Merge branch 'main' into cxymds/fix-1927-durable-checkpoint

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

.github/workflows/nightly-gnu.yml

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

Cargo.lock

@@ -9564,6 +9564,7 @@ dependencies = [
  "serde",
  "serde_json",
  "serial_test",
+ "sha2 0.11.0",
  "temp-env",
  "tempfile",
  "thiserror 2.0.20",

crates/ecstore/src/api/mod.rs

@@ -440,12 +440,6 @@ pub mod rebalance {
         RebalanceMeta, RebalanceStats, RebalanceStopPropagationRecord, decode_rebalance_stop_propagation_record,
         encode_rebalance_stop_propagation_record,
     };
-
-    #[cfg(feature = "test-util")]
-    pub mod test_util {
-        pub use crate::services::rebalance::PausedRebalanceEntryTestFixture;
-        pub use crate::services::rebalance::test_store_with_persisted_rebalance_meta;
-    }
 }
 
 pub mod rio {

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

@@ -4324,16 +4324,6 @@ pub async fn expire_transitioned_object(
     lc_event: &lifecycle::Event,
     _src: &LcEventSrc,
     bucket_incarnation_id: Uuid,
-) -> Result<ObjectInfo, std::io::Error> {
-    expire_transitioned_object_with_lock_lost_signal(api, oi, lc_event, bucket_incarnation_id, None).await
-}
-
-async fn expire_transitioned_object_with_lock_lost_signal(
-    api: Arc<ECStore>,
-    oi: &ObjectInfo,
-    lc_event: &lifecycle::Event,
-    bucket_incarnation_id: Uuid,
-    lock_lost_signal: Option<Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
 ) -> Result<ObjectInfo, std::io::Error> {
     let publication_guard = lifecycle_expiry_publication_guard(&api, oi, bucket_incarnation_id)
         .await
@@ -4345,9 +4335,6 @@ async fn expire_transitioned_object_with_lock_lost_signal(
     let mut opts = transitioned_object_delete_opts(oi, lc_event.action, versioned, version_suspended, bucket_incarnation_id)
         .map_err(std::io::Error::other)?;
     opts.add_namespace_lock_guard(&publication_guard);
-    if let Some(signal) = lock_lost_signal {
-        opts.add_namespace_lock_lost_signal(signal);
-    }
     opts.delete_replication_config_snapshot = Some(Arc::new(snapshot));
     //let tags = LcAuditEvent::new(src, lcEvent).Tags();
     if lc_event.action.delete_restored() {
@@ -5006,32 +4993,12 @@ pub async fn apply_expiry_on_transitioned_object(
     lc_event: &lifecycle::Event,
     src: &LcEventSrc,
     bucket_incarnation_id: Uuid,
-) -> bool {
-    apply_expiry_on_transitioned_object_with_lock_lost_signal(api, oi, lc_event, src, bucket_incarnation_id, None).await
-}
-
-async fn apply_expiry_on_transitioned_object_with_lock_lost_signal(
-    api: Arc<ECStore>,
-    oi: &ObjectInfo,
-    lc_event: &lifecycle::Event,
-    _src: &LcEventSrc,
-    bucket_incarnation_id: Uuid,
-    lock_lost_signal: Option<Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
 ) -> bool {
     if lc_event.action.delete_all() {
-        return apply_expiry_on_non_transitioned_objects_with_lock_lost_signal(
-            api,
-            oi,
-            lc_event,
-            bucket_incarnation_id,
-            lock_lost_signal,
-        )
-        .await;
+        return apply_expiry_on_non_transitioned_objects(api, oi, lc_event, src, bucket_incarnation_id).await;
     }
     let time_ilm = Metrics::time_ilm(lc_event.action);
-    if let Err(_err) =
-        expire_transitioned_object_with_lock_lost_signal(api, oi, lc_event, bucket_incarnation_id, lock_lost_signal).await
-    {
+    if let Err(_err) = expire_transitioned_object(api, oi, lc_event, src, bucket_incarnation_id).await {
         return false;
     }
     time_ilm(1)();
@@ -5045,16 +5012,6 @@ pub async fn apply_expiry_on_non_transitioned_objects(
     lc_event: &lifecycle::Event,
     _src: &LcEventSrc,
     bucket_incarnation_id: Uuid,
-) -> bool {
-    apply_expiry_on_non_transitioned_objects_with_lock_lost_signal(api, oi, lc_event, bucket_incarnation_id, None).await
-}
-
-async fn apply_expiry_on_non_transitioned_objects_with_lock_lost_signal(
-    api: Arc<ECStore>,
-    oi: &ObjectInfo,
-    lc_event: &lifecycle::Event,
-    bucket_incarnation_id: Uuid,
-    lock_lost_signal: Option<Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
 ) -> bool {
     let Some(publication_guard) = lifecycle_expiry_publication_guard(&api, oi, bucket_incarnation_id).await else {
         return false;
@@ -5085,9 +5042,6 @@ async fn apply_expiry_on_non_transitioned_objects_with_lock_lost_signal(
         ..Default::default()
     };
     opts.add_namespace_lock_guard(&publication_guard);
-    if let Some(signal) = lock_lost_signal {
-        opts.add_namespace_lock_lost_signal(signal);
-    }
 
     if lc_event.action.delete_versioned() {
         opts.version_id = oi.version_id.map(|v| v.to_string());
@@ -5169,61 +5123,6 @@ async fn enqueue_expiry_rule_with_incarnation(
     expiry_state.enqueue_by_days(oi, event, src, bucket_incarnation_id)
 }
 
-fn lifecycle_expiry_object_matches(current: &ObjectInfo, expected: &ObjectInfo) -> bool {
-    current.version_id == expected.version_id
-        && current.data_dir == expected.data_dir
-        && current.mod_time == expected.mod_time
-        && current.etag == expected.etag
-        && current.delete_marker == expected.delete_marker
-        && current.transitioned_object.name == expected.transitioned_object.name
-        && current.transitioned_object.version_id == expected.transitioned_object.version_id
-        && current.transitioned_object.tier == expected.transitioned_object.tier
-        && current.transitioned_object.status == expected.transitioned_object.status
-        && current.restore_expires == expected.restore_expires
-}
-
-pub(crate) async fn apply_expiry_rule_for_data_movement(
-    api: Arc<ECStore>,
-    event: &lifecycle::Event,
-    src: &LcEventSrc,
-    oi: &ObjectInfo,
-    lock_lost_signal: Option<Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
-) -> bool {
-    let Ok(_lifecycle_guard) = api.acquire_bucket_lifecycle_read_lock(&oi.bucket).await else {
-        return false;
-    };
-    let Ok(bucket_incarnation_id) = api.bucket_incarnation_id_from_disk(&oi.bucket).await else {
-        return false;
-    };
-    let current = match api
-        .get_object_info(
-            &oi.bucket,
-            &oi.name,
-            &ObjectOptions {
-                version_id: oi.version_id.map(|version_id| version_id.to_string()),
-                versioned: oi.version_id.is_some(),
-                expected_bucket_incarnation_id: Some(bucket_incarnation_id),
-                ..Default::default()
-            },
-        )
-        .await
-    {
-        Ok(current) => current,
-        Err(_) => return false,
-    };
-    if !lifecycle_expiry_object_matches(&current, oi) {
-        return false;
-    }
-
-    if oi.transitioned_object.status.is_empty() {
-        apply_expiry_on_non_transitioned_objects_with_lock_lost_signal(api, oi, event, bucket_incarnation_id, lock_lost_signal)
-            .await
-    } else {
-        apply_expiry_on_transitioned_object_with_lock_lost_signal(api, oi, event, src, bucket_incarnation_id, lock_lost_signal)
-            .await
-    }
-}
-
 pub(crate) async fn apply_expiry_rule_in(api: Arc<ECStore>, event: &lifecycle::Event, src: &LcEventSrc, oi: &ObjectInfo) -> bool {
     let Ok(_lifecycle_guard) = api.acquire_bucket_lifecycle_read_lock(&oi.bucket).await else {
         return false;
@@ -5247,7 +5146,17 @@ pub(crate) async fn apply_expiry_rule_in(api: Arc<ECStore>, event: &lifecycle::E
         Ok(current) => current,
         Err(_) => return false,
     };
-    if !lifecycle_expiry_object_matches(&current, oi) {
+    if current.version_id != oi.version_id
+        || current.data_dir != oi.data_dir
+        || current.mod_time != oi.mod_time
+        || current.etag != oi.etag
+        || current.delete_marker != oi.delete_marker
+        || current.transitioned_object.name != oi.transitioned_object.name
+        || current.transitioned_object.version_id != oi.transitioned_object.version_id
+        || current.transitioned_object.tier != oi.transitioned_object.tier
+        || current.transitioned_object.status != oi.transitioned_object.status
+        || current.restore_expires != oi.restore_expires
+    {
         return false;
     }
     enqueue_expiry_rule_with_incarnation(event, src, oi, bucket_incarnation_id).await

crates/ecstore/src/core/pools.rs

@@ -19,8 +19,8 @@ use crate::bucket::{
         LifecycleExpiryConfigs,
         bucket_lifecycle_audit::LcEventSrc,
         bucket_lifecycle_ops::{
-            LifecycleOps, apply_expiry_on_transitioned_object, apply_expiry_rule_for_data_movement, apply_expiry_rule_in,
-            eval_action_from_lifecycle, lifecycle_delete_all_versions_blocked_by_replication,
+            LifecycleOps, apply_expiry_on_transitioned_object, apply_expiry_rule_in, eval_action_from_lifecycle,
+            lifecycle_delete_all_versions_blocked_by_replication,
         },
         get_expiry_configs,
         lifecycle::IlmAction,
@@ -28,9 +28,7 @@ use crate::bucket::{
     metadata_sys,
 };
 use crate::cache_value::metacache_set::{ListPathRawOptions, list_path_raw};
-use crate::config::com::{
-    CONFIG_PREFIX, read_config, read_config_no_lock, save_config, save_config_with_opts, save_config_with_opts_quiet,
-};
+use crate::config::com::{CONFIG_PREFIX, read_config, read_config_no_lock, save_config, save_config_with_opts};
 use crate::data_movement;
 use crate::data_movement::backpressure::{self, DataMovementOperation};
 use crate::data_usage::DATA_USAGE_CACHE_NAME;
@@ -50,6 +48,7 @@ use crate::storage_api_contracts::{
     admin::StorageAdminApi,
     bucket::{BucketOperations, BucketOptions, MakeBucketOptions},
     heal::HealOperations as _,
+    namespace::NamespaceLocking as _,
     object::{EcstoreObjectIO, ObjectIO as _, ObjectOperations as _},
 };
 use crate::{core::sets::Sets, store::ECStore};
@@ -1087,7 +1086,7 @@ fn resolve_start_decommission_pool_meta_reload_result(result: Result<()>) -> Res
     resolve_decommission_pool_meta_reload_result(result, "start_decommission")
 }
 
-fn activation_rebalance_meta_lock_error(err: rustfs_lock::LockError) -> Error {
+fn decommission_rebalance_meta_lock_error(err: rustfs_lock::LockError) -> Error {
     match err {
         rustfs_lock::LockError::QuorumNotReached { required, achieved } => Error::NamespaceLockQuorumUnavailable {
             mode: "write",
@@ -1097,12 +1096,12 @@ fn activation_rebalance_meta_lock_error(err: rustfs_lock::LockError) -> Error {
             achieved,
         },
         other => Error::other(format!(
-            "failed to acquire rebalance activation lock on {RUSTFS_META_BUCKET}/{REBAL_META_NAME}: {other}"
+            "failed to acquire rebalance metadata write lock before decommission start on {RUSTFS_META_BUCKET}/{REBAL_META_NAME}: {other}"
         )),
     }
 }
 
-fn activation_pool_meta_lock_error(err: rustfs_lock::LockError) -> Error {
+fn decommission_pool_meta_lock_error(err: rustfs_lock::LockError) -> Error {
     match err {
         rustfs_lock::LockError::QuorumNotReached { required, achieved } => Error::NamespaceLockQuorumUnavailable {
             mode: "write",
@@ -1112,227 +1111,11 @@ fn activation_pool_meta_lock_error(err: rustfs_lock::LockError) -> Error {
             achieved,
         },
         other => Error::other(format!(
-            "failed to acquire pool activation lock on {RUSTFS_META_BUCKET}/{POOL_META_NAME}: {other}"
+            "failed to acquire pool metadata write lock before decommission start on {RUSTFS_META_BUCKET}/{POOL_META_NAME}: {other}"
         )),
     }
 }
 
-pub(crate) struct PoolRebalanceActivationFence {
-    pool_meta_guard: rustfs_lock::NamespaceLockGuard,
-    rebalance_meta_guard: rustfs_lock::NamespaceLockGuard,
-    #[cfg(test)]
-    forced_lost: Arc<AtomicBool>,
-}
-
-impl PoolRebalanceActivationFence {
-    pub(crate) fn ensure_held(&self) -> Result<()> {
-        #[cfg(test)]
-        let forced_lost = self.forced_lost.load(Ordering::Acquire);
-        #[cfg(not(test))]
-        let forced_lost = false;
-        if forced_lost || self.pool_meta_guard.is_lock_lost() || self.rebalance_meta_guard.is_lock_lost() {
-            return Err(Error::other("activation lock lost before metadata commit or worker admission"));
-        }
-
-        Ok(())
-    }
-
-    pub(crate) fn add_namespace_lock_fence(&self, opts: &mut ObjectOptions) {
-        opts.add_namespace_lock_guard(&self.pool_meta_guard);
-        opts.add_namespace_lock_guard(&self.rebalance_meta_guard);
-    }
-
-    #[cfg(test)]
-    fn force_lost_for_test(&self) {
-        self.forced_lost.store(true, Ordering::Release);
-    }
-}
-
-pub(crate) async fn acquire_pool_rebalance_activation_locks<S>(pool: Arc<S>) -> Result<PoolRebalanceActivationFence>
-where
-    S: crate::storage_api_contracts::namespace::NamespaceLocking<
-            Error = Error,
-            NamespaceLock = rustfs_lock::NamespaceLockWrapper,
-        >,
-{
-    // Activation lock order is always pool.bin -> rebalance.bin.
-    let pool_meta_lock = pool.new_ns_lock(RUSTFS_META_BUCKET, POOL_META_NAME).await?;
-    let pool_meta_guard = pool_meta_lock
-        .get_write_lock(get_lock_acquire_timeout())
-        .await
-        .map_err(activation_pool_meta_lock_error)?;
-    let rebalance_meta_lock = pool.new_ns_lock(RUSTFS_META_BUCKET, REBAL_META_NAME).await?;
-    let rebalance_meta_guard = rebalance_meta_lock
-        .get_write_lock(get_lock_acquire_timeout())
-        .await
-        .map_err(activation_rebalance_meta_lock_error)?;
-
-    Ok(PoolRebalanceActivationFence {
-        pool_meta_guard,
-        rebalance_meta_guard,
-        #[cfg(test)]
-        forced_lost: Arc::new(AtomicBool::new(false)),
-    })
-}
-
-#[cfg(test)]
-#[derive(Clone, Copy, Debug, Eq, PartialEq)]
-pub(crate) enum PoolActivationStartKind {
-    Rebalance,
-    Decommission,
-}
-
-#[cfg(test)]
-struct PoolActivationDurableSaveBarrierState {
-    pool_key: usize,
-    arrived: tokio::sync::Notify,
-    release: tokio::sync::Notify,
-}
-
-#[cfg(test)]
-static POOL_ACTIVATION_DURABLE_SAVE_BARRIER: std::sync::OnceLock<
-    std::sync::Mutex<Option<Arc<PoolActivationDurableSaveBarrierState>>>,
-> = std::sync::OnceLock::new();
-
-#[cfg(test)]
-pub(crate) struct PoolActivationDurableSaveBarrier {
-    state: Arc<PoolActivationDurableSaveBarrierState>,
-}
-
-#[cfg(test)]
-fn pool_activation_test_pool_key<S>(pool: &Arc<S>) -> usize {
-    Arc::as_ptr(pool).cast::<()>() as usize
-}
-
-#[cfg(test)]
-impl PoolActivationDurableSaveBarrier {
-    pub(crate) fn install<S>(pool: &Arc<S>) -> Self {
-        let state = Arc::new(PoolActivationDurableSaveBarrierState {
-            pool_key: pool_activation_test_pool_key(pool),
-            arrived: tokio::sync::Notify::new(),
-            release: tokio::sync::Notify::new(),
-        });
-        let mut barrier = POOL_ACTIVATION_DURABLE_SAVE_BARRIER
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("pool activation durable save barrier should not be poisoned");
-        assert!(barrier.is_none(), "pool activation durable save barrier must be unique");
-        *barrier = Some(Arc::clone(&state));
-        Self { state }
-    }
-
-    pub(crate) async fn wait_until_paused(&self) {
-        tokio::time::timeout(std::time::Duration::from_secs(30), self.state.arrived.notified())
-            .await
-            .expect("activation should reach the post-durable-save barrier");
-    }
-
-    pub(crate) fn release_after_fence_loss(&self) {
-        self.state.release.notify_one();
-    }
-}
-
-#[cfg(test)]
-impl Drop for PoolActivationDurableSaveBarrier {
-    fn drop(&mut self) {
-        self.state.release.notify_one();
-        let mut barrier = POOL_ACTIVATION_DURABLE_SAVE_BARRIER
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("pool activation durable save barrier should not be poisoned");
-        if barrier.as_ref().is_some_and(|state| Arc::ptr_eq(state, &self.state)) {
-            *barrier = None;
-        }
-    }
-}
-
-#[cfg(test)]
-pub(crate) async fn pause_pool_activation_after_durable_save<S>(pool: &Arc<S>, fence: &PoolRebalanceActivationFence) {
-    let pool_key = pool_activation_test_pool_key(pool);
-    let barrier = {
-        let mut barrier = POOL_ACTIVATION_DURABLE_SAVE_BARRIER
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("pool activation durable save barrier should not be poisoned");
-        if barrier.as_ref().is_some_and(|state| state.pool_key == pool_key) {
-            barrier.take()
-        } else {
-            None
-        }
-    };
-    if let Some(barrier) = barrier {
-        barrier.arrived.notify_one();
-        barrier.release.notified().await;
-        fence.force_lost_for_test();
-    }
-}
-
-#[cfg(test)]
-struct PoolActivationStartProbeState {
-    kind: PoolActivationStartKind,
-    attempted: std::sync::atomic::AtomicBool,
-    notify: tokio::sync::Notify,
-}
-
-#[cfg(test)]
-static POOL_ACTIVATION_START_PROBES: std::sync::OnceLock<std::sync::Mutex<Vec<Arc<PoolActivationStartProbeState>>>> =
-    std::sync::OnceLock::new();
-
-#[cfg(test)]
-pub(crate) struct PoolActivationStartProbe {
-    state: Arc<PoolActivationStartProbeState>,
-}
-
-#[cfg(test)]
-impl PoolActivationStartProbe {
-    pub(crate) fn install(kind: PoolActivationStartKind) -> Self {
-        let state = Arc::new(PoolActivationStartProbeState {
-            kind,
-            attempted: std::sync::atomic::AtomicBool::new(false),
-            notify: tokio::sync::Notify::new(),
-        });
-        POOL_ACTIVATION_START_PROBES
-            .get_or_init(|| std::sync::Mutex::new(Vec::new()))
-            .lock()
-            .expect("pool activation start probe should not be poisoned")
-            .push(Arc::clone(&state));
-        Self { state }
-    }
-
-    pub(crate) async fn wait_until_attempted(&self) {
-        while !self.state.attempted.load(Ordering::Acquire) {
-            self.state.notify.notified().await;
-        }
-    }
-}
-
-#[cfg(test)]
-impl Drop for PoolActivationStartProbe {
-    fn drop(&mut self) {
-        let mut probes = POOL_ACTIVATION_START_PROBES
-            .get_or_init(|| std::sync::Mutex::new(Vec::new()))
-            .lock()
-            .expect("pool activation start probe should not be poisoned");
-        probes.retain(|state| !Arc::ptr_eq(state, &self.state));
-    }
-}
-
-#[cfg(test)]
-pub(crate) fn observe_pool_activation_start_attempt(kind: PoolActivationStartKind) {
-    let probes = POOL_ACTIVATION_START_PROBES
-        .get_or_init(|| std::sync::Mutex::new(Vec::new()))
-        .lock()
-        .expect("pool activation start probe should not be poisoned")
-        .iter()
-        .filter(|state| state.kind == kind)
-        .cloned()
-        .collect::<Vec<_>>();
-    for state in probes {
-        state.attempted.store(true, Ordering::Release);
-        state.notify.notify_one();
-    }
-}
-
 fn rollback_decommission_pool_meta(pool_meta: &mut PoolMeta, previous_pool_meta: PoolMeta) {
     *pool_meta = previous_pool_meta;
 }
@@ -2317,67 +2100,6 @@ impl PoolMeta {
         Ok(())
     }
 
