ci(mint): pin manual image default

* fix(scanner): isolate corrupt cycle state

* fix(scanner): preserve newer state during recovery reset

* fix(scanner): fence recovery reset state

* fix(scanner): reject terminal recovery epochs

* fix(scanner): reject trailing cycle state bytes

* fix(scanner): reject terminal leadership epochs

* fix(scanner): retain recovery wake notifications

* fix(scanner): recover from oversized markers

.github/workflows/mint.yml

@@ -45,13 +45,6 @@
 # docker-capable self-hosted `dind-sm-standard-2` label was the alternative but
 # has fewer cores and reintroduces fleet-state risk for no reliability gain.
 
-# DISABLED. This workflow is switched off in the repository's Actions settings
-# (state: disabled_manually) and does not run on any trigger, including its cron
-# and workflow_dispatch. That state lives in GitHub's UI and is invisible when
-# reading this file, which has already misled at least one audit — hence this
-# banner. Re-enabling is a UI action; anyone doing so should first check that the
-# workflow still matches the current CI layout. See rustfs/backlog#1603.
-#
 name: mint
 
 on:
@@ -70,9 +63,9 @@ on:
           - core
           - full
       mint-image:
-        description: "Mint image reference"
+        description: "Mint image reference (empty = pinned default)"
         required: false
-        default: "minio/mint:edge"
+        default: ""
   schedule:
     # Weekly, after the Sunday s3-tests full sweep (starts 02:00 UTC, up to
     # 3h) has finished, so the two never contend for the same runner pool.

crates/ecstore/src/api/mod.rs

@@ -317,8 +317,6 @@ pub mod config {
 }
 
 pub mod data_usage {
-    #[cfg(feature = "test-util")]
-    pub use crate::data_usage::seed_bucket_usage_memory_for_test;
     pub use crate::data_usage::{
         DATA_USAGE_CACHE_NAME, apply_bucket_usage_memory_overlay, compute_bucket_usage,
         init_compression_total_memory_from_backend, invalidate_admin_data_usage_snapshot_cache,
@@ -330,6 +328,8 @@ pub mod data_usage {
         remove_bucket_usage_from_backend, replace_bucket_usage_memory_from_info, store_compression_total_in_backend,
         store_data_usage_in_backend,
     };
+    #[cfg(feature = "test-util")]
+    pub use crate::data_usage::{get_bucket_usage_memory, seed_bucket_usage_memory_for_test};
 }
 
 pub mod disk {

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

@@ -2855,7 +2855,7 @@ fn replicate_object_info_from_object_info(
         .map(|v| OffsetDateTime::parse(&v, &Rfc3339).unwrap_or(OffsetDateTime::UNIX_EPOCH));
     let mut rstate = oi.replication_state();
     rstate.replicate_decision_str = dsc.to_string();
-    let asz = oi.get_actual_size().unwrap_or_default();
+    let asz = oi.get_actual_size_or_physical();
     let ssec = replication_object_is_ssec_encrypted(&oi.user_defined);
     let checksum = if ssec { oi.checksum.clone() } else { None };
 

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

@@ -1412,7 +1412,7 @@ pub async fn get_heal_replicate_object_info(oi: &ObjectInfo, rcfg: &ReplicationC
     };
     let mut replication_state = oi.replication_state();
     replication_state.replicate_decision_str = dsc.to_string();
-    let actual_size = oi.get_actual_size().unwrap_or_default();
+    let actual_size = oi.get_actual_size_or_physical();
 
     Ok(ReplicateObjectInfo {
         name: oi.name.clone(),

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

@@ -389,7 +389,7 @@ fn replication_source_object(object_info: &ObjectInfo) -> ReplicationSourceObjec
             .map(|mod_time| OffsetDateTime::from_unix_timestamp(mod_time.unix_timestamp()).unwrap_or(mod_time)),
         version_id: object_info.version_id.map(|version_id| version_id.to_string()),
         etag: object_info.etag.as_deref(),
-        actual_size: object_info.get_actual_size().unwrap_or_default(),
+        actual_size: object_info.get_actual_size_or_physical(),
         delete_marker: object_info.delete_marker,
         content_type: object_info.content_type.as_deref(),
         content_encoding: object_info.content_encoding.as_deref(),
@@ -542,6 +542,20 @@ mod tests {
         assert!(replication_target_head_is_newer_null_version(&source, &target));
     }
 
+    #[test]
+    fn replication_source_uses_physical_size_for_unknown_compressed_object() {
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+        let source = ObjectInfo {
+            size: 128,
+            actual_size: -1,
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+
+        assert_eq!(replication_source_object(&source).actual_size, 128);
+    }
+
     #[test]
     fn replication_target_head_content_matches_compare_etag_only() {
         let source = ObjectInfo {

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

@@ -50,10 +50,10 @@ use rustfs_protos::evict_failed_connection;
 use rustfs_protos::proto_gen::node_service::RenamePartRequest;
 use rustfs_protos::proto_gen::node_service::{
     BatchReadVersionRequest, BatchReadVersionResponse, CheckPartsRequest, DeletePathsRequest, DeleteRequest,
-    DeleteVersionRequest, DeleteVersionsRequest, DeleteVolumeRequest, DiskInfoRequest, ListDirRequest, ListVolumesRequest,
-    MakeVolumeRequest, MakeVolumesRequest, PreparePartTransactionRequest, ReadAllRequest, ReadMetadataRequest,
-    ReadMultipleRequest, ReadMultipleResponse, ReadPartsRequest, ReadVersionRequest, ReadXlRequest, RenameDataRequest,
-    RenameFileRequest, SettlePartTransactionRequest, SnapshotLeaseReleaseRequest, SnapshotLeaseRenewRequest,
+    DeleteVersionRequest, DeleteVersionsRequest, DeleteVersionsResponse, DeleteVolumeRequest, DiskInfoRequest, ListDirRequest,
+    ListVolumesRequest, MakeVolumeRequest, MakeVolumesRequest, PreparePartTransactionRequest, ReadAllRequest,
+    ReadMetadataRequest, ReadMultipleRequest, ReadMultipleResponse, ReadPartsRequest, ReadVersionRequest, ReadXlRequest,
+    RenameDataRequest, RenameFileRequest, SettlePartTransactionRequest, SnapshotLeaseReleaseRequest, SnapshotLeaseRenewRequest,
     SnapshotLeaseRequest, SnapshotLeaseResponse, StatVolumeRequest, UpdateMetadataRequest, VerifyFileRequest, WriteAllRequest,
     WriteMetadataRequest, node_service_client::NodeServiceClient,
 };
@@ -112,6 +112,28 @@ const EVENT_REMOTE_DISK_RPC: &str = "remote_disk_rpc";
 const SNAPSHOT_LEASE_PROTOCOL_VERSION: u32 = 1;
 pub const REMOTE_SNAPSHOT_LEASE_TTL: Duration = Duration::from_secs(60);
 
+fn decode_delete_versions_errors(response: DeleteVersionsResponse, expected_len: usize) -> Vec<Option<Error>> {
+    if !response.item_errors.is_empty() {
+        if response.item_errors.len() != expected_len {
+            return vec![Some(Error::other("malformed delete_versions item errors")); expected_len];
+        }
+        return response
+            .item_errors
+            .into_iter()
+            .map(|error| (error.code != 0).then(|| error.into()))
+            .collect();
+    }
+
+    if response.errors.len() != expected_len {
+        return vec![Some(Error::other("malformed delete_versions errors")); expected_len];
+    }
+    response
+        .errors
+        .into_iter()
+        .map(|error| (!error.is_empty()).then(|| Error::other(error)))
+        .collect()
+}
+
 fn snapshot_lease_token_from_response(response: SnapshotLeaseResponse) -> Result<SnapshotLeaseToken> {
     if !response.success {
         return Err(response.error.unwrap_or_default().into());
@@ -2406,8 +2428,6 @@ impl DiskAPI for RemoteDisk {
             return errors;
         }
 
-        // TODO(backlog): replace string errors with typed `StorageError` variants
-
         let result = self
             .execute_with_timeout(
                 || async {
@@ -2439,17 +2459,7 @@ impl DiskAPI for RemoteDisk {
             }
             return errors;
         }
-        response
-            .errors
-            .iter()
-            .map(|error| {
-                if error.is_empty() {
-                    None
-                } else {
-                    Some(Error::other(error.to_string()))
-                }
-            })
-            .collect()
+        decode_delete_versions_errors(response, versions.len())
     }
 
     #[tracing::instrument(level = "trace", skip_all)]
@@ -3760,6 +3770,63 @@ mod tests {
 
     static INIT: Once = Once::new();
 
+    #[test]
+    fn delete_versions_response_preserves_typed_item_errors() {
+        let errors = decode_delete_versions_errors(
+            DeleteVersionsResponse {
+                success: true,
+                errors: vec!["file not found".to_string(), String::new()],
+                error: None,
+                item_errors: vec![
+                    rustfs_protos::proto_gen::node_service::Error {
+                        code: DiskError::FileNotFound.to_u32(),
+                        error_info: "file not found".to_string(),
+                    },
+                    rustfs_protos::proto_gen::node_service::Error::default(),
+                ],
+            },
+            2,
+        );
+
+        assert!(matches!(errors.as_slice(), [Some(DiskError::FileNotFound), None]));
+    }
+
+    #[test]
+    fn delete_versions_response_accepts_legacy_string_errors() {
+        let errors = decode_delete_versions_errors(
+            DeleteVersionsResponse {
+                success: true,
+                errors: vec!["legacy error".to_string(), String::new()],
+                error: None,
+                item_errors: Vec::new(),
+            },
+            2,
+        );
+
+        assert_eq!(errors.len(), 2);
+        assert_eq!(errors[0].as_ref().map(ToString::to_string).as_deref(), Some("io error legacy error"));
+        assert!(errors[1].is_none());
+    }
+
+    #[test]
+    fn delete_versions_response_rejects_misaligned_item_errors() {
+        let errors = decode_delete_versions_errors(
+            DeleteVersionsResponse {
+                success: true,
+                errors: vec!["file not found".to_string()],
+                error: None,
+                item_errors: vec![rustfs_protos::proto_gen::node_service::Error {
+                    code: DiskError::FileNotFound.to_u32(),
+                    error_info: "file not found".to_string(),
+                }],
+            },
+            2,
+        );
+
+        assert_eq!(errors.len(), 2);
+        assert!(errors.iter().all(Option::is_some));
+    }
+
     #[test]
     fn disk_mutation_digest_marks_rolling_compatibility() {
         let mut request = Request::new(());

crates/ecstore/src/core/pools.rs

@@ -3416,9 +3416,6 @@ impl ECStore {
             )
             .await?;
 
-            let source_cleanup_mutation_fence = self
-                .acquire_decommission_source_cleanup_fence(bucket.as_str(), entry.name.as_str(), set.as_ref())
-                .await?;
             let cleanup_result = data_movement::cleanup_source_entry_if_unchanged(
                 set.clone(),
                 bucket.as_str(),
@@ -3430,7 +3427,6 @@ impl ECStore {
                     lifecycle_guard: bucket_incarnation_fence
                         .as_ref()
                         .and_then(|guard| guard.namespace_lock_guard()),
-                    object_mutation_fence: Some(&source_cleanup_mutation_fence),
                 },
                 "decommission",
             )
@@ -3532,22 +3528,6 @@ impl ECStore {
         Ok(())
     }
 
-    #[cfg(test)]
-    pub(crate) async fn decommission_entry_for_test(
-        self: &Arc<Self>,
-        idx: usize,
-        entry: MetaCacheEntry,
-        bucket: String,
-        set: Arc<SetDisks>,
-    ) -> Result<()> {
-        let worker_permit = Arc::new(Semaphore::new(1))
-            .acquire_owned()
-            .await
-            .map_err(|err| Error::other(format!("decommission test worker permit acquire failed: {err}")))?;
-        self.decommission_entry(CancellationToken::new(), idx, entry, bucket, set, worker_permit, None, None, None, None)
-            .await
-    }
-
     #[tracing::instrument(skip(self, rx))]
     async fn decommission_pool(
         self: &Arc<Self>,
@@ -4484,20 +4464,15 @@ impl ECStore {
     ) -> Result<()> {
         warn!("decommission_object: start {} {}", &bucket, &rd.object_info.name);
         let object_name = rd.object_info.name.clone();
-        let mut migration = tokio::task::JoinSet::new();
-        migration.spawn(data_movement::migrate_decommission_object(
+        let result = data_movement::migrate_object(
             self,
             pool_idx,
             bucket.clone(),
             rd,
             expected_bucket_incarnation_id,
             "decommission_object",
-        ));
-        let result = migration
-            .join_next()
-            .await
-            .ok_or_else(|| Error::other("decommission migration task was not started"))?
-            .map_err(|err| Error::other(format!("decommission migration task join error: {err}")))?;
+        )
+        .await;
         if result.is_ok() {
             warn!("decommission_object: migrated {} {}", &bucket, &object_name);
         }

crates/ecstore/src/core/sets.rs

@@ -21,7 +21,7 @@ use crate::storage_api_contracts::{
     bucket::{BucketInfo, BucketOperations, BucketOptions, DeleteBucketOptions, MakeBucketOptions},
     list::{StorageListObjectVersionsInfo, StorageListObjectsV2Info, StorageObjectInfoOrErr, StorageWalkOptions},
     multipart::{CompletePart, ListMultipartsInfo, ListPartsInfo, MultipartInfo, MultipartUploadResult, PartInfo},
-    object::{DeletedObject, ObjectIO as _, ObjectOperations as _, ObjectToDelete},
+    object::{DeleteAccounting, DeletedObject, ObjectIO as _, ObjectOperations as _, ObjectToDelete},
     range::HTTPRangeSpec,
 };
 use crate::{
@@ -414,6 +414,66 @@ fn apply_delete_objects_results(
     }
 }
 
+fn apply_delete_accounting_results(
+    accounting: &mut [Option<DeleteAccounting>],
+    set_objects: &[DelObj],
+    set_accounting: &[Option<DeleteAccounting>],
+) {
+    for (obj, value) in set_objects.iter().zip(set_accounting.iter()) {
+        accounting[obj.orig_idx] = value.clone();
+    }
+}
+
+impl Sets {
+    pub(crate) async fn delete_objects_with_accounting(
+        &self,
+        bucket: &str,
+        objects: Vec<ObjectToDelete>,
+        opts: ObjectOptions,
+    ) -> (Vec<DeletedObject>, Vec<Option<Error>>, Vec<Option<DeleteAccounting>>) {
+        let mut del_objects = vec![DeletedObject::default(); objects.len()];
+        let mut del_errs = vec![None; objects.len()];
+        let mut accounting = vec![None; objects.len()];
+        let mut set_obj_map = HashMap::new();
+
+        for (i, obj) in objects.iter().enumerate() {
+            let idx = self.get_hashed_set_index(obj.object_name.as_str());
+            set_obj_map.entry(idx).or_insert_with(Vec::new).push(DelObj {
+                orig_idx: i,
+                obj: obj.clone(),
+            });
+        }
+
+        let max_concurrent = set_obj_map.len().min(num_cpus::get()).max(1);
+        let semaphore = Arc::new(tokio::sync::Semaphore::new(max_concurrent));
+        let mut futures = FuturesUnordered::new();
+        let bucket = bucket.to_owned();
+
+        for (set_index, set_objects) in set_obj_map {
+            let disks = self.get_disks(set_index);
+            let objects = set_objects.iter().map(|entry| entry.obj.clone()).collect::<Vec<_>>();
+            let bucket = bucket.clone();
+            let opts = opts.clone();
+            let semaphore = semaphore.clone();
+            futures.push(async move {
+                let _permit = semaphore
+                    .acquire_owned()
+                    .await
+                    .expect("delete_objects semaphore should remain open");
+                let (deleted, errors, accounting) = disks.delete_objects_with_accounting(&bucket, objects, opts).await;
+                (set_objects, deleted, errors, accounting)
+            });
+        }
+
+        while let Some((set_objects, deleted, errors, set_accounting)) = futures.next().await {
+            apply_delete_objects_results(&mut del_objects, &mut del_errs, &set_objects, &deleted, errors);
+            apply_delete_accounting_results(&mut accounting, &set_objects, &set_accounting);
+        }
+
+        (del_objects, del_errs, accounting)
+    }
+}
+
 #[async_trait::async_trait]
 impl crate::storage_api_contracts::object::ObjectIO for Sets {
     type Error = Error;
@@ -655,65 +715,8 @@ impl crate::storage_api_contracts::object::ObjectOperations for Sets {
         objects: Vec<ObjectToDelete>,
         opts: ObjectOptions,
     ) -> (Vec<DeletedObject>, Vec<Option<Error>>) {
-        // Default return value
-        let mut del_objects = vec![DeletedObject::default(); objects.len()];
-
-        let mut del_errs = Vec::with_capacity(objects.len());
-        for _ in 0..objects.len() {
-            del_errs.push(None)
-        }
-
-        let mut set_obj_map = HashMap::new();
-
-        // hash key
-        for (i, obj) in objects.iter().enumerate() {
-            let idx = self.get_hashed_set_index(obj.object_name.as_str());
-
-            if !set_obj_map.contains_key(&idx) {
-                set_obj_map.insert(
-                    idx,
-                    vec![DelObj {
-                        // set_idx: idx,
-                        orig_idx: i,
-                        obj: obj.clone(),
-                    }],
-                );
-            } else if let Some(val) = set_obj_map.get_mut(&idx) {
-                val.push(DelObj {
-                    // set_idx: idx,
-                    orig_idx: i,
-                    obj: obj.clone(),
-                });
-            }
-        }
-
-        let max_concurrent = set_obj_map.len().min(num_cpus::get()).max(1);
-        let semaphore = Arc::new(tokio::sync::Semaphore::new(max_concurrent));
-        let mut futures = FuturesUnordered::new();
-        let bucket = bucket.to_string();
-
-        for (k, v) in set_obj_map {
-            let disks = self.get_disks(k);
-            let objs: Vec<ObjectToDelete> = v.iter().map(|v| v.obj.clone()).collect();
-            let bucket = bucket.clone();
-            let opts = opts.clone();
-            let semaphore = semaphore.clone();
-
-            futures.push(async move {
-                let _permit = semaphore
-                    .acquire_owned()
-                    .await
-                    .expect("delete_objects semaphore should remain open");
-                let (dobjects, errs) = disks.delete_objects(&bucket, objs, opts).await;
-                (v, dobjects, errs)
-            });
-        }
-
-        while let Some((v, dobjects, errs)) = futures.next().await {
-            apply_delete_objects_results(&mut del_objects, &mut del_errs, &v, &dobjects, errs);
-        }
-
-        (del_objects, del_errs)
+        let (deleted, errors, _) = self.delete_objects_with_accounting(bucket, objects, opts).await;
+        (deleted, errors)
     }
 
     #[tracing::instrument(skip(self))]

crates/ecstore/src/data_movement/mod.rs

@@ -26,7 +26,7 @@ use crate::storage_api_contracts::{
     namespace::NamespaceLocking as _,
     object::{HTTPPreconditions, ObjectOperations as _},
 };
-use crate::store::{ECStore, ObjectLockDiagGuard, SourceCleanupMutationFence};
+use crate::store::ECStore;
 use bytes::Bytes;
 use rustfs_filemeta::{FileInfo, FileInfoVersions, ObjectPartInfo};
 use rustfs_rio::{EtagResolvable, HashReader, HashReaderDetector, Index, TryGetIndex};
@@ -856,6 +856,7 @@ fn is_equivalent_data_movement_object(source: &ObjectInfo, target: &ObjectInfo)
 fn is_superseding_unversioned_data_movement_object(source: &ObjectInfo, target: &ObjectInfo) -> bool {
     is_unversioned_data_movement_object(source)
         && is_unversioned_data_movement_object(target)
+        && !target.delete_marker
         && source
             .mod_time
             .zip(target.mod_time)
@@ -1027,7 +1028,6 @@ pub(crate) enum SourceCleanupError {
 pub(crate) struct SourceCleanupBucketFence<'a> {
     pub(crate) expected_incarnation_id: Option<uuid::Uuid>,
     pub(crate) lifecycle_guard: Option<&'a rustfs_lock::NamespaceLockGuard>,
-    pub(crate) object_mutation_fence: Option<&'a SourceCleanupMutationFence>,
 }
 
 fn ensure_source_cleanup_versions_match(
@@ -1065,9 +1065,7 @@ pub(crate) async fn ensure_source_cleanup_versions_unchanged(
 struct SourceCleanupDeleteBarrierState {
     bucket: String,
     object: String,
-    fence_pending: tokio::sync::Notify,
     arrived: tokio::sync::Notify,
-    is_paused: AtomicBool,
     release: tokio::sync::Notify,
 }
 
@@ -1081,7 +1079,7 @@ pub(crate) struct SourceCleanupDeleteBarrier {
 }
 
 #[cfg(test)]
-static SOURCE_CLEANUP_DELETE_BARRIERS: std::sync::OnceLock<std::sync::Mutex<Vec<Arc<SourceCleanupDeleteBarrierState>>>> =
+static SOURCE_CLEANUP_DELETE_BARRIER: std::sync::OnceLock<std::sync::Mutex<Option<Arc<SourceCleanupDeleteBarrierState>>>> =
     std::sync::OnceLock::new();
 
