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

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

.config/nextest.toml

@@ -325,8 +325,8 @@ slow-timeout = { period = "60s", terminate-after = 2, grace-period = "10s" }
 #
 # Wired by .github/workflows/e2e-replication-nightly.yml (schedule +
 # workflow_dispatch), which builds the rustfs binary once, installs awscurl so
-# the STS dual-node test actually exercises its path (the test fails when
-# awscurl is absent), and routes scheduled failures
+# the STS dual-node test actually exercises its path (it skips gracefully with
+# a visible log line when awscurl is absent), and routes scheduled failures
 # through .github/actions/schedule-failure-issue (ci-8). Explicit division of
 # labor with e2e-full: these tests run only in the consolidated nightly
 # workflow, not in the merge/main lane.

.github/workflows/ci.yml

@@ -681,19 +681,6 @@ jobs:
           cache-save-if: 'false'
           install-build-packaging-tools: 'false'
 
-      - name: Set up Python
-        uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
-        with:
-          python-version: "3.12"
-
-      - name: Install awscurl
-        run: |
-          python3 -m pip install --user --upgrade pip "awscurl==0.44"
-          echo "AWSCURL_PATH=$HOME/.local/bin/awscurl" >> "$GITHUB_ENV"
-
-      - name: Verify awscurl
-        run: test -x "$AWSCURL_PATH"
-
       # Download after the cache restore so the freshly built binary from the
       # build job always wins over anything restored into target/debug.
       - name: Download debug binary
@@ -816,20 +803,6 @@ jobs:
       - name: Verify awscurl
         run: test -x "$AWSCURL_PATH"
 
-      - name: Install mc
-        env:
-          MC_VERSION: RELEASE.2025-08-13T08-35-41Z
-          MC_SHA256: 01f866e9c5f9b87c2b09116fa5d7c06695b106242d829a8bb32990c00312e891
-        run: |
-          MC_BINARY="mc.linux-amd64.${MC_VERSION}"
-          curl -fsSLo "$RUNNER_TEMP/mc" "https://github.com/minio/mc/releases/download/${MC_VERSION}/${MC_BINARY}"
-          echo "${MC_SHA256}  $RUNNER_TEMP/mc" | sha256sum --check --status
-          chmod +x "$RUNNER_TEMP/mc"
-          echo "$RUNNER_TEMP" >> "$GITHUB_PATH"
-
-      - name: Verify mc
-        run: mc --version
-
       - name: Install Vault
         run: |
           VAULT_VERSION="1.17.6"

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

@@ -75,7 +75,11 @@ jobs:
           cache-save-if: ${{ github.ref == 'refs/heads/main' }}
           install-build-packaging-tools: 'false'
 
-      # The STS dual-node test requires awscurl and fails if it is unavailable.
+      # awscurl lets the STS dual-node test actually exercise its path. Without
+      # it the test skips gracefully with a visible log line
+      # (`awscurl_available()` in crates/e2e_test/src/common.rs), so the lane
+      # still passes — installing it just upgrades that one test from skip to
+      # real coverage.
       - name: Set up Python
         uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6.3.0
         with:
@@ -83,7 +87,7 @@ jobs:
 
       - name: Install awscurl
         run: |
-          python3 -m pip install --user --upgrade pip "awscurl==0.44"
+          python3 -m pip install --user --upgrade pip awscurl
           echo "AWSCURL_PATH=$HOME/.local/bin/awscurl" >> "$GITHUB_ENV"
 
