Merge remote-tracking branch 'origin/main' into cxymds/fix-rebalance-multipart-retry

# Conflicts:
#	crates/ecstore/src/set_disk/ops/object.rs
#	crates/ecstore/src/set_disk/read.rs

crates/ecstore/Cargo.toml

@@ -32,6 +32,11 @@ workspace = true
 
 [features]
 default = []
+# Compiles the controlled list-objects namespace-journal chaos injector into a
+# production binary (it is always available to tests). Off by default so the
+# RUSTFS_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_* env vars cannot rewrite journal
+# state in a stock build (backlog#1832).
+list-chaos = []
 rio-v2 = ["dep:rustfs-rio-v2"]
 hotpath = [
     "hotpath/hotpath",

crates/ecstore/benches/single_block_non_inline_benchmark.rs

@@ -69,6 +69,7 @@ fn build_non_inline_writers(config: &BenchConfig) -> Vec<Option<BitrotWriterWrap
 fn bench_single_block_non_inline_fast_path(c: &mut Criterion) {
     let configs = vec![
         BenchConfig::new(4 * 1024, 4, 2, 128 * 1024),
+        BenchConfig::new(16 * 1024, 4, 2, 128 * 1024),
         BenchConfig::new(64 * 1024, 4, 2, 128 * 1024),
         BenchConfig::new(128 * 1024, 4, 2, 128 * 1024),
     ];
@@ -112,7 +113,12 @@ fn bench_single_block_non_inline_fast_path(c: &mut Criterion) {
                 rt.block_on(async {
                     erasure
                         .clone()
-                        .encode_single_block_non_inline(reader, &mut writers, config.data_shards)
+                        .encode_single_block_non_inline_with_size_hint(
+                            reader,
+                            &mut writers,
+                            config.data_shards,
+                            config.payload_size,
+                        )
                         .await
                         .expect("single block candidate benchmark");
                 });

crates/ecstore/src/api/mod.rs

@@ -61,9 +61,11 @@ pub mod bucket {
                 delete_manual_transition_scope_admission_if_current, load_manual_transition_job_record,
                 load_manual_transition_job_record_with_etag, load_manual_transition_scope_admission,
                 manual_transition_job_lease_expired, manual_transition_scope_admission_lease_expired,
-                manual_transition_scope_key, persist_manual_transition_job_progress, renew_manual_transition_job_lease,
-                request_manual_transition_job_cancel, save_manual_transition_job_record,
-                save_manual_transition_job_record_if_current, save_manual_transition_scope_admission_if_absent,
+                manual_transition_scope_key, persist_manual_transition_job_progress,
+                persist_manual_transition_job_progress_if_owned, renew_manual_transition_job_lease,
+                renew_manual_transition_job_lease_if_owned, request_manual_transition_job_cancel,
+                save_manual_transition_job_record, save_manual_transition_job_record_if_current,
+                save_manual_transition_scope_admission_if_absent, update_manual_transition_job_record,
             };
         }
 
@@ -344,7 +346,7 @@ pub mod disk {
     }
 
     pub mod error {
-        pub use crate::disk::error::{BitrotErrorType, DiskError, Error, FileAccessDeniedWithContext, Result};
+        pub use crate::disk::error::{DiskError, Error, FileAccessDeniedWithContext, Result};
     }
 
     pub mod error_reduce {

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

@@ -27,9 +27,10 @@ use crate::bucket::lifecycle::manual_transition_job::{
     ManualTransitionWorkerResult, claim_manual_transition_scope_admission, delete_manual_transition_scope_admission_if_current,
     load_manual_transition_job_record, load_manual_transition_job_record_with_etag, load_manual_transition_pending_task_records,
     manual_transition_job_id_from_record_object_name, manual_transition_job_lease_expired,
-    manual_transition_worker_result_task_key, persist_manual_transition_job_progress, reconcile_manual_transition_worker_results,
-    record_manual_transition_worker_result, record_manual_transition_worker_result_with_reason,
-    renew_manual_transition_job_lease, save_manual_transition_job_record_if_current, save_manual_transition_task_if_absent,
+    manual_transition_worker_result_task_key, persist_manual_transition_job_progress_if_owned,
+    reconcile_manual_transition_worker_results_if_owned, record_manual_transition_worker_result,
+    record_manual_transition_worker_result_with_reason, renew_manual_transition_job_lease_if_owned,
+    save_manual_transition_job_record_if_current, save_manual_transition_task_if_absent, update_manual_transition_job_record,
 };
 use crate::bucket::lifecycle::replication_sink;
 use crate::bucket::lifecycle::replication_sink::{
@@ -78,8 +79,8 @@ use rustfs_common::metrics::{
 };
 use rustfs_config::{
     DEFAULT_TRANSITION_QUEUE_CAPACITY, DEFAULT_TRANSITION_QUEUE_SEND_TIMEOUT_MS, DEFAULT_TRANSITION_WORKERS_ABSOLUTE_MAX,
-    DEFAULT_TRANSITION_WORKERS_CAP, ENV_TRANSITION_QUEUE_CAPACITY, ENV_TRANSITION_QUEUE_SEND_TIMEOUT_MS, ENV_TRANSITION_WORKERS,
-    ENV_TRANSITION_WORKERS_ABSOLUTE_MAX,
+    DEFAULT_TRANSITION_WORKERS_CAP, ENV_MAX_EXPIRY_WORKERS, ENV_TRANSITION_QUEUE_CAPACITY, ENV_TRANSITION_QUEUE_SEND_TIMEOUT_MS,
+    ENV_TRANSITION_WORKERS, ENV_TRANSITION_WORKERS_ABSOLUTE_MAX,
 };
 use rustfs_data_usage::TierStats;
 use rustfs_filemeta::{
@@ -2016,18 +2017,25 @@ fn is_slow_down(err: &Error) -> bool {
     matches!(err, Error::SlowDown)
 }
 
-pub async fn init_background_expiry(api: Arc<ECStore>) {
-    let mut workers = get_env_usize("RUSTFS_MAX_EXPIRY_WORKERS", std::cmp::min(num_cpus::get(), 16));
-    //globalILMConfig.getExpirationWorkers()
-    if let Ok(env_expiration_workers) = env::var("_RUSTFS_ILM_EXPIRATION_WORKERS")
-        && let Ok(num_expirations) = env_expiration_workers.parse::<usize>()
-    {
-        workers = num_expirations;
+/// Resolves the expiry worker count from the single documented knob,
+/// `RUSTFS_MAX_EXPIRY_WORKERS`: a set, parsable, non-zero value wins;
+/// anything else falls back to `min(cpus, 16)`. The historical
+/// `_RUSTFS_ILM_EXPIRATION_WORKERS` silent override and the
+/// `RUSTFS_DEFAULT_EXPIRY_WORKERS` zero-fallback were undocumented, unset in
+/// every known deployment, and are removed (backlog#1832).
+fn expiry_worker_count() -> usize {
+    let default = std::cmp::min(num_cpus::get(), 16);
+    match env::var(ENV_MAX_EXPIRY_WORKERS) {
+        Ok(value) => match value.parse::<usize>() {
+            Ok(workers) if workers > 0 => workers,
+            _ => default,
+        },
+        Err(_) => default,
     }
+}
 
-    if workers == 0 {
-        workers = get_env_usize("RUSTFS_DEFAULT_EXPIRY_WORKERS", 8);
-    }
+pub async fn init_background_expiry(api: Arc<ECStore>) {
+    let workers = expiry_worker_count();
 
     ExpiryState::resize_workers(workers, api.clone()).await;
     let _ = spawn_tier_free_version_recovery_once(api.clone(), &TIER_FREE_VERSION_RECOVERY_STARTED);
@@ -2212,7 +2220,18 @@ async fn recover_manual_transition_job(
 
     let recovery_unknown_snapshot = ManualTransitionQueueSnapshot::default();
     if record.scan_completed {
-        let reconciled = reconcile_manual_transition_worker_results(api.clone(), job_id, recovery_unknown_snapshot).await?;
+        let reconciled = match reconcile_manual_transition_worker_results_if_owned(
+            api.clone(),
+            job_id,
+            record.lease_id,
+            recovery_unknown_snapshot,
+        )
+        .await
+        {
+            Ok(record) => record,
+            Err(Error::PreconditionFailed) => return Ok(ManualTransitionJobRecoveryOutcome::Skipped),
+            Err(err) => return Err(err),
+        };
         if reconciled.is_terminal() {
             release_manual_transition_recovery_admission(api, &reconciled).await;
             return match reconciled.state {
@@ -2265,34 +2284,41 @@ async fn recover_manual_transition_job(
         replay,
         ManualTransitionPendingTaskReplay::Queued | ManualTransitionPendingTaskReplay::Deferred
     ) {
-        spawn_manual_transition_recovery_heartbeat(api, job_id);
+        spawn_manual_transition_recovery_heartbeat(api, job_id, recovery_lease_id);
         return Ok(ManualTransitionJobRecoveryOutcome::Resumed);
     }
 
-    let (mut record, etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
-    if record.mark_unknown_if_worker_results_lost(recovery_unknown_snapshot)
-        || record.mark_unknown_if_recovery_would_skip_pending_page(recovery_unknown_snapshot)
+    let mut marked_unknown = false;
+    let record = match update_manual_transition_job_record(api.clone(), job_id, Some(recovery_lease_id), |record| {
+        marked_unknown = record.mark_unknown_if_worker_results_lost(recovery_unknown_snapshot)
+            || record.mark_unknown_if_recovery_would_skip_pending_page(recovery_unknown_snapshot);
+        marked_unknown
+    })
+    .await
     {
-        return match save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await {
-            Ok(()) => {
-                release_manual_transition_recovery_admission(api, &record).await;
-                Ok(ManualTransitionJobRecoveryOutcome::Unknown)
-            }
-            Err(Error::PreconditionFailed) => Ok(ManualTransitionJobRecoveryOutcome::Skipped),
-            Err(err) => Err(err),
-        };
+        Ok(record) => record,
+        Err(Error::PreconditionFailed) => return Ok(ManualTransitionJobRecoveryOutcome::Skipped),
+        Err(err) => return Err(err),
+    };
+    if marked_unknown {
+        release_manual_transition_recovery_admission(api, &record).await;
+        return Ok(ManualTransitionJobRecoveryOutcome::Unknown);
     }
 
     let mut options = record.resume_options();
     options.job_id = Some(job_id);
     options.cancel_check = Some(manual_transition_recovery_cancel_check(api.clone(), job_id));
-    options.progress_sink = Some(manual_transition_recovery_progress_sink(api.clone(), job_id));
+    options.progress_sink = Some(manual_transition_recovery_progress_sink(api.clone(), job_id, recovery_lease_id));
     let result = enqueue_transition_for_existing_objects_scoped(api.clone(), &record.bucket, options).await;
-    let final_record = finalize_recovered_manual_transition_job(api.clone(), job_id, result).await?;
+    let final_record = match finalize_recovered_manual_transition_job(api.clone(), job_id, recovery_lease_id, result).await {
+        Ok(record) => record,
+        Err(Error::PreconditionFailed) => return Ok(ManualTransitionJobRecoveryOutcome::Skipped),
+        Err(err) => return Err(err),
+    };
     if final_record.is_terminal() {
         release_manual_transition_recovery_admission(api, &final_record).await;
     } else {
-        spawn_manual_transition_recovery_heartbeat(api, job_id);
+        spawn_manual_transition_recovery_heartbeat(api, job_id, recovery_lease_id);
     }
     Ok(ManualTransitionJobRecoveryOutcome::Resumed)
 }
@@ -2376,11 +2402,11 @@ fn manual_transition_recovery_cancel_check(api: Arc<ECStore>, job_id: Uuid) -> M
     })
 }
 
-fn manual_transition_recovery_progress_sink(api: Arc<ECStore>, job_id: Uuid) -> ManualTransitionProgressSink {
+fn manual_transition_recovery_progress_sink(api: Arc<ECStore>, job_id: Uuid, lease_id: Uuid) -> ManualTransitionProgressSink {
     Arc::new(move |report| {
         let api = api.clone();
         Box::pin(async move {
-            persist_manual_transition_job_progress(api, job_id, &report, manual_transition_queue_snapshot())
+            persist_manual_transition_job_progress_if_owned(api, job_id, lease_id, &report, manual_transition_queue_snapshot())
                 .await
                 .map(|_| ())
         })
@@ -2390,24 +2416,20 @@ fn manual_transition_recovery_progress_sink(api: Arc<ECStore>, job_id: Uuid) ->
 async fn finalize_recovered_manual_transition_job(
     api: Arc<ECStore>,
     job_id: Uuid,
+    expected_lease_id: Uuid,
     result: Result<ManualTransitionRunReport, Error>,
 ) -> Result<ManualTransitionJobRecord, Error> {
-    for _ in 0..4 {
-        let (mut record, etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
+    update_manual_transition_job_record(api, job_id, Some(expected_lease_id), |record| {
         if record.is_terminal() {
-            return Ok(record);
+            return false;
         }
         match &result {
             Ok(report) => record.complete(report.clone(), manual_transition_queue_snapshot()),
             Err(err) => record.fail(format!("manual transition recovery failed: {err}")),
         }
-        match save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await {
-            Ok(()) => return Ok(record),
-            Err(Error::PreconditionFailed) => continue,
-            Err(err) => return Err(err),
-        }
-    }
-    Err(Error::PreconditionFailed)
+        true
+    })
+    .await
 }
 
 async fn release_manual_transition_recovery_admission(api: Arc<ECStore>, record: &ManualTransitionJobRecord) {
@@ -2426,18 +2448,20 @@ async fn release_manual_transition_recovery_admission(api: Arc<ECStore>, record:
     }
 }
 
-fn spawn_manual_transition_recovery_heartbeat(api: Arc<ECStore>, job_id: Uuid) {
+fn spawn_manual_transition_recovery_heartbeat(api: Arc<ECStore>, job_id: Uuid, lease_id: Uuid) {
     tokio::spawn(async move {
         let mut interval = tokio::time::interval(std::time::Duration::from_secs(5));
         loop {
             interval.tick().await;
-            match renew_manual_transition_job_lease(api.clone(), job_id, manual_transition_queue_snapshot()).await {
+            match renew_manual_transition_job_lease_if_owned(api.clone(), job_id, lease_id, manual_transition_queue_snapshot())
+                .await
+            {
                 Ok(record) if record.is_terminal() => {
                     release_manual_transition_recovery_admission(api, &record).await;
                     return;
                 }
                 Ok(_) => {}
-                Err(Error::ConfigNotFound) => return,
+                Err(Error::ConfigNotFound | Error::PreconditionFailed) => return,
                 Err(err) => {
                     warn!(
                         event = EVENT_LIFECYCLE_WORKER_STATE,
@@ -2455,23 +2479,18 @@ fn spawn_manual_transition_recovery_heartbeat(api: Arc<ECStore>, job_id: Uuid) {
 }
 
 async fn abandon_manual_transition_recovery_lease(api: Arc<ECStore>, job_id: Uuid, lease_id: Uuid) -> Result<(), Error> {
-    for _ in 0..4 {
-        let (mut record, etag) = match load_manual_transition_job_record_with_etag(api.clone(), job_id).await {
-            Ok(record) => record,
-            Err(Error::ConfigNotFound) => return Ok(()),
-            Err(err) => return Err(err),
-        };
-        if record.lease_id != lease_id || record.is_terminal() {
-            return Ok(());
+    match update_manual_transition_job_record(api, job_id, Some(lease_id), |record| {
+        if record.is_terminal() {
+            return false;
         }
         record.abandon_recovery_lease(lease_id);
-        match save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await {
-            Ok(()) => return Ok(()),
-            Err(Error::PreconditionFailed) => continue,
-            Err(err) => return Err(err),
-        }
+        true
+    })
+    .await
+    {
+        Ok(_) | Err(Error::ConfigNotFound | Error::PreconditionFailed) => Ok(()),
+        Err(err) => Err(err),
     }
-    Ok(())
 }
 
 fn tier_free_version_recovery_enabled() -> bool {
@@ -5075,6 +5094,7 @@ pub async fn apply_lifecycle_action(event: &lifecycle::Event, src: &LcEventSrc,
 
 #[cfg(test)]
 mod tests {
+    use super::expiry_worker_count;
     use super::{
         DATE_EXPIRY_EXISTING_OBJECTS_GRACE_SECS, DEFAULT_TRANSITION_QUEUE_CAPACITY, DEFAULT_TRANSITION_WORKERS_ABSOLUTE_MAX,
         DEFAULT_TRANSITION_WORKERS_CAP, EVENT_LIFECYCLE_EVALUATION_FAILED, EVENT_LIFECYCLE_EXPIRED_DETECTED,
@@ -5090,12 +5110,13 @@ mod tests {
         lifecycle_rule_has_date_expiration, manual_transition_duration_elapsed, manual_transition_has_more_after_limit,
         manual_transition_recovery_progress_sink, manual_transition_version_marker, manual_transition_worker_failure_reason,
         mark_delete_opts_skip_decommissioned_on_remote_success, merge_stale_multipart_candidate,
-        persist_manual_transition_job_progress, persist_manual_transition_page_checkpoint, recover_manual_transition_job,
-        recover_manual_transition_jobs, resolve_tier_free_version_recovery_enabled, resolve_transition_queue_capacity,
-        resolve_transition_queue_send_timeout, resolve_transition_worker_count, resolve_transition_workers_absolute_max,
-        run_tier_free_version_recovery_loop, select_restore_s3_location, set_lifecycle_observability_observer,
-        set_recovered_free_version_enqueue_observer, should_defer_date_expiry_for_recent_config_update,
-        transitioned_cleanup_tuple, transitioned_object_delete_opts, wait_for_tier_free_version_recovery,
+        persist_manual_transition_job_progress_if_owned, persist_manual_transition_page_checkpoint,
+        recover_manual_transition_job, recover_manual_transition_jobs, resolve_tier_free_version_recovery_enabled,
+        resolve_transition_queue_capacity, resolve_transition_queue_send_timeout, resolve_transition_worker_count,
+        resolve_transition_workers_absolute_max, run_tier_free_version_recovery_loop, select_restore_s3_location,
+        set_lifecycle_observability_observer, set_recovered_free_version_enqueue_observer,
+        should_defer_date_expiry_for_recent_config_update, transitioned_cleanup_tuple, transitioned_object_delete_opts,
+        wait_for_tier_free_version_recovery,
     };
     #[cfg(feature = "test-util")]
     use super::{delete_free_version_remote_object_then, encode_dir_object, get_transitioned_object_reader_with_tier_manager};
@@ -5105,18 +5126,19 @@ mod tests {
     };
     use crate::bucket::lifecycle::config_boundary;
     use crate::bucket::lifecycle::manual_transition_job::{
-        ManualTransitionJobRecord, ManualTransitionJobState, ManualTransitionScopeAdmission, ManualTransitionScopeAdmissionClaim,
-        ManualTransitionTaskRecord, ManualTransitionWorkerFailureReason, ManualTransitionWorkerResult,
-        ManualTransitionWorkerResultRecord, claim_manual_transition_scope_admission,
+        ManualTransitionJobCasBarrier, ManualTransitionJobRecord, ManualTransitionJobState, ManualTransitionScopeAdmission,
+        ManualTransitionScopeAdmissionClaim, ManualTransitionTaskRecord, ManualTransitionWorkerFailureReason,
+        ManualTransitionWorkerResult, ManualTransitionWorkerResultRecord, claim_manual_transition_scope_admission,
         delete_manual_transition_scope_admission_if_current, legacy_manual_transition_scope_key,
-        load_manual_transition_job_record, load_manual_transition_scope_admission,
+        load_manual_transition_job_record, load_manual_transition_job_record_with_etag, load_manual_transition_scope_admission,
         load_manual_transition_scope_admission_with_etag, load_manual_transition_task_record,
         manual_transition_scope_record_object_name, manual_transition_worker_result_object_name,
         manual_transition_worker_result_task_key, reconcile_manual_transition_worker_results,
         record_manual_transition_worker_result, record_manual_transition_worker_result_with_reason,
-        renew_manual_transition_job_lease, request_manual_transition_job_cancel, save_manual_transition_job_record,
-        save_manual_transition_scope_admission_if_absent, save_manual_transition_scope_admission_if_current,
-        save_manual_transition_task_if_absent, save_manual_transition_worker_result_if_absent,
+        renew_manual_transition_job_lease_if_owned, request_manual_transition_job_cancel, save_manual_transition_job_record,
+        save_manual_transition_job_record_if_current, save_manual_transition_scope_admission_if_absent,
+        save_manual_transition_scope_admission_if_current, save_manual_transition_task_if_absent,
+        save_manual_transition_worker_result_if_absent,
     };
     use crate::bucket::lifecycle::replication_sink::{ReplicationStatusType, VersionPurgeStatusType};
     use crate::bucket::lifecycle::runtime_boundary as runtime_sources;
@@ -5156,6 +5178,7 @@ mod tests {
     #[cfg(feature = "test-util")]
     use http::HeaderMap;
     use rustfs_common::metrics::{IlmAction, global_metrics};
+    use rustfs_config::ENV_MAX_EXPIRY_WORKERS;
     use rustfs_config::ENV_TRANSITION_WORKERS_ABSOLUTE_MAX;
     use rustfs_data_usage::TierStats;
     use rustfs_filemeta::{FileInfo, FileMeta};
@@ -7154,6 +7177,63 @@ mod tests {
         }
     }
 
+    // SAFETY: same contract as with_transition_worker_env — only used from
+    // `#[serial]` tests, so no concurrent reader/writer can access the process
+    // environment while `env::set_var`/`env::remove_var` is active.
+    #[allow(unsafe_code)]
+    fn with_expiry_worker_env<F>(value: Option<&str>, test_fn: F)
+    where
+        F: FnOnce(),
+    {
+        let original = env::var_os(ENV_MAX_EXPIRY_WORKERS);
+
+        match value {
+            Some(v) => unsafe {
+                env::set_var(ENV_MAX_EXPIRY_WORKERS, v);
+            },
+            None => unsafe {
+                env::remove_var(ENV_MAX_EXPIRY_WORKERS);
+            },
+        }
+
+        let result = std::panic::catch_unwind(std::panic::AssertUnwindSafe(test_fn));
+
+        match original {
+            Some(v) => unsafe {
+                env::set_var(ENV_MAX_EXPIRY_WORKERS, v);
+            },
+            None => unsafe {
+                env::remove_var(ENV_MAX_EXPIRY_WORKERS);
+            },
+        }
+
+        if let Err(e) = result {
+            std::panic::resume_unwind(e);
+        }
+    }
+
+    /// backlog#1832: the single expiry knob must resolve all four env states
+    /// (unset / zero / valid / garbage); the removed `_RUSTFS_ILM_EXPIRATION_WORKERS`
+    /// override and `RUSTFS_DEFAULT_EXPIRY_WORKERS` fallback must stay gone.
+    #[test]
+    #[serial]
+    fn expiry_worker_count_resolves_all_env_states() {
+        let default = std::cmp::min(num_cpus::get(), 16);
+
+        with_expiry_worker_env(None, || {
+            assert_eq!(expiry_worker_count(), default, "unset env must fall back to min(cpus, 16)");
+        });
+        with_expiry_worker_env(Some("0"), || {
+            assert_eq!(expiry_worker_count(), default, "zero must fall back instead of spawning zero workers");
+        });
+        with_expiry_worker_env(Some("4"), || {
+            assert_eq!(expiry_worker_count(), 4, "a valid positive value must win");
+        });
+        with_expiry_worker_env(Some("not-a-number"), || {
+            assert_eq!(expiry_worker_count(), default, "garbage must fall back to the default");
+        });
+    }
+
     // SAFETY: this helper is only used from `#[serial]` tests and those tests run under a
     // single-thread runtime (`worker_threads = 1`), so no concurrent reader/writer can access
     // process environment while `env::set_var`/`env::remove_var` is active.
@@ -8554,9 +8634,10 @@ mod tests {
             ..Default::default()
         };
 
-        let persisted = persist_manual_transition_job_progress(ecstore.clone(), job_id, &report, queue_snapshot)
-            .await
-            .expect("page checkpoint should persist to the job record");
+        let persisted =
+            persist_manual_transition_job_progress_if_owned(ecstore.clone(), job_id, record.lease_id, &report, queue_snapshot)
+                .await
+                .expect("page checkpoint should persist to the job record");
 
         assert_eq!(persisted.state, ManualTransitionJobState::Running);
         assert_eq!(persisted.report.scanned, 1000);
@@ -8575,6 +8656,232 @@ mod tests {
         assert_eq!(admission.updated_at_unix_nanos, loaded.updated_at_unix_nanos);
     }
 
+    #[tokio::test]
+    #[serial]
+    async fn manual_transition_progress_retries_heartbeat_cas_without_losing_checkpoint() {
+        let (_paths, ecstore) = setup_test_env().await;
+        let job_id = Uuid::new_v4();
+        let options = ManualTransitionRunOptions {
+            prefix: "logs/".to_string(),
+            ..Default::default()
+        };
+        let record = ManualTransitionJobRecord::new(job_id, "manual-progress-cas-bucket", &options, "owner-a");
+        save_manual_transition_job_record(ecstore.clone(), &record)
+            .await
+            .expect("running job record should save");
+        save_manual_transition_scope_admission_if_absent(ecstore.clone(), &ManualTransitionScopeAdmission::from_job(&record))
+            .await
+            .expect("running scope admission should save");
+        let lease_id = record.lease_id;
+        let barrier = ManualTransitionJobCasBarrier::install(job_id);
+        let progress_store = ecstore.clone();
+        let progress = tokio::spawn(async move {
+            persist_manual_transition_job_progress_if_owned(
+                progress_store,
+                job_id,
+                lease_id,
+                &ManualTransitionRunReport {
+                    bucket: "manual-progress-cas-bucket".to_string(),
+                    prefix: "logs/".to_string(),
+                    scanned: 1000,
+                    eligible: 900,
+                    enqueued: 800,
+                    continuation_token: Some("opaque-page-cursor".to_string()),
+                    ..Default::default()
+                },
+                ManualTransitionQueueSnapshot {
+                    queued: 7,
+                    active: 3,
+                    ..Default::default()
+                },
+            )
+            .await
+        });
+        barrier.wait_until_paused().await;
+
+        let heartbeat = renew_manual_transition_job_lease_if_owned(
+            ecstore.clone(),
+            job_id,
+            lease_id,
+            ManualTransitionQueueSnapshot {
+                queued: 2,
+                active: 1,
+                ..Default::default()
+            },
+        )
+        .await
+        .expect("heartbeat should win the first CAS write");
+        barrier.release();
+        let checkpointed = progress
+            .await
+            .expect("progress task should join")
+            .expect("progress should retry its stale ETag");
+
+        assert_eq!(checkpointed.lease_id, heartbeat.lease_id);
+        assert_eq!(checkpointed.report.scanned, 1000);
+        assert_eq!(checkpointed.report.eligible, 900);
+        assert_eq!(checkpointed.report.enqueued, 800);
+        assert_eq!(checkpointed.report.continuation_token.as_deref(), Some("opaque-page-cursor"));
+        assert_eq!(checkpointed.queue_snapshot.queued, 7);
+        assert_eq!(checkpointed.queue_snapshot.active, 3);
+    }
+
+    #[tokio::test]
+    #[serial]
+    async fn manual_transition_progress_rejects_stale_recovery_lease() {
+        let (_paths, ecstore) = setup_test_env().await;
+        let job_id = Uuid::new_v4();
+        let record = ManualTransitionJobRecord::new(
+            job_id,
+            "manual-progress-stale-lease-bucket",
+            &ManualTransitionRunOptions::default(),
+            "owner-a",
+        );
+        let stale_lease_id = record.lease_id;
+        save_manual_transition_job_record(ecstore.clone(), &record)
+            .await
+            .expect("running job record should save");
+
+        let (mut recovered, etag) = load_manual_transition_job_record_with_etag(ecstore.clone(), job_id)
+            .await
+            .expect("running job record should load");
+        recovered.lease_id = Uuid::new_v4();
+        recovered.owner_id = "owner-b".to_string();
+        save_manual_transition_job_record_if_current(ecstore.clone(), &recovered, &etag)
+            .await
+            .expect("recovery owner should replace the lease");
+
+        let error = persist_manual_transition_job_progress_if_owned(
+            ecstore.clone(),
+            job_id,
+            stale_lease_id,
+            &ManualTransitionRunReport {
+                scanned: 1000,
+                continuation_token: Some("stale-owner-cursor".to_string()),
+                ..Default::default()
+            },
+            ManualTransitionQueueSnapshot::default(),
+        )
+        .await
+        .expect_err("the stale owner must not update the recovered job");
+        let heartbeat_error = renew_manual_transition_job_lease_if_owned(
+            ecstore.clone(),
+            job_id,
+            stale_lease_id,
+            ManualTransitionQueueSnapshot::default(),
+        )
+        .await
+        .expect_err("the stale owner must not renew the recovered job");
+
+        assert_eq!(error, Error::PreconditionFailed);
+        assert_eq!(heartbeat_error, Error::PreconditionFailed);
+        let loaded = load_manual_transition_job_record(ecstore, job_id)
+            .await
+            .expect("recovered job record should load");
+        assert_eq!(loaded.lease_id, recovered.lease_id);
+        assert_eq!(loaded.owner_id, "owner-b");
+        assert_eq!(loaded.report.scanned, 0);
+        assert!(loaded.report.continuation_token.is_none());
+    }
+
+    #[tokio::test]
+    #[serial]
+    async fn manual_transition_reconcile_rejects_lease_takeover_during_cas() {
+        let (_paths, ecstore) = setup_test_env().await;
+        let job_id = Uuid::new_v4();
+        let bucket = format!("manual-reconcile-lease-race-{}", job_id.simple());
+        let mut record = ManualTransitionJobRecord::new(job_id, &bucket, &ManualTransitionRunOptions::default(), "owner-a");
+        record.scan_completed = true;
+        let stale_lease_id = record.lease_id;
+        save_manual_transition_job_record(ecstore.clone(), &record)
+            .await
+            .expect("running job record should save");
+        let task_key = manual_transition_worker_result_task_key(&bucket, "logs/a", None);
+        let task = ManualTransitionTaskRecord::new(job_id, &task_key, &bucket, "logs/a", None, "WARM");
+        assert!(
+            save_manual_transition_task_if_absent(ecstore.clone(), &task)
+                .await
+                .expect("task journal marker should save")
+        );
+
+        let barrier = ManualTransitionJobCasBarrier::install(job_id);
+        let heartbeat_store = ecstore.clone();
+        let heartbeat = tokio::spawn(async move {
+            renew_manual_transition_job_lease_if_owned(
+                heartbeat_store,
+                job_id,
+                stale_lease_id,
+                ManualTransitionQueueSnapshot::default(),
+            )
+            .await
+        });
+        barrier.wait_until_paused().await;
+
+        let (mut recovered, etag) = load_manual_transition_job_record_with_etag(ecstore.clone(), job_id)
+            .await
+            .expect("running job record should load during reconciliation");
+        recovered.lease_id = Uuid::new_v4();
+        recovered.owner_id = "owner-b".to_string();
+        save_manual_transition_job_record_if_current(ecstore.clone(), &recovered, &etag)
+            .await
+            .expect("recovery owner should replace the lease");
+        barrier.release();
+
+        let error = heartbeat
+            .await
+            .expect("heartbeat task should join")
+            .expect_err("stale reconciliation must reject the recovery lease");
+        assert_eq!(error, Error::PreconditionFailed);
+        let loaded = load_manual_transition_job_record(ecstore, job_id)
+            .await
+            .expect("recovered job record should load");
+        assert_eq!(loaded.lease_id, recovered.lease_id);
+        assert_eq!(loaded.owner_id, "owner-b");
+        assert_eq!(loaded.state, ManualTransitionJobState::Running);
+        assert_eq!(loaded.report.enqueued, 0);
+    }
+
+    #[tokio::test]
+    #[serial]
+    async fn manual_transition_progress_does_not_regress_newer_admission_lease() {
+        let (_paths, ecstore) = setup_test_env().await;
+        let job_id = Uuid::new_v4();
+        let record = ManualTransitionJobRecord::new(
+            job_id,
+            "manual-progress-admission-order-bucket",
+            &ManualTransitionRunOptions::default(),
+            "owner-a",
+        );
+        save_manual_transition_job_record(ecstore.clone(), &record)
+            .await
+            .expect("running job record should save");
+        let mut newer_admission = ManualTransitionScopeAdmission::from_job(&record);
+        newer_admission.lease_expires_at_unix_nanos = newer_admission.lease_expires_at_unix_nanos.saturating_add(60_000_000_000);
+        newer_admission.updated_at_unix_nanos = newer_admission.updated_at_unix_nanos.saturating_add(60_000_000_000);
+        save_manual_transition_scope_admission_if_absent(ecstore.clone(), &newer_admission)
+            .await
+            .expect("newer scope admission should save");
+
+        persist_manual_transition_job_progress_if_owned(
+            ecstore.clone(),
+            job_id,
+            record.lease_id,
+            &ManualTransitionRunReport {
+                scanned: 1000,
+                ..Default::default()
+            },
+            ManualTransitionQueueSnapshot::default(),
+        )
+        .await
+        .expect("progress should preserve the newer admission lease");
+
+        let admission = load_manual_transition_scope_admission(ecstore, &record.scope_key)
+            .await
+            .expect("scope admission should load");
+        assert_eq!(admission.lease_expires_at_unix_nanos, newer_admission.lease_expires_at_unix_nanos);
+        assert_eq!(admission.updated_at_unix_nanos, newer_admission.updated_at_unix_nanos);
+    }
+
     #[tokio::test]
     async fn manual_transition_page_checkpoint_persists_resume_cursor() {
         let observed = Arc::new(StdMutex::new(Vec::new()));
@@ -8639,7 +8946,7 @@ mod tests {
             .await
             .expect("expired scope admission should save");
         let checkpoint_options = ManualTransitionRunOptions {
-            progress_sink: Some(manual_transition_recovery_progress_sink(ecstore.clone(), job_id)),
+            progress_sink: Some(manual_transition_recovery_progress_sink(ecstore.clone(), job_id, record.lease_id)),
             ..options
         };
         let report = ManualTransitionRunReport {
@@ -8728,7 +9035,7 @@ mod tests {
             prefix: prefix.to_string(),
             tier: Some("WARM".to_string()),
             dry_run: true,
-            progress_sink: Some(manual_transition_recovery_progress_sink(ecstore.clone(), job_id)),
+            progress_sink: Some(manual_transition_recovery_progress_sink(ecstore.clone(), job_id, record.lease_id)),
             ..Default::default()
         };
         let final_report = enqueue_transition_for_existing_objects_scoped(ecstore.clone(), &bucket, production_path_options)
@@ -9428,9 +9735,14 @@ mod tests {
             "new worker result marker must be created"
         );
 
