Merge branch 'main' into cxymds/fix-1935-size-resolution-unknown-run2

Signed-off-by: houseme <housemecn@gmail.com>

.config/e2e-full-selection.txt

@@ -1,2 +1,2 @@
-sha256-darwin=b8549d3362a69cca01c2a81f548bb06d5142d8a9ab4509487a656c8b3db1c164
-sha256-linux=7ecd054965b4afa070af6deefdc37b5ca9f6a9b488dd5eef1ad0877378365b2f
+sha256-darwin=9f767b37ed8b1c82da62ea441462d75487785c8086e56f08fb6f6cd89c6e2e52
+sha256-linux=fbdaf42b220958d4b1e8880e0f8b5a7992d38e21051bb60596dd4538424757d6

.github/workflows/nightly-gnu.yml

@@ -39,11 +39,10 @@ jobs:
     env:
       FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
     steps:
-      - name: Checkout main branch
+      - name: Checkout repository
         uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
         with:
           persist-credentials: false
-          ref: main
 
       - name: Setup Rust environment
         uses: ./.github/actions/setup
@@ -89,11 +88,10 @@ jobs:
       # either casing.
       NO_PROXY: 127.0.0.1,localhost
     steps:
-      - name: Checkout main branch
+      - name: Checkout repository
         uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
         with:
           persist-credentials: false
-          ref: main
 
       - name: Setup Rust environment
         uses: ./.github/actions/setup
@@ -178,11 +176,10 @@ jobs:
       FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
       NO_PROXY: 127.0.0.1,localhost
     steps:
-      - name: Checkout main branch
+      - name: Checkout repository
         uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7
         with:
           persist-credentials: false
-          ref: main
 
       - name: Setup Rust environment
         uses: ./.github/actions/setup

CLAUDE.md

@@ -30,7 +30,8 @@ make build-docker BUILD_OS=ubuntu22.04
 - Crate membership: `Cargo.toml` `[workspace].members`
 - Architecture, layering, crate map: [ARCHITECTURE.md](ARCHITECTURE.md)
 - Migration guardrails & readiness contracts: [docs/architecture/](docs/architecture/README.md)
-- CI gates: `.github/workflows/ci.yml` (source of truth; never copy its steps into docs)
+- CI workflow steps: `.github/workflows/`; event, timeout, and required-status
+  matrix: [docs/testing/ci-gates.md](docs/testing/ci-gates.md)
 - Test-layer taxonomy, per-layer entry commands, serial/nextest rules, flake
   policy: [docs/testing/README.md](docs/testing/README.md)
 - Tier/ILM transition debugging (xl.meta inspection, versionId tracing):

CONTRIBUTING.md

@@ -70,6 +70,8 @@ make pre-pr
 
 > For the full test-layer taxonomy (unit / ecstore black-box / e2e / s3s-e2e / S3 compatibility / chaos / fuzz / bench), each layer's entry command, the naming conventions the migration gate depends on, and the serial/nextest rules, see [docs/testing/README.md](docs/testing/README.md).
 
+> For the event, timeout, required-status, and local reproduction matrix, see [docs/testing/ci-gates.md](docs/testing/ci-gates.md).
+
 ### 🔒 Automated Pre-commit Hooks
 #### What `make pre-commit` and `make pre-pr` actually run
 

Cargo.lock

@@ -1858,9 +1858,9 @@ dependencies = [
 
 [[package]]
 name = "cc"
-version = "1.4.3"
+version = "1.4.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "509591b7bcd67f4ef775afad7662703b4935daaa6ec0e5605cfb1090b32a2b6d"
+checksum = "0ad534f4357a5264cce5019c989cf66a4f0dc4e0d1b1d15f8aacec0ff7360273"
 dependencies = [
  "find-msvc-tools",
  "jobserver",
@@ -2522,12 +2522,6 @@ dependencies = [
  "subtle",
 ]
 
-[[package]]
-name = "cty"
-version = "0.2.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b365fabc795046672053e29c954733ec3b05e4be654ab130fe8f1f94d7051f35"
-
 [[package]]
 name = "curve25519-dalek"
 version = "4.1.3"
@@ -5988,15 +5982,6 @@ version = "0.2.16"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b6d2cec3eae94f9f509c767b45932f1ada8350c4bdb85af2fcab4a3c14807981"
 
-[[package]]
-name = "libmimalloc-sys"
-version = "0.1.49"
-source = "git+https://github.com/xonatius/mimalloc_rust.git?rev=6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11#6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11"
-dependencies = [
- "cc",
- "cty",
-]
-
 [[package]]
 name = "libredox"
 version = "0.1.20"
@@ -6397,14 +6382,6 @@ dependencies = [
  "synstructure 0.13.2",
 ]
 
-[[package]]
-name = "mimalloc"
-version = "0.1.52"
-source = "git+https://github.com/xonatius/mimalloc_rust.git?rev=6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11#6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11"
-dependencies = [
- "libmimalloc-sys",
-]
-
 [[package]]
 name = "mime"
 version = "0.3.17"
@@ -9162,13 +9139,11 @@ dependencies = [
  "insta",
  "jiff",
  "libc",
- "libmimalloc-sys",
  "libsystemd",
  "matchit 0.9.2",
  "md-5 0.11.0",
  "metrics",
  "metrics-util",
- "mimalloc",
  "mime_guess",
  "opentelemetry",
  "opentelemetry_sdk",
@@ -9204,6 +9179,8 @@ dependencies = [
  "rustfs-lock",
  "rustfs-log-analyzer",
  "rustfs-madmin",
+ "rustfs-mimalloc",
+ "rustfs-mimalloc-sys",
  "rustfs-notify",
  "rustfs-object-capacity",
  "rustfs-object-data-cache",
@@ -9875,6 +9852,24 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "rustfs-mimalloc"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a406f4aa07084301d485beec873af6dccc8e3f8762da244743df92038b1db1a6"
+dependencies = [
+ "rustfs-mimalloc-sys",
+]
+
+[[package]]
+name = "rustfs-mimalloc-sys"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3051b819175f58445d4c369a72f0ab88149f3885ba8bea2aff3be01f53fe7cd"
+dependencies = [
+ "cc",
+]
+
 [[package]]
 name = "rustfs-notify"
 version = "1.0.0-rc.3"

Cargo.toml

@@ -350,8 +350,8 @@ russh-sftp = "2.4.0"
 dav-server = "0.11.0"
 
 # Performance Analysis and Memory Profiling
-mimalloc = { version = "0.1.52", git = "https://github.com/xonatius/mimalloc_rust.git", rev = "6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11" }
-libmimalloc-sys = { version = "0.1.49", git = "https://github.com/xonatius/mimalloc_rust.git", rev = "6d4c41bb10c6d9da1d1b6f07b38c4cc051667f11", features = ["extended"] }
+rustfs-mimalloc = { version = "0.5.0" }
+rustfs-mimalloc-sys = { version = "0.5.0" }
 hotpath = { version = "0.23.3", default-features = false }
 # Snapshot testing for output format regression detection
 insta = { version = "1.48" }

crates/data-usage/src/data_usage.rs

@@ -307,6 +307,38 @@ pub struct DiskUsageStatus {
     pub snapshot_exists: bool,
 }
 
+/// A bounded reconciliation record for an object whose logical size could not
+/// be trusted at the scanner boundary. The scanner persists these records in
+/// its cache; keeping the model here avoids a second, incompatible accounting
+/// representation in storage-facing crates.
+#[derive(Debug, Default, Clone, Serialize, Deserialize, PartialEq, Eq)]
+pub struct SizeReconciliationEntry {
+    /// Stable object/version identity key (not a metrics label).
+    pub key: String,
+    pub bucket: String,
+    pub object: String,
+    #[serde(default)]
+    pub version_id: Option<String>,
+    #[serde(default)]
+    pub generation: Option<String>,
+    /// Structured reason label; raw metadata values must never be stored here.
+    pub reason: String,
+    #[serde(default)]
+    pub physical_size: Option<u64>,
+    #[serde(default)]
+    pub first_seen: u64,
+    #[serde(default)]
+    pub attempts: u32,
+}
+
+/// Object scope refreshed by one scanner pass. Existing debts in this scope
+/// are removed before the pass's unresolved records are inserted.
+#[derive(Debug, Default, Clone, PartialEq, Eq)]
+pub struct SizeReconciliationScope {
+    pub bucket: String,
+    pub object: String,
+}
+
 /// Size summary for a single object or group of objects
 #[derive(Debug, Default, Clone)]
 pub struct SizeSummary {
@@ -336,6 +368,16 @@ pub struct SizeSummary {
     pub repl_target_stats: HashMap<String, ReplTargetSizeSummary>,
     /// Per-tier accounting, keyed by storage class or remote tier name
     pub tier_stats: HashMap<String, TierStats>,
+    /// Size-resolution debts observed while scanning this summary.
+    pub size_reconciliation: Vec<SizeReconciliationEntry>,
+    /// True when the per-object summary exceeded its bounded debt buffer.
+    /// Callers must retain prior ledger entries rather than treating the
+    /// partial list as a complete refresh.
+    pub size_reconciliation_truncated: bool,
+    /// Object scopes refreshed by this summary. They let the durable ledger
+    /// remove versions that resolved without allocating one key per healthy
+    /// version on the hot path.
+    pub reconciliation_scopes: Vec<SizeReconciliationScope>,
 }
 
 /// Replication target size summary
@@ -833,7 +875,8 @@ impl DataUsageEntry {
 ///
 /// The canonical wire format is written by the hand-written map-encoded
 /// `Serialize` on the scanner-side `DataUsageCacheInfo`
-/// (`crates/scanner/src/data_usage_define.rs`), which carries 16 fields.
+/// (`crates/scanner/src/data_usage_define.rs`), which carries the original 16
+/// fields plus an optional reconciliation field.
 /// This type decodes only the shared subset and is deliberately not
 /// `Serialize`: a derived (array) encoding of this 6-field subset would
 /// corrupt the cache for scanner readers, so no write path may exist here.
@@ -1777,6 +1820,51 @@ impl SizeSummary {
             entry.pending_count = entry.pending_count.saturating_add(stats.pending_count);
             entry.failed_count = entry.failed_count.saturating_add(stats.failed_count);
         }
+
+        for entry in &other.size_reconciliation {
+            self.record_size_reconciliation(entry.clone());
+        }
+        self.size_reconciliation_truncated |= other.size_reconciliation_truncated;
+        for scope in &other.reconciliation_scopes {
+            self.record_reconciliation_scope(&scope.bucket, &scope.object);
+        }
+    }
+
+    /// Add one reconciliation debt, coalescing repeated observations in the
+    /// same object summary. The scanner cache applies its own larger bound.
+    pub fn record_size_reconciliation(&mut self, entry: SizeReconciliationEntry) {
+        const MAX_SUMMARY_RECONCILIATION_ENTRIES: usize = 1024;
+        if let Some(existing) = self.size_reconciliation.iter_mut().find(|value| value.key == entry.key) {
+            existing.reason = entry.reason;
+            existing.physical_size = entry.physical_size;
+            existing.generation = entry.generation;
+            existing.version_id = entry.version_id;
+            return;
+        }
+        if self.size_reconciliation.len() < MAX_SUMMARY_RECONCILIATION_ENTRIES {
+            self.size_reconciliation.push(entry);
+        } else {
+            self.size_reconciliation_truncated = true;
+        }
+    }
+
+    /// Mark one object scope as refreshed. Duplicate scopes are suppressed so
+    /// merging summaries remains bounded and deterministic.
+    pub fn record_reconciliation_scope(&mut self, bucket: &str, object: &str) {
+        if !self
+            .reconciliation_scopes
+            .iter()
+            .any(|scope| scope.bucket == bucket && scope.object == object)
+        {
+            if self.reconciliation_scopes.len() >= 1024 {
+                self.size_reconciliation_truncated = true;
+                return;
+            }
+            self.reconciliation_scopes.push(SizeReconciliationScope {
+                bucket: bucket.to_string(),
+                object: object.to_string(),
+            });
+        }
     }
 }
 

crates/e2e_test/src/security_boundary_test.rs

@@ -17,14 +17,14 @@
 //! These tests verify that RustFS properly enforces security-sensitive
 //! controls by issuing real requests against a running server and asserting
 //! the concrete outcome of each control:
-//! - DoS protection (oversized tagging payloads, out-of-range multipart part numbers)
+//! - DoS protection (oversized tagging payloads, excessive multipart parts)
 //! - SSRF prevention (internal/private endpoints rejected for tiering)
 //! - Race condition handling (concurrent writes converge without corruption)
 
 use crate::common::{RustFSTestEnvironment, awscurl_available, awscurl_put, init_logging};
 use aws_sdk_s3::error::ProvideErrorMetadata;
 use aws_sdk_s3::primitives::ByteStream;
-use aws_sdk_s3::types::{Tag, Tagging};
+use aws_sdk_s3::types::{CompletedMultipartUpload, CompletedPart, Tag, Tagging};
 use std::error::Error;
 use tracing::info;
 
@@ -88,9 +88,9 @@ async fn test_large_xml_body_rejection() -> Result<(), Box<dyn Error + Send + Sy
     Ok(())
 }
 
-/// Multipart part numbers above the S3 limit must be rejected.
+/// Excessive multipart parts must be rejected.
 #[tokio::test]
-async fn test_multipart_part_number_above_limit() -> Result<(), Box<dyn Error + Send + Sync>> {
+async fn test_excessive_multipart_parts() -> Result<(), Box<dyn Error + Send + Sync>> {
     init_logging();
     let mut env = RustFSTestEnvironment::new().await?;
     env.start_rustfs_server(vec![]).await?;
@@ -108,23 +108,18 @@ async fn test_multipart_part_number_above_limit() -> Result<(), Box<dyn Error +
 
     let upload_id = create_result.upload_id().expect("upload_id should be present").to_string();
 
-    client
-        .upload_part()
-        .bucket(&bucket_name)
-        .key("test-large")
-        .upload_id(&upload_id)
-        .part_number(10000)
-        .body(ByteStream::from_static(b"upper-bound part"))
-        .send()
-        .await?;
+    // Try to complete with too many parts (should be rejected).
+    let mut parts = Vec::new();
+    for i in 1..=10001 {
+        parts.push(CompletedPart::builder().part_number(i).e_tag(format!("etag-{i}")).build());
+    }
 
     let result = client
-        .upload_part()
+        .complete_multipart_upload()
         .bucket(&bucket_name)
         .key("test-large")
         .upload_id(&upload_id)
-        .part_number(10001)
-        .body(ByteStream::from_static(b"out-of-range part"))
+        .multipart_upload(CompletedMultipartUpload::builder().set_parts(Some(parts)).build())
         .send()
         .await;
 
