity_info_overrides_for_test(other_store.id, vec![capacities]);
@@ -5374,7 +5502,7 @@ mod decommission_lock_order_tests {
             )
             .await
             .expect("replica conflict recovery must be bounded")
-            .expect("identical native replica should finish migration on the reloaded node");
+            .expect("identical existing replica should finish migration on the reloaded node");
             let mut reconciled = crate::core::pools::PoolMeta::default();
             reconciled
                 .load_no_lock_from_replicas(other_store.pools.clone())
@@ -5404,14 +5532,14 @@ mod decommission_lock_order_tests {
                 .await
                 .expect_err("the source should be cleaned only after equivalent-target capacity reconciliation");
             assert!(crate::error::is_err_object_not_found(&missing));
-            let mut target_reader = other_store.pools[2]
+            let mut target_reader = other_store.pools[target_pool_index]
                 .get_object_reader(RUSTFS_META_BUCKET, object, None, HeaderMap::new(), &ObjectOptions::default())
                 .await
                 .expect("the surviving replica should remain readable");
             assert_eq!(
                 target_reader.object_info.mod_time,
                 Some(target_time),
-                "recovery must not overwrite the native target"
+                "recovery must not overwrite the existing target"
             );
             let mut actual = Vec::new();
             target_reader
@@ -5419,6 +5547,20 @@ mod decommission_lock_order_tests {
                 .await
                 .expect("read surviving ledger bytes");
             assert_eq!(actual, body);
+            if existing_outside_reservation {
+                let (_, outside_body, outside_time) = replicas.last().expect("unreserved existing replica");
+                let mut outside = other_store.pools[2]
+                    .get_object_reader(RUSTFS_META_BUCKET, object, None, HeaderMap::new(), &ObjectOptions::default())
+                    .await
+                    .expect("migration must leave the unreserved existing replica intact");
+                assert_eq!(outside.object_info.mod_time, Some(*outside_time));
+                let mut actual = Vec::new();
+                outside
+                    .read_to_end(&mut actual)
+                    .await
+                    .expect("read the untouched unreserved replica");
+                assert_eq!(&actual, outside_body);
+            }
         });
     }
 

crates/ecstore/src/data_movement/mod.rs

@@ -15,6 +15,7 @@
 // #730: data-movement migration keeps staged cleanup helpers until copy paths converge.
 
 pub(crate) mod backpressure;
+pub(crate) mod scanner_backlog;
 
 use crate::core::pools::{DecommissionCapacityOwner, decommission_capacity_mutation_id};
 use crate::error::{
@@ -984,24 +985,6 @@ fn is_superseding_unversioned_data_movement_object(source: &ObjectInfo, target:
             .is_some_and(|(source_time, target_time)| target_time > source_time)
 }
 
-fn is_equivalent_scanner_backlog_replica(source: &ObjectInfo, target: &ObjectInfo, compare_part_checksums: bool) -> bool {
-    // Scanner publishes this exact payload to surviving sets with CAS. Each
-    // set assigns its own write time; that timestamp is not a ledger generation.
-    // Accept only an identical, known unversioned identity, never a different
-    // record based on timestamp ordering or a similarly named user object.
-    source.bucket == crate::disk::RUSTFS_META_BUCKET
-        && target.bucket == source.bucket
-        && source.name == "buckets/.scanner-pause-backlog.json"
-        && target.name == source.name
-        && is_unversioned_data_movement_object(source)
-        && is_unversioned_data_movement_object(target)
-        && !source.delete_marker
-        && source.mod_time.is_some()
-        && target.mod_time.is_some()
-        && source.etag.as_ref().is_some_and(|etag| !etag.is_empty())
-        && is_equivalent_data_movement_object_identity(source, target, false, compare_part_checksums)
-}
-
 fn is_data_movement_upload_takeover_target(source: &ObjectInfo, target: &ObjectInfo, compare_part_checksums: bool) -> bool {
     let identity = data_movement_upload_identity(source);
     source.mod_time.is_some()
@@ -1217,7 +1200,7 @@ struct SourceCleanupDeleteBarrierState {
     dead_code,
     reason = "installed by set_disk object tests behind `--features test-util` (backlog#1823)"
 )]
-pub(crate) struct SourceCleanupDeleteBarrier {
+pub struct SourceCleanupDeleteBarrier {
     state: Arc<SourceCleanupDeleteBarrierState>,
 }
 
@@ -1231,7 +1214,7 @@ static SOURCE_CLEANUP_DELETE_BARRIERS: std::sync::OnceLock<std::sync::Mutex<Vec<
     reason = "installed by set_disk object tests behind `--features test-util` (backlog#1823)"
 )]
 impl SourceCleanupDeleteBarrier {
-    pub(crate) fn install(bucket: &str, object: &str) -> Self {
+    pub fn install(bucket: &str, object: &str) -> Self {
         let state = Arc::new(SourceCleanupDeleteBarrierState {
             bucket: bucket.to_string(),
             object: object.to_string(),
@@ -1254,7 +1237,7 @@ impl SourceCleanupDeleteBarrier {
         Self { state }
     }
 
-    pub(crate) async fn wait_until_paused(&self) {
+    pub async fn wait_until_paused(&self) {
         tokio::time::timeout(StdDuration::from_secs(30), self.state.arrived.notified())
             .await
             .expect("source cleanup should reach the pre-delete barrier");
@@ -1270,7 +1253,7 @@ impl SourceCleanupDeleteBarrier {
         self.state.is_paused.load(Ordering::Acquire)
     }
 
-    pub(crate) fn release(&self) {
+    pub fn release(&self) {
         self.state.release.notify_one();
     }
 }
@@ -1449,7 +1432,8 @@ fn resolve_data_movement_overwrite_resume_result_for(
     target_pool_idx: usize,
     compare_part_checksums: bool,
 ) -> Result<bool> {
-    if !should_check_data_movement_overwrite_resume(err)
+    if scanner_backlog::is_scanner_pause_backlog(&source.bucket, &source.name)
+        || !should_check_data_movement_overwrite_resume(err)
         || !should_check_data_movement_resume_target(src_pool_idx, target_pool_idx)
     {
         return Ok(false);
@@ -1471,9 +1455,7 @@ fn resolve_data_movement_overwrite_resume_result_for(
         return Ok(true);
     }
 
-    Ok(matches!(err, Error::PreconditionFailed)
+    Ok(matches!(err, Error::PreconditionFailed) && is_superseding_unversioned_data_movement_object(source, &target))
-        && (is_equivalent_scanner_backlog_replica(source, &target, compare_part_checksums)
-            || is_superseding_unversioned_data_movement_object(source, &target)))
 }
 
 #[derive(Clone, Copy)]
@@ -1521,9 +1503,27 @@ fn data_movement_part_stage_error(
     bucket: &str,
     object: &str,
     part_number: usize,
-    err: impl std::fmt::Display,
+    err: Error,
 ) -> Error {
-    Error::other(format!("{op_label}: {stage} failed for {bucket}/{object} part {part_number}: {err}"))
+    let rendered = format!("{op_label}: {stage} failed for {bucket}/{object} part {part_number}: {err}");
+    if matches!(&err, Error::DecommissionCapacityBlocked { .. }) {
+        return data_movement_context_error(rendered, err);
+    }
+    // A missing target part is not evidence that the source can be deleted.
+    // Keep other part errors opaque to the source-cleanup classifiers.
+    Error::other(rendered)
+}
+
+#[cfg(test)]
+pub(crate) fn data_movement_part_stage_error_for_test(
+    op_label: &str,
+    stage: &str,
+    bucket: &str,
+    object: &str,
+    part_number: usize,
+    err: Error,
+) -> Error {
+    data_movement_part_stage_error(op_label, stage, bucket, object, part_number, err)
 }
 
 fn is_data_movement_part_read_error(err: &Error) -> bool {
@@ -1628,6 +1628,9 @@ async fn migrate_object_inner(
     capacity_owner: Option<DecommissionCapacityOwner>,
     mutation_fence: Option<DecommissionFixedReadAnchor>,
 ) -> Result<()> {
+    if scanner_backlog::is_scanner_pause_backlog(&bucket, &rd.object_info.name) {
+        return Err(Error::other("scanner pause backlog requires native retirement handoff"));
+    }
     let mut mutation_fence = mutation_fence;
     let object_info = rd.object_info.clone();
     let capacity_owner = capacity_owner.map(|owner| {
@@ -2428,8 +2431,15 @@ mod tests {
         let err =
             data_movement_part_stage_error("rebalance_object", "put_object_part", "bucket-a", "object-a", 7, Error::SlowDown);
         let message = err.to_string();
-        assert!(message.contains("rebalance_object: put_object_part failed for bucket-a/object-a part 7"));
+        assert_eq!(
-        assert!(message.contains(Error::SlowDown.to_string().as_str()));
+            message,
+            Error::other(format!(
+                "rebalance_object: put_object_part failed for bucket-a/object-a part 7: {}",
+                Error::SlowDown
+            ))
+            .to_string()
+        );
+        assert!(data_movement_stage_source(&err).is_none());
     }
 
     #[test]
@@ -3329,16 +3339,25 @@ mod tests {
     }
 
     #[test]
-    fn test_scanner_backlog_resume_accepts_identical_native_replica_with_older_write_time() {
+    fn test_scanner_backlog_resume_requires_native_cohort_proof_even_for_identical_payload() {
         let (source, target) = scanner_backlog_replica_pair();
         assert!(!is_owned_data_movement_target(&target), "native scanner writes are not migration copies");
         assert!(!is_equivalent_data_movement_object(&source, &target));
         assert!(
-            scanner_backlog_precondition_resumes(&source, target),
+            !scanner_backlog_precondition_resumes(&source, target),
-            "identical ledger payloads have replica-local write times, not distinct committed generations"
+            "a single identical replica cannot prove native cohort authority"
         );
     }
 
+    #[test]
+    fn test_scanner_backlog_resume_rejects_newer_timestamp_and_full_single_replica_identity() {
+        let (source, mut target) = scanner_backlog_replica_pair();
+        target.mod_time = source.mod_time.map(|time| time + time::Duration::SECOND);
+        target.etag = Some("different-native-ledger".to_string());
+        assert!(!scanner_backlog_precondition_resumes(&source, target));
+        assert!(!scanner_backlog_precondition_resumes(&source, source.clone()));
+    }
+
     #[test]
     fn test_scanner_backlog_resume_rejects_changed_payload_or_metadata() {
         let (source, target) = scanner_backlog_replica_pair();

crates/ecstore/src/data_movement/scanner_backlog.rs

@@ -0,0 +1,292 @@
+// Copyright 2024 RustFS Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use crate::disk::RUSTFS_META_BUCKET;
+use crate::error::{Error, Result, is_err_object_not_found, is_err_version_not_found};
+use crate::object_api::ObjectOptions;
+use crate::object_api::{ObjectInfo, PutObjReader, WriteCompletion};
+use crate::set_disk::SetDisks;
+use crate::storage_api_contracts::object::HTTPPreconditions;
+use crate::storage_api_contracts::object::ObjectIO as _;
+use futures::future::join_all;
+use http::HeaderMap;
+use std::sync::{Arc, OnceLock};
+use tokio::io::AsyncReadExt;
+
+pub const MAX_SCANNER_PAUSE_BACKLOG_BYTES: u64 = 64 * 1024;
+pub(crate) const SCANNER_PAUSE_BACKLOG_PATH: &str = "buckets/.scanner-pause-backlog.json";
+
+/// A bounded, storage-fenced native replica. Only a confirmed missing object
+/// has no payload; read failures never enter the Scanner verifier.
+pub struct ScannerPauseBacklogRetirementReplica {
+    pub pool_index: usize,
+    pub set_index: usize,
+    pub data: Option<Vec<u8>>,
+}
+
+/// Native records for a membership handoff. Existing durable ledgers are
+/// preserved; an empty native bootstrap may initialize its first ledger.
+pub struct ScannerPauseBacklogRetirementPlan {
+    pub seed_record: Option<Vec<u8>>,
+    pub commit_record: Vec<u8>,
+    pub stable_record: Vec<u8>,
+}
+
+pub type ScannerPauseBacklogRetirementPlanner =
+    fn(usize, &[ScannerPauseBacklogRetirementReplica]) -> std::result::Result<Option<ScannerPauseBacklogRetirementPlan>, String>;
+
+static RETIREMENT_PLANNER: OnceLock<ScannerPauseBacklogRetirementPlanner> = OnceLock::new();
+
+/// Install the stateless native record planner before storage starts workers.
+/// The scanner runtime switch does not control this storage safety check.
+pub fn register_scanner_pause_backlog_retirement_planner(planner: ScannerPauseBacklogRetirementPlanner) {
+    RETIREMENT_PLANNER.get_or_init(|| planner);
+}
+
+pub(crate) fn is_scanner_pause_backlog(bucket: &str, object: &str) -> bool {
+    bucket == RUSTFS_META_BUCKET && object == SCANNER_PAUSE_BACKLOG_PATH
+}
+
+pub(crate) struct ScannerPauseBacklogRetirementRead {
+    pub replica: ScannerPauseBacklogRetirementReplica,
+    pub etag: Option<String>,
+}
+
+impl ScannerPauseBacklogRetirementRead {
+    pub(crate) fn preconditions(&self) -> HTTPPreconditions {
+        match &self.etag {
+            Some(etag) => HTTPPreconditions {
+                if_match: Some(etag.clone()),
+                ..Default::default()
+            },
+            None => HTTPPreconditions {
+                if_none_match: Some("*".to_string()),
+                ..Default::default()
+            },
+        }
+    }
+}
+
+async fn read_replica(set: Arc<SetDisks>) -> Result<ScannerPauseBacklogRetirementRead> {
+    let mut replica = ScannerPauseBacklogRetirementReplica {
+        pool_index: set.pool_index,
+        set_index: set.set_index,
+        data: None,
+    };
+    let reader = match set
+        .get_object_reader(
+            RUSTFS_META_BUCKET,
+            SCANNER_PAUSE_BACKLOG_PATH,
+            None,
+            HeaderMap::new(),
+            &ObjectOptions {
+                no_lock: true,
+                ..Default::default()
+            },
+        )
+        .await
+    {
+        Ok(reader) => reader,
+        Err(err) if is_err_object_not_found(&err) || is_err_version_not_found(&err) => {
+            return Ok(ScannerPauseBacklogRetirementRead { replica, etag: None });
+        }
+        Err(err) => return Err(err),
+    };
+    let info = &reader.object_info;
+    if info.version_id.is_some_and(|version| !version.is_nil())
+        || info.delete_marker
+        || info.is_dir
+        || info.etag.as_ref().is_none_or(String::is_empty)
+        || info.size < 0
+        || info.size > MAX_SCANNER_PAUSE_BACKLOG_BYTES as i64
+    {
+        return Err(Error::other("scanner pause backlog retirement found an unsupported replica identity"));
+    }
+    let etag = info.etag.clone();
+    let expected_size = info.size as usize;
+    let mut data = Vec::new();
+    reader
+        .take(MAX_SCANNER_PAUSE_BACKLOG_BYTES + 1)
+        .read_to_end(&mut data)
+        .await?;
+    if data.len() != expected_size || data.len() > MAX_SCANNER_PAUSE_BACKLOG_BYTES as usize {
+        return Err(Error::other("scanner pause backlog retirement replica has an invalid payload length"));
+    }
+    replica.data = Some(data);
+    Ok(ScannerPauseBacklogRetirementRead { replica, etag })
+}
+
+/// The caller retains the fixed object write lock and durable topology read
+/// fence through both this snapshot and physical source cleanup.
+pub(crate) async fn read_scanner_pause_backlog_retirement_replicas(
+    source_pool_index: usize,
+    source_set_index: usize,
+    sets: Vec<Arc<SetDisks>>,
+) -> Result<Vec<ScannerPauseBacklogRetirementRead>> {
+    let replicas = join_all(sets.into_iter().map(read_replica))
+        .await
+        .into_iter()
+        .collect::<Result<Vec<_>>>()?;
+    if !replicas.iter().any(|read| {
+        read.replica.pool_index == source_pool_index && read.replica.set_index == source_set_index && read.replica.data.is_some()
+    }) {
+        return Err(Error::other("scanner pause backlog retirement current source replica is missing"));
+    }
+    Ok(replicas)
+}
+
+pub(crate) fn plan_scanner_pause_backlog_retirement(
+    source_pool_index: usize,
+    replicas: &[ScannerPauseBacklogRetirementRead],
+) -> Result<Option<ScannerPauseBacklogRetirementPlan>> {
+    let planner = RETIREMENT_PLANNER
+        .get()
+        .ok_or_else(|| Error::other("scanner pause backlog native retirement planner is unavailable"))?;
+    let snapshots = replicas
+        .iter()
+        .map(|read| ScannerPauseBacklogRetirementReplica {
+            pool_index: read.replica.pool_index,
+            set_index: read.replica.set_index,
+            data: read.replica.data.clone(),
+        })
+        .collect::<Vec<_>>();
+    planner(source_pool_index, &snapshots).map_err(Error::other)
+}
+
+/// The native writer and retirement handoff use the same conditional, full-tail
+/// write. Their callers retain object and durable membership fences until return.
+pub(crate) async fn persist_native_scanner_pause_backlog_replica(
+    set: Arc<SetDisks>,
+    data: Vec<u8>,
+    preconditions: HTTPPreconditions,
+    mut opts: ObjectOptions,
+    _phase: &'static str,
+) -> Result<ObjectInfo> {
+    if data.len() > MAX_SCANNER_PAUSE_BACKLOG_BYTES as usize {
+        return Err(Error::other("scanner pause backlog exceeds its size bound"));
+    }
+    opts.max_parity = true;
+    opts.write_completion = WriteCompletion::TailDrained;
+    opts.http_preconditions = Some(preconditions);
+    #[cfg(feature = "test-util")]
+    let fault = test_util::matching_write(&set, _phase)?;
+    let result = set
+        .put_object(RUSTFS_META_BUCKET, SCANNER_PAUSE_BACKLOG_PATH, &mut PutObjReader::from_vec(data), &opts)
+        .await;
+    #[cfg(feature = "test-util")]
+    if result.is_ok()
+        && let Some(fault) = fault
+    {
+        fault.arrived.notify_one();
+        fault.release.notified().await;
+    }
+    result
+}
+
+#[cfg(feature = "test-util")]
+pub mod test_util {
+    use super::*;
+    use std::sync::Mutex;
+    use std::sync::atomic::{AtomicUsize, Ordering};
+    use tokio::sync::Notify;
+
+    #[derive(Debug, thiserror::Error)]
+    #[error("injected native scanner backlog {phase} write failure")]
+    struct InjectedWriteFailure {
+        phase: &'static str,
+    }
+
+    pub(super) struct WriteFault {
+        set: Arc<SetDisks>,
+        phase: &'static str,
+        remaining: AtomicUsize,
+        fail_before_write: bool,
+        pub(super) arrived: Notify,
+        pub(super) release: Notify,
+    }
+
+    static WRITE_FAULTS: Mutex<Vec<Arc<WriteFault>>> = Mutex::new(Vec::new());
+
+    /// Scope a one-shot fault to the actual set instance, so other stores and
+    /// concurrent tests keep using the ordinary native persistence path.
+    pub struct NativeScannerPauseBacklogWriteFault {
+        state: Arc<WriteFault>,
+    }
+
+    impl NativeScannerPauseBacklogWriteFault {
+        fn install(set: Arc<SetDisks>, phase: &'static str, nth: usize, fail_before_write: bool) -> Self {
+            assert!(nth > 0);
+            let state = Arc::new(WriteFault {
+                set,
+                phase,
+                remaining: AtomicUsize::new(nth),
+                fail_before_write,
+                arrived: Notify::new(),
+                release: Notify::new(),
+            });
+            let mut faults = WRITE_FAULTS.lock().unwrap();
+            assert!(
+                !faults
+                    .iter()
+                    .any(|fault| Arc::ptr_eq(&fault.set, &state.set) && fault.phase == phase)
+            );
+            faults.push(Arc::clone(&state));
+            Self { state }
+        }
+
+        pub fn fail_before_write(set: Arc<SetDisks>, phase: &'static str, nth: usize) -> Self {
+            Self::install(set, phase, nth, true)
+        }
+
+        pub fn pause_after_write(set: Arc<SetDisks>, phase: &'static str) -> Self {
+            Self::install(set, phase, 1, false)
+        }
+
+        pub async fn wait_until_paused(&self) {
+            self.state.arrived.notified().await;
+        }
+
+        pub fn release(&self) {
+            self.state.release.notify_one();
+        }
+    }
+
+    impl Drop for NativeScannerPauseBacklogWriteFault {
+        fn drop(&mut self) {
+            self.release();
+            WRITE_FAULTS.lock().unwrap().retain(|fault| !Arc::ptr_eq(fault, &self.state));
+        }
+    }
+
+    pub(super) fn matching_write(set: &Arc<SetDisks>, phase: &'static str) -> Result<Option<Arc<WriteFault>>> {
+        let fault = WRITE_FAULTS
+            .lock()
+            .unwrap()
+            .iter()
+            .find(|fault| Arc::ptr_eq(&fault.set, set) && fault.phase == phase)
+            .cloned();
+        let Some(fault) = fault else { return Ok(None) };
+        if fault
+            .remaining
+            .fetch_update(Ordering::AcqRel, Ordering::Acquire, |remaining| remaining.checked_sub(1))
+            != Ok(1)
+        {
+            return Ok(None);
+        }
+        if fault.fail_before_write {
+            return Err(Error::other(InjectedWriteFailure { phase }));
+        }
+        Ok(Some(fault))
+    }
+}

crates/ecstore/src/data_usage/mod.rs

@@ -3397,6 +3397,7 @@ mod tests {
             scan_plan_digest: Some([1; 32]),
             complete: false,
             tombstone: false,
+            segment_invalidation_proof: None,
         }];
         partial.buckets_usage.insert(
             "bucket".to_string(),
@@ -3469,6 +3470,7 @@ mod tests {
                 scan_plan_digest: Some([1; 32]),
                 complete: true,
                 tombstone: false,
+                segment_invalidation_proof: None,
             }],
             ..Default::default()
         };

crates/ecstore/src/disk/error.rs

@@ -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]

crates/ecstore/src/disk/error_reduce.rs

@@ -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");

crates/ecstore/src/disk/local.rs

@@ -22543,6 +22543,112 @@ mod test {
         );
     }
 
+    #[cfg(unix)]
+    #[tokio::test]
+    async fn conditional_mrf_manifest_storage_full_keeps_recovery_anchors() {
+        use tempfile::tempdir;
+
+        const MRF_COMMIT_MANIFEST_SLOT_0: &str = ".heal-mrf-commit.0.bin";
+        const MRF_SCOPED_JOURNAL_PATH: &str = "buckets/.heal/mrf/journal-scoped.bin";
+
+        let _mode = durability_mode_override::set(DurabilityMode::Relaxed);
+        let dir = tempdir().expect("temp dir should be created");
+        let endpoint = Endpoint::try_from(dir.path().to_str().expect("temp dir should be utf8")).expect("endpoint should parse");
+        let disk = LocalDisk::new(&endpoint, false).await.expect("local disk should be created");
+        let previous_manifest = Bytes::from_static(b"mrf-committed-manifest-v1");
+        let successor_manifest = Bytes::from_static(b"mrf-committed-manifest-v2");
+        let legacy_journal = Bytes::from_static(b"legacy-mrf-journal-records");
+
+        assert_eq!(
+            disk.compare_and_update_file(RUSTFS_META_BUCKET, MRF_COMMIT_MANIFEST_SLOT_0, None, Some(previous_manifest.clone()),)
+                .await
+                .expect("previous MRF manifest should commit"),
+            ConditionalFileUpdate::Updated
+        );
+        disk.write_all(RUSTFS_META_BUCKET, MRF_SCOPED_JOURNAL_PATH, legacy_journal.clone())
+            .await
+            .expect("legacy MRF journal should be retained");
+
+        let manifest_path = disk
+            .get_object_path(RUSTFS_META_BUCKET, MRF_COMMIT_MANIFEST_SLOT_0)
+            .expect("MRF manifest path should resolve");
+        let parent = manifest_path.parent().expect("MRF manifest path should have a parent");
+        os::fsync_dir_recorder::set_failure(parent, ErrorKind::StorageFull);
+
+        let err = disk
+            .compare_and_update_file(
+                RUSTFS_META_BUCKET,
+                MRF_COMMIT_MANIFEST_SLOT_0,
+                Some(previous_manifest.clone()),
+                Some(successor_manifest),
+            )
+            .await
+            .expect_err("storage-full fsync failure must fail the MRF manifest successor commit");
+        assert!(matches!(err, DiskError::Io(ref err) if err.kind() == ErrorKind::StorageFull));
+        assert_eq!(
+            disk.read_all(RUSTFS_META_BUCKET, MRF_COMMIT_MANIFEST_SLOT_0)
+                .await
+                .expect("previous committed MRF manifest should remain readable after storage-full rollback"),
+            previous_manifest
+        );
+        assert_eq!(
+            disk.read_all(RUSTFS_META_BUCKET, MRF_SCOPED_JOURNAL_PATH)
+                .await
+                .expect("legacy MRF journal should remain readable after storage-full manifest publication failure"),
+            legacy_journal
+        );
+    }
+
+    #[cfg(unix)]
+    #[tokio::test]
+    async fn conditional_mrf_manifest_storage_full_delete_keeps_recovery_anchors() {
+        use tempfile::tempdir;
+
+        const MRF_COMMIT_MANIFEST_SLOT_0: &str = ".heal-mrf-commit.0.bin";
+        const MRF_SCOPED_JOURNAL_PATH: &str = "buckets/.heal/mrf/journal-scoped.bin";
+
+        let _mode = durability_mode_override::set(DurabilityMode::Relaxed);
+        let dir = tempdir().expect("temp dir should be created");
+        let endpoint = Endpoint::try_from(dir.path().to_str().expect("temp dir should be utf8")).expect("endpoint should parse");
+        let disk = LocalDisk::new(&endpoint, false).await.expect("local disk should be created");
+        let committed_manifest = Bytes::from_static(b"mrf-committed-manifest-v1");
+        let legacy_journal = Bytes::from_static(b"legacy-mrf-journal-records");
+
+        assert_eq!(
+            disk.compare_and_update_file(RUSTFS_META_BUCKET, MRF_COMMIT_MANIFEST_SLOT_0, None, Some(committed_manifest.clone()),)
+                .await
+                .expect("committed MRF manifest should publish"),
+            ConditionalFileUpdate::Updated
+        );
+        disk.write_all(RUSTFS_META_BUCKET, MRF_SCOPED_JOURNAL_PATH, legacy_journal.clone())
+            .await
+            .expect("legacy MRF journal should be retained");
+
+        let manifest_path = disk
+            .get_object_path(RUSTFS_META_BUCKET, MRF_COMMIT_MANIFEST_SLOT_0)
+            .expect("MRF manifest path should resolve");
+        let parent = manifest_path.parent().expect("MRF manifest path should have a parent");
+        os::fsync_dir_recorder::set_failure(parent, ErrorKind::StorageFull);
+
+        let err = disk
+            .compare_and_update_file(RUSTFS_META_BUCKET, MRF_COMMIT_MANIFEST_SLOT_0, Some(committed_manifest.clone()), None)
+            .await
+            .expect_err("storage-full fsync failure must fail the MRF manifest cleanup delete");
+        assert!(matches!(err, DiskError::Io(ref err) if err.kind() == ErrorKind::StorageFull));
+        assert_eq!(
+            disk.read_all(RUSTFS_META_BUCKET, MRF_COMMIT_MANIFEST_SLOT_0)
+                .await
+                .expect("committed MRF manifest should be restored after failed cleanup delete"),
+            committed_manifest
+        );
+        assert_eq!(
+            disk.read_all(RUSTFS_META_BUCKET, MRF_SCOPED_JOURNAL_PATH)
+                .await
+                .expect("legacy MRF journal should remain readable after failed cleanup delete"),
+            legacy_journal
+        );
+    }
+
     #[cfg(unix)]
     #[tokio::test]
     async fn conditional_file_update_dir_fsync_failure_removes_new_file_without_anchor() {

crates/ecstore/src/disk/os.rs

@@ -452,13 +452,13 @@ pub(crate) mod windows_rename_test_hooks {
 
 /// Test-only hooks into the destination-parent walk of rename preparation.
 ///
-/// The prune race lives between two syscalls inside
+/// Pruning and Windows sharing races live between syscalls inside
 /// [`mkdir_all_below_existing_base_std`], so only an injection at that exact
 /// point reproduces it deterministically. Hooks are keyed by the absolute path
 /// of the component just opened and queued per path: a retrying preparation
 /// visits the same component again, so a test models a pruner that keeps
 /// walking upward by queueing one hook per visit.
-#[cfg(all(test, unix))]
+#[cfg(all(test, any(unix, windows)))]
 pub(crate) mod prepare_rename_test_hooks {
     use super::*;
 