 #[cfg(test)]
@@ -1094,22 +1092,15 @@ impl SourceCleanupDeleteBarrier {
         let state = Arc::new(SourceCleanupDeleteBarrierState {
             bucket: bucket.to_string(),
             object: object.to_string(),
-            fence_pending: tokio::sync::Notify::new(),
             arrived: tokio::sync::Notify::new(),
-            is_paused: AtomicBool::new(false),
             release: tokio::sync::Notify::new(),
         });
-        let mut barriers = SOURCE_CLEANUP_DELETE_BARRIERS
-            .get_or_init(|| std::sync::Mutex::new(Vec::new()))
+        let mut slot = SOURCE_CLEANUP_DELETE_BARRIER
+            .get_or_init(|| std::sync::Mutex::new(None))
             .lock()
             .expect("source cleanup delete barrier mutex should not poison");
-        assert!(
-            !barriers
-                .iter()
-                .any(|barrier| barrier.bucket == bucket && barrier.object == object),
-            "source cleanup delete barrier must be unique per object"
-        );
-        barriers.push(Arc::clone(&state));
+        assert!(slot.is_none(), "source cleanup delete barrier must be unique");
+        *slot = Some(Arc::clone(&state));
         Self { state }
     }
 
@@ -1119,58 +1110,35 @@ impl SourceCleanupDeleteBarrier {
             .expect("source cleanup should reach the pre-delete barrier");
     }
 
-    pub(crate) async fn wait_until_fence_pending(&self) {
-        tokio::time::timeout(StdDuration::from_secs(30), self.state.fence_pending.notified())
-            .await
-            .expect("source cleanup should attempt the fixed mutation fence");
-    }
-
-    pub(crate) fn is_paused(&self) -> bool {
-        self.state.is_paused.load(Ordering::Acquire)
-    }
-
     pub(crate) fn release(&self) {
         self.state.release.notify_one();
     }
 }
 
-#[cfg(test)]
-pub(crate) fn notify_source_cleanup_mutation_fence_pending(bucket: &str, object: &str) {
-    let barrier = SOURCE_CLEANUP_DELETE_BARRIERS
-        .get_or_init(|| std::sync::Mutex::new(Vec::new()))
-        .lock()
-        .expect("source cleanup delete barrier mutex should not poison")
-        .iter()
-        .find(|barrier| barrier.bucket == bucket && barrier.object == object)
-        .cloned();
-    if let Some(barrier) = barrier {
-        barrier.fence_pending.notify_one();
-    }
-}
-
 #[cfg(test)]
 impl Drop for SourceCleanupDeleteBarrier {
     fn drop(&mut self) {
         self.state.release.notify_one();
-        let mut barriers = SOURCE_CLEANUP_DELETE_BARRIERS
-            .get_or_init(|| std::sync::Mutex::new(Vec::new()))
+        let mut slot = SOURCE_CLEANUP_DELETE_BARRIER
+            .get_or_init(|| std::sync::Mutex::new(None))
             .lock()
             .expect("source cleanup delete barrier mutex should not poison");
-        barriers.retain(|state| !Arc::ptr_eq(state, &self.state));
+        if slot.as_ref().is_some_and(|state| Arc::ptr_eq(state, &self.state)) {
+            *slot = None;
+        }
     }
 }
 
 #[cfg(test)]
 async fn pause_source_cleanup_before_delete(bucket: &str, object: &str) {
-    let barrier = SOURCE_CLEANUP_DELETE_BARRIERS
-        .get_or_init(|| std::sync::Mutex::new(Vec::new()))
+    let barrier = SOURCE_CLEANUP_DELETE_BARRIER
+        .get_or_init(|| std::sync::Mutex::new(None))
         .lock()
         .expect("source cleanup delete barrier mutex should not poison")
-        .iter()
-        .find(|barrier| barrier.bucket == bucket && barrier.object == object)
+        .as_ref()
+        .filter(|barrier| barrier.bucket == bucket && barrier.object == object)
         .cloned();
     if let Some(barrier) = barrier {
-        barrier.is_paused.store(true, Ordering::Release);
         barrier.arrived.notify_one();
         barrier.release.notified().await;
     }
@@ -1186,20 +1154,11 @@ pub(crate) async fn cleanup_source_entry_if_unchanged(
     op_label: &str,
 ) -> std::result::Result<ObjectInfo, SourceCleanupError> {
     let cleanup_key = encode_dir_object(object);
-    let source_guard = if bucket_fence
-        .object_mutation_fence
-        .is_some_and(SourceCleanupMutationFence::source_lock_covered)
-    {
-        None
-    } else {
-        let ns_lock = set.new_ns_lock(bucket, cleanup_key.as_str()).await?;
-        Some(
-            ns_lock
-                .get_write_lock(get_lock_acquire_timeout())
-                .await
-                .map_err(Error::from)?,
-        )
-    };
+    let ns_lock = set.new_ns_lock(bucket, cleanup_key.as_str()).await?;
+    let _guard = ns_lock
+        .get_write_lock(get_lock_acquire_timeout())
+        .await
+        .map_err(Error::from)?;
 
     if bucket_fence
         .lifecycle_guard
@@ -1209,14 +1168,6 @@ pub(crate) async fn cleanup_source_entry_if_unchanged(
             "{op_label}: bucket incarnation fence was lost before source cleanup"
         ))));
     }
-    if bucket_fence
-        .object_mutation_fence
-        .is_some_and(SourceCleanupMutationFence::is_lock_lost)
-    {
-        return Err(SourceCleanupError::Storage(Error::other(format!(
-            "{op_label}: object mutation fence was lost before source cleanup"
-        ))));
-    }
 
     ensure_source_cleanup_versions_unchanged(set.clone(), bucket, object, expected, allowed_missing, op_label).await?;
 
@@ -1231,12 +1182,7 @@ pub(crate) async fn cleanup_source_entry_if_unchanged(
         expected_bucket_incarnation_id: bucket_fence.expected_incarnation_id,
         ..Default::default()
     };
-    if let Some(source_guard) = source_guard.as_ref() {
-        opts.add_namespace_lock_guard(source_guard);
-    }
-    if let Some(object_mutation_fence) = bucket_fence.object_mutation_fence {
-        object_mutation_fence.add_namespace_lock_fence(&mut opts);
-    }
+    opts.add_namespace_lock_guard(&_guard);
     if let Some(bucket_lifecycle_guard) = bucket_fence.lifecycle_guard {
         opts.add_bucket_lifecycle_lock_guard(bucket_lifecycle_guard);
     }
@@ -1384,37 +1330,6 @@ fn data_movement_part_upload_failure_stage(err: &Error) -> &'static str {
     }
 }
 
-pub(crate) async fn migrate_decommission_object(
-    store: Arc<ECStore>,
-    pool_idx: usize,
-    bucket: String,
-    rd: GetObjectReader,
-    source_bucket_incarnation_id: Option<uuid::Uuid>,
-    op_label: &str,
-) -> Result<()> {
-    let source = rd.object_info.clone();
-    let _mutation_fence = store
-        .acquire_decommission_object_mutation_fence(&bucket, &source.name)
-        .await?;
-    let current = find_data_movement_target_info(store.as_ref(), pool_idx, &bucket, &source)
-        .await?
-        .ok_or(Error::FileNotFound)?;
-    if !is_equivalent_data_movement_object_identity(&source, &current, true, false) {
-        return Err(Error::FileNotFound);
-    }
-
-    migrate_object_inner(
-        store,
-        pool_idx,
-        bucket,
-        rd,
-        source_bucket_incarnation_id,
-        op_label,
-        Some(&_mutation_fence),
-    )
-    .await
-}
-
 pub(crate) async fn migrate_object(
     store: Arc<ECStore>,
     pool_idx: usize,
@@ -1422,18 +1337,6 @@ pub(crate) async fn migrate_object(
     rd: GetObjectReader,
     source_bucket_incarnation_id: Option<uuid::Uuid>,
     op_label: &str,
-) -> Result<()> {
-    migrate_object_inner(store, pool_idx, bucket, rd, source_bucket_incarnation_id, op_label, None).await
-}
-
-async fn migrate_object_inner(
-    store: Arc<ECStore>,
-    pool_idx: usize,
-    bucket: String,
-    rd: GetObjectReader,
-    source_bucket_incarnation_id: Option<uuid::Uuid>,
-    op_label: &str,
-    mutation_fence: Option<&ObjectLockDiagGuard>,
 ) -> Result<()> {
     let object_info = rd.object_info.clone();
     let has_part_checksums = object_info
@@ -1447,7 +1350,7 @@ async fn migrate_object_inner(
         let mut new_multipart_opts = data_movement_new_multipart_opts(&object_info, pool_idx);
         new_multipart_opts.expected_bucket_incarnation_id = source_bucket_incarnation_id;
         let (res, target_pool_idx, expected_bucket_incarnation_id) = match store
-            .handle_new_multipart_upload_with_pool_idx(&bucket, &object_info.name, &new_multipart_opts, mutation_fence)
+            .handle_new_multipart_upload_with_pool_idx(&bucket, &object_info.name, &new_multipart_opts)
             .await
         {
             Ok(res) => res,
@@ -1545,7 +1448,7 @@ async fn migrate_object_inner(
             if let Err(err) = store
                 .clone()
                 .complete_multipart_upload_for_data_movement(
-                    (target_pool_idx, mutation_fence),
+                    target_pool_idx,
                     &bucket,
                     &object_info.name,
                     &res.upload_id,
@@ -1706,7 +1609,7 @@ async fn migrate_object_inner(
     let mut put_opts = data_movement_put_object_opts(&object_info, pool_idx);
     put_opts.expected_bucket_incarnation_id = source_bucket_incarnation_id;
     let (target_pool_idx, put_result) = store
-        .put_object_for_data_movement(&bucket, &object_info.name, &mut data, &put_opts, mutation_fence)
+        .put_object_for_data_movement(&bucket, &object_info.name, &mut data, &put_opts)
         .await
         .map_err(|err| data_movement_stage_error(op_label, "prepare_put_object", &bucket, &object_info.name, err))?;
     if let Err(err) = put_result {
@@ -3638,47 +3541,25 @@ mod tests {
     }
 
     #[test]
-    fn test_precondition_conflict_accepts_only_newer_null_delete_marker() {
-        for version_id in [None, Some(Uuid::nil())] {
-            let source = ObjectInfo {
-                version_id,
-                size: 128,
-                etag: Some("etag-source".to_string()),
-                mod_time: Some(OffsetDateTime::UNIX_EPOCH),
-                ..Default::default()
-            };
-            let target = ObjectInfo {
-                delete_marker: true,
-                etag: None,
-                mod_time: OffsetDateTime::UNIX_EPOCH.checked_add(time::Duration::SECOND),
-                ..source.clone()
-            };
+    fn test_precondition_conflict_rejects_newer_delete_marker() {
+        let source = ObjectInfo {
+            size: 128,
+            etag: Some("etag-source".to_string()),
+            mod_time: Some(OffsetDateTime::UNIX_EPOCH),
+            ..Default::default()
+        };
+        let target = ObjectInfo {
+            delete_marker: true,
+            etag: None,
+            mod_time: OffsetDateTime::UNIX_EPOCH.checked_add(time::Duration::SECOND),
+            ..source.clone()
+        };
 
-            assert!(
-                resolve_data_movement_overwrite_resume_result(
-                    &Error::PreconditionFailed,
-                    Ok(Some(target.clone())),
-                    &source,
-                    0,
-                    1,
-                )
-                .expect("newer null delete marker should be evaluated")
-            );
+        let should_resume =
+            resolve_data_movement_overwrite_resume_result(&Error::PreconditionFailed, Ok(Some(target)), &source, 0, 1)
+                .expect("delete marker conflict should be evaluated");
 
-            let mut same_time = target.clone();
-            same_time.mod_time = source.mod_time;
-            assert!(
-                !resolve_data_movement_overwrite_resume_result(&Error::PreconditionFailed, Ok(Some(same_time)), &source, 0, 1,)
-                    .expect("same-generation null delete marker should be rejected")
-            );
-
-            let mut versioned = target;
-            versioned.version_id = Some(Uuid::new_v4());
-            assert!(
-                !resolve_data_movement_overwrite_resume_result(&Error::PreconditionFailed, Ok(Some(versioned)), &source, 0, 1,)
-                    .expect("a UUID delete marker must not erase a null source version")
-            );
-        }
+        assert!(!should_resume);
     }
 
     #[test]

crates/ecstore/src/data_usage/mod.rs

@@ -1391,7 +1391,37 @@ impl BucketUsageAccumulator {
 }
 
 pub fn quota_object_size(object: &ObjectInfo) -> Result<u64, Error> {
-    let logical_size = u64::try_from(object.get_actual_size().map_err(Error::other)?).map_err(|_| Error::PartMissingOrCorrupt)?;
+    // A compressed object may carry -1 while the transformed size is unknown
+    // (legacy streaming sentinel). In that case the persisted physical size
+    // is still a valid accounting floor; every other negative value is corrupt.
+    // An explicit negative `actual-size` metadata value is corrupt, however:
+    // the sentinel is only valid in the in-memory/object-part field written by
+    // the legacy streaming path, not as a persisted declared size.
+    let compressed = object.is_compressed();
+    if object.actual_size < -1 || (object.actual_size == -1 && !compressed) {
+        return Err(Error::PartMissingOrCorrupt);
+    }
+    if object
+        .parts
+        .iter()
+        .any(|part| part.actual_size < -1 || (part.actual_size < 0 && !compressed))
+    {
+        return Err(Error::PartMissingOrCorrupt);
+    }
+    let declared_actual_size = rustfs_utils::http::get_str(&object.user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE)
+        .filter(|value| !value.is_empty());
+    if declared_actual_size
+        .as_deref()
+        .and_then(|value| value.parse::<i64>().ok())
+        .is_some_and(|size| size < 0)
+    {
+        return Err(Error::PartMissingOrCorrupt);
+    }
+    let logical_size = match object.get_actual_size().map_err(Error::other)? {
+        size if size == -1 && compressed && declared_actual_size.is_none() => None,
+        size if size >= 0 => Some(u64::try_from(size).map_err(|_| Error::PartMissingOrCorrupt)?),
+        _ => return Err(Error::PartMissingOrCorrupt),
+    };
     let persisted_part_size = if object.parts.is_empty() {
         u64::try_from(object.size).map_err(|_| Error::PartMissingOrCorrupt)?
     } else {
@@ -1399,12 +1429,8 @@ pub fn quota_object_size(object: &ObjectInfo) -> Result<u64, Error> {
             // Compressed streaming objects persist -1 when the transformed
             // part size is unknown. The physical part size remains a valid
             // quota floor; reject only non-negative values that overflow.
-            let actual_size = if part.actual_size < 0 {
-                if object.is_compressed() {
-                    0
-                } else {
-                    return Err(Error::PartMissingOrCorrupt);
-                }
+            let actual_size = if part.actual_size == -1 {
+                0
             } else {
                 u64::try_from(part.actual_size).map_err(|_| Error::PartMissingOrCorrupt)?
             };
@@ -1412,7 +1438,7 @@ pub fn quota_object_size(object: &ObjectInfo) -> Result<u64, Error> {
             total.checked_add(part_size).ok_or(Error::PartMissingOrCorrupt)
         })?
     };
-    Ok(logical_size.max(persisted_part_size))
+    Ok(logical_size.unwrap_or(0).max(persisted_part_size))
 }
 
 type UsageVersionPage = StorageListObjectVersionsInfo<ObjectInfo>;
@@ -3320,6 +3346,80 @@ mod tests {
         );
     }
 
+    #[test]
+    fn quota_object_size_accepts_compressed_unknown_actual_size_sentinel() {
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(
+            &mut metadata,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            "klauspost/compress/s2".to_string(),
+        );
+        let object = ObjectInfo {
+            size: 400,
+            actual_size: -1,
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+
+        assert_eq!(quota_object_size(&object).expect("compressed sentinel is valid"), 400);
+    }
+
+    #[test]
+    fn quota_object_size_rejects_compressed_part_sum_overflow() {
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(
+            &mut metadata,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            "klauspost/compress/s2".to_string(),
+        );
+        let object = ObjectInfo {
+            size: 1,
+            user_defined: Arc::new(metadata),
+            parts: Arc::new(vec![
+                rustfs_filemeta::ObjectPartInfo {
+                    actual_size: i64::MAX,
+                    ..Default::default()
+                },
+                rustfs_filemeta::ObjectPartInfo {
+                    actual_size: 1,
+                    ..Default::default()
+                },
+            ]),
+            ..Default::default()
+        };
+
+        assert!(matches!(quota_object_size(&object), Err(Error::Io(_))));
+    }
+
+    #[test]
+    fn quota_object_size_rejects_negative_values_other_than_the_compressed_sentinel() {
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(
+            &mut metadata,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            "klauspost/compress/s2".to_string(),
+        );
+        let corrupt_object = ObjectInfo {
+            size: 400,
+            actual_size: -2,
+            user_defined: Arc::new(metadata.clone()),
+            ..Default::default()
+        };
+        assert!(matches!(quota_object_size(&corrupt_object), Err(Error::PartMissingOrCorrupt)));
+
+        let corrupt_part = ObjectInfo {
+            size: 400,
+            user_defined: Arc::new(metadata),
+            parts: Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
+                size: 400,
+                actual_size: -2,
+                ..Default::default()
+            }]),
+            ..Default::default()
+        };
+        assert!(matches!(quota_object_size(&corrupt_part), Err(Error::PartMissingOrCorrupt)));
+    }
+
     #[tokio::test]
     #[serial]
     async fn live_bucket_usage_refreshes_are_coalesced_only_while_in_flight() {

crates/ecstore/src/disk/local.rs

@@ -858,7 +858,6 @@ const EVENT_DISK_LOCAL_DIRECT_IO_FALLBACK: &str = "disk_local_direct_io_fallback
 #[cfg(target_os = "linux")]
 const EVENT_DISK_LOCAL_URING_LATCH_OFF: &str = "disk_local_uring_latch_off";
 const EVENT_DISK_LOCAL_DELETE_FAILED: &str = "disk_local_delete_failed";
-const EVENT_DISK_LOCAL_DELETE_ROLLBACK_FAILED: &str = "disk_local_delete_rollback_failed";
 const EVENT_DISK_LOCAL_CHECK_PARTS: &str = "disk_local_check_parts";
 const EVENT_DISK_LOCAL_ACCESS_FAILED: &str = "disk_local_access_failed";
 const EVENT_DISK_LOCAL_VOLUME_SETUP_FAILED: &str = "disk_local_volume_setup_failed";
@@ -6107,43 +6106,6 @@ impl LocalDisk {
         Ok((bytes, modtime))
     }
 
-    async fn write_missing_delete_marker(
-        &self,
-        volume: &str,
-        path: &str,
-        fi: FileInfo,
-        object_dir: &Path,
-        xl_path: &Path,
-        rollback_dir: Option<Uuid>,
-    ) -> Result<()> {
-        if let Some(rollback_dir) = rollback_dir {
-            let rollback_path = object_dir.join(rollback_dir.to_string());
-            fs::create_dir_all(&rollback_path).await.map_err(to_file_error)?;
-            fs::write(rollback_path.join(DELETE_MARKER_ROLLBACK_FILE), [])
-                .await
-                .map_err(to_file_error)?;
-        }
-        if let Err(err) = self.write_metadata("", volume, path, fi).await {
-            if let Some(rollback_dir) = rollback_dir
-                && let Err(restore_err) = restore_delete_rollback(object_dir, xl_path, rollback_dir, &self.publication_root).await
-            {
-                warn!(
-                    event = EVENT_DISK_LOCAL_DELETE_ROLLBACK_FAILED,
-                    component = LOG_COMPONENT_ECSTORE,
-                    subsystem = LOG_SUBSYSTEM_DISK_LOCAL,
-                    result = "failed",
-                    volume,
-                    path,
-                    rollback_dir = %rollback_dir,
-                    error = ?restore_err,
-                    "Disk local delete rollback failed"
-                );
-            }
-            return Err(err);
-        }
-        Ok(())
-    }
-
     async fn delete_versions_internal(&self, volume: &str, path: &str, fis: &[FileInfo], opts: &DeleteOptions) -> Result<()> {
         let volume_dir = self.io_get_bucket_path(volume)?;
         let xlpath = self.io_get_object_path(volume, format!("{path}/{STORAGE_FORMAT_FILE}").as_str())?;
@@ -6161,20 +6123,7 @@ impl LocalDisk {
             return restore_metadata_backup(object_dir, &xlpath, rollback_dir, &self.publication_root).await;
         }
 