@@ -138,13 +133,7 @@ async fn test_multipart_part_number_above_limit() -> Result<(), Box<dyn Error +
         .await;
     let _ = client.delete_bucket().bucket(&bucket_name).send().await;
 
-    let err = result.expect_err("server must reject excessive multipart parts");
-    let code = err.as_service_error().and_then(ProvideErrorMetadata::code);
-    assert_eq!(
-        code,
-        Some("InvalidArgument"),
-        "Part number 10001 should be rejected with InvalidArgument, got code {code:?}, err: {err:?}"
-    );
+    assert!(result.is_err(), "Server should reject excessive multipart parts");
 
     env.stop_server();
     Ok(())
@@ -228,8 +217,12 @@ async fn test_concurrent_object_operations() -> Result<(), Box<dyn Error + Send
 /// Internal/private endpoints must be rejected as remote tier backends (SSRF).
 ///
 /// This issues a real admin AddTier call (`PUT /rustfs/admin/v3/tier`) for each
-/// internal/private endpoint and asserts the request reaches the outbound URL
-/// guard. Connectivity or credential failures do not prove SSRF protection.
+/// internal/private endpoint and asserts the server rejects it (non-2xx, so the
+/// signed request helper returns an error). An internal endpoint must never be
+/// accepted as a tier backend. The rejection may originate from explicit
+/// SSRF/internal-address filtering or from the backend connectivity/credential
+/// validation performed during AddTier; either way the security-relevant
+/// outcome — the internal endpoint is not accepted — is asserted here.
 ///
 /// The admin API is exercised via signed `awscurl` requests, matching the
 /// pattern used by the other admin-API E2E tests in this crate; the test is
@@ -270,13 +263,10 @@ async fn test_tiering_url_validation() -> Result<(), Box<dyn Error + Send + Sync
         })
         .to_string();
 
-        let err = awscurl_put(&tier_url, &body, &env.access_key, &env.secret_key)
-            .await
-            .expect_err("AddTier must reject internal endpoints");
-        let rendered = err.to_string();
+        let result = awscurl_put(&tier_url, &body, &env.access_key, &env.secret_key).await;
         assert!(
-            rendered.contains("TierAddFailed") && rendered.contains("tier endpoint is not allowed"),
-            "AddTier rejected {endpoint} outside the outbound URL guard: {rendered}"
+            result.is_err(),
+            "AddTier must reject internal endpoint {endpoint}, but it was accepted: {result:?}"
         );
     }
 

crates/ecstore/src/io_support/bitrot.rs

@@ -784,6 +784,24 @@ pub(crate) fn create_deferred_bitrot_reader_with_stripe_handle(
 ///
 /// # Returns
 /// A Result containing the BitrotWriterWrapper or an error
+/// Size hint handed to `DiskAPI::create_file` for a bitrot-wrapped shard.
+///
+/// A known length is grown by one checksum per shard so the on-disk file size
+/// matches what the bitrot writer emits. A negative length is the
+/// unknown-size sentinel (`HashReader::SIZE_PRESERVE_LAYER`, used by SSE and
+/// compression) and must be preserved: `RemoteDisk::create_file` forwards it
+/// in the `put_file_stream` query, and the receiver only treats `size > 0` as
+/// a fixed body length when locating the authenticated trailer. Clamping it
+/// to `0` would claim an empty body and misframe the stream. `0` stays `0`
+/// because a genuinely empty object still means an empty body.
+fn bitrot_create_file_size(length: i64, shard_size: usize, checksum_algo: &HashAlgorithm) -> i64 {
+    if length <= 0 {
+        return length;
+    }
+    let length = length as usize;
+    (length.div_ceil(shard_size) * checksum_algo.size() + length) as i64
+}
+
 pub async fn create_bitrot_writer(
     is_inline_buffer: bool,
     disk: Option<&DiskStore>,
@@ -796,12 +814,7 @@ pub async fn create_bitrot_writer(
     let writer = if is_inline_buffer {
         CustomWriter::new_inline_buffer()
     } else if let Some(disk) = disk {
-        let length = if length > 0 {
-            let length = length as usize;
-            (length.div_ceil(shard_size) * checksum_algo.size() + length) as i64
-        } else {
-            0
-        };
+        let length = bitrot_create_file_size(length, shard_size, &checksum_algo);
 
         let file = disk.create_file("", volume, path, length).await?;
         #[cfg(feature = "hotpath")]
@@ -820,6 +833,25 @@ mod tests {
     use rustfs_rio::ChunkReader;
     use std::collections::VecDeque;
 
+    #[test]
+    fn bitrot_create_file_size_grows_known_length_by_checksums() {
+        // 10 bytes over 4-byte shards = 3 shards, each followed by a 32-byte hash.
+        assert_eq!(bitrot_create_file_size(10, 4, &HashAlgorithm::HighwayHash256), 10 + 3 * 32);
+        assert_eq!(bitrot_create_file_size(10, 4, &HashAlgorithm::None), 10);
+    }
+
+    #[test]
+    fn bitrot_create_file_size_keeps_empty_and_unknown_distinct() {
+        assert_eq!(bitrot_create_file_size(0, 4, &HashAlgorithm::HighwayHash256), 0);
+        // SSE/compression streams advertise SIZE_PRESERVE_LAYER (-1); the remote
+        // put_file_stream receiver relies on a non-positive size to parse the auth
+        // trailer from the stream tail, so the sentinel must survive untouched.
+        assert_eq!(
+            bitrot_create_file_size(rustfs_rio::HashReader::SIZE_PRESERVE_LAYER, 4, &HashAlgorithm::HighwayHash256),
+            rustfs_rio::HashReader::SIZE_PRESERVE_LAYER
+        );
+    }
+
     struct TestChunkReader {
         chunks: VecDeque<Bytes>,
     }

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

@@ -2124,26 +2124,13 @@ impl SetDisks {
 
             let put_object_size = known_put_object_storage_size(data.size());
             let shard_file_size_raw = erasure.shard_file_size(put_object_size);
-            let is_inline_buffer =
-                storage_class_config.should_inline(shard_file_size_raw, erasure.data_shards, opts.versioned);
+            let is_inline_buffer = storage_class_config.should_inline(shard_file_size_raw, erasure.data_shards, opts.versioned);
 
             let collect_stage_timing = rustfs_io_metrics::put_stage_metrics_enabled() || issue3031_diag_enabled();
             let shard_file_size = shard_file_size_raw;
             let shard_size = erasure.shard_size();
             let write_path = classify_put_write_path(is_inline_buffer, put_object_size, fi.erasure.block_size);
             let direct_inline_commit = matches!(write_path, SmallWritePath::Inline);
-            {
-                use std::io::Write;
-                let msg = format!(
-                    "INLINE_DEBUG: bucket={} obj={} size={} shard_fs={} ds={} bs={} inline={} direct={} path={} iblock={} ver={}\n",
-                    bucket, object, put_object_size, shard_file_size_raw, erasure.data_shards, fi.erasure.block_size,
-                    is_inline_buffer, direct_inline_commit, write_path.metric_label(), storage_class_config.inline_block(), opts.versioned
-                );
-                if let Ok(mut f) = std::fs::OpenOptions::new().create(true).append(true).open("/tmp/rustfs_inline_debug.log") {
-                    let _ = f.write_all(msg.as_bytes());
-                }
-                let _ = std::io::stderr().write_all(msg.as_bytes());
-            }
             rustfs_io_metrics::record_put_object_path(write_path.metric_label());
             let writer_setup_stage_start = collect_stage_timing.then(Instant::now);
             let (mut writers, errors) = if direct_inline_commit {

crates/ecstore/src/store/object.rs

@@ -3194,7 +3194,7 @@ impl ECStore {
 
         // Default return value
         let mut del_objects = vec![DeletedObject::default(); objects.len()];
-        let mut accounting = vec![None; objects.len()];
+        let accounting = vec![None; objects.len()];
 
         let mut del_errs = Vec::with_capacity(objects.len());
         for _ in 0..objects.len() {

crates/ecstore/src/store/rebalance/support.rs

@@ -271,7 +271,7 @@ pub(super) fn resolve_latest_object_info_candidates(
             .filter(|candidate| latest_candidate_mod_time(candidate) == Some(latest_mod_time))
             .collect::<Vec<_>>();
 
-        latest_candidates.sort_by(|left, right| right.idx.cmp(&left.idx));
+        latest_candidates.sort_by_key(|candidate| std::cmp::Reverse(candidate.idx));
 
         let Some(winner) = latest_candidates.first() else {
             return Err(Error::ErasureReadQuorum);

crates/kms/tests/vault_ha_failover_live.rs

@@ -16,14 +16,14 @@
 //!
 //! `scripts/test/vault_ha_kms_live.sh` owns the official Vault containers and
 //! kills the active node while this test continuously decrypts through a
-//! surviving standby. KV2 and Transit requests must remain successful, use a
-//! bounded number of attempts, and leave the circuit and in-flight gauges at
-//! zero after a new leader is elected.
+//! surviving standby. KV2 and Transit must recover after the bounded circuit
+//! interval, use a bounded number of attempts, and leave the circuit and
+//! in-flight gauges at zero after a new leader is elected.
 
 use std::collections::HashMap;
 use std::path::{Path, PathBuf};
-use std::sync::Arc;
 use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
+use std::sync::{Arc, Mutex};
 use std::time::Duration;
 
 use metrics_util::MetricKind;
@@ -43,6 +43,11 @@ const OPERATION_ATTEMPTS: &str = "rustfs_kms_backend_operation_attempts";
 const IN_FLIGHT: &str = "rustfs_kms_backend_in_flight";
 const CIRCUIT_OPEN: &str = "rustfs_kms_backend_circuit_open";
 const MAX_ATTEMPTS: u32 = 10;
+const ATTEMPT_TIMEOUT: Duration = Duration::from_secs(2);
+const HEALTHY_PROGRESS_TIMEOUT: Duration = Duration::from_secs(20);
+// The circuit remains open for 30s after five failed attempts.
+const POST_FAILOVER_PROGRESS_TIMEOUT: Duration = Duration::from_secs(35);
+const FAILOVER_ERROR_POLL_INTERVAL: Duration = Duration::from_millis(100);
 
 type MetricEntry = (
     metrics_util::CompositeKey,
@@ -64,7 +69,7 @@ fn config(backend: KmsBackend, backend_config: BackendConfig) -> KmsConfig {
         backend,
         backend_config,
         allow_insecure_dev_defaults: true,
-        timeout: Duration::from_secs(2),
+        timeout: ATTEMPT_TIMEOUT,
         retry_attempts: MAX_ATTEMPTS,
         enable_cache: false,
         ..KmsConfig::default()
@@ -164,14 +169,31 @@ fn retryable_failures(snapshot: &[MetricEntry], operation: &str) -> u64 {
         .sum()
 }
 
-async fn wait_for_count(counter: &AtomicU64, minimum: u64, description: &str) {
-    tokio::time::timeout(Duration::from_secs(20), async {
+async fn wait_for_count(
+    counter: &AtomicU64,
+    failure: &Mutex<Option<String>>,
+    minimum: u64,
+    description: &str,
+    timeout: Duration,
+) {
+    tokio::time::timeout(timeout, async {
         while counter.load(Ordering::SeqCst) < minimum {
+            if let Some(error) = failure.lock().expect("decrypt failure lock poisoned").as_ref() {
+                panic!(
+                    "{description} worker failed after {} successful decrypts: {error}",
+                    counter.load(Ordering::SeqCst)
+                );
+            }
             tokio::time::sleep(Duration::from_millis(25)).await;
         }
     })
     .await
-    .unwrap_or_else(|_| panic!("timed out waiting for {description}"));
+    .unwrap_or_else(|_| {
+        panic!(
+            "timed out after {timeout:?} waiting for {description}: completed {}, expected {minimum}",
+            counter.load(Ordering::SeqCst)
+        )
+    });
 }
 
 async fn wait_for_file(path: &Path, description: &str) {
@@ -189,7 +211,8 @@ async fn decrypt_loop<B: KmsBackendTrait + Send + Sync + 'static>(
     request: DecryptRequest,
     expected: Vec<u8>,
     completed: Arc<AtomicU64>,
-    failed: Arc<AtomicBool>,
+    allow_failover_errors: Arc<AtomicBool>,
+    failure: Arc<Mutex<Option<String>>>,
     stop: CancellationToken,
 ) {
     while !stop.is_cancelled() {
@@ -197,8 +220,18 @@ async fn decrypt_loop<B: KmsBackendTrait + Send + Sync + 'static>(
             Ok(response) if response.plaintext == expected => {
                 completed.fetch_add(1, Ordering::SeqCst);
             }
-            Ok(_) | Err(_) => {
-                failed.store(true, Ordering::SeqCst);
+            Ok(_) => {
+                *failure.lock().expect("decrypt failure lock poisoned") =
+                    Some("decrypt returned unexpected plaintext".to_string());
+                return;
+            }
+            Err(rustfs_kms::KmsError::BackendError { .. } | rustfs_kms::KmsError::OperationTimedOut { .. })
+                if allow_failover_errors.load(Ordering::SeqCst) =>
+            {
+                tokio::time::sleep(FAILOVER_ERROR_POLL_INTERVAL).await;
+            }
+            Err(error) => {
+                *failure.lock().expect("decrypt failure lock poisoned") = Some(error.to_string());
                 return;
             }
         }
@@ -296,7 +329,9 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
     );
 
     let stop = CancellationToken::new();
-    let failed = Arc::new(AtomicBool::new(false));
+    let allow_failover_errors = Arc::new(AtomicBool::new(false));
+    let kv2_failure = Arc::new(Mutex::new(None));
+    let transit_failure = Arc::new(Mutex::new(None));
     let kv2_completed = Arc::new(AtomicU64::new(0));
     let transit_completed = Arc::new(AtomicU64::new(0));
     let kv2_worker = tokio::spawn(decrypt_loop(
@@ -304,7 +339,8 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
         kv2_request,
         kv2_data_key.plaintext_key,
         Arc::clone(&kv2_completed),
-        Arc::clone(&failed),
+        Arc::clone(&allow_failover_errors),
+        Arc::clone(&kv2_failure),
         stop.clone(),
     ));
     let transit_worker = tokio::spawn(decrypt_loop(
@@ -312,12 +348,21 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
         transit_request,
         transit_data_key.plaintext_key,
         Arc::clone(&transit_completed),
-        Arc::clone(&failed),
+        Arc::clone(&allow_failover_errors),
+        Arc::clone(&transit_failure),
         stop.clone(),
     ));
 