-    async fn save_no_lock_with_activation_fence<S>(
-        &self,
-        pools: Vec<Arc<S>>,
-        activation_fence: PoolRebalanceActivationFence,
-    ) -> Result<()>
-    where
-        S: EcstoreObjectIO,
-    {
-        let data = self.encode_config_data()?;
-        if data.is_empty() {
-            return Ok(());
-        }
-        let mut pools = pools.into_iter();
-        let Some(canonical_pool) = pools.next() else {
-            return Ok(());
-        };
-        let mut opts = ObjectOptions {
-            max_parity: true,
-            no_lock: true,
-            ..Default::default()
-        };
-        activation_fence.add_namespace_lock_fence(&mut opts);
-        activation_fence.ensure_held()?;
-        #[cfg(test)]
-        let barrier_pool = canonical_pool.clone();
-        save_config_with_opts(canonical_pool, POOL_META_NAME, data.clone(), &opts).await?;
-        #[cfg(test)]
-        pause_pool_activation_after_durable_save(&barrier_pool, &activation_fence).await;
-
-        // Pool zero is canonical. Once its save succeeds, later writes only
-        // replicate committed state and must not reuse the admission fence.
-        // Failed replicas remain repairable by the next full pool metadata save.
-        drop(activation_fence);
-        for (pool_index, pool) in pools.enumerate() {
-            if let Err(err) = save_config_with_opts_quiet(
-                pool,
-                POOL_META_NAME,
-                data.clone(),
-                &ObjectOptions {
-                    max_parity: true,
-                    no_lock: true,
-                    ..Default::default()
-                },
-            )
-            .await
-            {
-                warn!(
-                    event = EVENT_DECOMMISSION_STATE,
-                    component = LOG_COMPONENT_ECSTORE,
-                    subsystem = LOG_SUBSYSTEM_POOLS,
-                    pool_index = pool_index + 1,
-                    state = "activation_replica_repair_pending",
-                    error = %err,
-                    "Decommission activation replica repair pending"
-                );
-            }
-        }
-
-        Ok(())
-    }
-
     pub fn decommission_cancel(&mut self, idx: usize) -> bool {
         if let Some(stats) = self.pools.get_mut(idx) {
             if let Some(d) = &stats.decommission {
@@ -2975,85 +2697,6 @@ fn lifecycle_action_skips_heal_version(action: IlmAction) -> bool {
     action.delete()
 }
 
-#[cfg(test)]
-struct LifecycleDataMovementMutationBarrierState {
-    bucket: String,
-    object: String,
-    arrived: tokio::sync::Notify,
-    release: tokio::sync::Notify,
-}
-
-#[cfg(test)]
-pub(crate) struct LifecycleDataMovementMutationBarrier {
-    state: Arc<LifecycleDataMovementMutationBarrierState>,
-}
-
-#[cfg(test)]
-static LIFECYCLE_DATA_MOVEMENT_MUTATION_BARRIER: std::sync::OnceLock<
-    std::sync::Mutex<Option<Arc<LifecycleDataMovementMutationBarrierState>>>,
-> = std::sync::OnceLock::new();
-
-#[cfg(test)]
-impl LifecycleDataMovementMutationBarrier {
-    pub(crate) fn install(bucket: &str, object: &str) -> Self {
-        let state = Arc::new(LifecycleDataMovementMutationBarrierState {
-            bucket: bucket.to_string(),
-            object: object.to_string(),
-            arrived: tokio::sync::Notify::new(),
-            release: tokio::sync::Notify::new(),
-        });
-        let mut slot = LIFECYCLE_DATA_MOVEMENT_MUTATION_BARRIER
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("lifecycle data movement mutation barrier should not be poisoned");
-        assert!(slot.is_none(), "lifecycle data movement mutation barrier must be unique");
-        *slot = Some(Arc::clone(&state));
-        Self { state }
-    }
-
-    pub(crate) async fn wait_until_paused(&self) {
-        tokio::time::timeout(std::time::Duration::from_secs(30), self.state.arrived.notified())
-            .await
-            .expect("lifecycle data movement should reach its mutation boundary");
-    }
-
-    pub(crate) fn release(&self) {
-        self.state.release.notify_one();
-    }
-}
-
-#[cfg(test)]
-impl Drop for LifecycleDataMovementMutationBarrier {
-    fn drop(&mut self) {
-        self.state.release.notify_one();
-        let mut slot = LIFECYCLE_DATA_MOVEMENT_MUTATION_BARRIER
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("lifecycle data movement mutation barrier should not be poisoned");
-        if slot.as_ref().is_some_and(|state| Arc::ptr_eq(state, &self.state)) {
-            *slot = None;
-        }
-    }
-}
-
-#[cfg(test)]
-async fn pause_lifecycle_data_movement_mutation(bucket: &str, object: &str, has_run_fence_signal: bool) {
-    if !has_run_fence_signal {
-        return;
-    }
-    let barrier = LIFECYCLE_DATA_MOVEMENT_MUTATION_BARRIER
-        .get_or_init(|| std::sync::Mutex::new(None))
-        .lock()
-        .expect("lifecycle data movement mutation barrier should not be poisoned")
-        .as_ref()
-        .filter(|barrier| barrier.bucket == bucket && barrier.object == object)
-        .cloned();
-    if let Some(barrier) = barrier {
-        barrier.arrived.notify_one();
-        barrier.release.notified().await;
-    }
-}
-
 fn resolve_data_movement_lifecycle_expiry_result(action: IlmAction, apply_actions: bool, applied: bool) -> Result<bool> {
     if !apply_actions || applied {
         return Ok(true);
@@ -3073,7 +2716,6 @@ pub(crate) async fn should_skip_lifecycle_for_data_movement(
     object_lock_config: Option<&ObjectLockConfiguration>,
     apply_actions: bool,
     event_source: &LcEventSrc,
-    lock_lost_signal: Option<Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
 ) -> Result<bool> {
     let Some(lifecycle_config) = lifecycle_config else {
         return Ok(false);
@@ -3089,31 +2731,16 @@ pub(crate) async fn should_skip_lifecycle_for_data_movement(
                 let Ok(bucket_incarnation_id) = store.bucket_incarnation_id_from_disk(bucket).await else {
                     return Ok(false);
                 };
-                let _ = match lock_lost_signal {
-                    Some(signal) => {
-                        apply_expiry_rule_for_data_movement(store, &event, event_source, &object_info, Some(signal)).await
-                    }
-                    None => {
-                        apply_expiry_on_transitioned_object(store, &object_info, &event, event_source, bucket_incarnation_id)
-                            .await
-                    }
-                };
+                let _ =
+                    apply_expiry_on_transitioned_object(store, &object_info, &event, event_source, bucket_incarnation_id).await;
             }
             Ok(false)
         }
         action if lifecycle_action_removes_data_movement_version(action) => {
-            #[cfg(test)]
-            pause_lifecycle_data_movement_mutation(bucket, &version.name, lock_lost_signal.is_some()).await;
             if lifecycle_delete_all_versions_blocked_by_replication(store.clone(), bucket, &object_info.name, action).await? {
                 return Ok(false);
             }
-            let applied = !apply_actions
-                || match lock_lost_signal {
-                    Some(signal) => {
-                        apply_expiry_rule_for_data_movement(store, &event, event_source, &object_info, Some(signal)).await
-                    }
-                    None => apply_expiry_rule_in(store, &event, event_source, &object_info).await,
-                };
+            let applied = !apply_actions || apply_expiry_rule_in(store, &event, event_source, &object_info).await;
             resolve_data_movement_lifecycle_expiry_result(action, apply_actions, applied)
         }
         _ => Ok(false),
@@ -3250,9 +2877,16 @@ impl ECStore {
             .first()
             .cloned()
             .ok_or_else(|| Error::other("decommission start rebalance metadata load failed: no storage pools available"))?;
-        #[cfg(test)]
-        observe_pool_activation_start_attempt(PoolActivationStartKind::Decommission);
-        let activation_fence = acquire_pool_rebalance_activation_locks(rebalance_pool.clone()).await?;
+        let pool_meta_lock = rebalance_pool.new_ns_lock(RUSTFS_META_BUCKET, POOL_META_NAME).await?;
+        let _pool_meta_guard = pool_meta_lock
+            .get_write_lock(get_lock_acquire_timeout())
+            .await
+            .map_err(decommission_pool_meta_lock_error)?;
+        let ns_lock = rebalance_pool.new_ns_lock(RUSTFS_META_BUCKET, REBAL_META_NAME).await?;
+        let _guard = ns_lock
+            .get_write_lock(get_lock_acquire_timeout())
+            .await
+            .map_err(decommission_rebalance_meta_lock_error)?;
 
         let mut rebalance_meta = RebalanceMeta::new();
         match rebalance_meta
@@ -3297,10 +2931,7 @@ impl ECStore {
             latest_pool_meta.queue_buckets(idx, decom_buckets.clone());
         }
 
-        activation_fence.ensure_held()?;
-        latest_pool_meta
-            .save_no_lock_with_activation_fence(self.pools.clone(), activation_fence)
-            .await?;
+        latest_pool_meta.save_no_lock(self.pools.clone()).await?;
         {
             let mut pool_meta = self.pool_meta.write().await;
             *pool_meta = latest_pool_meta;
@@ -3314,11 +2945,6 @@ impl ECStore {
         ensure_decommission_not_rebalancing(self.is_rebalance_conflicting_with_decommission().await)
     }
 
-    async fn ensure_decommission_rebalance_idle_after_refresh_under_start_gate(&self) -> Result<()> {
-        self.load_rebalance_meta_under_start_gate().await?;
-        ensure_decommission_not_rebalancing(self.is_rebalance_conflicting_with_decommission().await)
-    }
-
     pub async fn status(&self, idx: usize) -> Result<PoolStatus> {
         let space_info = self.get_decommission_pool_space_info(idx).await?;
 
@@ -4213,7 +3839,6 @@ impl ECStore {
                     object_lock_config.as_ref(),
                     true,
                     &LcEventSrc::Decom,
-                    None,
                 )
                 .await
             })
@@ -4544,7 +4169,6 @@ impl ECStore {
                         lifecycle_guard: bucket_incarnation_fence
                             .as_ref()
                             .and_then(|guard| guard.namespace_lock_guard()),
-                        namespace_lock_lost_signal: None,
                         object_mutation_fence: Some(&source_cleanup_mutation_fence),
                     },
                     "decommission",
@@ -5361,11 +4985,7 @@ impl ECStore {
         validate_start_decommission_request(&indices, self.single_pool())?;
 
         self.ensure_decommission_rebalance_idle_after_refresh().await?;
-        #[cfg(test)]
-        let endpoints = self.instance_endpoints().unwrap_or_else(|| self.endpoints());
-        #[cfg(not(test))]
-        let endpoints = self.endpoints();
-        ensure_decommission_start_local_leader(&endpoints, &indices)?;
+        ensure_decommission_start_local_leader(&self.endpoints(), &indices)?;
 
         for idx in indices.iter().copied() {
             ensure_valid_decommission_pool_index(self.pools.len(), idx)?;
@@ -5400,8 +5020,7 @@ impl ECStore {
         }
 
         let _start_guard = self.start_gate.lock().await;
-        self.ensure_decommission_rebalance_idle_after_refresh_under_start_gate()
-            .await?;
+        self.ensure_decommission_rebalance_idle_after_refresh().await?;
 
         let all_space_infos = self.get_decommission_all_pool_space_infos().await?;
         self.cancel_decommission_routines_and_wait(&indices).await;
@@ -5595,7 +5214,6 @@ impl ECStore {
                                 object_lock_config.as_ref(),
                                 false,
                                 &LcEventSrc::Decom,
-                                None,
                             )
                             .await
                             {
@@ -5676,128 +5294,6 @@ mod tests {
     use crate::bucket::replication::{ReplicationState, ReplicationStatusType};
     use serde::Serialize;
 
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn decommission_activation_replicates_commit_after_post_save_fence_loss() {
-        let (_temp_dirs, store, _other_store) = crate::services::rebalance::test_two_pool_stores(None).await;
-        let barrier = PoolActivationDurableSaveBarrier::install(&store.pools[0]);
-        let start_store = Arc::clone(&store);
-        let start_task = tokio::spawn(async move {
-            start_store
-                .save_current_pool_meta_for_decommission_start(
-                    &[0],
-                    vec![(
-                        0,
-                        PoolSpaceInfo {
-                            free: 50,
-                            total: 100,
-                            used: 50,
-                        },
-                    )],
-                    Vec::new(),
-                )
-                .await
-        });
-
-        barrier.wait_until_paused().await;
-        barrier.release_after_fence_loss();
-        tokio::time::timeout(std::time::Duration::from_secs(30), start_task)
-            .await
-            .expect("decommission activation should finish after its canonical commit")
-            .expect("decommission activation task should not panic")
-            .expect("post-commit fence loss must not report the committed activation as failed");
-
-        let local = store.pool_meta.read().await;
-        assert!(pool_meta_has_active_decommission(&local));
-        drop(local);
-
-        for pool in &store.pools {
-            let mut persisted = PoolMeta::default();
-            persisted
-                .load_no_lock(pool.clone())
-                .await
-                .expect("every pool should retain readable committed decommission metadata");
-            assert!(
-                pool_meta_has_active_decommission(&persisted),
-                "every pool must adopt the canonical committed activation"
-            );
-        }
-    }
-
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn decommission_activation_adopts_canonical_commit_after_replica_failure() {
-        let (_temp_dirs, store, _other_store) = crate::services::rebalance::test_two_pool_stores(None).await;
-        let barrier = PoolActivationDurableSaveBarrier::install(&store.pools[0]);
-        let start_store = Arc::clone(&store);
-        let start_task = tokio::spawn(async move {
-            start_store
-                .save_current_pool_meta_for_decommission_start(
-                    &[0],
-                    vec![(
-                        0,
-                        PoolSpaceInfo {
-                            free: 50,
-                            total: 100,
-                            used: 50,
-                        },
-                    )],
-                    Vec::new(),
-                )
-                .await
-        });
-
-        barrier.wait_until_paused().await;
-        let mut replica_disks = Vec::new();
-        for set in &store.pools[1].disk_set {
-            let mut disks = set.disks.write().await;
-            let saved = std::mem::take(&mut *disks);
-            *disks = vec![None; saved.len()];
-            replica_disks.push(saved);
-        }
-        barrier.release_after_fence_loss();
-
-        tokio::time::timeout(std::time::Duration::from_secs(30), start_task)
-            .await
-            .expect("decommission activation should finish after its canonical commit")
-            .expect("decommission activation task should not panic")
-            .expect("a replica save failure must not report the committed activation as failed");
-
-        for (set, disks) in store.pools[1].disk_set.iter().zip(replica_disks) {
-            *set.disks.write().await = disks;
-        }
-
-        let local = store.pool_meta.read().await;
-        assert!(pool_meta_has_active_decommission(&local));
-        drop(local);
-
-        let mut canonical = PoolMeta::default();
-        canonical
-            .load_no_lock(store.pools[0].clone())
-            .await
-            .expect("the canonical committed decommission metadata should remain readable");
-        assert!(pool_meta_has_active_decommission(&canonical));
-
-        let mut replica = PoolMeta::default();
-        replica
-            .load_no_lock(store.pools[1].clone())
-            .await
-            .expect("the stale replica metadata should remain readable after disks recover");
-        assert!(!pool_meta_has_active_decommission(&replica));
-
-        let worker_cancel = CancellationToken::new();
-        store
-            .spawn_decommission_routines(Arc::clone(&store), worker_cancel.clone(), vec![0])
-            .await
-            .expect("the committed activation should admit its decommission worker");
-        let admitted_cancel = store.decommission_cancelers.read().await[0]
-            .clone()
-            .expect("the admitted decommission worker should have a cancellation token");
-        assert!(!admitted_cancel.is_cancelled());
-        worker_cancel.cancel();
-        assert!(admitted_cancel.is_cancelled());
-    }
-
     #[test]
     fn ensure_pool_not_left_in_cmdline_after_decommission_allows_active_pool() {
         assert!(ensure_pool_not_left_in_cmdline_after_decommission(0, "http://node{1...4}/disk{1...4}", false).is_ok());
@@ -6855,12 +6351,11 @@ mod pools_tests {
     use super::{
         DECOMMISSION_ENTRY_CONCURRENCY_DEFAULT_CAP, DECOMMISSION_ENTRY_CONCURRENCY_HARD_CAP, DECOMMISSION_ENTRY_QUEUE_HARD_CAP,
         DECOMMISSION_PROGRESS_SAVE_INTERVAL, DECOMMISSION_PROGRESS_SAVE_ITEM_THRESHOLD, DecomBucketInfo, DecommissionCanceler,
-        DecommissionEntryEnqueueResult, DecommissionStartPoolState, DecommissionTerminalState, ListCallback, POOL_META_NAME,
-        PoolDecommissionInfo, PoolMeta, PoolSpaceInfo, PoolStatus, QueuedDecommissionEntry, REBAL_META_NAME,
-        acquire_pool_rebalance_activation_locks, apply_decommission_status_space_info, await_decommission_worker,
-        bind_decommission_cancelers, bind_missing_decommission_cancelers, cancel_decommission_canceler,
-        clamp_decommission_entry_concurrency, classify_decommission_terminal_state, count_decommission_item,
-        decommission_cancel_signal_result, decommission_entry_queue_capacity, decommission_item_size,
+        DecommissionEntryEnqueueResult, DecommissionStartPoolState, DecommissionTerminalState, ListCallback,
+        PoolDecommissionInfo, PoolMeta, PoolSpaceInfo, PoolStatus, QueuedDecommissionEntry, apply_decommission_status_space_info,
+        await_decommission_worker, bind_decommission_cancelers, bind_missing_decommission_cancelers,
+        cancel_decommission_canceler, clamp_decommission_entry_concurrency, classify_decommission_terminal_state,
+        count_decommission_item, decommission_cancel_signal_result, decommission_entry_queue_capacity, decommission_item_size,
         decommission_meta_bucket_options, decommission_start_pool_state, dedup_indices, default_decommission_bucket_concurrency,
         default_decommission_entry_concurrency, drain_decommission_entry_queue, enqueue_decommission_entry,
         ensure_decommission_cancel_allowed, ensure_decommission_clear_allowed, ensure_decommission_generation,
@@ -6901,45 +6396,15 @@ mod pools_tests {
     use rustfs_filemeta::{FileInfo, FileInfoVersions, MetaCacheEntry, ObjectPartInfo};
     use rustfs_filemeta::{MetaCacheEntries, MetadataResolutionParams};
     use rustfs_rio::Index;
-    use std::future::Future;
     use std::sync::{
-        Arc, Mutex,
+        Arc,
         atomic::{AtomicBool, AtomicUsize, Ordering},
     };
-    use std::task::{Context, Poll};
     use std::time::Duration as StdDuration;
     use time::{Duration, OffsetDateTime};
     use tokio::sync::Semaphore;
     use tokio_util::sync::CancellationToken;
 