-        let (data, _) = match self.read_all_data_with_dmtime(volume, volume_dir.as_path(), &xlpath).await {
-            Ok(data) => data,
-            Err(DiskError::FileNotFound) => {
-                // `deleted` alone can be an explicit marker purge; only
-                // `mark_deleted` may create metadata that was not present.
-                let Some(delete_marker) = fis.iter().find(|fi| fi.deleted && fi.mark_deleted).cloned() else {
-                    return Err(DiskError::FileNotFound);
-                };
-                return self
-                    .write_missing_delete_marker(volume, path, delete_marker, object_dir, &xlpath, opts.old_data_dir)
-                    .await;
-            }
-            Err(err) => return Err(err),
-        };
+        let (data, _) = self.read_all_data_with_dmtime(volume, volume_dir.as_path(), &xlpath).await?;
 
         if data.is_empty() {
             return Err(DiskError::FileNotFound);
@@ -10473,9 +10422,29 @@ impl DiskAPI for LocalDisk {
                 }
 
                 if fi.deleted && force_del_marker {
-                    return self
-                        .write_missing_delete_marker(volume, path, fi, file_path.as_path(), &xl_path, rollback_dir)
-                        .await;
+                    if let Some(rollback_dir) = rollback_dir {
+                        let rollback_path = file_path.join(rollback_dir.to_string());
+                        fs::create_dir_all(&rollback_path).await.map_err(to_file_error)?;
+                        fs::write(rollback_path.join(DELETE_MARKER_ROLLBACK_FILE), [])
+                            .await
+                            .map_err(to_file_error)?;
+                    }
+                    if let Err(err) = self.write_metadata("", volume, path, fi).await {
+                        if let Some(rollback_dir) = rollback_dir
+                            && let Err(restore_err) =
+                                restore_delete_rollback(file_path.as_path(), &xl_path, rollback_dir, &self.publication_root).await
+                        {
+                            warn!(
+                                volume,
+                                path,
+                                rollback_dir = %rollback_dir,
+                                error = ?restore_err,
+                                "failed to restore metadata after delete marker commit error"
+                            );
+                        }
+                        return Err(err);
+                    }
+                    return Ok(());
                 }
 
                 return if fi.version_id.is_some() {

crates/ecstore/src/object_api/types.rs

@@ -24,7 +24,7 @@ use crate::storage_api_contracts::{
 pub struct NamespaceLockFence {
     signals: Arc<Vec<Arc<rustfs_lock::distributed_lock::LockLostSignal>>>,
     #[cfg(test)]
-    forced_lost: Arc<Vec<Arc<std::sync::atomic::AtomicBool>>>,
+    forced_lost: Arc<std::sync::atomic::AtomicBool>,
 }
 
 impl Debug for NamespaceLockFence {
@@ -40,17 +40,13 @@ impl NamespaceLockFence {
         Self {
             signals: Arc::default(),
             #[cfg(test)]
-            forced_lost: Arc::new(vec![Arc::new(std::sync::atomic::AtomicBool::new(false))]),
+            forced_lost: Arc::new(std::sync::atomic::AtomicBool::new(false)),
         }
     }
 
     pub(crate) fn is_lock_lost(&self) -> bool {
         #[cfg(test)]
-        if self
-            .forced_lost
-            .iter()
-            .any(|lost| lost.load(std::sync::atomic::Ordering::Acquire))
-        {
+        if self.forced_lost.load(std::sync::atomic::Ordering::Acquire) {
             return true;
         }
         self.signals.iter().any(|signal| signal.is_lost())
@@ -61,26 +57,27 @@ impl NamespaceLockFence {
     }
 
     fn extend(&mut self, other: &Self) {
-        if !Arc::ptr_eq(&self.signals, &other.signals) {
-            Arc::make_mut(&mut self.signals).extend(other.signals.iter().cloned());
+        if Arc::ptr_eq(&self.signals, &other.signals) {
+            return;
         }
+        Arc::make_mut(&mut self.signals).extend(other.signals.iter().cloned());
         #[cfg(test)]
-        if !Arc::ptr_eq(&self.forced_lost, &other.forced_lost) {
-            Arc::make_mut(&mut self.forced_lost).extend(other.forced_lost.iter().cloned());
+        if other.forced_lost.load(std::sync::atomic::Ordering::Acquire) {
+            self.forced_lost.store(true, std::sync::atomic::Ordering::Release);
         }
     }
 
     #[cfg(test)]
     pub(crate) fn lost_for_test() -> Self {
         let fence = Self::new();
-        fence.forced_lost[0].store(true, std::sync::atomic::Ordering::Release);
+        fence.forced_lost.store(true, std::sync::atomic::Ordering::Release);
         fence
     }
 
     #[cfg(test)]
     pub(crate) fn loss_handle_for_test() -> (Self, Arc<std::sync::atomic::AtomicBool>) {
         let fence = Self::new();
-        (fence.clone(), Arc::clone(&fence.forced_lost[0]))
+        (fence.clone(), Arc::clone(&fence.forced_lost))
     }
 }
 
@@ -414,13 +411,6 @@ impl ObjectOptions {
         self.namespace_lock_fence.get_or_insert_with(NamespaceLockFence::new);
     }
 
-    #[cfg(test)]
-    pub(crate) fn add_namespace_lock_fence_for_test(&mut self, fence: &NamespaceLockFence) {
-        self.namespace_lock_fence
-            .get_or_insert_with(NamespaceLockFence::new)
-            .extend(fence);
-    }
-
     pub(crate) fn ensure_lifecycle_delete_all_journal(&mut self) {
         self.lifecycle_delete_all_journal
             .get_or_insert_with(|| Arc::new(parking_lot::Mutex::new(LifecycleDeleteAllJournalState::default())));
@@ -699,6 +689,9 @@ impl ObjectInfo {
     }
 
     pub fn get_actual_size(&self) -> std::io::Result<i64> {
+        if self.actual_size < -1 || (self.actual_size == -1 && !self.is_compressed()) {
+            return Err(std::io::Error::other("invalid negative actual size"));
+        }
         if self.actual_size > 0 {
             return Ok(self.actual_size);
         }
@@ -710,10 +703,25 @@ impl ObjectInfo {
                 let size = size_str.parse::<i64>().map_err(|e| std::io::Error::other(e.to_string()))?;
                 return Ok(size);
             }
-            let mut actual_size = 0;
-            self.parts.iter().for_each(|part| {
-                actual_size += part.actual_size;
-            });
+            if self.actual_size == -1 && self.parts.is_empty() {
+                return Ok(-1);
+            }
+            let mut actual_size = 0_i64;
+            let mut unknown = false;
+            for part in self.parts.iter() {
+                match part.actual_size {
+                    -1 => unknown = true,
+                    size if size >= 0 => {
+                        actual_size = actual_size
+                            .checked_add(size)
+                            .ok_or_else(|| std::io::Error::other("compressed actual size overflow"))?;
+                    }
+                    _ => return Err(std::io::Error::other("invalid negative compressed part size")),
+                }
+            }
+            if unknown {
+                return Ok(-1);
+            }
             if actual_size == 0 && actual_size != self.size {
                 return Err(std::io::Error::other(format!("invalid decompressed size {} {}", actual_size, self.size)));
             }
@@ -728,6 +736,18 @@ impl ObjectInfo {
         Ok(self.size)
     }
 
+    /// Returns a non-negative size for client and replication boundaries.
+    ///
+    /// Compressed legacy metadata can retain the internal `-1` unknown-size
+    /// sentinel. Those boundaries cannot emit a negative length, so they use
+    /// the persisted physical size while quota accounting keeps the sentinel
+    /// distinction in [`crate::data_usage::quota_object_size`].
+    pub fn get_actual_size_or_physical(&self) -> i64 {
+        self.get_actual_size()
+            .map(|size| if size >= 0 { size } else { self.size.max(0) })
+            .unwrap_or_else(|_| self.size.max(0))
+    }
+
     pub fn from_file_info(fi: &FileInfo, bucket: &str, object: &str, versioned: bool) -> ObjectInfo {
         let mut version_id = fi.version_id;
 

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

@@ -334,7 +334,6 @@ impl ECStore {
                             lifecycle_guard: bucket_incarnation_fence
                                 .as_ref()
                                 .and_then(|guard| guard.namespace_lock_guard()),
-                            ..Default::default()
                         },
                         "rebalance",
                     ),

crates/ecstore/src/set_disk/mod.rs

@@ -97,7 +97,7 @@ use crate::storage_api_contracts::{
         CompletePart, ListMultipartsInfo, ListPartsInfo, MultipartInfo, MultipartOperations as _, MultipartUploadResult, PartInfo,
     },
     namespace::NamespaceLocking as _,
-    object::{DeletedObject, HTTPPreconditions, ObjectIO as _, ObjectOperations as _, ObjectToDelete},
+    object::{DeleteAccounting, DeletedObject, HTTPPreconditions, ObjectIO as _, ObjectOperations as _, ObjectToDelete},
     range::HTTPRangeSpec,
 };
 use crate::store::utils::is_reserved_or_invalid_bucket;
@@ -735,12 +735,8 @@ pub(crate) use core::io_primitives::disk_call_counters;
 mod ctx;
 mod metadata;
 mod ops;
-#[cfg(test)]
-pub(crate) use ops::multipart::NewMultipartUploadCommitObservation;
 #[cfg(any(test, feature = "test-util"))]
 pub use ops::multipart::{MultipartCommitBarrier, MultipartCommitPause};
-#[cfg(test)]
-pub(crate) use ops::object::DeleteObjectCommitBarrier;
 #[cfg(feature = "test-util")]
 pub(crate) use ops::object::TransitionCleanupStoreBarrier as SetDiskTransitionCleanupStoreBarrier;
 pub(crate) use ops::object::body_cache_plaintext_len;
@@ -3029,16 +3025,6 @@ pub struct SetDisks {
     storage_class_config_override: Arc<std::sync::RwLock<Option<Arc<storageclass::Config>>>>,
 }
 
-// DistributedLock sends the raw ObjectKey to its clients; LockRegistry clones
-// each endpoint's canonical Arc, so an exact Arc set identifies the lock domain.
-pub(crate) fn same_distributed_lock_domain(left: &[Arc<dyn LockClient>], right: &[Arc<dyn LockClient>]) -> bool {
-    left.iter()
-        .all(|left_client| right.iter().any(|right_client| Arc::ptr_eq(left_client, right_client)))
-        && right
-            .iter()
-            .all(|right_client| left.iter().any(|left_client| Arc::ptr_eq(left_client, right_client)))
-}
-
 const ERASURE_CACHE_MAX_ENTRIES: usize = 32;
 
 #[derive(Clone, Copy, Debug, Eq, Hash, PartialEq)]
@@ -3614,15 +3600,6 @@ impl SetDisks {
         &self.ctx
     }
 
-    /// Whether both sets' namespace-lock implementations cover the same object key.
-    pub(crate) async fn shares_namespace_lock_domain(&self, other: &Self) -> bool {
-        match (self.ctx.is_dist_erasure().await, other.ctx.is_dist_erasure().await) {
-            (false, false) => Arc::ptr_eq(&self.local_lock_manager, &other.local_lock_manager),
-            (true, true) => same_distributed_lock_domain(&self.lockers, &other.lockers),
-            _ => false,
-        }
-    }
-
     /// The lock manager this set actually uses (test-only; Phase 5 Slice 3).
     #[cfg(test)]
     pub(crate) fn local_lock_manager_for_test(&self) -> &Arc<rustfs_lock::GlobalLockManager> {
@@ -4607,11 +4584,11 @@ fn should_preserve_delete_replication_state(opts: &ObjectOptions) -> bool {
 }
 
 fn should_force_delete_marker_for_missing_version(opts: &ObjectOptions) -> bool {
-    opts.delete_marker || ((opts.versioned || opts.version_suspended) && opts.version_id.is_none() && !opts.data_movement)
+    opts.delete_marker || (opts.versioned && opts.version_id.is_none() && !opts.data_movement)
 }
 
 fn resolve_delete_version_state(opts: &ObjectOptions, goi: &ObjectInfo, version_found: bool) -> (bool, bool) {
-    let mut mark_delete = goi.version_id.is_some() || ((opts.versioned || opts.version_suspended) && opts.version_id.is_none());
+    let mut mark_delete = goi.version_id.is_some() || (opts.versioned && opts.version_id.is_none());
     let mut delete_marker = opts.versioned;
 
     if opts.version_id.is_some() {

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

@@ -32,8 +32,6 @@ use crate::crash_inject::{self, CrashPoint};
 use crate::multipart_listing::paginate_multipart_listing;
 use futures::{StreamExt, stream};
 use std::future::Future;
-#[cfg(test)]
-use std::sync::atomic::AtomicBool;
 #[cfg(any(test, feature = "test-util"))]
 use std::sync::atomic::{AtomicUsize, Ordering};
 use std::time::Duration;
@@ -67,7 +65,6 @@ impl StaleMultipartCleanupGuard {
 #[cfg(any(test, feature = "test-util"))]
 #[derive(Clone, Copy, PartialEq, Eq)]
 pub enum MultipartCommitPause {
-    NewUploadBeforeLockLost,
     PutPartBeforeLockAcquire,
     PutPartBeforeLockLost,
     PutPartAfterRename,
@@ -159,72 +156,6 @@ impl Drop for MultipartCommitBarrier {
     }
 }
 
-#[cfg(test)]
-struct NewMultipartUploadCommitObservationState {
-    bucket: String,
-    object: String,
-    committed: AtomicBool,
-}
-
-#[cfg(test)]
-pub(crate) struct NewMultipartUploadCommitObservation {
-    state: Arc<NewMultipartUploadCommitObservationState>,
-}
-
-#[cfg(test)]
-static NEW_MULTIPART_UPLOAD_COMMIT_OBSERVATION: std::sync::OnceLock<
-    std::sync::Mutex<Option<Arc<NewMultipartUploadCommitObservationState>>>,
-> = std::sync::OnceLock::new();
-
-#[cfg(test)]
-impl NewMultipartUploadCommitObservation {
-    pub(crate) fn install(bucket: &str, object: &str) -> Self {
-        let state = Arc::new(NewMultipartUploadCommitObservationState {
-            bucket: bucket.to_string(),
-            object: object.to_string(),
-            committed: AtomicBool::new(false),
-        });
-        let mut slot = NEW_MULTIPART_UPLOAD_COMMIT_OBSERVATION
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("new multipart upload commit observation mutex should not poison");
-        assert!(slot.is_none(), "new multipart upload commit observation must be unique");
-        *slot = Some(Arc::clone(&state));
-        Self { state }
-    }
-
-    pub(crate) fn committed(&self) -> bool {
-        self.state.committed.load(Ordering::Acquire)
-    }
-}
-
-#[cfg(test)]
-impl Drop for NewMultipartUploadCommitObservation {
-    fn drop(&mut self) {
-        let mut slot = NEW_MULTIPART_UPLOAD_COMMIT_OBSERVATION
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("new multipart upload commit observation mutex should not poison");
-        if slot.as_ref().is_some_and(|state| Arc::ptr_eq(state, &self.state)) {
-            *slot = None;
-        }
-    }
-}
-
-#[cfg(test)]
-fn observe_new_multipart_upload_commit(bucket: &str, object: &str) {
-    let state = NEW_MULTIPART_UPLOAD_COMMIT_OBSERVATION
-        .get_or_init(|| std::sync::Mutex::new(None))
-        .lock()
-        .expect("new multipart upload commit observation mutex should not poison")
-        .as_ref()
-        .filter(|state| state.bucket == bucket && state.object == object)
-        .cloned();
-    if let Some(state) = state {
-        state.committed.store(true, Ordering::Release);
-    }
-}
-
 #[cfg(any(test, feature = "test-util"))]
 async fn pause_multipart_commit(bucket: &str, object: &str, pause: MultipartCommitPause) {
     let barrier = {
@@ -1684,30 +1615,6 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
 
         let upload_path = Self::get_multipart_upload_dir(bucket, object, upload_uuid.as_str(), opts.data_movement);
 
-        #[cfg(any(test, feature = "test-util"))]
-        pause_multipart_commit(bucket, object, MultipartCommitPause::NewUploadBeforeLockLost).await;
-        if _object_lock_guard.as_ref().is_some_and(|guard| guard.is_lock_lost()) {
-            return Err(StorageError::NamespaceLockQuorumUnavailable {
-                mode: "new_multipart_upload_commit",
-                bucket: bucket.to_string(),
-                object: object.to_string(),
-                required: 1,
-                achieved: 0,
-            });
-        }
-        if opts
-            .namespace_lock_fence
-            .as_ref()
-            .is_some_and(NamespaceLockFence::is_lock_lost)
-        {
-            return Err(StorageError::NamespaceLockQuorumUnavailable {
-                mode: "new_multipart_upload_outer_lock",
-                bucket: bucket.to_string(),
-                object: object.to_string(),
-                required: 1,
-                achieved: 0,
-            });
-        }
         ensure_multipart_bucket_lifecycle_lock_held(bucket, object, opts)?;
         Self::write_unique_file_info(
             &shuffle_disks,
@@ -1719,8 +1626,6 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
         )
         .await
         .map_err(|e| to_object_err(e.into(), vec![bucket, object]))?;
-        #[cfg(test)]
-        observe_new_multipart_upload_commit(bucket, object);
 
         // evalDisks
 

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

@@ -45,6 +45,7 @@ use crate::bucket::replication::{
     DeleteReplicationConfigSnapshot, ReplicationLifecycleBridge, ReplicationStatusType, VersionPurgeStatusType,
     replication_state_to_filemeta, replication_status_from_filemeta, version_purge_status_to_filemeta,
 };
+use crate::data_usage::quota_object_size;
 use crate::diagnostics::get::GetObjectFailureReason;
 use crate::disk::{DataDirDeleteStatus, OldCurrentSize};
 use crate::error::is_err_invalid_upload_id;
@@ -2483,7 +2484,6 @@ impl SetDisks {
                         })
                         .await?,
                     );
-                    notify_put_object_commit_namespace_acquired(bucket, object);
                 }
                 #[cfg(not(any(test, feature = "test-util")))]
                 {
@@ -4631,7 +4631,6 @@ struct PutObjectCommitBarrierState {
     arrived: tokio::sync::Notify,
     release: tokio::sync::Notify,
     namespace_pending: tokio::sync::Notify,
-    namespace_acquired: std::sync::atomic::AtomicBool,
 }
 
 #[cfg(any(test, feature = "test-util"))]
@@ -4653,7 +4652,6 @@ impl PutObjectCommitBarrier {
             arrived: tokio::sync::Notify::new(),
             release: tokio::sync::Notify::new(),
             namespace_pending: tokio::sync::Notify::new(),
-            namespace_acquired: std::sync::atomic::AtomicBool::new(false),
         });
         let mut slot = PUT_OBJECT_COMMIT_BARRIER
             .get_or_init(|| std::sync::Mutex::new(Vec::new()))
@@ -4688,10 +4686,6 @@ impl PutObjectCommitBarrier {
             .await
             .expect("put object should wait for the namespace lock after leaving the commit barrier");
     }
-
-    pub fn namespace_acquired(&self) -> bool {
-        self.state.namespace_acquired.load(std::sync::atomic::Ordering::Acquire)
-    }
 }
 
 #[cfg(any(test, feature = "test-util"))]
@@ -4748,22 +4742,6 @@ fn notify_put_object_commit_namespace_pending(bucket: &str, object: &str) {
     }
 }
 
-#[cfg(any(test, feature = "test-util"))]
-fn notify_put_object_commit_namespace_acquired(bucket: &str, object: &str) {
-    let barrier = PUT_OBJECT_COMMIT_BARRIER
-        .get_or_init(|| std::sync::Mutex::new(Vec::new()))
-        .lock()
-        .expect("put object commit barrier mutex should not poison")
-        .iter()
-        .find(|barrier| {
-            barrier.bucket == bucket && barrier.object == object && barrier.pause == PutObjectCommitPause::BeforeNamespace
-        })
-        .cloned();
-    if let Some(barrier) = barrier {
-        barrier.namespace_acquired.store(true, std::sync::atomic::Ordering::Release);
-    }
-}
-
 #[cfg(test)]
 struct DeleteObjectCommitBarrierState {
     bucket: String,
@@ -4773,7 +4751,7 @@ struct DeleteObjectCommitBarrierState {
 }
 
 #[cfg(test)]
-pub(crate) struct DeleteObjectCommitBarrier {
+struct DeleteObjectCommitBarrier {
     state: Arc<DeleteObjectCommitBarrierState>,
 }
 
@@ -4783,7 +4761,7 @@ static DELETE_OBJECT_COMMIT_BARRIER: std::sync::OnceLock<std::sync::Mutex<Option
 
 #[cfg(test)]
 impl DeleteObjectCommitBarrier {
-    pub(crate) fn install(bucket: &str, object: &str) -> Self {
+    fn install(bucket: &str, object: &str) -> Self {
         let state = Arc::new(DeleteObjectCommitBarrierState {
             bucket: bucket.to_string(),
             object: object.to_string(),
@@ -4799,13 +4777,13 @@ impl DeleteObjectCommitBarrier {
         Self { state }
     }
 
-    pub(crate) async fn wait_until_paused(&self) {
+    async fn wait_until_paused(&self) {
         tokio::time::timeout(Duration::from_secs(30), self.state.arrived.notified())
             .await
             .expect("delete object should reach the deterministic commit barrier");
     }
 