-    wait_for_count(&kv2_completed, 2, "two healthy KV2 decrypts").await;
-    wait_for_count(&transit_completed, 2, "two healthy Transit decrypts").await;
+    wait_for_count(&kv2_completed, &kv2_failure, 2, "two healthy KV2 decrypts", HEALTHY_PROGRESS_TIMEOUT).await;
+    wait_for_count(
+        &transit_completed,
+        &transit_failure,
+        2,
+        "two healthy Transit decrypts",
+        HEALTHY_PROGRESS_TIMEOUT,
+    )
+    .await;
+    allow_failover_errors.store(true, Ordering::SeqCst);
     std::fs::write(&marker, b"ready").expect("publish failover readiness marker");
 
     wait_for_file(&elected, "the replacement Vault leader").await;
@@ -326,18 +371,39 @@ async fn exercise_failover(snapshotter: &Snapshotter) {
 
     let kv2_after_election = kv2_completed.load(Ordering::SeqCst) + 2;
     let transit_after_election = transit_completed.load(Ordering::SeqCst) + 2;
-    wait_for_count(&kv2_completed, kv2_after_election, "post-failover KV2 decrypts").await;
-    wait_for_count(&transit_completed, transit_after_election, "post-failover Transit decrypts").await;
+    wait_for_count(
+        &kv2_completed,
+        &kv2_failure,
+        kv2_after_election,
+        "post-failover KV2 decrypts",
+        POST_FAILOVER_PROGRESS_TIMEOUT,
+    )
+    .await;
+    wait_for_count(
+        &transit_completed,
+        &transit_failure,
+        transit_after_election,
+        "post-failover Transit decrypts",
+        POST_FAILOVER_PROGRESS_TIMEOUT,
+    )
+    .await;
 
     stop.cancel();
     kv2_worker.await.expect("KV2 decrypt worker must join");
     transit_worker.await.expect("Transit decrypt worker must join");
-    assert!(!failed.load(Ordering::SeqCst), "no decrypt may fail or return different plaintext");
+    assert!(
+        kv2_failure.lock().expect("KV2 failure lock poisoned").is_none(),
+        "no KV2 decrypt may fail or return different plaintext"
+    );
+    assert!(
+        transit_failure.lock().expect("Transit failure lock poisoned").is_none(),
+        "no Transit decrypt may fail or return different plaintext"
+    );
 }
 
 #[test]
 #[ignore = "requires a real three-node Vault Raft cluster; run scripts/test/vault_ha_kms_live.sh"]
-fn vault_raft_leader_failure_preserves_kv2_and_transit_decrypts() {
+fn vault_raft_leader_failure_recovers_kv2_and_transit_decrypts() {
     let recorder = DebuggingRecorder::new();
     let snapshotter = recorder.snapshotter();
     metrics::with_local_recorder(&recorder, || {
@@ -349,11 +415,6 @@ fn vault_raft_leader_failure_preserves_kv2_and_transit_decrypts() {
     });
     let snapshot = snapshotter.snapshot().into_vec();
 
-    assert_eq!(
-        counter_value(&snapshot, OPERATIONS_TOTAL, &[("outcome", "circuit_open")]),
-        0,
-        "a bounded leader election must not open the circuit"
-    );
     assert_eq!(
         counter_value(&snapshot, OPERATIONS_TOTAL, &[("outcome", "budget_exhausted")]),
         0,

crates/lifecycle/src/core.rs

@@ -54,6 +54,7 @@ const ERR_LIFECYCLE_INVALID_EXPIRED_OBJECT_ALL_VERSIONS: &str =
     "Days must be a positive integer and Date must not be specified inside Expiration with ExpiredObjectAllVersions";
 const ERR_LIFECYCLE_INVALID_DEL_MARKER_EXPIRATION_DAYS: &str = "Days must be a positive integer with DelMarkerExpiration";
 const ERR_LIFECYCLE_INVALID_RULE_ID_TOO_LONG: &str = "Rule ID must be at most 255 characters";
+const ERR_LIFECYCLE_INVALID_RULE_ID_EMPTY: &str = "Rule ID must not be empty";
 const ERR_LIFECYCLE_INVALID_RULE_STATUS: &str = "Rule status must be either Enabled or Disabled";
 const ERR_LIFECYCLE_DEL_MARKER_WITH_TAGS: &str = "Rule with DelMarkerExpiration cannot have tags based filtering";
 const ERR_LIFECYCLE_EXPIRED_OBJECT_DELETE_MARKER_WITH_TAGS: &str =
@@ -402,10 +403,13 @@ impl Lifecycle for BucketLifecycleConfiguration {
                     NoncurrentVersionTransitionOps::validate(transition)?;
                 }
             }
-            if let Some(id) = &r.id
-                && id.len() > 255
-            {
-                return Err(std::io::Error::other(ERR_LIFECYCLE_INVALID_RULE_ID_TOO_LONG));
+            if let Some(id) = &r.id {
+                if id.is_empty() {
+                    return Err(std::io::Error::other(ERR_LIFECYCLE_INVALID_RULE_ID_EMPTY));
+                }
+                if id.len() > 255 {
+                    return Err(std::io::Error::other(ERR_LIFECYCLE_INVALID_RULE_ID_TOO_LONG));
+                }
             }
             r.validate()?;
             if let Some(object_lock_enabled) = lr.object_lock_enabled.as_ref()
@@ -3730,6 +3734,31 @@ mod tests {
             .expect("empty prefix with filter should be valid");
     }
 
+    #[tokio::test]
+    async fn validate_rejects_empty_rule_id() {
+        let lc = BucketLifecycleConfiguration {
+            expiry_updated_at: None,
+            rules: vec![LifecycleRule {
+                status: ExpirationStatus::from_static(ExpirationStatus::ENABLED),
+                expiration: Some(LifecycleExpiration {
+                    days: Some(30),
+                    ..Default::default()
+                }),
+                abort_incomplete_multipart_upload: None,
+                del_marker_expiration: None,
+                filter: None,
+                id: Some(String::new()),
+                noncurrent_version_expiration: None,
+                noncurrent_version_transitions: None,
+                prefix: None,
+                transitions: None,
+            }],
+        };
+
+        let error = lc.validate(&ObjectLockConfiguration::default()).await.unwrap_err();
+        assert_eq!(error.to_string(), ERR_LIFECYCLE_INVALID_RULE_ID_EMPTY);
+    }
+
     // --- TASK-004 tests: ExpiredObjectAllVersions ---
 
     #[tokio::test]

crates/scanner/src/data_usage_define.rs

@@ -29,7 +29,8 @@ use rustfs_config::ENV_SCANNER_CACHE_SAVE_TIMEOUT_SECS;
 pub use rustfs_data_usage::{
     AllTierStats, BucketTargetUsageInfo, BucketUsageInfo, DATA_USAGE_OBJECT_NAME, DATA_USAGE_OBSERVED_OBJECT_NAME,
     DataUsageEntry, DataUsageHash, DataUsageHashMap, DataUsageInfo, LEGACY_DATA_USAGE_OBJECT_NAME, PrefixUsageEntry,
-    PrefixUsageQuery, PrefixUsageSummary, ReplTargetSizeSummary, SizeSummary, TierStats, hash_path, prefix_usage_in_cache,
+    PrefixUsageQuery, PrefixUsageSummary, ReplTargetSizeSummary, SizeReconciliationEntry, SizeReconciliationScope, SizeSummary,
+    TierStats, hash_path, prefix_usage_in_cache,
 };
 use rustfs_utils::path::{SLASH_SEPARATOR, path_join_buf};
 use tokio::time::{Duration, Instant, sleep, timeout};
@@ -192,6 +193,10 @@ const MAX_DATA_USAGE_CACHE_DEPTH: usize = 1024;
 pub trait ScannerSizeSummaryExt {
     /// Fold one object's contribution into the summary, including its tier.
     fn actions_accounting(&mut self, oi: &ObjectInfo, size: i64, actual_size: i64);
+    /// Fold counters and physical tier usage for an object whose metadata is
+    /// valid but whose logical size is currently unavailable. Logical totals
+    /// stay unchanged.
+    fn actions_accounting_unknown(&mut self, oi: &ObjectInfo);
 }
 
 impl ScannerSizeSummaryExt for SizeSummary {
@@ -225,6 +230,34 @@ impl ScannerSizeSummaryExt for SizeSummary {
             });
         }
     }
+
+    fn actions_accounting_unknown(&mut self, oi: &ObjectInfo) {
+        if oi.delete_marker {
+            self.delete_markers = self.delete_markers.saturating_add(1);
+            return;
+        }
+
+        if oi.version_id.is_some_and(|v| !v.is_nil()) {
+            self.versions = self.versions.saturating_add(1);
+        }
+
+        if oi.transitioned_object.free_version {
+            return;
+        }
+
+        let tier = if oi.transitioned_object.status == TRANSITION_COMPLETE {
+            oi.transitioned_object.tier.clone()
+        } else {
+            oi.storage_class.clone().unwrap_or_else(|| storageclass::STANDARD.to_string())
+        };
+        if let Some(tier_stats) = self.tier_stats.get_mut(&tier) {
+            *tier_stats = tier_stats.add(&TierStats {
+                total_size: u64::try_from(oi.size).unwrap_or(0),
+                num_versions: 1,
+                num_objects: u64::from(oi.is_latest),
+            });
+        }
+    }
 }
 
 // ===== Cache-related data structures =====
@@ -344,6 +377,10 @@ pub struct DataUsageCacheInfo {
     pub scan_plan_digest: Option<DataUsageScanPlanDigest>,
     #[serde(default)]
     pub cache_key_format: u16,
+    /// Bounded durable debts for versions whose logical size was not trusted.
+    /// The map key is an identity key, never a user-controlled metric label.
+    #[serde(default)]
+    pub size_reconciliation: HashMap<String, SizeReconciliationEntry>,
 }
 
 impl Serialize for DataUsageCacheInfo {
@@ -353,7 +390,8 @@ impl Serialize for DataUsageCacheInfo {
     {
         // Keep this metadata map-encoded so older readers can ignore fields
         // appended by newer scanner versions during rolling upgrades.
-        let mut state = serializer.serialize_map(Some(16))?;
+        let field_count = 16 + usize::from(!self.size_reconciliation.is_empty());
+        let mut state = serializer.serialize_map(Some(field_count))?;
         state.serialize_entry("name", &self.name)?;
         state.serialize_entry("next_cycle", &self.next_cycle)?;
         state.serialize_entry("leader_epoch", &self.leader_epoch)?;
@@ -370,6 +408,9 @@ impl Serialize for DataUsageCacheInfo {
         state.serialize_entry("snapshot_complete", &self.snapshot_complete)?;
         state.serialize_entry("scan_plan_digest", &self.scan_plan_digest)?;
         state.serialize_entry("cache_key_format", &self.cache_key_format)?;
+        if !self.size_reconciliation.is_empty() {
+            state.serialize_entry("size_reconciliation", &self.size_reconciliation)?;
+        }
         state.end()
     }
 }
@@ -428,14 +469,18 @@ impl DataUsageCache {
                     self.checked_flatten(name).is_some()
                 });
         if !reusable {
-            let pending_heals = if self.info.name == name {
-                std::mem::take(&mut self.info.pending_heals)
+            let (pending_heals, size_reconciliation) = if self.info.name == name {
+                (
+                    std::mem::take(&mut self.info.pending_heals),
+                    std::mem::take(&mut self.info.size_reconciliation),
+                )
             } else {
-                Vec::new()
+                (Vec::new(), HashMap::new())
             };
             *self = Self::default();
             self.info.name = name.to_string();
             self.info.pending_heals = pending_heals;
+            self.info.size_reconciliation = size_reconciliation;
         }
 
         self.info.next_cycle = next_cycle;

crates/scanner/src/data_usage_define/tests.rs

@@ -673,6 +673,34 @@ fn size_summary_actions_accounting_accumulates_tier_stats() {
     );
 }
 
+#[test]
+fn size_summary_unknown_accounting_keeps_physical_tier_and_version_only() {
+    let mut summary = SizeSummary::default();
+    summary
+        .tier_stats
+        .insert(storageclass::STANDARD.to_string(), TierStats::default());
+    let object = ObjectInfo {
+        size: 12,
+        storage_class: Some(storageclass::STANDARD.to_string()),
+        version_id: Some(uuid::Uuid::new_v4()),
+        is_latest: true,
+        ..Default::default()
+    };
+
+    summary.actions_accounting_unknown(&object);
+
+    assert_eq!(summary.total_size, 0, "unknown logical size must not become zero or physical bytes");
+    assert_eq!(summary.versions, 1);
+    assert_eq!(
+        summary.tier_stats.get(storageclass::STANDARD),
+        Some(&TierStats {
+            total_size: 12,
+            num_versions: 1,
+            num_objects: 1,
+        })
+    );
+}
+
 #[test]
 fn test_data_usage_entry_merge_sums_failed_objects() {
     let mut left = DataUsageEntry {
@@ -1079,6 +1107,16 @@ fn data_usage_cache_prepare_for_scan_preserves_pending_heal_only_progress() {
             scan_plan_digest: Some(TEST_PLAN_DIGEST),
             cache_key_format: DATA_USAGE_CACHE_KEY_FORMAT,
             pending_heals: vec![pending_heal.clone()],
+            size_reconciliation: HashMap::from([(
+                "size-key".to_string(),
+                SizeReconciliationEntry {
+                    key: "size-key".to_string(),
+                    bucket: "bucket".to_string(),
+                    object: "prefix/object".to_string(),
+                    reason: "invalid_declared_size".to_string(),
+                    ..Default::default()
+                },
+            )]),
             ..Default::default()
         },
         ..Default::default()
@@ -1088,6 +1126,7 @@ fn data_usage_cache_prepare_for_scan_preserves_pending_heal_only_progress() {
 
     assert_eq!(outcome, DataUsageCachePrepareOutcome::Reused);
     assert_eq!(cache.info.pending_heals, vec![pending_heal]);
+    assert!(cache.info.size_reconciliation.contains_key("size-key"));
     assert!(cache.cache.is_empty());
     assert!(!cache.info.snapshot_complete);
 }

crates/scanner/src/scanner_folder.rs

@@ -20,8 +20,9 @@ use std::time::{Duration, Instant, SystemTime};
 
 use crate::ReplTargetSizeSummary;
 use crate::data_usage_define::{
-    DATA_USAGE_SCAN_CHECKPOINT_VERSION, DataUsageCache, DataUsageEntry, DataUsageHash, DataUsageHashMap, DataUsageScanCheckpoint,
-    DataUsageScanCheckpointReason, PendingScannerHeal, PendingScannerHealKind, ScannerSizeSummaryExt, SizeSummary, hash_path,
+    DATA_USAGE_SCAN_CHECKPOINT_VERSION, DataUsageCache, DataUsageCacheInfo, DataUsageEntry, DataUsageHash, DataUsageHashMap,
+    DataUsageScanCheckpoint, DataUsageScanCheckpointReason, PendingScannerHeal, PendingScannerHealKind, ScannerSizeSummaryExt,
+    SizeReconciliationEntry, SizeSummary, hash_path,
 };
 use crate::error::ScannerError;
 use crate::runtime_config::{
@@ -97,6 +98,9 @@ const METRIC_SCANNER_EXCESS_FOLDERS_TOTAL: &str = "rustfs_scanner_excess_folders
 const METRIC_SCANNER_PENDING_HEAL_PRUNE_TOTAL: &str = "rustfs_scanner_pending_heal_prune_total";
 const METRIC_SCANNER_PENDING_HEAL_MALFORMED_TOTAL: &str = "rustfs_scanner_pending_heal_malformed_total";
 const MAX_PENDING_SCANNER_HEAL_RETRIES_PER_BUCKET: usize = 128;
+const MAX_SIZE_RECONCILIATION_ENTRIES_PER_BUCKET: usize = 10_000;
+const MAX_SIZE_RECONCILIATION_BYTES_PER_BUCKET: usize = 8 * 1024 * 1024;
+const MAX_SIZE_RECONCILIATION_AGE_SECS: u64 = 7 * 24 * 60 * 60;
 
