chore(ecstore): drop the disk dead_code blanket

Removing the blanket exposes 36 items in the lowest storage layer: 7 deleted, 29 kept with reasoned item-level allows. That is the smallest deletion share of this burn-down, and the reason is a verification limit rather than a judgement call. disk/local.rs carries 141 `#[cfg(target_os = "linux")]` sites — the densest platform gating in the tree, because O_DIRECT and io_uring only exist there. The direct-I/O cluster (six ENV_RUSTFS_OBJECT_DIRECT_IO_* constants plus is_direct_io_read_enabled, is_direct_io_write_enabled, get_direct_io_read_threshold, direct_write_staging_capacity, direct_write_tail_split and DIRECT_WRITE_STAGING_BYTES) reads as dead on macOS purely because its production callers at local.rs:1766, 3114 and 4605 sit inside Linux-gated blocks. direct_write_staging_capacity even documents itself as "Platform-independent (no O_DIRECT), so it is unit-tested on any host". Deleting those would leave every local check green — 4096 tests pass, clippy is clean, make pre-commit exits 0 — and break the Linux build in CI, because all four local lanes compile for aarch64-apple-darwin. Cross-checking locally is not available either: cargo check --target x86_64-unknown-linux-gnu fails in the aws-lc-sys build script for want of a Linux C cross-compiler. Their allows name the platform reason so the next reader on a non-Linux host does not repeat the investigation. Deleted, all in files with no target_os gating at all (os.rs, disk_store.rs): - HealthDiskCtxKey and HealthDiskCtxValue with its private log_success. Note that DiskHealthTracker::log_success is a different method of the same name and is live from cluster/rpc/peer_s3_client.rs and remote_disk.rs — the two have to be told apart by type, not by name. - LocalDiskWrapper::new_with_health and check_id. - os.rs file_exists and lock_destination_directory_for_path_access. Kept with allows: DiskHealthTracker's set_faulty, mark_offline, waiting_count and last_success have test callers in remote_disk.rs, so they only look dead in the lib target. to_disk_error, remove_all and sync_dir_files are asserted by their own files' tests. The reclaim, mmap and path-cache field groups are written but never read back. Placement follows the same rule as the earlier roots: per-method allows inside impl DiskHealthTracker and impl LocalDisk, since both are mostly live and a block-level allow would be a smaller version of the blanket this issue removes. Struct-level allows are used only where the warning covers that struct's own fields. The three cached_read_env! functions take their allow inside the macro invocation, before the fn line, because the macro forwards $(#[$meta:meta])* onto the generated item. Verification, four lanes warning-free: default, --tests, --features rio-v2 --tests, --features test-util --tests. cargo nextest run -p rustfs-ecstore 4096 passed; clippy --lib --tests -D warnings clean; make pre-commit exit 0. The Linux lane is not covered locally and is left to CI. Ref rustfs/backlog#1823 (step 2).
fix(ecstore): mint bucket-target ARNs in the madmin arn:minio partition (#6128 )
2026-08-16 09:58:21 +00:00 · 2026-08-16 11:53:10 +08:00 · 2026-08-16 10:28:45 +08:00 · 2026-08-16 10:28:11 +08:00 · 2026-08-16 10:27:59 +08:00 · 2026-08-16 09:50:16 +08:00
140 changed files with 21664 additions and 4652 deletions
@@ -252,10 +252,16 @@ test-group = 'ecstore-serial-flaky'
 # cluster, so it keeps the lane's parallel-safe / no-external-dependency
 # properties. The RustFS warm backend has no loopback guard (that guard is
 # replication-only), so it needs no opt-in env for its 127.0.0.1 tier target.
+#
+# Disk compression (backlog#1848): the `compression` module joins the smoke
+# lane so the multipart disk-compression roundtrips (restored after
+# rustfs/rustfs#5169 disabled them) have PR-lane signal, not just merge-gate.
+# Single-node servers on random ports with isolated temp dirs — meets the
+# admission criteria unchanged.
 [profile.e2e-smoke]
 default-filter = """
    package(e2e_test) & (
-        test(/^(delete_marker_migration_semantics|version_id_regression|list_objects_v2_pagination|list_object_versions_regression|list_objects_duplicates|list_buckets_double_slash|list_buckets_auth|list_buckets_iam_filter|leading_slash_key|special_chars|create_bucket_region|delete_objects_versioning|head_object_consistency|head_object_range|copy_object_metadata|copy_object_tagging|copy_source_invalid_date|content_encoding|multipart_storage_class|storage_class_capability|ssec_copy|anonymous_access|bucket_policy_check|presigned_negative|negative_sigv4|admin_auth|notification_webhook|tls_hot_reload|console_smoke|admin_iam_crud|admin_pools|sts_query_compat)_test::|^fake_s3_target::/)
+        test(/^(delete_marker_migration_semantics|version_id_regression|list_objects_v2_pagination|list_object_versions_regression|list_objects_duplicates|list_buckets_double_slash|list_buckets_auth|list_buckets_iam_filter|leading_slash_key|special_chars|create_bucket_region|delete_objects_versioning|head_object_consistency|head_object_range|copy_object_metadata|copy_object_tagging|copy_source_invalid_date|content_encoding|compression|multipart_storage_class|storage_class_capability|ssec_copy|anonymous_access|bucket_policy_check|presigned_negative|negative_sigv4|admin_auth|notification_webhook|tls_hot_reload|console_smoke|admin_iam_crud|admin_pools|sts_query_compat)_test::|^fake_s3_target::/)
        | test(/^replication_extension_test::(test_replication_check_succeeds_with_remote_target|test_replication_check_rejects_target_without_object_lock|test_set_remote_target_rejects_unversioned_source_bucket|test_replication_check_rejects_unversioned_source_bucket|test_replication_check_rejects_missing_replication_config|test_replication_check_rejects_invalid_bucket|test_set_remote_target_rejects_same_bucket_on_same_deployment|test_set_remote_target_rejects_unversioned_target_bucket|test_set_remote_target_update_requires_arn|test_set_remote_target_update_rejects_missing_target|test_set_remote_target_rejects_invalid_target_url|test_set_remote_target_rejects_self_signed_https_target_without_skip_tls_verify|test_set_remote_target_rejects_private_ca_https_target_without_ca_cert_pem|test_list_remote_targets_rejects_empty_bucket|test_list_remote_targets_rejects_invalid_bucket|test_remove_remote_target_rejects_missing_target|test_remove_remote_target_rejects_missing_arn|test_remove_remote_target_rejects_invalid_bucket|test_remove_remote_target_rejects_target_used_by_replication|test_delete_bucket_replication_removes_remote_target)$/)
        | test(/^reliant::lifecycle::/)
        | test(/^reliant::tiering::/)
@@ -94,6 +94,7 @@ jobs:
      short_sha: ${{ steps.check.outputs.short_sha }}
      is_prerelease: ${{ steps.check.outputs.is_prerelease }}
      create_latest: ${{ steps.check.outputs.create_latest }}
+      source_ref: ${{ steps.check.outputs.source_ref }}
    steps:
      - name: Checkout repository
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
@@ -118,6 +119,7 @@ jobs:
          short_sha=""
          is_prerelease=false
          create_latest=false
+          source_ref="$GITHUB_SHA"

          if [[ "${{ github.event_name }}" == "workflow_run" ]]; then
            # Triggered by build workflow completion
@@ -137,6 +139,7 @@ jobs:
            # Extract version info from commit message or use commit SHA
            # Use Git to generate consistent short SHA (ensures uniqueness like build.yml)
            short_sha=$(git rev-parse --short "$HEAD_SHA")
+            source_ref="$HEAD_SHA"

            # Determine build type based on triggering workflow event and ref
            triggering_event="$TRIGGERING_EVENT"
@@ -261,6 +264,23 @@ jobs:
                echo "⚠️  Only release versions (latest, v1.0.0, 1.0.0) and prereleases (v1.0.0-alpha1, 1.0.0-beta2) are supported"
                ;;
            esac
+
+            if [[ "$should_build" == true && "$input_version" != "latest" ]]; then
+              tag_ref="refs/tags/$input_version"
+              if ! git ls-remote --exit-code origin "$tag_ref" >/dev/null 2>&1; then
+                if [[ "$input_version" == v* ]]; then
+                  tag_ref="refs/tags/${input_version#v}"
+                else
+                  tag_ref="refs/tags/v$input_version"
+                fi
+              fi
+
+              if ! git ls-remote --exit-code origin "$tag_ref" >/dev/null 2>&1; then
+                echo "❌ Release tag not found for Docker build: $input_version"
+                exit 1
+              fi
+              source_ref="$tag_ref"
+            fi
          fi

          {
@@ -271,6 +291,7 @@ jobs:
            echo "short_sha=$short_sha"
            echo "is_prerelease=$is_prerelease"
            echo "create_latest=$create_latest"
+            echo "source_ref=$source_ref"
          } >> "$GITHUB_OUTPUT"

          echo "🐳 Docker Build Summary:"
@@ -281,6 +302,7 @@ jobs:
          echo "  - Short SHA: $short_sha"
          echo "  - Is prerelease: $is_prerelease"
          echo "  - Create latest: $create_latest"
+          echo "  - Source ref: $source_ref"

  # Build multi-arch Docker images
  # Strategy: Build images using pre-built binaries from dl.rustfs.com
@@ -308,6 +330,7 @@ jobs:
        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
        with:
          persist-credentials: false
+          ref: ${{ needs.build-check.outputs.source_ref }}

      - name: Login to Docker Hub
        uses: docker/login-action@c94ce9fb468520275223c153574b00df6fe4bcc9 # v3
@@ -397,7 +420,8 @@ jobs:
          LABELS="org.opencontainers.image.title=RustFS"
          LABELS="$LABELS,org.opencontainers.image.description=RustFS distributed object storage system"
          LABELS="$LABELS,org.opencontainers.image.version=$VERSION"
-          LABELS="$LABELS,org.opencontainers.image.revision=${{ github.sha }}"
+          SOURCE_REVISION="$(git rev-parse HEAD)"
+          LABELS="$LABELS,org.opencontainers.image.revision=$SOURCE_REVISION"
          LABELS="$LABELS,org.opencontainers.image.source=${{ github.server_url }}/${{ github.repository }}"
          LABELS="$LABELS,org.opencontainers.image.created=$(date -u +'%Y-%m-%dT%H:%M:%SZ')"
          LABELS="$LABELS,org.opencontainers.image.build-type=$BUILD_TYPE"
@@ -278,6 +278,7 @@ checksum = "312c1ea69e5fe9966e0029fb95aca8790100b85aff4f0d3b00a9337c74069a9c"
 dependencies = [
 "bigdecimal",
 "bon",
+ "crc32fast",
 "digest 0.11.3",
 "log",
 "miniz_oxide 0.9.1",
@@ -289,9 +290,11 @@ dependencies = [
 "serde",
 "serde_bytes",
 "serde_json",
+ "snap",
 "strum",
 "thiserror 2.0.20",
 "uuid",
+ "zstd",
 ]

 [[package]]
@@ -3761,7 +3764,7 @@ checksum = "d0881ea181b1df73ff77ffaaf9c7544ecc11e82fba9b5f27b262a3c73a332555"

 [[package]]
 name = "e2e_test"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "anyhow",
 "astral-tokio-tar",
@@ -9090,7 +9093,7 @@ dependencies = [

 [[package]]
 name = "rustfs"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "aes-gcm",
 "anyhow",
@@ -9200,6 +9203,7 @@ dependencies = [
 "serial_test",
 "sha2 0.11.0",
 "shadow-rs",
+ "snap",
 "socket2",
 "subtle",
 "sysinfo",
@@ -9227,7 +9231,7 @@ dependencies = [

 [[package]]
 name = "rustfs-audit"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-trait",
 "const-str",
@@ -9250,7 +9254,7 @@ dependencies = [

 [[package]]
 name = "rustfs-checksums"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "base64-simd",
 "bytes",
@@ -9266,7 +9270,7 @@ dependencies = [

 [[package]]
 name = "rustfs-common"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "chrono",
 "hotpath",
@@ -9284,7 +9288,7 @@ dependencies = [

 [[package]]
 name = "rustfs-concurrency"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "hotpath",
 "insta",
@@ -9297,7 +9301,7 @@ dependencies = [

 [[package]]
 name = "rustfs-config"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "const-str",
 "hotpath",
@@ -9307,7 +9311,7 @@ dependencies = [

 [[package]]
 name = "rustfs-credentials"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "base64-simd",
 "hmac 0.13.0",
@@ -9321,7 +9325,7 @@ dependencies = [

 [[package]]
 name = "rustfs-crypto"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "aes-gcm",
 "argon2",
@@ -9342,7 +9346,7 @@ dependencies = [

 [[package]]
 name = "rustfs-data-usage"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "hotpath",
 "rmp-serde",
@@ -9352,7 +9356,7 @@ dependencies = [

 [[package]]
 name = "rustfs-ecstore"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "arc-swap",
 "async-channel",
@@ -9491,7 +9495,7 @@ dependencies = [

 [[package]]
 name = "rustfs-extension-schema"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "hotpath",
 "serde",
@@ -9501,7 +9505,7 @@ dependencies = [

 [[package]]
 name = "rustfs-filemeta"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "arc-swap",
 "byteorder",
@@ -9528,7 +9532,7 @@ dependencies = [

 [[package]]
 name = "rustfs-heal"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-trait",
 "base64 0.23.1",
@@ -9559,7 +9563,7 @@ dependencies = [

 [[package]]
 name = "rustfs-iam"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "arc-swap",
 "async-trait",
@@ -9600,7 +9604,7 @@ dependencies = [

 [[package]]
 name = "rustfs-io-core"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "bytes",
 "hotpath",
@@ -9613,7 +9617,7 @@ dependencies = [

 [[package]]
 name = "rustfs-io-metrics"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "criterion",
 "hotpath",
@@ -9677,7 +9681,7 @@ dependencies = [

 [[package]]
 name = "rustfs-keystone"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "bytes",
 "futures",
@@ -9704,7 +9708,7 @@ dependencies = [

 [[package]]
 name = "rustfs-kms"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "aes-gcm",
 "anyhow",
@@ -9753,7 +9757,7 @@ dependencies = [

 [[package]]
 name = "rustfs-lifecycle"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-trait",
 "hotpath",
@@ -9776,7 +9780,7 @@ dependencies = [

 [[package]]
 name = "rustfs-lock"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-trait",
 "compact_str",
@@ -9799,7 +9803,7 @@ dependencies = [

 [[package]]
 name = "rustfs-log-analyzer"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "chrono",
 "flate2",
@@ -9818,7 +9822,7 @@ dependencies = [

 [[package]]
 name = "rustfs-madmin"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "hotpath",
 "humantime",
@@ -9833,7 +9837,7 @@ dependencies = [

 [[package]]
 name = "rustfs-notify"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "arc-swap",
 "async-trait",
@@ -9868,7 +9872,7 @@ dependencies = [

 [[package]]
 name = "rustfs-object-capacity"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "criterion",
 "futures",
@@ -9888,7 +9892,7 @@ dependencies = [

 [[package]]
 name = "rustfs-object-data-cache"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "bytes",
 "criterion",
@@ -9905,7 +9909,7 @@ dependencies = [

 [[package]]
 name = "rustfs-obs"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "chrono",
 "crossbeam-channel",
@@ -9960,7 +9964,7 @@ dependencies = [

 [[package]]
 name = "rustfs-policy"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-trait",
 "base64-simd",
@@ -9991,7 +9995,7 @@ dependencies = [

 [[package]]
 name = "rustfs-protocols"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "astral-tokio-tar",
 "async-compression",
@@ -10053,7 +10057,7 @@ dependencies = [

 [[package]]
 name = "rustfs-protos"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "flatbuffers",
 "hotpath",
@@ -10077,7 +10081,7 @@ dependencies = [

 [[package]]
 name = "rustfs-replication"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "byteorder",
 "bytes",
@@ -10095,7 +10099,7 @@ dependencies = [

 [[package]]
 name = "rustfs-rio"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "aes-gcm",
 "arc-swap",
@@ -10133,7 +10137,7 @@ dependencies = [

 [[package]]
 name = "rustfs-rio-v2"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "aes-gcm",
 "bytes",
@@ -10156,7 +10160,7 @@ dependencies = [

 [[package]]
 name = "rustfs-s3-ops"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "hotpath",
 "rustfs-s3-types",
@@ -10164,7 +10168,7 @@ dependencies = [

 [[package]]
 name = "rustfs-s3-types"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "hotpath",
 "serde",
@@ -10173,7 +10177,7 @@ dependencies = [

 [[package]]
 name = "rustfs-s3select-api"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-trait",
 "bytes",
@@ -10203,7 +10207,7 @@ dependencies = [

 [[package]]
 name = "rustfs-s3select-query"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-recursion",
 "async-trait",
@@ -10222,7 +10226,7 @@ dependencies = [

 [[package]]
 name = "rustfs-scanner"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-trait",
 "bytes",
@@ -10262,7 +10266,7 @@ dependencies = [

 [[package]]
 name = "rustfs-security-governance"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "hotpath",
 "thiserror 2.0.20",
@@ -10270,7 +10274,7 @@ dependencies = [

 [[package]]
 name = "rustfs-signer"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "base64-simd",
 "bytes",
@@ -10288,7 +10292,7 @@ dependencies = [

 [[package]]
 name = "rustfs-storage-api"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-trait",
 "hotpath",
@@ -10303,7 +10307,7 @@ dependencies = [

 [[package]]
 name = "rustfs-targets"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "arc-swap",
 "async-nats",
@@ -10357,7 +10361,7 @@ dependencies = [

 [[package]]
 name = "rustfs-test-utils"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "hotpath",
 "rustfs-data-usage",
@@ -10373,7 +10377,7 @@ dependencies = [

 [[package]]
 name = "rustfs-tls-runtime"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "arc-swap",
 "hotpath",
@@ -10394,7 +10398,7 @@ dependencies = [

 [[package]]
 name = "rustfs-trusted-proxies"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "async-trait",
 "axum",
@@ -10431,7 +10435,7 @@ dependencies = [

 [[package]]
 name = "rustfs-utils"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "base64-simd",
 "blake2",
@@ -10473,7 +10477,7 @@ dependencies = [

 [[package]]
 name = "rustfs-zip"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 dependencies = [
 "astral-tokio-tar",
 "async-compression",
@@ -69,7 +69,7 @@ edition = "2024"
 license = "Apache-2.0"
 repository = "https://github.com/rustfs/rustfs"
 rust-version = "1.97.1"
-version = "1.0.0-rc.1"
+version = "1.0.0-rc.2"
 homepage = "https://rustfs.com"
 description = "RustFS is a high-performance distributed object storage software built using Rust, one of the most popular languages worldwide. "
 keywords = ["RustFS", "Minio", "object-storage", "filesystem", "s3"]
@@ -86,52 +86,52 @@ redundant_clone = "warn"

 [workspace.dependencies]
 # RustFS Internal Crates
-rustfs = { path = "./rustfs", version = "1.0.0-rc.1" }
-rustfs-heal = { path = "crates/heal", version = "1.0.0-rc.1" }
-rustfs-audit = { path = "crates/audit", version = "1.0.0-rc.1" }
-rustfs-checksums = { path = "crates/checksums", version = "1.0.0-rc.1" }
-rustfs-common = { path = "crates/common", version = "1.0.0-rc.1" }
-rustfs-data-usage = { path = "crates/data-usage", version = "1.0.0-rc.1" }
-rustfs-config = { path = "./crates/config", version = "1.0.0-rc.1" }
-rustfs-concurrency = { path = "./crates/concurrency", version = "1.0.0-rc.1" }
-rustfs-credentials = { path = "crates/credentials", version = "1.0.0-rc.1" }
-rustfs-crypto = { path = "crates/crypto", version = "1.0.0-rc.1" }
-rustfs-ecstore = { path = "crates/ecstore", version = "1.0.0-rc.1" }
-rustfs-filemeta = { path = "crates/filemeta", version = "1.0.0-rc.1" }
-rustfs-iam = { path = "crates/iam", version = "1.0.0-rc.1" }
-rustfs-keystone = { path = "crates/keystone", version = "1.0.0-rc.1" }
-rustfs-lifecycle = { path = "crates/lifecycle", version = "1.0.0-rc.1" }
-rustfs-kms = { path = "crates/kms", version = "1.0.0-rc.1" }
-rustfs-lock = { path = "crates/lock", version = "1.0.0-rc.1" }
-rustfs-madmin = { path = "crates/madmin", version = "1.0.0-rc.1" }
-rustfs-notify = { path = "crates/notify", version = "1.0.0-rc.1" }
-rustfs-io-metrics = { path = "crates/io-metrics", version = "1.0.0-rc.1" }
-rustfs-io-core = { path = "crates/io-core", version = "1.0.0-rc.1" }
-rustfs-object-capacity = { path = "crates/object-capacity", version = "1.0.0-rc.1" }
-rustfs-object-data-cache = { path = "crates/object-data-cache", version = "1.0.0-rc.1", default-features = false }
-rustfs-log-analyzer = { path = "crates/log-analyzer", version = "1.0.0-rc.1" }
-rustfs-obs = { path = "crates/obs", version = "1.0.0-rc.1" }
-rustfs-policy = { path = "crates/policy", version = "1.0.0-rc.1" }
-rustfs-protos = { path = "crates/protos", version = "1.0.0-rc.1" }
-rustfs-protocols = { path = "crates/protocols", version = "1.0.0-rc.1" }
-rustfs-replication = { path = "crates/replication", version = "1.0.0-rc.1" }
-rustfs-rio = { path = "crates/rio", version = "1.0.0-rc.1" }
-rustfs-rio-v2 = { path = "crates/rio-v2", version = "1.0.0-rc.1" }
-rustfs-s3-types = { path = "crates/s3-types", version = "1.0.0-rc.1" }
-rustfs-s3-ops = { path = "crates/s3-ops", version = "1.0.0-rc.1" }
-rustfs-s3select-api = { path = "crates/s3select-api", version = "1.0.0-rc.1" }
-rustfs-s3select-query = { path = "crates/s3select-query", version = "1.0.0-rc.1" }
-rustfs-scanner = { path = "crates/scanner", version = "1.0.0-rc.1" }
-rustfs-security-governance = { path = "crates/security-governance", version = "1.0.0-rc.1" }
-rustfs-extension-schema = { path = "crates/extension-schema", version = "1.0.0-rc.1" }
-rustfs-signer = { path = "crates/signer", version = "1.0.0-rc.1" }
-rustfs-storage-api = { path = "crates/storage-api", version = "1.0.0-rc.1" }
-rustfs-trusted-proxies = { path = "crates/trusted-proxies", version = "1.0.0-rc.1" }
-rustfs-targets = { path = "crates/targets", version = "1.0.0-rc.1" }
-rustfs-test-utils = { path = "crates/test-utils", version = "1.0.0-rc.1" }
-rustfs-tls-runtime = { path = "crates/tls-runtime", version = "1.0.0-rc.1" }
-rustfs-utils = { path = "crates/utils", version = "1.0.0-rc.1" }
-rustfs-zip = { path = "./crates/zip", version = "1.0.0-rc.1" }
+rustfs = { path = "./rustfs", version = "1.0.0-rc.2" }
+rustfs-heal = { path = "crates/heal", version = "1.0.0-rc.2" }
+rustfs-audit = { path = "crates/audit", version = "1.0.0-rc.2" }
+rustfs-checksums = { path = "crates/checksums", version = "1.0.0-rc.2" }
+rustfs-common = { path = "crates/common", version = "1.0.0-rc.2" }
+rustfs-data-usage = { path = "crates/data-usage", version = "1.0.0-rc.2" }
+rustfs-config = { path = "./crates/config", version = "1.0.0-rc.2" }
+rustfs-concurrency = { path = "./crates/concurrency", version = "1.0.0-rc.2" }
+rustfs-credentials = { path = "crates/credentials", version = "1.0.0-rc.2" }
+rustfs-crypto = { path = "crates/crypto", version = "1.0.0-rc.2" }
+rustfs-ecstore = { path = "crates/ecstore", version = "1.0.0-rc.2" }
+rustfs-filemeta = { path = "crates/filemeta", version = "1.0.0-rc.2" }
+rustfs-iam = { path = "crates/iam", version = "1.0.0-rc.2" }
+rustfs-keystone = { path = "crates/keystone", version = "1.0.0-rc.2" }
+rustfs-lifecycle = { path = "crates/lifecycle", version = "1.0.0-rc.2" }
+rustfs-kms = { path = "crates/kms", version = "1.0.0-rc.2" }
+rustfs-lock = { path = "crates/lock", version = "1.0.0-rc.2" }
+rustfs-madmin = { path = "crates/madmin", version = "1.0.0-rc.2" }
+rustfs-notify = { path = "crates/notify", version = "1.0.0-rc.2" }
+rustfs-io-metrics = { path = "crates/io-metrics", version = "1.0.0-rc.2" }
+rustfs-io-core = { path = "crates/io-core", version = "1.0.0-rc.2" }
+rustfs-object-capacity = { path = "crates/object-capacity", version = "1.0.0-rc.2" }
+rustfs-object-data-cache = { path = "crates/object-data-cache", version = "1.0.0-rc.2", default-features = false }
+rustfs-log-analyzer = { path = "crates/log-analyzer", version = "1.0.0-rc.2" }
+rustfs-obs = { path = "crates/obs", version = "1.0.0-rc.2" }
+rustfs-policy = { path = "crates/policy", version = "1.0.0-rc.2" }
+rustfs-protos = { path = "crates/protos", version = "1.0.0-rc.2" }
+rustfs-protocols = { path = "crates/protocols", version = "1.0.0-rc.2" }
+rustfs-replication = { path = "crates/replication", version = "1.0.0-rc.2" }
+rustfs-rio = { path = "crates/rio", version = "1.0.0-rc.2" }
+rustfs-rio-v2 = { path = "crates/rio-v2", version = "1.0.0-rc.2" }
+rustfs-s3-types = { path = "crates/s3-types", version = "1.0.0-rc.2" }
+rustfs-s3-ops = { path = "crates/s3-ops", version = "1.0.0-rc.2" }
+rustfs-s3select-api = { path = "crates/s3select-api", version = "1.0.0-rc.2" }
+rustfs-s3select-query = { path = "crates/s3select-query", version = "1.0.0-rc.2" }
+rustfs-scanner = { path = "crates/scanner", version = "1.0.0-rc.2" }
+rustfs-security-governance = { path = "crates/security-governance", version = "1.0.0-rc.2" }
+rustfs-extension-schema = { path = "crates/extension-schema", version = "1.0.0-rc.2" }
+rustfs-signer = { path = "crates/signer", version = "1.0.0-rc.2" }
+rustfs-storage-api = { path = "crates/storage-api", version = "1.0.0-rc.2" }
+rustfs-trusted-proxies = { path = "crates/trusted-proxies", version = "1.0.0-rc.2" }
+rustfs-targets = { path = "crates/targets", version = "1.0.0-rc.2" }
+rustfs-test-utils = { path = "crates/test-utils", version = "1.0.0-rc.2" }
+rustfs-tls-runtime = { path = "crates/tls-runtime", version = "1.0.0-rc.2" }
+rustfs-utils = { path = "crates/utils", version = "1.0.0-rc.2" }
+rustfs-zip = { path = "./crates/zip", version = "1.0.0-rc.2" }

 # Async Runtime and Networking
 async-channel = "2.5.0"
@@ -171,7 +171,7 @@ tower = { version = "0.5.3" }
 tower-http = { version = "0.7.0" }

 # Serialization and Data Formats
-apache-avro = "0.22.0"
+apache-avro = { version = "0.22.0", features = ["snappy", "zstandard"] }
 bytes = { version = "1.12.1" }
 bytesize = "2.7.0"
 byteorder = "1.5.0"
@@ -116,7 +116,7 @@ chown -R 10001:10001 data logs
 docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:latest

 # Using specific version
-docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.0-rc.1
+docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.0-rc.2
 ```

 If you use [podman](https://github.com/containers/podman) instead of docker, you can install the RustFS with the below command
@@ -113,7 +113,7 @@ chown -R 10001:10001 data logs
 docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:latest

 # 使用指定版本运行
-docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.0-rc.1
+docker run -d -p 9000:9000 -p 9001:9001 -v $(pwd)/data:/data -v $(pwd)/logs:/logs rustfs/rustfs:1.0.0-rc.2
 ```

 如果您通过绑定挂载启用 TLS 证书目录，也请用同样方式准备该目录：
@@ -67,7 +67,10 @@ fn configured_capture_log_path(temp_dir: &str) -> Option<String> {
    capture_log_path(Path::new(&log_dir), temp_dir).map(|path| path.to_string_lossy().into_owned())
 }

-fn capture_command_logs(command: &mut Command, log_path: Option<&str>) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+pub(crate) fn capture_command_logs(
+    command: &mut Command,
+    log_path: Option<&str>,
+) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let Some(log_path) = log_path else {
        return Ok(());
    };
@@ -2,6 +2,7 @@

 use crate::common::{RustFSTestEnvironment, init_logging, rustfs_binary_path};
 use aws_sdk_s3::primitives::ByteStream;
+use aws_sdk_s3::types::{CompletedMultipartUpload, CompletedPart};
 use serial_test::serial;
 use std::fs;
 use std::path::PathBuf;
@@ -25,6 +26,15 @@ fn generate_compressible_data(size: usize) -> Vec<u8> {
    data
 }

+/// Deterministic 2048-byte-period binary pattern that compresses extremely well: every part
+/// yields many compressed blocks, which is exactly the shape that reproduced the mid-payload
+/// Pending truncation (rustfs/rustfs#5957).
+fn generate_high_ratio_binary_data(size: usize, seed: u8) -> Vec<u8> {
+    (0..size)
+        .map(|i| ((i as u64).wrapping_mul(2_654_435_761).wrapping_add(seed as u64) >> 3) as u8)
+        .collect()
+}
+
 fn find_part_files(temp_dir: &str, bucket: &str, object_key: &str) -> Vec<PathBuf> {
    let bucket_path = PathBuf::from(temp_dir).join(bucket);
    let mut part_files = Vec::new();
@@ -55,9 +65,14 @@ async fn start_rustfs_with_compression(env: &mut RustFSTestEnvironment) -> Resul
    env.cleanup_existing_processes().await?;

    let binary_path = rustfs_binary_path();
-    let process = Command::new(&binary_path)
+    // Route the child's stdout/stderr through the shared RUSTFS_E2E_LOG_DIR
+    // capture (survives the temp-dir cleanup on Drop and is uploaded as a CI
+    // artifact); without the env var the child inherits stdio as before.
+    let mut command = Command::new(&binary_path);
+    command
        .env("RUSTFS_CONSOLE_ENABLE", "false")
        .env("RUSTFS_COMPRESSION_ENABLED", "true")
+        .env("RUSTFS_COMPRESSION_MULTIPART_ENABLED", "true")
        .args([
            "--address",
            &env.address,
@@ -66,8 +81,9 @@ async fn start_rustfs_with_compression(env: &mut RustFSTestEnvironment) -> Resul
            "--secret-key",
            &env.secret_key,
            &env.temp_dir,
-        ])
-        .spawn()?;
+        ]);
+    crate::common::capture_command_logs(&mut command, env.capture_log_path.as_deref())?;
+    let process = command.spawn()?;

    env.process = Some(process);

@@ -154,3 +170,647 @@ async fn test_compression_roundtrip() -> Result<(), Box<dyn std::error::Error +
    env.stop_server();
    Ok(())
 }
+
+const MULTIPART_COMPRESSION_BUCKET: &str = "compression-multipart-bucket";
+const MPU_PART1_SIZE: usize = 5 * 1024 * 1024;
+const MPU_PART2_SIZE: usize = 1024 * 1024;
+
+async fn multipart_upload(
+    client: &aws_sdk_s3::Client,
+    bucket: &str,
+    key: &str,
+    parts: &[&[u8]],
+) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    let create = client.create_multipart_upload().bucket(bucket).key(key).send().await?;
+    let upload_id = create.upload_id().ok_or("missing upload id")?.to_string();
+
+    let mut completed_parts = Vec::with_capacity(parts.len());
+    for (i, part) in parts.iter().enumerate() {
+        let part_number = (i + 1) as i32;
+        let upload = client
+            .upload_part()
+            .bucket(bucket)
+            .key(key)
+            .upload_id(&upload_id)
+            .part_number(part_number)
+            .body(ByteStream::from(part.to_vec()))
+            .send()
+            .await?;
+        completed_parts.push(
+            CompletedPart::builder()
+                .part_number(part_number)
+                .e_tag(upload.e_tag().unwrap_or_default())
+                .build(),
+        );
+    }
+
+    client
+        .complete_multipart_upload()
+        .bucket(bucket)
+        .key(key)
+        .upload_id(&upload_id)
+        .multipart_upload(CompletedMultipartUpload::builder().set_parts(Some(completed_parts)).build())
+        .send()
+        .await?;
+    Ok(())
+}
+
+async fn fetch_range(
+    client: &aws_sdk_s3::Client,
+    bucket: &str,
+    key: &str,
+    range: &str,
+) -> Result<Vec<u8>, Box<dyn std::error::Error + Send + Sync>> {
+    let response = client.get_object().bucket(bucket).key(key).range(range).send().await?;
+    Ok(response.body.collect().await?.into_bytes().to_vec())
+}
+
+/// Multipart disk compression roundtrip: parts are written as independent
+/// compressed streams and every GET shape must reassemble the original bytes
+/// (rustfs/rustfs#5957: multipart uploads previously bypassed disk compression
+/// entirely).
+#[tokio::test]
+#[serial]
+async fn test_compression_multipart_roundtrip() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+    info!("Starting multipart compression roundtrip test");
+
+    let mut env = RustFSTestEnvironment::new().await?;
+    start_rustfs_with_compression(&mut env).await?;
+
+    let client = env.create_s3_client();
+    env.create_test_bucket(MULTIPART_COMPRESSION_BUCKET).await?;
+
+    let object_key = "multipart-compressible.txt";
+    let part1 = generate_compressible_data(MPU_PART1_SIZE);
+    let part2 = generate_compressible_data(MPU_PART2_SIZE);
+    let mut original_data = part1.clone();
+    original_data.extend_from_slice(&part2);
+    let total_size = original_data.len();
+
+    multipart_upload(&client, MULTIPART_COMPRESSION_BUCKET, object_key, &[&part1, &part2]).await?;
+
+    let head_response = client
+        .head_object()
+        .bucket(MULTIPART_COMPRESSION_BUCKET)
+        .key(object_key)
+        .send()
+        .await?;
+    assert_eq!(
+        head_response.content_length().unwrap_or(0) as usize,
+        total_size,
+        "Content-Length should be the logical object size"
+    );
+
+    let part_files = find_part_files(&env.temp_dir, MULTIPART_COMPRESSION_BUCKET, object_key);
+    assert!(!part_files.is_empty(), "expected on-disk part files for the multipart object");
+    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    assert!(
+        total_physical_size < (total_size / 2) as u64,
+        "Physical size {total_physical_size} should be well below original size {total_size} (multipart compression applied)"
+    );
+    info!("Multipart physical storage size: {total_physical_size} bytes (compressed from {total_size} bytes)");
+
+    // Full GET must reassemble both independently compressed parts.
+    let get_response = client
+        .get_object()
+        .bucket(MULTIPART_COMPRESSION_BUCKET)
+        .key(object_key)
+        .send()
+        .await?;
+    let downloaded = get_response.body.collect().await?.into_bytes();
+    assert_eq!(downloaded.len(), total_size);
+    assert_eq!(&downloaded[..], &original_data[..], "full GET data mismatch");
+
+    // Range fully inside part 1.
+    let range_inside_part1 = fetch_range(&client, MULTIPART_COMPRESSION_BUCKET, object_key, "bytes=1024-999423").await?;
+    assert_eq!(&range_inside_part1[..], &original_data[1024..999424], "part-1 range mismatch");
+
+    // Range crossing the part boundary.
+    let boundary_start = MPU_PART1_SIZE - 128 * 1024;
+    let boundary_end = MPU_PART1_SIZE + 128 * 1024 - 1;
+    let range_crossing = fetch_range(
+        &client,
+        MULTIPART_COMPRESSION_BUCKET,
+        object_key,
+        &format!("bytes={boundary_start}-{boundary_end}"),
+    )
+    .await?;
+    assert_eq!(
+        &range_crossing[..],
+        &original_data[boundary_start..boundary_end + 1],
+        "boundary-crossing range mismatch"
+    );
+
+    // Range fully inside part 2.
+    let part2_start = MPU_PART1_SIZE + 4096;
+    let part2_end = MPU_PART1_SIZE + 256 * 1024 - 1;
+    let range_inside_part2 = fetch_range(
+        &client,
+        MULTIPART_COMPRESSION_BUCKET,
+        object_key,
+        &format!("bytes={part2_start}-{part2_end}"),
+    )
+    .await?;
+    assert_eq!(
+        &range_inside_part2[..],
+        &original_data[part2_start..part2_end + 1],
+        "part-2 range mismatch"
+    );
+
+    // Suffix range (last 128 KiB, entirely in part 2).
+    let suffix_len = 128 * 1024;
+    let suffix = fetch_range(&client, MULTIPART_COMPRESSION_BUCKET, object_key, &format!("bytes=-{suffix_len}")).await?;
+    assert_eq!(&suffix[..], &original_data[total_size - suffix_len..], "suffix range mismatch");
+
+    // partNumber GETs must return each original part.
+    for (part_number, expected) in [(1, &part1), (2, &part2)] {
+        let response = client
+            .get_object()
+            .bucket(MULTIPART_COMPRESSION_BUCKET)
+            .key(object_key)
+            .part_number(part_number)
+            .send()
+            .await?;
+        let body = response.body.collect().await?.into_bytes();
+        assert_eq!(&body[..], &expected[..], "partNumber={part_number} GET mismatch");
+    }
+
+    info!("Multipart compression roundtrip test passed");
+    env.delete_test_bucket(MULTIPART_COMPRESSION_BUCKET).await?;
+    env.stop_server();
+    Ok(())
+}
+
+const MPU_HIGH_RATIO_BUCKET: &str = "compression-mpu-high-ratio-bucket";
+
+/// High-ratio binary multipart payload: the object key is on the compression allow-list, so the
+/// disk-compression path runs and each part is stored as many compressed blocks — the shape that
+/// reproduced the mid-payload Pending truncation (rustfs/rustfs#5957). Every GET shape must return
+/// the exact original bytes, and the stored size must show the data really was compressed.
+#[tokio::test]
+#[serial]
+async fn test_compression_multipart_high_ratio_binary_roundtrip() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+    info!("Starting multipart high-ratio binary compression roundtrip test");
+
+    let mut env = RustFSTestEnvironment::new().await?;
+    start_rustfs_with_compression(&mut env).await?;
+
+    let client = env.create_s3_client();
+    env.create_test_bucket(MPU_HIGH_RATIO_BUCKET).await?;
+
+    let object_key = "multipart-high-ratio.txt";
+    let part1 = generate_high_ratio_binary_data(MPU_PART1_SIZE, 7);
+    let part2 = generate_high_ratio_binary_data(MPU_PART2_SIZE, 61);
+    let mut original_data = part1.clone();
+    original_data.extend_from_slice(&part2);
+    let total_size = original_data.len();
+
+    multipart_upload(&client, MPU_HIGH_RATIO_BUCKET, object_key, &[&part1, &part2]).await?;
+
+    let head_response = client
+        .head_object()
+        .bucket(MPU_HIGH_RATIO_BUCKET)
+        .key(object_key)
+        .send()
+        .await?;
+    assert_eq!(
+        head_response.content_length().unwrap_or(0) as usize,
+        total_size,
+        "Content-Length should be the logical object size"
+    );
+
+    // This pattern compresses to roughly 1/50 of its logical size, so a comfortably loose 2x
+    // margin still proves the parts were stored compressed rather than raw or double-encoded.
+    let part_files = find_part_files(&env.temp_dir, MPU_HIGH_RATIO_BUCKET, object_key);
+    assert!(!part_files.is_empty(), "expected on-disk part files for the multipart object");
+    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    assert!(
+        total_physical_size < (total_size as u64) / 2,
+        "Physical size {total_physical_size} should be far below the logical size {total_size} for high-ratio data"
+    );
+    info!("High-ratio multipart physical storage size: {total_physical_size} bytes (logical {total_size} bytes)");
+
+    info!("step: full GET");
+    let get_response = client
+        .get_object()
+        .bucket(MPU_HIGH_RATIO_BUCKET)
+        .key(object_key)
+        .send()
+        .await?;
+    let downloaded = get_response.body.collect().await?.into_bytes();
+    assert_eq!(downloaded.len(), total_size);
+    assert_eq!(&downloaded[..], &original_data[..], "full GET data mismatch");
+
+    // Range crossing the part boundary.
+    info!("step: boundary range GET");
+    let boundary_start = MPU_PART1_SIZE - 128 * 1024;
+    let boundary_end = MPU_PART1_SIZE + 128 * 1024 - 1;
+    let range_crossing = fetch_range(
+        &client,
+        MPU_HIGH_RATIO_BUCKET,
+        object_key,
+        &format!("bytes={boundary_start}-{boundary_end}"),
+    )
+    .await?;
+    assert_eq!(
+        &range_crossing[..],
+        &original_data[boundary_start..boundary_end + 1],
+        "boundary-crossing range mismatch"
+    );
+
+    // partNumber GET for the trailing part.
+    info!("step: partNumber GET");
+    let part2_response = client
+        .get_object()
+        .bucket(MPU_HIGH_RATIO_BUCKET)
+        .key(object_key)
+        .part_number(2)
+        .send()
+        .await?;
+    let part2_body = part2_response.body.collect().await?.into_bytes();
+    assert_eq!(&part2_body[..], &part2[..], "partNumber=2 GET mismatch");
+
+    info!("Multipart high-ratio binary compression roundtrip test passed");
+    env.delete_test_bucket(MPU_HIGH_RATIO_BUCKET).await?;
+    env.stop_server();
+    Ok(())
+}
+
+const MPU_COPY_COMPRESSION_BUCKET: &str = "compression-mpu-copy-bucket";
+const MPU_COPY_SOURCE_SIZE: usize = 6 * 1024 * 1024;
+const MPU_COPY_RANGE_LEN: usize = 5 * 1024 * 1024;
+
+/// UploadPartCopy feeds a part from an already stored (and already compressed) object. The copied
+/// range must be decompressed on read and re-compressed into the destination part, so the final
+/// object has to match "source prefix + uploaded tail" byte for byte.
+#[tokio::test]
+#[serial]
+async fn test_compression_multipart_upload_part_copy_roundtrip() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+    info!("Starting multipart upload-part-copy compression roundtrip test");
+
+    let mut env = RustFSTestEnvironment::new().await?;
+    start_rustfs_with_compression(&mut env).await?;
+
+    let client = env.create_s3_client();
+    env.create_test_bucket(MPU_COPY_COMPRESSION_BUCKET).await?;
+
+    // Source object: a plain PUT that goes through the single-stream compression path.
+    let source_key = "copy-source.txt";
+    let source_data = generate_compressible_data(MPU_COPY_SOURCE_SIZE);
+    client
+        .put_object()
+        .bucket(MPU_COPY_COMPRESSION_BUCKET)
+        .key(source_key)
+        .body(ByteStream::from(source_data.clone()))
+        .send()
+        .await?;
+
+    // Destination object: part 1 copied from the source, part 2 uploaded directly.
+    let target_key = "copy-target.txt";
+    let part2 = generate_compressible_data(MPU_PART2_SIZE);
+    let mut expected_data = source_data[..MPU_COPY_RANGE_LEN].to_vec();
+    expected_data.extend_from_slice(&part2);
+    let total_size = expected_data.len();
+
+    let create = client
+        .create_multipart_upload()
+        .bucket(MPU_COPY_COMPRESSION_BUCKET)
+        .key(target_key)
+        .send()
+        .await?;
+    let upload_id = create.upload_id().ok_or("missing upload id")?.to_string();
+
+    let copy_part = client
+        .upload_part_copy()
+        .bucket(MPU_COPY_COMPRESSION_BUCKET)
+        .key(target_key)
+        .upload_id(&upload_id)
+        .part_number(1)
+        .copy_source(format!("{MPU_COPY_COMPRESSION_BUCKET}/{source_key}"))
+        .copy_source_range(format!("bytes=0-{}", MPU_COPY_RANGE_LEN - 1))
+        .send()
+        .await?;
+    let copy_etag = copy_part
+        .copy_part_result()
+        .and_then(|r| r.e_tag())
+        .ok_or("missing copy part etag")?
+        .to_string();
+
+    let uploaded_part = client
+        .upload_part()
+        .bucket(MPU_COPY_COMPRESSION_BUCKET)
+        .key(target_key)
+        .upload_id(&upload_id)
+        .part_number(2)
+        .body(ByteStream::from(part2.clone()))
+        .send()
+        .await?;
+
+    client
+        .complete_multipart_upload()
+        .bucket(MPU_COPY_COMPRESSION_BUCKET)
+        .key(target_key)
+        .upload_id(&upload_id)
+        .multipart_upload(
+            CompletedMultipartUpload::builder()
+                .parts(CompletedPart::builder().part_number(1).e_tag(copy_etag).build())
+                .parts(
+                    CompletedPart::builder()
+                        .part_number(2)
+                        .e_tag(uploaded_part.e_tag().unwrap_or_default())
+                        .build(),
+                )
+                .build(),
+        )
+        .send()
+        .await?;
+
+    let head_response = client
+        .head_object()
+        .bucket(MPU_COPY_COMPRESSION_BUCKET)
+        .key(target_key)
+        .send()
+        .await?;
+    assert_eq!(
+        head_response.content_length().unwrap_or(0) as usize,
+        total_size,
+        "Content-Length should be the logical object size"
+    );
+
+    let part_files = find_part_files(&env.temp_dir, MPU_COPY_COMPRESSION_BUCKET, target_key);
+    assert!(!part_files.is_empty(), "expected on-disk part files for the copied object");
+    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    assert!(
+        total_physical_size < (total_size / 2) as u64,
+        "Physical size {total_physical_size} should be well below original size {total_size} (copied part compression applied)"
+    );
+
+    let get_response = client
+        .get_object()
+        .bucket(MPU_COPY_COMPRESSION_BUCKET)
+        .key(target_key)
+        .send()
+        .await?;
+    let downloaded = get_response.body.collect().await?.into_bytes();
+    assert_eq!(downloaded.len(), total_size);
+    assert_eq!(&downloaded[..], &expected_data[..], "copied multipart GET data mismatch");
+
+    info!("Multipart upload-part-copy compression roundtrip test passed");
+    env.delete_test_bucket(MPU_COPY_COMPRESSION_BUCKET).await?;
+    env.stop_server();
+    Ok(())
+}
+
+const MPU_THREE_PARTS_BUCKET: &str = "compression-mpu-three-parts-bucket";
+const MPU_THREE_PARTS_TAIL_SIZE: usize = 512 * 1024;
+
+/// Three-part upload with uneven part sizes: each partNumber GET must map back to exactly one
+/// compressed part stream, and a suffix range must resolve inside the trailing part.
+#[tokio::test]
+#[serial]
+async fn test_compression_multipart_three_parts_part_number_gets() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    init_logging();
+    info!("Starting three-part multipart compression partNumber test");
+
+    let mut env = RustFSTestEnvironment::new().await?;
+    start_rustfs_with_compression(&mut env).await?;
+
+    let client = env.create_s3_client();
+    env.create_test_bucket(MPU_THREE_PARTS_BUCKET).await?;
+
+    let object_key = "multipart-three-parts.txt";
+    let part1 = generate_compressible_data(MPU_PART1_SIZE);
+    let part2 = generate_compressible_data(MPU_PART1_SIZE);
+    let part3 = generate_compressible_data(MPU_THREE_PARTS_TAIL_SIZE);
+    let mut original_data = part1.clone();
+    original_data.extend_from_slice(&part2);
+    original_data.extend_from_slice(&part3);
+    let total_size = original_data.len();
+
+    multipart_upload(&client, MPU_THREE_PARTS_BUCKET, object_key, &[&part1, &part2, &part3]).await?;
+
+    let head_response = client
+        .head_object()
+        .bucket(MPU_THREE_PARTS_BUCKET)
+        .key(object_key)
+        .send()
+        .await?;
+    assert_eq!(
+        head_response.content_length().unwrap_or(0) as usize,
+        total_size,
+        "Content-Length should be the logical object size"
+    );
+
+    let part_files = find_part_files(&env.temp_dir, MPU_THREE_PARTS_BUCKET, object_key);
+    assert!(!part_files.is_empty(), "expected on-disk part files for the multipart object");
+    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    assert!(
+        total_physical_size < (total_size / 2) as u64,
+        "Physical size {total_physical_size} should be well below original size {total_size} (multipart compression applied)"
+    );
+
+    // Every partNumber GET must return exactly the bytes of the corresponding uploaded part.
+    for (part_number, expected) in [(1, &part1), (2, &part2), (3, &part3)] {
+        let response = client
+            .get_object()
+            .bucket(MPU_THREE_PARTS_BUCKET)
+            .key(object_key)
+            .part_number(part_number)
+            .send()
+            .await?;
+        let body = response.body.collect().await?.into_bytes();
+        assert_eq!(&body[..], &expected[..], "partNumber={part_number} GET mismatch");
+    }
+
+    // Suffix range (last 64 KiB) resolves inside the trailing part.
+    let suffix_len = 64 * 1024;
+    let suffix = fetch_range(&client, MPU_THREE_PARTS_BUCKET, object_key, &format!("bytes=-{suffix_len}")).await?;
+    assert_eq!(&suffix[..], &original_data[total_size - suffix_len..], "suffix range mismatch");
+
+    info!("Three-part multipart compression partNumber test passed");
+    env.delete_test_bucket(MPU_THREE_PARTS_BUCKET).await?;
+    env.stop_server();
+    Ok(())
+}
+
+const MPU_SSE_COMPRESSION_BUCKET: &str = "compression-mpu-sse-bucket";
+
+async fn start_rustfs_with_compression_and_sse(
+    env: &mut RustFSTestEnvironment,
+) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    use base64::Engine;
+    env.cleanup_existing_processes().await?;
+
+    let binary_path = rustfs_binary_path();
+    let master_key = base64::engine::general_purpose::STANDARD.encode([0x42u8; 32]);
+    // Server output goes to a file inside the per-test temp dir so a failing
+    // run can be diagnosed from the child's logs.
+    let server_log = std::fs::File::create(format!("{}/server.log", env.temp_dir))?;
+    let server_log_err = server_log.try_clone()?;
+    let process = Command::new(&binary_path)
+        .env("RUSTFS_CONSOLE_ENABLE", "false")
+        .env("RUSTFS_COMPRESSION_ENABLED", "true")
+        .env("RUSTFS_COMPRESSION_MULTIPART_ENABLED", "true")
+        .env("RUSTFS_SSE_S3_MASTER_KEY", master_key)
+        .env("RUST_LOG", "rustfs=info,rustfs_ecstore=info")
+        .stdout(std::process::Stdio::from(server_log))
+        .stderr(std::process::Stdio::from(server_log_err))
+        .args([
+            "--address",
+            &env.address,
+            "--access-key",
+            &env.access_key,
+            "--secret-key",
+            &env.secret_key,
+            &env.temp_dir,
+        ])
+        .spawn()?;
+
+    env.process = Some(process);
+
+    info!("Waiting for RustFS server with compression + SSE-S3 enabled on {}", env.address);
+    for i in 0..30 {
+        if TcpStream::connect(&env.address).await.is_ok() {
+            info!("RustFS server is ready after {} attempts", i + 1);
+            return Ok(());
+        }
+        if i == 29 {
+            return Err("RustFS server failed to become ready".into());
+        }
+        sleep(Duration::from_secs(1)).await;
+    }
+    Ok(())
+}
+
+/// SSE-S3 + disk compression multipart: each part is compressed and then encrypted, and every GET
+/// shape must still return the original plaintext bytes. Physical size must shrink because the
+/// compression runs before encryption.
+#[tokio::test]
+#[serial]
+async fn test_compression_multipart_sse_s3_roundtrip() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    use aws_sdk_s3::types::ServerSideEncryption;
+
+    init_logging();
+    info!("Starting SSE-S3 multipart compression roundtrip test");
+
+    let mut env = RustFSTestEnvironment::new().await?;
+    start_rustfs_with_compression_and_sse(&mut env).await?;
+
+    let client = env.create_s3_client();
+    env.create_test_bucket(MPU_SSE_COMPRESSION_BUCKET).await?;
+
+    let object_key = "multipart-sse-compressible.txt";
+    let part1 = generate_compressible_data(MPU_PART1_SIZE);
+    let part2 = generate_compressible_data(MPU_PART2_SIZE);
+    let mut original_data = part1.clone();
+    original_data.extend_from_slice(&part2);
+    let total_size = original_data.len();
+
+    let create = client
+        .create_multipart_upload()
+        .bucket(MPU_SSE_COMPRESSION_BUCKET)
+        .key(object_key)
+        .server_side_encryption(ServerSideEncryption::Aes256)
+        .send()
+        .await?;
+    let upload_id = create.upload_id().ok_or("missing upload id")?.to_string();
+
+    let mut completed_parts = Vec::new();
+    for (i, part) in [&part1, &part2].into_iter().enumerate() {
+        let part_number = (i + 1) as i32;
+        let upload = client
+            .upload_part()
+            .bucket(MPU_SSE_COMPRESSION_BUCKET)
+            .key(object_key)
+            .upload_id(&upload_id)
+            .part_number(part_number)
+            .body(ByteStream::from(part.clone()))
+            .send()
+            .await?;
+        completed_parts.push(
+            CompletedPart::builder()
+                .part_number(part_number)
+                .e_tag(upload.e_tag().unwrap_or_default())
+                .build(),
+        );
+    }
+
+    client
+        .complete_multipart_upload()
+        .bucket(MPU_SSE_COMPRESSION_BUCKET)
+        .key(object_key)
+        .upload_id(&upload_id)
+        .multipart_upload(CompletedMultipartUpload::builder().set_parts(Some(completed_parts)).build())
+        .send()
+        .await?;
+
+    let head_response = client
+        .head_object()
+        .bucket(MPU_SSE_COMPRESSION_BUCKET)
+        .key(object_key)
+        .send()
+        .await?;
+    assert_eq!(
+        head_response.content_length().unwrap_or(0) as usize,
+        total_size,
+        "Content-Length should be the logical object size"
+    );
+    assert_eq!(
+        head_response.server_side_encryption(),
+        Some(&ServerSideEncryption::Aes256),
+        "HEAD must report SSE-S3"
+    );
+
+    let part_files = find_part_files(&env.temp_dir, MPU_SSE_COMPRESSION_BUCKET, object_key);
+    assert!(!part_files.is_empty(), "expected on-disk part files for the multipart object");
+    let total_physical_size: u64 = part_files.iter().filter_map(|p| fs::metadata(p).ok()).map(|m| m.len()).sum();
+    assert!(
+        total_physical_size < (total_size / 2) as u64,
+        "Physical size {total_physical_size} should be well below original size {total_size} (compress-then-encrypt applied)"
+    );
+
+    let get_response = client
+        .get_object()
+        .bucket(MPU_SSE_COMPRESSION_BUCKET)
+        .key(object_key)
+        .send()
+        .await?;
+    let downloaded = get_response.body.collect().await?.into_bytes();
+    assert_eq!(downloaded.len(), total_size);
+    assert_eq!(&downloaded[..], &original_data[..], "SSE-S3 multipart full GET data mismatch");
+
+    // Range crossing the part boundary must decrypt and decompress across parts.
+    let boundary_start = MPU_PART1_SIZE - 64 * 1024;
+    let boundary_end = MPU_PART1_SIZE + 64 * 1024 - 1;
+    let range_crossing = fetch_range(
+        &client,
+        MPU_SSE_COMPRESSION_BUCKET,
+        object_key,
+        &format!("bytes={boundary_start}-{boundary_end}"),
+    )
+    .await?;
+    assert_eq!(
+        &range_crossing[..],
+        &original_data[boundary_start..boundary_end + 1],
+        "SSE-S3 boundary-crossing range mismatch"
+    );
+
+    // partNumber GET for the trailing part.
+    let part2_response = client
+        .get_object()
+        .bucket(MPU_SSE_COMPRESSION_BUCKET)
+        .key(object_key)
+        .part_number(2)
+        .send()
+        .await?;
+    let part2_body = part2_response.body.collect().await?.into_bytes();
+    assert_eq!(&part2_body[..], &part2[..], "SSE-S3 partNumber=2 GET mismatch");
+
+    info!("SSE-S3 multipart compression roundtrip test passed");
+    env.delete_test_bucket(MPU_SSE_COMPRESSION_BUCKET).await?;
+    env.stop_server();
+    Ok(())
+}
@@ -76,6 +76,18 @@ const SOURCE_MTIME_HEADERS: [&str; 2] = ["x-rustfs-source-mtime", "x-minio-sourc
 const SOURCE_REPLICATION_REQUEST_HEADERS: [&str; 2] =
    ["x-rustfs-source-replication-request", "x-minio-source-replication-request"];
 const SOURCE_ETAG_HEADERS: [&str; 2] = ["x-rustfs-source-etag", "x-minio-source-etag"];
+const SOURCE_TAGGING_TIMESTAMP_HEADERS: [&str; 2] = [
+    "x-rustfs-source-replication-tagging-timestamp",
+    "x-minio-source-replication-tagging-timestamp",
+];
+const SOURCE_RETENTION_TIMESTAMP_HEADERS: [&str; 2] = [
+    "x-rustfs-source-replication-retention-timestamp",
+    "x-minio-source-replication-retention-timestamp",
+];
+const SOURCE_LEGALHOLD_TIMESTAMP_HEADERS: [&str; 2] = [
+    "x-rustfs-source-replication-legalhold-timestamp",
+    "x-minio-source-replication-legalhold-timestamp",
+];
 const RESERVED_BUCKET_PREFIXES: [&str; 3] = ["xn--", "sthree-", "amzn-s3-demo-"];
 const RESERVED_BUCKET_SUFFIXES: [&str; 6] = ["-s3alias", "--ol-s3", ".mrap", "--x-s3", "--table-s3", "-an"];

@@ -118,6 +130,25 @@ pub enum FaultAction {
    WrongEtag,
 }

+/// Replication LWW timestamp headers observed on a request, journaled so
+/// sender-side tests can assert what a real target would receive.
+#[derive(Debug, Clone, Default, PartialEq, Eq)]
+pub struct ReplicationTimestampHeaders {
+    pub tagging: Option<String>,
+    pub retention: Option<String>,
+    pub legalhold: Option<String>,
+}
+
+impl ReplicationTimestampHeaders {
+    fn from_headers(headers: &HeaderMap) -> Self {
+        Self {
+            tagging: header_value(headers, &SOURCE_TAGGING_TIMESTAMP_HEADERS).map(bounded_journal_value),
+            retention: header_value(headers, &SOURCE_RETENTION_TIMESTAMP_HEADERS).map(bounded_journal_value),
+            legalhold: header_value(headers, &SOURCE_LEGALHOLD_TIMESTAMP_HEADERS).map(bounded_journal_value),
+        }
+    }
+}
+
 /// Credential-free request metadata retained for deterministic assertions.
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct RequestRecord {
@@ -131,6 +162,7 @@ pub struct RequestRecord {
    pub part_number: Option<i32>,
    pub content_length: Option<u64>,
    pub consumed_bytes: Option<usize>,
+    pub replication_timestamps: ReplicationTimestampHeaders,
    pub fault: Option<FaultAction>,
 }

@@ -536,7 +568,15 @@ impl S3Access for FaultAccess {
            .get(CONTENT_LENGTH)
            .and_then(|value| value.to_str().ok())
            .and_then(|value| value.parse().ok());
-        let fault = record_request(&self.control, operation, context.method().clone(), parsed, content_length);
+        let replication_timestamps = ReplicationTimestampHeaders::from_headers(context.headers());
+        let fault = record_request(
+            &self.control,
+            operation,
+            context.method().clone(),
+            parsed,
+            content_length,
+            replication_timestamps,
+        );
        if let Some(RequestFault {
            action: FaultAction::Status(status),
            ..
@@ -589,6 +629,7 @@ fn record_request(
    method: Method,
    parsed: ParsedRequest,
    content_length: Option<u64>,
+    replication_timestamps: ReplicationTimestampHeaders,
 ) -> Option<RequestFault> {
    let mut state = lock(control);
    let action = parsed
@@ -613,6 +654,7 @@ fn record_request(
        part_number: parsed.part_number,
        content_length,
        consumed_bytes: None,
+        replication_timestamps,
        fault: action.clone(),
    });
    action.map(|action| RequestFault { sequence, action })
@@ -1699,6 +1741,52 @@ mod tests {
            .await?)
    }

+    #[tokio::test]
+    async fn journals_replication_timestamp_headers() -> Result<(), BoxError> {
+        let target = FakeS3Target::start().await?;
+        target.create_bucket("target-bucket");
+        let client = client(&target);
+
+        client
+            .put_object()
+            .bucket("target-bucket")
+            .key("plain")
+            .body(ByteStream::from_static(b"plain"))
+            .send()
+            .await?;
+        client
+            .put_object()
+            .bucket("target-bucket")
+            .key("stamped")
+            .body(ByteStream::from_static(b"stamped"))
+            .customize()
+            .map_request(move |mut request| {
+                let headers = request.headers_mut();
+                headers.insert("x-rustfs-source-replication-tagging-timestamp", "2026-01-02T03:04:05Z");
+                headers.insert("x-minio-source-replication-retention-timestamp", "2026-01-02T03:04:06Z");
+                headers.insert("x-rustfs-source-replication-legalhold-timestamp", "2026-01-02T03:04:07Z");
+                Ok::<_, std::convert::Infallible>(request)
+            })
+            .send()
+            .await?;
+
+        let requests = target.requests();
+        let plain = requests
+            .iter()
+            .find(|record| record.operation == Operation::PutObject && record.key.as_deref() == Some("plain"))
+            .expect("plain PUT must be journaled");
+        assert_eq!(plain.replication_timestamps, ReplicationTimestampHeaders::default());
+
+        let stamped = requests
+            .iter()
+            .find(|record| record.operation == Operation::PutObject && record.key.as_deref() == Some("stamped"))
+            .expect("stamped PUT must be journaled");
+        assert_eq!(stamped.replication_timestamps.tagging.as_deref(), Some("2026-01-02T03:04:05Z"));
+        assert_eq!(stamped.replication_timestamps.retention.as_deref(), Some("2026-01-02T03:04:06Z"));
+        assert_eq!(stamped.replication_timestamps.legalhold.as_deref(), Some("2026-01-02T03:04:07Z"));
+        Ok(())
+    }
+
    macro_rules! assert_sdk_error {
        ($error:expr, $status:expr, $code:expr) => {{
            let error = &$error;
@@ -2985,6 +3073,7 @@ mod tests {
                        part_number: None,
                    },
                    Some(0),
+                    ReplicationTimestampHeaders::default(),
                );
            }
            let records = lock(&control).requests.clone();
@@ -3006,6 +3095,7 @@ mod tests {
                    part_number: None,
                },
                None,
+                ReplicationTimestampHeaders::default(),
            );
            {
                let bounded_records = lock(&bounded_control);
@@ -1828,33 +1828,36 @@ async fn four_node_compressed_inline_fallback() -> TestResult {
    Ok(())
 }

+/// Multipart disk compression is live again, so a compression-enabled cluster classifies multipart objects as compressed and the roundtrip (full GET plus partNumber GET) must still return the original bytes.
+/// Reverting the multipart compression fix must fail this test.
 #[tokio::test]
 #[serial]
-async fn four_node_multipart_ignores_disk_compression_fallback() -> TestResult {
+async fn four_node_multipart_disk_compression_roundtrip() -> TestResult {
    init_logging();

    let collector = OtlpMetricCollector::start().await?;
    let mut cluster = RustFSTestClusterEnvironment::new(4).await?;
    configure_reader_metric_cluster(&mut cluster, &collector);
    cluster.set_env("RUSTFS_COMPRESSION_ENABLED", "true");
+    cluster.set_env("RUSTFS_COMPRESSION_MULTIPART_ENABLED", "true");
    cluster.start().await?;