-    pub(crate) fn release(&self) {
+    fn release(&self) {
         self.state.release.notify_one();
     }
 }
@@ -5678,7 +5656,18 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
         objects: Vec<ObjectToDelete>,
         opts: ObjectOptions,
     ) -> (Vec<DeletedObject>, Vec<Option<Error>>) {
+        let (deleted, errors, _) = self.delete_objects_with_accounting(bucket, objects, opts).await;
+        (deleted, errors)
+    }
+
+    async fn delete_objects_with_accounting(
+        &self,
+        bucket: &str,
+        objects: Vec<ObjectToDelete>,
+        opts: ObjectOptions,
+    ) -> (Vec<DeletedObject>, Vec<Option<Error>>, Vec<Option<DeleteAccounting>>) {
         let mut del_objects = vec![DeletedObject::default(); objects.len()];
+        let mut accounting = vec![None; objects.len()];
         let delete_config_snapshot = opts
             .delete_replication_config_snapshot
             .clone()
@@ -5768,7 +5757,7 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                                 *item = Some(Error::other(message.clone()));
                             }
                         }
-                        return (del_objects, del_errs);
+                        return (del_objects, del_errs, accounting);
                     }
                 },
             }
@@ -5815,6 +5804,22 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
             let source_missing = gerr
                 .as_ref()
                 .is_some_and(|err| is_err_object_not_found(err) || is_err_version_not_found(err));
+            // Resolve accounting from the generation selected under this
+            // object's write lock. A request-layer pre-stat is only an
+            // optimization and cannot identify a concurrent overwrite.
+            let (accounting_size, accounting_version_id, removed_current_object) = if source_missing
+                || dobj.synthetic_version_id
+                || set_disk_delete_creates_delete_marker(&check_opts)
+                || goi.delete_marker
+            {
+                (None, None, false)
+            } else {
+                (
+                    quota_object_size(&goi).ok(),
+                    goi.version_id.filter(|version_id| !version_id.is_nil()),
+                    (dobj.version_id.is_none() || is_explicit_null_version(dobj.version_id)) && !dobj.synthetic_version_id,
+                )
+            };
             // Normalize both sides before comparing. `goi.version_id` is the
             // client-facing identity, where `from_file_info` synthesizes
             // `Some(Uuid::nil())` for a null version on a versioned or
@@ -5900,7 +5905,6 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
             if dobj.version_id.is_none() && (version_suspended || versioned) {
                 vr.mod_time = Some(OffsetDateTime::now_utc());
                 vr.deleted = true;
-                vr.mark_deleted = true;
                 if versioned {
                     vr.version_id = Some(Uuid::new_v4());
                 }
@@ -5944,7 +5948,12 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                     },
                     replication_state: vr.replication_state_internal.clone(),
                     ..Default::default()
-                }
+                };
+                accounting[i] = Some(DeleteAccounting {
+                    size: accounting_size,
+                    version_id: accounting_version_id,
+                    removed_current_object,
+                });
             }
 
             // Only add to vers_map if we hold the lock
@@ -5990,7 +5999,7 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                     });
                 }
             }
-            return (del_objects, del_errs);
+            return (del_objects, del_errs, accounting);
         }
 
         let mut persisted_journal_entries = Vec::with_capacity(journal_entries.len());
@@ -6228,7 +6237,16 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
             }
         }
 
-        (del_objects, del_errs)
+        // An accounting identity is actionable only when the delete result is
+        // successful. Never let a failed commit (including a partial quorum
+        // failure) reach the request-layer fast delta path.
+        for (index, err) in del_errs.iter().enumerate() {
+            if err.is_some() {
+                accounting[index] = None;
+            }
+        }
+
+        (del_objects, del_errs, accounting)
     }
 
     #[tracing::instrument(skip(self))]
@@ -6557,6 +6575,12 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
 
         let mut obj_info = ObjectInfo::from_file_info(&dfi, bucket, object, opts.versioned || opts.version_suspended);
         obj_info.size = goi.size;
+        // Keep the committed source metadata on the internal delete result so
+        // the request layer can derive canonical accounting for this exact
+        // generation. Delete responses do not expose these fields.
+        obj_info.actual_size = goi.actual_size;
+        obj_info.user_defined = Arc::clone(&goi.user_defined);
+        obj_info.parts = Arc::clone(&goi.parts);
         obj_info.user_tags = Arc::clone(&goi.user_tags);
         self.invalidate_get_object_metadata_cache(bucket, object).await;
         Ok(obj_info)
@@ -7848,6 +7872,113 @@ mod replication_quota_safety_tests {
         assert_eq!(stored.get_actual_size().expect("stored logical size should parse"), 1);
     }
 
+    #[tokio::test]
+    async fn delete_returns_canonical_compressed_accounting_size() {
+        let (_temp_dirs, disks, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "compressed-delete-accounting";
+        for disk in &disks {
+            disk.make_volume(bucket).await.expect("bucket volume should be created");
+        }
+
+        let mut user_defined = HashMap::new();
+        insert_str(
+            &mut user_defined,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            "klauspost/compress/s2".to_string(),
+        );
+        insert_str(&mut user_defined, SUFFIX_ACTUAL_SIZE, "1000".to_string());
+        let mut reader = PutObjReader::new(
+            HashReader::from_stream(Cursor::new(vec![0x5a; 400]), 400, 1000, None, None, false)
+                .expect("compressed fixture reader should be valid"),
+        );
+        set_disks
+            .put_object(
+                bucket,
+                "object",
+                &mut reader,
+                &ObjectOptions {
+                    user_defined,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("compressed object should be written");
+
+        let (deleted, errors, accounting) = set_disks
+            .delete_objects_with_accounting(
+                bucket,
+                vec![ObjectToDelete {
+                    object_name: "object".to_string(),
+                    ..Default::default()
+                }],
+                ObjectOptions {
+                    object_lock_config_snapshot: Some(Arc::new(ObjectLockConfigSnapshot::new(
+                        ObjectLockConfigState::ConfirmedAbsent,
+                    ))),
+                    ..Default::default()
+                },
+            )
+            .await;
+
+        assert!(errors[0].is_none(), "compressed delete should succeed: {:?}", errors[0]);
+        assert!(deleted[0].found, "the committed object must be reported as found");
+        assert_eq!(accounting[0].as_ref().and_then(|value| value.size), Some(1000));
+        assert!(accounting[0].as_ref().is_some_and(|value| value.version_id.is_none()));
+        assert!(accounting[0].as_ref().is_some_and(|value| value.removed_current_object));
+    }
+
+    #[tokio::test]
+    async fn suspended_delete_marker_does_not_return_body_accounting() {
+        let (_temp_dirs, disks, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "suspended-delete-accounting";
+        for disk in &disks {
+            disk.make_volume(bucket).await.expect("bucket volume should be created");
+        }
+
+        let mut user_defined = HashMap::new();
+        insert_str(
+            &mut user_defined,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            "klauspost/compress/s2".to_string(),
+        );
+        insert_str(&mut user_defined, SUFFIX_ACTUAL_SIZE, "1000".to_string());
+        let mut reader = PutObjReader::new(
+            HashReader::from_stream(Cursor::new(vec![0x5a; 400]), 400, 1000, None, None, false)
+                .expect("compressed fixture reader should be valid"),
+        );
+        let suspended_opts = ObjectOptions {
+            version_suspended: true,
+            delete_replication_config_snapshot: Some(Arc::new(DeleteReplicationConfigSnapshot::from_configs_for_test(
+                s3s::dto::VersioningConfiguration {
+                    status: Some(s3s::dto::BucketVersioningStatus::from_static(s3s::dto::BucketVersioningStatus::SUSPENDED)),
+                    ..Default::default()
+                },
+                None,
+            ))),
+            user_defined,
+            object_lock_config_snapshot: Some(Arc::new(ObjectLockConfigSnapshot::new(ObjectLockConfigState::ConfirmedAbsent))),
+            ..Default::default()
+        };
+        set_disks
+            .put_object(bucket, "object", &mut reader, &suspended_opts)
+            .await
+            .expect("compressed object should be written");
+
+        let (deleted, errors, accounting) = set_disks
+            .delete_objects_with_accounting(
+                bucket,
+                vec![ObjectToDelete {
+                    object_name: "object".to_string(),
+                    ..Default::default()
+                }],
+                suspended_opts,
+            )
+            .await;
+        assert!(errors[0].is_none(), "suspended delete should create a marker: {:?}", errors[0]);
+        assert!(deleted[0].delete_marker);
+        assert!(accounting[0].is_none(), "a delete marker must not carry body accounting");
+    }
+
     #[tokio::test]
     async fn direct_put_cannot_persist_a_tiny_logical_size() {
         let (_temp_dirs, disks, set_disks) = hermetic_set_disks(4).await;
@@ -11660,7 +11791,6 @@ mod transition_upload_integrity_tests {
                 crate::data_movement::SourceCleanupBucketFence {
                     expected_incarnation_id: None,
                     lifecycle_guard: Some(&bucket_guard),
-                    ..Default::default()
                 },
                 "test_data_movement",
             )

crates/ecstore/src/storage_api_contracts/mod.rs

@@ -62,8 +62,8 @@ pub(crate) mod object {
     use super::{Debug, Error, FileInfo, GetObjectReader, ObjectInfo, ObjectOptions, PutObjReader};
     use crate::storage_api_contracts::range::HTTPRangeSpec;
     pub(crate) use rustfs_storage_api::{
-        DeletedObject, HTTPPreconditions, ObjectIO, ObjectLockDeleteOptions, ObjectLockRetentionOptions, ObjectOperations,
-        ObjectPreconditionError, ObjectPreconditionPart, ObjectPreconditionState, ObjectToDelete,
+        DeleteAccounting, DeletedObject, HTTPPreconditions, ObjectIO, ObjectLockDeleteOptions, ObjectLockRetentionOptions,
+        ObjectOperations, ObjectPreconditionError, ObjectPreconditionPart, ObjectPreconditionState, ObjectToDelete,
     };
 
     pub(crate) trait EcstoreObjectIO:

crates/ecstore/src/store/mod.rs

@@ -151,7 +151,7 @@ pub(crate) mod init_format;
 pub(crate) mod list_objects;
 mod multipart;
 mod object;
-pub(crate) use object::{ObjectLockDiagGuard, SourceCleanupMutationFence};
+pub(crate) use object::ObjectLockDiagGuard;
 pub use object::{
     PrepareSelectObjectSnapshotError, PreparedGetObjectReader, SelectObjectSnapshot, SelectObjectSnapshotReadError,
     SnapshotConsistencyError,

crates/ecstore/src/store/multipart.rs

@@ -400,7 +400,7 @@ impl ECStore {
         object: &str,
         opts: &ObjectOptions,
     ) -> Result<MultipartUploadResult> {
-        self.handle_new_multipart_upload_with_pool_idx(bucket, object, opts, None)
+        self.handle_new_multipart_upload_with_pool_idx(bucket, object, opts)
             .await
             .map(|(res, _, _)| res)
     }
@@ -410,22 +410,20 @@ impl ECStore {
         bucket: &str,
         object: &str,
         opts: &ObjectOptions,
-        mutation_fence: Option<&ObjectLockDiagGuard>,
     ) -> Result<(MultipartUploadResult, usize, Option<Uuid>)> {
         check_new_multipart_args(bucket, object)?;
-        let (mut opts, _bucket_lifecycle_guard) = self.guard_multipart_bucket_incarnation(bucket, opts).await?;
+        let (opts, _bucket_lifecycle_guard) = self.guard_multipart_bucket_incarnation(bucket, opts).await?;
+        let opts = &opts;
 
         if self.single_pool() {
-            self.apply_decommission_target_mutation_fence(0, object, &mut opts, mutation_fence)
-                .await;
             return self.pools[0]
-                .new_multipart_upload(bucket, object, &opts)
+                .new_multipart_upload(bucket, object, opts)
                 .await
                 .map(|res| (res, 0, opts.expected_bucket_incarnation_id));
         }
 
         if opts.data_movement && opts.version_id.is_some() {
-            let idx = self.select_data_movement_pool_idx(bucket, object, -1, &opts, false).await?;
+            let idx = self.select_data_movement_pool_idx(bucket, object, -1, opts, false).await?;
             if idx == opts.src_pool_idx {
                 return Err(StorageError::DataMovementOverwriteErr(
                     bucket.to_owned(),
@@ -433,9 +431,7 @@ impl ECStore {
                     opts.version_id.clone().unwrap_or_default(),
                 ));
             }
-            self.apply_decommission_target_mutation_fence(idx, object, &mut opts, mutation_fence)
-                .await;
-            let res = self.pools[idx].new_multipart_upload(bucket, object, &opts).await?;
+            let res = self.pools[idx].new_multipart_upload(bucket, object, opts).await?;
             return Ok((res, idx, opts.expected_bucket_incarnation_id));
         }
 
@@ -458,9 +454,7 @@ impl ECStore {
             .await?;
 
             if !res.uploads.is_empty() {
-                self.apply_decommission_target_mutation_fence(idx, object, &mut opts, mutation_fence)
-                    .await;
-                let res = self.pools[idx].new_multipart_upload(bucket, object, &opts).await?;
+                let res = self.pools[idx].new_multipart_upload(bucket, object, opts).await?;
                 return Ok((res, idx, opts.expected_bucket_incarnation_id));
             }
         }
@@ -473,9 +467,7 @@ impl ECStore {
             ));
         }
 
-        self.apply_decommission_target_mutation_fence(idx, object, &mut opts, mutation_fence)
-            .await;
-        let res = self.pools[idx].new_multipart_upload(bucket, object, &opts).await?;
+        let res = self.pools[idx].new_multipart_upload(bucket, object, opts).await?;
         Ok((res, idx, opts.expected_bucket_incarnation_id))
     }
 
@@ -712,14 +704,13 @@ impl ECStore {
 
     pub(crate) async fn complete_multipart_upload_for_data_movement(
         self: Arc<Self>,
-        target: (usize, Option<&ObjectLockDiagGuard>),
+        target_pool_idx: usize,
         bucket: &str,
         object: &str,
         upload_id: &str,
         uploaded_parts: Vec<CompletePart>,
         opts: &ObjectOptions,
     ) -> Result<ObjectInfo> {
-        let (target_pool_idx, mutation_fence) = target;
         check_complete_multipart_args(bucket, object, upload_id)?;
         if !opts.data_movement {
             return Err(Error::other("targeted multipart completion requires data_movement options"));
@@ -748,8 +739,6 @@ impl ECStore {
             snapshot.add_lock_fences(&mut opts);
             opts.object_lock_config_snapshot = Some(snapshot);
         }
-        self.apply_decommission_target_mutation_fence(target_pool_idx, object, &mut opts, mutation_fence)
-            .await;
         #[cfg(test)]
         pause_data_movement_multipart_before_selected_completion(bucket).await;
         let pool = self

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

@@ -73,7 +73,7 @@ pub(super) fn resolve_rebalance_delete_from_all_pools_result(
     object: &str,
 ) -> Result<ObjectInfo> {
     result.map_err(|err| {
-        if matches!(&err, Error::PreconditionFailed | Error::PrefixAccessDenied(_, _)) {
+        if err == Error::PreconditionFailed {
             err
         } else {
             Error::other(format!("failed to delete rebalance source object {bucket}/{object}: {err}"))
@@ -86,7 +86,7 @@ fn is_ignorable_rebalance_delete_error(err: &Error) -> bool {
 }
 
 fn rebalance_delete_pool_error(pool_idx: usize, bucket: &str, object: &str, err: Error) -> Error {
-    if matches!(&err, Error::PreconditionFailed | Error::PrefixAccessDenied(_, _)) {
+    if err == Error::PreconditionFailed {
         err
     } else {
         Error::other(format!("pool {pool_idx} delete failed for {bucket}/{object}: {err}"))
@@ -191,18 +191,6 @@ mod tests {
         assert_eq!(err, Error::PreconditionFailed);
     }
 
-    #[test]
-    fn rebalance_delete_result_preserves_prefix_access_denied() {
-        let err = resolve_rebalance_delete_from_all_pools_result(
-            Err(Error::PrefixAccessDenied("bucket".to_owned(), "object".to_owned())),
-            "bucket",
-            "object",
-        )
-        .expect_err("prefix access denial should remain structured");
-
-        assert_eq!(err, Error::PrefixAccessDenied("bucket".to_owned(), "object".to_owned()));
-    }
-
     #[test]
     fn rebalance_delete_pool_result_preserves_precondition_failed() {
         let err = resolve_rebalance_delete_from_all_pools_results(
@@ -217,19 +205,4 @@ mod tests {
 
         assert_eq!(err, Error::PreconditionFailed);
     }
-
-    #[test]
-    fn rebalance_delete_pool_result_preserves_prefix_access_denied() {
-        let err = resolve_rebalance_delete_from_all_pools_results(
-            vec![RebalanceDeletePoolResult {
-                pool_idx: 0,
-                result: Err(Error::PrefixAccessDenied("bucket".to_owned(), "object".to_owned())),
-            }],
-            "bucket",
-            "object",
-        )
-        .expect_err("prefix access denial should remain structured");
-
-        assert_eq!(err, Error::PrefixAccessDenied("bucket".to_owned(), "object".to_owned()));
-    }
 }

crates/protos/src/generated/proto_gen/node_service.rs

@@ -722,6 +722,10 @@ pub struct DeleteVersionsResponse {
     pub errors: ::prost::alloc::vec::Vec<::prost::alloc::string::String>,
     #[prost(message, optional, tag = "3")]
     pub error: ::core::option::Option<Error>,
+    /// Senders dual-write the legacy strings and typed entries. Receivers prefer typed entries
+    /// when present and fall back to strings for peers that predate this field. Code zero means success.
+    #[prost(message, repeated, tag = "4")]
+    pub item_errors: ::prost::alloc::vec::Vec<Error>,
 }
 #[derive(Clone, PartialEq, Eq, Hash, ::prost::Message)]
 pub struct ReadMultipleRequest {

crates/protos/src/node.proto

@@ -493,6 +493,9 @@ message DeleteVersionsResponse {
   bool success = 1;
   repeated string errors = 2;
   optional Error error = 3;
+  // Senders dual-write the legacy strings and typed entries. Receivers prefer typed entries
+  // when present and fall back to strings for peers that predate this field. Code zero means success.
+  repeated Error item_errors = 4;
 }
 
 message ReadMultipleRequest {

crates/scanner/src/data_usage_define.rs

@@ -125,6 +125,34 @@ pub(crate) async fn read_config_with_revision<S: ScannerObjectIO>(
     }
 }
 
+/// Read only the object revision without materializing its body.
+pub(crate) async fn read_config_revision<S: ScannerObjectIO>(store: Arc<S>, path: &str) -> StorageResult<DataUsageCacheRevision> {
+    match store
+        .get_object_reader(
+            RUSTFS_META_BUCKET,
+            path,
+            None,
+            HeaderMap::new(),
+            &ObjectOptions {
+                no_lock: true,
+                ..Default::default()
+            },
+        )
+        .await
+    {
+        Ok(reader) => reader
+            .object_info
+            .etag
+            .filter(|etag| !etag.is_empty())
+            .map(DataUsageCacheRevision::Etag)
+            .ok_or_else(|| StorageError::other(format!("scanner config object {path} has no ETag"))),
+        Err(Error::FileNotFound | Error::VolumeNotFound | Error::ObjectNotFound(_, _) | Error::BucketNotFound(_)) => {
+            Ok(DataUsageCacheRevision::Missing)
+        }
+        Err(err) => Err(err),
+    }
+}
+
 #[derive(Clone, Debug)]
 pub(crate) struct DataUsageCacheRevisions {
     main: DataUsageCacheRevision,
@@ -146,6 +174,11 @@ pub static LEGACY_DATA_USAGE_OBJ_NAME_PATH: LazyLock<String> =
 pub static DATA_USAGE_BLOOM_NAME_PATH: LazyLock<String> =
     LazyLock::new(|| format!("{BUCKET_META_PREFIX}{SLASH_SEPARATOR}{DATA_USAGE_BLOOM_NAME}"));
 
+/// Durable companion object for a cycle-state object which cannot be decoded.
+/// The primary object is deliberately never replaced or deleted by recovery.
+pub static DATA_USAGE_BLOOM_RECOVERY_PATH: LazyLock<String> =
+    LazyLock::new(|| format!("{}.recovery-required.json", DATA_USAGE_BLOOM_NAME_PATH.as_str()));
+
 pub static BACKGROUND_HEAL_INFO_PATH: LazyLock<String> =
     LazyLock::new(|| format!("{BUCKET_META_PREFIX}{SLASH_SEPARATOR}.background-heal.json"));
 

crates/scanner/src/data_usage_define/persistence.rs

@@ -74,7 +74,7 @@ impl DataUsageCache {
         let loaded = Self::load_cache(store.clone(), name).await?;
         let backup = match loaded.backup_revision {
             Some(revision) => Some(revision),
-            None => match Self::revision_for_path(store, &backup_path).await {
+            None => match read_config_revision(store, &backup_path).await {
                 Ok(revision) => Some(revision),
                 Err(err) => {
                     counter!(METRIC_CACHE_BACKUP_REVISION_FAILURE_TOTAL).increment(1);
@@ -336,33 +336,6 @@ impl DataUsageCache {
         }
     }
 