@@ -4208,11 +4208,14 @@ pub(crate) fn mkdir_all_below_existing_base_std(
         let mut handles = Vec::with_capacity(capacity);
         handles.push(publication_root.directory.clone());
         let mut guard = ExistingBaseDirectoryGuard::new(handles);
-        for component in base_relative.components() {
+        let mut components = base_relative
+            .components()
+            .map(|component| (component, FILE_OPEN))
+            .chain(relative.components().map(|component| (component, FILE_OPEN_IF)))
+            .filter(|(component, _)| !matches!(component, Component::CurDir))
+            .peekable();
+        while let Some((component, disposition)) = components.next() {
             let Component::Normal(component) = component else {
-                if matches!(component, Component::CurDir) {
-                    continue;
-                }
                 return Err(io::Error::new(
                     io::ErrorKind::InvalidInput,
                     "rename base directory contains an invalid path component",
@@ -4222,40 +4225,22 @@ pub(crate) fn mkdir_all_below_existing_base_std(
                 .handles
                 .last()
                 .ok_or_else(|| io::Error::other("Windows publication root guard is empty"))?;
-            let child = open_windows_directory_component(parent, component, FILE_OPEN)?;
+            // The kernel opens the final parent for write during a relative
-            guard.handles.push(child);
+            // rename. Share writes from its first open: a temporary read-only
-        }
+            // share would block another rename into the same trash directory.
-        for component in relative.components() {
+            // Ancestors stay strict and no handle shares delete access, keeping
-            let Component::Normal(component) = component else {
+            // every retained directory identity pinned.
-                continue;
+            let share_access = if components.peek().is_none() {
+                FILE_SHARE_READ | FILE_SHARE_WRITE
+            } else {
+                FILE_SHARE_READ
             };
-            let parent = guard
+            let child = open_windows_relative_directory_component(parent, component, disposition, share_access)?;
-                .handles
-                .last()
-                .ok_or_else(|| io::Error::other("Windows base directory guard is empty"))?;
-            let child = open_windows_directory_component(parent, component, FILE_OPEN_IF)?;
             guard.handles.push(child);
-        }
+            #[cfg(test)]
-
+            if components.peek().is_none() {
-        // Windows resolves a handle-relative rename by opening the target for
+                prepare_rename_test_hooks::run_after_component_opened(dir_path);
-        // write. Keep every ancestor strict, but let that internal open share
+            }
-        // the final parent. Delete sharing remains omitted, so the retained
-        // directory entry cannot be renamed or removed during publication.
-        if guard.handles.len() > 1 {
-            let component = dir_path
-                .file_name()
-                .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidInput, "rename destination parent must have a name"))?;
-            let parent_index = guard.handles.len() - 2;
-            let parent = guard
-                .handles
-                .get(parent_index)
-                .ok_or_else(|| io::Error::other("Windows destination guard lost its parent handle"))?;
-            let rename_parent =
-                open_windows_relative_directory_component(parent, component, FILE_OPEN, FILE_SHARE_READ | FILE_SHARE_WRITE)?;
-            *guard
-                .handles
-                .last_mut()
-                .ok_or_else(|| io::Error::other("Windows destination guard is empty"))? = rename_parent;
         }
 
         Ok(guard)
@@ -5446,6 +5431,111 @@ mod tests {
         assert_eq!(std::fs::read(dst).expect("read committed metadata"), b"metadata");
     }
 
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn rename_all_concurrent_trash_renames_remove_every_rollback_directory() {
+        let temp_dir = tempdir().expect("create temp dir");
+        let trash = temp_dir.path().join(".rustfs.sys/tmp/.trash");
+        std::fs::create_dir_all(&trash).expect("create trash directory");
+        let publication_root = PublicationRoot::new(temp_dir.path()).expect("open publication root");
+        let sources: Vec<_> = (0..16)
+            .map(|index| {
+                let source = temp_dir
+                    .path()
+                    .join(format!("bucket/{index}.mp4"))
+                    .join(uuid::Uuid::new_v4().to_string());
+                std::fs::create_dir_all(&source).expect("create rollback directory");
+                std::fs::write(source.join("xl.meta.bkp"), b"rollback metadata").expect("write metadata backup");
+                source
+            })
+            .collect();
+
+        let results = futures::future::join_all(sources.iter().enumerate().map(|(index, source)| {
+            super::rename_all_ignore_missing_source(source, trash.join(index.to_string()), &trash, &publication_root)
+        }))
+        .await;
+
+        for (index, (source, result)) in sources.iter().zip(results).enumerate() {
+            result.expect("concurrent rollback cleanup must reach the shared trash directory");
+            assert!(!source.exists(), "rollback cleanup must not leave a directory in the bucket");
+            assert_eq!(
+                std::fs::read(trash.join(index.to_string()).join("xl.meta.bkp")).expect("read moved metadata backup"),
+                b"rollback metadata",
+                "trash staging must retain the complete backup"
+            );
+        }
+    }
+
+    #[cfg(windows)]
+    #[test]
+    fn windows_trash_rename_succeeds_during_concurrent_parent_preparation() {
+        use std::os::windows::fs::OpenOptionsExt;
+        use std::sync::atomic::{AtomicBool, Ordering};
+        use windows_sys::Win32::{
+            Foundation::{ERROR_SHARING_VIOLATION, GENERIC_WRITE},
+            Storage::FileSystem::{
+                DELETE, FILE_FLAG_BACKUP_SEMANTICS, FILE_FLAG_OPEN_REPARSE_POINT, FILE_SHARE_DELETE, FILE_SHARE_READ,
+                FILE_SHARE_WRITE,
+            },
+        };
+
+        // Exercise the final parent both in the existing base walk and in the
+        // creatable suffix walk. Neither may briefly deny write sharing.
+        for nested_parent in [false, true] {
+            let temp_dir = tempdir().expect("create temp dir");
+            let tmp = temp_dir.path().join(".rustfs.sys/tmp");
+            let trash = tmp.join(".trash");
+            std::fs::create_dir_all(&trash).expect("create trash directory");
+            let base = if nested_parent { &tmp } else { &trash };
+            let publication_root = PublicationRoot::new(temp_dir.path()).expect("open publication root");
+            let sources = ["first", "second"].map(|name| {
+                let source = temp_dir
+                    .path()
+                    .join("bucket")
+                    .join(name)
+                    .join(uuid::Uuid::new_v4().to_string());
+                std::fs::create_dir_all(&source).expect("create rollback directory");
+                std::fs::write(source.join("xl.meta.bkp"), name.as_bytes()).expect("write metadata backup");
+                source
+            });
+            let destinations = [trash.join("first"), trash.join("second")];
+            let first_preparation = prepare_rename_with_retry(&sources[0], &destinations[0], base, &publication_root)
+                .expect("prepare the first trash rename");
+            let first_source = sources[0].clone();
+            let first_destination = destinations[0].clone();
+            let interleaved = Arc::new(AtomicBool::new(false));
+            let interleaved_hook = Arc::clone(&interleaved);
+            prepare_rename_test_hooks::queue_after_component_opened(&trash, move || {
+                interleaved_hook.store(true, Ordering::Release);
+                rename_prepared(&first_source, &first_destination, &first_preparation)
+                    .expect("another preparation's first parent handle must allow the pending trash rename");
+            });
+
+            let second_preparation = prepare_rename_with_retry(&sources[1], &destinations[1], base, &publication_root)
+                .expect("prepare the second trash rename");
+            assert!(interleaved.load(Ordering::Acquire), "the competing parent-open window must be exercised");
+            for (path, access) in [(&tmp, GENERIC_WRITE), (&trash, DELETE)] {
+                let err = std::fs::OpenOptions::new()
+                    .access_mode(access)
+                    .share_mode(FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE)
+                    .custom_flags(FILE_FLAG_BACKUP_SEMANTICS | FILE_FLAG_OPEN_REPARSE_POINT)
+                    .open(path)
+                    .expect_err("ancestor writes and final-parent deletion must remain excluded");
+                assert_eq!(
+                    err.raw_os_error(),
+                    Some(i32::try_from(ERROR_SHARING_VIOLATION).expect("Windows error code must fit i32"))
+                );
+            }
+            rename_prepared(&sources[1], &destinations[1], &second_preparation).expect("publish the second trash rename");
+            for (index, payload) in [b"first".as_slice(), b"second".as_slice()].into_iter().enumerate() {
+                assert!(!sources[index].exists(), "both rollback directories must leave the bucket");
+                assert_eq!(
+                    std::fs::read(destinations[index].join("xl.meta.bkp")).expect("read the staged backup"),
+                    payload
+                );
+            }
+        }
+    }
+
     #[cfg(windows)]
     #[tokio::test]
     async fn windows_rename_all_supports_same_parent_publication() {

crates/ecstore/src/runtime/sources.rs

@@ -14,7 +14,7 @@
 
 use std::{
     collections::{HashMap, HashSet},
-    sync::{Arc, OnceLock},
+    sync::{Arc, LazyLock, OnceLock, RwLock as StdRwLock},
     time::SystemTime,
 };
 
@@ -57,6 +57,13 @@ use uuid::Uuid;
 const TEST_RPC_SECRET: &str = "test-rpc-secret";
 
 pub(crate) type WorkloadSnapshotProviderRef = Arc<dyn WorkloadAdmissionSnapshotProvider + Send + Sync>;
+pub type ScannerDirtyUsageMutationObserver = Arc<dyn Fn(&str, &str, ScannerDirtyUsageMutationSource) + Send + Sync + 'static>;
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum ScannerDirtyUsageMutationSource {
+    Replication,
+    TierExpiration,
+}
 
 #[derive(Clone, Default)]
 pub(crate) struct LockRegistry {
@@ -88,6 +95,8 @@ impl LockRegistry {
 }
 
 static WORKLOAD_ADMISSION_SNAPSHOT_PROVIDER: OnceLock<WorkloadSnapshotProviderRef> = OnceLock::new();
+static SCANNER_DIRTY_USAGE_MUTATION_OBSERVER: LazyLock<StdRwLock<Option<ScannerDirtyUsageMutationObserver>>> =
+    LazyLock::new(|| StdRwLock::new(None));
 
 pub(crate) fn set_workload_admission_snapshot_provider(
     provider: WorkloadSnapshotProviderRef,
@@ -99,6 +108,28 @@ pub(crate) fn workload_admission_snapshot_provider() -> Option<WorkloadSnapshotP
     WORKLOAD_ADMISSION_SNAPSHOT_PROVIDER.get().cloned()
 }
 
+pub fn set_scanner_dirty_usage_mutation_observer(
+    observer: Option<ScannerDirtyUsageMutationObserver>,
+) -> Option<ScannerDirtyUsageMutationObserver> {
+    let mut slot = SCANNER_DIRTY_USAGE_MUTATION_OBSERVER
+        .write()
+        .unwrap_or_else(|poisoned| poisoned.into_inner());
+    std::mem::replace(&mut *slot, observer)
+}
+
+pub(crate) fn notify_scanner_dirty_usage_mutation(bucket: &str, object: &str, source: ScannerDirtyUsageMutationSource) {
+    if bucket.is_empty() || object.is_empty() {
+        return;
+    }
+    let observer = SCANNER_DIRTY_USAGE_MUTATION_OBSERVER
+        .read()
+        .unwrap_or_else(|poisoned| poisoned.into_inner())
+        .clone();
+    if let Some(observer) = observer {
+        observer(bucket, object, source);
+    }
+}
+
 pub(crate) fn record_erasure_write_quorum_failure(stage: &'static str, dominant_error: &'static str) {
     global_internode_metrics().record_erasure_write_quorum_failure(stage, dominant_error);
 }
@@ -570,36 +601,26 @@ pub(crate) async fn initialize_local_disk_maps(
     Ok(())
 }
 
-#[cfg(test)]
-tokio::task_local! {
-    pub(crate) static TEST_TIER_CONFIG_MGR: Arc<tokio::sync::RwLock<TierConfigMgr>>;
-}
-
 pub(crate) async fn init_tier_config_mgr(store: Arc<ECStore>) -> Result<()> {
-    #[cfg(not(test))]
     let handle = get_global_tier_config_mgr();
-    #[cfg(test)]
+    TierConfigMgr::reload_handle(&handle, store.clone()).await?;
-    let handle = TEST_TIER_CONFIG_MGR
-        .try_with(Arc::clone)
-        .unwrap_or_else(|_| get_global_tier_config_mgr());
-    let initial_reload = TierConfigMgr::reload_handle(&handle, store.clone()).await;
-    if initial_reload.is_err() {
-        // Keep local recovery active when the initial snapshot cannot be loaded.
-        TierConfigMgr::request_committed_mutation_refresh(&handle).await;
-    }
     tokio::spawn(TierConfigMgr::refresh_tier_config_handle(handle, store));
-    initial_reload.map_err(Error::from)
+    Ok(())
 }
 
 #[cfg(test)]
 mod tests {
     use super::{
-        LockRegistry, clear_local_disk_id_map_for_test, local_disk_path_by_id, local_node_name, reconcile_local_disk_ids,
+        LockRegistry, ScannerDirtyUsageMutationSource, clear_local_disk_id_map_for_test, local_disk_path_by_id, local_node_name,
-        replace_local_disk_id, set_local_node_name,
+        notify_scanner_dirty_usage_mutation, reconcile_local_disk_ids, replace_local_disk_id, set_local_node_name,
+        set_scanner_dirty_usage_mutation_observer,
     };
     use crate::disk::endpoint::Endpoint;
     use rustfs_lock::{LocalClient, LockClient};
-    use std::{collections::HashMap, sync::Arc};
+    use std::{
+        collections::HashMap,
+        sync::{Arc, Mutex},
+    };
     use uuid::Uuid;
 
     fn url_endpoint(raw: &str) -> Endpoint {
@@ -634,6 +655,42 @@ mod tests {
         assert!(Arc::ptr_eq(&clients[1], &client_b));
     }
 
+    #[test]
+    #[serial_test::serial(scanner_dirty_usage_mutation_observer)]
+    fn scanner_dirty_usage_mutation_observer_filters_empty_identity_and_preserves_source() {
+        let observed = Arc::new(Mutex::new(Vec::new()));
+        let observed_clone = Arc::clone(&observed);
+        let previous = set_scanner_dirty_usage_mutation_observer(Some(Arc::new(move |bucket, object, source| {
+            observed_clone.lock().expect("observer lock should not be poisoned").push((
+                bucket.to_string(),
+                object.to_string(),
+                source,
+            ));
+        })));
+
+        notify_scanner_dirty_usage_mutation("photos", "2026/image.jpg", ScannerDirtyUsageMutationSource::Replication);
+        notify_scanner_dirty_usage_mutation("", "2026/empty-bucket.jpg", ScannerDirtyUsageMutationSource::TierExpiration);
+        notify_scanner_dirty_usage_mutation("photos", "", ScannerDirtyUsageMutationSource::TierExpiration);
+        notify_scanner_dirty_usage_mutation("archive", "expired.bin", ScannerDirtyUsageMutationSource::TierExpiration);
+        set_scanner_dirty_usage_mutation_observer(previous);
+
+        assert_eq!(
+            *observed.lock().expect("observer lock should not be poisoned"),
+            vec![
+                (
+                    "photos".to_string(),
+                    "2026/image.jpg".to_string(),
+                    ScannerDirtyUsageMutationSource::Replication
+                ),
+                (
+                    "archive".to_string(),
+                    "expired.bin".to_string(),
+                    ScannerDirtyUsageMutationSource::TierExpiration
+                ),
+            ]
+        );
+    }
+
     #[tokio::test]
     #[serial_test::serial]
     async fn local_node_name_round_trips_through_common_runtime_helper() {

crates/ecstore/src/services/notification_sys.rs

@@ -568,6 +568,10 @@ pub(crate) fn tier_delete_journal_topology_generation(proof: &TierDeleteJournalF
     stable_tier_delete_journal_topology_generation(&proof.token.topology_fingerprint)
 }
 
+pub(crate) fn cross_pool_fence_topology_generation(proof: &CrossPoolFenceFleetProofToken) -> String {
+    stable_tier_delete_journal_topology_generation(&proof.0.topology_fingerprint)
+}
+
 /// Acquire one non-cloneable authority that must span the complete reconcile
 /// effect window, including its final strong readback.
 pub async fn acquire_legacy_transition_state_reconcile_fleet_proof() -> Option<LegacyTransitionStateReconcileFleetProofToken> {
@@ -1120,6 +1124,14 @@ pub(crate) fn install_remote_version_state_fleet_proof_for_test(topology_fingerp
     RemoteVersionStateFleetProofGuard
 }
 
+#[cfg(all(test, feature = "test-util"))]
+pub(crate) fn install_current_remote_version_state_fleet_proof_for_test() -> RemoteVersionStateFleetProofGuard {
+    let topology = REMOTE_VERSION_STATE_PROBE_TOPOLOGY
+        .get()
+        .expect("the test store must bind its fleet topology before installing a writer proof");
+    install_remote_version_state_fleet_proof_for_test(topology)
+}
+
 #[cfg(all(test, feature = "test-util"))]
 pub(crate) struct TransitionTransactionCompactionFleetProofGuard;
 

crates/ecstore/src/services/rebalance/control.rs

@@ -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

crates/ecstore/src/services/rebalance/entry.rs

@@ -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() {

crates/ecstore/src/services/rebalance/rebalance_unit_tests.rs

@@ -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"))));

crates/ecstore/src/services/rebalance/worker.rs

@@ -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,
+                );
+            }
+        }
+    }
+}

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

@@ -653,6 +653,26 @@ impl MockWarmBackend {
 
 #[async_trait]
 impl WarmBackend for MockWarmBackend {
+    async fn probe_legacy_metadata(
+        &self,
+        object: &str,
+        remote_version: Option<&str>,
+    ) -> Result<super::warm_backend::LegacyTransitionStateProbe, std::io::Error> {
+        use super::warm_backend::LegacyTransitionStateProbe as Probe;
+        let candidate = match remote_version {
+            Some(version) if !version.is_empty() => self.probe_transition_version(object, version).await?,
+            _ => self.probe_transition_candidate(object).await?,
+        };
+        Ok(match candidate {
+            TransitionCandidateProbe::Missing => Probe::Missing,
+            TransitionCandidateProbe::UnversionedPresent => Probe::UnversionedPresent,
+            TransitionCandidateProbe::VersionedPresent(version) if version == "null" => Probe::SuspendedNullPresent,
+            TransitionCandidateProbe::VersionedPresent(version) => Probe::VersionedPresent(version),
+            TransitionCandidateProbe::Ambiguous => Probe::Ambiguous,
+            TransitionCandidateProbe::Unsupported => Probe::Unsupported,
+        })
+    }
+
     fn validate_remote_version_id(&self, remote_version_id: &str) -> Result<(), std::io::Error> {
         if remote_version_id.is_empty() {
             return Ok(());
@@ -874,8 +894,9 @@ pub async fn register_mock_tier_backend(handle: &Arc<RwLock<TierConfigMgr>>, tie
             ..Default::default()
         },
     );
-    tier_config_mgr
+    drop(tier_config_mgr);
-        .install_test_driver(tier_name, Box::new(backend))
+    TierConfigMgr::install_test_driver_in(handle, tier_name, Box::new(backend))
+        .await
         .expect("mock tier driver should install");
 }
 

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

@@ -2322,6 +2322,14 @@ struct SharedWarmBackendProxy(SharedWarmBackend);
 
 #[async_trait::async_trait]
 impl WarmBackend for SharedWarmBackendProxy {
+    async fn probe_legacy_metadata(
+        &self,
+        object: &str,
+        remote_version: Option<&str>,
+    ) -> io::Result<crate::services::tier::warm_backend::LegacyTransitionStateProbe> {
+        self.0.probe_legacy_metadata(object, remote_version).await
+    }
+
     async fn validate(&self) -> io::Result<()> {
         self.0.validate().await
     }
@@ -2490,6 +2498,27 @@ impl TierOperationLease {
         self.inner.driver.probe_transition_version(object, remote_version_id).await
     }
 
+    pub(crate) async fn probe_legacy_transition_state(
+        &self,
+        object: &str,
+        remote_version: Option<&str>,
+    ) -> io::Result<crate::services::tier::warm_backend::LegacyTransitionStateProbe> {
+        let Some(reconciler) = self
+            .inner
+            .reconciler
+            .get_or_try_init(|| async {
+                crate::services::tier::warm_backend::new_transition_candidate_reconciler(&self.inner.tier_config)
+                    .await
+                    .map(|reconciler| reconciler.map(Arc::from))
+            })
+            .await
+            .map_err(|err| io::Error::other(err.message))?
+        else {
+            return self.inner.driver.probe_legacy_metadata(object, remote_version).await;
+        };
+        reconciler.probe_legacy_transition_state(object, remote_version).await
+    }
+
     pub(crate) fn is_current_generation(&self) -> bool {
         lock_unpoisoned(&self.runtime)
             .generations
@@ -6013,6 +6042,19 @@ impl TierConfigMgr {
         Ok(())
     }
 
+    #[cfg(any(test, feature = "test-util"))]
+    pub(crate) async fn install_test_driver_in(
+        handle: &Arc<RwLock<Self>>,
+        tier_name: &str,
+        driver: WarmBackendImpl,
+    ) -> std::result::Result<(), AdminError> {
+        let mut manager = handle.write().await;
+        // Register the generation runtime before installing the mock so its
+        // explicit lack of a network reconciler survives the first lease.
+        tier_driver_runtime(handle, &manager);
+        manager.install_test_driver(tier_name, driver)
+    }
+
     #[cfg(any(test, feature = "test-util"))]
     pub(crate) fn install_test_driver(
         &mut self,