       - name: Verify awscurl

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/README.md

@@ -123,7 +123,7 @@ via `create_s3_client(idx)` / `create_all_clients()`. See
 | `find_available_port` | Random free port (isolation primitive) |
 | `rustfs_binary_path` / `_with_features` | Locate/build the binary; honors `RUSTFS_BUILD_FEATURES` |
 | `requested_rustfs_build_features` / `rustfs_build_feature_enabled` | Feature-gate a test to what the binary was built with |
-| `execute_awscurl` / `awscurl_post` / `_get` / `_put` / `_delete` / `awscurl_post_sts_form_urlencoded` | Admin/STS API calls via `awscurl`; missing binaries are test failures |
+| `awscurl_available` + `execute_awscurl` / `awscurl_post` / `_get` / `_put` / `_delete` / `awscurl_post_sts_form_urlencoded` | Admin/STS API calls via `awscurl` (skip gracefully when absent) |
 | `replication_fast_env` | Env vars that shrink replication timers (from repl-4); pass to `start_rustfs_server_with_env` |
 | `local_http_client` / `init_logging` | Loopback HTTP client; idempotent tracing init |
 | `RustFSTestClusterEnvironment` (`new`/`start`/`start_node`/`stop_node`/`create_all_clients`) | Multi-node harness |
@@ -189,7 +189,7 @@ cargo nextest run --profile e2e-smoke -p e2e_test
 cargo nextest run --profile e2e-full -p e2e_test
 # Cluster fault nightly lane
 cargo nextest run --profile e2e-nightly -p e2e_test
-# Replication nightly lane; awscurl is required for STS paths
+# Replication nightly lane; install awscurl so STS paths do not skip
 cargo nextest run --profile e2e-repl-nightly -p e2e_test
 # Fixed-port protocol nightly lane
 RUSTFS_BUILD_FEATURES=ftps,webdav,sftp \
@@ -221,8 +221,9 @@ The `s3s-e2e` CI job selects a random `RUSTFS_TEST_PORT` (see the `e2e-tests`
 job) to dodge this; local single-node tests already use random ports, so a
 lingering orphan is usually the cause of a spurious bind failure.
 
-**`awscurl` not found.** `awscurl`-dependent tests fail closed with a process
-spawn error. Install the pinned CI version before running their profiles.
+**`awscurl` not found.** `awscurl`-dependent tests skip gracefully with a
+visible log line (`awscurl_available()`); install `awscurl` to actually run
+them.
 
 ## Related
 
@@ -257,9 +258,10 @@ A test module may join the smoke filter only if every test in it is:
 2. **Single-node** — spawns its own server via
    `RustFSTestEnvironment`/`start_rustfs_server` on a random port with an
    isolated temp dir. No `RustFSTestClusterEnvironment`, no fixed ports.
-3. **Hermetic dependencies** — no pre-started server at `localhost:9000`, no
-   Vault, and no fixed protocol ports. Any required CLI must be pinned and
-   installed by the workflow; a missing CLI must fail the test.
+3. **Dependency-free** — no pre-started server at `localhost:9000`, no Vault,
+   no fixed protocol ports. Tools that may be absent on the runner (e.g.
+   `awscurl`) are acceptable only when the test skips gracefully with a
+   visible log line (see `bucket_policy_check_test.rs`).
 4. **Not `#[ignore]`** — ignored tests are activation work (backlog#1149
    ci-13 / backlog#1148 ilm-3), not smoke candidates.
 

crates/e2e_test/src/bucket_policy_check_test.rs

@@ -52,6 +52,10 @@ fn create_user_client(env: &RustFSTestEnvironment, access_key: &str, secret_key:
 #[tokio::test]
 async fn test_bucket_policy_authenticated_user() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if !crate::common::awscurl_available() {
+        info!("Skipping test_bucket_policy_authenticated_user because awscurl is not available");
+        return Ok(());
+    }
     info!("Starting test_bucket_policy_authenticated_user...");
 
     let mut env = RustFSTestEnvironment::new().await?;

crates/e2e_test/src/common.rs

@@ -494,6 +494,17 @@ fn awscurl_binary_path() -> PathBuf {
         .unwrap_or_else(|| PathBuf::from("awscurl"))
 }
 
+pub fn awscurl_available() -> bool {
+    let path = awscurl_binary_path();
+    if path.components().count() > 1 || path.is_absolute() {
+        return path.is_file();
+    }
+
+    std::env::var_os("PATH")
+        .map(|paths| std::env::split_paths(&paths).any(|dir| dir.join(&path).is_file()))
+        .unwrap_or(false)
+}
+
 // Global initialization
 static INIT: Once = Once::new();
 

crates/e2e_test/src/existing_object_tag_policy_test.rs

@@ -16,7 +16,9 @@
 //! session policy** (`Policy` parameter) via `awscurl --service sts` with explicit
 //! `Content-Type: application/x-www-form-urlencoded` on `POST /`.
 