-    async fn revision_for_path<S: ScannerObjectIO>(store: Arc<S>, path: &str) -> StorageResult<DataUsageCacheRevision> {
-        match store
-            .get_object_reader(
-                RUSTFS_META_BUCKET,
-                path,
-                None,
-                HeaderMap::new(),
-                &ObjectOptions {
-                    no_lock: true,
-                    ..Default::default()
-                },
-            )
-            .await
-        {
-            Ok(reader) => reader
-                .object_info
-                .etag
-                .filter(|etag| !etag.is_empty())
-                .map(DataUsageCacheRevision::Etag)
-                .ok_or_else(|| StorageError::other(format!("scanner cache object {path} has no ETag"))),
-            Err(Error::FileNotFound | Error::VolumeNotFound | Error::ObjectNotFound(_, _) | Error::BucketNotFound(_)) => {
-                Ok(DataUsageCacheRevision::Missing)
-            }
-            Err(err) => Err(err),
-        }
-    }
-
     pub(super) fn cache_save_timeout() -> Duration {
         crate::runtime_config::scanner_cache_save_timeout()
     }

crates/scanner/src/lib.rs

@@ -75,7 +75,10 @@ pub use remote_scanner::{
 };
 pub use runtime_config::{apply_scanner_runtime_config, scanner_runtime_config_status, validate_scanner_runtime_config};
 pub use rustfs_common::last_minute;
-pub use scanner::{ScannerCycleScheduleStatus, init_data_scanner, scanner_cycle_schedule_status, scanner_topology_digest};
+pub use scanner::{
+    ScannerCycleRecoveryMarker, ScannerCycleRecoveryStatus, ScannerCycleScheduleStatus, init_data_scanner,
+    reset_scanner_cycle_recovery, scanner_cycle_recovery_status, scanner_cycle_schedule_status, scanner_topology_digest,
+};
 pub use scanner_io::{
     ScannerDirtyUsageAckError, ScannerDirtyUsageState, acknowledge_dirty_usage_generation, clear_dirty_usage_bucket,
     record_dirty_usage_bucket, record_scanner_maintenance_change, scanner_activity_epoch, scanner_dirty_usage_state,

crates/scanner/src/scanner.rs

@@ -20,7 +20,7 @@ use std::sync::{Arc, LazyLock, RwLock};
 
 use crate::data_usage_define::{
     BACKGROUND_HEAL_INFO_PATH, DATA_USAGE_BLOOM_NAME_PATH, DATA_USAGE_OBJ_NAME_PATH, DATA_USAGE_OBSERVED_OBJ_NAME_PATH,
-    DataUsageCache, DataUsageCacheRevision, LEGACY_DATA_USAGE_OBJ_NAME_PATH, read_config_with_revision,
+    DataUsageCache, DataUsageCacheRevision, LEGACY_DATA_USAGE_OBJ_NAME_PATH, read_config_revision, read_config_with_revision,
 };
 use crate::runtime_config::{
     ScannerRuntimeConfig, ScannerRuntimeConfigSource, refresh_scanner_runtime_config_from_global, scanner_bitrot_cycle,
@@ -54,9 +54,7 @@ use rustfs_config::{ENV_SCANNER_CYCLE, ENV_SCANNER_SPEED, ENV_SCANNER_START_DELA
 use rustfs_data_usage::observed_data_usage_is_newer;
 use serde::{Deserialize, Serialize};
 use sha2::{Digest as _, Sha256};
-#[cfg(test)]
-use tokio::sync::Notify;
-use tokio::sync::mpsc;
+use tokio::sync::{Notify, mpsc};
 use tokio::time::{Duration, Instant};
 use tokio_util::sync::CancellationToken;
 use tokio_util::task::AbortOnDropHandle;
@@ -104,6 +102,13 @@ const CLEAN_IDLE_BACKOFF_FACTOR: u32 = 2;
 /// unavailable peer cannot drive a tight retry loop.
 const SCANNER_RETRY_BASE_INTERVAL: Duration = Duration::from_secs(5);
 const SCANNER_RETRY_MAX_INTERVAL: Duration = Duration::from_secs(30 * 60);
+/// A transient backend outage remains self-healing after the short retry
+/// budget is exhausted, but the probe is intentionally sparse until storage
+/// recovers or an operator reset wakes the scanner.
+const SCANNER_CYCLE_RECOVERY_PAUSED_INTERVAL: Duration = Duration::from_secs(5 * 60);
+/// Permanent recovery states still get a sparse status probe so a reset that
+/// races the wait registration cannot leave the scanner asleep forever.
+const SCANNER_CYCLE_RECOVERY_BLOCKED_PROBE_INTERVAL: Duration = Duration::from_secs(5 * 60);
 const SCANNER_LEADER_LOCK_POLL_INTERVAL: Duration = Duration::from_secs(1);
 #[cfg(not(test))]
 const SCANNER_LOCK_LOSS_SHUTDOWN_TIMEOUT: Duration = Duration::from_secs(30);
@@ -125,6 +130,12 @@ type ScannerCycleStatePersistTestHook = (u64, Arc<Notify>);
 static SCANNER_CYCLE_STATE_PERSIST_TEST_HOOK: LazyLock<StdMutex<Option<ScannerCycleStatePersistTestHook>>> =
     LazyLock::new(|| StdMutex::new(None));
 
+static SCANNER_CYCLE_RECOVERY_WAKE: LazyLock<Notify> = LazyLock::new(Notify::new);
+
+pub(super) fn notify_scanner_cycle_recovery_wake() {
+    SCANNER_CYCLE_RECOVERY_WAKE.notify_one();
+}
+
 #[cfg(test)]
 struct ScannerCycleStatePersistTestHookGuard;
 
@@ -576,19 +587,21 @@ pub async fn init_data_scanner(ctx: CancellationToken, storeapi: Arc<ECStore>) {
             tokio::time::sleep(sleep_time).await;
         }
 
+        let mut transient_backoff = ScannerRetryBackoff::default();
+        let mut recovery_retry_count = 0_u32;
         loop {
             if ctx_clone.is_cancelled() {
                 break;
             }
 
-            if let Err(e) = run_data_scanner_with_maintenance_state(
+            let run_result = run_data_scanner_with_maintenance_state(
                 ctx_clone.clone(),
                 storeapi_clone.clone(),
                 startup_features,
                 startup_maintenance_generation,
             )
-            .await
-            {
+            .await;
+            if let Err(e) = &run_result {
                 error!(
                     target: "rustfs::scanner",
                     event = EVENT_SCANNER_CYCLE_STATE,
@@ -599,11 +612,52 @@ pub async fn init_data_scanner(ctx: CancellationToken, storeapi: Arc<ECStore>) {
                     "Scanner runtime iteration failed"
                 );
             }
+            let recovery_status = scanner_cycle_recovery_status();
+            if recovery_status.retryable {
+                recovery_retry_count = recovery_retry_count.saturating_add(1);
+                let _ = record_scanner_cycle_recovery_retry(recovery_retry_count);
+            } else {
+                recovery_retry_count = 0;
+            }
+
+            let recovery_status = scanner_cycle_recovery_status();
+            if recovery_status.state == "paused" {
+                transient_backoff.record_retryable_cycle(false);
+                tokio::select! {
+                    _ = ctx_clone.cancelled() => break,
+                    _ = SCANNER_CYCLE_RECOVERY_WAKE.notified() => {},
+                    _ = tokio::time::sleep(SCANNER_CYCLE_RECOVERY_PAUSED_INTERVAL) => {},
+                }
+                recovery_retry_count = 0;
+                continue;
+            }
+            if !recovery_status.retryable
+                && matches!(recovery_status.state.as_str(), "blocked" | "recovery-required" | "cleanup-pending")
+            {
+                transient_backoff.record_retryable_cycle(false);
+                tokio::select! {
+                    _ = ctx_clone.cancelled() => break,
+                    _ = SCANNER_CYCLE_RECOVERY_WAKE.notified() => {},
+                    _ = tokio::time::sleep(SCANNER_CYCLE_RECOVERY_BLOCKED_PROBE_INTERVAL) => {},
+                }
+                continue;
+            }
+
+            let retry_delay = if recovery_status.retryable || run_result.is_err() {
+                transient_backoff.record_retryable_cycle(true);
+                transient_backoff
+                    .retry_interval(scanner_cycle_interval())
+                    .unwrap_or(SCANNER_RETRY_BASE_INTERVAL)
+            } else {
+                transient_backoff.record_retryable_cycle(false);
+                randomized_cycle_delay()
+            };
             // Backoff before retrying after lock contention or scanner-level failures.
             // Keep this cancellation-aware so shutdown is not delayed by backoff sleep.
             tokio::select! {
                 _ = ctx_clone.cancelled() => break,
-                _ = tokio::time::sleep(randomized_cycle_delay()) => {}
+                _ = SCANNER_CYCLE_RECOVERY_WAKE.notified() => {},
+                _ = tokio::time::sleep(retry_delay) => {}
             }
         }
     });
@@ -1606,40 +1660,22 @@ async fn run_data_scanner_with_maintenance_state(
         observe_scanner_activity(&storeapi, distributed, &mut scanner_activity_seen).await;
     }
 
-    let (buf, mut cycle_revision) = match read_config_with_revision(storeapi.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str()).await {
-        Ok((buf, revision)) => (buf.unwrap_or_default(), revision),
-        Err(err) => {
-            error!(
-                target: "rustfs::scanner",
-                event = EVENT_SCANNER_PERSIST_STATE,
-                component = LOG_COMPONENT_SCANNER,
-                subsystem = LOG_SUBSYSTEM_RUNTIME,
-                path = %&*DATA_USAGE_BLOOM_NAME_PATH,
-                state = "revision_load_failed",
-                error = %err,
-                "Scanner cycle state revision load failed"
-            );
-            global_metrics().set_cycle(None).await;
-            return Ok(());
-        }
-    };
-    let (mut cycle_info, mut leader_epoch) = match decode_scanner_cycle_state_for_startup(&buf) {
-        Ok(state) => state,
-        Err(err) => {
-            error!(
-                target: "rustfs::scanner",
-                event = EVENT_SCANNER_PERSIST_STATE,
-                component = LOG_COMPONENT_SCANNER,
-                subsystem = LOG_SUBSYSTEM_RUNTIME,
-                path = %&*DATA_USAGE_BLOOM_NAME_PATH,
-                state = "cycle_decode_failed",
-                error = %err,
-                "Scanner stopped because persisted cycle state is invalid"
-            );
-            global_metrics().set_cycle(None).await;
-            return Ok(());
-        }
-    };
+    let (mut cycle_info, mut leader_epoch, mut cycle_revision) =
+        match load_scanner_cycle_state_for_startup(storeapi.clone()).await {
+            ScannerCycleStateStartup::Ready {
+                cycle,
+                leader_epoch,
+                revision,
+            } => (cycle, leader_epoch, revision),
+            ScannerCycleStateStartup::Blocked => {
+                global_metrics().set_cycle(None).await;
+                return Ok(());
+            }
+            ScannerCycleStateStartup::Transient(err) => {
+                global_metrics().set_cycle(None).await;
+                return Err(err);
+            }
+        };
     let usage_floor = match persisted_usage_floor(storeapi.clone()).await {
         Ok(floor) => floor,
         Err(err) => {
@@ -2219,7 +2255,12 @@ pub(crate) use activity::{
 pub(crate) use activity::{ScannerCycleOutcome, scanner_cycle_outcome_with_pending_maintenance};
 #[cfg(test)]
 pub(crate) use cycle_state::encode_scanner_cycle_fence_for_test;
-pub(crate) use cycle_state::{current_scanner_leader_epoch, decode_persisted_scanner_cycle_fence};
+pub use cycle_state::{
+    ScannerCycleRecoveryMarker, ScannerCycleRecoveryStatus, reset_scanner_cycle_recovery, scanner_cycle_recovery_status,
+};
+pub(crate) use cycle_state::{
+    current_scanner_leader_epoch, decode_persisted_scanner_cycle_fence, load_scanner_cycle_state_for_startup,
+};
 pub use heal_info::{BackgroundHealInfo, read_background_heal_info, save_background_heal_info};
 pub use usage_store::store_data_usage_in_backend;
 

crates/scanner/src/scanner/leadership.rs

@@ -196,7 +196,7 @@ pub(super) async fn claim_scanner_leadership(
         if ctx.is_cancelled() {
             return false;
         }
-        let Some(claimed_epoch) = persisted_epoch.checked_add(1) else {
+        let Some(claimed_epoch) = persisted_epoch.checked_add(1).filter(|epoch| *epoch < u64::MAX) else {
             error!(
                 target: "rustfs::scanner",
                 event = EVENT_SCANNER_PERSIST_STATE,

crates/scanner/src/scanner/tests.rs

@@ -15,11 +15,12 @@
 use super::*;
 use crate::EcstoreResult;
 use crate::{
-    Endpoint, EndpointServerPools, Endpoints, InstanceContext, PoolEndpoints, ScannerGetObjectReader as GetObjectReader,
-    ScannerObjectInfo as ObjectInfo, ScannerObjectOptions as ObjectOptions, ScannerPutObjReader as PutObjReader,
-    init_bucket_metadata_sys_for_scanner_tests, init_ecstore_config_for_scanner_tests, init_local_disks_with_instance_ctx,
+    DATA_USAGE_BLOOM_RECOVERY_PATH, Endpoint, EndpointServerPools, Endpoints, InstanceContext, PoolEndpoints,
+    ScannerGetObjectReader as GetObjectReader, ScannerObjectInfo as ObjectInfo, ScannerObjectOptions as ObjectOptions,
+    ScannerPutObjReader as PutObjReader, init_bucket_metadata_sys_for_scanner_tests, init_ecstore_config_for_scanner_tests,
+    init_local_disks_with_instance_ctx,
 };
-use std::collections::HashMap;
+use std::collections::{HashMap, HashSet};
 use std::io::Cursor;
 use std::task::Poll;
 use temp_env::{with_var, with_var_unset};
@@ -117,6 +118,15 @@ async fn scanner_cycle_lock_fence_bounds_uncooperative_shutdown() {
     assert!(cycle_ctx.is_cancelled());
 }
 
+#[tokio::test]
+async fn scanner_cycle_recovery_wake_survives_wait_registration_race() {
+    notify_scanner_cycle_recovery_wake();
+
+    tokio::time::timeout(Duration::from_secs(1), SCANNER_CYCLE_RECOVERY_WAKE.notified())
+        .await
+        .expect("recovery wake should retain a permit until the waiter registers");
+}
+
 struct ScannerDefaultSpeedGuard;
 
 impl ScannerDefaultSpeedGuard {
@@ -151,6 +161,7 @@ impl Drop for ScannerDefaultCycleGuard {
 struct MemoryConfigStore {
     objects: Mutex<HashMap<String, Vec<u8>>>,
     revisions: Mutex<HashMap<String, u64>>,
+    non_regular_objects: Mutex<HashSet<String>>,
     fail_put_number: Mutex<HashMap<String, usize>>,
     object_not_found_put_number: Mutex<HashMap<String, usize>>,
     error_after_commit_put_number: Mutex<HashMap<String, usize>>,
@@ -191,12 +202,16 @@ impl crate::storage_api::scanner_io::ObjectIO for MemoryConfigStore {
             .get(&key)
             .cloned()
             .ok_or(EcstoreError::FileNotFound)?;
-        let revision = *self.revisions.lock().await.entry(key).or_insert(1);
+        let data_len = i64::try_from(data.len()).expect("memory test object length should fit in i64");
+        let revision = *self.revisions.lock().await.entry(key.clone()).or_insert(1);
+        let is_dir = self.non_regular_objects.lock().await.contains(&key);
 
         Ok(GetObjectReader {
             stream: Box::new(Cursor::new(data)),
             object_info: ObjectInfo {
                 etag: Some(format!("memory-{revision}")),
+                size: data_len,
+                is_dir,
                 ..Default::default()
             },
             buffered_body: None,
@@ -797,6 +812,10 @@ fn scanner_cycle_state_decodes_legacy_and_fenced_formats() {
     let (fenced_cycle, fenced_epoch) = decode_scanner_cycle_state(&fenced).expect("fenced cycle state should decode");
     assert_eq!(fenced_cycle.next, 13);
     assert_eq!(fenced_epoch, 7);
+
+    let mut trailing = fenced;
+    trailing.push(0);
+    assert!(decode_scanner_cycle_state(&trailing).is_err());
 }
 
 #[test]
@@ -823,6 +842,840 @@ fn scanner_startup_fails_closed_on_nonempty_corrupt_cycle_state() {
     assert!(encode_scanner_cycle_state(&exhausted, 7).is_err());
 }
 