 // --- scanner excess alerts as S3 notification events (rustfs/backlog#1868) --
 //
@@ -364,7 +368,7 @@ impl PendingScannerAccounting<'_> {
     fn apply(self, size_summary: &mut SizeSummary, cumulative_size: &mut i64, queued: bool) {
         let size = if queued { self.expired_size } else { self.retained_size };
         size_summary.actions_accounting(self.object, size, self.retained_size);
-        *cumulative_size += size;
+        *cumulative_size = cumulative_size.saturating_add(size);
     }
 }
 
@@ -671,10 +675,65 @@ pub struct FolderScanner {
     skip_heal: Arc<std::sync::atomic::AtomicBool>,
     local_disk: Arc<Disk>,
     pending_heals_changed: bool,
+    pending_size_reconciliation_keys: HashSet<String>,
+    pending_size_reconciliation_scopes: HashSet<String>,
+    pending_size_reconciliation_truncated: bool,
     #[cfg(test)]
     list_path_raw_options_observer: Option<mpsc::UnboundedSender<ListPathRawTimeoutSnapshot>>,
 }
 
+fn size_reconciliation_entry_bytes(entry: &SizeReconciliationEntry) -> usize {
+    entry.key.len()
+        + entry.bucket.len()
+        + entry.object.len()
+        + entry.version_id.as_deref().map_or(0, str::len)
+        + entry.generation.as_deref().map_or(0, str::len)
+        + entry.reason.len()
+        + std::mem::size_of::<u64>()
+        + std::mem::size_of::<u32>()
+}
+
+fn size_reconciliation_scope_key(bucket: &str, object: &str) -> String {
+    format!("{}:{}|{}:{}", bucket.len(), bucket, object.len(), object)
+}
+
+fn prune_size_reconciliation(info: &mut DataUsageCacheInfo, now: u64) {
+    info.size_reconciliation.retain(|key, entry| {
+        if entry.first_seen == 0 || entry.first_seen > now {
+            entry.first_seen = now;
+        }
+        key == &entry.key
+            && entry.key.len() <= 4096
+            && entry.bucket.len() <= 512
+            && entry.object.len() <= 512
+            && entry.version_id.as_deref().is_none_or(|value| value.len() <= 64)
+            && entry.generation.as_deref().is_none_or(|value| value.len() <= 64)
+            && entry.reason.len() <= 64
+            && now.saturating_sub(entry.first_seen) <= MAX_SIZE_RECONCILIATION_AGE_SECS
+    });
+
+    while info.size_reconciliation.len() > MAX_SIZE_RECONCILIATION_ENTRIES_PER_BUCKET
+        || info
+            .size_reconciliation
+            .values()
+            .map(size_reconciliation_entry_bytes)
+            .sum::<usize>()
+            > MAX_SIZE_RECONCILIATION_BYTES_PER_BUCKET
+    {
+        let oldest = info
+            .size_reconciliation
+            .iter()
+            .min_by(|(left_key, left), (right_key, right)| {
+                left.first_seen.cmp(&right.first_seen).then_with(|| left_key.cmp(right_key))
+            })
+            .map(|(key, _)| key.clone());
+        let Some(oldest) = oldest else {
+            break;
+        };
+        info.size_reconciliation.remove(&oldest);
+    }
+}
+
 impl FolderScanner {
     fn now_secs() -> u64 {
         SystemTime::now()
@@ -748,6 +807,60 @@ impl FolderScanner {
         }
     }
 
+    /// Apply the per-object size-resolution ledger updates in one place. The
+    /// scanner cache is the durable boundary; both working copies are updated
+    /// so an incremental publication cannot lose a debt or its resolution.
+    fn apply_size_reconciliation(&mut self, summary: &SizeSummary) {
+        let now = Self::now_secs();
+        self.pending_size_reconciliation_keys
+            .extend(summary.size_reconciliation.iter().map(|entry| entry.key.clone()));
+        self.pending_size_reconciliation_scopes.extend(
+            summary
+                .reconciliation_scopes
+                .iter()
+                .map(|scope| size_reconciliation_scope_key(&scope.bucket, &scope.object)),
+        );
+        self.pending_size_reconciliation_truncated |= summary.size_reconciliation_truncated;
+
+        for info in [&mut self.new_cache.info, &mut self.update_cache.info] {
+            for incoming in &summary.size_reconciliation {
+                if let Some(existing) = info.size_reconciliation.get_mut(&incoming.key) {
+                    existing.reason = incoming.reason.clone();
+                    existing.physical_size = incoming.physical_size;
+                    existing.generation = incoming.generation.clone();
+                    existing.version_id = incoming.version_id.clone();
+                    existing.attempts = existing.attempts.saturating_add(1);
+                    continue;
+                }
+
+                if size_reconciliation_entry_bytes(incoming) > MAX_SIZE_RECONCILIATION_BYTES_PER_BUCKET {
+                    continue;
+                }
+
+                let mut entry = incoming.clone();
+                entry.first_seen = now;
+                entry.attempts = 1;
+                info.size_reconciliation.insert(entry.key.clone(), entry);
+            }
+        }
+    }
+
+    fn finish_size_reconciliation_batch(&mut self) {
+        let now = Self::now_secs();
+        let current_keys = std::mem::take(&mut self.pending_size_reconciliation_keys);
+        let scopes = std::mem::take(&mut self.pending_size_reconciliation_scopes);
+        let truncated = std::mem::replace(&mut self.pending_size_reconciliation_truncated, false);
+
+        for info in [&mut self.new_cache.info, &mut self.update_cache.info] {
+            if !truncated {
+                info.size_reconciliation.retain(|key, entry| {
+                    !scopes.contains(&size_reconciliation_scope_key(&entry.bucket, &entry.object)) || current_keys.contains(key)
+                });
+            }
+            prune_size_reconciliation(info, now);
+        }
+    }
+
     fn record_scan_resume_hint(&mut self, folder: &str) {
         self.new_cache.info.scan_resume_after = Some(folder.to_string());
         self.update_cache.info.scan_resume_after = Some(folder.to_string());
@@ -1426,6 +1539,7 @@ impl FolderScanner {
                 abandoned_children.remove(&path_join_buf(&[&item.bucket, &item.object_path()]));
 
                 apply_scanner_size_summary(into, &sz);
+                self.apply_size_reconciliation(&sz);
                 into.objects += 1;
                 object_count += 1;
                 self.budget.record_object_scanned();
@@ -2105,6 +2219,7 @@ impl FolderScanner {
             }
         }
 
+        self.finish_size_reconciliation_batch();
         done_folder();
         let scanned_objects = u64::try_from(into.objects).unwrap_or(u64::MAX);
         emit_scanner_folder_trace(&self.root, &folder.name, scanned_objects, trace_started_at, "completed");
@@ -2190,10 +2305,17 @@ pub async fn scan_data_folder(
         skip_heal,
         local_disk,
         pending_heals_changed: false,
+        pending_size_reconciliation_keys: HashSet::new(),
+        pending_size_reconciliation_scopes: HashSet::new(),
+        pending_size_reconciliation_truncated: false,
         #[cfg(test)]
         list_path_raw_options_observer: None,
     };
 
+    let now = FolderScanner::now_secs();
+    prune_size_reconciliation(&mut scanner.new_cache.info, now);
+    prune_size_reconciliation(&mut scanner.update_cache.info, now);
+
     // Check if context is cancelled
     if ctx.is_cancelled() {
         return Err(ScannerError::Other("Operation cancelled".to_string()));
@@ -2217,7 +2339,9 @@ pub async fn scan_data_folder(
             new_cache.force_compact(DATA_SCANNER_COMPACT_AT_CHILDREN);
             new_cache.info.last_update = Some(SystemTime::now());
             new_cache.info.next_cycle = cache.info.next_cycle;
-            let unresolved_objects = root.failed_objects > 0 || !new_cache.info.failed_objects.is_empty();
+            let unresolved_objects = root.failed_objects > 0
+                || !new_cache.info.failed_objects.is_empty()
+                || !new_cache.info.size_reconciliation.is_empty();
             new_cache.info.snapshot_complete = !unresolved_objects;
             let had_scan_checkpoint = cache.info.scan_checkpoint.is_some() || new_cache.info.scan_checkpoint.is_some();
             new_cache.info.scan_resume_after = None;
@@ -2245,7 +2369,7 @@ pub async fn scan_data_folder(
                 if root_has_progress {
                     new_cache.replace_hashed(&root_hash, &None, &root);
                 }
-                if partial_cache_is_useful(&root, pending_heals_changed) {
+                if partial_cache_is_useful(&root, pending_heals_changed) || !new_cache.info.size_reconciliation.is_empty() {
                     if new_cache.root().is_some() {
                         new_cache.force_compact(DATA_SCANNER_COMPACT_AT_CHILDREN);
                     }

crates/scanner/src/scanner_folder/item_actions.rs

@@ -13,6 +13,7 @@
 // limitations under the License.
 /// Per-object scan actions: ScannerItem, the get-size failure policy, and the heal/ILM admission helpers.
 use super::*;
+use sha2::{Digest as _, Sha256};
 
 /// Cached folder information for scanning
 #[derive(Clone, Debug)]
@@ -32,6 +33,263 @@ pub(super) enum GetSizeFailureAction {
     HealMetadata { object: String },
 }
 