-    let bucket = "inline-multipart-compression-fallback";
+    let bucket = "inline-multipart-compression-roundtrip";
    cluster.create_test_bucket(bucket).await?;
    let client = cluster.create_s3_client(0)?;
-    let key = "multipart/compression-disabled.txt";
+    let key = "multipart/compressed.txt";
    let (body, second_part, etag) = put_two_part_multipart(&client, bucket, key).await?;

    assert_reader_path(
        &collector,
        &client,
-        ReaderPathExpectation::for_class(ReaderObject::new(bucket, key, &body, etag.as_deref(), None), LEGACY_DUPLEX, MULTIPART),
+        ReaderPathExpectation::for_class(ReaderObject::new(bucket, key, &body, etag.as_deref(), None), LEGACY_DUPLEX, COMPRESSED),
    )
    .await?;
    assert_part_number_reader_path(
        &collector,
        &client,
-        PartNumberReaderPathExpectation::new(bucket, key, &second_part, body.len(), MULTIPART, LEGACY_DUPLEX),
+        PartNumberReaderPathExpectation::new(bucket, key, &second_part, body.len(), COMPRESSED, LEGACY_DUPLEX),
    )
    .await?;

@@ -1871,6 +1874,7 @@ async fn four_node_mixed_msgpack_compat_mode_preserves_fallback_controls() -> Te
    let sse_master_key = base64::engine::general_purpose::STANDARD.encode([0x42u8; 32]);
    cluster.set_env("RUSTFS_SSE_S3_MASTER_KEY", sse_master_key);
    cluster.set_env("RUSTFS_COMPRESSION_ENABLED", "true");
+    cluster.set_env("RUSTFS_COMPRESSION_MULTIPART_ENABLED", "true");
    configure_mixed_msgpack_cluster(&mut cluster, &collector)?;
    cluster.start().await?;

@@ -1890,14 +1894,21 @@ async fn four_node_mixed_msgpack_compat_mode_preserves_fallback_controls() -> Te
        ReaderPathExpectation::for_class(
            ReaderObject::new(bucket, multipart_key, &multipart_body, multipart_etag.as_deref(), None),
            LEGACY_DUPLEX,
-            MULTIPART,
+            COMPRESSED,
        ),
    )
    .await?;
    assert_part_number_reader_path(
        &collector,
        &client,
-        PartNumberReaderPathExpectation::new(bucket, multipart_key, &second_part, multipart_body.len(), MULTIPART, LEGACY_DUPLEX),
+        PartNumberReaderPathExpectation::new(
+            bucket,
+            multipart_key,
+            &second_part,
+            multipart_body.len(),
+            COMPRESSED,
+            LEGACY_DUPLEX,
+        ),
    )
    .await?;
    assert_msgpack_decode_observed(&collector, &decode_before).await?;
@@ -2353,7 +2364,11 @@ async fn four_node_mixed_msgpack_compat_mode_preserves_fallback_controls_during_
    hot_client.create_bucket().bucket(bucket).send().await?;
    put_lifecycle_with_transition_retry(&hot_client, bucket, &tier_name).await?;

-    let key = "transition/mixed-multipart.bin";
+    // `.zip` sits on the disk-compression exclusion list: this test pins
+    // msgpack compat controls across ILM transition, and a compressed object
+    // would classify as `compressed` instead of `remote` (and the warm-tier
+    // read path does not decode compression — tracked separately).
+    let key = "transition/mixed-multipart.zip";
    let (body, second_part, etag) = put_two_part_multipart(&hot_client, bucket, key).await?;
    wait_for_transition(&hot_client, bucket, key, &tier_name).await?;
    assert!(
@@ -97,7 +97,7 @@ async fn start_enforcing_ilm_server(env: &mut LocalKMSTestEnvironment) -> TestRe

    let envs = [
        ("RUSTFS_KMS_ALLOW_INSECURE_DEV_DEFAULTS", "true"),
-        ("RUSTFS_KMS_ENFORCE_SSE_KEY_POLICY", "false"),
+        ("RUSTFS_KMS_ENFORCE_SSE_KEY_POLICY", "true"),
        ("RUSTFS_SCANNER_CYCLE", "1"),
        ("RUSTFS_ILM_PROCESS_TIME", "1"),
        ("RUSTFS_ILM_DEBUG_DAY_SECS", "2"),
@@ -486,7 +486,6 @@ async fn ilm_expiration_on_sse_kms_bucket_under_enforcement() -> TestResult {
 /// depend on scanner scheduling; the 1s scanner cycle stays on as a backstop.
 #[tokio::test]
 #[serial]
-#[ignore = "pins rustfs/rustfs#6025: GET on a transitioned managed-SSE object silently returns corrupt bytes (fails with enforcement on AND off, so it is not an authorization regression); un-ignore with the fix"]
 async fn ilm_transition_on_sse_kms_bucket_under_enforcement_reads_back() -> TestResult {
    init_logging();

@@ -131,16 +131,19 @@ pub mod bucket {
    }

    pub mod metadata_sys {
+        #[cfg(feature = "test-util")]
+        pub use crate::bucket::metadata_sys::ConfigWriteLockProbe;
        pub use crate::bucket::metadata_sys::{
            BucketMetadataMutationGuard, BucketMetadataSys, ObjectLockConfigState, acquire_bucket_metadata_transaction_lock,
-            capture_bucket_metadata_incarnation, delete, delete_if_incarnation, get, get_accelerate_config, get_bucket_policy,
+            acquire_bucket_metadata_transaction_lock_for_incarnation, capture_bucket_metadata_incarnation, delete,
+            delete_if_incarnation, delete_under_transaction_lock, get, get_accelerate_config, get_bucket_policy,
            get_bucket_policy_raw, get_bucket_targets_config, get_config_from_disk, get_cors_config, get_durability_config,
            get_global_bucket_metadata_sys, get_lifecycle_config, get_logging_config, get_notification_config,
            get_object_lock_config, get_object_lock_config_state, get_public_access_block_config, get_quota_config,
            get_replication_config, get_request_payment_config, get_sse_config, get_tagging_config, get_versioning_config,
            get_website_config, init_bucket_metadata_sys, list_bucket_targets, reload_bucket_metadata, remove_bucket_metadata,
            set_bucket_metadata, update, update_bucket_targets_under_transaction_lock, update_config_with, update_if_incarnation,
-            update_under_transaction_lock,
+            update_quota_if_incarnation, update_under_transaction_lock,
        };
    }

@@ -182,17 +185,18 @@ pub mod bucket {
            mrf_backlog_observability_snapshot,
        };
        pub use crate::bucket::replication::{
-            BucketReplicationResyncStatus, BucketReplicationStats, BucketStats, DeleteReplicationConfigSnapshot,
-            DeletedObjectReplicationInfo, DurableMrfBacklog, DynReplicationPool, MrfOpKind, MrfReplicateEntry,
-            MustReplicateOptions, ObjectOpts, REMOTE_TARGET_CAPABILITY_CONTRACT_VERSION, REMOTE_TARGET_UNSUPPORTED_FIELDS,
-            REMOTE_TARGET_WRITABLE_FIELDS, REPLICATE_INCOMING_DELETE, REPLICATION_CAPABILITY_CONTRACT_VERSION,
-            REPLICATION_READ_ONLY_HISTORICAL_FIELDS, REPLICATION_WRITABLE_FIELDS, ReplicateDecision, ReplicateObjectInfo,
-            ReplicationBatchAdmission, ReplicationConfig, ReplicationConfigStructureError, ReplicationConfigurationExt,
-            ReplicationDeleteScheduleInput, ReplicationDeleteStateSource, ReplicationHealQueueResult, ReplicationObjectBridge,
-            ReplicationObjectIO, ReplicationOperation, ReplicationPoolTrait, ReplicationPriority, ReplicationQueueAdmission,
-            ReplicationScannerBridge, ReplicationState, ReplicationStats, ReplicationStatusType, ReplicationStorage,
-            ReplicationTargetValidationError, ReplicationType, ResyncOpts, ResyncStatusType, RuntimeReplicationTargetBacklog,
-            TargetReplicationResyncStatus, VersionPurgeStatusType, commit_force_delete_intent, complete_force_delete_intent,
+            BucketReplicationResyncStatus, BucketReplicationStat, BucketReplicationStats, BucketStats,
+            DeleteReplicationConfigSnapshot, DeletedObjectReplicationInfo, DurableMrfBacklog, DynReplicationPool, InQueueMetric,
+            MrfOpKind, MrfReplicateEntry, MustReplicateOptions, ObjectOpts, REMOTE_TARGET_CAPABILITY_CONTRACT_VERSION,
+            REMOTE_TARGET_UNSUPPORTED_FIELDS, REMOTE_TARGET_WRITABLE_FIELDS, REPLICATE_INCOMING_DELETE,
+            REPLICATION_CAPABILITY_CONTRACT_VERSION, REPLICATION_READ_ONLY_HISTORICAL_FIELDS, REPLICATION_WRITABLE_FIELDS,
+            ReplicateDecision, ReplicateObjectInfo, ReplicationBatchAdmission, ReplicationConfig,
+            ReplicationConfigStructureError, ReplicationConfigurationExt, ReplicationDeleteScheduleInput,
+            ReplicationDeleteStateSource, ReplicationHealQueueResult, ReplicationObjectBridge, ReplicationObjectIO,
+            ReplicationOperation, ReplicationPoolTrait, ReplicationPriority, ReplicationQueueAdmission, ReplicationScannerBridge,
+            ReplicationState, ReplicationStats, ReplicationStatusType, ReplicationStorage, ReplicationTargetValidationError,
+            ReplicationType, ResyncOpts, ResyncStatusType, RuntimeReplicationTargetBacklog, TargetReplicationResyncStatus,
+            VersionPurgeStatusType, XferStats, commit_force_delete_intent, complete_force_delete_intent,
            delete_replication_state_from_config, delete_replication_version_id, get_global_replication_pool,
            get_global_replication_stats, init_background_replication, invalid_replication_config_status_field,
            persist_force_delete_intent, read_durable_mrf_backlog, replication_state_to_filemeta, replication_status_to_filemeta,
@@ -276,7 +280,9 @@ pub mod cluster {
 }

 pub mod compression {
-    pub use crate::io_support::compress::{MIN_DISK_COMPRESSIBLE_SIZE, is_disk_compressible, is_disk_compression_enabled};
+    pub use crate::io_support::compress::{
+        MIN_DISK_COMPRESSIBLE_SIZE, is_disk_compressible, is_disk_compression_enabled, is_multipart_disk_compression_enabled,
+    };
 }

 pub mod config {
@@ -316,7 +322,7 @@ pub mod 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,
        invalidate_data_usage_snapshot_cache, live_bucket_usage_computations, load_admin_data_usage_from_backend_cached,
-        load_compression_total_from_memory, load_data_usage_from_backend, load_data_usage_from_backend_cached,
+        load_compression_total_from_memory, load_data_usage_from_backend, load_data_usage_from_backend_cached, quota_object_size,
        record_bucket_delete_marker_memory, record_bucket_object_delete_memory, record_bucket_object_version_write_memory,
        record_bucket_object_write_memory, record_bucket_object_write_unknown_previous_memory, record_compression_total_memory,
        refresh_bucket_usage_from_object_layer, refresh_versioned_bucket_usage_from_object_layer,
@@ -403,8 +409,11 @@ pub mod metrics {
 }

 pub mod notification {
+    #[cfg(any(test, feature = "test-util"))]
+    pub use crate::services::notification_sys::rotate_cross_pool_fence_fleet_proof_for_test;
    pub use crate::services::notification_sys::{
-        NotificationPeerErr, NotificationSys, get_global_notification_sys, new_global_notification_sys,
+        CrossPoolFenceFleetProofToken, NotificationPeerErr, NotificationSys, acquire_cross_pool_fence_fleet_proof,
+        cross_pool_fence_fleet_proof_matches, get_global_notification_sys, new_global_notification_sys,
        start_remote_version_state_fleet_probe,
    };
 }
@@ -464,7 +473,8 @@ pub mod set_disk {

    #[cfg(feature = "test-util")]
    pub mod test_util {
-        pub use crate::set_disk::{PutObjectCommitBarrier, PutObjectCommitPause};
+        pub use crate::bucket::quota::reservation::fail_next_quota_ledger_save_for_test;
+        pub use crate::set_disk::{MultipartCommitBarrier, MultipartCommitPause, PutObjectCommitBarrier, PutObjectCommitPause};
    }
 }

@@ -58,7 +58,9 @@ use rustfs_utils::http::{
 };
 use rustfs_utils::http::{
    SUFFIX_FORCE_DELETE, SUFFIX_SOURCE_DELETEMARKER, SUFFIX_SOURCE_ETAG, SUFFIX_SOURCE_MTIME, SUFFIX_SOURCE_REPLICATION_CHECK,
-    SUFFIX_SOURCE_REPLICATION_REQUEST, SUFFIX_SOURCE_VERSION_ID, insert_header,
+    SUFFIX_SOURCE_REPLICATION_LEGALHOLD_TIMESTAMP, SUFFIX_SOURCE_REPLICATION_REQUEST,
+    SUFFIX_SOURCE_REPLICATION_RETENTION_TIMESTAMP, SUFFIX_SOURCE_REPLICATION_TAGGING_TIMESTAMP, SUFFIX_SOURCE_VERSION_ID,
+    insert_header,
 };
 use rustls_pki_types::pem::PemObject;
 use serde::{Deserialize, Serialize};
@@ -1476,9 +1478,12 @@ impl Default for AdvancedPutOptions {
            replication_status: ReplicationStatusType::Pending,
            source_mtime: OffsetDateTime::now_utc(),
            replication_request: false,
-            retention_timestamp: OffsetDateTime::now_utc(),
-            tagging_timestamp: OffsetDateTime::now_utc(),
-            legalhold_timestamp: OffsetDateTime::now_utc(),
+            // UNIX_EPOCH means "never modified": header() must not emit a
+            // timestamp header for it, otherwise a receiver would treat an
+            // unset category as a modification made right now.
+            retention_timestamp: OffsetDateTime::UNIX_EPOCH,
+            tagging_timestamp: OffsetDateTime::UNIX_EPOCH,
+            legalhold_timestamp: OffsetDateTime::UNIX_EPOCH,
            replication_validity_check: false,
        }
    }
@@ -1675,6 +1680,16 @@ impl PutObjectOptions {
            );
        }

+        for (suffix, timestamp) in [
+            (SUFFIX_SOURCE_REPLICATION_TAGGING_TIMESTAMP, self.internal.tagging_timestamp),
+            (SUFFIX_SOURCE_REPLICATION_RETENTION_TIMESTAMP, self.internal.retention_timestamp),
+            (SUFFIX_SOURCE_REPLICATION_LEGALHOLD_TIMESTAMP, self.internal.legalhold_timestamp),
+        ] {
+            if timestamp.unix_timestamp() != 0 {
+                insert_header(&mut header, suffix, timestamp.format(&Rfc3339).unwrap_or_default());
+            }
+        }
+
        if self.internal.replication_request {
            insert_header(&mut header, SUFFIX_SOURCE_REPLICATION_REQUEST, "true");
        }
@@ -2842,6 +2857,57 @@ mod tests {
        );
    }

+    #[test]
+    fn put_object_headers_carry_replication_timestamp_headers() {
+        // MinIO receivers resolve concurrent tag/retention/legal-hold edits by
+        // last-writer-wins on these headers (object-api-options.go parses them
+        // as RFC3339); a replica without them loses every conflict resolution.
+        let mut opts = PutObjectOptions::default();
+        opts.internal.replication_request = true;
+        let tagging = OffsetDateTime::from_unix_timestamp(1_700_000_001).expect("valid timestamp");
+        let retention = OffsetDateTime::from_unix_timestamp(1_700_000_002).expect("valid timestamp");
+        let legalhold = OffsetDateTime::from_unix_timestamp(1_700_000_003).expect("valid timestamp");
+        opts.internal.tagging_timestamp = tagging;
+        opts.internal.retention_timestamp = retention;
+        opts.internal.legalhold_timestamp = legalhold;
+
+        let header = opts.header();
+        for (suffix, expected) in [
+            ("source-replication-tagging-timestamp", tagging),
+            ("source-replication-retention-timestamp", retention),
+            ("source-replication-legalhold-timestamp", legalhold),
+        ] {
+            assert_eq!(
+                rustfs_utils::http::get_header(&header, suffix).as_deref(),
+                Some(expected.format(&Rfc3339).expect("RFC3339 timestamp").as_str()),
+                "replication put requests must carry the {suffix} header"
+            );
+        }
+    }
+
+    #[test]
+    fn put_object_headers_omit_unset_replication_timestamps() {
+        // UNIX_EPOCH means "never modified on the source"; sending it would
+        // make the receiver treat an unset category as a fresh modification.
+        let mut opts = PutObjectOptions::default();
+        opts.internal.replication_request = true;
+        opts.internal.tagging_timestamp = OffsetDateTime::UNIX_EPOCH;
+        opts.internal.retention_timestamp = OffsetDateTime::UNIX_EPOCH;
+        opts.internal.legalhold_timestamp = OffsetDateTime::UNIX_EPOCH;
+
+        let header = opts.header();
+        for suffix in [
+            "source-replication-tagging-timestamp",
+            "source-replication-retention-timestamp",
+            "source-replication-legalhold-timestamp",
+        ] {
+            assert!(
+                rustfs_utils::http::get_header(&header, suffix).is_none(),
+                "unset {suffix} must not be sent to replication targets"
+            );
+        }
+    }
+
    #[tokio::test]
    async fn get_remote_target_client_internal_rejects_loopback_endpoint() {
        let sys = BucketTargetSys::default();
@@ -46,15 +46,13 @@ use crate::bucket::lifecycle::transition_transaction::run_transition_transaction
 use crate::bucket::object_lock::ObjectLockApi;
 use crate::bucket::versioning::VersioningApi as _;
 use crate::bucket::versioning_sys::BucketVersioningSys;
-use crate::client::object_api_utils::new_getobjectreader;
 use crate::disk::error::DiskError;
 use crate::disk::{DeleteOptions, Disk, DiskAPI, RUSTFS_META_BUCKET, RUSTFS_META_MULTIPART_BUCKET, STORAGE_FORMAT_FILE};
 use crate::error::Error;
 use crate::error::StorageError;
-use crate::error::{
-    error_resp_to_object_err, is_err_object_not_found, is_err_read_quorum, is_err_version_not_found, is_network_or_host_down,
-};
+use crate::error::{is_err_object_not_found, is_err_read_quorum, is_err_version_not_found, is_network_or_host_down};
 use crate::object_api::{GetObjectReader, ObjectInfo, ObjectOptions};
+use crate::object_api::{ObjectEncryptionResolver, ReadPlan};
 use crate::services::tier::{
    tier::{TierConfigMgr, TierOperationLease, tier_destination_id_from_metadata},
    warm_backend::WarmBackendGetOpts,
@@ -4400,9 +4398,10 @@ pub async fn get_transitioned_object_reader(
    h: &HeaderMap,
    oi: &ObjectInfo,
    opts: &ObjectOptions,
+    resolver: Option<&dyn ObjectEncryptionResolver>,
 ) -> Result<GetObjectReader, std::io::Error> {
    let tier_config_mgr = runtime_sources::tier_config_mgr_handle();
-    get_transitioned_object_reader_with_tier_manager(bucket, object, rs, h, oi, opts, &tier_config_mgr).await
+    get_transitioned_object_reader_with_tier_manager(bucket, object, rs, h, oi, opts, &tier_config_mgr, resolver).await
 }

 fn validate_transition_remote_version(oi: &ObjectInfo) -> Result<bool, std::io::Error> {
@@ -4422,6 +4421,10 @@ fn validate_transition_remote_version(oi: &ObjectInfo) -> Result<bool, std::io::
    }
 }

+// The resolver joins the tier manager as the second injected port this read
+// needs; grouping the request half into a struct would churn every call site of
+// a bug fix.
+#[allow(clippy::too_many_arguments)]
 pub(crate) async fn get_transitioned_object_reader_with_tier_manager(
    bucket: &str,
    object: &str,
@@ -4430,6 +4433,7 @@ pub(crate) async fn get_transitioned_object_reader_with_tier_manager(
    oi: &ObjectInfo,
    opts: &ObjectOptions,
    tier_config_mgr: &Arc<RwLock<TierConfigMgr>>,
+    resolver: Option<&dyn ObjectEncryptionResolver>,
 ) -> Result<GetObjectReader, std::io::Error> {
    validate_transition_remote_version(oi)?;
    let expected_identity = tier_destination_id_from_metadata(&oi.user_defined)?;
@@ -4447,11 +4451,16 @@ pub(crate) async fn get_transitioned_object_reader_with_tier_manager(

    tgt_client.validate_remote_version_id(&oi.transitioned_object.version_id)?;

-    let ret = new_getobjectreader(rs, oi, opts, h);
-    if let Err(err) = ret {
-        return Err(error_resp_to_object_err(err, vec![bucket, object]));
-    }
-    let (get_fn, off, length) = ret.expect("get_transitioned_object_reader should succeed after error check");
+    // The same read plan the local path uses, so the tier fetch is positioned in
+    // the object's *stored* coordinate system and the stream is handed the same
+    // decrypt/decompress transforms. Reading an encrypted object's ciphertext
+    // through a plaintext-coordinate range and skipping the transform is how a
+    // transitioned SSE object used to come back as silently corrupt bytes of the
+    // right length (rustfs/rustfs#6025).
+    let plan = ReadPlan::build_for_request(rs.clone(), oi, opts, h, resolver)
+        .await
+        .map_err(|err| std::io::Error::other(format!("building the read plan for {bucket}/{object} failed: {err}")))?;
+    let (off, length) = (plan.storage_offset() as i64, plan.storage_length());
    let mut gopts = WarmBackendGetOpts::default();

    if off >= 0 && length >= 0 {
@@ -4488,7 +4497,10 @@ pub(crate) async fn get_transitioned_object_reader_with_tier_manager(
            );
            e
        })?;
-    Ok(attach_tier_operation_lease(get_fn(reader, h.clone()), tgt_client))
+    let object_reader = plan
+        .into_object_reader(Box::new(reader), oi)
+        .map_err(|err| std::io::Error::other(format!("wrapping the tier stream for {bucket}/{object} failed: {err}")))?;
+    Ok(attach_tier_operation_lease(object_reader, tgt_client))
 }

 struct TierOperationLeaseReader {
@@ -5776,6 +5788,7 @@ mod tests {
            &object_info,
            &ObjectOptions::default(),
            &manager,
+            None,
        )
        .await
        .expect("transitioned reader should open");
@@ -5840,6 +5853,7 @@ mod tests {
            &object_info,
            &ObjectOptions::default(),
            &manager,
+            None,
        )
        .await
        {
@@ -5880,6 +5894,7 @@ mod tests {
            &object_info,
            &ObjectOptions::default(),
            &manager,
+            None,
        )
        .await
        {
@@ -6117,6 +6132,7 @@ mod tests {
            &oi,
            &ObjectOptions::default(),
            &manager,
+            None,
        )
        .await
        {
@@ -6140,6 +6156,7 @@ mod tests {
            &oi,
            &ObjectOptions::default(),
            &manager,
+            None,
        )
        .await
        {
@@ -50,6 +50,72 @@ use uuid::Uuid;

 const BUCKET_METADATA_REFRESH_INTERVAL: Duration = Duration::from_secs(15 * 60);

+#[cfg(any(test, feature = "test-util"))]
+struct ConfigWriteLockProbeState {
+    bucket: String,
+    arrived: tokio::sync::Notify,
+}
+
+#[cfg(any(test, feature = "test-util"))]
+static CONFIG_WRITE_LOCK_PROBES: std::sync::OnceLock<StdMutex<Vec<Arc<ConfigWriteLockProbeState>>>> = std::sync::OnceLock::new();
+
+#[cfg(any(test, feature = "test-util"))]
+pub struct ConfigWriteLockProbe {
+    state: Arc<ConfigWriteLockProbeState>,
+}
+
+#[cfg(any(test, feature = "test-util"))]
+impl ConfigWriteLockProbe {
+    pub fn install(bucket: &str) -> Self {
+        let state = Arc::new(ConfigWriteLockProbeState {
+            bucket: bucket.to_string(),
+            arrived: tokio::sync::Notify::new(),
+        });
+        let mut probes = CONFIG_WRITE_LOCK_PROBES
+            .get_or_init(|| StdMutex::new(Vec::new()))
+            .lock()
+            .expect("config write lock probe mutex should not poison");
+        assert!(
+            !probes.iter().any(|current| current.bucket == state.bucket),
+            "config write lock probe must be unique for a bucket"
+        );
+        probes.push(Arc::clone(&state));
+        drop(probes);
+        Self { state }
+    }
+
+    pub async fn wait_until_attempted(&self) {
+        tokio::time::timeout(Duration::from_secs(30), self.state.arrived.notified())
+            .await
+            .expect("bucket config update should attempt the transaction lock");
+    }
+}
+
+#[cfg(any(test, feature = "test-util"))]
+impl Drop for ConfigWriteLockProbe {
+    fn drop(&mut self) {
+        let mut probes = CONFIG_WRITE_LOCK_PROBES
+            .get_or_init(|| StdMutex::new(Vec::new()))
+            .lock()
+            .expect("config write lock probe mutex should not poison");
+        probes.retain(|state| !Arc::ptr_eq(state, &self.state));
+    }
+}
+
+#[cfg(any(test, feature = "test-util"))]
+fn notify_config_write_lock_attempt(bucket: &str) {
+    let probe = CONFIG_WRITE_LOCK_PROBES
+        .get_or_init(|| StdMutex::new(Vec::new()))
+        .lock()
+        .expect("config write lock probe mutex should not poison")
+        .iter()
+        .find(|probe| probe.bucket == bucket)
+        .cloned();
+    if let Some(probe) = probe {
+        probe.arrived.notify_one();
+    }
+}
+
 #[derive(Clone, Copy)]
 enum MetadataLoadMode {
    Initial,
@@ -590,6 +656,41 @@ pub async fn update_under_transaction_lock(
    update_under_config_write_guard(get_bucket_metadata_sys()?, guard, config_file, data).await
 }

+/// Clear one config file while the caller holds this bucket's transaction lock.
+pub async fn delete_under_transaction_lock(
+    guard: &BucketMetadataMutationGuard,
+    bucket: &str,
+    config_file: &str,
+) -> Result<OffsetDateTime> {
+    guard.ensure_valid(bucket)?;
+    delete_under_config_write_guard(get_bucket_metadata_sys()?, guard, config_file).await
+}
+
+pub async fn update_quota_if_incarnation(
+    bucket: &str,
+    data: Vec<u8>,
+    expected_incarnation_id: Uuid,
+    proof: &crate::services::notification_sys::CrossPoolFenceFleetProofToken,
+) -> Result<OffsetDateTime> {
+    let sys = get_bucket_metadata_sys()?;
+    let guard = Box::pin(acquire_config_write_guard_for_incarnation(
+        sys.clone(),
+        bucket,
+        Some(expected_incarnation_id),
+    ))
+    .await?;
+    if !crate::services::notification_sys::cross_pool_fence_fleet_proof_matches(proof) {
+        return Err(Error::NamespaceLockQuorumUnavailable {
+            mode: "quota_capability",
+            bucket: bucket.to_string(),
+            object: rustfs_config::QUOTA_CONFIG_FILE.to_string(),
+            required: 1,
+            achieved: 0,
+        });
+    }
+    update_under_config_write_guard(sys, &guard, rustfs_config::QUOTA_CONFIG_FILE, data).await
+}
+
 pub async fn update_bucket_targets_under_transaction_lock(
    guard: &BucketMetadataMutationGuard,
    bucket: &str,
@@ -704,6 +805,14 @@ pub async fn acquire_bucket_metadata_transaction_lock(bucket: &str) -> Result<Bu
    acquire_config_write_guard(get_bucket_metadata_sys()?, bucket).await
 }

+/// Acquire the bucket transaction lock only if its incarnation still matches.
+pub async fn acquire_bucket_metadata_transaction_lock_for_incarnation(
+    bucket: &str,
+    expected_incarnation_id: Uuid,
+) -> Result<BucketMetadataMutationGuard> {
+    acquire_config_write_guard_for_incarnation(get_bucket_metadata_sys()?, bucket, Some(expected_incarnation_id)).await
+}
+
 pub(crate) async fn acquire_bucket_metadata_transaction_lock_in(
    ctx: &crate::runtime::instance::InstanceContext,
    bucket: &str,
@@ -734,7 +843,26 @@ async fn acquire_transaction_lock_with_sys(
    let lock = api
        .new_ns_lock(RUSTFS_META_BUCKET, &bucket_metadata_transaction_lock_key(bucket))
        .await?;
-    Ok(lock.get_write_lock(crate::set_disk::get_lock_acquire_timeout()).await?)
+    let acquire = lock.get_write_lock(crate::set_disk::get_lock_acquire_timeout());
+    #[cfg(any(test, feature = "test-util"))]
+    {
+        tokio::pin!(acquire);
+        let mut notified = false;
+        let guard = futures::future::poll_fn(|cx| match std::future::Future::poll(acquire.as_mut(), cx) {
+            std::task::Poll::Pending => {
+                if !notified {
+                    notify_config_write_lock_attempt(bucket);
+                    notified = true;
+                }
+                std::task::Poll::Pending
+            }
+            std::task::Poll::Ready(result) => std::task::Poll::Ready(result),
+        })
+        .await?;
+        Ok(guard)
+    }
+    #[cfg(not(any(test, feature = "test-util")))]
+    Ok(acquire.await?)
 }

 /// The lock resource name is deliberately still the `bucket-targets` one it
@@ -889,6 +1017,37 @@ pub(crate) async fn get_object_lock_config_and_incarnation_from_disk_in(
    }
 }

+/// Re-read the quota configuration and bucket incarnation from the same
+/// authoritative metadata blob while the caller holds the bucket metadata
+/// transaction read lock.
+pub(crate) async fn get_quota_config_and_incarnation_from_disk_in(
+    ctx: &crate::runtime::instance::InstanceContext,
+    bucket: &str,
+) -> Result<(Option<BucketQuota>, Uuid, OffsetDateTime)> {
+    let bucket_meta_sys_lock = bucket_metadata_sys_of(ctx)?;
+    let bucket_meta_sys = bucket_meta_sys_lock.read().await.clone();
+
+    match bucket_meta_sys
+        .read_authoritative_metadata_from_disk_under_transaction_lock(bucket)
+        .await?
+    {
+        BucketMetadataAuthority::Authoritative(metadata)
+            if metadata.bucket_incarnation_sidecar && !metadata.bucket_incarnation_id.is_nil() =>
+        {
+            Ok((
+                metadata.quota_config.clone(),
+                metadata.bucket_incarnation_id,
+                metadata.quota_config_updated_at,
+            ))
+        }
+        BucketMetadataAuthority::Authoritative(_) => {
+            Err(Error::other(format!("bucket incarnation metadata is not authoritative: {bucket}")))
+        }
+        BucketMetadataAuthority::MissingBucket => Err(Error::BucketNotFound(bucket.to_string())),
+        BucketMetadataAuthority::Fabricated => Err(Error::other(format!("bucket quota metadata is not authoritative: {bucket}"))),
+    }
+}
+
 pub async fn get_replication_config(bucket: &str) -> Result<(ReplicationConfiguration, OffsetDateTime)> {
    let bucket_meta_sys_lock = get_bucket_metadata_sys()?;
    let bucket_meta_sys = bucket_meta_sys_lock.read().await;
@@ -52,6 +52,7 @@ impl QuotaChecker {
    ) -> Result<QuotaCheckResult, QuotaError> {
        let start_time = Instant::now();
        let quota_config = self.get_quota_config(bucket).await?;
+        let uses_durable_reservations = quota_config.uses_durable_reservations();

        // If no quota limit is set, allow operation
        let quota_limit = match quota_config.quota {
@@ -67,6 +68,7 @@ impl QuotaChecker {
                    quota_limit: None,
                    operation_size,
                    remaining: None,
+                    uses_durable_reservations,
                });
            }
            Some(q) => q,
@@ -74,14 +76,17 @@ impl QuotaChecker {

        let current_usage = self.get_real_time_usage(bucket).await?;

+        let admission_size = if uses_durable_reservations { 0 } else { operation_size };
        let expected_usage = match operation {
-            QuotaOperation::PutObject | QuotaOperation::PostObject | QuotaOperation::CopyObject => current_usage + operation_size,
+            QuotaOperation::PutObject | QuotaOperation::PostObject | QuotaOperation::CopyObject => {
+                current_usage.saturating_add(admission_size)
+            }
            QuotaOperation::DeleteObject => current_usage.saturating_sub(operation_size),
        };

        let allowed = match operation {
            QuotaOperation::PutObject | QuotaOperation::PostObject | QuotaOperation::CopyObject => {
-                quota_config.check_operation_allowed(current_usage, operation_size)
+                quota_config.check_operation_allowed(current_usage, admission_size)
            }
            QuotaOperation::DeleteObject => true,
        };
@@ -105,6 +110,7 @@ impl QuotaChecker {
            quota_limit: Some(quota_limit),
            operation_size,
            remaining,
+            uses_durable_reservations,
        };

        let duration = start_time.elapsed();
@@ -158,6 +164,26 @@ impl QuotaChecker {
            .await
    }

+    pub async fn set_durable_quota_config_if_incarnation(
+        &mut self,
+        bucket: &str,
+        quota: BucketQuota,
+        expected_incarnation_id: uuid::Uuid,
+        proof: &crate::services::notification_sys::CrossPoolFenceFleetProofToken,
+    ) -> Result<OffsetDateTime, QuotaError> {
+        let json_data = serde_json::to_vec(&quota).map_err(|e| QuotaError::InvalidConfig {
+            reason: format!("Failed to serialize quota config: {}", e),
+        })?;
+        let start_time = Instant::now();
+        let updated_at =
+            crate::bucket::metadata_sys::update_quota_if_incarnation(bucket, json_data, expected_incarnation_id, proof)
+                .await
+                .map_err(QuotaError::StorageError)?;
+
+        rustfs_common::metrics::Metrics::inc_time(Metric::QuotaSync, start_time.elapsed());
+        Ok(updated_at)
+    }
+
    async fn set_quota_config_for_incarnation(
        &mut self,
        bucket: &str,
@@ -355,6 +381,7 @@ mod tests {
            quota_limit: None,
            operation_size: 1024,
            remaining: None,
+            uses_durable_reservations: false,
        };

        assert!(result.allowed);
@@ -378,4 +405,13 @@ mod tests {
        let allowed = quota.check_operation_allowed(512, 1024);
        assert!(!allowed);
    }
+
+    #[test]
+    fn legacy_quota_rejects_full_operation_while_v1_defers_net_growth() {
+        let legacy: BucketQuota = serde_json::from_str(r#"{"quota":5}"#).expect("legacy quota should parse");
+        let durable = BucketQuota::new(Some(5));
+
+        assert!(!legacy.check_operation_allowed(4, 2));
+        assert!(durable.uses_durable_reservations());
+    }
 }
@@ -13,40 +13,100 @@
 // limitations under the License.

 pub mod checker;
+pub(crate) mod reservation;

 use crate::error::Result;
 use rustfs_config::{
    QUOTA_API_PATH, QUOTA_EXCEEDED_ERROR_CODE, QUOTA_INTERNAL_ERROR_CODE, QUOTA_INVALID_CONFIG_ERROR_CODE,
    QUOTA_NOT_FOUND_ERROR_CODE,
 };
-use serde::{Deserialize, Serialize};
+use serde::{Deserialize, Deserializer, Serialize, Serializer, de::Error as _};
 use thiserror::Error;
 use time::OffsetDateTime;

 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
 pub enum QuotaType {
-    /// Hard quota: reject immediately when exceeded
+    /// Hard quota accounting.
    #[default]
    #[serde(alias = "HARD", alias = "hard")]
    Hard,
 }

+pub(crate) const QUOTA_RESERVATION_PROTOCOL_V1: u32 = 1;
+
 /// Bucket quota configuration. quota_type defaults to Hard when omitted.
-#[derive(Debug, Deserialize, Serialize, Default, Clone, PartialEq)]
+#[derive(Debug, Default, Clone, PartialEq)]
 pub struct BucketQuota {
-    #[serde(default)]
    pub quota: Option<u64>,
    /// Defaults to Hard when missing.
-    #[serde(default)]
    pub quota_type: QuotaType,
+    /// Optional durable reservation protocol. The wire format gives older
+    /// nodes a zero hard quota so a mixed-version fleet fails closed.
+    pub reservation_protocol: Option<u32>,
    /// Timestamp when this quota configuration was set (for audit purposes)
-    #[serde(default, with = "time::serde::rfc3339::option")]
    pub created_at: Option<OffsetDateTime>,
    /// Accept updated_at for compatibility; not used.
-    #[serde(default, with = "time::serde::rfc3339::option", skip_serializing_if = "Option::is_none")]
    pub updated_at: Option<OffsetDateTime>,
 }

+#[derive(Deserialize, Serialize)]
+struct BucketQuotaWire {
+    #[serde(default)]
+    quota: Option<u64>,
+    #[serde(default)]
+    quota_type: QuotaType,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    reservation_protocol: Option<u32>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    reservation_quota: Option<u64>,
+    #[serde(default, with = "time::serde::rfc3339::option")]
+    created_at: Option<OffsetDateTime>,
+    #[serde(default, with = "time::serde::rfc3339::option", skip_serializing_if = "Option::is_none")]
+    updated_at: Option<OffsetDateTime>,
+}
+
+impl Serialize for BucketQuota {
+    fn serialize<S>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        let durable = self.uses_durable_reservations();
+        BucketQuotaWire {
+            quota: if durable { Some(0) } else { self.quota },
+            quota_type: self.quota_type.clone(),
+            reservation_protocol: self.reservation_protocol,
+            reservation_quota: if durable { self.quota } else { None },
+            created_at: self.created_at,
+            updated_at: self.updated_at,
+        }
+        .serialize(serializer)
+    }
+}
+
+impl<'de> Deserialize<'de> for BucketQuota {
+    fn deserialize<D>(deserializer: D) -> std::result::Result<Self, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        let wire = BucketQuotaWire::deserialize(deserializer)?;
+        let quota = if wire.reservation_protocol == Some(QUOTA_RESERVATION_PROTOCOL_V1) {
+            Some(
+                wire.reservation_quota
+                    .ok_or_else(|| D::Error::custom("reservation_quota is required for reservation protocol v1"))?,
+            )
+        } else {
+            wire.quota
+        };
+        Ok(Self {
+            quota,
+            quota_type: wire.quota_type,
+            reservation_protocol: wire.reservation_protocol,
+            created_at: wire.created_at,
+            updated_at: wire.updated_at,
+        })
+    }
+}
+
 impl BucketQuota {
    /// Serialize to JSON bytes. Same format as parse_all_configs.
    pub fn marshal_msg(&self) -> Result<Vec<u8>> {
@@ -63,6 +123,7 @@ impl BucketQuota {
        Self {
            quota,
            quota_type: QuotaType::Hard,
+            reservation_protocol: quota.map(|_| QUOTA_RESERVATION_PROTOCOL_V1),
            created_at: Some(now),
            updated_at: None,
        }
@@ -72,7 +133,19 @@ impl BucketQuota {
        self.quota
    }

+    pub fn uses_durable_reservations(&self) -> bool {
+        self.reservation_protocol == Some(QUOTA_RESERVATION_PROTOCOL_V1)
+    }
+
+    pub fn has_unsupported_reservation_protocol(&self) -> bool {
+        self.reservation_protocol
+            .is_some_and(|version| version != QUOTA_RESERVATION_PROTOCOL_V1)
+    }
+
    pub fn check_operation_allowed(&self, current_usage: u64, operation_size: u64) -> bool {
+        if operation_size == 0 {
+            return true;
+        }
        if let Some(quota_limit) = self.quota {
            current_usage.saturating_add(operation_size) <= quota_limit
        } else {
@@ -94,6 +167,7 @@ pub struct QuotaCheckResult {
    pub quota_limit: Option<u64>,
    pub operation_size: u64,
    pub remaining: Option<u64>,
+    pub uses_durable_reservations: bool,
 }

 #[derive(Debug)]
@@ -210,7 +284,59 @@ mod tests {
        let buf = q.marshal_msg().expect("marshal");
        let restored = BucketQuota::unmarshal(&buf).expect("unmarshal");
        assert_eq!(q.quota, restored.quota);
-        assert_eq!(q.quota_type, restored.quota_type);
+        assert_eq!(restored.quota_type, QuotaType::Hard);
+        assert_eq!(restored.reservation_protocol, Some(QUOTA_RESERVATION_PROTOCOL_V1));
+    }
+
+    #[test]
+    fn clearing_quota_keeps_the_legacy_compatible_type() {
+        let quota = BucketQuota::new(None);
+
+        assert_eq!(quota.quota_type, QuotaType::Hard);
+        assert_eq!(quota.reservation_protocol, None);
+        assert!(!quota.uses_durable_reservations());
+    }
+
+    #[test]
+    fn durable_quota_makes_legacy_nodes_fail_closed() {
+        let json = serde_json::to_vec(&BucketQuota::new(Some(2048))).expect("durable quota should serialize");
+        let quota: BucketQuota = serde_json::from_slice(&json).expect("current quota version should parse");
+        assert!(quota.uses_durable_reservations());
+        assert_eq!(quota.quota, Some(2048));
+
+        #[derive(Deserialize)]
+        enum LegacyQuotaType {
+            Hard,
+        }
+        #[derive(Deserialize)]
+        struct LegacyBucketQuota {
+            #[allow(dead_code)]
+            quota: Option<u64>,
+            #[allow(dead_code)]
+            quota_type: LegacyQuotaType,
+        }
+        let legacy = serde_json::from_slice::<LegacyBucketQuota>(&json)
+            .expect("legacy readers should ignore the reservation protocol field");
+        assert_eq!(legacy.quota, Some(0));
+        assert!(matches!(legacy.quota_type, LegacyQuotaType::Hard));
+    }
+
+    #[test]
+    fn unknown_reservation_protocol_does_not_activate_v1() {
+        let quota: BucketQuota =
+            serde_json::from_str(r#"{"quota":0,"quota_type":"Hard","reservation_protocol":2,"reservation_quota":2048}"#)
+                .expect("future protocol should remain parseable");
+
+        assert!(!quota.uses_durable_reservations());
+        assert!(quota.has_unsupported_reservation_protocol());
+    }
+
+    #[test]
+    fn reservation_protocol_v1_requires_reservation_quota() {
+        let err = serde_json::from_str::<BucketQuota>(r#"{"quota":0,"quota_type":"Hard","reservation_protocol":1}"#)
+            .expect_err("v1 without its authoritative quota must fail closed");
+
+        assert!(err.to_string().contains("reservation_quota is required"));
    }

    /// unmarshal accepts format without quota_type
@@ -81,6 +81,6 @@ pub use replication_queue_boundary::{
 pub use replication_resync_boundary::{BucketReplicationResyncStatus, ResyncOpts, TargetReplicationResyncStatus};
 pub use replication_scanner_bridge::ReplicationScannerBridge;
 pub use replication_state::{ReplicationStats, RuntimeReplicationTargetBacklog};
-pub use replication_stats_boundary::{BucketReplicationStats, BucketStats};
+pub use replication_stats_boundary::{BucketReplicationStat, BucketReplicationStats, BucketStats, InQueueMetric, XferStats};
 pub use replication_storage_boundary::{ReplicationObjectIO, ReplicationStorage};
 pub(crate) use replication_target_config_bridge::ReplicationTargetConfigBridge;
@@ -704,6 +704,12 @@ impl ReplicationStats {
        } else {
            BucketReplicationStats::new()
        };
+        // Stamp the serializable failure windows from the live samples: the
+        // samples themselves do not cross the peer-RPC wire, so this snapshot
+        // is what cluster aggregation and the metrics endpoints see.
+        for stat in replication_stats.stats.values_mut() {
+            stat.fail_stats.refresh_windows();
+        }
        let uptime = if cache.contains_key(bucket) {
            SystemTime::now()
                .duration_since(SystemTime::UNIX_EPOCH)
@@ -15,7 +15,9 @@
 #[cfg(test)]
 pub(crate) use rustfs_replication::FailStats;
 pub(crate) use rustfs_replication::{
-    ActiveWorkerStat, BucketReplicationStat, InQueueMetric, ProxyMetric, ProxyStatsCache, QueueCache, ReplicationMetricScope,
-    SRMetricsSummary, XferStats,
+    ActiveWorkerStat, ProxyMetric, ProxyStatsCache, QueueCache, ReplicationMetricScope, SRMetricsSummary,
 };
-pub use rustfs_replication::{BucketReplicationStats, BucketStats};
+// Public so the admin wire DTOs (rustfs/src/admin/replication_metrics_wire.rs)
+// can project the internal stats onto the minio-go response shapes through
+// the storage_api facade chain.
+pub use rustfs_replication::{BucketReplicationStat, BucketReplicationStats, BucketStats, InQueueMetric, XferStats};
@@ -27,8 +27,10 @@ use rustfs_utils::http::{
    AMZ_OBJECT_TAGGING, AMZ_SERVER_SIDE_ENCRYPTION, AMZ_SERVER_SIDE_ENCRYPTION_KMS_CONTEXT, AMZ_SERVER_SIDE_ENCRYPTION_KMS_ID,
    AMZ_STORAGE_CLASS, AMZ_TAG_COUNT, CACHE_CONTROL, CONTENT_DISPOSITION, CONTENT_ENCODING, CONTENT_LANGUAGE, CONTENT_TYPE,
    HeaderExt as _, SUFFIX_OBJECTLOCK_LEGALHOLD_TIMESTAMP, SUFFIX_OBJECTLOCK_RETENTION_TIMESTAMP,
-    SUFFIX_REPLICATION_ACTUAL_OBJECT_SIZE, SUFFIX_REPLICATION_SSEC_CRC, SUFFIX_TAGGING_TIMESTAMP, get_str, insert_header_map,
-    is_internal_key, is_object_encryption_marker, is_replication_stripped_encryption_key, ssec_replication_transport_header,
+    SUFFIX_REPLICATION_ACTUAL_OBJECT_SIZE, SUFFIX_REPLICATION_SSEC_CRC, SUFFIX_SOURCE_REPLICATION_LEGALHOLD_TIMESTAMP,
+    SUFFIX_SOURCE_REPLICATION_RETENTION_TIMESTAMP, SUFFIX_SOURCE_REPLICATION_TAGGING_TIMESTAMP, SUFFIX_TAGGING_TIMESTAMP,
+    get_str, insert_header_map, is_internal_key, is_object_encryption_marker, is_replication_stripped_encryption_key,
+    ssec_replication_transport_header,
 };
 use time::OffsetDateTime;
 use time::format_description::well_known::Rfc3339;
@@ -119,6 +121,27 @@ fn classify_replication_source_encryption(metadata: &HashMap<String, String>) ->
    }
 }

+fn is_legacy_source_replication_timestamp_key(key: &str) -> bool {
+    fn has_prefix_and_suffix(key: &str, prefix: &str, suffix: &str) -> bool {
+        let key = key.as_bytes();
+        key.len() == prefix.len() + suffix.len()
+            && key[..prefix.len()].eq_ignore_ascii_case(prefix.as_bytes())
+            && key[prefix.len()..].eq_ignore_ascii_case(suffix.as_bytes())
+    }
+
+    [
+        SUFFIX_SOURCE_REPLICATION_TAGGING_TIMESTAMP,
+        SUFFIX_SOURCE_REPLICATION_RETENTION_TIMESTAMP,
+        SUFFIX_SOURCE_REPLICATION_LEGALHOLD_TIMESTAMP,
+    ]
+    .iter()
+    .any(|suffix| {
+        ["x-rustfs-", "x-minio-"]
+            .iter()
+            .any(|prefix| has_prefix_and_suffix(key, prefix, suffix))
+    })
+}
+
 pub(crate) fn replication_object_is_ssec_encrypted(user_defined: &HashMap<String, String>) -> bool {
    rustfs_replication::is_ssec_encrypted(user_defined)
 }
@@ -176,6 +199,11 @@ pub(crate) fn replication_put_object_options(sc: &str, object_info: &ObjectInfo)
            continue;
        }

+        if is_legacy_source_replication_timestamp_key(key) {
+            meta.insert(format!("x-amz-meta-{key}"), value.to_string());
+            continue;
+        }
+
        if is_internal_key(key) || is_standard_header(key) {
            continue;
        }
@@ -259,15 +287,23 @@ pub(crate) fn replication_put_object_options(sc: &str, object_info: &ObjectInfo)

        if !tags.is_empty() {
            put_options.user_tags = tags;
-            put_options.internal.tagging_timestamp =
-                if let Some(timestamp) = get_str(&object_info.user_defined, SUFFIX_TAGGING_TIMESTAMP) {
-                    OffsetDateTime::parse(&timestamp, &Rfc3339)
-                        .map_err(|err| Error::other(format!("Failed to parse tagging timestamp: {err}")))?
-                } else {
-                    object_info.mod_time.unwrap_or(OffsetDateTime::UNIX_EPOCH)
-                };
        }
    }
+    // Load the stored tagging timestamp independently of whether any tags
+    // remain: DeleteObjectTagging leaves the object tagless but stamps this
+    // key, and the deletion's LWW timestamp must still reach the replica.
+    // With no stored key, fall back to mod_time only while tags exist
+    // (MinIO parity); a tagless object without the key was never tagged and
+    // keeps the epoch default (no header).
+    put_options.internal.tagging_timestamp = if let Some(timestamp) = get_str(&object_info.user_defined, SUFFIX_TAGGING_TIMESTAMP)
+    {
+        OffsetDateTime::parse(&timestamp, &Rfc3339)
+            .map_err(|err| Error::other(format!("Failed to parse tagging timestamp: {err}")))?
+    } else if !put_options.user_tags.is_empty() {
+        object_info.mod_time.unwrap_or(OffsetDateTime::UNIX_EPOCH)
+    } else {
+        OffsetDateTime::UNIX_EPOCH
+    };

    let metadata = &*object_info.user_defined;

@@ -283,13 +319,15 @@ pub(crate) fn replication_put_object_options(sc: &str, object_info: &ObjectInfo)
        put_options.cache_control = cache_control.to_string();
    }

-    if let Some(mode) = metadata.lookup(AMZ_OBJECT_LOCK_MODE) {
+    if let Some(mode) = metadata.lookup(AMZ_OBJECT_LOCK_MODE).filter(|mode| !mode.is_empty()) {
        put_options.mode = Some(ObjectLockRetentionMode::from(mode.to_uppercase().as_str()));
    }

    if let Some(retain_until_date) = metadata.lookup(AMZ_OBJECT_LOCK_RETAIN_UNTIL_DATE) {
-        put_options.retain_until_date = OffsetDateTime::parse(retain_until_date, &Rfc3339)
-            .map_err(|err| Error::other(format!("Failed to parse retain until date: {err}")))?;
+        if !retain_until_date.is_empty() {
+            put_options.retain_until_date = OffsetDateTime::parse(retain_until_date, &Rfc3339)
+                .map_err(|err| Error::other(format!("Failed to parse retain until date: {err}")))?;
+        }
        put_options.internal.retention_timestamp =
            if let Some(timestamp) = get_str(&object_info.user_defined, SUFFIX_OBJECTLOCK_RETENTION_TIMESTAMP) {
                OffsetDateTime::parse(&timestamp, &Rfc3339).unwrap_or(OffsetDateTime::UNIX_EPOCH)
@@ -694,6 +732,110 @@ mod tests {
        assert!(options.internal.replication_request);
    }

+    /// DeleteObjectTagging leaves the object tagless but stamps the
+    /// tagging-timestamp internal key; the deletion's LWW timestamp must
+    /// still be loaded (and therefore sent) so the replica can order the
+    /// deletion against concurrent tag edits.
+    #[test]
+    fn replication_put_options_carry_tagging_timestamp_after_tag_deletion() {
+        let mut metadata = std::collections::HashMap::new();
+        rustfs_utils::http::insert_str(&mut metadata, SUFFIX_TAGGING_TIMESTAMP, "2026-01-02T03:04:05Z".to_string());
+
+        let object_info = ObjectInfo {
+            user_defined: Arc::new(metadata),
+            user_tags: Arc::new(String::new()),
+            mod_time: Some(OffsetDateTime::UNIX_EPOCH),
+            version_id: Some(Uuid::nil()),
+            ..Default::default()
+        };
+
+        let (options, _) = replication_put_object_options("", &object_info).expect("build put options");
+
+        assert!(options.user_tags.is_empty());
+        assert_eq!(
+            options.internal.tagging_timestamp,
+            OffsetDateTime::parse("2026-01-02T03:04:05Z", &Rfc3339).expect("valid timestamp"),
+            "the stored tagging timestamp must load independently of remaining tags"
+        );
+
+        // A tagless object without the stored key was never tagged: the epoch
+        // default keeps the header unsent.
+        let untagged = ObjectInfo {
+            user_tags: Arc::new(String::new()),
+            mod_time: Some(OffsetDateTime::from_unix_timestamp(1_700_000_000).expect("timestamp")),
+            version_id: Some(Uuid::nil()),
+            ..Default::default()
+        };
+        let (options, _) = replication_put_object_options("", &untagged).expect("build put options");
+        assert_eq!(options.internal.tagging_timestamp, OffsetDateTime::UNIX_EPOCH);
+    }
+
+    #[test]
+    fn replication_put_options_do_not_promote_legacy_user_timestamp_metadata() {
+        let legacy_keys = [
+            "x-rustfs-source-replication-tagging-timestamp",
+            "x-rustfs-source-replication-retention-timestamp",
+            "x-rustfs-source-replication-legalhold-timestamp",
+            "x-minio-source-replication-tagging-timestamp",
+            "x-minio-source-replication-retention-timestamp",
+            "x-minio-source-replication-legalhold-timestamp",
+        ];
+        let object_info = ObjectInfo {
+            user_defined: Arc::new(
+                legacy_keys
+                    .iter()
+                    .map(|key| (key.to_string(), "2099-01-02T03:04:05Z".to_string()))
+                    .collect(),
+            ),
+            ..Default::default()
+        };
+
+        let (options, _) = replication_put_object_options("", &object_info).expect("build put options");
+
+        for legacy_key in legacy_keys {
+            assert!(!options.user_metadata.contains_key(legacy_key));
+            assert_eq!(
+                options
+                    .user_metadata
+                    .get(&format!("x-amz-meta-{legacy_key}"))
+                    .map(String::as_str),
+                Some("2099-01-02T03:04:05Z")
+            );
+        }
+        assert_eq!(options.internal.tagging_timestamp, OffsetDateTime::UNIX_EPOCH);
+        assert_eq!(options.internal.retention_timestamp, OffsetDateTime::UNIX_EPOCH);
+        assert_eq!(options.internal.legalhold_timestamp, OffsetDateTime::UNIX_EPOCH);
+    }
+
+    #[test]
+    fn replication_put_options_carry_retention_timestamp_after_clear() {
+        let mut metadata = HashMap::from([
+            (AMZ_OBJECT_LOCK_MODE.to_string(), String::new()),
+            (AMZ_OBJECT_LOCK_RETAIN_UNTIL_DATE.to_string(), String::new()),
+        ]);
+        rustfs_utils::http::insert_str(&mut metadata, SUFFIX_OBJECTLOCK_RETENTION_TIMESTAMP, "2026-01-02T03:04:05Z".to_string());
+        let object_info = ObjectInfo {
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+
+        let (options, _) = replication_put_object_options("", &object_info).expect("retention clear must replicate");
+
+        assert!(options.mode.is_none());
+        assert_eq!(options.retain_until_date, OffsetDateTime::UNIX_EPOCH);
+        assert_eq!(
+            options.internal.retention_timestamp,
+            OffsetDateTime::parse("2026-01-02T03:04:05Z", &Rfc3339).expect("valid timestamp")
+        );
+        let headers = options.header();
+        assert!(!headers.contains_key(AMZ_OBJECT_LOCK_MODE));
+        assert!(!headers.contains_key(AMZ_OBJECT_LOCK_RETAIN_UNTIL_DATE));
+        assert_eq!(
+            rustfs_utils::http::get_header(&headers, SUFFIX_SOURCE_REPLICATION_RETENTION_TIMESTAMP).as_deref(),
+            Some("2026-01-02T03:04:05Z")
+        );
+    }
+
    #[test]
    fn replication_put_options_strip_encryption_metadata_from_plaintext_objects() {
        use rustfs_utils::http::object_encryption_keys::{INTERNAL_ENCRYPTION_ORIGINAL_SIZE_HEADER, SSEC_ORIGINAL_SIZE_HEADER};
@@ -40,7 +40,14 @@ impl ARN {

 impl Display for ARN {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "arn:rustfs:{}:{}:{}:{}", self.arn_type, self.region, self.id, self.bucket)
+        // The `minio` partition is deliberate: madmin-go's ParseARN
+        // hard-rejects any other partition, so native mc/madmin tooling can
+        // only decode remote-target ARNs minted in this form (backlog#1675
+        // P1-7). Legacy `arn:rustfs:` ARNs persisted by older releases stay
+        // readable via the FromStr whitelist below; runtime matching between
+        // targets and replication rules is by full-string equality, so mixed
+        // partitions coexist safely.
+        write!(f, "arn:minio:{}:{}:{}:{}", self.arn_type, self.region, self.id, self.bucket)
    }
 }

@@ -48,7 +55,12 @@ impl FromStr for ARN {
    type Err = std::io::Error;

    fn from_str(s: &str) -> Result<Self, Self::Err> {
-        if !s.starts_with("arn:rustfs:") {
+        // Partition whitelist, not just an `arn:` check: `BucketTargetType::
+        // from_str(...).unwrap_or_default()` below never fails, so this is
+        // the only structural gate rejecting foreign ARNs. `arn:rustfs:` is
+        // the legacy partition and must stay accepted forever (persisted
+        // bucket-targets.json / replication configs from older releases).
+        if !s.starts_with("arn:minio:") && !s.starts_with("arn:rustfs:") {
            return Err(std::io::Error::new(std::io::ErrorKind::InvalidInput, "Invalid ARN format"));
        }

@@ -101,14 +113,50 @@ mod tests {
    }

    /// RustFS commonly generates ARNs with an empty region:
-    /// `arn:rustfs:replication::<deployment_id>:<bucket>`.
+    /// `arn:minio:replication::<deployment_id>:<bucket>`.
    #[test]
    fn from_str_handles_empty_region_segment() {
-        let parsed = ARN::from_str("arn:rustfs:replication::depl-123:bucket-a").expect("valid ARN must parse");
+        let parsed = ARN::from_str("arn:minio:replication::depl-123:bucket-a").expect("valid ARN must parse");

        assert_eq!(parsed.arn_type, BucketTargetType::ReplicationService);
        assert_eq!(parsed.region, "", "region segment is empty in this form");
        assert_eq!(parsed.id, "depl-123");
        assert_eq!(parsed.bucket, "bucket-a");
    }
+
+    /// madmin-go's `ParseARN` hard-rejects anything that does not start with
+    /// `arn:minio:`, so generated ARNs must use the `minio` partition or the
+    /// native mc/madmin tooling cannot decode remote-target listings.
+    #[test]
+    fn display_emits_minio_partition() {
+        let arn = ARN::new(
+            BucketTargetType::ReplicationService,
+            "depl-123".to_string(),
+            String::new(),
+            "bucket-a".to_string(),
+        );
+
+        assert_eq!(arn.to_string(), "arn:minio:replication::depl-123:bucket-a");
+    }
+
+    /// Persisted bucket-targets.json files from older RustFS releases carry
+    /// `arn:rustfs:` ARNs; the legacy partition must stay parseable forever.
+    #[test]
+    fn from_str_accepts_legacy_rustfs_partition() {
+        let parsed = ARN::from_str("arn:rustfs:replication:us-east-1:depl-123:bucket-a").expect("legacy ARN must parse");
+
+        assert_eq!(parsed.arn_type, BucketTargetType::ReplicationService);
+        assert_eq!(parsed.region, "us-east-1");
+        assert_eq!(parsed.id, "depl-123");
+        assert_eq!(parsed.bucket, "bucket-a");
+    }
+
+    /// The partition whitelist is the only structural gate: `BucketTargetType::
+    /// from_str(...).unwrap_or_default()` never fails, so any 6-segment string
+    /// would otherwise parse as `type=None`.
+    #[test]
+    fn from_str_rejects_unknown_partition() {
+        assert!(ARN::from_str("arn:aws:replication::depl-123:bucket-a").is_err());
+        assert!(ARN::from_str("not-an-arn").is_err());
+    }
 }
@@ -15,6 +15,7 @@
 use crate::disk::disk_store::{get_drive_walkdir_peek_timeout, get_drive_walkdir_stall_timeout};
 use crate::disk::error::DiskError;
 use crate::disk::{self, DiskAPI, DiskStore, WalkDirOptions};
+use futures::future::join_all;
 use metrics::counter;
 use rustfs_filemeta::{MetaCacheEntries, MetaCacheEntry, MetacacheReader, is_io_eof};
 use std::{
@@ -655,6 +656,7 @@ async fn list_path_raw_inner(
            errs.push(None);
        }
        let mut pending_entries: Vec<Option<MetaCacheEntry>> = vec![None; readers.len()];
+        let mut peek_outcomes: Vec<Option<PeekOutcome>> = std::iter::repeat_with(|| None).take(readers.len()).collect();

        loop {
            let mut current = MetaCacheEntry::default();
@@ -676,6 +678,21 @@ async fn list_path_raw_inner(
            let mut has_err = 0;
            let mut agree = 0;

+            // Start every missing head read in the same round so one stalled
+            // disk cannot multiply the wait budget by the erasure-set width.
+            // Outcomes are still consumed below in stable disk-index order.
+            let concurrent_peeks = readers.iter_mut().enumerate().filter_map(|(i, reader)| {
+                if errs[i].is_some() || pending_entries[i].is_some() {
+                    return None;
+                }
+
+                let cancel = &revjob_rx;
+                Some(async move { (i, peek_with_timeout(cancel, reader, peek_timeout).await) })
+            });
+            for (i, outcome) in join_all(concurrent_peeks).await {
+                peek_outcomes[i] = Some(outcome);
+            }
+
            for (i, r) in readers.iter_mut().enumerate() {
                if errs[i].is_some() {
                    has_err += 1;
@@ -685,7 +702,10 @@ async fn list_path_raw_inner(
                let entry = if let Some(entry) = pending_entries[i].take() {
                    entry
                } else {
-                    match peek_with_timeout(&revjob_rx, r, peek_timeout).await {
+                    let Some(outcome) = peek_outcomes[i].take() else {
+                        return Err(DiskError::Unexpected);
+                    };
+                    match outcome {
                        PeekOutcome::Ready(res) => {
                            if let Some(entry) = res {
                                // info!("read entry disk: {}, name: {}", i, entry.name);
@@ -1295,6 +1315,36 @@ mod tests {
        assert_eq!(err, DiskError::Timeout);
    }

+    #[tokio::test(start_paused = true)]
+    async fn list_path_raw_bounds_multiple_stalled_readers_by_one_peek_deadline() {
+        let peek_timeout = Duration::from_millis(20);
+        let started = tokio::time::Instant::now();
+        let err = list_path_raw(
+            CancellationToken::new(),
+            ListPathRawOptions {
+                disks: vec![None, None, None, None],
+                min_disks: 1,
+                test_reader_behaviors: vec![
+                    TestReaderBehavior::Stall,
+                    TestReaderBehavior::Stall,
+                    TestReaderBehavior::Stall,
+                    TestReaderBehavior::Stall,
+                ],
+                peek_timeout: Some(peek_timeout),
+                ..Default::default()
+            },
+        )
+        .await
+        .expect_err("all stalled readers should fail the listing");
+
+        assert_eq!(err, DiskError::Timeout);
+        assert_eq!(
+            started.elapsed(),
+            peek_timeout,
+            "reader deadlines must overlap instead of accumulating once per disk"
+        );
+    }
+
    #[tokio::test]
    async fn list_path_raw_waits_past_producer_stall_for_slow_progressing_reader() {
        let entry = MetaCacheEntry {
@@ -248,6 +248,16 @@ fn decode_remote_version_state_capability(expected_member: &str, result: &[u8])
    Ok(server_epoch)
 }

+fn decode_cross_pool_fence_capability(expected_member: &str, result: &[u8]) -> Result<(u32, Uuid)> {
+    let version = result
+        .get(..4)
+        .and_then(|value| value.try_into().ok())
+        .map(u32::from_be_bytes)
+        .ok_or_else(|| Error::other("peer returned an invalid cross-pool fence capability version"))?;
+    let epoch = decode_remote_version_state_capability(expected_member, &result[4..])?;
+    Ok((version, epoch))
+}
+
 #[derive(Clone, Debug)]
 pub struct PeerLiveEventsBatch {
    pub events: Vec<u8>,
@@ -1288,6 +1298,16 @@ impl PeerRestClient {
        Ok((self.topology_member.clone(), epoch))
    }