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

@@ -89,6 +89,18 @@ pub enum TransitionCandidateProbe {
     Unsupported,
 }
 
+/// Live evidence for repairing legacy metadata. Ordinary candidate GETs do not
+/// establish the bucket's versioning model and cannot supply this authority.
+#[derive(Clone, Debug, Eq, PartialEq)]
+pub enum LegacyTransitionStateProbe {
+    Missing,
+    UnversionedPresent,
+    SuspendedNullPresent,
+    VersionedPresent(String),
+    Ambiguous,
+    Unsupported,
+}
+
 #[derive(Clone, Copy)]
 pub(crate) struct TransitionCandidateIdentity {
     pub transaction_id: uuid::Uuid,
@@ -97,6 +109,14 @@ pub(crate) struct TransitionCandidateIdentity {
 
 #[async_trait::async_trait]
 pub(crate) trait TransitionCandidateReconciler {
+    async fn probe_legacy_transition_state(
+        &self,
+        _object: &str,
+        _remote_version: Option<&str>,
+    ) -> Result<LegacyTransitionStateProbe, std::io::Error> {
+        Ok(LegacyTransitionStateProbe::Unsupported)
+    }
+
     async fn probe_transition_candidate_for(
         &self,
         object: &str,
@@ -106,6 +126,14 @@ pub(crate) trait TransitionCandidateReconciler {
 
 #[async_trait::async_trait]
 pub trait WarmBackend {
+    async fn probe_legacy_metadata(
+        &self,
+        _object: &str,
+        _remote_version: Option<&str>,
+    ) -> Result<LegacyTransitionStateProbe, std::io::Error> {
+        Ok(LegacyTransitionStateProbe::Unsupported)
+    }
+
     async fn validate(&self) -> Result<(), std::io::Error> {
         Ok(())
     }
@@ -448,6 +476,18 @@ impl MeteredWarmBackend {
 
 #[async_trait::async_trait]
 impl WarmBackend for MeteredWarmBackend {
+    async fn probe_legacy_metadata(
+        &self,
+        object: &str,
+        remote_version: Option<&str>,
+    ) -> Result<LegacyTransitionStateProbe, std::io::Error> {
+        let result = self.inner.probe_legacy_metadata(object, remote_version).await;
+        if matches!(result, Ok(LegacyTransitionStateProbe::Unsupported)) {
+            return result;
+        }
+        Self::record(TierRequestOperation::Probe, result)
+    }
+
     /// Delegated without a counter: only one backend issues a remote request
     /// here, and every other one takes the trait default, so a `validate`
     /// counter would mostly record requests that never happened.
@@ -524,6 +564,18 @@ struct MeteredTransitionCandidateReconciler {
 
 #[async_trait::async_trait]
 impl TransitionCandidateReconciler for MeteredTransitionCandidateReconciler {
+    async fn probe_legacy_transition_state(
+        &self,
+        object: &str,
+        remote_version: Option<&str>,
+    ) -> Result<LegacyTransitionStateProbe, std::io::Error> {
+        let result = self.inner.probe_legacy_transition_state(object, remote_version).await;
+        if matches!(result, Ok(LegacyTransitionStateProbe::Unsupported)) {
+            return result;
+        }
+        MeteredWarmBackend::record(TierRequestOperation::Probe, result)
+    }
+
     async fn probe_transition_candidate_for(
         &self,
         object: &str,

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

@@ -101,6 +101,19 @@ impl WarmBackend for WarmBackendMinIO {
 
 #[async_trait::async_trait]
 impl crate::services::tier::warm_backend::TransitionCandidateReconciler for WarmBackendMinIO {
+    async fn probe_legacy_transition_state(
+        &self,
+        object: &str,
+        remote_version: Option<&str>,
+    ) -> Result<super::warm_backend::LegacyTransitionStateProbe, std::io::Error> {
+        crate::services::tier::warm_backend::TransitionCandidateReconciler::probe_legacy_transition_state(
+            &self.0,
+            object,
+            remote_version,
+        )
+        .await
+    }
+
     async fn probe_transition_candidate_for(
         &self,
         object: &str,

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

@@ -146,6 +146,19 @@ impl WarmBackend for WarmBackendRustFS {
 
 #[async_trait::async_trait]
 impl crate::services::tier::warm_backend::TransitionCandidateReconciler for WarmBackendRustFS {
+    async fn probe_legacy_transition_state(
+        &self,
+        object: &str,
+        remote_version: Option<&str>,
+    ) -> Result<super::warm_backend::LegacyTransitionStateProbe, std::io::Error> {
+        crate::services::tier::warm_backend::TransitionCandidateReconciler::probe_legacy_transition_state(
+            &self.0,
+            object,
+            remote_version,
+        )
+        .await
+    }
+
     async fn probe_transition_candidate_for(
         &self,
         object: &str,

crates/ecstore/src/services/tier/warm_backend_s3.rs

@@ -376,7 +376,6 @@ struct TransitionCandidateVersions {
 }
 
 impl TransitionCandidateVersions {
-    #[cfg(test)]
     fn extend(&mut self, remote_object: &str, versions: &ListVersionsResult) {
         for version in versions.versions.iter().filter(|version| version.key == remote_object) {
             if self.version_id.is_some() {
@@ -512,6 +511,20 @@ mod tests {
     }
 
     async fn candidate_probe_fixture() -> Option<(WarmBackendS3, tokio::task::JoinHandle<Vec<String>>)> {
+        scripted_probe_fixture([
+                "HTTP/1.1 206 Partial Content\r\nContent-Length: 1\r\nx-amz-version-id: opaque-version\r\nConnection: close\r\n\r\nx",
+                "HTTP/1.1 206 Partial Content\r\nContent-Length: 1\r\nConnection: close\r\n\r\nx",
+                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 63\r\nConnection: close\r\n\r\n<Error><Code>NoSuchKey</Code><Message>missing</Message></Error>",
+                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 66\r\nConnection: close\r\n\r\n<Error><Code>NoSuchObject</Code><Message>missing</Message></Error>",
+                "HTTP/1.1 403 Forbidden\r\nContent-Type: application/xml\r\nContent-Length: 65\r\nConnection: close\r\n\r\n<Error><Code>AccessDenied</Code><Message>denied</Message></Error>",
+                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 63\r\nConnection: close\r\n\r\n<Error><Code>NoSuchKey</Code><Message>missing</Message></Error>",
+                "HTTP/1.1 416 Range Not Satisfiable\r\nContent-Type: application/xml\r\nContent-Length: 72\r\nConnection: close\r\n\r\n<Error><Code>InvalidRange</Code><Message>empty version</Message></Error>",
+                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 67\r\nConnection: close\r\n\r\n<Error><Code>NoSuchVersion</Code><Message>missing</Message></Error>",
+                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 63\r\nConnection: close\r\n\r\n<Error><Code>NoSuchKey</Code><Message>missing</Message></Error>",
+            ].into_iter().map(str::to_owned).collect()).await
+    }
+
+    async fn scripted_probe_fixture(responses: Vec<String>) -> Option<(WarmBackendS3, tokio::task::JoinHandle<Vec<String>>)> {
         let listener = match tokio::net::TcpListener::bind("127.0.0.1:0").await {
             Ok(listener) => listener,
             Err(err) if err.kind() == std::io::ErrorKind::PermissionDenied => return None,
@@ -522,17 +535,6 @@ mod tests {
             .expect("listener local address should be available")
             .to_string();
         let fixture = tokio::spawn(async move {
-            let responses = [
-                "HTTP/1.1 206 Partial Content\r\nContent-Length: 1\r\nx-amz-version-id: opaque-version\r\nConnection: close\r\n\r\nx",
-                "HTTP/1.1 206 Partial Content\r\nContent-Length: 1\r\nConnection: close\r\n\r\nx",
-                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 63\r\nConnection: close\r\n\r\n<Error><Code>NoSuchKey</Code><Message>missing</Message></Error>",
-                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 66\r\nConnection: close\r\n\r\n<Error><Code>NoSuchObject</Code><Message>missing</Message></Error>",
-                "HTTP/1.1 403 Forbidden\r\nContent-Type: application/xml\r\nContent-Length: 65\r\nConnection: close\r\n\r\n<Error><Code>AccessDenied</Code><Message>denied</Message></Error>",
-                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 63\r\nConnection: close\r\n\r\n<Error><Code>NoSuchKey</Code><Message>missing</Message></Error>",
-                "HTTP/1.1 416 Range Not Satisfiable\r\nContent-Type: application/xml\r\nContent-Length: 72\r\nConnection: close\r\n\r\n<Error><Code>InvalidRange</Code><Message>empty version</Message></Error>",
-                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 67\r\nConnection: close\r\n\r\n<Error><Code>NoSuchVersion</Code><Message>missing</Message></Error>",
-                "HTTP/1.1 404 Not Found\r\nContent-Type: application/xml\r\nContent-Length: 63\r\nConnection: close\r\n\r\n<Error><Code>NoSuchKey</Code><Message>missing</Message></Error>",
-            ];
             let mut requests = Vec::new();
             for response in responses {
                 let (mut stream, _) = listener.accept().await.expect("fixture should accept candidate GET");
@@ -673,6 +675,117 @@ mod tests {
         assert!(requests[8].to_ascii_lowercase().contains("?versionid=historical-version"));
     }
 
+    fn legacy_probe_xml_response(body: &str) -> String {
+        format!(
+            "HTTP/1.1 200 OK\r\nContent-Type: application/xml\r\nContent-Length: {}\r\nConnection: close\r\n\r\n{body}",
+            body.len()
+        )
+    }
+
+    fn legacy_probe_versioning_response(status: &str) -> String {
+        let state = if status.is_empty() {
+            String::new()
+        } else {
+            format!("<Status>{status}</Status>")
+        };
+        legacy_probe_xml_response(&format!(
+            "<VersioningConfiguration xmlns=\"http://s3.amazonaws.com/doc/2006-03-01/\">{state}</VersioningConfiguration>"
+        ))
+    }
+
+    fn legacy_probe_versions_response(versions: &[&str]) -> String {
+        let versions = versions.iter().map(|version| format!(
+            "<Version><Key>archive/object</Key><VersionId>{version}</VersionId><IsLatest>true</IsLatest><LastModified>2026-09-01T00:00:00Z</LastModified><ETag>\"legacy-etag\"</ETag><Size>7</Size><StorageClass>STANDARD</StorageClass></Version>"
+        )).collect::<String>();
+        legacy_probe_xml_response(&format!(
+            "<ListVersionsResult xmlns=\"http://s3.amazonaws.com/doc/2006-03-01/\"><Name>bucket</Name><Prefix>archive/object</Prefix><KeyMarker/><VersionIdMarker/><MaxKeys>1000</MaxKeys><IsTruncated>false</IsTruncated>{versions}</ListVersionsResult>"
+        ))
+    }
+
+    #[tokio::test]
+    async fn legacy_transition_state_probe_verifies_disabled_suspended_and_enabled_responses() {
+        use super::super::warm_backend::LegacyTransitionStateProbe as Probe;
+        for (initial, confirmed, version, expected) in [
+            ("", "", "null", Probe::UnversionedPresent),
+            ("Suspended", "Suspended", "null", Probe::SuspendedNullPresent),
+            ("Enabled", "Enabled", "version-a", Probe::VersionedPresent("version-a".to_string())),
+            ("Enabled", "Enabled", "null", Probe::SuspendedNullPresent),
+            ("", "", "unexpected-version", Probe::Ambiguous),
+            ("Suspended", "Enabled", "null", Probe::Ambiguous),
+        ] {
+            let responses = vec![
+                legacy_probe_versioning_response(initial),
+                legacy_probe_versions_response(&[version]),
+                legacy_probe_versioning_response(confirmed),
+            ];
+            let (backend, fixture) = scripted_probe_fixture(responses)
+                .await
+                .expect("legacy probe loopback fixture");
+            let result =
+                tokio::time::timeout(Duration::from_secs(10), backend.probe_legacy_transition_state("archive/object", None))
+                    .await
+                    .expect("legacy probe must finish")
+                    .expect("legacy probe should decode provider XML");
+            assert_eq!(result, expected, "initial={initial} confirmed={confirmed} version={version}");
+            let requests = fixture.await.expect("legacy probe fixture should finish");
+            assert_eq!(requests.len(), 3);
+            assert!(requests.iter().all(|request| request.starts_with("GET ")));
+            assert!(requests[0].lines().next().expect("request line").contains("versioning"));
+            assert!(requests[1].lines().next().expect("request line").contains("versions"));
+            assert!(requests[2].lines().next().expect("request line").contains("versioning"));
+        }
+    }
+
+    #[tokio::test]
+    async fn legacy_transition_state_probe_preserves_historical_exact_version() {
+        use super::super::warm_backend::LegacyTransitionStateProbe as Probe;
+        let responses = vec![
+            legacy_probe_versioning_response("Enabled"),
+            "HTTP/1.1 206 Partial Content\r\nContent-Length: 1\r\nx-amz-version-id: historical-version\r\nConnection: close\r\n\r\nx".to_string(),
+            legacy_probe_versioning_response("Enabled"),
+        ];
+        let (backend, fixture) = scripted_probe_fixture(responses).await.expect("exact legacy probe fixture");
+        assert_eq!(
+            backend
+                .probe_legacy_transition_state("archive/object", Some("historical-version"))
+                .await
+                .expect("exact version proof"),
+            Probe::VersionedPresent("historical-version".to_string())
+        );
+        let requests = fixture.await.expect("exact probe fixture should finish");
+        assert!(
+            requests[1]
+                .lines()
+                .next()
+                .expect("request line")
+                .contains("versionId=historical-version")
+        );
+        assert!(requests[1].to_ascii_lowercase().contains("range: bytes=0-0"));
+        assert!(requests.iter().all(|request| request.starts_with("GET ")));
+    }
+
+    #[tokio::test]
+    async fn legacy_transition_state_probe_retains_multiple_candidates() {
+        use super::super::warm_backend::LegacyTransitionStateProbe as Probe;
+        let responses = vec![
+            legacy_probe_versioning_response("Enabled"),
+            legacy_probe_versions_response(&["version-a", "version-b"]),
+        ];
+        let (backend, fixture) = scripted_probe_fixture(responses)
+            .await
+            .expect("ambiguous legacy probe fixture");
+        assert_eq!(
+            backend
+                .probe_legacy_transition_state("archive/object", None)
+                .await
+                .expect("ambiguous proof"),
+            Probe::Ambiguous
+        );
+        let requests = fixture.await.expect("ambiguous fixture should finish");
+        assert_eq!(requests.len(), 2);
+        assert!(requests.iter().all(|request| request.starts_with("GET ")));
+    }
+
     fn list_versions(versions: &[(&str, &str)], delete_markers: &[(&str, &str)], is_truncated: bool) -> ListVersionsResult {
         ListVersionsResult {
             versions: versions
@@ -921,6 +1034,56 @@ impl WarmBackend for WarmBackendS3 {
 
 #[async_trait::async_trait]
 impl TransitionCandidateReconciler for WarmBackendS3 {
+    async fn probe_legacy_transition_state(
+        &self,
+        object: &str,
+        remote_version: Option<&str>,
+    ) -> Result<super::warm_backend::LegacyTransitionStateProbe, std::io::Error> {
+        use super::warm_backend::LegacyTransitionStateProbe as Probe;
+
+        let initial_versioning = self.remote_bucket_versioning().await?;
+        let candidate = if let Some(version) = remote_version.filter(|version| !version.is_empty()) {
+            validate_remote_version_id(version)?;
+            match self.probe_transition_version(object, version).await? {
+                TransitionCandidateProbe::VersionedPresent(actual) if actual == version => Some(actual),
+                TransitionCandidateProbe::Missing => return Ok(Probe::Missing),
+                _ => return Ok(Probe::Ambiguous),
+            }
+        } else {
+            let remote_object = self.get_dest(object);
+            let mut opts = ListObjectsOptions::default();
+            opts.set("prefix", &remote_object);
+            opts.set("max-keys", "1000");
+            let mut key_marker = String::new();
+            let mut version_marker = String::new();
+            let mut candidates = TransitionCandidateVersions::default();
+            let mut complete = false;
+            // This is one synchronous record inspection, not an unbounded
+            // remote history scan. The caller also bounds the whole probe.
+            for _ in 0..128 {
+                let page = self
+                    .client
+                    .list_object_versions_query(&self.bucket, &opts, &key_marker, &version_marker, "")
+                    .await?;
+                candidates.extend(&remote_object, &page);
+                if candidates.ambiguous {
+                    return Ok(Probe::Ambiguous);
+                }
+                if !page.is_truncated {
+                    complete = true;
+                    break;
+                }
+                advance_version_markers(&mut key_marker, &mut version_marker, &page)?;
+            }
+            if !complete {
+                return Ok(Probe::Ambiguous);
+            }
+            candidates.version_id
+        };
+        let confirmed_versioning = self.remote_bucket_versioning().await?;
+        classify_legacy_transition_state(candidate.as_deref(), initial_versioning, confirmed_versioning)
+    }
+
     async fn probe_transition_candidate_for(
         &self,
         object: &str,
@@ -931,3 +1094,32 @@ impl TransitionCandidateReconciler for WarmBackendS3 {
             .await
     }
 }
+
+fn classify_legacy_transition_state(
+    candidate: Option<&str>,
+    initial: RemoteBucketVersioning,
+    confirmed: RemoteBucketVersioning,
+) -> Result<super::warm_backend::LegacyTransitionStateProbe, std::io::Error> {
+    use super::warm_backend::LegacyTransitionStateProbe as Probe;
+    if initial != confirmed {
+        return Ok(Probe::Ambiguous);
+    }
+    let Some(version) = candidate else {
+        return Ok(Probe::Missing);
+    };
+    if !version.is_empty() {
+        validate_remote_version_id(version)?;
+        if uuid::Uuid::parse_str(version).is_ok_and(|id| id.is_nil()) {
+            return Err(std::io::Error::new(
+                std::io::ErrorKind::InvalidData,
+                "legacy tier probe returned a nil version identifier",
+            ));
+        }
+    }
+    Ok(match (confirmed, version) {
+        (RemoteBucketVersioning::Disabled, "" | "null") => Probe::UnversionedPresent,
+        (RemoteBucketVersioning::Disabled, _) | (_, "") => Probe::Ambiguous,
+        (_, "null") => Probe::SuspendedNullPresent,
+        (_, version) => Probe::VersionedPresent(version.to_string()),
+    })
+}

crates/ecstore/src/set_disk/metadata.rs

@@ -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:
+            // A consistent layout can require more replies than the initial
-            // enough disks answered with valid-looking metadata that simply
+            // half-set probe. Reaching that probe alone is not corruption;
-            // cannot be reconciled (corrupt/foreign entries — retrying cannot
+            // only invalid or conflicting healthy replies establish that.
-            // help, and heal should see Corrupt, rustfs#5801) versus too few
-            // healthy answers (a genuine quorum condition where retry may
-            // succeed once disks recover).
             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]

crates/ecstore/src/set_disk/mod.rs

@@ -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;
@@ -3888,6 +3889,31 @@ pub struct SetDisks {
     >,
 }
 
+/// Read every physical copy before selecting a version quorum. A minority
+/// legacy record is still evidence and must not disappear behind a majority
+/// not-found result. Only an explicit file/volume absence produces `None`;
+/// an unreadable disk cannot prove that no conflicting copy exists.
+pub(crate) async fn read_legacy_transition_state_metadata_copies(
+    set: &SetDisks,
+    bucket: &str,
+    object: &str,
+) -> std::result::Result<Vec<Option<Vec<u8>>>, DiskError> {
+    let disk_object = rustfs_utils::path::encode_dir_object(object);
+    let disks = set.get_disks_internal().await;
+    if disks.is_empty() {
+        return Err(DiskError::DiskNotFound);
+    }
+
+    let (copies, errs) = SetDisks::read_all_raw_file_info(&disks, bucket, disk_object.as_str(), false).await;
+    for err in errs.into_iter().flatten() {
+        if !matches!(err, DiskError::FileNotFound | DiskError::FileVersionNotFound | DiskError::VolumeNotFound) {
+            return Err(err);
+        }
+    }
+
+    Ok(copies.into_iter().map(|copy| copy.map(|copy| copy.buf)).collect())
+}
+
 // DistributedLock sends the raw ObjectKey to its clients; LockRegistry clones
 // each endpoint's canonical Arc, so an exact Arc set identifies the lock domain.
 pub(crate) fn same_distributed_lock_domain(left: &[Arc<dyn LockClient>], right: &[Arc<dyn LockClient>]) -> bool {

crates/ecstore/src/set_disk/ops/bucket.rs

@@ -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 mut info = self.stat_bucket_with_quorum(bucket, BucketInfoQuorum::Write).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;
         if let Ok(sys) = metadata_sys::get(bucket).await {
-            versioning = sys.versioning();
+            info.versioning = sys.versioning();
-            object_locking = sys.object_locking();
+            info.object_locking = sys.object_locking();
         }
-
+        Ok(info)
-        infos
-            .into_iter()
-            .flatten()
-            .next()
-            .map(|info| BucketInfo {
-                name: info.name,
-                created: info.created,
-                versioning,
-                object_locking,
-                ..Default::default()
-            })
-            .ok_or(Error::VolumeNotFound)
     }
 
     #[tracing::instrument(skip(self))]

crates/ecstore/src/set_disk/ops/object.rs

@@ -9209,6 +9209,11 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
         let dest_obj = transaction.remote_object.clone();
         let mut transition_meta = (*oi.user_defined).clone();
         rustfs_utils::http::remove_str(&mut transition_meta, rustfs_utils::http::SUFFIX_PART_CHECKSUMS);
+        // The tier holds opaque stored bytes. Its metadata must not be treated
+        // as a second object header set: forwarding SSE intent or wrapped DEKs
+        // would request a second encryption pass and disclose local envelope
+        // material to the remote provider.
+        transition_meta.retain(|key, _| !rustfs_utils::http::is_replication_stripped_encryption_key(key));
         transition_meta.insert("name".to_string(), object.to_string());
         rustfs_utils::http::metadata_compat::insert_str(
             &mut transition_meta,
@@ -9765,13 +9770,16 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                     part_opts.part_number = Some(part_info.number);
                     #[cfg(test)]
                     fail_restore_multipart_at(RestoreMultipartFailurePoint::InvalidPartSize)?;
-                    if part_info.actual_size <= 0 {
+                    if part_info.size == 0 {
-                        return Err(Error::other(format!("invalid multipart restore part size {}", part_info.actual_size)));
+                        return Err(Error::other(format!("invalid multipart restore stored part size {}", part_info.size)));
                     }
+                    let stored_part_size = i64::try_from(part_info.size).map_err(|_| {
+                        Error::other(format!("multipart restore stored part size exceeds i64: {}", part_info.size))
+                    })?;
                     #[cfg(test)]
                     fail_restore_multipart_at(RestoreMultipartFailurePoint::RangeOverflow)?;
                     let part_end = part_offset
-                        .checked_add(part_info.actual_size - 1)
+                        .checked_add(stored_part_size - 1)
                         .ok_or_else(|| Error::other("multipart restore part range overflow".to_string()))?;
                     let rs = Some(HTTPRangeSpec {
                         is_suffix_length: false,
@@ -9799,13 +9807,19 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                     #[cfg(test)]
                     fail_restore_multipart_at(RestoreMultipartFailurePoint::HashReader)?;
                     let hash_reader =
-                        HashReader::from_stream(reader, part_info.actual_size, part_info.actual_size, None, None, false)?;
+                        HashReader::from_stream(reader, stored_part_size, part_info.actual_size, None, None, false)?;
                     let mut p_reader = PutObjReader::new(hash_reader);
                     #[cfg(test)]
                     fail_restore_multipart_at(RestoreMultipartFailurePoint::PutPart)?;
+                    // `ropts` carries the object's ETag so the single-part copy-back
+                    // keeps it (the writer only ever sees stored bytes). A part write
+                    // must not inherit that object-level value, or every restored part
+                    // would be recorded under the same ETag; each part keeps its own.
+                    let mut part_write_opts = ropts.clone();
+                    part_write_opts.preserve_etag = Some(part_info.etag.clone()).filter(|etag| !etag.is_empty());
                     let p_info = self_
                         .clone()
-                        .put_object_part(bucket, object, &res.upload_id, part_info.number, &mut p_reader, &ropts)
+                        .put_object_part(bucket, object, &res.upload_id, part_info.number, &mut p_reader, &part_write_opts)
                         .await?;
                     #[cfg(test)]
                     let p_info = if restore_multipart_failure_is(RestoreMultipartFailurePoint::SizeMismatch) {
@@ -9815,7 +9829,7 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                     } else {
                         p_info
                     };
-                    if p_info.size as i64 != part_info.actual_size {
+                    if p_info.size as i64 != stored_part_size {
                         return Err(Error::other(ObjectApiError::InvalidObjectState(GenericError {
                             bucket: bucket.to_string(),
                             object: object.to_string(),
@@ -9847,6 +9861,11 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                     user_defined: restore_commit_metadata,
                     no_lock: false,
                     decommission_capacity_admission: opts.decommission_capacity_admission.clone(),
+                    // The composite ETag would otherwise be recomputed from the
+                    // parts as they were written back, which for an encrypted or
+                    // compressed object digests stored bytes rather than the
+                    // object's public ETag.
+                    preserve_etag: oi.etag.clone(),
                     ..Default::default()
                 };
                 self_
@@ -13488,6 +13507,303 @@ mod transition_commit_failure_tests {
         );
     }
 