-use crate::common::{RustFSTestEnvironment, awscurl_delete, awscurl_post_sts_form_urlencoded, awscurl_put, init_logging};
+use crate::common::{
+    RustFSTestEnvironment, awscurl_available, awscurl_delete, awscurl_post_sts_form_urlencoded, awscurl_put, init_logging,
+};
 use aws_sdk_s3::config::{Credentials, Region};
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::{Delete, ObjectIdentifier, Tag, Tagging};
@@ -173,6 +175,11 @@ async fn cleanup_bucket_and_object(admin: &Client, bucket: &str, key: &str) {
 #[tokio::test]
 async fn test_e2e_iam_policy_existing_object_tag_get_object() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if !awscurl_available() {
+        info!("Skipping test_e2e_iam_policy_existing_object_tag_get_object: awscurl not available");
+        return Ok(());
+    }
+
     let suffix = Uuid::new_v4();
     let user = format!("e2eiamtag-{suffix}");
     let user_secret = "longSecretKeyForTest123!";
@@ -226,6 +233,11 @@ async fn test_e2e_iam_policy_existing_object_tag_get_object() -> Result<(), Box<
 #[tokio::test]
 async fn test_e2e_bucket_policy_existing_object_tag_get_object() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if !awscurl_available() {
+        info!("Skipping test_e2e_bucket_policy_existing_object_tag_get_object: awscurl not available");
+        return Ok(());
+    }
+
     let suffix = Uuid::new_v4();
     let user = format!("e2ebptag-{suffix}");
     let user_secret = "longSecretKeyForTest456!";
@@ -282,6 +294,11 @@ async fn test_e2e_bucket_policy_existing_object_tag_get_object() -> Result<(), B
 #[tokio::test]
 async fn test_e2e_sts_assume_role_session_policy_existing_object_tag() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if !awscurl_available() {
+        info!("Skipping test_e2e_sts_assume_role_session_policy_existing_object_tag: awscurl not available");
+        return Ok(());
+    }
+
     let suffix = Uuid::new_v4();
     let parent = format!("e2e-sts-par-{suffix}");
     let parent_secret = "longSecretKeyForParentSts99!";
@@ -353,6 +370,11 @@ async fn test_e2e_sts_assume_role_session_policy_existing_object_tag() -> Result
 #[tokio::test]
 async fn test_e2e_sts_session_policy_delete_objects_object_prefix_only() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if !awscurl_available() {
+        info!("Skipping test_e2e_sts_session_policy_delete_objects_object_prefix_only: awscurl not available");
+        return Ok(());
+    }
+
     let suffix = Uuid::new_v4();
     let parent = format!("e2e-sts-del-par-{suffix}");
     let parent_secret = "longSecretKeyForParentDelete99!";

crates/e2e_test/src/kms/common.rs

@@ -22,7 +22,9 @@
 //! - KMS backend configuration (Local and Vault)
 //! - SSE encryption testing utilities
 
-use crate::common::{RustFSTestEnvironment, awscurl_get, awscurl_post, init_logging as common_init_logging, local_http_client};
+use crate::common::{
+    RustFSTestEnvironment, awscurl_available, awscurl_get, awscurl_post, init_logging as common_init_logging, local_http_client,
+};
 use aws_sdk_s3::Client;
 use aws_sdk_s3::primitives::ByteStream;
 use aws_sdk_s3::types::ServerSideEncryption;
@@ -57,6 +59,15 @@ pub fn init_logging() {
     // Additional KMS-specific logging configuration can be added here if needed
 }
 
+pub fn skip_if_kms_admin_tool_unavailable(test_name: &str) -> bool {
+    if awscurl_available() {
+        return false;
+    }
+
+    info!("Skipping {} because awscurl is not available in PATH", test_name);
+    true
+}
+
 pub fn sse_customer_key_md5_base64(key: &str) -> String {
     let mut hasher = Md5::new();
     hasher.update(key.as_bytes());
@@ -479,6 +490,10 @@ pub async fn test_kms_key_management(
     access_key: &str,
     secret_key: &str,
 ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    if skip_if_kms_admin_tool_unavailable("test_kms_key_management") {
+        return Ok(());
+    }
+
     info!("Testing KMS key management APIs");
 
     // Test CreateKey

crates/e2e_test/src/kms/kms_local_test.rs

@@ -20,7 +20,8 @@
 //! - Complete encryption/decryption lifecycle
 
 use super::common::{
-    LocalKMSTestEnvironment, get_kms_status, sse_customer_key_md5_base64, test_kms_key_management, test_sse_c_encryption,
+    LocalKMSTestEnvironment, get_kms_status, skip_if_kms_admin_tool_unavailable, sse_customer_key_md5_base64,
+    test_kms_key_management, test_sse_c_encryption,
 };
 use crate::common::{TEST_BUCKET, init_logging};
 use tracing::{error, info};
@@ -28,6 +29,9 @@ use tracing::{error, info};
 #[tokio::test]
 async fn test_local_kms_end_to_end() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if skip_if_kms_admin_tool_unavailable("test_local_kms_end_to_end") {
+        return Ok(());
+    }
     info!("Starting Local KMS End-to-End Test");
 