+    pub async fn probe_cross_pool_fence(&self, topology_fingerprint: String) -> Result<(String, u32, Uuid)> {
+        let mut probe = rustfs_protos::CROSS_POOL_FENCE_CAPABILITY_PROBE_PREFIX.to_vec();
+        probe.extend_from_slice(Uuid::new_v4().as_bytes());
+        let result = self
+            .heal_control(rustfs_protos::HEAL_CONTROL_PROTOCOL_VERSION, topology_fingerprint, probe)
+            .await?;
+        let (supported_version, epoch) = decode_cross_pool_fence_capability(&self.topology_member, &result)?;
+        Ok((self.topology_member.clone(), supported_version, epoch))
+    }
+
    pub async fn load_bucket_metadata(&self, bucket: &str, scanner_maintenance_change: bool) -> Result<()> {
        self.finalize_result(
            async {
@@ -2738,6 +2758,24 @@ mod tests {
        assert!(decode_remote_version_state_capability("node-a:9000", &nil).is_err());
    }

+    #[test]
+    fn cross_pool_fence_capability_decoder_fails_closed() {
+        let epoch = Uuid::new_v4();
+        let result = rustfs_protos::encode_cross_pool_fence_capability(1, "node-a:9000", epoch.as_bytes())
+            .expect("small capability response should encode");
+        assert_eq!(
+            decode_cross_pool_fence_capability("node-a:9000", &result).expect("valid capability should decode"),
+            (1, epoch)
+        );
+        for malformed in [&[][..], &[0, 0, 0][..], &result[..result.len() - 1]] {
+            assert!(decode_cross_pool_fence_capability("node-a:9000", malformed).is_err());
+        }
+        assert!(decode_cross_pool_fence_capability("node-b:9000", &result).is_err());
+        let nil = rustfs_protos::encode_cross_pool_fence_capability(1, "node-a:9000", Uuid::nil().as_bytes())
+            .expect("small capability response should encode");
+        assert!(decode_cross_pool_fence_capability("node-a:9000", &nil).is_err());
+    }
+
    struct TierMutationResponseFixture<'a> {
        version: u32,
        phase: TierMutationRpcPhase,
@@ -1359,6 +1359,71 @@ fn validate_decoded_file_info(file_info: &FileInfo) -> Result<()> {
    file_info.validate_for_metadata_read().map_err(Into::into)
 }

+impl RemoteDisk {
+    #[tracing::instrument(level = "trace", skip_all)]
+    pub(crate) async fn rename_data_borrowed(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+    ) -> Result<RenameDataResp> {
+        trace!(
+            event = EVENT_REMOTE_DISK_RPC,
+            component = LOG_COMPONENT_ECSTORE,
+            subsystem = LOG_SUBSYSTEM_REMOTE_DISK,
+            endpoint = %self.endpoint,
+            src_volume,
+            src_path,
+            dst_volume,
+            dst_path,
+            op = "rename_data",
+            state = "started",
+            "Remote disk RPC started"
+        );
+
+        self.execute_with_timeout_for_op(
+            "rename_data",
+            || async {
+                let file_info = compat_json(fi)?;
+                let file_info_bin = encode_file_info_msgpack(fi)?;
+                let mut client = self
+                    .get_client()
+                    .await
+                    .map_err(|err| Error::other(format!("can not get client, err: {err}")))?;
+                let mut request = Request::new(RenameDataRequest {
+                    disk: self.endpoint.to_string(),
+                    src_volume: src_volume.to_string(),
+                    src_path: src_path.to_string(),
+                    file_info,
+                    dst_volume: dst_volume.to_string(),
+                    dst_path: dst_path.to_string(),
+                    file_info_bin: file_info_bin.into(),
+                });
+                let canonical_body = rustfs_protos::canonical_rename_data_request_body(request.get_ref());
+                attach_mutation_body_digest(&mut request, canonical_body, "rename_data")?;
+
+                let response = client.rename_data(request).await?.into_inner();
+
+                if !response.success {
+                    return Err(response.error.unwrap_or_default().into());
+                }
+
+                let rename_data_resp = decode_msgpack_or_json::<RenameDataResp>(
+                    &response.rename_data_resp_bin,
+                    &response.rename_data_resp,
+                    "RenameDataResp",
+                )?;
+
+                Ok(rename_data_resp)
+            },
+            get_max_timeout_duration(),
+        )
+        .await
+    }
+}
+
 #[async_trait::async_trait]
 impl DiskAPI for RemoteDisk {
    #[tracing::instrument(level = "trace", skip_all)]
@@ -2284,58 +2349,8 @@ impl DiskAPI for RemoteDisk {
        dst_volume: &str,
        dst_path: &str,
    ) -> Result<RenameDataResp> {
-        trace!(
-            event = EVENT_REMOTE_DISK_RPC,
-            component = LOG_COMPONENT_ECSTORE,
-            subsystem = LOG_SUBSYSTEM_REMOTE_DISK,
-            endpoint = %self.endpoint,
-            src_volume,
-            src_path,
-            dst_volume,
-            dst_path,
-            op = "rename_data",
-            state = "started",
-            "Remote disk RPC started"
-        );
-
-        self.execute_with_timeout_for_op(
-            "rename_data",
-            || async {
-                let file_info = compat_json(&fi)?;
-                let file_info_bin = encode_file_info_msgpack(&fi)?;
-                let mut client = self
-                    .get_client()
-                    .await
-                    .map_err(|err| Error::other(format!("can not get client, err: {err}")))?;
-                let mut request = Request::new(RenameDataRequest {
-                    disk: self.endpoint.to_string(),
-                    src_volume: src_volume.to_string(),
-                    src_path: src_path.to_string(),
-                    file_info,
-                    dst_volume: dst_volume.to_string(),
-                    dst_path: dst_path.to_string(),
-                    file_info_bin: file_info_bin.into(),
-                });
-                let canonical_body = rustfs_protos::canonical_rename_data_request_body(request.get_ref());
-                attach_mutation_body_digest(&mut request, canonical_body, "rename_data")?;
-
-                let response = client.rename_data(request).await?.into_inner();
-
-                if !response.success {
-                    return Err(response.error.unwrap_or_default().into());
-                }
-
-                let rename_data_resp = decode_msgpack_or_json::<RenameDataResp>(
-                    &response.rename_data_resp_bin,
-                    &response.rename_data_resp,
-                    "RenameDataResp",
-                )?;
-
-                Ok(rename_data_resp)
-            },
-            get_max_timeout_duration(),
-        )
-        .await
+        self.rename_data_borrowed(src_volume, src_path, &fi, dst_volume, dst_path)
+            .await
    }

    #[tracing::instrument(level = "trace", skip_all)]
@@ -12,6 +12,16 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+//! Per-disk usage snapshots persisted under the metadata bucket.
+//!
+//! **Nothing calls into this module.** It landed complete with tests in #5307
+//! (2026-07-27) and its aggregation entry point,
+//! [`crate::data_usage::aggregate_local_snapshots`], has never had a caller in
+//! the tree's history. The live data-usage path is
+//! `load_data_usage_from_backend` / `store_data_usage_in_backend`. The items
+//! below therefore carry individual `dead_code` allows rather than a module
+//! blanket, so the gap stays greppable until it is either wired up or removed.
+
 use crate::data_usage::BucketUsageInfo;
 use crate::disk::RUSTFS_META_BUCKET;
 use crate::error::{Error, Result};
@@ -26,10 +36,12 @@ pub const DATA_USAGE_DIR: &str = "datausage";
 /// Directory used to store incremental scan state files under the metadata bucket.
 pub const DATA_USAGE_STATE_DIR: &str = "datausage/state";
 /// Snapshot file format version, allows forward compatibility if the structure evolves.
+#[allow(dead_code, reason = "unwired local usage-snapshot feature; see module docs (backlog#1823)")]
 pub const LOCAL_USAGE_SNAPSHOT_VERSION: u32 = 1;

 /// Additional metadata describing which disk produced the snapshot.
 #[derive(Debug, Clone, Serialize, Deserialize, Default)]
+#[allow(dead_code, reason = "unwired local usage-snapshot feature; see module docs (backlog#1823)")]
 pub struct LocalUsageSnapshotMeta {
    /// Disk UUID stored as a string for simpler serialization.
    pub disk_id: String,
@@ -43,6 +55,7 @@ pub struct LocalUsageSnapshotMeta {

 /// Usage snapshot produced by a single disk.
 #[derive(Debug, Clone, Serialize, Deserialize, Default)]
+#[allow(dead_code, reason = "unwired local usage-snapshot feature; see module docs (backlog#1823)")]
 pub struct LocalUsageSnapshot {
    /// Format version recorded in the snapshot.
    pub format_version: u32,
@@ -64,6 +77,7 @@ pub struct LocalUsageSnapshot {
    pub objects_total_size: u64,
 }

+#[allow(dead_code, reason = "unwired local usage-snapshot feature; see module docs (backlog#1823)")]
 impl LocalUsageSnapshot {
    /// Create an empty snapshot with the default format version filled in.
    pub fn new(meta: LocalUsageSnapshotMeta) -> Self {
@@ -99,11 +113,13 @@ impl LocalUsageSnapshot {
 }

 /// Build the snapshot file name `<disk-id>.json`.
+#[allow(dead_code, reason = "unwired local usage-snapshot feature; see module docs (backlog#1823)")]
 pub fn snapshot_file_name(disk_id: &str) -> String {
    format!("{disk_id}.json")
 }

 /// Build the object path relative to `RUSTFS_META_BUCKET`, e.g. `datausage/<disk-id>.json`.
+#[allow(dead_code, reason = "unwired local usage-snapshot feature; see module docs (backlog#1823)")]
 pub fn snapshot_object_path(disk_id: &str) -> String {
    format!("{}/{}", DATA_USAGE_DIR, snapshot_file_name(disk_id))
 }
@@ -119,11 +135,13 @@ pub fn data_usage_state_dir(root: &Path) -> PathBuf {
 }

 /// Build the absolute path to the snapshot file for the provided disk ID.
+#[allow(dead_code, reason = "unwired local usage-snapshot feature; see module docs (backlog#1823)")]
 pub fn snapshot_path(root: &Path, disk_id: &str) -> PathBuf {
    data_usage_dir(root).join(snapshot_file_name(disk_id))
 }

 /// Read a snapshot from disk if it exists.
+#[allow(dead_code, reason = "unwired local usage-snapshot feature; see module docs (backlog#1823)")]
 pub async fn read_snapshot(root: &Path, disk_id: &str) -> Result<Option<LocalUsageSnapshot>> {
    let path = snapshot_path(root, disk_id);
    match fs::read(&path).await {
@@ -138,6 +156,7 @@ pub async fn read_snapshot(root: &Path, disk_id: &str) -> Result<Option<LocalUsa
 }

 /// Persist a snapshot to disk, creating directories as needed and overwriting any existing file.
+#[allow(dead_code, reason = "unwired local usage-snapshot feature; see module docs (backlog#1823)")]
 pub async fn write_snapshot(root: &Path, disk_id: &str, snapshot: &LocalUsageSnapshot) -> Result<()> {
    let dir = data_usage_dir(root);
    fs::create_dir_all(&dir).await.map_err(Error::other)?;
@@ -13,7 +13,6 @@
 // limitations under the License.

 // #730: scanner/data-usage state is partially migrated and still owns staged cache helpers.
-#![allow(dead_code)]

 pub mod local_snapshot;

@@ -34,8 +33,8 @@ use crate::{
 pub use local_snapshot::{LocalUsageSnapshot, read_snapshot as read_local_snapshot, snapshot_path};
 use rustfs_data_usage::{
    BucketTargetUsageInfo, BucketUsageInfo, CompressionTotalInfo, DATA_USAGE_OBJECT_NAME, DATA_USAGE_OBSERVED_OBJECT_NAME,
-    DataUsageCache, DataUsageEntry, DataUsageInfo, DiskUsageStatus, LEGACY_DATA_USAGE_OBJECT_NAME, SizeHistogram, SizeSummary,
-    VersionsHistogram, observed_data_usage_is_newer,
+    DataUsageCache, DataUsageInfo, DiskUsageStatus, LEGACY_DATA_USAGE_OBJECT_NAME, SizeHistogram, VersionsHistogram,
+    observed_data_usage_is_newer,
 };
 use rustfs_io_metrics::record_system_path_failure;
 use rustfs_utils::path::SLASH_SEPARATOR;
@@ -55,7 +54,6 @@ use tracing::{debug, error, info, instrument};
 // Data usage storage constants
 pub const DATA_USAGE_ROOT: &str = SLASH_SEPARATOR;
 const DATA_COMPRESSION_TOTAL_NAME: &str = ".compression.json";
-const DATA_USAGE_BLOOM_NAME: &str = ".bloomcycle.bin";
 pub const DATA_USAGE_CACHE_NAME: &str = ".usage-cache.bin";
 const DATA_USAGE_CACHE_TTL_SECS: u64 = 30;
 const LIVE_BUCKET_USAGE_MAX_ENTRIES: u64 = 1024;
@@ -313,11 +311,6 @@ lazy_static::lazy_static! {
        LEGACY_DATA_USAGE_OBJECT_NAME
    );
    static ref LEGACY_DATA_USAGE_OBJ_BACKUP_PATH: String = format!("{}.bkp", LEGACY_DATA_USAGE_OBJ_NAME_PATH.as_str());
-    pub static ref DATA_USAGE_BLOOM_NAME_PATH: String = format!("{}{}{}",
-        crate::disk::BUCKET_META_PREFIX,
-        SLASH_SEPARATOR,
-        DATA_USAGE_BLOOM_NAME
-    );
    pub static ref DATA_COMPRESSION_TOTAL_NAME_PATH: String = format!("{}{}{}",
        crate::disk::BUCKET_META_PREFIX,
        SLASH_SEPARATOR,
@@ -858,6 +851,10 @@ async fn resolve_loaded_snapshot_pair_with_source(
    }
 }

+#[allow(
+    dead_code,
+    reason = "primary/backup snapshot fallback asserted by this file's tests (backlog#1823)"
+)]
 async fn resolve_loaded_snapshot(
    primary: Result<Vec<u8>, Error>,
    backup: impl Future<Output = Result<Vec<u8>, Error>>,
@@ -1187,6 +1184,10 @@ pub async fn invalidate_admin_data_usage_snapshot_cache() {
 }

 /// Aggregate usage information from local disk snapshots.
+#[allow(
+    dead_code,
+    reason = "reached only through aggregate_local_snapshots, which has no caller (backlog#1823)"
+)]
 fn merge_snapshot(aggregated: &mut DataUsageInfo, mut snapshot: LocalUsageSnapshot, latest_update: &mut Option<SystemTime>) {
    if let Some(update) = snapshot.last_update
        && latest_update.is_none_or(|current| update > current)
@@ -1220,6 +1221,10 @@ fn merge_snapshot(aggregated: &mut DataUsageInfo, mut snapshot: LocalUsageSnapsh
    }
 }

+#[allow(
+    dead_code,
+    reason = "entry point of the local usage-snapshot feature, which has had no caller since it landed in #5307 (backlog#1823)"
+)]
 pub async fn aggregate_local_snapshots(store: Arc<ECStore>) -> Result<(Vec<DiskUsageStatus>, DataUsageInfo), Error> {
    let mut aggregated = DataUsageInfo::default();
    let mut latest_update: Option<SystemTime> = None;
@@ -1355,7 +1360,7 @@ impl BucketUsageAccumulator {
            return Ok(());
        }

-        let object_size = object.size.max(0) as u64;
+        let object_size = quota_object_size(object)?;
        self.current_live_versions = self.current_live_versions.saturating_add(1);
        self.size_histogram.add(object_size);
        self.total_size = self.total_size.saturating_add(object_size);
@@ -1385,6 +1390,31 @@ impl BucketUsageAccumulator {
    }
 }

+pub fn quota_object_size(object: &ObjectInfo) -> Result<u64, Error> {
+    let logical_size = u64::try_from(object.get_actual_size().map_err(Error::other)?).map_err(|_| Error::PartMissingOrCorrupt)?;
+    let persisted_part_size = if object.parts.is_empty() {
+        u64::try_from(object.size).map_err(|_| Error::PartMissingOrCorrupt)?
+    } else {
+        object.parts.iter().try_fold(0_u64, |total, part| {
+            // Compressed streaming objects persist -1 when the transformed
+            // part size is unknown. The physical part size remains a valid
+            // quota floor; reject only non-negative values that overflow.
+            let actual_size = if part.actual_size < 0 {
+                if object.is_compressed() {
+                    0
+                } else {
+                    return Err(Error::PartMissingOrCorrupt);
+                }
+            } else {
+                u64::try_from(part.actual_size).map_err(|_| Error::PartMissingOrCorrupt)?
+            };
+            let part_size = actual_size.max(u64::try_from(part.size).map_err(|_| Error::PartMissingOrCorrupt)?);
+            total.checked_add(part_size).ok_or(Error::PartMissingOrCorrupt)
+        })?
+    };
+    Ok(logical_size.max(persisted_part_size))
+}
+
 type UsageVersionPage = StorageListObjectVersionsInfo<ObjectInfo>;

 pub async fn compute_bucket_usage(store: Arc<ECStore>, bucket_name: &str) -> Result<BucketUsageInfo, Error> {
@@ -1742,11 +1772,6 @@ pub async fn record_bucket_object_write_unknown_previous_memory(bucket: &str, ne
    entry.pending_scanner_position = None;
 }

-/// Fast in-memory increment for immediate quota consistency.
-pub async fn increment_bucket_usage_memory(bucket: &str, size_increment: u64) {
-    record_bucket_object_write_memory(bucket, None, size_increment).await;
-}
-
 /// Fast in-memory update for successful object deletes.
 pub async fn record_bucket_object_delete_memory(bucket: &str, deleted_size: u64, removed_current_object: bool) {
    ensure_bucket_usage_cached(bucket).await;
@@ -1789,11 +1814,6 @@ pub async fn record_bucket_delete_marker_memory(bucket: &str) {
    entry.pending_scanner_position = None;
 }

-/// Fast in-memory decrement for immediate quota consistency
-pub async fn decrement_bucket_usage_memory(bucket: &str, size_decrement: u64) {
-    record_bucket_object_delete_memory(bucket, size_decrement, size_decrement > 0).await;
-}
-
 /// Get bucket usage from the authoritative cache for this topology.
 async fn get_persisted_bucket_usage(bucket: &str) -> Option<u64> {
    let store = runtime_sources::object_store_handle()?;
@@ -1988,91 +2008,6 @@ pub async fn apply_bucket_usage_memory_overlay(data_usage_info: &mut DataUsageIn
    apply_bucket_usage_memory_overlay_if_authoritative(data_usage_info, authoritative).await;
 }

-/// Sync memory cache with backend data (called by scanner)
-pub async fn sync_memory_cache_with_backend() -> Result<(), Error> {
-    if let Some(store) = runtime_sources::object_store_handle() {
-        match load_data_usage_from_backend(store.clone()).await {
-            Ok(data_usage_info) => {
-                replace_bucket_usage_memory_from_info(&data_usage_info).await;
-            }
-            Err(e) => {
-                debug!("Failed to sync memory cache with backend: {}", e);
-            }
-        }
-    }
-    Ok(())
-}
-
-/// Create a data usage cache entry from size summary
-pub fn create_cache_entry_from_summary(summary: &SizeSummary) -> DataUsageEntry {
-    let mut entry = DataUsageEntry::default();
-    entry.add_sizes(summary);
-    entry
-}
-
-/// Convert data usage cache to DataUsageInfo
-pub fn cache_to_data_usage_info(
-    cache: &DataUsageCache,
-    path: &str,
-    buckets: &[crate::storage_api_contracts::bucket::BucketInfo],
-) -> DataUsageInfo {
-    let e = match cache.find(path) {
-        Some(e) => e,
-        None => return DataUsageInfo::default(),
-    };
-    let flat = cache.flatten(&e);
-
-    let mut buckets_usage = HashMap::new();
-    for bucket in buckets.iter() {
-        let e = match cache.find(&bucket.name) {
-            Some(e) => e,
-            None => continue,
-        };
-        let flat = cache.flatten(&e);
-        let mut bui = BucketUsageInfo {
-            size: flat.size as u64,
-            versions_count: flat.versions as u64,
-            objects_count: flat.objects as u64,
-            delete_markers_count: flat.delete_markers as u64,
-            object_size_histogram: flat.obj_sizes.to_map(),
-            object_versions_histogram: flat.obj_versions.to_map(),
-            ..Default::default()
-        };
-
-        if let Some(rs) = &flat.replication_stats {
-            bui.replica_size = rs.replica_size;
-            bui.replica_count = rs.replica_count;
-
-            for (arn, stat) in rs.targets.iter() {
-                bui.replication_info.insert(
-                    arn.clone(),
-                    BucketTargetUsageInfo {
-                        replication_pending_size: stat.pending_size,
-                        replicated_size: stat.replicated_size,
-                        replication_failed_size: stat.failed_size,
-                        replication_pending_count: stat.pending_count,
-                        replication_failed_count: stat.failed_count,
-                        replicated_count: stat.replicated_count,
-                        ..Default::default()
-                    },
-                );
-            }
-        }
-        buckets_usage.insert(bucket.name.clone(), bui);
-    }
-
-    DataUsageInfo {
-        last_update: cache.info.last_update,
-        objects_total_count: flat.objects as u64,
-        versions_total_count: flat.versions as u64,
-        delete_markers_total_count: flat.delete_markers as u64,
-        objects_total_size: flat.size as u64,
-        buckets_count: e.children.len() as u64,
-        buckets_usage,
-        ..Default::default()
-    }
-}
-
 // Helper functions for DataUsageCache operations
 pub async fn load_data_usage_cache(store: &crate::set_disk::SetDisks, name: &str) -> crate::error::Result<DataUsageCache> {
    use crate::disk::{BUCKET_META_PREFIX, RUSTFS_META_BUCKET};
@@ -3124,6 +3059,102 @@ mod tests {
        assert_eq!(usage.object_versions_histogram.get("BETWEEN_1000_AND_10000"), Some(&1));
    }

+    #[test]
+    fn bucket_usage_uses_the_larger_of_logical_and_physical_size() {
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(
+            &mut metadata,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            "klauspost/compress/s2".to_string(),
+        );
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "4096".to_string());
+        let object = ObjectInfo {
+            name: "compressed".to_string(),
+            size: 128,
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+        let mut usage = BucketUsageAccumulator::default();
+        usage
+            .record("bucket", &object)
+            .expect("valid compressed metadata should be counted");
+        assert_eq!(usage.finish().size, 4096);
+
+        let mut framed_metadata = HashMap::new();
+        rustfs_utils::http::insert_str(
+            &mut framed_metadata,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            "klauspost/compress/s2".to_string(),
+        );
+        rustfs_utils::http::insert_str(&mut framed_metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "1".to_string());
+        let framed = ObjectInfo {
+            name: "framed".to_string(),
+            size: 17,
+            user_defined: Arc::new(framed_metadata),
+            ..Default::default()
+        };
+        assert_eq!(quota_object_size(&framed).expect("physical framing must remain quota-accounted"), 17);
+
+        let legacy_compressed_part = ObjectInfo {
+            name: "legacy-compressed-part".to_string(),
+            size: 1,
+            user_defined: Arc::new((*framed.user_defined).clone()),
+            parts: Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
+                size: 1,
+                actual_size: -1,
+                ..Default::default()
+            }]),
+            ..Default::default()
+        };
+        assert_eq!(
+            quota_object_size(&legacy_compressed_part).expect("unknown compressed part size is a valid sentinel"),
+            1
+        );
+
+        let uncompressed_negative_part = ObjectInfo {
+            name: "uncompressed-negative-part".to_string(),
+            size: 1,
+            parts: Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
+                size: 1,
+                actual_size: -1,
+                ..Default::default()
+            }]),
+            ..Default::default()
+        };
+        assert!(matches!(quota_object_size(&uncompressed_negative_part), Err(Error::PartMissingOrCorrupt)));
+
+        let mut corrupt_metadata = (*object.user_defined).clone();
+        rustfs_utils::http::insert_str(&mut corrupt_metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "-1".to_string());
+        let corrupt = ObjectInfo {
+            user_defined: Arc::new(corrupt_metadata),
+            ..object
+        };
+        assert!(matches!(quota_object_size(&corrupt), Err(Error::PartMissingOrCorrupt)));
+
+        let mut poisoned_metadata = HashMap::new();
+        rustfs_utils::http::insert_str(
+            &mut poisoned_metadata,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            "klauspost/compress/s2".to_string(),
+        );
+        rustfs_utils::http::insert_str(&mut poisoned_metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "1".to_string());
+        let poisoned = ObjectInfo {
+            name: "legacy-swift-metadata".to_string(),
+            size: 4096,
+            user_defined: Arc::new(poisoned_metadata),
+            parts: Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
+                size: 4096,
+                actual_size: 4096,
+                ..Default::default()
+            }]),
+            ..Default::default()
+        };
+        assert_eq!(
+            quota_object_size(&poisoned).expect("persisted part accounting must bound legacy user metadata"),
+            4096
+        );
+    }
+
    #[tokio::test]
    #[serial]
    async fn live_bucket_usage_refreshes_are_coalesced_only_while_in_flight() {
@@ -190,6 +190,17 @@ pub(crate) const GET_METADATA_CACHE_REASON_VERSION_SUSPENDED: &str = "version_su
 pub(crate) const GET_METADATA_CACHE_REASON_VERSIONED: &str = "versioned";
 pub(crate) const GET_METADATA_EARLY_STOP_REASON_CONFLICTING_METADATA: &str = "conflicting_metadata";
 pub(crate) const GET_METADATA_EARLY_STOP_REASON_DELETE_MARKER: &str = "delete_marker";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_BODY_VERIFY: &str = "data_read_inline_body_verify";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_DELETED: &str = "data_read_inline_deleted";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_GEOMETRY: &str = "data_read_inline_geometry";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_IDENTITY_MISMATCH: &str = "data_read_inline_identity_mismatch";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_PAYLOAD: &str = "data_read_inline_missing_payload";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_SHARD: &str = "data_read_inline_missing_shard";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_NOT_INLINE: &str = "data_read_inline_not_inline";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_PART_SHAPE: &str = "data_read_inline_part_shape";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_REMOTE: &str = "data_read_inline_remote";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_SIZE: &str = "data_read_inline_size";
+pub(crate) const GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_TRANSFORMED: &str = "data_read_inline_transformed";
 pub(crate) const GET_METADATA_EARLY_STOP_REASON_ERROR: &str = "error";
 pub(crate) const GET_METADATA_EARLY_STOP_REASON_INSUFFICIENT_QUORUM: &str = "insufficient_quorum";
 pub(crate) const GET_METADATA_EARLY_STOP_REASON_NOT_FOUND: &str = "not_found";
@@ -551,6 +562,32 @@ mod tests {
        assert_eq!(GET_METADATA_CACHE_REASON_VERSIONED, "versioned");
        assert_eq!(GET_METADATA_EARLY_STOP_REASON_CONFLICTING_METADATA, "conflicting_metadata");
        assert_eq!(GET_METADATA_EARLY_STOP_REASON_DELETE_MARKER, "delete_marker");
+        assert_eq!(
+            GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_BODY_VERIFY,
+            "data_read_inline_body_verify"
+        );
+        assert_eq!(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_DELETED, "data_read_inline_deleted");
+        assert_eq!(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_GEOMETRY, "data_read_inline_geometry");
+        assert_eq!(
+            GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_IDENTITY_MISMATCH,
+            "data_read_inline_identity_mismatch"
+        );
+        assert_eq!(
+            GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_PAYLOAD,
+            "data_read_inline_missing_payload"
+        );
+        assert_eq!(
+            GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_SHARD,
+            "data_read_inline_missing_shard"
+        );
+        assert_eq!(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_NOT_INLINE, "data_read_inline_not_inline");
+        assert_eq!(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_PART_SHAPE, "data_read_inline_part_shape");
+        assert_eq!(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_REMOTE, "data_read_inline_remote");
+        assert_eq!(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_SIZE, "data_read_inline_size");
+        assert_eq!(
+            GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_TRANSFORMED,
+            "data_read_inline_transformed"
+        );
        assert_eq!(GET_METADATA_EARLY_STOP_REASON_ERROR, "error");
        assert_eq!(GET_METADATA_EARLY_STOP_REASON_INSUFFICIENT_QUORUM, "insufficient_quorum");
        assert_eq!(GET_METADATA_EARLY_STOP_REASON_NOT_FOUND, "not_found");
@@ -241,6 +241,40 @@ pub fn get_drive_list_dir_timeout() -> Duration {
    )
 }

+pub(crate) trait DiskStoreRenameDataExt {
+    async fn rename_data_borrowed(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+    ) -> Result<RenameDataResp>;
+}
+
+impl DiskStoreRenameDataExt for LocalDiskWrapper {
+    async fn rename_data_borrowed(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+    ) -> Result<RenameDataResp> {
+        self.track_disk_health_mutation(
+            "rename_data",
+            DiskMetricMutation::Write,
+            || async {
+                self.disk
+                    .rename_data_borrowed(src_volume, src_path, fi, dst_volume, dst_path)
+                    .await
+            },
+            get_max_timeout_duration(),
+        )
+        .await
+    }
+}
+
 pub fn get_drive_walkdir_timeout() -> Duration {
    get_drive_timeout_duration(
        rustfs_config::ENV_DRIVE_WALKDIR_TIMEOUT_SECS,
@@ -603,14 +637,23 @@ impl Default for DiskOperationMetrics {
 }

 impl DiskOperationMetrics {
+    #[allow(
+        dead_code,
+        reason = "internal metrics recorder reached only from record() below (backlog#1823)"
+    )]
    fn record_call(&mut self) {
        self.lifetime_calls.fetch_add(1, Ordering::Relaxed);
    }

+    #[allow(
+        dead_code,
+        reason = "internal metrics recorder reached only from record() below (backlog#1823)"
+    )]
    fn record_latency(&mut self, now_sec: u64, elapsed: Duration) {
        self.record_latency_atomic(now_sec, elapsed);
    }

+    #[allow(dead_code, reason = "metrics roll-up with no caller in this port (backlog#1823)")]
    fn record(&mut self, now_sec: u64, elapsed: Duration) {
        self.record_call();
        self.record_latency(now_sec, elapsed);
@@ -736,6 +779,7 @@ impl DiskHealthTracker {
    }

    /// Set disk as faulty
+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    pub fn set_faulty(&self) {
        self.status.store(DISK_HEALTH_FAULTY, Ordering::Release);
    }
@@ -816,6 +860,7 @@ impl DiskHealthTracker {
        became_offline
    }

+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    pub fn mark_offline(&self, endpoint: &Endpoint, reason: &'static str) -> bool {
        let current = self.runtime_state();
        if current == RuntimeDriveHealthState::Offline {
@@ -946,11 +991,13 @@ impl DiskHealthTracker {
    }

    /// Get waiting operations count
+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    pub fn waiting_count(&self) -> u32 {
        self.waiting.load(Ordering::Relaxed)
    }

    /// Get last success timestamp
+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    pub fn last_success(&self) -> i64 {
        self.last_success.load(Ordering::Acquire)
    }
@@ -992,21 +1039,6 @@ impl Default for DiskHealthTracker {
    }
 }

-/// Health check context key for tracking disk operations
-#[derive(Debug, Clone)]
-struct HealthDiskCtxKey;
-
-#[derive(Debug)]
-struct HealthDiskCtxValue {
-    last_success: Arc<AtomicI64>,
-}
-
-impl HealthDiskCtxValue {
-    fn log_success(&self) {
-        self.last_success.store(current_unix_nanos(), Ordering::Relaxed);
-    }
-}
-
 /// LocalDiskWrapper wraps a DiskStore with health tracking capabilities.
 /// This is similar to Go's xlStorageDiskIDCheck.
 #[derive(Debug, Clone)]
@@ -1038,10 +1070,6 @@ impl LocalDiskWrapper {
        )
    }

-    pub(crate) fn new_with_health(disk: Arc<LocalDisk>, health_check: bool, health: Arc<DiskHealthTracker>) -> Self {
-        Self::new_with_health_and_metrics(disk, health_check, health, Arc::new(DiskHealthMetricEpoch::default()))
-    }
-
    pub(crate) fn new_with_reconnect_state(
        disk: Arc<LocalDisk>,
        health_check: bool,
@@ -1404,20 +1432,6 @@ impl LocalDiskWrapper {
        }
    }

-    async fn check_id(&self, want_id: Option<Uuid>) -> Result<()> {
-        if want_id.is_none() {
-            return Ok(());
-        }
-
-        let stored_disk_id = self.disk.get_disk_id().await?;
-
-        if stored_disk_id != want_id {
-            return Err(Error::other(format!("Disk ID mismatch wanted {want_id:?}, got {stored_disk_id:?}")));
-        }
-
-        Ok(())
-    }
-
    /// Check if disk ID is stale
    async fn check_disk_stale(&self) -> Result<()> {
        let Some(current_disk_id) = *self.disk_id.read().await else {
@@ -1985,13 +1999,8 @@ impl DiskAPI for LocalDiskWrapper {
        dst_volume: &str,
        dst_path: &str,
    ) -> Result<RenameDataResp> {
-        self.track_disk_health_mutation(
-            "rename_data",
-            DiskMetricMutation::Write,
-            || async { self.disk.rename_data(src_volume, src_path, fi, dst_volume, dst_path).await },
-            get_max_timeout_duration(),
-        )
-        .await
+        self.rename_data_borrowed(src_volume, src_path, &fi, dst_volume, dst_path)
+            .await
    }

    async fn list_dir(&self, origvolume: &str, volume: &str, dir_path: &str, count: i32) -> Result<Vec<String>> {
@@ -48,6 +48,7 @@ pub fn to_volume_error(io_err: std::io::Error) -> std::io::Error {
    }
 }

+#[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
 pub fn to_disk_error(io_err: std::io::Error) -> std::io::Error {
    match io_err.kind() {
        std::io::ErrorKind::NotFound => DiskError::DiskNotFound.into(),
@@ -178,6 +178,7 @@ pub async fn remove(path: impl AsRef<Path>) -> io::Result<()> {
    }
 }

+#[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
 pub async fn remove_all(path: impl AsRef<Path>) -> io::Result<()> {
    // Try remove_file first; fall back to remove_dir_all if it's a directory
    match fs::remove_file(path.as_ref()).await {
@@ -27,17 +27,18 @@ use crate::disk::{
    BUCKET_META_PREFIX, CHECK_PART_FILE_CORRUPT, CHECK_PART_FILE_NOT_FOUND, CHECK_PART_SUCCESS, CHECK_PART_UNKNOWN,
    CHECK_PART_VOLUME_NOT_FOUND, CheckPartsResp, ConditionalFileUpdate, DataDirDeleteStatus, DeleteOptions, DiskAPI, DiskInfo,
    DiskInfoOptions, DiskLocation, DiskMetrics, FileInfoVersions, FileReader, FileWriter, MmapCopyStageMetrics, OldCurrentSize,
-    PART_TRANSACTION_NEW_META, PART_TRANSACTION_OLD_META, PART_TRANSACTION_ROLLBACK, PartTransactionAction, RUSTFS_META_BUCKET,
-    RUSTFS_META_TMP_BUCKET, RUSTFS_META_TMP_DELETED_BUCKET, ReadMultipleReq, ReadMultipleResp, ReadOptions, RenameDataResp,
-    STORAGE_FORMAT_FILE, STORAGE_FORMAT_FILE_BACKUP, SnapshotLeaseToken, UpdateMetadataOpts, VolumeInfo, WalkDirOptions,
-    conv_part_err_to_int,
+    PART_TRANSACTION_NEW_META, PART_TRANSACTION_OLD_META, PART_TRANSACTION_ROLLBACK, PartTransactionAction,
+    QUOTA_MUTATION_FENCE_METADATA_SUFFIX, RUSTFS_META_BUCKET, RUSTFS_META_TMP_BUCKET, RUSTFS_META_TMP_DELETED_BUCKET,
+    ReadMultipleReq, ReadMultipleResp, ReadOptions, RenameDataResp, STORAGE_FORMAT_FILE, STORAGE_FORMAT_FILE_BACKUP,
+    SnapshotLeaseToken, UpdateMetadataOpts, VolumeInfo, WalkDirOptions, conv_part_err_to_int,
    endpoint::Endpoint,
    error::{DiskError, Error, FileAccessDeniedWithContext, Result},
    error_conv::{to_access_error, to_file_error, to_unformatted_disk_error, to_volume_error},
    format::FormatV3,
    fs::{O_APPEND, O_CREATE, O_RDONLY, O_TRUNC, O_WRONLY, access, lstat, lstat_std, remove, remove_all_std, remove_std, rename},
-    os,
+    is_quota_mutation_fence_path, os,
    os::{check_path_length, is_dir_not_empty_error, is_empty_dir, is_root_disk, rename_all, rename_all_ignore_missing_source},
+    quota_mutation_fence_path,
 };
 use crate::erasure::coding::{self, bitrot_verify};
 use crate::runtime::sources as runtime_sources;
@@ -60,9 +61,7 @@ use std::collections::HashMap;
 use std::collections::HashSet;
 use std::fmt::Debug;
 use std::io::{Error as IoError, SeekFrom};
-#[cfg(target_os = "linux")]
-use std::sync::atomic::AtomicBool;
-use std::sync::atomic::{AtomicU32, Ordering};
+use std::sync::atomic::{AtomicBool, AtomicU32, AtomicUsize, Ordering};
 use std::sync::{Arc, OnceLock};
 use std::time::Duration;
 use std::{
@@ -666,6 +665,7 @@ async fn remove_empty_directory_tree_under_mount_lease(
 }

 #[cfg(unix)]
+#[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
 async fn remove_empty_directory_tree_with(
    root: &Path,
    before_descend: impl FnMut(&Path) -> std::io::Result<()>,
@@ -1017,13 +1017,29 @@ fn record_direct_read_page_fault_delta(path: &'static str, stage: &'static str,
 /// When enabled, shard reads bypass the page cache using O_DIRECT flag.
 /// Requires aligned buffers (typically 512 bytes or 4096 bytes).
 /// Default: false (uses page cache via mmap/pread).
+#[allow(
+    dead_code,
+    reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)"
+)]
 const ENV_RUSTFS_OBJECT_DIRECT_IO_READ_ENABLE: &str = "RUSTFS_OBJECT_DIRECT_IO_READ_ENABLE";
+#[allow(
+    dead_code,
+    reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)"
+)]
 const DEFAULT_RUSTFS_OBJECT_DIRECT_IO_READ_ENABLE: bool = false;

 /// Minimum shard size threshold for O_DIRECT reads.
 /// Only shards larger than this threshold will use O_DIRECT.
 /// Default: 4MB.
+#[allow(
+    dead_code,
+    reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)"
+)]
 const ENV_RUSTFS_OBJECT_DIRECT_IO_READ_THRESHOLD: &str = "RUSTFS_OBJECT_DIRECT_IO_READ_THRESHOLD";
+#[allow(
+    dead_code,
+    reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)"
+)]
 const DEFAULT_RUSTFS_OBJECT_DIRECT_IO_READ_THRESHOLD: usize = 4 * 1024 * 1024;

 /// Enable O_DIRECT for erasure shard / multipart part data writes (Linux only).
@@ -1037,7 +1053,15 @@ const DEFAULT_RUSTFS_OBJECT_DIRECT_IO_READ_THRESHOLD: usize = 4 * 1024 * 1024;
 /// EINVAL/EOPNOTSUPP (tmpfs, overlayfs, 9p, ...) latch the path off and fall
 /// back to buffered writes for the whole disk. Non-Linux always falls back.
 /// Default: false (buffered writes via the page cache, as before).
+#[allow(
+    dead_code,
+    reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)"
+)]
 const ENV_RUSTFS_OBJECT_DIRECT_IO_WRITE_ENABLE: &str = "RUSTFS_OBJECT_DIRECT_IO_WRITE_ENABLE";
+#[allow(
+    dead_code,
+    reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)"
+)]
 const DEFAULT_RUSTFS_OBJECT_DIRECT_IO_WRITE_ENABLE: bool = false;
 const ENV_RUSTFS_OBJECT_MMAP_POPULATE_ENABLE: &str = "RUSTFS_OBJECT_MMAP_POPULATE_ENABLE";
 const DEFAULT_RUSTFS_OBJECT_MMAP_POPULATE_ENABLE: bool = false;
@@ -1096,12 +1120,14 @@ macro_rules! cached_read_env {

 cached_read_env! {
    /// Check if O_DIRECT reads are enabled.
+    #[allow(dead_code, reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)")]
    fn is_direct_io_read_enabled() -> bool =
        rustfs_utils::get_env_bool(ENV_RUSTFS_OBJECT_DIRECT_IO_READ_ENABLE, DEFAULT_RUSTFS_OBJECT_DIRECT_IO_READ_ENABLE);
 }

 cached_read_env! {
    /// Check if O_DIRECT shard/part data writes are enabled.
+    #[allow(dead_code, reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)")]
    fn is_direct_io_write_enabled() -> bool =
        rustfs_utils::get_env_bool(ENV_RUSTFS_OBJECT_DIRECT_IO_WRITE_ENABLE, DEFAULT_RUSTFS_OBJECT_DIRECT_IO_WRITE_ENABLE);
 }
@@ -1457,6 +1483,7 @@ pub(crate) fn effective_durability(volume: &str) -> DurabilityMode {

 cached_read_env! {
    /// Get the O_DIRECT read threshold size.
+    #[allow(dead_code, reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)")]
    fn get_direct_io_read_threshold() -> usize =
        rustfs_utils::get_env_usize(ENV_RUSTFS_OBJECT_DIRECT_IO_READ_THRESHOLD, DEFAULT_RUSTFS_OBJECT_DIRECT_IO_READ_THRESHOLD);
 }
@@ -1674,12 +1701,20 @@ impl DirectIoWriteState {
 /// Target staging size for O_DIRECT writes, rounded up to the DIO alignment.
 /// Bounds the per-writer aligned bounce buffer and batches many shard blocks
 /// into one positioned write to keep the syscall count low.
+#[allow(
+    dead_code,
+    reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)"
+)]
 const DIRECT_WRITE_STAGING_BYTES: usize = 1024 * 1024;

 /// Aligned bounce-buffer capacity for a given DIO alignment: the target staging
 /// size rounded up to a whole multiple of `align` so the buffer address, every
 /// flushed batch length, and every write offset stay alignment-correct.
 /// Platform-independent (no O_DIRECT), so it is unit-tested on any host.
+#[allow(
+    dead_code,
+    reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)"
+)]
 fn direct_write_staging_capacity(align: usize) -> usize {
    debug_assert!(align.is_power_of_two() && align >= 512);
    DIRECT_WRITE_STAGING_BYTES.div_ceil(align) * align
@@ -1688,6 +1723,10 @@ fn direct_write_staging_capacity(align: usize) -> usize {
 /// Split `filled` staged bytes into the alignment-sized prefix written with
 /// O_DIRECT and the sub-alignment tail written buffered. Platform-independent,
 /// so the tail-boundary math is unit-tested on any host.
+#[allow(
+    dead_code,
+    reason = "platform-conditional: production callers are inside #[cfg(target_os = \"linux\")] blocks, so this reads as dead on non-Linux hosts (backlog#1823)"
+)]
 fn direct_write_tail_split(filled: usize, align: usize) -> (usize, usize) {
    let aligned = filled - (filled % align);
    (aligned, filled - aligned)
@@ -2029,14 +2068,17 @@ static RENAME_DATA_REMOVE_DST_BASE_BEFORE_COMMIT: std::sync::Mutex<Option<(Strin
 #[cfg(test)]
 type InlinePreparationHook = Box<dyn FnOnce() + Send>;
 #[cfg(test)]
+type RenameDataPublicationHookKey = (PathBuf, String, String);
+#[cfg(test)]
 static INLINE_PREPARATION_BEFORE_BACKUP: std::sync::LazyLock<std::sync::Mutex<HashMap<String, InlinePreparationHook>>> =
    std::sync::LazyLock::new(|| std::sync::Mutex::new(HashMap::new()));
 #[cfg(test)]
 static INLINE_BEFORE_FILE_SYNC_ADMISSION: std::sync::LazyLock<std::sync::Mutex<HashMap<String, InlinePreparationHook>>> =
    std::sync::LazyLock::new(|| std::sync::Mutex::new(HashMap::new()));
 #[cfg(test)]
-static RENAME_DATA_AFTER_FIRST_PUBLICATION: std::sync::LazyLock<std::sync::Mutex<HashMap<String, InlinePreparationHook>>> =
-    std::sync::LazyLock::new(|| std::sync::Mutex::new(HashMap::new()));
+static RENAME_DATA_AFTER_FIRST_PUBLICATION: std::sync::LazyLock<
+    std::sync::Mutex<HashMap<RenameDataPublicationHookKey, InlinePreparationHook>>,
+> = std::sync::LazyLock::new(|| std::sync::Mutex::new(HashMap::new()));
 #[cfg(test)]
 static OWNED_FILE_WRITE_BEFORE_OPEN: std::sync::LazyLock<std::sync::Mutex<HashMap<PathBuf, InlinePreparationHook>>> =
    std::sync::LazyLock::new(|| std::sync::Mutex::new(HashMap::new()));
@@ -2108,11 +2150,11 @@ fn set_inline_before_file_sync_admission(dst_path: &str, hook: impl FnOnce() + S
 }

 #[cfg(test)]
-fn set_rename_data_after_first_publication(dst_path: &str, hook: impl FnOnce() + Send + 'static) {
+fn set_rename_data_after_first_publication(root: &Path, dst_volume: &str, dst_path: &str, hook: impl FnOnce() + Send + 'static) {
    RENAME_DATA_AFTER_FIRST_PUBLICATION
        .lock()
        .expect("test publication hook lock should not be poisoned")
-        .insert(dst_path.to_string(), Box::new(hook));
+        .insert((root.to_path_buf(), dst_volume.to_string(), dst_path.to_string()), Box::new(hook));
 }

 #[cfg(test)]
@@ -2140,6 +2182,7 @@ fn set_delete_version_fail_after_data_staged(path: &str) {
 }

 #[cfg(test)]
+#[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
 pub(crate) fn set_delete_version_fail_after_commit(root: &Path, path: &str) {
    DELETE_VERSION_FAIL_AFTER_COMMIT
        .lock()
@@ -2264,11 +2307,11 @@ fn run_inline_before_file_sync_admission(dst_path: &str) {
 }

 #[cfg(test)]
-fn run_rename_data_after_first_publication(dst_path: &str) {
+fn run_rename_data_after_first_publication(root: &Path, dst_volume: &str, dst_path: &str) {
    let hook = RENAME_DATA_AFTER_FIRST_PUBLICATION
        .lock()
        .expect("test publication hook lock should not be poisoned")
-        .remove(dst_path);
+        .remove(&(root.to_path_buf(), dst_volume.to_string(), dst_path.to_string()));
    if let Some(hook) = hook {
        hook();
    }
@@ -2366,9 +2409,6 @@ async fn remove_dst_base_before_commit(
 #[cfg(not(test))]
 fn run_inline_preparation_before_backup(_dst_path: &str) {}

-#[cfg(not(test))]
-fn run_rename_data_after_first_publication(_dst_path: &str) {}
-
 #[cfg(not(test))]
 fn should_fail_after_delete_data_staged(_path: &str) -> bool {
    false
@@ -2448,6 +2488,10 @@ enum SyncMode {
    FileOnly,
 }

+#[allow(
+    dead_code,
+    reason = "reclaim bookkeeping fields written by Drop but never read back (backlog#1823)"
+)]
 struct FileCacheReclaimWriter {
    inner: File,
    reclaim_len: usize,
@@ -2455,6 +2499,10 @@ struct FileCacheReclaimWriter {
    reclaimed: bool,
 }

+#[allow(
+    dead_code,
+    reason = "reclaim bookkeeping fields written by Drop but never read back (backlog#1823)"
+)]
 struct FileCacheReclaimReader {
    inner: File,
    reclaim_offset: u64,
@@ -2520,6 +2568,10 @@ impl<R: AsyncRead + Unpin> AsyncRead for StallTimeoutReader<R> {
    }
 }

+#[allow(
+    dead_code,
+    reason = "reclaim metrics emitter reached only from the Linux-gated reclaim paths (backlog#1823)"
+)]
 fn record_file_cache_reclaim_success(kind: &'static str, reclaim_len: usize, started: std::time::Instant) {
    // Runs per read-stream page-cache reclaim window; skip the whole emission
    // (three metric-key constructions) when general metrics are disabled.
@@ -3072,6 +3124,7 @@ impl LocalIoBackend for StdBackend {
            use memmap2::MmapOptions;
            use std::time::{Duration as StdDuration, Instant as StdInstant};

+            #[allow(dead_code, reason = "mmap copy result slot kept beside the mapping it owns (backlog#1823)")]
            struct MmapCopyReadResult {
                bytes: Bytes,
                access_check_duration: StdDuration,
@@ -4705,6 +4758,10 @@ fn build_local_io_backend(root: PathBuf) -> Arc<dyn LocalIoBackend> {
    Arc::new(StdBackend::new(root))
 }

+#[allow(
+    dead_code,
+    reason = "path cache and cwd slots retained beside the disk root they derive from (backlog#1823)"
+)]
 pub struct LocalDisk {
    pub root: PathBuf,
    publication_root: os::PublicationRoot,
@@ -4756,6 +4813,25 @@ struct SnapshotLeaseEntry {
    tokens: HashSet<SnapshotLeaseToken>,
    pending_delete: Option<DeleteOptions>,
    deleting: bool,
+    mutation_fence: Option<Arc<QuotaMutationFenceState>>,
+}
+
+#[derive(Default)]
+struct QuotaMutationFenceState {
+    revoked: AtomicBool,
+    running: AtomicUsize,
+    notify: Notify,
+}
+
+struct QuotaMutationFenceClaim {
+    state: Arc<QuotaMutationFenceState>,
+}
+
+impl Drop for QuotaMutationFenceClaim {
+    fn drop(&mut self) {
+        self.state.running.fetch_sub(1, Ordering::AcqRel);
+        self.state.notify.notify_waiters();
+    }
 }

 #[derive(Default)]
@@ -5472,6 +5548,7 @@ impl LocalDisk {
        Ok(Self::resolve_abs_path_from(&self.root, path.as_ref()))
    }

+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    fn io_resolve_abs_path(&self, path: impl AsRef<Path>) -> PathBuf {
        let path_ref = path.as_ref();
        let path_str = path_ref.to_string_lossy();
@@ -5549,15 +5626,18 @@ impl LocalDisk {
    }

    // Check if a path is valid
+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    fn check_valid_path<P: AsRef<Path>>(&self, path: P) -> Result<()> {
        check_local_disk_valid_path(self.io_root(), path)
    }

+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    fn reject_symlink_components(&self, path: &Path) -> Result<()> {
        reject_local_disk_symlink_components(self.io_root(), path)
    }

    // Batch path generation with single lock acquisition
+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    fn get_object_paths_batch(&self, requests: &[(String, String)]) -> Result<Vec<PathBuf>> {
        let mut results = Vec::with_capacity(requests.len());
        let mut cache_misses = Vec::new();
@@ -6470,6 +6550,7 @@ impl LocalDisk {
        Ok(f)
    }

+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    async fn open_file_read_only(&self, path: impl AsRef<Path>) -> Result<File> {
        let f = super::fs::open_file(path.as_ref(), O_RDONLY).await.map_err(to_file_error)?;
        Ok(f)
@@ -7320,6 +7401,34 @@ fn normalize_path_components(path: impl AsRef<Path>) -> PathBuf {
 }

 impl LocalDisk {
+    async fn claim_quota_mutation_fence(
+        &self,
+        volume: &str,
+        path: &str,
+        token: SnapshotLeaseToken,
+    ) -> Result<Arc<QuotaMutationFenceClaim>> {
+        let key = SnapshotLeaseKey {
+            volume: RUSTFS_META_BUCKET.to_string(),
+            path: quota_mutation_fence_path(volume, path),
+        };
+        let state = {
+            let registry = self.snapshot_leases.lock().await;
+            let entry = registry.entries.get(&key).ok_or(DiskError::FileNotFound)?;
+            let state = entry.mutation_fence.as_ref().ok_or(DiskError::FileNotFound)?;
+            if !entry.tokens.contains(&token) || state.revoked.load(Ordering::Acquire) {
+                return Err(DiskError::FileNotFound);
+            }
+            state.running.fetch_add(1, Ordering::AcqRel);
+            Arc::clone(state)
+        };
+        if state.revoked.load(Ordering::Acquire) {
+            state.running.fetch_sub(1, Ordering::AcqRel);
+            state.notify.notify_waiters();
+            return Err(DiskError::FileNotFound);
+        }
+        Ok(Arc::new(QuotaMutationFenceClaim { state }))
+    }
+
    async fn reserve_version_delete(&self, volume: &str, object: &str, data_dir: Uuid, rollback_dir: Uuid) -> Result<bool> {
        let path = format!("{object}/{data_dir}");
        let data_path = self.io_get_object_path(volume, &path)?;
@@ -8643,17 +8752,41 @@ impl DiskAPI for LocalDisk {
        &self,
        src_volume: &str,
        src_path: &str,
-        mut fi: FileInfo,
+        fi: FileInfo,
        dst_volume: &str,
        dst_path: &str,
    ) -> Result<RenameDataResp> {
        crate::hp_guard!("LocalDisk::rename_data");
+        let mut fi = fi;
        // A non-force DeleteBucket must not remove a directory while a local
        // object commit is publishing into it. The peer's empty scan remains
        // optimistic; this lease establishes the local commit/delete order and
        // remains owned by any blocking syscall that outlives async cancellation.
        let destination_object_path = self.io_get_object_path(dst_volume, dst_path)?;
+        let quota_fence_token =
+            match rustfs_utils::http::metadata_compat::get_consistent_str(&fi.metadata, QUOTA_MUTATION_FENCE_METADATA_SUFFIX) {
+                Some(value) => {
+                    let token = Uuid::parse_str(value).map_err(|_| DiskError::FileCorrupt)?;
+                    Some(SnapshotLeaseToken::from_slice(token.as_bytes())?)
+                }
+                None if rustfs_utils::http::metadata_compat::contains_key_str(
+                    &fi.metadata,
+                    QUOTA_MUTATION_FENCE_METADATA_SUFFIX,
+                ) =>
+                {
+                    return Err(DiskError::FileCorrupt);
+                }
+                None => None,
+            };
+        rustfs_utils::http::metadata_compat::remove_str(&mut fi.metadata, QUOTA_MUTATION_FENCE_METADATA_SUFFIX);
+        let quota_fence_claim = match quota_fence_token {
+            Some(token) => Some(self.claim_quota_mutation_fence(dst_volume, dst_path, token).await?),
+            None => None,
+        };
        let mutation_lease = os::acquire_rename_data_mutation_lease(&self.root, dst_volume, &destination_object_path).await;
+        if let Some(claim) = quota_fence_claim {
+            mutation_lease.attach_external_guard(claim);
+        }
        if fi.is_legacy_indexed_delete_marker() {
            fi.erasure.index = 0;
        }
@@ -8946,8 +9079,9 @@ impl DiskAPI for LocalDisk {
                .await?;
                return Err(err);
            }
+            #[cfg(test)]
            if has_data_dir_path.is_some() {
-                run_rename_data_after_first_publication(dst_path);
+                run_rename_data_after_first_publication(&self.root, dst_volume, dst_path);
            }

            // Crash-consistency injection: hard power loss after the data dir
@@ -9380,7 +9514,8 @@ impl DiskAPI for LocalDisk {
                    let _ = remove_file_if_exists(staged_backup);
                    return Err(err);
                }
-                run_rename_data_after_first_publication(dst_path);
+                #[cfg(test)]
+                run_rename_data_after_first_publication(&self.root, dst_volume, dst_path);
                if sync {
                    file_sync_admission = Some(
                        os::acquire_file_sync_admission(self.file_sync_permits.clone())
@@ -9643,11 +9778,26 @@ impl DiskAPI for LocalDisk {
    }

    async fn acquire_snapshot_lease(&self, volume: &str, path: &str) -> Result<SnapshotLeaseToken> {
-        let file_path = self.io_get_object_path(volume, path)?;
        let key = SnapshotLeaseKey {
            volume: volume.to_string(),
            path: path.to_string(),
        };
+        if volume == RUSTFS_META_BUCKET && is_quota_mutation_fence_path(path) {
+            let mut registry = self.snapshot_leases.lock().await;
+            let entry = registry.entries.entry(key).or_default();
+            let state = entry
+                .mutation_fence
+                .get_or_insert_with(|| Arc::new(QuotaMutationFenceState::default()));
+            if state.revoked.load(Ordering::Acquire) {
+                return Err(DiskError::FileNotFound);
+            }
+            let token = SnapshotLeaseToken::new();
+            entry.tokens.insert(token);
+            return Ok(token);
+        }
+
+        let file_path = self.io_get_object_path(volume, path)?;
+        let _mutation_lease = os::acquire_rename_data_mutation_lease(&self.root, volume, &file_path).await;
        let token = {
            let mut registry = self.snapshot_leases.lock().await;
            if registry.entries.get(&key).is_some_and(|entry| entry.deleting) {
@@ -9675,6 +9825,48 @@ impl DiskAPI for LocalDisk {
            volume: volume.to_string(),
            path: path.to_string(),
        };
+        if volume == RUSTFS_META_BUCKET && is_quota_mutation_fence_path(path) {
+            if !token.is_revoke_all() {
+                let mut registry = self.snapshot_leases.lock().await;
+                let Some(entry) = registry.entries.get_mut(&key) else {
+                    return Ok(());
+                };
+                entry.tokens.remove(&token);
+                let removable = entry.tokens.is_empty()
+                    && entry
+                        .mutation_fence
+                        .as_ref()
+                        .is_none_or(|state| state.running.load(Ordering::Acquire) == 0);
+                if removable {
+                    registry.entries.remove(&key);
+                }
+                return Ok(());
+            }
+            let state = {
+                let mut registry = self.snapshot_leases.lock().await;
+                let Some(entry) = registry.entries.get_mut(&key) else {
+                    return Ok(());
+                };
+                let Some(state) = entry.mutation_fence.as_ref().cloned() else {
+                    registry.entries.remove(&key);
+                    return Ok(());
+                };
+                state.revoked.store(true, Ordering::Release);
+                entry.tokens.clear();
+                state
+            };
+            loop {
+                let notified = state.notify.notified();
+                tokio::pin!(notified);
+                notified.as_mut().enable();
+                if state.running.load(Ordering::Acquire) == 0 {
+                    break;
+                }
+                notified.await;
+            }
+            self.snapshot_leases.lock().await.entries.remove(&key);
+            return Ok(());
+        }
        let opts = {
            let mut registry = self.snapshot_leases.lock().await;
            let Some(entry) = registry.entries.get_mut(&key) else {
@@ -10453,6 +10645,19 @@ impl DiskAPI for LocalDisk {
    }
 }

+impl LocalDisk {
+    pub(crate) async fn rename_data_borrowed(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+    ) -> Result<RenameDataResp> {
+        <Self as DiskAPI>::rename_data(self, src_volume, src_path, fi.clone(), dst_volume, dst_path).await
+    }
+}
+
 async fn wait_for_startup_cleanup_signal(
    startup_cleanup_ready: &AtomicU32,
    startup_cleanup_notify: &Notify,
@@ -13007,7 +13212,7 @@ mod test {
        let replacement_staging_parent_for_hook = replacement_staging_parent.clone();
        let staged_metadata_for_hook = staged_metadata.clone();
        let replacement_staged_metadata_for_hook = replacement_staged_metadata.clone();
-        set_rename_data_after_first_publication(object, move || {
+        set_rename_data_after_first_publication(&disk.root, bucket, object, move || {
            std::fs::rename(&object_dir_for_hook, &replacement_dir_for_hook)
                .expect_err("the destination object identity must remain pinned until xl.meta commits");
            std::fs::rename(&staging_parent_for_hook, &replacement_staging_parent_for_hook)
@@ -13274,7 +13479,7 @@ mod test {
        let replacement_dir_for_hook = replacement_dir.clone();
        let staged_metadata_for_hook = staged_metadata.clone();
        let replacement_staged_metadata_for_hook = replacement_staged_metadata.clone();
-        set_rename_data_after_first_publication(object, move || {
+        set_rename_data_after_first_publication(&disk.root, bucket, object, move || {
            std::fs::rename(&object_dir_for_hook, &replacement_dir_for_hook)
                .expect_err("the destination object identity must remain pinned after publishing its rollback backup");
            std::fs::rename(&staged_metadata_for_hook, &replacement_staged_metadata_for_hook)
@@ -13640,7 +13845,7 @@ mod test {

        let (entered_tx, entered_rx) = mpsc::channel();
        let (release_tx, release_rx) = mpsc::channel();
-        set_rename_data_after_first_publication(object, move || {
+        set_rename_data_after_first_publication(&disk.root, bucket, object, move || {
            entered_tx.send(()).expect("signal first publication");
            release_rx.recv().expect("wait while delete_volume is blocked");
        });
@@ -14234,7 +14439,7 @@ mod test {

        let (published_tx, published_rx) = mpsc::channel();
        let (release_tx, release_rx) = mpsc::channel();
-        set_rename_data_after_first_publication(object, move || {
+        set_rename_data_after_first_publication(&disk.root, bucket, object, move || {
            published_tx.send(()).expect("signal backup publication");
            release_rx.recv().expect("wait for lock-order assertion");
        });
@@ -18889,6 +19094,48 @@ mod test {
        assert!(matches!(disk.read_all(volume, &first_part).await, Err(DiskError::FileNotFound)));
    }

+    #[tokio::test]
+    async fn quota_mutation_fence_revoke_waits_for_active_claim_and_rejects_late_claims() {
+        use tempfile::tempdir;
+
+        let root_dir = tempdir().expect("temp dir should be created");
+        let endpoint = Endpoint::try_from(root_dir.path().to_string_lossy().as_ref()).expect("endpoint should parse");
+        let disk = Arc::new(LocalDisk::new(&endpoint, false).await.expect("local disk should be created"));
+        let bucket = "quota-fence-volume";
+        let object = "object";
+        let fence_path = quota_mutation_fence_path(bucket, object);
+        let token = disk
+            .acquire_snapshot_lease(RUSTFS_META_BUCKET, &fence_path)
+            .await
+            .expect("quota mutation token should be prepared");
+        let claim = disk
+            .claim_quota_mutation_fence(bucket, object, token)
+            .await
+            .expect("prepared token should be claimable");
+
+        let release_disk = Arc::clone(&disk);
+        let mut release = tokio::spawn(async move {
+            release_disk
+                .release_snapshot_lease(RUSTFS_META_BUCKET, &fence_path, SnapshotLeaseToken::revoke_all())
+                .await
+        });
+        assert!(
+            tokio::time::timeout(Duration::from_millis(50), &mut release).await.is_err(),
+            "revoke must wait until an already claimed mutation has finished"
+        );
+
+        drop(claim);
+        tokio::time::timeout(Duration::from_secs(1), release)
+            .await
+            .expect("revoke should wake after the final claim drops")
+            .expect("revoke task should not panic")
+            .expect("revoke should succeed");
+        assert!(matches!(
+            disk.claim_quota_mutation_fence(bucket, object, token).await,
+            Err(DiskError::FileNotFound)
+        ));
+    }
+
    #[tokio::test]
    async fn delete_version_keeps_later_part_until_snapshot_release() {
        use tempfile::tempdir;
@@ -13,7 +13,6 @@
 // limitations under the License.