-        let renewed = renew_manual_transition_job_lease(ecstore.clone(), job_id, ManualTransitionQueueSnapshot::default())
-            .await
-            .expect("heartbeat should reconcile marker before unknown fallback");
+        let renewed = renew_manual_transition_job_lease_if_owned(
+            ecstore.clone(),
+            job_id,
+            record.lease_id,
+            ManualTransitionQueueSnapshot::default(),
+        )
+        .await
+        .expect("heartbeat should reconcile marker before unknown fallback");
 
         assert_eq!(renewed.state, ManualTransitionJobState::Completed);
         assert_eq!(renewed.report.transition_completed, 1);
@@ -9473,9 +9785,14 @@ mod tests {
             "new worker result marker must be created"
         );
 
-        let renewed = renew_manual_transition_job_lease(ecstore.clone(), job_id, ManualTransitionQueueSnapshot::default())
-            .await
-            .expect("heartbeat should reconcile task and result journals");
+        let renewed = renew_manual_transition_job_lease_if_owned(
+            ecstore.clone(),
+            job_id,
+            record.lease_id,
+            ManualTransitionQueueSnapshot::default(),
+        )
+        .await
+        .expect("heartbeat should reconcile task and result journals");
 
         assert_eq!(renewed.state, ManualTransitionJobState::Completed);
         assert_eq!(renewed.report.enqueued, 1);
@@ -9790,9 +10107,10 @@ mod tests {
             .await
             .expect("running scope admission should save");
 
-        let checkpointed = persist_manual_transition_job_progress(
+        let checkpointed = persist_manual_transition_job_progress_if_owned(
             ecstore.clone(),
             job_id,
+            record.lease_id,
             &ManualTransitionRunReport {
                 bucket: bucket.to_string(),
                 prefix: "logs/".to_string(),
@@ -9881,7 +10199,7 @@ mod tests {
             compensation_running: 1,
         };
 
-        let renewed = renew_manual_transition_job_lease(ecstore.clone(), job_id, queue_snapshot)
+        let renewed = renew_manual_transition_job_lease_if_owned(ecstore.clone(), job_id, record.lease_id, queue_snapshot)
             .await
             .expect("running job heartbeat should persist queue pressure status");
 
@@ -9932,9 +10250,14 @@ mod tests {
             .await
             .expect("running job admission should save");
 
-        let renewed = renew_manual_transition_job_lease(ecstore.clone(), job_id, ManualTransitionQueueSnapshot::default())
-            .await
-            .expect("lost worker result should persist unknown state");
+        let renewed = renew_manual_transition_job_lease_if_owned(
+            ecstore.clone(),
+            job_id,
+            record.lease_id,
+            ManualTransitionQueueSnapshot::default(),
+        )
+        .await
+        .expect("lost worker result should persist unknown state");
 
         assert_eq!(renewed.state, ManualTransitionJobState::Unknown);
         assert!(renewed.completed_at_unix_nanos.is_some());
@@ -11524,7 +11847,6 @@ mod tests {
     }
 
     #[tokio::test]
-    #[ignore = "requires isolated global object layer state"]
     #[serial]
     async fn ecstore_new_succeeds_on_fresh_local_volumes() {
         let test_base_dir = format!("/tmp/rustfs_ecstore_empty_boot_{}", Uuid::new_v4());

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

@@ -86,6 +86,21 @@ where
     com::save_config_with_opts(api, file, data, opts).await
 }
 
+pub(crate) async fn save_config_with_opts_quiet<S>(api: Arc<S>, file: &str, data: Vec<u8>, opts: &ObjectOptions) -> Result<()>
+where
+    S: ObjectIO<
+            Error = Error,
+            RangeSpec = HTTPRangeSpec,
+            HeaderMap = HeaderMap,
+            ObjectOptions = ObjectOptions,
+            ObjectInfo = ObjectInfo,
+            GetObjectReader = GetObjectReader,
+            PutObjectReader = PutObjReader,
+        >,
+{
+    com::save_config_with_opts_quiet(api, file, data, opts).await
+}
+
 pub(crate) async fn delete_config<S>(api: Arc<S>, file: &str) -> Result<()>
 where
     S: ObjectOperations<

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

@@ -45,6 +45,104 @@ const MANUAL_TRANSITION_JOB_LEASE_SECONDS: i128 = 60;
 const MANUAL_TRANSITION_LEGACY_SCOPE_SCAN_LIMIT: i32 = 1000;
 const MANUAL_TRANSITION_TASK_SCAN_LIMIT: i32 = 1000;
 const MANUAL_TRANSITION_WORKER_RESULT_SCAN_LIMIT: i32 = 1000;
+const MANUAL_TRANSITION_JOB_CAS_RETRIES: usize = 4;
+
+#[cfg(test)]
+struct ManualTransitionJobCasBarrierState {
+    job_id: Uuid,
+    paused: std::sync::atomic::AtomicBool,
+    arrived: tokio::sync::Notify,
+    release: tokio::sync::Semaphore,
+}
+
+#[cfg(test)]
+pub(crate) struct ManualTransitionJobCasBarrier {
+    state: Arc<ManualTransitionJobCasBarrierState>,
+}
+
+#[cfg(test)]
+static MANUAL_TRANSITION_JOB_CAS_BARRIER: std::sync::OnceLock<std::sync::Mutex<Option<Arc<ManualTransitionJobCasBarrierState>>>> =
+    std::sync::OnceLock::new();
+
+#[cfg(test)]
+impl ManualTransitionJobCasBarrier {
+    pub(crate) fn install(job_id: Uuid) -> Self {
+        let state = Arc::new(ManualTransitionJobCasBarrierState {
+            job_id,
+            paused: std::sync::atomic::AtomicBool::new(false),
+            arrived: tokio::sync::Notify::new(),
+            release: tokio::sync::Semaphore::new(0),
+        });
+        let mut slot = MANUAL_TRANSITION_JOB_CAS_BARRIER
+            .get_or_init(|| std::sync::Mutex::new(None))
+            .lock()
+            .expect("manual transition progress CAS barrier mutex should not poison");
+        assert!(
+            slot.is_none(),
+            "manual transition job CAS barrier must be installed by one test at a time"
+        );
+        *slot = Some(Arc::clone(&state));
+        drop(slot);
+        Self { state }
+    }
+
+    pub(crate) async fn wait_until_paused(&self) {
+        tokio::time::timeout(std::time::Duration::from_secs(30), async {
+            loop {
+                let arrived = self.state.arrived.notified();
+                if self.state.paused.load(std::sync::atomic::Ordering::Acquire) {
+                    return;
+                }
+                arrived.await;
+            }
+        })
+        .await
+        .expect("manual transition job update should reach the deterministic CAS barrier");
+    }
+
+    pub(crate) fn release(&self) {
+        self.state.release.add_permits(1);
+    }
+}
+
+#[cfg(test)]
+impl Drop for ManualTransitionJobCasBarrier {
+    fn drop(&mut self) {
+        self.release();
+        let mut slot = MANUAL_TRANSITION_JOB_CAS_BARRIER
+            .get_or_init(|| std::sync::Mutex::new(None))
+            .lock()
+            .expect("manual transition progress CAS barrier mutex should not poison");
+        if slot.as_ref().is_some_and(|state| Arc::ptr_eq(state, &self.state)) {
+            *slot = None;
+        }
+    }
+}
+
+#[cfg(test)]
+async fn pause_manual_transition_job_before_first_cas(job_id: Uuid) {
+    let barrier = MANUAL_TRANSITION_JOB_CAS_BARRIER
+        .get_or_init(|| std::sync::Mutex::new(None))
+        .lock()
+        .expect("manual transition progress CAS barrier mutex should not poison")
+        .as_ref()
+        .filter(|barrier| barrier.job_id == job_id)
+        .cloned();
+    if let Some(barrier) = barrier
+        && barrier
+            .paused
+            .compare_exchange(false, true, std::sync::atomic::Ordering::AcqRel, std::sync::atomic::Ordering::Acquire)
+            .is_ok()
+    {
+        barrier.arrived.notify_one();
+        barrier
+            .release
+            .acquire()
+            .await
+            .expect("manual transition job CAS barrier should remain open")
+            .forget();
+    }
+}
 
 fn is_false(value: &bool) -> bool {
     !*value
@@ -148,7 +246,6 @@ impl ManualTransitionJobRecord {
 
     pub fn fail(&mut self, error: impl Into<String>) {
         self.state = ManualTransitionJobState::Failed;
-        self.report.tier_failure = self.report.tier_failure.saturating_add(1);
         self.error = Some(error.into());
         self.mark_updated_terminal();
     }
@@ -1040,7 +1137,7 @@ pub async fn save_manual_transition_job_record_if_current(
     }
     let object = manual_transition_job_record_object_name(job.job_id).map_err(manual_transition_job_store_error)?;
     let data = job.encode().map_err(manual_transition_job_store_error)?;
-    config_boundary::save_config_with_opts(
+    config_boundary::save_config_with_opts_quiet(
         api,
         &object,
         data,
@@ -1056,6 +1153,54 @@ pub async fn save_manual_transition_job_record_if_current(
     .await
 }
 
+/// Applies a job-record mutation with optimistic concurrency control.
+///
+/// The mutation returns whether the record needs to be persisted. When a lease
+/// is supplied, ownership is checked again after every conflicting write.
+pub async fn update_manual_transition_job_record<F>(
+    api: Arc<ECStore>,
+    job_id: Uuid,
+    expected_lease_id: Option<Uuid>,
+    update: F,
+) -> EcstoreResult<ManualTransitionJobRecord>
+where
+    F: FnMut(&mut ManualTransitionJobRecord) -> bool,
+{
+    update_manual_transition_job_record_from(api, job_id, expected_lease_id, None, update).await
+}
+
+async fn update_manual_transition_job_record_from<F>(
+    api: Arc<ECStore>,
+    job_id: Uuid,
+    expected_lease_id: Option<Uuid>,
+    mut current: Option<(ManualTransitionJobRecord, String)>,
+    mut update: F,
+) -> EcstoreResult<ManualTransitionJobRecord>
+where
+    F: FnMut(&mut ManualTransitionJobRecord) -> bool,
+{
+    for _ in 0..MANUAL_TRANSITION_JOB_CAS_RETRIES {
+        let (mut record, etag) = match current.take() {
+            Some(current) => current,
+            None => load_manual_transition_job_record_with_etag(api.clone(), job_id).await?,
+        };
+        if expected_lease_id.is_some_and(|lease_id| record.lease_id != lease_id) {
+            return Err(Error::PreconditionFailed);
+        }
+        if !update(&mut record) {
+            return Ok(record);
+        }
+        #[cfg(test)]
+        pause_manual_transition_job_before_first_cas(job_id).await;
+        match save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await {
+            Ok(()) => return Ok(record),
+            Err(Error::PreconditionFailed) => continue,
+            Err(err) => return Err(err),
+        }
+    }
+    Err(Error::PreconditionFailed)
+}
+
 pub(crate) async fn save_manual_transition_worker_result_if_absent(
     api: Arc<ECStore>,
     record: &ManualTransitionWorkerResultRecord,
@@ -1314,99 +1459,113 @@ pub async fn reconcile_manual_transition_worker_results(
     api: Arc<ECStore>,
     job_id: Uuid,
     queue_snapshot: ManualTransitionQueueSnapshot,
+) -> EcstoreResult<ManualTransitionJobRecord> {
+    reconcile_manual_transition_worker_results_inner(api, job_id, None, queue_snapshot, false).await
+}
+
+pub(crate) async fn reconcile_manual_transition_worker_results_if_owned(
+    api: Arc<ECStore>,
+    job_id: Uuid,
+    expected_lease_id: Uuid,
+    queue_snapshot: ManualTransitionQueueSnapshot,
+) -> EcstoreResult<ManualTransitionJobRecord> {
+    reconcile_manual_transition_worker_results_inner(api, job_id, Some(expected_lease_id), queue_snapshot, false).await
+}
+
+async fn reconcile_manual_transition_worker_results_inner(
+    api: Arc<ECStore>,
+    job_id: Uuid,
+    expected_lease_id: Option<Uuid>,
+    queue_snapshot: ManualTransitionQueueSnapshot,
+    mark_missing_results_unknown: bool,
 ) -> EcstoreResult<ManualTransitionJobRecord> {
     let task_stats = match scan_manual_transition_task_journal(api.clone(), job_id).await? {
         ManualTransitionTaskJournal::Stats(stats) => stats,
         ManualTransitionTaskJournal::Corrupt(error) => {
-            return mark_manual_transition_job_unknown_for_task_journal_error(api, job_id, error, queue_snapshot).await;
+            return mark_manual_transition_job_unknown_for_task_journal_error(
+                api,
+                job_id,
+                expected_lease_id,
+                error,
+                queue_snapshot,
+            )
+            .await;
         }
     };
     let stats = match scan_manual_transition_worker_result_journal(api.clone(), job_id).await? {
         ManualTransitionWorkerResultJournal::Stats(stats) => stats,
         ManualTransitionWorkerResultJournal::Corrupt(error) => {
-            return mark_manual_transition_job_unknown_for_worker_result_journal_error(api, job_id, error, queue_snapshot).await;
+            return mark_manual_transition_job_unknown_for_worker_result_journal_error(
+                api,
+                job_id,
+                expected_lease_id,
+                error,
+                queue_snapshot,
+            )
+            .await;
         }
     };
-    for _ in 0..4 {
-        let (mut record, etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
-        let changed = record.apply_worker_result_counts(
+    let mut changed = false;
+    let record = update_manual_transition_job_record(api.clone(), job_id, expected_lease_id, |record| {
+        let counts_changed = record.apply_worker_result_counts(
             stats.stats.completed,
             stats.stats.failed,
             &stats.stats.tier_failure_by_reason,
             task_stats.queued,
             queue_snapshot,
         );
-        if !changed {
-            return Ok(record);
-        }
-        match save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await {
-            Ok(()) => {
-                if record.is_terminal() {
-                    delete_manual_transition_scope_admission_if_current(
-                        api.clone(),
-                        &record.scope_key,
-                        record.job_id,
-                        record.lease_id,
-                    )
-                    .await?;
-                } else {
-                    renew_manual_transition_scope_admission_from_job(api, &record).await?;
-                }
-                return Ok(record);
-            }
-            Err(Error::PreconditionFailed) => continue,
-            Err(err) => return Err(err),
-        }
+        let became_unknown = mark_missing_results_unknown && record.mark_unknown_if_worker_results_lost(queue_snapshot);
+        changed = counts_changed || became_unknown;
+        changed
+    })
+    .await?;
+    if !changed {
+        return Ok(record);
     }
-    Err(Error::PreconditionFailed)
+    if record.is_terminal() {
+        delete_manual_transition_scope_admission_if_current(api, &record.scope_key, record.job_id, record.lease_id).await?;
+    } else {
+        renew_manual_transition_scope_admission_from_job(api, &record).await?;
+    }
+    Ok(record)
 }
 
 async fn mark_manual_transition_job_unknown_for_task_journal_error(
     api: Arc<ECStore>,
     job_id: Uuid,
+    expected_lease_id: Option<Uuid>,
     error: String,
     queue_snapshot: ManualTransitionQueueSnapshot,
 ) -> EcstoreResult<ManualTransitionJobRecord> {
-    for _ in 0..4 {
-        let (mut record, etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
-        if !record.mark_unknown_for_task_journal_error(error.clone(), queue_snapshot) {
-            return Ok(record);
-        }
-        match save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await {
-            Ok(()) => {
-                delete_manual_transition_scope_admission_if_current(api, &record.scope_key, record.job_id, record.lease_id)
-                    .await?;
-                return Ok(record);
-            }
-            Err(Error::PreconditionFailed) => continue,
-            Err(err) => return Err(err),
-        }
+    let mut changed = false;
+    let record = update_manual_transition_job_record(api.clone(), job_id, expected_lease_id, |record| {
+        changed = record.mark_unknown_for_task_journal_error(error.clone(), queue_snapshot);
+        changed
+    })
+    .await?;
+    if changed && record.is_terminal() {
+        delete_manual_transition_scope_admission_if_current(api, &record.scope_key, record.job_id, record.lease_id).await?;
     }
-    Err(Error::PreconditionFailed)
+    Ok(record)
 }
 
 async fn mark_manual_transition_job_unknown_for_worker_result_journal_error(
     api: Arc<ECStore>,
     job_id: Uuid,
+    expected_lease_id: Option<Uuid>,
     error: String,
     queue_snapshot: ManualTransitionQueueSnapshot,
 ) -> EcstoreResult<ManualTransitionJobRecord> {
-    for _ in 0..4 {
-        let (mut record, etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
-        if !record.mark_unknown_for_worker_result_journal_error(error.clone(), queue_snapshot) {
-            return Ok(record);
-        }
-        match save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await {
-            Ok(()) => {
-                delete_manual_transition_scope_admission_if_current(api, &record.scope_key, record.job_id, record.lease_id)
-                    .await?;
-                return Ok(record);
-            }
-            Err(Error::PreconditionFailed) => continue,
-            Err(err) => return Err(err),
-        }
+    let mut changed = false;
+    let record = update_manual_transition_job_record(api.clone(), job_id, expected_lease_id, |record| {
+        changed = record.mark_unknown_for_worker_result_journal_error(error.clone(), queue_snapshot);
+        changed
+    })
+    .await?;
+    if changed && record.is_terminal() {
+        delete_manual_transition_scope_admission_if_current(api, &record.scope_key, record.job_id, record.lease_id).await?;
     }
-    Err(Error::PreconditionFailed)
+    Ok(record)
 }
 
 pub async fn save_manual_transition_scope_admission_if_absent(
@@ -1603,19 +1762,14 @@ async fn find_active_legacy_manual_transition_scope_conflict(
 }
 
 pub async fn request_manual_transition_job_cancel(api: Arc<ECStore>, job_id: Uuid) -> EcstoreResult<ManualTransitionJobRecord> {
-    for _ in 0..4 {
-        let (mut record, etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
+    update_manual_transition_job_record(api, job_id, None, |record| {
         if record.is_terminal() || record.cancel_requested {
-            return Ok(record);
+            return false;
         }
         record.mark_cancel_requested();
-        match save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await {
-            Ok(()) => return Ok(record),
-            Err(Error::PreconditionFailed) => continue,
-            Err(err) => return Err(err),
-        }
-    }
-    Err(Error::PreconditionFailed)
+        true
+    })
+    .await
 }
 
 pub async fn persist_manual_transition_job_progress(
@@ -1624,10 +1778,39 @@ pub async fn persist_manual_transition_job_progress(
     report: &ManualTransitionRunReport,
     queue_snapshot: ManualTransitionQueueSnapshot,
 ) -> EcstoreResult<ManualTransitionJobRecord> {
-    let (mut record, etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
-    record.update_running_progress(report.clone(), queue_snapshot);
-    save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await?;
-    renew_manual_transition_scope_admission_from_job(api, &record).await?;
+    let current = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
+    persist_manual_transition_job_progress_inner(api, job_id, current.0.lease_id, Some(current), report, queue_snapshot).await
+}
+
+pub async fn persist_manual_transition_job_progress_if_owned(
+    api: Arc<ECStore>,
+    job_id: Uuid,
+    expected_lease_id: Uuid,
+    report: &ManualTransitionRunReport,
+    queue_snapshot: ManualTransitionQueueSnapshot,
+) -> EcstoreResult<ManualTransitionJobRecord> {
+    persist_manual_transition_job_progress_inner(api, job_id, expected_lease_id, None, report, queue_snapshot).await
+}
+
+async fn persist_manual_transition_job_progress_inner(
+    api: Arc<ECStore>,
+    job_id: Uuid,
+    expected_lease_id: Uuid,
+    current: Option<(ManualTransitionJobRecord, String)>,
+    report: &ManualTransitionRunReport,
+    queue_snapshot: ManualTransitionQueueSnapshot,
+) -> EcstoreResult<ManualTransitionJobRecord> {
+    let record = update_manual_transition_job_record_from(api.clone(), job_id, Some(expected_lease_id), current, |record| {
+        if record.state != ManualTransitionJobState::Running {
+            return false;
+        }
+        record.update_running_progress(report.clone(), queue_snapshot);
+        true
+    })
+    .await?;
+    if record.state == ManualTransitionJobState::Running {
+        renew_manual_transition_scope_admission_from_job(api, &record).await?;
+    }
     Ok(record)
 }
 
@@ -1661,25 +1844,58 @@ pub async fn renew_manual_transition_job_lease(
     job_id: Uuid,
     queue_snapshot: ManualTransitionQueueSnapshot,
 ) -> EcstoreResult<ManualTransitionJobRecord> {
-    let (mut record, mut etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
-    if record.state == ManualTransitionJobState::Running {
-        if record.scan_completed && queue_snapshot.queued == 0 && queue_snapshot.active == 0 {
-            record = reconcile_manual_transition_worker_results(api.clone(), job_id, queue_snapshot).await?;
-            if record.is_terminal() || !record.report.worker_transition_pending() {
-                return Ok(record);
+    let current = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
+    renew_manual_transition_job_lease_inner(api, job_id, current.0.lease_id, Some(current), queue_snapshot).await
+}
+
+pub async fn renew_manual_transition_job_lease_if_owned(
+    api: Arc<ECStore>,
+    job_id: Uuid,
+    expected_lease_id: Uuid,
+    queue_snapshot: ManualTransitionQueueSnapshot,
+) -> EcstoreResult<ManualTransitionJobRecord> {
+    renew_manual_transition_job_lease_inner(api, job_id, expected_lease_id, None, queue_snapshot).await
+}
+
+async fn renew_manual_transition_job_lease_inner(
+    api: Arc<ECStore>,
+    job_id: Uuid,
+    expected_lease_id: Uuid,
+    current: Option<(ManualTransitionJobRecord, String)>,
+    queue_snapshot: ManualTransitionQueueSnapshot,
+) -> EcstoreResult<ManualTransitionJobRecord> {
+    let (current, current_etag) = match current {
+        Some(current) => current,
+        None => load_manual_transition_job_record_with_etag(api.clone(), job_id).await?,
+    };
+    if current.lease_id != expected_lease_id {
+        return Err(Error::PreconditionFailed);
+    }
+    if current.state != ManualTransitionJobState::Running {
+        return Ok(current);
+    }
+    if current.scan_completed && queue_snapshot.queued == 0 && queue_snapshot.active == 0 {
+        return reconcile_manual_transition_worker_results_inner(api, job_id, Some(expected_lease_id), queue_snapshot, true)
+            .await;
+    }
+    let record = update_manual_transition_job_record_from(
+        api.clone(),
+        job_id,
+        Some(expected_lease_id),
+        Some((current, current_etag)),
+        |record| {
+            if record.state != ManualTransitionJobState::Running {
+                return false;
             }
-            (record, etag) = load_manual_transition_job_record_with_etag(api.clone(), job_id).await?;
-        }
-        let became_terminal = record.mark_unknown_if_worker_results_lost(queue_snapshot);
-        if !became_terminal {
             record.renew_lease(queue_snapshot);
-        }
-        save_manual_transition_job_record_if_current(api.clone(), &record, &etag).await?;
-        if became_terminal {
-            delete_manual_transition_scope_admission_if_current(api, &record.scope_key, record.job_id, record.lease_id).await?;
-        } else {
-            renew_manual_transition_scope_admission_from_job(api, &record).await?;
-        }
+            true
+        },
+    )
+    .await?;
+    if record.is_terminal() {
+        delete_manual_transition_scope_admission_if_current(api, &record.scope_key, record.job_id, record.lease_id).await?;
+    } else if record.state == ManualTransitionJobState::Running {
+        renew_manual_transition_scope_admission_from_job(api, &record).await?;
     }
     Ok(record)
 }
@@ -1688,15 +1904,31 @@ async fn renew_manual_transition_scope_admission_from_job(
     api: Arc<ECStore>,
     record: &ManualTransitionJobRecord,
 ) -> EcstoreResult<()> {
-    if let Ok((admission, admission_etag)) =
-        load_manual_transition_scope_admission_with_etag(api.clone(), &record.scope_key).await
-        && admission.job_id == record.job_id
-        && admission.lease_id == record.lease_id
-    {
-        let renewed_admission = ManualTransitionScopeAdmission::from_job(record);
-        save_manual_transition_scope_admission_if_current(api, &renewed_admission, &admission_etag).await?;
+    for _ in 0..MANUAL_TRANSITION_JOB_CAS_RETRIES {
+        let (admission, admission_etag) =
+            match load_manual_transition_scope_admission_with_etag(api.clone(), &record.scope_key).await {
+                Ok(admission) => admission,
+                Err(Error::ConfigNotFound) => return Ok(()),
+                Err(err) => return Err(err),
+            };
+        if admission.job_id != record.job_id || admission.lease_id != record.lease_id {
+            return Err(Error::PreconditionFailed);
+        }
+        let mut renewed_admission = ManualTransitionScopeAdmission::from_job(record);
+        renewed_admission.lease_expires_at_unix_nanos = renewed_admission
+            .lease_expires_at_unix_nanos
+            .max(admission.lease_expires_at_unix_nanos);
+        renewed_admission.updated_at_unix_nanos = renewed_admission.updated_at_unix_nanos.max(admission.updated_at_unix_nanos);
+        if renewed_admission == admission {
+            return Ok(());
+        }
+        match save_manual_transition_scope_admission_if_current(api.clone(), &renewed_admission, &admission_etag).await {
+            Ok(()) => return Ok(()),
+            Err(Error::PreconditionFailed) => continue,
+            Err(err) => return Err(err),
+        }
     }
-    Ok(())
+    Err(Error::PreconditionFailed)
 }
 
 pub async fn delete_manual_transition_scope_admission_if_current(
@@ -2386,14 +2618,14 @@ mod tests {
     }
 