-    #[derive(Debug)]
-    struct ActivationLockRecorder {
-        lock_manager: Arc<rustfs_lock::GlobalLockManager>,
-        owner: &'static str,
-        resources: Mutex<Vec<String>>,
-    }
-
-    #[async_trait::async_trait]
-    impl crate::storage_api_contracts::namespace::NamespaceLocking for ActivationLockRecorder {
-        type Error = Error;
-        type NamespaceLock = rustfs_lock::NamespaceLockWrapper;
-
-        async fn new_ns_lock(&self, bucket: &str, object: &str) -> crate::error::Result<Self::NamespaceLock> {
-            self.resources
-                .lock()
-                .expect("activation lock recorder should not be poisoned")
-                .push(object.to_string());
-            Ok(rustfs_lock::NamespaceLockWrapper::new(
-                rustfs_lock::NamespaceLock::with_local_manager(
-                    "activation-lock-test".to_string(),
-                    Arc::clone(&self.lock_manager),
-                ),
-                rustfs_lock::ObjectKey::new(bucket, object),
-                self.owner.to_string(),
-            ))
-        }
-    }
-
     fn noop_decommission_list_callback() -> ListCallback {
         Arc::new(|_| Box::pin(async {}))
     }
@@ -6988,48 +6453,6 @@ mod pools_tests {
         }
     }
 
-    #[tokio::test]
-    async fn test_activation_fence_uses_one_lock_order_and_serializes_callers() {
-        let manager = Arc::new(rustfs_lock::GlobalLockManager::new());
-        let first = Arc::new(ActivationLockRecorder {
-            lock_manager: Arc::clone(&manager),
-            owner: "first",
-            resources: Mutex::new(Vec::new()),
-        });
-        let second = Arc::new(ActivationLockRecorder {
-            lock_manager: manager,
-            owner: "second",
-            resources: Mutex::new(Vec::new()),
-        });
-
-        let first_guards = acquire_pool_rebalance_activation_locks(first.clone())
-            .await
-            .expect("first activation should acquire both locks");
-        assert_eq!(
-            *first
-                .resources
-                .lock()
-                .expect("activation lock recorder should not be poisoned"),
-            vec![POOL_META_NAME.to_string(), REBAL_META_NAME.to_string()]
-        );
-
-        let mut second_acquire = Box::pin(acquire_pool_rebalance_activation_locks(second.clone()));
-        let mut context = Context::from_waker(futures::task::noop_waker_ref());
-        assert!(matches!(second_acquire.as_mut().poll(&mut context), Poll::Pending));
-
-        drop(first_guards);
-        second_acquire
-            .await
-            .expect("second activation should acquire both locks after the first releases them");
-        assert_eq!(
-            *second
-                .resources
-                .lock()
-                .expect("activation lock recorder should not be poisoned"),
-            vec![POOL_META_NAME.to_string(), REBAL_META_NAME.to_string()]
-        );
-    }
-
     #[test]
     fn test_apply_decommission_status_space_info_adds_idle_pool_usage() {
         let status = apply_decommission_status_space_info(

crates/ecstore/src/core/sets.rs

@@ -1241,16 +1241,8 @@ pub(crate) async fn make_local_two_set_sets() -> (Vec<tempfile::TempDir>, Arc<Se
     make_local_two_set_sets_with_ctx(bootstrap_ctx()).await
 }
 
-#[cfg(any(test, feature = "test-util"))]
+#[cfg(test)]
 pub(crate) async fn make_local_two_set_sets_with_ctx(ctx: Arc<InstanceContext>) -> (Vec<tempfile::TempDir>, Arc<Sets>) {
-    make_local_two_set_sets_for_pool_with_ctx(ctx, 0).await
-}
-
-#[cfg(any(test, feature = "test-util"))]
-pub(crate) async fn make_local_two_set_sets_for_pool_with_ctx(
-    ctx: Arc<InstanceContext>,
-    pool_idx: usize,
-) -> (Vec<tempfile::TempDir>, Arc<Sets>) {
     use crate::layout::endpoint::Endpoint;
     use rustfs_lock::client::local::LocalClient;
 
@@ -1266,7 +1258,7 @@ pub(crate) async fn make_local_two_set_sets_for_pool_with_ctx(
             let temp_dir = tempfile::tempdir().expect("tempdir should be created");
             let mut endpoint = Endpoint::try_from(temp_dir.path().to_str().expect("tempdir path should be utf8"))
                 .expect("endpoint should parse");
-            endpoint.set_pool_index(pool_idx);
+            endpoint.set_pool_index(0);
             endpoint.set_set_index(set_index);
             endpoint.set_disk_index(disk_index);
             let disk = new_disk(
@@ -1302,7 +1294,7 @@ pub(crate) async fn make_local_two_set_sets_for_pool_with_ctx(
                 2,
                 1,
                 set_index,
-                pool_idx,
+                0,
                 endpoints,
                 format.clone(),
                 lockers,
@@ -1315,7 +1307,7 @@ pub(crate) async fn make_local_two_set_sets_for_pool_with_ctx(
     let sets = Arc::new(Sets {
         id: format.id,
         disk_set: disk_sets,
-        pool_idx,
+        pool_idx: 0,
         endpoints: PoolEndpoints {
             legacy: false,
             set_count: 2,

crates/ecstore/src/data_movement/mod.rs

@@ -1023,11 +1023,10 @@ pub(crate) enum SourceCleanupError {
     Storage(#[from] Error),
 }
 
-#[derive(Clone, Default)]
+#[derive(Clone, Copy, Default)]
 pub(crate) struct SourceCleanupBucketFence<'a> {
     pub(crate) expected_incarnation_id: Option<uuid::Uuid>,
     pub(crate) lifecycle_guard: Option<&'a rustfs_lock::NamespaceLockGuard>,
-    pub(crate) namespace_lock_lost_signal: Option<Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
     pub(crate) object_mutation_fence: Option<&'a SourceCleanupMutationFence>,
 }
 
@@ -1062,7 +1061,7 @@ pub(crate) async fn ensure_source_cleanup_versions_unchanged(
     ensure_source_cleanup_versions_match(expected, &current, allowed_missing)
 }
 
-#[cfg(any(test, feature = "test-util"))]
+#[cfg(test)]
 struct SourceCleanupDeleteBarrierState {
     bucket: String,
     object: String,
@@ -1072,7 +1071,7 @@ struct SourceCleanupDeleteBarrierState {
     release: tokio::sync::Notify,
 }
 
-#[cfg(any(test, feature = "test-util"))]
+#[cfg(test)]
 #[allow(
     dead_code,
     reason = "installed by set_disk object tests behind `--features test-util` (backlog#1823)"
@@ -1081,11 +1080,11 @@ pub(crate) struct SourceCleanupDeleteBarrier {
     state: Arc<SourceCleanupDeleteBarrierState>,
 }
 
-#[cfg(any(test, feature = "test-util"))]
+#[cfg(test)]
 static SOURCE_CLEANUP_DELETE_BARRIERS: std::sync::OnceLock<std::sync::Mutex<Vec<Arc<SourceCleanupDeleteBarrierState>>>> =
     std::sync::OnceLock::new();
 
-#[cfg(any(test, feature = "test-util"))]
+#[cfg(test)]
 #[allow(
     dead_code,
     reason = "installed by set_disk object tests behind `--features test-util` (backlog#1823)"
@@ -1149,7 +1148,7 @@ pub(crate) fn notify_source_cleanup_mutation_fence_pending(bucket: &str, object:
     }
 }
 
-#[cfg(any(test, feature = "test-util"))]
+#[cfg(test)]
 impl Drop for SourceCleanupDeleteBarrier {
     fn drop(&mut self) {
         self.state.release.notify_one();
@@ -1161,7 +1160,7 @@ impl Drop for SourceCleanupDeleteBarrier {
     }
 }
 
-#[cfg(any(test, feature = "test-util"))]
+#[cfg(test)]
 async fn pause_source_cleanup_before_delete(bucket: &str, object: &str) {
     let barrier = SOURCE_CLEANUP_DELETE_BARRIERS
         .get_or_init(|| std::sync::Mutex::new(Vec::new()))
@@ -1221,7 +1220,7 @@ pub(crate) async fn cleanup_source_entry_if_unchanged(
 
     ensure_source_cleanup_versions_unchanged(set.clone(), bucket, object, expected, allowed_missing, op_label).await?;
 
-    #[cfg(any(test, feature = "test-util"))]
+    #[cfg(test)]
     pause_source_cleanup_before_delete(bucket, object).await;
 
     let mut opts = ObjectOptions {
@@ -1241,9 +1240,6 @@ pub(crate) async fn cleanup_source_entry_if_unchanged(
     if let Some(bucket_lifecycle_guard) = bucket_fence.lifecycle_guard {
         opts.add_bucket_lifecycle_lock_guard(bucket_lifecycle_guard);
     }
-    if let Some(signal) = bucket_fence.namespace_lock_lost_signal {
-        opts.add_namespace_lock_lost_signal(signal);
-    }
     let result = set.delete_object(bucket, cleanup_key.as_str(), opts).await;
     if result.is_ok() {
         crate::store::list_objects::observe_scanner_namespace_mutations(bucket, 1);
@@ -1414,13 +1410,11 @@ pub(crate) async fn migrate_decommission_object(
         rd,
         source_bucket_incarnation_id,
         op_label,
-        None,
         Some(&_mutation_fence),
     )
     .await
 }
 
-#[cfg(test)]
 pub(crate) async fn migrate_object(
     store: Arc<ECStore>,
     pool_idx: usize,
@@ -1429,33 +1423,9 @@ pub(crate) async fn migrate_object(
     source_bucket_incarnation_id: Option<uuid::Uuid>,
     op_label: &str,
 ) -> Result<()> {
-    migrate_object_with_lock_lost_signal(store, pool_idx, bucket, rd, source_bucket_incarnation_id, op_label, None).await
+    migrate_object_inner(store, pool_idx, bucket, rd, source_bucket_incarnation_id, op_label, None).await
 }
 
-#[allow(clippy::too_many_arguments)]
-pub(crate) async fn migrate_object_with_lock_lost_signal(
-    store: Arc<ECStore>,
-    pool_idx: usize,
-    bucket: String,
-    rd: GetObjectReader,
-    source_bucket_incarnation_id: Option<uuid::Uuid>,
-    op_label: &str,
-    lock_lost_signal: Option<Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
-) -> Result<()> {
-    migrate_object_inner(
-        store,
-        pool_idx,
-        bucket,
-        rd,
-        source_bucket_incarnation_id,
-        op_label,
-        lock_lost_signal,
-        None,
-    )
-    .await
-}
-
-#[allow(clippy::too_many_arguments)]
 async fn migrate_object_inner(
     store: Arc<ECStore>,
     pool_idx: usize,
@@ -1463,7 +1433,6 @@ async fn migrate_object_inner(
     rd: GetObjectReader,
     source_bucket_incarnation_id: Option<uuid::Uuid>,
     op_label: &str,
-    lock_lost_signal: Option<Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
     mutation_fence: Option<&ObjectLockDiagGuard>,
 ) -> Result<()> {
     let object_info = rd.object_info.clone();
@@ -1477,9 +1446,6 @@ async fn migrate_object_inner(
     if should_use_multipart_data_movement(&object_info, has_part_checksums) {
         let mut new_multipart_opts = data_movement_new_multipart_opts(&object_info, pool_idx);
         new_multipart_opts.expected_bucket_incarnation_id = source_bucket_incarnation_id;
-        if let Some(signal) = lock_lost_signal.as_ref() {
-            new_multipart_opts.add_namespace_lock_lost_signal(Arc::clone(signal));
-        }
         let (res, target_pool_idx, expected_bucket_incarnation_id) = match store
             .handle_new_multipart_upload_with_pool_idx(&bucket, &object_info.name, &new_multipart_opts, mutation_fence)
             .await
@@ -1524,7 +1490,7 @@ async fn migrate_object_inner(
                             err,
                         )
                     })?;
-                let mut part_opts = ObjectOptions {
+                let part_opts = ObjectOptions {
                     part_number: Some(part.number),
                     preserve_etag: Some(part.etag.clone()),
                     data_movement: true,
@@ -1532,9 +1498,6 @@ async fn migrate_object_inner(
                     expected_bucket_incarnation_id,
                     ..Default::default()
                 };
-                if let Some(signal) = lock_lost_signal.as_ref() {
-                    part_opts.add_namespace_lock_lost_signal(Arc::clone(signal));
-                }
                 let pi = match store
                     .put_object_part_for_data_movement(
                         target_pool_idx,
@@ -1579,9 +1542,6 @@ async fn migrate_object_inner(
                     )
                 })?;
             complete_multipart_opts.expected_bucket_incarnation_id = expected_bucket_incarnation_id;
-            if let Some(signal) = lock_lost_signal.as_ref() {
-                complete_multipart_opts.add_namespace_lock_lost_signal(Arc::clone(signal));
-            }
             if let Err(err) = store
                 .clone()
                 .complete_multipart_upload_for_data_movement(
@@ -1630,18 +1590,18 @@ async fn migrate_object_inner(
 
         if multipart_result.is_ok() && should_abort_multipart_upload(&abort_multipart_flag) {
             let abort_result = store
-                .abort_multipart_upload_for_data_movement(target_pool_idx, &bucket, &object_info.name, &res.upload_id, &{
-                    let mut opts = ObjectOptions {
+                .abort_multipart_upload_for_data_movement(
+                    target_pool_idx,
+                    &bucket,
+                    &object_info.name,
+                    &res.upload_id,
+                    &ObjectOptions {
                         data_movement: true,
                         src_pool_idx: pool_idx,
                         expected_bucket_incarnation_id,
                         ..Default::default()
-                    };
-                    if let Some(signal) = lock_lost_signal.as_ref() {
-                        opts.add_namespace_lock_lost_signal(Arc::clone(signal));
-                    }
-                    opts
-                })
+                    },
+                )
                 .await;
             match abort_result {
                 Ok(()) => return Ok(()),
@@ -1699,18 +1659,18 @@ async fn migrate_object_inner(
         if let Err(primary_err) = multipart_result {
             if should_abort_multipart_upload(&abort_multipart_flag) {
                 return match store
-                    .abort_multipart_upload_for_data_movement(target_pool_idx, &bucket, &object_info.name, &res.upload_id, &{
-                        let mut opts = ObjectOptions {
+                    .abort_multipart_upload_for_data_movement(
+                        target_pool_idx,
+                        &bucket,
+                        &object_info.name,
+                        &res.upload_id,
+                        &ObjectOptions {
                             data_movement: true,
                             src_pool_idx: pool_idx,
                             expected_bucket_incarnation_id,
                             ..Default::default()
-                        };
-                        if let Some(signal) = lock_lost_signal.as_ref() {
-                            opts.add_namespace_lock_lost_signal(Arc::clone(signal));
-                        }
-                        opts
-                    })
+                        },
+                    )
                     .await
                 {
                     Ok(()) => Err(primary_err),
@@ -1745,9 +1705,6 @@ async fn migrate_object_inner(
 
     let mut put_opts = data_movement_put_object_opts(&object_info, pool_idx);
     put_opts.expected_bucket_incarnation_id = source_bucket_incarnation_id;
-    if let Some(signal) = lock_lost_signal {
-        put_opts.add_namespace_lock_lost_signal(signal);
-    }
     let (target_pool_idx, put_result) = store
         .put_object_for_data_movement(&bucket, &object_info.name, &mut data, &put_opts, mutation_fence)
         .await

crates/ecstore/src/disk/local.rs

@@ -8000,10 +8000,15 @@ impl DiskAPI for LocalDisk {
             use std::io::Write as _;
 
             let file_path = self.io_get_object_path(volume, path)?;
-            let lock_path = file_path.with_extension("rustfs-cas.lock");
             let path = path.to_string();
             let sync_metadata = effective_durability(volume).syncs_commit_metadata();
             return Ok(tokio::task::spawn_blocking(move || {
+                // A persistent directory lock bounds metadata growth. Removing
+                // per-target lock files can split flock ownership across inodes.
+                let lock_path = file_path
+                    .parent()
+                    .ok_or_else(|| std::io::Error::new(ErrorKind::InvalidInput, "conditional file has no parent"))?
+                    .join(".rustfs-cas.lock");
                 let lock = std::fs::OpenOptions::new()
                     .create(true)
                     .truncate(false)
@@ -8068,7 +8073,25 @@ impl DiskAPI for LocalDisk {
             .map_err(DiskError::from)??);
         }
 
-        #[cfg(not(unix))]
+        #[cfg(windows)]
+        {
+            let file_path = self.io_get_object_path(volume, path)?;
+            let sync_metadata = effective_durability(volume).syncs_commit_metadata();
+            let publication_root = self.publication_root.clone();
+            return Ok(tokio::task::spawn_blocking(move || {
+                os::compare_and_update_control_file(
+                    &file_path,
+                    expected.as_deref(),
+                    replacement.as_deref(),
+                    sync_metadata,
+                    &publication_root,
+                )
+            })
+            .await
+            .map_err(DiskError::from)??);
+        }
+
+        #[cfg(not(any(unix, windows)))]
         {
             let _ = (volume, path, expected, replacement);
             Err(DiskError::MethodNotAllowed)
@@ -21823,9 +21846,9 @@ mod test {
         assert!(matches!(results[1].as_ref().unwrap_err(), DiskError::Io(_)));
     }
 