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub(super) enum SizeResolutionReason {
+    CompressedSizeUnknown,
+    InvalidPhysicalSize,
+    UnsupportedCompression,
+    InvalidObjectSize,
+    InvalidPartSize,
+    InvalidDeclaredSize,
+    SizeOverflowOrMismatch,
+}
+
+impl SizeResolutionReason {
+    fn as_str(self) -> &'static str {
+        match self {
+            Self::CompressedSizeUnknown => "compressed_size_unknown",
+            Self::InvalidPhysicalSize => "invalid_physical_size",
+            Self::UnsupportedCompression => "unsupported_compression",
+            Self::InvalidObjectSize => "invalid_object_size",
+            Self::InvalidPartSize => "invalid_part_size",
+            Self::InvalidDeclaredSize => "invalid_declared_size",
+            Self::SizeOverflowOrMismatch => "size_overflow_or_mismatch",
+        }
+    }
+}
+
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub(super) enum SizeResolution {
+    Known { logical: i64, physical: i64 },
+    Unknown { physical: i64, reason: SizeResolutionReason },
+    Corrupt { physical: i64, reason: SizeResolutionReason },
+}
+
+impl SizeResolution {
+    fn known_size(&self) -> Option<i64> {
+        match self {
+            Self::Known { logical, .. } => Some(*logical),
+            Self::Unknown { .. } | Self::Corrupt { .. } => None,
+        }
+    }
+}
+
+fn size_reconciliation_key(oi: &ObjectInfo, reason: SizeResolutionReason) -> String {
+    let version = oi
+        .version_id
+        .filter(|version| !version.is_nil())
+        .map(|version| version.to_string())
+        .unwrap_or_default();
+    let generation = oi
+        .data_dir
+        .filter(|generation| !generation.is_nil())
+        .map(|generation| generation.to_string())
+        .unwrap_or_default();
+    // Length-prefix each component so an object key containing the separator
+    // cannot alias another identity. S3 keys are bounded in normal operation;
+    // oversized persisted values use a digest so a corrupt metadata record
+    // cannot grow the ledger without bound.
+    fn component(value: &str) -> String {
+        const MAX_COMPONENT_LEN: usize = 512;
+        if value.len() <= MAX_COMPONENT_LEN {
+            return format!("{}:{}", value.len(), value);
+        }
+        let digest = Sha256::digest(value.as_bytes());
+        let digest = hex_simd::encode_to_string(digest, hex_simd::AsciiCase::Lower);
+        format!("hash:{}:{}", value.len(), digest)
+    }
+    format!(
+        "{}|{}|{}|{}|{}",
+        component(&oi.bucket),
+        component(&oi.name),
+        component(&version),
+        component(&generation),
+        component(reason.as_str())
+    )
+}
+
+pub(super) fn bounded_reconciliation_field(value: &str) -> String {
+    const MAX_FIELD_LEN: usize = 512;
+    if value.len() <= MAX_FIELD_LEN {
+        return value.to_string();
+    }
+    let digest = hex_simd::encode_to_string(Sha256::digest(value.as_bytes()), hex_simd::AsciiCase::Lower);
+    let prefix_len = MAX_FIELD_LEN - 65;
+    let prefix = value
+        .char_indices()
+        .take_while(|(offset, ch)| offset.saturating_add(ch.len_utf8()) <= prefix_len)
+        .map(|(_, ch)| ch)
+        .collect::<String>();
+    format!("{}~{}", prefix, digest)
+}
+
+fn record_size_resolution(summary: &mut SizeSummary, oi: &ObjectInfo, resolution: &SizeResolution) {
+    match resolution {
+        SizeResolution::Known { .. } => {}
+        SizeResolution::Unknown { physical, reason } | SizeResolution::Corrupt { physical, reason } => {
+            summary.record_size_reconciliation(SizeReconciliationEntry {
+                key: size_reconciliation_key(oi, *reason),
+                bucket: bounded_reconciliation_field(&oi.bucket),
+                object: bounded_reconciliation_field(&oi.name),
+                version_id: oi
+                    .version_id
+                    .filter(|version| !version.is_nil())
+                    .map(|version| version.to_string()),
+                generation: oi
+                    .data_dir
+                    .filter(|generation| !generation.is_nil())
+                    .map(|generation| generation.to_string()),
+                reason: reason.as_str().to_string(),
+                physical_size: u64::try_from(*physical).ok(),
+                first_seen: 0,
+                attempts: 0,
+            });
+        }
+    }
+}
+
+/// Resolve the size metadata once at the scanner trust boundary. A compressed
+/// -1 sentinel is valid legacy metadata, but it cannot participate in normal
+/// logical-size accounting or size-filtered lifecycle rules.
+pub(super) fn resolve_size(oi: &ObjectInfo) -> SizeResolution {
+    let physical = oi.size;
+    if physical < 0 {
+        return SizeResolution::Corrupt {
+            physical,
+            reason: SizeResolutionReason::InvalidPhysicalSize,
+        };
+    }
+
+    let compressed = match oi.compression_read_plan() {
+        Ok((_, _, compressed)) => compressed,
+        Err(_) => {
+            return SizeResolution::Corrupt {
+                physical,
+                reason: SizeResolutionReason::UnsupportedCompression,
+            };
+        }
+    };
+
+    if oi.actual_size < -1 || (oi.actual_size == -1 && !compressed) {
+        return SizeResolution::Corrupt {
+            physical,
+            reason: SizeResolutionReason::InvalidObjectSize,
+        };
+    }
+
+    // Match ObjectInfo::get_actual_size: a positive in-memory value is the
+    // authoritative decoded size. Stale declared/part metadata must not turn
+    // an otherwise valid object into a false corruption report.
+    if oi.actual_size > 0 {
+        return SizeResolution::Known {
+            logical: oi.actual_size,
+            physical,
+        };
+    }
+
+    if oi
+        .parts
+        .iter()
+        .any(|part| part.actual_size < -1 || (part.actual_size < 0 && !compressed))
+    {
+        return SizeResolution::Corrupt {
+            physical,
+            reason: SizeResolutionReason::InvalidPartSize,
+        };
+    }
+
+    let declared = rustfs_utils::http::get_str(&oi.user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE);
+    let declared = match declared {
+        Some(value) if value.is_empty() => {
+            return SizeResolution::Corrupt {
+                physical,
+                reason: SizeResolutionReason::InvalidDeclaredSize,
+            };
+        }
+        Some(value) => match value.parse::<i64>() {
+            Ok(value) if value >= 0 => Some(value),
+            _ => {
+                return SizeResolution::Corrupt {
+                    physical,
+                    reason: SizeResolutionReason::InvalidDeclaredSize,
+                };
+            }
+        },
+        None => None,
+    };
+
+    let logical = match oi.get_actual_size() {
+        Ok(size) if size == -1 && compressed && declared.is_none() => {
+            return SizeResolution::Unknown {
+                physical,
+                reason: SizeResolutionReason::CompressedSizeUnknown,
+            };
+        }
+        Ok(size) if size >= 0 => size,
+        Ok(_) | Err(_) => {
+            return SizeResolution::Corrupt {
+                physical,
+                reason: SizeResolutionReason::SizeOverflowOrMismatch,
+            };
+        }
+    };
+
+    if compressed && logical == 0 && physical != 0 && oi.parts.is_empty() && declared.is_none() {
+        return SizeResolution::Corrupt {
+            physical,
+            reason: SizeResolutionReason::SizeOverflowOrMismatch,
+        };
+    }
+
+    SizeResolution::Known { logical, physical }
+}
+
+fn resolve_sizes(object_infos: &[ObjectInfo]) -> Vec<SizeResolution> {
+    object_infos.iter().map(resolve_size).collect()
+}
+
+fn lifecycle_rule_has_size_filter(lifecycle: &BucketLifecycleConfiguration, rule_id: &str) -> bool {
+    let filter_has_size = |filter: &s3s::dto::LifecycleRuleFilter| {
+        filter.object_size_greater_than.is_some()
+            || filter.object_size_less_than.is_some()
+            || filter
+                .and
+                .as_ref()
+                .is_some_and(|and| and.object_size_greater_than.is_some() || and.object_size_less_than.is_some())
+    };
+    lifecycle
+        .rules
+        .iter()
+        .find(|rule| {
+            if rule_id.is_empty() {
+                rule.id.as_deref().is_none_or(str::is_empty)
+            } else {
+                rule.id.as_deref() == Some(rule_id)
+            }
+        })
+        .and_then(|rule| rule.filter.as_ref())
+        .is_some_and(filter_has_size)
+}
+
+fn lifecycle_event_allowed(resolution: &SizeResolution, event: &Event, lifecycle: &BucketLifecycleConfiguration) -> bool {
+    match resolution {
+        // Missing or invalid logical size only defers actions whose selected
+        // rule actually depends on that size. Time/version-only actions retain
+        // their existing semantics, including intrinsic events without a rule ID.
+        SizeResolution::Unknown { .. } | SizeResolution::Corrupt { .. } => {
+            !lifecycle_rule_has_size_filter(lifecycle, &event.rule_id)
+        }
+        SizeResolution::Known { .. } => true,
+    }
+}
+
+/// A successful newer-noncurrent batch consumes both known and unresolved
+/// versions from the retained-version alert count. The two accounting paths
+/// are separate because only known sizes can contribute byte totals.
+fn remaining_versions_after_queued_noncurrent(remaining_versions: usize, known_count: usize, unknown_count: usize) -> usize {
+    remaining_versions.saturating_sub(known_count.saturating_add(unknown_count))
+}
+
 /// How the corrupt-metadata branch records the repair after attempting an
 /// MRF intent (backlog#1894 axis A).
 #[derive(Debug, PartialEq, Eq)]
@@ -319,34 +577,48 @@ impl ScannerItem {
             "Scanner lifecycle evaluation started"
         );
 
+        let resolved_sizes = resolve_sizes(&object_infos);
+        if let Some(first) = object_infos.first() {
+            size_summary.record_reconciliation_scope(
+                &bounded_reconciliation_field(&first.bucket),
+                &bounded_reconciliation_field(&first.name),
+            );
+        }
+        for (oi, resolution) in object_infos.iter().zip(resolved_sizes.iter()) {
+            record_size_resolution(size_summary, oi, resolution);
+        }
+        let has_corrupt_size = resolved_sizes
+            .iter()
+            .any(|resolution| matches!(resolution, SizeResolution::Corrupt { .. }));
+
         // `versioning_config` is resolved once per object by the caller
         // (`get_size`) and handed in; only `prefix_enabled` is consulted here.
 
-        let Some(lifecycle) = self.lifecycle.as_ref() else {
-            let mut cumulative_size = 0;
-            for oi in object_infos.iter() {
-                let actual_size = match oi.get_actual_size() {
-                    Ok(size) => size,
-                    Err(_) => {
-                        warn!(
-                            target: "rustfs::scanner::folder",
-                            event = EVENT_SCANNER_LIFECYCLE_ACTION,
-                            component = LOG_COMPONENT_SCANNER,
-                            subsystem = LOG_SUBSYSTEM_LIFECYCLE,
-                            bucket = %self.bucket,
-                            object = %oi.name,
-                            state = "size_lookup_failed",
-                            "Scanner lifecycle action used fallback size"
-                        );
+        let Some(lifecycle) = self.lifecycle.clone() else {
+            let mut cumulative_size: i64 = 0;
+            for (oi, resolved_size) in object_infos.iter().zip(resolved_sizes.iter()) {
+                let accounting_size = match resolved_size {
+                    SizeResolution::Known { logical, .. } => *logical,
+                    // A valid compressed legacy sentinel has no logical size,
+                    // but heal and replication still need to run. The
+                    // physical size is only an input to those operations; it
+                    // is not folded into the logical total below.
+                    SizeResolution::Unknown { physical, .. } => {
+                        self.heal_actions(oi, *physical, size_summary).await;
+                        size_summary.actions_accounting_unknown(oi);
+                        continue;
+                    }
+                    SizeResolution::Corrupt { .. } => {
+                        size_summary.actions_accounting_unknown(oi);
                         continue;
                     }
                 };
 
-                let size = self.heal_actions(oi, actual_size, size_summary).await;
+                let size = self.heal_actions(oi, accounting_size, size_summary).await;
 
-                size_summary.actions_accounting(oi, size, actual_size);
+                size_summary.actions_accounting(oi, size, accounting_size);
 
-                cumulative_size += size;
+                cumulative_size = cumulative_size.saturating_add(size);
             }
 
             self.alert_excessive_versions(object_infos.len(), cumulative_size);
@@ -400,25 +672,108 @@ impl ScannerItem {
         let mut to_delete_objs: Vec<ObjectToDelete> = Vec::new();
         let mut noncurrent_events: Vec<Event> = Vec::new();
         let mut noncurrent_accounting: Vec<PendingScannerAccounting<'_>> = Vec::new();
+        let mut noncurrent_unknown: Vec<&ObjectInfo> = Vec::new();
         let mut cumulative_size = 0;
         let mut remaining_versions = object_infos.len();
         'eventLoop: {
             for (i, event) in events.iter().enumerate() {
                 let oi = &object_infos[i];
-                let actual_size = match oi.get_actual_size() {
-                    Ok(size) => size,
-                    Err(_) => {
-                        warn!(
-                            target: "rustfs::scanner::folder",
-                            event = EVENT_SCANNER_LIFECYCLE_ACTION,
-                            component = LOG_COMPONENT_SCANNER,
-                            subsystem = LOG_SUBSYSTEM_LIFECYCLE,
-                            bucket = %self.bucket,
-                            object = %oi.name,
-                            state = "size_lookup_failed",
-                            "Scanner lifecycle action used fallback size"
-                        );
-                        0
+                let known_size = resolved_sizes[i].known_size();
+                if has_corrupt_size
+                    && matches!(
+                        event.action,
+                        IlmAction::DeleteAllVersionsAction | IlmAction::DelMarkerDeleteAllVersionsAction
+                    )
+                {
+                    // An all-version delete would also remove a corrupt
+                    // sibling that could not be reconciled safely.
+                    continue;
+                }
+                if !lifecycle_event_allowed(&resolved_sizes[i], event, &lifecycle) {
+                    // An unknown logical size must not make an otherwise
+                    // non-destructive scan disappear from heal/physical-tier
+                    // accounting. Size-filtered or deferred events remain
+                    // pending, so retain the version-only physical counters.
+                    if let SizeResolution::Unknown { physical, .. } = &resolved_sizes[i] {
+                        self.heal_actions(oi, *physical, size_summary).await;
+                        size_summary.actions_accounting_unknown(oi);
+                    }
+                    continue;
+                }
+                let actual_size = match known_size {
+                    Some(size) => size,
+                    None => {
+                        match event.action {
+                            IlmAction::DeleteAction
+                            | IlmAction::DeleteRestoredAction
+                            | IlmAction::DeleteRestoredVersionAction
+                            | IlmAction::DeleteAllVersionsAction
+                            | IlmAction::DelMarkerDeleteAllVersionsAction => {
+                                let done_ilm = Metrics::time_ilm(event.action);
+                                let trace_started_at = trace_start_instant();
+                                let queued = apply_expiry_rule(event, &LcEventSrc::Scanner, oi).await;
+                                emit_scanner_ilm_action_trace(&self.bucket, &oi.name, event.action, 1, queued, trace_started_at);
+                                if record_scanner_ilm_action_if_queued(global_metrics(), event.action, 1, queued) {
+                                    done_ilm(1)();
+                                    if event.action == IlmAction::DeleteAllVersionsAction
+                                        || event.action == IlmAction::DelMarkerDeleteAllVersionsAction
+                                    {
+                                        remaining_versions = 0;
+                                    }
+                                } else if matches!(
+                                    event.action,
+                                    IlmAction::DeleteAction
+                                        | IlmAction::DeleteRestoredAction
+                                        | IlmAction::DeleteRestoredVersionAction
+                                ) {
+                                    size_summary.actions_accounting_unknown(oi);
+                                } else {
+                                    size_summary.actions_accounting_unknown(oi);
+                                    for (j, retained) in object_infos.iter().enumerate().skip(i + 1) {
+                                        match &resolved_sizes[j] {
+                                            SizeResolution::Known { logical, .. } => PendingScannerAccounting {
+                                                object: retained,
+                                                retained_size: *logical,
+                                                expired_size: 0,
+                                            }
+                                            .apply(size_summary, &mut cumulative_size, false),
+                                            SizeResolution::Unknown { .. } => {
+                                                size_summary.actions_accounting_unknown(retained);
+                                            }
+                                            SizeResolution::Corrupt { .. } => {}
+                                        }
+                                    }
+                                }
+                            }
+                            IlmAction::DeleteVersionAction => {
+                                if let Some(opt) = object_opts.get(i) {
+                                    to_delete_objs.push(ObjectToDelete {
+                                        object_name: opt.name.clone(),
+                                        version_id: opt.version_id,
+                                        ..Default::default()
+                                    });
+                                    noncurrent_events.push(event.clone());
+                                    noncurrent_unknown.push(oi);
+                                }
+                            }
+                            IlmAction::TransitionAction | IlmAction::TransitionVersionAction => {
+                                let trace_started_at = trace_start_instant();
+                                let queued = apply_transition_rule(event, &LcEventSrc::Scanner, oi).await;
+                                emit_scanner_ilm_action_trace(&self.bucket, &oi.name, event.action, 1, queued, trace_started_at);
+                                if record_scanner_ilm_action_if_queued(global_metrics(), event.action, 1, queued) {
+                                    let done_ilm = Metrics::time_ilm(event.action);
+                                    done_ilm(1)();
+                                }
+                                size_summary.actions_accounting_unknown(oi);
+                            }
+                            IlmAction::NoneAction | IlmAction::ActionCount => {
+                                if let SizeResolution::Unknown { physical, .. } = &resolved_sizes[i] {
+                                    self.heal_actions(oi, *physical, size_summary).await;
+                                }
+                                size_summary.actions_accounting_unknown(oi);
+                            }
+                        }
+                        continue;
                     }
                 };
 