     #[test]
-    fn manual_transition_job_record_failure_counts_tier_failure() {
+    fn manual_transition_job_record_control_plane_failure_does_not_count_tier_failure() {
         let options = ManualTransitionRunOptions::default();
         let mut record = ManualTransitionJobRecord::new(Uuid::new_v4(), "bucket", &options, TEST_OWNER);
 
         record.fail("missing tier");
 
         assert_eq!(record.state, ManualTransitionJobState::Failed);
-        assert_eq!(record.report.tier_failure, 1);
+        assert_eq!(record.report.tier_failure, 0);
         assert_eq!(record.error.as_deref(), Some("missing tier"));
     }
 

crates/ecstore/src/bucket/metadata.rs

@@ -1311,7 +1311,7 @@ mod test {
         assert!(bm.object_locking(), "object lock active via parsed config");
     }
 
-    /// backlog#580: KNOWN GAP (weisd 2026-03-06 "inline_data 前缀不同"). RustFS's
+    /// backlog#580: KNOWN GAP (flagged 2026-03-06: "inline_data 前缀不同"). RustFS's
     /// inline-data extraction does not yet recover the object body from a
     /// MinIO-written bucket-metadata object: `into_fileinfo(read_data=true).data`
     /// returns bytes that are not the `.metadata.bin` blob (no `format|version`
@@ -1319,7 +1319,7 @@ mod test {
     /// inline-data framing is handled on the read path.
     /// backlog#580: prove RustFS reads a MinIO-written **inlined** bucket-metadata
     /// object end-to-end. MinIO stores inline data as `[bitrot hash][object body]`
-    /// (the "`inline_data` 前缀不同" that weisd flagged on 2026-03-06 is that
+    /// (the "`inline_data` 前缀不同" gap flagged on 2026-03-06 is that
     /// bitrot prefix, not a format incompatibility). Running the raw inline shard
     /// through RustFS's `BitrotReader` with the default `HighwayHash256S` must
     /// verify the checksum and yield the exact `.metadata.bin` blob.

crates/ecstore/src/client/api_bucket_policy.rs

@@ -1,171 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-#![allow(unused_imports)]
-#![allow(unused_variables)]
-#![allow(unused_mut)]
-#![allow(unused_assignments)]
-#![allow(unused_must_use)]
-#![allow(clippy::all)]
-
-use http::{HeaderMap, StatusCode};
-use http_body_util::BodyExt;
-use hyper::body::Body;
-use hyper::body::Bytes;
-use std::collections::HashMap;
-
-use crate::client::{
-    api_error_response::http_resp_to_error_response,
-    transition_api::{ReaderImpl, RequestMetadata, TransitionClient},
-};
-use rustfs_utils::hash::EMPTY_STRING_SHA256_HASH;
-
-impl TransitionClient {
-    pub async fn set_bucket_policy(&self, bucket_name: &str, policy: &str) -> Result<(), std::io::Error> {
-        if policy == "" {
-            return self.remove_bucket_policy(bucket_name).await;
-        }
-
-        self.put_bucket_policy(bucket_name, policy).await
-    }
-
-    pub async fn put_bucket_policy(&self, bucket_name: &str, policy: &str) -> Result<(), std::io::Error> {
-        let mut url_values = HashMap::new();
-        url_values.insert("policy".to_string(), "".to_string());
-
-        let mut req_metadata = RequestMetadata {
-            bucket_name: bucket_name.to_string(),
-            query_values: url_values,
-            content_body: ReaderImpl::Body(Bytes::from(policy.as_bytes().to_vec())),
-            content_length: policy.len() as i64,
-            object_name: "".to_string(),
-            custom_header: HeaderMap::new(),
-            content_md5_base64: "".to_string(),
-            content_sha256_hex: "".to_string(),
-            stream_sha256: false,
-            trailer: HeaderMap::new(),
-            pre_sign_url: Default::default(),
-            add_crc: Default::default(),
-            extra_pre_sign_header: Default::default(),
-            bucket_location: Default::default(),
-            expires: Default::default(),
-        };
-
-        let resp = self.execute_method(http::Method::PUT, &mut req_metadata).await?;
-        //defer closeResponse(resp)
-
-        let resp_status = resp.status();
-        let h = resp.headers().clone();
-
-        //if resp != nil {
-        if resp_status != StatusCode::NO_CONTENT && resp.status() != StatusCode::OK {
-            return Err(std::io::Error::other(http_resp_to_error_response(
-                resp_status,
-                &h,
-                vec![],
-                bucket_name,
-                "",
-            )));
-        }
-        //}
-        Ok(())
-    }
-
-    pub async fn remove_bucket_policy(&self, bucket_name: &str) -> Result<(), std::io::Error> {
-        let mut url_values = HashMap::new();
-        url_values.insert("policy".to_string(), "".to_string());
-
-        let resp = self
-            .execute_method(
-                http::Method::DELETE,
-                &mut RequestMetadata {
-                    bucket_name: bucket_name.to_string(),
-                    query_values: url_values,
-                    content_sha256_hex: EMPTY_STRING_SHA256_HASH.to_string(),
-                    object_name: "".to_string(),
-                    custom_header: HeaderMap::new(),
-                    content_body: ReaderImpl::Body(Bytes::new()),
-                    content_length: 0,
-                    content_md5_base64: "".to_string(),
-                    stream_sha256: false,
-                    trailer: HeaderMap::new(),
-                    pre_sign_url: Default::default(),
-                    add_crc: Default::default(),
-                    extra_pre_sign_header: Default::default(),
-                    bucket_location: Default::default(),
-                    expires: Default::default(),
-                },
-            )
-            .await?;
-        //defer closeResponse(resp)
-
-        let resp_status = resp.status();
-        let h = resp.headers().clone();
-
-        if resp_status != StatusCode::NO_CONTENT {
-            return Err(std::io::Error::other(http_resp_to_error_response(
-                resp_status,
-                &h,
-                vec![],
-                bucket_name,
-                "",
-            )));
-        }
-
-        Ok(())
-    }
-
-    pub async fn get_bucket_policy(&self, bucket_name: &str) -> Result<String, std::io::Error> {
-        let bucket_policy = self.get_bucket_policy_inner(bucket_name).await?;
-        Ok(bucket_policy)
-    }
-
-    pub async fn get_bucket_policy_inner(&self, bucket_name: &str) -> Result<String, std::io::Error> {
-        let mut url_values = HashMap::new();
-        url_values.insert("policy".to_string(), "".to_string());
-
-        let resp = self
-            .execute_method(
-                http::Method::GET,
-                &mut RequestMetadata {
-                    bucket_name: bucket_name.to_string(),
-                    query_values: url_values,
-                    content_sha256_hex: EMPTY_STRING_SHA256_HASH.to_string(),
-                    object_name: "".to_string(),
-                    custom_header: HeaderMap::new(),
-                    content_body: ReaderImpl::Body(Bytes::new()),
-                    content_length: 0,
-                    content_md5_base64: "".to_string(),
-                    stream_sha256: false,
-                    trailer: HeaderMap::new(),
-                    pre_sign_url: Default::default(),
-                    add_crc: Default::default(),
-                    extra_pre_sign_header: Default::default(),
-                    bucket_location: Default::default(),
-                    expires: Default::default(),
-                },
-            )
-            .await?;
-
-        let mut body_vec = Vec::new();
-        let mut body = resp.into_body();
-        while let Some(frame) = body.frame().await {
-            let frame = frame.map_err(|e| std::io::Error::new(std::io::ErrorKind::Other, e.to_string()))?;
-            if let Some(data) = frame.data_ref() {
-                body_vec.extend_from_slice(data);
-            }
-        }
-        let policy = String::from_utf8_lossy(&body_vec).to_string();
-        Ok(policy)
-    }
-}

crates/ecstore/src/client/api_get_object_acl.rs

@@ -1,199 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-#![allow(unused_imports)]
-#![allow(unused_variables)]
-#![allow(unused_mut)]
-#![allow(unused_assignments)]
-#![allow(unused_must_use)]
-#![allow(clippy::all)]
-
-use crate::client::{
-    api_error_response::http_resp_to_error_response,
-    api_get_options::GetObjectOptions,
-    transition_api::{ObjectInfo, ReaderImpl, RequestMetadata, TransitionClient},
-};
-use bytes::Bytes;
-use http::{HeaderMap, HeaderValue};
-use http_body_util::BodyExt;
-use rustfs_config::MAX_S3_CLIENT_RESPONSE_SIZE;
-use rustfs_utils::EMPTY_STRING_SHA256_HASH;
-use s3s::dto::Owner;
-use std::collections::HashMap;
-
-#[derive(Clone, Debug, Default, serde::Serialize, serde::Deserialize)]
-pub struct Grantee {
-    pub id: String,
-    pub display_name: String,
-    pub uri: String,
-}
-
-#[derive(Clone, Debug, Default, serde::Serialize, serde::Deserialize)]
-pub struct Grant {
-    pub grantee: Grantee,
-    pub permission: String,
-}
-
-#[derive(Debug, Default, serde::Serialize, serde::Deserialize)]
-pub struct AccessControlList {
-    pub grant: Vec<Grant>,
-    pub permission: String,
-}
-
-#[derive(Debug, Default, serde::Deserialize)]
-pub struct AccessControlPolicy {
-    #[serde(skip)]
-    owner: Owner,
-    pub access_control_list: AccessControlList,
-}
-
-impl TransitionClient {
-    pub async fn get_object_acl(&self, bucket_name: &str, object_name: &str) -> Result<ObjectInfo, std::io::Error> {
-        let mut url_values = HashMap::new();
-        url_values.insert("acl".to_string(), "".to_string());
-        let mut resp = self
-            .execute_method(
-                http::Method::GET,
-                &mut RequestMetadata {
-                    bucket_name: bucket_name.to_string(),
-                    object_name: object_name.to_string(),
-                    query_values: url_values,
-                    custom_header: HeaderMap::new(),
-                    content_sha256_hex: EMPTY_STRING_SHA256_HASH.to_string(),
-                    content_body: ReaderImpl::Body(Bytes::new()),
-                    content_length: 0,
-                    content_md5_base64: "".to_string(),
-                    stream_sha256: false,
-                    trailer: HeaderMap::new(),
-                    pre_sign_url: Default::default(),
-                    add_crc: Default::default(),
-                    extra_pre_sign_header: Default::default(),
-                    bucket_location: Default::default(),
-                    expires: Default::default(),
-                },
-            )
-            .await?;
-
-        let resp_status = resp.status();
-        let h = resp.headers().clone();
-
-        let mut body_vec = Vec::new();
-        let mut body = resp.into_body();
-        while let Some(frame) = body.frame().await {
-            let frame = frame.map_err(|e| std::io::Error::new(std::io::ErrorKind::Other, e.to_string()))?;
-            if let Some(data) = frame.data_ref() {
-                body_vec.extend_from_slice(data);
-            }
-        }
-
-        if resp_status != http::StatusCode::OK {
-            return Err(std::io::Error::other(http_resp_to_error_response(
-                resp_status,
-                &h,
-                body_vec,
-                bucket_name,
-                object_name,
-            )));
-        }
-
-        let mut res = match quick_xml::de::from_str::<AccessControlPolicy>(&String::from_utf8(body_vec).unwrap()) {
-            Ok(result) => result,
-            Err(err) => {
-                return Err(std::io::Error::other(err.to_string()));
-            }
-        };
-
-        let mut obj_info = self
-            .stat_object(bucket_name, object_name, &GetObjectOptions::default())
-            .await?;
-
-        obj_info.owner.display_name = res.owner.display_name.clone();
-        obj_info.owner.id = res.owner.id.clone();
-
-        //obj_info.grant.extend(res.access_control_list.grant);
-
-        let canned_acl = get_canned_acl(&res);
-        if canned_acl != "" {
-            obj_info
-                .metadata
-                .insert("X-Amz-Acl", HeaderValue::from_str(&canned_acl).unwrap());
-            return Ok(obj_info);
-        }
-
-        let grant_acl = get_amz_grant_acl(&res);
-        /*for (k, v) in grant_acl {
-            obj_info.metadata.insert(HeaderName::from_bytes(k.as_bytes()).unwrap(), HeaderValue::from_str(&v.to_string()).unwrap());
-        }*/
-
-        Ok(obj_info)
-    }
-}
-
-fn get_canned_acl(ac_policy: &AccessControlPolicy) -> String {
-    let grants = ac_policy.access_control_list.grant.clone();
-
-    if grants.len() == 1 {
-        if grants[0].grantee.uri == "" && grants[0].permission == "FULL_CONTROL" {
-            return "private".to_string();
-        }
-    } else if grants.len() == 2 {
-        for g in grants {
-            if g.grantee.uri == "http://acs.amazonaws.com/groups/global/AuthenticatedUsers" && &g.permission == "READ" {
-                return "authenticated-read".to_string();
-            }
-            if g.grantee.uri == "http://acs.amazonaws.com/groups/global/AllUsers" && &g.permission == "READ" {
-                return "public-read".to_string();
-            }
-            if g.permission == "READ" && g.grantee.id == ac_policy.owner.id.clone().unwrap() {
-                return "bucket-owner-read".to_string();
-            }
-        }
-    } else if grants.len() == 3 {
-        for g in grants {
-            if g.grantee.uri == "http://acs.amazonaws.com/groups/global/AllUsers" && g.permission == "WRITE" {
-                return "public-read-write".to_string();
-            }
-        }
-    }
-    "".to_string()
-}
-
-pub fn get_amz_grant_acl(ac_policy: &AccessControlPolicy) -> HashMap<String, Vec<String>> {
-    let grants = ac_policy.access_control_list.grant.clone();
-    let mut res = HashMap::<String, Vec<String>>::new();
-
-    for g in grants {
-        let mut id = "id=".to_string();
-        id.push_str(&g.grantee.id);
-        let permission: &str = &g.permission;
-        match permission {
-            "READ" => {
-                res.entry("X-Amz-Grant-Read".to_string()).or_insert(vec![]).push(id);
-            }
-            "WRITE" => {
-                res.entry("X-Amz-Grant-Write".to_string()).or_insert(vec![]).push(id);
-            }
-            "READ_ACP" => {
-                res.entry("X-Amz-Grant-Read-Acp".to_string()).or_insert(vec![]).push(id);
-            }
-            "WRITE_ACP" => {
-                res.entry("X-Amz-Grant-Write-Acp".to_string()).or_insert(vec![]).push(id);
-            }
-            "FULL_CONTROL" => {
-                res.entry("X-Amz-Grant-Full-Control".to_string()).or_insert(vec![]).push(id);
-            }
-            _ => (),
-        }
-    }
-    res
-}

crates/ecstore/src/client/api_get_object_attributes.rs

@@ -1,266 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-#![allow(unused_imports)]
-#![allow(unused_variables)]
-#![allow(unused_mut)]
-#![allow(unused_assignments)]
-#![allow(unused_must_use)]
-#![allow(clippy::all)]
-
-use http::{HeaderMap, HeaderValue};
-use std::collections::HashMap;
-use time::OffsetDateTime;
-
-use crate::client::constants::{GET_OBJECT_ATTRIBUTES_MAX_PARTS, GET_OBJECT_ATTRIBUTES_TAGS, ISO8601_DATEFORMAT};
-use crate::client::{
-    api_get_object_acl::AccessControlPolicy,
-    transition_api::{ReaderImpl, RequestMetadata, TransitionClient},
-};
-use http_body_util::BodyExt;
-use hyper::body::Body;
-use hyper::body::Bytes;
-use hyper::body::Incoming;
-use rustfs_config::MAX_S3_CLIENT_RESPONSE_SIZE;
-use rustfs_utils::EMPTY_STRING_SHA256_HASH;
-use s3s::header::{X_AMZ_MAX_PARTS, X_AMZ_OBJECT_ATTRIBUTES, X_AMZ_PART_NUMBER_MARKER, X_AMZ_VERSION_ID};
-
-pub struct ObjectAttributesOptions {
-    pub max_parts: i64,
-    pub version_id: String,
-    pub part_number_marker: i64,
-    //server_side_encryption: encrypt::ServerSide,
-}
-
-pub struct ObjectAttributes {
-    pub version_id: String,
-    pub last_modified: OffsetDateTime,
-    pub object_attributes_response: ObjectAttributesResponse,
-}
-
-impl ObjectAttributes {
-    fn new() -> Self {
-        Self {
-            version_id: "".to_string(),
-            last_modified: OffsetDateTime::now_utc(),
-            object_attributes_response: ObjectAttributesResponse::new(),
-        }
-    }
-}
-
-#[derive(Debug, Default, serde::Deserialize)]
-pub struct Checksum {
-    checksum_crc32: String,
-    checksum_crc32c: String,
-    checksum_sha1: String,
-    checksum_sha256: String,
-}
-
-impl Checksum {
-    fn new() -> Self {
-        Self {
-            checksum_crc32: "".to_string(),
-            checksum_crc32c: "".to_string(),
-            checksum_sha1: "".to_string(),
-            checksum_sha256: "".to_string(),
-        }
-    }
-}
-
-#[derive(Debug, Default, serde::Deserialize)]
-pub struct ObjectParts {
-    pub parts_count: i64,
-    pub part_number_marker: i64,
-    pub next_part_number_marker: i64,
-    pub max_parts: i64,
-    is_truncated: bool,
-    parts: Vec<ObjectAttributePart>,
-}
-
-impl ObjectParts {
-    fn new() -> Self {
-        Self {
-            parts_count: 0,
-            part_number_marker: 0,
-            next_part_number_marker: 0,
-            max_parts: 0,
-            is_truncated: false,
-            parts: Vec::new(),
-        }
-    }
-}
-
-#[derive(Debug, Default, serde::Deserialize)]
-pub struct ObjectAttributesResponse {
-    pub etag: String,
-    pub storage_class: String,
-    pub object_size: i64,
-    pub checksum: Checksum,
-    pub object_parts: ObjectParts,
-}
-
-impl ObjectAttributesResponse {
-    fn new() -> Self {
-        Self {
-            etag: "".to_string(),
-            storage_class: "".to_string(),
-            object_size: 0,
-            checksum: Checksum::new(),
-            object_parts: ObjectParts::new(),
-        }
-    }
-}
-
-#[derive(Debug, Default, serde::Deserialize)]
-struct ObjectAttributePart {
-    checksum_crc32: String,
-    checksum_crc32c: String,
-    checksum_sha1: String,
-    checksum_sha256: String,
-    part_number: i64,
-    size: i64,
-}
-
-impl ObjectAttributes {
-    pub async fn parse_response(&mut self, h: &HeaderMap, body_vec: Vec<u8>) -> Result<(), std::io::Error> {
-        let last_modified = h
-            .get("Last-Modified")
-            .ok_or_else(|| std::io::Error::other("missing Last-Modified header"))?
-            .to_str()
-            .map_err(|e| std::io::Error::other(format!("invalid Last-Modified header: {e}")))?;
-        let mod_time = OffsetDateTime::parse(last_modified, ISO8601_DATEFORMAT)
-            .map_err(|e| std::io::Error::other(format!("invalid Last-Modified date: {e}")))?;
-        self.last_modified = mod_time;
-
-        let version_id = h
-            .get(X_AMZ_VERSION_ID)
-            .ok_or_else(|| std::io::Error::other("missing version ID header"))?
-            .to_str()
-            .map_err(|e| std::io::Error::other(format!("invalid version ID header: {e}")))?;
-        self.version_id = version_id.to_string();
-
-        let body_str = String::from_utf8(body_vec).map_err(|e| std::io::Error::other(format!("invalid UTF-8 body: {e}")))?;
-        let mut response = match quick_xml::de::from_str::<ObjectAttributesResponse>(&body_str) {
-            Ok(result) => result,
-            Err(err) => {
-                return Err(std::io::Error::other(err.to_string()));
-            }
-        };
-        self.object_attributes_response = response;
-
-        Ok(())
-    }
-}
-
-impl TransitionClient {
-    pub async fn get_object_attributes(
-        &self,
-        bucket_name: &str,
-        object_name: &str,
-        opts: ObjectAttributesOptions,
-    ) -> Result<ObjectAttributes, std::io::Error> {
-        let mut url_values = HashMap::new();
-        url_values.insert("attributes".to_string(), "".to_string());
-        if opts.version_id != "" {
-            url_values.insert("versionId".to_string(), opts.version_id);
-        }
-
-        let mut headers = HeaderMap::new();
-        headers.insert(
-            X_AMZ_OBJECT_ATTRIBUTES,
-            HeaderValue::from_str(GET_OBJECT_ATTRIBUTES_TAGS).expect("valid header value"),
-        );
-
-        if opts.part_number_marker > 0 {
-            headers.insert(
-                X_AMZ_PART_NUMBER_MARKER,
-                HeaderValue::from_str(&opts.part_number_marker.to_string()).expect("valid header value"),
-            );
-        }
-
-        if opts.max_parts > 0 {
-            headers.insert(
-                X_AMZ_MAX_PARTS,
-                HeaderValue::from_str(&opts.max_parts.to_string()).expect("valid header value"),
-            );
-        } else {
-            headers.insert(
-                X_AMZ_MAX_PARTS,
-                HeaderValue::from_str(&GET_OBJECT_ATTRIBUTES_MAX_PARTS.to_string()).expect("valid header value"),
-            );
-        }
-
-        /*if opts.server_side_encryption.is_some() {
-            opts.server_side_encryption.Marshal(headers);
-        }*/
-
-        let mut resp = self
-            .execute_method(
-                http::Method::HEAD,
-                &mut RequestMetadata {
-                    bucket_name: bucket_name.to_string(),
-                    object_name: object_name.to_string(),
-                    query_values: url_values,
-                    custom_header: headers,
-                    content_sha256_hex: EMPTY_STRING_SHA256_HASH.to_string(),
-                    content_md5_base64: "".to_string(),
-                    content_body: ReaderImpl::Body(Bytes::new()),
-                    content_length: 0,
-                    stream_sha256: false,
-                    trailer: HeaderMap::new(),
-                    pre_sign_url: Default::default(),
-                    add_crc: Default::default(),
-                    extra_pre_sign_header: Default::default(),
-                    bucket_location: Default::default(),
-                    expires: Default::default(),
-                },
-            )
-            .await?;
-
-        let resp_status = resp.status();
-        let h = resp.headers().clone();
-        let has_etag = h.get("ETag").and_then(|v| v.to_str().ok()).unwrap_or("");
-        if !has_etag.is_empty() {
-            return Err(std::io::Error::other(
-                "get_object_attributes is not supported by the current endpoint version",
-            ));
-        }
-
-        let mut body_vec = Vec::new();
-        let mut body = resp.into_body();
-        while let Some(frame) = body.frame().await {
-            let frame = frame.map_err(|e| std::io::Error::new(std::io::ErrorKind::Other, e.to_string()))?;
-            if let Some(data) = frame.data_ref() {
-                body_vec.extend_from_slice(data);
-            }
-        }
-
-        if resp_status != http::StatusCode::OK {
-            let err_body =
-                String::from_utf8(body_vec).map_err(|e| std::io::Error::other(format!("invalid UTF-8 error body: {e}")))?;
-            let mut er = match quick_xml::de::from_str::<AccessControlPolicy>(&err_body) {
-                Ok(result) => result,
-                Err(err) => {
-                    return Err(std::io::Error::other(err.to_string()));
-                }
-            };
-
-            return Err(std::io::Error::other(er.access_control_list.permission));
-        }
-
-        let mut oa = ObjectAttributes::new();
-        oa.parse_response(&h, body_vec).await?;
-
-        Ok(oa)
-    }
-}

crates/ecstore/src/client/api_get_object_file.rs

@@ -1,159 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use std::io;
-use std::path::{Path, PathBuf};
-
-#[cfg(not(windows))]
-use std::os::unix::fs::PermissionsExt;
-
-use tokio::fs::{self, OpenOptions};
-use tokio::io::{AsyncSeekExt, AsyncWriteExt, SeekFrom};
-
-use crate::client::{
-    api_error_response::err_invalid_argument, api_get_options::GetObjectOptions, transition_api::TransitionClient,
-};
-
-async fn prepare_download_target(file_path: &Path) -> io::Result<()> {
-    match fs::metadata(file_path).await {
-        Ok(metadata) if metadata.is_dir() => {
-            return Err(io::Error::other(err_invalid_argument("filename is a directory.")));
-        }
-        Ok(_) => {}
-        Err(err) if err.kind() == io::ErrorKind::NotFound => {}
-        Err(err) => return Err(err),
-    }
-
-    if let Some(parent) = file_path.parent()
-        && !parent.as_os_str().is_empty()
-    {
-        fs::create_dir_all(parent).await?;
-
-        #[cfg(not(windows))]
-        {
-            let mut permissions = fs::metadata(parent).await?.permissions();
-            permissions.set_mode(0o700);
-            fs::set_permissions(parent, permissions).await?;
-        }
-    }
-
-    Ok(())
-}
-
-fn build_part_path(file_path: &Path) -> PathBuf {
-    PathBuf::from(format!("{}.part.rustfs", file_path.display()))
-}
-
-async fn open_download_part_file(file_part_path: &Path) -> io::Result<tokio::fs::File> {
-    let mut options = OpenOptions::new();
-    options.create(true).truncate(false).read(true).write(true);
-
-    #[cfg(not(windows))]
-    options.mode(0o600);
-
-    options.open(file_part_path).await
-}
-
-async fn cleanup_part_file(file_part_path: &Path) {
-    let _ = fs::remove_file(file_part_path).await;
-}
-
-impl TransitionClient {
-    pub async fn fget_object(
-        &self,
-        bucket_name: &str,
-        object_name: &str,
-        file_path: &str,
-        mut opts: GetObjectOptions,
-    ) -> Result<(), io::Error> {
-        let file_path = Path::new(file_path);
-        prepare_download_target(file_path).await?;
-
-        let file_part_path = build_part_path(file_path);
-        let mut file_part = open_download_part_file(&file_part_path).await?;
-        let existing_len = file_part.metadata().await?.len();
-        if existing_len > 0 {
-            opts.set_range(existing_len as i64, 0)?;
-            file_part.seek(SeekFrom::Start(existing_len)).await?;
-        }
-
-        let (_object_info, _headers, mut object_reader) = self.get_object_inner(bucket_name, object_name, &opts).await?;
-        if let Err(err) = tokio::io::copy(&mut object_reader, &mut file_part).await {
-            cleanup_part_file(&file_part_path).await;
-            return Err(err);
-        }
-
-        if let Err(err) = file_part.flush().await {
-            cleanup_part_file(&file_part_path).await;
-            return Err(err);
-        }
-        drop(file_part);
-
-        if let Err(err) = fs::rename(&file_part_path, file_path).await {
-            cleanup_part_file(&file_part_path).await;
-            return Err(err);
-        }
-
-        Ok(())
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use tempfile::tempdir;
-
-    #[tokio::test]
-    async fn prepare_download_target_allows_missing_file_and_creates_parent_dirs() {
-        let dir = tempdir().expect("temp dir");
-        let target = dir.path().join("nested").join("object.bin");
-
-        prepare_download_target(&target)
-            .await
-            .expect("missing target should be accepted");
-
-        assert!(target.parent().expect("parent").exists(), "parent directory should be created");
-        assert!(
-            fs::metadata(&target).await.is_err(),
-            "preparing the target should not create the final file eagerly"
-        );
-    }
-
-    #[tokio::test]
-    async fn prepare_download_target_rejects_directory_paths() {
-        let dir = tempdir().expect("temp dir");
-        let target_dir = dir.path().join("download-dir");
-        fs::create_dir_all(&target_dir).await.expect("target dir");
-
-        let err = prepare_download_target(&target_dir)
-            .await
-            .expect_err("directory targets must be rejected");
-
-        assert!(err.to_string().contains("directory"), "unexpected error for directory target: {err}");
-    }
-
-    #[tokio::test]
-    async fn open_download_part_file_creates_part_file() {
-        let dir = tempdir().expect("temp dir");
-        let target = dir.path().join("object.bin");
-        let part_path = build_part_path(&target);
-
-        let file = open_download_part_file(&part_path)
-            .await
-            .expect("part file should be created");
-        drop(file);
-
-        assert!(part_path.exists(), "part file should exist after creation");
-    }
-}

crates/ecstore/src/client/api_restore.rs

@@ -1,134 +0,0 @@
-// Copyright 2024 RustFS Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-#![allow(unused_imports)]
-#![allow(unused_variables)]
-#![allow(unused_mut)]
-#![allow(unused_assignments)]
-#![allow(unused_must_use)]
-#![allow(clippy::all)]
-
-use crate::client::{
-    api_error_response::{err_invalid_argument, http_resp_to_error_response},
-    api_get_object_acl::AccessControlList,
-    api_get_options::GetObjectOptions,
-    transition_api::{ObjectInfo, ReadCloser, ReaderImpl, RequestMetadata, TransitionClient, to_object_info},
-};
-use http::HeaderMap;
-use http_body_util::BodyExt;
-use hyper::body::Body;
-use hyper::body::Bytes;
-use s3s::dto::RestoreRequest;
-use std::collections::HashMap;
-use std::io::Cursor;
-use tokio::io::BufReader;
-
-const TIER_STANDARD: &str = "Standard";
-const TIER_BULK: &str = "Bulk";
-const TIER_EXPEDITED: &str = "Expedited";
-
-#[derive(Debug, Default, serde::Serialize, serde::Deserialize)]
-pub struct Encryption {
-    pub encryption_type: String,
-    pub kms_context: String,
-    pub kms_key_id: String,
-}
-
-#[derive(Debug, Default, serde::Serialize, serde::Deserialize)]
-pub struct MetadataEntry {
-    pub name: String,
-    pub value: String,
-}
-
-#[derive(Debug, Default, serde::Serialize)]
-pub struct S3 {
-    pub access_control_list: AccessControlList,
-    pub bucket_name: String,
-    pub prefix: String,
-    pub canned_acl: String,
-    pub encryption: Encryption,
-    pub storage_class: String,
-    //tagging:            Tags,
-    pub user_metadata: MetadataEntry,
-}
-
-impl TransitionClient {
-    pub async fn restore_object(
-        &self,
-        bucket_name: &str,
-        object_name: &str,
-        version_id: &str,
-        restore_req: &RestoreRequest,
-    ) -> Result<(), std::io::Error> {
-        /*let restore_request = match quick_xml::se::to_string(restore_req) {
-            Ok(buf) => buf,
-            Err(e) => {
-                return Err(std::io::Error::other(e));
-            }
-        };*/
-        let restore_request = "".to_string();
-        let restore_request_bytes = restore_request.as_bytes().to_vec();
-
-        let mut url_values = HashMap::new();
-        url_values.insert("restore".to_string(), "".to_string());
-        if version_id != "" {
-            url_values.insert("versionId".to_string(), version_id.to_string());
-        }
-
-        let restore_request_buffer = Bytes::from(restore_request_bytes.clone());
-        let resp = self
-            .execute_method(
-                http::Method::HEAD,
-                &mut RequestMetadata {
-                    bucket_name: bucket_name.to_string(),
-                    object_name: object_name.to_string(),
-                    query_values: url_values,
-                    custom_header: HeaderMap::new(),
-                    content_sha256_hex: "".to_string(), //sum_sha256_hex(&restore_request_bytes),
-                    content_md5_base64: "".to_string(), //sum_md5_base64(&restore_request_bytes),
-                    content_body: ReaderImpl::Body(restore_request_buffer),
-                    content_length: restore_request_bytes.len() as i64,
-                    stream_sha256: false,
-                    trailer: HeaderMap::new(),
-                    pre_sign_url: Default::default(),
-                    add_crc: Default::default(),
-                    extra_pre_sign_header: Default::default(),
-                    bucket_location: Default::default(),
-                    expires: Default::default(),
-                },
-            )
-            .await?;
-
-        let resp_status = resp.status();
-        let h = resp.headers().clone();
-
-        let mut body_vec = Vec::new();
-        let mut body = resp.into_body();
-        while let Some(frame) = body.frame().await {
-            let frame = frame.map_err(|e| std::io::Error::new(std::io::ErrorKind::Other, e.to_string()))?;
-            if let Some(data) = frame.data_ref() {
-                body_vec.extend_from_slice(data);
-            }
-        }
-        if resp_status != http::StatusCode::ACCEPTED && resp_status != http::StatusCode::OK {
-            return Err(std::io::Error::other(http_resp_to_error_response(
-                resp_status,
-                &h,
-                body_vec,
-                bucket_name,
-                "",
-            )));
-        }
-        Ok(())
-    }
-}

crates/ecstore/src/client/checksum.rs

@@ -27,12 +27,24 @@ use crate::client::utils::base64_decode;
 use crate::client::utils::base64_encode;
 use crate::client::{api_put_object::PutObjectOptions, api_s3_datatypes::ObjectPart};
 use crate::{disk::DiskAPI, object_api::GetObjectReader};
+// s3s::header has no CRC64NVME constant yet; the canonical RustFS copy lives
+// in rustfs-utils' headers module.
+use rustfs_utils::http::headers::AMZ_CHECKSUM_CRC64NVME;
 use s3s::header::{
     X_AMZ_CHECKSUM_ALGORITHM, X_AMZ_CHECKSUM_CRC32, X_AMZ_CHECKSUM_CRC32C, X_AMZ_CHECKSUM_SHA1, X_AMZ_CHECKSUM_SHA256,
 };
 
 use enumset::{EnumSet, EnumSetType, enum_set};
 