-    #[cfg(unix)]
+    #[cfg(any(unix, windows))]
     #[tokio::test]
-    async fn conditional_file_update_never_deletes_a_new_owner() {
+    async fn windows_and_unix_conditional_file_update_never_deletes_a_new_owner() {
         use tempfile::tempdir;
 
         let dir = tempdir().expect("temp dir should be created");
@@ -21856,8 +21879,18 @@ mod test {
             disk.read_all(RUSTFS_META_BUCKET, HEALING_MARKER_PATH)
                 .await
                 .expect("new owner marker should remain"),
-            owner_b
+            owner_b.clone()
         );
+        assert_eq!(
+            disk.compare_and_update_file(RUSTFS_META_BUCKET, HEALING_MARKER_PATH, Some(owner_b), None)
+                .await
+                .expect("current owner should remove marker"),
+            ConditionalFileUpdate::Updated
+        );
+        assert!(matches!(
+            disk.read_all(RUSTFS_META_BUCKET, HEALING_MARKER_PATH).await,
+            Err(DiskError::FileNotFound)
+        ));
     }
 
     #[cfg(unix)]
@@ -21872,7 +21905,10 @@ mod test {
         let marker_path = disk
             .get_object_path(RUSTFS_META_BUCKET, HEALING_MARKER_PATH)
             .expect("marker path should resolve");
-        let lock_path = marker_path.with_extension("rustfs-cas.lock");
+        let lock_path = marker_path
+            .parent()
+            .expect("marker path should have a parent")
+            .join(".rustfs-cas.lock");
         let lock = std::fs::OpenOptions::new()
             .create(true)
             .truncate(false)
@@ -21893,6 +21929,40 @@ mod test {
         assert!(matches!(err, DiskError::Io(ref err) if err.kind() == ErrorKind::WouldBlock));
     }
 
+    #[cfg(windows)]
+    #[tokio::test]
+    async fn windows_conditional_file_update_returns_would_block_when_marker_lock_is_contended() {
+        let dir = tempfile::tempdir().expect("temp dir should be created");
+        let endpoint = Endpoint::try_from(dir.path().to_str().expect("temp dir should be utf8")).expect("endpoint should parse");
+        let disk = LocalDisk::new(&endpoint, false).await.expect("local disk should be created");
+        ensure_test_volume(&disk, RUSTFS_META_BUCKET).await;
+        let marker_path = disk
+            .get_object_path(RUSTFS_META_BUCKET, HEALING_MARKER_PATH)
+            .expect("marker path should resolve");
+        let lock_path = marker_path
+            .parent()
+            .expect("marker path should have a parent")
+            .join(".rustfs-cas.lock");
+        let lock = std::fs::OpenOptions::new()
+            .create(true)
+            .truncate(false)
+            .read(true)
+            .write(true)
+            .open(lock_path)
+            .expect("marker lock should open");
+        lock.try_lock().expect("marker lock should be held");
+
+        let err = tokio::time::timeout(
+            Duration::from_secs(1),
+            disk.compare_and_update_file(RUSTFS_META_BUCKET, HEALING_MARKER_PATH, None, Some(Bytes::from_static(b"owner"))),
+        )
+        .await
+        .expect("contended conditional update must not block")
+        .expect_err("contended conditional update must retry");
+
+        assert!(matches!(err, DiskError::Io(ref err) if err.kind() == ErrorKind::WouldBlock));
+    }
+
     #[cfg(target_os = "linux")]
     #[tokio::test]
     async fn replacement_io_paths_stay_under_the_mount_lease() {

crates/ecstore/src/disk/os.rs

@@ -12,6 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#[cfg(windows)]
+use crate::disk::ConditionalFileUpdate;
 use crate::disk::error::DiskError;
 use crate::disk::error::Result;
 use crate::disk::error_conv::to_file_error;
@@ -3458,6 +3460,89 @@ fn read_windows_relative_file(file_path: &Path, parent_guard: &ExistingBaseDirec
     Ok(Some(data))
 }
 
+#[cfg(windows)]
+pub(crate) fn compare_and_update_control_file(
+    file_path: &Path,
+    expected: Option<&[u8]>,
+    replacement: Option<&[u8]>,
+    sync_metadata: bool,
+    publication_root: &PublicationRoot,
+) -> io::Result<ConditionalFileUpdate> {
+    use windows_sys::{
+        Wdk::Storage::FileSystem::{
+            FILE_NON_DIRECTORY_FILE, FILE_OPEN, FILE_OPEN_IF, FILE_OPEN_REPARSE_POINT, FILE_SYNCHRONOUS_IO_NONALERT,
+        },
+        Win32::Storage::FileSystem::{
+            DELETE, FILE_ATTRIBUTE_NORMAL, FILE_READ_ATTRIBUTES, FILE_SHARE_READ, FILE_SHARE_WRITE, FILE_WRITE_DATA, SYNCHRONIZE,
+        },
+    };
+
+    let parent = file_path
+        .parent()
+        .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidInput, "conditional file has no parent"))?;
+    let parent_guard = lock_windows_directory_tree(parent, Some(parent), publication_root)?;
+    let lock = open_windows_relative(
+        parent_guard.last_handle()?,
+        std::ffi::OsStr::new(".rustfs-cas.lock"),
+        SYNCHRONIZE | FILE_READ_ATTRIBUTES | FILE_WRITE_DATA,
+        FILE_SHARE_READ | FILE_SHARE_WRITE,
+        FILE_OPEN_IF,
+        FILE_NON_DIRECTORY_FILE | FILE_OPEN_REPARSE_POINT | FILE_SYNCHRONOUS_IO_NONALERT,
+        FILE_ATTRIBUTE_NORMAL,
+        true,
+    )?;
+    validate_windows_owned_file(&lock)?;
+    match lock.as_file().try_lock() {
+        Ok(()) => {}
+        Err(std::fs::TryLockError::WouldBlock) => return Err(io::Error::from(io::ErrorKind::WouldBlock)),
+        Err(std::fs::TryLockError::Error(err)) => return Err(err),
+    }
+
+    let current = read_windows_relative_file(file_path, &parent_guard)?;
+    let matches = match (&current, expected) {
+        (None, None) => true,
+        (Some(current), Some(expected)) => current.as_slice() == expected,
+        _ => false,
+    };
+    if !matches {
+        return Ok(match current {
+            None => ConditionalFileUpdate::Missing,
+            Some(_) => ConditionalFileUpdate::Mismatch,
+        });
+    }
+
+    match replacement {
+        Some(replacement) => RenameDestinationPathGuard {
+            directory: parent.to_path_buf(),
+            _directory_guard: parent_guard,
+        }
+        .write_file_for_path_access(file_path, replacement, sync_metadata, sync_metadata)?,
+        None => {
+            let file_name = file_path
+                .file_name()
+                .ok_or_else(|| io::Error::new(io::ErrorKind::InvalidInput, "conditional file must have a name"))?;
+            let file = open_windows_relative(
+                parent_guard.last_handle()?,
+                file_name,
+                DELETE | SYNCHRONIZE | FILE_READ_ATTRIBUTES,
+                FILE_SHARE_READ,
+                FILE_OPEN,
+                FILE_NON_DIRECTORY_FILE | FILE_OPEN_REPARSE_POINT | FILE_SYNCHRONOUS_IO_NONALERT,
+                0,
+                true,
+            )?;
+            validate_windows_owned_file(&file)?;
+            set_windows_file_delete_on_close(&file, true)?;
+            drop(file);
+            if sync_metadata {
+                fsync_dir_std(parent)?;
+            }
+        }
+    }
+
+    Ok(ConditionalFileUpdate::Updated)
+}
+
 #[cfg(windows)]
 fn open_windows_directory_component(
     parent: &WindowsDirectoryHandle,

crates/ecstore/src/io_support/bitrot.rs

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

crates/ecstore/src/object_api/types.rs

@@ -84,61 +84,6 @@ impl NamespaceLockFence {
     }
 }
 
-#[cfg(test)]
-static NAMESPACE_LOCK_SIGNAL_TEST_FENCES: std::sync::OnceLock<std::sync::Mutex<Vec<(usize, NamespaceLockFence)>>> =
-    std::sync::OnceLock::new();
-
-#[cfg(test)]
-pub(crate) struct NamespaceLockSignalTestFence {
-    signal_key: usize,
-}
-
-#[cfg(test)]
-impl NamespaceLockSignalTestFence {
-    pub(crate) fn install_with_loss_handle(
-        signal: &Arc<rustfs_lock::distributed_lock::LockLostSignal>,
-        loss_handle: Arc<std::sync::atomic::AtomicBool>,
-    ) -> Self {
-        let fence = NamespaceLockFence {
-            signals: Arc::default(),
-            forced_lost: Arc::new(vec![loss_handle]),
-        };
-        let signal_key = Arc::as_ptr(signal) as usize;
-        let mut fences = NAMESPACE_LOCK_SIGNAL_TEST_FENCES
-            .get_or_init(|| std::sync::Mutex::new(Vec::new()))
-            .lock()
-            .expect("namespace lock signal test fence should not be poisoned");
-        assert!(
-            !fences.iter().any(|(key, _)| *key == signal_key),
-            "namespace lock signal test fence must be unique"
-        );
-        fences.push((signal_key, fence));
-        Self { signal_key }
-    }
-}
-
-#[cfg(test)]
-impl Drop for NamespaceLockSignalTestFence {
-    fn drop(&mut self) {
-        let mut fences = NAMESPACE_LOCK_SIGNAL_TEST_FENCES
-            .get_or_init(|| std::sync::Mutex::new(Vec::new()))
-            .lock()
-            .expect("namespace lock signal test fence should not be poisoned");
-        fences.retain(|(key, _)| *key != self.signal_key);
-    }
-}
-
-#[cfg(test)]
-pub(crate) fn namespace_lock_signal_test_fence_is_lost(signal: &Arc<rustfs_lock::distributed_lock::LockLostSignal>) -> bool {
-    NAMESPACE_LOCK_SIGNAL_TEST_FENCES
-        .get_or_init(|| std::sync::Mutex::new(Vec::new()))
-        .lock()
-        .expect("namespace lock signal test fence should not be poisoned")
-        .iter()
-        .find(|(key, _)| *key == Arc::as_ptr(signal) as usize)
-        .is_some_and(|(_, fence)| fence.is_lock_lost())
-}
-
 #[derive(Debug)]
 pub struct ObjectLockConfigSnapshot {
     store_id: Option<Uuid>,
@@ -460,23 +405,9 @@ impl ObjectOptions {
     }
 
     pub(crate) fn add_namespace_lock_lost_signal(&mut self, signal: Arc<rustfs_lock::distributed_lock::LockLostSignal>) {
-        #[cfg(test)]
-        let test_fence = NAMESPACE_LOCK_SIGNAL_TEST_FENCES
-            .get_or_init(|| std::sync::Mutex::new(Vec::new()))
-            .lock()
-            .expect("namespace lock signal test fence should not be poisoned")
-            .iter()
-            .find(|(key, _)| *key == Arc::as_ptr(&signal) as usize)
-            .map(|(_, fence)| fence.clone());
         self.namespace_lock_fence
             .get_or_insert_with(NamespaceLockFence::new)
             .add_signal(signal);
-        #[cfg(test)]
-        if let Some(test_fence) = test_fence {
-            self.namespace_lock_fence
-                .get_or_insert_with(NamespaceLockFence::new)
-                .extend(&test_fence);
-        }
     }
 
     pub(crate) fn ensure_namespace_lock_fence(&mut self) {

crates/ecstore/src/runtime/sources.rs

@@ -119,8 +119,8 @@ pub(crate) fn endpoint_erasure_set_count() -> Option<usize> {
     endpoint_pools().map(|endpoints| endpoints.es_count())
 }
 
-pub(crate) fn endpoint_pool_is_local(endpoints: &EndpointServerPools, pool_index: usize) -> bool {
-    endpoints
+pub(crate) fn endpoint_pool_is_local(pool_index: usize) -> bool {
+    get_global_endpoints()
         .as_ref()
         .get(pool_index)
         .is_some_and(|pool| pool.endpoints.as_ref().first().is_some_and(|endpoint| endpoint.is_local))

crates/ecstore/src/services/notification_sys.rs

@@ -1121,21 +1121,9 @@ impl NotificationSys {
             }
         }
 
-        let local_rebalance_id = match expected_rebalance_id {
-            Some(expected_id) => Some(expected_id.to_owned()),
-            None => store.current_rebalance_id().await,
-        };
-        match store.stop_rebalance_for_id(local_rebalance_id.as_deref()).await {
+        match store.stop_rebalance_for_id(expected_rebalance_id).await {
             Ok(_) => {
-                let save_result = match local_rebalance_id.as_deref() {
-                    Some(expected_id) => {
-                        store
-                            .save_rebalance_stats_for_id(usize::MAX, RebalSaveOpt::StoppedAt, expected_id)
-                            .await
-                    }
-                    None => Ok(()),
-                };
-                if let Err(err) = save_result {
+                if let Err(err) = store.save_rebalance_stats(usize::MAX, RebalSaveOpt::StoppedAt).await {
                     error!(
                         event = EVENT_NOTIFICATION_PEER_PROPAGATION,
                         component = LOG_COMPONENT_ECSTORE,

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

@@ -102,20 +102,11 @@ pub(crate) trait MigrationBackend: Send + Sync {
 pub(crate) struct RebalanceMigrationBackend<'a> {
     source: &'a SetDisks,
     store: &'a ECStore,
-    lock_lost_signal: Option<std::sync::Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
 }
 
 impl<'a> RebalanceMigrationBackend<'a> {
-    pub(crate) fn new(
-        source: &'a SetDisks,
-        store: &'a ECStore,
-        lock_lost_signal: Option<std::sync::Arc<rustfs_lock::distributed_lock::LockLostSignal>>,
-    ) -> Self {
-        Self {
-            source,
-            store,
-            lock_lost_signal,
-        }
+    pub(crate) fn new(source: &'a SetDisks, store: &'a ECStore) -> Self {
+        Self { source, store }
     }
 }
 
@@ -139,11 +130,7 @@ impl MigrationBackend for RebalanceMigrationBackend<'_> {
         fi: &FileInfo,
         opts: &ObjectOptions,
     ) -> Result<()> {
-        let mut opts = opts.clone();
-        if let Some(signal) = self.lock_lost_signal.as_ref() {
-            opts.add_namespace_lock_lost_signal(std::sync::Arc::clone(signal));
-        }
-        self.store.decommission_tiered_object(bucket, object, fi, &opts).await
+        self.store.decommission_tiered_object(bucket, object, fi, opts).await
     }
 }
 

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

@@ -12,8 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#[cfg(test)]
-use crate::disk::DiskAPI;
 use crate::error::{Error, Result};
 use crate::object_api::{GetObjectReader, ObjectInfo, ObjectOptions, PutObjReader};
 use tokio::time::Duration;
@@ -47,8 +45,6 @@ mod runtime;
 mod types;
 mod worker;
 
-#[cfg(feature = "test-util")]
-pub use entry::test_util::PausedRebalanceEntryTestFixture;
 pub(crate) use meta::is_rebalance_conflicting_with_decommission;
 pub use meta::{decode_rebalance_stop_propagation_record, encode_rebalance_stop_propagation_record};
 pub use types::{
@@ -57,91 +53,5 @@ pub use types::{
 };
 use types::{RebalanceBucketConfigs, RebalanceBucketOutcome, RebalanceEntryOutcome};
 
-#[cfg(any(test, feature = "test-util"))]
-pub async fn test_store_with_persisted_rebalance_meta(
-    meta: RebalanceMeta,
-) -> (Vec<tempfile::TempDir>, std::sync::Arc<crate::store::ECStore>) {
-    let ctx = std::sync::Arc::new(crate::runtime::instance::InstanceContext::new());
-    let (temp_dirs, pool) = crate::core::sets::make_local_two_set_sets_with_ctx(ctx.clone()).await;
-    meta.save(pool.clone())
-        .await
-        .expect("rebalance test metadata should be persisted");
-    let endpoint_pools: crate::layout::endpoints::EndpointServerPools = vec![pool.endpoints.clone()].into();
-    let store = std::sync::Arc::new(crate::store::ECStore {
-        id: uuid::Uuid::new_v4(),
-        disk_map: std::collections::HashMap::new(),
-        pools: vec![pool],
-        peer_sys: crate::cluster::rpc::S3PeerSys::new_with_instance_ctx(&endpoint_pools, ctx.clone()),
-        pool_meta: tokio::sync::RwLock::new(crate::core::pools::PoolMeta::default()),
-        rebalance_meta: tokio::sync::RwLock::new(Some(meta)),
-        decommission_cancelers: tokio::sync::RwLock::new(vec![None]),
-        start_gate: tokio::sync::Mutex::new(()),
-        pool_meta_save_gate: tokio::sync::Mutex::new(()),
-        ctx,
-        bucket_fence_registry: std::sync::Arc::default(),
-    });
-    (temp_dirs, store)
-}
-
-#[cfg(test)]
-pub(crate) async fn test_two_pool_stores(
-    rebalance_meta: Option<RebalanceMeta>,
-) -> (
-    Vec<tempfile::TempDir>,
-    std::sync::Arc<crate::store::ECStore>,
-    std::sync::Arc<crate::store::ECStore>,
-) {
-    use crate::core::pools::PoolMeta;
-    use crate::layout::endpoints::{EndpointServerPools, SetupType};
-
-    let ctx = std::sync::Arc::new(crate::runtime::instance::InstanceContext::new());
-    ctx.update_erasure_type(SetupType::DistErasure).await;
-    let (mut temp_dirs, first_pool) =
-        crate::core::sets::make_local_two_set_sets_for_pool_with_ctx(std::sync::Arc::clone(&ctx), 0).await;
-    let (second_temp_dirs, second_pool) =
-        crate::core::sets::make_local_two_set_sets_for_pool_with_ctx(std::sync::Arc::clone(&ctx), 1).await;
-    temp_dirs.extend(second_temp_dirs);
-    let pools = vec![first_pool, second_pool];
-    {
-        let local_disk_map = ctx.local_disk_map();
-        let mut local_disk_map = local_disk_map.write().await;
-        for pool in &pools {
-            for set in &pool.disk_set {
-                for disk in set.disks.read().await.iter().flatten() {
-                    local_disk_map.insert(disk.endpoint().to_string(), Some(disk.clone()));
-                }
-            }
-        }
-    }
-    let pool_meta = PoolMeta::new(&pools, &PoolMeta::default());
-    pool_meta
-        .save(pools.clone())
-        .await
-        .expect("baseline pool metadata should be persisted");
-    if let Some(meta) = rebalance_meta.as_ref() {
-        meta.save(pools[0].clone())
-            .await
-            .expect("active rebalance metadata should be persisted");
-    }
-    let endpoint_pools: EndpointServerPools = pools.iter().map(|pool| pool.endpoints.clone()).collect::<Vec<_>>().into();
-    ctx.set_endpoints(endpoint_pools.clone());
-    let make_store = || {
-        std::sync::Arc::new(crate::store::ECStore {
-            id: uuid::Uuid::new_v4(),
-            disk_map: std::collections::HashMap::new(),
-            pools: pools.clone(),
-            peer_sys: crate::cluster::rpc::S3PeerSys::new_with_instance_ctx(&endpoint_pools, std::sync::Arc::clone(&ctx)),
-            pool_meta: tokio::sync::RwLock::new(pool_meta.clone()),
-            rebalance_meta: tokio::sync::RwLock::new(rebalance_meta.clone()),
-            decommission_cancelers: tokio::sync::RwLock::new(vec![None, None]),
-            start_gate: tokio::sync::Mutex::new(()),
-            pool_meta_save_gate: tokio::sync::Mutex::new(()),
-            ctx: std::sync::Arc::clone(&ctx),
-            bucket_fence_registry: std::sync::Arc::default(),
-        })
-    };
-    (temp_dirs, make_store(), make_store())
-}
-
 #[cfg(test)]
 mod rebalance_unit_tests;