@@ -446,36 +801,24 @@ impl ScannerItem {
                             done_ilm(1)();
                             remaining_versions = 0;
                         } else {
-                            PendingScannerAccounting {
-                                object: oi,
-                                retained_size: actual_size,
-                                expired_size: 0,
-                            }
-                            .apply(size_summary, &mut cumulative_size, false);
-                            for retained in object_infos.iter().skip(i + 1) {
-                                let retained_size = match retained.get_actual_size() {
-                                    Ok(size) => size,
-                                    Err(_) => {
-                                        warn!(
-                                            target: "rustfs::scanner::folder",
-                                            event = EVENT_SCANNER_LIFECYCLE_ACTION,
-                                            component = LOG_COMPONENT_SCANNER,
-                                            subsystem = LOG_SUBSYSTEM_LIFECYCLE,
-                                            bucket = %self.bucket,
-                                            object = %retained.name,
-                                            state = "size_lookup_failed",
-                                            "Scanner lifecycle action used fallback size"
-                                        );
-                                        0
-                                    }
-                                };
+                            if let Some(actual_size) = known_size {
                                 PendingScannerAccounting {
-                                    object: retained,
-                                    retained_size,
+                                    object: oi,
+                                    retained_size: actual_size,
                                     expired_size: 0,
                                 }
                                 .apply(size_summary, &mut cumulative_size, false);
                             }
+                            for (j, retained) in object_infos.iter().enumerate().skip(i + 1) {
+                                if let Some(retained_size) = resolved_sizes[j].known_size() {
+                                    PendingScannerAccounting {
+                                        object: retained,
+                                        retained_size,
+                                        expired_size: 0,
+                                    }
+                                    .apply(size_summary, &mut cumulative_size, false);
+                                }
+                            }
                         }
                         break 'eventLoop;
                     }
@@ -511,11 +854,13 @@ impl ScannerItem {
                                 version_id: opt.version_id,
                                 ..Default::default()
                             });
-                            noncurrent_accounting.push(PendingScannerAccounting {
-                                object: oi,
-                                retained_size: actual_size,
-                                expired_size: 0,
-                            });
+                            if let Some(actual_size) = known_size {
+                                noncurrent_accounting.push(PendingScannerAccounting {
+                                    object: oi,
+                                    retained_size: actual_size,
+                                    expired_size: 0,
+                                });
+                            }
                             account_now = false;
                         }
                         noncurrent_events.push(event.clone());
@@ -548,7 +893,7 @@ impl ScannerItem {
 
                 if account_now {
                     size_summary.actions_accounting(oi, size, actual_size);
-                    cumulative_size += size;
+                    cumulative_size = cumulative_size.saturating_add(size);
                 }
             }
         }
@@ -576,11 +921,20 @@ impl ScannerItem {
             }
             if record_scanner_ilm_action_if_queued(global_metrics(), action, count, queued) {
                 done_ilm(count)();
-                remaining_versions = remaining_versions.saturating_sub(noncurrent_accounting.len());
+                remaining_versions = remaining_versions_after_queued_noncurrent(
+                    remaining_versions,
+                    noncurrent_accounting.len(),
+                    noncurrent_unknown.len(),
+                );
             }
             for pending in noncurrent_accounting {
                 pending.apply(size_summary, &mut cumulative_size, queued);
             }
+            if !queued {
+                for object in noncurrent_unknown {
+                    size_summary.actions_accounting_unknown(object);
+                }
+            }
         }
         self.alert_excessive_versions(remaining_versions, cumulative_size);
     }
@@ -929,4 +1283,394 @@ mod tests {
         assert_eq!(item.object_name, "object");
         assert_eq!(item.object_path(), "object");
     }
+
+    #[test]
+    fn size_resolution_rejects_negative_overflow_and_unknown_compression() {
+        let compressed = |actual_size: i64, declared: Option<&str>| {
+            let mut user_defined = HashMap::new();
+            rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+            if let Some(declared) = declared {
+                rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, declared.to_string());
+            }
+            ObjectInfo {
+                size: 12,
+                actual_size,
+                user_defined: Arc::new(user_defined),
+                ..Default::default()
+            }
+        };
+
+        let normal = ObjectInfo {
+            size: 12,
+            actual_size: 10,
+            ..Default::default()
+        };
+        assert_eq!(
+            resolve_size(&normal),
+            SizeResolution::Known {
+                logical: 10,
+                physical: 12
+            }
+        );
+
+        let stale_declared_metadata = ObjectInfo {
+            size: 12,
+            actual_size: 10,
+            user_defined: Arc::new(HashMap::from([("x-rustfs-internal-actual-size".to_string(), "not-a-size".to_string())])),
+            parts: Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
+                actual_size: -2,
+                ..Default::default()
+            }]),
+            ..Default::default()
+        };
+        assert_eq!(
+            resolve_size(&stale_declared_metadata),
+            SizeResolution::Known {
+                logical: 10,
+                physical: 12
+            }
+        );
+
+        assert_eq!(
+            resolve_size(&compressed(0, Some("9"))),
+            SizeResolution::Known {
+                logical: 9,
+                physical: 12
+            }
+        );
+        assert_eq!(
+            resolve_size(&compressed(-1, None)),
+            SizeResolution::Unknown {
+                physical: 12,
+                reason: SizeResolutionReason::CompressedSizeUnknown,
+            }
+        );
+        assert!(matches!(
+            resolve_size(&compressed(0, Some("not-a-size"))),
+            SizeResolution::Corrupt {
+                reason: SizeResolutionReason::InvalidDeclaredSize,
+                ..
+            }
+        ));
+        assert!(matches!(
+            resolve_size(&ObjectInfo {
+                size: 12,
+                actual_size: -2,
+                ..Default::default()
+            }),
+            SizeResolution::Corrupt { .. }
+        ));
+        assert!(matches!(resolve_size(&compressed(0, Some("-1"))), SizeResolution::Corrupt { .. }));
+        assert!(matches!(resolve_size(&compressed(0, Some(""))), SizeResolution::Corrupt { .. }));
+
+        let unsupported = {
+            let mut object = compressed(0, None);
+            let mut metadata = (*object.user_defined).clone();
+            rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "unsupported".to_string());
+            object.user_defined = Arc::new(metadata);
+            object
+        };
+        assert!(matches!(resolve_size(&unsupported), SizeResolution::Corrupt { .. }));
+
+        let invalid_part = {
+            let mut object = compressed(0, None);
+            object.parts = Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
+                size: 12,
+                actual_size: -2,
+                ..Default::default()
+            }]);
+            object
+        };
+        assert!(matches!(resolve_size(&invalid_part), SizeResolution::Corrupt { .. }));
+
+        let overflow = {
+            let mut object = compressed(0, None);
+            object.parts = Arc::new(vec![
+                rustfs_filemeta::ObjectPartInfo {
+                    size: 1,
+                    actual_size: i64::MAX,
+                    ..Default::default()
+                },
+                rustfs_filemeta::ObjectPartInfo {
+                    size: 1,
+                    actual_size: 1,
+                    ..Default::default()
+                },
+            ]);
+            object
+        };
+        assert!(matches!(resolve_size(&overflow), SizeResolution::Corrupt { .. }));
+
+        let mismatch = compressed(0, None);
+        assert!(matches!(resolve_size(&mismatch), SizeResolution::Corrupt { .. }));
+        assert_eq!(
+            resolve_size(&ObjectInfo {
+                size: 0,
+                actual_size: 0,
+                ..Default::default()
+            }),
+            SizeResolution::Known { logical: 0, physical: 0 }
+        );
+    }
+
+    #[test]
+    fn size_resolution_records_and_replays_one_identity() {
+        let version_id = uuid::Uuid::new_v4();
+        let generation = uuid::Uuid::new_v4();
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "not-a-number".to_string());
+        let corrupt = ObjectInfo {
+            bucket: "bucket".to_string(),
+            name: "object".to_string(),
+            size: 12,
+            version_id: Some(version_id),
+            data_dir: Some(generation),
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+
+        let mut summary = SizeSummary::default();
+        let resolution = resolve_size(&corrupt);
+        record_size_resolution(&mut summary, &corrupt, &resolution);
+        record_size_resolution(&mut summary, &corrupt, &resolution);
+        assert_eq!(summary.size_reconciliation.len(), 1);
+        assert_eq!(summary.size_reconciliation[0].reason, "invalid_declared_size");
+        assert_eq!(summary.size_reconciliation[0].physical_size, Some(12));
+
+        let known = ObjectInfo {
+            actual_size: 12,
+            user_defined: Arc::new(HashMap::new()),
+            ..corrupt.clone()
+        };
+        record_size_resolution(&mut summary, &known, &resolve_size(&known));
+        summary.record_reconciliation_scope(&known.bucket, &known.name);
+        assert_eq!(summary.reconciliation_scopes.len(), 1);
+        assert_eq!(summary.reconciliation_scopes[0].bucket, "bucket");
+    }
+
+    #[test]
+    fn malformed_size_has_same_ilm_accounting() {
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "invalid".to_string());
+        let object = ObjectInfo {
+            bucket: "bucket".to_string(),
+            name: "object".to_string(),
+            size: 12,
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+        let resolution = resolve_size(&object);
+        let mut without_ilm = SizeSummary::default();
+        let mut with_ilm = SizeSummary::default();
+        record_size_resolution(&mut without_ilm, &object, &resolution);
+        record_size_resolution(&mut with_ilm, &object, &resolution);
+        assert_eq!(without_ilm.size_reconciliation, with_ilm.size_reconciliation);
+        assert_eq!(without_ilm.total_size, 0);
+        assert_eq!(with_ilm.total_size, 0);
+        assert!(without_ilm.tier_stats.is_empty());
+        assert!(with_ilm.tier_stats.is_empty());
+    }
+
+    #[test]
+    fn size_resolution_parses_once_per_version() {
+        let objects = vec![
+            ObjectInfo {
+                bucket: "bucket".to_string(),
+                name: "one".to_string(),
+                size: 1,
+                actual_size: 1,
+                ..Default::default()
+            },
+            ObjectInfo {
+                bucket: "bucket".to_string(),
+                name: "two".to_string(),
+                size: 2,
+                actual_size: -2,
+                ..Default::default()
+            },
+        ];
+        let resolutions = resolve_sizes(&objects);
+        assert_eq!(resolutions.len(), objects.len());
+        assert!(matches!(resolutions[0], SizeResolution::Known { logical: 1, .. }));
+        assert!(matches!(resolutions[1], SizeResolution::Corrupt { .. }));
+    }
+
+    #[test]
+    fn queued_unknown_noncurrent_versions_are_removed_from_alert_count() {
+        assert_eq!(remaining_versions_after_queued_noncurrent(3, 1, 2), 0);
+        assert_eq!(remaining_versions_after_queued_noncurrent(7, 2, 1), 4);
+        assert_eq!(remaining_versions_after_queued_noncurrent(usize::MAX, usize::MAX, usize::MAX), 0);
+    }
+
+    #[test]
+    fn malformed_size_blocks_size_dependent_transition_but_allows_time_only_expiry() {
+        let size_filtered = BucketLifecycleConfiguration {
+            rules: vec![s3s::dto::LifecycleRule {
+                status: s3s::dto::ExpirationStatus::from_static(s3s::dto::ExpirationStatus::ENABLED),
+                expiration: None,
+                abort_incomplete_multipart_upload: None,
+                del_marker_expiration: None,
+                id: Some("size".to_string()),
+                filter: Some(s3s::dto::LifecycleRuleFilter {
+                    object_size_greater_than: Some(1),
+                    ..Default::default()
+                }),
+                noncurrent_version_expiration: None,
+                noncurrent_version_transitions: None,
+                prefix: None,
+                transitions: None,
+            }],
+            ..Default::default()
+        };
+        let unknown = SizeResolution::Unknown {
+            physical: 12,
+            reason: SizeResolutionReason::CompressedSizeUnknown,
+        };
+        let size_event = Event {
+            action: IlmAction::DeleteAction,
+            rule_id: "size".to_string(),
+            ..Default::default()
+        };
+        assert!(!lifecycle_event_allowed(&unknown, &size_event, &size_filtered));
+        assert!(!lifecycle_event_allowed(
+            &unknown,
+            &Event {
+                action: IlmAction::TransitionAction,
+                rule_id: "size".to_string(),
+                ..Default::default()
+            },
+            &size_filtered
+        ));
+        let mixed_filters = BucketLifecycleConfiguration {
+            rules: vec![
+                size_filtered.rules[0].clone(),
+                s3s::dto::LifecycleRule {
+                    status: s3s::dto::ExpirationStatus::from_static(s3s::dto::ExpirationStatus::ENABLED),
+                    expiration: None,
+                    abort_incomplete_multipart_upload: None,
+                    del_marker_expiration: None,
+                    id: Some("time".to_string()),
+                    filter: None,
+                    noncurrent_version_expiration: None,
+                    noncurrent_version_transitions: None,
+                    prefix: None,
+                    transitions: None,
+                },
+            ],
+            ..Default::default()
+        };
+        assert!(lifecycle_event_allowed(
+            &unknown,
+            &Event {
+                action: IlmAction::DeleteAction,
+                rule_id: "time".to_string(),
+                ..Default::default()
+            },
+            &mixed_filters
+        ));
+        assert!(lifecycle_event_allowed(
+            &unknown,
+            &Event {
+                action: IlmAction::TransitionAction,
+                ..Default::default()
+            },
+            &BucketLifecycleConfiguration::default()
+        ));
+        assert!(lifecycle_event_allowed(
+            &SizeResolution::Corrupt {
+                physical: 12,
+                reason: SizeResolutionReason::InvalidDeclaredSize,
+            },
+            &Event {
+                action: IlmAction::DeleteAction,
+                ..Default::default()
+            },
+            &BucketLifecycleConfiguration::default()
+        ));
+        assert!(!lifecycle_event_allowed(
+            &SizeResolution::Corrupt {
+                physical: 12,
+                reason: SizeResolutionReason::InvalidDeclaredSize,
+            },
+            &Event {
+                action: IlmAction::DeleteAction,
+                rule_id: "size".to_string(),
+                ..Default::default()
+            },
+            &size_filtered
+        ));
+        assert!(lifecycle_rule_has_size_filter(
+            &BucketLifecycleConfiguration {
+                rules: vec![s3s::dto::LifecycleRule {
+                    status: s3s::dto::ExpirationStatus::from_static(s3s::dto::ExpirationStatus::ENABLED),
+                    expiration: None,
+                    abort_incomplete_multipart_upload: None,
+                    del_marker_expiration: None,
+                    id: None,
+                    filter: Some(s3s::dto::LifecycleRuleFilter {
+                        object_size_greater_than: Some(1),
+                        ..Default::default()
+                    }),
+                    noncurrent_version_expiration: None,
+                    noncurrent_version_transitions: None,
+                    prefix: None,
+                    transitions: None,
+                }],
+                ..Default::default()
+            },
+            ""
+        ));
+        assert!(lifecycle_event_allowed(
+            &SizeResolution::Known {
+                logical: 10,
+                physical: 12,
+            },
+            &Event {
+                action: IlmAction::DeleteAllVersionsAction,
+                ..Default::default()
+            },
+            &BucketLifecycleConfiguration::default()
+        ));
+        assert!(lifecycle_event_allowed(
+            &unknown,
+            &Event {
+                action: IlmAction::DeleteAction,
+                rule_id: "time-only".to_string(),
+                ..Default::default()
+            },
+            &BucketLifecycleConfiguration::default()
+        ));
+    }
+
+    #[tokio::test]
+    async fn long_object_size_reconciliation_scope_uses_bounded_identity() {
+        let object_name = "o".repeat(600);
+        let mut item = scanner_item_with_prefix("");
+        item.object_name = object_name.clone();
+
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+        let object = ObjectInfo {
+            bucket: item.bucket.clone(),
+            name: object_name.clone(),
+            size: 12,
+            actual_size: -1,
+            version_id: Some(uuid::Uuid::new_v4()),
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+        let mut summary = SizeSummary::default();
+        item.apply_actions(vec![object], None, VersioningConfiguration::default(), &mut summary)
+            .await;
+
+        let bounded_bucket = bounded_reconciliation_field(&item.bucket);
+        let bounded_object = bounded_reconciliation_field(&object_name);
+        assert_eq!(summary.reconciliation_scopes[0].bucket, bounded_bucket);
+        assert_eq!(summary.reconciliation_scopes[0].object, bounded_object);
+        assert_eq!(summary.size_reconciliation[0].object, bounded_object);
+        assert_eq!(summary.versions, 1);
+        assert_eq!(summary.total_size, 0);
+    }
 }