+/// One of three deliberately separate checksum registries (backlog#1833):
+/// this enum is the MinIO-port client's wire vocabulary and stops at the
+/// standard S3 set (CRC64NVME is its newest member; the RustFS extensions do
+/// not exist on this client path). The streaming-hash registry lives in
+/// `rustfs_checksums::ChecksumAlgorithm` (crates/checksums/src/lib.rs) and
+/// the on-disk xl.meta bitset in `rustfs_rio::ChecksumType`
+/// (crates/rio/src/checksum.rs, varint bits are append-only). When adding an
+/// algorithm, extend all three (or record why not) — they do not derive from
+/// each other.
 #[derive(Debug, EnumSetType, Default)]
 #[enumset(repr = "u8")]
 pub enum ChecksumMode {
@@ -57,8 +69,6 @@ lazy_static! {
     static ref C_ChecksumFullObjectCRC32C: EnumSet<ChecksumMode> =
         enum_set!(ChecksumMode::ChecksumCRC32C | ChecksumMode::ChecksumFullObject);
 }
-const AMZ_CHECKSUM_CRC64NVME: &str = "x-amz-checksum-crc64nvme";
-
 impl ChecksumMode {
     //pub const CRC64_NVME_POLYNOMIAL: i64 = 0xad93d23594c93659;
 

crates/ecstore/src/client/constants.rs

@@ -37,6 +37,3 @@ pub const TOTAL_WORKERS: i64 = 4;
 pub const SIGN_V4_ALGORITHM: &str = "AWS4-HMAC-SHA256";
 pub const ISO8601_DATEFORMAT: &[FormatItem<'_>] =
     format_description!("[year]-[month]-[day]T[hour]:[minute]:[second].[subsecond]Z");
-
-pub const GET_OBJECT_ATTRIBUTES_TAGS: &str = "ETag,Checksum,StorageClass,ObjectSize,ObjectParts";
-pub const GET_OBJECT_ATTRIBUTES_MAX_PARTS: i64 = 1000;

crates/ecstore/src/client/mod.rs

@@ -16,12 +16,8 @@
 #![allow(dead_code)]
 
 pub mod admin_handler_utils;
-pub mod api_bucket_policy;
 pub mod api_error_response;
 pub mod api_get_object;
-pub mod api_get_object_acl;
-pub mod api_get_object_attributes;
-pub mod api_get_object_file;
 pub mod api_get_options;
 pub mod api_list;
 pub mod api_put_object;
@@ -29,7 +25,6 @@ pub mod api_put_object_common;
 pub mod api_put_object_multipart;
 pub mod api_put_object_streaming;
 pub mod api_remove;
-pub mod api_restore;
 pub mod api_s3_datatypes;
 pub mod api_stat;
 pub mod bucket_cache;

crates/ecstore/src/client/transition_api.rs

@@ -1006,16 +1006,6 @@ impl TransitionCore {
         client.abort_multipart_upload(bucket_name, object, upload_id).await
     }
 
-    pub async fn get_bucket_policy(&self, bucket_name: &str) -> Result<String, std::io::Error> {
-        let client = self.0.clone();
-        client.get_bucket_policy(bucket_name).await
-    }
-
-    pub async fn put_bucket_policy(&self, bucket_name: &str, bucket_policy: &str) -> Result<(), std::io::Error> {
-        let client = self.0.clone();
-        client.put_bucket_policy(bucket_name, bucket_policy).await
-    }
-
     pub async fn get_object(
         &self,
         bucket_name: &str,

crates/ecstore/src/diagnostics/admin_server_info.rs

@@ -12,7 +12,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use crate::cluster::rpc::{TonicInterceptor, gen_tonic_signature_interceptor, node_service_time_out_client};
+use crate::cluster::rpc::{
+    ScannerBucketListing, TonicInterceptor, gen_tonic_signature_interceptor, node_service_time_out_client,
+};
 use crate::data_usage::{DATA_USAGE_CACHE_NAME, DATA_USAGE_ROOT, load_data_usage_from_backend_cached};
 use crate::error::{Error, Result};
 use crate::{
@@ -23,6 +25,7 @@ use crate::{
 
 use crate::data_usage::load_data_usage_cache;
 use crate::storage_api_contracts::admin::StorageAdminApi;
+use crate::storage_api_contracts::bucket::BucketOptions;
 use rustfs_common::heal_channel::DriveState;
 use rustfs_madmin::{
     BackendDisks, Disk, ErasureSetInfo, ITEM_INITIALIZING, ITEM_OFFLINE, ITEM_ONLINE, ITEM_UNKNOWN, InfoMessage, MemStats,
@@ -74,6 +77,19 @@ fn apply_data_usage_result(
     }
 }
 
+fn apply_bucket_namespace_count(result: Result<ScannerBucketListing>, buckets: &mut rustfs_madmin::Buckets) {
+    if let Ok(listing) = result
+        && listing.topology_complete
+    {
+        let count = listing.buckets.iter().filter(|bucket| !bucket.name.starts_with('.')).count();
+        let Ok(count) = u64::try_from(count) else {
+            return;
+        };
+        buckets.count = count;
+        buckets.error = None;
+    }
+}
+
 // pub const ITEM_OFFLINE: &str = "offline";
 // pub const ITEM_INITIALIZING: &str = "initializing";
 // pub const ITEM_ONLINE: &str = "online";
@@ -285,6 +301,18 @@ pub async fn get_server_info(get_pools: bool) -> InfoMessage {
             &mut delete_markers,
             &mut usage,
         );
+        if buckets.error.is_some() {
+            apply_bucket_namespace_count(
+                store
+                    .list_bucket_for_scanner(&BucketOptions {
+                        cached: true,
+                        no_metadata: true,
+                        ..Default::default()
+                    })
+                    .await,
+                &mut buckets,
+            );
+        }
 
         let after3 = OffsetDateTime::now_utc();
 
@@ -705,12 +733,13 @@ mod tests {
         endpoints::{EndpointServerPools, Endpoints, PoolEndpoints},
     };
     use crate::runtime::sources as runtime_sources;
+    use crate::storage_api_contracts::bucket::BucketInfo;
     use rustfs_madmin::{Disk, ITEM_OFFLINE, ITEM_ONLINE, ITEM_UNKNOWN, ServerProperties};
 
     use super::{
-        DATA_USAGE_ROOT, DATA_USAGE_UNAVAILABLE_ERROR, apply_data_usage_result, apply_erasure_set_usage,
-        get_local_server_property, get_online_offline_disks_stats, get_server_info, reconcile_servers_with_endpoint_topology,
-        server_topology_completeness_report,
+        DATA_USAGE_ROOT, DATA_USAGE_UNAVAILABLE_ERROR, apply_bucket_namespace_count, apply_data_usage_result,
+        apply_erasure_set_usage, get_local_server_property, get_online_offline_disks_stats, get_server_info,
+        reconcile_servers_with_endpoint_topology, server_topology_completeness_report,
     };
 
     fn disk_with_state(endpoint: &str, state: &str) -> Disk {
@@ -960,6 +989,75 @@ mod tests {
         assert_eq!(usage.error.as_deref(), Some(DATA_USAGE_UNAVAILABLE_ERROR));
     }
 
+    #[test]
+    fn live_bucket_namespace_count_survives_unavailable_data_usage() {
+        let mut buckets = rustfs_madmin::Buckets {
+            count: 0,
+            error: Some(DATA_USAGE_UNAVAILABLE_ERROR.to_string()),
+        };
+
+        apply_bucket_namespace_count(
+            Ok(crate::cluster::rpc::ScannerBucketListing {
+                buckets: vec![
+                    BucketInfo {
+                        name: "bucket-a".to_string(),
+                        ..Default::default()
+                    },
+                    BucketInfo {
+                        name: ".rustfs.sys".to_string(),
+                        ..Default::default()
+                    },
+                    BucketInfo {
+                        name: "bucket-b".to_string(),
+                        ..Default::default()
+                    },
+                ],
+                set_buckets: Vec::new(),
+                topology_complete: true,
+            }),
+            &mut buckets,
+        );
+
+        assert_eq!(buckets.count, 2);
+        assert_eq!(buckets.error, None);
+    }
+
+    #[test]
+    fn incomplete_bucket_namespace_lookup_preserves_usage_state() {
+        let mut buckets = rustfs_madmin::Buckets {
+            count: 7,
+            error: Some(DATA_USAGE_UNAVAILABLE_ERROR.to_string()),
+        };
+
+        apply_bucket_namespace_count(
+            Ok(crate::cluster::rpc::ScannerBucketListing {
+                buckets: vec![BucketInfo {
+                    name: "bucket-a".to_string(),
+                    ..Default::default()
+                }],
+                set_buckets: Vec::new(),
+                topology_complete: false,
+            }),
+            &mut buckets,
+        );
+
+        assert_eq!(buckets.count, 7);
+        assert_eq!(buckets.error.as_deref(), Some(DATA_USAGE_UNAVAILABLE_ERROR));
+    }
+
+    #[test]
+    fn failed_bucket_namespace_lookup_preserves_usage_state() {
+        let mut buckets = rustfs_madmin::Buckets {
+            count: 7,
+            error: Some(DATA_USAGE_UNAVAILABLE_ERROR.to_string()),
+        };
+
+        apply_bucket_namespace_count(Err(crate::error::Error::DiskNotFound), &mut buckets);
+
+        assert_eq!(buckets.count, 7);
+        assert_eq!(buckets.error.as_deref(), Some(DATA_USAGE_UNAVAILABLE_ERROR));
+    }
+
     #[test]
     fn incomplete_erasure_set_cache_is_not_reported_as_zero() {
         let mut cache = rustfs_data_usage::DataUsageCache::default();

crates/ecstore/src/disk/error.rs

@@ -113,6 +113,9 @@ pub enum DiskError {
     #[error("bit-rot hash algorithm is invalid")]
     BitrotHashAlgoInvalid,
 
+    /// Never constructed locally by RustFS (only reachable through wire
+    /// decoding, and no current node sends it). The wire code is kept for
+    /// cross-version compatibility — do not renumber or remove (backlog#1831).
     #[error("Rename across devices not allowed, please fix your backend configuration")]
     CrossDeviceLink,
 
@@ -143,6 +146,9 @@ pub enum DiskError {
     #[error("io error {0}")]
     Io(#[source] io::Error),
 
+    /// Never constructed locally by RustFS (only reachable through wire
+    /// decoding, and no current node sends it). The wire code is kept for
+    /// cross-version compatibility — do not renumber or remove (backlog#1831).
     #[error("source stalled")]
     SourceStalled,
 
@@ -642,19 +648,6 @@ impl Hash for DiskError {
 // is currently commented out to avoid complexity. These can be re-enabled
 // when needed for specific disk quorum checking and error aggregation logic.
 
-/// Bitrot errors
-#[derive(Debug, thiserror::Error)]
-pub enum BitrotErrorType {
-    #[error("bitrot checksum verification failed")]
-    BitrotChecksumMismatch { expected: String, got: String },
-}
-
-impl From<BitrotErrorType> for DiskError {
-    fn from(e: BitrotErrorType) -> Self {
-        DiskError::other(e)
-    }
-}
-
 /// Context wrapper for file access errors
 #[derive(Debug, thiserror::Error)]
 pub struct FileAccessDeniedWithContext {
@@ -869,19 +862,6 @@ mod tests {
         let _disk_error: DiskError = json_error.into();
     }
 
-    #[test]
-    fn test_bitrot_error_type() {
-        let bitrot_error = BitrotErrorType::BitrotChecksumMismatch {
-            expected: "abc123".to_string(),
-            got: "def456".to_string(),
-        };
-
-        assert!(bitrot_error.to_string().contains("bitrot checksum verification failed"));
-
-        let disk_error: DiskError = bitrot_error.into();
-        assert!(matches!(disk_error, DiskError::Io(_)));
-    }
-
     #[test]
     fn test_file_access_denied_with_context() {
         let path = PathBuf::from("/test/path");

crates/ecstore/src/erasure/coding/bitrot.rs

@@ -18,7 +18,11 @@ use std::io::IoSlice;
 use std::time::Duration;
 use tokio::io::{AsyncRead, AsyncReadExt, AsyncWrite, AsyncWriteExt};
 use tracing::error;
-use uuid::Uuid;
+
+const LOG_COMPONENT_ECSTORE: &str = "ecstore";
+const LOG_SUBSYSTEM_ERASURE: &str = "erasure";
+const EVENT_BITROT_SHORT_SHARD_READ: &str = "bitrot_short_shard_read";
+const EVENT_BITROT_HASH_MISMATCH: &str = "bitrot_hash_mismatch";
 
 /// A shard source that may already hold its bytes in memory.
 ///
@@ -73,7 +77,6 @@ pin_project! {
         buf: Vec<u8>,
         skip_verify: bool,
         last_verify_duration: Duration,
-        id: Uuid,
     }
 }
 
@@ -90,7 +93,6 @@ where
             buf: Vec::new(),
             skip_verify,
             last_verify_duration: Duration::ZERO,
-            id: Uuid::new_v4(),
         }
     }
 
@@ -118,7 +120,7 @@ where
 
         let need = self.hash_algo.size() + want;
         self.read_scratch_block(need, want).await?;
-        let (data, verify) = split_and_verify(&self.hash_algo, self.skip_verify, &self.buf[..need], &self.id)?;
+        let (data, verify) = split_and_verify(&self.hash_algo, self.skip_verify, &self.buf[..need])?;
         out.copy_from_slice(data);
         self.last_verify_duration = verify;
         Ok(want)
@@ -157,7 +159,7 @@ where
         }
         let filled = fill(&mut self.inner, &mut self.buf[..need]).await?;
         if filled < need {
-            return Err(short_shard_read(&self.id, filled.saturating_sub(self.hash_algo.size()), want));
+            return Err(short_shard_read(filled.saturating_sub(self.hash_algo.size()), want));
         }
         Ok(())
     }
@@ -166,15 +168,23 @@ where
     /// buffer returns its length, a short read is UnexpectedEof (backlog#799 B2).
     fn finish_len(&self, data_len: usize, want: usize) -> std::io::Result<usize> {
         if data_len < want {
-            return Err(short_shard_read(&self.id, data_len, want));
+            return Err(short_shard_read(data_len, want));
         }
         Ok(data_len)
     }
 }
 
 /// A truncated shard is `UnexpectedEof`, not a short success (backlog#799 B2).
-fn short_shard_read(id: &Uuid, got: usize, want: usize) -> std::io::Error {
-    error!("bitrot reader short shard read: id={id} got {got} of {want} bytes");
+fn short_shard_read(got: usize, want: usize) -> std::io::Error {
+    error!(
+        event = EVENT_BITROT_SHORT_SHARD_READ,
+        component = LOG_COMPONENT_ECSTORE,
+        subsystem = LOG_SUBSYSTEM_ERASURE,
+        state = "failed",
+        got,
+        want,
+        "short shard read: got {got} of {want} bytes"
+    );
     std::io::Error::new(std::io::ErrorKind::UnexpectedEof, format!("short shard read: got {got} of {want} bytes"))
 }
 
@@ -184,12 +194,7 @@ fn short_shard_read(id: &Uuid, got: usize, want: usize) -> std::io::Error {
 /// hash never reaches the caller's buffer. The verify duration is returned
 /// rather than stored so this stays a free function usable while `self` is
 /// borrowed for the block.
-fn split_and_verify<'a>(
-    hash_algo: &HashAlgorithm,
-    skip_verify: bool,
-    block: &'a [u8],
-    id: &Uuid,
-) -> std::io::Result<(&'a [u8], Duration)> {
+fn split_and_verify<'a>(hash_algo: &HashAlgorithm, skip_verify: bool, block: &'a [u8]) -> std::io::Result<(&'a [u8], Duration)> {
     let (hash, data) = block.split_at(hash_algo.size());
     if skip_verify {
         return Ok((data, Duration::ZERO));
@@ -198,7 +203,14 @@ fn split_and_verify<'a>(
     let actual_hash = hash_algo.hash_encode(data);
     let verify = verify_start.elapsed();
     if actual_hash.as_ref() != hash {
-        error!("bitrot reader hash mismatch, id={id} data_len={}", data.len());
+        error!(
+            event = EVENT_BITROT_HASH_MISMATCH,
+            component = LOG_COMPONENT_ECSTORE,
+            subsystem = LOG_SUBSYSTEM_ERASURE,
+            state = "failed",
+            data_len = data.len(),
+            "bitrot hash mismatch"
+        );
         return Err(std::io::Error::new(std::io::ErrorKind::InvalidData, "bitrot hash mismatch"));
     }
     Ok((data, verify))
@@ -254,7 +266,7 @@ where
         // `need` bytes returns `None` and falls through to the scratch path,
         // keeping the short-read contract.
         if let Some(block) = self.inner.try_take_block(need) {
-            let (data, verify) = split_and_verify(&self.hash_algo, self.skip_verify, &block, &self.id)?;
+            let (data, verify) = split_and_verify(&self.hash_algo, self.skip_verify, &block)?;
             out.extend_from_slice(data);
             self.last_verify_duration = verify;
             return Ok(want);
@@ -264,7 +276,7 @@ where
         // the sink differs (`extend_from_slice` into `out` instead of
         // `copy_from_slice` into a pre-zeroed buffer).
         self.read_scratch_block(need, want).await?;
-        let (data, verify) = split_and_verify(&self.hash_algo, self.skip_verify, &self.buf[..need], &self.id)?;
+        let (data, verify) = split_and_verify(&self.hash_algo, self.skip_verify, &self.buf[..need])?;
         out.extend_from_slice(data);
         self.last_verify_duration = verify;
         Ok(want)

crates/ecstore/src/erasure/coding/decode.rs

@@ -29,6 +29,7 @@ use crate::set_disk::shard_source::{ShardReadCost, ShardStripeSource, StripeRead
 use futures::FutureExt;
 use futures::stream::{FuturesUnordered, StreamExt};
 use pin_project_lite::pin_project;
+use smallvec::{SmallVec, smallvec};
 use std::future::Future;
 use std::io;
 use std::io::ErrorKind;
@@ -40,9 +41,15 @@ use tracing::{debug, error, warn};
 
 type ShardReadFuture<'a> = Pin<Box<dyn Future<Output = (usize, ShardReadCost, Result<Vec<u8>, Error>, bool)> + Send + 'a>>;
 
+const INLINE_SHARD_SLOTS: usize = 32;
+type ShardBuffers = SmallVec<[Option<Vec<u8>>; INLINE_SHARD_SLOTS]>;
+type ShardErrors = SmallVec<[Option<Error>; INLINE_SHARD_SLOTS]>;
+type ShardIndexes = SmallVec<[usize; INLINE_SHARD_SLOTS]>;
+type ActiveReaders = SmallVec<[bool; INLINE_SHARD_SLOTS]>;
+
 /// One stripe's worth of shard buffers plus the per-shard read errors, as
 /// returned by `ParallelReader::read` / `read_stripe_timed`.
-type StripeReadOutput = (Vec<Option<Vec<u8>>>, Vec<Option<Error>>);
+type StripeReadOutput = (ShardBuffers, ShardErrors);
 
 const ENV_RUSTFS_SHARD_LOCALITY_SCHEDULING: &str = "RUSTFS_SHARD_LOCALITY_SCHEDULING";
 const ENV_RUSTFS_GET_SHARD_LOCALITY_PREFERENCE_ENABLE: &str = "RUSTFS_GET_SHARD_LOCALITY_PREFERENCE_ENABLE";
@@ -390,7 +397,7 @@ pub(crate) struct ParallelReader<R> {
     // start, parity slots only once a data shard is missing/dead. Unengaged
     // parity stays an unopened deferred reader; `deferred_handles[i]` realigns
     // it to the current stripe when it is engaged mid-object (backlog#923).
-    engaged: Vec<bool>,
+    engaged: SmallVec<[bool; INLINE_SHARD_SLOTS]>,
     deferred_handles: Vec<Option<DeferredReaderStripeHandle>>,
     stripe_index: usize,
 }
@@ -573,7 +580,7 @@ where
         // behavior. With the gate on, only data slots start engaged; parity is
         // engaged on demand, stripe-aligned through its deferred handle.
         let data_shards_only = get_lockstep_data_shards_only_enabled();
-        let engaged = (0..readers.len())
+        let engaged: SmallVec<_> = (0..readers.len())
             .map(|index| !data_shards_only || index < e.data_shards)
             .collect();
         ParallelReader {
@@ -612,7 +619,7 @@ where
 fn record_shard_read_result(
     shards: &mut [Option<Vec<u8>>],
     errs: &mut [Option<Error>],
-    retire_readers: &mut Vec<usize>,
+    retire_readers: &mut ShardIndexes,
     success: &mut usize,
     successful_costs: &mut ShardReadCostCounts,
     i: usize,
@@ -637,7 +644,7 @@ fn record_shard_read_result(
     }
 }
 
-fn retire_abandoned_readers(errs: &mut [Option<Error>], retire_readers: &mut Vec<usize>, active_readers: &[bool]) {
+fn retire_abandoned_readers(errs: &mut [Option<Error>], retire_readers: &mut ShardIndexes, active_readers: &[bool]) {
     for (i, active) in active_readers.iter().enumerate() {
         if !*active {
             continue;
@@ -692,7 +699,7 @@ where
     R: crate::erasure::coding::ShardSource,
 {
     #[hotpath::measure(impl_type = "ParallelReader")]
-    pub async fn read(&mut self) -> (Vec<Option<Vec<u8>>>, Vec<Option<Error>>) {
+    pub async fn read(&mut self) -> StripeReadOutput {
         // On the reconstruction-verifying GET path, read every live shard reader
         // in lockstep so all readers advance one block per stripe and stay
         // mutually aligned. The adaptive data-first path below only reads
@@ -716,7 +723,7 @@ where
         };
 
         if shard_size == 0 {
-            return (vec![None; num_readers], vec![None; num_readers]);
+            return (smallvec![None; num_readers], smallvec![None; num_readers]);
         }
 
         // Advance to the next stripe so the following read() computes the correct
@@ -727,8 +734,8 @@ where
         // is only read above to derive `shard_size`, so advancing here is safe.
         self.offset += shard_size;
 
-        let mut shards: Vec<Option<Vec<u8>>> = vec![None; num_readers];
-        let mut errs = vec![None; num_readers];
+        let mut shards: ShardBuffers = smallvec![None; num_readers];
+        let mut errs: ShardErrors = smallvec![None; num_readers];
         let read_costs = self.read_costs.as_slice();
         let locality_preference_enabled = self.locality_preference_enabled;
         let low_cost_available = self
@@ -759,11 +766,11 @@ where
 
         self.buffers.ensure_slots(num_readers);
 
-        let mut retire_readers = Vec::new();
+        let mut retire_readers = ShardIndexes::new();
         if num_readers >= self.data_shards {
             let mut reader_iter = ReaderLaunchIter::new(&mut self.readers, read_costs, locality_preference_enabled);
             let mut sets = FuturesUnordered::new();
-            let mut active_readers = vec![false; num_readers];
+            let mut active_readers: ActiveReaders = smallvec![false; num_readers];
             let stripe_read_start = self.metrics_path.map(|_| Instant::now());
             let mut scheduled = 0usize;
             for _ in 0..self.data_shards {
@@ -1023,7 +1030,7 @@ where
     /// stripe would reintroduce the desync. A parity reader that cannot be
     /// realigned (no pending deferred handle) is likewise retired instead of
     /// being read out of position.
-    async fn read_lockstep(&mut self) -> (Vec<Option<Vec<u8>>>, Vec<Option<Error>>) {
+    async fn read_lockstep(&mut self) -> StripeReadOutput {
         let num_readers = self.readers.len();
         let shard_size = if self.offset + self.shard_size > self.shard_file_size {
             self.shard_file_size - self.offset
@@ -1031,8 +1038,8 @@ where
             self.shard_size
         };
 
-        let mut shards: Vec<Option<Vec<u8>>> = vec![None; num_readers];
-        let mut errs: Vec<Option<Error>> = vec![None; num_readers];
+        let mut shards: ShardBuffers = smallvec![None; num_readers];
+        let mut errs: ShardErrors = smallvec![None; num_readers];
         if shard_size == 0 {
             return (shards, errs);
         }
@@ -1071,7 +1078,7 @@ where
         // Pre-claim per-slot buffers so the `self.readers` borrow below stays
         // disjoint from `self.buffers`; `Some(buffer)` also records which slots
         // participate, avoiding a per-stripe sidecar allocation.
-        let mut bufs: Vec<Option<Vec<u8>>> = Vec::with_capacity(num_readers);
+        let mut bufs: ShardBuffers = SmallVec::with_capacity(num_readers);
         for i in 0..num_readers {
             bufs.push(if self.engaged[i] && self.readers[i].is_some() {
                 Some(self.buffers.take(i, shard_size))
@@ -1086,7 +1093,7 @@ where
         let locality_preference_enabled = self.locality_preference_enabled;
         let stripe_read_start = metrics_path.map(|_| Instant::now());
 
-        let mut retire_readers = Vec::new();
+        let mut retire_readers = ShardIndexes::new();
         let mut scheduled = 0usize;
         let mut success = 0usize;
         let mut completed = 0usize;
@@ -1351,10 +1358,7 @@ fn get_data_block_len(shards: &[Option<Vec<u8>>], data_blocks: usize) -> usize {
 /// stripe-read stage timer. Factored out so the depth-1 prefetch loop and the
 /// serial loop time reads identically. A free `async fn` (rather than a closure)
 /// so the returned future's borrow of `reader` is correctly tied to the call.
-async fn read_stripe_timed<R>(
-    reader: &mut ParallelReader<R>,
-    stage_metrics_enabled: bool,
-) -> (Vec<Option<Vec<u8>>>, Vec<Option<Error>>)
+async fn read_stripe_timed<R>(reader: &mut ParallelReader<R>, stage_metrics_enabled: bool) -> StripeReadOutput
 where
     R: crate::erasure::coding::ShardSource,
 {
@@ -1967,6 +1971,32 @@ mod tests {
 
     type BoxedShardReader = crate::io_support::bitrot::ShardReader;
 