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

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use super::control::{fail_next_rebalance_activation_save_for_test, validate_rebalance_disk_stats_coverage};
+use super::control::validate_rebalance_disk_stats_coverage;
 use super::meta::{
     RebalanceMetaMergeOutcome, RebalanceTerminalEvent, apply_rebalance_save_option, apply_rebalance_terminal_event,
     apply_stopped_at, classify_rebalance_terminal_event, clone_arc_by_index, clone_first_arc, clone_rebalance_pool_stats,
@@ -32,11 +32,7 @@ use super::migration::{
     MigrationBackend, MigrationVersionResult, migrate_entry_version, migrate_entry_version_with_retry_wait,
     rebalance_delete_marker_opts,
 };
-use super::runtime::{
-    RebalanceLocalActivationOutcome, commit_local_rebalance_worker_activation,
-    commit_local_rebalance_worker_activation_candidate, should_fail_repeated_rebalance_bucket_defer,
-    source_cleanup_defer_attempt, stage_local_rebalance_worker_activation,
-};
+use super::runtime::{should_fail_repeated_rebalance_bucket_defer, source_cleanup_defer_attempt};
 use super::worker::{
     RebalanceEntryCleanupResult, ensure_rebalance_listing_disks_available, is_transient_rebalance_error,
     parse_rebalance_max_attempts, rebalance_listing_retry_delay, rebalance_migration_retry_delay,
@@ -52,7 +48,6 @@ use super::worker::{
 use super::{
     DiskStat, GetObjectReader, ObjectInfo, ObjectOptions, RebalSaveOpt, RebalStatus, RebalanceBucketConfigs,
     RebalanceBucketOutcome, RebalanceCleanupWarnings, RebalanceEntryOutcome, RebalanceInfo, RebalanceMeta, RebalanceStats,
-    RebalanceStopPropagationRecord,
 };
 use super::{REBALANCE_DEFERRED_ENTRY_ERROR_PREFIX, REBALANCE_SOURCE_CLEANUP_DEFERRED_ERROR_PREFIX};
 use crate::bucket::replication::{ReplicationState, ReplicationStatusType, replication_state_to_filemeta};
@@ -2713,281 +2708,6 @@ async fn test_start_rebalance_for_id_rejects_stopped_metadata() {
     assert!(err.to_string().contains("was stopped before start"));
 }
 
-#[test]
-fn test_stopped_activation_state_prevents_worker_token_commit() {
-    let mut meta = RebalanceMeta {
-        id: "rebalance-a".to_string(),
-        stopped_at: Some(OffsetDateTime::now_utc()),
-        pool_stats: vec![RebalanceStats {
-            participating: true,
-            info: RebalanceInfo {
-                status: RebalStatus::Started,
-                ..Default::default()
-            },
-            ..Default::default()
-        }],
-        ..Default::default()
-    };
-
-    let outcome = commit_local_rebalance_worker_activation(&mut meta, "rebalance-a", tokio_util::sync::CancellationToken::new())
-        .expect("stopped metadata should produce a non-start outcome");
-    assert_eq!(outcome, RebalanceLocalActivationOutcome::NotStartedTerminal);
-    assert!(meta.cancel.is_none(), "stopped rebalance must not receive a worker token");
-}
-
-#[test]
-fn test_rebalance_activation_candidate_does_not_clobber_replacement_token() {
-    let mut local = RebalanceMeta {
-        id: "rebalance-a".to_string(),
-        pool_stats: vec![RebalanceStats {
-            participating: true,
-            buckets: vec!["bucket-a".to_string()],
-            info: RebalanceInfo {
-                status: RebalStatus::Started,
-                ..Default::default()
-            },
-            ..Default::default()
-        }],
-        ..Default::default()
-    };
-    let (candidate, outcome, must_persist) =
-        stage_local_rebalance_worker_activation(&local, "rebalance-a", CancellationToken::new(), OffsetDateTime::UNIX_EPOCH)
-            .expect("active activation candidate should be staged");
-    assert_eq!(outcome, RebalanceLocalActivationOutcome::Started);
-    assert!(!must_persist);
-
-    let replacement = CancellationToken::new();
-    local.cancel = Some(replacement.clone());
-    let err = commit_local_rebalance_worker_activation_candidate(&mut local, "rebalance-a", None, candidate)
-        .expect_err("a replacement token must reject the stale activation candidate");
-    assert!(err.to_string().contains("worker token changed"));
-    assert_eq!(local.cancel.as_ref(), Some(&replacement));
-}
-
-#[tokio::test]
-#[serial_test::serial]
-async fn test_rebalance_start_save_failure_retries_persisted_completed_state() {
-    let active = RebalanceMeta {
-        id: "rebalance-real-save-completed".to_string(),
-        percent_free_goal: 0.5,
-        pool_stats: vec![RebalanceStats {
-            participating: true,
-            init_free_space: 400,
-            init_capacity: 1_000,
-            bytes: 100,
-            info: RebalanceInfo {
-                status: RebalStatus::Started,
-                ..Default::default()
-            },
-            ..Default::default()
-        }],
-        ..Default::default()
-    };
-    let (_temp_dirs, store) = super::test_store_with_persisted_rebalance_meta(active).await;
-
-    fail_next_rebalance_activation_save_for_test("rebalance-real-save-completed");
-    let err = store
-        .start_rebalance_under_gate()
-        .await
-        .expect_err("the injected first activation save must fail through the real start path");
-    assert!(err.to_string().contains("injected rebalance activation save failure"));
-    {
-        let local = store.rebalance_meta.read().await;
-        let local = local.as_ref().expect("local rebalance metadata should remain present");
-        assert_eq!(local.pool_stats[0].info.status, RebalStatus::Started);
-        assert!(local.cancel.is_none(), "failed persistence must not publish a worker token");
-    }
-    let mut after_failure = RebalanceMeta::new();
-    after_failure
-        .load(store.pools[0].clone())
-        .await
-        .expect("active metadata should remain readable after the failed save");
-    assert_eq!(after_failure.pool_stats[0].info.status, RebalStatus::Started);
-
-    store
-        .start_rebalance_under_gate()
-        .await
-        .expect("the real start path must retry and persist the terminal candidate");
-
-    let mut persisted = RebalanceMeta::new();
-    persisted
-        .load(store.pools[0].clone())
-        .await
-        .expect("retry-persisted completed metadata should be readable");
-    assert_eq!(persisted.pool_stats[0].info.status, RebalStatus::Completed);
-    let local = store.rebalance_meta.read().await;
-    let local = local.as_ref().expect("local rebalance metadata should remain present");
-    assert_eq!(local.pool_stats[0].info.status, RebalStatus::Completed);
-    assert!(local.cancel.is_none());
-}
-
-#[tokio::test]
-#[serial_test::serial]
-async fn test_rebalance_start_save_failure_retries_persisted_stopped_state() {
-    let active = RebalanceMeta {
-        id: "rebalance-real-save-stopped".to_string(),
-        pool_stats: vec![RebalanceStats {
-            participating: true,
-            info: RebalanceInfo {
-                status: RebalStatus::Started,
-                ..Default::default()
-            },
-            ..Default::default()
-        }],
-        ..Default::default()
-    };
-    let (_temp_dirs, store) = super::test_store_with_persisted_rebalance_meta(active).await;
-    let stopped_at = OffsetDateTime::from_unix_timestamp(1_000).expect("test timestamp should be valid");
-    {
-        let mut local = store.rebalance_meta.write().await;
-        let local = local.as_mut().expect("local rebalance metadata should remain present");
-        local.stopped_at = Some(stopped_at);
-        local.pool_stats[0].info.status = RebalStatus::Stopped;
-        local.pool_stats[0].info.end_time = Some(stopped_at);
-    }
-
-    fail_next_rebalance_activation_save_for_test("rebalance-real-save-stopped");
-    let err = store
-        .start_rebalance_under_gate()
-        .await
-        .expect_err("the injected first stopped-state save must fail through the real start path");
-    assert!(err.to_string().contains("injected rebalance activation save failure"));
-    let mut after_failure = RebalanceMeta::new();
-    after_failure
-        .load(store.pools[0].clone())
-        .await
-        .expect("active metadata should remain readable after the failed save");
-    assert_eq!(after_failure.pool_stats[0].info.status, RebalStatus::Started);
-    {
-        let local = store.rebalance_meta.read().await;
-        let local = local.as_ref().expect("local rebalance metadata should remain present");
-        assert_eq!(local.pool_stats[0].info.status, RebalStatus::Stopped);
-        assert!(local.cancel.is_none(), "failed persistence must not publish a worker token");
-    }
-
-    store
-        .start_rebalance_under_gate()
-        .await
-        .expect("the real start path must retry and persist the stopped candidate");
-
-    let mut persisted = RebalanceMeta::new();
-    persisted
-        .load(store.pools[0].clone())
-        .await
-        .expect("retry-persisted stopped metadata should be readable");
-    assert_eq!(persisted.stopped_at, Some(stopped_at));
-    assert_eq!(persisted.pool_stats[0].info.status, RebalStatus::Stopped);
-    let local = store.rebalance_meta.read().await;
-    let local = local.as_ref().expect("local rebalance metadata should remain present");
-    assert_eq!(local.pool_stats[0].info.status, RebalStatus::Stopped);
-    assert!(local.cancel.is_none());
-}
-
-#[tokio::test]
-async fn test_old_worker_cannot_mutate_replacement_rebalance_state() {
-    let meta = RebalanceMeta {
-        id: "rebalance-b".to_string(),
-        pool_stats: vec![RebalanceStats {
-            participating: true,
-            buckets: vec!["bucket-a".to_string()],
-            info: RebalanceInfo {
-                status: RebalStatus::Started,
-                ..Default::default()
-            },
-            ..Default::default()
-        }],
-        ..Default::default()
-    };
-    let store = test_store_with_rebalance_meta(meta);
-    let fi = FileInfo {
-        size: 128,
-        ..Default::default()
-    };
-
-    for err in [
-        store
-            .next_rebal_bucket(0, "rebalance-a")
-            .await
-            .expect_err("old worker must not read replacement work"),
-        store
-            .bucket_rebalance_done(0, "bucket-a".to_string(), "rebalance-a")
-            .await
-            .expect_err("old worker must not complete replacement bucket"),
-        store
-            .update_pool_stats_batch_for_rebalance(0, "bucket-a".to_string(), &[&fi], "rebalance-a")
-            .await
-            .expect_err("old worker must not update replacement stats"),
-        store
-            .check_if_rebalance_done(0, "rebalance-a")
-            .await
-            .expect_err("old worker must not complete replacement pool"),
-        store
-            .save_rebalance_stats_for_id(0, RebalSaveOpt::Stats, "rebalance-a")
-            .await
-            .expect_err("old save task must not persist replacement metadata"),
-        store
-            .save_rebalance_stats_for_id(usize::MAX, RebalSaveOpt::StoppedAt, "rebalance-a")
-            .await
-            .expect_err("old stop path must not persist replacement metadata"),
-    ] {
-        assert!(err.to_string().contains("stale rebalance worker rejected"));
-    }
-
-    let meta = store.rebalance_meta.read().await;
-    let meta = meta.as_ref().expect("replacement metadata should remain present");
-    assert_eq!(meta.id, "rebalance-b");
-    assert!(meta.pool_stats[0].rebalanced_buckets.is_empty());
-    assert_eq!(meta.pool_stats[0].bytes, 0);
-    assert_eq!(meta.pool_stats[0].info.status, RebalStatus::Started);
-    assert!(meta.stopped_at.is_none());
-}
-
-#[tokio::test]
-async fn test_rebalance_metadata_reload_under_start_gate_does_not_reacquire_gate() {
-    let store = test_store_with_rebalance_meta(RebalanceMeta::default());
-    let _start_guard = store.start_gate.lock().await;
-
-    let err = store
-        .load_rebalance_meta_under_start_gate()
-        .await
-        .expect_err("empty test store should reach the metadata load without waiting on start_gate again");
-
-    assert!(err.to_string().contains("no pools available"));
-}
-
-#[tokio::test]
-async fn test_stale_stop_propagation_cannot_mutate_replacement_rebalance() {
-    let meta = RebalanceMeta {
-        id: "rebalance-b".to_string(),
-        pool_stats: vec![RebalanceStats {
-            participating: true,
-            info: RebalanceInfo {
-                status: RebalStatus::Started,
-                ..Default::default()
-            },
-            ..Default::default()
-        }],
-        ..Default::default()
-    };
-    let store = test_store_with_rebalance_meta(meta);
-    let record = RebalanceStopPropagationRecord {
-        stop_failures: vec!["old rebalance stop failed".to_string()],
-        ..Default::default()
-    };
-
-    let err = store
-        .record_rebalance_stop_propagation("rebalance-a", record)
-        .await
-        .expect_err("old propagation failure must not mutate replacement metadata");
-
-    assert!(err.to_string().contains("stale rebalance worker rejected"));
-    let meta = store.rebalance_meta.read().await;
-    let meta = meta.as_ref().expect("replacement metadata should remain present");
-    assert_eq!(meta.id, "rebalance-b");
-    assert!(meta.last_refreshed_at.is_none());
-    assert!(meta.pool_stats[0].info.last_error.is_none());
-}
-
 fn test_store_with_rebalance_meta(meta: RebalanceMeta) -> Arc<crate::store::ECStore> {
     let endpoint_pools: crate::layout::endpoints::EndpointServerPools = Vec::new().into();
     Arc::new(crate::store::ECStore {

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

@@ -1,4 +1,3 @@
-use super::control::RebalanceWorkerActivationFence;
 use super::meta::{
     apply_rebalance_save_option, apply_rebalance_terminal_event, classify_rebalance_terminal_event, clone_first_arc,
     complete_rebalance_pools_at_goal, complete_rebalance_pools_with_empty_queue, ensure_valid_rebalance_pool_index,
@@ -40,97 +39,9 @@ pub(super) fn source_cleanup_defer_attempt(deferred_attempts: &mut HashMap<Strin
     *attempts
 }
 
-#[derive(Clone, Copy, Debug, PartialEq, Eq)]
-pub(super) enum RebalanceLocalActivationOutcome {
-    Started,
-    NotStartedTerminal,
-}
-
-pub(super) fn commit_local_rebalance_worker_activation(
-    meta: &mut super::RebalanceMeta,
-    expected_id: &str,
-    cancel: CancellationToken,
-) -> Result<RebalanceLocalActivationOutcome> {
-    if meta.id != expected_id {
-        return Err(Error::other(format!(
-            "rebalance metadata changed before local worker activation: expected {expected_id}, found {}",
-            meta.id
-        )));
-    }
-    if meta.stopped_at.is_some() || !is_rebalance_in_progress(meta) {
-        return Ok(RebalanceLocalActivationOutcome::NotStartedTerminal);
-    }
-    meta.cancel = Some(cancel);
-    Ok(RebalanceLocalActivationOutcome::Started)
-}
-
-pub(super) fn stage_local_rebalance_worker_activation(
-    meta: &super::RebalanceMeta,
-    expected_id: &str,
-    cancel: CancellationToken,
-    now: OffsetDateTime,
-) -> Result<(super::RebalanceMeta, RebalanceLocalActivationOutcome, bool)> {
-    let mut candidate = meta.clone();
-    let completed_at_goal = complete_rebalance_pools_at_goal(&mut candidate, now);
-    let completed_empty_queue = complete_rebalance_pools_with_empty_queue(&mut candidate, now);
-    let outcome = commit_local_rebalance_worker_activation(&mut candidate, expected_id, cancel)?;
-    let must_persist =
-        completed_at_goal || completed_empty_queue || outcome == RebalanceLocalActivationOutcome::NotStartedTerminal;
-    Ok((candidate, outcome, must_persist))
-}
-
-pub(super) fn commit_local_rebalance_worker_activation_candidate(
-    current: &mut super::RebalanceMeta,
-    expected_id: &str,
-    expected_cancel: Option<&CancellationToken>,
-    candidate: super::RebalanceMeta,
-) -> Result<()> {
-    if current.id != expected_id || candidate.id != expected_id {
-        return Err(Error::other(format!(
-            "rebalance metadata changed before local worker activation commit: expected {expected_id}, found {}",
-            current.id
-        )));
-    }
-    if !Arc::ptr_eq(&current.activation_gate, &candidate.activation_gate) {
-        return Err(Error::other(format!(
-            "rebalance activation gate changed before local worker activation commit: {expected_id}"
-        )));
-    }
-    if current.cancel.as_ref() != expected_cancel {
-        return Err(Error::other(format!(
-            "rebalance worker token changed before local worker activation commit: {expected_id}"
-        )));
-    }
-    *current = candidate;
-    Ok(())
-}
-
-pub(super) fn rollback_local_rebalance_worker_activation(
-    meta: Option<&mut super::RebalanceMeta>,
-    expected_id: &str,
-    activation_token: &CancellationToken,
-) -> bool {
-    let Some(meta) = meta else {
-        return false;
-    };
-    if meta.id != expected_id || meta.cancel.as_ref() != Some(activation_token) {
-        return false;
-    }
-    if let Some(cancel) = meta.cancel.take() {
-        cancel.cancel();
-        return true;
-    }
-    false
-}
-
 impl ECStore {
     #[tracing::instrument(skip_all)]
     pub async fn start_rebalance(self: &Arc<Self>) -> Result<()> {
-        let _start_guard = self.start_gate.lock().await;
-        self.start_rebalance_under_gate().await
-    }
-
-    pub(super) async fn start_rebalance_under_gate(self: &Arc<Self>) -> Result<()> {
         info!(
             event = EVENT_REBALANCE_STATE,
             component = LOG_COMPONENT_ECSTORE,
@@ -138,27 +49,12 @@ impl ECStore {
             state = "starting",
             "Starting rebalance"
         );
-        let expected_id: Arc<str> = {
-            let rebalance_meta = self.rebalance_meta.read().await;
-            Arc::from(rebalance_meta.as_ref().ok_or(Error::ConfigNotFound)?.id.as_str())
-        };
-        let pool = clone_first_arc(self.pools.as_slice(), "start_rebalance: no pools available")?;
-        let activation_fence = match self
-            .fence_rebalance_worker_activation(pool.clone(), expected_id.as_ref())
-            .await?
-        {
-            RebalanceWorkerActivationFence::Ready(fence) => fence,
-            RebalanceWorkerActivationFence::NotStartedTerminal => return Ok(()),
-        };
-
         let decommission_running = self.is_decommission_running().await;
+        // let rebalance_meta = self.rebalance_meta.read().await;
 
         let cancel_tx = CancellationToken::new();
         let rx = cancel_tx.clone();
-        let activation_outcome;
-        let candidate;
-        let expected_cancel;
-        let must_persist;
+        let mut meta_to_save = None;
 