crates/scanner/src/scanner_folder/tests.rs

@@ -326,6 +326,9 @@ async fn build_test_scanner() -> (FolderScanner, std::path::PathBuf) {
         skip_heal: Arc::new(AtomicBool::new(false)),
         local_disk: disk,
         pending_heals_changed: false,
+        pending_size_reconciliation_keys: HashSet::new(),
+        pending_size_reconciliation_scopes: HashSet::new(),
+        pending_size_reconciliation_truncated: false,
         list_path_raw_options_observer: None,
     };
 
@@ -388,6 +391,66 @@ async fn test_record_failed_ttl_zero_noop() {
     assert!(!scanner.should_skip_failed("path2"));
 }
 
+#[tokio::test]
+async fn malformed_size_reconciliation_replays_after_restart() {
+    let (mut scanner, temp_dir) = build_test_scanner().await;
+    let _guard = TestGuard::new(60, 100, &mut scanner, temp_dir);
+
+    let entry = SizeReconciliationEntry {
+        key: "1:b|6:object|0:|0:".to_string(),
+        bucket: "b".to_string(),
+        object: "object".to_string(),
+        reason: "invalid_declared_size".to_string(),
+        physical_size: Some(12),
+        ..Default::default()
+    };
+    let mut summary = SizeSummary::default();
+    summary.record_size_reconciliation(entry.clone());
+    summary.record_reconciliation_scope("b", "object");
+    scanner.apply_size_reconciliation(&summary);
+    scanner.apply_size_reconciliation(&summary);
+
+    assert_eq!(scanner.new_cache.info.size_reconciliation.len(), 1);
+    assert_eq!(scanner.update_cache.info.size_reconciliation.len(), 1);
+    assert_eq!(scanner.new_cache.info.size_reconciliation[&entry.key].attempts, 2);
+
+    let encoded = rmp_serde::to_vec_named(&scanner.new_cache.info).expect("size ledger should encode");
+    let decoded: crate::data_usage_define::DataUsageCacheInfo =
+        rmp_serde::from_slice(&encoded).expect("size ledger should decode");
+    assert_eq!(decoded.size_reconciliation.len(), 1);
+    assert_eq!(decoded.size_reconciliation[&entry.key].reason, "invalid_declared_size");
+
+    let mut resolved = SizeSummary::default();
+    resolved.record_reconciliation_scope("b", "object");
+    scanner.apply_size_reconciliation(&resolved);
+    assert!(scanner.new_cache.info.size_reconciliation.is_empty());
+    assert!(scanner.update_cache.info.size_reconciliation.is_empty());
+}
+
+#[tokio::test]
+async fn malformed_size_reconciliation_clears_bounded_long_object_scope() {
+    let (mut scanner, temp_dir) = build_test_scanner().await;
+    let _guard = TestGuard::new(60, 100, &mut scanner, temp_dir);
+    let long_object = "o".repeat(600);
+    let bounded_object = item_actions::bounded_reconciliation_field(&long_object);
+    let entry = SizeReconciliationEntry {
+        key: "long-object-key".to_string(),
+        bucket: "b".to_string(),
+        object: bounded_object,
+        reason: "invalid_declared_size".to_string(),
+        ..Default::default()
+    };
+    let mut summary = SizeSummary::default();
+    summary.record_size_reconciliation(entry);
+    scanner.apply_size_reconciliation(&summary);
+    assert_eq!(scanner.new_cache.info.size_reconciliation.len(), 1);
+
+    let mut resolved = SizeSummary::default();
+    resolved.record_reconciliation_scope("b", &long_object);
+    scanner.apply_size_reconciliation(&resolved);
+    assert!(scanner.new_cache.info.size_reconciliation.is_empty());
+}
+
 #[test]
 fn test_classify_get_size_failure_marks_metadata_heal_object_path() {
     let temp_dir = std::env::temp_dir();

deny.toml

@@ -43,9 +43,6 @@ allow-git = [
     # RustFS fork carrying presigned expiry and constant-time authentication fixes.
     # owner: rustfs-maintainers  review: 2026-10
     "https://github.com/rustfs/s3s.git",
-    # MiMalloc fork pinned for hotpath allocation counting support.
-    # owner: houseme  review: 2026-10
-    "https://github.com/xonatius/mimalloc_rust.git",
 ]
 
 [bans]

docs/testing/ci-gates.md

@@ -0,0 +1,149 @@
+# CI gate matrix
+
+This file is the source of truth for which validation runs on each event, its
+configured wall-clock budget, and whether it can block a merge. Test taxonomy,
+naming, and nextest serialization rules remain in [README.md](README.md); e2e
+membership and counts remain in
+[e2e-suite-inventory.md](e2e-suite-inventory.md).
+
+The distinction between **required** and **report-only** is load-bearing:
+a failing job blocks a merge only when its exact check name is present in the
+live `main` ruleset. A workflow name, a `merge_group` trigger, or a red PR check
+does not make a job required by itself.
+
+## Required merge checks
+
+The live `main` ruleset (`6436880`) currently requires exactly these contexts:
+
+| Required context | Producer | Validation |
+|---|---|---|
+| `CLA Check` | `.github/workflows/cla.yml` | Contributor agreement |
+| `Quick Checks` | `.github/workflows/ci.yml` | Formatting and repository guard scripts |
+| `Test and Lint` | `.github/workflows/ci.yml` | Clippy, workspace nextest excluding `e2e_test`, doctests, and migration proofs |
+
+For pull requests limited to the paths excluded by the main CI workflow,
+`.github/workflows/ci-docs-only.yml` reports `Quick Checks` and
+`Test and Lint` under the same names. It runs the real quick checks and the
+planning-document guard; it does not claim that Rust compilation or runtime
+tests ran. Despite the workflow name, these paths also include selected deploy,
+workflow, and lock files.
+
+Verify the live rule rather than trusting this snapshot before changing merge
+policy:
+
+```bash
+gh api repos/rustfs/rustfs/rulesets/6436880 \
+  --jq '.rules[] | select(.type == "required_status_checks") | .parameters'
+```
+
+The ruleset currently has `strict_required_status_checks_policy=false`.
+`Continuous Integration` accepts `merge_group` events and runs `e2e-full` for
+them, but `End-to-End Tests (full merge gate)` is not currently a required
+context. Therefore the repository is prepared to test a merge-queue SHA, but
+the workflow alone does not prove that every merge passed that lane.
+
+## Pull request and merge matrix
+
+Budgets below are job `timeout-minutes`, not typical runtimes. “Report-only”
+means the result is visible and actionable but is not in the live required
+context list.
+
+| Event | Validation | Budget | Merge status | Reproduction |
+|---|---|---:|---|---|
+| PR, non-doc change | `Quick Checks` | 10 min | Required | `make pre-commit` (broader local umbrella) |
+| PR, non-doc change | `Test and Lint` | 90 min | Required | `cargo nextest run --profile ci --all --exclude e2e_test` |
+| PR, non-doc change | `Typos` | 10 min | Report-only | `typos` |
+| PR, non-doc change | `ILM Integration (serial)` | 90 min | Report-only | Use the exact command in `.github/workflows/ci.yml` |
+| PR, non-doc change | rio-v2 / swift / sftp test-and-lint variants | 90 min each | Report-only | `cargo nextest run` with the workflow's feature set |
+| PR, non-doc change | `Build RustFS Debug Binary` | 30 min | Report-only; prerequisite for black-box lanes | `cargo build -p rustfs --bins` |
+| PR, non-doc change | `io_uring Integration (real)` | 30 min | Report-only | `cargo test -p rustfs-ecstore --lib uring_ -- --test-threads=1 --nocapture` |
+| PR, non-doc change | `End-to-End Tests` (`e2e-smoke` plus `s3s-e2e`) | 30 min | Report-only | `cargo nextest run --profile e2e-smoke -p e2e_test`; then `./scripts/e2e-run.sh ./target/debug/rustfs <data-dir>` |
+| PR, non-doc change | `S3 Implemented Tests` | 60 min | Report-only | Build `rustfs`, then run `scripts/s3-tests/run.sh` with `DEPLOY_MODE=binary`, `TEST_MODE=single`, and `MAXFAIL=0` |
+| PR, non-doc change | `S3 Lifecycle Behavior Tests` | 30 min | Report-only | Use the accelerated scanner environment in `.github/workflows/ci.yml` with `scripts/s3-tests/run.sh` |
+| PR touching dependency or workflow inputs | Cargo Deny / Workflow Pin Report / Dependency Review | 20 / 5 / 30 min | Report-only | `cargo deny check`; `scripts/security/check_workflow_pins.sh` |
+| PR touching architecture rules or architecture docs | `Architecture Migration Rules` | 10 min | Report-only | `scripts/check_architecture_migration_rules.sh` |
+| PR touching Nix or workspace manifests | `Nix Build & Check` | 60 min | Report-only | `nix flake check` |
+| PR limited to main-CI-excluded paths | companion `Quick Checks` and `Test and Lint` | 10 min each | Required | `git diff --check`; `make doc-paths-check` when documentation paths changed |
+| `merge_group` | Standard CI plus `e2e-full` | 55 min for `e2e-full` | Standard required contexts only; `e2e-full` report-only | `cargo nextest run --profile e2e-full -p e2e_test` |
+| Push to `main` | Standard CI plus `e2e-full` | 55 min for `e2e-full` | Post-merge detection | Same as `merge_group` |
+| PR touching fuzz inputs or harness paths | Build plus five 60-second fuzz smoke targets | 60 min build; 30 min per target | Report-only | `MAX_TOTAL_TIME=60 ./scripts/fuzz/run.sh` |
+| PR touching selected ecstore disk/format paths | `Rename Safety` on Windows | 60 min | Report-only | Run the four `cargo test -p rustfs-ecstore --lib <filter>` commands in `windows-filesystem.yml` on Windows |
+
+The authoritative e2e filters live in `.config/nextest.toml`; extend a profile
+instead of adding a second ad-hoc selector. Before a profile runs,
+`scripts/check_test_wiring.py` compares its exact membership to the committed
+digest so a silent test drop fails closed.
+
+## Scheduled and manual validation
+
+Scheduled lanes are independent fault domains. They do not block a pull
+request, but their workflow-local gate can fail the run and scheduled failures
+are routed to the shared failure-issue action. The scheduled-validation
+watchdog and freshness workflow separately detect incomplete runs and missing
+schedules.
+
+| Cadence (UTC unless noted) | Workflow / validation | Budget | Verdict and artifacts | Reproduction |
+|---|---|---:|---|---|
+| Daily 02:17 | Fuzz: five nightly corpus targets | 60 min build; 60 min per target | Gate; corpus/crash artifacts, scheduled failure alert | `MAX_TOTAL_TIME=<seconds> ./scripts/fuzz/run.sh` |
+| Daily 03:17 | MinIO interop (EC + SSE read parity) | 40 min | Gate; scheduled failure alert | Dispatch `minio-interop.yml` or follow its pinned Docker fixture steps |
+| Daily 04:29 | Replication / cluster-fault / protocol e2e | 45 / 90 / 90 min | Three independent gates; JUnit, membership, and server logs | `cargo nextest run --profile e2e-repl-nightly -p e2e_test`; `--profile e2e-nightly`; `-j 1 --profile e2e-protocols` |
+| Daily 06:31 | Warp performance A/B | 180 min | Regression budget gate; A/B summaries and server logs | `bash scripts/run_hotpath_warp_abba.sh --help` |
+| Daily 00:07 Asia/Shanghai (16:07 UTC previous day) | Nightly GNU build and Vault lanes | 150 / 90 / 60 min | Build, live Vault, and HA failover gates | Use the commands and pinned Vault images in `nightly-gnu.yml` |
+| Daily 03:23 | Security Audit | 20 / 5 min, plus 30 min on PR dependency review | Cargo Deny and workflow-pin gates; scheduled failure alert | `cargo deny check`; `scripts/security/check_workflow_pins.sh` |
+| Daily 23:47 | Scheduled Validation Freshness | 10 min | Fails when a critical schedule was never created or is stale | Dispatch `scheduled-validation-freshness.yml` |
+| Sunday 00:11 | Full `Continuous Integration` matrix | Per-job budgets above | Weekly variant coverage, including dormant rio-v2 binary/e2e lanes | Dispatch `ci.yml` |
+| Sunday 01:13 | Seven-platform build matrix | 150 min per platform | Build/package integrity; scheduled failure alert | Dispatch `build.yml` with an exact platform set |
+| Sunday 02:19 | Ceph s3-tests full sweep: single and real four-node, four shards each | 180 min per shard | Compatibility gate; report, JUnit, exact node IDs, and server logs | `scripts/s3-tests/run.sh` against an existing single or distributed target |
+| Sunday 06:41 | Mint | 120 min | **Report-only by design**; per-suite PASS/FAIL/NA and raw `log.json` | Reproduce the pinned Docker sequence in `mint.yml` or dispatch it |
+| Sunday 07:43 | Workspace line coverage | 120 min | Report-only trend; lcov and JSON retained 90 days | `make coverage` |
+| Monthly, day 1 06:37 | Runner Hygiene | 15 min | Validates runner ephemerality; scheduled failure alert | Dispatch `runner-hygiene.yml` |
+
+Manual `workflow_dispatch` exists for the scheduled workflows above. Manual
+runs are debugging evidence and intentionally do not open scheduled-failure
+issues. A manual performance run may explicitly allow a known regression; that
+override must not be treated as an ordinary passing baseline.
+
+## Release validation
+
+Release validation is post-merge and tag-driven; it does not substitute for a
+pull-request gate.
+
+| Event | Validation | Budget | Result |
+|---|---|---:|---|
+| Push to `main` or weekly schedule | `Build and Release` platform matrix | 150 min per platform | Build artifacts for all selected targets; no release publication on a main push |
+| Valid release or preview tag | `Build and Release` plus asset checks | 150 min per platform | Draft release, checksummed assets, and publish step |
+| Successful non-preview release-tag build | Docker image build and image scan | 60 min build; 30 min scan | Multi-architecture images plus vulnerability report |
+| Successful release-tag build | DEB/RPM packaging | 30 min per architecture | Packages and checksum files uploaded to the release |
+| Successful non-preview release-tag build | Helm template test and package | 30 min build; 30 min publish | Versioned chart and repository index |
+
+Use an exact preview tag for end-to-end release rehearsal. Manual dispatches
+are backfill/debug paths and do not prove the automatic `workflow_run` chain.
+
+## Evidence requirements
+
+A green check is useful only when it proves the intended behavior ran:
+
+- Record the exact commit SHA and run URL.
+- Separate product failure from runner prerequisites, service readiness, and
+  cancellation. Repair the precondition, then rerun the exact workload.
+- Preserve membership manifests, JUnit, raw compatibility logs, seeds, and
+  server logs where the workflow provides them.
+- For a bug fix or a new fault checker, provide sensitivity evidence: the old
+  behavior or an intentional mutation must fail the new oracle, and the fixed
+  behavior must pass it.
+- Never promote a report-only lane to required from one green run. Require at
+  least 14 days and 30 representative pull requests with at least 99% complete
+  execution, then update the ruleset and this table together.
+
+## Change checklist
+
+Update this file in the same pull request when any of these change:
+
+- workflow triggers, job names, timeouts, or nextest profile ownership;
+- required status contexts or strict/merge-queue policy;
+- scheduled cadence, alert routing, artifact contract, or local reproduction;
+- report-only versus gating semantics.
+
+Do not copy per-module test counts here. Update
+[e2e-suite-inventory.md](e2e-suite-inventory.md) and its enforced membership
+digest instead.