+    #[test]
+    fn shard_scratch_stays_inline_through_the_common_limit_and_spills_safely() {
+        let inline: ShardBuffers = smallvec![None; INLINE_SHARD_SLOTS];
+        assert!(!inline.spilled(), "the common shard-count boundary must not allocate");
+
+        let spilled: ShardBuffers = smallvec![None; INLINE_SHARD_SLOTS + 1];
+        assert!(spilled.spilled(), "larger supported shard counts must fall back to the heap");
+        assert_eq!(spilled.len(), INLINE_SHARD_SLOTS + 1);
+    }
+
+    #[tokio::test]
+    async fn parallel_reader_preserves_slot_count_above_inline_capacity() {
+        const DATA_SHARDS: usize = INLINE_SHARD_SLOTS;
+        const TOTAL_SHARDS: usize = INLINE_SHARD_SLOTS + 1;
+        let readers = std::iter::repeat_with(|| None).take(TOTAL_SHARDS).collect();
+        let erasure = Erasure::new(DATA_SHARDS, 1, DATA_SHARDS);
+        let mut reader: ParallelReader<Cursor<Vec<u8>>> = ParallelReader::new(readers, erasure, 0, DATA_SHARDS);
+
+        let (shards, errors) = reader.read().await;
+
+        assert!(shards.spilled());
+        assert!(errors.spilled());
+        assert_eq!(shards.len(), TOTAL_SHARDS);
+        assert_eq!(errors.len(), TOTAL_SHARDS);
+    }
+
     /// Counts the raw bytes pulled from a shard stream, to prove which shards
     /// a decode path actually touches (backlog#923 call-count evidence).
     struct CountingShardReader {
@@ -2343,6 +2373,19 @@ mod tests {
         assert_eq!(err.expect("range beyond total length should fail").kind(), ErrorKind::InvalidInput);
     }
 
+    #[tokio::test]
+    async fn test_erasure_decode_zero_length_does_not_read_or_emit() {
+        let erasure = Erasure::new(2, 1, 64);
+        let readers: Vec<Option<BitrotReader<Cursor<Vec<u8>>>>> = vec![None, None, None];
+        let mut output = Vec::new();
+
+        let (written, err) = erasure.decode(&mut output, readers, 0, 0, 0).await;
+
+        assert_eq!(written, 0);
+        assert!(err.is_none());
+        assert!(output.is_empty());
+    }
+
     #[tokio::test]
     async fn test_erasure_decode_with_read_costs_restores_missing_data_shard_range() {
         const DATA_SHARDS: usize = 2;

crates/ecstore/src/erasure/coding/encode.rs

@@ -91,6 +91,11 @@ fn use_bytesmut_ingest() -> bool {
     })
 }
 
+fn small_ingest_capacity(erasure: &Erasure, size_hint: usize) -> usize {
+    let data_len = size_hint.min(erasure.block_size);
+    erasure.encoded_capacity_for_data_len(data_len).min(erasure.block_size)
+}
+
 /// Keeps the encoder producer scoped to its parent future. Tokio detaches a
 /// task when its `JoinHandle` is dropped, so the producer must be aborted when
 /// an upload is cancelled before the encode pipeline finishes.
@@ -540,13 +545,14 @@ impl Erasure {
         writers: &mut [Option<BitrotWriterWrapper>],
         quorum: usize,
         require_single_block: bool,
+        size_hint: usize,
     ) -> std::io::Result<(R, usize)>
     where
         R: AsyncRead + Send + Sync + Unpin,
     {
         use tokio::io::AsyncReadExt;
 
-        let mut buf = Vec::with_capacity(self.block_size);
+        let mut buf = Vec::with_capacity(small_ingest_capacity(&self, size_hint));
         let total = if require_single_block {
             let read_limit = self
                 .block_size
@@ -880,7 +886,24 @@ impl Erasure {
     where
         R: AsyncRead + Send + Sync + Unpin,
     {
-        self.encode_small_direct(reader, writers, quorum, false).await
+        let size_hint = self.block_size;
+        self.encode_small_direct(reader, writers, quorum, false, size_hint).await
+    }
+
+    /// Size-aware inline fast path. `size_hint` only controls the bounded initial
+    /// allocation; reads remain authoritative.
+    #[hotpath::measure(impl_type = "Erasure")]
+    pub async fn encode_inline_small_with_size_hint<R>(
+        self: Arc<Self>,
+        reader: R,
+        writers: &mut [Option<BitrotWriterWrapper>],
+        quorum: usize,
+        size_hint: usize,
+    ) -> std::io::Result<(R, usize)>
+    where
+        R: AsyncRead + Send + Sync + Unpin,
+    {
+        self.encode_small_direct(reader, writers, quorum, false, size_hint).await
     }
 
     /// Fast path for single-block non-inline objects: avoids the producer/consumer
@@ -895,7 +918,24 @@ impl Erasure {
     where
         R: AsyncRead + Send + Sync + Unpin,
     {
-        self.encode_small_direct(reader, writers, quorum, true).await
+        let size_hint = self.block_size;
+        self.encode_small_direct(reader, writers, quorum, true, size_hint).await
+    }
+
+    /// Size-aware single-block fast path. `size_hint` only controls the bounded
+    /// initial allocation; reads remain authoritative.
+    #[hotpath::measure(impl_type = "Erasure")]
+    pub async fn encode_single_block_non_inline_with_size_hint<R>(
+        self: Arc<Self>,
+        reader: R,
+        writers: &mut [Option<BitrotWriterWrapper>],
+        quorum: usize,
+        size_hint: usize,
+    ) -> std::io::Result<(R, usize)>
+    where
+        R: AsyncRead + Send + Sync + Unpin,
+    {
+        self.encode_small_direct(reader, writers, quorum, true, size_hint).await
     }
 }
 
@@ -2293,7 +2333,10 @@ mod tests {
 
         let erasure = Arc::new(Erasure::new(1, 0, 16));
         let reader = tokio::io::BufReader::new(Cursor::new(Vec::<u8>::new()));
-        let (_reader, total) = erasure.encode_inline_small(reader, &mut writers, 1).await.unwrap();
+        let (_reader, total) = erasure
+            .encode_inline_small_with_size_hint(reader, &mut writers, 1, 0)
+            .await
+            .unwrap();
 
         assert_eq!(total, 0);
         // No shutdown was called, so nothing should be committed
@@ -2325,7 +2368,10 @@ mod tests {
         let payload = b"hello inline small";
         let erasure = Arc::new(Erasure::new(DATA_SHARDS, PARITY_SHARDS, BLOCK_SIZE));
         let reader = tokio::io::BufReader::new(Cursor::new(payload.to_vec()));
-        let (_reader, total) = erasure.encode_inline_small(reader, &mut writers, DATA_SHARDS).await.unwrap();
+        let (_reader, total) = erasure
+            .encode_inline_small_with_size_hint(reader, &mut writers, DATA_SHARDS, 1)
+            .await
+            .unwrap();
 
         assert_eq!(total, payload.len());
         // All shards must have received data (shutdown flushed the bitrot header + shard bytes)
@@ -2392,7 +2438,7 @@ mod tests {
         let erasure = Arc::new(Erasure::new(DATA_SHARDS, PARITY_SHARDS, BLOCK_SIZE));
         let reader = tokio::io::BufReader::new(Cursor::new(payload));
         let err = erasure
-            .encode_single_block_non_inline(reader, &mut writers, DATA_SHARDS)
+            .encode_single_block_non_inline_with_size_hint(reader, &mut writers, DATA_SHARDS, BLOCK_SIZE)
             .await
             .expect_err("single-block fast path must reject oversized readers");
 
@@ -2403,6 +2449,21 @@ mod tests {
         }
     }
 
+    #[test]
+    fn small_ingest_capacity_uses_bounded_size_hint() {
+        let erasure = Erasure::new(4, 2, 1024 * 1024);
+        assert_eq!(small_ingest_capacity(&erasure, 0), 0);
+        assert_eq!(small_ingest_capacity(&erasure, 4 * 1024), 6 * 1024);
+        assert_eq!(small_ingest_capacity(&erasure, 16 * 1024), 24 * 1024);
+        assert_eq!(small_ingest_capacity(&erasure, usize::MAX), 1024 * 1024);
+
+        let legacy = Erasure::new_with_options(4, 2, 1024 * 1024, true);
+        assert_eq!(small_ingest_capacity(&legacy, 4 * 1024), 6 * 1024);
+
+        let high_parity = Erasure::new(4, 12, 1024 * 1024);
+        assert_eq!(small_ingest_capacity(&high_parity, usize::MAX), 1024 * 1024);
+    }
+
     #[tokio::test]
     async fn read_full_buf_or_eof_returns_none_on_empty_reader() {
         let mut reader = Cursor::new(Vec::<u8>::new());

crates/ecstore/src/erasure/coding/erasure.rs

@@ -968,6 +968,15 @@ impl Erasure {
         self.data_shards + self.parity_shards
     }
 
+    pub(crate) fn encoded_capacity_for_data_len(&self, data_len: usize) -> usize {
+        let shard_size_fn = if self.uses_legacy {
+            calc_shard_size_legacy
+        } else {
+            calc_shard_size
+        };
+        shard_size_fn(data_len, self.data_shards).saturating_mul(self.total_shard_count())
+    }
+
     /// Whether the erasure dimensions are safe for the shard/offset arithmetic.
     ///
     /// `block_size` and `data_shards` come straight from on-disk metadata; a

crates/ecstore/src/io_support/bitrot.rs

@@ -120,26 +120,41 @@ struct BitrotReaderSource {
 
 impl BitrotReaderSource {
     async fn open(self) -> disk::error::Result<Option<BoxedObjectReader>> {
-        if let Some(data) = self.inline_data {
-            let mut rd = Cursor::new(data);
-            let offset = u64::try_from(self.offset).map_err(|_| DiskError::FileCorrupt)?;
-            rd.set_position(offset);
-            Ok(Some(ShardReader::InMemory(rd)))
-        } else if let Some(disk) = self.disk {
-            open_disk_reader(
-                &disk,
-                &self.bucket,
-                &self.path,
-                self.offset,
-                self.length,
-                self.use_mmap_read,
-                self.stage_metrics.map(|metrics| metrics.path),
-            )
+        open_reader_source(
+            self.inline_data,
+            self.disk.as_ref(),
+            &self.bucket,
+            &self.path,
+            self.offset,
+            self.length,
+            self.use_mmap_read,
+            self.stage_metrics.map(|metrics| metrics.path),
+        )
+        .await
+    }
+}
+
+#[allow(clippy::too_many_arguments)]
+async fn open_reader_source(
+    inline_data: Option<Bytes>,
+    disk: Option<&DiskStore>,
+    bucket: &str,
+    path: &str,
+    offset: usize,
+    length: usize,
+    use_mmap_read: bool,
+    metrics_path: Option<&'static str>,
+) -> disk::error::Result<Option<BoxedObjectReader>> {
+    if let Some(data) = inline_data {
+        let mut reader = Cursor::new(data);
+        reader.set_position(u64::try_from(offset).map_err(|_| DiskError::FileCorrupt)?);
+        Ok(Some(ShardReader::InMemory(reader)))
+    } else if let Some(disk) = disk {
+        open_disk_reader(disk, bucket, path, offset, length, use_mmap_read, metrics_path)
             .await
             .map(Some)
-        } else {
-            Ok(None)
-        }
+    } else {
+        Ok(None)
     }
 }
 
@@ -623,22 +638,22 @@ async fn create_bitrot_reader_from_bytes_with_stage_metrics(
 
     let reader_construction_start = stage_metrics_enabled.then(Instant::now);
     let (offset, length) = bitrot_encoded_range(offset, length, shard_size, checksum_algo.clone());
-    let source = BitrotReaderSource {
-        inline_data,
-        disk: disk.cloned(),
-        bucket: bucket.to_string(),
-        path: path.to_string(),
-        offset,
-        length,
-        use_mmap_read,
-        stage_metrics,
-    };
     if let Some(metrics) = stage_metrics {
         record_get_stage_duration_if_enabled(metrics.path, metrics.reader_construction_stage, reader_construction_start);
     }
 
     let file_open_start = stage_metrics_enabled.then(Instant::now);
-    let reader = source.open().await?;
+    let reader = open_reader_source(
+        inline_data,
+        disk,
+        bucket,
+        path,
+        offset,
+        length,
+        use_mmap_read,
+        stage_metrics.map(|metrics| metrics.path),
+    )
+    .await?;
     if let Some(metrics) = stage_metrics {
         record_get_stage_duration_if_enabled(metrics.path, metrics.file_open_stage, file_open_start);
     }
@@ -698,11 +713,12 @@ pub(crate) fn create_deferred_bitrot_reader_with_stripe_handle(
 ) -> (BitrotReader<ShardReader>, DeferredReaderStripeHandle) {
     let stripe_stride = shard_size + checksum_algo.size();
     let (offset, length) = bitrot_encoded_range(offset, length, shard_size, checksum_algo.clone());
+    let inline_source = inline_data.is_some();
     let source = BitrotReaderSource {
         inline_data,
         disk,
-        bucket: bucket.to_string(),
-        path: path.to_string(),
+        bucket: if inline_source { String::new() } else { bucket.to_string() },
+        path: if inline_source { String::new() } else { path.to_string() },
         offset,
         length,
         use_mmap_read,

crates/ecstore/src/layout/format.rs

@@ -234,11 +234,17 @@ mod test {
 
     #[test]
     fn test_format_v1() {
+        // A freshly created format must survive a serialize -> parse roundtrip
+        // unchanged (identity on every on-disk field).
         let format = FormatV3::new(1, 4);
+        let serialized = serde_json::to_string(&format).expect("FormatV3 must serialize to JSON");
+        let reparsed = FormatV3::try_from(serialized.as_str()).expect("serialized FormatV3 must parse back");
+        assert_eq!(reparsed, format);
 
-        let str = serde_json::to_string(&format);
-        println!("{str:?}");
-
+        // minio-file-format-compat: this literal pins the on-disk format.json
+        // shape (erasure version "1", distributionAlgo "CRCMOD"). `this` always
+        // carries the disk's own UUID in real format.json files; a JSON null
+        // there was never parseable and never written by MinIO or RustFS.
         let data = r#"
         {
             "version": "1",
@@ -246,7 +252,7 @@ mod test {
             "id": "321b3874-987d-4c15-8fa5-757c956b1243",
             "xl": {
                 "version": "1",
-                "this": null,
+                "this": "8ab9a908-f869-4f1f-8e42-eb067ffa7eb5",
                 "sets": [
                     [
                         "8ab9a908-f869-4f1f-8e42-eb067ffa7eb5",
@@ -259,9 +265,23 @@ mod test {
             }
         }"#;
 
-        let p = FormatV3::try_from(data);
+        let parsed = FormatV3::try_from(data).expect("pinned v1 format.json literal must keep parsing");
 
-        println!("{p:?}");
+        assert_eq!(parsed.version, FormatMetaVersion::V1);
+        assert_eq!(parsed.format, FormatBackend::Erasure);
+        assert_eq!(
+            parsed.id,
+            Uuid::parse_str("321b3874-987d-4c15-8fa5-757c956b1243").expect("literal id is a valid UUID")
+        );
+        assert_eq!(parsed.erasure.version, FormatErasureVersion::V1);
+        assert_eq!(
+            parsed.erasure.this,
+            Uuid::parse_str("8ab9a908-f869-4f1f-8e42-eb067ffa7eb5").expect("literal this is a valid UUID")
+        );
+        assert_eq!(parsed.erasure.sets.len(), 1);
+        assert_eq!(parsed.erasure.sets[0].len(), 4);
+        assert_eq!(parsed.erasure.sets[0][0], parsed.erasure.this);
+        assert_eq!(parsed.erasure.distribution_algo, DistributionAlgoVersion::V1);
     }
 
     #[test]

crates/ecstore/src/set_disk/core/io_primitives.rs

@@ -58,6 +58,7 @@ use crate::io_support::bitrot::{
     create_deferred_bitrot_reader_with_stripe_handle, object_mmap_read_enabled, object_mmap_read_max_length,
 };
 use crate::set_disk::shard_source::ShardReadCost;
+use futures::FutureExt as _;
 use futures::stream::{FuturesUnordered, StreamExt};
 use metrics::counter;
 use std::{
@@ -221,7 +222,7 @@ impl MetadataFanoutDiagnostics {
         self.observations.iter().filter(|observation| observation.ignored).count()
     }
 
-    pub(in crate::set_disk) fn error_responses(&self) -> usize {
+    pub(in crate::set_disk) fn non_valid_responses(&self) -> usize {
         self.total_responses().saturating_sub(self.valid_responses())
     }
 
@@ -272,7 +273,7 @@ impl MetadataFanoutDiagnostics {
             self.total_responses(),
             self.valid_responses(),
             self.ignored_responses(),
-            self.error_responses(),
+            self.non_valid_responses(),
         );
         for observation in &self.observations {
             rustfs_io_metrics::record_get_object_metadata_response(path, observation.outcome);
@@ -2863,8 +2864,6 @@ impl SetDisks {
                 file_info.validate_for_erasure_write()?;
             }
         }
-        let mut futures = Vec::with_capacity(disks.len());
-
         let mut errs = Vec::with_capacity(disks.len());
 
         let src_bucket = Arc::new(src_bucket.to_string());
@@ -2872,48 +2871,65 @@ impl SetDisks {
         let dst_bucket = Arc::new(dst_bucket.to_string());
         let dst_object = Arc::new(dst_object.to_string());
 
-        for (i, (disk, file_info)) in disks.iter().zip(file_infos.iter()).enumerate() {
-            let mut file_info = file_info.clone();
-            let disk = disk.clone();
-            let src_bucket = src_bucket.clone();
-            let src_object = src_object.clone();
-            let dst_object = dst_object.clone();
-            let dst_bucket = dst_bucket.clone();
+        let disk_count = disks.len();
+        let fanout_disks = disks.to_vec();
+        let fanout_file_infos = file_infos.to_vec();
+        let fanout_src_bucket = src_bucket.clone();
+        let fanout_src_object = src_object.clone();
+        let fanout_dst_bucket = dst_bucket.clone();
+        let fanout_dst_object = dst_object.clone();
+        // Keep one coordinator task so a cancelled caller cannot drop partially
+        // completed disk mutations. Per-disk futures stay ordered in `join_all`,
+        // preserving slot-indexed quorum and convergence accounting without a
+        // scheduler task for every disk.
+        let fanout = tokio::spawn(async move {
+            let futures = fanout_disks
+                .into_iter()
+                .zip(fanout_file_infos)
+                .enumerate()
+                .map(|(i, (disk, mut file_info))| {
+                    let src_bucket = fanout_src_bucket.clone();
+                    let src_object = fanout_src_object.clone();
+                    let dst_object = fanout_dst_object.clone();
+                    let dst_bucket = fanout_dst_bucket.clone();
 
-            futures.push(tokio::spawn(async move {
-                // Test-only introspection guard: counts this task as in-flight for
-                // the whole body. Compiles to `()` in production (no behavior).
-                #[allow(clippy::let_unit_value)]
-                let _fanout_task_guard = Self::rename_fanout_task_guard(&dst_object);
+                    std::panic::AssertUnwindSafe(async move {
+                        // Test-only introspection guard: counts this operation as
+                        // in-flight for the whole body. Compiles to `()` in production.
+                        #[allow(clippy::let_unit_value)]
+                        let _fanout_task_guard = Self::rename_fanout_task_guard(&dst_object);
 
-                let Some(disk) = disk else {
-                    return Err(DiskError::DiskNotFound);
-                };
+                        let Some(disk) = disk else {
+                            return Err(DiskError::DiskNotFound);
+                        };
 
-                let is_delete_marker = file_info.is_canonical_delete_marker();
-                if file_info.erasure.index == 0 {
-                    file_info.erasure.index = i + 1;
-                }
+                        let is_delete_marker = file_info.is_canonical_delete_marker();
+                        if file_info.erasure.index == 0 {
+                            file_info.erasure.index = i + 1;
+                        }
 
-                if !is_delete_marker && !file_info.has_valid_erasure_geometry() {
-                    return Err(DiskError::FileCorrupt);
-                }
+                        if !is_delete_marker && !file_info.has_valid_erasure_geometry() {
+                            return Err(DiskError::FileCorrupt);
+                        }
 
-                // Test-only awaitable pause point right before the disk rename.
-                // A no-op immediately-ready future in production.
-                Self::rename_fanout_barrier(&dst_object, i, rename_fanout_barrier_phase::RENAME).await;
+                        // Test-only awaitable pause point right before the disk rename.
+                        // A no-op immediately-ready future in production.
+                        Self::rename_fanout_barrier(&dst_object, i, rename_fanout_barrier_phase::RENAME).await;
 
-                disk.rename_data(&src_bucket, &src_object, file_info, &dst_bucket, &dst_object)
-                    .await
-            }));
-        }
+                        disk.rename_data(&src_bucket, &src_object, file_info, &dst_bucket, &dst_object)
+                            .await
+                    })
+                    .catch_unwind()
+                });
+            join_all(futures).await
+        });
 
-        let mut disk_versions = vec![None; disks.len()];
-        let mut data_dirs = vec![None; disks.len()];
-        let mut cleanup_data_dirs = vec![None; disks.len()];
-        let mut old_current_sizes = vec![None; disks.len()];
+        let mut disk_versions = vec![None; disk_count];
+        let mut data_dirs = vec![None; disk_count];
+        let mut cleanup_data_dirs = vec![None; disk_count];
+        let mut old_current_sizes = vec![None; disk_count];
 
-        let results = join_all(futures).await;
+        let results = fanout.await.map_err(|_| DiskError::Unexpected)?;
 
         for (idx, result) in results.iter().enumerate() {
             match result.as_ref().map_err(|_| DiskError::Unexpected)? {
@@ -5877,6 +5893,51 @@ mod tests {
         drop(dirs);
     }
 
+    #[tokio::test]
+    async fn rename_fanout_drains_after_caller_cancellation() {
+        const DISKS: usize = 4;
+        let bucket = "rename-cancel-bucket";
+        let object = "rename-cancel-object";
+        let (dirs, disks) = call_counter_local_disks(bucket, DISKS).await;
+        let marker = metadata_test_delete_marker(object, Uuid::new_v4(), OffsetDateTime::now_utc());
+        let file_infos = vec![marker; DISKS];
+        let tracker = rename_fanout_barrier::observe_tasks(object);
+        let barrier = rename_fanout_barrier::arm(object, 0, rename_fanout_barrier::PHASE_RENAME);
+
+        let rename =
+            tokio::spawn(
+                async move { SetDisks::rename_data(&disks, bucket, object, &file_infos, bucket, object, DISKS - 1).await },
+            );
+        tokio::time::timeout(BARRIER_PAUSE_GUARD, barrier.wait_until_paused())
+            .await
+            .expect("rename fan-out must reach the armed barrier");
+        rename.abort();
+        assert!(
+            rename
+                .await
+                .expect_err("aborted caller should report cancellation")
+                .is_cancelled(),
+            "caller task should be cancelled, not panic"
+        );
+        assert!(tracker.running() >= 1, "the coordinator must retain in-flight disk mutations");
+
+        barrier.release();
+        tokio::time::timeout(BARRIER_PAUSE_GUARD, async {
+            while tracker.running() != 0 {
+                tokio::task::yield_now().await;
+            }
+        })
+        .await
+        .expect("cancelled caller's disk mutations must drain");
+
+        for (idx, dir) in dirs.iter().enumerate() {
+            assert!(
+                dir.path().join(bucket).join(object).join(STORAGE_FORMAT_FILE).exists(),
+                "disk {idx} must finish the rename after caller cancellation"
+            );
+        }
+    }
+
     /// Demo / regression guard for the barrier on the commit (old-data-dir)
     /// cleanup fan-out. Serves the same #1312/#1319 "no background disk write
     /// after release" shape, on the reclamation path that runs *after* a write is
@@ -6047,7 +6108,7 @@ mod tests {
         assert_eq!(diagnostics.total_responses(), 3);
         assert_eq!(diagnostics.valid_responses(), 1);
         assert_eq!(diagnostics.ignored_responses(), 1);
-        assert_eq!(diagnostics.error_responses(), 2);
+        assert_eq!(diagnostics.non_valid_responses(), 2);
         assert_eq!(diagnostics.first_response_latency(), Some(Duration::from_millis(10)));
         assert_eq!(diagnostics.first_valid_response_latency(), Some(Duration::from_millis(30)));
         assert_eq!(diagnostics.slowest_response_latency(), Some(Duration::from_millis(30)));

crates/ecstore/src/set_disk/mod.rs

@@ -639,9 +639,11 @@ const ENV_RUSTFS_GET_CODEC_STREAMING_ENABLE: &str = "RUSTFS_GET_CODEC_STREAMING_
 const DEFAULT_RUSTFS_GET_CODEC_STREAMING_ENABLE: bool = true;
 
 const ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE: &str = "RUSTFS_GET_CODEC_STREAMING_MIN_SIZE";
-const DEFAULT_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE: usize = MI_B;
+// Meet the direct-memory path at its default ceiling. Codec streaming remains
+// rollout-gated and starts where the eager small-object path ends.
+const DEFAULT_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE: usize = DEFAULT_RUSTFS_GET_SMALL_OBJECT_DIRECT_MEMORY_THRESHOLD;
 const ENV_RUSTFS_GET_CODEC_STREAMING_RUSTFS_MIN_SIZE: &str = "RUSTFS_GET_CODEC_STREAMING_RUSTFS_MIN_SIZE";
-const DEFAULT_RUSTFS_GET_CODEC_STREAMING_RUSTFS_MIN_SIZE: usize = MI_B;
+const DEFAULT_RUSTFS_GET_CODEC_STREAMING_RUSTFS_MIN_SIZE: usize = DEFAULT_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE;
 
 const ENV_RUSTFS_GET_CODEC_STREAMING_ENGINE: &str = "RUSTFS_GET_CODEC_STREAMING_ENGINE";
 const DEFAULT_RUSTFS_GET_CODEC_STREAMING_ENGINE: &str = GET_CODEC_STREAMING_ENGINE_LEGACY;
@@ -733,10 +735,41 @@ mod transition_matrix_tests;
 
 pub use ops::heal_walk::HealWalkVersion;
 
+pub(in crate::set_disk) enum GetObjectMetadata<T> {
+    Owned(T),
+    Shared(Arc<T>),
+}
+
+impl<T> std::ops::Deref for GetObjectMetadata<T> {
+    type Target = T;
+
+    fn deref(&self) -> &Self::Target {
+        match self {
+            Self::Owned(value) => value,
+            Self::Shared(value) => value,
+        }
+    }
+}
+
+impl<T: Clone> GetObjectMetadata<T> {
+    fn into_owned(self) -> T {
+        match self {
+            Self::Owned(value) => value,
+            Self::Shared(value) => Arc::try_unwrap(value).unwrap_or_else(|value| (*value).clone()),
+        }
+    }
+}
+
+type GetObjectFileInfo = (
+    GetObjectMetadata<FileInfo>,
+    GetObjectMetadata<Vec<FileInfo>>,
+    GetObjectMetadata<Vec<Option<DiskStore>>>,
+);
+
 pub(crate) struct PreparedGetObjectMetadata {
-    fi: FileInfo,
-    files: Vec<FileInfo>,
-    disks: Vec<Option<DiskStore>>,
+    fi: GetObjectMetadata<FileInfo>,
+    files: GetObjectMetadata<Vec<FileInfo>>,
+    disks: GetObjectMetadata<Vec<Option<DiskStore>>>,
     object_info: Option<ObjectInfo>,
 }
 
@@ -805,9 +838,9 @@ mod prepared_get_object_metadata_tests {
     #[tokio::test]
     async fn prepared_metadata_is_consumed_exactly_once() {
         let metadata = PreparedGetObjectMetadata {
-            fi: FileInfo::default(),
-            files: Vec::new(),
-            disks: Vec::new(),
+            fi: GetObjectMetadata::Owned(FileInfo::default()),
+            files: GetObjectMetadata::Owned(Vec::new()),
+            disks: GetObjectMetadata::Owned(Vec::new()),
             object_info: None,
         };
 
@@ -2334,6 +2367,8 @@ pub struct SetDisks {
     pub default_parity_count: usize,
     pub set_index: usize,
     pub pool_index: usize,
+    /// Stable namespace shared by every object lock created for this set.
+    set_lock_namespace: Arc<str>,
     pub format: FormatV3,
     disk_health_cache: Arc<RwLock<Vec<Option<DiskHealthEntry>>>>,
     get_object_metadata_cache: moka::future::Cache<GetObjectMetadataCacheKey, Arc<GetObjectMetadataCacheEntry>>,
@@ -2463,13 +2498,13 @@ impl Hash for GetObjectMetadataCacheKey {
     }
 }
 
-#[derive(Clone, Debug)]
+#[derive(Debug)]
 struct GetObjectMetadataCacheEntry {
     #[allow(dead_code)] // Kept for debugging; moka handles TTL internally
     created_at: Instant,
-    fi: FileInfo,
-    parts_metadata: Vec<FileInfo>,
-    online_disks: Vec<Option<DiskStore>>,
+    fi: Arc<FileInfo>,
+    parts_metadata: Arc<Vec<FileInfo>>,
+    online_disks: Arc<Vec<Option<DiskStore>>>,
     read_quorum: usize,
 }
 
@@ -2735,6 +2770,7 @@ impl SetDisks {
         instance_ctx: Arc<InstanceContext>,
     ) -> Arc<Self> {
         let ctx = instance_ctx;
+        let set_lock_namespace: Arc<str> = format!("set-{pool_index}-{set_index}").into();
         Arc::new(SetDisks {
             locker_owner,
             disks,
@@ -2742,6 +2778,7 @@ impl SetDisks {
             default_parity_count,
             set_index,
             pool_index,
+            set_lock_namespace,
             format,
             set_endpoints,
             disk_health_cache: Arc::new(RwLock::new(Vec::new())),
@@ -3190,23 +3227,28 @@ async fn try_read_inline_data_shards_direct(
         return None;
     }
 