         {
             let mut rebalance_meta = self.rebalance_meta.write().await;
@@ -178,70 +74,25 @@ impl ECStore {
                 );
                 return Ok(());
             }
-            expected_cancel = meta.cancel.clone();
-            (candidate, activation_outcome, must_persist) = stage_local_rebalance_worker_activation(
-                meta,
-                expected_id.as_ref(),
-                cancel_tx.clone(),
-                OffsetDateTime::now_utc(),
+            let now = OffsetDateTime::now_utc();
+            if complete_rebalance_pools_at_goal(meta, now) {
+                meta_to_save = Some(meta.clone());
+            }
+            if complete_rebalance_pools_with_empty_queue(meta, now) {
+                meta_to_save = Some(meta.clone());
+            }
+            meta.cancel = Some(cancel_tx);
+
+            drop(rebalance_meta);
+        }
+
+        if let Some(meta) = meta_to_save {
+            let pool = clone_first_arc(self.pools.as_slice(), "start_rebalance: no pools available")?;
+            resolve_rebalance_meta_save_result(
+                self.save_rebalance_meta_with_merge(pool, &meta, "start_rebalance complete pools at goal")
+                    .await,
+                "start_rebalance complete pools at goal",
             )?;
-            if let Err(err) = activation_fence.ensure_held() {
-                cancel_tx.cancel();
-                return Err(err);
-            }
-            if !must_persist
-                && let Err(err) = commit_local_rebalance_worker_activation_candidate(
-                    meta,
-                    expected_id.as_ref(),
-                    expected_cancel.as_ref(),
-                    candidate.clone(),
-                )
-            {
-                cancel_tx.cancel();
-                return Err(err);
-            }
-        }
-
-        if must_persist {
-            let save_result = resolve_rebalance_meta_save_result(
-                self.save_rebalance_meta_under_activation_fence(
-                    pool,
-                    &candidate,
-                    "start_rebalance persist activation candidate",
-                    activation_fence.as_ref(),
-                    expected_id.as_ref(),
-                )
-                .await,
-                "start_rebalance persist activation candidate",
-            );
-            if let Err(err) = save_result {
-                cancel_tx.cancel();
-                return Err(err);
-            }
-            let mut rebalance_meta = self.rebalance_meta.write().await;
-            let Some(meta) = rebalance_meta.as_mut() else {
-                cancel_tx.cancel();
-                return Err(Error::ConfigNotFound);
-            };
-            if let Err(err) = commit_local_rebalance_worker_activation_candidate(
-                meta,
-                expected_id.as_ref(),
-                expected_cancel.as_ref(),
-                candidate,
-            ) {
-                cancel_tx.cancel();
-                return Err(err);
-            }
-        }
-        if !must_persist && let Err(err) = activation_fence.ensure_held() {
-            let mut rebalance_meta = self.rebalance_meta.write().await;
-            rollback_local_rebalance_worker_activation(rebalance_meta.as_mut(), expected_id.as_ref(), &rx);
-            return Err(err);
-        }
-        drop(activation_fence);
-
-        if activation_outcome != RebalanceLocalActivationOutcome::Started {
-            return Ok(());
         }
 
         let participants = if let Some(ref meta) = *self.rebalance_meta.read().await {
@@ -259,8 +110,6 @@ impl ECStore {
         };
 
         if !participants.iter().any(|participating| *participating) {
-            let mut rebalance_meta = self.rebalance_meta.write().await;
-            rollback_local_rebalance_worker_activation(rebalance_meta.as_mut(), expected_id.as_ref(), &rx);
             debug!(
                 event = EVENT_REBALANCE_STATE,
                 component = LOG_COMPONENT_ECSTORE,
@@ -272,11 +121,6 @@ impl ECStore {
             return Ok(());
         }
 
-        #[cfg(test)]
-        let endpoints = self.instance_endpoints().unwrap_or_else(|| self.endpoints());
-        #[cfg(not(test))]
-        let endpoints = self.endpoints();
-
         let mut workers_started = 0usize;
         for (idx, participating) in participants.iter().enumerate() {
             if !*participating {
@@ -292,7 +136,7 @@ impl ECStore {
                 continue;
             }
 
-            if !runtime_sources::endpoint_pool_is_local(&endpoints, idx) {
+            if !runtime_sources::endpoint_pool_is_local(idx) {
                 debug!(
                     event = EVENT_REBALANCE_STATE,
                     component = LOG_COMPONENT_ECSTORE,
@@ -308,10 +152,9 @@ impl ECStore {
             let pool_idx = idx;
             let store = self.clone();
             let rx_clone = rx.clone();
-            let worker_id = Arc::clone(&expected_id);
             workers_started += 1;
             tokio::spawn(async move {
-                if let Err(err) = store.rebalance_buckets(rx_clone, pool_idx, worker_id).await {
+                if let Err(err) = store.rebalance_buckets(rx_clone, pool_idx).await {
                     error!(
                         event = EVENT_REBALANCE_STATE,
                         component = LOG_COMPONENT_ECSTORE,
@@ -335,8 +178,6 @@ impl ECStore {
         }
 
         if workers_started == 0 {
-            let mut rebalance_meta = self.rebalance_meta.write().await;
-            rollback_local_rebalance_worker_activation(rebalance_meta.as_mut(), expected_id.as_ref(), &rx);
             debug!(
                 event = EVENT_REBALANCE_STATE,
                 component = LOG_COMPONENT_ECSTORE,
@@ -360,14 +201,13 @@ impl ECStore {
     }
 
     #[tracing::instrument(skip(self, rx))]
-    async fn rebalance_buckets(self: &Arc<Self>, rx: CancellationToken, pool_index: usize, rebalance_id: Arc<str>) -> Result<()> {
+    async fn rebalance_buckets(self: &Arc<Self>, rx: CancellationToken, pool_index: usize) -> Result<()> {
         ensure_valid_rebalance_pool_index(self.pools.len(), pool_index)?;
 
         let (done_tx, mut done_rx) = tokio::sync::mpsc::channel::<Result<()>>(1);
 
         // Save rebalance metadata periodically
         let store = self.clone();
-        let save_rebalance_id = Arc::clone(&rebalance_id);
         let save_task = tokio::spawn(async move {
             let mut timer = tokio::time::interval_at(Instant::now() + Duration::from_secs(30), Duration::from_secs(10));
             let mut msg: String;
@@ -381,11 +221,6 @@ impl ECStore {
                         let terminal_event = classify_rebalance_terminal_event(result, now);
                         msg = terminal_event.message().to_string();
                         let mut rebalance_meta = store.rebalance_meta.write().await;
-                        super::control::ensure_rebalance_run_id(
-                            rebalance_meta.as_ref(),
-                            save_rebalance_id.as_ref(),
-                            "apply rebalance terminal event",
-                        )?;
                         if let Some(meta) = rebalance_meta.as_mut() {
                             let meta_stopped = meta.stopped_at.is_some();
                             if let Some(pool_stat) = meta.pool_stats.get_mut(pool_index) {
@@ -434,10 +269,7 @@ impl ECStore {
                     }
                 }
 
-                if let Err(err) = store
-                    .save_rebalance_stats_for_id(pool_index, RebalSaveOpt::Stats, save_rebalance_id.as_ref())
-                    .await
-                {
+                if let Err(err) = store.save_rebalance_stats(pool_index, RebalSaveOpt::Stats).await {
                     let wrapped = Error::other(format!("rebalance save_task stats save failed for pool {pool_index}: {err}"));
                     error!("{} err: {:?}", msg, wrapped);
                     if quit {
@@ -503,7 +335,7 @@ impl ECStore {
                 break;
             }
 
-            let next_bucket = match self.next_rebal_bucket(pool_index, rebalance_id.as_ref()).await {
+            let next_bucket = match self.next_rebal_bucket(pool_index).await {
                 Ok(bucket) => bucket,
                 Err(err) => {
                     error!(
@@ -535,8 +367,7 @@ impl ECStore {
                 );
 
                 let outcome = match resolve_rebalance_bucket_result(
-                    self.rebalance_bucket(rx.clone(), bucket.clone(), pool_index, Arc::clone(&rebalance_id))
-                        .await,
+                    self.rebalance_bucket(rx.clone(), bucket.clone(), pool_index).await,
                     pool_index,
                     &bucket,
                 ) {
@@ -599,7 +430,7 @@ impl ECStore {
                         "Deferred rebalance bucket after transient object failures"
                     );
                     if let Err(err) = self
-                        .defer_rebalance_bucket(pool_index, bucket.clone(), last_error.clone(), rebalance_id.as_ref())
+                        .defer_rebalance_bucket(pool_index, bucket.clone(), last_error.clone())
                         .await
                     {
                         error!(
@@ -663,7 +494,7 @@ impl ECStore {
                     "Completed rebalance bucket"
                 );
                 source_cleanup_deferred_attempts.remove(&bucket);
-                if let Err(err) = self.bucket_rebalance_done(pool_index, bucket, rebalance_id.as_ref()).await {
+                if let Err(err) = self.bucket_rebalance_done(pool_index, bucket).await {
                     error!(
                         event = EVENT_REBALANCE_BUCKET,
                         component = LOG_COMPONENT_ECSTORE,
@@ -724,9 +555,8 @@ impl ECStore {
         final_result
     }
 
-    pub(super) async fn check_if_rebalance_done(&self, pool_index: usize, expected_id: &str) -> Result<bool> {
+    pub(super) async fn check_if_rebalance_done(&self, pool_index: usize) -> bool {
         let mut rebalance_meta = self.rebalance_meta.write().await;
-        super::control::ensure_rebalance_worker_active(rebalance_meta.as_ref(), expected_id, "check rebalance completion")?;
 
         if let Some(meta) = rebalance_meta.as_mut()
             && let Some(pool_stat) = meta.pool_stats.get_mut(pool_index)
@@ -741,7 +571,7 @@ impl ECStore {
                     state = "already_completed",
                     "Rebalance pool is already completed"
                 );
-                return Ok(true);
+                return true;
             }
 
             // Mark pool rebalance as done only after it reaches the PercentFreeGoal.
@@ -771,30 +601,19 @@ impl ECStore {
                     percent_free = pfi,
                     "Marked rebalance pool completed"
                 );
-                return Ok(true);
+                return true;
             }
         }
 
-        Ok(false)
+        false
     }
 }
 
 impl ECStore {
     #[tracing::instrument(skip(self))]
     pub async fn save_rebalance_stats(&self, pool_idx: usize, opt: RebalSaveOpt) -> Result<()> {
-        self.save_rebalance_stats_inner(pool_idx, opt, None).await
-    }
-
-    pub async fn save_rebalance_stats_for_id(&self, pool_idx: usize, opt: RebalSaveOpt, expected_id: &str) -> Result<()> {
-        self.save_rebalance_stats_inner(pool_idx, opt, Some(expected_id)).await
-    }
-
-    async fn save_rebalance_stats_inner(&self, pool_idx: usize, opt: RebalSaveOpt, expected_id: Option<&str>) -> Result<()> {
         let meta_to_save = {
             let mut rebalance_meta = self.rebalance_meta.write().await;
-            if let Some(expected_id) = expected_id {
-                super::control::ensure_rebalance_run_id(rebalance_meta.as_ref(), expected_id, "save rebalance stats")?;
-            }
             let Some(meta) = rebalance_meta.as_mut() else {
                 return Ok(());
             };
@@ -816,14 +635,10 @@ impl ECStore {
             "Rebalance metadata save requested"
         );
         let stage = format!("save_rebalance_stats for pool {pool_idx} opt {opt:?}");
-        let save_result = match expected_id {
-            Some(expected_id) => {
-                self.save_rebalance_meta_for_id_with_merge(pool, &meta_to_save, stage.as_str(), expected_id)
-                    .await
-            }
-            None => self.save_rebalance_meta_with_merge(pool, &meta_to_save, stage.as_str()).await,
-        };
-        resolve_rebalance_meta_save_result(save_result, stage.as_str())?;
+        resolve_rebalance_meta_save_result(
+            self.save_rebalance_meta_with_merge(pool, &meta_to_save, stage.as_str()).await,
+            stage.as_str(),
+        )?;
 
         Ok(())
     }

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

@@ -144,8 +144,6 @@ pub struct RebalanceMeta {
     #[serde(skip)]
     pub cancel: Option<CancellationToken>, // To be invoked on rebalance-stop
     #[serde(skip)]
-    pub activation_gate: std::sync::Arc<tokio::sync::RwLock<()>>,
-    #[serde(skip)]
     pub last_refreshed_at: Option<OffsetDateTime>,
     #[serde(rename = "stopTs")]
     pub stopped_at: Option<OffsetDateTime>, // Time when rebalance-stop was issued

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

@@ -100,17 +100,17 @@ pub(super) fn resolve_rebalance_meta_save_result(result: Result<()>, stage: &str
     result.map_err(|err| Error::other(format!("rebalance meta save failed during {stage}: {err}")))
 }
 
-pub(super) fn rebalance_meta_lock_error(err: rustfs_lock::LockError, mode: &'static str) -> Error {
+pub(super) fn rebalance_meta_lock_error(err: rustfs_lock::LockError) -> Error {
     match err {
         rustfs_lock::LockError::QuorumNotReached { required, achieved } => Error::NamespaceLockQuorumUnavailable {
-            mode,
+            mode: "write",
             bucket: crate::disk::RUSTFS_META_BUCKET.to_string(),
             object: REBAL_META_NAME.to_string(),
             required,
             achieved,
         },
         other => Error::other(format!(
-            "failed to acquire rebalance metadata {mode} lock on {}/{}: {other}",
+            "failed to acquire rebalance metadata write lock on {}/{}: {other}",
             crate::disk::RUSTFS_META_BUCKET,
             REBAL_META_NAME
         )),

crates/ecstore/src/set_disk/mod.rs

@@ -735,9 +735,6 @@ pub(crate) use core::io_primitives::disk_call_counters;
 mod ctx;
 mod metadata;
 mod ops;
-
-#[cfg(test)]
-pub(crate) use ops::hermetic_set_disks_isolated;
 #[cfg(test)]
 pub(crate) use ops::multipart::NewMultipartUploadCommitObservation;
 #[cfg(any(test, feature = "test-util"))]
@@ -1452,21 +1449,6 @@ mod prepared_get_object_metadata_tests {
 }
 
 impl SetDisks {
-    #[cfg(test)]
-    async fn pause_tiered_metadata_commit(bucket: &str, object: &str) {
-        let barrier = TIERED_METADATA_COMMIT_BARRIER
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("tiered metadata commit barrier should not be poisoned")
-            .as_ref()
-            .filter(|barrier| barrier.bucket == bucket && barrier.object == object)
-            .cloned();
-        if let Some(barrier) = barrier {
-            barrier.arrived.notify_one();
-            barrier.release.notified().await;
-        }
-    }
-
     pub(crate) async fn prepare_get_object_metadata(
         &self,
         bucket: &str,
@@ -4735,8 +4717,6 @@ impl SetDisks {
         )?;
         let fi = build_tiered_decommission_file_info(bucket, object, fi, layout);
         let write_quorum = layout.write_quorum;
-        #[cfg(test)]
-        Self::pause_tiered_metadata_commit(bucket, object).await;
         if _lock_guard.as_ref().is_some_and(|guard| guard.is_lock_lost())
             || opts
                 .namespace_lock_fence
@@ -4784,66 +4764,6 @@ impl SetDisks {
     }
 }
 
-#[cfg(test)]
-struct TieredMetadataCommitBarrierState {
-    bucket: String,
-    object: String,
-    arrived: tokio::sync::Notify,
-    release: tokio::sync::Notify,
-}
-
-#[cfg(test)]
-pub(crate) struct TieredMetadataCommitBarrier {
-    state: Arc<TieredMetadataCommitBarrierState>,
-}
-
-#[cfg(test)]
-static TIERED_METADATA_COMMIT_BARRIER: std::sync::OnceLock<std::sync::Mutex<Option<Arc<TieredMetadataCommitBarrierState>>>> =
-    std::sync::OnceLock::new();
-
-#[cfg(test)]
-impl TieredMetadataCommitBarrier {
-    pub(crate) fn install(bucket: &str, object: &str) -> Self {
-        let state = Arc::new(TieredMetadataCommitBarrierState {
-            bucket: bucket.to_string(),
-            object: object.to_string(),
-            arrived: tokio::sync::Notify::new(),
-            release: tokio::sync::Notify::new(),
-        });
-        let mut slot = TIERED_METADATA_COMMIT_BARRIER
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("tiered metadata commit barrier should not be poisoned");
-        assert!(slot.is_none(), "tiered metadata commit barrier must be unique");
-        *slot = Some(Arc::clone(&state));
-        Self { state }
-    }
-
-    pub(crate) async fn wait_until_paused(&self) {
-        tokio::time::timeout(Duration::from_secs(30), self.state.arrived.notified())
-            .await
-            .expect("tiered metadata write should reach its deterministic commit barrier");
-    }
-
-    pub(crate) fn release(&self) {
-        self.state.release.notify_one();
-    }
-}
-
-#[cfg(test)]
-impl Drop for TieredMetadataCommitBarrier {
-    fn drop(&mut self) {
-        self.state.release.notify_one();
-        let mut slot = TIERED_METADATA_COMMIT_BARRIER
-            .get_or_init(|| std::sync::Mutex::new(None))
-            .lock()
-            .expect("tiered metadata commit barrier should not be poisoned");
-        if slot.as_ref().is_some_and(|state| Arc::ptr_eq(state, &self.state)) {
-            *slot = None;
-        }
-    }
-}
-
 #[derive(Debug, PartialEq, Eq)]
 struct ObjProps {
     successor_mod_time: Option<OffsetDateTime>,

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

@@ -25,6 +25,3 @@ pub(crate) mod list;
 pub(crate) mod locking;
 pub(crate) mod multipart;
 pub(crate) mod object;
-
-#[cfg(test)]
-pub(crate) use object::hermetic_set_disks_support::hermetic_set_disks_isolated;

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

@@ -86,8 +86,6 @@ struct MultipartCommitBarrierState {
     pause: MultipartCommitPause,
     expected_arrivals: usize,
     arrivals: AtomicUsize,
-    #[cfg(test)]
-    committed: AtomicBool,
     arrived: tokio::sync::Notify,
     release: tokio::sync::Semaphore,
 }
@@ -115,8 +113,6 @@ impl MultipartCommitBarrier {
             pause,
             expected_arrivals,
             arrivals: AtomicUsize::new(0),
-            #[cfg(test)]
-            committed: AtomicBool::new(false),
             arrived: tokio::sync::Notify::new(),
             release: tokio::sync::Semaphore::new(0),
         });
@@ -147,11 +143,6 @@ impl MultipartCommitBarrier {
     pub fn release(&self) {
         self.state.release.add_permits(self.state.expected_arrivals);
     }
-
-    #[cfg(test)]
-    pub(crate) fn commit_observed(&self) -> bool {
-        self.state.committed.load(Ordering::Acquire)
-    }
 }
 
 #[cfg(any(test, feature = "test-util"))]
@@ -272,20 +263,6 @@ async fn pause_multipart_commit(bucket: &str, object: &str, pause: MultipartComm
     }
 }
 