+#[tokio::test]
+async fn corrupt_cycle_state_is_quarantined_once() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let state_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_NAME_PATH.as_str());
+    store.objects.lock().await.insert(state_key.clone(), vec![1]);
+    store.revisions.lock().await.insert(state_key.clone(), 7);
+
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store.clone()).await,
+        ScannerCycleStateStartup::Blocked
+    ));
+    let marker_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str());
+    let marker_data = store
+        .objects
+        .lock()
+        .await
+        .get(&marker_key)
+        .cloned()
+        .expect("corrupt state must leave a durable recovery marker");
+    let marker: ScannerCycleRecoveryMarker = serde_json::from_slice(&marker_data).expect("marker should be valid JSON");
+    assert_eq!(marker.primary_revision, "memory-7");
+    assert_eq!(marker.path, DATA_USAGE_BLOOM_NAME_PATH.as_str());
+    assert_eq!(marker.quarantine_path, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str());
+    assert_eq!(marker.classification, "corrupt");
+
+    // A second startup sees the matching marker before consuming the poison body.
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store.clone()).await,
+        ScannerCycleStateStartup::Blocked
+    ));
+
+    // Replacing the primary object advances its revision; the stale marker must
+    // not quarantine the newer, valid state.
+    let cycle = CurrentCycle {
+        next: 9,
+        ..Default::default()
+    };
+    let encoded = encode_scanner_cycle_state(&cycle, 3).expect("valid state should encode");
+    store.objects.lock().await.insert(state_key.clone(), encoded);
+    store.revisions.lock().await.insert(state_key, 8);
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store).await,
+        ScannerCycleStateStartup::Ready {
+            cycle: CurrentCycle { next: 9, .. },
+            leader_epoch: 3,
+            ..
+        }
+    ));
+}
+
+#[tokio::test]
+async fn empty_cycle_state_object_is_quarantined_as_corrupt() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let state_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_NAME_PATH.as_str());
+    store.objects.lock().await.insert(state_key.clone(), Vec::new());
+    store.revisions.lock().await.insert(state_key, 6);
+
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store).await,
+        ScannerCycleStateStartup::Blocked
+    ));
+    assert_eq!(scanner_cycle_recovery_status().classification.as_deref(), Some("corrupt"));
+    assert!(
+        scanner_cycle_recovery_status()
+            .reason
+            .as_deref()
+            .is_some_and(|reason| reason.contains("empty"))
+    );
+}
+
+#[tokio::test]
+async fn future_cycle_state_schema_is_recovery_required() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let state_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_NAME_PATH.as_str());
+    let mut future = 17_u64.to_le_bytes().to_vec();
+    future.extend_from_slice(b"RSCYC999");
+    future.extend_from_slice(&4_u64.to_le_bytes());
+    future.extend_from_slice(&[0x90]);
+    store.objects.lock().await.insert(state_key.clone(), future);
+    store.revisions.lock().await.insert(state_key, 13);
+
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store).await,
+        ScannerCycleStateStartup::Blocked
+    ));
+    assert_eq!(scanner_cycle_recovery_status().classification.as_deref(), Some("future_schema"));
+}
+
+#[tokio::test]
+async fn concurrent_leaders_cannot_quarantine_newer_cycle_state() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let state_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_NAME_PATH.as_str());
+    store.objects.lock().await.insert(state_key.clone(), vec![1]);
+    store.revisions.lock().await.insert(state_key, 4);
+
+    let (first, second) = tokio::join!(
+        load_scanner_cycle_state_for_startup(store.clone()),
+        load_scanner_cycle_state_for_startup(store.clone()),
+    );
+    assert!(matches!(first, ScannerCycleStateStartup::Blocked));
+    assert!(matches!(second, ScannerCycleStateStartup::Blocked));
+
+    let marker_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str());
+    let marker_data = store
+        .objects
+        .lock()
+        .await
+        .get(&marker_key)
+        .cloned()
+        .expect("one contender must publish the recovery marker");
+    let marker: ScannerCycleRecoveryMarker = serde_json::from_slice(&marker_data).expect("marker should decode");
+    assert_eq!(marker.primary_revision, "memory-4");
+}
+
+#[tokio::test]
+async fn cleanup_pending_marker_blocks_a_rewritten_primary_after_restart() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let state_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_NAME_PATH.as_str());
+    let marker_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str());
+    let encoded = encode_scanner_cycle_state(
+        &CurrentCycle {
+            next: 12,
+            ..Default::default()
+        },
+        8,
+    )
+    .expect("valid state should encode");
+    store.objects.lock().await.insert(state_key.clone(), encoded);
+    store.revisions.lock().await.insert(state_key, 22);
+    let marker = ScannerCycleRecoveryMarker {
+        schema_version: 1,
+        primary_revision: "memory-21".to_string(),
+        generation: 11,
+        leader_epoch: 7,
+        classification: "corrupt".to_string(),
+        first_detected_at_unix_secs: 1,
+        last_attempt_at_unix_secs: 2,
+        retry_count: 1,
+        reason: "reset in progress".to_string(),
+        path: DATA_USAGE_BLOOM_NAME_PATH.clone(),
+        quarantine_path: DATA_USAGE_BLOOM_RECOVERY_PATH.clone(),
+        state: "cleanup-pending".to_string(),
+    };
+    store
+        .objects
+        .lock()
+        .await
+        .insert(marker_key.clone(), serde_json::to_vec(&marker).expect("marker should encode"));
+    store.revisions.lock().await.insert(marker_key, 3);
+
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store).await,
+        ScannerCycleStateStartup::Blocked
+    ));
+    assert_eq!(scanner_cycle_recovery_status().state, "cleanup-pending");
+}
+
+#[test]
+fn full_rescan_reset_accepts_unknown_marker_fields_without_trusting_cursor() {
+    let marker = br#"{
+        "schema_version": 99,
+        "primary_revision": "memory-7",
+        "generation": 9000,
+        "leader_epoch": 9000,
+        "classification": "new-future-classification",
+        "first_detected_at_unix_secs": 1,
+        "last_attempt_at_unix_secs": 2,
+        "retry_count": 9,
+        "reason": "future marker",
+        "path": "buckets/.bloomcycle.bin",
+        "quarantine_path": "buckets/.bloomcycle.bin.recovery-required.json",
+        "future_field": {"cursor": "untrusted"}
+    }"#;
+    let decoded =
+        super::cycle_state::decode_recovery_marker_for_reset(marker, &DataUsageCacheRevision::Etag("memory-3".to_string()))
+            .expect("full-rescan compatibility decoder should accept additive fields");
+    assert_eq!(decoded.primary_revision, "memory-7");
+    assert_eq!(decoded.classification, "future_schema");
+    assert_eq!(decoded.generation, 0);
+    assert_eq!(decoded.leader_epoch, 0);
+    assert_eq!(decoded.state, "blocked");
+
+    let malformed =
+        super::cycle_state::decode_recovery_marker_for_reset(b"{not-json", &DataUsageCacheRevision::Etag("memory-4".to_string()))
+            .expect("a full-rescan reset must recover even when the marker is malformed");
+    assert!(malformed.primary_revision.is_empty());
+    assert_eq!(malformed.classification, "future_schema");
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rebuilds_after_malformed_marker_without_trusting_cursor() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), vec![0xff, 0x00, 0x01])
+        .await
+        .expect("corrupt cycle state should be persisted");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), br#"{not-json"#.to_vec())
+        .await
+        .expect("malformed marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("full-rescan reset should recover malformed marker");
+
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("rebuilt cycle state should remain durable");
+    let (cycle, leader_epoch) = decode_scanner_cycle_state(&state).expect("rebuilt cycle state should decode");
+    assert_eq!(cycle.next, 0, "reset must use the verified usage floor, not marker cursor");
+    assert_eq!(leader_epoch, 1);
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_ignores_epoch_from_malformed_future_primary() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    let mut future_primary = vec![0; 24];
+    future_primary[8..16].copy_from_slice(b"RSCY9999");
+    future_primary[16..24].copy_from_slice(&u64::MAX.to_le_bytes());
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), future_primary)
+        .await
+        .expect("future cycle state should be persisted");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), br#"{not-json"#.to_vec())
+        .await
+        .expect("malformed marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("full-rescan reset should recover malformed future state");
+
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("rebuilt cycle state should remain durable");
+    let (_, leader_epoch) = decode_scanner_cycle_state(&state).expect("rebuilt cycle state should decode");
+    assert_eq!(leader_epoch, 1, "invalid persisted bytes must not raise the recovery epoch");
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn ecstore_exact_recovery_marker_delete_honors_etag() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), b"marker-v1".to_vec())
+        .await
+        .expect("initial recovery marker should be persisted");
+    let (_, stale_revision) = read_config_with_revision(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str())
+        .await
+        .expect("initial marker revision should load");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), b"marker-v2".to_vec())
+        .await
+        .expect("replacement recovery marker should be persisted");
+
+    let delete_result = store
+        .delete_config_object(
+            RUSTFS_META_BUCKET,
+            DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(),
+            ObjectOptions {
+                http_preconditions: Some(stale_revision.preconditions()),
+                ..Default::default()
+            },
+        )
+        .await;
+    assert!(matches!(delete_result, Err(EcstoreError::PreconditionFailed)));
+    assert_eq!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str())
+            .await
+            .expect("replacement marker should remain durable"),
+        b"marker-v2"
+    );
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rejects_corrupt_primary_under_stale_blocked_marker() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    let corrupt_primary = vec![0xff, 0x00, 0x01];
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), corrupt_primary.clone())
+        .await
+        .expect("corrupt cycle state should be persisted");
+    let (_, primary_revision) = read_config_with_revision(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("primary revision should load");
+    let marker = ScannerCycleRecoveryMarker {
+        schema_version: 1,
+        primary_revision: "memory-stale".to_string(),
+        generation: 1,
+        leader_epoch: 1,
+        classification: "corrupt".to_string(),
+        first_detected_at_unix_secs: 1,
+        last_attempt_at_unix_secs: 2,
+        retry_count: 1,
+        reason: "blocked primary changed".to_string(),
+        path: DATA_USAGE_BLOOM_NAME_PATH.clone(),
+        quarantine_path: DATA_USAGE_BLOOM_RECOVERY_PATH.clone(),
+        state: "blocked".to_string(),
+    };
+    let marker_data = serde_json::to_vec(&marker).expect("blocked marker should encode");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), marker_data.clone())
+        .await
+        .expect("blocked marker should be persisted");
+
+    assert!(
+        reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+            .await
+            .is_err(),
+        "a strict marker must fail closed when its primary revision changed"
+    );
+    assert_eq!(
+        read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+            .await
+            .expect("primary should remain readable"),
+        corrupt_primary
+    );
+    assert_eq!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str())
+            .await
+            .expect("blocked marker should remain durable"),
+        marker_data
+    );
+    assert!(!matches!(primary_revision, DataUsageCacheRevision::Missing));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_preserves_valid_primary_when_marker_is_malformed() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    let primary = CurrentCycle {
+        next: 42,
+        ..Default::default()
+    };
+    let old_primary_data = encode_scanner_cycle_state(&primary, 7).expect("valid cycle state should encode");
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), old_primary_data.clone())
+        .await
+        .expect("valid cycle state should be persisted");
+    let (_, old_primary_revision) = read_config_with_revision(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("primary state revision should load");
+    let old_usage = DataUsageInfo {
+        scanner_epoch: Some(7),
+        scanner_cycle: Some(41),
+        ..Default::default()
+    };
+    let old_usage_data = serde_json::to_vec(&old_usage).expect("usage snapshot should encode");
+    save_config(store.clone(), DATA_USAGE_OBJ_NAME_PATH.as_str(), old_usage_data.clone())
+        .await
+        .expect("usage snapshot should be persisted");
+    let (_, old_usage_revision) = read_config_with_revision(store.clone(), DATA_USAGE_OBJ_NAME_PATH.as_str())
+        .await
+        .expect("usage snapshot revision should load");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), b"{not-json".to_vec())
+        .await
+        .expect("malformed marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("reset should clear a stale malformed marker");
+
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("valid primary should remain durable");
+    let (cycle, leader_epoch) = decode_scanner_cycle_state(&state).expect("primary cycle state should decode");
+    assert_eq!(cycle.next, 42, "reset must not regress an independently fenced primary");
+    assert_eq!(leader_epoch, 8, "reset must advance the preserved primary epoch");
+    let stale_primary_save = save_config_with_preconditions(
+        store.clone(),
+        DATA_USAGE_BLOOM_NAME_PATH.as_str(),
+        old_primary_data,
+        old_primary_revision.preconditions(),
+    )
+    .await;
+    assert!(matches!(stale_primary_save, Err(EcstoreError::PreconditionFailed)));
+    let usage = read_config(store.clone(), DATA_USAGE_OBJ_NAME_PATH.as_str())
+        .await
+        .expect("usage epoch fence should remain durable");
+    assert_eq!(
+        serde_json::from_slice::<DataUsageInfo>(&usage)
+            .expect("fenced usage should decode")
+            .scanner_epoch,
+        Some(8)
+    );
+    let stale_save = save_config_with_preconditions(
+        store.clone(),
+        DATA_USAGE_OBJ_NAME_PATH.as_str(),
+        old_usage_data,
+        old_usage_revision.preconditions(),
+    )
+    .await;
+    assert!(matches!(stale_save, Err(EcstoreError::PreconditionFailed)));
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_resumes_cleanup_pending_preserved_primary() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    let completed_at = Utc::now();
+    let primary = CurrentCycle {
+        current: 3,
+        next: 42,
+        cycle_completed: vec![completed_at],
+        started: completed_at,
+    };
+    save_config(
+        store.clone(),
+        DATA_USAGE_BLOOM_NAME_PATH.as_str(),
+        encode_scanner_cycle_state(&primary, 7).expect("valid cycle state should encode"),
+    )
+    .await
+    .expect("valid cycle state should be persisted");
+    let usage = DataUsageInfo {
+        scanner_epoch: Some(7),
+        scanner_cycle: Some(41),
+        ..Default::default()
+    };
+    save_config(
+        store.clone(),
+        DATA_USAGE_OBJ_NAME_PATH.as_str(),
+        serde_json::to_vec(&usage).expect("usage snapshot should encode"),
+    )
+    .await
+    .expect("usage snapshot should be persisted");
+    let marker = ScannerCycleRecoveryMarker {
+        schema_version: 1,
+        primary_revision: "memory-old".to_string(),
+        generation: 41,
+        leader_epoch: 7,
+        classification: "corrupt".to_string(),
+        first_detected_at_unix_secs: 1,
+        last_attempt_at_unix_secs: 2,
+        retry_count: 1,
+        reason: "reset in progress".to_string(),
+        path: DATA_USAGE_BLOOM_NAME_PATH.clone(),
+        quarantine_path: DATA_USAGE_BLOOM_RECOVERY_PATH.clone(),
+        state: "cleanup-pending".to_string(),
+    };
+    save_config(
+        store.clone(),
+        DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(),
+        serde_json::to_vec(&marker).expect("marker should encode"),
+    )
+    .await
+    .expect("cleanup marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("reset should resume a cleanup-pending preserved primary");
+
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("preserved cycle state should remain durable");
+    let (cycle, leader_epoch) = decode_scanner_cycle_state(&state).expect("cycle state should decode");
+    assert_eq!(cycle.current, 3, "cleanup retry must preserve the in-progress cursor");
+    assert_eq!(cycle.next, 42);
+    assert_eq!(cycle.cycle_completed, vec![completed_at]);
+    assert_eq!(cycle.started, completed_at);
+    assert_eq!(leader_epoch, 8);
+    let usage = read_config(store.clone(), DATA_USAGE_OBJ_NAME_PATH.as_str())
+        .await
+        .expect("usage epoch fence should remain durable");
+    assert_eq!(
+        serde_json::from_slice::<DataUsageInfo>(&usage)
+            .expect("usage should decode")
+            .scanner_epoch,
+        Some(8)
+    );
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rebuilds_oversized_regular_primary_with_malformed_marker() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), vec![0; 1024 * 1024 + 1])
+        .await
+        .expect("oversized cycle state should be persisted");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), b"{not-json".to_vec())
+        .await
+        .expect("malformed marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("explicit full-rescan reset should replace an oversized regular primary");
+
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("rebuilt cycle state should remain durable");
+    let (cycle, leader_epoch) = decode_scanner_cycle_state(&state).expect("rebuilt cycle state should decode");
+    assert_eq!(cycle.next, 0);
+    assert_eq!(leader_epoch, 1);
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rebuilds_oversized_primary_after_cleanup_marker() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), vec![0; 1024 * 1024 + 1])
+        .await
+        .expect("oversized cycle state should be persisted");
+    let (_, primary_revision) = read_config_with_revision(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("primary revision should load");
+    let marker = ScannerCycleRecoveryMarker {
+        schema_version: 1,
+        primary_revision: match primary_revision {
+            DataUsageCacheRevision::Etag(etag) => etag,
+            DataUsageCacheRevision::Missing => panic!("primary revision should be present"),
+        },
+        generation: 1,
+        leader_epoch: 1,
+        classification: "corrupt".to_string(),
+        first_detected_at_unix_secs: 1,
+        last_attempt_at_unix_secs: 2,
+        retry_count: 1,
+        reason: "reset in progress".to_string(),
+        path: DATA_USAGE_BLOOM_NAME_PATH.clone(),
+        quarantine_path: DATA_USAGE_BLOOM_RECOVERY_PATH.clone(),
+        state: "cleanup-pending".to_string(),
+    };
+    save_config(
+        store.clone(),
+        DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(),
+        serde_json::to_vec(&marker).expect("cleanup marker should encode"),
+    )
+    .await
+    .expect("cleanup marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("cleanup retry should rebuild an oversized primary");
+
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("rebuilt cycle state should remain durable");
+    let (cycle, leader_epoch) = decode_scanner_cycle_state(&state).expect("rebuilt cycle state should decode");
+    assert_eq!(cycle.next, 0);
+    assert_eq!(leader_epoch, 1);
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rebuilds_with_oversized_marker() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), vec![0xff, 0x00, 0x01])
+        .await
+        .expect("corrupt cycle state should be persisted");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), vec![b'x'; 64 * 1024 + 1])
+        .await
+        .expect("oversized recovery marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("full-rescan reset should recover an oversized marker");
+
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("rebuilt cycle state should remain durable");
+    let (_, leader_epoch) = decode_scanner_cycle_state(&state).expect("rebuilt cycle state should decode");
+    assert_eq!(leader_epoch, 1);
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rebuilds_with_empty_marker() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), vec![0xff, 0x00, 0x01])
+        .await
+        .expect("corrupt cycle state should be persisted");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), Vec::new())
+        .await
+        .expect("empty recovery marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("full-rescan reset should recover an empty marker");
+
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("rebuilt cycle state should remain durable");
+    let (_, leader_epoch) = decode_scanner_cycle_state(&state).expect("rebuilt cycle state should decode");
+    assert_eq!(leader_epoch, 1);
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_keeps_cleanup_marker_when_preserved_epoch_is_exhausted() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    let primary = CurrentCycle {
+        next: 42,
+        ..Default::default()
+    };
+    save_config(
+        store.clone(),
+        DATA_USAGE_BLOOM_NAME_PATH.as_str(),
+        encode_scanner_cycle_state(&primary, u64::MAX).expect("valid cycle state should encode"),
+    )
+    .await
+    .expect("valid cycle state should be persisted");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), b"{not-json".to_vec())
+        .await
+        .expect("malformed marker should be persisted");
+
+    assert!(
+        reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+            .await
+            .is_err()
+    );
+
+    let marker = read_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str())
+        .await
+        .expect("cleanup marker should remain durable");
+    assert_eq!(
+        serde_json::from_slice::<ScannerCycleRecoveryMarker>(&marker)
+            .expect("cleanup marker should decode")
+            .state,
+        "cleanup-pending"
+    );
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store).await,
+        ScannerCycleStateStartup::Blocked
+    ));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rejects_preserved_epoch_that_would_be_terminal() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    let primary = CurrentCycle {
+        next: 42,
+        ..Default::default()
+    };
+    save_config(
+        store.clone(),
+        DATA_USAGE_BLOOM_NAME_PATH.as_str(),
+        encode_scanner_cycle_state(&primary, u64::MAX - 1).expect("valid cycle state should encode"),
+    )
+    .await
+    .expect("valid cycle state should be persisted");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), b"{not-json".to_vec())
+        .await
+        .expect("malformed marker should be persisted");
+
+    assert!(
+        reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+            .await
+            .is_err(),
+        "reset must not persist the terminal leader epoch"
+    );
+
+    let marker = read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str())
+        .await
+        .expect("cleanup marker should remain durable");
+    assert_eq!(
+        serde_json::from_slice::<ScannerCycleRecoveryMarker>(&marker)
+            .expect("cleanup marker should decode")
+            .state,
+        "cleanup-pending"
+    );
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rejects_usage_floor_that_would_be_terminal() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), vec![0xff, 0x00, 0x01])
+        .await
+        .expect("corrupt cycle state should be persisted");
+    save_config(
+        store.clone(),
+        DATA_USAGE_OBJ_NAME_PATH.as_str(),
+        serde_json::to_vec(&DataUsageInfo {
+            scanner_epoch: Some(u64::MAX - 1),
+            ..Default::default()
+        })
+        .expect("usage floor should encode"),
+    )
+    .await
+    .expect("usage floor should be persisted");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), b"{not-json".to_vec())
+        .await
+        .expect("malformed marker should be persisted");
+
+    assert!(
+        reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+            .await
+            .is_err(),
+        "reset must not persist the terminal leader epoch"
+    );
+    assert_eq!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str())
+            .await
+            .expect("recovery marker should remain durable"),
+        b"{not-json"
+    );
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rebuilds_empty_primary_with_malformed_marker() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    save_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str(), Vec::new())
+        .await
+        .expect("empty cycle state should be persisted");
+    save_config(store.clone(), DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(), b"{not-json".to_vec())
+        .await
+        .expect("malformed marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("explicit full-rescan reset should replace an empty primary");
+
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("rebuilt cycle state should remain durable");
+    let (cycle, leader_epoch) = decode_scanner_cycle_state(&state).expect("rebuilt cycle state should decode");
+    assert_eq!(cycle.next, 0);
+    assert_eq!(leader_epoch, 1);
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn full_rescan_reset_rebuilds_when_primary_cycle_state_is_missing() {
+    let (_temp_dir, store) = setup_scanner_cycle_store().await;
+    let marker = ScannerCycleRecoveryMarker {
+        schema_version: 1,
+        primary_revision: "memory-missing".to_string(),
+        generation: u64::MAX,
+        leader_epoch: u64::MAX,
+        classification: "corrupt".to_string(),
+        first_detected_at_unix_secs: 1,
+        last_attempt_at_unix_secs: 2,
+        retry_count: 0,
+        reason: "missing primary".to_string(),
+        path: DATA_USAGE_BLOOM_NAME_PATH.clone(),
+        quarantine_path: DATA_USAGE_BLOOM_RECOVERY_PATH.clone(),
+        state: "blocked".to_string(),
+    };
+    save_config(
+        store.clone(),
+        DATA_USAGE_BLOOM_RECOVERY_PATH.as_str(),
+        serde_json::to_vec(&marker).expect("marker should encode"),
+    )
+    .await
+    .expect("marker should be persisted");
+
+    reset_scanner_cycle_recovery(CancellationToken::new(), store.clone())
+        .await
+        .expect("full-rescan reset should recreate missing primary");
+    let state = read_config(store.clone(), DATA_USAGE_BLOOM_NAME_PATH.as_str())
+        .await
+        .expect("missing primary should be rebuilt");
+    let (cycle, leader_epoch) = decode_scanner_cycle_state(&state).expect("rebuilt cycle state should decode");
+    assert_eq!(cycle.next, 0);
+    assert_eq!(leader_epoch, 1);
+    assert!(matches!(
+        read_config(store, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()).await,
+        Err(EcstoreError::ConfigNotFound)
+    ));
+}
+
+#[tokio::test]
+async fn corrupt_cycle_state_rename_or_marker_failure_stays_recovery_required() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let state_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_NAME_PATH.as_str());
+    let marker_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str());
+    store.objects.lock().await.insert(state_key.clone(), vec![1]);
+    store.revisions.lock().await.insert(state_key, 9);
+    store.fail_put_number.lock().await.insert(marker_key, 1);
+
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store.clone()).await,
+        ScannerCycleStateStartup::Transient(_)
+    ));
+    let status = scanner_cycle_recovery_status();
+    assert_eq!(status.state, "recovery-required");
+    assert!(status.retryable);
+    assert!(
+        store
+            .objects
+            .lock()
+            .await
+            .contains_key(&memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_NAME_PATH.as_str()))
+    );
+}
+
+#[tokio::test]
+async fn oversized_or_symlinked_cycle_state_is_rejected() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_NAME_PATH.as_str());
+    store.objects.lock().await.insert(key.clone(), vec![0; 1024 * 1024 + 1]);
+    store.revisions.lock().await.insert(key.clone(), 11);
+
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store.clone()).await,
+        ScannerCycleStateStartup::Blocked
+    ));
+    assert_eq!(scanner_cycle_recovery_status().classification.as_deref(), Some("corrupt"));
+    assert!(
+        scanner_cycle_recovery_status()
+            .reason
+            .as_deref()
+            .is_some_and(|reason| reason.contains("oversized"))
+    );
+
+    let marker_key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_BLOOM_RECOVERY_PATH.as_str());
+    store.objects.lock().await.remove(&marker_key);
+    store.objects.lock().await.insert(key.clone(), vec![1]);
+    store.revisions.lock().await.insert(key.clone(), 12);
+    store.non_regular_objects.lock().await.insert(key);
+    // The object contract exposes a non-regular object as `is_dir`; local
+    // backends reject symlink/reparse entries before they become an object.
+    assert!(matches!(
+        load_scanner_cycle_state_for_startup(store).await,
+        ScannerCycleStateStartup::Blocked
+    ));
+}
+
 #[tokio::test]
 async fn scanner_startup_uses_primary_and_backup_usage_floor() {
     let store = Arc::new(MemoryConfigStore::default());
@@ -855,6 +1708,31 @@ async fn scanner_startup_uses_primary_and_backup_usage_floor() {
     assert_eq!(epoch, 11);
 }
 