rustfs/Cargo.toml

@@ -336,13 +336,13 @@ opentelemetry = { workspace = true }
 tracing-opentelemetry = { workspace = true }
 # Data structures
 hashbrown = { workspace = true, features = ["serde", "rayon"] }
-mimalloc = { workspace = true }
+rustfs-mimalloc = { workspace = true }
 
 [target.'cfg(target_os = "linux")'.dependencies]
 libsystemd.workspace = true
 
 [target.'cfg(not(target_os = "windows"))'.dependencies]
-libmimalloc-sys.workspace = true
+rustfs-mimalloc-sys.workspace = true
 
 [dev-dependencies]
 uuid = { workspace = true, features = ["v4", "v5", "fast-rng", "macro-diagnostics"] }

rustfs/src/allocator_reclaim.rs

@@ -369,14 +369,8 @@ pub fn allocator_reclaim_controller_snapshot(ctx: &CancellationToken) -> Allocat
 }
 
 #[cfg(not(target_os = "windows"))]
-#[allow(unsafe_code)]
 fn collect_allocator_memory(force: bool) -> Result<(), String> {
-    // SAFETY: `mi_collect` is provided by the active global allocator backend
-    // on this target family. It is explicitly intended to reclaim retained
-    // pages/segments and does not require additional invariants from the caller.
-    unsafe {
-        libmimalloc_sys::mi_collect(force);
-    }
+    rustfs_mimalloc::MiMalloc::collect(force);
     Ok(())
 }
 

rustfs/src/main.rs

@@ -26,22 +26,22 @@ struct MiMallocAllocator;
 unsafe impl GlobalAlloc for MiMallocAllocator {
     unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
         // SAFETY: the caller upholds GlobalAlloc's contract for layout.
-        unsafe { mimalloc::MiMalloc.alloc(layout) }
+        unsafe { rustfs_mimalloc::MiMalloc.alloc(layout) }
     }
 
     unsafe fn alloc_zeroed(&self, layout: Layout) -> *mut u8 {
         // SAFETY: the caller upholds GlobalAlloc's contract for layout.
-        unsafe { mimalloc::MiMalloc.alloc_zeroed(layout) }
+        unsafe { rustfs_mimalloc::MiMalloc.alloc_zeroed(layout) }
     }
 
     unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
         // SAFETY: ptr and layout came from this allocator and are forwarded unchanged.
-        unsafe { mimalloc::MiMalloc.dealloc(ptr, layout) }
+        unsafe { rustfs_mimalloc::MiMalloc.dealloc(ptr, layout) }
     }
 
     unsafe fn realloc(&self, ptr: *mut u8, layout: Layout, new_size: usize) -> *mut u8 {
         // SAFETY: ptr and layout came from this allocator and are forwarded unchanged.
-        unsafe { mimalloc::MiMalloc.realloc(ptr, layout, new_size) }
+        unsafe { rustfs_mimalloc::MiMalloc.realloc(ptr, layout, new_size) }
     }
 }
 
@@ -51,7 +51,7 @@ static GLOBAL: hotpath::CountingAllocator<MiMallocAllocator> = hotpath::Counting
 
 #[cfg(not(all(feature = "hotpath", feature = "hotpath-alloc")))]
 #[global_allocator]
-static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc;
+static GLOBAL: rustfs_mimalloc::MiMalloc = rustfs_mimalloc::MiMalloc;
 
 fn main() {
     let _hotpath_guard = hotpath::HotpathGuardBuilder::new("main").build();
@@ -71,8 +71,9 @@ mod tests {
         allocation.extend_from_slice(&[7_u8; 64]);
 
         assert_eq!(allocation.len(), 64);
+        let heap = rustfs_mimalloc::heap::Heap::main();
         // SAFETY: the live Vec pointer is valid to inspect for heap ownership.
-        assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(allocation.as_ptr().cast()) });
+        assert!(unsafe { heap.contains(allocation.as_ptr()) });
     }
 
     #[test]
@@ -85,12 +86,13 @@ mod tests {
         let layout = Layout::from_size_align(32, 8).expect("valid test allocation layout");
         let grown_layout = Layout::from_size_align(64, 8).expect("valid grown test allocation layout");
         let allocator = super::MiMallocAllocator;
+        let heap = rustfs_mimalloc::heap::Heap::main();
 
         // SAFETY: The pointer is checked for null before use and later released
         // through the same allocator with the corresponding layout.
         let ptr = unsafe { allocator.alloc_zeroed(layout) };
         assert!(!ptr.is_null());
-        assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(ptr.cast()) });
+        assert!(unsafe { heap.contains(ptr) });
         assert!(unsafe { std::slice::from_raw_parts(ptr, 32).iter().all(|byte| *byte == 0) });
 
         // SAFETY: `ptr` was allocated by `allocator` with `layout`; on failure
@@ -102,7 +104,7 @@ mod tests {
             panic!("mimalloc realloc failed in allocator smoke test");
         }
 
-        assert!(unsafe { libmimalloc_sys::mi_is_in_heap_region(grown_ptr.cast()) });
+        assert!(unsafe { heap.contains(grown_ptr) });
         // SAFETY: `grown_ptr` was reallocated by `allocator` and is released
         // with the matching grown layout.
         unsafe { allocator.dealloc(grown_ptr, grown_layout) };

rustfs/src/memory_observability.rs

@@ -17,10 +17,7 @@ use rustfs_io_metrics::{
     record_cpu_usage, record_memory_usage, record_process_memory_split,
 };
 use serde::Serialize;
-#[cfg(any(test, not(target_os = "windows")))]
 use serde_json::Value;
-#[cfg(not(target_os = "windows"))]
-use std::ffi::CStr;
 use std::path::Path;
 use std::sync::{Arc, Mutex, OnceLock};
 use std::time::Duration;
@@ -231,7 +228,18 @@ fn read_cgroup_memory_snapshot() -> Option<CgroupMemorySnapshot> {
     read_cgroup_v2().or_else(read_cgroup_v1)
 }
 
-#[cfg(any(test, not(target_os = "windows")))]
+fn read_allocator_memory_snapshot() -> Option<AllocatorMemorySnapshot> {
+    let json = rustfs_mimalloc::MiMalloc::stats_json();
+    if json.is_empty() {
+        return None;
+    }
+    let observation = parse_mimalloc_stats_json(&json)?;
+    Some(AllocatorMemorySnapshot {
+        backend: crate::allocator_reclaim::allocator_backend(),
+        observation,
+    })
+}
+
 fn numeric_json_value(value: &Value) -> Option<u64> {
     match value {
         Value::Number(number) => number
@@ -242,7 +250,6 @@ fn numeric_json_value(value: &Value) -> Option<u64> {
     }
 }
 
-#[cfg(any(test, not(target_os = "windows")))]
 fn numeric_json_field(value: &Value, field: &str) -> Option<u64> {
     match value {
         Value::Object(fields) => fields
@@ -254,7 +261,6 @@ fn numeric_json_field(value: &Value, field: &str) -> Option<u64> {
     }
 }
 
-#[cfg(any(test, not(target_os = "windows")))]
 fn mimalloc_stat_field(value: &Value, metric: &str, field: &str) -> Option<u64> {
     match value {
         Value::Object(fields) => {
@@ -271,12 +277,10 @@ fn mimalloc_stat_field(value: &Value, metric: &str, field: &str) -> Option<u64>
     }
 }
 
-#[cfg(any(test, not(target_os = "windows")))]
 fn mimalloc_stat_current(value: &Value, metric: &str) -> Option<u64> {
     mimalloc_stat_field(value, metric, "current")
 }
 
-#[cfg(any(test, not(target_os = "windows")))]
 fn mimalloc_stat_sum(value: &Value, metrics: &[&str], field: &str) -> Option<u64> {
     metrics
         .iter()
@@ -285,7 +289,6 @@ fn mimalloc_stat_sum(value: &Value, metrics: &[&str], field: &str) -> Option<u64
         .filter(|value| *value > 0)
 }
 
-#[cfg(any(test, not(target_os = "windows")))]
 fn parse_mimalloc_stats_json(stats_json: &str) -> Option<AllocatorMemoryObservation> {
     let value = serde_json::from_str::<Value>(stats_json).ok()?;
     let malloc_metrics = ["malloc_normal", "malloc_huge"];
@@ -312,33 +315,6 @@ fn parse_mimalloc_stats_json(stats_json: &str) -> Option<AllocatorMemoryObservat
     }
 }
 
-#[cfg(not(target_os = "windows"))]
-#[allow(unsafe_code)]
-fn read_allocator_memory_snapshot() -> Option<AllocatorMemorySnapshot> {
-    // SAFETY: `mi_stats_get_json` returns a null-terminated JSON buffer owned by
-    // mimalloc when called with a null input buffer. The mimalloc API requires
-    // freeing that buffer with `mi_free`; parsing finishes before the buffer is freed.
-    let observation = unsafe {
-        let stats_ptr = libmimalloc_sys::mi_stats_get_json(0, std::ptr::null_mut());
-        if stats_ptr.is_null() {
-            return None;
-        }
-
-        let observation = CStr::from_ptr(stats_ptr).to_str().ok().and_then(parse_mimalloc_stats_json);
-        libmimalloc_sys::mi_free(stats_ptr.cast());
-        observation?
-    };
-    Some(AllocatorMemorySnapshot {
-        backend: crate::allocator_reclaim::allocator_backend(),
-        observation,
-    })
-}
-
-#[cfg(target_os = "windows")]
-fn read_allocator_memory_snapshot() -> Option<AllocatorMemorySnapshot> {
-    None
-}
-
 fn configured_memory_observability_interval_secs() -> u64 {
     rustfs_utils::get_env_u64(ENV_MEMORY_OBSERVABILITY_INTERVAL_SECS, DEFAULT_MEMORY_OBSERVABILITY_INTERVAL_SECS).max(1)
 }
@@ -566,6 +542,13 @@ mod tests {
         assert_eq!(parse_mimalloc_stats_json(r#"{ "allocator": "unknown" }"#), None);
     }
 
+    #[test]
+    fn read_allocator_memory_snapshot_uses_mimalloc_stats_json() {
+        let snapshot = super::read_allocator_memory_snapshot();
+        #[cfg(not(target_os = "windows"))]
+        assert!(snapshot.is_some(), "allocator snapshot should be available on non-Windows");
+    }
+
     #[test]
     fn memory_observability_snapshot_reports_disabled_when_metrics_are_disabled() {
         let snapshot = build_memory_observability_status_snapshot(false, 15, false);

scripts/test/vault_ha_kms_live.sh

@@ -241,7 +241,7 @@ env \
   RUSTFS_TEST_VAULT_FAILOVER_MARKER="$MARKER" \
   RUSTFS_TEST_VAULT_OLD_LEADER="$OLD_LEADER" \
   cargo test -p rustfs-kms --test vault_ha_failover_live \
-    vault_raft_leader_failure_preserves_kv2_and_transit_decrypts -- \
+    vault_raft_leader_failure_recovers_kv2_and_transit_decrypts -- \
     --ignored --nocapture --test-threads=1 &
 TEST_PID=$!