 // #730: disk abstractions still carry staged health and direct-I/O migration paths.
-#![allow(dead_code)]

 pub mod disk_store;
 pub mod endpoint;
@@ -55,6 +54,7 @@ pub fn part_transaction_path(part_path: &str) -> String {

 use crate::cluster::rpc::RemoteDisk;
 use crate::cluster::rpc::build_internode_data_transport_from_env;
+use crate::disk::disk_store::DiskStoreRenameDataExt;
 use crate::disk::disk_store::LocalDiskWrapper;
 use crate::disk::health_state::RuntimeDriveHealthState;
 use crate::disk::local::ScanGuard;
@@ -72,6 +72,28 @@ use time::OffsetDateTime;
 use tokio::io::{AsyncRead, AsyncWrite};
 use uuid::Uuid;

+const QUOTA_MUTATION_FENCE_PREFIX: &str = "tmp/quota-mutation-fences/";
+pub(crate) const QUOTA_MUTATION_FENCE_METADATA_SUFFIX: &str = "quota-mutation-fence-token";
+
+pub(crate) fn quota_mutation_fence_path(bucket: &str, object: &str) -> String {
+    use sha2::{Digest, Sha256};
+
+    let mut input = Vec::with_capacity(bucket.len() + object.len() + 1);
+    input.extend_from_slice(bucket.as_bytes());
+    input.push(0);
+    input.extend_from_slice(object.as_bytes());
+    let digest = Sha256::digest(input);
+    format!(
+        "{QUOTA_MUTATION_FENCE_PREFIX}{}",
+        hex_simd::encode_to_string(digest, hex_simd::AsciiCase::Lower)
+    )
+}
+
+pub(crate) fn is_quota_mutation_fence_path(path: &str) -> bool {
+    path.strip_prefix(QUOTA_MUTATION_FENCE_PREFIX)
+        .is_some_and(|digest| digest.len() == 64 && digest.bytes().all(|byte| byte.is_ascii_hexdigit()))
+}
+
 pub type DiskStore = Arc<Disk>;

 pub type FileReader = Box<dyn AsyncRead + Send + Sync + Unpin>;
@@ -96,6 +118,20 @@ impl SnapshotLeaseToken {
    pub fn as_bytes(&self) -> &[u8; 16] {
        self.0.as_bytes()
    }
+
+    pub(crate) fn as_uuid(self) -> Uuid {
+        self.0
+    }
+
+    #[doc(hidden)]
+    pub fn revoke_all() -> Self {
+        Self(Uuid::nil())
+    }
+
+    #[doc(hidden)]
+    pub fn is_revoke_all(self) -> bool {
+        self.0.is_nil()
+    }
 }

 impl Default for SnapshotLeaseToken {
@@ -398,10 +434,8 @@ impl DiskAPI for Disk {
        dst_volume: &str,
        dst_path: &str,
    ) -> Result<RenameDataResp> {
-        match self {
-            Disk::Local(local_disk) => local_disk.rename_data(src_volume, src_path, fi, dst_volume, dst_path).await,
-            Disk::Remote(remote_disk) => remote_disk.rename_data(src_volume, src_path, fi, dst_volume, dst_path).await,
-        }
+        self.rename_data_borrowed(src_volume, src_path, &fi, dst_volume, dst_path)
+            .await
    }

    #[tracing::instrument(level = "trace", skip_all)]
@@ -631,6 +665,30 @@ impl DiskAPI for Disk {
    }
 }

+impl Disk {
+    pub(crate) async fn rename_data_borrowed(
+        &self,
+        src_volume: &str,
+        src_path: &str,
+        fi: &FileInfo,
+        dst_volume: &str,
+        dst_path: &str,
+    ) -> Result<RenameDataResp> {
+        match self {
+            Disk::Local(local_disk) => {
+                local_disk
+                    .rename_data_borrowed(src_volume, src_path, fi, dst_volume, dst_path)
+                    .await
+            }
+            Disk::Remote(remote_disk) => {
+                remote_disk
+                    .rename_data_borrowed(src_volume, src_path, fi, dst_volume, dst_path)
+                    .await
+            }
+        }
+    }
+}
+
 impl Disk {
    pub async fn ns_scanner_server_epoch(&self) -> Result<Option<Uuid>> {
        match self {
@@ -1055,6 +1113,10 @@ pub struct DiskInfo {
 }

 #[derive(Clone, Debug, Default)]
+#[allow(
+    dead_code,
+    reason = "MinIO-parity disk info shape with no constructor in this port (backlog#1823)"
+)]
 pub struct Info {
    pub total: u64,
    pub free: u64,
@@ -306,12 +306,20 @@ fn disk_namespace_mutation_lock(path: &Path) -> Arc<NamespaceMutationLock> {
 pub(crate) struct NamespaceMutationLease {
    _namespace_guard: OwnedMutexGuard<()>,
    _volume_guard: Option<OwnedRwLockReadGuard<()>>,
+    external_guard: Mutex<Option<Arc<dyn Send + Sync>>>,
+}
+
+impl NamespaceMutationLease {
+    pub(crate) fn attach_external_guard(&self, guard: Arc<dyn Send + Sync>) {
+        *self.external_guard.lock() = Some(guard);
+    }
 }

 async fn acquire_namespace_mutation_lease(path: &Path) -> Arc<NamespaceMutationLease> {
    Arc::new(NamespaceMutationLease {
        _namespace_guard: disk_namespace_mutation_lock(path).lock_owned().await,
        _volume_guard: None,
+        external_guard: Mutex::new(None),
    })
 }

@@ -327,6 +335,7 @@ pub(crate) async fn acquire_rename_data_mutation_lease(
    Arc::new(NamespaceMutationLease {
        _namespace_guard: namespace_guard,
        _volume_guard: Some(volume_guard),
+        external_guard: Mutex::new(None),
    })
 }

@@ -562,6 +571,10 @@ fn regular_files(dir: &Path) -> io::Result<Vec<PathBuf>> {

 /// Fdatasync every regular file directly inside `dir`, then fsync the directory
 /// itself.
+#[allow(
+    dead_code,
+    reason = "reached only through sync_dir_files, whose callers are tests (backlog#1823)"
+)]
 pub fn sync_dir_files_std(dir: impl AsRef<Path>) -> io::Result<()> {
    for entry in std::fs::read_dir(dir.as_ref())? {
        let entry = entry?;
@@ -574,6 +587,7 @@ pub fn sync_dir_files_std(dir: impl AsRef<Path>) -> io::Result<()> {

 /// Async wrapper around [`sync_dir_files_std`]. Large directories flush files
 /// concurrently, bounded both per directory and process-wide.
+#[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
 pub async fn sync_dir_files(dir: impl AsRef<Path>) -> io::Result<()> {
    sync_dir_files_with_limiter(dir, Arc::new(Semaphore::new(MAX_PARALLEL_FILE_SYNCS))).await
 }
@@ -1800,10 +1814,6 @@ impl RenameCommitGuard {
        })
    }

-    pub(crate) fn lock_destination_directory_for_path_access(&self, directory: &Path) -> io::Result<RenameDestinationPathGuard> {
-        self.destination_directory_guard(directory, false)
-    }
-
    pub(crate) fn create_destination_directory_for_path_access(
        &self,
        directory: &Path,
@@ -2849,13 +2859,6 @@ pub async fn os_mkdir_all(dir_path: impl AsRef<Path>, base_dir: impl AsRef<Path>
    Ok(())
 }

-/// Check if a file exists.
-/// Returns true if the file exists, false otherwise.
-#[tracing::instrument(level = "debug", skip_all)]
-pub fn file_exists(path: impl AsRef<Path>) -> bool {
-    std::fs::metadata(path.as_ref()).map(|_| true).unwrap_or(false)
-}
-
 /// Whether an [`io::Error`] means "the directory is not empty".
 ///
 /// POSIX lets `rmdir`/`rename` report a non-empty directory as either
@@ -1075,9 +1075,6 @@ pub struct GenericError {

 #[derive(Debug, thiserror::Error, PartialEq, Eq)]
 pub enum ObjectApiError {
-    #[error("BackendDown")]
-    BackendDown(String),
-
    #[error("The operation is not valid for the current state of the object {}/{}({})", .0.bucket, .0.object, .0.version_id)]
    InvalidObjectState(GenericError),
 }
@@ -1094,72 +1091,6 @@ pub struct ErrorResponse {
    pub host_id: String,
 }

-pub fn error_resp_to_object_err(err: ErrorResponse, params: Vec<&str>) -> std::io::Error {
-    let mut bucket = "";
-    let mut object = "";
-    let mut version_id = "";
-    if !params.is_empty() {
-        bucket = params[0];
-    }
-    if params.len() >= 2 {
-        object = params[1];
-    }
-    if params.len() >= 3 {
-        version_id = params[2];
-    }
-
-    if is_network_or_host_down(&err.to_string(), false) {
-        return std::io::Error::other(ObjectApiError::BackendDown(format!("{err}")));
-    }
-
-    let err_ = std::io::Error::other(err.to_string());
-    let r_err = err;
-    let err;
-    let bucket = bucket.to_string();
-    let object = object.to_string();
-    let version_id = version_id.to_string();
-
-    match r_err.code {
-        S3ErrorCode::BucketNotEmpty => {
-            err = std::io::Error::other(StorageError::BucketNotEmpty("".to_string()).to_string());
-        }
-        S3ErrorCode::InvalidBucketName => {
-            err = std::io::Error::other(StorageError::BucketNameInvalid(bucket));
-        }
-        S3ErrorCode::InvalidPart => {
-            err = std::io::Error::other(StorageError::InvalidPart(0, bucket, object /* , version_id */));
-        }
-        S3ErrorCode::NoSuchBucket => {
-            err = std::io::Error::other(StorageError::BucketNotFound(bucket));
-        }
-        S3ErrorCode::NoSuchKey => {
-            if !object.is_empty() {
-                err = std::io::Error::other(StorageError::ObjectNotFound(bucket, object));
-            } else {
-                err = std::io::Error::other(StorageError::BucketNotFound(bucket));
-            }
-        }
-        S3ErrorCode::NoSuchVersion => {
-            if !object.is_empty() {
-                err = std::io::Error::other(StorageError::ObjectNotFound(bucket, object)); //, version_id);
-            } else {
-                err = std::io::Error::other(StorageError::BucketNotFound(bucket));
-            }
-        }
-        S3ErrorCode::AccessDenied => {
-            err = std::io::Error::other(StorageError::PrefixAccessDenied(bucket, object));
-        }
-        S3ErrorCode::NoSuchUpload => {
-            err = std::io::Error::other(StorageError::InvalidUploadID(bucket, object, version_id));
-        }
-        _ => {
-            err = err_;
-        }
-    }
-
-    err
-}
-
 #[cfg(test)]
 mod tests {
    use super::*;
@@ -20,6 +20,10 @@ use std::sync::atomic::AtomicI64;
 /// this type never grew past its counter. `total_events` is read by the
 /// notifier's log line but nothing increments it, so that field reports zero.
 #[derive(Default)]
+#[allow(
+    dead_code,
+    reason = "held only by the dead ecstore EventNotifier; see services/event_notification.rs (backlog#1823)"
+)]
 pub struct TargetList {
    pub total_events: AtomicI64,
 }
@@ -31,6 +31,13 @@ pub const ENV_DISK_COMPRESSION_MIME_TYPES: &str = "RUSTFS_COMPRESSION_MIME_TYPES
 // Environment variable for additional extensions to exclude from compression (comma-separated, e.g. ".foo,.bar")
 pub const ENV_ADDED_EXCLUDE_COMPRESS_EXTENSIONS: &str = "RUSTFS_ADDED_EXCLUDE_COMPRESS_EXTENSIONS";

+// Environment variable to additionally enable disk compression for multipart uploads.
+// Default off: nodes from before the resumable decompressor fix fail transient reads of
+// compressed objects, so multipart compression stays dark until the operator confirms the
+// fleet has converged on a fixed build.
+// RUSTFS_COMPAT_TODO(multipart-compression-default-off-window): staged rollout switch for restored multipart compression, flipping the default to enabled on retirement. Remove after the minimum supported direct-upgrade release ships the resumable DecompressReader.
+pub const ENV_DISK_COMPRESSION_MULTIPART_ENABLED: &str = "RUSTFS_COMPRESSION_MULTIPART_ENABLED";
+
 pub const DEFAULT_DISK_COMPRESS_EXTENSIONS: &str = ".txt,.log,.csv,.json,.tar,.xml,.bin";
 pub const DEFAULT_DISK_COMPRESS_MIME_TYPES: &str = "text/*,application/json,application/xml,binary/octet-stream";

@@ -171,6 +178,21 @@ pub fn is_disk_compression_enabled() -> bool {
    DISK_COMPRESSION_CONFIG.get_or_init(parse_disk_compression_config).enabled
 }

+// Parsed once at first use, mirroring DISK_COMPRESSION_CONFIG.
+static MULTIPART_DISK_COMPRESSION_ENABLED: OnceLock<bool> = OnceLock::new();
+
+/// Whether multipart uploads may advertise disk compression. Requires the
+/// regular disk-compression gates to pass as well; this is the staged-rollout
+/// switch that keeps multipart compression dark during rolling upgrades from
+/// builds whose decompressor was not yet resumable.
+pub fn is_multipart_disk_compression_enabled() -> bool {
+    *MULTIPART_DISK_COMPRESSION_ENABLED.get_or_init(|| {
+        env::var(ENV_DISK_COMPRESSION_MULTIPART_ENABLED)
+            .map(|s| matches!(s.to_ascii_lowercase().as_str(), "true" | "on" | "1"))
+            .unwrap_or(false)
+    })
+}
+
 fn is_disk_compressible_with_config(headers: &http::HeaderMap, object_name: &str, config: &DiskCompressionConfig) -> bool {
    // Check if disk compression is enabled (read once at first use, then fixed for process lifetime)
    if !config.enabled {
@@ -22,7 +22,7 @@ use crate::bucket::replication::{
 use crate::bucket::versioning::VersioningApi as _;
 use crate::config::storageclass;
 use crate::error::{Error, Result};
-use crate::io_support::rio::{HashReader, LimitReader};
+use crate::io_support::rio::{HardLimitReader, HashReader};
 use crate::storage_api_contracts::{
    lifecycle::{ExpirationOptions, TransitionedObject},
    range::HTTPRangeSpec,
@@ -479,7 +479,15 @@ enum ReadTransform {
    },
 }

-struct ReadPlan {
+/// How an object's stored bytes must be fetched and transformed to serve a
+/// request.
+///
+/// Public so callers that fetch the stored bytes from somewhere other than the
+/// local erasure set — the remote-tier read path — can position their own fetch
+/// with [`ReadPlan::storage_offset`] / [`ReadPlan::storage_length`] and then
+/// hand the resulting stream to [`ReadPlan::into_object_reader`], instead of
+/// reimplementing the transform decisions (rustfs/rustfs#6025).
+pub struct ReadPlan {
    storage_offset: usize,
    storage_length: i64,
    object_size: i64,
@@ -487,6 +495,43 @@ struct ReadPlan {
 }

 impl ReadPlan {
+    /// Byte offset into the object's **stored** bytes where the fetch must
+    /// start. Encrypted and compressed objects address their storage in a
+    /// different coordinate system than the plaintext range the caller asked
+    /// for, which is exactly the distinction this plan resolves.
+    pub fn storage_offset(&self) -> usize {
+        self.storage_offset
+    }
+
+    /// Number of **stored** bytes the fetch must deliver, in the same
+    /// coordinate system as [`Self::storage_offset`].
+    pub fn storage_length(&self) -> i64 {
+        self.storage_length
+    }
+
+    /// Build the plan for a request without consuming a stream, so a caller
+    /// that has to issue its own positioned fetch can read the offsets first.
+    pub async fn build_for_request(
+        rs: Option<HTTPRangeSpec>,
+        oi: &ObjectInfo,
+        opts: &ObjectOptions,
+        h: &HeaderMap<HeaderValue>,
+        resolver: Option<&dyn ObjectEncryptionResolver>,
+    ) -> Result<Self> {
+        Self::build_with_resolver(rs, oi, opts, h, resolver).await
+    }
+
+    /// Wrap `reader` — the stored bytes this plan asked for, already positioned
+    /// at [`Self::storage_offset`] — in the transforms that turn them into the
+    /// bytes the caller requested.
+    pub fn into_object_reader(
+        self,
+        reader: Box<dyn AsyncRead + Unpin + Send + Sync>,
+        oi: &ObjectInfo,
+    ) -> Result<GetObjectReader> {
+        self.into_reader(reader, oi).map(|(reader, _, _)| reader)
+    }
+
    #[cfg(test)]
    async fn build(rs: Option<HTTPRangeSpec>, oi: &ObjectInfo, opts: &ObjectOptions, h: &HeaderMap<HeaderValue>) -> Result<Self> {
        Self::build_with_resolver(rs, oi, opts, h, Some(&tests::TEST_RESOLVER)).await
@@ -500,8 +545,17 @@ impl ReadPlan {
        resolver: Option<&dyn ObjectEncryptionResolver>,
    ) -> Result<Self> {
        let mut rs = rs;
+        // A part number addresses the object's PLAINTEXT bytes. A restore read
+        // serves the stored representation instead (see
+        // [`restore_request_active`]), where that synthesized range would be
+        // reinterpreted as a storage range and truncate an encrypted or
+        // compressed payload by exactly its encoding overhead — the copy-back
+        // then fails its length check partway through
+        // (rustfs/rustfs#6025). An explicit caller range is already in storage
+        // coordinates on that path and is still honored.
        if let Some(part_number) = opts.part_number
            && rs.is_none()
+            && !restore_request_active(opts)
        {
            rs = http_range_spec_from_object_info(oi, part_number);
        }
@@ -754,7 +808,7 @@ impl ReadPlan {
                        }
                    }
                } else {
-                    Box::new(LimitReader::new(dec_reader, total_plaintext_size))
+                    Box::new(HardLimitReader::new(dec_reader, decompressed_length))
                };

                let mut object_info = oi.clone();
@@ -846,7 +900,7 @@ impl ReadPlan {
                            )?;
                            Box::new(ranged_reader)
                        } else {
-                            Box::new(LimitReader::new(decompressed_reader, total_plaintext_size))
+                            Box::new(HardLimitReader::new(decompressed_reader, total_plaintext_size_i64))
                        }
                    } else if plaintext_offset > 0 || plaintext_length != total_plaintext_size_i64 {
                        Box::new(RangedDecompressReader::new(
@@ -856,7 +910,7 @@ impl ReadPlan {
                            total_plaintext_size,
                        )?)
                    } else {
-                        Box::new(LimitReader::new(decrypted_reader, total_plaintext_size))
+                        Box::new(HardLimitReader::new(decrypted_reader, total_plaintext_size_i64))
                    };

                let mut object_info = oi.clone();
@@ -1727,6 +1781,423 @@ mod tests {
        assert_eq!(actual, b"fghijkl");
    }

+    /// Compresses one multipart part exactly like the write path does
+    /// (`WritePlan::with_compression` wraps each part in its own
+    /// `compression_reader`), returning the on-disk bytes and the storage-format
+    /// compression index.
+    async fn compressed_part_fixture(data: &[u8]) -> (Vec<u8>, Option<Bytes>) {
+        use crate::io_support::rio::TryGetIndex as _;
+        let mut compressor =
+            crate::io_support::rio::compression_reader(Cursor::new(data.to_vec()), CompressionAlgorithm::default(), false);
+        let mut compressed = Vec::new();
+        compressor.read_to_end(&mut compressed).await.expect("compress part stream");
+        let index = compressor
+            .try_get_index()
+            .map(crate::io_support::rio::compression_index_storage_bytes);
+        (compressed, index)
+    }
+
+    struct CompressedMultipartFixture {
+        object_info: ObjectInfo,
+        stored: Vec<u8>,
+        plaintext: Vec<u8>,
+    }
+
+    /// Builds the on-disk representation of a compressed multipart object: each
+    /// part is an independent compressed stream and the storage layer serves
+    /// their concatenation.
+    async fn compressed_multipart_fixture(part_sizes: &[usize]) -> CompressedMultipartFixture {
+        let pattern = b"compressed multipart read path fixture data ";
+        let mut plaintext = Vec::new();
+        let mut stored = Vec::new();
+        let mut parts = Vec::with_capacity(part_sizes.len());
+
+        for (i, part_size) in part_sizes.iter().enumerate() {
+            let mut part_plaintext = Vec::with_capacity(*part_size);
+            while part_plaintext.len() < *part_size {
+                part_plaintext.extend_from_slice(pattern);
+                part_plaintext.push(i as u8);
+            }
+            part_plaintext.truncate(*part_size);
+
+            let (compressed, index) = compressed_part_fixture(&part_plaintext).await;
+            parts.push(ObjectPartInfo {
+                number: i + 1,
+                size: compressed.len(),
+                actual_size: *part_size as i64,
+                index,
+                ..Default::default()
+            });
+            stored.extend_from_slice(&compressed);
+            plaintext.extend_from_slice(&part_plaintext);
+        }
+
+        let mut user_defined = HashMap::new();
+        rustfs_utils::http::insert_str(
+            &mut user_defined,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            crate::io_support::rio::compression_metadata_value(CompressionAlgorithm::default()),
+        );
+        rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, plaintext.len().to_string());
+
+        let object_info = ObjectInfo {
+            bucket: "test-bucket".to_string(),
+            name: "compressed-multipart".to_string(),
+            size: stored.len() as i64,
+            etag: Some(format!("6bcf86bed8807b8e78f0fc6e0a53079d-{}", part_sizes.len())),
+            parts: Arc::new(parts),
+            user_defined: Arc::new(user_defined),
+            ..Default::default()
+        };
+
+        CompressedMultipartFixture {
+            object_info,
+            stored,
+            plaintext,
+        }
+    }
+
+    /// Plans the read once to learn the storage window, then serves exactly that
+    /// window — mirroring how `set_disk` feeds the erasure read into the
+    /// returned reader.
+    async fn read_compressed_multipart(
+        fixture: &CompressedMultipartFixture,
+        rs: Option<HTTPRangeSpec>,
+        opts: &ObjectOptions,
+    ) -> Vec<u8> {
+        let headers = HeaderMap::new();
+        let (_, offset, length) =
+            GetObjectReader::new(Box::new(Cursor::new(Vec::new())), rs.clone(), &fixture.object_info, opts, &headers)
+                .await
+                .expect("plan compressed multipart read");
+
+        let end = offset + usize::try_from(length).expect("storage window length must be non-negative");
+        assert!(
+            end <= fixture.stored.len(),
+            "planned storage window {offset}..{end} exceeds stored stream of {} bytes",
+            fixture.stored.len()
+        );
+        let window = fixture.stored[offset..end].to_vec();
+
+        let (mut reader, replay_offset, replay_length) =
+            GetObjectReader::new(Box::new(Cursor::new(window)), rs, &fixture.object_info, opts, &headers)
+                .await
+                .expect("build compressed multipart reader");
+        assert_eq!((replay_offset, replay_length), (offset, length), "read plan must be deterministic");
+
+        reader.read_all().await.expect("read compressed multipart stream")
+    }
+
+    /// Byte pattern with a 2 KiB period: it compresses extremely well while
+    /// looking nothing like ASCII fixtures. Mirrors the e2e generator that
+    /// exposed a truncated full GET on high-ratio multipart payloads.
+    fn high_ratio_binary_payload(size: usize, seed: u8) -> Vec<u8> {
+        (0..size)
+            .map(|i| ((i as u64).wrapping_mul(2_654_435_761).wrapping_add(seed as u64) >> 3) as u8)
+            .collect()
+    }
+
+    #[tokio::test]
+    async fn compressed_multipart_full_get_handles_high_ratio_binary_payload() {
+        let part_sizes = [5 * 1024 * 1024_usize, 1024 * 1024];
+        let mut plaintext = Vec::new();
+        let mut stored = Vec::new();
+        let mut parts = Vec::with_capacity(part_sizes.len());
+
+        for (i, part_size) in part_sizes.iter().enumerate() {
+            let part_plaintext = high_ratio_binary_payload(*part_size, if i == 0 { 7 } else { 61 });
+            let (compressed, index) = compressed_part_fixture(&part_plaintext).await;
+            parts.push(ObjectPartInfo {
+                number: i + 1,
+                size: compressed.len(),
+                actual_size: *part_size as i64,
+                index,
+                ..Default::default()
+            });
+            stored.extend_from_slice(&compressed);
+            plaintext.extend_from_slice(&part_plaintext);
+        }
+
+        let mut user_defined = HashMap::new();
+        rustfs_utils::http::insert_str(
+            &mut user_defined,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            crate::io_support::rio::compression_metadata_value(CompressionAlgorithm::default()),
+        );
+        rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, plaintext.len().to_string());
+        let fixture = CompressedMultipartFixture {
+            object_info: ObjectInfo {
+                bucket: "test-bucket".to_string(),
+                name: "high-ratio-multipart".to_string(),
+                size: stored.len() as i64,
+                etag: Some("6bcf86bed8807b8e78f0fc6e0a53079d-2".to_string()),
+                parts: Arc::new(parts),
+                user_defined: Arc::new(user_defined),
+                ..Default::default()
+            },
+            stored,
+            plaintext,
+        };
+
+        let read = read_compressed_multipart(&fixture, None, &ObjectOptions::default()).await;
+
+        assert_eq!(read.len(), fixture.plaintext.len(), "full GET must return the logical size");
+        assert_eq!(read, fixture.plaintext, "high-ratio multipart payload must survive the roundtrip");
+    }
+
+    /// Full GET over a compressed multipart object must decode across part
+    /// boundaries: every part is an independent compressed stream (this is also
+    /// the on-disk shape written by builds before rustfs/rustfs#5169 disabled
+    /// multipart compression, so this pins legacy-object readability).
+    #[tokio::test]
+    async fn compressed_multipart_full_get_decodes_across_part_boundaries() {
+        let fixture = compressed_multipart_fixture(&[3 * 1024 * 1024, 2 * 1024 * 1024, 512 * 1024]).await;
+
+        let read = read_compressed_multipart(&fixture, None, &ObjectOptions::default()).await;
+
+        assert_eq!(read.len(), fixture.plaintext.len(), "full GET must return the logical size");
+        assert_eq!(read, fixture.plaintext, "full GET must reassemble all parts");
+    }
+
+    #[tokio::test]
+    async fn compressed_multipart_range_get_crosses_part_boundary() {
+        let fixture = compressed_multipart_fixture(&[3 * 1024 * 1024, 2 * 1024 * 1024]).await;
+        let boundary = 3 * 1024 * 1024_i64;
+        let rs = HTTPRangeSpec {
+            is_suffix_length: false,
+            start: boundary - 100_000,
+            end: boundary + 100_000 - 1,
+        };
+
+        let read = read_compressed_multipart(&fixture, Some(rs), &ObjectOptions::default()).await;
+
+        let expected = &fixture.plaintext[(boundary - 100_000) as usize..(boundary + 100_000) as usize];
+        assert_eq!(read, expected, "boundary-crossing range must splice both parts");
+    }
+
+    #[tokio::test]
+    async fn compressed_multipart_range_get_seeks_into_later_part() {
+        let fixture = compressed_multipart_fixture(&[3 * 1024 * 1024, 4 * 1024 * 1024]).await;
+        // Deep inside part 2 so the plan skips part 1 entirely and (when the
+        // part carries an index) seeks within part 2.
+        let start = 3 * 1024 * 1024_i64 + 2 * 1024 * 1024_i64 + 137;
+        let rs = HTTPRangeSpec {
+            is_suffix_length: false,
+            start,
+            end: start + 64 * 1024 - 1,
+        };
+
+        let read = read_compressed_multipart(&fixture, Some(rs), &ObjectOptions::default()).await;
+
+        let expected = &fixture.plaintext[start as usize..(start + 64 * 1024) as usize];
+        assert_eq!(read, expected, "range inside a later part must decode from that part");
+    }
+
+    /// Parts written without a compression index (small parts skip the index in
+    /// the rio-v2 backend) must still be rangeable: the plan starts at the part
+    /// boundary and skips decompressed bytes.
+    #[tokio::test]
+    async fn compressed_multipart_range_get_works_without_part_indexes() {
+        let mut fixture = compressed_multipart_fixture(&[1024 * 1024, 1024 * 1024]).await;
+        let parts = fixture
+            .object_info
+            .parts
+            .iter()
+            .map(|part| ObjectPartInfo {
+                index: None,
+                ..part.clone()
+            })
+            .collect::<Vec<_>>();
+        fixture.object_info.parts = Arc::new(parts);
+
+        let start = 1024 * 1024_i64 + 4096;
+        let rs = HTTPRangeSpec {
+            is_suffix_length: false,
+            start,
+            end: start + 32 * 1024 - 1,
+        };
+
+        let read = read_compressed_multipart(&fixture, Some(rs), &ObjectOptions::default()).await;
+
+        let expected = &fixture.plaintext[start as usize..(start + 32 * 1024) as usize];
+        assert_eq!(read, expected, "index-less parts must fall back to part-boundary skip");
+    }
+
+    #[tokio::test]
+    async fn compressed_multipart_part_number_get_returns_single_part() {
+        let part_sizes = [3 * 1024 * 1024, 2 * 1024 * 1024, 512 * 1024];
+        let fixture = compressed_multipart_fixture(&part_sizes).await;
+
+        let mut logical_offset = 0_usize;
+        for (i, part_size) in part_sizes.iter().enumerate() {
+            let opts = ObjectOptions {
+                part_number: Some(i + 1),
+                ..Default::default()
+            };
+
+            let read = read_compressed_multipart(&fixture, None, &opts).await;
+
+            let expected = &fixture.plaintext[logical_offset..logical_offset + part_size];
+            assert_eq!(read.len(), *part_size, "partNumber={} GET must return the part's logical size", i + 1);
+            assert_eq!(read, expected, "partNumber={} GET must return the original part bytes", i + 1);
+            logical_offset += part_size;
+        }
+    }
+
+    #[tokio::test]
+    async fn compressed_multipart_suffix_range_reads_tail() {
+        let fixture = compressed_multipart_fixture(&[3 * 1024 * 1024, 1024 * 1024]).await;
+        let suffix_len = 128 * 1024_i64;
+        let rs = HTTPRangeSpec {
+            is_suffix_length: true,
+            start: suffix_len,
+            end: -1,
+        };
+
+        let read = read_compressed_multipart(&fixture, Some(rs), &ObjectOptions::default()).await;
+
+        let expected = &fixture.plaintext[fixture.plaintext.len() - suffix_len as usize..];
+        assert_eq!(read, expected, "suffix range must return the tail of the last part");
+    }
+
+    /// Builds an SSE-C + disk-compression multipart object exactly like the
+    /// write path: each part is compressed into its own stream and then
+    /// encrypted with the per-part key schedule. The fixture is
+    /// legacy-encryption-specific (`rustfs_rio::EncryptReader`), matching the
+    /// pre-existing `build_legacy_ssec_multipart_fixture` shape, while the
+    /// compression layer follows the active backend feature.
+    async fn compressed_encrypted_multipart_fixture(key_bytes: [u8; 32], part_sizes: &[usize]) -> CompressedMultipartFixture {
+        let pattern = b"compressed encrypted multipart fixture data ";
+        let mut plaintext = Vec::new();
+        let mut stored = Vec::new();
+        let mut parts = Vec::with_capacity(part_sizes.len());
+
+        for (i, part_size) in part_sizes.iter().enumerate() {
+            let part_number = i + 1;
+            let mut part_plaintext = Vec::with_capacity(*part_size);
+            while part_plaintext.len() < *part_size {
+                part_plaintext.extend_from_slice(pattern);
+                part_plaintext.push(part_number as u8);
+            }
+            part_plaintext.truncate(*part_size);
+
+            let (compressed, index) = compressed_part_fixture(&part_plaintext).await;
+            let mut part_cipher = Vec::new();
+            rustfs_rio::EncryptReader::new_multipart(Cursor::new(compressed), key_bytes, LEGACY_FIXTURE_BASE_NONCE, part_number)
+                .read_to_end(&mut part_cipher)
+                .await
+                .expect("encrypt compressed fixture part");
+
+            parts.push(ObjectPartInfo {
+                number: part_number,
+                size: part_cipher.len(),
+                actual_size: *part_size as i64,
+                index,
+                ..Default::default()
+            });
+            stored.extend_from_slice(&part_cipher);
+            plaintext.extend_from_slice(&part_plaintext);
+        }
+
+        let mut user_defined = legacy_ssec_multipart_metadata(key_bytes, plaintext.len());
+        rustfs_utils::http::insert_str(
+            &mut user_defined,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            crate::io_support::rio::compression_metadata_value(CompressionAlgorithm::default()),
+        );
+        rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, plaintext.len().to_string());
+
+        let object_info = ObjectInfo {
+            bucket: "test-bucket".to_string(),
+            name: "compressed-encrypted-multipart".to_string(),
+            size: stored.len() as i64,
+            etag: Some(format!("6bcf86bed8807b8e78f0fc6e0a53079d-{}", part_sizes.len())),
+            parts: Arc::new(parts),
+            user_defined: Arc::new(user_defined),
+            ..Default::default()
+        };
+
+        CompressedMultipartFixture {
+            object_info,
+            stored,
+            plaintext,
+        }
+    }
+
+    async fn read_compressed_encrypted_multipart(
+        fixture: &CompressedMultipartFixture,
+        key_bytes: [u8; 32],
+        rs: Option<HTTPRangeSpec>,
+        opts: &ObjectOptions,
+    ) -> Vec<u8> {
+        let headers = ssec_headers_from_key(key_bytes);
+        let (_, offset, length) =
+            GetObjectReader::new(Box::new(Cursor::new(Vec::new())), rs.clone(), &fixture.object_info, opts, &headers)
+                .await
+                .expect("plan compressed encrypted multipart read");
+
+        let end = offset + usize::try_from(length).expect("storage window length must be non-negative");
+        assert!(
+            end <= fixture.stored.len(),
+            "planned storage window {offset}..{end} exceeds stored stream of {} bytes",
+            fixture.stored.len()
+        );
+        let window = fixture.stored[offset..end].to_vec();
+
+        let (mut reader, replay_offset, replay_length) =
+            GetObjectReader::new(Box::new(Cursor::new(window)), rs, &fixture.object_info, opts, &headers)
+                .await
+                .expect("build compressed encrypted multipart reader");
+        assert_eq!((replay_offset, replay_length), (offset, length), "read plan must be deterministic");
+
+        reader.read_all().await.expect("read compressed encrypted multipart stream")
+    }
+
+    #[tokio::test]
+    async fn compressed_encrypted_multipart_full_get_roundtrip() {
+        let key_bytes = [0x6Eu8; 32];
+        let fixture = compressed_encrypted_multipart_fixture(key_bytes, &[3 * 1024 * 1024, 1024 * 1024]).await;
+
+        let read = read_compressed_encrypted_multipart(&fixture, key_bytes, None, &ObjectOptions::default()).await;
+
+        assert_eq!(read.len(), fixture.plaintext.len(), "full GET must return the logical size");
+        assert_eq!(read, fixture.plaintext, "SSE-C + compression full GET must reassemble all parts");
+    }
+
+    #[tokio::test]
+    async fn compressed_encrypted_multipart_range_crosses_part_boundary() {
+        let key_bytes = [0x6Eu8; 32];
+        let fixture = compressed_encrypted_multipart_fixture(key_bytes, &[3 * 1024 * 1024, 1024 * 1024]).await;
+        let boundary = 3 * 1024 * 1024_i64;
+        let rs = HTTPRangeSpec {
+            is_suffix_length: false,
+            start: boundary - 65_536,
+            end: boundary + 65_536 - 1,
+        };
+
+        let read = read_compressed_encrypted_multipart(&fixture, key_bytes, Some(rs), &ObjectOptions::default()).await;
+
+        let expected = &fixture.plaintext[(boundary - 65_536) as usize..(boundary + 65_536) as usize];
+        assert_eq!(read, expected, "SSE-C + compression boundary-crossing range must splice both parts");
+    }
+
+    #[tokio::test]
+    async fn compressed_encrypted_multipart_part_number_get_returns_single_part() {
+        let key_bytes = [0x6Eu8; 32];
+        let part_sizes = [3 * 1024 * 1024, 1024 * 1024];
+        let fixture = compressed_encrypted_multipart_fixture(key_bytes, &part_sizes).await;
+
+        let opts = ObjectOptions {
+            part_number: Some(2),
+            ..Default::default()
+        };
+        let read = read_compressed_encrypted_multipart(&fixture, key_bytes, None, &opts).await;
+
+        let expected = &fixture.plaintext[part_sizes[0]..];
+        assert_eq!(read.len(), part_sizes[1], "partNumber=2 GET must return the part's logical size");
+        assert_eq!(read, expected, "partNumber=2 GET must return the original part bytes");
+    }
+
    #[tokio::test]
    async fn test_get_object_reader_rejects_ssec_read_without_headers() {
        let object_info = ObjectInfo {
@@ -277,6 +277,12 @@ pub struct ObjectOptions {
    /// fence avoids recursively acquiring the read lock behind a queued writer.
    pub bucket_lifecycle_lock_fence: Option<NamespaceLockFence>,
    pub replication_request: bool,
+    /// Source-cluster LWW timestamps carried by an authorized replication
+    /// request; None when the source never modified the category. Only the
+    /// replication-authorized options builders may set these.
+    pub replication_tagging_timestamp: Option<OffsetDateTime>,
+    pub replication_retention_timestamp: Option<OffsetDateTime>,
+    pub replication_legalhold_timestamp: Option<OffsetDateTime>,
    /// Authorized SSE-C replication passthrough: the body is already
    /// ciphertext, so the write path must not encrypt or compress it and
    /// stores the restored encryption metadata verbatim. Only the
@@ -23,6 +23,10 @@ use std::sync::{Arc, Mutex};
 use std::time::{Duration, Instant};
 use tokio::task::JoinSet;

+#[allow(
+    dead_code,
+    reason = "default operation label for the test-only AsyncBatchProcessor::new (backlog#1823)"
+)]
 const BATCH_PROCESSOR_OPERATION_CUSTOM: &str = "custom";
 const BATCH_PROCESSOR_OPERATION_READ: &str = "read";
 const BATCH_PROCESSOR_OPERATION_WRITE: &str = "write";
@@ -211,6 +215,7 @@ pub struct AsyncBatchProcessor {
 }

 impl AsyncBatchProcessor {
+    #[allow(dead_code, reason = "constructor used only by this file's tests (backlog#1823)")]
    pub fn new(max_concurrent: usize) -> Self {
        Self::new_with_operation(max_concurrent, BATCH_PROCESSOR_OPERATION_CUSTOM)
    }
@@ -26,11 +26,26 @@ use std::sync::atomic::Ordering;
 use tokio::sync::RwLock;
 use tracing::warn;

+/// Dead ecstore-side notification skeleton.
+///
+/// The working notification stack is `rustfs-notify`, whose own `EventNotifier`
+/// is the one bucket configuration actually drives. Nothing calls the methods
+/// below; `init_bucket_targets` even logs that it is a no-op in this build.
+/// Removing it means also retiring the `InstanceContext` slot that holds it
+/// (backlog#939 Phase 5), so it is left explicit here rather than half-removed.
+#[allow(
+    dead_code,
+    reason = "ecstore-side notification skeleton superseded by rustfs-notify; see module note (backlog#1823)"
+)]
 pub struct EventNotifier {
    target_list: TargetList,
    //bucket_rules_map: HashMap<String , HashMap<EventName, Rules>>,
 }

+#[allow(
+    dead_code,
+    reason = "ecstore-side notification skeleton superseded by rustfs-notify; see module note (backlog#1823)"
+)]
 impl EventNotifier {
    pub fn new() -> Arc<RwLock<Self>> {
        Arc::new(RwLock::new(Self {
@@ -13,7 +13,6 @@
 // limitations under the License.

 // #730: background service owners still contain staged notification/rebalance/tier paths.
-#![allow(dead_code)]

 pub(crate) mod batch_processor;
 pub(crate) mod event_notification;
@@ -44,12 +44,14 @@ const CONSECUTIVE_FAILURE_THRESHOLD: u32 = 3;
 const LOG_COMPONENT_ECSTORE: &str = "ecstore";
 const LOG_SUBSYSTEM_NOTIFICATION: &str = "notification";
 const EVENT_NOTIFICATION_PEER_PROPAGATION: &str = "notification_peer_propagation";
+const EVENT_NOTIFICATION_CAPABILITY_PROBE: &str = "notification_capability_probe";
 const SCANNER_ACTIVITY_PROBE_TIMEOUT: Duration = Duration::from_secs(5);
 const TIER_CONFIG_RELOAD_RETRY_BASE: Duration = Duration::from_millis(100);
 const TIER_CONFIG_RELOAD_RETRY_CAP: Duration = Duration::from_secs(5);
 const REMOTE_VERSION_STATE_PROBE_INTERVAL: Duration = Duration::from_secs(10);
 const REMOTE_VERSION_STATE_PROBE_TIMEOUT: Duration = Duration::from_secs(5);
 const REMOTE_VERSION_STATE_PROOF_TTL: Duration = Duration::from_secs(30);
+const CROSS_POOL_FENCE_SUPPORTED_VERSION: u32 = 1;

 /// Cached result from the last successful admin call to a peer.
 struct PeerAdminCache {
@@ -95,15 +97,15 @@ lazy_static! {
 }

 #[derive(Clone)]
-struct RemoteVersionStateFleetProof {
+struct FleetCapabilityProof {
    topology_fingerprint: String,
    peer_epochs: Arc<BTreeMap<String, Uuid>>,
    expires_at: Instant,
 }

-impl RemoteVersionStateFleetProof {
-    fn token(&self) -> RemoteVersionStateFleetProofToken {
-        RemoteVersionStateFleetProofToken {
+impl FleetCapabilityProof {
+    fn token(&self) -> FleetCapabilityProofToken {
+        FleetCapabilityProofToken {
            topology_fingerprint: self.topology_fingerprint.clone(),
            peer_epochs: self.peer_epochs.clone(),
        }
@@ -111,37 +113,41 @@ impl RemoteVersionStateFleetProof {
 }

 #[derive(Clone, PartialEq, Eq)]
-pub(crate) struct RemoteVersionStateFleetProofToken {
+struct FleetCapabilityProofToken {
    topology_fingerprint: String,
    peer_epochs: Arc<BTreeMap<String, Uuid>>,
 }

 #[derive(Default)]
-struct RemoteVersionStateFleetProofState {
-    proof: Option<RemoteVersionStateFleetProof>,
+struct FleetCapabilityProofState {
+    proof: Option<FleetCapabilityProof>,
    topology_conflict: bool,
 }

-static REMOTE_VERSION_STATE_FLEET_PROOF: OnceLock<std::sync::RwLock<RemoteVersionStateFleetProofState>> = OnceLock::new();
+#[derive(Clone, PartialEq, Eq)]
+pub(crate) struct RemoteVersionStateFleetProofToken(FleetCapabilityProofToken);
+
+#[derive(Clone, PartialEq, Eq)]
+pub struct CrossPoolFenceFleetProofToken(FleetCapabilityProofToken);
+
+static REMOTE_VERSION_STATE_FLEET_PROOF: OnceLock<std::sync::RwLock<FleetCapabilityProofState>> = OnceLock::new();
+static CROSS_POOL_FENCE_FLEET_PROOF: OnceLock<std::sync::RwLock<FleetCapabilityProofState>> = OnceLock::new();
 static REMOTE_VERSION_STATE_PROBE_TOPOLOGY: OnceLock<String> = OnceLock::new();

-fn remote_version_state_fleet_proof_slot() -> &'static std::sync::RwLock<RemoteVersionStateFleetProofState> {
-    REMOTE_VERSION_STATE_FLEET_PROOF.get_or_init(|| std::sync::RwLock::new(RemoteVersionStateFleetProofState::default()))
+fn cross_pool_fence_fleet_proof_slot() -> &'static std::sync::RwLock<FleetCapabilityProofState> {
+    CROSS_POOL_FENCE_FLEET_PROOF.get_or_init(|| std::sync::RwLock::new(FleetCapabilityProofState::default()))
 }

-fn replace_remote_version_state_fleet_proof(proof: Option<RemoteVersionStateFleetProof>) {
-    replace_remote_version_state_fleet_proof_in(remote_version_state_fleet_proof_slot(), proof);
+fn remote_version_state_fleet_proof_slot() -> &'static std::sync::RwLock<FleetCapabilityProofState> {
+    REMOTE_VERSION_STATE_FLEET_PROOF.get_or_init(|| std::sync::RwLock::new(FleetCapabilityProofState::default()))
 }

-fn replace_remote_version_state_fleet_proof_in(
-    slot: &std::sync::RwLock<RemoteVersionStateFleetProofState>,
-    proof: Option<RemoteVersionStateFleetProof>,
-) {
+fn replace_fleet_capability_proof(slot: &std::sync::RwLock<FleetCapabilityProofState>, proof: Option<FleetCapabilityProof>) {
    slot.write().unwrap_or_else(std::sync::PoisonError::into_inner).proof = proof;
 }

-fn publish_remote_version_state_probe_result(
-    slot: &std::sync::RwLock<RemoteVersionStateFleetProofState>,
+fn publish_fleet_capability_probe_result(
+    slot: &std::sync::RwLock<FleetCapabilityProofState>,
    topology_fingerprint: &str,
    result: Result<BTreeMap<String, Uuid>>,
    observed_at: Instant,
@@ -155,7 +161,7 @@ fn publish_remote_version_state_probe_result(
                .filter(|proof| proof.topology_fingerprint == topology_fingerprint && proof.peer_epochs.as_ref() == &peer_epochs)
                .map(|proof| Arc::clone(&proof.peer_epochs))
                .unwrap_or_else(|| Arc::new(peer_epochs));
-            state.proof = Some(RemoteVersionStateFleetProof {
+            state.proof = Some(FleetCapabilityProof {
                topology_fingerprint: topology_fingerprint.to_string(),
                peer_epochs,
                expires_at: observed_at + REMOTE_VERSION_STATE_PROOF_TTL,
@@ -163,7 +169,7 @@ fn publish_remote_version_state_probe_result(
            None
        }
        Err(err) => {
-            replace_remote_version_state_fleet_proof_in(slot, None);
+            replace_fleet_capability_proof(slot, None);
            Some(err)
        }
    }
@@ -174,27 +180,60 @@ pub(crate) fn acquire_remote_version_state_fleet_proof() -> Option<RemoteVersion
    let state = remote_version_state_fleet_proof_slot()
        .read()
        .unwrap_or_else(std::sync::PoisonError::into_inner);
-    acquire_remote_version_state_fleet_proof_from(&state, expected_topology, Instant::now())
+    acquire_fleet_capability_proof_from(&state, expected_topology, Instant::now()).map(RemoteVersionStateFleetProofToken)
 }

-fn acquire_remote_version_state_fleet_proof_from(
-    state: &RemoteVersionStateFleetProofState,
+fn acquire_fleet_capability_proof_from(
+    state: &FleetCapabilityProofState,
    expected_topology: &str,
    now: Instant,
-) -> Option<RemoteVersionStateFleetProofToken> {
-    if state.topology_conflict || !remote_version_state_fleet_proof_valid_at(state.proof.as_ref(), expected_topology, now) {
+) -> Option<FleetCapabilityProofToken> {
+    if state.topology_conflict || !fleet_capability_proof_valid_at(state.proof.as_ref(), expected_topology, now) {
        return None;
    }
-    state.proof.as_ref().map(RemoteVersionStateFleetProof::token)
+    state.proof.as_ref().map(FleetCapabilityProof::token)
 }

 pub(crate) fn remote_version_state_fleet_proof_matches(proof: &RemoteVersionStateFleetProofToken) -> bool {
+    fleet_capability_proof_matches(remote_version_state_fleet_proof_slot(), &proof.0)
+}
+
+pub fn acquire_cross_pool_fence_fleet_proof() -> Option<CrossPoolFenceFleetProofToken> {
+    let expected_topology = REMOTE_VERSION_STATE_PROBE_TOPOLOGY.get()?;
+    let state = cross_pool_fence_fleet_proof_slot()
+        .read()
+        .unwrap_or_else(std::sync::PoisonError::into_inner);
+    acquire_fleet_capability_proof_from(&state, expected_topology, Instant::now()).map(CrossPoolFenceFleetProofToken)
+}
+
+pub fn cross_pool_fence_fleet_proof_matches(proof: &CrossPoolFenceFleetProofToken) -> bool {
+    fleet_capability_proof_matches(cross_pool_fence_fleet_proof_slot(), &proof.0)
+}
+
+#[cfg(any(test, feature = "test-util"))]
+pub fn rotate_cross_pool_fence_fleet_proof_for_test() -> bool {
+    let mut state = cross_pool_fence_fleet_proof_slot()
+        .write()
+        .unwrap_or_else(std::sync::PoisonError::into_inner);
+    let Some(current) = state.proof.as_ref() else {
+        return false;
+    };
+    state.proof = Some(FleetCapabilityProof {
+        topology_fingerprint: current.topology_fingerprint.clone(),
+        peer_epochs: Arc::new(current.peer_epochs.as_ref().clone()),
+        expires_at: current.expires_at,
+    });
+    true
+}
+
+fn fleet_capability_proof_matches(
+    slot: &std::sync::RwLock<FleetCapabilityProofState>,
+    proof: &FleetCapabilityProofToken,
+) -> bool {
    let Some(expected_topology) = REMOTE_VERSION_STATE_PROBE_TOPOLOGY.get() else {
        return false;
    };
-    let state = remote_version_state_fleet_proof_slot()
-        .read()
-        .unwrap_or_else(std::sync::PoisonError::into_inner);
+    let state = slot.read().unwrap_or_else(std::sync::PoisonError::into_inner);
    if state.topology_conflict {
        return false;
    }
@@ -206,13 +245,17 @@ pub(crate) fn remote_version_state_fleet_proof_matches(proof: &RemoteVersionStat
    })
 }

+fn fleet_capability_proof_valid_at(proof: Option<&FleetCapabilityProof>, expected_topology: &str, now: Instant) -> bool {
+    proof.is_some_and(|proof| proof.topology_fingerprint == expected_topology && now < proof.expires_at)
+}
+
 #[cfg(test)]
 pub(crate) struct RemoteVersionStateFleetProofGuard;

 #[cfg(test)]
 impl Drop for RemoteVersionStateFleetProofGuard {
    fn drop(&mut self) {
-        replace_remote_version_state_fleet_proof(None);
+        replace_fleet_capability_proof(remote_version_state_fleet_proof_slot(), None);
    }
 }

@@ -227,7 +270,7 @@ pub(crate) fn install_remote_version_state_fleet_proof_for_test(topology_fingerp
        Err(_) => panic!("remote version state test topology is already bound to another fingerprint"),
    }
    let peer_epochs = BTreeMap::new();
-    if let Some(err) = publish_remote_version_state_probe_result(
+    if let Some(err) = publish_fleet_capability_probe_result(
        remote_version_state_fleet_proof_slot(),
        topology_fingerprint,
        Ok(peer_epochs),
@@ -238,14 +281,6 @@ pub(crate) fn install_remote_version_state_fleet_proof_for_test(topology_fingerp
    RemoteVersionStateFleetProofGuard
 }

-fn remote_version_state_fleet_proof_valid_at(
-    proof: Option<&RemoteVersionStateFleetProof>,
-    expected_topology: &str,
-    now: Instant,
-) -> bool {
-    proof.is_some_and(|proof| proof.topology_fingerprint == expected_topology && now < proof.expires_at)
-}
-
 fn insert_remote_version_state_peer(peer_epochs: &mut BTreeMap<String, Uuid>, peer: String, epoch: Uuid) -> Result<()> {
    if epoch.is_nil() || peer_epochs.values().any(|existing| *existing == epoch) || peer_epochs.insert(peer, epoch).is_some() {
        return Err(Error::other("remote version state capability peer identity is invalid"));
@@ -256,11 +291,11 @@ fn insert_remote_version_state_peer(peer_epochs: &mut BTreeMap<String, Uuid>, pe
 pub fn start_remote_version_state_fleet_probe(topology_fingerprint: String) {
    if REMOTE_VERSION_STATE_PROBE_TOPOLOGY.set(topology_fingerprint.clone()).is_err() {
        if REMOTE_VERSION_STATE_PROBE_TOPOLOGY.get() != Some(&topology_fingerprint) {
-            let mut state = remote_version_state_fleet_proof_slot()
-                .write()
-                .unwrap_or_else(std::sync::PoisonError::into_inner);
-            state.topology_conflict = true;
-            state.proof = None;
+            for slot in [remote_version_state_fleet_proof_slot(), cross_pool_fence_fleet_proof_slot()] {
+                let mut state = slot.write().unwrap_or_else(std::sync::PoisonError::into_inner);
+                state.topology_conflict = true;
+                state.proof = None;
+            }
        }
        return;
    }
@@ -281,13 +316,23 @@ pub fn start_remote_version_state_fleet_probe(topology_fingerprint: String) {
                }
                None => Err(Error::other("remote version state fleet capability notification system is unavailable")),
            };
+            let fence_result = match get_global_notification_sys() {
+                Some(notification_sys) => timeout(
+                    REMOTE_VERSION_STATE_PROBE_TIMEOUT,
+                    notification_sys.probe_cross_pool_fence_fleet(&topology_fingerprint),
+                )
+                .await
+                .unwrap_or_else(|_| Err(Error::other("cross-pool fence fleet capability probe timed out"))),
+                None => Err(Error::other("cross-pool fence fleet capability notification system is unavailable")),
+            };
            let topology_conflict = remote_version_state_fleet_proof_slot()
                .read()
                .unwrap_or_else(std::sync::PoisonError::into_inner)
                .topology_conflict;
            if topology_conflict {
-                replace_remote_version_state_fleet_proof(None);
-            } else if let Some(err) = publish_remote_version_state_probe_result(
+                replace_fleet_capability_proof(remote_version_state_fleet_proof_slot(), None);
+                replace_fleet_capability_proof(cross_pool_fence_fleet_proof_slot(), None);
+            } else if let Some(err) = publish_fleet_capability_probe_result(
                remote_version_state_fleet_proof_slot(),
                &topology_fingerprint,
                result,
@@ -295,6 +340,24 @@ pub fn start_remote_version_state_fleet_probe(topology_fingerprint: String) {
            ) {
                debug!(error = %err, "remote version state fleet capability probe failed closed");
            }
+            if !topology_conflict
+                && let Some(err) = publish_fleet_capability_probe_result(
+                    cross_pool_fence_fleet_proof_slot(),
+                    &topology_fingerprint,
+                    fence_result,
+                    Instant::now(),
+                )
+            {
+                debug!(
+                    event = EVENT_NOTIFICATION_CAPABILITY_PROBE,
+                    component = LOG_COMPONENT_ECSTORE,
+                    subsystem = LOG_SUBSYSTEM_NOTIFICATION,
+                    capability = "cross_pool_fence_v1",
+                    state = "failed_closed",
+                    error = %err,
+                    "notification capability probe"
+                );
+            }
            sleep(REMOTE_VERSION_STATE_PROBE_INTERVAL).await;
        }
    });
@@ -362,6 +425,27 @@ impl NotificationSys {
        }
        Ok(peer_epochs)
    }
+
+    async fn probe_cross_pool_fence_fleet(&self, topology_fingerprint: &str) -> Result<BTreeMap<String, Uuid>> {
+        if self.peer_clients.len() != self.peer_topology_hosts.len() {
+            return Err(Error::other("cross-pool fence capability fleet membership is incomplete"));
+        }
+        let probes = self.peer_clients.iter().map(|client| async {
+            let client = client
+                .as_ref()
+                .ok_or_else(|| Error::other("cross-pool fence capability peer is unreachable"))?;
+            client.probe_cross_pool_fence(topology_fingerprint.to_string()).await
+        });
+        let mut peer_epochs = BTreeMap::new();
+        for result in join_all(probes).await {
+            let (peer, version, epoch) = result?;
+            if version < CROSS_POOL_FENCE_SUPPORTED_VERSION {
+                return Err(Error::other("cross-pool fence capability version is unsupported"));
+            }
+            insert_remote_version_state_peer(&mut peer_epochs, peer, epoch)?;
+        }
+        Ok(peer_epochs)
+    }
 }

 pub struct NotificationPeerErr {
@@ -1539,6 +1623,7 @@ impl NotificationSys {
        workers.peers.remove(host);
    }

+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    fn tier_config_reload_worker_active(&self, host: &str) -> bool {
        self.tier_config_reload_workers
            .lock()
@@ -1712,6 +1797,7 @@ where
        .map_err(|_| Error::other(format!("scanner activity peer {host} timed out after {timeout_duration:?}")))?
 }

+#[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
 async fn call_peer_with_timeout<F, Fut>(
    timeout_dur: Duration,
    host_label: &str,
@@ -2177,16 +2263,16 @@ mod tests {
        let now = Instant::now();
        let mut peer_epochs = BTreeMap::new();
        peer_epochs.insert("peer-a".to_string(), Uuid::new_v4());
-        let proof = RemoteVersionStateFleetProof {
+        let proof = FleetCapabilityProof {
            topology_fingerprint: "topology-a".to_string(),
            peer_epochs: Arc::new(peer_epochs),
            expires_at: now + Duration::from_secs(1),
        };

-        assert!(remote_version_state_fleet_proof_valid_at(Some(&proof), "topology-a", now));
-        assert!(!remote_version_state_fleet_proof_valid_at(Some(&proof), "topology-b", now));
-        assert!(!remote_version_state_fleet_proof_valid_at(Some(&proof), "topology-a", proof.expires_at));
-        assert!(!remote_version_state_fleet_proof_valid_at(None, "topology-a", now));
+        assert!(fleet_capability_proof_valid_at(Some(&proof), "topology-a", now));
+        assert!(!fleet_capability_proof_valid_at(Some(&proof), "topology-b", now));
+        assert!(!fleet_capability_proof_valid_at(Some(&proof), "topology-a", proof.expires_at));
+        assert!(!fleet_capability_proof_valid_at(None, "topology-a", now));
    }

    #[test]
@@ -2200,25 +2286,25 @@ mod tests {
    #[test]
    fn remote_version_state_fleet_proof_accepts_single_node_membership() {
        let now = Instant::now();
-        let proof = RemoteVersionStateFleetProof {
+        let proof = FleetCapabilityProof {
            topology_fingerprint: "topology-a".to_string(),
            peer_epochs: Arc::new(BTreeMap::new()),
            expires_at: now + Duration::from_secs(1),
        };

-        assert!(remote_version_state_fleet_proof_valid_at(Some(&proof), "topology-a", now));
+        assert!(fleet_capability_proof_valid_at(Some(&proof), "topology-a", now));
    }

    #[test]
    fn remote_version_state_fleet_proof_token_changes_with_process_epoch() {
        let now = Instant::now();
-        let proof = RemoteVersionStateFleetProof {
+        let proof = FleetCapabilityProof {
            topology_fingerprint: "topology-a".to_string(),
            peer_epochs: Arc::new(BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())])),
            expires_at: now + Duration::from_secs(1),
        };
        let captured = proof.token();
-        let restarted = RemoteVersionStateFleetProof {
+        let restarted = FleetCapabilityProof {
            topology_fingerprint: proof.topology_fingerprint.clone(),
            peer_epochs: Arc::new(BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())])),
            expires_at: proof.expires_at,
@@ -2229,11 +2315,11 @@ mod tests {

    #[test]
    fn remote_version_state_fleet_proof_renewal_preserves_only_same_epoch_token() {
-        let slot = std::sync::RwLock::new(RemoteVersionStateFleetProofState::default());
+        let slot = std::sync::RwLock::new(FleetCapabilityProofState::default());
        let now = Instant::now();
        let epoch = Uuid::new_v4();
        let peers = BTreeMap::from([("peer-a".to_string(), epoch)]);
-        assert!(publish_remote_version_state_probe_result(&slot, "topology-a", Ok(peers.clone()), now).is_none());
+        assert!(publish_fleet_capability_probe_result(&slot, "topology-a", Ok(peers.clone()), now).is_none());
        let original = slot
            .read()
            .expect("proof slot should not poison")
@@ -2242,9 +2328,7 @@ mod tests {
            .expect("successful probe should publish proof")
            .token();

-        assert!(
-            publish_remote_version_state_probe_result(&slot, "topology-a", Ok(peers), now + Duration::from_millis(1)).is_none()
-        );
+        assert!(publish_fleet_capability_probe_result(&slot, "topology-a", Ok(peers), now + Duration::from_millis(1)).is_none());
        let renewed = slot
            .read()
            .expect("proof slot should not poison")
@@ -2256,8 +2340,7 @@ mod tests {

        let restarted = BTreeMap::from([("peer-a".to_string(), Uuid::new_v4())]);
        assert!(
-            publish_remote_version_state_probe_result(&slot, "topology-a", Ok(restarted), now + Duration::from_millis(2))
-                .is_none()
+            publish_fleet_capability_probe_result(&slot, "topology-a", Ok(restarted), now + Duration::from_millis(2)).is_none()
        );
        let replaced = slot
            .read()
@@ -2272,18 +2355,18 @@ mod tests {
    #[test]
    fn remote_version_state_fleet_proof_conflict_revokes_atomic_snapshot() {
        let now = Instant::now();
-        let mut state = RemoteVersionStateFleetProofState {
-            proof: Some(RemoteVersionStateFleetProof {
+        let mut state = FleetCapabilityProofState {
+            proof: Some(FleetCapabilityProof {
                topology_fingerprint: "topology-a".to_string(),
                peer_epochs: Arc::new(BTreeMap::new()),
                expires_at: now + Duration::from_secs(1),
            }),
            topology_conflict: false,
        };
-        assert!(acquire_remote_version_state_fleet_proof_from(&state, "topology-a", now).is_some());
+        assert!(acquire_fleet_capability_proof_from(&state, "topology-a", now).is_some());

        state.topology_conflict = true;
-        assert!(acquire_remote_version_state_fleet_proof_from(&state, "topology-a", now).is_none());
+        assert!(acquire_fleet_capability_proof_from(&state, "topology-a", now).is_none());
    }

    #[test]
@@ -2299,19 +2382,19 @@ mod tests {

    #[test]
    fn remote_version_state_fleet_probe_failure_revokes_previous_proof() {
-        let slot = std::sync::RwLock::new(RemoteVersionStateFleetProofState::default());
+        let slot = std::sync::RwLock::new(FleetCapabilityProofState::default());
        let now = Instant::now();
        let peer_epochs = BTreeMap::from([("node-a:9000".to_string(), Uuid::new_v4())]);
-        assert!(publish_remote_version_state_probe_result(&slot, "topology-a", Ok(peer_epochs), now).is_none());
+        assert!(publish_fleet_capability_probe_result(&slot, "topology-a", Ok(peer_epochs), now).is_none());
        assert!(slot.read().expect("proof slot should not poison").proof.is_some());

        assert!(
-            publish_remote_version_state_probe_result(&slot, "topology-a", Err(Error::other("peer unavailable")), now,).is_some()
+            publish_fleet_capability_probe_result(&slot, "topology-a", Err(Error::other("peer unavailable")), now,).is_some()
        );
        assert!(slot.read().expect("proof slot should not poison").proof.is_none());

        let peer_epochs = BTreeMap::from([("node-a:9000".to_string(), Uuid::new_v4())]);
-        assert!(publish_remote_version_state_probe_result(&slot, "topology-a", Ok(peer_epochs), now).is_none());
+        assert!(publish_fleet_capability_probe_result(&slot, "topology-a", Ok(peer_epochs), now).is_none());
        assert!(slot.read().expect("proof slot should not poison").proof.is_some());
    }

@@ -864,6 +864,10 @@ pub(super) fn merge_rebalance_meta(remote: &mut RebalanceMeta, local: &Rebalance
    RebalanceMetaMergeOutcome::Merged
 }

+#[allow(
+    dead_code,
+    reason = "stop-transition helper retained beside stop_rebalance_meta_snapshot; no caller yet (backlog#1823)"
+)]
 pub(super) fn mark_started_rebalance_pools_stopped(meta: &mut RebalanceMeta, stop_time: OffsetDateTime) {
    for pool_stat in meta.pool_stats.iter_mut() {
        if pool_stat.info.status == RebalStatus::Started {
@@ -964,6 +968,7 @@ pub(super) fn rollback_rebalance_start_meta_snapshot_for_id(
    })
 }

+#[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
 pub(super) fn stop_rebalance_meta_snapshot(meta: Option<&mut RebalanceMeta>, now: OffsetDateTime) -> Option<RebalanceMeta> {
    let meta = meta?;
    stop_rebalance_state(meta, now);
@@ -171,6 +171,7 @@ where
 }

 #[allow(clippy::too_many_arguments)]
+#[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
 pub(super) async fn migrate_entry_version_with_retry_wait<Backend, F, Fut, D, DFut, W, WFut>(
    set: &Backend,
    bucket: String,
@@ -1,5 +1,4 @@
 use serde::{Deserialize, Serialize};
-use std::sync::Arc;
 use time::OffsetDateTime;
 use tokio_util::sync::CancellationToken;

@@ -32,8 +31,6 @@ pub struct RebalanceStats {
    pub cleanup_warnings: RebalanceCleanupWarnings,
 }