+    /// backlog#2368 B5: the tier stores opaque bytes, so the archive request
+    /// must not carry the object's encryption headers. Forwarding them made
+    /// every S3 target reject an SSE-C archive outright, asked the target to
+    /// encrypt an SSE-KMS object a second time under a key id it does not own,
+    /// and handed the wrapped DEK to a third-party provider.
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn transition_does_not_forward_encryption_metadata_to_the_tier() {
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "transition-encryption-metadata-bucket";
+        let object = "object.bin";
+        for disk in &disk_stores {
+            disk.make_volume(bucket).await.expect("bucket volume should be created");
+        }
+
+        let encryption_metadata = [
+            ("x-amz-server-side-encryption", "aws:kms"),
+            ("x-amz-server-side-encryption-aws-kms-key-id", "arn:aws:kms:us-east-1:123:key/abc"),
+            ("x-amz-server-side-encryption-customer-algorithm", "AES256"),
+            (rustfs_utils::http::INTERNAL_ENCRYPTION_KEY_HEADER, "d3JhcHBlZC1kZWs="),
+            (rustfs_utils::http::INTERNAL_ENCRYPTION_IV_HEADER, "AAAAAAAAAAAAAAAA"),
+            (rustfs_utils::http::INTERNAL_ENCRYPTION_ALGORITHM_HEADER, "AES256"),
+        ];
+        let mut user_defined: HashMap<String, String> = encryption_metadata
+            .iter()
+            .map(|(key, value)| ((*key).to_string(), (*value).to_string()))
+            .collect();
+        user_defined.insert("x-amz-meta-owner".to_string(), "finance".to_string());
+
+        let mut reader = PutObjReader::from_vec(b"stored bytes the tier keeps opaque ".repeat(64));
+        set_disks
+            .put_object(
+                bucket,
+                object,
+                &mut reader,
+                &ObjectOptions {
+                    user_defined,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("the encrypted source object should be written");
+        let original = set_disks
+            .get_object_info(bucket, object, &ObjectOptions::default())
+            .await
+            .expect("the source object should be readable");
+
+        let tier_name = format!("COLDTIER{}", &Uuid::new_v4().simple().to_string()[..8]).to_uppercase();
+        let backend = register_mock_tier(&runtime_sources::global_tier_config_mgr(), &tier_name).await;
+        set_disks
+            .transition_object(
+                bucket,
+                object,
+                &ObjectOptions {
+                    no_lock: true,
+                    transition: TransitionOptions {
+                        status: TRANSITION_PENDING.to_string(),
+                        tier: tier_name,
+                        etag: original.etag.clone().unwrap_or_default(),
+                        ..Default::default()
+                    },
+                    version_id: original.version_id.map(|version| version.to_string()),
+                    mod_time: original.mod_time,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("the encrypted object should transition");
+
+        let transitioned = set_disks
+            .get_object_info(bucket, object, &ObjectOptions::default())
+            .await
+            .expect("the transitioned object should be readable");
+        let remote_metadata = backend
+            .metadata(&transitioned.transitioned_object.name)
+            .await
+            .expect("the tier must have received the object");
+
+        for (key, _) in encryption_metadata {
+            assert!(
+                !remote_metadata.keys().any(|stored| stored.eq_ignore_ascii_case(key)),
+                "transition must not forward {key} to the tier: {remote_metadata:?}"
+            );
+        }
+        assert!(
+            remote_metadata
+                .iter()
+                .any(|(key, value)| key.eq_ignore_ascii_case("x-amz-meta-owner") && value == "finance"),
+            "ordinary user metadata must still travel to the tier: {remote_metadata:?}"
+        );
+
+        // Read-through and restore both resolve encryption locally, so the
+        // stripped keys must survive untouched in the local metadata.
+        for (key, value) in encryption_metadata {
+            assert_eq!(
+                transitioned.user_defined.get(key).map(String::as_str),
+                Some(value),
+                "the local copy must keep {key}"
+            );
+        }
+    }
+
+    /// Deterministic bytes that do not repeat with a short period, so a slice
+    /// taken at the wrong offset cannot coincidentally compare equal.
+    fn stored_representation_bytes(seed: u32, len: usize) -> Vec<u8> {
+        (0..len)
+            .map(|index| {
+                let mixed = (index as u32).wrapping_add(seed).wrapping_mul(2_654_435_761);
+                (mixed >> 13) as u8
+            })
+            .collect()
+    }
+
+    /// Compares two stored representations without dumping megabytes of bytes
+    /// into the failure output.
+    fn assert_stored_representation_eq(actual: &[u8], expected: &[u8], what: &str) {
+        assert_eq!(actual.len(), expected.len(), "{what}: stored length differs");
+        if let Some(offset) = actual.iter().zip(expected).position(|(left, right)| left != right) {
+            panic!(
+                "{what}: stored bytes differ at offset {offset} (found {:#04x}, expected {:#04x})",
+                actual[offset], expected[offset]
+            );
+        }
+    }
+
+    async fn read_stored_representation(set_disks: &Arc<SetDisks>, bucket: &str, object: &str) -> Vec<u8> {
+        let mut reader = set_disks
+            .get_object_reader(
+                bucket,
+                object,
+                None,
+                HeaderMap::new(),
+                &ObjectOptions {
+                    no_lock: true,
+                    raw_data_movement_read: true,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("stored-representation reader should open");
+        let mut body = Vec::new();
+        reader.stream.read_to_end(&mut body).await.expect("stored body should drain");
+        body
+    }
+
+    /// backlog#2368 B3: the multipart restore loop addresses the tier in STORED
+    /// coordinates. Accumulating each part's PLAINTEXT length instead handed
+    /// every part a misaligned slice of the remote object whose length still
+    /// satisfied the range, the `HashReader` and the completion size check, so
+    /// the copy-back reported success while silently replacing the bytes.
+    ///
+    /// The fixture reproduces the encrypted geometry — a stored form LONGER
+    /// than the plaintext it encodes — because that is what keeps a
+    /// plaintext-coordinate range inside the tier object and makes the
+    /// corruption silent rather than a short read.
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn multipart_restore_copies_the_stored_representation_back_verbatim() {
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "restore-multipart-stored-coordinates-bucket";
+        let object = "object.bin";
+        for disk in &disk_stores {
+            disk.make_volume(bucket).await.expect("bucket volume should be created");
+        }
+
+        // The minimum-part-size gate reads the PLAINTEXT length, so part one
+        // clears 5 MiB there while its stored form carries encoding overhead.
+        let part_shapes = [(6 * 1024 * 1024_usize, 9_216_usize), (256 * 1024_usize, 512_usize)];
+        let mut user_defined = HashMap::new();
+        user_defined.insert(rustfs_utils::http::INTERNAL_ENCRYPTION_ALGORITHM_HEADER.to_string(), "AES256".to_string());
+        user_defined.insert(
+            rustfs_utils::http::INTERNAL_ENCRYPTION_IV_HEADER.to_string(),
+            "AAAAAAAAAAAAAAAA".to_string(),
+        );
+
+        let upload = set_disks
+            .new_multipart_upload(
+                bucket,
+                object,
+                &ObjectOptions {
+                    user_defined: user_defined.clone(),
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("multipart upload should be created");
+
+        let mut uploaded_parts = Vec::new();
+        let mut expected_stored = Vec::new();
+        for (index, (plaintext_len, overhead)) in part_shapes.iter().enumerate() {
+            let stored = stored_representation_bytes(index as u32 * 7 + 1, plaintext_len + overhead);
+            expected_stored.extend_from_slice(&stored);
+            let stored_len = stored.len() as i64;
+            let hash_reader =
+                HashReader::from_stream(std::io::Cursor::new(stored), stored_len, *plaintext_len as i64, None, None, false)
+                    .expect("hash reader over the stored representation");
+            let mut reader = PutObjReader::new(hash_reader);
+            let info = set_disks
+                .put_object_part(bucket, object, &upload.upload_id, index + 1, &mut reader, &ObjectOptions::default())
+                .await
+                .expect("stored part should be staged");
+            assert_eq!(info.size as i64, stored_len, "a part is stored in its encoded length");
+            uploaded_parts.push(CompletePart {
+                part_num: info.part_num,
+                etag: info.etag,
+                ..Default::default()
+            });
+        }
+
+        let original = set_disks
+            .clone()
+            .complete_multipart_upload(bucket, object, &upload.upload_id, uploaded_parts, &ObjectOptions::default())
+            .await
+            .expect("source multipart upload should complete");
+        let original_parts: Vec<(usize, usize, i64, String)> = original
+            .parts
+            .iter()
+            .map(|part| (part.number, part.size, part.actual_size, part.etag.clone()))
+            .collect();
+        for (_, size, actual_size, _) in &original_parts {
+            assert!(
+                *size as i64 > *actual_size,
+                "the fixture must keep the two coordinate systems apart: stored {size} vs plaintext {actual_size}"
+            );
+        }
+        let stored_before = read_stored_representation(&set_disks, bucket, object).await;
+        assert_stored_representation_eq(&stored_before, &expected_stored, "the fixture must store its encoded bytes verbatim");
+
+        let tier_name = format!("COLDTIER{}", &Uuid::new_v4().simple().to_string()[..8]).to_uppercase();
+        register_mock_tier(&runtime_sources::global_tier_config_mgr(), &tier_name).await;
+        set_disks
+            .transition_object(
+                bucket,
+                object,
+                &ObjectOptions {
+                    no_lock: true,
+                    transition: TransitionOptions {
+                        status: TRANSITION_PENDING.to_string(),
+                        tier: tier_name,
+                        etag: original.etag.clone().unwrap_or_default(),
+                        ..Default::default()
+                    },
+                    version_id: original.version_id.map(|version| version.to_string()),
+                    mod_time: original.mod_time,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("multipart source should transition before restore");
+
+        let operation_id = Uuid::new_v4();
+        set_disks
+            .put_object_metadata(
+                bucket,
+                object,
+                &ObjectOptions {
+                    eval_metadata: Some(restore_metadata(operation_id, true)),
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("the restore generation should be installed");
+        let mut restore_opts = ObjectOptions::default();
+        restore_opts.transition.restore_request.days = Some(1);
+        restore_opts.user_defined = restore_operation_id_metadata(operation_id);
+        set_disks
+            .clone()
+            .restore_transitioned_object(bucket, object, &restore_opts)
+            .await
+            .expect("multipart restore should complete");
+
+        let stored_after = read_stored_representation(&set_disks, bucket, object).await;
+        assert_stored_representation_eq(
+            &stored_after,
+            &expected_stored,
+            "a multipart restore must copy the stored representation back verbatim",
+        );
+
+        let restored = set_disks
+            .get_object_info(bucket, object, &ObjectOptions::default())
+            .await
+            .expect("the restored object should be readable");
+        let restored_parts: Vec<(usize, usize, i64, String)> = restored
+            .parts
+            .iter()
+            .map(|part| (part.number, part.size, part.actual_size, part.etag.clone()))
+            .collect();
+        assert_eq!(
+            restored_parts, original_parts,
+            "restore must rebuild the same part layout, sizes and part ETags"
+        );
+        assert_eq!(restored.size, original.size, "restore must keep the stored object size");
+        // backlog#2369 P7.1: the copy-back digests stored bytes, so the object's
+        // public ETag has to be carried over rather than recomputed.
+        assert_eq!(restored.etag, original.etag, "restore must preserve the object ETag");
+    }
+
     #[tokio::test]
     #[serial_test::serial]
     async fn restore_failure_after_snapshot_cleans_exact_generation_and_returns_primary_error() {

crates/ecstore/src/store/bucket.rs

@@ -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};
@@ -770,19 +770,57 @@ impl ECStore {
         Ok(())
     }
 
+    /// Prove a live bucket generation before repairing missing expansion volumes.
+    /// Unlike request validation, repair only needs one erasure set to confirm
+    /// existence; an incomplete expansion set is precisely what repair fixes.
+    /// Callers must hold the bucket namespace lock through the subsequent heal.
+    pub(crate) async fn bucket_exists_for_heal(&self, bucket: &str) -> Result<bool> {
+        let results = futures::future::join_all(
+            self.bucket_sets()
+                .map(|(_, _, set)| async move { set.get_bucket_info(bucket, &BucketOptions::default()).await }),
+        )
+        .await;
+        let mut first_error = None;
+        for result in results {
+            match result {
+                Ok(_) => return Ok(true),
+                Err(err) if is_err_strict_volume_not_found(&err) => {}
+                Err(err) if first_error.is_none() => first_error = Some(err),
+                Err(_) => {}
+            }
+        }
+        match first_error {
+            Some(err) => Err(err),
+            None => Ok(false),
+        }
+    }
+
     #[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 =
+        let mut scoped_results = futures::future::join_all(self.bucket_sets().map(|(pool_index, set_index, set)| async move {
-            futures::future::join_all(self.bucket_sets().map(|(pool_index, set_index, set)| async move {
+            let result = match quorum {
-                (pool_index, set_index, set.get_bucket_info(bucket, opts).await)
+                BucketInfoQuorum::Read => set.stat_bucket_with_quorum(bucket, quorum).await,
-            }))
+                BucketInfoQuorum::Write => set.get_bucket_info(bucket, opts).await,
-            .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 +844,11 @@ 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 +861,35 @@ 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 +1120,7 @@ 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 +1147,7 @@ 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 +1431,18 @@ 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 +1457,7 @@ 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 +1478,12 @@ mod tests {
         )
         .await
         .expect("multi-pool ECStore should initialize");
-        let storage_class =
+        let mut storage_class_kvs = rustfs_config::server_config::KVS::new();
-            crate::config::storageclass::lookup_config_for_pools_without_env(&rustfs_config::server_config::KVS::new(), &[4, 4])
+        if let Some(parity) = standard_parity {
-                .expect("multi-pool storage class should match both four-disk pools");
+            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());
@@ -2050,6 +2132,98 @@ mod tests {
             .expect("metadata initialization should recreate the bucket volume in the new pool");
     }
 
+    #[tokio::test]
+    #[serial]
+    async fn bucket_metadata_init_repairs_half_created_expansion_pool() {
+        // Check both pool orders: an incomplete set must not hide a later
+        // complete set, and a complete set must not weaken request validation.
+        for complete_pool in 0..2 {
+            let (temp_dir, ecstore) = setup_multi_pool_bucket_test_env().await;
+            let bucket = format!("partial-expansion-{}", Uuid::new_v4().simple());
+            for pool_index in 0..2 {
+                let present_disks = if pool_index == complete_pool { 4 } else { 2 };
+                for disk_index in 0..present_disks {
+                    tokio::fs::create_dir(
+                        temp_dir
+                            .path()
+                            .join(format!("pool{pool_index}-disk{disk_index}"))
+                            .join(&bucket),
+                    )
+                    .await
+                    .expect("fixture bucket volume should be created");
+                }
+            }
+            assert_eq!(
+                ecstore
+                    .get_bucket_info_from_sets(&bucket, &BucketOptions::default())
+                    .await
+                    .expect_err("request validation must reject a half-created expansion set"),
+                StorageError::ErasureWriteQuorum
+            );
+
+            metadata_sys::init_bucket_metadata_sys(ecstore.clone(), vec![bucket.clone()]).await;
+
+            for pool_index in 0..2 {
+                for disk_index in 0..4 {
+                    assert!(
+                        temp_dir
+                            .path()
+                            .join(format!("pool{pool_index}-disk{disk_index}"))
+                            .join(&bucket)
+                            .is_dir(),
+                        "metadata initialization must heal every missing expansion volume"
+                    );
+                }
+            }
+            ecstore
+                .get_bucket_info_from_sets(&bucket, &BucketOptions::default())
+                .await
+                .expect("strict request validation should succeed after volume repair");
+        }
+    }
+
+    #[tokio::test]
+    #[serial]
+    async fn bucket_metadata_init_does_not_combine_partial_set_evidence() {
+        let (temp_dir, ecstore) = setup_multi_pool_bucket_test_env().await;
+        let bucket = format!("no-quorum-expansion-{}", Uuid::new_v4().simple());
+        for pool_index in 0..2 {
+            for disk_index in 0..2 {
+                tokio::fs::create_dir(
+                    temp_dir
+                        .path()
+                        .join(format!("pool{pool_index}-disk{disk_index}"))
+                        .join(&bucket),
+                )
+                .await
+                .expect("fixture bucket volume should be created");
+            }
+        }
+        assert_eq!(
+            ecstore
+                .bucket_exists_for_heal(&bucket)
+                .await
+                .expect_err("repair must require a complete quorum within one set"),
+            StorageError::ErasureWriteQuorum
+        );
+
+        metadata_sys::init_bucket_metadata_sys(ecstore.clone(), vec![bucket.clone()]).await;
+
+        for pool_index in 0..2 {
+            for disk_index in 0..4 {
+                assert_eq!(
+                    temp_dir
+                        .path()
+                        .join(format!("pool{pool_index}-disk{disk_index}"))
+                        .join(&bucket)
+                        .is_dir(),
+                    disk_index < 2,
+                    "unproven bucket generations must not recreate missing volumes"
+                );
+            }
+        }
+    }
+
     #[tokio::test]
     #[serial]
     async fn bucket_metadata_init_does_not_recreate_stale_bucket_name() {
@@ -2067,6 +2241,218 @@ 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 +2476,7 @@ 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 +2484,7 @@ 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 +2496,7 @@ 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]

crates/ecstore/src/store/heal.rs

@@ -18,7 +18,6 @@ use crate::services::rebalance::{REBAL_META_NAME, RebalStatus};
 use crate::set_disk::get_lock_acquire_timeout;
 use crate::storage_api_contracts::heal::HealOperations as _;
 use crate::storage_api_contracts::namespace::NamespaceLocking as _;
-use rustfs_heal_contracts::heal_channel::DriveState;
 use rustfs_lock::NamespaceLockGuard;
 use std::collections::BTreeSet;
 use tracing::trace;
@@ -379,79 +378,6 @@ impl ECStore {
         Ok(result)
     }
 
-    /// Heal every pool metadata owner in the selected scope without allowing
-    /// one healthy pool to hide another pool's failed repair.
-    pub async fn heal_pool_metadata(&self, opts: &HealOpts) -> Result<Vec<HealResultItem>> {
-        let scopes = self.heal_erasure_set_scopes(opts).await?;
-        let mut results = Vec::new();
-        for (pool_index, set_index) in scopes {
-            if !self.replacement_pool_metadata_applies(pool_index, set_index)? {
-                continue;
-            }
-            let set = &self.pools[pool_index].disk_set[set_index];
-            let targets = set.set_endpoints.iter().map(ToString::to_string).collect::<Vec<_>>();
-            if targets.is_empty()
-                || targets.len() != set.set_drive_count
-                || targets.iter().collect::<BTreeSet<_>>().len() != targets.len()
-            {
-                return Err(Error::SlowDown);
-            }
-            // Administrative remove/no-lock options apply to user objects,
-            // never to the cluster's authoritative metadata transaction.
-            let metadata_opts = HealOpts {
-                dry_run: opts.dry_run,
-                recreate: opts.recreate,
-                scan_mode: opts.scan_mode,
-                pool: Some(pool_index),
-                set: Some(set_index),
-                ..Default::default()
-            };
-            let (result, error) = self
-                .handle_heal_object(RUSTFS_META_BUCKET, POOL_META_NAME, "", &metadata_opts)
-                .await?;
-            if let Some(error) = error {
-                return Err(error);
-            }
-            if !opts.dry_run {
-                let ok_state = DriveState::Ok.to_string();
-                let complete = result.after.drives.len() == targets.len()
-                    && targets.iter().all(|target| {
-                        let mut outcomes = result.after.drives.iter().filter(|drive| drive.endpoint == *target);
-                        outcomes.next().is_some_and(|drive| drive.state == ok_state) && outcomes.next().is_none()
-                    });
-                if !complete
-                    || !set
-                        .replacement_targets_have_version(RUSTFS_META_BUCKET, POOL_META_NAME, "", &targets)
-                        .await?
-                {
-                    return Err(Error::SlowDown);
-                }
-            }
-            results.push(result);
-        }
-        Ok(results)
-    }
-
-    /// Whether this replacement set owns the pool's metadata replica.
-    ///
-    /// Pool metadata follows normal object placement within each pool. A valid
-    /// non-owner set has no replica to repair; missing metadata on the owner
-    /// set still requires healing and target-specific readback.
-    pub fn replacement_pool_metadata_applies(&self, pool_index: usize, set_index: usize) -> Result<bool> {
-        let pool = self
-            .pools
-            .get(pool_index)
-            .ok_or_else(|| invalid_heal_pool_index(pool_index, self.pools.len()))?;
-        let selected = pool.get_disks_for_heal_object(
-            POOL_META_NAME,
-            &HealOpts {
-                set: Some(set_index),
-                ..Default::default()
-            },
-        )?;
-        Ok(Arc::ptr_eq(&selected, &pool.get_disks_by_key(POOL_META_NAME)))
-    }
-
     #[instrument(skip(self, targets), fields(pool_index, set_index, target_count = targets.len()))]
     pub async fn replacement_targets_have_version(
         &self,
@@ -903,249 +829,6 @@ mod tests {
         }
     }
 