+#[tokio::test]
+async fn scanner_usage_floor_ignores_older_backup_after_primary_epoch_fence() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let backup_path = format!("{}.bkp", DATA_USAGE_OBJ_NAME_PATH.as_str());
+    for (path, epoch, cycle) in [(DATA_USAGE_OBJ_NAME_PATH.as_str(), 8, 100), (backup_path.as_str(), 7, 10_000)] {
+        store.objects.lock().await.insert(
+            memory_config_key(RUSTFS_META_BUCKET, path),
+            serde_json::to_vec(&DataUsageInfo {
+                scanner_epoch: Some(epoch),
+                scanner_cycle: Some(cycle),
+                ..Default::default()
+            })
+            .expect("usage snapshot should encode"),
+        );
+    }
+
+    assert_eq!(
+        persisted_usage_floor(store).await.expect("usage floor should load"),
+        PersistedUsageFloor {
+            next_cycle: 101,
+            leader_epoch: 8,
+        }
+    );
+}
+
 #[test]
 fn scanner_startup_treats_incomplete_usage_snapshot_as_cold() {
     let mut legacy = complete_usage_with_bucket_count(Some(std::time::SystemTime::now()), 1);
@@ -987,6 +1865,15 @@ async fn scanner_usage_floor_fails_closed_on_corrupt_or_exhausted_usage_state()
 
     assert!(persisted_usage_floor(store.clone()).await.is_err());
 
+    store.objects.lock().await.insert(
+        memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_OBJ_NAME_PATH.as_str()),
+        br#"{}"#.to_vec(),
+    );
+    assert!(
+        persisted_usage_floor(store.clone()).await.is_err(),
+        "a structurally incomplete usage snapshot must not be treated as an empty floor"
+    );
+
     store.objects.lock().await.insert(
         memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_OBJ_NAME_PATH.as_str()),
         serde_json::to_vec(&DataUsageInfo {
@@ -1244,6 +2131,22 @@ async fn test_leadership_claim_preserves_usage_epoch_floor_across_old_epoch_conf
     assert_eq!(store.put_counts.lock().await.get(&key), Some(&3));
 }
 
+#[tokio::test]
+async fn test_leadership_claim_rejects_terminal_epoch() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let ctx = CancellationToken::new();
+    let mut revision = DataUsageCacheRevision::Missing;
+    let mut cycle = CurrentCycle {
+        next: 12,
+        ..Default::default()
+    };
+    let mut persisted_epoch = u64::MAX - 1;
+
+    assert!(!claim_scanner_leadership(&ctx, store.clone(), &mut cycle, &mut revision, &mut persisted_epoch).await);
+    assert_eq!(persisted_epoch, u64::MAX - 1);
+    assert!(read_config(store, &DATA_USAGE_BLOOM_NAME_PATH).await.is_err());
+}
+
 #[tokio::test]
 async fn test_leadership_claim_confirms_commit_after_returned_error() {
     let store = Arc::new(MemoryConfigStore::default());
@@ -2975,6 +3878,24 @@ fn superseded_retry_backoff_grows_from_the_default_cycle() {
     }
 }
 
+#[tokio::test(start_paused = true)]
+async fn corrupt_cycle_state_backoff_uses_virtual_clock() {
+    let mut backoff = ScannerRetryBackoff::default();
+    backoff.record_retryable_cycle(true);
+    let first_delay = backoff
+        .retry_interval(Duration::from_secs(60))
+        .expect("the first recovery retry should be scheduled");
+    assert_eq!(first_delay, Duration::from_secs(5));
+
+    let deadline = Instant::now() + first_delay;
+    assert!(Instant::now() < deadline);
+    tokio::time::advance(first_delay).await;
+    assert!(Instant::now() >= deadline);
+
+    backoff.record_retryable_cycle(true);
+    assert_eq!(backoff.retry_interval(Duration::from_secs(60)), Some(Duration::from_secs(10)));
+}
+
 #[test]
 fn scanner_cycle_wait_plan_drives_growth_resets_and_bitrot_cap() {
     let runtime_config = ScannerRuntimeConfig {

crates/storage-api/src/lib.rs

@@ -76,6 +76,7 @@ pub use bucket::{BucketInfo, BucketOperations, BucketOptions, DeleteBucketOption
 pub use capability::{CapabilitySnapshotError, CapabilityState, CapabilityStatus};
 pub use error::{StorageErrorCode, StorageResult};
 pub use multipart::{CompletePart, ListMultipartsInfo, ListPartsInfo, MultipartInfo, MultipartUploadResult, PartInfo};
+pub use object::DeleteAccounting;
 pub use object::ObjectLockDeleteOptions;
 pub use object::{DeletedObject, ObjectToDelete};
 pub use object::{ExpirationOptions, TransitionedObject};

crates/storage-api/src/object.rs

@@ -218,6 +218,17 @@ pub struct DeletedObject {
     pub force_delete_generation: Option<i64>,
 }
 
+/// Accounting identity returned by the internal commit-time delete path.
+///
+/// This is carried separately from [`DeletedObject`] so adding quota details
+/// does not change the source shape of the public S3 delete result contract.
+#[derive(Debug, Default, Clone, PartialEq, Eq)]
+pub struct DeleteAccounting {
+    pub size: Option<u64>,
+    pub version_id: Option<Uuid>,
+    pub removed_current_object: bool,
+}
+
 impl DeletedObject {
     pub fn version_purge_status(&self) -> VersionPurgeStatusType {
         self.replication_state
@@ -341,6 +352,19 @@ pub trait ObjectOperations: Send + Sync + fmt::Debug {
         objects: Vec<Self::ObjectToDelete>,
         opts: Self::ObjectOptions,
     ) -> (Vec<Self::DeletedObject>, Vec<Option<Self::Error>>);
+    /// Delete objects and optionally return commit-time accounting identities.
+    /// The default preserves the ordinary delete contract for implementations
+    /// that do not expose storage-level accounting details.
+    async fn delete_objects_with_accounting(
+        &self,
+        bucket: &str,
+        objects: Vec<Self::ObjectToDelete>,
+        opts: Self::ObjectOptions,
+    ) -> (Vec<Self::DeletedObject>, Vec<Option<Self::Error>>, Vec<Option<DeleteAccounting>>) {
+        let object_count = objects.len();
+        let (deleted, errors) = self.delete_objects(bucket, objects, opts).await;
+        (deleted, errors, vec![None; object_count])
+    }
     async fn put_object_metadata(
         &self,
         bucket: &str,

rustfs/src/admin/handlers/mod.rs

@@ -126,6 +126,7 @@ mod tests {
         let _list_remote_target_handler = replication::ListRemoteTargetHandler {};
         let _remove_remote_target_handler = replication::RemoveRemoteTargetHandler {};
         let _scanner_status_handler = scanner::ScannerStatusHandler {};
+        let _scanner_cycle_state_reset_handler = scanner::ScannerCycleStateResetHandler {};
         let _ilm_expiry_status_handler = scanner::IlmExpiryStatusHandler {};
         let _manual_transition_handler = ilm_transition::ManualTransitionRunHandler {};
         let _manual_transition_status_handler = ilm_transition::ManualTransitionJobStatusHandler {};

rustfs/src/admin/handlers/scanner.rs

@@ -13,8 +13,11 @@
 // limitations under the License.
 
 use crate::admin::auth::authorize_admin_request;
+use crate::admin::handlers::supervise_admin_mutation;
 use crate::admin::router::{AdminOperation, Operation, S3Router};
-use crate::admin::runtime_sources::current_scanner_metrics_report;
+use crate::admin::runtime_sources::{
+    app_context_from_req, current_object_store_handle_for_context, current_scanner_metrics_report,
+};
 use crate::module_switches::{ENV_SCANNER_ENABLED, scanner_enabled_from_env};
 use crate::server::ADMIN_PREFIX;
 use chrono::Utc;
@@ -22,11 +25,13 @@ use http::{HeaderMap, HeaderValue};
 use hyper::{Method, StatusCode};
 use matchit::Params;
 use rustfs_common::metrics::{ScannerLifecycleExpirySnapshot, ScannerMaintenanceControlSnapshot, ScannerMetricsReport};
+use rustfs_config::MAX_ADMIN_REQUEST_BODY_SIZE;
 use rustfs_credentials::Credentials;
 use rustfs_policy::policy::action::{Action, AdminAction};
 use s3s::header::CONTENT_TYPE;
 use s3s::{Body, S3Error, S3ErrorCode, S3Request, S3Response, S3Result, s3_error};
-use serde::Serialize;
+use serde::{Deserialize, Serialize};
+use tokio_util::sync::CancellationToken;
 
 const JSON_CONTENT_TYPE: &str = "application/json";
 
@@ -38,6 +43,13 @@ struct ScannerStatusResponse {
     metrics: ScannerMetricsReport,
     cycle_schedule: rustfs_scanner::ScannerCycleScheduleStatus,
     runtime_config: rustfs_scanner::runtime_config::ScannerRuntimeConfigStatus,
+    cycle_recovery: rustfs_scanner::ScannerCycleRecoveryStatus,
+}
+
+#[derive(Debug, Deserialize)]
+#[serde(deny_unknown_fields)]
+struct ScannerCycleResetRequest {
+    mode: String,
 }
 
 #[derive(Debug, Serialize)]
@@ -117,6 +129,7 @@ fn scanner_status_response(
         metrics,
         cycle_schedule,
         runtime_config,
+        cycle_recovery: rustfs_scanner::scanner::scanner_cycle_recovery_status(),
     }
 }
 
@@ -144,6 +157,11 @@ pub fn register_scanner_route(r: &mut S3Router<AdminOperation>) -> std::io::Resu
         format!("{ADMIN_PREFIX}/v3/scanner/status").as_str(),
         AdminOperation(&ScannerStatusHandler {}),
     )?;
+    r.insert(
+        Method::POST,
+        format!("{ADMIN_PREFIX}/v3/scanner/cycle-state/reset").as_str(),
+        AdminOperation(&ScannerCycleStateResetHandler {}),
+    )?;
     r.insert(
         Method::GET,
         format!("{ADMIN_PREFIX}/v3/ilm/expiry/status").as_str(),
@@ -163,6 +181,13 @@ async fn validate_scanner_status_request(req: &S3Request<Body>) -> S3Result<Cred
     authorize_admin_request(req, vec![Action::AdminAction(AdminAction::ServerInfoAdminAction)]).await
 }
 
+async fn validate_scanner_reset_request(req: &S3Request<Body>) -> S3Result<Credentials> {
+    if req.credentials.is_none() {
+        return Err(s3_error!(InvalidRequest, "missing credentials"));
+    }
+    authorize_admin_request(req, vec![Action::AdminAction(AdminAction::ConfigUpdateAdminAction)]).await
+}
+
 fn json_response(body: Vec<u8>) -> S3Result<S3Response<(StatusCode, Body)>> {
     let mut headers = HeaderMap::new();
     let content_type = HeaderValue::from_str(JSON_CONTENT_TYPE)
@@ -192,6 +217,37 @@ impl Operation for ScannerStatusHandler {
 
 pub struct IlmExpiryStatusHandler {}
 
+pub struct ScannerCycleStateResetHandler {}
+
+#[async_trait::async_trait]
+impl Operation for ScannerCycleStateResetHandler {
+    async fn call(&self, mut req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
+        let _cred = validate_scanner_reset_request(&req).await?;
+        let body = req
+            .input
+            .store_all_limited(MAX_ADMIN_REQUEST_BODY_SIZE)
+            .await
+            .map_err(|err| S3Error::with_message(S3ErrorCode::InvalidRequest, format!("invalid reset request body: {err}")))?;
+        let reset = serde_json::from_slice::<ScannerCycleResetRequest>(&body)
+            .map_err(|err| S3Error::with_message(S3ErrorCode::InvalidRequest, format!("invalid reset request body: {err}")))?;
+        if reset.mode != "full-rescan" {
+            return Err(S3Error::with_message(S3ErrorCode::InvalidRequest, "reset mode must be full-rescan"));
+        }
+        let context = app_context_from_req(&req)
+            .ok_or_else(|| S3Error::with_message(S3ErrorCode::InternalError, "storage layer not initialized"))?;
+        let store = current_object_store_handle_for_context(Some(context.as_ref()))
+            .ok_or_else(|| S3Error::with_message(S3ErrorCode::InternalError, "storage layer not initialized"))?;
+        supervise_admin_mutation("scanner cycle state reset", async move {
+            rustfs_scanner::scanner::reset_scanner_cycle_recovery(CancellationToken::new(), store)
+                .await
+                .map_err(|err| S3Error::with_message(S3ErrorCode::InternalError, err.to_string()))?;
+            Ok::<_, S3Error>(())
+        })
+        .await?;
+        json_response(br#"{"status":"reset","mode":"full-rescan"}"#.to_vec())
+    }
+}
+
 #[async_trait::async_trait]
 impl Operation for IlmExpiryStatusHandler {
     async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
@@ -237,6 +293,38 @@ mod tests {
         assert_eq!(err.message(), Some("missing credentials"));
     }
 
+    #[tokio::test]
+    async fn scanner_reset_gate_rejects_missing_credentials() {
+        let req = S3Request {
+            input: Body::from(String::new()),
+            method: Method::POST,
+            uri: http::Uri::from_static("/rustfs/admin/v3/scanner/cycle-state/reset"),
+            headers: HeaderMap::new(),
+            extensions: http::Extensions::new(),
+            credentials: None,
+            region: None,
+            service: None,
+            trailing_headers: None,
+        };
+
+        let err = validate_scanner_reset_request(&req)
+            .await
+            .expect_err("a reset request without credentials must be rejected");
+        assert_eq!(err.code(), &S3ErrorCode::InvalidRequest);
+        assert_eq!(err.message(), Some("missing credentials"));
+    }
+
+    #[test]
+    fn admin_reset_requires_full_rescan_or_verified_cursor() {
+        let full_rescan: ScannerCycleResetRequest =
+            serde_json::from_str(r#"{"mode":"full-rescan"}"#).expect("full rescan must be accepted");
+        assert_eq!(full_rescan.mode, "full-rescan");
+        let cursor: ScannerCycleResetRequest =
+            serde_json::from_str(r#"{"mode":"cursor"}"#).expect("mode validation belongs to the handler");
+        assert_ne!(cursor.mode, "full-rescan");
+        assert!(serde_json::from_str::<ScannerCycleResetRequest>(r#"{"mode":"full-rescan","cursor":"untrusted"}"#).is_err());
+    }
+
     #[test]
     fn scanner_disabled_reason_reports_startup_env_key() {
         assert_eq!(scanner_disabled_reason(true), None);
@@ -304,6 +392,11 @@ mod tests {
         assert_eq!(encoded["cycle_schedule"]["effective_interval_seconds"], 0);
         assert_eq!(encoded["cycle_schedule"]["clean_idle_backoff_enabled"], false);
         assert_eq!(encoded["cycle_schedule"]["clean_idle_backoff_multiplier"], 1);
+        assert_eq!(encoded["cycle_recovery"]["state"], "healthy");
+        assert_eq!(
+            encoded["cycle_recovery"]["quarantine_path"],
+            rustfs_scanner::DATA_USAGE_BLOOM_RECOVERY_PATH.as_str()
+        );
     }
 
     #[test]

rustfs/src/admin/route_policy.rs

@@ -428,6 +428,12 @@ pub const ADMIN_ROUTE_POLICY_SPECS: &[AdminRouteSpec] = &[
     admin(HttpMethod::Get, "/rustfs/admin/v3/config", CONFIG_UPDATE, RouteRiskLevel::High),
     admin(HttpMethod::Put, "/rustfs/admin/v3/config", CONFIG_UPDATE, RouteRiskLevel::High),
     admin(HttpMethod::Get, "/rustfs/admin/v3/scanner/status", SERVER_INFO, RouteRiskLevel::Sensitive),
+    admin(
+        HttpMethod::Post,
+        "/rustfs/admin/v3/scanner/cycle-state/reset",
+        CONFIG_UPDATE,
+        RouteRiskLevel::High,
+    ),
     admin(
         HttpMethod::Get,
         "/rustfs/admin/v3/ilm/expiry/status",
@@ -2020,6 +2026,12 @@ mod tests {
         assert_not_action(HttpMethod::Get, "/rustfs/admin/v3/ilm/expiry/status", SET_TIER);
     }
 
+    #[test]
+    fn route_policy_requires_config_update_for_scanner_cycle_reset() {
+        assert_action(HttpMethod::Post, "/rustfs/admin/v3/scanner/cycle-state/reset", CONFIG_UPDATE);
+        assert_not_action(HttpMethod::Post, "/rustfs/admin/v3/scanner/cycle-state/reset", SERVER_INFO);
+    }
+
     #[test]
     fn route_policy_uses_tier_actions_for_transition_routes() {
         assert_action(HttpMethod::Post, "/rustfs/admin/v3/ilm/transition/run", SET_TIER);

rustfs/src/admin/route_registration_test.rs

@@ -243,6 +243,7 @@ fn expected_admin_route_matrix() -> Vec<RouteMatrixEntry> {
         admin_route(Method::GET, "/v3/config"),
         admin_route(Method::PUT, "/v3/config"),
         admin_route(Method::GET, "/v3/scanner/status"),
+        admin_route(Method::POST, "/v3/scanner/cycle-state/reset"),
         admin_route(Method::GET, "/v3/audit/target/list"),
         admin_route_sample(
             Method::PUT,
@@ -879,6 +880,7 @@ fn test_register_routes_cover_representative_admin_paths() {
     assert_route(&router, Method::GET, &admin_path("/v3/config"));
     assert_route(&router, Method::PUT, &admin_path("/v3/config"));
     assert_route(&router, Method::GET, &admin_path("/v3/scanner/status"));
+    assert_route(&router, Method::POST, &admin_path("/v3/scanner/cycle-state/reset"));
     assert_route(&router, Method::GET, &admin_path("/v3/ilm/expiry/status"));
     assert_route(&router, Method::POST, &admin_path("/v3/ilm/transition/run"));
     assert_route(
@@ -1367,6 +1369,7 @@ fn test_admin_alias_paths_match_existing_admin_routes() {
         (Method::GET, compat_admin_alias_path("/v3/config")),
         (Method::PUT, compat_admin_alias_path("/v3/config")),
         (Method::GET, compat_admin_alias_path("/v3/scanner/status")),
+        (Method::POST, compat_admin_alias_path("/v3/scanner/cycle-state/reset")),
         (Method::GET, compat_admin_alias_path("/v3/ilm/expiry/status")),
     ] {
         assert!(

rustfs/src/app/bucket_usecase.rs

@@ -1021,7 +1021,7 @@ fn build_list_objects_v2_metadata_output(
                 object: Object {
                     key: Some(encode_list_objects_v2_value(&object.name, encoding_type)),
                     last_modified: object.mod_time.map(Timestamp::from),
-                    size: Some(object.get_actual_size().unwrap_or_default()),
+                    size: Some(object.get_actual_size_or_physical()),
                     e_tag: object.etag.clone().map(|etag| to_s3s_etag(&etag)),
                     storage_class: Some(ObjectStorageClass::from(
                         object

rustfs/src/app/object_usecase.rs

@@ -3969,6 +3969,55 @@ fn delete_creates_delete_marker(opts: &ObjectOptions) -> bool {
     opts.version_id.is_none() && opts.versioned && !opts.version_suspended
 }
 