-#[cfg(test)]
-fn observe_multipart_commit(bucket: &str, object: &str, pause: MultipartCommitPause) {
-    let slot = MULTIPART_COMMIT_BARRIER
-        .get_or_init(|| std::sync::Mutex::new(None))
-        .lock()
-        .expect("multipart commit barrier mutex should not poison");
-    if let Some(barrier) = slot
-        .as_ref()
-        .filter(|barrier| barrier.bucket == bucket && barrier.object == object && barrier.pause == pause)
-    {
-        barrier.committed.store(true, Ordering::Release);
-    }
-}
-
 fn map_upload_id_metadata_error(bucket: &str, object: &str, upload_id: &str, err: DiskError) -> Error {
     if err == DiskError::FileNotFound {
         return StorageError::InvalidUploadID(bucket.to_owned(), object.to_owned(), upload_id.to_owned());
@@ -1343,19 +1320,6 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
             pause_multipart_commit(bucket, object, MultipartCommitPause::PutPartBeforeLockLost).await;
             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)?;
-            if opts
-                .namespace_lock_fence
-                .as_ref()
-                .is_some_and(NamespaceLockFence::is_lock_lost)
-            {
-                return Err(StorageError::NamespaceLockQuorumUnavailable {
-                    mode: "put_object_part_outer_lock",
-                    bucket: bucket.to_string(),
-                    object: object.to_string(),
-                    required: 1,
-                    achieved: 0,
-                });
-            }
             ensure_multipart_bucket_lifecycle_lock_held(bucket, object, opts)?;
 
             let _ = self
@@ -1376,8 +1340,6 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
                     }),
                 )
                 .await?;
-            #[cfg(test)]
-            observe_multipart_commit(bucket, object, MultipartCommitPause::PutPartBeforeLockLost);
 
             #[cfg(test)]
             pause_multipart_commit(bucket, object, MultipartCommitPause::PutPartAfterRename).await;
@@ -1758,10 +1720,7 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for SetDisks {
         .await
         .map_err(|e| to_object_err(e.into(), vec![bucket, object]))?;
         #[cfg(test)]
-        {
-            observe_multipart_commit(bucket, object, MultipartCommitPause::NewUploadBeforeLockLost);
-            observe_new_multipart_upload_commit(bucket, object);
-        }
+        observe_new_multipart_upload_commit(bucket, object);
 
         // evalDisks
 

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

@@ -6519,8 +6519,6 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
         };
 
         let find_vid = Uuid::new_v4();
-        #[cfg(test)]
-        pause_delete_object_commit(bucket, object).await;
 
         if mark_delete && (opts.versioned || opts.version_suspended) {
             if !delete_marker {
@@ -6589,6 +6587,8 @@ impl crate::storage_api_contracts::object::ObjectOperations for SetDisks {
             dfi.set_skip_tier_free_version();
         }
 
+        #[cfg(test)]
+        pause_delete_object_commit(bucket, object).await;
         ensure_delete_commit_locks_held(_lock_guard.as_ref(), bucket, object, &opts)?;
         self.delete_object_version(bucket, object, &dfi, opts.delete_marker)
             .await
@@ -7750,7 +7750,9 @@ pub(in crate::set_disk::ops) mod hermetic_set_disks_support {
     /// for tests that never touch context-resolved services registered on the
     /// ambient context (tier config manager, expiry state, ...), because the
     /// isolated context starts every one of those cells fresh.
-    pub(crate) async fn hermetic_set_disks_isolated(disk_count: usize) -> (Vec<TempDir>, Vec<DiskStore>, Arc<SetDisks>) {
+    pub(in crate::set_disk::ops) async fn hermetic_set_disks_isolated(
+        disk_count: usize,
+    ) -> (Vec<TempDir>, Vec<DiskStore>, Arc<SetDisks>) {
         hermetic_set_disks_for_pool_with_default_parity_isolated(disk_count, 0, disk_count / 2).await
     }
 
@@ -11813,7 +11815,6 @@ mod transition_upload_integrity_tests {
                 crate::data_movement::SourceCleanupBucketFence {
                     expected_incarnation_id: None,
                     lifecycle_guard: Some(&bucket_guard),
-                    namespace_lock_lost_signal: None,
                     ..Default::default()
                 },
                 "test_data_movement",

crates/heal/Cargo.toml

@@ -91,6 +91,7 @@ metrics = { workspace = true }
 base64 = { workspace = true }
 bytes = { workspace = true }
 crc-fast = { workspace = true }
+sha2 = { workspace = true }
 
 [dev-dependencies]
 serde_json = { workspace = true, features = ["raw_value"] }

crates/heal/src/heal/erasure_healer.rs

@@ -373,6 +373,11 @@ impl ErasureSetHealer {
         set_disk_id: &str,
         buckets: &[String],
     ) -> Result<(ResumeManager, CheckpointManager)> {
+        if self.replacement_task_id.is_none() && CheckpointManager::is_blocked(&self.disk, task_id).await {
+            return Err(Error::TaskExecutionFailed {
+                message: format!("Resume task {task_id} has a blocked checkpoint"),
+            });
+        }
         // check if resume state exists
         let has_resume_state = if self.replacement_task_id.is_some() {
             ResumeManager::has_replacement_intent(&self.disk, task_id).await

crates/heal/src/heal/resume.rs

@@ -51,6 +51,7 @@ const RESUME_STATE_FILE: &str = "ahm_resume_state.json";
 const REPLACEMENT_INTENT_FILE: &str = "ahm_replacement_intent.json";
 const RESUME_PROGRESS_FILE: &str = "ahm_progress.json";
 pub(super) const RESUME_CHECKPOINT_FILE: &str = "ahm_checkpoint.json";
+pub(super) const RESUME_CHECKPOINT_BLOCKED_FILE: &str = "ahm_checkpoint.blocked";
 const REPLACEMENT_COMPLETION_PROOF_FILE: &str = "ahm_replacement_completion_proof.json";
 const REPLACEMENT_RECOVERY_DIR: &str = "ahm-replacement";
 const REPLACEMENT_INTENT_SEAL_FILE: &str = "ahm_replacement_intent_seal";

crates/heal/src/heal/resume/checkpoint.rs

@@ -13,26 +13,31 @@
 // limitations under the License.
 
 use crate::{Error, Result};
+use base64::Engine as _;
 use serde::{Deserialize, Serialize};
+use sha2::{Digest, Sha256};
 use std::collections::HashSet;
 use std::path::Path;
 use std::sync::{Arc, Mutex};
 use std::time::{SystemTime, UNIX_EPOCH};
-use tokio::sync::RwLock;
+use tokio::sync::{Mutex as AsyncMutex, RwLock};
 use tracing::{debug, warn};
 
-use super::super::{BUCKET_META_PREFIX, DiskStore, HealDiskExt as _, RUSTFS_META_BUCKET};
+use super::super::storage_api::owner::{EcstoreConditionalFileUpdate, EcstoreDiskAPI, EcstoreDiskBytes};
+use super::super::{BUCKET_META_PREFIX, DiskStore, HealDiskExt, RUSTFS_META_BUCKET};
 use super::{
-    LOG_COMPONENT_HEAL, LOG_SUBSYSTEM_RESUME, PersistThrottle, RESUME_CHECKPOINT_FILE, delete_resume_file, path_to_str,
-    validate_resume_task_id,
+    LOG_COMPONENT_HEAL, LOG_SUBSYSTEM_RESUME, PersistThrottle, RESUME_CHECKPOINT_BLOCKED_FILE, RESUME_CHECKPOINT_FILE,
+    delete_resume_file, path_to_str, validate_resume_task_id,
 };
 
 const EVENT_HEAL_CHECKPOINT_STATE: &str = "heal_checkpoint_state";
+const RESUME_CHECKPOINT_DIGEST_FILE: &str = "ahm_checkpoint.sha256";
+const CHECKPOINT_PER_VERSION_SCHEMA: u32 = 5;
 
 /// Current on-disk schema version for `ResumeCheckpoint`. Same rationale as
 /// `CURRENT_RESUME_SCHEMA`: pre-per-version dedup identities are not comparable
 /// to the new `compose_key` identities, so a stale checkpoint is discarded.
-pub(super) const CURRENT_CHECKPOINT_SCHEMA: u32 = 5;
+pub(super) const CURRENT_CHECKPOINT_SCHEMA: u32 = 6;
 
 /// resume checkpoint
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -57,6 +62,11 @@ pub struct ResumeCheckpoint {
     pub failed_objects: HashSet<String>,
     /// skipped objects
     pub skipped_objects: HashSet<String>,
+    /// Integrity digest over the checkpoint with this field set to `None`.
+    /// Keeping it in the checkpoint makes the payload and its authentication
+    /// record one CAS generation instead of two independently-written files.
+    #[serde(default)]
+    pub integrity_digest: Option<String>,
 }
 
 impl ResumeCheckpoint {
@@ -70,6 +80,7 @@ impl ResumeCheckpoint {
             processed_objects: HashSet::new(),
             failed_objects: HashSet::new(),
             skipped_objects: HashSet::new(),
+            integrity_digest: None,
         }
     }
 
@@ -116,17 +127,111 @@ pub struct CheckpointManager {
     disk: DiskStore,
     checkpoint: Arc<RwLock<ResumeCheckpoint>>,
     throttle: Mutex<PersistThrottle>,
+    save_lock: AsyncMutex<()>,
+    last_saved: Mutex<Option<EcstoreDiskBytes>>,
 }
 
 impl CheckpointManager {
+    fn blocked_path(task_id: &str) -> std::path::PathBuf {
+        Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_BLOCKED_FILE}"))
+    }
+
+    /// Return whether a checkpoint was permanently isolated after a malformed
+    /// or unsupported snapshot was observed.
+    pub(crate) async fn is_blocked(disk: &DiskStore, task_id: &str) -> bool {
+        if validate_resume_task_id(task_id).is_err() {
+            return false;
+        }
+        let blocked_path = Self::blocked_path(task_id);
+        let Ok(path) = path_to_str(&blocked_path) else {
+            return false;
+        };
+        match HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, path).await {
+            Ok(_) => true,
+            Err(crate::heal::DiskError::FileNotFound) => false,
+            Err(_) => true,
+        }
+    }
+
+    /// Validate the checkpoint while enumerating resumable state. This reads
+    /// the checkpoint once and also isolates malformed or unsupported data.
+    pub(crate) async fn is_resumable(disk: &DiskStore, task_id: &str) -> Result<bool> {
+        validate_resume_task_id(task_id)?;
+        if Self::is_blocked(disk, task_id).await {
+            return Err(Error::InvalidCheckpoint(format!("Resume task {task_id} has a blocked checkpoint")));
+        }
+        let file_path = Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"));
+        let Ok(path) = path_to_str(&file_path) else {
+            return Err(Error::InvalidCheckpoint("Resume checkpoint path is not valid UTF-8".to_string()));
+        };
+        match HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, path).await {
+            Ok(bytes) if bytes.is_empty() => Ok(true),
+            Ok(bytes) => Self::load_from_data(disk.clone(), task_id, bytes.to_vec())
+                .await
+                .map(|_| true),
+            Err(crate::heal::DiskError::FileNotFound) => Ok(true),
+            Err(error) => Err(error.into()),
+        }
+    }
+
+    async fn block_invalid_snapshot(disk: &DiskStore, task_id: &str) {
+        // This marker is intentionally version-agnostic: an unsupported reader
+        // must stop selector retries until an operator cleans up the snapshot.
+        let blocked_path = Self::blocked_path(task_id);
+        let Ok(path) = path_to_str(&blocked_path) else {
+            return;
+        };
+        let result = EcstoreDiskAPI::compare_and_update_file(
+            disk.as_ref(),
+            RUSTFS_META_BUCKET,
+            path,
+            None,
+            Some(EcstoreDiskBytes::from_static(b"blocked")),
+        )
+        .await;
+        match result {
+            Ok(EcstoreConditionalFileUpdate::Updated | EcstoreConditionalFileUpdate::Mismatch) => {}
+            Ok(EcstoreConditionalFileUpdate::Missing) => warn!(
+                target: "rustfs::heal::resume",
+                event = EVENT_HEAL_CHECKPOINT_STATE,
+                component = LOG_COMPONENT_HEAL,
+                subsystem = LOG_SUBSYSTEM_RESUME,
+                task_id,
+                state = "blocked_marker_write_failed",
+                error = "marker target disappeared",
+                "Heal checkpoint could not persist its blocked marker"
+            ),
+            Err(error) => warn!(
+                target: "rustfs::heal::resume",
+                event = EVENT_HEAL_CHECKPOINT_STATE,
+                component = LOG_COMPONENT_HEAL,
+                subsystem = LOG_SUBSYSTEM_RESUME,
+                task_id,
+                state = "blocked_marker_write_failed",
+                error = %error,
+                "Heal checkpoint could not persist its blocked marker"
+            ),
+        }
+    }
+
     /// create new checkpoint manager
     pub async fn new(disk: DiskStore, task_id: String) -> Result<Self> {
         validate_resume_task_id(&task_id)?;
+        let checkpoint_volume = format!("{RUSTFS_META_BUCKET}/{BUCKET_META_PREFIX}");
+        if let Err(error) = EcstoreDiskAPI::make_volume(disk.as_ref(), &checkpoint_volume).await
+            && error != crate::heal::DiskError::VolumeExists
+        {
+            return Err(Error::TaskExecutionFailed {
+                message: format!("Failed to create checkpoint volume: {error}"),
+            });
+        }
         let checkpoint = ResumeCheckpoint::new(task_id);
         let manager = Self {
             disk,
             checkpoint: Arc::new(RwLock::new(checkpoint)),
             throttle: Mutex::new(PersistThrottle::new()),
+            save_lock: AsyncMutex::new(()),
+            last_saved: Mutex::new(None),
         };
 
         // save initial checkpoint
@@ -140,6 +245,7 @@ impl CheckpointManager {
                 error = %e,
                 "Heal checkpoint persistence failed"
             );
+            return Err(e);
         }
         Ok(manager)
     }
@@ -148,11 +254,22 @@ impl CheckpointManager {
     pub async fn load_from_disk(disk: DiskStore, task_id: &str) -> Result<Self> {
         validate_resume_task_id(task_id)?;
         let checkpoint_data = Self::read_checkpoint_file(&disk, task_id).await?;
-        let mut checkpoint: ResumeCheckpoint =
-            serde_json::from_slice(&checkpoint_data).map_err(|e| Error::TaskExecutionFailed {
-                message: format!("Failed to deserialize checkpoint: {e}"),
-            })?;
+        Self::load_from_data(disk, task_id, checkpoint_data).await
+    }
+
+    async fn load_from_data(disk: DiskStore, task_id: &str, checkpoint_data: Vec<u8>) -> Result<Self> {
+        validate_resume_task_id(task_id)?;
+        let mut checkpoint: ResumeCheckpoint = match serde_json::from_slice(&checkpoint_data) {
+            Ok(checkpoint) => checkpoint,
+            Err(error) => {
+                Self::block_invalid_snapshot(&disk, task_id).await;
+                return Err(Error::TaskExecutionFailed {
+                    message: format!("Failed to deserialize checkpoint: {error}"),
+                });
+            }
+        };
         if checkpoint.task_id != task_id {
+            Self::block_invalid_snapshot(&disk, task_id).await;
             return Err(Error::TaskExecutionFailed {
                 message: "Resume checkpoint task id does not match filename".to_string(),
             });
@@ -163,6 +280,7 @@ impl CheckpointManager {
         // identities. Discard the stale sets and position, then stamp the
         // current schema so the scan restarts cleanly.
         if checkpoint.schema_version > CURRENT_CHECKPOINT_SCHEMA {
+            Self::block_invalid_snapshot(&disk, task_id).await;
             return Err(Error::TaskExecutionFailed {
                 message: format!(
                     "Checkpoint schema {} is newer than supported schema {CURRENT_CHECKPOINT_SCHEMA}",
@@ -170,7 +288,45 @@ impl CheckpointManager {
                 ),
             });
         }
-        if checkpoint.schema_version < CURRENT_CHECKPOINT_SCHEMA {
+
+        let integrity_verified = if let Some(expected) = checkpoint.integrity_digest.as_deref() {
+            let actual = Self::checkpoint_digest(&Self::serialize_without_digest(&checkpoint)?);
+            if expected != actual {
+                Self::block_invalid_snapshot(&disk, task_id).await;
+                return Err(Error::InvalidCheckpoint(format!(
+                    "Resume checkpoint digest does not match task {task_id}"
+                )));
+            }
+            true
+        } else if checkpoint.schema_version >= CURRENT_CHECKPOINT_SCHEMA {
+            Self::block_invalid_snapshot(&disk, task_id).await;
+            return Err(Error::InvalidCheckpoint(format!(
+                "Resume checkpoint digest is missing for task {task_id}"
+            )));
+        } else {
+            let digest_path = Self::digest_path(task_id);
+            let digest_path = path_to_str(&digest_path)?;
+            match HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, digest_path).await {
+                Ok(expected) => {
+                    let actual = Self::checkpoint_digest(&checkpoint_data);
+                    if expected.as_ref() != actual.as_bytes() {
+                        Self::block_invalid_snapshot(&disk, task_id).await;
+                        return Err(Error::InvalidCheckpoint(format!(
+                            "Resume checkpoint digest does not match task {task_id}"
+                        )));
+                    }
+                    true
+                }
+                Err(crate::heal::DiskError::FileNotFound) => false,
+                Err(error) => {
+                    return Err(Error::TaskExecutionFailed {
+                        message: format!("Failed to read checkpoint digest: {error}"),
+                    });
+                }
+            }
+        };
+
+        if checkpoint.schema_version < CHECKPOINT_PER_VERSION_SCHEMA || !integrity_verified {
             warn!(
                 target: "rustfs::heal::resume",
                 event = EVENT_HEAL_CHECKPOINT_STATE,
@@ -187,13 +343,15 @@ impl CheckpointManager {
             checkpoint.skipped_objects.clear();
             checkpoint.current_bucket_index = 0;
             checkpoint.current_object_index = 0;
-            checkpoint.schema_version = CURRENT_CHECKPOINT_SCHEMA;
         }
+        checkpoint.schema_version = CURRENT_CHECKPOINT_SCHEMA;
 
         Ok(Self {
             disk,
             checkpoint: Arc::new(RwLock::new(checkpoint)),
             throttle: Mutex::new(PersistThrottle::new()),
+            save_lock: AsyncMutex::new(()),
+            last_saved: Mutex::new(Some(EcstoreDiskBytes::from(checkpoint_data))),
         })
     }
 
@@ -204,7 +362,7 @@ impl CheckpointManager {
         }
         let file_path = Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"));
         match path_to_str(&file_path) {
-            Ok(path_str) => match disk.read_all(RUSTFS_META_BUCKET, path_str).await {
+            Ok(path_str) => match HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, path_str).await {
                 Ok(data) => !data.is_empty(),
                 Err(_) => false,
             },
@@ -292,6 +450,8 @@ impl CheckpointManager {
 
         let checkpoint_file = Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"));
         delete_resume_file(&self.disk, &checkpoint_file).await?;
+        delete_resume_file(&self.disk, &Self::digest_path(&task_id)).await?;
+        delete_resume_file(&self.disk, &Self::blocked_path(&task_id)).await?;
 
         debug!(
             target: "rustfs::heal::resume",
@@ -307,21 +467,130 @@ impl CheckpointManager {
 