     // Create LocalKMS test environment

crates/e2e_test/src/kms/kms_vault_test.rs

@@ -22,8 +22,8 @@ use crate::common::{TEST_BUCKET, init_logging};
 use tracing::{error, info};
 
 use super::common::{
-    VAULT_KEY_NAME, VaultTestEnvironment, get_kms_status, sse_customer_key_md5_base64, start_kms,
-    test_all_multipart_encryption_types, test_error_scenarios, test_kms_key_management, test_sse_c_encryption,
+    VAULT_KEY_NAME, VaultTestEnvironment, get_kms_status, skip_if_kms_admin_tool_unavailable, sse_customer_key_md5_base64,
+    start_kms, test_all_multipart_encryption_types, test_error_scenarios, test_kms_key_management, test_sse_c_encryption,
     test_sse_kms_encryption, test_sse_s3_encryption,
 };
 
@@ -62,6 +62,9 @@ impl VaultKmsTestContext {
 #[tokio::test]
 async fn test_vault_kms_end_to_end() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if skip_if_kms_admin_tool_unavailable("test_vault_kms_end_to_end") {
+        return Ok(());
+    }
     info!("Starting Vault KMS End-to-End Test with default key {}", VAULT_KEY_NAME);
 
     let context = VaultKmsTestContext::new().await?;
@@ -114,6 +117,9 @@ async fn test_vault_kms_end_to_end() -> Result<(), Box<dyn std::error::Error + S
 #[tokio::test]
 async fn test_vault_kms_key_isolation() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if skip_if_kms_admin_tool_unavailable("test_vault_kms_key_isolation") {
+        return Ok(());
+    }
     info!("Starting Vault KMS SSE-C key isolation test");
 
     let context = VaultKmsTestContext::new().await?;
@@ -197,6 +203,9 @@ async fn test_vault_kms_key_isolation() -> Result<(), Box<dyn std::error::Error
 #[tokio::test]
 async fn test_vault_kms_large_file() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if skip_if_kms_admin_tool_unavailable("test_vault_kms_large_file") {
+        return Ok(());
+    }
     info!("Starting Vault KMS large file SSE-S3 test");
 
     let context = VaultKmsTestContext::new().await?;
@@ -258,6 +267,9 @@ async fn test_vault_kms_large_file() -> Result<(), Box<dyn std::error::Error + S
 #[tokio::test]
 async fn test_vault_kms_multipart_upload() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if skip_if_kms_admin_tool_unavailable("test_vault_kms_multipart_upload") {
+        return Ok(());
+    }
     info!("Starting Vault KMS multipart upload encryption suite");
 
     let context = VaultKmsTestContext::new().await?;
@@ -285,6 +297,9 @@ async fn test_vault_kms_multipart_upload() -> Result<(), Box<dyn std::error::Err
 #[tokio::test]
 async fn test_vault_kms_key_operations() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if skip_if_kms_admin_tool_unavailable("test_vault_kms_key_operations") {
+        return Ok(());
+    }
     info!("Starting Vault KMS key operations test (CRUD)");
 
     let context = VaultKmsTestContext::new().await?;

crates/e2e_test/src/mc_mirror_small_bucket_test.rs

@@ -41,6 +41,13 @@ async fn create_issue_3107_fixture(root: &Path) -> TestResult {
     Ok(())
 }
 
+fn mc_available() -> bool {
+    Command::new("mc")
+        .arg("--version")
+        .output()
+        .is_ok_and(|output| output.status.success())
+}
+
 fn run_mc(args: &[&str]) -> TestResult {
     let output = Command::new("mc").args(args).output()?;
     if !output.status.success() {
@@ -68,7 +75,10 @@ fn count_files(root: &Path) -> usize {
 async fn test_mc_mirror_small_bucket_completes_without_list_timeout() -> TestResult {
     crate::common::init_logging();
     info!("Starting issue #3107 mc mirror regression test");
-    run_mc(&["--version"])?;
+    if !mc_available() {
+        info!("Skipping issue #3107 mc mirror regression test because mc is not installed");
+        return Ok(());
+    }
 
     let mut env = RustFSTestEnvironment::new().await?;
     env.start_rustfs_server(vec![]).await?;

crates/e2e_test/src/multipart_auth_test.rs

@@ -4278,6 +4278,10 @@ async fn test_signed_put_object_extract_preserves_pax_metadata_and_version_id()
 async fn test_signed_put_object_extract_authorizes_each_pax_privilege_and_retention_conditions()
 -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
     init_logging();
+    if !crate::common::awscurl_available() {
+        return Ok(());
+    }
+
     let mut env = RustFSTestEnvironment::new().await?;
     env.start_rustfs_server(vec![]).await?;
 

crates/e2e_test/src/quota_test.rs

@@ -18,6 +18,15 @@ use http::{Method, StatusCode};
 use tokio::time::{Duration, sleep, timeout};
 use tracing::{debug, info};
 