-    #[tokio::test]
-    async fn replacement_pool_metadata_applies_to_the_written_replica_in_each_pool() {
-        let mut store = minimal_heal_store().await;
-        for pool_index in 0..store.pools.len() {
-            assert!(
-                store
-                    .replacement_pool_metadata_applies(pool_index, 0)
-                    .expect("a valid single-set pool should have a metadata owner")
-            );
-        }
-        store.ctx = Arc::new(InstanceContext::new());
-        for algorithm in [
-            crate::disk::format::DistributionAlgoVersion::V1,
-            crate::disk::format::DistributionAlgoVersion::V2,
-            crate::disk::format::DistributionAlgoVersion::V3,
-        ] {
-            let mut temp_dirs = Vec::new();
-            for pool_index in 0..store.pools.len() {
-                let (dirs, mut pool) =
-                    crate::core::sets::make_local_two_set_sets_for_pool_with_ctx(Arc::clone(&store.ctx), pool_index).await;
-                temp_dirs.extend(dirs);
-                Arc::get_mut(&mut pool)
-                    .expect("fixture pool should have one owner")
-                    .distribution_algo = algorithm.clone();
-                store.pools[pool_index] = pool;
-            }
-            for (pool_index, pool) in store.pools.iter().enumerate() {
-                let mut required_sets = 0;
-                for set_index in 0..pool.disk_set.len() {
-                    required_sets += usize::from(
-                        store
-                            .replacement_pool_metadata_applies(pool_index, set_index)
-                            .expect("valid replacement topology should be classified before metadata exists"),
-                    );
-                }
-                assert_eq!(required_sets, 1, "missing metadata cannot exempt the owner set");
-                save_config(pool.clone(), POOL_META_NAME, b"pool metadata placement".to_vec())
-                    .await
-                    .expect("normal config writes should persist one metadata replica per pool");
-                for (set_index, set) in pool.disk_set.iter().enumerate() {
-                    let applies = store
-                        .replacement_pool_metadata_applies(pool_index, set_index)
-                        .expect("valid replacement topology should be classified");
-                    let disks = set.disks.read().await.clone();
-                    for disk in disks.iter().flatten() {
-                        let replica = disk.read_xl(RUSTFS_META_BUCKET, POOL_META_NAME, false).await;
-                        if applies {
-                            replica.expect("the metadata owner must match actual persisted shards");
-                        } else {
-                            assert!(
-                                matches!(replica, Err(crate::disk::error::DiskError::FileNotFound)),
-                                "non-owner sets must have no persisted metadata shard; observed error: {:?}",
-                                replica.as_ref().err()
-                            );
-                        }
-                    }
-                }
-                assert!(
-                    store
-                        .replacement_pool_metadata_applies(pool_index, pool.disk_set.len())
-                        .is_err()
-                );
-            }
-        }
-        assert!(store.replacement_pool_metadata_applies(store.pools.len(), 0).is_err());
-    }
-
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn ordinary_pool_metadata_heal_repairs_each_owner_and_preserves_dry_run() {
-        let (_temp_dirs, store, _other_store) = test_two_pool_stores(None).await;
-        let first_missing = remove_pool_meta_shard(&store, 0).await;
-        let second_missing = remove_pool_meta_shard(&store, 1).await;
-        let destructive_options = HealOpts {
-            remove: true,
-            no_lock: true,
-            ..Default::default()
-        };
-        let results = store
-            .heal_pool_metadata(&HealOpts {
-                dry_run: true,
-                ..destructive_options
-            })
-            .await
-            .expect("dry-run should inspect both metadata owners without requiring a commit");
-        assert_eq!(results.len(), 2);
-        assert!(
-            first_missing
-                .read_xl(RUSTFS_META_BUCKET, POOL_META_NAME, false)
-                .await
-                .is_err()
-        );
-        assert!(
-            second_missing
-                .read_xl(RUSTFS_META_BUCKET, POOL_META_NAME, false)
-                .await
-                .is_err()
-        );
-
-        let lock = store.pools[0]
-            .new_ns_lock(RUSTFS_META_BUCKET, POOL_META_NAME)
-            .await
-            .expect("metadata namespace lock should be available");
-        let guard = lock
-            .get_read_lock(get_lock_acquire_timeout())
-            .await
-            .expect("a metadata reader should hold the shared fence");
-        let error = temp_env::async_with_vars(
-            [(rustfs_config::ENV_OBJECT_LOCK_ACQUIRE_TIMEOUT, Some("1"))],
-            store.heal_pool_metadata(&destructive_options),
-        )
-        .await
-        .expect_err("administrative no-lock cannot bypass the metadata write fence");
-        assert!(matches!(error, Error::Lock(rustfs_lock::LockError::Timeout { .. })));
-        drop(guard);
-        let results = store
-            .heal_pool_metadata(&destructive_options)
-            .await
-            .expect("every metadata owner should be repaired");
-        assert_eq!(results.len(), 2);
-        assert!(first_missing.read_xl(RUSTFS_META_BUCKET, POOL_META_NAME, false).await.is_ok());
-        assert!(
-            second_missing
-                .read_xl(RUSTFS_META_BUCKET, POOL_META_NAME, false)
-                .await
-                .is_ok()
-        );
-    }
-
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn ordinary_pool_metadata_heal_does_not_hide_missing_later_pool() {
-        let (_temp_dirs, store, _other_store) = test_two_pool_stores(None).await;
-        delete_config(store.pools[1].clone(), POOL_META_NAME)
-            .await
-            .expect("the second pool metadata replica should be removed");
-
-        let error = store
-            .heal_pool_metadata(&HealOpts::default())
-            .await
-            .expect_err("the healthy first pool must not hide the second owner's missing replica");
-
-        assert!(!matches!(error, Error::NoHealRequired));
-        let second_set = store.pools[1].get_disks_by_key(POOL_META_NAME);
-        for disk in second_set.disks.read().await.iter().flatten() {
-            assert!(disk.read_xl(RUSTFS_META_BUCKET, POOL_META_NAME, false).await.is_err());
-        }
-    }
-
-    #[tokio::test]
-    async fn ordinary_pool_metadata_heal_skips_only_valid_non_owner_sets() {
-        let mut store = minimal_heal_store().await;
-        store.ctx = Arc::new(InstanceContext::new());
-        for algorithm in [
-            crate::disk::format::DistributionAlgoVersion::V1,
-            crate::disk::format::DistributionAlgoVersion::V2,
-            crate::disk::format::DistributionAlgoVersion::V3,
-        ] {
-            let mut temp_dirs = Vec::new();
-            for pool_index in 0..store.pools.len() {
-                let (dirs, mut pool) =
-                    crate::core::sets::make_local_two_set_sets_for_pool_with_ctx(Arc::clone(&store.ctx), pool_index).await;
-                temp_dirs.extend(dirs);
-                Arc::get_mut(&mut pool)
-                    .expect("fixture pool should have one owner")
-                    .distribution_algo = algorithm.clone();
-                store.pools[pool_index] = pool;
-            }
-            for pool_index in 0..store.pools.len() {
-                let owner = (0..store.pools[pool_index].disk_set.len())
-                    .find(|set_index| {
-                        store
-                            .replacement_pool_metadata_applies(pool_index, *set_index)
-                            .expect("valid metadata placement")
-                    })
-                    .expect("every pool must have one metadata owner");
-                let non_owner = 1 - owner;
-                assert!(
-                    store
-                        .heal_pool_metadata(&HealOpts {
-                            pool: Some(pool_index),
-                            set: Some(non_owner),
-                            ..Default::default()
-                        })
-                        .await
-                        .expect("valid non-owner should need no metadata write")
-                        .is_empty()
-                );
-                assert!(
-                    store
-                        .heal_pool_metadata(&HealOpts {
-                            pool: Some(pool_index),
-                            set: Some(owner),
-                            ..Default::default()
-                        })
-                        .await
-                        .is_err(),
-                    "an owner with no authoritative metadata must fail"
-                );
-                assert!(
-                    store
-                        .heal_pool_metadata(&HealOpts {
-                            pool: Some(pool_index),
-                            set: Some(2),
-                            ..Default::default()
-                        })
-                        .await
-                        .is_err(),
-                    "invalid sets cannot claim the non-owner exemption"
-                );
-            }
-        }
-        assert!(
-            store
-                .heal_pool_metadata(&HealOpts {
-                    pool: Some(2),
-                    ..Default::default()
-                })
-                .await
-                .is_err()
-        );
-    }
-
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn ordinary_pool_metadata_heal_requires_every_owner_endpoint() {
-        let (_temp_dirs, store, _other_store) = test_two_pool_stores(None).await;
-        let owner = store.pools[0].get_disks_by_key(POOL_META_NAME);
-        let offline_disk = owner.disks.write().await[0]
-            .take()
-            .expect("fixture owner disk should start online");
-
-        let result = store
-            .heal_pool_metadata(&HealOpts {
-                pool: Some(0),
-                ..Default::default()
-            })
-            .await;
-
-        assert!(result.is_err(), "a surviving metadata shard must not hide an offline owner endpoint");
-        owner.disks.write().await[0] = Some(offline_disk);
-    }
-
     async fn remove_pool_meta_shard(store: &ECStore, pool_idx: usize) -> DiskStore {
         let target_set = store.pools[pool_idx].get_disks_by_key(POOL_META_NAME);
         let missing_disk = target_set.disks.read().await[0]

crates/ecstore/src/store/init.rs

@@ -95,7 +95,6 @@ fn preflight_startup_rpc_secret_with(
     }
 }
 
-const LOCAL_DECOMMISSION_INITIAL_RESUME_DELAY: Duration = Duration::from_secs(60 * 3);
 const LOCAL_DECOMMISSION_RESUME_RETRY_DELAY: Duration = Duration::from_secs(30);
 const LOCAL_DECOMMISSION_WATCHDOG_INTERVAL: Duration = Duration::from_secs(30);
 const LOCAL_DECOMMISSION_WATCHDOG_MAX_RETRY_DELAY: Duration = Duration::from_secs(60 * 5);
@@ -280,20 +279,26 @@ where
     }
 }
 
+async fn reconcile_local_decommission_after_init(store: &Arc<ECStore>, rx: CancellationToken) -> Result<()> {
+    store
+        .ensure_pool_meta_side_effects_safe("decommission worker recovery blocked while pool metadata requires recovery")
+        .await?;
+    if store.has_active_local_decommission_worker().await {
+        return Ok(());
+    }
+    store.refresh_pool_status_meta().await?;
+    let resume_required = pool_meta_has_active_decommission(&*store.pool_meta.read().await);
+    if resume_required {
+        crate::core::pools::acquire_pool_activation_fleet_proof(&store.ctx).await?;
+    }
+    store.spawn_missing_local_decommission_routines_with_token(rx).await
+}
+
 async fn supervise_local_decommission_after_init(store: Arc<ECStore>, rx: CancellationToken) {
     run_local_decommission_watchdog(rx.clone(), || {
         let store = store.clone();
         let worker_rx = rx.clone();
-        async move {
+        async move { reconcile_local_decommission_after_init(&store, worker_rx).await }
-            store
-                .ensure_pool_meta_side_effects_safe("decommission worker recovery blocked while pool metadata requires recovery")
-                .await?;
-            if store.has_active_local_decommission_worker().await {
-                return Ok(());
-            }
-            store.refresh_pool_status_meta().await?;
-            store.spawn_missing_local_decommission_routines_with_token(worker_rx).await
-        }
     })
     .await;
 }
@@ -784,14 +789,9 @@ impl ECStore {
             );
         }
         if has_local_decommission_leadership {
-            let store = self.clone();
+            // The watchdog checks recovery safety and retries transient failures.
-            let decommission_rx = rx.clone();
+            // Resume persisted work without an unconditional cold-start delay.
-            tokio::spawn(async move {
+            tokio::spawn(supervise_local_decommission_after_init(self.clone(), rx.clone()));
-                if !wait_for_local_decommission_resume_delay(&decommission_rx, LOCAL_DECOMMISSION_INITIAL_RESUME_DELAY).await {
-                    return;
-                }
-                supervise_local_decommission_after_init(store, decommission_rx).await;
-            });
         }
 
         let recovery_store = self.clone();
@@ -2104,6 +2104,44 @@ mod tests {
         );
     }
 
+    #[tokio::test(start_paused = true)]
+    async fn test_local_decommission_watchdog_cancelled_start_does_not_reconcile() {
+        let rx = CancellationToken::new();
+        rx.cancel();
+        run_local_decommission_watchdog(rx, || async {
+            panic!("cancelled startup must not schedule persisted work");
+        })
+        .await;
+    }
+
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn test_local_decommission_recovery_waits_for_live_fleet_proof_before_reserving_worker() {
+        let (_temp_dirs, store, _other_store) = crate::services::rebalance::test_two_pool_stores(None).await;
+        mark_test_pool_decommissioning(&store, 0).await;
+        assert!(store.ctx.is_dist_erasure().await);
+        let worker_rx = CancellationToken::new();
+
+        {
+            let _proof_guard = crate::services::notification_sys::without_cross_pool_fence_fleet_proof_for_test();
+            let err = super::reconcile_local_decommission_after_init(&store, worker_rx.clone())
+                .await
+                .expect_err("cold distributed recovery must wait for live fleet proof");
+            assert!(
+                crate::core::pools::is_pool_activation_fleet_proof_error(&err),
+                "recovery must reach the live fleet proof gate: {err:?}"
+            );
+            assert!(store.decommission_cancelers.read().await.iter().all(Option::is_none));
+            assert!(pool_meta_has_active_decommission(&*store.pool_meta.read().await));
+        }
+
+        super::reconcile_local_decommission_after_init(&store, worker_rx.clone())
+            .await
+            .expect("restored fleet proof should admit the persisted worker");
+        assert!(store.has_active_local_decommission_worker().await);
+        worker_rx.cancel();
+    }
+
     #[tokio::test(start_paused = true)]
     async fn test_local_decommission_watchdog_retries_general_failures_until_cancelled() {
         let rx = CancellationToken::new();
@@ -2115,11 +2153,13 @@ mod tests {
                 let attempts = attempts.clone();
                 let rx = rx.clone();
                 async move {
-                    if attempts.fetch_add(1, Ordering::SeqCst) == 0 {
+                    match attempts.fetch_add(1, Ordering::SeqCst) {
-                        Err(StorageError::SlowDown)
+                        0 => Err(StorageError::other("pool activation requires a live fleet capability proof")),
-                    } else {
+                        1 => Err(StorageError::SlowDown),
-                        rx.cancel();
+                        _ => {
-                        Ok(())
+                            rx.cancel();
+                            Ok(())
+                        }
                     }
                 }
             }
@@ -2128,8 +2168,11 @@ mod tests {
         tokio::task::yield_now().await;
         assert_eq!(attempts.load(Ordering::SeqCst), 1);
         tokio::time::advance(LOCAL_DECOMMISSION_RESUME_RETRY_DELAY).await;
-        task.await.expect("watchdog task should exit after cancellation");
+        tokio::task::yield_now().await;
         assert_eq!(attempts.load(Ordering::SeqCst), 2);
+        tokio::time::advance(local_decommission_watchdog_retry_delay(2)).await;
+        task.await.expect("watchdog task should exit after cancellation");
+        assert_eq!(attempts.load(Ordering::SeqCst), 3);
     }
 
     #[tokio::test(start_paused = true)]
@@ -12760,6 +12803,227 @@ mod tests {
         body
     }
 
+    #[cfg(feature = "test-util")]
+    #[test]
+    #[serial_test::serial(storage_class_env)]
+    fn legacy_transition_state_inspection_and_apply_keep_all_disk_copies_unchanged() {
+        run_large_stack_async_test("legacy-state-reconcile-inspection", legacy_transition_state_inspection_and_apply_case);
+    }
+
+    #[cfg(feature = "test-util")]
+    async fn legacy_transition_state_inspection_and_apply_case() {
+        use crate::bucket::lifecycle::legacy_transition_state_reconcile::{
+            LegacyTransitionStateReconcileOutcome as Outcome, LegacyTransitionStateReconcileRequest,
+            LegacyTransitionStateReconcileSelector,
+        };
+        for (remote_version, expected_state) in [
+            ("", rustfs_filemeta::TransitionVersionState::KnownDisabled),
+            ("null", rustfs_filemeta::TransitionVersionState::SuspendedNull),
+            ("opaque-version", rustfs_filemeta::TransitionVersionState::Exact),
+        ] {
+            let temp_dir = tempfile::tempdir().expect("legacy reconcile store directory");
+            let (ctx, store, _shutdown) =
+                without_storage_class_env(build_isolated_test_store(temp_dir.path(), "legacy-state-reconcile-inspect", &[4]))
+                    .await;
+            crate::bucket::metadata_sys::init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+            let tier_name = "LEGACY-RECONCILE";
+            let backend = register_mock_tier(&ctx.tier_config_mgr(), tier_name).await;
+            backend.set_put_remote_version(Some(remote_version.to_string())).await;
+            let bucket = "legacy-state-reconcile-bucket";
+            let object = "archive.bin";
+            store
+                .make_bucket(bucket, &MakeBucketOptions::default())
+                .await
+                .expect("create legacy fixture bucket");
+            let mut reader = PutObjReader::from_vec(b"legacy reconcile body".repeat(1024));
+            let source = store
+                .put_object(bucket, object, &mut reader, &ObjectOptions::default())
+                .await
+                .expect("write source");
+            {
+                // Create the fixture under the existing remote-version writer
+                // gate. This does not authorize legacy metadata reconciliation.
+                let _proof = crate::services::notification_sys::install_current_remote_version_state_fleet_proof_for_test();
+                temp_env::async_with_vars(
+                    [
+                        (rustfs_config::ENV_TIER_REMOTE_VERSION_STATE_WRITE, Some("true")),
+                        (rustfs_config::ENV_TIER_REMOTE_VERSION_STATE_FLEET_CONFIRMED, Some("true")),
+                    ],
+                    store.transition_object(
+                        bucket,
+                        object,
+                        &ObjectOptions {
+                            transition: TransitionOptions {
+                                status: TRANSITION_PENDING.to_string(),
+                                tier: tier_name.to_string(),
+                                etag: source.etag.clone().expect("source ETag"),
+                                ..Default::default()
+                            },
+                            mod_time: source.mod_time,
+                            ..Default::default()
+                        },
+                    ),
+                )
+                .await
+                .expect("transition source");
+            }
+            assert!(
+                crate::services::notification_sys::acquire_legacy_transition_state_reconcile_fleet_proof()
+                    .await
+                    .is_none(),
+                "fixture setup must not grant the missing reconciliation write capability"
+            );
+            let selector = LegacyTransitionStateReconcileSelector {
+                bucket: bucket.to_string(),
+                object: object.to_string(),
+                version_id: "null".to_string(),
+            };
+            if expected_state == rustfs_filemeta::TransitionVersionState::Exact {
+                backend
+                    .set_transition_candidate_probe_override(Some(
+                        crate::services::tier::warm_backend::TransitionCandidateProbe::Ambiguous,
+                    ))
+                    .await;
+            }
+            let converged = store
+                .inspect_legacy_transition_state(selector.clone())
+                .await
+                .expect("inspect an already explicit transition");
+            assert_eq!(converged.outcome, Outcome::Migrated, "{converged:?}");
+            assert!(!converged.changed);
+            backend.set_transition_candidate_probe_override(None).await;
+            rewrite_transitioned_xlmeta_as_legacy_unknown(temp_dir.path(), 0, bucket, object, remote_version.is_empty()).await;
+            let paths = (0..4)
+                .map(|disk| {
+                    temp_dir
+                        .path()
+                        .join(format!("pool0/set0/disk{disk}/{bucket}/{object}/{STORAGE_FORMAT_FILE}"))
+                })
+                .collect::<Vec<_>>();
+            let mut original = Vec::new();
+            for path in &paths {
+                original.push(tokio::fs::read(path).await.expect("original xl.meta"));
+            }
+            backend.clear_op_log().await;
+            let inspection = store.inspect_legacy_transition_state(selector.clone());
+            assert!(
+                std::mem::size_of_val(&inspection) <= 4 * 1024,
+                "admin inspection future must remain stack-bounded"
+            );
+            let inspected = inspection.await.expect("inspect legacy state");
+            assert_eq!(inspected.outcome, Outcome::ReadyToMigrate, "{inspected:?}");
+            assert!(!inspected.readiness.post_ready, "current fleet cannot authorize conditional writes");
+            let target = inspected.target.expect("live probe should establish one model");
+            assert_eq!(target.state, expected_state);
+            let request = LegacyTransitionStateReconcileRequest {
+                confirm: true,
+                selector,
+                source: inspected.source.expect("immutable source"),
+                original_sets: inspected.original_sets,
+                target,
+                reconciliation_digest: inspected.reconciliation_digest.expect("expected tuple digest"),
+            };
+            let mut tampered = request.clone();
+            tampered.source.remote_object.push_str("-other");
+            let probes_before = backend.op_log().await.len();
+            let rejected = store
+                .reconcile_legacy_transition_state(tampered)
+                .await
+                .expect("reject tampered tuple");
+            assert_eq!(rejected.outcome, Outcome::Corrupt);
+            assert_eq!(backend.op_log().await.len(), probes_before, "invalid digest must not probe the backend");
+            let applied = store
+                .reconcile_legacy_transition_state(request)
+                .await
+                .expect("apply must report unavailable write authority");
+            assert_eq!(applied.outcome, Outcome::BackendUnavailable, "{applied:?}");
+            assert_eq!(applied.reason_code, "write_fence_unavailable");
+            assert!(!applied.changed);
+            for (path, expected) in paths.iter().zip(&original) {
+                assert_eq!(tokio::fs::read(path).await.expect("xl.meta after inspection"), *expected);
+            }
+            assert_eq!(backend.remove_count().await, 0);
+            assert!(
+                backend
+                    .op_log()
+                    .await
+                    .iter()
+                    .all(|operation| matches!(operation, MockWarmOp::Probe { .. }))
+            );
+
+            backend.set_unreachable(true).await;
+            let unavailable = store
+                .inspect_legacy_transition_state(LegacyTransitionStateReconcileSelector {
+                    bucket: bucket.to_string(),
+                    object: object.to_string(),
+                    version_id: "null".to_string(),
+                })
+                .await
+                .expect("unreachable tier is a diagnostic outcome");
+            assert_eq!(unavailable.outcome, Outcome::BackendUnavailable);
+            assert!(!unavailable.changed);
+            backend.set_unreachable(false).await;
+            for candidate in ["", "00000000-0000-0000-0000-000000000000", "bad\nversion"] {
+                backend
+                    .set_transition_candidate_probe_override(Some(
+                        crate::services::tier::warm_backend::TransitionCandidateProbe::VersionedPresent(candidate.to_string()),
+                    ))
+                    .await;
+                let invalid_proof = store
+                    .inspect_legacy_transition_state(LegacyTransitionStateReconcileSelector {
+                        bucket: bucket.to_string(),
+                        object: object.to_string(),
+                        version_id: "null".to_string(),
+                    })
+                    .await
+                    .expect("invalid backend proof is a diagnostic outcome");
+                assert_eq!(invalid_proof.outcome, Outcome::BackendUnavailable, "{invalid_proof:?}");
+                assert!(invalid_proof.target.is_none());
+            }
+            backend.set_transition_candidate_probe_override(None).await;
+
+            backend.clear_op_log().await;
+            for path in &paths[1..] {
+                tokio::fs::remove_file(path)
+                    .await
+                    .expect("hide majority metadata copies in fixture");
+            }
+            let minority = store
+                .inspect_legacy_transition_state(LegacyTransitionStateReconcileSelector {
+                    bucket: bucket.to_string(),
+                    object: object.to_string(),
+                    version_id: "null".to_string(),
+                })
+                .await
+                .expect("inspect minority legacy record");
+            assert_eq!(
+                minority.outcome,
+                Outcome::BackendUnavailable,
+                "a minority owner must remain visible: {minority:?}"
+            );
+            assert!(
+                backend.op_log().await.is_empty(),
+                "unproven metadata quorum cannot initiate a remote probe"
+            );
+            for (path, bytes) in paths.iter().zip(&original) {
+                tokio::fs::write(path, bytes).await.expect("restore fixture copies");
+            }
+            tokio::fs::write(&paths[0], b"corrupt-xl-meta")
+                .await
+                .expect("inject corrupt copy");
+            let corrupt = store
+                .inspect_legacy_transition_state(LegacyTransitionStateReconcileSelector {
+                    bucket: bucket.to_string(),
+                    object: object.to_string(),
+                    version_id: "null".to_string(),
+                })
+                .await
+                .expect("inspect corrupt legacy record");
+            assert_eq!(corrupt.outcome, Outcome::Corrupt, "{corrupt:?}");
+            assert!(backend.op_log().await.is_empty(), "corruption must fail before backend I/O");
+        }
+    }
+
     #[cfg(feature = "test-util")]
     #[tokio::test]
     #[serial_test::serial(storage_class_env)]
@@ -18932,124 +19196,6 @@ mod tests {
         .await;
     }
 
-    #[cfg(feature = "test-util")]
-    #[test]
-    #[serial_test::serial(storage_class_env)]
-    fn tier_config_init_recovers_after_initial_reload_failure_without_another_mutation() {
-        run_large_stack_async_test("tier-config-init-recovery", || async {
-            use crate::services::tier::tier::{TIER_DRIVER_TEST_FACTORY, TierDriverTestFactory};
-
-            let temp_dir = tempfile::tempdir().expect("create tier startup recovery store dir");
-            let (ctx, store, _shutdown) = build_isolated_test_store(temp_dir.path(), "tier-startup-recovery", &[4]).await;
-            let manager = ctx.tier_config_mgr();
-            let candidate = TierConfigMgr::new();
-            let tier_name = "STARTUP-RECOVERY";
-            let backend = register_mock_tier(&candidate, tier_name).await;
-            let backend_identity = TierConfigMgr::acquire_operation_lease(&candidate, tier_name)
-                .await
-                .expect("candidate backend identity should resolve")
-                .backend_identity();
-            let candidate_digest = {
-                let candidate = candidate.read().await;
-                candidate
-                    .save_tiering_config(store.clone())
-                    .await
-                    .expect("committed tier config should persist");
-                tier_config_candidate_digest(&candidate).expect("committed candidate digest should build")
-            };
-            let config_path = format!("{}/{}", com::CONFIG_PREFIX, TIER_CONFIG_FILE);
-            let config_bytes = com::read_config(store.clone(), &config_path)
-                .await
-                .expect("committed tier config bytes should load");
-            let config_etag = store
-                .get_object_info(RUSTFS_META_BUCKET, &config_path, &ObjectOptions::default())
-                .await
-                .expect("committed tier config metadata should load")
-                .etag
-                .expect("committed tier config should have an ETag");
-            let mutation_id = uuid::Uuid::new_v4();
-            let intent = TierMutationIntent {
-                mutation_id,
-                revision: 2,
-                kind: TierMutationIntentKind::Add,
-                state: TierMutationIntentState::Committed,
-                old_config_etag: None,
-                committed_config_etag: Some(config_etag),
-                candidate_digest,
-                affected_targets: vec![TierMutationIntentTarget {
-                    tier_name: tier_name.to_string(),
-                    old_backend_identity: None,
-                    new_backend_identity: Some(backend_identity),
-                }],
-                expires_at_unix_nanos: 1,
-            };
-            // Persist the restart state directly: no peer Commit or runtime block
-            // installation may leave a notification that masks a missing startup wakeup.
-            save_tier_mutation_intent_record(store.clone(), &intent)
-                .await
-                .expect("committed restart intent should persist without notifying the manager");
-            com::save_config(store.clone(), &config_path, vec![0])
-                .await
-                .expect("controlled invalid config should persist");
-
-            runtime_sources::TEST_TIER_CONFIG_MGR
-                .scope(manager.clone(), async {
-                    let err = runtime_sources::init_tier_config_mgr(store.clone())
-                        .await
-                        .expect_err("initial reload must report the controlled config parse failure");
-                    assert!(err.to_string().contains("tierConfigInit: no data"), "unexpected startup failure: {err}");
-                })
-                .await;
-            assert!(manager.read().await.tiers.is_empty(), "failed startup must not publish the candidate");
-            let blocked = match TierConfigMgr::acquire_operation_lease(&manager, tier_name).await {
-                Ok(_) => panic!("failed startup must retain the recovered committed fence"),
-                Err(err) => err,
-            };
-            assert!(TierConfigMgr::operation_lease_blocked_by_mutation(&blocked));
-
-            // Restore only the durable bytes. Recovery must come from the worker
-            // started by init, after the test-only handle scope has already ended.
-            com::save_config(store.clone(), &config_path, config_bytes)
-                .await
-                .expect("restoring committed config bytes should remove the startup failure");
-            // Reload only publishes this Add into an empty manager, so it has
-            // no replaced backend to construct or probe. Only the lease check
-            // below needs a driver; keep its factory on the observing task.
-            let driver_factory: TierDriverTestFactory = Arc::new(move |_| Ok(Box::new(backend.clone())));
-            TIER_DRIVER_TEST_FACTORY
-                .scope(driver_factory, async {
-                    tokio::time::timeout(Duration::from_secs(10), async {
-                        loop {
-                            match TierConfigMgr::acquire_operation_lease(&manager, tier_name).await {
-                                Ok(lease) => {
-                                    drop(lease);
-                                    match load_tier_mutation_intent_record(store.clone(), mutation_id).await {
-                                        Err(Error::ConfigNotFound) => break,
-                                        Ok(retained) => assert_eq!(retained, intent),
-                                        Err(err) => panic!("committed recovery intent should remain readable: {err}"),
-                                    }
-                                }
-                                Err(err) => assert!(
-                                    TierConfigMgr::operation_lease_blocked_by_mutation(&err),
-                                    "recovery must retain the fence until the tier becomes available: {err}"
-                                ),
-                            }
-                            tokio::time::sleep(Duration::from_millis(10)).await;
-                        }
-                    })
-                    .await
-                    .expect("startup worker must publish the tier and clean its fence without another mutation");
-                })
-                .await;
-            let recovered = manager.read().await;
-            assert_eq!(
-                tier_config_candidate_digest(&recovered).expect("recovered config digest should build"),
-                candidate_digest,
-                "startup recovery must publish the committed configuration"
-            );
-        });
-    }
-
     #[cfg(feature = "test-util")]
     #[tokio::test]
     #[serial_test::serial(storage_class_env)]

crates/ecstore/src/store/list_objects.rs

@@ -164,6 +164,16 @@ pub fn max_keys_plus_one(max_keys: i32, add_one: bool) -> i32 {
     max_keys
 }
 