-    let mut body = Vec::with_capacity(object_size);
-    let mut remaining = object_size;
-    for reader in readers.iter_mut().take(data_shards) {
+    let shards_needed = object_size.div_ceil(read_length);
+    if shards_needed > data_shards {
+        return None;
+    }
+    let encoded_capacity = read_length.checked_mul(shards_needed)?;
+    let mut body = Vec::with_capacity(encoded_capacity);
+    for reader in readers.iter_mut().take(shards_needed) {
         let reader = reader.as_mut()?;
-        let mut shard = vec![0u8; read_length];
-        let Ok(read) = reader.read(&mut shard).await else {
+        let Ok(read) = reader.read_appending(&mut body, read_length).await else {
             return None;
         };
         if read != read_length {
             return None;
         }
 
-        let take = remaining.min(shard.len());
-        body.extend_from_slice(&shard[..take]);
-        remaining -= take;
-        if remaining == 0 {
-            return Some(Bytes::from(body));
+        if body.len() >= object_size {
+            let body = Bytes::from(body);
+            return Some(if body.len() == object_size {
+                body
+            } else {
+                body.slice(..object_size)
+            });
         }
     }
 
@@ -4934,6 +4976,28 @@ mod tests {
         );
     }
 
+    #[tokio::test]
+    async fn new_ns_lock_reuses_the_set_namespace_allocation() {
+        let ctx = Arc::new(InstanceContext::new());
+        ctx.update_erasure_type(SetupType::Erasure).await;
+        let set = make_test_set_disks_with_ctx(Vec::new(), ctx).await;
+
+        assert_eq!(&*set.set_lock_namespace, "set-0-0");
+        let before = Arc::strong_count(&set.set_lock_namespace);
+        let lock = set
+            .new_ns_lock("bucket", "object")
+            .await
+            .expect("namespace lock should be created");
+
+        assert_eq!(
+            Arc::strong_count(&set.set_lock_namespace),
+            before + 1,
+            "each lock should share the set namespace instead of formatting a new String"
+        );
+        drop(lock);
+        assert_eq!(Arc::strong_count(&set.set_lock_namespace), before);
+    }
+
     struct SetupTypeGuard {
         previous: SetupType,
     }
@@ -8974,10 +9038,17 @@ mod tests {
         ));
     }
 
-    async fn inline_bitrot_files_for_payload(payload: &[u8]) -> (coding::Erasure, Vec<FileInfo>, usize, HashAlgorithm) {
-        let erasure = coding::Erasure::new(4, 2, 1024 * 1024);
+    async fn inline_bitrot_files_for_payload_with_mode(
+        payload: &[u8],
+        uses_legacy: bool,
+    ) -> (coding::Erasure, Vec<FileInfo>, usize, HashAlgorithm) {
+        let erasure = coding::Erasure::new_with_options(4, 2, 1024 * 1024, uses_legacy);
         let read_length = erasure.shard_file_offset(0, payload.len(), payload.len());
-        let checksum_algo = HashAlgorithm::HighwayHash256S;
+        let checksum_algo = if uses_legacy {
+            HashAlgorithm::HighwayHash256SLegacy
+        } else {
+            HashAlgorithm::HighwayHash256S
+        };
         let shards = erasure.encode_data(payload).expect("payload should encode");
         let mut files = Vec::with_capacity(shards.len());
 
@@ -8999,6 +9070,10 @@ mod tests {
         (erasure, files, read_length, checksum_algo)
     }
 
+    async fn inline_bitrot_files_for_payload(payload: &[u8]) -> (coding::Erasure, Vec<FileInfo>, usize, HashAlgorithm) {
+        inline_bitrot_files_for_payload_with_mode(payload, false).await
+    }
+
     fn inline_data_shard_fileinfo(
         name: &str,
         data_blocks: usize,
@@ -9078,15 +9153,41 @@ mod tests {
         assert_eq!(body.as_ref(), payload);
     }
 
+    #[tokio::test]
+    async fn inline_data_shards_direct_read_reassembles_legacy_payload_with_padding() {
+        let payload = b"legacy inline payload whose size is not divisible by the data shard count";
+        let (erasure, files, read_length, checksum_algo) = inline_bitrot_files_for_payload_with_mode(payload, true).await;
+        assert_ne!(payload.len() % erasure.data_shards, 0, "test payload must exercise EC padding");
+        let mut readers = build_inline_bitrot_readers(
+            &files,
+            erasure.data_shards,
+            "bucket",
+            "object",
+            read_length,
+            erasure.shard_size(),
+            &checksum_algo,
+            false,
+        )
+        .await
+        .expect("legacy inline bitrot readers should build");
+
+        let body = try_read_inline_data_shards_direct(&mut readers, erasure.data_shards, read_length, payload.len())
+            .await
+            .expect("legacy data shard direct read should succeed");
+
+        assert_eq!(body.len(), payload.len());
+        assert_eq!(body.as_ref(), payload);
+    }
+
     #[tokio::test]
     async fn inline_data_shards_direct_read_rejects_corrupt_shard() {
         let payload = b"small inline object payload that will be corrupted";
         let (erasure, mut files, read_length, checksum_algo) = inline_bitrot_files_for_payload(payload).await;
-        let first = files[0].data.as_mut().expect("first shard should exist");
-        let mut corrupted = first.to_vec();
+        let second = files[1].data.as_mut().expect("second shard should exist");
+        let mut corrupted = second.to_vec();
         let last = corrupted.last_mut().expect("encoded shard should not be empty");
         *last ^= 0xff;
-        *first = Bytes::from(corrupted);
+        *second = Bytes::from(corrupted);
 
         let mut readers = build_inline_bitrot_readers(
             &files,
@@ -9103,7 +9204,7 @@ mod tests {
 
         let body = try_read_inline_data_shards_direct(&mut readers, 4, read_length, payload.len()).await;
 
-        assert!(body.is_none());
+        assert!(body.is_none(), "a later corrupt shard must discard the already-appended body prefix");
     }
 
     #[test]

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

@@ -39,16 +39,9 @@ impl crate::storage_api_contracts::namespace::NamespaceLocking for SetDisks {
             // Calculate quorum based on lockers count (majority)
             let lockers_count = self.lockers.len();
             let write_quorum = if lockers_count > 1 { (lockers_count / 2) + 1 } else { 1 };
-            NamespaceLock::with_clients_and_quorum(
-                format!("set-{}-{}", self.pool_index, self.set_index),
-                self.lockers.clone(),
-                write_quorum,
-            )
+            NamespaceLock::with_clients_and_quorum_shared(self.set_lock_namespace.clone(), self.lockers.clone(), write_quorum)
         } else {
-            NamespaceLock::Local(LocalLock::new(
-                format!("set-{}-{}", self.pool_index, self.set_index),
-                self.local_lock_manager.clone(),
-            ))
+            NamespaceLock::with_local_manager_shared(self.set_lock_namespace.clone(), self.local_lock_manager.clone())
         };
 
         let resource = ObjectKey {

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

@@ -162,6 +162,22 @@ fn map_upload_id_metadata_error(bucket: &str, object: &str, upload_id: &str, err
     err.into()
 }
 
+/// Abort a multipart commit when the guard's refresh heartbeat has observed a
+/// refresh-quorum loss (backlog#899 Phase 2): a stale holder must not race a
+/// concurrent committer past its fenced commit point.
+fn fence_commit_on_lock_loss(guard: Option<&ObjectLockDiagGuard>, mode: &'static str, lock_path: &str) -> Result<()> {
+    if guard.is_some_and(|guard| guard.is_lock_lost()) {
+        return Err(StorageError::NamespaceLockQuorumUnavailable {
+            mode,
+            bucket: RUSTFS_META_MULTIPART_BUCKET.to_string(),
+            object: lock_path.to_string(),
+            required: 1,
+            achieved: 0,
+        });
+    }
+    Ok(())
+}
+
 fn multipart_bucket_incarnation_id(metadata: &HashMap<String, String>) -> Result<Option<Uuid>> {
     let Some(value) = rustfs_utils::http::metadata_compat::get_consistent_str(metadata, SUFFIX_BUCKET_INCARNATION_ID) else {
         if rustfs_utils::http::metadata_compat::contains_key_str(metadata, SUFFIX_BUCKET_INCARNATION_ID) {
@@ -975,12 +991,17 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
 
             let write_path = classify_multipart_part_write_path(multipart_part_size, fi.erasure.block_size);
             rustfs_io_metrics::record_put_object_path(write_path.multipart_metric_label());
+            let small_size_hint = if matches!(write_path, SmallWritePath::SingleBlockNonInline) {
+                usize::try_from(multipart_part_size).map_err(Error::other)?
+            } else {
+                0
+            };
             let encode_stage_start = rustfs_io_metrics::put_stage_metrics_enabled().then(Instant::now);
 
             let (reader, w_size) = match write_path {
                 SmallWritePath::SingleBlockNonInline => {
                     Arc::clone(&erasure)
-                        .encode_single_block_non_inline(stream, &mut writers, write_quorum)
+                        .encode_single_block_non_inline_with_size_hint(stream, &mut writers, write_quorum, small_size_hint)
                         .await?
                 }
                 SmallWritePath::PipelineBatchedLarge => {
@@ -1087,29 +1108,38 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
             }
 
             let part_path = format!("{}/{}/{}", upload_id_path, fi.data_dir.unwrap_or_default(), part_suffix);
+            let part_lock_path = format!("{upload_id_path}/{part_suffix}");
 
             #[cfg(test)]
             pause_multipart_commit(bucket, object, MultipartCommitPause::PutPartBeforeLockAcquire).await;
-            // Serialize only the commit (rename_part), not the whole upload. Each
-            // concurrent stream writes to its own unique temp dir (see `tmp_part`
-            // above), so the encode/stream phase never conflicts and must stay
-            // lock-free — holding a lock across it would serialize slow re-transmits
-            // of the same part and defeat the S3 "last finisher wins" semantics
-            // (it also caused UploadPart lock-acquire timeouts). The mixed-generation
-            // hazard is confined to rename_part, where two temp parts are moved
-            // cross-disk onto the SAME final part_path: interleaving there can leave
-            // shards from two generations, each individually bitrot-valid, that only
-            // surface as silent corruption at read time (backlog#853). A write lock
-            // scoped to the uploadId namespace makes each commit atomic across disks,
-            // so the last committer wins consistently. A guarded completion takes
-            // the object lock before this upload lock to preserve global ordering.
-            let _upload_commit_guard = if opts.no_lock {
-                None
+            // Serialize only same-part commits (rename_part), not the whole upload.
+            // Each concurrent stream writes to its own unique temp dir (see
+            // `tmp_part` above), so the encode/stream phase never conflicts and must
+            // stay lock-free — holding a lock across it would serialize slow
+            // re-transmits of the same part and defeat the S3 "last finisher wins"
+            // semantics. The mixed-generation hazard is confined to rename_part,
+            // where two temp parts are moved cross-disk onto the SAME final
+            // part_path: interleaving there can leave shards from two generations,
+            // each individually bitrot-valid, that only surface as silent corruption
+            // at read time (backlog#853). A write lock scoped to this part number
+            // makes each same-part commit atomic across disks, so the last committer
+            // wins consistently, while different part numbers commit onto disjoint
+            // part paths and stay concurrent (issue#5961 — an uploadId-wide write
+            // lock serialized them into 503 lock-acquire timeouts). The shared
+            // uploadId read lock keeps completion/abort (which take the uploadId
+            // write lock) from racing any in-flight part commit; a guarded
+            // completion takes the object lock before the upload lock to preserve
+            // global ordering.
+            let (_upload_commit_guard, _part_commit_guard) = if opts.no_lock {
+                (None, None)
             } else {
-                Some(
-                    self.acquire_write_lock_diag("put_object_part_commit", RUSTFS_META_MULTIPART_BUCKET, &upload_id_path)
-                        .await?,
-                )
+                let upload_guard = self
+                    .acquire_read_lock_diag("put_object_part_commit", RUSTFS_META_MULTIPART_BUCKET, &upload_id_path)
+                    .await?;
+                let part_guard = self
+                    .acquire_write_lock_diag("put_object_part_commit", RUSTFS_META_MULTIPART_BUCKET, &part_lock_path)
+                    .await?;
+                (Some(upload_guard), Some(part_guard))
             };
             let (commit_fi, _) = self.check_upload_id_exists(bucket, object, upload_id, false).await?;
             ensure_data_movement_upload_access(&commit_fi, bucket, object, upload_id, opts)?;
@@ -1124,15 +1154,8 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
             .await?;
             #[cfg(test)]
             pause_multipart_commit(bucket, object, MultipartCommitPause::PutPartBeforeLockLost).await;
-            if _upload_commit_guard.as_ref().is_some_and(|guard| guard.is_lock_lost()) {
-                return Err(StorageError::NamespaceLockQuorumUnavailable {
-                    mode: "put_object_part_commit",
-                    bucket: RUSTFS_META_MULTIPART_BUCKET.to_string(),
-                    object: upload_id_path.clone(),
-                    required: 1,
-                    achieved: 0,
-                });
-            }
+            fence_commit_on_lock_loss(_upload_commit_guard.as_ref(), "put_object_part_commit", &upload_id_path)?;
+            fence_commit_on_lock_loss(_part_commit_guard.as_ref(), "put_object_part_commit", &part_lock_path)?;
             ensure_multipart_bucket_lifecycle_lock_held(bucket, object, opts)?;
 
             let _ = self
@@ -1156,6 +1179,7 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
 
             #[cfg(test)]
             pause_multipart_commit(bucket, object, MultipartCommitPause::PutPartAfterRename).await;
+            drop(_part_commit_guard);
             drop(_upload_commit_guard);
 
             let ret: PartInfo = PartInfo {
@@ -3988,6 +4012,268 @@ mod tests {
             .expect("abort should delete the upload after UploadPart releases the lock");
     }
 
+    #[tokio::test(flavor = "multi_thread")]
+    #[serial]
+    async fn put_object_part_different_part_numbers_commit_concurrently() {
+        use tokio::io::AsyncReadExt as _;
+
+        const PART1_SIZE: usize = 5 * 1024 * 1024; // non-final parts must be >= 5MiB to complete
+        const PART2_SIZE: usize = 4096;
+
+        let manager = Arc::new(rustfs_lock::GlobalLockManager::new());
+        let locker: Arc<dyn LockClient> = Arc::new(LocalClient::with_manager(manager));
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks_with_lockers(4, 0, 2, vec![locker]).await;
+        let bucket = "multipart-concurrent-part-numbers-bucket";
+        let object = "object";
+        make_bucket_on_all(&disk_stores, bucket).await;
+        let upload = set_disks
+            .new_multipart_upload(bucket, object, &ObjectOptions::default())
+            .await
+            .expect("multipart upload should be created");
+        let upload_id = upload.upload_id;
+        let _setup_type_guard = SetupTypeGuard::switch_to(SetupType::DistErasure).await;
+        // issue#5961: the barrier releases only once BOTH commits are paused
+        // inside their commit sections, so reaching wait_until_paused proves the
+        // two part numbers held their commit locks concurrently. Under an
+        // uploadId-wide exclusive commit lock the second put errors at the 5s
+        // lock-acquire timeout instead of arriving, and wait_until_paused fails
+        // deterministically. No wall-clock bound on the success path.
+        let barrier = MultipartCommitBarrier::install_for_arrivals(bucket, object, MultipartCommitPause::PutPartAfterRename, 2);
+
+        let put1_store = set_disks.clone();
+        let put1_upload_id = upload_id.clone();
+        let put1 = tokio::spawn(async move {
+            let mut reader = PutObjReader::from_vec(vec![0x51; PART1_SIZE]);
+            put1_store
+                .put_object_part(bucket, object, &put1_upload_id, 1, &mut reader, &ObjectOptions::default())
+                .await
+        });
+        let put2_store = set_disks.clone();
+        let put2_upload_id = upload_id.clone();
+        let put2 = tokio::spawn(async move {
+            let mut reader = PutObjReader::from_vec(vec![0x52; PART2_SIZE]);
+            put2_store
+                .put_object_part(bucket, object, &put2_upload_id, 2, &mut reader, &ObjectOptions::default())
+                .await
+        });
+        barrier.wait_until_paused().await;
+
+        barrier.release();
+        let part1 = put1
+            .await
+            .expect("part 1 task should not panic")
+            .expect("part 1 should commit after the barrier is released");
+        let part2 = put2
+            .await
+            .expect("part 2 task should not panic")
+            .expect("part 2 should commit after the barrier is released");
+        assert_eq!(part1.part_num, 1);
+        assert_eq!(part2.part_num, 2);
+
+        set_disks
+            .clone()
+            .complete_multipart_upload(
+                bucket,
+                object,
+                &upload_id,
+                vec![
+                    CompletePart {
+                        part_num: part1.part_num,
+                        etag: part1.etag.clone(),
+                        ..Default::default()
+                    },
+                    CompletePart {
+                        part_num: part2.part_num,
+                        etag: part2.etag.clone(),
+                        ..Default::default()
+                    },
+                ],
+                &ObjectOptions::default(),
+            )
+            .await
+            .expect("completion should succeed with both concurrently committed parts");
+
+        let mut reader = set_disks
+            .get_object_reader(bucket, object, None, HeaderMap::new(), &ObjectOptions::default())
+            .await
+            .expect("completed object should open");
+        let mut body = Vec::new();
+        reader
+            .stream
+            .read_to_end(&mut body)
+            .await
+            .expect("completed object should stream fully");
+        assert_eq!(body.len(), PART1_SIZE + PART2_SIZE);
+        assert!(body[..PART1_SIZE].iter().all(|b| *b == 0x51), "part 1 bytes must round-trip");
+        assert!(body[PART1_SIZE..].iter().all(|b| *b == 0x52), "part 2 bytes must round-trip");
+    }
+
+    #[tokio::test(flavor = "multi_thread")]
+    #[serial]
+    async fn put_object_part_same_part_retries_serialize_on_part_lock() {
+        use tokio::io::AsyncReadExt as _;
+
+        let manager = Arc::new(rustfs_lock::GlobalLockManager::new());
+        let signaling = Arc::new(SignalingLockClient::new(Arc::new(LocalClient::with_manager(manager))));
+        let lockers: Vec<Arc<dyn LockClient>> = vec![signaling.clone()];
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks_with_lockers(4, 0, 2, lockers).await;
+        let bucket = "multipart-same-part-retry-bucket";
+        let object = "object";
+        make_bucket_on_all(&disk_stores, bucket).await;
+        let upload = set_disks
+            .new_multipart_upload(bucket, object, &ObjectOptions::default())
+            .await
+            .expect("multipart upload should be created");
+        let upload_id = upload.upload_id;
+        let upload_id_path = SetDisks::get_upload_id_dir(bucket, object, &upload_id);
+        let part_lock_path = format!("{upload_id_path}/part.1");
+        let _setup_type_guard = SetupTypeGuard::switch_to(SetupType::DistErasure).await;
+        let barrier = MultipartCommitBarrier::install(bucket, object, MultipartCommitPause::PutPartAfterRename);
+
+        let first_store = set_disks.clone();
+        let first_upload_id = upload_id.clone();
+        let first = tokio::spawn(async move {
+            let mut reader = PutObjReader::from_vec(vec![0x53; 4096]);
+            first_store
+                .put_object_part(bucket, object, &first_upload_id, 1, &mut reader, &ObjectOptions::default())
+                .await
+        });
+        barrier.wait_until_paused().await;
+
+        // The paused commit must hold its part lock EXCLUSIVELY: even a shared
+        // probe on the part key has to time out. This pins the write-ness of the
+        // part lock — a shared part lock would let two same-part rename_part
+        // calls interleave into mixed-generation shards (backlog#853).
+        let probe = set_disks
+            .new_ns_lock(RUSTFS_META_MULTIPART_BUCKET, &part_lock_path)
+            .await
+            .expect("part namespace lock should be created")
+            .get_read_lock(Duration::from_secs(1))
+            .await;
+        assert!(
+            probe.is_err(),
+            "the in-flight part commit must hold an exclusive write lock on its part key"
+        );
+
+        signaling.set_target(rustfs_lock::ObjectKey::new(RUSTFS_META_MULTIPART_BUCKET, part_lock_path));
+        let retry_store = set_disks.clone();
+        let retry_upload_id = upload_id.clone();
+        let retry = tokio::spawn(async move {
+            let mut reader = PutObjReader::from_vec(vec![0x54; 4096]);
+            retry_store
+                .put_object_part(bucket, object, &retry_upload_id, 1, &mut reader, &ObjectOptions::default())
+                .await
+        });
+        signaling.wait_for_attempts(1).await;
+        tokio::task::yield_now().await;
+        assert!(
+            !retry.is_finished(),
+            "a retry of the same part number must wait for the in-flight commit (backlog#853)"
+        );
+
+        barrier.release();
+        first
+            .await
+            .expect("first attempt task should not panic")
+            .expect("first attempt should commit after the barrier is released");
+        let retry_part = retry
+            .await
+            .expect("retry task should not panic")
+            .expect("the retry should commit after the first attempt releases the part lock");
+
+        set_disks
+            .clone()
+            .complete_multipart_upload(
+                bucket,
+                object,
+                &upload_id,
+                vec![CompletePart {
+                    part_num: retry_part.part_num,
+                    etag: retry_part.etag.clone(),
+                    ..Default::default()
+                }],
+                &ObjectOptions::default(),
+            )
+            .await
+            .expect("the last committed retry must win the final part generation");
+        let mut reader = set_disks
+            .get_object_reader(bucket, object, None, HeaderMap::new(), &ObjectOptions::default())
+            .await
+            .expect("completed object should open");
+        let mut body = Vec::new();
+        reader
+            .stream
+            .read_to_end(&mut body)
+            .await
+            .expect("completed object should stream fully");
+        assert_eq!(body, vec![0x54; 4096], "the retry's generation must be the one served");
+    }
+
+    #[tokio::test(start_paused = true)]
+    #[serial]
+    async fn put_object_part_fences_part_lock_loss_before_rename() {
+        let target = Arc::new(std::sync::RwLock::new(None));
+        let refresh_calls = Arc::new(AtomicUsize::new(0));
+        let lockers: Vec<Arc<dyn LockClient>> = (0..4)
+            .map(|_| {
+                Arc::new(SelectiveLockLossClient::new(Arc::clone(&target), Arc::clone(&refresh_calls))) as Arc<dyn LockClient>
+            })
+            .collect();
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks_with_lockers(4, 0, 2, lockers).await;
+        let bucket = "multipart-put-part-part-lock-loss-bucket";
+        let object = "object";
+        make_bucket_on_all(&disk_stores, bucket).await;
+        let upload = set_disks
+            .new_multipart_upload(bucket, object, &ObjectOptions::default())
+            .await
+            .expect("multipart upload should be created");
+        let upload_id = upload.upload_id;
+        let upload_id_path = SetDisks::get_upload_id_dir(bucket, object, &upload_id);
+        let part_lock_path = format!("{upload_id_path}/part.1");
+        *target.write().expect("lock-loss target should be writable") =
+            Some(rustfs_lock::ObjectKey::new(RUSTFS_META_MULTIPART_BUCKET, part_lock_path.clone()));
+        let _setup_type_guard = SetupTypeGuard::switch_to(SetupType::DistErasure).await;
+        let barrier = MultipartCommitBarrier::install(bucket, object, MultipartCommitPause::PutPartBeforeLockLost);
+
+        let put_store = set_disks.clone();
+        let put_upload_id = upload_id.clone();
+        let put = tokio::spawn(async move {
+            let mut reader = PutObjReader::from_vec(vec![0x47; 4096]);
+            put_store
+                .put_object_part(bucket, object, &put_upload_id, 1, &mut reader, &ObjectOptions::default())
+                .await
+        });
+        barrier.wait_until_paused().await;
+        tokio::time::advance(Duration::from_secs(11)).await;
+        tokio::task::yield_now().await;
+        assert!(
+            refresh_calls.load(Ordering::Acquire) > 0,
+            "part lock heartbeat should reach the test client"
+        );
+        barrier.release();
+
+        let err = put
+            .await
+            .expect("UploadPart task should not panic")
+            .expect_err("UploadPart must fail after losing the part lock");
+        match err {
+            StorageError::NamespaceLockQuorumUnavailable {
+                bucket: lock_bucket,
+                object: lock_object,
+                ..
+            } => {
+                assert_eq!(lock_bucket, RUSTFS_META_MULTIPART_BUCKET);
+                assert_eq!(lock_object, part_lock_path);
+            }
+            other => panic!("unexpected lock-loss error: {other:?}"),
+        }
+        let listed = set_disks
+            .list_object_parts(bucket, object, &upload_id, None, MAX_PARTS_COUNT, &ObjectOptions::default())
+            .await
+            .expect("part lock loss before rename must leave the upload readable");
+        assert!(listed.parts.is_empty(), "part lock loss before rename must not publish the part");
+    }
+
     #[tokio::test(start_paused = true)]
     #[serial]
     async fn put_object_part_fences_upload_lock_loss_before_rename() {

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

@@ -56,6 +56,22 @@ use http::HeaderValue;
 use rustfs_utils::path::decode_dir_object;
 use std::future::Future;
 
+#[inline]
+fn duration_millis_f64(duration: std::time::Duration) -> f64 {
+    duration.as_secs_f64() * 1000.0
+}
+
+#[cfg(test)]
+mod duration_metrics_tests {
+    use super::duration_millis_f64;
+    use std::time::Duration;
+
+    #[test]
+    fn duration_millis_preserves_sub_millisecond_precision() {
+        assert_eq!(duration_millis_f64(Duration::from_micros(125)), 0.125);
+    }
+}
+
 fn is_restore_control_metadata(key: &str) -> bool {
     key.eq_ignore_ascii_case(X_AMZ_RESTORE.as_str())
         || key.eq_ignore_ascii_case(rustfs_utils::http::headers::AMZ_RESTORE_EXPIRY_DAYS)
@@ -734,8 +750,8 @@ impl crate::storage_api_contracts::object::ObjectIO for SetDisks {
                 0,
                 object_info.size,
                 &mut output,
-                fi,
-                files,
+                fi.into_owned(),
+                files.into_owned(),
                 &disks,
                 self.set_index,
                 self.pool_index,
@@ -851,8 +867,8 @@ impl crate::storage_api_contracts::object::ObjectIO for SetDisks {
                 offset,
                 length,
                 &mut writer,
-                fi,
-                files,
+                fi.into_owned(),
+                files.into_owned(),
                 &disks,
                 set_index,
                 pool_index,
@@ -1107,8 +1123,12 @@ impl SetDisks {
                 writers.push(w);
                 errors.push(e);
             }
-            let writer_setup_ms = writer_setup_stage_start.elapsed().as_millis() as u64;
-            rustfs_io_metrics::record_put_object_stage_duration("set_disk_writer_setup", writer_setup_ms as f64);
+            let writer_setup_elapsed = writer_setup_stage_start.elapsed();
+            let writer_setup_ms = writer_setup_elapsed.as_millis() as u64;
+            rustfs_io_metrics::record_put_object_stage_duration(
+                "set_disk_writer_setup",
+                duration_millis_f64(writer_setup_elapsed),
+            );
 
             let nil_count = errors.iter().filter(|&e| e.is_none()).count();
             if nil_count < write_quorum {
@@ -1138,11 +1158,16 @@ impl SetDisks {
 
             let write_path = classify_put_write_path(is_inline_buffer, put_object_size, fi.erasure.block_size);
             rustfs_io_metrics::record_put_object_path(write_path.metric_label());
+            let small_size_hint = if matches!(write_path, SmallWritePath::Inline | SmallWritePath::SingleBlockNonInline) {
+                usize::try_from(put_object_size).map_err(Error::other)?
+            } else {
+                0
+            };
 
             let encode_stage_start = Instant::now();
             let (reader, w_size) = match write_path {
                 SmallWritePath::Inline => match Arc::clone(&erasure)
-                    .encode_inline_small(stream, &mut writers, write_quorum)
+                    .encode_inline_small_with_size_hint(stream, &mut writers, write_quorum, small_size_hint)
                     .await
                 {
                     Ok((r, w)) => (r, w),
@@ -1152,7 +1177,7 @@ impl SetDisks {
                     }
                 },
                 SmallWritePath::SingleBlockNonInline => match Arc::clone(&erasure)
-                    .encode_single_block_non_inline(stream, &mut writers, write_quorum)
+                    .encode_single_block_non_inline_with_size_hint(stream, &mut writers, write_quorum, small_size_hint)
                     .await
                 {
                     Ok((r, w)) => (r, w),
@@ -1178,8 +1203,9 @@ impl SetDisks {
                     }
                 },
             };
-            let encode_ms = encode_stage_start.elapsed().as_millis() as u64;
-            rustfs_io_metrics::record_put_object_stage_duration("set_disk_encode", encode_ms as f64);
+            let encode_elapsed = encode_stage_start.elapsed();
+            let encode_ms = encode_elapsed.as_millis() as u64;
+            rustfs_io_metrics::record_put_object_stage_duration("set_disk_encode", duration_millis_f64(encode_elapsed));
 
             let _ = mem::replace(&mut data.stream, reader);
             // if let Err(err) = close_bitrot_writers(&mut writers).await {
@@ -1497,8 +1523,18 @@ impl SetDisks {
                     let _ = rustfs_common::heal_channel::send_heal_request(request).await;
                 });
             }
-            let rename_stage_ms = rename_stage_start.elapsed().as_millis() as u64;
-            rustfs_io_metrics::record_put_object_stage_duration("set_disk_rename", rename_stage_ms as f64);
+
+            let rename_stage_elapsed = rename_stage_start.elapsed();
+            let rename_stage_ms = rename_stage_elapsed.as_millis() as u64;
+
+            self.invalidate_get_object_metadata_cache(bucket, object).await;
+
+            // `rename_data` has completed the authoritative quorum commit. The
+            // exact old-data-dir reclamation below is best-effort space cleanup;
+            // it must not serialize the next operation on this object.
+            drop(object_lock_guard);
+
+            rustfs_io_metrics::record_put_object_stage_duration("set_disk_rename", duration_millis_f64(rename_stage_elapsed));
             if (rename_stage_ms as u128) >= SET_DISK_COMMIT_TAIL_WARN_THRESHOLD_MS {
                 warn!(
                     event = EVENT_SET_DISK_COMMIT_TAIL_SLOW,
@@ -1527,9 +1563,13 @@ impl SetDisks {
                 let cleanup = self
                     .commit_rename_data_dir(&cleanup_disks, bucket, object, &old_dir.to_string(), &committed_dir, write_quorum)
                     .await;
-                let cleanup_ms = cleanup_stage_start.elapsed().as_millis() as u64;
+                let cleanup_elapsed = cleanup_stage_start.elapsed();
+                let cleanup_ms = cleanup_elapsed.as_millis() as u64;
                 cleanup_stage_ms = Some(cleanup_ms);
-                rustfs_io_metrics::record_put_object_stage_duration("set_disk_old_data_cleanup", cleanup_ms as f64);
+                rustfs_io_metrics::record_put_object_stage_duration(
+                    "set_disk_old_data_cleanup",
+                    duration_millis_f64(cleanup_elapsed),
+                );
                 self.report_old_data_dir_cleanup(bucket, object, &old_dir.to_string(), &cleanup)
                     .await;
                 if (cleanup_ms as u128) >= SET_DISK_COMMIT_TAIL_WARN_THRESHOLD_MS {
@@ -1550,8 +1590,6 @@ impl SetDisks {
                 }
             }
 