-pub type RStats = Vec<Arc<RebalanceStats>>;
-
 #[derive(Debug, Default)]
 pub(super) struct RebalanceBucketConfigs {
    pub(super) bucket_incarnation_id: Option<uuid::Uuid>,
@@ -30,6 +30,5 @@ pub mod warm_backend_minio;
 pub mod warm_backend_r2;
 pub mod warm_backend_rustfs;
 pub mod warm_backend_s3;
-pub mod warm_backend_s3sdk;
 pub mod warm_backend_tencent;
 pub mod warm_backend_wasabi;
@@ -488,6 +488,7 @@ impl TierCandidateMutation {
        targets
    }

+    #[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
    fn affected_targets(
        &self,
        manager: &TierConfigMgr,
@@ -802,6 +803,7 @@ fn tier_persisted_reference_blocks_any_target(
        .any(|target| tier_persisted_reference_blocks_target(tier_name, backend_identity, target))
 }

+#[allow(dead_code, reason = "asserted by this file's tests (backlog#1823)")]
 fn tier_object_blocks_target_rebind(object: &ObjectInfo, target: &TierMutationIntentTarget) -> io::Result<bool> {
    tier_object_blocks_any_target_rebind(object, std::slice::from_ref(target))
 }
@@ -2726,14 +2728,6 @@ impl TierConfigMgr {
        Self::publish_candidate_owned(handle, candidate, driver_tier.map(str::to_string), update).await
    }

-    fn begin_publish_transition(
-        handle: &Arc<RwLock<Self>>,
-        manager: &mut Self,
-        candidate: &Self,
-    ) -> std::result::Result<TierPublishTransition, AdminError> {
-        Self::begin_publish_transition_with_allowed_mutation_blocks(handle, manager, candidate, None)
-    }
-
    fn begin_publish_transition_with_allowed_mutation_blocks(
        handle: &Arc<RwLock<Self>>,
        manager: &mut Self,
@@ -2819,14 +2813,6 @@ impl TierConfigMgr {
        })
    }

-    async fn publish_candidate_inner(
-        handle: &Arc<RwLock<Self>>,
-        candidate: Self,
-        driver_tier: Option<&str>,
-    ) -> std::result::Result<(), AdminError> {
-        Self::publish_candidate_inner_with_allowed_mutation_blocks(handle, candidate, driver_tier, None).await
-    }
-
    async fn publish_candidate_inner_with_allowed_mutation_blocks(
        handle: &Arc<RwLock<Self>>,
        candidate: Self,
@@ -2939,6 +2925,7 @@ impl TierConfigMgr {
        admin_err
    }

+    #[allow(dead_code, reason = "reached only through #[cfg(test)] helpers in this file (backlog#1823)")]
    async fn publish_candidate_owned(
        handle: &Arc<RwLock<Self>>,
        candidate: Self,
@@ -3541,6 +3528,7 @@ impl TierConfigMgr {
        Self::update_candidate_with_config_lock(handle, api, TierCandidateMutation::Remove(tier_name.to_string(), force)).await
    }

+    #[allow(dead_code, reason = "reached only through #[cfg(test)] helpers in this file (backlog#1823)")]
    async fn remove_and_save_with<S>(
        handle: &Arc<RwLock<Self>>,
        api: Arc<S>,
@@ -3574,6 +3562,7 @@ impl TierConfigMgr {
        Self::update_candidate_with_config_lock(handle, api, TierCandidateMutation::Clear(force)).await
    }

+    #[allow(dead_code, reason = "reached only through #[cfg(test)] helpers in this file (backlog#1823)")]
    async fn clear_and_save_with<S>(
        handle: &Arc<RwLock<Self>>,
        api: Arc<S>,
@@ -3612,6 +3601,10 @@ impl TierConfigMgr {
    }

    #[cfg(test)]
+    #[allow(
+        dead_code,
+        reason = "lease accounting asserted by a bucket_lifecycle_ops test behind `--features test-util` (backlog#1823)"
+    )]
    pub(crate) async fn active_operation_lease_count(handle: &Arc<RwLock<Self>>, tier_name: &str) -> usize {
        let manager = handle.read().await;
        let Some(runtime) = registered_tier_driver_runtime(&manager) else {
@@ -3717,10 +3710,6 @@ impl TierConfigMgr {
        Ok(())
    }

-    fn retire_driver(&mut self, tier_name: &str) {
-        self.revoke_driver(tier_name);
-    }
-
    fn revoke_all_drivers(&mut self) {
        if let Some(runtime) = registered_tier_driver_runtime(self) {
            let mut runtime = lock_unpoisoned(&runtime);
@@ -3884,6 +3873,7 @@ impl TierConfigMgr {
        self.save_config(api, &config_file, data).await
    }

+    #[allow(dead_code, reason = "reached only through #[cfg(test)] helpers in this file (backlog#1823)")]
    async fn save_tiering_config_if_current<S>(
        &self,
        api: Arc<S>,
@@ -305,6 +305,10 @@ impl TierMutationIntent {
    }
 }

+#[allow(
+    dead_code,
+    reason = "intent-record persistence asserted by store::init tests (backlog#1823)"
+)]
 pub(crate) fn tier_mutation_intent_record_object_name(mutation_id: Uuid) -> Result<String> {
    tier_mutation_intent_record_object_name_with_prefix(TIER_MUTATION_INTENT_RECORD_PREFIX, mutation_id)
 }
@@ -317,6 +321,10 @@ fn tier_mutation_intent_record_object_name_with_prefix(prefix: &str, mutation_id
    Ok(format!("{}/{}/{}/{}.json", prefix, &mutation_key[..2], &mutation_key[2..4], mutation_key))
 }

+#[allow(
+    dead_code,
+    reason = "intent-record persistence asserted by store::init tests (backlog#1823)"
+)]
 pub(crate) fn tier_mutation_intent_id_from_record_object_name(object: &str) -> Result<Uuid> {
    tier_mutation_intent_id_from_record_object_name_with_prefix(TIER_MUTATION_INTENT_RECORD_PREFIX, object)
 }
@@ -355,6 +363,10 @@ fn tier_mutation_intent_id_from_record_object_name_with_prefix(prefix: &str, obj
    Uuid::parse_str(mutation_key).map_err(|_| TierMutationIntentError::Corrupt("intent record path has invalid uuid"))
 }

+#[allow(
+    dead_code,
+    reason = "intent-record persistence asserted by store::init tests (backlog#1823)"
+)]
 pub(crate) async fn save_tier_mutation_intent_record<S>(api: Arc<S>, intent: &TierMutationIntent) -> EcstoreResult<()>
 where
    S: EcstoreObjectIO,
@@ -446,6 +458,10 @@ where
    Ok((intent, etag))
 }

+#[allow(
+    dead_code,
+    reason = "intent-record persistence asserted by store::init tests (backlog#1823)"
+)]
 pub(crate) async fn save_tier_mutation_intent_record_if_current<S>(
    api: Arc<S>,
    intent: &TierMutationIntent,
@@ -41,10 +41,7 @@ use crate::services::tier::{
 };
 use tracing::warn;

-const MAX_MULTIPART_PUT_OBJECT_SIZE: i64 = 1024 * 1024 * 1024 * 1024 * 5;
-const MAX_PARTS_COUNT: i64 = 10000;
 const _MAX_PART_SIZE: i64 = 1024 * 1024 * 1024 * 5;
-const MIN_PART_SIZE: i64 = 1024 * 1024 * 128;

 fn parse_generation(remote_version: &str) -> Result<Option<i64>, Error> {
    if remote_version.is_empty() {
@@ -64,7 +61,6 @@ pub struct WarmBackendGCS {
    pub control: Arc<StorageControl>,
    pub bucket: String,
    pub prefix: String,
-    pub storage_class: String,
 }

 impl WarmBackendGCS {
@@ -104,7 +100,6 @@ impl WarmBackendGCS {
            control,
            bucket: conf.bucket.clone(),
            prefix: conf.prefix.strip_suffix("/").unwrap_or(&conf.prefix).to_owned(),
-            storage_class: "".to_string(),
        })
    }

@@ -33,8 +33,6 @@ use crate::client::{
    transition_api::{BucketLookupType, Options, TransitionClient, TransitionCore},
    transition_api::{ReadCloser, ReaderImpl},
 };
-use crate::error::ErrorResponse;
-use crate::error::error_resp_to_object_err;
 use crate::services::tier::{
    tier_config::TierS3,
    warm_backend::{
@@ -1,200 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-#![allow(unused_imports)]
-#![allow(unused_variables)]
-#![allow(unused_mut)]
-#![allow(unused_assignments)]
-#![allow(unused_must_use)]
-#![allow(clippy::all)]
-
-use std::collections::HashMap;
-use std::sync::Arc;
-use url::Url;
-
-use aws_config::meta::region::RegionProviderChain;
-use aws_sdk_s3::Client;
-use aws_sdk_s3::config::{Credentials, Region};
-use aws_sdk_s3::primitives::ByteStream;
-
-use crate::client::{
-    api_get_options::GetObjectOptions,
-    api_put_object::PutObjectOptions,
-    api_remove::RemoveObjectOptions,
-    transition_api::{ReadCloser, ReaderImpl},
-};
-use crate::error::ErrorResponse;
-use crate::error::error_resp_to_object_err;
-use crate::services::tier::{
-    tier_config::TierS3,
-    warm_backend::{WarmBackend, WarmBackendGetOpts},
-};
-
-pub struct WarmBackendS3 {
-    pub client: Arc<Client>,
-    pub bucket: String,
-    pub prefix: String,
-    pub storage_class: String,
-}
-
-impl WarmBackendS3 {
-    pub async fn new(conf: &TierS3, tier: &str) -> Result<Self, std::io::Error> {
-        let u = match Url::parse(&conf.endpoint) {
-            Ok(u) => u,
-            Err(err) => {
-                return Err(std::io::Error::other(err.to_string()));
-            }
-        };
-
-        if conf.aws_role_web_identity_token_file == "" && conf.aws_role_arn != ""
-            || conf.aws_role_web_identity_token_file != "" && conf.aws_role_arn == ""
-        {
-            return Err(std::io::Error::other("both the token file and the role ARN are required"));
-        } else if conf.access_key == "" && conf.secret_key != "" || conf.access_key != "" && conf.secret_key == "" {
-            return Err(std::io::Error::other("both the access and secret keys are required"));
-        } else if conf.aws_role
-            && (conf.aws_role_web_identity_token_file != ""
-                || conf.aws_role_arn != ""
-                || conf.access_key != ""
-                || conf.secret_key != "")
-        {
-            return Err(std::io::Error::other(
-                "AWS Role cannot be activated with static credentials or the web identity token file",
-            ));
-        } else if conf.bucket == "" {
-            return Err(std::io::Error::other("no bucket name was provided"));
-        }
-
-        let creds;
-        if conf.access_key != "" && conf.secret_key != "" {
-            creds = Credentials::new(
-                conf.access_key.clone(), // access_key_id
-                conf.secret_key.clone(), // secret_access_key
-                None,                    // session_token (optional)
-                None,
-                "Static",
-            );
-        } else {
-            return Err(std::io::Error::other("insufficient parameters for S3 backend authentication"));
-        }
-        let region_provider = RegionProviderChain::default_provider().or_else(Region::new(conf.region.clone()));
-        #[allow(deprecated)]
-        let config = aws_config::from_env()
-            .endpoint_url(conf.endpoint.clone())
-            .region(region_provider)
-            .credentials_provider(creds)
-            .load()
-            .await;
-        let client = Client::new(&config);
-        let client = Arc::new(client);
-        Ok(Self {
-            client,
-            bucket: conf.bucket.clone(),
-            prefix: conf.prefix.clone().trim_matches('/').to_string(),
-            storage_class: conf.storage_class.clone(),
-        })
-    }
-
-    pub fn get_dest(&self, object: &str) -> String {
-        let mut dest_obj = object.to_string();
-        if self.prefix != "" {
-            dest_obj = format!("{}/{}", &self.prefix, object);
-        }
-        return dest_obj;
-    }
-}
-
-#[async_trait::async_trait]
-impl WarmBackend for WarmBackendS3 {
-    async fn put_with_meta(
-        &self,
-        object: &str,
-        r: ReaderImpl,
-        length: i64,
-        meta: HashMap<String, String>,
-    ) -> Result<String, std::io::Error> {
-        let client = self.client.clone();
-        let Ok(res) = client
-            .put_object()
-            .bucket(&self.bucket)
-            .key(&self.get_dest(object))
-            .body(match r {
-                ReaderImpl::Body(content_body) => ByteStream::from(content_body.to_vec()),
-                ReaderImpl::ObjectBody(mut content_body) => ByteStream::from(content_body.read_all().await?),
-            })
-            .send()
-            .await
-        else {
-            return Err(std::io::Error::other("put_object error"));
-        };
-
-        Ok(res.version_id().unwrap_or("").to_string())
-    }
-
-    async fn put(&self, object: &str, r: ReaderImpl, length: i64) -> Result<String, std::io::Error> {
-        self.put_with_meta(object, r, length, HashMap::new()).await
-    }
-
-    async fn get(&self, object: &str, rv: &str, opts: WarmBackendGetOpts) -> Result<ReadCloser, std::io::Error> {
-        let client = self.client.clone();
-        let mut req = client.get_object().bucket(&self.bucket).key(&self.get_dest(object));
-
-        if !rv.is_empty() {
-            req = req.version_id(rv);
-        }
-
-        if opts.start_offset >= 0 && opts.length > 0 {
-            let end = opts
-                .start_offset
-                .checked_add(opts.length)
-                .and_then(|v| v.checked_sub(1))
-                .ok_or_else(|| std::io::Error::other("invalid range: overflow"))?;
-            req = req.range(format!("bytes={}-{}", opts.start_offset, end));
-        }
-
-        let res = req.send().await.map_err(|e| std::io::Error::other(e.to_string()))?;
-
-        Ok(ReadCloser::new(std::io::Cursor::new(
-            res.body.collect().await.map(|data| data.into_bytes().to_vec())?,
-        )))
-    }
-
-    async fn remove(&self, object: &str, rv: &str) -> Result<(), std::io::Error> {
-        let client = self.client.clone();
-        let mut req = client.delete_object().bucket(&self.bucket).key(&self.get_dest(object));
-
-        if !rv.is_empty() {
-            req = req.version_id(rv);
-        }
-
-        req.send().await.map_err(|e| std::io::Error::other(e.to_string()))?;
-
-        Ok(())
-    }
-
-    async fn in_use(&self) -> Result<bool, std::io::Error> {
-        let client = self.client.clone();
-        let Ok(res) = client
-            .list_objects_v2()
-            .bucket(&self.bucket)
-            //.max_keys(10)
-            //.into_paginator()
-            .send()
-            .await
-        else {
-            return Err(std::io::Error::other("list_objects_v2 error"));
-        };
-
-        Ok(res.common_prefixes.unwrap_or_default().len() > 0 || res.contents.unwrap_or_default().len() > 0)
-    }
-}
@@ -32,20 +32,28 @@ use crate::diagnostics::get::{
    GET_METADATA_CACHE_REASON_NOT_READ_DATA, GET_METADATA_CACHE_REASON_PART_NUMBER,
    GET_METADATA_CACHE_REASON_RAW_DATA_MOVEMENT_READ, GET_METADATA_CACHE_REASON_USABLE, GET_METADATA_CACHE_REASON_VERSION_ID,
    GET_METADATA_CACHE_REASON_VERSION_SUSPENDED, GET_METADATA_CACHE_REASON_VERSIONED,
-    GET_METADATA_EARLY_STOP_REASON_CONFLICTING_METADATA, GET_METADATA_EARLY_STOP_REASON_DELETE_MARKER,
-    GET_METADATA_EARLY_STOP_REASON_ERROR, GET_METADATA_EARLY_STOP_REASON_INSUFFICIENT_QUORUM,
-    GET_METADATA_EARLY_STOP_REASON_NOT_FOUND, GET_METADATA_EARLY_STOP_REASON_UNSAFE_REQUEST,
-    GET_METADATA_EARLY_STOP_REASON_VALID_QUORUM, GET_METADATA_EARLY_STOP_REASON_VERSION_MATCH_QUORUM,
-    GET_METADATA_EARLY_STOP_REASON_VERSION_NOT_FOUND, GET_METADATA_RESPONSE_CORRUPT, GET_METADATA_RESPONSE_DISK_NOT_FOUND,
-    GET_METADATA_RESPONSE_ERROR, GET_METADATA_RESPONSE_IGNORED, GET_METADATA_RESPONSE_NOT_FOUND, GET_METADATA_RESPONSE_TIMEOUT,
-    GET_METADATA_RESPONSE_VALID, GET_METADATA_RESPONSE_VERSION_NOT_FOUND, GET_OBJECT_PATH_CODEC_STREAMING,
-    GET_OBJECT_PATH_DIRECT_MEMORY, GET_OBJECT_PATH_INTERNAL_META, GET_OBJECT_PATH_LEGACY_DUPLEX, GET_OBJECT_PATH_SET_DISK,
-    GET_STAGE_DECODE, GET_STAGE_METADATA_CACHE_LOOKUP, GET_STAGE_METADATA_RESOLVE, GET_STAGE_RANGE, GET_STAGE_READER_SETUP,
+    GET_METADATA_EARLY_STOP_REASON_CONFLICTING_METADATA, GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_BODY_VERIFY,
+    GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_DELETED, GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_GEOMETRY,
+    GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_IDENTITY_MISMATCH,
+    GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_PAYLOAD,
+    GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_SHARD, GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_NOT_INLINE,
+    GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_PART_SHAPE, GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_REMOTE,
+    GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_SIZE, GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_TRANSFORMED,
+    GET_METADATA_EARLY_STOP_REASON_DELETE_MARKER, GET_METADATA_EARLY_STOP_REASON_ERROR,
+    GET_METADATA_EARLY_STOP_REASON_INSUFFICIENT_QUORUM, GET_METADATA_EARLY_STOP_REASON_NOT_FOUND,
+    GET_METADATA_EARLY_STOP_REASON_UNSAFE_REQUEST, GET_METADATA_EARLY_STOP_REASON_VALID_QUORUM,
+    GET_METADATA_EARLY_STOP_REASON_VERSION_MATCH_QUORUM, GET_METADATA_EARLY_STOP_REASON_VERSION_NOT_FOUND,
+    GET_METADATA_RESPONSE_CORRUPT, GET_METADATA_RESPONSE_DISK_NOT_FOUND, GET_METADATA_RESPONSE_ERROR,
+    GET_METADATA_RESPONSE_IGNORED, GET_METADATA_RESPONSE_NOT_FOUND, GET_METADATA_RESPONSE_TIMEOUT, GET_METADATA_RESPONSE_VALID,
+    GET_METADATA_RESPONSE_VERSION_NOT_FOUND, GET_OBJECT_PATH_CODEC_STREAMING, GET_OBJECT_PATH_DIRECT_MEMORY,
+    GET_OBJECT_PATH_INTERNAL_META, GET_OBJECT_PATH_LEGACY_DUPLEX, GET_OBJECT_PATH_SET_DISK, GET_STAGE_DECODE,
+    GET_STAGE_METADATA_CACHE_LOOKUP, GET_STAGE_METADATA_RESOLVE, GET_STAGE_RANGE, GET_STAGE_READER_SETUP,
    GET_STAGE_READER_SETUP_DROP_PENDING, GET_STAGE_READER_SETUP_SCHEDULE, GET_STAGE_READER_SETUP_WAIT_QUORUM,
    GET_STAGE_READER_TASK_BITROT_READER_INIT, GET_STAGE_READER_TASK_FILE_OPEN, GET_STAGE_READER_TASK_READER_CONSTRUCTION,
    GetObjectFailureReason, classify_disk_error, get_stage_timer_if_enabled, record_get_object_pipeline_failure,
    record_get_object_pipeline_failure_for_path, record_get_stage_duration_if_enabled,
 };
+use crate::disk::disk_store::DiskStoreRenameDataExt;
 use crate::disk::local::DELETE_DATA_DIR_MARKER_PREFIX;
 use crate::disk::{
    DataDirDeleteStatus, OldCurrentSize, PART_TRANSACTION_NEW_META, PART_TRANSACTION_OLD_META, PART_TRANSACTION_ROLLBACK,
@@ -590,7 +598,7 @@ impl MetadataQuorumAccumulator {
        })
    }

-    pub(in crate::set_disk) fn default_write_quorum(&self) -> usize {
+    pub(crate) fn default_write_quorum(&self) -> usize {
        if self.default_parity_count == 0 || self.default_parity_count >= self.total_disks {
            return self.total_disks;
        }
@@ -651,36 +659,49 @@ pub(in crate::set_disk) fn metadata_early_stop_candidate_matches(left: &FileInfo
        && left.erasure.distribution == right.erasure.distribution
 }

-pub(in crate::set_disk) async fn data_read_early_stop_inline_body_verified(
+pub(in crate::set_disk) async fn data_read_early_stop_inline_body_miss_reason(
    bucket: &str,
    object: &str,
    candidate: &FileInfo,
    parts_metadata: &[FileInfo],
    disks: &[Option<DiskStore>],
-) -> bool {
-    if !candidate.inline_data()
-        || candidate.is_compressed()
+) -> Option<&'static str> {
+    // `inline_data` excludes remote objects; this diagnostic reports them separately.
+    if !rustfs_utils::http::contains_key_str(&candidate.metadata, rustfs_utils::http::SUFFIX_INLINE_DATA) {
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_NOT_INLINE);
+    }
+    if candidate.is_compressed()
        || candidate
            .metadata
            .keys()
            .any(|key| rustfs_utils::http::is_object_encryption_marker(key))
-        || candidate.is_remote()
-        || candidate.deleted
-        || candidate.size <= 0
-        || candidate.parts.len() != 1
-        || !candidate.has_valid_erasure_geometry()
    {
-        return false;
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_TRANSFORMED);
+    }
+    if candidate.is_remote() {
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_REMOTE);
+    }
+    if candidate.deleted {
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_DELETED);
+    }
+    if candidate.size <= 0 {
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_SIZE);
+    }
+    if candidate.parts.len() != 1 {
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_PART_SHAPE);
+    }
+    if !candidate.has_valid_erasure_geometry() {
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_GEOMETRY);
    }

    let Ok(object_size) = usize::try_from(candidate.size) else {
-        return false;
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_SIZE);
    };
    if candidate.parts.first().is_none_or(|part| part.size != object_size) {
-        return false;
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_PART_SHAPE);
    }
    if !can_try_inline_data_shards_direct(object_size, candidate.erasure.block_size) {
-        return false;
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_SIZE);
    }

    let Ok(erasure) = coding::Erasure::try_new_with_options(
@@ -689,18 +710,18 @@ pub(in crate::set_disk) async fn data_read_early_stop_inline_body_verified(
        candidate.erasure.block_size,
        candidate.uses_legacy_checksum,
    ) else {
-        return false;
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_GEOMETRY);
    };
-    let Some(data_files) =
-        collect_inline_data_shard_fileinfos_by_index(parts_metadata, candidate, erasure.data_shards, |index| {
+    let data_files =
+        match collect_inline_data_shard_fileinfos_by_index_or_reason(parts_metadata, candidate, erasure.data_shards, |index| {
            disks.get(index).is_some_and(Option::is_some)
-        })
-    else {
-        return false;
-    };
+        }) {
+            Ok(data_files) => data_files,
+            Err(reason) => return Some(reason),
+        };

    let Some(part) = candidate.parts.first() else {
-        return false;
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_PART_SHAPE);
    };
    let checksum_info = candidate.erasure.get_checksum_info(part.number);
    let checksum_algo = if candidate.uses_legacy_checksum && checksum_info.algorithm == HashAlgorithm::HighwayHash256S {
@@ -720,12 +741,70 @@ pub(in crate::set_disk) async fn data_read_early_stop_inline_body_verified(
    let Ok(mut readers) =
        build_inline_bitrot_readers_from_refs(&data_files, bucket, object, read_length, shard_size, &checksum_algo, false).await
    else {
-        return false;
+        return Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_BODY_VERIFY);
    };

-    try_read_inline_data_shards_direct(&mut readers, erasure.data_shards, read_length, object_size)
-        .await
-        .is_some_and(|body| body.len() == object_size)
+    match try_read_inline_data_shards_direct(&mut readers, erasure.data_shards, read_length, object_size).await {
+        Some(body) if body.len() == object_size => None,
+        _ => Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_BODY_VERIFY),
+    }
+}
+
+fn data_read_inline_missing_shards_are_pending(
+    candidate: &FileInfo,
+    parts_metadata: &[FileInfo],
+    errors: &[Option<DiskError>],
+    disks: &[Option<DiskStore>],
+    fanout_order: &[usize],
+    scheduled_fanout_len: usize,
+) -> bool {
+    let Ok(erasure) = coding::Erasure::try_new_with_options(
+        candidate.erasure.data_blocks,
+        candidate.erasure.parity_blocks,
+        candidate.erasure.block_size,
+        candidate.uses_legacy_checksum,
+    ) else {
+        return false;
+    };
+    let distribution = &candidate.erasure.distribution;
+    let mut data_shards_seen_or_pending = vec![false; erasure.data_shards];
+    let mut missing_pending_data_shards = 0usize;
+
+    for (disk_index, file_info) in parts_metadata.iter().enumerate() {
+        let Some(&block_index) = distribution.get(disk_index) else {
+            return false;
+        };
+        if block_index == 0 || block_index > erasure.data_shards {
+            continue;
+        }
+        if !disks.get(disk_index).is_some_and(Option::is_some) {
+            return false;
+        }
+
+        let data_slot = block_index - 1;
+        if file_info.name.is_empty() {
+            let scheduled_and_not_failed = fanout_order
+                .get(..scheduled_fanout_len)
+                .is_some_and(|scheduled_disks| scheduled_disks.contains(&disk_index))
+                && errors.get(disk_index).is_some_and(Option::is_none);
+            if scheduled_and_not_failed {
+                data_shards_seen_or_pending[data_slot] = true;
+                missing_pending_data_shards = missing_pending_data_shards.saturating_add(1);
+                continue;
+            }
+            return false;
+        }
+        if file_info.erasure.index != block_index
+            || !file_info.has_valid_erasure_geometry()
+            || !metadata_early_stop_candidate_matches(file_info, candidate)
+            || file_info.data.as_ref().is_none_or(|data| data.is_empty())
+        {
+            return false;
+        }
+        data_shards_seen_or_pending[data_slot] = true;
+    }
+
+    missing_pending_data_shards > 0 && data_shards_seen_or_pending.into_iter().all(|seen_or_pending| seen_or_pending)
 }

 pub(in crate::set_disk) fn classify_metadata_response_error(err: &DiskError) -> &'static str {
@@ -2468,6 +2547,7 @@ impl SetDisks {
        let mut next_fanout_index = 0usize;
        let mut scheduled_count = 0usize;
        let mut force_full_wait = false;
+        let mut final_miss_reason_override = None;
        let spawn_read_version =
            |join_set: &mut JoinSet<(usize, disk::error::Result<FileInfo>, Duration)>, index: usize, disk: Option<DiskStore>| {
                let task_opts = opts;
@@ -2510,6 +2590,7 @@ impl SetDisks {
        }

        while let Some(result) = join_set.join_next().await {
+            let mut defer_pending_inline_data_shard = false;
            match result {
                Ok((index, res, elapsed)) => match res {
                    Ok(file_info) => {
@@ -2540,17 +2621,43 @@ impl SetDisks {
                    .or_else(|| accumulator.version_early_stop_decision())
            {
                let should_return_early = if read_data {
-                    let allow_data_read_early_stop = match accumulator.candidate.as_ref() {
-                        Some(candidate) => {
-                            data_read_early_stop_inline_body_verified(bucket.as_ref(), object.as_ref(), candidate, &ress, disks)
-                                .await
+                    match accumulator.candidate.as_ref() {
+                        Some(candidate) => match data_read_early_stop_inline_body_miss_reason(
+                            bucket.as_ref(),
+                            object.as_ref(),
+                            candidate,
+                            &ress,
+                            disks,
+                        )
+                        .await
+                        {
+                            None => true,
+                            Some(reason) => {
+                                final_miss_reason_override = Some(reason);
+                                if bounded_fanout
+                                    && reason == GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_SHARD
+                                    && data_read_inline_missing_shards_are_pending(
+                                        candidate,
+                                        &ress,
+                                        &errors,
+                                        disks,
+                                        &fanout_order,
+                                        next_fanout_index,
+                                    )
+                                {
+                                    defer_pending_inline_data_shard = true;
+                                } else {
+                                    force_full_wait = true;
+                                }
+                                false
+                            }
+                        },
+                        None => {
+                            force_full_wait = true;
+                            final_miss_reason_override = Some(GET_METADATA_EARLY_STOP_REASON_INSUFFICIENT_QUORUM);
+                            false
                        }
-                        None => false,
-                    };
-                    if !allow_data_read_early_stop {
-                        force_full_wait = true;
                    }
-                    allow_data_read_early_stop
                } else {
                    true
                };
@@ -2587,6 +2694,7 @@ impl SetDisks {
            let pending_responses = join_set.len();
            let should_hedge_single_pending_data_read = read_data
                && !force_full_wait
+                && !defer_pending_inline_data_shard
                && pending_responses == 1
                && accumulator.can_still_reach_early_stop_with_pending(pending_responses);
            if bounded_fanout && force_full_wait {
@@ -2599,6 +2707,7 @@ impl SetDisks {
                    next_fanout_index = next_fanout_index.saturating_add(1);
                }
            } else if bounded_fanout
+                && !defer_pending_inline_data_shard
                && next_fanout_index < disks.len()
                && (!accumulator.can_still_reach_early_stop_with_pending(pending_responses)
                    || should_hedge_single_pending_data_read)
@@ -2612,7 +2721,12 @@ impl SetDisks {
            }
        }

-        rustfs_io_metrics::record_get_object_metadata_early_stop_miss(metrics_path, accumulator.final_miss_reason());
+        let accumulator_miss_reason = accumulator.final_miss_reason();
+        let final_miss_reason = match (final_miss_reason_override, accumulator_miss_reason) {
+            (Some(reason), GET_METADATA_EARLY_STOP_REASON_INSUFFICIENT_QUORUM) => reason,
+            _ => accumulator_miss_reason,
+        };
+        rustfs_io_metrics::record_get_object_metadata_early_stop_miss(metrics_path, final_miss_reason);
        rustfs_io_metrics::record_get_object_metadata_early_stop_saved_responses(metrics_path, 0);
        rustfs_io_metrics::record_get_object_metadata_fanout_lifecycle(metrics_path, scheduled_count, scheduled_count, 0);
        let diagnostics = MetadataFanoutDiagnostics::new(fanout_start.elapsed(), observations);
@@ -3011,12 +3125,41 @@ impl RenameConvergence {
    }
 }

+pub(in crate::set_disk) struct RenameDataCommit {
+    pub(in crate::set_disk) online_disks: Vec<Option<DiskStore>>,
+    pub(in crate::set_disk) convergence: RenameConvergence,
+    pub(in crate::set_disk) data_dir: Option<Uuid>,
+    pub(in crate::set_disk) cleanup_disks: Vec<Option<DiskStore>>,
+    pub(in crate::set_disk) old_current_size: Option<OldCurrentSize>,
+    pub(in crate::set_disk) committed_file_info: FileInfo,
+}
+
+type RenameDataLegacyTuple = (
+    Vec<Option<DiskStore>>,
+    RenameConvergence,
+    Option<Uuid>,
+    Vec<Option<DiskStore>>,
+    Option<OldCurrentSize>,
+);
+
+impl RenameDataCommit {
+    fn into_legacy_tuple(self) -> RenameDataLegacyTuple {
+        (
+            self.online_disks,
+            self.convergence,
+            self.data_dir,
+            self.cleanup_disks,
+            self.old_current_size,
+        )
+    }
+}
+
 impl SetDisks {
    pub(in crate::set_disk) fn default_read_quorum(&self) -> usize {
        self.set_drive_count - self.default_parity_count
    }

-    pub(in crate::set_disk) fn default_write_quorum(&self) -> usize {
+    pub(crate) fn default_write_quorum(&self) -> usize {
        let mut data_count = self.set_drive_count - self.default_parity_count;
        if data_count == self.default_parity_count {
            data_count += 1
@@ -3025,6 +3168,97 @@ impl SetDisks {
        data_count
    }

+    pub(in crate::set_disk) async fn prepare_quota_mutation_fences(
+        disks: &[Option<DiskStore>],
+        bucket: &str,
+        object: &str,
+        write_quorum: usize,
+    ) -> crate::error::Result<(Vec<Option<DiskStore>>, Vec<Option<SnapshotLeaseToken>>)> {
+        let fence_path = crate::disk::quota_mutation_fence_path(bucket, object);
+        let results = join_all(disks.iter().map(|disk| {
+            let disk = disk.clone();
+            let fence_path = fence_path.clone();
+            async move {
+                let disk = disk?;
+                match disk.acquire_snapshot_lease(RUSTFS_META_BUCKET, &fence_path).await {
+                    Ok(token) => Some((disk, token)),
+                    Err(_) => None,
+                }
+            }
+        }))
+        .await;
+        if results.iter().flatten().count() < write_quorum {
+            for (disk, token) in results.iter().flatten() {
+                let _ = disk.release_snapshot_lease(RUSTFS_META_BUCKET, &fence_path, *token).await;
+            }
+            return Err(StorageError::ErasureWriteQuorum);
+        }
+        let mut fenced_disks = Vec::with_capacity(results.len());
+        let mut tokens = Vec::with_capacity(results.len());
+        for result in results {
+            match result {
+                Some((disk, token)) => {
+                    fenced_disks.push(Some(disk));
+                    tokens.push(Some(token));
+                }
+                None => {
+                    fenced_disks.push(None);
+                    tokens.push(None);
+                }
+            }
+        }
+        Ok((fenced_disks, tokens))
+    }
+
+    pub(in crate::set_disk) async fn release_quota_mutation_fences(
+        disks: &[Option<DiskStore>],
+        tokens: &[Option<SnapshotLeaseToken>],
+        bucket: &str,
+        object: &str,
+        write_quorum: usize,
+    ) -> crate::error::Result<()> {
+        let fence_path = crate::disk::quota_mutation_fence_path(bucket, object);
+        let results = join_all(disks.iter().zip(tokens).filter_map(|(disk, token)| {
+            let disk = disk.as_ref()?.clone();
+            let token = (*token)?;
+            let fence_path = fence_path.clone();
+            Some(async move { disk.release_snapshot_lease(RUSTFS_META_BUCKET, &fence_path, token).await })
+        }))
+        .await;
+        if results.iter().filter(|result| result.is_ok()).count() < write_quorum {
+            return Err(StorageError::ErasureWriteQuorum);
+        }
+        Ok(())
+    }
+
+    pub(in crate::set_disk) fn assign_rename_data_indexes(file_infos: &mut [FileInfo]) {
+        for (index, file_info) in file_infos.iter_mut().enumerate() {
+            if file_info.erasure.index == 0 {
+                file_info.erasure.index = index + 1;
+            }
+        }
+    }
+
+    pub(in crate::set_disk) async fn abort_quota_reservation_after_fence(
+        reservation: crate::bucket::quota::reservation::QuotaReservation,
+        disks: &[Option<DiskStore>],
+        tokens: &[Option<SnapshotLeaseToken>],
+        bucket: &str,
+        object: &str,
+        write_quorum: usize,
+        fenced: bool,
+    ) {
+        let safe_to_abort = !fenced
+            || Self::release_quota_mutation_fences(disks, tokens, bucket, object, write_quorum)
+                .await
+                .is_ok();
+        if safe_to_abort {
+            reservation.abort().await;
+        } else {
+            reservation.defer_after_fence();
+        }
+    }
+
    #[tracing::instrument(level = "debug", skip(disks, file_infos))]
    #[allow(clippy::type_complexity)]
    pub(in crate::set_disk) async fn rename_data(
@@ -3035,13 +3269,22 @@ impl SetDisks {
        dst_bucket: &str,
        dst_object: &str,
        write_quorum: usize,
-    ) -> disk::error::Result<(
-        Vec<Option<DiskStore>>,
-        RenameConvergence,
-        Option<Uuid>,
-        Vec<Option<DiskStore>>,
-        Option<OldCurrentSize>,
-    )> {
+    ) -> disk::error::Result<RenameDataLegacyTuple> {
+        Self::rename_data_owned(disks, src_bucket, src_object, file_infos.to_vec(), dst_bucket, dst_object, write_quorum)
+            .await
+            .map(RenameDataCommit::into_legacy_tuple)
+    }
+
+    #[tracing::instrument(level = "debug", skip(disks, file_infos))]
+    pub(in crate::set_disk) async fn rename_data_owned(
+        disks: &[Option<DiskStore>],
+        src_bucket: &str,
+        src_object: &str,
+        file_infos: Vec<FileInfo>,
+        dst_bucket: &str,
+        dst_object: &str,
+        write_quorum: usize,
+    ) -> disk::error::Result<RenameDataCommit> {
        if let Some(file_info) = disks
            .iter()
            .zip(file_infos.iter())
@@ -3066,7 +3309,7 @@ impl SetDisks {

        let disk_count = disks.len();
        let fanout_disks = disks.to_vec();
-        let fanout_file_infos = file_infos.to_vec();
+        let fanout_file_infos = file_infos;
        let fanout_src_bucket = src_bucket.clone();
        let fanout_src_object = src_object.clone();
        let fanout_dst_bucket = dst_bucket.clone();
@@ -3078,9 +3321,9 @@ impl SetDisks {
        let fanout = tokio::spawn(async move {
            let futures = fanout_disks
                .into_iter()
-                .zip(fanout_file_infos)
+                .zip(fanout_file_infos.iter())
                .enumerate()
-                .map(|(i, (disk, mut file_info))| {
+                .map(|(i, (disk, file_info))| {
                    let src_bucket = fanout_src_bucket.clone();
                    let src_object = fanout_src_object.clone();
                    let dst_object = fanout_dst_object.clone();
@@ -3097,11 +3340,15 @@ impl SetDisks {
                        };

                        let is_delete_marker = file_info.is_canonical_delete_marker();
-                        if file_info.erasure.index == 0 {
-                            file_info.erasure.index = i + 1;
-                        }
-
-                        if !is_delete_marker && !file_info.has_valid_erasure_geometry() {
+                        let mut local_file_info;
+                        let file_info = if file_info.erasure.index == 0 {
+                            local_file_info = file_info.clone();
+                            local_file_info.erasure.index = i + 1;
+                            &local_file_info
+                        } else {
+                            file_info
+                        };
+                        if file_info.erasure.index == 0 || (!is_delete_marker && !file_info.has_valid_erasure_geometry()) {
                            return Err(DiskError::FileCorrupt);
                        }

@@ -3109,12 +3356,13 @@ impl SetDisks {
                        // A no-op immediately-ready future in production.
                        Self::rename_fanout_barrier(&dst_object, i, rename_fanout_barrier_phase::RENAME).await;

-                        disk.rename_data(&src_bucket, &src_object, file_info, &dst_bucket, &dst_object)
+                        disk.rename_data_borrowed(&src_bucket, &src_object, file_info, &dst_bucket, &dst_object)
                            .await
                    })
                    .catch_unwind()
                });
-            join_all(futures).await
+            let results = join_all(futures).await;
+            (results, fanout_file_infos)
        });

        let mut disk_versions = vec![None; disk_count];
@@ -3122,7 +3370,7 @@ impl SetDisks {
        let mut cleanup_data_dirs = vec![None; disk_count];
        let mut old_current_sizes = vec![None; disk_count];

-        let results = fanout.await.map_err(|_| DiskError::Unexpected)?;
+        let (results, mut file_infos) = fanout.await.map_err(|_| DiskError::Unexpected)?;

        for (idx, result) in results.iter().enumerate() {
            match result {
@@ -3178,7 +3426,7 @@ impl SetDisks {
                }

                if let Some(disk) = disks[i].as_ref() {
-                    let fi = file_infos[i].clone();
+                    let fi = std::mem::take(&mut file_infos[i]);
                    let old_data_dir = data_dirs[i];
                    let disk = disk.clone();
                    let dst_bucket = dst_bucket.clone();
@@ -3301,6 +3549,8 @@ impl SetDisks {
        let convergence = Self::classify_rename_convergence(&disk_versions, &errs);
        let old_current_size = Self::reduce_common_old_current_size(&old_current_sizes, write_quorum);
        let online_disks = Self::eval_disks(disks, &errs);
+        let committed_slot = online_disks.iter().position(Option::is_some).ok_or(DiskError::Unexpected)?;
+        let committed_file_info = std::mem::take(&mut file_infos[committed_slot]);
        let cleanup_disks = if let Some(data_dir) = data_dir {
            disks
                .iter()
@@ -3318,7 +3568,14 @@ impl SetDisks {
            vec![None; disks.len()]
        };

-        Ok((online_disks, convergence, data_dir, cleanup_disks, old_current_size))
+        Ok(RenameDataCommit {
+            online_disks,
+            convergence,
+            data_dir,
+            cleanup_disks,
+            old_current_size,
+            committed_file_info,
+        })
    }

    /// rustfs/backlog#1009: reduce the per-disk observations of the
@@ -5608,9 +5865,20 @@ mod tests {
        object: &str,
        payload: &[u8],
        uses_legacy_checksum: bool,
+    ) -> Vec<FileInfo> {
+        inline_metadata_fanout_fileinfos_with_geometry(bucket, object, payload, uses_legacy_checksum, 2, 2).await
+    }
+
+    async fn inline_metadata_fanout_fileinfos_with_geometry(
+        bucket: &str,
+        object: &str,
+        payload: &[u8],
+        uses_legacy_checksum: bool,
+        data_shards: usize,
+        parity_shards: usize,
    ) -> Vec<FileInfo> {
        let distribution_key = metadata_distribution_key(bucket, object);
-        let mut base = FileInfo::new(&distribution_key, 2, 2);
+        let mut base = FileInfo::new(&distribution_key, data_shards, parity_shards);
        base.volume = bucket.to_string();
        base.name = object.to_string();
        base.size = i64::try_from(payload.len()).expect("test payload should fit i64");
@@ -5673,6 +5941,21 @@ mod tests {
        install_inline_metadata_fanout_files(disks, bucket, object, files).await;
    }

+    async fn install_inline_metadata_fanout_fileinfo_with_geometry(
+        disks: &[Option<DiskStore>],
+        bucket: &str,
+        object: &str,
+        payload: &[u8],
+        data_shards: usize,
+        parity_shards: usize,
+        mutate: impl FnOnce(&mut [FileInfo]),
+    ) {
+        let mut files =
+            inline_metadata_fanout_fileinfos_with_geometry(bucket, object, payload, false, data_shards, parity_shards).await;
+        mutate(&mut files);
+        install_inline_metadata_fanout_files(disks, bucket, object, files).await;
+    }
+
    async fn install_inline_metadata_fanout_files(disks: &[Option<DiskStore>], bucket: &str, object: &str, files: Vec<FileInfo>) {
        let distribution = files
            .first()
@@ -5893,6 +6176,118 @@ mod tests {
        drop(dirs);
    }

+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn bounded_metadata_early_stop_waits_for_pending_inline_data_shard() {
+        const DISKS: usize = 6;
+        const DATA_SHARDS: usize = 4;
+        const PARITY_SHARDS: usize = 2;
+        let bucket = "bounded-inline-data-get-pending-shard-bucket";
+        let object =
+            object_with_initial_data_shards(bucket, "bounded-inline-data-get-pending-shard-object", DATA_SHARDS, DATA_SHARDS);
+        let (dirs, disks) = call_counter_local_disks(bucket, DISKS).await;
+        install_inline_metadata_fanout_fileinfo_with_geometry(
+            &disks,
+            bucket,
+            &object,
+            b"verified inline payload",
+            DATA_SHARDS,
+            PARITY_SHARDS,
+            |_| {},
+        )
+        .await;
+
+        temp_env::async_with_vars(
+            [
+                ("RUSTFS_GET_METADATA_EARLY_STOP_ENABLE", Some("true")),
+                ("RUSTFS_GET_METADATA_DATA_READ_EARLY_STOP_ENABLE", Some("true")),
+                ("RUSTFS_GET_METADATA_EARLY_STOP_BOUNDED_FANOUT", Some("true")),
+            ],
+            async {
+                let fanout_order = bounded_metadata_fanout_order(bucket, &object, DISKS, PARITY_SHARDS);
+                let distribution_key = metadata_distribution_key(bucket, &object);
+                let distribution = FileInfo::new(&distribution_key, DATA_SHARDS, PARITY_SHARDS)
+                    .erasure
+                    .distribution;
+                let paused_data_disk = *fanout_order
+                    .iter()
+                    .take(DATA_SHARDS)
+                    .find(|disk_index| {
+                        distribution
+                            .get(**disk_index)
+                            .is_some_and(|block_index| (1..=DATA_SHARDS).contains(block_index))
+                    })
+                    .expect("initial fanout should include a data shard to pause");
+                let hedged_parity_disk = fanout_order[DATA_SHARDS];
+                let unscheduled_parity_disk = fanout_order[DATA_SHARDS + 1];
+
+                let barrier = rename_fanout_barrier::arm(&object, paused_data_disk, rename_fanout_barrier::PHASE_READ_VERSION);
+                let tracker = rename_fanout_barrier::observe_tasks(&object);
+                let calls = disk_call_counters::observe(&object);
+                let disks_for_read = disks.clone();
+                let object_for_read = object.clone();
+                let mut read = tokio::spawn(async move {
+                    SetDisks::read_all_fileinfo_observed(
+                        &disks_for_read,
+                        bucket,
+                        bucket,
+                        &object_for_read,
+                        "",
+                        true,
+                        false,
+                        false,
+                        true,
+                        PARITY_SHARDS,
+                    )
+                    .await
+                });
+
+                tokio::time::timeout(BARRIER_PAUSE_GUARD, barrier.wait_until_paused())
+                    .await
+                    .expect("initial data shard should pause before returning");
+                tokio::time::timeout(BARRIER_PAUSE_GUARD, async {
+                    while calls.for_disk(disk_call_counters::KIND_READ_VERSION, hedged_parity_disk) == 0 {
+                        tokio::task::yield_now().await;
+                    }
+                })
+                .await
+                .expect("bounded fanout should hedge one parity disk while the data shard is pending");
+
+                assert!(
+                    tokio::time::timeout(BARRIER_PAUSE_GUARD, &mut read).await.is_err(),
+                    "inline data-read early-stop must wait for a scheduled missing data shard instead of forcing full wait"
+                );
+
+                barrier.release();
+                let (parts_metadata, errs, diagnostics) = read
+                    .await
+                    .expect("metadata read task should not panic")
+                    .expect("pending data shard should let the inline verifier finish");
+
+                assert_eq!(
+                    calls.total(disk_call_counters::KIND_READ_VERSION),
+                    5,
+                    "pending data-shard defer should not schedule the final parity disk"
+                );
+                assert_eq!(
+                    calls.for_disk(disk_call_counters::KIND_READ_VERSION, unscheduled_parity_disk),
+                    0,
+                    "the remaining parity disk must stay unissued when pending data verification succeeds"
+                );
+                assert_eq!(
+                    tracker.running(),
+                    0,
+                    "early-stop should drain spawned read_version tasks before returning"
+                );
+                assert_eq!(diagnostics.total_responses(), 5);
+                assert_eq!(parts_metadata.iter().filter(|fi| fi.name == object).count(), 5);
+                assert!(errs.iter().all(Option::is_none));
+            },
+        )
+        .await;
+
+        drop(dirs);
+    }
+
    #[tokio::test]
    async fn data_read_early_stop_verifies_legacy_inline_checksum_payload() {
        let bucket = "legacy-inline-data-get-fanout-bucket";
@@ -5923,11 +6318,133 @@ mod tests {
            .clone();

        assert!(
-            data_read_early_stop_inline_body_verified(bucket, object, &candidate, &parts_metadata, &disks).await,
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &candidate, &parts_metadata, &disks)
+                .await
+                .is_none(),
            "legacy inline metadata must use the legacy bitrot shard sizing and checksum algorithm"
        );
    }

+    #[tokio::test]
+    async fn data_read_early_stop_reports_inline_miss_reasons() {
+        let bucket = "inline-data-get-miss-reason-bucket";
+        let object = "inline-data-get-miss-reason-object";
+        let payload = b"verified inline payload";
+        let (_dirs, disks) = call_counter_local_disks(bucket, 4).await;
+        let files = inline_metadata_fanout_fileinfos_with_mode(bucket, object, payload, false).await;
+        let distribution = files
+            .first()
+            .map(|file| file.erasure.distribution.clone())
+            .expect("fixture should include metadata");
+        let order = bounded_metadata_fanout_order(bucket, object, 4, 2);
+        let mut parts_metadata = vec![FileInfo::default(); 4];
+        for disk_index in order.into_iter().take(3) {
+            let block_index = distribution
+                .get(disk_index)
+                .copied()
+                .expect("fixture distribution should cover every disk");
+            parts_metadata[disk_index] = files
+                .get(block_index.checked_sub(1).expect("erasure block indexes are one-based"))
+                .expect("fixture should include every distributed shard")
+                .clone();
+        }
+        let candidate = parts_metadata
+            .iter()
+            .find(|file| file.name == object)
+            .expect("fixture should include observed metadata")
+            .clone();
+        let data_disk = distribution
+            .iter()
+            .position(|block_index| *block_index == 1)
+            .expect("fixture distribution should include first data shard");
+
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &candidate, &parts_metadata, &disks).await,
+            None
+        );
+
+        let mut not_inline = candidate.clone();
+        rustfs_utils::http::remove_str(&mut not_inline.metadata, rustfs_utils::http::SUFFIX_INLINE_DATA);
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &not_inline, &parts_metadata, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_NOT_INLINE)
+        );
+
+        let mut remote = candidate.clone();
+        remote.transition_status = TRANSITION_COMPLETE.to_string();
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &remote, &parts_metadata, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_REMOTE)
+        );
+
+        let mut transformed = candidate.clone();
+        rustfs_utils::http::insert_str(&mut transformed.metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &transformed, &parts_metadata, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_TRANSFORMED)
+        );
+
+        let mut deleted = candidate.clone();
+        deleted.deleted = true;
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &deleted, &parts_metadata, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_DELETED)
+        );
+
+        let mut zero_size = candidate.clone();
+        zero_size.size = 0;
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &zero_size, &parts_metadata, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_SIZE)
+        );
+
+        let mut multipart = candidate.clone();
+        multipart.parts.push(multipart.parts[0].clone());
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &multipart, &parts_metadata, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_PART_SHAPE)
+        );
+
+        let mut invalid_geometry = candidate.clone();
+        invalid_geometry.erasure.data_blocks = 0;
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &invalid_geometry, &parts_metadata, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_GEOMETRY)
+        );
+
+        let mut missing_shard = parts_metadata.clone();
+        missing_shard[data_disk] = FileInfo::default();
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &candidate, &missing_shard, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_SHARD)
+        );
+
+        let mut missing_payload = parts_metadata.clone();
+        missing_payload[data_disk].data = None;
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &candidate, &missing_payload, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_PAYLOAD)
+        );
+
+        let mut identity_mismatch = parts_metadata.clone();
+        identity_mismatch[data_disk].version_id = Some(Uuid::new_v4());
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &candidate, &identity_mismatch, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_IDENTITY_MISMATCH)
+        );
+
+        let mut corrupt = parts_metadata.clone();
+        if let Some(data) = corrupt[data_disk].data.as_mut() {
+            let mut corrupt_data = data.to_vec();
+            corrupt_data[0] ^= 0x01;
+            *data = Bytes::from(corrupt_data);
+        }
+        assert_eq!(
+            data_read_early_stop_inline_body_miss_reason(bucket, object, &candidate, &corrupt, &disks).await,
+            Some(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_BODY_VERIFY)
+        );
+    }
+
    #[test]
    #[serial_test::serial]
    fn metadata_fanout_lifecycle_records_real_early_stop_abort() {
@@ -6017,7 +6534,7 @@ mod tests {
                &[
                    ("path", GET_OBJECT_PATH_INTERNAL_META),
                    ("decision", "miss"),
-                    ("reason", GET_METADATA_EARLY_STOP_REASON_INSUFFICIENT_QUORUM),
+                    ("reason", GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_BODY_VERIFY),
                ],
            ),
            1,
@@ -6029,7 +6546,7 @@ mod tests {
                &[
                    ("path", GET_OBJECT_PATH_LEGACY_DUPLEX),
                    ("decision", "miss"),
-                    ("reason", GET_METADATA_EARLY_STOP_REASON_INSUFFICIENT_QUORUM),
+                    ("reason", GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_BODY_VERIFY),
                ],
            ),
            0,
@@ -59,7 +59,10 @@ use crate::client::{object_api_utils::get_raw_etag, transition_api::ReaderImpl};
 use crate::cluster::rpc::heal_bucket_local_on_disks;
 use crate::data_usage::record_compression_total_memory;
 use crate::diagnostics::get::{
-    GET_CODEC_STREAMING_OBJECT_CLASS_PLAIN_SINGLE_PART, GET_OBJECT_PATH_BODY_CACHE, GET_OBJECT_PATH_CODEC_STREAMING,
+    GET_CODEC_STREAMING_OBJECT_CLASS_PLAIN_SINGLE_PART, GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_GEOMETRY,
+    GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_IDENTITY_MISMATCH,
+    GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_PAYLOAD,
+    GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_SHARD, GET_OBJECT_PATH_BODY_CACHE, GET_OBJECT_PATH_CODEC_STREAMING,
    GET_OBJECT_PATH_CODEC_STREAMING_LEGACY_ENGINE, GET_OBJECT_PATH_CODEC_STREAMING_RUSTFS_ENGINE, GET_OBJECT_PATH_DIRECT_MEMORY,
    GET_OBJECT_PATH_EMPTY, GET_OBJECT_PATH_INLINE_DIRECT, GET_OBJECT_PATH_INTERNAL_META, GET_OBJECT_PATH_LEGACY_DUPLEX,
    GET_OBJECT_PATH_REMOTE_TRANSITION, GET_OBJECT_PATH_SET_DISK, GET_STAGE_DECODE, GET_STAGE_EMIT, GET_STAGE_INLINE_PREPARE,
@@ -719,8 +722,8 @@ pub(crate) use core::io_primitives::disk_call_counters;
 mod ctx;
 mod metadata;
 mod ops;
-#[cfg(test)]
-pub(crate) use ops::multipart::{MultipartCommitBarrier, MultipartCommitPause};
+#[cfg(any(test, feature = "test-util"))]
+pub use ops::multipart::{MultipartCommitBarrier, MultipartCommitPause};
 #[cfg(feature = "test-util")]
 pub(crate) use ops::object::TransitionCleanupStoreBarrier as SetDiskTransitionCleanupStoreBarrier;
 pub(crate) use ops::object::body_cache_plaintext_len;
@@ -3866,8 +3869,17 @@ fn collect_inline_data_shard_fileinfos_by_index<'a>(
    parts_metadata: &'a [FileInfo],
    fi: &FileInfo,
    data_shards: usize,
-    mut disk_is_online: impl FnMut(usize) -> bool,
+    disk_is_online: impl FnMut(usize) -> bool,
 ) -> Option<Vec<&'a FileInfo>> {
+    collect_inline_data_shard_fileinfos_by_index_or_reason(parts_metadata, fi, data_shards, disk_is_online).ok()
+}
+
+fn collect_inline_data_shard_fileinfos_by_index_or_reason<'a>(
+    parts_metadata: &'a [FileInfo],
+    fi: &FileInfo,
+    data_shards: usize,
+    mut disk_is_online: impl FnMut(usize) -> bool,
+) -> std::result::Result<Vec<&'a FileInfo>, &'static str> {
    let distribution = &fi.erasure.distribution;
    let mut data_files = vec![None; data_shards];

@@ -3875,27 +3887,35 @@ fn collect_inline_data_shard_fileinfos_by_index<'a>(
        if !disk_is_online(disk_index) {
            continue;
        }
-        let block_index = *distribution.get(disk_index)?;
+        let Some(&block_index) = distribution.get(disk_index) else {
+            return Err(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_GEOMETRY);
+        };
        if block_index == 0 || block_index > data_shards {
            continue;
        }
+        if file_info.name.is_empty() {
+            return Err(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_SHARD);
+        }
        if file_info.erasure.index != block_index {
-            continue;
+            return Err(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_IDENTITY_MISMATCH);
        }
        if !file_info.has_valid_erasure_geometry() {
-            continue;
+            return Err(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_GEOMETRY);
        }
        if !core::io_primitives::metadata_early_stop_candidate_matches(file_info, fi) {
-            continue;
+            return Err(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_IDENTITY_MISMATCH);
        }
        if file_info.data.as_ref().is_none_or(|data| data.is_empty()) {
-            continue;
+            return Err(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_PAYLOAD);
        }

        data_files[block_index - 1] = Some(file_info);
    }

-    data_files.into_iter().collect()
+    data_files
+        .into_iter()
+        .collect::<Option<Vec<_>>>()
+        .ok_or(GET_METADATA_EARLY_STOP_REASON_DATA_READ_INLINE_MISSING_SHARD)
 }

 impl SetDisks {
@@ -13,6 +13,7 @@
 // limitations under the License.

 use super::super::*;
+use crate::disk::disk_store::DiskStoreRenameDataExt;
 use crate::io_support::bitrot::object_mmap_read_enabled;
 use crate::storage_api_contracts::namespace::NamespaceLocking as _;
 use tracing::trace;
@@ -1164,10 +1165,10 @@ impl SetDisks {
                                let rename_result = if should_fail_heal_rename(bucket, object, index) {
                                    Err(DiskError::Unexpected)
                                } else {
-                                    disk.rename_data(
+                                    disk.rename_data_borrowed(
                                        RUSTFS_META_TMP_BUCKET,
                                        &tmp_id,
-                                        parts_metadata[index].clone(),
+                                        &parts_metadata[index],
                                        bucket,
                                        object,
                                    )
@@ -27,11 +27,12 @@ use super::object::{
    object_transaction_fencing_requested, old_data_cleanup_receipt_path, read_object_transaction_epoch_fence,
    verify_object_transaction_epoch_fence,
 };
+use crate::bucket::quota::reservation;
 use crate::crash_inject::{self, CrashPoint};
 use crate::multipart_listing::paginate_multipart_listing;
 use futures::{StreamExt, stream};
 use std::future::Future;
-#[cfg(test)]
+#[cfg(any(test, feature = "test-util"))]
 use std::sync::atomic::{AtomicUsize, Ordering};
 use std::time::Duration;
 use tokio::task::JoinSet;
@@ -61,20 +62,21 @@ impl StaleMultipartCleanupGuard {
    }
 }

-#[cfg(test)]
+#[cfg(any(test, feature = "test-util"))]
 #[derive(Clone, Copy, PartialEq, Eq)]
-pub(crate) enum MultipartCommitPause {
+pub enum MultipartCommitPause {
    PutPartBeforeLockAcquire,
    PutPartBeforeLockLost,
    PutPartAfterRename,
    BeforeLockLost,
+    BeforeQuotaRename,
    BeforeTransactionEpochVerify,
    BeforeObjectPublication,
    AfterObjectPublication,
    AfterRename,
 }

-#[cfg(test)]
+#[cfg(any(test, feature = "test-util"))]
 struct MultipartCommitBarrierState {
    bucket: String,
    object: String,
@@ -85,27 +87,22 @@ struct MultipartCommitBarrierState {
    release: tokio::sync::Semaphore,
 }

-#[cfg(test)]
-pub(crate) struct MultipartCommitBarrier {
+#[cfg(any(test, feature = "test-util"))]
+pub struct MultipartCommitBarrier {
    state: Arc<MultipartCommitBarrierState>,
 }

-#[cfg(test)]
+#[cfg(any(test, feature = "test-util"))]
 static MULTIPART_COMMIT_BARRIER: std::sync::OnceLock<std::sync::Mutex<Option<Arc<MultipartCommitBarrierState>>>> =
    std::sync::OnceLock::new();

-#[cfg(test)]
+#[cfg(any(test, feature = "test-util"))]
 impl MultipartCommitBarrier {
-    pub(crate) fn install(bucket: &str, object: &str, pause: MultipartCommitPause) -> Self {
+    pub fn install(bucket: &str, object: &str, pause: MultipartCommitPause) -> Self {
        Self::install_for_arrivals(bucket, object, pause, 1)
    }

-    pub(crate) fn install_for_arrivals(
-        bucket: &str,
-        object: &str,
-        pause: MultipartCommitPause,
-        expected_arrivals: usize,
-    ) -> Self {
+    pub fn install_for_arrivals(bucket: &str, object: &str, pause: MultipartCommitPause, expected_arrivals: usize) -> Self {
        assert!(expected_arrivals > 0, "multipart commit barrier must wait for at least one arrival");
        let state = Arc::new(MultipartCommitBarrierState {
            bucket: bucket.to_string(),
@@ -126,7 +123,7 @@ impl MultipartCommitBarrier {
        Self { state }
    }

-    pub(crate) async fn wait_until_paused(&self) {
+    pub async fn wait_until_paused(&self) {
        tokio::time::timeout(Duration::from_secs(30), async {
            loop {
                let arrived = self.state.arrived.notified();
@@ -140,12 +137,12 @@ impl MultipartCommitBarrier {
        .expect("multipart completion should reach the deterministic commit barrier");
    }

-    pub(crate) fn release(&self) {
+    pub fn release(&self) {
        self.state.release.add_permits(self.state.expected_arrivals);
    }
 }

-#[cfg(test)]
+#[cfg(any(test, feature = "test-util"))]
 impl Drop for MultipartCommitBarrier {
    fn drop(&mut self) {
        self.release();
@@ -159,7 +156,7 @@ impl Drop for MultipartCommitBarrier {
    }
 }

-#[cfg(test)]
+#[cfg(any(test, feature = "test-util"))]
 async fn pause_multipart_commit(bucket: &str, object: &str, pause: MultipartCommitPause) {
    let barrier = {
        let mut slot = MULTIPART_COMMIT_BARRIER
@@ -1892,6 +1889,41 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {

        fi.parts = Vec::with_capacity(uploaded_parts.len());

+        let quota_context = reservation::begin(
+            &self.ctx,
+            bucket,
+            object,
+            opts.quota_admission,
+            opts.data_movement,
+            self.pool_index,
+            self.set_index,
+        )
+        .await?;
+        let quota_mutation_fence = quota_context.is_enforced() || opts.quota_admission.is_some();
+        let preserve_replication_ciphertext = opts.replication_request
+            && contains_key_str(&fi.metadata, rustfs_utils::http::SUFFIX_REPLICATION_PRESERVE_CIPHERTEXT);
+        if quota_context.is_enforced() && preserve_replication_ciphertext {
+            return Err(Error::PartMissingOrCorrupt);
+        }
+        let transformed_object = fi.is_compressed() || should_persist_encryption_original_size(&fi.metadata);
+        let data_movement_actual_size = if opts.data_movement {
+            rustfs_utils::http::get_consistent_str(&opts.user_defined, SUFFIX_ACTUAL_SIZE)
+                .map(|value| {
+                    value
+                        .parse::<i64>()
+                        .ok()
+                        .filter(|value| *value >= 0)
+                        .ok_or(Error::PartMissingOrCorrupt)
+                })
+                .transpose()?
+        } else {
+            None
+        };
+        let data_movement_actual_size_u64 = data_movement_actual_size
+            .map(u64::try_from)
+            .transpose()
+            .map_err(|_| Error::PartMissingOrCorrupt)?;
+
        let mut object_size: usize = 0;
        let mut object_actual_size: i64 = 0;

@@ -2018,17 +2050,29 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
                checksum_combined.extend_from_slice(cs.raw.as_slice());
            }