+fn list_versions_scan_limit(max_keys: i32, has_version_marker: bool) -> i32 {
+    if max_keys <= 0 {
+        return 0;
+    }
+
+    // The marker object's versions may all be filtered out after gathering.
+    // Reserve its raw entry in addition to the next-page lookahead entry.
+    max_keys_plus_one(max_keys, true) + i32::from(has_version_marker)
+}
+
 #[derive(Debug, Clone, Copy, Eq, PartialEq)]
 enum GatherResultsState {
     LimitReached,
@@ -2139,15 +2149,19 @@ fn build_list_versions_next_marker(
         // here; advertise it as the literal `null` marker so a resumed listing
         // parses it back to `VersionMarker::Null` instead of a nil UUID that
         // `find_version_index` can never match (issue #6745).
-        (
+        let version_marker = if last.is_dir && last.mod_time.is_none() {
-            Some(append_list_cache_id_to_marker(last.name.clone(), cache_id)),
+            // A CommonPrefix has no version to resume; a version marker would
+            // make the next page include this same prefix again.
+            None
+        } else {
             Some(
                 last.version_id
                     .filter(|v| !v.is_nil())
                     .map(|v| v.to_string())
                     .unwrap_or_else(|| "null".to_string()),
-            ),
+            )
-        )
+        };
+        (Some(append_list_cache_id_to_marker(last.name.clone(), cache_id)), version_marker)
     } else if let Some(last_prefix) = prefixes.last() {
         (Some(append_list_cache_id_to_marker(last_prefix.clone(), cache_id)), None)
     } else {
@@ -2866,6 +2880,20 @@ fn listing_entries_supplement_target(
         return None;
     }
 
+    if let Some(directory) = entries.0.iter().flatten().find(|entry| entry.is_dir()) {
+        let directory_copies = entries
+            .0
+            .iter()
+            .flatten()
+            .filter(|entry| entry.is_dir() && entry.name == directory.name)
+            .count();
+        // A committed child may have some of its directory copies only on
+        // fallback disks, just like object metadata in a partial primary sample.
+        if directory_copies < resolver.dir_quorum {
+            return Some(directory.name.clone());
+        }
+    }
+
     for (idx, entry) in entries.0.iter().enumerate() {
         let Some(entry) = entry.as_ref().filter(|entry| entry.is_object()) else {
             continue;
@@ -4018,8 +4046,7 @@ impl ECStore {
             None
         };
 
-        let effective_max_keys = if max_keys <= 0 { 0 } else { max_keys_plus_one(max_keys, true) };
+        let effective_max_keys = list_versions_scan_limit(max_keys, has_version_marker);
-        // Always request max_keys + 1 to detect if there are more results
         let mut opts = ListPathOptions {
             bucket: bucket.to_owned(),
             prefix: prefix.to_owned(),
@@ -5325,7 +5352,7 @@ impl Sets {
             None
         };
 
-        let effective_max_keys = if max_keys <= 0 { 0 } else { max_keys_plus_one(max_keys, true) };
+        let effective_max_keys = list_versions_scan_limit(max_keys, has_version_marker);
         let mut opts = ListPathOptions {
             bucket: bucket.to_owned(),
             prefix: prefix.to_owned(),
@@ -6034,7 +6061,7 @@ impl SetDisks {
 
         let has_version_marker = version_marker.is_some();
         let version_marker = version_marker.map(parse_version_marker).transpose()?;
-        let effective_max_keys = if max_keys <= 0 { 0 } else { max_keys_plus_one(max_keys, true) };
+        let effective_max_keys = list_versions_scan_limit(max_keys, has_version_marker);
         let mut opts = ListPathOptions {
             bucket: bucket.to_owned(),
             prefix: prefix.to_owned(),
@@ -6248,7 +6275,7 @@ impl SetDisks {
             None
         };
 
-        let effective_max_keys = if max_keys <= 0 { 0 } else { max_keys_plus_one(max_keys, true) };
+        let effective_max_keys = list_versions_scan_limit(max_keys, has_version_marker);
         let mut opts = ListPathOptions {
             bucket: bucket.to_owned(),
             prefix: prefix.to_owned(),
@@ -7441,6 +7468,153 @@ mod test {
         assert!(cancel.is_cancelled());
     }
 
+    #[test]
+    fn list_versions_pagination_scan_limit_boundaries() {
+        for has_version_marker in [false, true] {
+            assert_eq!(super::list_versions_scan_limit(-1, has_version_marker), 0);
+            assert_eq!(super::list_versions_scan_limit(0, has_version_marker), 0);
+            let marker_slot = i32::from(has_version_marker);
+            assert_eq!(super::list_versions_scan_limit(1, has_version_marker), 2 + marker_slot);
+            assert_eq!(super::list_versions_scan_limit(MAX_OBJECT_LIST, has_version_marker), 1001 + marker_slot);
+            assert_eq!(super::list_versions_scan_limit(i32::MAX, has_version_marker), 1001 + marker_slot);
+        }
+    }
+
+    #[tokio::test]
+    async fn list_versions_pagination_does_not_require_an_empty_final_page() {
+        use crate::bucket::metadata_sys::{init_bucket_metadata_sys, test_support::isolated_store_over_temp_disks};
+        use crate::storage_api_contracts::bucket::{BucketOperations as _, MakeBucketOptions};
+
+        let (dirs, store) = isolated_store_over_temp_disks().await;
+        let bucket = "version-pagination-bucket";
+        init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+        store
+            .make_bucket(bucket, &MakeBucketOptions::default())
+            .await
+            .expect("pagination bucket should be created");
+        let mod_time = time::OffsetDateTime::from_unix_timestamp(1_705_312_300).expect("valid timestamp");
+
+        for kind in ["objects", "deletes", "null", "mixed", "delimiter"] {
+            let count = if kind == "mixed" { 5 } else { 10 };
+            let mut expected = Vec::new();
+            for index in 0..count {
+                let name = if kind == "delimiter" && index % 2 == 1 {
+                    format!("{kind}/testobject-{index:02}/child")
+                } else {
+                    format!("{kind}/testobject-{index:02}")
+                };
+                let entry = match kind {
+                    "deletes" => test_delete_marker_meta_entry(&name, mod_time),
+                    "null" => test_object_meta_entry(&name),
+                    "mixed" => test_object_with_delete_marker_meta_entry(&name, mod_time, mod_time + time::Duration::SECOND),
+                    _ => test_object_meta_entry_with_erasure_versions(&name, &[(mod_time, "etag", 2, 2)]),
+                };
+                for dir in &dirs {
+                    let object_dir = dir.path().join(bucket).join(&name);
+                    tokio::fs::create_dir_all(&object_dir)
+                        .await
+                        .expect("pagination object directory should be created");
+                    tokio::fs::write(object_dir.join(STORAGE_FORMAT_FILE), &entry.metadata)
+                        .await
+                        .expect("pagination metadata should be written");
+                }
+                if kind == "delimiter" && index % 2 == 1 {
+                    expected.push((name.trim_end_matches("child").to_owned(), None, false));
+                } else {
+                    let versions = entry.file_info_versions(bucket).expect("fixture versions should decode");
+                    expected.extend(
+                        versions
+                            .versions
+                            .iter()
+                            .map(|version| (name.clone(), version.version_id, version.deleted)),
+                    );
+                }
+            }
+            let prefix = format!("{kind}/");
+            let delimiter = (kind == "delimiter").then(|| "/".to_owned());
+            // Exercise each public/internal entry point with the reported page size.
+            // The store entry point also covers exact and one-over limit boundaries.
+            for (layer, max_keys) in [(0, 0), (0, 1), (0, 5), (0, 9), (0, 10), (0, 11), (1, 5), (2, 5), (3, 5)] {
+                if layer == 3 && delimiter.is_some() {
+                    continue;
+                }
+                let mut marker = None;
+                let mut version_marker = None;
+                let expected_pages = if max_keys == 0 {
+                    1
+                } else {
+                    10usize.div_ceil(usize::try_from(max_keys).expect("positive page size"))
+                };
+                let mut actual = Vec::new();
+                for page in 0..expected_pages {
+                    let result = match layer {
+                        0 => {
+                            store
+                                .clone()
+                                .inner_list_object_versions(bucket, &prefix, marker, version_marker, delimiter.clone(), max_keys)
+                                .await
+                        }
+                        1 => {
+                            store.pools[0]
+                                .clone()
+                                .inner_list_object_versions(bucket, &prefix, marker, version_marker, delimiter.clone(), max_keys)
+                                .await
+                        }
+                        2 => {
+                            store.pools[0].disk_set[0]
+                                .clone()
+                                .inner_list_object_versions(bucket, &prefix, marker, version_marker, delimiter.clone(), max_keys)
+                                .await
+                        }
+                        _ => {
+                            store.pools[0].disk_set[0]
+                                .clone()
+                                .inner_list_object_versions_for_recursive_delete(
+                                    bucket,
+                                    &prefix,
+                                    marker,
+                                    version_marker,
+                                    max_keys,
+                                )
+                                .await
+                        }
+                    }
+                    .expect("version page should list successfully");
+                    let page_size = usize::try_from(max_keys).expect("nonnegative page size");
+                    assert_eq!(result.objects.len() + result.prefixes.len(), (10 - page * page_size).min(page_size));
+                    let has_more = page + 1 < expected_pages;
+                    assert_eq!(result.is_truncated, has_more, "{kind}, layer {layer}, max_keys {max_keys}, page {page}");
+                    assert_eq!(
+                        result.next_marker.is_some(),
+                        has_more,
+                        "key marker must exist only when another page exists"
+                    );
+                    if !has_more {
+                        assert!(
+                            result.next_version_idmarker.is_none(),
+                            "the final page must not advertise a version marker"
+                        );
+                    }
+                    actual.extend(
+                        result
+                            .objects
+                            .into_iter()
+                            .map(|object| (object.name, object.version_id, object.delete_marker)),
+                    );
+                    actual.extend(result.prefixes.into_iter().map(|prefix| (prefix, None, false)));
+                    marker = result.next_marker;
+                    version_marker = result.next_version_idmarker;
+                }
+                // Objects and CommonPrefixes are serialized separately; compare their
+                // identities without relying on their relative position in the response.
+                actual.sort();
+                let mut expected = if max_keys == 0 { Vec::new() } else { expected.clone() };
+                expected.sort();
+                assert_eq!(actual, expected, "{kind}, layer {layer}, max_keys {max_keys}");
+            }
+        }
+    }
+
     #[test]
     fn version_marker_is_applied_only_when_key_marker_entry_is_present() {
         let version_marker = Some(VersionMarker::Null);
@@ -9448,6 +9622,71 @@ mod test {
         assert!(supplemented.is_latest_delete_marker());
     }
 
+    #[tokio::test]
+    async fn latest_listing_supplement_checks_fallback_disks_for_common_prefix_quorum() {
+        let mut fallback_disks = Vec::new();
+        let mut fallback_tempdirs = Vec::new();
+        for index in 0..4 {
+            let tempdir = tempfile::tempdir().expect("fallback tempdir should be created");
+            let endpoint = Endpoint::try_from(tempdir.path().to_str().expect("fallback path should be utf8"))
+                .expect("fallback endpoint should parse");
+            let disk = new_disk(
+                &endpoint,
+                &DiskOption {
+                    cleanup: false,
+                    health_check: false,
+                },
+            )
+            .await
+            .expect("fallback disk should be created");
+            disk.make_volume("bucket").await.expect("fallback bucket should be created");
+            for copies in [3, 4] {
+                if index < copies {
+                    let object = format!("quux-{copies}/thud");
+                    let entry = test_object_meta_entry(&object);
+                    disk.write_all("bucket", &format!("{object}/{STORAGE_FORMAT_FILE}"), bytes::Bytes::from(entry.metadata))
+                        .await
+                        .expect("fallback child metadata should be written");
+                }
+            }
+            fallback_disks.push(disk);
+            fallback_tempdirs.push(tempdir);
+        }
+        let supplement = ListingSupplement::new(
+            ListingSupplementOptions {
+                bucket: "bucket".to_owned(),
+                path: String::new(),
+                recursive: false,
+                incl_deleted: false,
+                skip_hidden_prefix_check: false,
+                filter_prefix: None,
+                forward_to: None,
+                per_disk_limit: 100,
+                skip_total_timeout: true,
+                walkdir_timeout: None,
+                walkdir_stall_timeout: None,
+            },
+            Arc::new(fallback_disks),
+            FallbackClaimTracker::default(),
+        );
+        // A 16-drive EC:4 set asks 12 primary disks. A committed write may
+        // exist on eight primary disks and all four remaining fallback disks.
+        let resolver = list_metadata_resolution_params("bucket".to_owned(), 4, 12, false, 0);
+        for fallback_copies in [3, 4] {
+            let prefix = format!("quux-{fallback_copies}/");
+            let mut primary = vec![Some(test_dir_meta_entry(&prefix)); 8];
+            primary.extend([None, None, None, None]);
+            let entry =
+                resolve_listing_entries_with_supplement(MetaCacheEntries(primary), resolver.clone(), true, supplement.clone())
+                    .await;
+            assert_eq!(
+                entry.map(|entry| entry.name),
+                (fallback_copies == 4).then_some(prefix),
+                "the common prefix needs all twelve copies, including fallback disks"
+            );
+        }
+    }
+
     #[test]
     fn latest_listing_supplement_keeps_a_subquorum_delete_marker_hidden() {
         let object_mod_time = time::OffsetDateTime::from_unix_timestamp(1_705_312_300).expect("valid timestamp");

crates/ecstore/src/store/object.rs

@@ -3079,12 +3079,7 @@ impl ECStore {
         let store = Arc::clone(self);
         let write = async move {
             let object = "buckets/.scanner-pause-backlog.json";
-            let mut opts = ObjectOptions {
+            let mut opts = ObjectOptions::default();
-                max_parity: true,
-                http_preconditions: Some(preconditions),
-                write_completion: crate::object_api::WriteCompletion::TailDrained,
-                ..Default::default()
-            };
             // Match migration: fixed object namespace -> durable pool metadata ->
             // actual replica namespace. The replica need not be the hash-routed set.
             let object_guard = if store.single_pool() {
@@ -3110,9 +3105,14 @@ impl ECStore {
             } else {
                 None
             };
-            let result = set
+            let result = crate::data_movement::scanner_backlog::persist_native_scanner_pause_backlog_replica(
-                .put_object(RUSTFS_META_BUCKET, object, &mut PutObjReader::from_vec(data), &opts)
+                set,
-                .await;
+                data,
+                preconditions,
+                opts,
+                "publish",
+            )
+            .await;
             drop(capacity_guard);
             drop(object_guard);
             result
@@ -3747,29 +3747,37 @@ impl ECStore {
         opts: &ObjectOptions,
         no_lock: bool,
     ) -> Result<usize> {
+        let capacity_owner = DecommissionCapacityOwner::from_options(opts);
         match self
             .get_pool_info_existing_with_opts(bucket, object, &data_movement_pool_lookup_opts(opts, no_lock))
             .await
         {
-            Ok((pinfo, _)) => Ok(pinfo.index),
+            Ok((pinfo, _)) => {
+                if let Some(owner) = capacity_owner {
+                    if self.is_decommission_capacity_target_reserved(owner, pinfo.index).await? {
+                        return Ok(pinfo.index);
+                    }
+                } else {
+                    return Ok(pinfo.index);
+                }
+            }
             Err(err) => {
                 if !is_err_object_not_found(&err) && !is_err_version_not_found(&err) {
                     return Err(err);
                 }
-
-                if let Some(owner) = DecommissionCapacityOwner::from_options(opts) {
-                    let expected_data_bytes = opts
-                        .capacity_expected_data_bytes()
-                        .or_else(|| usize::try_from(size).ok())
-                        .unwrap_or_default();
-                    return self
-                        .select_decommission_capacity_target_pool(owner, expected_data_bytes)
-                        .await;
-                }
-
-                self.get_available_pool_idx(bucket, object, size).await.ok_or(Error::DiskFull)
             }
         }
+        if let Some(owner) = capacity_owner {
+            let expected_data_bytes = opts
+                .capacity_expected_data_bytes()
+                .or_else(|| usize::try_from(size).ok())
+                .unwrap_or_default();
+            return self
+                .select_decommission_capacity_target_pool(owner, expected_data_bytes)
+                .await;
+        }
+
+        self.get_available_pool_idx(bucket, object, size).await.ok_or(Error::DiskFull)
     }
 
     async fn find_data_movement_target_info(
@@ -4251,8 +4259,10 @@ impl ECStore {
     }
 
     /// Return metadata for DELETE preflight, including an explicitly addressed
-    /// delete marker. Read APIs must keep using `get_object_info`; authorization
+    /// delete marker. GET/HEAD may also use this metadata-only lookup to enrich
-    /// and Object Lock enforcement still belong to the caller and locked delete.
+    /// an already failed read with marker headers, never to serve marker data.
+    /// Normal reads must keep using `get_object_info`; authorization and Object
+    /// Lock enforcement still belong to the caller and locked delete.
     #[instrument(level = "trace", skip_all)]
     pub async fn get_object_info_for_delete(&self, bucket: &str, object: &str, opts: &ObjectOptions) -> Result<ObjectInfo> {
         self.get_object_info_snapshot(bucket, object, opts, true).await

crates/heal/Cargo.toml

@@ -31,6 +31,7 @@ workspace = true
 
 [features]
 default = []
+test-util = []
 hotpath = [
     "hotpath/hotpath",
     "hotpath/tokio",
@@ -104,6 +105,7 @@ walkdir = { workspace = true }
 http = { workspace = true }
 temp-env = { workspace = true, features = ["async_closure"] }
 tokio = { workspace = true, features = ["test-util", "fs", "rt-multi-thread"] }
+chrono = { workspace = true }
 
 [lib]
 doctest = false

crates/heal/src/heal/channel.rs

@@ -847,7 +847,7 @@ mod tests {
 
     fn create_test_heal_manager() -> Arc<HealManager> {
         let storage: Arc<dyn HealStorageAPI> = Arc::new(MockStorage);
-        Arc::new(HealManager::new(storage, None))
+        Arc::new(HealManager::new_without_root_recovery_for_test(storage, None))
     }
 
     #[test]
@@ -1481,7 +1481,7 @@ mod tests {
     #[tokio::test]
     async fn test_process_start_request_returns_admission_result() {
         let storage: Arc<dyn HealStorageAPI> = Arc::new(MockStorage);
-        let manager = Arc::new(HealManager::new(
+        let manager = Arc::new(HealManager::new_without_root_recovery_for_test(
             storage,
             Some(HealConfig {
                 queue_size: 1,
@@ -1826,7 +1826,7 @@ mod tests {
 
     #[tokio::test]
     async fn test_process_query_request_reports_displaced_terminal_detail() {
-        let heal_manager = Arc::new(HealManager::new(
+        let heal_manager = Arc::new(HealManager::new_without_root_recovery_for_test(
             Arc::new(MockStorage),
             Some(HealConfig {
                 queue_size: 1,

crates/heal/src/heal/erasure_healer.rs

@@ -851,7 +851,7 @@ impl ErasureSetHealer {
         }
 
         if failed_objects == 0 && skipped_objects == 0 && failed_buckets == 0 {
-            self.heal_pool_metadata(
+            self.heal_replacement_pool_metadata(
                 set_disk_id,
                 &mut ErasureSetPassCounters {
                     processed_objects: &mut processed_objects,
@@ -950,43 +950,26 @@ impl ErasureSetHealer {
         Ok(())
     }
 
-    async fn heal_pool_metadata(
+    async fn heal_replacement_pool_metadata(
         &self,
         set_disk_id: &str,
         counters: &mut ErasureSetPassCounters<'_>,
         resume_manager: &ResumeManager,
         checkpoint_manager: &CheckpointManager,
     ) -> Result<()> {
-        let mut metadata_opts = self.heal_opts;
+        let target_endpoints = if self.pool_metadata_target_endpoints.is_empty() {
-        metadata_opts.remove = false;
-        metadata_opts.no_lock = false;
-        if self.replacement_task_id.is_none() {
-            let (pool_index, set_index) = crate::heal::utils::parse_set_disk_id(set_disk_id)?;
-            if metadata_opts.pool.is_some_and(|pool| pool != pool_index) || metadata_opts.set.is_some_and(|set| set != set_index)
-            {
-                return Err(Error::TaskExecutionFailed {
-                    message: format!("Pool metadata scope does not match resumed set {set_disk_id}"),
-                });
-            }
-            metadata_opts.pool = Some(pool_index);
-            metadata_opts.set = Some(set_index);
-        }
-        let target_endpoints = if self.replacement_task_id.is_some() || self.pool_metadata_target_endpoints.is_empty() {
             self.target_endpoints.as_ref()
         } else {
             self.pool_metadata_target_endpoints.as_ref()
         };
-        let target_scoped_recreate = !metadata_opts.dry_run && metadata_opts.recreate && !target_endpoints.is_empty();
+        let target_scoped_recreate = !self.heal_opts.dry_run && self.heal_opts.recreate && !target_endpoints.is_empty();
-        let ordinary_heal = self.replacement_task_id.is_none() && !target_scoped_recreate;
+        if self.replacement_task_id.is_none() && !target_scoped_recreate {
-        if !ordinary_heal {
+            return Ok(());
-            if target_endpoints.is_empty() {
+        }
-                return Err(Error::TaskExecutionFailed {
+        if target_endpoints.is_empty() {
-                    message: "Replacement pool metadata heal requires target endpoints".to_string(),
+            return Err(Error::TaskExecutionFailed {
-                });
+                message: "Replacement pool metadata heal requires target endpoints".to_string(),
-            }
+            });
-            if !self.storage.replacement_pool_metadata_applies(&metadata_opts).await? {
-                return Ok(());
-            }
         }
 
         let object_key = format!("{RUSTFS_META_BUCKET}/{POOL_META_NAME}");
@@ -1005,88 +988,61 @@ impl ErasureSetHealer {
             .set_current_item(Some(RUSTFS_META_BUCKET.to_string()), Some(POOL_META_NAME.to_string()))
             .await?;
 