-            drop(object_lock_guard); // drop object lock guard to release the lock
-
             for (i, op_disk) in online_disks.iter().enumerate() {
                 if let Some(disk) = op_disk
                     && disk.is_online().await
@@ -1650,10 +1688,6 @@ impl SetDisks {
             );
         }
 
-        if result.is_ok() {
-            self.invalidate_get_object_metadata_cache(bucket, object).await;
-        }
-
         if issue3031_diag_enabled() {
             warn!(
                 target: "rustfs_ecstore::set_disk",
@@ -3169,9 +3203,10 @@ impl SetDisks {
         // quorum, failing write quorum on update_object_meta (backlog#872).
         let mut read_opts = opts.clone();
         read_opts.include_part_checksums = true;
-        let (mut fi, _, disks) = self
+        let (fi, _, disks) = self
             .get_object_fileinfo_gated(bucket, object, &read_opts, false, false)
             .await?;
+        let mut fi = fi.into_owned();
 
         fi.metadata.insert(AMZ_OBJECT_TAGGING.to_owned(), tags.to_owned());
         if let Some(eval_metadata) = &opts.eval_metadata {
@@ -3194,7 +3229,7 @@ impl SetDisks {
             });
         }
 
-        self.update_object_meta(bucket, object, fi.clone(), disks.as_slice()).await?;
+        self.update_object_meta(bucket, object, fi.clone(), &disks).await?;
 
         Ok(ObjectInfo::from_file_info(&fi, bucket, object, opts.versioned || opts.version_suspended))
     }
@@ -4619,9 +4654,10 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
 
         let mut transition_read_opts = opts.clone();
         transition_read_opts.include_part_checksums = true;
-        let (mut fi, meta_arr, online_disks) = self
+        let (fi, meta_arr, online_disks) = self
             .get_object_fileinfo(bucket, object, &transition_read_opts, true, false)
             .await?;
+        let mut fi = fi.into_owned();
         /*if err != nil {
             return Err(to_object_err(err, vec![bucket, object]));
         }*/
@@ -4736,7 +4772,7 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                 cloned_fi.size,
                 &mut writer,
                 cloned_fi,
-                meta_arr,
+                meta_arr.into_owned(),
                 &online_disks,
                 set_index,
                 pool_index,
@@ -4861,7 +4897,7 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
         };
         self.invalidate_get_object_metadata_cache(bucket, object).await;
         let current = self.get_object_fileinfo(bucket, object, &commit_opts, true, false).await;
-        let (mut current_fi, _, _) = match current {
+        let (current_fi, _, _) = match current {
             Ok(current) => current,
             Err(err) => {
                 drop(transition_lock_guard);
@@ -4872,6 +4908,7 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
                 return Err(err);
             }
         };
+        let mut current_fi = current_fi.into_owned();
         let source_matches = current_fi.version_id == fi.version_id
             && current_fi.data_dir == fi.data_dir
             && current_fi.mod_time == fi.mod_time
@@ -5596,7 +5633,7 @@ mod get_object_downstream_close_accounting_tests {
         let previous_gate = rustfs_io_metrics::get_stage_metrics_enabled();
         rustfs_io_metrics::set_get_stage_metrics_enabled(true);
 
-        let (decode_failures, emit_failures) = metrics::with_local_recorder(&recorder, || {
+        let (decode_failures, emit_failures, legacy_fanout, internal_fanout) = metrics::with_local_recorder(&recorder, || {
             runtime.block_on(async {
                 let (_temp_dirs, _disk_stores, set_disks) = hermetic_set_disks(4).await;
                 let bucket = "get-downstream-close-accounting";
@@ -5664,6 +5701,14 @@ mod get_object_downstream_close_accounting_tests {
                             ("reason", GetObjectFailureReason::DownstreamClosed.as_str()),
                         ],
                     ),
+                    recorder.histogram_values(
+                        "rustfs_io_get_object_metadata_fanout_total_responses",
+                        &[("path", GET_OBJECT_PATH_LEGACY_DUPLEX)],
+                    ),
+                    recorder.histogram_values(
+                        "rustfs_io_get_object_metadata_fanout_total_responses",
+                        &[("path", GET_OBJECT_PATH_INTERNAL_META)],
+                    ),
                 )
             })
         });
@@ -5671,6 +5716,11 @@ mod get_object_downstream_close_accounting_tests {
 
         assert!(decode_failures > 0, "the producer must expose the downstream close at decode");
         assert_eq!(emit_failures, 0, "downstream closure must not be counted as an emit failure");
+        assert_eq!(legacy_fanout, vec![4.0], "ordinary object fanout must retain the legacy_duplex path");
+        assert!(
+            internal_fanout.is_empty(),
+            "ordinary object fanout must not be attributed to internal_meta"
+        );
     }
 
     #[test]
@@ -5684,7 +5734,7 @@ mod get_object_downstream_close_accounting_tests {
         let previous_gate = rustfs_io_metrics::get_stage_metrics_enabled();
         rustfs_io_metrics::set_get_stage_metrics_enabled(true);
 
-        let (internal_missing, legacy_unknown) = metrics::with_local_recorder(&recorder, || {
+        let (internal_missing, legacy_unknown, internal_fanout, legacy_fanout) = metrics::with_local_recorder(&recorder, || {
             runtime.block_on(async {
                 let (_temp_dirs, _disk_stores, set_disks) = hermetic_set_disks(4).await;
                 let options = ObjectOptions {
@@ -5720,6 +5770,14 @@ mod get_object_downstream_close_accounting_tests {
                             ("reason", GetObjectFailureReason::Unknown.as_str()),
                         ],
                     ),
+                    recorder.histogram_values(
+                        "rustfs_io_get_object_metadata_fanout_error_responses",
+                        &[("path", GET_OBJECT_PATH_INTERNAL_META)],
+                    ),
+                    recorder.histogram_values(
+                        "rustfs_io_get_object_metadata_fanout_error_responses",
+                        &[("path", GET_OBJECT_PATH_LEGACY_DUPLEX)],
+                    ),
                 )
             })
         });
@@ -5730,6 +5788,8 @@ mod get_object_downstream_close_accounting_tests {
             legacy_unknown, 0,
             "internal metadata miss must not be attributed to legacy_duplex/unknown"
         );
+        assert_eq!(internal_fanout, vec![4.0], "internal metadata fanout must retain its path label");
+        assert!(legacy_fanout.is_empty(), "internal metadata fanout must not leak into legacy_duplex");
     }
 }
 
@@ -6396,9 +6456,9 @@ mod transition_commit_failure_tests {
                 cache_key.clone(),
                 Arc::new(GetObjectMetadataCacheEntry {
                     created_at: Instant::now(),
-                    fi: fi.clone(),
-                    parts_metadata,
-                    online_disks,
+                    fi: Arc::new((*fi).clone()),
+                    parts_metadata: Arc::new(parts_metadata.into_owned()),
+                    online_disks: Arc::new(online_disks.into_owned()),
                     read_quorum: 2,
                 }),
             )
@@ -9018,8 +9078,10 @@ mod put_object_tmp_cleanup_tests {
     use super::hermetic_set_disks_support::hermetic_set_disks_isolated as hermetic_set_disks;
     use super::*;
     use crate::disk::DiskAPI as _;
+    use crate::set_disk::core::io_primitives::rename_fanout_barrier;
     use std::time::Duration;
     use tempfile::TempDir;
+    use tokio::io::AsyncReadExt;
 
     /// Large enough that the erasure shards are written as real tmp files
     /// (never inlined into xl.meta), so both tests exercise actual cleanup.
@@ -9104,6 +9166,168 @@ mod put_object_tmp_cleanup_tests {
         drop(temp_dirs);
     }
 
+    #[tokio::test]
+    async fn committed_put_releases_namespace_lock_before_old_data_cleanup() {
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "put-commit-lock-window";
+        let object = "commit-lock-window-object";
+        for disk in &disk_stores {
+            disk.make_volume(bucket).await.expect("bucket volume should be created");
+        }
+
+        let mut initial_reader = PutObjReader::from_vec(vec![b'0'; TEST_OBJECT_SIZE]);
+        set_disks
+            .put_object(bucket, object, &mut initial_reader, &ObjectOptions::default())
+            .await
+            .expect("initial object should be committed");
+        let mut initial = set_disks
+            .get_object_reader(bucket, object, None, HeaderMap::new(), &ObjectOptions::default())
+            .await
+            .expect("initial object should populate the metadata cache");
+        let mut initial_body = Vec::new();
+        initial
+            .stream
+            .read_to_end(&mut initial_body)
+            .await
+            .expect("initial body should drain");
+        assert_eq!(initial_body, vec![b'0'; TEST_OBJECT_SIZE]);
+
+        let cleanup_barrier = rename_fanout_barrier::arm(object, 0, rename_fanout_barrier::PHASE_CLEANUP);
+        let first_store = Arc::clone(&set_disks);
+        let first = tokio::spawn(async move {
+            let mut reader = PutObjReader::from_vec(vec![b'1'; TEST_OBJECT_SIZE]);
+            first_store
+                .put_object(bucket, object, &mut reader, &ObjectOptions::default())
+                .await
+        });
+        tokio::time::timeout(Duration::from_secs(30), cleanup_barrier.wait_until_paused())
+            .await
+            .expect("first overwrite should reach old-data cleanup");
+
+        let mut committed = tokio::time::timeout(
+            Duration::from_secs(30),
+            set_disks.get_object_reader(bucket, object, None, HeaderMap::new(), &ObjectOptions::default()),
+        )
+        .await
+        .expect("GET should not wait for old-data cleanup")
+        .expect("committed overwrite should be readable during old-data cleanup");
+        let mut committed_body = Vec::new();
+        committed
+            .stream
+            .read_to_end(&mut committed_body)
+            .await
+            .expect("committed overwrite body should drain");
+        assert_eq!(committed_body, vec![b'1'; TEST_OBJECT_SIZE]);
+
+        let second_commit_barrier = PutObjectCommitBarrier::install(bucket, object, PutObjectCommitPause::AfterNamespace);
+        let second_store = Arc::clone(&set_disks);
+        let second = tokio::spawn(async move {
+            let mut reader = PutObjReader::from_vec(vec![b'2'; TEST_OBJECT_SIZE]);
+            second_store
+                .put_object(bucket, object, &mut reader, &ObjectOptions::default())
+                .await
+        });
+        tokio::time::timeout(Duration::from_secs(30), second_commit_barrier.wait_until_paused())
+            .await
+            .expect("second overwrite should acquire the namespace lock during cleanup");
+
+        cleanup_barrier.release();
+        first
+            .await
+            .expect("first overwrite task should join")
+            .expect("first overwrite should remain successful after cleanup");
+        drop(cleanup_barrier);
+        second_commit_barrier.release();
+        second
+            .await
+            .expect("second overwrite task should join")
+            .expect("second overwrite should commit after acquiring the released namespace lock");
+
+        let mut reader = set_disks
+            .get_object_reader(bucket, object, None, HeaderMap::new(), &ObjectOptions::default())
+            .await
+            .expect("the latest overwrite should be readable");
+        let mut body = Vec::new();
+        reader.stream.read_to_end(&mut body).await.expect("latest body should drain");
+        assert_eq!(body, vec![b'2'; TEST_OBJECT_SIZE]);
+    }
+
+    #[tokio::test]
+    async fn cancelled_post_commit_cleanup_does_not_retain_namespace_lock() {
+        let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;
+        let bucket = "put-commit-lock-cancelled-cleanup";
+        let object = "commit-lock-cancelled-cleanup-object";
+        for disk in &disk_stores {
+            disk.make_volume(bucket).await.expect("bucket volume should be created");
+        }
+
+        let mut initial_reader = PutObjReader::from_vec(vec![b'0'; TEST_OBJECT_SIZE]);
+        set_disks
+            .put_object(bucket, object, &mut initial_reader, &ObjectOptions::default())
+            .await
+            .expect("initial object should be committed");
+
+        let cleanup_tasks = rename_fanout_barrier::observe_tasks(object);
+        let cleanup_barrier = rename_fanout_barrier::arm(object, 0, rename_fanout_barrier::PHASE_CLEANUP);
+        let first_store = Arc::clone(&set_disks);
+        let first = tokio::spawn(async move {
+            let mut reader = PutObjReader::from_vec(vec![b'1'; TEST_OBJECT_SIZE]);
+            first_store
+                .put_object(bucket, object, &mut reader, &ObjectOptions::default())
+                .await
+        });
+        tokio::time::timeout(Duration::from_secs(30), cleanup_barrier.wait_until_paused())
+            .await
+            .expect("first overwrite should reach old-data cleanup");
+
+        let second_commit_barrier = PutObjectCommitBarrier::install(bucket, object, PutObjectCommitPause::AfterNamespace);
+        let second_store = Arc::clone(&set_disks);
+        let second = tokio::spawn(async move {
+            let mut reader = PutObjReader::from_vec(vec![b'2'; TEST_OBJECT_SIZE]);
+            second_store
+                .put_object(bucket, object, &mut reader, &ObjectOptions::default())
+                .await
+        });
+        tokio::time::timeout(Duration::from_secs(30), second_commit_barrier.wait_until_paused())
+            .await
+            .expect("second overwrite should acquire the namespace lock before cancellation");
+
+        first.abort();
+        assert!(
+            first
+                .await
+                .expect_err("the first request should be cancelled during cleanup")
+                .is_cancelled()
+        );
+        assert!(
+            cleanup_tasks.running() >= 1,
+            "cancelled cleanup must remain observable until its disk task drains"
+        );
+        cleanup_barrier.release();
+        tokio::time::timeout(Duration::from_secs(30), async {
+            while cleanup_tasks.running() != 0 {
+                tokio::task::yield_now().await;
+            }
+        })
+        .await
+        .expect("cancelled cleanup disk tasks should drain");
+        drop(cleanup_barrier);
+
+        second_commit_barrier.release();
+        second
+            .await
+            .expect("second overwrite task should join")
+            .expect("second overwrite should survive the earlier request cancellation");
+
+        let mut reader = set_disks
+            .get_object_reader(bucket, object, None, HeaderMap::new(), &ObjectOptions::default())
+            .await
+            .expect("the latest overwrite should be readable");
+        let mut body = Vec::new();
+        reader.stream.read_to_end(&mut body).await.expect("latest body should drain");
+        assert_eq!(body, vec![b'2'; TEST_OBJECT_SIZE]);
+    }
+
     #[tokio::test]
     async fn put_object_no_lock_aborts_after_outer_namespace_lock_loss() {
         let (_temp_dirs, disk_stores, set_disks) = hermetic_set_disks(4).await;

crates/ecstore/src/set_disk/read.rs

@@ -30,12 +30,12 @@ use crate::diagnostics::get::{
     GET_METADATA_RESPONSE_CORRUPT, GET_METADATA_RESPONSE_DISK_NOT_FOUND, GET_METADATA_RESPONSE_ERROR,
     GET_METADATA_RESPONSE_IGNORED, GET_METADATA_RESPONSE_NOT_FOUND, GET_METADATA_RESPONSE_TIMEOUT, GET_METADATA_RESPONSE_VALID,
     GET_METADATA_RESPONSE_VERSION_NOT_FOUND, GET_OBJECT_PATH_CODEC_STREAMING, GET_OBJECT_PATH_DIRECT_MEMORY,
-    GET_OBJECT_PATH_LEGACY_DUPLEX, GET_OBJECT_PATH_SET_DISK, GET_STAGE_DECODE, GET_STAGE_METADATA_CACHE_LOOKUP,
-    GET_STAGE_METADATA_RESOLVE, GET_STAGE_RANGE, GET_STAGE_READER_SETUP, GET_STAGE_READER_SETUP_DROP_PENDING,
-    GET_STAGE_READER_SETUP_SCHEDULE, GET_STAGE_READER_SETUP_WAIT_QUORUM, GET_STAGE_READER_TASK_BITROT_READER_INIT,
-    GET_STAGE_READER_TASK_FILE_OPEN, GET_STAGE_READER_TASK_READER_CONSTRUCTION, GetObjectFailureReason, classify_disk_error,
-    get_stage_timer_if_enabled, mark_get_object_downstream_closed, record_get_object_pipeline_failure,
-    record_get_object_pipeline_failure_for_path, record_get_stage_duration_if_enabled,
+    GET_OBJECT_PATH_INTERNAL_META, GET_OBJECT_PATH_LEGACY_DUPLEX, GET_OBJECT_PATH_SET_DISK, GET_STAGE_DECODE,
+    GET_STAGE_METADATA_CACHE_LOOKUP, GET_STAGE_METADATA_RESOLVE, GET_STAGE_RANGE, GET_STAGE_READER_SETUP,
+    GET_STAGE_READER_SETUP_DROP_PENDING, GET_STAGE_READER_SETUP_SCHEDULE, GET_STAGE_READER_SETUP_WAIT_QUORUM,
+    GET_STAGE_READER_TASK_BITROT_READER_INIT, GET_STAGE_READER_TASK_FILE_OPEN, GET_STAGE_READER_TASK_READER_CONSTRUCTION,
+    GetObjectFailureReason, classify_disk_error, get_stage_timer_if_enabled, mark_get_object_downstream_closed,
+    record_get_object_pipeline_failure, record_get_object_pipeline_failure_for_path, record_get_stage_duration_if_enabled,
 };
 use crate::erasure::coding::BitrotReader;
 use crate::io_support::bitrot::{
@@ -116,9 +116,9 @@ impl SetDisks {
             .then_some(GET_METADATA_CACHE_REASON_DIST_ERASURE)
     }
 
-    async fn cached_get_object_fileinfo(&self, bucket: &str, object: &str) -> Option<GetObjectMetadataCacheEntry> {
+    async fn cached_get_object_fileinfo(&self, bucket: &str, object: &str) -> Option<Arc<GetObjectMetadataCacheEntry>> {
         match self.lookup_cached_get_object_fileinfo(bucket, object).await {
-            MetadataCacheLookup::Hit(entry) => Some((*entry).clone()),
+            MetadataCacheLookup::Hit(entry) => Some(entry),
             MetadataCacheLookup::Miss | MetadataCacheLookup::RejectedInsufficientQuorum => None,
         }
     }
@@ -180,9 +180,9 @@ impl SetDisks {
         let key = GetObjectMetadataCacheKey::new(bucket, object, generation);
         let entry = Arc::new(GetObjectMetadataCacheEntry {
             created_at: Instant::now(),
-            fi: fi.clone(),
-            parts_metadata: parts_metadata.to_vec(),
-            online_disks: online_disks.to_vec(),
+            fi: Arc::new(fi.clone()),
+            parts_metadata: Arc::new(parts_metadata.to_vec()),
+            online_disks: Arc::new(online_disks.to_vec()),
             read_quorum,
         });
         self.insert_get_object_metadata_cache_entry_after_insert(key, generation, entry, || {})
@@ -257,7 +257,7 @@ impl SetDisks {
         opts: &ObjectOptions,
         read_data: bool,
         caller_allows_early_stop: bool,
-    ) -> Result<(FileInfo, Vec<FileInfo>, Vec<Option<DiskStore>>)> {
+    ) -> Result<GetObjectFileInfo> {
         self.get_object_fileinfo_gated(bucket, object, opts, read_data, caller_allows_early_stop)
             .await
     }
@@ -274,7 +274,7 @@ impl SetDisks {
         opts: &ObjectOptions,
         read_data: bool,
         allow_early_stop: bool,
-    ) -> Result<(FileInfo, Vec<FileInfo>, Vec<Option<DiskStore>>)> {
+    ) -> Result<GetObjectFileInfo> {
         let vid = opts.version_id.clone().unwrap_or_default();
         let stage_metrics_enabled = rustfs_io_metrics::get_stage_metrics_enabled();
 
@@ -300,7 +300,11 @@ impl SetDisks {
                         GET_STAGE_METADATA_CACHE_LOOKUP,
                         metadata_cache_lookup_start,
                     );
-                    return Ok((cached.fi.clone(), cached.parts_metadata.clone(), cached.online_disks.clone()));
+                    return Ok((
+                        GetObjectMetadata::Shared(Arc::clone(&cached.fi)),
+                        GetObjectMetadata::Shared(Arc::clone(&cached.parts_metadata)),
+                        GetObjectMetadata::Shared(Arc::clone(&cached.online_disks)),
+                    ));
                 }
                 MetadataCacheLookup::Miss => {
                     rustfs_io_metrics::record_get_object_metadata_cache_decision(
@@ -349,7 +353,12 @@ impl SetDisks {
             self.default_parity_count,
         )
         .await?;
-        metadata_fanout_diagnostics.record(GET_OBJECT_PATH_LEGACY_DUPLEX);
+        let metadata_metrics_path = if crate::bucket::utils::is_meta_bucketname(bucket) {
+            GET_OBJECT_PATH_INTERNAL_META
+        } else {
+            GET_OBJECT_PATH_LEGACY_DUPLEX
+        };
+        metadata_fanout_diagnostics.record(metadata_metrics_path);
         let metadata_fanout_complete = metadata_fanout_diagnostics.total_responses() >= disks.len();
         // warn!("get_object_fileinfo parts_metadata {:?}", &parts_metadata);
         // warn!("get_object_fileinfo {}/{} errs {:?}", bucket, object, &errs);
@@ -396,7 +405,7 @@ impl SetDisks {
                 rustfs_utils::http::remove_str(&mut metadata.metadata, rustfs_utils::http::SUFFIX_PART_CHECKSUMS);
             }
         }
-        metadata_fanout_diagnostics.record_quorum_candidate_latency(GET_OBJECT_PATH_LEGACY_DUPLEX, fileinfo_selection_quorum);
+        metadata_fanout_diagnostics.record_quorum_candidate_latency(metadata_metrics_path, fileinfo_selection_quorum);
         if errs.iter().any(|err| err.is_some()) {
             let version_id = resolved_read_repair_version_id(&fi, opts.version_id.as_deref());
             submit_read_repair_heal(
@@ -427,7 +436,11 @@ impl SetDisks {
 
         // let online_disks: Vec<Option<DiskStore>> = op_online_disks.iter().filter(|v| v.is_some()).cloned().collect();
 
-        Ok((fi, parts_metadata, op_online_disks))
+        Ok((
+            GetObjectMetadata::Owned(fi),
+            GetObjectMetadata::Owned(parts_metadata),
+            GetObjectMetadata::Owned(op_online_disks),
+        ))
     }
 
     #[hotpath::measure(impl_type = "SetDisks")]
@@ -2696,6 +2709,39 @@ mod metadata_cache_tests {
         assert_eq!(cached.read_quorum, 0);
     }
 
+    #[tokio::test]
+    async fn get_object_fileinfo_cache_hit_shares_cached_metadata() {
+        let set = new_metadata_cache_test_set().await;
+        let fi = valid_test_fileinfo("object");
+        let parts_metadata = vec![fi.clone()];
+        let online_disks = Vec::new();
+        let generation = set.get_object_metadata_cache_generation("bucket", "object");
+        set.cache_get_object_fileinfo(("bucket", "object"), generation, &fi, &parts_metadata, &online_disks, 0)
+            .await;
+        let cached = set
+            .cached_get_object_fileinfo("bucket", "object")
+            .await
+            .expect("fresh cache entry should be returned");
+
+        let (returned_fi, returned_parts_metadata, returned_online_disks) = set
+            .get_object_fileinfo("bucket", "object", &ObjectOptions::default(), true, false)
+            .await
+            .expect("cache-backed metadata lookup should succeed");
+
+        assert!(
+            matches!(returned_fi, GetObjectMetadata::Shared(ref value) if Arc::ptr_eq(value, &cached.fi)),
+            "cache hits must share FileInfo ownership"
+        );
+        assert!(
+            matches!(returned_parts_metadata, GetObjectMetadata::Shared(ref value) if Arc::ptr_eq(value, &cached.parts_metadata)),
+            "cache hits must share the metadata vector"
+        );
+        assert!(
+            matches!(returned_online_disks, GetObjectMetadata::Shared(ref value) if Arc::ptr_eq(value, &cached.online_disks)),
+            "cache hits must share the online-disk vector"
+        );
+    }
+
     #[tokio::test]
     async fn get_object_metadata_cache_rejects_deleted_and_invalid_fileinfo() {
         let set = new_metadata_cache_test_set().await;
@@ -2735,9 +2781,9 @@ mod metadata_cache_tests {
                 ),
                 Arc::new(GetObjectMetadataCacheEntry {
                     created_at: Instant::now(),
-                    fi: fi.clone(),
-                    parts_metadata: vec![fi],
-                    online_disks: vec![None],
+                    fi: Arc::new(fi.clone()),
+                    parts_metadata: Arc::new(vec![fi]),
+                    online_disks: Arc::new(vec![None]),
                     read_quorum: 1,
                 }),
             )
@@ -2751,9 +2797,6 @@ mod metadata_cache_tests {
 
     #[tokio::test]
     async fn get_object_metadata_cache_rejects_stale_entries() {
-        // moka handles TTL expiry automatically via time_to_live(250ms).
-        // This test verifies that entries inserted with the cache API are retrievable
-        // while fresh, and that the cache API works correctly.
         let set = new_metadata_cache_test_set().await;
         let fi = valid_test_fileinfo("object");
 
@@ -2765,6 +2808,14 @@ mod metadata_cache_tests {
             set.cached_get_object_fileinfo("bucket", "object").await.is_some(),
             "freshly inserted entry should be returned"
         );
+
+        tokio::time::timeout(GET_OBJECT_METADATA_CACHE_TTL + Duration::from_secs(1), async {
+            while set.cached_get_object_fileinfo("bucket", "object").await.is_some() {
+                tokio::time::sleep(Duration::from_millis(10)).await;
+            }
+        })
+        .await
+        .expect("metadata cache entry should expire after its TTL");
     }
 
     #[tokio::test]
@@ -2826,9 +2877,13 @@ mod metadata_cache_tests {
         barrier.wait_until_paused().await;
         set.invalidate_get_object_metadata_cache(bucket, object).await;
         barrier.release();
-        read.await
+        let (fi, parts_metadata, online_disks) = read
+            .await
             .expect("metadata read task should not panic")
             .expect("metadata fanout should still return its selected FileInfo");
+        assert!(matches!(fi, GetObjectMetadata::Owned(_)));
+        assert!(matches!(parts_metadata, GetObjectMetadata::Owned(_)));
+        assert!(matches!(online_disks, GetObjectMetadata::Owned(_)));
 
         assert!(
             set.get_object_metadata_cache
@@ -2875,9 +2930,9 @@ mod metadata_cache_tests {
         let key = GetObjectMetadataCacheKey::new("bucket", "object", generation);
         let entry = Arc::new(GetObjectMetadataCacheEntry {
             created_at: Instant::now(),
-            fi: fi.clone(),
-            parts_metadata: vec![fi],
-            online_disks: Vec::new(),
+            fi: Arc::new(fi.clone()),
+            parts_metadata: Arc::new(vec![fi]),
+            online_disks: Arc::new(Vec::new()),
             read_quorum: 0,
         });
 
@@ -2985,9 +3040,9 @@ mod metadata_cache_tests {
         let entry = |fi: FileInfo| {
             Arc::new(GetObjectMetadataCacheEntry {
                 created_at: Instant::now(),
-                parts_metadata: vec![fi.clone()],
-                fi,
-                online_disks: Vec::new(),
+                parts_metadata: Arc::new(vec![fi.clone()]),
+                fi: Arc::new(fi),
+                online_disks: Arc::new(Vec::new()),
                 read_quorum: 0,
             })
         };
@@ -3437,7 +3492,7 @@ mod tests {
         );
         assert_eq!(diagnostics.total_responses(), 9);
         assert_eq!(diagnostics.valid_responses(), 1);
-        assert_eq!(diagnostics.error_responses(), 8);
+        assert_eq!(diagnostics.non_valid_responses(), 8);
     }
 