-            object_size += ext_part.size;
-            if opts.quota_admission.is_some() && ext_part.actual_size < 0 {
+            object_size = object_size.checked_add(ext_part.size).ok_or(Error::PartMissingOrCorrupt)?;
+            let unknown_actual_size_allowed = opts.data_movement && transformed_object && data_movement_actual_size.is_some()
+                || opts.replication_request && !quota_context.is_enforced();
+            if ext_part.actual_size < 0 && !unknown_actual_size_allowed {
                return Err(Error::PartMissingOrCorrupt);
            }
+            let normalized_actual_size = if ext_part.actual_size >= 0 && !transformed_object {
+                ext_part
+                    .actual_size
+                    .max(i64::try_from(ext_part.size).map_err(|_| Error::PartMissingOrCorrupt)?)
+            } else {
+                ext_part.actual_size
+            };
            object_actual_size = object_actual_size
-                .checked_add(ext_part.actual_size)
+                .checked_add(normalized_actual_size)
                .ok_or(Error::PartMissingOrCorrupt)?;
-
-            fi.parts.push(completed_multipart_object_part(p.part_num, ext_part));
+            let mut completed_part = completed_multipart_object_part(p.part_num, ext_part);
+            completed_part.actual_size = normalized_actual_size;
+            fi.parts.push(completed_part);
+        }
+        if !transformed_object && data_movement_actual_size.is_some_and(|actual_size| actual_size < object_actual_size) {
+            return Err(Error::PartMissingOrCorrupt);
        }
-
        if let Some(wtcs) = opts.want_checksum.as_ref() {
            if checksum_type.full_object_requested() {
                if wtcs.encoded != checksum.encoded {
@@ -2053,15 +2097,35 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
            }
        }

-        if let Some(admission) = opts.quota_admission {
-            let quota_operation_size = u64::try_from(object_actual_size).map_err(|_| Error::PartMissingOrCorrupt)?;
-            if quota_operation_size > admission.remaining() {
-                return Err(Error::QuotaExceeded {
-                    current: admission.current_usage(),
-                    limit: admission.quota_limit(),
-                });
-            }
-        }
+        let declared_replication_actual_size = opts
+            .replication_request
+            .then(|| get_str(&opts.user_defined, SUFFIX_ACTUAL_OBJECT_SIZE_CAP))
+            .flatten();
+        let replication_actual_size = if opts.replication_request && quota_context.is_enforced() {
+            let observed_size = u64::try_from(object_actual_size).map_err(|_| Error::PartMissingOrCorrupt)?;
+            let declared_cap = declared_replication_actual_size
+                .as_deref()
+                .map(|value| value.parse::<u64>().map_err(|_| Error::PartMissingOrCorrupt))
+                .transpose()?
+                .unwrap_or(0);
+            let declared_encryption_size = rustfs_utils::http::get_object_encryption_original_size(&fi.metadata)
+                .map_err(Error::other)?
+                .map(u64::try_from)
+                .transpose()
+                .map_err(|_| Error::PartMissingOrCorrupt)?
+                .unwrap_or(0);
+            Some(observed_size.max(declared_cap).max(declared_encryption_size))
+        } else {
+            None
+        };
+        let data_movement_quota_size = data_movement_actual_size_u64.filter(|_| quota_context.is_enforced());
+        let quota_new_size = match data_movement_quota_size.or(replication_actual_size) {
+            Some(size) => size.max(u64::try_from(object_size).map_err(|_| Error::PartMissingOrCorrupt)?),
+            None if quota_context.is_enforced() => u64::try_from(object_actual_size)
+                .map_err(|_| Error::PartMissingOrCorrupt)?
+                .max(u64::try_from(object_size).map_err(|_| Error::PartMissingOrCorrupt)?),
+            None => 0,
+        };
        if let Some(rc_crc) = get_header_map(&opts.user_defined, SUFFIX_REPLICATION_SSEC_CRC) {
            if let Ok(rc_crc_bytes) = base64_simd::STANDARD.decode_to_vec(&rc_crc) {
                fi.checksum = Some(Bytes::from(rc_crc_bytes));
@@ -2113,20 +2177,6 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
            );
        }

-        let data_movement_actual_size = if opts.data_movement {
-            rustfs_utils::http::get_consistent_str(&opts.user_defined, SUFFIX_ACTUAL_SIZE)
-                .map(|value| {
-                    value
-                        .parse::<i64>()
-                        .ok()
-                        .filter(|value| *value >= 0)
-                        .ok_or_else(|| Error::other("data movement actual size metadata is invalid"))
-                })
-                .transpose()?
-        } else {
-            None
-        };
-
        if let Some(actual_size) = data_movement_actual_size {
            insert_str(&mut fi.metadata, SUFFIX_ACTUAL_SIZE, actual_size.to_string());
            if persist_encryption_original_size {
@@ -2134,7 +2184,13 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
                    .insert("x-rustfs-encryption-original-size".to_string(), actual_size.to_string());
            }
        } else if opts.replication_request {
-            if let Some(actual_size) = get_str(&opts.user_defined, SUFFIX_ACTUAL_OBJECT_SIZE_CAP) {
+            if let Some(actual_size) = replication_actual_size {
+                insert_str(&mut fi.metadata, SUFFIX_ACTUAL_SIZE, actual_size.to_string());
+                if persist_encryption_original_size {
+                    fi.metadata
+                        .insert("x-rustfs-encryption-original-size".to_string(), actual_size.to_string());
+                }
+            } else if let Some(actual_size) = declared_replication_actual_size {
                insert_str(&mut fi.metadata, SUFFIX_ACTUAL_SIZE, actual_size.clone());
                if persist_encryption_original_size {
                    fi.metadata
@@ -2324,8 +2380,41 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
        } else {
            None
        };
+
+        let quota_old_size = if quota_context.is_enforced() {
+            if opts.data_movement {
+                quota_new_size
+            } else {
+                reservation::replaced_logical_size(&self, bucket, object, opts).await?
+            }
+        } else {
+            0
+        };
+        let quota_reservation = quota_context.reserve(quota_old_size, quota_new_size).await?;
+        let (commit_disks, quota_fence_tokens) = if quota_mutation_fence {
+            match Self::prepare_quota_mutation_fences(&shuffle_disks, bucket, object, write_quorum).await {
+                Ok((disks, tokens)) => {
+                    for (metadata, token) in parts_metadatas.iter_mut().zip(tokens.iter().copied()) {
+                        if let Some(token) = token {
+                            insert_str(
+                                &mut metadata.metadata,
+                                crate::disk::QUOTA_MUTATION_FENCE_METADATA_SUFFIX,
+                                token.as_uuid().to_string(),
+                            );
+                        }
+                    }
+                    (disks, tokens)
+                }
+                Err(err) => {
+                    quota_reservation.abort().await;
+                    return Err(err);
+                }
+            }
+        } else {
+            (shuffle_disks.clone(), vec![None; shuffle_disks.len()])
+        };
        let transaction_epoch =
-            transaction_epoch_fence.map(|_| assign_object_transaction_epoch(&shuffle_disks, &mut parts_metadatas));
+            transaction_epoch_fence.map(|_| assign_object_transaction_epoch(&commit_disks, &mut parts_metadatas));

        let commit_set = self.clone();
        let commit_bucket = bucket.to_owned();
@@ -2333,49 +2422,152 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
        let commit_upload_id = upload_id.to_owned();
        let commit_upload_id_path = upload_id_path.clone();
        let commit_version_suspended = opts.version_suspended;
+        let commit_versioned = opts.versioned;
+        let commit_version_id = opts.version_id.clone();
+        let commit_namespace_lock_fence = opts.namespace_lock_fence.clone();
+        let commit_bucket_lifecycle_lock_fence = opts.bucket_lifecycle_lock_fence.clone();
        let commit_is_versioned = opts.versioned || opts.version_suspended;
        let commit_capacity_scope_token = opts.capacity_scope_token;
        let commit_object_lock_guard = object_lock_guard.take();
-        let detach_commit_owner = commit_object_lock_guard.is_some() || upload_guard.is_some();
+        let detach_commit_owner = commit_object_lock_guard.is_some() || upload_guard.is_some() || quota_mutation_fence;
        let commit = async move {
            let _object_lock_guard = commit_object_lock_guard;
            let _upload_guard = upload_guard;
+            let mut quota_reservation = quota_reservation;
            let complete_tail_stage_start = rustfs_io_metrics::put_stage_metrics_enabled().then(Instant::now);

-            // Crash-consistency injection: hard power loss after the upload is fully
-            // staged and locked but before the authoritative rename_data commit. No
-            // disk has moved the staged data, so a crash here must leave any prior
-            // committed version byte-for-byte intact (rustfs/backlog#864) and the
-            // upload fully retryable. Compiles to a no-op outside `#[cfg(test)]`.
-            if crash_inject::should_crash_at(CrashPoint::MultipartBeforeCommitRename, &commit_object) {
-                return Err(StorageError::Unexpected);
+            let pre_rename_result: Result<()> = async {
+                // Crash-consistency injection: hard power loss after the upload is fully
+                // staged and locked but before the authoritative rename_data commit. No
+                // disk has moved the staged data, so a crash here must leave any prior
+                // committed version byte-for-byte intact (rustfs/backlog#864) and the
+                // upload fully retryable. Compiles to a no-op outside `#[cfg(test)]`.
+                if crash_inject::should_crash_at(CrashPoint::MultipartBeforeCommitRename, &commit_object) {
+                    return Err(StorageError::Unexpected);
+                }
+                quota_reservation.mark_commit_started().await?;
+                #[cfg(any(test, feature = "test-util"))]
+                pause_multipart_commit(&commit_bucket, &commit_object, MultipartCommitPause::BeforeQuotaRename).await;
+                if quota_reservation.is_lock_lost()
+                    || !quota_reservation.capability_proof_matches()
+                    || _object_lock_guard.as_ref().is_some_and(|guard| guard.is_lock_lost())
+                    || _upload_guard.as_ref().is_some_and(|guard| guard.is_lock_lost())
+                    || commit_namespace_lock_fence
+                        .as_ref()
+                        .is_some_and(NamespaceLockFence::is_lock_lost)
+                    || commit_bucket_lifecycle_lock_fence
+                        .as_ref()
+                        .is_some_and(NamespaceLockFence::is_lock_lost)
+                {
+                    return Err(StorageError::NamespaceLockQuorumUnavailable {
+                        mode: "quota_reservation",
+                        bucket: commit_bucket.clone(),
+                        object: commit_object.clone(),
+                        required: 1,
+                        achieved: 0,
+                    });
+                }
+                let restore_opts = ObjectOptions {
+                    version_id: commit_version_id.clone(),
+                    versioned: commit_versioned,
+                    version_suspended: commit_version_suspended,
+                    no_lock: true,
+                    ..Default::default()
+                };
+                commit_set
+                    .require_current_restore_operation_id(
+                        &commit_bucket,
+                        &commit_object,
+                        &restore_opts,
+                        expected_restore_operation_id,
+                        "complete_multipart_upload_quota_reservation",
+                    )
+                    .await?;
+                if let Some(proof) = transaction_fencing_proof.as_ref()
+                    && !object_transaction_fencing_fleet_proof_matches(proof)
+                {
+                    return Err(Error::other(
+                        "object transaction fencing fleet capability changed during complete_multipart_upload",
+                    ));
+                }
+                if let Some(expected) = transaction_epoch_fence {
+                    #[cfg(test)]
+                    pause_multipart_commit(&commit_bucket, &commit_object, MultipartCommitPause::BeforeTransactionEpochVerify)
+                        .await;
+                    verify_object_transaction_epoch_fence(&commit_set, &commit_bucket, &commit_object, expected).await?;
+                }
+                if quota_reservation.is_lock_lost()
+                    || !quota_reservation.capability_proof_matches()
+                    || _object_lock_guard.as_ref().is_some_and(|guard| guard.is_lock_lost())
+                    || _upload_guard.as_ref().is_some_and(|guard| guard.is_lock_lost())
+                    || commit_namespace_lock_fence
+                        .as_ref()
+                        .is_some_and(NamespaceLockFence::is_lock_lost)
+                    || commit_bucket_lifecycle_lock_fence
+                        .as_ref()
+                        .is_some_and(NamespaceLockFence::is_lock_lost)
+                {
+                    return Err(StorageError::NamespaceLockQuorumUnavailable {
+                        mode: "quota_reservation",
+                        bucket: commit_bucket.clone(),
+                        object: commit_object.clone(),
+                        required: 1,
+                        achieved: 0,
+                    });
+                }
+                Ok(())
+            }
+            .await;
+            if let Err(err) = pre_rename_result {
+                SetDisks::abort_quota_reservation_after_fence(
+                    quota_reservation,
+                    &commit_disks,
+                    &quota_fence_tokens,
+                    &commit_bucket,
+                    &commit_object,
+                    write_quorum,
+                    quota_mutation_fence,
+                )
+                .await;
+                return Err(err);
            }

            // The trailing `_` drops the rename_data old-size backfill
            // (rustfs/backlog#1009): CompleteMultipartUpload keeps its pre-commit
            // `get_object_info` lookup, so the backfill has no consumer here yet.
-            if let Some(proof) = transaction_fencing_proof.as_ref()
-                && !object_transaction_fencing_fleet_proof_matches(proof)
-            {
-                return Err(Error::other(
-                    "object transaction fencing fleet capability changed during complete_multipart_upload",
-                ));
-            }
-            if let Some(expected) = transaction_epoch_fence {
-                #[cfg(test)]
-                pause_multipart_commit(&commit_bucket, &commit_object, MultipartCommitPause::BeforeTransactionEpochVerify).await;
-                verify_object_transaction_epoch_fence(&commit_set, &commit_bucket, &commit_object, expected).await?;
-            }
-            let (online_disks, convergence, op_old_dir, cleanup_disks, _) = SetDisks::rename_data(
-                &shuffle_disks,
+            Self::assign_rename_data_indexes(&mut parts_metadatas);
+            let rename_result = SetDisks::rename_data_owned(
+                &commit_disks,
                RUSTFS_META_MULTIPART_BUCKET,
                &commit_upload_id_path,
-                &parts_metadatas,
+                parts_metadatas,
                &commit_bucket,
                &commit_object,
                write_quorum,
            )
-            .await?;
+            .await;
+            if quota_mutation_fence {
+                let _ = SetDisks::release_quota_mutation_fences(
+                    &commit_disks,
+                    &quota_fence_tokens,
+                    &commit_bucket,
+                    &commit_object,
+                    write_quorum,
+                )
+                .await;
+            }
+            if rename_result.is_ok() {
+                quota_reservation.commit().await;
+            }
+            let rename_commit = match rename_result {
+                Ok(result) => result,
+                Err(err) => return Err(err.into()),
+            };
+            let online_disks = rename_commit.online_disks;
+            let convergence = rename_commit.convergence;
+            let op_old_dir = rename_commit.data_dir;
+            let cleanup_disks = rename_commit.cleanup_disks;
+            let committed_file_info = rename_commit.committed_file_info;

            // Detach admission before any post-commit await: client cancellation
            // must not couple durable convergence repair to cleanup work.
@@ -2420,9 +2612,7 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
                return Err(StorageError::Unexpected);
            }

-            if let Some(committed_slot) = online_disks.iter().position(Option::is_some) {
-                fi = parts_metadatas[committed_slot].clone();
-            }
+            fi = committed_file_info;
            let committed_dir = fi.data_dir.unwrap_or_default().to_string();

            commit_set.record_capacity_scope_if_needed(commit_capacity_scope_token, &online_disks);
@@ -2927,12 +3117,34 @@ mod tests {
        content: &[u8],
        actual_size: i64,
    ) -> CompletePart {
+        put_test_part_with_opts(
+            set_disks,
+            bucket,
+            object,
+            upload_id,
+            part_number,
+            (content, actual_size),
+            &ObjectOptions::default(),
+        )
+        .await
+    }
+
+    async fn put_test_part_with_opts(
+        set_disks: &Arc<SetDisks>,
+        bucket: &str,
+        object: &str,
+        upload_id: &str,
+        part_number: usize,
+        part: (&[u8], i64),
+        opts: &ObjectOptions,
+    ) -> CompletePart {
+        let (content, actual_size) = part;
        let mut reader = PutObjReader::new(
            HashReader::from_stream(Cursor::new(content.to_vec()), content.len() as i64, actual_size, None, None, false)
                .expect("hash reader should be constructed"),
        );
        let part = set_disks
-            .put_object_part(bucket, object, upload_id, part_number, &mut reader, &ObjectOptions::default())
+            .put_object_part(bucket, object, upload_id, part_number, &mut reader, opts)
            .await
            .expect("uploading the part should succeed");
        CompletePart {
@@ -3162,7 +3374,7 @@ mod tests {
        let (upload_id, parts) =
            stage_upload_with_create_opts(&set_disks, bucket, object, &payload, &ObjectOptions::default()).await;
        let mut denied_opts = ObjectOptions::default();
-        assert!(denied_opts.set_quota_admission(100, 4195));
+        assert!(denied_opts.set_quota_admission(100, 4180));

        let err = set_disks
            .clone()
@@ -3173,7 +3385,7 @@ mod tests {
            err,
            StorageError::QuotaExceeded {
                current: 100,
-                limit: 4195
+                limit: 4180
            }
        ));

@@ -3191,7 +3403,7 @@ mod tests {
        );

        let mut allowed_opts = ObjectOptions::default();
-        assert!(allowed_opts.set_quota_admission(100, 4196));
+        assert!(allowed_opts.set_quota_admission(100, 4181));
        let completed = set_disks
            .clone()
            .complete_multipart_upload(bucket, object, &upload_id, parts, &allowed_opts)
@@ -3232,6 +3444,120 @@ mod tests {
        );
    }

+    #[tokio::test]
+    async fn replication_quota_uses_server_observed_part_size_as_lower_bound() {
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "multipart-replication-quota-bucket";
+        let object = "object";
+        make_bucket_on_all(&disk_stores, bucket).await;
+
+        let mut create_opts = ObjectOptions::default();
+        insert_str(&mut create_opts.user_defined, SUFFIX_COMPRESSION, "S2".to_string());
+        let upload = set_disks
+            .new_multipart_upload(bucket, object, &create_opts)
+            .await
+            .expect("replication multipart upload should be created");
+        let part = put_test_part(&set_disks, bucket, object, &upload.upload_id, 1, &[0x72; 4096], 1).await;
+        let mut complete_opts = ObjectOptions {
+            replication_request: true,
+            ..Default::default()
+        };
+        assert!(complete_opts.set_quota_admission(0, 4095));
+
+        let err = set_disks
+            .clone()
+            .complete_multipart_upload(bucket, object, &upload.upload_id, vec![part.clone()], &complete_opts)
+            .await
+            .expect_err("a forged tiny replication logical size must not reduce quota admission");
+        assert!(matches!(err, StorageError::QuotaExceeded { current: 0, limit: 4095 }));
+        assert!(
+            set_disks
+                .check_upload_id_exists(bucket, object, &upload.upload_id, false)
+                .await
+                .is_ok(),
+            "quota rejection must leave replicated multipart parts retryable"
+        );
+
+        assert!(complete_opts.set_quota_admission(0, 4096));
+        let completed = set_disks
+            .clone()
+            .complete_multipart_upload(bucket, object, &upload.upload_id, vec![part], &complete_opts)
+            .await
+            .expect("the physical safety boundary should admit the transformed replica");
+        assert_eq!(completed.get_actual_size().expect("replica logical size should parse"), 1);
+    }
+
+    #[tokio::test]
+    async fn direct_multipart_quota_uses_server_observed_part_size_as_lower_bound() {
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "multipart-direct-quota-bucket";
+        let object = "object";
+        make_bucket_on_all(&disk_stores, bucket).await;
+
+        let upload = set_disks
+            .new_multipart_upload(bucket, object, &ObjectOptions::default())
+            .await
+            .expect("multipart upload should be created");
+        let part = put_test_part(&set_disks, bucket, object, &upload.upload_id, 1, &[0x73; 4096], 1).await;
+        let mut complete_opts = ObjectOptions::default();
+        assert!(complete_opts.set_quota_admission(0, 4095));
+
+        let err = set_disks
+            .clone()
+            .complete_multipart_upload(bucket, object, &upload.upload_id, vec![part], &complete_opts)
+            .await
+            .expect_err("a forged tiny direct logical size must not reduce quota admission");
+        assert!(matches!(err, StorageError::QuotaExceeded { current: 0, limit: 4095 }));
+        assert!(
+            set_disks
+                .check_upload_id_exists(bucket, object, &upload.upload_id, false)
+                .await
+                .is_ok(),
+            "quota rejection must leave direct multipart parts retryable"
+        );
+    }
+
+    #[tokio::test]
+    async fn quota_rejects_ciphertext_replication_without_a_server_observed_logical_size() {
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "multipart-compressed-ciphertext-quota-bucket";
+        let object = "object";
+        make_bucket_on_all(&disk_stores, bucket).await;
+
+        let mut create_opts = ObjectOptions::default();
+        insert_str(&mut create_opts.user_defined, SUFFIX_COMPRESSION, "S2".to_string());
+        insert_str(
+            &mut create_opts.user_defined,
+            rustfs_utils::http::SUFFIX_REPLICATION_PRESERVE_CIPHERTEXT,
+            "true".to_string(),
+        );
+        let upload = set_disks
+            .new_multipart_upload(bucket, object, &create_opts)
+            .await
+            .expect("ciphertext multipart upload should be created");
+        let payload = vec![0x74; 4096];
+        let part = put_test_part(&set_disks, bucket, object, &upload.upload_id, 1, &payload, 1).await;
+        let mut complete_opts = ObjectOptions {
+            replication_request: true,
+            ..Default::default()
+        };
+        assert!(complete_opts.set_quota_admission(0, u64::MAX));
+
+        let err = set_disks
+            .clone()
+            .complete_multipart_upload(bucket, object, &upload.upload_id, vec![part], &complete_opts)
+            .await
+            .expect_err("ciphertext replication has no server-observed logical quota size");
+        assert!(matches!(err, StorageError::PartMissingOrCorrupt));
+        assert!(
+            set_disks
+                .check_upload_id_exists(bucket, object, &upload.upload_id, false)
+                .await
+                .is_ok(),
+            "rejection must leave ciphertext multipart parts retryable"
+        );
+    }
+
    #[tokio::test]
    async fn complete_multipart_quota_rejects_invalid_logical_sizes() {
        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
@@ -3554,7 +3880,6 @@ mod tests {
    async fn data_movement_complete_accepts_unknown_compressed_part_actual_size() {
        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
        let bucket = "data-movement-unknown-actual-size-bucket";
-        let object = "object";
        make_bucket_on_all(&disk_stores, bucket).await;

        let mut metadata = HashMap::new();
@@ -3568,6 +3893,63 @@ mod tests {
            rustfs_utils::http::SUFFIX_DATA_MOVEMENT_UPLOAD,
            "source-generation".to_string(),
        );
+
+        for (object, quota_limit) in [("without-quota", None), ("with-quota", Some(u64::MAX))] {
+            let create_opts = ObjectOptions {
+                data_movement: true,
+                user_defined: metadata.clone(),
+                ..Default::default()
+            };
+            let upload = set_disks
+                .new_multipart_upload(bucket, object, &create_opts)
+                .await
+                .expect("data movement upload should be created");
+
+            let mut completed_parts = Vec::new();
+            for (number, actual_size) in [(1, -1), (2, 1)] {
+                let mut reader = PutObjReader::new(
+                    HashReader::from_stream(Cursor::new(vec![number as u8]), 1, actual_size, None, None, false)
+                        .expect("part reader should be constructed"),
+                );
+                let part = set_disks
+                    .put_object_part(bucket, object, &upload.upload_id, number, &mut reader, &create_opts)
+                    .await
+                    .expect("data movement part should be written");
+                completed_parts.push(CompletePart {
+                    part_num: number,
+                    etag: part.etag,
+                    ..Default::default()
+                });
+            }
+
+            let missing_size_err = set_disks
+                .clone()
+                .complete_multipart_upload(bucket, object, &upload.upload_id, completed_parts.clone(), &create_opts)
+                .await
+                .expect_err("data movement completion must require an authoritative total size");
+            assert!(matches!(missing_size_err, StorageError::PartMissingOrCorrupt));
+
+            let mut complete_opts = create_opts.clone();
+            rustfs_utils::http::insert_str(
+                &mut complete_opts.user_defined,
+                rustfs_utils::http::SUFFIX_ACTUAL_SIZE,
+                "2".to_string(),
+            );
+            if let Some(quota_limit) = quota_limit {
+                assert!(complete_opts.set_quota_admission(0, quota_limit));
+            }
+            let completion = set_disks
+                .clone()
+                .complete_multipart_upload(bucket, object, &upload.upload_id, completed_parts, &complete_opts)
+                .await;
+
+            let completed = completion.expect("data movement completion should accept the persisted unknown-size sentinel");
+
+            assert_eq!(completed.parts[0].actual_size, -1);
+            assert_eq!(completed.get_actual_size().expect("completed object actual size"), 2);
+        }
+
+        let object = "all-unknown-with-quota";
        let create_opts = ObjectOptions {
            data_movement: true,
            user_defined: metadata.clone(),
@@ -3577,43 +3959,93 @@ mod tests {
            .new_multipart_upload(bucket, object, &create_opts)
            .await
            .expect("data movement upload should be created");
-
-        let mut completed_parts = Vec::new();
-        for (number, actual_size) in [(1, -1), (2, 1)] {
-            let mut reader = PutObjReader::new(
-                HashReader::from_stream(Cursor::new(vec![number as u8]), 1, actual_size, None, None, false)
-                    .expect("part reader should be constructed"),
-            );
-            let part = set_disks
-                .put_object_part(bucket, object, &upload.upload_id, number, &mut reader, &create_opts)
-                .await
-                .expect("data movement part should be written");
-            completed_parts.push(CompletePart {
-                part_num: number,
-                etag: part.etag,
-                ..Default::default()
-            });
-        }
-
-        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "2".to_string());
+        let part = put_test_part_with_opts(&set_disks, bucket, object, &upload.upload_id, 1, (&[0x40], -1), &create_opts).await;
+        let mut complete_opts = create_opts;
+        rustfs_utils::http::insert_str(&mut complete_opts.user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "2".to_string());
+        assert!(complete_opts.set_quota_admission(0, u64::MAX));
        let completed = set_disks
            .clone()
-            .complete_multipart_upload(
-                bucket,
-                object,
-                &upload.upload_id,
-                completed_parts,
-                &ObjectOptions {
-                    data_movement: true,
-                    user_defined: metadata,
-                    ..Default::default()
-                },
-            )
+            .complete_multipart_upload(bucket, object, &upload.upload_id, vec![part], &complete_opts)
            .await
-            .expect("data movement completion should accept the persisted unknown-size sentinel");
-
+            .expect("quota must accept an all-unknown data movement upload with an authoritative total");
        assert_eq!(completed.parts[0].actual_size, -1);
        assert_eq!(completed.get_actual_size().expect("completed object actual size"), 2);
+
+        let object = "legacy-zero-fallback";
+        let create_opts = ObjectOptions {
+            data_movement: true,
+            user_defined: metadata.clone(),
+            ..Default::default()
+        };
+        let upload = set_disks
+            .new_multipart_upload(bucket, object, &create_opts)
+            .await
+            .expect("data movement upload should be created");
+        let part =
+            put_test_part_with_opts(&set_disks, bucket, object, &upload.upload_id, 1, (&[0x41; 128], 128), &create_opts).await;
+        let mut complete_opts = create_opts;
+        rustfs_utils::http::insert_str(
+            &mut complete_opts.user_defined,
+            rustfs_utils::http::SUFFIX_ACTUAL_SIZE,
+            "100".to_string(),
+        );
+        assert!(complete_opts.set_quota_admission(0, u64::MAX));
+        let completed = set_disks
+            .clone()
+            .complete_multipart_upload(bucket, object, &upload.upload_id, vec![part], &complete_opts)
+            .await
+            .expect("legacy zero-to-physical part fallback must remain migratable");
+        assert_eq!(completed.parts[0].actual_size, 128);
+        assert_eq!(completed.get_actual_size().expect("completed object actual size"), 100);
+
+        let object = "inconsistent-untransformed-total";
+        let mut untransformed_metadata = metadata.clone();
+        rustfs_utils::http::remove_str(&mut untransformed_metadata, rustfs_utils::http::SUFFIX_COMPRESSION);
+        let create_opts = ObjectOptions {
+            data_movement: true,
+            user_defined: untransformed_metadata,
+            ..Default::default()
+        };
+        let upload = set_disks
+            .new_multipart_upload(bucket, object, &create_opts)
+            .await
+            .expect("data movement upload should be created");
+        let part = put_test_part_with_opts(&set_disks, bucket, object, &upload.upload_id, 1, (&[0x42; 2], 2), &create_opts).await;
+        let mut complete_opts = create_opts;
+        rustfs_utils::http::insert_str(&mut complete_opts.user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "1".to_string());
+        let err = set_disks
+            .clone()
+            .complete_multipart_upload(bucket, object, &upload.upload_id, vec![part], &complete_opts)
+            .await
+            .expect_err("authoritative total below known logical part sizes must fail closed");
+        assert!(matches!(err, StorageError::PartMissingOrCorrupt));
+
+        for (object, declared_size) in [("invalid-total", "invalid"), ("negative-total", "-1")] {
+            let create_opts = ObjectOptions {
+                data_movement: true,
+                user_defined: metadata.clone(),
+                ..Default::default()
+            };
+            let upload = set_disks
+                .new_multipart_upload(bucket, object, &create_opts)
+                .await
+                .expect("data movement upload should be created");
+            let part =
+                put_test_part_with_opts(&set_disks, bucket, object, &upload.upload_id, 1, (&[0x41], 1), &create_opts).await;
+            let mut complete_opts = create_opts.clone();
+            rustfs_utils::http::insert_str(
+                &mut complete_opts.user_defined,
+                rustfs_utils::http::SUFFIX_ACTUAL_SIZE,
+                declared_size.to_string(),
+            );
+
+            let err = set_disks
+                .clone()
+                .complete_multipart_upload(bucket, object, &upload.upload_id, vec![part], &complete_opts)
+                .await
+                .expect_err("invalid authoritative total size must fail closed");
+            assert!(matches!(err, StorageError::PartMissingOrCorrupt));
+        }
    }

    async fn assert_complete_first_linearizes(bucket: &'static str, object: &'static str, create_opts: ObjectOptions) {
@@ -1174,6 +1174,32 @@ mod tests {
        }
    }

+    #[tokio::test]
+    #[serial_test::serial(storage_class_env)]
+    async fn quota_object_fence_ignores_an_unrelated_offline_pool() {
+        let temp_dir = tempfile::tempdir().expect("create quota fence store dir");
+        let (_ctx, store, shutdown) =
+            without_storage_class_env(build_isolated_test_store(temp_dir.path(), "quota-object-fence", &[4, 4])).await;
+        crate::bucket::metadata_sys::init_bucket_metadata_sys(store.clone(), Vec::new()).await;
+        let bucket = format!("quota-object-fence-{}", uuid::Uuid::new_v4());
+        let object = "object.bin";
+        store
+            .make_bucket(&bucket, &MakeBucketOptions::default())
+            .await
+            .expect("create quota fence bucket");
+        store.pools[1].disk_set[0].disks.write().await.fill(None);
+
+        crate::bucket::quota::reservation::fence_namespace_mutations_for_test(&store, &bucket, object, Some((0, 0)))
+            .await
+            .expect("the selected pool fence should ignore an unrelated offline pool");
+        let err = crate::bucket::quota::reservation::fence_namespace_mutations_for_test(&store, &bucket, object, None)
+            .await
+            .expect_err("legacy reservations must conservatively fence every pool");
+        assert!(matches!(err, StorageError::ErasureWriteQuorum));
+
+        shutdown.cancel();
+    }
+
    #[tokio::test]
    #[serial_test::serial(storage_class_env)]
    async fn tag_updates_skip_active_rebalance_source_pool() {
@@ -1,81 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use super::*;
-
-impl ECStore {
-    #[instrument(level = "trace", skip(self))]
-    #[allow(clippy::too_many_arguments)]
-    pub(super) async fn handle_list_objects_v2(
-        self: Arc<Self>,
-        bucket: &str,
-        prefix: &str,
-        continuation_token: Option<String>,
-        delimiter: Option<String>,
-        max_keys: i32,
-        fetch_owner: bool,
-        start_after: Option<String>,
-        incl_deleted: bool,
-    ) -> Result<ListObjectsV2Info> {
-        self.inner_list_objects_v2(
-            bucket,
-            prefix,
-            continuation_token,
-            delimiter,
-            max_keys,
-            fetch_owner,
-            start_after,
-            incl_deleted,
-        )
-        .await
-    }
-
-    #[instrument(skip(self))]
-    pub(super) async fn handle_list_object_versions(
-        self: Arc<Self>,
-        bucket: &str,
-        prefix: &str,
-        marker: Option<String>,
-        version_marker: Option<String>,
-        delimiter: Option<String>,
-        max_keys: i32,
-    ) -> Result<ListObjectVersionsInfo> {
-        self.inner_list_object_versions(bucket, prefix, marker, version_marker, delimiter, max_keys)
-            .await
-    }
-
-    pub(crate) async fn list_object_versions_for_lifecycle(
-        self: Arc<Self>,
-        bucket: &str,
-        prefix: &str,
-        marker: Option<String>,
-        version_marker: Option<String>,
-        delimiter: Option<String>,
-        max_keys: i32,
-    ) -> Result<ListObjectVersionsInfo> {
-        self.inner_list_object_versions_for_lifecycle(bucket, prefix, marker, version_marker, delimiter, max_keys)
-            .await
-    }
-
-    pub(super) async fn handle_walk(
-        self: Arc<Self>,
-        rx: CancellationToken,
-        bucket: &str,
-        prefix: &str,
-        result: tokio::sync::mpsc::Sender<ObjectInfoOrErr>,
-        opts: WalkOptions,
-    ) -> Result<()> {
-        self.walk_internal(rx, bucket, prefix, result, opts).await
-    }
-}
@@ -641,7 +641,7 @@ pub(crate) fn observe_scanner_namespace_mutations(bucket: &str, delta: u64) {
        .fetch_update(Ordering::AcqRel, Ordering::Acquire, |current| Some(current.saturating_add(delta)));
 }

-pub(super) async fn observe_list_objects_mutation(store: &ECStore, bucket: &str) -> u64 {
+pub(crate) async fn observe_list_objects_mutation(store: &ECStore, bucket: &str) -> u64 {
    observe_list_objects_mutations(store, bucket, 1).await.unwrap_or_default()
 }

@@ -3845,7 +3845,7 @@ impl ECStore {
        .await
    }

-    pub(crate) async fn inner_list_object_versions_for_lifecycle(
+    pub(crate) async fn list_object_versions_for_lifecycle(
        self: Arc<Self>,
        bucket: &str,
        prefix: &str,
@@ -148,7 +148,6 @@ mod heal_walk;
 pub use heal_walk::HealWalkVersion;
 mod init;
 pub(crate) mod init_format;
-mod list;
 pub(crate) mod list_objects;
 mod multipart;
 mod object;
@@ -601,7 +600,7 @@ impl crate::storage_api_contracts::list::ListOperations for ECStore {
        start_after: Option<String>,
        incl_deleted: bool,
    ) -> Result<ListObjectsV2Info> {
-        self.handle_list_objects_v2(
+        self.inner_list_objects_v2(
            bucket,
            prefix,
            continuation_token,
@@ -624,7 +623,7 @@ impl crate::storage_api_contracts::list::ListOperations for ECStore {
        delimiter: Option<String>,
        max_keys: i32,
    ) -> Result<ListObjectVersionsInfo> {
-        self.handle_list_object_versions(bucket, prefix, marker, version_marker, delimiter, max_keys)
+        self.inner_list_object_versions(bucket, prefix, marker, version_marker, delimiter, max_keys)
            .await
    }

@@ -636,7 +635,7 @@ impl crate::storage_api_contracts::list::ListOperations for ECStore {
        result: tokio::sync::mpsc::Sender<ObjectInfoOrErr>,
        opts: WalkOptions,
    ) -> Result<()> {
-        self.handle_walk(rx, bucket, prefix, result, opts).await
+        self.walk_internal(rx, bucket, prefix, result, opts).await
    }
 }

@@ -4741,9 +4741,10 @@ mod tests {
    #[tokio::test]
    #[serial_test::serial(body_cache_hook)]
    async fn select_snapshot_rejects_latest_versioned_delete_marker_during_prepare() {
-        let (_first_dirs, first_set) = make_local_set_disks(4, 2).await;
-        let (_second_dirs, second_set) = make_local_set_disks(4, 2).await;
-        let store = new_prepared_reader_test_store(&[Arc::clone(&first_set), Arc::clone(&second_set)]).await;
+        let ctx = Arc::new(crate::runtime::instance::InstanceContext::new());
+        let (_first_dirs, first_set) = make_local_set_disks_with_ctx(4, 2, Arc::clone(&ctx)).await;
+        let (_second_dirs, second_set) = make_local_set_disks_with_ctx(4, 2, Arc::clone(&ctx)).await;
+        let store = new_prepared_reader_test_store_with_ctx(&[Arc::clone(&first_set), Arc::clone(&second_set)], ctx).await;
        let bucket = "select-snapshot-latest-delete-marker";
        let object = "versioned-object.bin";
        let versioned_opts = ObjectOptions {
@@ -157,6 +157,18 @@ fn ecstore_implements_storage_list_operations_contract() {
    assert!(storage_list_operations_type_name::<ECStore>().ends_with("::ECStore"));
 }

+#[test]
+fn ecstore_pools_expose_storage_list_operations_contract() {
+    fn assert_contract(store: &ECStore) {
+        let future = store.pools[0]
+            .clone()
+            .list_objects_v2("bucket", "", None, None, 1, false, None, false);
+        drop(future);
+    }
+
+    let _ = assert_contract;
+}
+
 #[test]
 fn ecstore_implements_storage_multipart_operations_contract() {
    assert!(storage_multipart_operations_type_name::<ECStore>().ends_with("::ECStore"));
@@ -135,10 +135,7 @@ impl FileMeta {
        let i = buf.len() as u64;

        // check version, buf = buf[8..]
-        let (buf, _, _) = Self::check_xl2_v1(buf).map_err(|e| {
-            error!("failed to check XL2 v1 format: {}", e);
-            e
-        })?;
+        let (buf, _, _) = Self::check_xl2_v1(buf)?;

        if buf.len() < 5 {
            error!(
@@ -82,8 +82,8 @@ impl Error {
    /// Whether a heal operation can be retried without changing its inputs.
    pub(crate) fn is_recoverable_heal(&self) -> bool {
        match self {
-            Error::TaskCancelled => false,
-            Error::TaskTimeout | Error::TransientSkip { .. } => true,
+            Error::TaskCancelled | Error::TaskTimeout => false,
+            Error::TransientSkip { .. } => true,
            Error::Storage(err) => {
                err.is_quorum_error()
                    || matches!(
@@ -165,4 +165,9 @@ mod tests {
        assert!(Error::Storage(EcstoreError::DiskNotFound).is_recoverable_heal());
        assert!(Error::Storage(EcstoreError::VolumeNotFound).is_recoverable_heal());
    }
+
+    #[test]
+    fn task_timeout_is_terminal() {
+        assert!(!Error::TaskTimeout.is_recoverable_heal());
+    }
 }
@@ -673,6 +673,12 @@ fn retry_request_for_result(task: &HealTask, result: &Result<()>) -> Option<(Hea
    Some((request, delay, error))
 }

+async fn retry_request_for_result_with_budget(task: &HealTask, result: &Result<()>) -> Option<(HealRequest, Duration, String)> {
+    let (_, delay, error) = retry_request_for_result(task, result)?;
+    let request = task.retry_request_with_remaining_timeout().await.ok()?;
+    Some((request, delay, error))
+}
+
 fn recoverable_heal_retry_delay(retry_attempt: u32) -> Duration {
    let retry_attempt = retry_attempt.clamp(1, 5);
    let delay = Duration::from_secs(2_u64.saturating_pow(retry_attempt));
@@ -690,7 +696,7 @@ pub struct HealConfig {
    pub max_concurrent_heals: usize,
    /// Maximum concurrent heal tasks allowed for a single erasure set
    pub max_concurrent_per_set: usize,
-    /// Task timeout
+    /// Aggregate task execution timeout across recoverable retries
    pub task_timeout: Duration,
    /// Queue size
    pub queue_size: usize,
@@ -3106,7 +3112,7 @@ impl HealManager {
                        "Heal scheduler task started"
                    );
                    let result = task.execute().await;
-                    let retry_request = retry_request_for_result(task.as_ref(), &result);
+                    let retry_request = retry_request_for_result_with_budget(task.as_ref(), &result).await;
                    match &result {
                        Ok(_) => {
                            debug!(
@@ -4539,6 +4545,25 @@ mod tests {
        assert!(retry_error.contains("Lock acquisition timeout"));
    }

+    #[tokio::test]
+    async fn retry_request_for_result_preserves_remaining_timeout_budget() {
+        let storage: Arc<dyn HealStorageAPI> = Arc::new(MockStorage);
+        let mut request = HealRequest::object("retry-transition".to_string(), "object".to_string(), None);
+        request.options.timeout = Some(Duration::from_secs(60));
+        let task = HealTask::from_request(request, storage);
+        let result = task.execute().await;
+
+        let (retry_request, _, _) = retry_request_for_result_with_budget(&task, &result)
+            .await
+            .expect("read quorum failure should retain the unused timeout budget");
+        let remaining = retry_request
+            .options
+            .timeout
+            .expect("configured timeout should remain present");
+        assert!(remaining < Duration::from_secs(60));
+        assert!(remaining > Duration::from_secs(59));
+    }
+
    #[test]
    fn test_retry_request_for_incomplete_heal_rename() {
        let storage: Arc<dyn HealStorageAPI> = Arc::new(MockStorage);
@@ -6054,7 +6079,7 @@ mod tests {
        process_manager_queue_once(&manager).await;
        let defaulted_status = tokio::time::timeout(Duration::from_secs(1), async {
            loop {
-                if let Ok(status @ HealTaskStatus::Retrying { .. }) = manager.get_task_status(&defaulted_id).await {
+                if let Ok(status @ HealTaskStatus::Timeout) = manager.get_task_status(&defaulted_id).await {
                    break status;
                }
                tokio::task::yield_now().await;
@@ -6062,23 +6087,8 @@ mod tests {
        })
        .await
        .expect("configured timeout should finish the task");
-        assert!(matches!(defaulted_status, HealTaskStatus::Retrying { .. }));
-        assert_eq!(
-            manager
-                .retrying_heals
-                .lock()
-                .await
-                .get(&defaulted_id)
-                .expect("timed out task should retain its retry request")
-                .request
-                .options
-                .timeout,
-            Some(Duration::ZERO)
-        );
-        manager
-            .cancel_task(&defaulted_id)
-            .await
-            .expect("retrying timeout task should be cancelled");
+        assert_eq!(defaulted_status, HealTaskStatus::Timeout);
+        assert!(manager.retrying_heals.lock().await.get(&defaulted_id).is_none());

        let mut explicit = bucket_request("explicit-timeout", HealPriority::Normal, HealRequestSource::Admin);
        explicit.options.timeout = Some(Duration::from_secs(60));
@@ -196,7 +196,7 @@ pub struct HealOptions {
    /// Whether to skip namespace locking
    #[serde(default)]
    pub no_lock: bool,
-    /// Timeout
+    /// Aggregate execution timeout across recoverable manager retries
    pub timeout: Option<Duration>,
    /// pool index
    pub pool_index: Option<usize>,
@@ -442,6 +442,14 @@ impl HealTask {
        }
    }

+    pub(crate) async fn retry_request_with_remaining_timeout(&self) -> Result<HealRequest> {
+        let mut request = self.retry_request();
+        if self.options.timeout.is_some() {
+            request.options.timeout = self.remaining_timeout().await?;
+        }
+        Ok(request)
+    }
+
    pub(crate) fn from_replacement_recovery_request(
        request: HealRequest,
        storage: Arc<dyn HealStorageAPI>,
@@ -2657,6 +2665,36 @@ mod tests {

    use super::super::storage_api::status::BucketInfo;

+    #[tokio::test]
+    async fn retry_request_carries_remaining_timeout_budget() {
+        let storage: Arc<dyn HealStorageAPI> = Arc::new(MockStorage::default());
+        let mut request = HealRequest::bucket("bucket".to_string());
+        request.options.timeout = Some(Duration::from_secs(100));
+        let task = HealTask::from_request(request, storage.clone());
+        *task.task_start_instant.write().await = Some(Instant::now() - Duration::from_secs(40));
+
+        let retry = task
+            .retry_request_with_remaining_timeout()
+            .await
+            .expect("first retry should retain the unused timeout budget");
+        let first_remaining = retry.options.timeout.expect("configured timeout should remain present");
+        assert!(first_remaining <= Duration::from_secs(60));
+        assert!(first_remaining > Duration::from_secs(59));
+
+        let retry_task = HealTask::from_request(retry, storage);
+        *retry_task.task_start_instant.write().await = Some(Instant::now() - Duration::from_secs(20));
+        let second_retry = retry_task
+            .retry_request_with_remaining_timeout()
+            .await
+            .expect("second retry should retain only the unused aggregate budget");
+        let second_remaining = second_retry
+            .options
+            .timeout
+            .expect("configured timeout should remain present");
+        assert!(second_remaining <= Duration::from_secs(40));
+        assert!(second_remaining > Duration::from_secs(39));
+    }
+
    #[test]
    fn format_result_requires_every_requested_target_to_be_ok() {
        let result = HealResultItem {
@@ -225,8 +225,6 @@ async fn nothing_readable_leaves_the_bundle_unwrapped() {
            "artifact {} carries the raw on-disk record",
            artifact.path
        );
-        // A cheap structural check too: an encrypted payload is not JSON.
-        assert_ne!(payload.first(), Some(&b'{'), "artifact {} looks like plaintext JSON", artifact.path);
    }

    // The manifest itself is not encrypted, so assert directly that it carries
@@ -15,10 +15,8 @@
 use std::collections::HashMap;
 use std::hash::{Hash, Hasher};
 use std::sync::Arc;
-use std::sync::atomic::{AtomicBool, Ordering};
 use std::time::{Duration, SystemTime};
 use tokio::sync::RwLock;
-use tokio::time::Instant;

 use crate::{
    FastLockGuard, GlobalLockManager, LockClient, LockId, LockInfo, LockManager, LockMetadata, LockPriority, LockRequest,
@@ -28,51 +26,43 @@ use crate::{
 /// Default shard count for guard storage (must be power of 2)
 const DEFAULT_GUARD_SHARD_COUNT: usize = 64;

-type GuardShard = Arc<RwLock<HashMap<LockId, LocalGuardEntry>>>;
-type GuardStorage = Arc<Vec<GuardShard>>;
-
 /// Local lock client using FastLock with sharded guard storage for better concurrency
 #[derive(Debug)]
 pub struct LocalClient {
    /// Sharded guard storage to reduce lock contention
-    guard_storage: GuardStorage,
+    guard_storage: Vec<Arc<RwLock<HashMap<LockId, LocalGuardEntry>>>>,
    /// Mask for fast shard index calculation (shard_count - 1)
    shard_mask: usize,
    /// Optional lock manager (if None, uses global singleton)
    manager: Option<Arc<GlobalLockManager>>,
-    reaper_started: AtomicBool,
-    reaper_interval: Duration,
 }

 #[derive(Debug)]
 struct LocalGuardEntry {
    guard: FastLockGuard,
    expires_at: SystemTime,
-    deadline: Instant,
    ttl: Duration,
+    /// Owner recorded at acquire time; used only for reclaim diagnostics (#899).
+    owner: String,
 }

 impl LocalGuardEntry {
-    fn new(guard: FastLockGuard, ttl: Duration) -> Self {
+    fn new(guard: FastLockGuard, ttl: Duration, owner: String) -> Self {
        let now = SystemTime::now();
-        let monotonic_now = Instant::now();
        Self {
            guard,
-            expires_at: now.checked_add(ttl).unwrap_or(now),
-            deadline: monotonic_now.checked_add(ttl).unwrap_or(monotonic_now),
+            expires_at: now + ttl,
            ttl,
+            owner,
        }
    }

    fn is_expired(&self) -> bool {
-        self.deadline <= Instant::now()
+        self.expires_at <= SystemTime::now()
    }

    fn refresh(&mut self) {
-        let now = SystemTime::now();
-        let monotonic_now = Instant::now();
-        self.expires_at = now.checked_add(self.ttl).unwrap_or(now);
-        self.deadline = monotonic_now.checked_add(self.ttl).unwrap_or(monotonic_now);
+        self.expires_at = SystemTime::now() + self.ttl;
    }
 }

@@ -87,38 +77,26 @@ impl LocalClient {
    pub fn with_shard_count(shard_count: usize) -> Self {
        assert!(shard_count.is_power_of_two(), "Shard count must be power of 2");

-        let guard_storage: Vec<GuardShard> = (0..shard_count).map(|_| Arc::new(RwLock::new(HashMap::new()))).collect();
+        let guard_storage: Vec<Arc<RwLock<HashMap<LockId, LocalGuardEntry>>>> =
+            (0..shard_count).map(|_| Arc::new(RwLock::new(HashMap::new()))).collect();

-        Self::with_storage(Arc::new(guard_storage), None, crate::fast_lock::CLEANUP_INTERVAL)
-    }
-
-    fn with_storage(guard_storage: GuardStorage, manager: Option<Arc<GlobalLockManager>>, reaper_interval: Duration) -> Self {
-        let shard_count = guard_storage.len();
-        debug_assert!(shard_count.is_power_of_two());
        Self {
            guard_storage,
            shard_mask: shard_count - 1,
-            manager,
-            reaper_started: AtomicBool::new(false),
-            reaper_interval,
+            manager: None,
        }
    }

    /// Create new local client with a specific lock manager
    /// This allows simulating multi-node environments where each node has its own lock backend
    pub fn with_manager(manager: Arc<GlobalLockManager>) -> Self {
-        let guard_storage = (0..DEFAULT_GUARD_SHARD_COUNT)
-            .map(|_| Arc::new(RwLock::new(HashMap::new())))
-            .collect();
-        Self::with_storage(Arc::new(guard_storage), Some(manager), crate::fast_lock::CLEANUP_INTERVAL)
-    }
-
-    #[cfg(test)]
-    pub(crate) fn with_manager_and_reaper_interval(manager: Arc<GlobalLockManager>, reaper_interval: Duration) -> Self {
-        let guard_storage = (0..DEFAULT_GUARD_SHARD_COUNT)
-            .map(|_| Arc::new(RwLock::new(HashMap::new())))
-            .collect();
-        Self::with_storage(Arc::new(guard_storage), Some(manager), reaper_interval)
+        Self {
+            guard_storage: (0..DEFAULT_GUARD_SHARD_COUNT)
+                .map(|_| Arc::new(RwLock::new(HashMap::new())))
+                .collect(),
+            shard_mask: DEFAULT_GUARD_SHARD_COUNT - 1,
+            manager: Some(manager),
+        }
    }

    /// Get the lock manager (injected manager if available, otherwise global singleton)
@@ -140,60 +118,51 @@ impl LocalClient {
    }

    async fn reclaim_expired_guards_for_resource(&self, resource: &crate::ObjectKey) -> usize {
-        let expired_entries = Self::extract_expired_guards(&self.guard_storage, Some(resource)).await;
-        Self::release_reclaimed_guards(expired_entries, Some(resource))
-    }
+        let mut reclaimed = 0usize;

-    async fn extract_expired_guards(storage: &GuardStorage, resource: Option<&crate::ObjectKey>) -> Vec<LocalGuardEntry> {
-        let mut expired_entries = Vec::new();
-        for shard in storage.iter() {
-            let mut guards = shard.write().await;
-            expired_entries.extend(
-                guards
-                    .extract_if(|lock_id, entry| {
-                        resource.is_none_or(|resource| &lock_id.resource == resource) && entry.is_expired()
-                    })
-                    .map(|(_, entry)| entry),
-            );
-        }
-        expired_entries
-    }
+        for shard in &self.guard_storage {
+            let expired_entries = {
+                let mut guards = shard.write().await;
+                let mut retained = HashMap::with_capacity(guards.len());
+                let mut expired_entries = Vec::new();

-    fn release_reclaimed_guards(mut entries: Vec<LocalGuardEntry>, resource: Option<&crate::ObjectKey>) -> usize {
-        let reclaimed = entries.len();
-        for entry in &mut entries {
-            let _ = entry.guard.release();
-        }
-        if reclaimed > 0 {
-            for _ in 0..reclaimed {
+                for (lock_id, entry) in std::mem::take(&mut *guards) {
+                    if &lock_id.resource == resource && entry.is_expired() {
+                        expired_entries.push(entry);
+                    } else {
+                        retained.insert(lock_id, entry);
+                    }
+                }
+
+                *guards = retained;
+                expired_entries
+            };
+
+            for mut entry in expired_entries {
+                // An expired entry whose owner never refreshed it (a dead coordinator, #698) is
+                // reclaimed so a live contender can re-form quorum. With guard heartbeats in place
+                // (#899) a live owner keeps its entry from expiring, so reaching here means the
+                // lease genuinely lapsed. Surface it for observability; the reclaim decision itself
+                // is unchanged.
+                let since_last_refresh = entry
+                    .expires_at
+                    .checked_sub(entry.ttl)
+                    .and_then(|last_refresh| SystemTime::now().duration_since(last_refresh).ok())
+                    .unwrap_or(entry.ttl);
+                tracing::warn!(
+                    owner = %entry.owner,
+                    resource = %resource,
+                    ttl_ms = entry.ttl.as_millis() as u64,
+                    since_last_refresh_ms = since_last_refresh.as_millis() as u64,
+                    "reclaiming expired lock guard whose lease was not refreshed"
+                );
                rustfs_io_metrics::record_lock_reclaimed();
-            }
-            if let Some(resource) = resource {
-                tracing::debug!(event = "lock_guard_reclaimed", resource = %resource, count = reclaimed, "expired lock guards reclaimed");
-            } else {
-                tracing::debug!(event = "lock_guard_reaper_sweep", count = reclaimed, "expired lock guards reclaimed");
+                let _ = entry.guard.release();
+                reclaimed = reclaimed.saturating_add(1);
            }
        }
-        reclaimed
-    }

-    fn ensure_reaper(&self) {
-        if self.reaper_started.swap(true, Ordering::AcqRel) {
-            return;
-        }
-        let storage = Arc::downgrade(&self.guard_storage);
-        let interval = self.reaper_interval;
-        tokio::spawn(async move {
-            let mut ticker = tokio::time::interval(interval);
-            loop {
-                ticker.tick().await;
-                let Some(storage) = storage.upgrade() else {
-                    break;
-                };
-                let expired_entries = Self::extract_expired_guards(&storage, None).await;
-                Self::release_reclaimed_guards(expired_entries, None);
-            }
-        });
+        reclaimed
    }
 }

@@ -206,36 +175,28 @@ impl Default for LocalClient {
 #[async_trait::async_trait]
 impl LockClient for LocalClient {
    async fn acquire_lock(&self, request: &LockRequest) -> Result<LockResponse> {
-        self.ensure_reaper();
        let lock_manager = self.get_lock_manager();
        let reclaimed_before_acquire = self.reclaim_expired_guards_for_resource(&request.resource).await;
-        let acquire_deadline = Instant::now()
-            .checked_add(request.acquire_timeout)
-            .unwrap_or_else(Instant::now);

-        let build_lock_request = |acquire_timeout| match request.lock_type {
+        let build_lock_request = || match request.lock_type {
            LockType::Exclusive => crate::ObjectLockRequest::new_write(request.resource.clone(), request.owner.clone())
-                .with_acquire_timeout(acquire_timeout),
+                .with_acquire_timeout(request.acquire_timeout),
            LockType::Shared => crate::ObjectLockRequest::new_read(request.resource.clone(), request.owner.clone())
-                .with_acquire_timeout(acquire_timeout),
+                .with_acquire_timeout(request.acquire_timeout),
        };

        let mut retried_after_reclaim = reclaimed_before_acquire > 0;
        loop {
-            let remaining = acquire_deadline.saturating_duration_since(Instant::now());
-            if remaining.is_zero() {
-                return Ok(LockResponse::failure("Lock acquisition timeout", request.acquire_timeout));
-            }
-            match lock_manager.acquire_lock(build_lock_request(remaining)).await {
+            match lock_manager.acquire_lock(build_lock_request()).await {
                Ok(guard) => {
                    let lock_id = request.lock_id.clone();
                    let acquired_at = SystemTime::now();
-                    let expires_at = acquired_at.checked_add(request.ttl).unwrap_or(acquired_at);
+                    let expires_at = acquired_at + request.ttl;

                    {
                        let shard = self.get_shard(&lock_id);
                        let mut guards = shard.write().await;
-                        guards.insert(lock_id.clone(), LocalGuardEntry::new(guard, request.ttl));
+                        guards.insert(lock_id.clone(), LocalGuardEntry::new(guard, request.ttl, request.owner.clone()));
                    }

                    let lock_info = LockInfo {
@@ -356,111 +317,3 @@ impl LockClient for LocalClient {
        true
    }
 }
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use crate::{GlobalLockManager, LockClient, LockRequest, LockType};
-
-    fn request(resource: crate::ObjectKey, owner: &str, ttl: Duration) -> LockRequest {
-        LockRequest::new(resource, LockType::Exclusive, owner)
-            .with_ttl(ttl)
-            .with_acquire_timeout(Duration::from_millis(80))
-    }
-
-    async fn wait_until_reaped(client: &LocalClient, lock_id: &LockId) {
-        for _ in 0..80 {
-            if client.check_status(lock_id).await.unwrap().is_none() {
-                return;
-            }
-            tokio::time::sleep(Duration::from_millis(5)).await;
-        }
-        panic!("lock guard was not reaped before test deadline");
-    }
-
-    #[tokio::test(flavor = "current_thread")]
-    async fn expired_guard_is_reaped_without_resource_reacquire() {
-        let manager = Arc::new(GlobalLockManager::new());
-        let client = LocalClient::with_manager_and_reaper_interval(manager.clone(), Duration::from_millis(5));
-        let request = request(crate::ObjectKey::new("bucket", "unique-chunk"), "owner-a", Duration::from_millis(10));
-        let lock_id = request.lock_id.clone();
-
-        assert!(client.acquire_lock(&request).await.unwrap().success);
-        assert!(client.check_status(&lock_id).await.unwrap().is_some());
-        tokio::time::sleep(Duration::from_millis(15)).await;
-        wait_until_reaped(&client, &lock_id).await;
-
-        let direct = manager
-            .acquire_lock(crate::ObjectLockRequest::new_write(request.resource.clone(), "owner-b"))
-            .await;
-        assert!(direct.is_ok());
-    }
-
-    #[tokio::test(flavor = "current_thread")]
-    async fn sibling_client_cannot_reclaim_but_owner_reaper_releases_shared_lock() {
-        let manager = Arc::new(GlobalLockManager::new());
-        let owner = LocalClient::with_manager_and_reaper_interval(manager.clone(), Duration::from_millis(5));
-        let contender = LocalClient::with_manager_and_reaper_interval(manager, Duration::from_millis(5));
-        let request_a = request(crate::ObjectKey::new("bucket", "shared-resource"), "owner-a", Duration::from_millis(10));
-        assert!(owner.acquire_lock(&request_a).await.unwrap().success);
-
-        let request_b = request(request_a.resource.clone(), "owner-b", Duration::from_millis(20))
-            .with_acquire_timeout(Duration::from_millis(5));
-        assert!(!contender.acquire_lock(&request_b).await.unwrap().success);
-
-        tokio::time::sleep(Duration::from_millis(25)).await;
-        assert!(owner.check_status(&request_a.lock_id).await.unwrap().is_none());
-        assert!(contender.acquire_lock(&request_b).await.unwrap().success);
-    }
-
-    #[tokio::test(flavor = "current_thread")]
-    async fn refresh_wins_before_deadline_and_reaper_wins_after_deadline() {
-        let manager = Arc::new(GlobalLockManager::new());
-        let client = LocalClient::with_manager_and_reaper_interval(manager, Duration::from_millis(5));
-        let request = request(crate::ObjectKey::new("bucket", "refresh-race"), "owner-a", Duration::from_millis(25));
-        let lock_id = request.lock_id.clone();
-        assert!(client.acquire_lock(&request).await.unwrap().success);
-
-        tokio::time::sleep(Duration::from_millis(10)).await;
-        assert!(client.refresh(&lock_id).await.unwrap());
-        tokio::time::sleep(Duration::from_millis(15)).await;
-        assert!(client.check_status(&lock_id).await.unwrap().is_some());
-        wait_until_reaped(&client, &lock_id).await;
-    }
-
-    #[tokio::test(flavor = "current_thread")]
-    async fn zero_ttl_is_reaped_and_oversized_ttl_does_not_panic() {
-        let manager = Arc::new(GlobalLockManager::new());
-        let client = LocalClient::with_manager_and_reaper_interval(manager, Duration::from_millis(5));
-
-        let zero = request(crate::ObjectKey::new("bucket", "zero-ttl"), "owner-zero", Duration::ZERO);
-        let zero_id = zero.lock_id.clone();
-        assert!(client.acquire_lock(&zero).await.unwrap().success);
-        wait_until_reaped(&client, &zero_id).await;
-
-        let huge = request(crate::ObjectKey::new("bucket", "huge-ttl"), "owner-huge", Duration::MAX);
-        let huge_id = huge.lock_id.clone();
-        assert!(client.acquire_lock(&huge).await.unwrap().success);
-        wait_until_reaped(&client, &huge_id).await;
-    }
-
-    #[tokio::test(flavor = "current_thread")]
-    async fn acquire_retry_preserves_total_deadline() {
-        let manager = Arc::new(GlobalLockManager::new());
-        let client = LocalClient::with_manager_and_reaper_interval(manager, Duration::from_secs(60));
-        let first = request(crate::ObjectKey::new("bucket", "deadline-budget"), "owner-a", Duration::from_millis(10));
-        assert!(client.acquire_lock(&first).await.unwrap().success);
-
-        let second =
-            request(first.resource.clone(), "owner-b", Duration::from_millis(30)).with_acquire_timeout(Duration::from_millis(60));
-        let started = Instant::now();
-        let response = client.acquire_lock(&second).await.unwrap();
-        assert!(!response.success, "the first attempt consumed the caller's acquire budget");
-        assert!(
-            started.elapsed() < Duration::from_millis(100),
-            "reclaim retry must not double the acquire budget"
-        );
-        let recovered = client.acquire_lock(&second).await.unwrap();
-        assert!(recovered.success, "the reclaimed guard must be available to the next request");
-    }
-}
@@ -840,117 +840,6 @@ async fn test_namespace_lock_distributed_reclaims_expired_same_resource_after_fa
    );
 }