-        let result = if ordinary_heal {
+        let result = match self
-            match self.storage.heal_pool_metadata(&metadata_opts).await {
+            .storage
-                Ok(results) if results.is_empty() => return Ok(()),
+            .heal_object(RUSTFS_META_BUCKET, POOL_META_NAME, None, &self.heal_opts)
-                Ok(results) => {
+            .await
-                    let [result] = results.as_slice() else {
+        {
-                        return Err(Error::TaskExecutionFailed {
+            Ok((result, None)) if target_outcomes_complete(&result, target_endpoints) => {
-                            message: format!("Pool metadata returned multiple replicas for set {set_disk_id}"),
+                let object_size = result_object_size_u64(&result);
-                        });
+                match self
-                    };
+                    .storage
-                    (result_object_size_u64(result), Ok(()))
+                    .replacement_targets_have_version(RUSTFS_META_BUCKET, POOL_META_NAME, None, &self.heal_opts, target_endpoints)
+                    .await
+                {
+                    Ok(true) => (object_size, Ok(())),
+                    Ok(false) => (
+                        object_size,
+                        Err(Error::transient_skip(
+                            "Skipped replacement pool metadata heal because target readback did not confirm the committed version",
+                        )),
+                    ),
+                    Err(err) => (
+                        object_size,
+                        Err(Error::transient_skip(format!(
+                            "Skipped replacement pool metadata heal because target readback failed: {err}"
+                        ))),
+                    ),
                 }
-                Err(err @ Error::TaskCancelled) | Err(err @ Error::TaskTimeout) => return Err(err),
-                Err(err) => match Self::classify_heal_object_error(&err) {
-                    HealObjectOutcome::Absent | HealObjectOutcome::Transient => {
-                        (0, Err(Error::transient_skip(format!("Pool metadata heal must be retried: {err}"))))
-                    }
-                    HealObjectOutcome::Failed => (0, Err(err)),
-                },
             }
-        } else {
+            Ok((result, None)) => (
-            match self
+                result_object_size_u64(&result),
-                .storage
+                Err(Error::transient_skip(
-                .heal_object(RUSTFS_META_BUCKET, POOL_META_NAME, None, &metadata_opts)
+                    "Skipped replacement pool metadata heal because a replacement target was not committed",
-                .await
+                )),
-            {
+            ),
-                Ok((result, None)) if target_outcomes_complete(&result, target_endpoints) => {
+            Ok((result, Some(err))) => {
-                    let object_size = result_object_size_u64(&result);
+                let object_size = result_object_size_u64(&result);
-                    match self
+                match Self::classify_heal_object_error(&err) {
-                        .storage
-                        .replacement_targets_have_version(
-                            RUSTFS_META_BUCKET,
-                            POOL_META_NAME,
-                            None,
-                            &metadata_opts,
-                            target_endpoints,
-                        )
-                        .await
-                    {
-                        Ok(true) => (object_size, Ok(())),
-                        Ok(false) => (
-                            object_size,
-                            Err(Error::transient_skip(
-                                "Skipped replacement pool metadata heal because target readback did not confirm the committed version",
-                            )),
-                        ),
-                        Err(err) => (
-                            object_size,
-                            Err(Error::transient_skip(format!(
-                                "Skipped replacement pool metadata heal because target readback failed: {err}"
-                            ))),
-                        ),
-                    }
-                }
-                Ok((result, None)) => (
-                    result_object_size_u64(&result),
-                    Err(Error::transient_skip(
-                        "Skipped replacement pool metadata heal because a replacement target was not committed",
-                    )),
-                ),
-                Ok((result, Some(err))) => {
-                    let object_size = result_object_size_u64(&result);
-                    match Self::classify_heal_object_error(&err) {
-                        HealObjectOutcome::Absent | HealObjectOutcome::Transient => (
-                            object_size,
-                            Err(Error::transient_skip(format!(
-                                "Skipped replacement pool metadata heal due to transient error: {err}"
-                            ))),
-                        ),
-                        HealObjectOutcome::Failed => (object_size, Err(err)),
-                    }
-                }
-                Err(err @ Error::TaskCancelled) | Err(err @ Error::TaskTimeout) => return Err(err),
-                Err(err) => match Self::classify_heal_object_error(&err) {
                     HealObjectOutcome::Absent | HealObjectOutcome::Transient => (
-                        0,
+                        object_size,
                         Err(Error::transient_skip(format!(
                             "Skipped replacement pool metadata heal due to transient error: {err}"
                         ))),
                     ),
-                    HealObjectOutcome::Failed => (0, Err(err)),
+                    HealObjectOutcome::Failed => (object_size, Err(err)),
-                },
+                }
             }
+            Err(err @ Error::TaskCancelled) | Err(err @ Error::TaskTimeout) => return Err(err),
+            Err(err) => match Self::classify_heal_object_error(&err) {
+                HealObjectOutcome::Absent | HealObjectOutcome::Transient => (
+                    0,
+                    Err(Error::transient_skip(format!(
+                        "Skipped replacement pool metadata heal due to transient error: {err}"
+                    ))),
+                ),
+                HealObjectOutcome::Failed => (0, Err(err)),
+            },
         };
 
         let (object_size, result) = result;
@@ -1105,7 +1061,7 @@ impl ErasureSetHealer {
                     bucket = RUSTFS_META_BUCKET,
                     object = POOL_META_NAME,
                     state = "healed",
-                    "Pool metadata healed"
+                    "Replacement pool metadata healed"
                 );
                 CheckpointObjectOutcome::Processed
             }
@@ -1122,7 +1078,7 @@ impl ErasureSetHealer {
                     object = POOL_META_NAME,
                     state = "transient_skip",
                     error = %message,
-                    "Pool metadata heal skipped due to transient error"
+                    "Replacement pool metadata heal skipped due to transient error"
                 );
                 CheckpointObjectOutcome::Skipped
             }
@@ -1139,7 +1095,7 @@ impl ErasureSetHealer {
                     object = POOL_META_NAME,
                     state = "failed",
                     error = %err,
-                    "Pool metadata heal failed"
+                    "Replacement pool metadata heal failed"
                 );
                 CheckpointObjectOutcome::Failed
             }
@@ -1580,9 +1536,7 @@ impl ErasureSetHealer {
                         );
                         CheckpointObjectOutcome::Processed
                     }
-                    Err(err @ Error::TaskCancelled) | Err(err @ Error::TaskTimeout) => {
+                    Err(err @ Error::TaskCancelled) | Err(err @ Error::TaskTimeout) => return Err(err),
-                        return Err(err);
-                    }
                     Err(Error::TransientSkip { message }) => {
                         telemetry_unknown |= !increment_counter(skipped_objects);
                         telemetry_unknown |= !add_bytes(&mut bytes_processed, object_size);
@@ -2084,8 +2038,6 @@ mod resume_loop_tests {
     #[derive(Clone)]
     enum HealOutcome {
         Ok,
-        /// The object has no metadata on any disk in the selected set.
-        FileNotFound,
         /// The version vanished before heal ran (deleted mid-heal).
         VersionNotFound,
         /// A transient infrastructure condition (offline disk / unmet quorum):
@@ -2111,10 +2063,6 @@ mod resume_loop_tests {
         /// Target-specific physical readback evidence per `compose_key`; the
         /// fake models a healthy backend unless a test explicitly revokes it.
         replacement_commit_evidence: Mutex<HashMap<String, ReplacementCommitEvidence>>,
-        ordinary_pool_metadata_required: AtomicBool,
-        pool_metadata_opts: Mutex<Vec<HealOpts>>,
-        pool_metadata_not_applicable: AtomicBool,
-        fail_pool_metadata_scope: AtomicBool,
         lifecycle_expired: Mutex<HashSet<String>>,
         /// every heal_object call recorded as (name, version_id)
         heal_calls: Mutex<Vec<(String, Option<String>)>>,
@@ -2203,14 +2151,11 @@ mod resume_loop_tests {
         }
         async fn heal_object(
             &self,
-            bucket: &str,
+            _bucket: &str,
             object: &str,
             version_id: Option<&str>,
-            opts: &HealOpts,
+            _opts: &HealOpts,
         ) -> Result<(HealResultItem, Option<Error>)> {
-            if bucket == RUSTFS_META_BUCKET && object == POOL_META_NAME {
-                self.pool_metadata_opts.lock().expect("metadata options").push(*opts);
-            }
             self.heal_calls
                 .lock()
                 .unwrap()
@@ -2219,7 +2164,6 @@ mod resume_loop_tests {
             let outcome = self.outcomes.lock().unwrap().get(&key).cloned().unwrap_or(HealOutcome::Ok);
             match outcome {
                 HealOutcome::Ok => Ok((self.results.lock().unwrap().get(&key).cloned().unwrap_or_default(), None)),
-                HealOutcome::FileNotFound => Ok((HealResultItem::default(), Some(Error::Storage(EcstoreError::FileNotFound)))),
                 HealOutcome::VersionNotFound => {
                     Ok((HealResultItem::default(), Some(Error::Storage(EcstoreError::FileVersionNotFound))))
                 }
@@ -2233,30 +2177,6 @@ mod resume_loop_tests {
         async fn heal_format(&self, _dry: bool) -> Result<(HealResultItem, Option<Error>)> {
             Ok((HealResultItem::default(), None))
         }
-        async fn heal_pool_metadata(&self, opts: &HealOpts) -> Result<Vec<HealResultItem>> {
-            if !self.ordinary_pool_metadata_required.load(Ordering::SeqCst) {
-                return Ok(Vec::new());
-            }
-            if !self.replacement_pool_metadata_applies(opts).await? {
-                return Ok(Vec::new());
-            }
-            let (result, error) = self.heal_object(RUSTFS_META_BUCKET, POOL_META_NAME, None, opts).await?;
-            if let Some(error) = error {
-                return Err(error);
-            }
-            Ok(vec![result])
-        }
-        async fn replacement_pool_metadata_applies(&self, opts: &HealOpts) -> Result<bool> {
-            if self.fail_pool_metadata_scope.load(Ordering::SeqCst) {
-                return Err(Error::other("injected pool metadata scope failure"));
-            }
-            if self.pool_metadata_not_applicable.load(Ordering::SeqCst) {
-                assert_eq!(opts.pool, Some(0));
-                assert_eq!(opts.set, Some(1));
-                return Ok(false);
-            }
-            Ok(true)
-        }
         async fn replacement_targets_have_version(
             &self,
             _bucket: &str,
@@ -2748,115 +2668,6 @@ mod resume_loop_tests {
         assert!(state.completed, "successful data heal must be persisted before cleanup is attempted");
     }
 
-    #[tokio::test]
-    async fn ordinary_set_heals_pool_metadata_without_replacement_generation_or_targets() {
-        let env = make_env().await;
-        env.storage.ordinary_pool_metadata_required.store(true, Ordering::SeqCst);
-        env.storage
-            .set_result(POOL_META_NAME, None, replacement_target_ok_result("metadata-disk", POOL_META_NAME));
-        assert!(env.healer.replacement_task_id.is_none());
-        assert!(env.healer.target_endpoints.is_empty());
-
-        env.healer
-            .execute_heal_with_resume(&[], "pool_0_set_0", &env.resume, &env.checkpoint)
-            .await
-            .expect("ordinary set recovery should repair metadata even without user buckets");
-
-        assert_eq!(env.storage.calls(), vec![(POOL_META_NAME.to_string(), None)]);
-        {
-            let opts = env.storage.pool_metadata_opts.lock().expect("metadata options");
-            assert_eq!(opts.len(), 1);
-            assert_eq!((opts[0].pool, opts[0].set), (Some(0), Some(0)));
-        }
-        let state = env.resume.get_state().await;
-        assert!(state.completed);
-        assert_eq!(state.successful_objects, 1, "metadata must enter durable completion counters");
-    }
-
-    #[tokio::test]
-    async fn ordinary_set_pool_metadata_respects_non_owner_and_dry_run() {
-        let mut env = make_env().await;
-        env.storage.ordinary_pool_metadata_required.store(true, Ordering::SeqCst);
-        env.storage.pool_metadata_not_applicable.store(true, Ordering::SeqCst);
-        env.healer.heal_opts.pool = Some(0);
-        env.healer.heal_opts.set = Some(1);
-        env.healer
-            .execute_heal_with_resume(&[], "pool_0_set_1", &env.resume, &env.checkpoint)
-            .await
-            .expect("a valid non-owner set must not invent a metadata replica");
-        assert!(env.storage.calls().is_empty());
-        assert_eq!(env.resume.get_state().await.successful_objects, 0);
-
-        let mut env = make_env().await;
-        env.storage.ordinary_pool_metadata_required.store(true, Ordering::SeqCst);
-        env.healer.heal_opts.dry_run = true;
-        env.healer.heal_opts.remove = true;
-        env.healer.heal_opts.no_lock = true;
-        env.storage.set_replacement_commit_evidence(POOL_META_NAME, None, false);
-        env.healer
-            .execute_heal_with_resume(&[], "pool_0_set_0", &env.resume, &env.checkpoint)
-            .await
-            .expect("ordinary dry-run metadata work must not require a replacement commit");
-        assert_eq!(env.storage.calls(), vec![(POOL_META_NAME.to_string(), None)]);
-        let opts = env.storage.pool_metadata_opts.lock().expect("metadata options");
-        assert_eq!(opts.len(), 1);
-        assert!(opts[0].dry_run);
-        assert!(!opts[0].remove);
-        assert!(!opts[0].no_lock);
-    }
-
-    #[tokio::test]
-    async fn ordinary_set_missing_pool_metadata_preserves_retry_state() {
-        let env = make_env().await;
-        env.storage.ordinary_pool_metadata_required.store(true, Ordering::SeqCst);
-        env.storage.set_outcome(POOL_META_NAME, None, HealOutcome::FileNotFound);
-
-        let error = env
-            .healer
-            .execute_heal_with_resume(&[], "pool_0_set_0", &env.resume, &env.checkpoint)
-            .await
-            .expect_err("missing required pool metadata must prevent ordinary set completion");
-
-        assert!(matches!(error, Error::TransientSkip { .. }));
-        let state = env.resume.get_state().await;
-        assert!(!state.completed);
-        assert_eq!(state.retry_count, 1);
-        assert!(CheckpointManager::has_checkpoint(&env.healer.disk, &env.task_id).await);
-    }
-
-    #[tokio::test]
-    async fn ordinary_set_pool_metadata_timeout_keeps_control_error() {
-        let env = make_env().await;
-        env.storage.ordinary_pool_metadata_required.store(true, Ordering::SeqCst);
-        env.storage.set_outcome(POOL_META_NAME, None, HealOutcome::Timeout);
-
-        let error = env
-            .healer
-            .execute_heal_with_resume(&[], "pool_0_set_0", &env.resume, &env.checkpoint)
-            .await
-            .expect_err("metadata timeout must abort the ordinary set pass");
-
-        assert!(matches!(error, Error::TaskTimeout));
-        assert!(!env.resume.get_state().await.completed);
-    }
-
-    #[tokio::test]
-    async fn ordinary_set_pool_metadata_rejects_mismatched_explicit_scope() {
-        let mut env = make_env().await;
-        env.storage.ordinary_pool_metadata_required.store(true, Ordering::SeqCst);
-        env.healer.heal_opts.pool = Some(1);
-
-        let error = env
-            .healer
-            .execute_heal_with_resume(&[], "pool_0_set_0", &env.resume, &env.checkpoint)
-            .await
-            .expect_err("explicit metadata scope must agree with the resumed set");
-
-        assert!(matches!(error, Error::TaskExecutionFailed { .. }));
-        assert!(env.storage.calls().is_empty(), "scope mismatch must fail before metadata mutation");
-        assert!(!env.resume.get_state().await.completed);
-    }
-
     #[tokio::test]
     async fn replacement_completion_keeps_resume_artifacts_until_marker_cleanup() {
         let env = make_env_with_targets(vec!["replacement-a".to_string()]).await;
@@ -2911,111 +2722,6 @@ mod resume_loop_tests {
         drop(checkpoint);
     }
 
-    #[tokio::test]
-    async fn replacement_pool_metadata_non_owner_completes_but_missing_owner_retries() {
-        for owns_pool_metadata in [false, true] {
-            let env = make_env_with_targets(vec!["replacement-a".to_string()]).await;
-            let replacement_task_id = ResumeUtils::generate_task_id();
-            let set_index = usize::from(!owns_pool_metadata);
-            let set_disk_id = format!("pool_0_set_{set_index}");
-            ResumeManager::new_replacement_intent(
-                env.healer.disk.clone(),
-                replacement_task_id.clone(),
-                set_disk_id.clone(),
-                vec!["b".to_string()],
-                vec!["replacement-a".to_string()],
-                vec![crate::heal::resume::ReplacementTargetIdentity {
-                    endpoint: "replacement-a".to_string(),
-                    canonical_path: "/mnt/replacement-a".to_string(),
-                    physical_device_ids: vec!["device-a".to_string()],
-                    filesystem_identity: "1:2:3".to_string(),
-                }],
-            )
-            .await
-            .expect("replacement intent should persist");
-            env.storage
-                .pool_metadata_not_applicable
-                .store(!owns_pool_metadata, Ordering::SeqCst);
-            env.storage.set_outcome(POOL_META_NAME, None, HealOutcome::FileNotFound);
-            let healer = ErasureSetHealer::new(
-                env.storage.clone(),
-                Arc::new(RwLock::new(HealProgress::new())),
-                CancellationToken::new(),
-                env.healer.disk.clone(),
-                HealOpts {
-                    pool: Some(0),
-                    set: Some(set_index),
-                    ..Default::default()
-                },
-                HealRequestSource::AutoHeal,
-            )
-            .with_replacement_targets(vec!["replacement-a".to_string()], Some(replacement_task_id.clone()));
-
-            let result = healer.heal_erasure_set(&["b".to_string()], &set_disk_id).await;
-            let state = ResumeManager::load_replacement_intent(env.healer.disk.clone(), &replacement_task_id)
-                .await
-                .expect("replacement state must remain until marker cleanup")
-                .get_state()
-                .await;
-            if owns_pool_metadata {
-                let error = result.expect_err("missing metadata in the owner set must keep replacement incomplete");
-                assert!(error.to_string().contains("Replacement erasure set heal incomplete"));
-                assert!(!state.completed);
-                assert_eq!(state.replacement_phase, crate::heal::resume::ReplacementPhase::Intent);
-                assert_eq!(state.retry_count, 1);
-                assert_eq!(env.storage.calls(), vec![(POOL_META_NAME.to_string(), None)]);
-            } else {
-                result.expect("a non-owner set must complete without a pool metadata replica");
-                assert!(state.completed);
-                assert_eq!(state.replacement_phase, crate::heal::resume::ReplacementPhase::Verified);
-                assert_eq!(state.retry_count, 0);
-                assert!(env.storage.calls().is_empty(), "non-owner sets must not attempt pool metadata repair");
-            }
-        }
-    }
-
-    #[tokio::test]
-    async fn replacement_pool_metadata_unknown_scope_cannot_complete() {
-        let env = make_env_with_targets(vec!["replacement-a".to_string()]).await;
-        let healer = ErasureSetHealer::new(
-            env.storage.clone(),
-            Arc::new(RwLock::new(HealProgress::new())),
-            CancellationToken::new(),
-            env.healer.disk.clone(),
-            HealOpts {
-                pool: Some(0),
-                set: Some(0),
-                ..Default::default()
-            },
-            HealRequestSource::AutoHeal,
-        )
-        .with_replacement_targets(vec!["replacement-a".to_string()], Some("generation-a".to_string()));
-        env.storage.fail_pool_metadata_scope.store(true, Ordering::SeqCst);
-        env.storage
-            .set_result(POOL_META_NAME, None, replacement_target_ok_result("replacement-a", POOL_META_NAME));
-        let mut processed_objects = 0;
-        let mut successful_objects = 0;
-        let mut failed_objects = 0;
-        let mut skipped_objects = 0;
-        let error = healer
-            .heal_pool_metadata(
-                "pool_0_set_0",
-                &mut super::ErasureSetPassCounters {
-                    processed_objects: &mut processed_objects,
-                    successful_objects: &mut successful_objects,
-                    failed_objects: &mut failed_objects,
-                    skipped_objects: &mut skipped_objects,
-                },
-                &env.resume,
-                &env.checkpoint,
-            )
-            .await
-            .expect_err("unknown metadata placement must keep replacement incomplete");
-        assert!(error.to_string().contains("injected pool metadata scope failure"));
-        assert!(env.storage.calls().is_empty());
-        assert_eq!((processed_objects, successful_objects, failed_objects, skipped_objects), (0, 0, 0, 0));
-    }
-
     #[tokio::test]
     async fn replacement_pool_metadata_readback_failure_schedules_retry() {
         let env = make_env_with_targets(vec!["replacement-a".to_string()]).await;
@@ -3075,8 +2781,8 @@ mod resume_loop_tests {
             env.healer.disk.clone(),
             HealOpts {
                 recreate: true,
-                remove: true,
+                pool: Some(0),
-                no_lock: true,
+                set: Some(0),
                 ..Default::default()
             },
             HealRequestSource::Admin,
@@ -3092,43 +2798,6 @@ mod resume_loop_tests {
 
         assert!(env.resume.get_state().await.completed);
         assert_eq!(env.storage.calls(), vec![(POOL_META_NAME.to_string(), None)]);
-        let opts = env.storage.pool_metadata_opts.lock().expect("metadata options");
-        assert_eq!(opts.len(), 1);
-        assert_eq!((opts[0].pool, opts[0].set), (Some(0), Some(0)));
-        assert!(opts[0].recreate);
-        assert!(!opts[0].remove);
-        assert!(!opts[0].no_lock);
-    }
-
-    #[tokio::test]
-    async fn admin_recreate_pool_metadata_validates_owner_scope_before_io() {
-        for (non_owner, unknown_scope, pool) in [(true, false, None), (false, true, None), (false, false, Some(1))] {
-            let env = make_env_with_targets(vec!["replacement-a".to_string()]).await;
-            env.storage.pool_metadata_not_applicable.store(non_owner, Ordering::SeqCst);
-            env.storage.fail_pool_metadata_scope.store(unknown_scope, Ordering::SeqCst);
-            let healer = ErasureSetHealer::new(
-                env.storage.clone(),
-                Arc::new(RwLock::new(HealProgress::new())),
-                CancellationToken::new(),
-                env.healer.disk.clone(),
-                HealOpts {
-                    recreate: true,
-                    pool,
-                    ..Default::default()
-                },
-                HealRequestSource::Admin,
-            )
-            .with_pool_metadata_targets(vec!["replacement-a".to_string()]);
-
-            let set_disk_id = if non_owner { "pool_0_set_1" } else { "pool_0_set_0" };
-            let result = healer
-                .execute_heal_with_resume(&[], set_disk_id, &env.resume, &env.checkpoint)
-                .await;
-
-            assert_eq!(result.is_ok(), non_owner, "only a known non-owner may skip metadata: {result:?}");
-            assert!(env.storage.calls().is_empty(), "scope validation must precede metadata I/O");
-            assert_eq!(env.resume.get_state().await.completed, non_owner);
-        }
     }
 
     #[tokio::test]

crates/heal/src/heal/manager.rs

@@ -841,6 +841,10 @@ pub struct HealManager {
     replacement_recovery_anchors: Arc<std::sync::Mutex<HashMap<String, String>>>,
     /// Set IDs whose durable replacement metadata is corrupt or conflicting.
     replacement_recovery_blocked_sets: Arc<std::sync::Mutex<HashSet<String>>>,
+    /// Durable handoff of interrupted administrator root traversals.
+    root_recovery: Arc<root_recovery::RootHealRecovery>,
+    /// Keep forceStart's cancellation side effects inside the shutdown fence.
+    force_start_shutdown: Mutex<()>,
     /// Storage layer interface
     storage: Arc<dyn HealStorageAPI>,
     /// Cancel token
@@ -876,6 +880,7 @@ struct HealQueueContext<'a> {
     retrying_heals: &'a Arc<Mutex<HashMap<String, RetryingHeal>>>,
     mrf_repair_notice_targets: &'a Arc<StdMutex<HashMap<String, Vec<MrfRepairNoticeTarget>>>>,
     replacement_recovery_anchors: &'a Arc<std::sync::Mutex<HashMap<String, String>>>,
+    root_recovery: &'a Arc<root_recovery::RootHealRecovery>,
     config: &'a Arc<RwLock<HealConfig>>,
     statistics: &'a Arc<RwLock<HealStatistics>>,
     storage: &'a Arc<dyn HealStorageAPI>,
@@ -939,6 +944,10 @@ impl HealManager {
         matches!(request.source, HealRequestSource::Admin | HealRequestSource::Internal)
     }
 
+    fn queued_request_can_be_displaced(request: &HealRequest) -> bool {
+        !root_recovery::is_admin_heal_recovery(&request.heal_type, request.source)
+    }
+
     fn request_bypasses_mainline_throttle(request: &HealRequest) -> bool {
         request.force_start
             || matches!(request.source, HealRequestSource::Admin | HealRequestSource::Internal)
@@ -1075,11 +1084,15 @@ impl HealManager {
         let per_object_request = request.heal_type.is_per_object();
 
         if queue_len >= queue_capacity && !request.force_start {
-            if Self::can_displace_queued_work(&request) && queue.can_displace_lower_priority(request.priority) {
+            if Self::can_displace_queued_work(&request)
+                && queue.can_displace_lower_priority_where(request.priority, Self::queued_request_can_be_displaced)
+            {
                 let request_id = request.id.clone();
                 let priority = request.priority;
                 let source = request.source;
-                if let Some(displaced) = queue.push_displacing_lower_priority(request) {
+                if let Some(displaced) =
+                    queue.push_displacing_lower_priority_where(request, Self::queued_request_can_be_displaced)
+                {
                     publish_heal_queue_length(queue);
                     Self::record_admission_metric(source, HealAdmissionResult::Accepted, context);
                     demote_to_debug_when!(per_object_request, warn, target: "rustfs::heal::manager", {
@@ -1353,6 +1366,51 @@ impl HealManager {
         });
     }
 