     #[test]
@@ -3452,7 +3507,7 @@ mod tests {
         );
 
         assert_eq!(diagnostics.ignored_responses(), 2);
-        assert_eq!(diagnostics.error_responses(), 3);
+        assert_eq!(diagnostics.non_valid_responses(), 3);
         assert_eq!(diagnostics.observations[0].outcome, GET_METADATA_RESPONSE_DISK_NOT_FOUND);
         assert_eq!(diagnostics.observations[1].outcome, GET_METADATA_RESPONSE_IGNORED);
         assert_eq!(diagnostics.observations[2].outcome, GET_METADATA_RESPONSE_NOT_FOUND);
@@ -3520,7 +3575,7 @@ mod tests {
 
         assert_eq!(diagnostics.total_responses(), 3);
         assert_eq!(diagnostics.valid_responses(), 3);
-        assert_eq!(diagnostics.error_responses(), 0);
+        assert_eq!(diagnostics.non_valid_responses(), 0);
         assert!(
             diagnostics
                 .observations
@@ -5496,33 +5551,36 @@ mod tests {
     }
 
     #[test]
-    fn rustfs_codec_streaming_uses_conservative_default_min_size() {
-        temp_env::with_vars(
-            [
-                (ENV_RUSTFS_GET_CODEC_STREAMING_ENABLE, Some("true")),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_ENGINE, Some(GET_CODEC_STREAMING_ENGINE_RUSTFS)),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_ROLLOUT, Some("benchmark")),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_BODY_COMPAT_CONFIRMED, Some("true")),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_HEADER_COMPAT_CONFIRMED, Some("true")),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, None::<&str>),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_RUSTFS_MIN_SIZE, None::<&str>),
-            ],
-            || {
-                let below_threshold_fi = codec_streaming_test_fileinfo(512 * 1024, 1);
-                let below_threshold_object_info = codec_streaming_test_object_info(&below_threshold_fi);
-                assert_eq!(
-                    codec_streaming_reader_gate_for_test(&None, &below_threshold_object_info, &below_threshold_fi, true).decision,
-                    GetCodecStreamingDecision::Fallback(GetCodecStreamingFallbackReason::BelowMinSize)
-                );
+    fn codec_streaming_default_min_size_meets_direct_memory_ceiling() {
+        for engine in [None, Some(GET_CODEC_STREAMING_ENGINE_RUSTFS)] {
+            temp_env::with_vars(
+                [
+                    (ENV_RUSTFS_GET_CODEC_STREAMING_ENABLE, Some("true")),
+                    (ENV_RUSTFS_GET_CODEC_STREAMING_ENGINE, engine),
+                    (ENV_RUSTFS_GET_CODEC_STREAMING_ROLLOUT, Some("benchmark")),
+                    (ENV_RUSTFS_GET_CODEC_STREAMING_BODY_COMPAT_CONFIRMED, Some("true")),
+                    (ENV_RUSTFS_GET_CODEC_STREAMING_HEADER_COMPAT_CONFIRMED, Some("true")),
+                    (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, None::<&str>),
+                    (ENV_RUSTFS_GET_CODEC_STREAMING_RUSTFS_MIN_SIZE, None::<&str>),
+                ],
+                || {
+                    let below_threshold_fi = codec_streaming_test_fileinfo(128 * 1024 - 1, 1);
+                    let below_threshold_object_info = codec_streaming_test_object_info(&below_threshold_fi);
+                    assert_eq!(
+                        codec_streaming_reader_gate_for_test(&None, &below_threshold_object_info, &below_threshold_fi, true)
+                            .decision,
+                        GetCodecStreamingDecision::Fallback(GetCodecStreamingFallbackReason::BelowMinSize)
+                    );
 
-                let threshold_fi = codec_streaming_test_fileinfo(1_048_576, 1);
-                let threshold_object_info = codec_streaming_test_object_info(&threshold_fi);
-                assert_eq!(
-                    codec_streaming_reader_gate_for_test(&None, &threshold_object_info, &threshold_fi, true).decision,
-                    GetCodecStreamingDecision::Use
-                );
-            },
-        );
+                    let threshold_fi = codec_streaming_test_fileinfo(128 * 1024, 1);
+                    let threshold_object_info = codec_streaming_test_object_info(&threshold_fi);
+                    assert_eq!(
+                        codec_streaming_reader_gate_for_test(&None, &threshold_object_info, &threshold_fi, true).decision,
+                        GetCodecStreamingDecision::Use
+                    );
+                },
+            );
+        }
     }
 
     #[test]
@@ -5820,10 +5878,10 @@ mod tests {
                 (ENV_RUSTFS_GET_CODEC_STREAMING_ROLLOUT, None::<&str>),
                 (ENV_RUSTFS_GET_CODEC_STREAMING_BODY_COMPAT_CONFIRMED, None::<&str>),
                 (ENV_RUSTFS_GET_CODEC_STREAMING_HEADER_COMPAT_CONFIRMED, None::<&str>),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, Some("1")),
+                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, None::<&str>),
             ],
             || {
-                let fi = codec_streaming_test_fileinfo(1024, 1);
+                let fi = codec_streaming_test_fileinfo(128 * 1024, 1);
                 let object_info = codec_streaming_test_object_info(&fi);
 
                 assert_eq!(
@@ -5844,10 +5902,10 @@ mod tests {
                 (ENV_RUSTFS_GET_CODEC_STREAMING_ROLLOUT, Some("on")),
                 (ENV_RUSTFS_GET_CODEC_STREAMING_BODY_COMPAT_CONFIRMED, None::<&str>),
                 (ENV_RUSTFS_GET_CODEC_STREAMING_HEADER_COMPAT_CONFIRMED, None::<&str>),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, Some("1")),
+                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, None::<&str>),
             ],
             || {
-                let fi = codec_streaming_test_fileinfo(1024, 1);
+                let fi = codec_streaming_test_fileinfo(128 * 1024, 1);
                 let object_info = codec_streaming_test_object_info(&fi);
 
                 assert_eq!(
@@ -5865,10 +5923,10 @@ mod tests {
             [
                 (ENV_RUSTFS_GET_CODEC_STREAMING_ENABLE, Some("false")),
                 (ENV_RUSTFS_GET_CODEC_STREAMING_ROLLOUT, Some("on")),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, Some("1")),
+                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, None::<&str>),
             ],
             || {
-                let fi = codec_streaming_test_fileinfo(1024, 1);
+                let fi = codec_streaming_test_fileinfo(128 * 1024, 1);
                 let object_info = codec_streaming_test_object_info(&fi);
 
                 assert_eq!(
@@ -5948,10 +6006,10 @@ mod tests {
                 (ENV_RUSTFS_GET_CODEC_STREAMING_ROLLOUT_PCT, Some("0")),
                 (ENV_RUSTFS_GET_CODEC_STREAMING_BODY_COMPAT_CONFIRMED, Some("true")),
                 (ENV_RUSTFS_GET_CODEC_STREAMING_HEADER_COMPAT_CONFIRMED, Some("true")),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, Some("1")),
+                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, None::<&str>),
             ],
             || {
-                let fi = codec_streaming_test_fileinfo(1024, 1);
+                let fi = codec_streaming_test_fileinfo(128 * 1024, 1);
                 let object_info = codec_streaming_test_object_info(&fi);
 
                 assert_eq!(
@@ -5968,10 +6026,10 @@ mod tests {
                 (ENV_RUSTFS_GET_CODEC_STREAMING_ROLLOUT_PCT, Some("100")),
                 (ENV_RUSTFS_GET_CODEC_STREAMING_BODY_COMPAT_CONFIRMED, Some("true")),
                 (ENV_RUSTFS_GET_CODEC_STREAMING_HEADER_COMPAT_CONFIRMED, Some("true")),
-                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, Some("1")),
+                (ENV_RUSTFS_GET_CODEC_STREAMING_MIN_SIZE, None::<&str>),
             ],
             || {
-                let fi = codec_streaming_test_fileinfo(1024, 1);
+                let fi = codec_streaming_test_fileinfo(128 * 1024, 1);
                 let object_info = codec_streaming_test_object_info(&fi);
 
                 assert_eq!(

crates/ecstore/src/set_disk/replication.rs

@@ -78,9 +78,10 @@ impl SetDisks {
             include_part_checksums: true,
             ..Default::default()
         };
-        let (mut fi, _, disks) = self
+        let (fi, _, disks) = self
             .get_object_fileinfo_gated(bucket, object, &read_opts, false, false)
             .await?;
+        let mut fi = fi.into_owned();
         if let Some(expected_operation_id) = expected_operation_id {
             require_restore_operation_id(&fi.metadata, expected_operation_id)?;
         }
@@ -102,7 +103,7 @@ impl SetDisks {
             bucket,
             object,
             fi.clone(),
-            disks.as_slice(),
+            &disks,
             &UpdateMetadataOpts {
                 replace_user_metadata: true,
                 ..Default::default()
@@ -145,9 +146,10 @@ impl SetDisks {
             include_part_checksums: true,
             ..Default::default()
         };
-        let (mut fi, _, disks) = self
+        let (fi, _, disks) = self
             .get_object_fileinfo_gated(bucket, object, &read_opts, false, false)
             .await?;
+        let mut fi = fi.into_owned();
         if let Some(expected_operation_id) = expected_operation_id {
             match restore_operation_id_from_metadata(&fi.metadata)? {
                 Some(actual_operation_id) if actual_operation_id == expected_operation_id => {}
@@ -172,7 +174,7 @@ impl SetDisks {
             bucket,
             object,
             fi,
-            disks.as_slice(),
+            &disks,
             &UpdateMetadataOpts {
                 replace_user_metadata: true,
                 ..Default::default()

crates/ecstore/src/set_disk/shard_source.rs

@@ -117,16 +117,17 @@ impl StripeReadState {
         Self::from_parts_with_read_costs(shards, errors, &[], read_quorum)
     }
 
-    pub(crate) fn from_parts_with_read_costs(
-        shards: Vec<Option<Vec<u8>>>,
-        errors: Vec<Option<Error>>,
-        read_costs: &[ShardReadCost],
-        read_quorum: usize,
-    ) -> Self {
-        let slot_count = shards.len().max(errors.len());
-        let mut slots = Vec::with_capacity(slot_count);
+    pub(crate) fn from_parts_with_read_costs<S, E>(shards: S, errors: E, read_costs: &[ShardReadCost], read_quorum: usize) -> Self
+    where
+        S: IntoIterator<Item = Option<Vec<u8>>>,
+        S::IntoIter: ExactSizeIterator,
+        E: IntoIterator<Item = Option<Error>>,
+        E::IntoIter: ExactSizeIterator,
+    {
         let mut shards = shards.into_iter();
         let mut errors = errors.into_iter();
+        let slot_count = shards.len().max(errors.len());
+        let mut slots = Vec::with_capacity(slot_count);
         for index in 0..slot_count {
             let read_cost = read_costs.get(index).copied().unwrap_or(ShardReadCost::Unknown);
             slots.push(ShardSlot::with_read_cost(

crates/ecstore/src/store/list_objects.rs

@@ -309,9 +309,17 @@ const ENV_API_LIST_OBJECTS_INDEX_PROVIDER: &str = "RUSTFS_LIST_OBJECTS_INDEX_PRO
 const ENV_API_LIST_OBJECTS_INDEX_PROVIDER_PATH: &str = "RUSTFS_LIST_OBJECTS_INDEX_PROVIDER_PATH";
 const ENV_API_LIST_OBJECTS_INDEX_PROVIDER_GENERATION: &str = "RUSTFS_LIST_OBJECTS_INDEX_PROVIDER_GENERATION";
 const ENV_API_LIST_OBJECTS_NAMESPACE_JOURNAL_PATH: &str = "RUSTFS_LIST_OBJECTS_NAMESPACE_JOURNAL_PATH";
+// The chaos machinery below is compiled only for tests and the opt-in
+// `list-chaos` feature (backlog#1832): a production binary without the
+// feature carries no chaos symbols, so the two env vars cannot silently
+// rewrite a bucket's namespace-journal state.
+#[cfg(any(test, feature = "list-chaos"))]
 const ENV_API_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_ENABLED: &str = "RUSTFS_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_ENABLED";
+#[cfg(any(test, feature = "list-chaos"))]
 const ENV_API_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_BUCKET: &str = "RUSTFS_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_BUCKET";
+#[cfg(any(test, feature = "list-chaos"))]
 const ENV_API_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_SEQUENCE: &str = "RUSTFS_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_SEQUENCE";
+#[cfg(any(test, feature = "list-chaos"))]
 const ENV_API_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_STATUS: &str = "RUSTFS_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_STATUS";
 const ENV_API_LIST_OBJECTS_METADATA_FAST_ENABLED: &str = "RUSTFS_LIST_OBJECTS_METADATA_FAST_ENABLED";
 const ENV_API_LIST_OBJECTS_METADATA_FAST_STALENESS_MS: &str = "RUSTFS_LIST_OBJECTS_METADATA_FAST_STALENESS_MS";
@@ -552,7 +560,9 @@ static LIST_OBJECTS_MUTATION_SEQUENCE: AtomicU64 = AtomicU64::new(0);
 static SCANNER_NAMESPACE_MUTATION_GENERATION: AtomicU64 = AtomicU64::new(0);
 static LIST_OBJECTS_BUCKET_MUTATION_SEQUENCE: OnceCell<RwLock<HashMap<String, u64>>> = OnceCell::const_new();
 static LIST_OBJECTS_NAMESPACE_JOURNAL_DEGRADED_BUCKETS: OnceCell<RwLock<HashSet<String>>> = OnceCell::const_new();
+#[cfg(any(test, feature = "list-chaos"))]
 static LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_CONFIG: OnceCell<Option<NamespaceMutationJournalChaosConfig>> = OnceCell::const_new();
+#[cfg(any(test, feature = "list-chaos"))]
 static LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_APPLIED: OnceCell<RwLock<HashSet<String>>> = OnceCell::const_new();
 
 async fn persistent_key_only_index_cache() -> &'static RwLock<Option<PersistentKeyOnlyIndexCache>> {
@@ -579,6 +589,7 @@ async fn list_objects_namespace_journal_degraded_buckets() -> &'static RwLock<Ha
         .await
 }
 
+#[cfg(any(test, feature = "list-chaos"))]
 async fn list_objects_namespace_journal_chaos_config() -> Option<&'static NamespaceMutationJournalChaosConfig> {
     LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_CONFIG
         .get_or_init(|| async { namespace_mutation_journal_chaos_config_from_env() })
@@ -586,6 +597,7 @@ async fn list_objects_namespace_journal_chaos_config() -> Option<&'static Namesp
         .as_ref()
 }
 
+#[cfg(any(test, feature = "list-chaos"))]
 async fn list_objects_namespace_journal_chaos_applied() -> &'static RwLock<HashSet<String>> {
     LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_APPLIED
         .get_or_init(|| async { RwLock::new(HashSet::new()) })
@@ -681,6 +693,7 @@ enum NamespaceMutationJournalStatus {
 }
 
 impl NamespaceMutationJournalStatus {
+    #[cfg(any(test, feature = "list-chaos"))]
     fn from_env_value(value: &str) -> Option<Self> {
         if value.eq_ignore_ascii_case(LIST_OBJECTS_NAMESPACE_JOURNAL_STATUS_HEALTHY) {
             Some(Self::Healthy)
@@ -691,6 +704,7 @@ impl NamespaceMutationJournalStatus {
         }
     }
 
+    #[cfg(any(test, feature = "list-chaos"))]
     fn env_value(self) -> &'static str {
         match self {
             Self::Healthy => LIST_OBJECTS_NAMESPACE_JOURNAL_STATUS_HEALTHY,
@@ -712,6 +726,7 @@ struct NamespaceMutationJournalSnapshot {
     degraded: bool,
 }
 
+#[cfg(any(test, feature = "list-chaos"))]
 #[derive(Debug, Clone, PartialEq, Eq)]
 struct NamespaceMutationJournalChaosConfig {
     bucket: String,
@@ -795,30 +810,35 @@ fn list_objects_namespace_journal_root_from_env() -> Option<PathBuf> {
         .filter(|path| !path.as_os_str().is_empty())
 }
 
+#[cfg(any(test, feature = "list-chaos"))]
 fn namespace_mutation_journal_chaos_enabled_from_env() -> bool {
     std::env::var(ENV_API_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_ENABLED)
         .ok()
         .is_some_and(|value| value == "1" || value.eq_ignore_ascii_case("on") || value.eq_ignore_ascii_case("true"))
 }
 
+#[cfg(any(test, feature = "list-chaos"))]
 fn namespace_mutation_journal_chaos_bucket_from_env() -> Option<String> {
     std::env::var(ENV_API_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_BUCKET)
         .ok()
         .filter(|bucket| !bucket.is_empty())
 }
 
+#[cfg(any(test, feature = "list-chaos"))]
 fn namespace_mutation_journal_chaos_sequence_from_env() -> Option<u64> {
     std::env::var(ENV_API_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_SEQUENCE)
         .ok()
         .and_then(|value| value.parse::<u64>().ok())
 }
 
+#[cfg(any(test, feature = "list-chaos"))]
 fn namespace_mutation_journal_chaos_status_from_env() -> Option<NamespaceMutationJournalStatus> {
     std::env::var(ENV_API_LIST_OBJECTS_NAMESPACE_JOURNAL_CHAOS_STATUS)
         .ok()
         .and_then(|value| NamespaceMutationJournalStatus::from_env_value(&value))
 }
 
+#[cfg(any(test, feature = "list-chaos"))]
 fn namespace_mutation_journal_chaos_config_from_env() -> Option<NamespaceMutationJournalChaosConfig> {
     if !namespace_mutation_journal_chaos_enabled_from_env() {
         return None;
@@ -846,6 +866,7 @@ fn namespace_mutation_journal_chaos_config_from_env() -> Option<NamespaceMutatio
     })
 }
 
+#[cfg(any(test, feature = "list-chaos"))]
 fn namespace_mutation_journal_chaos_applied_key(bucket: &str, status: NamespaceMutationJournalStatus) -> String {
     let mut key = String::with_capacity(bucket.len() + 1 + status.env_value().len());
     key.push_str(bucket);
@@ -854,6 +875,13 @@ fn namespace_mutation_journal_chaos_applied_key(bucket: &str, status: NamespaceM
     key
 }
 
+/// Production no-op twin of the chaos injector: without `list-chaos` the
+/// injection point compiles to nothing (backlog#1832).
+#[cfg(not(any(test, feature = "list-chaos")))]
+#[inline]
+async fn maybe_apply_system_namespace_mutation_journal_chaos(_store: &ECStore, _bucket: &str, _default_sequence: u64) {}
+
+#[cfg(any(test, feature = "list-chaos"))]
 async fn maybe_apply_system_namespace_mutation_journal_chaos(store: &ECStore, bucket: &str, default_sequence: u64) {
     let Some(config) = list_objects_namespace_journal_chaos_config().await else {
         return;
@@ -9529,294 +9557,6 @@ mod test {
             .expect("a partial outage with a healthy set must not fail the walk");
     }
 
-    // use std::sync::Arc;
-
-    // use crate::cache_value::metacache_set::list_path_raw;
-    // use crate::cache_value::metacache_set::ListPathRawOptions;
-    // use crate::disk::endpoint::Endpoint;
-    // use crate::disk::error::is_err_eof;
-    // use crate::disk::format::FormatV3;
-    // use crate::disk::new_disk;
-    // use crate::disk::DiskAPI;
-    // use crate::disk::DiskOption;
-    // use crate::disk::MetaCacheEntries;
-    // use crate::disk::MetaCacheEntry;
-    // use crate::disk::WalkDirOptions;
-    // use crate::layout::endpoints::EndpointServerPools;
-    // use crate::error::Error;
-    // use crate::metacache::writer::MetacacheReader;
-    // use crate::set_disk::SetDisks;
-    // use crate::store::list_objects::ListPathOptions;
-    // use crate::store::list_objects::WalkOptions;
-    // use crate::store::list_objects::WalkVersionsSortOrder;
-    // use futures::future::join_all;
-    // use rustfs_lock::namespace_lock::NsLockMap;
-    // use tokio::sync::broadcast;
-    // use tokio::sync::mpsc;
-    // use tokio::sync::RwLock;
-    // use uuid::Uuid;
-
-    // #[tokio::test]
-    // async fn test_walk_dir() {
-    //     let mut ep = Endpoint::try_from("/Users/weisd/project/weisd/s3-rustfs/target/volume/test").unwrap();
-    //     ep.pool_idx = 0;
-    //     ep.set_idx = 0;
-    //     ep.disk_idx = 0;
-    //     ep.is_local = true;
-
-    //     let disk = new_disk(&ep, &DiskOption::default()).await.expect("init disk fail");
-
-    //     // let disk = match LocalDisk::new(&ep, false).await {
-    //     //     Ok(res) => res,
-    //     //     Err(err) => {
-    //     //         println!("LocalDisk::new err {:?}", err);
-    //     //         return;
-    //     //     }
-    //     // };
-
-    //     let (rd, mut wr) = tokio::io::duplex(64);
-
-    //     let job = tokio::spawn(async move {
-    //         let opts = WalkDirOptions {
-    //             bucket: "dada".to_owned(),
-    //             base_dir: "".to_owned(),
-    //             recursive: true,
-    //             ..Default::default()
-    //         };
-
-    //         println!("walk opts {:?}", opts);
-    //         if let Err(err) = disk.walk_dir(opts, &mut wr).await {
-    //             println!("walk_dir err {:?}", err);
-    //         }
-    //     });
-
-    //     let job2 = tokio::spawn(async move {
-    //         let mut mrd = MetacacheReader::new(rd);
-
-    //         loop {
-    //             match mrd.peek().await {
-    //                 Ok(res) => {
-    //                     if let Some(info) = res {
-    //                         println!("info {:?}", info.name)
-    //                     } else {
-    //                         break;
-    //                     }
-    //                 }
-    //                 Err(err) => {
-    //                     if is_err_eof(&err) {
-    //                         break;
-    //                     }
-
-    //                     println!("get err {:?}", err);
-    //                     break;
-    //                 }
-    //             }
-    //         }
-    //     });
-    //     join_all(vec![job, job2]).await;
-    // }
-
-    // #[tokio::test]
-    // async fn test_list_path_raw() {
-    //     let mut ep = Endpoint::try_from("/Users/weisd/project/weisd/s3-rustfs/target/volume/test").unwrap();
-    //     ep.pool_idx = 0;
-    //     ep.set_idx = 0;
-    //     ep.disk_idx = 0;
-    //     ep.is_local = true;
-
-    //     let disk = new_disk(&ep, &DiskOption::default()).await.expect("init disk fail");
-
-    //     // let disk = match LocalDisk::new(&ep, false).await {
-    //     //     Ok(res) => res,
-    //     //     Err(err) => {
-    //     //         println!("LocalDisk::new err {:?}", err);
-    //     //         return;
-    //     //     }
-    //     // };
-
-    //     let (_, rx) = broadcast::channel(1);
-    //     let bucket = "dada".to_owned();
-    //     let forward_to = None;
-    //     let disks = vec![Some(disk)];
-    //     let fallback_disks = Vec::new();
-
-    //     list_path_raw(
-    //         rx,
-    //         ListPathRawOptions {
-    //             disks,
-    //             fallback_disks,
-    //             bucket,
-    //             path: "".to_owned(),
-    //             recursice: true,
-    //             forward_to,
-    //             min_disks: 1,
-    //             report_not_found: false,
-    //             agreed: Some(Box::new(move |entry: MetaCacheEntry| {
-    //                 Box::pin(async move { println!("get entry: {}", entry.name) })
-    //             })),
-    //             partial: Some(Box::new(move |entries: MetaCacheEntries, _: &[Option<Error>]| {
-    //                 Box::pin(async move { println!("get entries: {:?}", entries) })
-    //             })),
-    //             finished: None,
-    //             ..Default::default()
-    //         },
-    //     )
-    //     .await
-    //     .unwrap();
-    // }
-
-    // #[tokio::test]
-    // async fn test_set_list_path() {
-    //     let mut ep = Endpoint::try_from("/Users/weisd/project/weisd/s3-rustfs/target/volume/test").unwrap();
-    //     ep.pool_idx = 0;
-    //     ep.set_idx = 0;
-    //     ep.disk_idx = 0;
-    //     ep.is_local = true;
-
-    //     let disk = new_disk(&ep, &DiskOption::default()).await.expect("init disk fail");
-    //     let _ = disk.set_disk_id(Some(Uuid::new_v4())).await;
-
-    //     let set = SetDisks {
-    //         lockers: Vec::new(),
-    //         locker_owner: String::new(),
-    //         ns_mutex: Arc::new(RwLock::new(NsLockMap::new(false))),
-    //         disks: RwLock::new(vec![Some(disk)]),
-    //         set_endpoints: Vec::new(),
-    //         set_drive_count: 1,
-    //         default_parity_count: 0,
-    //         set_index: 0,
-    //         pool_index: 0,
-    //         format: FormatV3::new(1, 1),
-    //     };
-
-    //     let (_tx, rx) = broadcast::channel(1);
-
-    //     let bucket = "dada".to_owned();
-
-    //     let opts = ListPathOptions {
-    //         bucket,
-    //         recursive: true,
-    //         ..Default::default()
-    //     };
-
-    //     let (sender, mut recv) = mpsc::channel(10);
-
-    //     set.list_path(rx, opts, sender).await.unwrap();
-
-    //     while let Some(entry) = recv.recv().await {
-    //         println!("get entry {:?}", entry.name)
-    //     }
-    // }
-
-    // #[tokio::test]
-    //walk() {
-    //     let server_address = "localhost:9000";
-
-    //     let (endpoint_pools, _setup_type) = EndpointServerPools::from_volumes(
-    //         server_address,
-    //         vec!["/Users/weisd/project/weisd/s3-rustfs/target/volume/test".to_string()],
-    //     )
-    //     .unwrap();
-
-    //     let store = ECStore::new(server_address.to_string(), endpoint_pools.clone())
-    //         .await
-    //         .unwrap();
-
-    //     let (_tx, rx) = broadcast::channel(1);
-
-    //     let bucket = "dada".to_owned();
-    //     let opts = ListPathOptions {
-    //         bucket,
-    //         recursive: true,
-    //         ..Default::default()
-    //     };
-
-    //     let (sender, mut recv) = mpsc::channel(10);
-
-    //     store.list_merged(rx, opts, sender).await.unwrap();
-
-    //     while let Some(entry) = recv.recv().await {
-    //         println!("get entry {:?}", entry.name)
-    //     }
-    // }
-
-    // #[tokio::test]
-    // async fn test_list_path() {
-    //     let server_address = "localhost:9000";
-
-    //     let (endpoint_pools, _setup_type) = EndpointServerPools::from_volumes(
-    //         server_address,
-    //         vec!["/Users/weisd/project/weisd/s3-rustfs/target/volume/test".to_string()],
-    //     )
-    //     .unwrap();
-
-    //     let store = ECStore::new(server_address.to_string(), endpoint_pools.clone())
-    //         .await
-    //         .unwrap();
-
-    //     let bucket = "dada".to_owned();
-    //     let opts = ListPathOptions {
-    //         bucket,
-    //         recursive: true,
-    //         limit: 100,
-
-    //         ..Default::default()
-    //     };
-
-    //     let ret = store.list_path(&opts).await.unwrap();
-    //     println!("ret {:?}", ret);
-    // }
-
-    // #[tokio::test]
-    // async fn test_list_objects_v2() {
-    //     let server_address = "localhost:9000";
-
-    //     let (endpoint_pools, _setup_type) = EndpointServerPools::from_volumes(
-    //         server_address,
-    //         vec!["/Users/weisd/project/weisd/s3-rustfs/target/volume/test".to_string()],
-    //     )
-    //     .unwrap();
-
-    //     let store = ECStore::new(server_address.to_string(), endpoint_pools.clone())
-    //         .await
-    //         .unwrap();
-
-    //     let ret = store.list_objects_v2("data", "", "", "", 100, false, "").await.unwrap();
-    //     println!("ret {:?}", ret);
-    // }
-
-    // #[tokio::test]
-    // async fn test_walk() {
-    //     let server_address = "localhost:9000";
-
-    //     let (endpoint_pools, _setup_type) = EndpointServerPools::from_volumes(
-    //         server_address,
-    //         vec!["/Users/weisd/project/weisd/s3-rustfs/target/volume/test".to_string()],
-    //     )
-    //     .unwrap();
-
-    //     let store = ECStore::new(server_address.to_string(), endpoint_pools.clone())
-    //         .await
-    //         .unwrap();
-
-    //     ECStore::init(store.clone()).await.unwrap();
-
-    //     let (_tx, rx) = broadcast::channel(1);
-
-    //     let bucket = ".rustfs.sys";
-    //     let prefix = "config/iam/sts/";
-
-    //     let (sender, mut recv) = mpsc::channel(10);
-
-    //     let opts = WalkOptions::default();
-
-    //     store.walk(rx, bucket, prefix, sender, opts).await.unwrap();
-
-    //     while let Some(entry) = recv.recv().await {
-    //         println!("get entry {:?}", entry)
-    //     }
-    // }
-
     #[tokio::test]
     async fn merge_entry_channels_produces_sorted_unique_output_from_two_channels() {
         let (tx_a, rx_a) = mpsc::channel(4);

crates/ecstore/src/store/object.rs

@@ -1781,7 +1781,7 @@ impl ECStore {
     ) -> Result<GetObjectReader> {
         check_get_obj_args(bucket, object)?;
 
-        let object = encode_dir_object(object);
+        let object = rustfs_utils::path::encode_dir_object_ref(object);
         let mut opts = opts.clone();
         let read_lock_guard = self
             .acquire_object_read_lock_if_needed("get_object", bucket, &object, &mut opts)
@@ -1789,14 +1789,14 @@ impl ECStore {
 
         let reader = if self.single_pool() {
             self.pools[0]
-                .get_object_reader(bucket, object.as_str(), range, h, &opts)
+                .get_object_reader(bucket, object.as_ref(), range, h, &opts)
                 .await?
         } else {
             let (_, idx) = self
                 .get_latest_accessible_object_info_with_idx(bucket, &object, &opts)
                 .await?;
             self.pools[idx]
-                .get_object_reader(bucket, object.as_str(), range, h, &opts)
+                .get_object_reader(bucket, object.as_ref(), range, h, &opts)
                 .await?
         };