+fn skip_without_awscurl() -> bool {
+    if crate::common::awscurl_available() {
+        return false;
+    }
+
+    info!("Skipping quota test because awscurl is not available");
+    true
+}
+
 /// Test environment setup for quota tests
 pub struct QuotaTestEnv {
     pub env: RustFSTestEnvironment,
@@ -267,6 +276,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_basic_operations() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         // Create test bucket
@@ -308,6 +320,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_admission_aws_chunked_declared_encoding() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
         env.create_bucket().await?;
 
@@ -356,6 +371,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_update_and_clear() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
@@ -388,6 +406,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_delete_operations() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
@@ -421,6 +442,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_usage_tracking() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
@@ -456,6 +480,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_statistics() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
@@ -486,6 +513,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_check_api() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
@@ -523,6 +553,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_multiple_buckets() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         // Create two buckets in the same environment
@@ -560,6 +593,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_error_handling() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
@@ -592,6 +628,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_http_endpoints() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
@@ -650,6 +689,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_normal_user_permissions() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
         env.create_bucket().await?;
 
@@ -702,6 +744,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_copy_operations() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
@@ -744,6 +789,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_batch_delete() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;
@@ -799,6 +847,9 @@ mod integration_tests {
     #[tokio::test]
     async fn test_quota_multipart_upload() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
         init_logging();
+        if skip_without_awscurl() {
+            return Ok(());
+        }
         let env = QuotaTestEnv::new().await?;
 
         env.create_bucket().await?;

crates/e2e_test/src/replication_extension_test.rs

@@ -13,8 +13,9 @@
 // limitations under the License.
 
 use crate::common::{
-    RustFSTestEnvironment, admin_create_user, awscurl_post_sts_form_urlencoded, init_logging, local_http_client,
-    replication_fast_env, rustfs_binary_path, signed_request, signed_request_with_client, signed_request_with_session_token,
+    RustFSTestEnvironment, admin_create_user, awscurl_available, awscurl_post_sts_form_urlencoded, init_logging,
+    local_http_client, replication_fast_env, rustfs_binary_path, signed_request, signed_request_with_client,
+    signed_request_with_session_token,
 };
 use crate::fake_s3_target::{
     FAKE_ACCESS_KEY, FAKE_SECRET_KEY, FakeS3Target, FaultAction as FakeTargetFault, Operation as FakeTargetOperation,
@@ -7280,6 +7281,11 @@ async fn test_site_replication_replicates_multiple_service_accounts_real_dual_no
 async fn test_site_replication_replicates_service_accounts_created_from_sts_session_real_dual_node() -> TestResult {
     init_logging();
 
+    if !awscurl_available() {
+        eprintln!("Skipping STS site replication service-account test because awscurl is unavailable");
+        return Ok(());
+    }
+
     let mut source_env = RustFSTestEnvironment::new().await?;
     source_env
         .start_rustfs_server_with_env(vec![], LOOPBACK_REPLICATION_TARGET_ENV)

crates/e2e_test/src/security_boundary_test.rs

@@ -21,7 +21,7 @@
 //! - SSRF prevention (internal/private endpoints rejected for tiering)
 //! - Race condition handling (concurrent writes converge without corruption)
 
-use crate::common::{RustFSTestEnvironment, awscurl_put, init_logging};
+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::{CompletedMultipartUpload, CompletedPart, Tag, Tagging};
@@ -225,11 +225,16 @@ async fn test_concurrent_object_operations() -> Result<(), Box<dyn Error + Send
 /// 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 full E2E
-/// lane installs and verifies the pinned `awscurl` prerequisite.
+/// pattern used by the other admin-API E2E tests in this crate; the test is
+/// skipped when `awscurl` is not installed.
 #[tokio::test]
 async fn test_tiering_url_validation() -> Result<(), Box<dyn Error + Send + Sync>> {
     init_logging();
+    if !awscurl_available() {
+        info!("Skipping tiering URL validation test because awscurl is not available");
+        return Ok(());
+    }
+
     let mut env = RustFSTestEnvironment::new().await?;
     env.start_rustfs_server(vec![]).await?;
 

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();