+    async fn start_root_recovery_terminal_gc(&self) {
+        let cancel = self.cancel_token.clone();
+        let root_recovery = self.root_recovery.clone();
+        tokio::spawn(async move {
+            let mut ticker = interval(RESUME_GC_INTERVAL);
+            loop {
+                tokio::select! {
+                    _ = cancel.cancelled() => break,
+                    _ = ticker.tick() => {
+                        match root_recovery.gc_terminal_receipts_once(SystemTime::now()).await {
+                            Ok(report) => {
+                                if report.pending_removed > 0 || report.terminals_removed > 0 || report.budget_exhausted {
+                                    debug!(
+                                        target: "rustfs::heal::manager",
+                                        event = EVENT_HEAL_RESUME_GC,
+                                        component = LOG_COMPONENT_HEAL,
+                                        subsystem = LOG_SUBSYSTEM_MANAGER,
+                                        state = "root_terminal_gc",
+                                        scanned = report.scanned,
+                                        retained = report.retained,
+                                        pending_removed = report.pending_removed,
+                                        terminals_removed = report.terminals_removed,
+                                        budget_exhausted = report.budget_exhausted,
+                                        "Root heal terminal receipt GC inspected durable state"
+                                    );
+                                }
+                            }
+                            Err(error) => {
+                                warn!(
+                                    target: "rustfs::heal::manager",
+                                    event = EVENT_HEAL_RESUME_GC,
+                                    component = LOG_COMPONENT_HEAL,
+                                    subsystem = LOG_SUBSYSTEM_MANAGER,
+                                    state = "root_terminal_gc_failed",
+                                    error = %error,
+                                    "Root heal terminal receipt GC failed"
+                                );
+                            }
+                        }
+                    }
+                }
+            }
+        });
+    }
+
     /// Create new HealManager
     pub fn new(storage: Arc<dyn HealStorageAPI>, config: Option<HealConfig>) -> Self {
         Self::new_with_workload_provider(storage, config, None)
@@ -1363,6 +1421,15 @@ impl HealManager {
         storage: Arc<dyn HealStorageAPI>,
         config: Option<HealConfig>,
         workload_provider: Option<WorkloadSnapshotProviderRef>,
+    ) -> Self {
+        Self::new_with_root_recovery(storage, config, workload_provider, Arc::new(root_recovery::RootHealRecovery::default()))
+    }
+
+    fn new_with_root_recovery(
+        storage: Arc<dyn HealStorageAPI>,
+        config: Option<HealConfig>,
+        workload_provider: Option<WorkloadSnapshotProviderRef>,
+        root_recovery: Arc<root_recovery::RootHealRecovery>,
     ) -> Self {
         let config = config.unwrap_or_default();
         Self {
@@ -1377,6 +1444,8 @@ impl HealManager {
             mrf_repair_notice_targets: Arc::new(StdMutex::new(HashMap::new())),
             replacement_recovery_anchors: Arc::new(std::sync::Mutex::new(HashMap::new())),
             replacement_recovery_blocked_sets: Arc::new(std::sync::Mutex::new(HashSet::new())),
+            root_recovery,
+            force_start_shutdown: Mutex::new(()),
             storage,
             cancel_token: CancellationToken::new(),
             statistics: Arc::new(RwLock::new(HealStatistics::new())),
@@ -1386,6 +1455,27 @@ impl HealManager {
         }
     }
 
+    #[cfg(any(test, feature = "test-util"))]
+    #[doc(hidden)]
+    pub fn new_without_root_recovery_for_test(storage: Arc<dyn HealStorageAPI>, config: Option<HealConfig>) -> Self {
+        Self::new_with_root_recovery(storage, config, None, Arc::new(root_recovery::RootHealRecovery::disabled_for_tests()))
+    }
+
+    #[cfg(any(test, feature = "test-util"))]
+    #[doc(hidden)]
+    pub fn new_with_workload_provider_without_root_recovery_for_test(
+        storage: Arc<dyn HealStorageAPI>,
+        config: Option<HealConfig>,
+        workload_provider: Option<WorkloadSnapshotProviderRef>,
+    ) -> Self {
+        Self::new_with_root_recovery(
+            storage,
+            config,
+            workload_provider,
+            Arc::new(root_recovery::RootHealRecovery::disabled_for_tests()),
+        )
+    }
+
     /// Start HealManager
     pub async fn start(&self) -> Result<()> {
         let mut state = self.state.write().await;
@@ -1412,6 +1502,23 @@ impl HealManager {
             "Heal manager starting"
         );
 
+        // Restore graceful-shutdown root responsibilities before automatic
+        // repair can admit overlapping work.
+        if let Err(error) = self.replay_root_heals().await {
+            // A missing owner or invalid root record must not block existing
+            // replacement recovery. Keep its file for a later restart after
+            // the owner is readable or the record has been repaired.
+            warn!(
+                target: "rustfs::heal::manager",
+                event = EVENT_HEAL_MANAGER_STATE,
+                component = LOG_COMPONENT_HEAL,
+                subsystem = LOG_SUBSYSTEM_MANAGER,
+                state = "root_recovery_deferred",
+                error = %error,
+                "Root heal restart recovery deferred"
+            );
+        }
+
         // start scheduler
         self.start_scheduler().await?;
 
@@ -1421,6 +1528,7 @@ impl HealManager {
 
         // Inspect resume artifacts in a bounded, fail-closed background task.
         self.start_resume_gc().await;
+        self.start_root_recovery_terminal_gc().await;
 
         // start auto disk scanner to heal unformatted disks
         if self.config.read().await.enable_auto_heal {
@@ -1449,6 +1557,7 @@ impl HealManager {
 
     /// Stop HealManager
     pub async fn stop(&self) -> Result<()> {
+        let _force_start_guard = self.force_start_shutdown.lock().await;
         info!(
             target: "rustfs::heal::manager",
             event = EVENT_HEAL_MANAGER_STATE,
@@ -1458,11 +1567,39 @@ impl HealManager {
             "Heal manager stopping"
         );
 
-        // cancel all tasks
+        // Keep scheduler, cancellation, and retry ownership stable until every
-        self.cancel_token.cancel();
+        // unfinished admin control-plane heal has a durable successor. A failed
-
+        // write must leave the manager running and the shutdown marker unclean.
-        // wait for all tasks to complete
         let mut active_heals = self.active_heals.lock().await;
+        let queue = self.heal_queue.lock().await;
+        let retrying = self.retrying_heals.lock().await;
+        for task in active_heals.values() {
+            if root_recovery::is_admin_heal_recovery(&task.heal_type, task.source) {
+                if task.get_status().await == HealTaskStatus::Completed {
+                    self.root_recovery.remove(&task.id, &task.heal_type, task.source).await?;
+                } else {
+                    let mut request = match task.retry_request_with_remaining_timeout().await {
+                        Ok(request) => request,
+                        Err(Error::TaskTimeout) => {
+                            let mut request = task.retry_request();
+                            request.options.timeout = Some(Duration::ZERO);
+                            request
+                        }
+                        Err(error) => return Err(error),
+                    };
+                    request.retry_attempts = task.retry_attempts;
+                    self.root_recovery.persist(&request).await?;
+                }
+            }
+        }
+        for request in queue.requests().chain(retrying.values().map(|retrying| &retrying.request)) {
+            self.root_recovery.persist(request).await?;
+        }
+        self.cancel_token.cancel();
+        drop(retrying);
+        drop(queue);
+
+        // cancel active workers after the durable handoff
         for task in active_heals.values() {
             if let Err(e) = task.cancel().await {
                 warn!(
@@ -1589,6 +1726,17 @@ impl HealManager {
         let admission_start = Instant::now();
         let source = request.source;
         let force_start = request.force_start;
+        // A forceStart must not retire an old durable owner if shutdown will
+        // reject its replacement. Hold the same gate through final admission.
+        let _force_start_guard = if source == HealRequestSource::Admin && force_start {
+            let guard = self.force_start_shutdown.lock().await;
+            if self.cancel_token.is_cancelled() {
+                return Err(Error::Other("Heal manager is stopping".to_string()));
+            }
+            Some(guard)
+        } else {
+            None
+        };
         // HS-06 forceStart semantics (admin only): MinIO stops the old task
         // first and then starts the new one. Cancel any active admin task
         // overlapping this request's path before entering admission, so the
@@ -1596,7 +1744,9 @@ impl HealManager {
         if request.source == HealRequestSource::Admin && request.force_start {
             let overlapping: Vec<String> = {
                 let active_heals = self.active_heals.lock().await;
-                active_heals
+                let queue = self.heal_queue.lock().await;
+                let retrying = self.retrying_heals.lock().await;
+                let mut ids = active_heals
                     .iter()
                     .filter(|(task_id, task)| {
                         task.source == HealRequestSource::Admin
@@ -1604,7 +1754,19 @@ impl HealManager {
                             && *task_id != &request.id
                     })
                     .map(|(task_id, _)| task_id.clone())
-                    .collect()
+                    .collect::<Vec<_>>();
+                ids.extend(
+                    queue
+                        .requests()
+                        .chain(retrying.values().map(|retrying| &retrying.request))
+                        .filter(|pending| {
+                            root_recovery::is_admin_heal_recovery(&pending.heal_type, pending.source)
+                                && heal_types_overlap(&request.heal_type, &pending.heal_type) != OverlapVerdict::Disjoint
+                                && pending.id != request.id
+                        })
+                        .map(|pending| pending.id.clone()),
+                );
+                ids
             };
             for task_id in overlapping {
                 match self.cancel_task(&task_id).await {
@@ -1618,17 +1780,16 @@ impl HealManager {
                         result = "force_start_cancelled_overlap",
                         "Admin forceStart cancelled an overlapping heal task"
                     ),
-                    Err(err) => warn!(
+                    Err(err) => return Err(err),
-                        target: "rustfs::heal::manager",
+                }
-                        event = EVENT_HEAL_QUEUE_ADMISSION,
+            }
-                        component = LOG_COMPONENT_HEAL,
+            // A failed or timed-out replay may have only its durable owner
-                        subsystem = LOG_SUBSYSTEM_MANAGER,
+            // left. Cancel only records that overlap this forced start.
-                        request_id = %request.id,
+            for pending in self.root_recovery.pending().await? {
-                        cancelled_task_id = %task_id,
+                if pending.id != request.id
-                        error = %err,
+                    && heal_types_overlap(&request.heal_type, &pending.heal_type) != OverlapVerdict::Disjoint
-                        result = "force_start_cancel_failed",
+                {
-                        "Admin forceStart failed to cancel an overlapping heal task"
+                    self.cancel_task(&pending.id).await?;
-                    ),
                 }
             }
         }
@@ -1641,6 +1802,9 @@ impl HealManager {
         // active -> retrying transitions can slip between duplicate checks.
         let lock_phase_start = Instant::now();
         let active_heals = self.active_heals.lock().await;
+        if self.cancel_token.is_cancelled() {
+            return Err(Error::Other("Heal manager is stopping".to_string()));
+        }
         #[cfg(test)]
         pause_duplicate_admission_after_active_lock(&request.id).await;
         let mut queue = self.heal_queue.lock().await;
@@ -1854,9 +2018,29 @@ impl HealManager {
             }
         }
 
+        let durable_handoff = root_recovery::is_admin_heal_recovery(&request.heal_type, request.source);
+        let durable_handoff_required = durable_handoff
+            && (request.force_start
+                || queue.len() < config.queue_size
+                || (Self::can_displace_queued_work(&request)
+                    && queue.can_displace_lower_priority_where(request.priority, Self::queued_request_can_be_displaced)));
+        // An admin receipt is a control-plane responsibility. Persist it before
+        // queue publication so a crash after admission can replay it.
+        if durable_handoff_required {
+            self.root_recovery.persist(&request).await?;
+        }
+
         let mut task_id = request.id.clone();
+        let request_id = request.id.clone();
+        let request_heal_type = request.heal_type.clone();
+        let request_source = request.source;
         let admission_decision = Self::admit_request_to_queue(&mut queue, request, &config, "submit");
         let admission = admission_decision.result;
+        if durable_handoff_required && !admission.is_admitted() {
+            self.root_recovery
+                .remove(&request_id, &request_heal_type, request_source)
+                .await?;
+        }
         if admission == HealAdmissionResult::Merged
             && let Some(queued_id) = queue.queued_request_id_for_dedup_key(&dedup_key)
         {
@@ -1927,7 +2111,7 @@ impl HealManager {
     /// Retrying, never Pending — then the queue, and finally a terminal
     /// completed entry. `heal_path` additionally constrains the map matches
     /// the way the `*_for_path` variants always have.
-    async fn lookup_task_state(&self, canonical_task_id: &str, heal_path: Option<&str>) -> TaskStateLookup {
+    async fn lookup_task_state(&self, canonical_task_id: &str, heal_path: Option<&str>) -> Result<TaskStateLookup> {
         let matches_path = |heal_type: &HealType| heal_path.is_none_or(|path| heal_type_matches_path(heal_type, path));
 
         {
@@ -1936,7 +2120,7 @@ impl HealManager {
                 .get(canonical_task_id)
                 .filter(|task| matches_path(&task.heal_type))
             {
-                return TaskStateLookup::Active(Arc::clone(task));
+                return Ok(TaskStateLookup::Active(Arc::clone(task)));
             }
         }
 
@@ -1946,7 +2130,7 @@ impl HealManager {
                 .get(canonical_task_id)
                 .filter(|retrying| matches_path(&retrying.request.heal_type))
             {
-                return TaskStateLookup::Retrying(retrying.status());
+                return Ok(TaskStateLookup::Retrying(retrying.status()));
             }
         }
 
@@ -1959,7 +2143,7 @@ impl HealManager {
             prune_completed_heal_statuses(&mut completed_heals);
             if let Some(completed) = completed_heals.get(canonical_task_id).filter(|c| matches_path(&c.heal_type)) {
                 if completed_status_is_retrying(&completed.status) {
-                    return TaskStateLookup::Completed(Arc::clone(completed));
+                    return Ok(TaskStateLookup::Completed(Arc::clone(completed)));
                 }
                 terminal_completed = Some(Arc::clone(completed));
             }
@@ -1972,7 +2156,7 @@ impl HealManager {
                 None => queue.contains_request_id(canonical_task_id),
             };
             if queued {
-                return TaskStateLookup::Queued;
+                return Ok(TaskStateLookup::Queued);
             }
         }
 
@@ -1985,15 +2169,55 @@ impl HealManager {
                 .cloned();
         }
 
-        match terminal_completed {
+        if terminal_completed.is_none()
+            && let Some(completed) = self.root_recovery.completed(canonical_task_id).await?
+            && matches_path(&completed.heal_type)
+        {
+            terminal_completed = Some(Arc::new(completed));
+        }
+
+        Ok(match terminal_completed {
             Some(completed) => TaskStateLookup::Completed(completed),
             None => TaskStateLookup::NotFound,
-        }
+        })
+    }
+
+    async fn publish_admin_terminal(
+        &self,
+        task_id: &str,
+        heal_type: &HealType,
+        source: HealRequestSource,
+        completed: &CompletedHealStatus,
+    ) -> Result<bool> {
+        self.root_recovery
+            .persist_terminal(task_id, heal_type, source, completed)
+            .await
+    }
+
+    async fn publish_admin_cancelled_terminal(
+        &self,
+        task_id: &str,
+        heal_type: &HealType,
+        source: HealRequestSource,
+    ) -> Result<bool> {
+        let completed = CompletedHealStatus {
+            outcome: None,
+            progress: None,
+            retained_bytes: std::sync::OnceLock::new(),
+            heal_type: heal_type.clone(),
+            status: HealTaskStatus::Cancelled,
+            result_items_truncated: false,
+            completed_at: SystemTime::now(),
+            seqed_items: Vec::new(),
+            next_seq: 0,
+            min_seq: 0,
+        };
+        self.publish_admin_terminal(task_id, heal_type, source, &completed).await
     }
 
     pub async fn get_task_status(&self, task_id: &str) -> Result<HealTaskStatus> {
         let canonical_task_id = self.canonical_task_id(task_id).await;
-        match self.lookup_task_state(&canonical_task_id, None).await {
+        match self.lookup_task_state(&canonical_task_id, None).await? {
             TaskStateLookup::Active(task) => Ok(task.get_status().await),
             TaskStateLookup::Retrying(status) => Ok(status),
             TaskStateLookup::Completed(completed) => Ok(completed.status.clone()),
@@ -2013,7 +2237,7 @@ impl HealManager {
     /// full-snapshot semantics.
     pub async fn get_task_report_since(&self, task_id: &str, since: Option<u64>) -> Result<HealTaskReport> {
         let canonical_task_id = self.canonical_task_id(task_id).await;
-        match self.lookup_task_state(&canonical_task_id, None).await {
+        match self.lookup_task_state(&canonical_task_id, None).await? {
             TaskStateLookup::Active(task) => Ok(active_task_report(&task, since).await),
             TaskStateLookup::Retrying(status) => Ok(empty_task_report(status)),
             TaskStateLookup::Completed(completed) => Ok(completed_task_report(&completed, since)),
@@ -2036,7 +2260,7 @@ impl HealManager {
         since: Option<u64>,
     ) -> Result<HealTaskReport> {
         let canonical_task_id = self.canonical_task_id(task_id).await;
-        match self.lookup_task_state(&canonical_task_id, Some(heal_path)).await {
+        match self.lookup_task_state(&canonical_task_id, Some(heal_path)).await? {
             TaskStateLookup::Active(task) => Ok(active_task_report(&task, since).await),
             TaskStateLookup::Retrying(status) => Ok(empty_task_report(status)),
             TaskStateLookup::Completed(completed) => Ok(completed_task_report(&completed, since)),
@@ -2059,7 +2283,7 @@ impl HealManager {
     /// recently completed task, a different token is invalid for that path.
     pub async fn get_task_status_for_path(&self, heal_path: &str, task_id: &str) -> Result<HealTaskStatus> {
         let canonical_task_id = self.canonical_task_id(task_id).await;
-        match self.lookup_task_state(&canonical_task_id, Some(heal_path)).await {
+        match self.lookup_task_state(&canonical_task_id, Some(heal_path)).await? {
             TaskStateLookup::Active(task) => Ok(task.get_status().await),
             TaskStateLookup::Retrying(status) => Ok(status),
             TaskStateLookup::Completed(completed) => Ok(completed.status.clone()),
@@ -2113,11 +2337,18 @@ impl HealManager {
         }
         drop(completed_heals);
 
-        let mut displaced_terminals = lock_displaced_terminals(&self.displaced_terminals);
+        {
-        prune_completed_heal_statuses(&mut displaced_terminals);
+            let mut displaced_terminals = lock_displaced_terminals(&self.displaced_terminals);
-        displaced_terminals
+            prune_completed_heal_statuses(&mut displaced_terminals);
-            .values()
+            if displaced_terminals
-            .any(|terminal| heal_type_matches_path(&terminal.heal_type, heal_path))
+                .values()
+                .any(|terminal| heal_type_matches_path(&terminal.heal_type, heal_path))
+            {
+                return true;
+            }
+        }
+
+        self.root_recovery.completed_matches_path(heal_path).await.unwrap_or(false)
     }
 
     /// Get task progress
@@ -2129,7 +2360,7 @@ impl HealManager {
 
     pub async fn get_task_progress(&self, task_id: &str) -> Result<HealProgress> {
         let canonical_task_id = self.canonical_task_id(task_id).await;
-        let progress = match self.lookup_task_state(&canonical_task_id, None).await {
+        let progress = match self.lookup_task_state(&canonical_task_id, None).await? {
             TaskStateLookup::Active(task) => Some(task.get_progress().await),
             TaskStateLookup::Completed(completed) => completed.progress.clone(),
             _ => None,
@@ -2145,8 +2376,11 @@ impl HealManager {
         {
             let mut active_heals = self.active_heals.lock().await;
             if let Some(task) = active_heals.get(&canonical_task_id) {
-                task.cancel().await?;
                 let completed = CompletedHealStatus::snapshot(task, HealTaskStatus::Cancelled).await;
+                self.publish_admin_terminal(&canonical_task_id, &task.heal_type, task.source, &completed)
+                    .await?;
+                self.root_recovery.remove(&task.id, &task.heal_type, task.source).await?;
+                task.cancel().await?;
                 publish_completed_heal(&self.completed_heals, &self.task_aliases, &canonical_task_id, completed, true).await;
                 active_heals.remove(&canonical_task_id);
                 publish_active_heal_count(&active_heals);
@@ -2168,6 +2402,13 @@ impl HealManager {
 
         {
             let mut retrying_heals = self.retrying_heals.lock().await;
+            if let Some(retrying) = retrying_heals.get(&canonical_task_id) {
+                self.publish_admin_cancelled_terminal(&canonical_task_id, &retrying.request.heal_type, retrying.request.source)
+                    .await?;
+                self.root_recovery
+                    .remove(&canonical_task_id, &retrying.request.heal_type, retrying.request.source)
+                    .await?;
+            }
             if let Some(retrying) = retrying_heals.remove(&canonical_task_id) {
                 retrying.cancel_token.cancel();
                 drop(retrying_heals);
@@ -2188,6 +2429,13 @@ impl HealManager {
         }
 
         let mut queue = self.heal_queue.lock().await;
+        if let Some(request) = queue.requests().find(|request| request.id == canonical_task_id) {
+            self.publish_admin_cancelled_terminal(&canonical_task_id, &request.heal_type, request.source)
+                .await?;
+            self.root_recovery
+                .remove(&request.id, &request.heal_type, request.source)
+                .await?;
+        }
         if queue.remove_request_id(&canonical_task_id).is_some() {
             publish_heal_queue_length(&queue);
             info!(
@@ -2205,6 +2453,10 @@ impl HealManager {
             return Ok(());
         }
 
+        drop(queue);
+        if self.root_recovery.cancel_pending(&canonical_task_id).await? {
+            return Ok(());
+        }
         Err(Error::TaskNotFound {
             task_id: task_id.to_string(),
         })
@@ -2223,8 +2475,11 @@ impl HealManager {
 
             for task_id in &task_ids {
                 if let Some(task) = active_heals.get(task_id) {
-                    task.cancel().await?;
                     let completed = CompletedHealStatus::snapshot(task, HealTaskStatus::Cancelled).await;
+                    self.publish_admin_terminal(task_id, &task.heal_type, task.source, &completed)
+                        .await?;
+                    self.root_recovery.remove(&task.id, &task.heal_type, task.source).await?;
+                    task.cancel().await?;
                     publish_completed_heal(&self.completed_heals, &self.task_aliases, task_id, completed, true).await;
                 }
                 active_heals.remove(task_id);
@@ -2251,6 +2506,13 @@ impl HealManager {
                 .collect::<Vec<_>>();
 
             for task_id in &task_ids {
+                if let Some(retrying) = retrying_heals.get(task_id) {
+                    self.publish_admin_cancelled_terminal(task_id, &retrying.request.heal_type, retrying.request.source)
+                        .await?;
+                    self.root_recovery
+                        .remove(task_id, &retrying.request.heal_type, retrying.request.source)
+                        .await?;
+                }
                 if let Some(retrying) = retrying_heals.remove(task_id) {
                     retrying.cancel_token.cancel();
                     cancelled += 1;
@@ -2272,6 +2534,21 @@ impl HealManager {
             }
         }
 
+        let queued_matches = {
+            let queue = self.heal_queue.lock().await;
+            queue
+                .requests()
+                .filter(|request| heal_type_matches_path(&request.heal_type, heal_path))
+                .cloned()
+                .collect::<Vec<_>>()
+        };
+        for request in &queued_matches {
+            self.publish_admin_cancelled_terminal(&request.id, &request.heal_type, request.source)
+                .await?;
+            self.root_recovery
+                .remove(&request.id, &request.heal_type, request.source)
+                .await?;
+        }
         let mut queue = self.heal_queue.lock().await;
         let queued_cancelled = queue.remove_matching(|request| heal_type_matches_path(&request.heal_type, heal_path));
         if !queued_cancelled.is_empty() {
@@ -2284,6 +2561,17 @@ impl HealManager {
             self.remove_mrf_repair_notice_targets_for_task(&request.id);
         }
 
+        for pending in self
+            .root_recovery
+            .pending()
+            .await?
+            .into_iter()
+            .filter(|pending| heal_type_matches_path(&pending.heal_type, heal_path))
+        {
+            if self.root_recovery.cancel_pending(&pending.id).await? {
+                cancelled += 1;
+            }
+        }
         if cancelled == 0 {
             return Err(Error::TaskNotFound {
                 task_id: heal_path.to_string(),
@@ -2395,6 +2683,7 @@ impl std::fmt::Debug for HealManager {
 
 mod auto_scan;
 mod queue;
+mod root_recovery;
 mod scheduler;
 mod unclean_shutdown;
 

crates/heal/src/heal/manager/queue.rs

@@ -288,16 +288,29 @@ impl PriorityHealQueue {
         QueuePushOutcome::Accepted
     }
 
-    pub(super) fn can_displace_lower_priority(&self, priority: HealPriority) -> bool {
+    pub(super) fn can_displace_lower_priority_where<F>(&self, priority: HealPriority, can_displace: F) -> bool
-        self.heap.iter().any(|item| item.priority < priority)
+    where
+        F: Fn(&HealRequest) -> bool,
+    {
+        self.heap
+            .iter()
+            .any(|item| item.priority < priority && can_displace(&item.request))
     }
 
-    pub(super) fn push_displacing_lower_priority(&mut self, request: HealRequest) -> Option<HealRequest> {
+    #[cfg(test)]
+    pub(super) fn can_displace_lower_priority(&self, priority: HealPriority) -> bool {
+        self.can_displace_lower_priority_where(priority, |_| true)
+    }
+
+    pub(super) fn push_displacing_lower_priority_where<F>(&mut self, request: HealRequest, can_displace: F) -> Option<HealRequest>
+    where
+        F: Fn(&HealRequest) -> bool,
+    {
         let mut retained = BinaryHeap::new();
         let mut displaced: Option<PriorityQueueItem> = None;
 
         while let Some(item) = self.heap.pop() {
-            if item.priority < request.priority {
+            if item.priority < request.priority && can_displace(&item.request) {
                 let should_displace = displaced
                     .as_ref()
                     .map(|current| {
@@ -337,6 +350,11 @@ impl PriorityHealQueue {
         displaced
     }
 
+    #[cfg(test)]
+    pub(super) fn push_displacing_lower_priority(&mut self, request: HealRequest) -> Option<HealRequest> {
+        self.push_displacing_lower_priority_where(request, |_| true)
+    }
+
     /// Get statistics about queue contents by priority
     pub(super) fn get_priority_stats(&self) -> HashMap<HealPriority, usize> {
         let mut stats = HashMap::new();