-#[tokio::test]
-async fn four_node_failed_release_converges_without_replica_repair() {
-    let managers = (0..4).map(|_| Arc::new(GlobalLockManager::new())).collect::<Vec<_>>();
-    let flaky_clients = managers
-        .iter()
-        .map(|manager| {
-            Arc::new(FlakyReleaseClient {
-                inner: LocalClient::with_manager_and_reaper_interval(manager.clone(), Duration::from_millis(5)),
-                failed_releases_remaining: AtomicUsize::new(usize::MAX),
-                release_attempts: AtomicUsize::new(0),
-            })
-        })
-        .collect::<Vec<_>>();
-    let clients = flaky_clients
-        .iter()
-        .map(|client| client.clone() as Arc<dyn LockClient>)
-        .collect::<Vec<_>>();
-    let lock = NamespaceLock::Distributed(DistributedLock::new("four-node-expired-lease".to_string(), clients, 3));
-    let resource = create_test_object_key("bucket", "object-four-node-expired");
-    let request = LockRequest::new(resource.clone(), LockType::Exclusive, "owner-a")
-        .with_acquire_timeout(Duration::from_millis(300))
-        .with_ttl(Duration::from_millis(40));
-
-    let mut guard = lock
-        .acquire_guard(&request)
-        .await
-        .expect("initial acquire should not error")
-        .expect("initial acquire should reach quorum");
-    assert!(guard.release(), "release should be acknowledged while RPC cleanup is pending");
-
-    for _ in 0..40 {
-        if flaky_clients.iter().all(|client| client.release_attempts() >= 3) {
-            break;
-        }
-        tokio::time::sleep(Duration::from_millis(5)).await;
-    }
-
-    let deadline = tokio::time::Instant::now() + Duration::from_secs(2);
-    loop {
-        let all_reaped =
-            futures::future::join_all(flaky_clients.iter().map(|client| client.inner.check_status(&request.lock_id)))
-                .await
-                .into_iter()
-                .all(|status| status.expect("status should not error").is_none());
-        if all_reaped {
-            break;
-        }
-        assert!(tokio::time::Instant::now() < deadline, "all four local lease entries must converge");
-        tokio::time::sleep(Duration::from_millis(10)).await;
-    }
-
-    for suffix in ["chunk-0", "chunk-1", ".rustfs.sys/multipart/upload-0"] {
-        for client in &flaky_clients {
-            let orphan = LockRequest::new(create_test_object_key("bucket", suffix), LockType::Exclusive, "orphan")
-                .with_ttl(Duration::from_millis(25));
-            assert!(client.inner.acquire_lock(&orphan).await.expect("orphan acquire").success);
-        }
-    }
-    tokio::time::sleep(Duration::from_millis(80)).await;
-
-    let recovered = lock
-        .acquire_guard(
-            &LockRequest::new(resource, LockType::Exclusive, "owner-b")
-                .with_acquire_timeout(Duration::from_millis(300))
-                .with_ttl(Duration::from_millis(40)),
-        )
-        .await
-        .expect("recovery acquire should not error")
-        .expect("four-node quorum should recover after local reapers run");
-    drop(recovered);
-}
-
-#[tokio::test]
-async fn four_node_stale_quorum_contention_respects_acquire_deadline() {
-    let managers = (0..4).map(|_| Arc::new(GlobalLockManager::new())).collect::<Vec<_>>();
-    let node_clients = managers
-        .iter()
-        .map(|manager| Arc::new(LocalClient::with_manager_and_reaper_interval(manager.clone(), Duration::from_millis(5))))
-        .collect::<Vec<_>>();
-    let resource = create_test_object_key("bucket", "stale-quorum");
-    let stale = LockRequest::new(resource.clone(), LockType::Exclusive, "stale-owner").with_ttl(Duration::from_millis(180));
-    for client in &node_clients {
-        assert!(client.acquire_lock(&stale).await.expect("stale acquire").success);
-    }
-
-    let clients = node_clients
-        .iter()
-        .map(|client| client.clone() as Arc<dyn LockClient>)
-        .collect::<Vec<_>>();
-    let lock = NamespaceLock::Distributed(DistributedLock::new("stale-quorum-deadline".to_string(), clients, 3));
-    let contender = LockRequest::new(resource.clone(), LockType::Exclusive, "new-owner")
-        .with_acquire_timeout(Duration::from_millis(150))
-        .with_ttl(Duration::from_millis(100));
-    let started = tokio::time::Instant::now();
-    let response = lock.acquire_guard(&contender).await.expect("contention should not error");
-    assert!(response.is_none(), "unexpired leases must not be force-reclaimed");
-    assert!(started.elapsed() < Duration::from_millis(350), "acquire must respect its deadline");
-
-    tokio::time::sleep(Duration::from_millis(80)).await;
-    let recovered = lock
-        .acquire_guard(
-            &LockRequest::new(resource, LockType::Exclusive, "new-owner")
-                .with_acquire_timeout(Duration::from_millis(300))
-                .with_ttl(Duration::from_millis(100)),
-        )
-        .await
-        .expect("post-expiry acquire should not error")
-        .expect("quorum should recover after local reapers clear stale leases");
-    drop(recovered);
-}
-
 #[tokio::test]
 async fn test_namespace_lock_distributed_retries_transient_acquire_timeout() {
    let managers = (0..3).map(|_| Arc::new(GlobalLockManager::new())).collect::<Vec<_>>();
@@ -103,11 +103,7 @@ pub(super) fn rules() -> Vec<Rule> {
                P2Degraded,
                "heal",
                "heal 任务调度/执行失败",
-                any([
-                    prefix("Heal task timeout"),
-                    prefix("Heal task execution failed"),
-                    contains("Heal manager is not running"),
-                ]),
+                any([prefix("Heal task timeout"), prefix("Heal task execution failed")]),
                "heal 任务调度/执行层故障。",
                "检查 heal 后台服务状态与资源压力。",
            )
@@ -748,9 +748,10 @@ async fn handle_authenticated_request(
                                }

                                for (key, value) in info.user_defined.iter() {
-                                    if key != "content-type" {
-                                        let header_name = format!("x-object-meta-{}", key);
-                                        response = response.header(header_name, value.as_str());
+                                    if key != "content-type"
+                                        && let Some(key) = object::swift_response_user_metadata_key(key)
+                                    {
+                                        response = response.header(format!("x-object-meta-{key}"), value.as_str());
                                    }
                                }

@@ -817,9 +818,10 @@ async fn handle_authenticated_request(

                    // Add custom metadata headers (X-Object-Meta-*)
                    for (key, value) in info.user_defined.iter() {
-                        if key != "content-type" {
-                            let header_name = format!("x-object-meta-{}", key);
-                            response = response.header(header_name, value.as_str());
+                        if key != "content-type"
+                            && let Some(key) = object::swift_response_user_metadata_key(key)
+                        {
+                            response = response.header(format!("x-object-meta-{key}"), value.as_str());
                        }
                    }

@@ -856,9 +858,10 @@ async fn handle_authenticated_request(

                    // Add custom metadata headers (X-Object-Meta-*)
                    for (key, value) in info.user_defined.iter() {
-                        if key != "content-type" {
-                            let header_name = format!("x-object-meta-{}", key);
-                            response = response.header(header_name, value.as_str());
+                        if key != "content-type"
+                            && let Some(key) = object::swift_response_user_metadata_key(key)
+                        {
+                            response = response.header(format!("x-object-meta-{key}"), value.as_str());
                        }
                    }

@@ -1168,9 +1171,10 @@ async fn handle_object_get(
                }

                for (key, value) in info.user_defined.iter() {
-                    if key != "content-type" {
-                        let header_name = format!("x-object-meta-{}", key);
-                        response = response.header(header_name, value.as_str());
+                    if key != "content-type"
+                        && let Some(key) = object::swift_response_user_metadata_key(key)
+                    {
+                        response = response.header(format!("x-object-meta-{key}"), value.as_str());
                    }
                }

@@ -1237,9 +1241,10 @@ async fn handle_object_get(
        if key == "x-delete-at" {
            // Add X-Delete-At header directly (not as X-Object-Meta-*)
            response = response.header("x-delete-at", value.as_str());
-        } else if key != "content-type" {
-            let header_name = format!("x-object-meta-{}", key);
-            response = response.header(header_name, value.as_str());
+        } else if key != "content-type"
+            && let Some(key) = object::swift_response_user_metadata_key(key)
+        {
+            response = response.header(format!("x-object-meta-{key}"), value.as_str());
        }
    }

@@ -1293,9 +1298,10 @@ async fn handle_object_head(
        if key == "x-delete-at" {
            // Add X-Delete-At header directly (not as X-Object-Meta-*)
            response = response.header("x-delete-at", value.as_str());
-        } else if key != "content-type" {
-            let header_name = format!("x-object-meta-{}", key);
-            response = response.header(header_name, value.as_str());
+        } else if key != "content-type"
+            && let Some(key) = object::swift_response_user_metadata_key(key)
+        {
+            response = response.header(format!("x-object-meta-{key}"), value.as_str());
        }
    }

@@ -67,10 +67,57 @@ const LOG_COMPONENT_PROTOCOLS: &str = "protocols";
 const LOG_SUBSYSTEM_SWIFT_OBJECT: &str = "swift_object";
 const EVENT_SWIFT_OBJECT_STORAGE_STATE: &str = "swift_object_storage_state";
 const SWIFT_DELETE_AT_METADATA: &str = "x-delete-at";
+const USER_METADATA_PREFIX: &str = "x-amz-meta-";

 /// Maximum object size in bytes (5GB - Swift default)
 const MAX_OBJECT_SIZE: i64 = 5 * 1024 * 1024 * 1024;

+fn stored_swift_user_metadata_key(key: &str) -> String {
+    if rustfs_utils::http::is_internal_key(key)
+        || rustfs_utils::http::starts_with_ignore_ascii_case(key, "x-amz-")
+        || rustfs_utils::http::starts_with_ignore_ascii_case(key, "x-rustfs-encryption-")
+        || rustfs_utils::http::starts_with_ignore_ascii_case(key, "x-minio-encryption-")
+    {
+        format!("{USER_METADATA_PREFIX}{key}")
+    } else {
+        key.to_string()
+    }
+}
+
+pub(super) fn swift_response_user_metadata_key(key: &str) -> Option<&str> {
+    if rustfs_utils::http::is_internal_key(key)
+        || rustfs_utils::http::starts_with_ignore_ascii_case(key, "x-rustfs-encryption-")
+        || rustfs_utils::http::starts_with_ignore_ascii_case(key, "x-minio-encryption-")
+    {
+        return None;
+    }
+    if let Some(unescaped) = key.strip_prefix(USER_METADATA_PREFIX)
+        && (rustfs_utils::http::is_internal_key(unescaped)
+            || rustfs_utils::http::starts_with_ignore_ascii_case(unescaped, "x-amz-")
+            || rustfs_utils::http::starts_with_ignore_ascii_case(unescaped, "x-rustfs-encryption-")
+            || rustfs_utils::http::starts_with_ignore_ascii_case(unescaped, "x-minio-encryption-"))
+    {
+        return Some(unescaped);
+    }
+    Some(key)
+}
+
+fn swift_user_metadata(headers: &HeaderMap) -> Option<HashMap<String, String>> {
+    let mut metadata = HashMap::new();
+    let mut present = false;
+    for (header_name, header_value) in headers.iter() {
+        let header_name = header_name.as_str().to_lowercase();
+        let Some(key) = header_name.strip_prefix("x-object-meta-") else {
+            continue;
+        };
+        present = true;
+        if let Ok(value) = header_value.to_str() {
+            metadata.insert(stored_swift_user_metadata_key(key), value.to_string());
+        }
+    }
+    present.then_some(metadata)
+}
+
 /// Object key translator for Swift object names
 ///
 /// Handles URL encoding/decoding and path normalization for Swift object keys.
@@ -303,15 +350,7 @@ where
    let bucket = mapper.swift_to_s3_bucket(container, &project_id);

    // 5. Extract Swift metadata from X-Object-Meta-* headers
-    let mut user_metadata = HashMap::new();
-    for (header_name, header_value) in headers.iter() {
-        let header_str = header_name.as_str().to_lowercase();
-        if let Some(meta_key) = header_str.strip_prefix("x-object-meta-")
-            && let Ok(value_str) = header_value.to_str()
-        {
-            user_metadata.insert(meta_key.to_string(), value_str.to_string());
-        }
-    }
+    let mut user_metadata = swift_user_metadata(headers).unwrap_or_default();

    // 6. Extract Content-Type if provided
    if let Some(content_type) = headers.get("content-type")
@@ -739,15 +778,7 @@ pub async fn update_object_metadata(
    }

    // 8. Extract new metadata from X-Object-Meta-* headers
-    let mut new_metadata = HashMap::new();
-    for (header_name, header_value) in headers.iter() {
-        let header_str = header_name.as_str().to_lowercase();
-        if let Some(meta_key) = header_str.strip_prefix("x-object-meta-")
-            && let Ok(value_str) = header_value.to_str()
-        {
-            new_metadata.insert(meta_key.to_string(), value_str.to_string());
-        }
-    }
+    let mut new_metadata = swift_user_metadata(headers).unwrap_or_default();

    // 9. Also update Content-Type if provided
    if let Some(content_type) = headers.get("content-type")
@@ -889,19 +920,8 @@ pub async fn copy_object(
    let mut new_metadata = (*src_info.user_defined).clone();

    // 11. If custom metadata headers provided, use those instead (Swift behavior)
-    let mut has_custom_meta = false;
-    for (header_name, header_value) in headers.iter() {
-        let header_str = header_name.as_str().to_lowercase();
-        if let Some(meta_key) = header_str.strip_prefix("x-object-meta-") {
-            if !has_custom_meta {
-                // First custom meta header - clear source metadata
-                new_metadata.clear();
-                has_custom_meta = true;
-            }
-            if let Ok(value_str) = header_value.to_str() {
-                new_metadata.insert(meta_key.to_string(), value_str.to_string());
-            }
-        }
+    if let Some(custom_metadata) = swift_user_metadata(headers) {
+        new_metadata = custom_metadata;
    }

    // 12. Also check for Content-Type override
@@ -1123,6 +1143,36 @@ mod tests {
        assert!(ObjectKeyMapper::validate_object_name("unicode-文件.txt").is_ok());
    }

+    #[test]
+    fn swift_user_metadata_cannot_materialize_internal_storage_keys() {
+        let mut headers = HeaderMap::new();
+        headers.insert("x-object-meta-x-rustfs-internal-actual-size", "1".parse().expect("valid metadata value"));
+        headers.insert("x-object-meta-description", "safe".parse().expect("valid metadata value"));
+        let metadata = swift_user_metadata(&headers).expect("custom metadata should be detected");
+
+        assert_eq!(metadata.get("x-amz-meta-x-rustfs-internal-actual-size").map(String::as_str), Some("1"));
+        assert_eq!(metadata.get("description").map(String::as_str), Some("safe"));
+        assert!(!metadata.contains_key("x-rustfs-internal-actual-size"));
+        assert_eq!(
+            stored_swift_user_metadata_key("x-minio-encryption-original-size"),
+            "x-amz-meta-x-minio-encryption-original-size"
+        );
+        assert_eq!(stored_swift_user_metadata_key("description"), "description");
+    }
+
+    #[test]
+    fn swift_user_metadata_response_mapping_is_reversible_and_filters_internal_keys() {
+        assert_eq!(
+            swift_response_user_metadata_key("x-amz-meta-x-rustfs-internal-actual-size"),
+            Some("x-rustfs-internal-actual-size")
+        );
+        assert_eq!(swift_response_user_metadata_key("x-amz-meta-x-amz-checksum"), Some("x-amz-checksum"));
+        assert_eq!(swift_response_user_metadata_key("x-amz-meta-description"), Some("x-amz-meta-description"));
+        assert_eq!(swift_response_user_metadata_key("description"), Some("description"));
+        assert_eq!(swift_response_user_metadata_key("x-rustfs-internal-actual-size"), None);
+        assert_eq!(swift_response_user_metadata_key("x-minio-internal-actual-size"), None);
+    }
+
    #[test]
    fn test_validate_object_name_empty() {
        let result = ObjectKeyMapper::validate_object_name("");
@@ -520,6 +520,14 @@ struct FailureSample {
 pub struct FailStats {
    pub count: i64,
    pub size: i64,
+    /// Rolling-window snapshots refreshed at collection time
+    /// ([`Self::refresh_windows`]). The raw samples (`recent`) are process
+    /// local (serde-skipped), so these fields are what survives the peer-RPC
+    /// wire and [`Self::merge`]-based cluster aggregation.
+    #[serde(default)]
+    pub last_minute: FailedMetric,
+    #[serde(default)]
+    pub last_hour: FailedMetric,
    #[serde(skip)]
    recent: VecDeque<FailureSample>,
 }
@@ -537,6 +545,17 @@ impl FailStats {
        self.prune(observed_at);
    }

+    /// Recompute the serializable rolling-window snapshots from the local
+    /// samples. Called at the collection point (per-node stats snapshot),
+    /// never on the failure hot path — the two deque scans are O(window) and
+    /// `add_size` runs under the bucket-stats write lock. Only meaningful on
+    /// the live per-node struct: a deserialized or merged struct has no
+    /// samples, and refreshing it would wipe the aggregated windows.
+    pub fn refresh_windows(&mut self) {
+        self.last_minute = self.recent_since(Duration::from_secs(60));
+        self.last_hour = self.recent_since(Duration::from_secs(3600));
+    }
+
    fn prune(&mut self, observed_at: Instant) {
        while self
            .recent
@@ -565,6 +584,16 @@ impl FailStats {
        Self {
            count: self.count.saturating_add(other.count),
            size: self.size.saturating_add(other.size),
+            // The window snapshots sum across nodes; the raw samples do not
+            // travel and stay empty on aggregated structs.
+            last_minute: FailedMetric {
+                count: self.last_minute.count.saturating_add(other.last_minute.count),
+                size: self.last_minute.size.saturating_add(other.last_minute.size),
+            },
+            last_hour: FailedMetric {
+                count: self.last_hour.count.saturating_add(other.last_hour.count),
+                size: self.last_hour.size.saturating_add(other.last_hour.size),
+            },
            recent: VecDeque::new(),
        }
    }
@@ -636,7 +665,9 @@ impl BucketReplicationStat {
    }

    pub fn update_xfer_rate(&mut self, size: i64, duration: Duration) {
-        if size > 1024 * 1024 {
+        // Same boundary as the worker-pool split and minio-go's
+        // Large/Small transfer-summary labels: >= 128 MiB is "large".
+        if size >= crate::runtime::MIN_LARGE_OBJ_SIZE {
            self.xfer_rate_lrg.add_size(size, duration);
        } else {
            self.xfer_rate_sml.add_size(size, duration);
@@ -71,6 +71,7 @@ where

    /// Optional: allow users to customize block_size
    pub fn with_block_size(inner: R, block_size: usize, compression_algorithm: CompressionAlgorithm) -> Self {
+        debug_assert!(block_size > 0, "CompressReader block_size must be non-zero");
        Self {
            inner,
            buffer: Vec::new(),
@@ -183,11 +184,21 @@ pin_project! {
        buffer: Vec<u8>,
        buffer_pos: usize,
        finished: bool,
+        // A previously surfaced stream error is sticky: without this, a caller
+        // that polls again after an error would restart at the header phase and
+        // read a truncated tail as a clean EOF, converting the error into a
+        // silently short body.
+        poisoned: bool,
        // Fields for saving header read progress across polls
        header_buf: [u8; 8],
        header_read: usize,
-        header_done: bool,
-        // Fields for saving compressed block read progress across polls
+        // Fields for saving compressed block read progress across polls.
+        // `compressed_len > 0` means a block payload is in flight: the header has
+        // been fully parsed and `compressed_read` bytes of the payload are already
+        // consumed from the inner stream. The header phase must not run again (and
+        // must not reset `compressed_read`) until this block completes, or a
+        // `Poll::Pending` in the middle of a payload would silently drop the bytes
+        // read so far and desynchronize the block framing.
        compressed_buf: Vec<u8>,
        compressed_read: usize,
        compressed_len: usize,
@@ -205,9 +216,9 @@ where
            buffer: Vec::new(),
            buffer_pos: 0,
            finished: false,
+            poisoned: false,
            header_buf: [0u8; 8],
            header_read: 0,
-            header_done: false,
            compressed_buf: Vec::new(),
            compressed_read: 0,
            compressed_len: 0,
@@ -236,54 +247,74 @@ where
        if *this.finished {
            return Poll::Ready(Ok(()));
        }
-        // Read header
-        while !*this.header_done && *this.header_read < HEADER_LEN {
-            let mut temp = [0u8; HEADER_LEN];
-            let mut temp_buf = ReadBuf::new(&mut temp[0..HEADER_LEN - *this.header_read]);
-            match this.inner.as_mut().poll_read(cx, &mut temp_buf) {
-                Poll::Pending => return Poll::Pending,
-                Poll::Ready(Ok(())) => {
-                    let n = temp_buf.filled().len();
-                    if n == 0 {
-                        break;
+        if *this.poisoned {
+            return Poll::Ready(Err(io::Error::new(io::ErrorKind::InvalidData, "decompress reader previously failed")));
+        }
+
+        if *this.compressed_len == 0 {
+            // Read the 8-byte block header, resuming across polls via `header_read`.
+            while *this.header_read < HEADER_LEN {
+                let mut temp = [0u8; HEADER_LEN];
+                let mut temp_buf = ReadBuf::new(&mut temp[0..HEADER_LEN - *this.header_read]);
+                match this.inner.as_mut().poll_read(cx, &mut temp_buf) {
+                    Poll::Pending => return Poll::Pending,
+                    Poll::Ready(Ok(())) => {
+                        let n = temp_buf.filled().len();
+                        if n == 0 {
+                            if *this.header_read == 0 {
+                                // Clean EOF on a block boundary.
+                                *this.finished = true;
+                                return Poll::Ready(Ok(()));
+                            }
+                            *this.poisoned = true;
+                            return Poll::Ready(Err(io::Error::new(
+                                io::ErrorKind::UnexpectedEof,
+                                "unexpected EOF while reading compressed block header",
+                            )));
+                        }
+                        this.header_buf[*this.header_read..*this.header_read + n].copy_from_slice(&temp_buf.filled()[..n]);
+                        *this.header_read += n;
+                    }
+                    Poll::Ready(Err(e)) => {
+                        // error!("DecompressReader poll_read: read header error: {e}");
+                        *this.poisoned = true;
+                        return Poll::Ready(Err(e));
                    }
-                    this.header_buf[*this.header_read..*this.header_read + n].copy_from_slice(&temp_buf.filled()[..n]);
-                    *this.header_read += n;
-                }
-                Poll::Ready(Err(e)) => {
-                    // error!("DecompressReader poll_read: read header error: {e}");
-                    return Poll::Ready(Err(e));
                }
            }
-            if *this.header_read < HEADER_LEN {
-                return Poll::Pending;
-            }
-        }
-        if !*this.header_done && *this.header_read == 0 {
-            return Poll::Ready(Ok(()));
-        }
-        let typ = this.header_buf[0];
-        let len = (this.header_buf[1] as usize) | ((this.header_buf[2] as usize) << 8) | ((this.header_buf[3] as usize) << 16);
-        let crc = (this.header_buf[4] as u32)
-            | ((this.header_buf[5] as u32) << 8)
-            | ((this.header_buf[6] as u32) << 16)
-            | ((this.header_buf[7] as u32) << 24);
-        *this.header_read = 0;
-        *this.header_done = true;

-        if typ == COMPRESS_TYPE_END {
+            let typ = this.header_buf[0];
+            let len =
+                (this.header_buf[1] as usize) | ((this.header_buf[2] as usize) << 8) | ((this.header_buf[3] as usize) << 16);
+            *this.header_read = 0;
+
+            // `CompressReader` never emits an end block — a stream terminates on
+            // inner EOF, which is what lets concatenated per-part streams decode as
+            // one. This branch is kept for streams that do carry the marker.
+            if typ == COMPRESS_TYPE_END {
+                *this.compressed_read = 0;
+                *this.compressed_len = 0;
+                *this.finished = true;
+                return Poll::Ready(Ok(()));
+            }
+            if typ != COMPRESS_TYPE_COMPRESSED && typ != COMPRESS_TYPE_UNCOMPRESSED {
+                // error!("DecompressReader unknown compression type: {typ}");
+                *this.poisoned = true;
+                return Poll::Ready(Err(io::Error::new(io::ErrorKind::InvalidData, "Unknown compression type")));
+            }
+            if len == 0 {
+                *this.poisoned = true;
+                return Poll::Ready(Err(io::Error::new(io::ErrorKind::InvalidData, "Invalid compressed block length")));
+            }
+
+            if this.compressed_buf.len() < len {
+                this.compressed_buf.resize(len, 0);
+            }
+            *this.compressed_len = len;
            *this.compressed_read = 0;
-            *this.compressed_len = 0;
-            *this.finished = true;
-            return Poll::Ready(Ok(()));
        }

-        if this.compressed_buf.len() < len {
-            this.compressed_buf.resize(len, 0);
-        }
-        *this.compressed_len = len;
-        *this.compressed_read = 0;
-
+        // Fill the in-flight block payload, resuming across polls via `compressed_read`.
        while *this.compressed_read < *this.compressed_len {
            let mut temp_buf = ReadBuf::new(&mut this.compressed_buf[*this.compressed_read..*this.compressed_len]);
            match this.inner.as_mut().poll_read(cx, &mut temp_buf) {
@@ -291,7 +322,13 @@ where
                Poll::Ready(Ok(())) => {
                    let n = temp_buf.filled().len();
                    if n == 0 {
-                        break;
+                        *this.compressed_read = 0;
+                        *this.compressed_len = 0;
+                        *this.poisoned = true;
+                        return Poll::Ready(Err(io::Error::new(
+                            io::ErrorKind::UnexpectedEof,
+                            "unexpected EOF while reading compressed block payload",
+                        )));
                    }
                    *this.compressed_read += n;
                }
@@ -299,10 +336,17 @@ where
                    // error!("DecompressReader poll_read: read compressed block error: {e}");
                    *this.compressed_read = 0;
                    *this.compressed_len = 0;
+                    *this.poisoned = true;
                    return Poll::Ready(Err(e));
                }
            }
        }
+
+        let typ = this.header_buf[0];
+        let crc = (this.header_buf[4] as u32)
+            | ((this.header_buf[5] as u32) << 8)
+            | ((this.header_buf[6] as u32) << 16)
+            | ((this.header_buf[7] as u32) << 24);
        let compressed_buf = &this.compressed_buf[..*this.compressed_len];
        // `compressed_buf`'s length comes from the untrusted 24-bit header length field, so it
        // can be shorter than 16 bytes. `uvarint` is safe on any slice length (reads at most 10
@@ -316,6 +360,7 @@ where
        if uvarint <= 0 || uvarint as usize > compressed_buf.len() {
            *this.compressed_read = 0;
            *this.compressed_len = 0;
+            *this.poisoned = true;
            return Poll::Ready(Err(io::Error::new(io::ErrorKind::InvalidData, "Invalid compressed block length prefix")));
        }
        let compressed_data = &compressed_buf[uvarint as usize..];
@@ -326,21 +371,29 @@ where
                    // error!("DecompressReader decompress_block error: {e}");
                    *this.compressed_read = 0;
                    *this.compressed_len = 0;
+                    *this.poisoned = true;
                    return Poll::Ready(Err(e));
                }
            }
-        } else if typ == COMPRESS_TYPE_UNCOMPRESSED {
-            compressed_data.to_vec()
        } else {
-            // error!("DecompressReader unknown compression type: {typ}");
+            // The header phase already rejected every type other than
+            // COMPRESS_TYPE_COMPRESSED / COMPRESS_TYPE_UNCOMPRESSED.
+            compressed_data.to_vec()
+        };
+        if decompressed.is_empty() {
+            // The writer never emits zero-length plaintext blocks; an empty
+            // decode surfacing as Ready(Ok) with no bytes would read as EOF and
+            // silently truncate the stream.
+            *this.poisoned = true;
            *this.compressed_read = 0;
            *this.compressed_len = 0;
-            return Poll::Ready(Err(io::Error::new(io::ErrorKind::InvalidData, "Unknown compression type")));
-        };
+            return Poll::Ready(Err(io::Error::new(io::ErrorKind::InvalidData, "Empty compressed block")));
+        }
        if decompressed.len() != uncompress_len as usize {
            // error!("DecompressReader decompressed length mismatch: {} != {}", decompressed.len(), uncompress_len);
            *this.compressed_read = 0;
            *this.compressed_len = 0;
+            *this.poisoned = true;
            return Poll::Ready(Err(io::Error::new(io::ErrorKind::InvalidData, "Decompressed length mismatch")));
        }
        let actual_crc = {
@@ -352,13 +405,13 @@ where
            // error!("DecompressReader CRC32 mismatch: actual {actual_crc} != expected {crc}");
            *this.compressed_read = 0;
            *this.compressed_len = 0;
+            *this.poisoned = true;
            return Poll::Ready(Err(io::Error::new(io::ErrorKind::InvalidData, "CRC32 mismatch")));
        }
        *this.buffer = decompressed;
        *this.buffer_pos = 0;
        *this.compressed_read = 0;
        *this.compressed_len = 0;
-        *this.header_done = false;
        let to_copy = min(buf.remaining(), this.buffer.len());
        buf.put_slice(&this.buffer[..to_copy]);
        *this.buffer_pos += to_copy;
@@ -493,6 +546,184 @@ mod tests {
        assert_eq!(&decompressed, &data);
    }

+    /// Wraps a reader so every other poll returns `Poll::Pending` and every
+    /// `Ready` poll serves at most `chunk` bytes. This is the shape a duplex
+    /// pipe produces when the erasure writer is slower than the decoder, which
+    /// is exactly what desynchronized the block framing before the resumable
+    /// payload state was added (rustfs/rustfs#5957 multipart GET truncation).
+    struct PendingChunkReader<R> {
+        inner: R,
+        chunk: usize,
+        pending_next: bool,
+    }
+
+    impl<R> PendingChunkReader<R> {
+        fn new(inner: R, chunk: usize) -> Self {
+            Self {
+                inner,
+                chunk,
+                pending_next: true,
+            }
+        }
+    }
+
+    impl<R: AsyncRead + Unpin> AsyncRead for PendingChunkReader<R> {
+        fn poll_read(
+            mut self: std::pin::Pin<&mut Self>,
+            cx: &mut std::task::Context<'_>,
+            buf: &mut tokio::io::ReadBuf<'_>,
+        ) -> std::task::Poll<std::io::Result<()>> {
+            if self.pending_next {
+                self.pending_next = false;
+                cx.waker().wake_by_ref();
+                return std::task::Poll::Pending;
+            }
+            self.pending_next = true;
+            let cap = self.chunk.min(buf.remaining());
+            let mut scratch = vec![0u8; cap];
+            let mut inner_buf = tokio::io::ReadBuf::new(&mut scratch);
+            match std::pin::Pin::new(&mut self.inner).poll_read(cx, &mut inner_buf) {
+                std::task::Poll::Ready(Ok(())) => {
+                    buf.put_slice(inner_buf.filled());
+                    std::task::Poll::Ready(Ok(()))
+                }
+                other => other,
+            }
+        }
+    }
+
+    fn patterned_payload(size: usize, seed: u8) -> Vec<u8> {
+        (0..size)
+            .map(|i| ((i as u64).wrapping_mul(2_654_435_761).wrapping_add(seed as u64) >> 3) as u8)
+            .collect()
+    }
+
+    /// Root-cause regression for the multipart compressed GET truncation: a
+    /// `Poll::Pending` in the middle of a block payload must not drop the bytes
+    /// already consumed. Before the resumable payload state, the decoder reset
+    /// `compressed_read` on every re-poll and surfaced
+    /// `LZ4 error: ERROR_frameType_unknown` mid-stream.
+    #[tokio::test]
+    async fn test_decompress_reader_survives_pending_mid_payload() {
+        let data = patterned_payload(100 * 1024, 7);
+        let mut compress_reader =
+            CompressReader::with_block_size(Cursor::new(data.clone()), 8192, CompressionAlgorithm::default());
+        let mut compressed = Vec::new();
+        compress_reader.read_to_end(&mut compressed).await.unwrap();
+
+        for chunk in [1usize, 3, 7, 8, 17, 1000, 8192] {
+            let inner = PendingChunkReader::new(Cursor::new(compressed.clone()), chunk);
+            let mut decompress_reader = DecompressReader::new(inner, CompressionAlgorithm::default());
+            let mut decompressed = Vec::new();
+            decompress_reader.read_to_end(&mut decompressed).await.unwrap();
+            assert_eq!(decompressed, data, "pending-chunked decode must be byte-exact for chunk={chunk}");
+        }
+    }
+
+    /// Two independently compressed streams concatenated back to back — the
+    /// on-disk shape of a compressed multipart object — must decode across the
+    /// stream boundary even when every poll can suspend mid-block.
+    #[tokio::test]
+    async fn test_decompress_reader_survives_pending_across_concatenated_streams() {
+        let part1 = patterned_payload(64 * 1024, 7);
+        let part2 = patterned_payload(24 * 1024, 61);
+
+        let mut stored = Vec::new();
+        for part in [&part1, &part2] {
+            let mut compress_reader =
+                CompressReader::with_block_size(Cursor::new(part.clone()), 8192, CompressionAlgorithm::default());
+            let mut compressed = Vec::new();
+            compress_reader.read_to_end(&mut compressed).await.unwrap();
+            stored.extend_from_slice(&compressed);
+        }
+
+        let mut expected = part1;
+        expected.extend_from_slice(&part2);
+
+        for chunk in [1usize, 5, 8, 13, 4096] {
+            let inner = PendingChunkReader::new(Cursor::new(stored.clone()), chunk);
+            let mut decompress_reader = DecompressReader::new(inner, CompressionAlgorithm::default());
+            let mut decompressed = Vec::new();
+            decompress_reader.read_to_end(&mut decompressed).await.unwrap();
+            assert_eq!(
+                decompressed, expected,
+                "concatenated part streams must decode byte-exact for chunk={chunk}"
+            );
+        }
+    }
+
+    /// After the first stream error, every further poll must keep failing.
+    /// Without the sticky poison a retrying caller would restart at the header
+    /// phase and read the truncated tail as a clean EOF — converting a hard
+    /// error into a silently short body.
+    #[tokio::test]
+    async fn test_decompress_reader_error_is_sticky() {
+        let data = patterned_payload(32 * 1024, 7);
+        let mut compress_reader = CompressReader::with_block_size(Cursor::new(data), 8192, CompressionAlgorithm::default());
+        let mut compressed = Vec::new();
+        compress_reader.read_to_end(&mut compressed).await.unwrap();
+        compressed.truncate(compressed.len() - 3);
+
+        let mut decompress_reader = DecompressReader::new(Cursor::new(compressed), CompressionAlgorithm::default());
+        let mut out = Vec::new();
+        let first = decompress_reader
+            .read_to_end(&mut out)
+            .await
+            .expect_err("truncated payload must error");
+        assert_eq!(first.kind(), std::io::ErrorKind::UnexpectedEof);
+
+        let mut retry = Vec::new();
+        let second = decompress_reader
+            .read_to_end(&mut retry)
+            .await
+            .expect_err("a poll after the first error must not turn into a clean EOF");
+        assert_eq!(second.kind(), std::io::ErrorKind::InvalidData);
+        assert!(retry.is_empty(), "no bytes may be produced after the stream failed");
+    }
+
+    /// A stream cut off in the middle of a block payload must fail with a clean
+    /// UnexpectedEof instead of decoding a short buffer.
+    #[tokio::test]
+    async fn test_decompress_reader_truncated_payload_is_unexpected_eof() {
+        let data = patterned_payload(32 * 1024, 7);
+        let mut compress_reader = CompressReader::with_block_size(Cursor::new(data), 8192, CompressionAlgorithm::default());
+        let mut compressed = Vec::new();
+        compress_reader.read_to_end(&mut compressed).await.unwrap();
+
+        compressed.truncate(compressed.len() - 3);
+        let mut decompress_reader = DecompressReader::new(Cursor::new(compressed), CompressionAlgorithm::default());
+        let mut out = Vec::new();
+        let err = decompress_reader
+            .read_to_end(&mut out)
+            .await
+            .expect_err("truncated payload must error");
+        assert_eq!(err.kind(), std::io::ErrorKind::UnexpectedEof);
+    }
+
+    /// A stream cut off in the middle of a block header must fail with a clean
+    /// UnexpectedEof instead of parsing a garbage header.
+    #[tokio::test]
+    async fn test_decompress_reader_truncated_header_is_unexpected_eof() {
+        let data = patterned_payload(12 * 1024, 7);
+        let mut compress_reader = CompressReader::with_block_size(Cursor::new(data), 8192, CompressionAlgorithm::default());
+        let mut compressed = Vec::new();
+        compress_reader.read_to_end(&mut compressed).await.unwrap();
+
+        // Keep the first full block plus 3 bytes of the next header.
+        let ln = (compressed[1] as usize) | ((compressed[2] as usize) << 8) | ((compressed[3] as usize) << 16);
+        let first_block_end = 8 + ln;
+        assert!(compressed.len() > first_block_end, "fixture must contain more than one block");
+        compressed.truncate(first_block_end + 3);
+
+        let mut decompress_reader = DecompressReader::new(Cursor::new(compressed), CompressionAlgorithm::default());
+        let mut out = Vec::new();
+        let err = decompress_reader
+            .read_to_end(&mut out)
+            .await
+            .expect_err("truncated header must error");
+        assert_eq!(err.kind(), std::io::ErrorKind::UnexpectedEof);
+    }
+
    // Regression: a corrupted block whose 24-bit length field is < 16 must not panic.
    // Header layout (HEADER_LEN = 8): [type, len_lo, len_mid, len_hi, crc0..crc3], then `len`
    // bytes of block body. Pre-fix, poll_read sliced `compressed_buf[0..16]` unconditionally,
@@ -518,6 +749,85 @@ mod tests {
        assert_eq!(res.unwrap_err().kind(), std::io::ErrorKind::InvalidData);
    }

+    // Header-level fail-closed matrix, built by hand so the decoder is exercised against bytes no
+    // encoder in this crate can produce. Header layout (HEADER_LEN = 8):
+    // [type, len_lo, len_mid, len_hi, crc0..crc3], then `len` body bytes = uvarint(plain_len) + data.
+    #[tokio::test]
+    async fn test_decompress_reader_header_validation_matrix() {
+        // Build a block whose body is `uvarint(plain.len()) + plain` (i.e. the
+        // COMPRESS_TYPE_UNCOMPRESSED shape), with the header CRC taken over the plaintext exactly
+        // like the production writer does.
+        fn build_raw_block(typ: u8, plain: &[u8], len_override: Option<usize>) -> Vec<u8> {
+            let crc = {
+                let mut hasher = crc_fast::Digest::new(crc_fast::CrcAlgorithm::Crc32IsoHdlc);
+                hasher.update(plain);
+                hasher.finalize() as u32
+            };
+            let mut uvarint_buf = [0u8; 10];
+            let int_len = put_uvarint(&mut uvarint_buf[..], plain.len() as u64);
+            let body_len = int_len + plain.len();
+            let len = len_override.unwrap_or(body_len);
+
+            let mut out = Vec::with_capacity(HEADER_LEN + body_len);
+            out.push(typ);
+            out.push((len & 0xFF) as u8);
+            out.push(((len >> 8) & 0xFF) as u8);
+            out.push(((len >> 16) & 0xFF) as u8);
+            out.extend_from_slice(&crc.to_le_bytes());
+            out.extend_from_slice(&uvarint_buf[..int_len]);
+            out.extend_from_slice(plain);
+            out
+        }
+
+        let plain = b"uncompressed passthrough payload";
+
+        // (a) A well-formed uncompressed block decodes to the plaintext verbatim.
+        let mut out = Vec::new();
+        DecompressReader::new(
+            Cursor::new(build_raw_block(COMPRESS_TYPE_UNCOMPRESSED, plain, None)),
+            CompressionAlgorithm::default(),
+        )
+        .read_to_end(&mut out)
+        .await
+        .expect("a well-formed uncompressed block must decode");
+        assert_eq!(out.as_slice(), plain.as_slice());
+
+        // (b) An unknown block type must be rejected instead of being treated as passthrough.
+        let mut out = Vec::new();
+        let err = DecompressReader::new(Cursor::new(build_raw_block(0x7E, plain, None)), CompressionAlgorithm::default())
+            .read_to_end(&mut out)
+            .await
+            .expect_err("unknown compression type must error");
+        assert_eq!(err.kind(), std::io::ErrorKind::InvalidData);
+        assert!(err.to_string().contains("Unknown compression type"), "got: {err}");
+
+        // (c) A zero-length block would stall the decoder, so it must be rejected up front.
+        let mut out = Vec::new();
+        let err = DecompressReader::new(
+            Cursor::new(build_raw_block(COMPRESS_TYPE_UNCOMPRESSED, plain, Some(0))),
+            CompressionAlgorithm::default(),
+        )
+        .read_to_end(&mut out)
+        .await
+        .expect_err("zero-length block must error");
+        assert_eq!(err.kind(), std::io::ErrorKind::InvalidData);
+        assert!(err.to_string().contains("Invalid compressed block length"), "got: {err}");
+
+        // (d) A block that decodes to zero plaintext bytes must be rejected: the
+        // writer never emits empty blocks, and an empty decode surfacing as
+        // Ready(Ok) with no bytes would read as EOF and silently truncate.
+        let mut out = Vec::new();
+        let err = DecompressReader::new(
+            Cursor::new(build_raw_block(COMPRESS_TYPE_UNCOMPRESSED, b"", None)),
+            CompressionAlgorithm::default(),
+        )
+        .read_to_end(&mut out)
+        .await
+        .expect_err("empty block must error");
+        assert_eq!(err.kind(), std::io::ErrorKind::InvalidData);
+        assert!(err.to_string().contains("Empty compressed block"), "got: {err}");
+    }
+
    // Directly exercises the length-prefix guard: an unterminated varint (all continuation bytes)
    // makes `uvarint` return 0, which must be rejected as an invalid length prefix.
    #[tokio::test]
@@ -24,12 +24,17 @@ pin_project! {
        #[pin]
        pub inner: R,
        remaining: i64,
+        scratch: Vec<u8>,
    }
 }

 impl<R> HardLimitReader<R> {
    pub fn new(inner: R, limit: i64) -> Self {
-        HardLimitReader { inner, remaining: limit }
+        HardLimitReader {
+            inner,
+            remaining: limit,
+            scratch: Vec::new(),
+        }
    }
 }

@@ -37,19 +42,21 @@ impl<R> AsyncRead for HardLimitReader<R>
 where
    R: AsyncRead,
 {
-    fn poll_read(mut self: Pin<&mut Self>, cx: &mut Context<'_>, buf: &mut ReadBuf<'_>) -> Poll<Result<()>> {
-        if self.remaining < 0 {
+    fn poll_read(self: Pin<&mut Self>, cx: &mut Context<'_>, buf: &mut ReadBuf<'_>) -> Poll<Result<()>> {
+        let mut this = self.project();
+        if *this.remaining < 0 {
            return Poll::Ready(Err(Error::other("input provided more bytes than specified")));
        }
-        let original_filled = buf.filled().len();
-        if self.remaining == 0 {
+        if buf.remaining() == 0 {
+            return Poll::Ready(Ok(()));
+        }
+        if *this.remaining == 0 {
            let mut discard = [0u8; 8192];
            let mut discard_buf = ReadBuf::new(&mut discard);
-            return match self.as_mut().project().inner.poll_read(cx, &mut discard_buf) {
+            return match this.inner.as_mut().poll_read(cx, &mut discard_buf) {
                Poll::Pending => Poll::Pending,
                Poll::Ready(Ok(())) => {
                    if discard_buf.filled().is_empty() {
-                        debug_assert_eq!(buf.filled().len(), original_filled);
                        Poll::Ready(Ok(()))
                    } else {
                        Poll::Ready(Err(Error::other("input provided more bytes than specified")))
@@ -58,30 +65,46 @@ where
                Poll::Ready(Err(err)) => Poll::Ready(Err(err)),
            };
        }
-        // Save the initial length
-        let before = original_filled;

-        // Poll the inner reader
-        let this = self.as_mut().project();
-        let poll = this.inner.poll_read(cx, buf);
-
-        if let Poll::Ready(Ok(())) = &poll {
-            let after = buf.filled().len();
-            let read = (after - before) as i64;
-            if read == 0 && *this.remaining > 0 {
-                return Poll::Ready(Err(Error::new(
-                    std::io::ErrorKind::UnexpectedEof,
-                    IncompleteBody {
-                        remaining: *this.remaining,
-                    },
-                )));
+        let remaining = match usize::try_from(*this.remaining) {
+            Ok(remaining) => remaining,
+            Err(_) => usize::MAX,
+        };
+        let allowed = remaining.min(buf.remaining());
+        let read = if allowed == buf.remaining() {
+            let before = buf.filled().len();
+            match this.inner.as_mut().poll_read(cx, buf) {
+                Poll::Pending => return Poll::Pending,
+                Poll::Ready(Err(err)) => return Poll::Ready(Err(err)),
+                Poll::Ready(Ok(())) => buf.filled().len() - before,
            }
-            *this.remaining -= read;
-            if *this.remaining < 0 {
-                return Poll::Ready(Err(Error::other("input provided more bytes than specified")));
+        } else {
+            this.scratch.resize(allowed, 0);
+            let mut scratch_buf = ReadBuf::new(&mut this.scratch[..allowed]);
+            match this.inner.as_mut().poll_read(cx, &mut scratch_buf) {
+                Poll::Pending => return Poll::Pending,
+                Poll::Ready(Err(err)) => return Poll::Ready(Err(err)),
+                Poll::Ready(Ok(())) => {
+                    let read = scratch_buf.filled().len();
+                    buf.put_slice(scratch_buf.filled());
+                    read
+                }
            }
+        };
+        if read == 0 {
+            return Poll::Ready(Err(Error::new(
+                std::io::ErrorKind::UnexpectedEof,
+                IncompleteBody {
+                    remaining: *this.remaining,
+                },
+            )));
        }
-        poll
+        let read = match i64::try_from(read) {
+            Ok(read) => read,
+            Err(_) => return Poll::Ready(Err(Error::other("read count exceeds i64::MAX"))),
+        };
+        *this.remaining -= read;
+        Poll::Ready(Ok(()))
    }
 }

@@ -140,7 +163,12 @@ mod tests {
        assert!(err.is_some());

        let err = err.unwrap();
-        assert_eq!(err.kind(), std::io::ErrorKind::Other);
+        assert_eq!(err.kind(), std::io::ErrorKind::UnexpectedEof);
+        assert!(
+            err.get_ref()
+                .and_then(|source| source.downcast_ref::<std::io::Error>())
+                .is_some_and(|source| source.to_string().contains("more bytes than specified"))
+        );
    }

    #[tokio::test]
@@ -155,6 +183,17 @@ mod tests {
        assert_eq!(&buf, data);
    }

+    #[tokio::test]
+    async fn test_hardlimit_reader_zero_capacity_read_does_not_consume_input() {
+        let mut reader = HardLimitReader::new(BufReader::new(&b"abc"[..]), 3);
+        let mut empty = [];
+
+        assert_eq!(reader.read(&mut empty).await.expect("zero-capacity read should succeed"), 0);
+        let mut out = Vec::new();
+        reader.read_to_end(&mut out).await.expect("input should remain readable");
+        assert_eq!(out, b"abc");
+    }
+
    #[tokio::test]
    async fn test_hardlimit_reader_short_input_returns_unexpected_eof() {
        let data = b"abc";
@@ -195,4 +234,18 @@ mod tests {
        assert_eq!(err.kind(), std::io::ErrorKind::Other);
        assert!(err.to_string().contains("more bytes than specified"));
    }
+
+    #[tokio::test]
+    async fn test_hardlimit_reader_caps_each_read_before_reporting_extra_bytes() {
+        let mut reader = HardLimitReader::new(BufReader::new(&b"abcdef"[..]), 3);
+        let mut out = Vec::new();
+
+        let err = reader
+            .read_to_end(&mut out)
+            .await
+            .expect_err("bytes beyond the declared limit must be rejected");
+
+        assert_eq!(out, b"abc");
+        assert!(err.to_string().contains("more bytes than specified"));
+    }
 }
@@ -138,6 +138,12 @@ impl std::fmt::Display for InternodeHttpErrorKind {
    }
 }

+#[derive(thiserror::Error, Debug, Clone, Copy, Eq, PartialEq)]
+#[error("internode body stalled for {timeout:?}")]
+pub struct BodyStalled {
+    pub timeout: Duration,
+}
+
 #[derive(Debug, Clone, Eq, PartialEq)]
 pub struct InternodeHttpRequestContext {
    method: String,
@@ -271,6 +277,10 @@ pub fn internode_http_timeout_error(method: &Method, url: &str) -> io::Error {
    internode_kind_error(method, url, internode_rpc_operation(url), InternodeHttpErrorKind::ConnectTimeout)
 }

+fn body_stalled_error(stall_timeout: Duration) -> io::Error {
+    Error::new(io::ErrorKind::TimedOut, BodyStalled { timeout: stall_timeout })
+}
+
 /// Clone an internode HTTP I/O error while retaining its structured classification.
 ///
 /// The underlying transport source is intentionally omitted because it is not
@@ -1085,10 +1095,7 @@ impl AsyncRead for HttpReader {
                    );
                    record_internode_stall_timeout(*this.track_internode_metrics, *this.internode_operation);
                    record_internode_error(*this.track_internode_metrics, *this.internode_operation);
-                    Poll::Ready(Err(Error::new(
-                        io::ErrorKind::TimedOut,
-                        "HttpReader stall timeout: no data received before deadline",
-                    )))
+                    Poll::Ready(Err(body_stalled_error(stall_timeout)))
                } else {
                    Poll::Pending
                }
@@ -1217,10 +1224,7 @@ impl ChunkReader for HttpChunkReader {
                        );
                        record_internode_stall_timeout(*this.track_internode_metrics, *this.internode_operation);
                        record_internode_error(*this.track_internode_metrics, *this.internode_operation);
-                        return Poll::Ready(Err(Error::new(
-                            io::ErrorKind::TimedOut,
-                            "HttpReader stall timeout: no data received before deadline",
-                        )));
+                        return Poll::Ready(Err(body_stalled_error(stall_timeout)));
                    }
                    return Poll::Pending;
                }
@@ -2379,6 +2383,46 @@ mod tests {
            Err(err) => err,
        };
        assert_eq!(err.kind(), io::ErrorKind::TimedOut);
+        let stalled = err
+            .get_ref()
+            .and_then(|source| source.downcast_ref::<BodyStalled>())
+            .expect("stall timeout should retain typed body-stalled source");
+        assert_eq!(stalled.timeout, Duration::from_millis(20));
+
+        handle.abort();
+    }
+
+    #[tokio::test]
+    async fn http_chunk_reader_stall_timeout_retains_typed_source() {
+        let state = TestState::default();
+        let Some((base_url, handle)) = start_test_server(state).await else {
+            return;
+        };
+        let url = base_url.replace("/stream", "/stall");
+        let mut reader =
+            HttpChunkReader::new_with_stall_timeout(url, Method::GET, HeaderMap::new(), None, Some(Duration::from_millis(20)))
+                .await
+                .expect("chunk reader should open");
+
+        let first = std::future::poll_fn(|cx| Pin::new(&mut reader).poll_read_chunk(cx, 64))
+            .await
+            .expect("initial body chunk should arrive")
+            .expect("initial body chunk should not be EOF");
+        assert_eq!(first, b"hello"[..]);
+
+        let err = tokio::time::timeout(
+            Duration::from_secs(1),
+            std::future::poll_fn(|cx| Pin::new(&mut reader).poll_read_chunk(cx, 64)),
+        )
+        .await
+        .expect("stall timeout should wake chunk reader")
+        .expect_err("chunk reader should return a timeout error");
+        assert_eq!(err.kind(), io::ErrorKind::TimedOut);
+        let stalled = err
+            .get_ref()
+            .and_then(|source| source.downcast_ref::<BodyStalled>())
+            .expect("chunk stall timeout should retain typed body-stalled source");
+        assert_eq!(stalled.timeout, Duration::from_millis(20));

        handle.abort();
    }
@@ -102,7 +102,7 @@ bytes.workspace = true
 hex-simd.workspace = true

 [dev-dependencies]
-tracing-subscriber = { workspace = true, features = ["env-filter", "time"] }
+tracing-subscriber = { workspace = true, features = ["json", "env-filter", "time"] }
 serial_test = { workspace = true }
 temp-env = { workspace = true }
 tempfile = { workspace = true }
@@ -65,6 +65,7 @@ const LOG_SUBSYSTEM_FOLDER: &str = "folder";
 const LOG_SUBSYSTEM_LIFECYCLE: &str = "lifecycle";
 const LOG_SUBSYSTEM_HEAL: &str = "heal";
 const EVENT_SCANNER_FOLDER_STATE: &str = "scanner_folder_state";
+const EVENT_SCANNER_METADATA_CORRUPT: &str = "scanner_metadata_corrupt";
 const EVENT_SCANNER_LIFECYCLE_ACTION: &str = "scanner_lifecycle_action";
 const EVENT_SCANNER_HEAL_ADMISSION: &str = "scanner_heal_admission";
 const EVENT_SCANNER_ALERT_STATE: &str = "scanner_alert_state";
@@ -2154,17 +2155,34 @@ impl FolderScanner {
                            self.record_failed(&item.path);

                            if should_log_failed_object(into.failed_objects) {
-                                warn!(
-                                    target: "rustfs::scanner::folder",
-                                    event = EVENT_SCANNER_FOLDER_STATE,
-                                    component = LOG_COMPONENT_SCANNER,
-                                    subsystem = LOG_SUBSYSTEM_FOLDER,
-                                    path = %item.path,
-                                    failed_objects = into.failed_objects,
-                                    state = "get_size_failed",
-                                    error = %e,
-                                    "Scanner folder failed to get object size"
-                                );
+                                if let GetSizeFailureAction::HealMetadata { object } = &failure_action {
+                                    error!(
+                                        target: "rustfs::scanner::folder",
+                                        event = EVENT_SCANNER_METADATA_CORRUPT,
+                                        component = LOG_COMPONENT_SCANNER,
+                                        subsystem = LOG_SUBSYSTEM_FOLDER,
+                                        drive = %self.local_disk.path().display(),
+                                        bucket = %item.bucket,
+                                        object = %object,
+                                        metadata_path = %item.path,
+                                        failed_objects = into.failed_objects,
+                                        state = "metadata_corrupt",
+                                        error = %e,
+                                        "Scanner detected corrupt object metadata"
+                                    );
+                                } else {
+                                    warn!(
+                                        target: "rustfs::scanner::folder",
+                                        event = EVENT_SCANNER_FOLDER_STATE,
+                                        component = LOG_COMPONENT_SCANNER,
+                                        subsystem = LOG_SUBSYSTEM_FOLDER,
+                                        path = %item.path,
+                                        failed_objects = into.failed_objects,
+                                        state = "get_size_failed",
+                                        error = %e,
+                                        "Scanner folder failed to get object size"
+                                    );
+                                }
                            }
                        }

@@ -3054,12 +3072,59 @@ mod tests {
    use crate::{DiskOption, Endpoint, STORAGE_FORMAT_FILE, TierStats, new_disk, storageclass};
    use rustfs_filemeta::{FileInfo, FileMeta};
    use serial_test::serial;
+    use std::io::Write;
    #[cfg(unix)]
    use std::os::unix::fs::{PermissionsExt, symlink};
+    use std::sync::Mutex;
    use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
    use temp_env::{with_var, with_var_unset};
+    use tracing_subscriber::fmt::MakeWriter;
    use uuid::Uuid;

+    #[derive(Clone, Default)]
+    struct CapturedLogs {
+        buffer: Arc<Mutex<Vec<u8>>>,
+    }
+
+    struct CapturedLogWriter {
+        buffer: Arc<Mutex<Vec<u8>>>,
+    }
+
+    impl CapturedLogs {
+        fn contents(&self) -> String {
+            let buffer = self
+                .buffer
+                .lock()
+                .expect("captured logs mutex should not be poisoned")
+                .clone();
+            String::from_utf8(buffer).expect("captured logs should be valid UTF-8")
+        }
+    }
+
+    impl Write for CapturedLogWriter {
+        fn write(&mut self, buf: &[u8]) -> std::io::Result<usize> {
+            self.buffer
+                .lock()
+                .expect("captured logs mutex should not be poisoned")
+                .extend_from_slice(buf);
+            Ok(buf.len())
+        }
+
+        fn flush(&mut self) -> std::io::Result<()> {
+            Ok(())
+        }
+    }
+
+    impl<'a> MakeWriter<'a> for CapturedLogs {
+        type Writer = CapturedLogWriter;
+
+        fn make_writer(&'a self) -> Self::Writer {
+            CapturedLogWriter {
+                buffer: Arc::clone(&self.buffer),
+            }
+        }
+    }
+
    #[test]
    fn scanner_size_summary_application_saturates_usage_counters() {
        let target = "arn:minio:replication::target".to_string();
@@ -4542,9 +4607,19 @@ mod tests {
        assert!(budget.entries_visited() >= 1);
    }

-    #[tokio::test]
+    #[tokio::test(flavor = "current_thread")]
    #[serial]
    async fn test_scan_folder_corrupt_xl_meta_stops_erasure_data_dir_descent() {
+        let logs = CapturedLogs::default();
+        let subscriber = tracing_subscriber::fmt()
+            .json()
+            .with_max_level(tracing::Level::ERROR)
+            .with_writer(logs.clone())
+            .with_ansi(false)
+            .without_time()
+            .finish();
+        let _subscriber_guard = tracing::subscriber::set_default(subscriber);
+
        let (mut scanner, temp_dir) = build_test_scanner().await;
        let _guard = TestGuard::new(60, 100, &mut scanner, temp_dir.clone());

@@ -4596,6 +4671,30 @@ mod tests {
        assert!(!budget.budget_elapsed());
        assert_eq!(budget.reason(), None);

+        let captured = logs.contents();
+        assert!(
+            !captured.contains("failed to check XL2 v1 format"),
+            "the context-free filemeta parser error must not be emitted"
+        );
+        let events = captured
+            .lines()
+            .map(|line| serde_json::from_str::<serde_json::Value>(line).expect("captured scanner log should be valid JSON"))
+            .filter(|line| line["fields"]["event"] == EVENT_SCANNER_METADATA_CORRUPT)
+            .collect::<Vec<_>>();
+        assert_eq!(
+            events.len(),
+            1,
+            "one corrupt metadata observation must emit one scanner-owned diagnostic event"
+        );
+        let fields = &events[0]["fields"];
+        assert_eq!(fields["component"], LOG_COMPONENT_SCANNER);
+        assert_eq!(fields["subsystem"], LOG_SUBSYSTEM_FOLDER);
+        assert_eq!(fields["drive"], temp_dir.to_string_lossy().as_ref());
+        assert_eq!(fields["bucket"], "bucket");
+        assert_eq!(fields["object"], "object");
+        assert_eq!(fields["metadata_path"], metadata_path.to_string_lossy().as_ref());
+        assert_eq!(fields["state"], "metadata_corrupt");
+
        let retry_budget = ScannerCycleBudget::new_with_progress_tracking(
            &parent,
            crate::scanner_budget::ScannerCycleBudgetConfig {
@@ -3849,17 +3849,6 @@ impl ScannerIODisk for Disk {
        let fivs = match meta.get_file_info_versions(item.bucket.as_str(), item.object_path().as_str(), false) {
            Ok(versions) => versions,
            Err(e) => {
-                error!(
-                    target: "rustfs::scanner::io",
-                    event = EVENT_SCANNER_DISK_BUCKET_STATE,
-                    component = LOG_COMPONENT_SCANNER,
-                    subsystem = LOG_SUBSYSTEM_IO,
-                    bucket = %item.bucket,
-                    object = %item.object_path(),
-                    state = "file_info_versions_failed",
-                    error = %e,
-                    "Scanner disk bucket failed to resolve file info versions"
-                );
                return Err(scanner_metadata_corrupt_error(
                    format!("failed to resolve file info versions: {e}"),
                    &item.bucket,
@@ -4023,7 +4012,7 @@ mod tests {
    use crate::scanner_budget::ScannerCycleBudgetConfig;
    use crate::scanner_folder::ScannerItem;
    use crate::storage_api::owner::{EcstoreRebalStatus, EcstoreRebalanceInfo, EcstoreRebalanceMeta, EcstoreRebalanceStats};
-    use crate::storage_api::scan::{BucketOperations as _, MakeBucketOptions, ObjectIO as _};
+    use crate::storage_api::scan::{BucketOperations as _, DeleteBucketOptions, MakeBucketOptions, ObjectIO as _};
    use crate::{
        DiskOption, ECStore, Endpoint, EndpointServerPools, Endpoints, InstanceContext, PoolEndpoints, ScannerObjectOptions,
        ScannerPutObjReader, init_bucket_metadata_sys_for_scanner_tests, init_ecstore_config_for_scanner_tests,
@@ -4248,16 +4237,21 @@ mod tests {
    async fn multi_pool_scanner_cycle_zero_fills_bucket_absent_from_first_pool() {
        let (_temp_dir, store) = setup_two_pool_scanner_store().await;
        let bucket = format!("scanner-second-pool-{}", Uuid::new_v4().simple());
-        store.pools[1].disk_set[0]
+        store
            .make_bucket(&bucket, &MakeBucketOptions::default())
            .await
-            .expect("bucket should be created only in the second pool");
+            .expect("bucket and its authoritative metadata should be created");
        let body = b"second-only";
        let mut reader = ScannerPutObjReader::from_vec(body.to_vec());
-        store.pools[1].disk_set[0]
+        store.pools[1]
            .put_object(&bucket, "pool-b", &mut reader, &ScannerObjectOptions::default())
            .await
            .expect("object should be written only to the second pool");
+        store.pools[0]
+            .delete_bucket(&bucket, &DeleteBucketOptions::default())
+            .await
+            .expect("bucket should be removed from the first pool only");
+        init_bucket_metadata_sys_for_scanner_tests(store.clone()).await;

        let ctx = CancellationToken::new();
        let budget = ScannerCycleBudget::new(&ctx, ScannerCycleBudgetConfig::default());
@@ -278,7 +278,7 @@ pub(crate) mod scan {
        SCANNER_ACTIVITY_PREVIOUS_PROTOCOL_VERSION,
    };
    #[cfg(test)]
-    pub(crate) use super::storage_contracts::{MakeBucketOptions, ObjectIO};
+    pub(crate) use super::storage_contracts::{DeleteBucketOptions, MakeBucketOptions, ObjectIO};
 }

 pub(crate) mod scanner_io {
@@ -50,6 +50,14 @@ pub const SUFFIX_SOURCE_DELETEMARKER: &str = "source-deletemarker";
 pub const SUFFIX_SOURCE_PROXY_REQUEST: &str = "source-proxy-request";
 pub const SUFFIX_SOURCE_REPLICATION_REQUEST: &str = "source-replication-request";
 pub const SUFFIX_SOURCE_REPLICATION_CHECK: &str = "source-replication-check";
+// LWW timestamps for replicated tag/retention/legal-hold modifications. MinIO
+// declares these with mixed case (internal/http/headers.go:
+// X-Minio-Source-Replication-Tagging-Timestamp / -Retention-Timestamp /
+// -LegalHold-Timestamp); HTTP header names compare case-insensitively, so the
+// lowercase suffix forms interoperate. Values are RFC3339 on the wire.
+pub const SUFFIX_SOURCE_REPLICATION_TAGGING_TIMESTAMP: &str = "source-replication-tagging-timestamp";
+pub const SUFFIX_SOURCE_REPLICATION_RETENTION_TIMESTAMP: &str = "source-replication-retention-timestamp";
+pub const SUFFIX_SOURCE_REPLICATION_LEGALHOLD_TIMESTAMP: &str = "source-replication-legalhold-timestamp";
 pub const SUFFIX_REPLICATION_SSEC_CRC: &str = "replication-ssec-crc";

 /// Returns true if the key is object-encryption metadata understood by RustFS or MinIO.
@@ -196,6 +204,27 @@ mod tests {
        assert_eq!(get_object_encryption_original_size(&metadata).expect("valid size"), Some(42));
    }

+    #[test]
+    fn replication_timestamp_headers_match_minio_wire_names() {
+        let mut headers = HeaderMap::new();
+        insert_header(&mut headers, SUFFIX_SOURCE_REPLICATION_TAGGING_TIMESTAMP, "2026-01-02T03:04:05Z");
+        insert_header(&mut headers, SUFFIX_SOURCE_REPLICATION_RETENTION_TIMESTAMP, "2026-01-02T03:04:06Z");
+        insert_header(&mut headers, SUFFIX_SOURCE_REPLICATION_LEGALHOLD_TIMESTAMP, "2026-01-02T03:04:07Z");
+
+        // The exact names MinIO's object-api-options.go reads (its Get()
+        // canonicalizes case, so a case-insensitive match is wire-equivalent).
+        for name in [
+            "X-Minio-Source-Replication-Tagging-Timestamp",
+            "X-Minio-Source-Replication-Retention-Timestamp",
+            "X-Minio-Source-Replication-LegalHold-Timestamp",
+            "x-rustfs-source-replication-tagging-timestamp",
+            "x-rustfs-source-replication-retention-timestamp",
+            "x-rustfs-source-replication-legalhold-timestamp",
+        ] {
+            assert!(headers.contains_key(name), "replication timestamp header {name} must be written");
+        }
+    }
+
    #[test]
    fn test_get_header() {
        let mut headers = HeaderMap::new();
@@ -659,9 +659,6 @@ mod test {
        // Port should be in valid range (u16 max is always <= 65535)
        assert!(port1 > 0);
        assert!(port2 > 0);
-
-        // Different calls should typically return different ports
-        assert_ne!(port1, port2);
    }

    #[test]
@@ -16,7 +16,6 @@ for later deletion.
 - `table-catalog-strong-snapshot-v1` durable strong catalog snapshot compatibility: version 1 writes continue during mixed-version rollout until operators confirm that every serving node reads version 2, and version 1 table/view identifier collisions remain available only for cleanup. Remove version 1 writes and collision cleanup after the minimum supported RustFS release reads version 2 and every retained durable strong snapshot is collision-free and has been upgraded to version 2.
 - `table-catalog-migration-fence-v1` durable strong migration fence compatibility: version 1 "PREPARING" fences did not distinguish a known-absent global strong snapshot from an unknown baseline, so retries read them but fail closed if the global snapshot is missing. Version 2 preserves the same JSON shape and records the pre-migration global snapshot ETag in the existing target_snapshot_etag field while the fence is "PREPARING". Remove version 1 reads after every supported direct-upgrade source writes version 2 fences and operators have completed or cancelled every older in-progress backing migration.
 - `table-catalog-backing-manifest-v1-wire-labels` durable strong backing manifest labels: version 1 published "STRONG_KV_WAL" and "CUT_OVER_LINEARIZABLE_READS" before the implementation was narrowed to the ETag-CAS durable snapshot backing. Internal names and operator documentation describe the implemented semantics, while version 1 responses retain those labels for existing clients. Replace the labels only in a new manifest version with an explicit client migration contract.
- `cross-pool-fence-v1` authenticated unsupported advertisement: predeployment servers recognize the versioned cross-pool fence capability probe but report support version 0, allowing a later all-peer probe to distinguish predeployment nodes without activating a second lock domain. Replace the unsupported advertisement only when composite lock acquisition, a cluster-wide activation fence, complete fleet proof, commit-time proof revalidation, and fail-closed revocation ship together.
 - `replacement-recovery-status-v1` replacement recovery status capability: new peers treat an unimplemented ReplacementRecoveryStatus RPC as an explicitly non-definitive rolling-upgrade response, so Admin v4 cannot claim distributed replacement completion from old peers. Remove the fallback after the minimum supported RustFS peer version implements ReplacementRecoveryStatus.
 - `table-catalog-dotted-namespace` Iceberg REST namespace path compatibility: existing RustFS clients use dotted namespace paths, while the standard multi-level contract uses the URL-encoded unit separator `%1F`. New servers accept both forms so a rolling upgrade does not invalidate existing catalog configuration. Remove the dotted fallback after the minimum supported RustFS release advertises `%1F` and all supported clients have refreshed their catalog configuration.
 - `rustfs-5509` FileInfo positional MessagePack decoding: beta.11 serialized 28 fields, while beta.12 inserted transition-version fields in the middle and serialized an incompatible 30-field array. New releases write named maps and retain readers for both shipped array layouts so direct and rolling upgrades can read either release. Remove the positional-array readers after every supported direct-upgrade release writes named maps and no retained RPC payload can contain a pre-map FileInfo array.
@@ -34,6 +33,7 @@ for later deletion.
 - `tonic-013-status-render` peer RPC failure classification: internode failures that reach a node only as text (a peer's error_info payload, a status flattened through format!) are classified by matching the rendering of an Unavailable gRPC status. Releases up to 1.0.0-alpha.38 shipped tonic 0.13, which rendered that status as "status: Unavailable, message: ..."; tonic 0.14 renders it as "code: 'The service is currently unavailable', message: ...". Both forms are matched so an older peer's relayed text still marks an unreachable peer offline. Remove the tonic 0.13 form after the minimum supported RustFS peer version ships tonic 0.14 or later.
 - `rustfs-5063` pre-beta.9 Local KMS recovery: persisted Local KMS configs from beta.8 and earlier predate the explicit insecure-development flag, and encrypted key files use the legacy SHA-256 KDF. Remove the config fallback after supported upgrades have rewritten or explicitly resaved all pre-beta.9 configs with the development-default field, and remove the legacy KDF after supported upgrades have rewritten all pre-beta.9 Local KMS key files with explicit at-rest protection.
 - `sse-local-dek-json-v1` legacy local SSE DEK decoding: releases before the JSON envelope wrote wrapped DEKs as `base64(nonce):base64(ciphertext)`, so readers retain that decoder while all new writes use the versioned JSON envelope. Remove the colon decoder after the minimum supported direct-upgrade release writes JSON envelopes and migration tooling has rewritten every retained legacy object.
+- `multipart-compression-default-off-window` staged multipart disk-compression rollout: releases before the resumable legacy decompressor fail transient reads of compressed objects under mid-payload suspension, so multipart uploads advertise the compression marker only when RUSTFS_COMPRESSION_MULTIPART_ENABLED is set in addition to RUSTFS_COMPRESSION_ENABLED, keeping rolling upgrades from creating new compressed multipart objects while pre-fix nodes may still serve reads. Flip the default to enabled (and retire the extra switch) after the minimum supported direct-upgrade release ships the resumable decompressor.