+fn delete_removes_current_object(opts: &ObjectOptions) -> bool {
+    delete_request_targets_current(
+        opts.version_id
+            .as_deref()
+            .and_then(|version_id| Uuid::parse_str(version_id).ok()),
+    )
+}
+
+fn delete_request_targets_current(version_id: Option<Uuid>) -> bool {
+    version_id.is_none() || version_id.is_some_and(|version_id| version_id.is_nil())
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum DeleteMemoryUpdate {
+    DeleteMarker,
+    Object { size: u64, removed_current_object: bool },
+}
+
+fn delete_memory_update(
+    creates_delete_marker: bool,
+    committed_delete_marker: bool,
+    requested_current: bool,
+    accounting_size: Option<u64>,
+    removed_current_object: bool,
+) -> Option<DeleteMemoryUpdate> {
+    if creates_delete_marker || (committed_delete_marker && requested_current) {
+        return Some(DeleteMemoryUpdate::DeleteMarker);
+    }
+
+    (!committed_delete_marker)
+        .then_some(accounting_size)
+        .flatten()
+        .map(|size| DeleteMemoryUpdate::Object {
+            size,
+            removed_current_object,
+        })
+}
+
+async fn apply_delete_memory_update(bucket: &str, update: Option<DeleteMemoryUpdate>) {
+    match update {
+        Some(DeleteMemoryUpdate::DeleteMarker) => record_bucket_delete_marker_memory(bucket).await,
+        Some(DeleteMemoryUpdate::Object {
+            size,
+            removed_current_object,
+        }) => record_bucket_object_delete_memory(bucket, size, removed_current_object).await,
+        None => {}
+    }
+}
+
 /// `DeleteObjects` is idempotent. A raw filesystem `NotFound` can cross the
 /// distributed delete path instead of its usual typed missing-object error.
 fn is_delete_objects_not_found(error: &EcstoreError) -> bool {
@@ -8409,8 +8458,6 @@ impl DefaultObjectUsecase {
             object: ObjectToDelete,
             versioned: bool,
             version_suspended: bool,
-            size: i64,
-            existing: Option<ObjectInfo>,
         }
 
         // Phase 2 (bounded concurrency, backlog#929 / HP-8): collect the
@@ -8428,32 +8475,23 @@ impl DefaultObjectUsecase {
                     skip_stat,
                 } = prepared;
                 let synthetic_version_id = object.version_id.is_none() && is_dir_object(&object.object_name);
-                let (goi, source_missing) = if skip_stat {
-                    (ObjectInfo::default(), false)
-                } else {
+                if !skip_stat {
                     match store_ref.get_object_info(bucket_ref, &object.object_name, &opts).await {
-                        Ok(res) => (res, false),
-                        Err(err) if is_err_object_not_found(&err) || is_err_version_not_found(&err) => {
-                            (ObjectInfo::default(), true)
-                        }
+                        Ok(_) => {}
+                        Err(err) if is_err_object_not_found(&err) || is_err_version_not_found(&err) => {}
                         Err(err) => return Err(ApiError::from(err)),
                     }
-                };
-
-                let size = goi.size;
+                }
 
                 if synthetic_version_id {
                     object.version_id = Some(Uuid::nil());
                 }
 
-                let existing = (!skip_stat && !source_missing).then_some(goi);
                 Ok::<_, ApiError>(AdmittedDelete {
                     idx,
                     object,
                     versioned: opts.versioned,
                     version_suspended: opts.version_suspended,
-                    size,
-                    existing,
                 })
             }))
             .buffered(DELETE_OBJECTS_PRE_STAT_CONCURRENCY)
@@ -8464,15 +8502,11 @@ impl DefaultObjectUsecase {
         // per-key success/failure reporting is unchanged.
         let mut object_to_delete = Vec::new();
         let mut object_to_delete_idx = Vec::new();
-        let mut object_sizes = Vec::new();
-        let mut existing_object_infos = Vec::new();
         let mut object_versioning = Vec::new();
         for admitted in admitted_deletes {
-            object_sizes.push(admitted.size);
             object_to_delete_idx.push(admitted.idx);
             object_versioning.push((admitted.versioned, admitted.version_suspended));
             object_to_delete.push(admitted.object);
-            existing_object_infos.push(admitted.existing);
         }
         let cache_adapter = self.object_data_cache();
         let cache_keys_before_delete = object_to_delete
@@ -8489,8 +8523,8 @@ impl DefaultObjectUsecase {
             ..Default::default()
         };
         apply_bucket_generation_guard(&req, &bucket, &mut storage_delete_opts)?;
-        let (dobjs, errs) = store
-            .delete_objects_with_tier_delete_journal(&bucket, object_to_delete.clone(), storage_delete_opts)
+        let (dobjs, errs, accounting) = store
+            .delete_objects_with_tier_delete_journal_and_accounting(&bucket, object_to_delete.clone(), storage_delete_opts)
             .await;
 
         let _manager = get_concurrency_manager();
@@ -8515,17 +8549,16 @@ impl DefaultObjectUsecase {
                     delete_results[didx].delete_object = Some(deleted_object.clone());
                     let (versioned, version_suspended) = object_versioning[i];
                     let creates_delete_marker = object_to_delete[i].version_id.is_none() && versioned && !version_suspended;
-                    if creates_delete_marker {
-                        record_bucket_delete_marker_memory(&bucket).await;
-                    } else {
-                        let size = object_sizes[i].max(0) as u64;
-                        record_bucket_object_delete_memory(
-                            &bucket,
-                            size,
-                            existing_object_infos[i].is_some() && object_to_delete[i].version_id.is_none(),
-                        )
-                        .await;
-                    }
+                    let committed_delete_marker = dobjs[i].delete_marker;
+                    let delete_accounting = accounting.get(i).and_then(Option::as_ref);
+                    let update = delete_memory_update(
+                        creates_delete_marker,
+                        committed_delete_marker,
+                        delete_request_targets_current(object_to_delete[i].version_id),
+                        delete_accounting.and_then(|value| value.size),
+                        delete_accounting.is_some_and(|value| value.removed_current_object),
+                    );
+                    apply_delete_memory_update(&bucket, update).await;
                 }
                 Err(error) => {
                     delete_results[didx].error = Some(error);
@@ -8803,12 +8836,24 @@ impl DefaultObjectUsecase {
             let _ = invalidate_object_data_cache_after_delete_success(&cache_adapter, &bucket, &key).await;
         }
 
-        // Fast in-memory update for immediate quota and admin usage consistency
-        if delete_creates_delete_marker(&opts) {
-            record_bucket_delete_marker_memory(&bucket).await;
+        // Fast in-memory update for immediate quota and admin usage consistency.
+        // Prefix/force deletes and synthetic directory entries do not carry one
+        // committed object identity; leave their cache delta to reconciliation.
+        let update = if force_delete || obj_info.name.is_empty() || synthetic_version_id {
+            None
         } else {
-            record_bucket_object_delete_memory(&bucket, obj_info.size.max(0) as u64, opts.version_id.is_none()).await;
-        }
+            // The storage commit returns this object's metadata while its
+            // generation lock is held. Never fall back to a pre-delete stat:
+            // an overwrite can commit between that stat and this delete.
+            delete_memory_update(
+                delete_creates_delete_marker(&opts),
+                obj_info.delete_marker,
+                opts.version_id.is_none(),
+                quota_object_size(&obj_info).ok(),
+                delete_removes_current_object(&opts),
+            )
+        };
+        apply_delete_memory_update(&bucket, update).await;
 
         if obj_info.name.is_empty() {
             if let Some((operation_id, target_arns, generation)) = force_delete_intent {
@@ -17861,6 +17906,158 @@ mod tests {
         assert!(!can_skip_delete_objects_pre_stat(false, &delete_marker_creating_opts(), false));
     }
 
+    #[test]
+    fn delete_accounting_recognizes_explicit_null_as_current_object() {
+        let opts = ObjectOptions {
+            version_id: Some(Uuid::nil().to_string()),
+            version_suspended: true,
+            ..Default::default()
+        };
+        assert!(delete_removes_current_object(&opts));
+        assert!(delete_request_targets_current(Some(Uuid::nil())));
+        assert!(!delete_request_targets_current(Some(Uuid::new_v4())));
+        assert!(!delete_removes_current_object(&ObjectOptions {
+            version_id: Some(Uuid::new_v4().to_string()),
+            ..Default::default()
+        }));
+    }
+
+    #[test]
+    fn compressed_object_delete_restores_usage_baseline() {
+        let mut metadata = HashMap::new();
+        insert_str(&mut metadata, SUFFIX_COMPRESSION, "klauspost/compress/s2".to_string());
+        let object = ObjectInfo {
+            size: 400,
+            actual_size: 1000,
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+        let accounting_size = quota_object_size(&object).expect("logical compressed size should be canonical");
+
+        assert_eq!(
+            delete_memory_update(false, false, true, Some(accounting_size), true),
+            Some(DeleteMemoryUpdate::Object {
+                size: 1000,
+                removed_current_object: true,
+            })
+        );
+    }
+
+    #[test]
+    fn invalid_accounting_metadata_is_reconciled_without_overflow() {
+        assert_eq!(delete_memory_update(false, false, true, None, true), None);
+        assert_eq!(
+            delete_memory_update(false, true, true, None, true),
+            Some(DeleteMemoryUpdate::DeleteMarker)
+        );
+    }
+
+    #[tokio::test]
+    #[serial_test::serial]
+    async fn compressed_delete_requests_restore_usage_baseline() {
+        use crate::app::storage_api::test::contract::bucket::{BucketOperations as _, DeleteBucketOptions, MakeBucketOptions};
+
+        let store = crate::app::gating_test_env::shared_gating_ecstore().await;
+        if current_app_context().is_none() {
+            crate::app::runtime_sources::install_test_app_context(Arc::clone(&store)).await;
+        }
+        let bucket = format!("compressed-delete-request-{}", Uuid::new_v4().simple());
+        store
+            .make_bucket(&bucket, &MakeBucketOptions::default())
+            .await
+            .expect("create compressed delete request bucket");
+
+        // Seed the process-local usage with the canonical logical bytes. The
+        // direct storage PUT below intentionally does not apply an app-layer
+        // usage delta; the two real DELETE requests must remove exactly this
+        // amount through their request-layer wiring.
+        crate::app::storage_api::test::data_usage::seed_bucket_usage_memory_for_test(&bucket, 2_000).await;
+
+        for object in ["single", "batch"] {
+            let mut metadata = HashMap::new();
+            insert_str(&mut metadata, SUFFIX_COMPRESSION, "klauspost/compress/s2".to_string());
+            insert_str(&mut metadata, SUFFIX_ACTUAL_SIZE, "1000".to_string());
+            let reader = HashReader::from_stream(std::io::Cursor::new(vec![0x5a; 400]), 400, 1000, None, None, false)
+                .expect("compressed fixture reader should be valid");
+            let mut reader = PutObjReader::new(reader);
+            store
+                .put_object(
+                    &bucket,
+                    object,
+                    &mut reader,
+                    &ObjectOptions {
+                        user_defined: metadata,
+                        ..Default::default()
+                    },
+                )
+                .await
+                .expect("compressed fixture object should be written");
+        }
+
+        let mut single_req = build_request(
+            DeleteObjectInput::builder()
+                .bucket(bucket.clone())
+                .key("single".to_string())
+                .build()
+                .expect("single delete input should build"),
+            Method::DELETE,
+        );
+        single_req.extensions.insert(crate::storage::access::ReqInfo {
+            cred: Some(rustfs_credentials::Credentials::default()),
+            is_owner: true,
+            ..Default::default()
+        });
+        DefaultObjectUsecase::from_global()
+            .execute_delete_object(single_req)
+            .await
+            .expect("single compressed delete should succeed");
+        assert_eq!(
+            crate::app::storage_api::test::data_usage::get_bucket_usage_memory(&bucket).await,
+            Some(1_000),
+            "single delete must subtract the logical accounting size"
+        );
+
+        let mut batch_req = build_request(
+            DeleteObjectsInput::builder()
+                .bucket(bucket.clone())
+                .delete(Delete {
+                    objects: vec![ObjectIdentifier {
+                        key: "batch".to_string(),
+                        ..Default::default()
+                    }],
+                    quiet: None,
+                })
+                .build()
+                .expect("batch delete input should build"),
+            Method::POST,
+        );
+        batch_req.extensions.insert(crate::storage::access::ReqInfo {
+            cred: Some(rustfs_credentials::Credentials::default()),
+            is_owner: true,
+            ..Default::default()
+        });
+        DefaultObjectUsecase::from_global()
+            .execute_delete_objects(batch_req)
+            .await
+            .expect("batch compressed delete should succeed");
+        assert_eq!(
+            crate::app::storage_api::test::data_usage::get_bucket_usage_memory(&bucket).await,
+            Some(0),
+            "batch delete must subtract the committed logical accounting size"
+        );
+
+        store
+            .delete_bucket(
+                &bucket,
+                &DeleteBucketOptions {
+                    force: true,
+                    ..Default::default()
+                },
+            )
+            .await
+            .expect("clean up compressed delete request bucket");
+    }
+
     #[tokio::test]
     async fn execute_get_object_attributes_returns_internal_error_when_store_uninitialized() {
         let input = GetObjectAttributesInput::builder()

rustfs/src/app/storage_api.rs

@@ -72,6 +72,11 @@ pub(crate) mod data_usage {
         compute_bucket_usage, live_bucket_usage_computations, seed_bucket_usage_memory_for_test, store_data_usage_in_backend,
     };
 
+    #[cfg(test)]
+    pub(crate) async fn get_bucket_usage_memory(bucket: &str) -> Option<u64> {
+        crate::storage::storage_api::ecstore_data_usage::get_bucket_usage_memory(bucket).await
+    }
+
     pub(crate) async fn record_bucket_object_delete_memory(bucket: &str, deleted_size: u64, removed_current_object: bool) {
         crate::storage::storage_api::ecstore_data_usage::record_bucket_object_delete_memory(
             bucket,
@@ -1233,7 +1238,10 @@ pub(crate) mod test {
 
     pub(crate) use super::access::ReqInfo;
     pub(crate) use super::options::VERSIONING_CONFIG_LOOKUPS;
-    pub(crate) use super::{bucket, data_usage, ecfs, object_utils, runtime};
+    pub(crate) use super::{bucket, ecfs, object_utils, runtime};
+    pub(crate) mod data_usage {
+        pub(crate) use super::super::data_usage::*;
+    }
     pub(crate) use crate::storage::storage_api::test_consumer::{get_global_bucket_metadata_sys, set_bucket_metadata};
     pub(crate) use crate::storage::storage_api::{
         ECStore, Endpoint, Endpoints, PoolEndpoints, StorageObjectInfo, StorageObjectOptions, StoragePutObjReader,

rustfs/src/storage/rpc/node_service/disk.rs

@@ -146,6 +146,29 @@ fn encode_file_info_msgpack(value: &FileInfo) -> std::result::Result<Vec<u8>, Di
     encode_msgpack_with_capacity(value, "FileInfo", FILE_INFO_MSGPACK_ENCODE_CAPACITY_HINT)
 }
 
+fn encode_delete_versions_errors(disk_errors: Vec<Option<DiskError>>) -> (Vec<String>, Vec<Error>) {
+    let mut errors = Vec::with_capacity(disk_errors.len());
+    let mut item_errors = Vec::with_capacity(disk_errors.len());
+    for error in disk_errors {
+        match error {
+            Some(error) => {
+                let code = match &error {
+                    DiskError::Io(source) if source.kind() == std::io::ErrorKind::NotFound => DiskError::FileNotFound.to_u32(),
+                    _ => error.to_u32(),
+                };
+                let error_info = error.to_string();
+                errors.push(error_info.clone());
+                item_errors.push(Error { code, error_info });
+            }
+            None => {
+                errors.push(String::new());
+                item_errors.push(Error::default());
+            }
+        }
+    }
+    (errors, item_errors)
+}
+
 fn encode_msgpack_named<T: serde::Serialize>(value: &T, value_name: &str) -> std::result::Result<Vec<u8>, DiskError> {
     let mut serializer = rmp_serde::Serializer::new(Vec::with_capacity(MSGPACK_ENCODE_CAPACITY_HINT)).with_struct_map();
     value
@@ -552,6 +575,7 @@ impl NodeService {
                             success: false,
                             errors: Vec::new(),
                             error: Some(DiskError::other(format!("decode FileInfoVersions failed: {err}")).into()),
+                            item_errors: Vec::new(),
                         }));
                     }
                 };
@@ -563,30 +587,26 @@ impl NodeService {
                         success: false,
                         errors: Vec::new(),
                         error: Some(DiskError::other(format!("decode DeleteOptions failed: {err}")).into()),
+                        item_errors: Vec::new(),
                     }));
                 }
             };
 
-            let errors = disk
-                .delete_versions(&request.volume, versions, opts)
-                .await
-                .into_iter()
-                .map(|error| match error {
-                    Some(e) => e.to_string(),
-                    None => "".to_string(),
-                })
-                .collect();
+            let (errors, item_errors) =
+                encode_delete_versions_errors(disk.delete_versions(&request.volume, versions, opts).await);
 
             Ok(Response::new(DeleteVersionsResponse {
                 success: true,
                 errors,
                 error: None,
+                item_errors,
             }))
         } else {
             Ok(Response::new(DeleteVersionsResponse {
                 success: false,
                 errors: Vec::new(),
                 error: Some(DiskError::other("cannot find disk".to_string()).into()),
+                item_errors: Vec::new(),
             }))
         }
     }
@@ -1612,8 +1632,8 @@ impl NodeService {
 mod tests {
     use super::{
         compat_response_json, decode_msgpack_or_json, decode_rename_data_request_file_info,
-        encode_batch_read_version_response_payloads, encode_file_info_msgpack, encode_msgpack, encode_msgpack_named,
-        encode_read_multiple_response_payloads, encode_rename_data_response_payloads,
+        encode_batch_read_version_response_payloads, encode_delete_versions_errors, encode_file_info_msgpack, encode_msgpack,
+        encode_msgpack_named, encode_read_multiple_response_payloads, encode_rename_data_response_payloads,
     };
     use crate::storage::rpc::node_service::make_server;
     use crate::storage::storage_api::ReadMultipleResp;
@@ -1632,6 +1652,18 @@ mod tests {
         count: u32,
     }
 
+    #[test]
+    fn delete_versions_response_dual_writes_typed_item_errors() {
+        let raw_not_found = super::DiskError::Io(std::io::Error::from(std::io::ErrorKind::NotFound));
+        let (errors, item_errors) = encode_delete_versions_errors(vec![Some(raw_not_found), None]);
+
+        assert!(errors[0].starts_with("io error "));
+        assert!(errors[1].is_empty());
+        assert_eq!(item_errors[0].code, super::DiskError::FileNotFound.to_u32());
+        assert_eq!(item_errors[0].error_info, errors[0]);
+        assert_eq!(item_errors[1].code, 0);
+    }
+
     #[tokio::test]
     #[serial]
     async fn handle_read_version_records_attribution_for_missing_disk() {

rustfs/src/storage/s3_api/bucket.rs

@@ -296,7 +296,10 @@ pub(crate) fn build_list_objects_v2_output(
             let mut obj = Object {
                 key: Some(key),
                 last_modified: v.mod_time.map(Timestamp::from),
-                size: Some(v.get_actual_size().unwrap_or_default()),
+                // Compressed legacy objects may retain an unknown (-1)
+                // logical-size sentinel; never expose that internal value in
+                // an S3 response.
+                size: Some(v.get_actual_size_or_physical()),
                 e_tag: v.etag.clone().map(|etag| to_s3s_etag(&etag)),
                 storage_class: v.storage_class.clone().map(ObjectStorageClass::from),
                 ..Default::default()
@@ -656,6 +659,45 @@ mod tests {
         assert_eq!(output.common_prefixes.as_ref().map(std::vec::Vec::len), Some(2));
     }
 
+    #[test]
+    fn list_objects_never_exposes_compressed_unknown_size_sentinel() {
+        let mut metadata = std::collections::HashMap::new();
+        rustfs_utils::http::insert_str(
+            &mut metadata,
+            rustfs_utils::http::SUFFIX_COMPRESSION,
+            "klauspost/compress/s2".to_string(),
+        );
+        let output = build_list_objects_v2_output(
+            ListObjectsV2Info {
+                objects: vec![ObjectInfo {
+                    name: "legacy-compressed".to_string(),
+                    size: 128,
+                    actual_size: -1,
+                    user_defined: std::sync::Arc::new(metadata),
+                    ..Default::default()
+                }],
+                ..Default::default()
+            },
+            false,
+            1000,
+            "bucket".to_string(),
+            String::new(),
+            None,
+            None,
+            None,
+            None,
+        );
+
+        assert_eq!(
+            output
+                .contents
+                .as_ref()
+                .and_then(|objects| objects.first())
+                .and_then(|object| object.size),
+            Some(128)
+        );
+    }
+
     #[test]
     fn list_responses_report_standard_for_legacy_label_only_file_metadata() {
         let version_id = Uuid::parse_str("11111111-2222-3333-4444-555555555555").expect("fixture version ID should be valid");

rustfs/src/storage/storage_api.rs

@@ -429,6 +429,8 @@ pub(crate) mod ecstore_config {
 }
 
 pub(crate) mod ecstore_data_usage {
+    #[cfg(test)]
+    pub(crate) use rustfs_ecstore::api::data_usage::get_bucket_usage_memory;
     pub(crate) use rustfs_ecstore::api::data_usage::{
         apply_bucket_usage_memory_overlay, init_compression_total_memory_from_backend, load_admin_data_usage_from_backend_cached,
         load_data_usage_from_backend, quota_object_size, record_bucket_delete_marker_memory, record_bucket_object_delete_memory,