 ## Review Checklist

@@ -111,7 +111,7 @@ inventory. Generic function-local names such as `CACHE`, `LOCK`, `INIT`, and
 | `GET_OBJECT_BUFFER_THRESHOLD_WARNED`, `GET_READER_STREAM_BUFFER_SIZE_OVERRIDE`, function-local `ENABLED`, `OBJECT_SEEK_SUPPORT_THRESHOLD`, `OBJECT_SEEK_SUPPORT_CONCURRENCY_THRESHOLDS` | `rustfs/src/app/object_usecase.rs` | Cache or constant / owner-local cache | Object GET/seek tuning caches and warning guards stay private to object usecase helpers. |
 | `SUPPORTED_HEADERS` | `rustfs/src/storage/options.rs` | Cache or constant / owner-local constant | Supported-header lookup state stays private to storage option parsing. |
 | `AUDIT_TARGET_SPECS`, `NOTIFICATION_TARGET_SPECS` | `rustfs/src/admin/handlers/audit.rs`, `rustfs/src/admin/handlers/event.rs`, `rustfs/src/admin/handlers/plugins_instances.rs` | Cache or constant / owner-local constant | Admin target descriptor tables stay private to their handler owners. |
-| `SITE_REPLICATION_PEER_CLIENT`, `SITE_REPLICATION_STATE_LOCK` | `rustfs/src/admin/handlers/site_replication.rs` | Process-global owner-local cache / guard | Site-replication peer client cache and state lock stay private to site-replication handlers. |
+| `SITE_REPLICATION_PEER_CLIENT` | `rustfs/src/admin/handlers/site_replication.rs` | Process-global owner-local cache | Site-replication peer client cache stays private to site-replication handlers. The state RMW transaction holds no process-local mutex — see `rustfs/src/admin/site_replication_state.rs`. |
 | `AUDIT_MODULE_ENABLED`, `NOTIFY_MODULE_ENABLED`, `PERSISTED_NOTIFY_MODULE_ENABLED`, `PERSISTED_AUDIT_MODULE_ENABLED`, `PERSISTED_MODULE_SWITCH_CONFIGURED` | `rustfs/src/server/audit.rs`, `rustfs/src/server/event.rs`, `rustfs/src/server/module_switch.rs` | Process-global owner-local toggles | Audit/notify module snapshots stay private to the server module switch owners. |
 | `DELETE_TAIL_TOTAL`, `DELETE_CLEANUP_TOTAL`, `DELETE_REPLICATION_TOTAL`, `DELETE_NOTIFY_TOTAL` | `rustfs/src/delete_tail_activity.rs` | Process-global owner-local counters | Delete-tail activity counters stay private behind delete-tail activity helpers. |
 | `EMBEDDED_SERVER_STARTED` | `rustfs/src/startup_lifecycle.rs` | Process-global owner-local guard | Embedded startup single-start protection stays private to startup lifecycle. |
@@ -61,17 +61,17 @@ catalog extension.

 | Area | Status | Covered behavior |
 |---|---|---|
-| Catalog config | Supported | `GET /v1/config` advertises RustFS catalog defaults and route capabilities. |
+| Catalog config | Supported | `GET /v1/config` advertises RustFS catalog defaults and only the supported OpenAPI REST paths in `endpoints`. RustFS administration, maintenance, migration, diagnostics, refs, and metadata-location extensions remain available but are not presented as standard Iceberg REST endpoints. |
 | Table bucket discovery | Supported | `PUT` and `GET /v1/buckets/{warehouse}` enable and inspect table bucket state. |
 | Namespaces | Supported | Create, list, load, existence check, and drop namespace routes are registered on both catalog prefixes. List responses support Iceberg REST `pageSize`/`pageToken` pagination with context-bound tokens and bounded catalog-store reads. Namespace identifiers are limited to 512 ASCII characters so persisted paths and stateless continuation tokens remain bounded. |
-| Tables | Supported | Create, register, list, load, existence check, commit, metadata-location get/update, and drop table routes are registered on both catalog prefixes. Table and view listings support Iceberg REST `pageSize`/`pageToken` pagination with context-bound tokens and bounded catalog-store reads. |
-| Commit CAS | Supported | Single-table commits validate base metadata, expected version token, referenced object existence, warehouse scope, and Iceberg commit requirements before advancing the current metadata pointer. Standard commits preserve the normal commit-token file name and use an immutable-table-scoped fallback when rename followed by source-name reuse would otherwise collide at the same generation and commit ID. |
+| Tables | Supported | Create, register, list, load, existence check, commit, metadata-location get/update, and drop table routes are registered on both catalog prefixes. Table and view listings support Iceberg REST `pageSize`/`pageToken` pagination with context-bound tokens and bounded catalog-store reads. Commit identifiers must match the URL resource; unknown requirements, updates, and snapshot operations fail as bad requests; staged create, register overwrite, purge-on-drop, and v3-only encryption-key updates return an explicit unsupported-operation response. Standard statistics, partition statistics, and schema/spec cleanup updates are accepted. |
+| Commit CAS | Supported | Single-table commits validate base metadata, expected version token, referenced object existence, warehouse scope, and Iceberg commit requirements before advancing the current metadata pointer. Externally supplied metadata transitions preserve monotonic column, partition, and sequence assignment watermarks and immutable definitions for retained schemas, partition specs, sort orders, and snapshots. Standard commits preserve the normal commit-token file name and use an immutable-table-scoped fallback when rename followed by source-name reuse would otherwise collide at the same generation and commit ID. The catalog does not advertise `idempotency-key-lifetime`; clients must treat standard mutation-wide `Idempotency-Key` semantics as unsupported. |
 | Commit recovery | Supported | Commit log, idempotency lookup, diagnostics, and recovery routes expose staged/finalization gaps and repair safe idempotency gaps without moving the table pointer. |
 | Snapshot refs | Supported | Refs can be listed, created or replaced, and deleted through catalog commits. `main` is protected and refs with explicit retention require forced delete. |
-| Iceberg views | Supported | Basic create, list, load, replace, existence check, and drop routes persist view metadata with view-scoped authorization. |
-| Table credentials endpoint | Supported | Returns an empty `storage-credentials` list by default. Returns table-scoped temporary credentials only when credential vending is enabled. |
+| Iceberg views | Supported | Basic create, list, load, replace, existence check, and drop routes persist view metadata with view-scoped authorization. Replace identifiers must match the URL resource, `schema-id: -1` resolves to the last added schema, one commit timestamp is used consistently, and only Iceberg view format version 1 is accepted. |
+| Table credentials endpoint | Supported | Returns an empty `storage-credentials` list by default. Returns table-scoped temporary credentials only when credential vending is enabled. Credential responses set `Cache-Control: no-store, private`, `Pragma: no-cache`, and `Expires: 0`. |
 | Catalog diagnostics and export | Supported | Exposes recovery state, consistency state, backing manifest, recoverable commit-log WAL state, strong backing migration target, single-active-writer policy, and scale validation matrix. |
-| Catalog import and rollback | Supported | Import/register and rollback use catalog validation and commit paths rather than direct pointer mutation. |
+| Catalog import and rollback | Supported | Import/register and online rollback use catalog validation and commit paths rather than direct pointer mutation. Online rollback accepts only a forward-safe metadata target that preserves assignment watermarks and retained definitions. Restoring an older target that lowers those watermarks is an offline disaster-recovery operation and requires every writer to be stopped. |
 | External catalog bridge | Supported operator path | Operator-supplied metadata pointer sync/import is supported for external catalog identity boundaries. Online vendor SDK polling and policy mirroring are not claimed. |
 | Multi-table transactions | Not claimed | RustFS currently claims single-table commit atomicity only. |

@@ -0,0 +1,218 @@
+# Object I/O (GET/PUT) tuning A/B matrix runbook
+
+> Scope: **parameter tuning** — measuring the effect of changing one
+> `RUSTFS_*` runtime knob at a time, against a fixed binary. This is
+> deliberately different from the code-change A/B gate in
+> [`hotpath-warp-ab-runbook.md`](hotpath-warp-ab-runbook.md) and the formal
+> ABBA validation in
+> [`hotpath-warp-abba-runbook.md`](hotpath-warp-abba-runbook.md), which compare
+> a baseline binary against a candidate binary.
+>
+> When a knob change turns out to need a code change, use those two runbooks
+> for the code-level validation and come back here for the knob-level sweep.
+
+## 1. What this runbook answers
+
+For each tuning knob it answers three questions:
+
+1. Which stage is actually slow — `set_disk_encode`, `set_disk_rename`,
+   `metadata_fanout`, `bitrot_verify`, etc.?
+2. Is the knob the real bottleneck, or is the stage slow for another reason?
+3. Does widening/loosening the knob buy throughput without an unacceptable
+   memory (RSS) or tail-latency regression?
+
+The core discipline is **one variable per A/B cell**. Never change two knobs in
+the same cell, or the result is unexplainable.
+
+## 2. Prerequisites
+
+- Linux bench host (or an ansible-managed cluster); a laptop smoke run is too
+  noisy to decide anything.
+- `warp` on `PATH` (or pass `--warp-bin` to the driver).
+- The observability metrics runtime **enabled**. The stage histograms below are
+  not emitted when `RUSTFS_OBS_METRICS_EXPORT_ENABLED=false` or the runtime is
+  otherwise off — see
+  [`hotpath-warp-ab-runbook.md`](hotpath-warp-ab-runbook.md) for the no-log /
+  no-monitor baseline env.
+- A warm, disposable data set. Recreate the bucket per run; do not bench against
+  production data.
+
+Load driver and gate are reused, not reimplemented:
+
+- `scripts/run_object_batch_bench_enhanced.sh` — warp driver with rounds,
+  median aggregation, `baseline_compare.csv`, and Prometheus service-metric
+  capture.
+- `scripts/hotpath_warp_ab_gate.sh` — relative budget gate over the deltas.
+- `scripts/run_hotpath_warp_ab.sh` — optional orchestrator when a knob needs
+  the full baseline-vs-candidate treatment (e.g. two different defaults).
+
+## 3. Fixed test conditions (lock before you start)
+
+Record these per run; a result without them is not reproducible:
+
+```text
+nodes, disks_per_node, total_disks, cpu_per_node, mem_per_node,
+network, erasure_set_drive_count, endpoint_mode (direct|lb),
+rustfs_commit_sha, warp --version, durability mode
+```
+
+Workload matrix (the same shapes the hotpath gate uses, expanded for the
+stage-breakdown object sizes):
+
+| Workload | mode | sizes |
+| --- | --- | --- |
+| small-fixed | put / get | 4KiB, 100KiB |
+| ec-boundary | put / get | 1MiB, 4MiB |
+| large-stream | put / get | 10MiB, 16MiB, 32MiB |
+| mixed | mixed | 256KiB |
+
+Concurrency ladder: `8, 16, 32, 64` (add `96, 128` on a bigger rig). Duration
+`120s`, `--rounds >= 3`, cooldown `>= 30s`.
+
+Isolate background noise before the sweep: scanner deep-verify, heal,
+replication, lifecycle transition, periodic capacity refresh — record whether
+each is on rather than silently assuming it is off.
+
+## 4. Measurement stack
+
+The code already instruments every stage below. Drive each A/B cell with these
+histograms (names verified against `crates/io-metrics/src/lib.rs`):
+
+- PUT stages: `rustfs_s3_put_object_stage_duration_ms{stage=...}` — compute
+  P50/P95/P99 per stage. Stages: `app_bucket_validate`, `app_sse_config_lookup`,
+  `app_object_lock_config_lookup`, `app_put_opts_build`, `app_prelookup`,
+  `ingress_prepare`, `app_encryption_prepare`, `app_replication_decision`,
+  `app_store_put`, `app_post_store_bookkeeping`, `app_capacity_update`,
+  `set_disk_writer_setup`, `set_disk_encode`, `set_disk_rename`,
+  `set_disk_old_data_cleanup`.
+- GET stages: `rustfs_io_get_object_stage_duration_seconds{path=..., stage=...}` —
+  the `path` label separates the read paths: `legacy_duplex`, `codec_streaming`,
+  `direct_memory`, `body_cache`, `inline_direct`, `internal_meta`,
+  `remote_transition`, `set_disk`, `empty`. Stages: `metadata`,
+  `metadata_cache_lookup`, `metadata_fanout`, `metadata_resolve`, `object_info`,
+  `path_decision`, `quorum_reached`, `range`, `reader_setup`,
+  `stripe_read`, `stripe_read_first_shard`, `stripe_read_quorum`, `decode`,
+  `reconstruct`, `emit`, `fill`, `output_poll`, `output_lock_wait`,
+  `bitrot_verify`, `first_byte`, `full_body`, `response_handoff`,
+  `lock_acquire`.
+- EC memory pressure: `rustfs_ec_encode_inflight_bytes_current` and the
+  allocator reclaim gauge; plus node RSS and CPU.
+
+Host telemetry (collect alongside every cell):
+
+```bash
+pidstat -durh 5 > telemetry/pidstat.txt &
+mpstat 5 > telemetry/mpstat.txt &
+iostat -xz 5 > telemetry/iostat.txt &
+```
+
+## 5. Tuning knob catalog
+
+Defaults are verified against `crates/config/src/constants/object.rs` and
+`crates/ecstore/src/erasure/coding/encode.rs`.
+
+### 5.1 PUT
+
+| Knob | Default | Controls | Validating stage | Risk if widened |
+| --- | --- | --- | --- | --- |
+| `RUSTFS_ERASURE_ENCODE_MAX_INFLIGHT_BYTES` | 32MiB | EC encode producer/consumer memory budget (blocks queued between encode and shard write) | `set_disk_encode` P95 + `rustfs_ec_encode_inflight_bytes_current` | RSS growth under high concurrency |
+| `RUSTFS_OBJECT_IO_BUFFER_SIZE` | 128KiB | Streaming read-in / write-out block size | `ingress_prepare`, `set_disk_encode` | Larger buffers = fewer polls, more resident memory |
+| `RUSTFS_OBJECT_DUPLEX_BUFFER_SIZE` | 4MiB | duplex pipe capacity (shared, but PUT path uses it less than GET) | `set_disk_encode` feed smoothness | Memory per in-flight request |
+| `RUSTFS_DURABILITY_MODE` / `RUSTFS_DRIVE_SYNC_ENABLE` | mode-dependent | per-shard fsync/sync discipline on commit | `set_disk_rename` P99 | Weakening it changes the durability contract — treat as a deliberate tradeoff, not a free win |
+| `RUSTFS_RUNTIME_WORKER_THREADS` / `RUSTFS_RUNTIME_MAX_BLOCKING_THREADS` | Tokio defaults | async workers + `spawn_blocking` pool feeding per-block encode | `set_disk_encode` P95 + mpstat | Oversubscription |
+
+### 5.2 GET
+
+| Knob | Default | Controls | Validating stage | Risk if enabled |
+| --- | --- | --- | --- | --- |
+| `RUSTFS_GET_CODEC_STREAMING_ROLLOUT` | `off` | switches the read path from `legacy_duplex` to the pull-based `ErasureDecodeReader` (`codec_streaming`) | compare `path="legacy_duplex"` vs `path="codec_streaming"` for `decode`/`emit`/`output_lock_wait`/`stripe_read` | behavioral change to the read path; rollout is `off` by default for a reason |
+| `RUSTFS_GET_CODEC_STREAMING_ENGINE` | `legacy` | `legacy` vs `rustfs` decode engine under the streaming reader | `reconstruct`/`decode` per `path` | engine swap on a correctness-critical path |
+| `RUSTFS_GET_CODEC_STREAMING_MULTIPART_ENABLE` | `false` | multipart objects on the streaming reader | same, multipart cells | wider format coverage |
+| `RUSTFS_GET_CODEC_STREAMING_DATA_BLOCKS_FIRST_ENABLE` (+ `_MAX_SIZE`, `_FIRST_READER_SETUP`) | `false` / 512KiB | prefer data-shard readers before parity | `stripe_read_first_shard`/`stripe_read_quorum` | shard-selection order change |
+| `RUSTFS_OBJECT_GET_SKIP_BITROT_VERIFY` | `false` | skip per-shard HighwayHash verify | `bitrot_verify` | **do not default on** — measures the theoretical ceiling only |
+| `RUSTFS_OBJECT_DUPLEX_BUFFER_SIZE` | 4MiB | legacy GET in-process pipe capacity | `output_lock_wait`/`output_poll` | memory per in-flight GET |
+| `RUSTFS_GET_SEEK_BUFFER_ENABLE` | `false` | in-memory seek buffer for small GET | `first_byte` | experimental, startup-latched — see [`get-path-experimental-switches.md`](get-path-experimental-switches.md) |
+| `RUSTFS_GET_OUTPUT_HANDOFF_ATTRIBUTION_ENABLE` | `false` | adds `response_handoff` attribution (metrics only) | `response_handoff` | small per-request bookkeeping cost |
+
+## 6. A/B matrix
+
+Run each row as an independent cell. Baseline is the shipped default; candidate
+is one knob moved. Everything else (topology, sizes, concurrency, rounds,
+durability) stays fixed.
+
+### 6.1 PUT
+
+| # | Knob | Baseline | Candidates | Signal metric | Judgement |
+| --- | --- | --- | --- | --- | --- |
+| P1 | encode in-flight | 32MiB | 48MiB, 64MiB, 96MiB | `set_disk_encode` P95 + throughput + RSS | throughput up AND `set_disk_encode` down AND RSS tolerable → budget was the binding constraint |
+| P2 | object I/O buffer | 128KiB | 256KiB, 512KiB, 1MiB | `ingress_prepare` + `set_disk_encode` | encode smooths with bounded RSS → upstream feed was too small |
+| P3 | duplex buffer | 4MiB | 8MiB, 16MiB | `set_disk_encode` feed variance | lower priority than P1/P2 |
+| P4 | blocking threads | default | 512, 768, 1024 | `set_disk_encode` P95 + mpstat | per-block `spawn_blocking` is scheduler-bound if P95 falls |
+| P5 | durability | current mode | weaker/stronger mode | `set_disk_rename` P99 | only as a **deliberate durability tradeoff**, never a silent default change |
+| P6 | set drive count | current | other valid widths | all PUT stages | highest cost — only after P1–P5, and only on a rebuildable topology |
+
+### 6.2 GET
+
+| # | Knob | Baseline | Candidates | Signal metric | Judgement |
+| --- | --- | --- | --- | --- | --- |
+| G1 | codec streaming rollout | `off` | `on` (pct ramp 10/50/100) | `path="legacy_duplex"` vs `path="codec_streaming"` for `decode`/`emit`/`output_lock_wait`/`stripe_read` + throughput | streaming beats duplex on `decode`+`output_lock_wait` and byte-for-byte output matches → candidate for default |
+| G2 | codec streaming engine | `legacy` | `rustfs` | `reconstruct`/`decode` per `path` | engine swap is neutral-or-better on CPU with identical bytes |
+| G3 | multipart streaming | `false` | `true` | multipart GET cells | only after G1 is stable on single-part |
+| G4 | data-blocks-first | `false` | `true` | `stripe_read_first_shard`/`stripe_read_quorum` | fewer shard reads without a correctness regression |
+| G5 | duplex buffer | 4MiB | 8MiB, 16MiB | `output_lock_wait`/`output_poll` (legacy path) | only if still on `legacy_duplex` |
+| G6 | skip bitrot verify | `false` | `true` | `bitrot_verify` | **ceiling measurement only**; do not carry into production |
+
+## 7. Execution sequence
+
+1. Freeze the conditions in §3 and record the provenance block.
+2. Run the baseline cell (all defaults) and capture stage histograms + host
+   telemetry.
+3. Pick the **one** most-likely knob from the analysis. For large-object PUT
+   that is almost always `set_disk_encode` → P1; for GET it is G1 (the
+   `legacy_duplex` → `codec_streaming` switch).
+4. Sweep that knob's candidate column one value at a time, same workload.
+5. Read the decision table in §8; if the stage did not move, the knob is not
+   the bottleneck — stop widening it and pick the next stage.
+
+Driver invocation for one PUT cell:
+
+```bash
+scripts/run_object_batch_bench_enhanced.sh \
+  --tool warp --endpoint http://127.0.0.1:9000 \
+  --access-key "$RUSTFS_ACCESS_KEY" --secret-key "$RUSTFS_SECRET_KEY" \
+  --bucket rustfs-put-tuning --warp-mode put \
+  --sizes 16MiB,32MiB --concurrency 32 --duration 120s --rounds 3 \
+  --out-dir target/bench/put-tuning-p1-64mib
+```
+
+## 8. Interpretation / decision table
+
+| Stage high | Most likely cause | Next action |
+| --- | --- | --- |
+| `set_disk_encode` | per-block EC encode scheduling + in-flight budget | P1 → P4 → P2, in that order |
+| `set_disk_rename` | commit tail (rename fan-out / RPC / fsync) | P5 (durability) and cluster tail analysis, not encode |
+| `set_disk_writer_setup` | per-disk `BitrotWriter` + temp-file create | disk/filesystem metadata; per-disk fan-out cost |
+| `set_disk_old_data_cleanup` | overwrite / versioned-object directory delete | confirm overwrite-vs-new-write; defer cleanup further |
+| `metadata_fanout` / `metadata_resolve` | cross-disk `xl.meta` read + quorum | metadata cache hit rate; small-object fixed cost |
+| `bitrot_verify` | HighwayHash verify on the read path | G6 ceiling only; do not default on |
+| `output_lock_wait` / `output_poll` | legacy duplex backpressure | G1 (move off duplex) or G5 |
+| `stripe_read*` | shard concurrency / selection | G4 shard-selection, disk/network tail |
+
+## 9. Guardrails
+
+- **Never weaken correctness for throughput**: read/write quorum, bitrot verify,
+  `xl.meta` validation, and durability (`RUSTFS_DURABILITY_MODE`) are integrity
+  contracts, not knobs. P5 and G6 are ceiling measurements and must be labelled
+  as such; do not carry their values into production without an explicit
+  durability/correctness decision.
+- **One variable per cell.** A cell that changes two knobs is thrown away.
+- **Memory is part of the result.** A throughput win with unbounded RSS growth
+  is a regression; record RSS and the EC in-flight gauge for every PUT cell.
+- **Startup-latched knobs** (`RUSTFS_GET_SEEK_BUFFER_ENABLE`,
+  `RUSTFS_GET_OUTPUT_HANDOFF_ATTRIBUTION_ENABLE`, and the codec-streaming
+  switches) require a process restart to change — see
+  [`get-path-experimental-switches.md`](get-path-experimental-switches.md).
+- **Archive the raw data.** Keep the `baseline_compare.csv`, `median_summary.csv`,
+  stage histograms, and host telemetry per cell; the conclusion must trace back
+  to them. Do not commit benchmark result snapshots to the repo — record them in
+  the issue tracker.
@@ -59,7 +59,7 @@
        {
          default = rustPlatform.buildRustPackage {
            pname = "rustfs";
-            version = "1.0.0-rc.1";
+            version = "1.0.0-rc.2";

            src = ./.;

@@ -2,8 +2,8 @@ apiVersion: v2
 name: rustfs
 description: RustFS helm chart to deploy RustFS on kubernetes cluster.
 type: application
-version: "1.0.0-rc.1"
-appVersion: "1.0.0-rc.1"
+version: "1.0.0-rc.2"
+appVersion: "1.0.0-rc.2"
 home: https://rustfs.com
 icon: https://media.sys.truenas.net/apps/rustfs/icons/icon.svg
 maintainers:
@@ -1,9 +1,9 @@
 %global _enable_debug_packages 0
 %global _empty_manifest_terminate_build 0
-%global prerelease rc.1
+%global prerelease rc.2
 Name:           rustfs
 Version:        1.0.0
-Release:        rc.1
+Release:        rc.2
 Summary:       High-performance distributed object storage for MinIO alternative

 License:        Apache-2.0
@@ -58,6 +58,9 @@ install %_builddir/%{name}-%{version}-%{prerelease}/target/%_arch/%_arch-unknown
 %_bindir/rustfs

 %changelog
+* Fri Aug 14 2026 overtrue <anzhengchao@gmail.com>
+- Update RPM package to RustFS 1.0.0-rc.2
+
 * Sat Aug 08 2026 overtrue <anzhengchao@gmail.com>
 - Update RPM package to RustFS 1.0.0-rc.1

@@ -278,6 +278,8 @@ rustfs-signer.workspace = true
 serde = { workspace = true, features = ["derive"] }
 serde_json = { workspace = true, features = ["raw_value"] }
 serde_urlencoded = { workspace = true }
+snap.workspace = true
+zstd.workspace = true

 # Cryptography and Security
 rustls = { workspace = true, default-features = false, features = ["aws-lc-rs", "logging", "tls12", "prefer-post-quantum", "std"] }
@@ -355,7 +357,6 @@ rcgen = { workspace = true }
 rustfs-test-utils.workspace = true
 # diagnose_e2e fixtures (archives are generated in-test, never checked in)
 zip = { workspace = true }
-zstd = { workspace = true }
 # Enables the shared MockWarmBackend / xl.meta assertion helpers exposed via
 # the ecstore `api::tier::test_util` facade module (rustfs/backlog#1148 ilm-6).
 rustfs-ecstore = { workspace = true, features = ["test-util"] }
@@ -40,7 +40,7 @@ use s3s::{Body, S3Request, S3Response, S3Result, s3_error};
 use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
 use std::sync::LazyLock;
-use tracing::{Span, error, info, warn};
+use tracing::{error, info, warn};

 const LOG_COMPONENT_ADMIN_API: &str = "admin_api";
 const LOG_SUBSYSTEM_AUDIT_TARGET: &str = "audit_target";
@@ -278,8 +278,6 @@ pub struct AuditTargetConfig {}
 #[async_trait::async_trait]
 impl Operation for AuditTargetConfig {
    async fn call(&self, req: S3Request<Body>, params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
-        let span = Span::current();
-        let _enter = span.enter();
        let (target_type, target_name) = extract_target_params(&params)?;

        authorize_audit_admin_request(&req, AdminAction::SetBucketTargetAction).await?;
@@ -345,8 +343,6 @@ pub struct ListAuditTargets {}
 #[async_trait::async_trait]
 impl Operation for ListAuditTargets {
    async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
-        let span = Span::current();
-        let _enter = span.enter();
        authorize_audit_admin_request(&req, AdminAction::GetBucketTargetAction).await?;

        let mut runtime_statuses = HashMap::new();
@@ -370,8 +366,6 @@ pub struct RemoveAuditTarget {}
 #[async_trait::async_trait]
 impl Operation for RemoveAuditTarget {
    async fn call(&self, req: S3Request<Body>, params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
-        let span = Span::current();
-        let _enter = span.enter();
        let (target_type, target_name) = extract_target_params(&params)?;

        authorize_audit_admin_request(&req, AdminAction::SetBucketTargetAction).await?;
@@ -838,6 +832,13 @@ mod tests {
            extract_block_between_markers(src, "impl Operation for ListAuditTargets", "pub struct RemoveAuditTarget");
        let delete_block = extract_block_between_markers(src, "impl Operation for RemoveAuditTarget", "#[cfg(test)]");

+        for block in [put_block, list_block, delete_block] {
+            assert!(
+                !block.contains(".enter()"),
+                "async audit handlers must rely on request-future instrumentation instead of holding span guards across awaits"
+            );
+        }
+
        assert!(
            put_block.contains("authorize_audit_admin_request(&req, AdminAction::SetBucketTargetAction).await?;"),
            "audit target writes should require SetBucketTargetAction"
@@ -465,6 +465,16 @@ impl Operation for ImportBucketMetadata {
            file_contents.push((file_path, content));
        }

+        let durable_quota_import = imported_quota_requires_fleet_proof(&file_contents)?;
+        let quota_fleet_proof =
+            if durable_quota_import {
+                Some(crate::admin::storage_api::acquire_cross_pool_fence_fleet_proof().ok_or_else(|| {
+                    s3_error!(ServiceUnavailable, "durable quota capability is not confirmed across the cluster")
+                })?)
+            } else {
+                None
+            };
+
        // Extract bucket names
        let mut bucket_names = Vec::new();
        for (file_path, _) in &file_contents {
@@ -707,21 +717,6 @@ impl Operation for ImportBucketMetadata {
                }

                BUCKET_QUOTA_CONFIG_FILE => {
-                    if let Err(e) = serde_json::from_slice::<BucketQuota>(&content) {
-                        warn!(
-                            event = EVENT_ADMIN_BUCKET_META_STATE,
-                            component = LOG_COMPONENT_ADMIN,
-                            subsystem = LOG_SUBSYSTEM_BUCKET_META,
-                            action = "import_bucket_metadata",
-                            result = "config_deserialize_failed",
-                            bucket = %bucket_name,
-                            config_name = %conf_name,
-                            error = %e,
-                            "admin bucket meta state"
-                        );
-                        continue;
-                    }
-
                    let metadata = match bucket_metadatas.get_mut(bucket_name) {
                        Some(m) => m,
                        None => continue,
@@ -830,10 +825,6 @@ impl Operation for ImportBucketMetadata {
            }
        }

-        // Persist the assembled metadata to disk. Prior to this, the import only mutated the
-        // in-memory `bucket_metadatas` map and returned 200, silently dropping every imported
-        // config. `metadata_sys::update` loads the on-disk metadata, overwrites the given config
-        // field and saves it, preserving any configs not present in the import archive.
        for (bucket_name, metadata) in &bucket_metadatas {
            for (config_file, data) in imported_configs_to_persist(metadata) {
                let site_replication_item = imported_config_to_site_replication_item(bucket_name, metadata, config_file, &data)?;
@@ -842,7 +833,21 @@ impl Operation for ImportBucketMetadata {
                } else {
                    None
                };
-                if let Err(e) = metadata_sys::update(bucket_name, config_file, data).await {
+                let persist_result = if config_file == BUCKET_QUOTA_CONFIG_FILE {
+                    let quota: BucketQuota =
+                        serde_json::from_slice(&data).map_err(|e| s3_error!(InvalidRequest, "invalid bucket quota: {e}"))?;
+                    if quota.uses_durable_reservations() {
+                        let proof = quota_fleet_proof.as_ref().ok_or_else(|| {
+                            s3_error!(ServiceUnavailable, "durable quota capability is not confirmed across the cluster")
+                        })?;
+                        metadata_sys::update_quota_if_incarnation(bucket_name, data, metadata.bucket_incarnation_id, proof).await
+                    } else {
+                        metadata_sys::update_if_incarnation(bucket_name, config_file, data, metadata.bucket_incarnation_id).await
+                    }
+                } else {
+                    metadata_sys::update_if_incarnation(bucket_name, config_file, data, metadata.bucket_incarnation_id).await
+                };
+                if let Err(e) = persist_result {
                    warn!(
                        event = EVENT_ADMIN_BUCKET_META_STATE,
                        component = LOG_COMPONENT_ADMIN,
@@ -885,6 +890,26 @@ impl Operation for ImportBucketMetadata {
    }
 }

+fn imported_quota_requires_fleet_proof(file_contents: &[(String, Vec<u8>)]) -> S3Result<bool> {
+    let mut durable = false;
+    for (file_path, content) in file_contents {
+        let mut parts = file_path.split(SLASH_SEPARATOR);
+        let Some(_bucket) = parts.next() else {
+            continue;
+        };
+        if parts.next() != Some(BUCKET_QUOTA_CONFIG_FILE) {
+            continue;
+        }
+        let quota: BucketQuota =
+            serde_json::from_slice(content).map_err(|e| s3_error!(InvalidRequest, "invalid bucket quota: {e}"))?;
+        if quota.has_unsupported_reservation_protocol() {
+            return Err(s3_error!(InvalidRequest, "unsupported bucket quota reservation protocol"));
+        }
+        durable |= quota.uses_durable_reservations();
+    }
+    Ok(durable)
+}
+
 /// The `(config_file, data)` pairs to persist for an imported bucket's metadata: every non-empty
 /// config field keyed by its on-disk config-file name, as owned data ready for
 /// `metadata_sys::update`. Empty fields are skipped so an import never overwrites an existing
@@ -1076,4 +1101,31 @@ mod import_persist_tests {

        assert!(!has_site_replication_item);
    }
+
+    #[test]
+    fn quota_import_preflight_rejects_invalid_and_unknown_protocols() {
+        let missing_limit = vec![(
+            format!("bucket/{BUCKET_QUOTA_CONFIG_FILE}"),
+            br#"{"quota":0,"reservation_protocol":1}"#.to_vec(),
+        )];
+        assert!(imported_quota_requires_fleet_proof(&missing_limit).is_err());
+
+        let unknown_protocol = vec![(
+            format!("bucket/{BUCKET_QUOTA_CONFIG_FILE}"),
+            br#"{"quota":0,"reservation_protocol":2,"reservation_quota":1024}"#.to_vec(),
+        )];
+        assert!(imported_quota_requires_fleet_proof(&unknown_protocol).is_err());
+    }
+
+    #[test]
+    fn quota_import_preflight_requires_proof_only_for_durable_quota() {
+        let legacy = vec![(format!("bucket/{BUCKET_QUOTA_CONFIG_FILE}"), br#"{"quota":1024}"#.to_vec())];
+        assert!(!imported_quota_requires_fleet_proof(&legacy).expect("legacy quota should remain compatible"));
+
+        let durable = vec![(
+            format!("bucket/{BUCKET_QUOTA_CONFIG_FILE}"),
+            serde_json::to_vec(&BucketQuota::new(Some(1024))).expect("durable quota should encode"),
+        )];
+        assert!(imported_quota_requires_fleet_proof(&durable).expect("durable quota should pass preflight"));
+    }
 }
@@ -43,7 +43,7 @@ use s3s::{Body, S3Request, S3Response, S3Result, s3_error};
 use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
 use std::sync::LazyLock;
-use tracing::{Span, error, info, warn};
+use tracing::{error, info, warn};

 const LOG_COMPONENT_ADMIN_API: &str = "admin_api";

@@ -333,8 +333,6 @@ pub struct NotificationTarget {}
 #[async_trait::async_trait]
 impl Operation for NotificationTarget {
    async fn call(&self, req: S3Request<Body>, params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
-        let span = Span::current();
-        let _enter = span.enter();
        let (target_type, target_name) = extract_target_params(&params)?;
        let context = app_context_from_req(&req);

@@ -401,8 +399,6 @@ pub struct ListNotificationTargets {}
 #[async_trait::async_trait]
 impl Operation for ListNotificationTargets {
    async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
-        let span = Span::current();
-        let _enter = span.enter();
        authorize_notification_admin_request(&req, AdminAction::GetBucketTargetAction).await?;
        refresh_persisted_module_switches_from_store().await.map_err(|err| {
            warn!(
@@ -439,8 +435,6 @@ pub struct ListTargetsArns {}
 #[async_trait::async_trait]
 impl Operation for ListTargetsArns {
    async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
-        let span = Span::current();
-        let _enter = span.enter();
        authorize_notification_admin_request(&req, AdminAction::GetBucketTargetAction).await?;
        if let Some(reason) = notification_target_operation_block_reason(
            "querying notification target ARNs for bucket associations from the console",
@@ -485,8 +479,6 @@ pub struct RemoveNotificationTarget {}
 #[async_trait::async_trait]
 impl Operation for RemoveNotificationTarget {
    async fn call(&self, req: S3Request<Body>, params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
-        let span = Span::current();
-        let _enter = span.enter();
        let (target_type, target_name) = extract_target_params(&params)?;
        let context = app_context_from_req(&req);

@@ -1006,6 +998,13 @@ mod tests {
            extract_block_between_markers(src, "impl Operation for ListTargetsArns", "pub struct RemoveNotificationTarget");
        let delete_block = extract_block_between_markers(src, "impl Operation for RemoveNotificationTarget", "fn extract_param");

+        for block in [put_block, list_block, arns_block, delete_block] {
+            assert!(
+                !block.contains(".enter()"),
+                "async notification handlers must rely on request-future instrumentation instead of holding span guards across awaits"
+            );
+        }
+
        assert!(
            put_block.contains("authorize_notification_admin_request(&req, AdminAction::SetBucketTargetAction).await?;"),
            "notification target writes should require SetBucketTargetAction"
@@ -22,6 +22,7 @@ use crate::admin::storage_api::bucket::metadata_sys::{self, BucketMetadataSys};
 use crate::admin::storage_api::bucket::quota::checker::QuotaChecker;
 use crate::admin::storage_api::bucket::quota::{BucketQuota, QuotaError, QuotaOperation};
 use crate::auth::{check_key_valid, get_session_token};
+use crate::error::ApiError;
 use crate::server::ADMIN_PREFIX;
 use hyper::{Method, StatusCode};
 use matchit::Params;
@@ -293,16 +294,36 @@ impl Operation for SetBucketQuotaHandler {
            return Err(s3_error!(InvalidArgument, "{}", rustfs_config::QUOTA_INVALID_TYPE_ERROR_MSG));
        }

+        let fleet_proof = if request.quota.is_some() {
+            Some(crate::admin::storage_api::acquire_cross_pool_fence_fleet_proof().ok_or_else(|| {
+                S3Error::with_message(
+                    s3s::S3ErrorCode::ServiceUnavailable,
+                    "durable quota capability is not confirmed across the cluster".to_string(),
+                )
+            })?)
+        } else {
+            None
+        };
+
        let quota = BucketQuota::new(request.quota);

        let metadata_sys_lock = bucket_metadata_from_context()
            .ok_or_else(|| s3_error!(InternalError, "{}", rustfs_config::QUOTA_METADATA_SYSTEM_ERROR_MSG))?;
        let mut quota_checker = QuotaChecker::new(metadata_sys_lock.clone());

-        let updated_at = quota_checker
-            .set_quota_config_if_incarnation(&bucket, quota.clone(), expected_incarnation_id)
-            .await
-            .map_err(|e| s3_error!(InternalError, "failed to set quota: {}", e))?;
+        let updated_at = match fleet_proof.as_ref() {
+            Some(fleet_proof) => {
+                quota_checker
+                    .set_durable_quota_config_if_incarnation(&bucket, quota.clone(), expected_incarnation_id, fleet_proof)
+                    .await
+            }
+            None => {
+                quota_checker
+                    .set_quota_config_if_incarnation(&bucket, quota.clone(), expected_incarnation_id)
+                    .await
+            }
+        }
+        .map_err(ApiError::from)?;

        if let Err(err) = site_replication_bucket_meta_hook(SRBucketMeta {
            bucket: bucket.clone(),
@@ -535,8 +535,13 @@ impl Operation for GetReplicationMetricsHandler {

        let bucket_stats = cluster_replication_stats(bucket, app_context_from_req(&req)).await;

-        let data = serde_json::to_vec(&bucket_stats.replication_stats)
-            .map_err(|_| S3Error::with_message(S3ErrorCode::InternalError, "serialize failed"))?;
+        // Same minio-go `replication.Metrics` wire shape as
+        // `?replication-metrics` — the internal snake_case stats are the peer
+        // RPC wire format and must not leak here.
+        let data = serde_json::to_vec(&crate::admin::replication_metrics_wire::MetricsWire::from(
+            &bucket_stats.replication_stats,
+        ))
+        .map_err(|_| S3Error::with_message(S3ErrorCode::InternalError, "serialize failed"))?;
        let mut headers = HeaderMap::new();
        headers.insert(CONTENT_TYPE, HeaderValue::from_static("application/json"));
        Ok(S3Response::with_headers((StatusCode::OK, Body::from(data)), headers))
@@ -1153,7 +1158,7 @@ struct MrfResponse {
 fn build_mrf_response(
    bucket: String,
    bucket_stats: &BucketStats,
-    durable: crate::admin::storage_api::replication::DurableMrfBacklog,
+    durable: &crate::admin::storage_api::replication::DurableMrfBacklog,
 ) -> MrfResponse {
    let observation_scope = if bucket_stats.replication_stats.cluster_complete {
        "cluster_aggregated"
@@ -1223,7 +1228,10 @@ fn build_mrf_response(
        total_failed_size,
        queued_count: queued.count,
        queued_size: queued.bytes,
-        per_object_entries_available: false,
+        // The default (non-aggregate) response mode streams the durable
+        // backlog per object, so the enumerable API exists whenever the
+        // backlog is readable.
+        per_object_entries_available: durable.available,
        runtime_stats_available: bucket_stats.replication_stats.provider_available,
        cluster_complete: bucket_stats.replication_stats.cluster_complete,
        observed_node_count: bucket_stats.replication_stats.observed_node_count,
@@ -1235,23 +1243,165 @@ fn build_mrf_response(
    }
 }

+/// One durable MRF backlog entry rendered for the default (madmin-compatible)
+/// stream. Field names are the exact json tags of madmin-go `ReplicationMRF`
+/// (replication-api.go), which `mc replicate backlog` decodes one JSON
+/// document at a time. `Size` and `TargetARNs` are RustFS extension keys with
+/// no madmin counterpart; Go decoders ignore unknown keys.
+#[derive(Debug, Serialize)]
+struct MrfEntryDocument {
+    /// The durable backlog is a cluster-shared ledger with no per-node
+    /// attribution, so the madmin `nodeName` tag is always empty.
+    #[serde(rename = "nodeName")]
+    node_name: String,
+    #[serde(rename = "bucket")]
+    bucket: String,
+    #[serde(rename = "object")]
+    object: String,
+    #[serde(rename = "versionId")]
+    version_id: String,
+    #[serde(rename = "retryCount")]
+    retry_count: i32,
+    #[serde(rename = "Size")]
+    size: i64,
+    #[serde(rename = "TargetARNs", skip_serializing_if = "Vec::is_empty")]
+    target_arns: Vec<String>,
+}
+
+/// Upper bound on the number of documents one stream response emits. The
+/// durable ledger is not bounded by the in-memory pending cap (recovery can
+/// persist far larger generations), and the body is buffered before send, so
+/// an unbounded read could stage hundreds of MB per request. The handler
+/// rejects a response beyond this bound instead of returning a partial 200.
+const REPLICATION_MRF_MAX_STREAM_ENTRIES: usize = 10_000;
+
+/// Project the durable backlog into madmin `ReplicationMRF` documents,
+/// scoped to `bucket` when it is non-empty (madmin allows an empty bucket to
+/// mean "across all buckets"), bounded by
+/// [`REPLICATION_MRF_MAX_STREAM_ENTRIES`]. Returns the documents and whether
+/// the backlog was truncated.
+fn mrf_entry_documents(
+    bucket: &str,
+    durable: &crate::admin::storage_api::replication::DurableMrfBacklog,
+) -> (Vec<MrfEntryDocument>, bool) {
+    let mut documents = Vec::new();
+    let mut truncated = false;
+    for entry in durable
+        .entries
+        .iter()
+        .filter(|entry| bucket.is_empty() || entry.bucket == bucket)
+    {
+        if documents.len() >= REPLICATION_MRF_MAX_STREAM_ENTRIES {
+            truncated = true;
+            break;
+        }
+        documents.push(MrfEntryDocument {
+            node_name: String::new(),
+            bucket: entry.bucket.clone(),
+            object: entry.object.clone(),
+            // Delete-marker purge entries track the marker version separately;
+            // fall back to it so those rows still carry a version identity.
+            // The nil UUID is RustFS's in-memory null-version sentinel and
+            // must leave as the S3 wire token, not a zero UUID.
+            version_id: entry
+                .version_id
+                .or(entry.delete_marker_version_id)
+                .map(|v| {
+                    if v.is_nil() {
+                        rustfs_filemeta::NULL_VERSION_ID.to_string()
+                    } else {
+                        v.to_string()
+                    }
+                })
+                .unwrap_or_default(),
+            retry_count: entry.retry_count,
+            size: entry.size,
+            target_arns: entry.target_arns.clone(),
+        });
+    }
+    (documents, truncated)
+}
+
+/// Render the MRF backlog as a response body.
+///
+/// Default (madmin-compatible) mode emits one `ReplicationMRF` JSON document
+/// per line with no envelope — madmin's `BucketReplicationMRF` reads the body
+/// with a `json.Decoder` loop, so an envelope object would decode as a single
+/// entry whose `"Bucket"` key case-insensitively matches
+/// `ReplicationMRF.Bucket` (a phantom row in `mc replicate backlog`), and an
+/// empty backlog must render an empty body so the loop ends on io.EOF with
+/// zero rows.
+///
+/// `aggregate=true` (RustFS extension) keeps the enveloped counter shape;
+/// backlog-source health (`RuntimeStatsAvailable`/`DurableBacklogAvailable`)
+/// is only representable there — an unreadable ledger fails the stream
+/// request outright in the handler (madmin only decodes the body of a 200,
+/// so an empty stream would read as a healthy zero-row backlog).
+fn render_mrf_backlog(
+    response: &MrfResponse,
+    durable: &crate::admin::storage_api::replication::DurableMrfBacklog,
+    aggregate: bool,
+) -> Result<(Vec<u8>, bool), serde_json::Error> {
+    if aggregate {
+        return Ok((serde_json::to_vec(response)?, false));
+    }
+
+    let (documents, truncated) = mrf_entry_documents(&response.bucket, durable);
+    let mut data = Vec::new();
+    for entry in documents {
+        serde_json::to_writer(&mut data, &entry)?;
+        data.push(b'\n');
+    }
+    Ok((data, truncated))
+}
+
+fn ensure_complete_mrf_stream(truncated: bool) -> S3Result<()> {
+    if truncated {
+        return Err(S3Error::with_message(
+            S3ErrorCode::ServiceUnavailable,
+            "durable MRF backlog exceeds the stream limit; narrow the bucket scope or drain the backlog".to_string(),
+        ));
+    }
+    Ok(())
+}
+
 /// `GET /v3/replication/mrf`
 ///
 /// Reports the failed-replication backlog (MinIO's MRF concept) for a bucket.
 ///
-/// Compatibility note: MinIO returns a stream of individual MRF entries. RustFS
-/// deliberately returns aggregate runtime and durable counters instead.
-/// `PerObjectEntriesAvailable` remains false until an enumerable API exists.
-/// `PerTargetDurableEntriesAvailable` is false when the durable backlog includes
-/// older entries that cannot be attributed to a target.
+/// The default response is a madmin-compatible stream of `ReplicationMRF`
+/// documents built from the durable backlog ledger (in-memory failures that
+/// have not been flushed yet — the persister runs every few seconds — are not
+/// visible). `?aggregate=true` (RustFS extension) returns the enveloped
+/// runtime + durable counter shape instead; `PerTargetDurableEntriesAvailable`
+/// is false there when the durable backlog includes older entries that cannot
+/// be attributed to a target.
+///
+/// The madmin `node` parameter is accepted but has no filtering effect: the
+/// durable ledger is cluster-shared with no per-node attribution, so every
+/// node serves the same (complete) backlog.
+///
+/// Authorization: the stream requires `admin:ReplicationDiff` (it enumerates
+/// object names and version ids, MinIO parity); `?aggregate=true` carries no
+/// object identities and requires only `admin:GetReplicationMetrics`.
 pub struct ReplicationMrfHandler {}

 #[async_trait::async_trait]
 impl Operation for ReplicationMrfHandler {
    async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
-        validate_replication_admin_request(&req, AdminAction::GetReplicationMetricsAction).await?;
-
        let queries = extract_query_params(&req.uri);
+        let aggregate = queries.get("aggregate").map(String::as_str) == Some("true");
+        // The default stream enumerates object names and version ids, which
+        // a metrics-only principal must not see; gate it on the same action
+        // MinIO uses for this endpoint. The aggregate counters carry no
+        // object identities and keep the metrics action.
+        let action = if aggregate {
+            AdminAction::GetReplicationMetricsAction
+        } else {
+            AdminAction::ReplicationDiff
+        };
+        validate_replication_admin_request(&req, action).await?;
+
        let Some(bucket) = queries.get("bucket").filter(|b| !b.is_empty()).cloned() else {
            return Err(s3_error!(InvalidRequest, "bucket is required"));
        };
@@ -1275,14 +1425,48 @@ impl Operation for ReplicationMrfHandler {
            return Err(ApiError::from(err).into());
        }

+        if let Some(node) = queries.get("node").filter(|node| !node.is_empty() && node.as_str() != "all") {
+            // The durable backlog ledger is cluster-shared with no per-node
+            // attribution, so a node-scoped request still sees the complete
+            // (superset) backlog.
+            debug!(node = %node, "replication mrf node filter has no effect on the cluster-shared backlog");
+        }
+
        let durable = crate::admin::storage_api::replication::read_durable_mrf_backlog(store).await;
        let bucket_stats = cluster_replication_stats(&bucket, app_context_from_req(&req)).await;
-        let response = build_mrf_response(bucket, &bucket_stats, durable);
+        let response = build_mrf_response(bucket, &bucket_stats, &durable);

-        let data = serde_json::to_vec(&response)
+        if !durable.available && !aggregate {
+            // The madmin stream has no envelope to carry source health, and
+            // madmin only decodes the body of a 200 — an empty stream would
+            // read as a clean, healthy zero-row backlog. Fail loudly instead;
+            // aggregate mode still reports the availability fields.
+            tracing::warn!(
+                bucket = %response.bucket,
+                "durable MRF backlog is unreadable; failing the stream request — use aggregate=true to see source health"
+            );
+            return Err(S3Error::with_message(
+                S3ErrorCode::ServiceUnavailable,
+                "durable MRF backlog is unreadable; retry, or use aggregate=true for source health".to_string(),
+            ));
+        }
+
+        let (data, truncated) = render_mrf_backlog(&response, &durable, aggregate)
            .map_err(|e| S3Error::with_message(S3ErrorCode::InternalError, format!("serialize failed: {e}")))?;
        let mut headers = HeaderMap::new();
        headers.insert(CONTENT_TYPE, HeaderValue::from_static("application/json"));
+        if truncated {
+            tracing::warn!(
+                event = "replication_mrf_stream_rejected",
+                component = "admin",
+                subsystem = "replication",
+                result = "rejected",
+                bucket = %response.bucket,
+                max_entries = REPLICATION_MRF_MAX_STREAM_ENTRIES,
+                "replication mrf stream exceeds the response limit"
+            );
+        }
+        ensure_complete_mrf_stream(truncated)?;
        Ok(S3Response::with_headers((StatusCode::OK, Body::from(data)), headers))
    }
 }
@@ -1292,7 +1476,8 @@ mod tests {
    use super::{
        REMOTE_TARGET_UNSUPPORTED_FIELDS, REMOTE_TARGET_WRITABLE_FIELDS, RemoteTargetCredentialsRequest, RemoteTargetRequest,
        ReplicationDiffEntry, SUPPORTED_REMOTE_TARGET_API, TargetUpdateOp, build_mrf_response, extract_query_params,
-        parse_remote_target_update_ops, render_replication_diff, unique_replication_peers, validate_remote_target_tls_settings,
+        parse_remote_target_update_ops, render_mrf_backlog, render_replication_diff, unique_replication_peers,
+        validate_remote_target_tls_settings,
    };
    use crate::admin::storage_api::bucket::target::{BucketTarget, LatencyStat};
    use crate::admin::storage_api::replication::{BucketStats, DurableMrfBacklog, MrfOpKind, MrfReplicateEntry};
@@ -1510,7 +1695,7 @@ mod tests {
            ],
        };

-        let response = build_mrf_response("bucket-a".to_string(), &stats, durable);
+        let response = build_mrf_response("bucket-a".to_string(), &stats, &durable);
        let json = serde_json::to_value(response).expect("MRF response should serialize");

        assert_eq!(json["TotalFailedCount"], 3);
@@ -1522,7 +1707,9 @@ mod tests {
        assert_eq!(json["RuntimeStatsAvailable"], true);
        assert_eq!(json["ClusterComplete"], false);
        assert_eq!(json["Targets"][0]["ObservationScope"], "partial_cluster");
-        assert_eq!(json["PerObjectEntriesAvailable"], false);
+        // The bare stream enumerates the durable backlog per object, so a
+        // readable backlog advertises the enumerable API.
+        assert_eq!(json["PerObjectEntriesAvailable"], true);
        assert_eq!(json["PerTargetDurableEntriesAvailable"], true);

        let targets = json["Targets"].as_array().expect("targets should serialize as an array");
@@ -1569,7 +1756,7 @@ mod tests {
            }],
        };

-        let response = build_mrf_response("bucket-a".to_string(), &stats, durable);
+        let response = build_mrf_response("bucket-a".to_string(), &stats, &durable);
        let json = serde_json::to_value(response).expect("MRF response should serialize");

        assert_eq!(json["DurableBacklogAvailable"], true);
@@ -1587,7 +1774,7 @@ mod tests {

    #[test]
    fn mrf_response_distinguishes_unavailable_sources_from_valid_zero() {
-        let unavailable = build_mrf_response("bucket-a".to_string(), &BucketStats::default(), DurableMrfBacklog::default());
+        let unavailable = build_mrf_response("bucket-a".to_string(), &BucketStats::default(), &DurableMrfBacklog::default());
        let unavailable_json = serde_json::to_value(unavailable).expect("unavailable response should serialize");
        assert_eq!(unavailable_json["RuntimeStatsAvailable"], false);
        assert_eq!(unavailable_json["DurableBacklogAvailable"], false);
@@ -1600,7 +1787,7 @@ mod tests {
        let valid_empty = build_mrf_response(
            "bucket-a".to_string(),
            &valid_empty_stats,
-            DurableMrfBacklog {
+            &DurableMrfBacklog {
                available: true,
                entries: Vec::new(),
            },
@@ -1613,6 +1800,153 @@ mod tests {
        assert_eq!(valid_empty_json["PerTargetDurableEntriesAvailable"], true);
    }

+    fn sample_durable_backlog() -> DurableMrfBacklog {
+        DurableMrfBacklog {
+            available: true,
+            entries: vec![
+                MrfReplicateEntry {
+                    bucket: "bucket-a".to_string(),
+                    object: "object-a".to_string(),
+                    version_id: Some(uuid::Uuid::from_u128(7)),
+                    retry_count: 2,
+                    size: 250,
+                    op: MrfOpKind::Object,
+                    target_arns: vec!["arn-a".to_string()],
+                    ..Default::default()
+                },
+                MrfReplicateEntry {
+                    bucket: "other-bucket".to_string(),
+                    object: "object-b".to_string(),
+                    version_id: None,
+                    retry_count: 0,
+                    size: 999,
+                    op: MrfOpKind::Object,
+                    target_arns: Vec::new(),
+                    ..Default::default()
+                },
+            ],
+        }
+    }
+
+    /// madmin's `BucketReplicationMRF` decodes the body one `ReplicationMRF`
+    /// JSON document at a time; the default response must therefore be a bare
+    /// document stream with madmin's exact json tags, not an envelope.
+    #[test]
+    fn mrf_stream_renders_bare_madmin_documents() {
+        let durable = sample_durable_backlog();
+        let response = build_mrf_response("bucket-a".to_string(), &BucketStats::default(), &durable);
+
+        let (body, _) = render_mrf_backlog(&response, &durable, false).expect("stream body should serialize");
+        let text = String::from_utf8(body).expect("body should be utf-8");
+        let lines: Vec<&str> = text.lines().filter(|line| !line.trim().is_empty()).collect();
+
+        // Only the entry matching the requested bucket is streamed.
+        assert_eq!(lines.len(), 1, "expected one MRF document, got: {text}");
+        let doc: serde_json::Value = serde_json::from_str(lines[0]).expect("each line should be a JSON document");
+        assert_eq!(doc["bucket"], "bucket-a");
+        assert_eq!(doc["object"], "object-a");
+        assert_eq!(doc["versionId"], uuid::Uuid::from_u128(7).to_string());
+        assert_eq!(doc["retryCount"], 2);
+        // madmin `ReplicationMRF` has a `nodeName` tag; the durable backlog is
+        // cluster-shared, so RustFS reports an empty node name.
+        assert_eq!(doc["nodeName"], "");
+        // The envelope keys must not leak into the stream: a `"Bucket"` key
+        // would case-insensitively populate `ReplicationMRF.Bucket` and render
+        // a phantom row in `mc replicate backlog`.
+        assert!(doc.get("Bucket").is_none());
+        assert!(doc.get("Targets").is_none());
+    }
+
+    /// An empty backlog must produce an empty body: madmin's decoder loop then
+    /// terminates on io.EOF with zero rows instead of one phantom row.
+    #[test]
+    fn mrf_stream_renders_empty_body_for_no_entries() {
+        let durable = DurableMrfBacklog {
+            available: true,
+            entries: Vec::new(),
+        };
+        let response = build_mrf_response("bucket-a".to_string(), &BucketStats::default(), &durable);
+
+        let (body, _) = render_mrf_backlog(&response, &durable, false).expect("stream body should serialize");
+        assert!(
+            body.is_empty(),
+            "empty backlog must serialize to an empty body, got: {}",
+            String::from_utf8_lossy(&body)
+        );
+    }
+
+    /// `?aggregate=true` (RustFS extension) keeps the enveloped counter shape.
+    #[test]
+    fn mrf_aggregate_envelope_retains_counters() {
+        let durable = sample_durable_backlog();
+        let response = build_mrf_response("bucket-a".to_string(), &BucketStats::default(), &durable);
+
+        let (body, _) = render_mrf_backlog(&response, &durable, true).expect("aggregate body should serialize");
+        let json: serde_json::Value = serde_json::from_slice(&body).expect("aggregate body should be one JSON object");
+        assert_eq!(json["Bucket"], "bucket-a");
+        assert_eq!(json["DurableCount"], 1);
+        assert_eq!(json["DurableBacklogAvailable"], true);
+        // The bare stream is an enumerable per-object API, so the aggregate
+        // shell now truthfully advertises it whenever the backlog is readable.
+        assert_eq!(json["PerObjectEntriesAvailable"], true);
+    }
+
+    /// The nil UUID is RustFS's in-memory null-version sentinel; the wire
+    /// token is `null`, never the zero UUID (second review round).
+    #[test]
+    fn mrf_stream_maps_nil_version_to_null_token() {
+        let durable = DurableMrfBacklog {
+            available: true,
+            entries: vec![MrfReplicateEntry {
+                bucket: "bucket-a".to_string(),
+                object: "null-version-object".to_string(),
+                version_id: Some(uuid::Uuid::nil()),
+                retry_count: 1,
+                size: 10,
+                op: MrfOpKind::Object,
+                ..Default::default()
+            }],
+        };
+        let response = build_mrf_response("bucket-a".to_string(), &BucketStats::default(), &durable);
+
+        let (body, truncated) = render_mrf_backlog(&response, &durable, false).expect("stream body should serialize");
+        assert!(!truncated);
+        let doc: serde_json::Value =
+            serde_json::from_str(String::from_utf8(body).expect("utf-8").lines().next().expect("one line"))
+                .expect("line should be a JSON document");
+        assert_eq!(doc["versionId"], "null");
+    }
+
+    /// The durable ledger is not bounded by the in-memory pending cap; the
+    /// stream must stop at the documented bound and signal truncation
+    /// (second review round).
+    #[test]
+    fn mrf_stream_truncates_at_the_documented_bound() {
+        let entries = (0..super::REPLICATION_MRF_MAX_STREAM_ENTRIES + 1)
+            .map(|index| MrfReplicateEntry {
+                bucket: "bucket-a".to_string(),
+                object: format!("object-{index}"),
+                retry_count: 1,
+                op: MrfOpKind::Object,
+                ..Default::default()
+            })
+            .collect();
+        let durable = DurableMrfBacklog {
+            available: true,
+            entries,
+        };
+        let response = build_mrf_response("bucket-a".to_string(), &BucketStats::default(), &durable);
+
+        let (body, truncated) = render_mrf_backlog(&response, &durable, false).expect("stream body should serialize");
+        assert!(truncated, "one entry past the bound must signal truncation");
+        assert_eq!(
+            String::from_utf8(body).expect("utf-8").lines().count(),
+            super::REPLICATION_MRF_MAX_STREAM_ENTRIES
+        );
+        let error = super::ensure_complete_mrf_stream(truncated).expect_err("partial streams must not return 200");
+        assert_eq!(error.code(), &s3s::S3ErrorCode::ServiceUnavailable);
+    }
+
    #[test]
    fn test_extract_query_params_decodes_percent_encoded_values() {
        let uri: Uri = "/rustfs/admin/v3/list-remote-targets?bucket=foo%2Fbar&flag=a+b"
@@ -29,6 +29,6 @@ impl Operation for RestLoadCredentialsHandler {
        let issuer = IamTableCredentialIssuer::from_request(&req)?;
        let response =
            load_credentials_response(&store, &warehouse, &namespace, &table, &issuer, Some(&principal.credentials)).await?;
-        build_json_response(StatusCode::OK, &response)
+        build_sensitive_json_response(StatusCode::OK, &response)
    }
 }
--- a/Show More
+++ b/Show More