     /// save checkpoint to disk
     async fn save_checkpoint(&self) -> Result<()> {
-        let checkpoint = self.checkpoint.read().await;
+        // Serialize saves and take the snapshot only after acquiring the lock:
+        // a slower writer must not publish a snapshot taken before a newer one.
+        let _save_guard = self.save_lock.lock().await;
+        let checkpoint = self.checkpoint.read().await.clone();
         validate_resume_task_id(&checkpoint.task_id)?;
-        let checkpoint_data = serde_json::to_vec(&*checkpoint).map_err(|e| Error::TaskExecutionFailed {
-            message: format!("Failed to serialize checkpoint: {e}"),
-        })?;
+        let unsigned_checkpoint_data = Self::serialize_without_digest(&checkpoint)?;
+        let digest = Self::checkpoint_digest(&unsigned_checkpoint_data);
+        let mut persisted_checkpoint = checkpoint.clone();
+        persisted_checkpoint.integrity_digest = Some(digest);
+        let checkpoint_data =
+            EcstoreDiskBytes::from(serde_json::to_vec(&persisted_checkpoint).map_err(|e| Error::TaskExecutionFailed {
+                message: format!("Failed to serialize checkpoint: {e}"),
+            })?);
 
         let file_path = Path::new(BUCKET_META_PREFIX).join(format!("{}_{}", checkpoint.task_id, RESUME_CHECKPOINT_FILE));
 
         let path_str = path_to_str(&file_path)?;
-        self.disk
-            .write_all(RUSTFS_META_BUCKET, path_str, checkpoint_data.into())
+        let last_saved = self
+            .last_saved
+            .lock()
+            .map_err(|_| Error::TaskExecutionFailed {
+                message: "Checkpoint save state lock is poisoned; refusing to save".to_string(),
+            })?
+            .clone();
+        let update = EcstoreDiskAPI::compare_and_update_file(
+            self.disk.as_ref(),
+            RUSTFS_META_BUCKET,
+            path_str,
+            last_saved.clone(),
+            Some(checkpoint_data.clone()),
+        )
+        .await
+        .map_err(|e| Error::TaskExecutionFailed {
+            message: format!("Failed to save checkpoint: {e}"),
+        })?;
+
+        let expected = match update {
+            EcstoreConditionalFileUpdate::Updated => None,
+            EcstoreConditionalFileUpdate::Missing => {
+                return Err(Error::TaskExecutionFailed {
+                    message: "Checkpoint was removed after this manager saved it; refusing to recreate it".to_string(),
+                });
+            }
+            EcstoreConditionalFileUpdate::Mismatch => {
+                // A healthy manager normally completes the CAS above without
+                // another read or JSON parse. Inspect only after a mismatch so
+                // corruption and future schemas cannot be overwritten blindly.
+                let existing = match HealDiskExt::read_all(self.disk.as_ref(), RUSTFS_META_BUCKET, path_str).await {
+                    Ok(existing) => existing,
+                    Err(crate::heal::DiskError::FileNotFound) => {
+                        return Err(Error::TaskExecutionFailed {
+                            message: "Checkpoint was removed after this manager saved it; refusing to recreate it".to_string(),
+                        });
+                    }
+                    Err(error) => {
+                        return Err(Error::TaskExecutionFailed {
+                            message: format!("Failed to inspect checkpoint after CAS mismatch: {error}"),
+                        });
+                    }
+                };
+
+                if existing.is_empty() && last_saved.is_none() {
+                    Some(existing)
+                } else {
+                    let current: ResumeCheckpoint = match serde_json::from_slice(&existing) {
+                        Ok(current) => current,
+                        Err(error) => {
+                            Self::block_invalid_snapshot(&self.disk, &checkpoint.task_id).await;
+                            return Err(Error::TaskExecutionFailed {
+                                message: format!("Existing checkpoint is corrupt: {error}"),
+                            });
+                        }
+                    };
+                    if current.task_id != checkpoint.task_id {
+                        Self::block_invalid_snapshot(&self.disk, &checkpoint.task_id).await;
+                        return Err(Error::TaskExecutionFailed {
+                            message: "Existing checkpoint task id does not match filename".to_string(),
+                        });
+                    }
+                    if current.schema_version > CURRENT_CHECKPOINT_SCHEMA {
+                        Self::block_invalid_snapshot(&self.disk, &checkpoint.task_id).await;
+                        return Err(Error::TaskExecutionFailed {
+                            message: format!(
+                                "Existing checkpoint schema {} is newer than supported schema {CURRENT_CHECKPOINT_SCHEMA}",
+                                current.schema_version
+                            ),
+                        });
+                    }
+                    if last_saved.as_ref().is_none_or(|saved| saved.as_ref() != existing.as_ref()) {
+                        return Err(Error::TaskExecutionFailed {
+                            message: "Checkpoint changed since this manager loaded it; refusing to overwrite newer progress"
+                                .to_string(),
+                        });
+                    }
+                    Some(existing)
+                }
+            }
+        };
+
+        if let Some(expected) = expected {
+            match EcstoreDiskAPI::compare_and_update_file(
+                self.disk.as_ref(),
+                RUSTFS_META_BUCKET,
+                path_str,
+                Some(expected),
+                Some(checkpoint_data.clone()),
+            )
             .await
             .map_err(|e| Error::TaskExecutionFailed {
-                message: format!("Failed to save checkpoint: {e}"),
-            })?;
+                message: format!("Failed to save checkpoint after CAS mismatch: {e}"),
+            })? {
+                EcstoreConditionalFileUpdate::Updated => {}
+                EcstoreConditionalFileUpdate::Missing | EcstoreConditionalFileUpdate::Mismatch => {
+                    return Err(Error::TaskExecutionFailed {
+                        message: "Checkpoint changed while saving; refusing to overwrite newer progress".to_string(),
+                    });
+                }
+            }
+        }
+
+        let mut last_saved = self.last_saved.lock().map_err(|_| Error::TaskExecutionFailed {
+            message: "Checkpoint save state lock is poisoned after save".to_string(),
+        })?;
+        *last_saved = Some(checkpoint_data);
 
         debug!(
             target: "rustfs::heal::resume",
@@ -341,11 +610,38 @@ impl CheckpointManager {
         let file_path = Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"));
 
         let path_str = path_to_str(&file_path)?;
-        disk.read_all(RUSTFS_META_BUCKET, path_str)
+        HealDiskExt::read_all(disk.as_ref(), RUSTFS_META_BUCKET, path_str)
             .await
             .map(|bytes| bytes.to_vec())
             .map_err(|e| Error::TaskExecutionFailed {
                 message: format!("Failed to read checkpoint file: {e}"),
             })
     }
+
+    fn serialize_without_digest(checkpoint: &ResumeCheckpoint) -> Result<Vec<u8>> {
+        let mut unsigned = checkpoint.clone();
+        unsigned.integrity_digest = None;
+        let mut value = serde_json::to_value(&unsigned).map_err(|e| Error::TaskExecutionFailed {
+            message: format!("Failed to serialize checkpoint: {e}"),
+        })?;
+        for field in ["processed_objects", "failed_objects", "skipped_objects"] {
+            let Some(values) = value.get_mut(field).and_then(serde_json::Value::as_array_mut) else {
+                return Err(Error::TaskExecutionFailed {
+                    message: format!("Failed to canonicalize checkpoint field: {field}"),
+                });
+            };
+            values.sort_by(|left, right| left.as_str().cmp(&right.as_str()));
+        }
+        serde_json::to_vec(&value).map_err(|e| Error::TaskExecutionFailed {
+            message: format!("Failed to serialize checkpoint: {e}"),
+        })
+    }
+
+    fn checkpoint_digest(checkpoint_data: &[u8]) -> String {
+        base64::engine::general_purpose::STANDARD.encode(Sha256::digest(checkpoint_data))
+    }
+
+    fn digest_path(task_id: &str) -> std::path::PathBuf {
+        Path::new(BUCKET_META_PREFIX).join(format!("{task_id}_{RESUME_CHECKPOINT_DIGEST_FILE}"))
+    }
 }

crates/heal/src/heal/resume/tests.rs

@@ -1600,6 +1600,32 @@ async fn test_checkpoint_schema_v4_discarded_on_load() {
     temp_dir.close().expect("remove schema test directory");
 }
 
+#[tokio::test]
+async fn downgraded_unsigned_checkpoint_resets_untrusted_progress() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone()).await.unwrap();
+    manager.add_processed_object("victim-a".to_string()).await.unwrap();
+    manager.update_position(2, 500).await.unwrap();
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let bytes = disk.read_all(RUSTFS_META_BUCKET, &checkpoint_path).await.unwrap();
+    let mut downgraded: serde_json::Value = serde_json::from_slice(&bytes).unwrap();
+    downgraded["schema_version"] = serde_json::json!(CURRENT_CHECKPOINT_SCHEMA - 1);
+    downgraded.as_object_mut().unwrap().remove("integrity_digest");
+    downgraded["processed_objects"] = serde_json::json!(["victim-b"]);
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, serde_json::to_vec(&downgraded).unwrap().into())
+        .await
+        .expect("write downgraded checkpoint");
+
+    let manager = CheckpointManager::load_from_disk(disk, &task_id).await.unwrap();
+    let checkpoint = manager.get_checkpoint().await;
+    assert_eq!(checkpoint.schema_version, CURRENT_CHECKPOINT_SCHEMA);
+    assert_eq!(checkpoint.current_bucket_index, 0);
+    assert_eq!(checkpoint.current_object_index, 0);
+    assert!(checkpoint.processed_objects.is_empty());
+    temp_dir.close().unwrap();
+}
+
 #[tokio::test]
 async fn current_normal_resume_schema_preserves_progress() {
     let (temp_dir, disk) = schema_test_disk().await;
@@ -1675,6 +1701,369 @@ async fn future_resume_and_checkpoint_schemas_are_rejected() {
     temp_dir.close().expect("remove schema test directory");
 }
 
+#[tokio::test]
+async fn checkpoint_save_does_not_replace_a_non_empty_truncated_snapshot() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let truncated = b"{\"schema_version\":5,\"task_id\":";
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, truncated.as_slice().into())
+        .await
+        .expect("write truncated checkpoint fixture");
+
+    let error = manager
+        .update_position(2, 7)
+        .await
+        .expect_err("a truncated checkpoint must fail closed during save");
+    assert!(error.to_string().contains("Existing checkpoint is corrupt"));
+    assert_eq!(
+        disk.read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+            .await
+            .expect("read truncated checkpoint fixture"),
+        truncated.as_slice()
+    );
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().expect("remove checkpoint save test directory");
+}
+
+#[tokio::test]
+async fn checkpoint_save_does_not_replace_a_future_schema_snapshot() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let mut future = ResumeCheckpoint::new(task_id.clone());
+    future.schema_version = CURRENT_CHECKPOINT_SCHEMA + 1;
+    let future_bytes = serde_json::to_vec(&future).expect("serialize future checkpoint fixture");
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, future_bytes.clone().into())
+        .await
+        .expect("write future checkpoint fixture");
+
+    let error = manager
+        .update_position(2, 7)
+        .await
+        .expect_err("a future schema must fail closed during save");
+    assert!(error.to_string().contains("Existing checkpoint schema"));
+    assert_eq!(
+        disk.read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+            .await
+            .expect("read future checkpoint fixture"),
+        future_bytes
+    );
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().expect("remove future schema test directory");
+}
+
+#[tokio::test]
+async fn checkpoint_digest_rejects_same_length_progress_tampering() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    manager
+        .add_processed_object("victim-a".to_string())
+        .await
+        .expect("persist checkpoint progress");
+    manager.update_position(1, 1).await.expect("flush checkpoint progress");
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let original = disk
+        .read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+        .await
+        .expect("read checkpoint fixture");
+    let tampered = original
+        .windows(b"victim-a".len())
+        .position(|window| window == b"victim-a")
+        .map(|index| {
+            let mut bytes = original.to_vec();
+            bytes[index..index + b"victim-a".len()].copy_from_slice(b"victim-b");
+            bytes
+        })
+        .expect("checkpoint should contain the processed object");
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, tampered.into())
+        .await
+        .expect("write tampered checkpoint fixture");
+
+    assert!(CheckpointManager::load_from_disk(disk.clone(), &task_id).await.is_err());
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().expect("remove digest test directory");
+}
+
+#[tokio::test]
+async fn checkpoint_integrity_survives_missing_legacy_sidecar() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone()).await.unwrap();
+    manager.update_position(2, 9).await.unwrap();
+
+    let digest_path = format!("{BUCKET_META_PREFIX}/{task_id}_ahm_checkpoint.sha256");
+    delete_resume_file(&disk, Path::new(&digest_path)).await.unwrap();
+
+    let restored = CheckpointManager::load_from_disk(disk, &task_id).await.unwrap();
+    let checkpoint = restored.get_checkpoint().await;
+    assert_eq!(checkpoint.current_bucket_index, 2);
+    assert_eq!(checkpoint.current_object_index, 9);
+    assert!(checkpoint.integrity_digest.is_some());
+    temp_dir.close().unwrap();
+}
+
+#[tokio::test]
+async fn checkpoint_integrity_survives_multi_object_reload() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone()).await.unwrap();
+    for index in 0..32 {
+        manager.add_processed_object(format!("processed-{index}")).await.unwrap();
+        manager.add_failed_object(format!("failed-{index}")).await.unwrap();
+        manager.add_skipped_object(format!("skipped-{index}")).await.unwrap();
+    }
+    manager.update_position(2, 9).await.unwrap();
+
+    CheckpointManager::load_from_disk(disk, &task_id)
+        .await
+        .expect("a healthy multi-object checkpoint must survive reload");
+    temp_dir.close().unwrap();
+}
+
+#[tokio::test]
+async fn checkpoint_integrity_rejects_a_removed_embedded_digest() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone()).await.unwrap();
+    manager.update_position(2, 9).await.unwrap();
+
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    let bytes = disk
+        .read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+        .await
+        .expect("read checkpoint fixture");
+    let mut value: serde_json::Value = serde_json::from_slice(&bytes).unwrap();
+    value["current_object_index"] = serde_json::json!(10);
+    value.as_object_mut().unwrap().remove("integrity_digest");
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, serde_json::to_vec(&value).unwrap().into())
+        .await
+        .expect("write tampered checkpoint fixture");
+
+    assert!(
+        CheckpointManager::load_from_disk(disk.clone(), &task_id).await.is_err(),
+        "a current checkpoint without its embedded digest must fail closed"
+    );
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().unwrap();
+}
+
+#[tokio::test]
+async fn new_checkpoint_manager_rebuilds_an_empty_snapshot() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+    disk.write_all(RUSTFS_META_BUCKET, &checkpoint_path, EcstoreDiskBytes::new())
+        .await
+        .expect("write empty checkpoint fixture");
+
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("a new manager must rebuild an empty checkpoint");
+    manager
+        .update_position(3, 11)
+        .await
+        .expect("rebuilt checkpoint must remain writable");
+    assert!(CheckpointManager::has_checkpoint(&disk, &task_id).await);
+    temp_dir.close().expect("remove empty checkpoint test directory");
+}
+
+#[tokio::test]
+async fn deleted_checkpoint_is_not_recreated_by_an_old_manager() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    manager.cleanup().await.expect("delete checkpoint fixture");
+
+    let error = manager
+        .update_position(1, 2)
+        .await
+        .expect_err("an old manager must not resurrect a deleted checkpoint");
+    assert!(error.to_string().contains("removed after this manager saved it"));
+    assert!(!CheckpointManager::has_checkpoint(&disk, &task_id).await);
+    temp_dir.close().expect("remove deleted checkpoint test directory");
+}
+
+#[cfg(unix)]
+#[tokio::test]
+async fn checkpoint_cleanup_leaves_no_task_specific_lock_artifact() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    let lock_path = Path::new(BUCKET_META_PREFIX)
+        .join(format!("{task_id}_{RESUME_CHECKPOINT_FILE}"))
+        .with_extension("rustfs-cas.lock");
+    let lock_path = temp_dir.path().join(RUSTFS_META_BUCKET).join(lock_path);
+
+    manager.cleanup().await.expect("delete checkpoint fixture");
+
+    assert!(
+        !lock_path.exists(),
+        "successful checkpoint cleanup must not leave a task-specific lock artifact"
+    );
+}
+
+#[tokio::test]
+async fn an_empty_blocked_marker_still_blocks_resume_selection() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let manager = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create checkpoint manager");
+    let blocked_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_BLOCKED_FILE}");
+    disk.write_all(RUSTFS_META_BUCKET, &blocked_path, EcstoreDiskBytes::new())
+        .await
+        .expect("write empty blocked marker fixture");
+
+    assert!(CheckpointManager::is_blocked(&disk, &task_id).await);
+    assert!(CheckpointManager::is_resumable(&disk, &task_id).await.is_err());
+    // Recovery requires replacing/cleaning the snapshot, then removing the
+    // marker; ordinary selector retries are intentionally not an unlock path.
+    manager.cleanup().await.expect("clean blocked checkpoint");
+    assert!(!CheckpointManager::is_blocked(&disk, &task_id).await);
+    temp_dir.close().expect("remove empty blocked marker test directory");
+}
+
+#[tokio::test]
+async fn resumable_selector_skips_healthy_tasks_with_blocked_markers() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let tasks = [
+        (ResumeUtils::generate_task_id(), EcstoreDiskBytes::new()),
+        (ResumeUtils::generate_task_id(), EcstoreDiskBytes::from_static(b"blocked")),
+    ];
+    for (task_id, marker) in &tasks {
+        ResumeManager::new(
+            disk.clone(),
+            task_id.clone(),
+            "erasure_set".to_string(),
+            "pool_0_set_0".to_string(),
+            vec!["bucket".to_string()],
+        )
+        .await
+        .expect("create healthy resume state");
+        CheckpointManager::new(disk.clone(), task_id.clone())
+            .await
+            .expect("create healthy checkpoint");
+        let checkpoint_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_FILE}");
+        let checkpoint_bytes = disk
+            .read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+            .await
+            .expect("read healthy checkpoint before blocking");
+        let marker_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_BLOCKED_FILE}");
+        disk.write_all(RUSTFS_META_BUCKET, &marker_path, marker.clone())
+            .await
+            .expect("write blocked marker");
+
+        assert!(ResumeUtils::get_resumable_tasks(&disk).await.is_err());
+        assert_eq!(
+            disk.read_all(RUSTFS_META_BUCKET, &checkpoint_path)
+                .await
+                .expect("read healthy checkpoint after blocking"),
+            checkpoint_bytes
+        );
+    }
+    temp_dir.close().expect("remove blocked selector test directory");
+}
+
+#[tokio::test]
+async fn stale_checkpoint_manager_cannot_overwrite_newer_progress() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let task_id = ResumeUtils::generate_task_id();
+    let first = CheckpointManager::new(disk.clone(), task_id.clone())
+        .await
+        .expect("create first checkpoint manager");
+    let second = CheckpointManager::load_from_disk(disk.clone(), &task_id)
+        .await
+        .expect("load second checkpoint manager");
+
+    second
+        .update_position(4, 20)
+        .await
+        .expect("persist newer checkpoint progress");
+    let error = first
+        .update_position(1, 3)
+        .await
+        .expect_err("stale checkpoint manager must not overwrite newer progress");
+    assert!(error.to_string().contains("newer progress"));
+
+    let persisted = CheckpointManager::load_from_disk(disk.clone(), &task_id)
+        .await
+        .expect("load newer checkpoint progress")
+        .get_checkpoint()
+        .await;
+    assert_eq!(persisted.current_bucket_index, 4);
+    assert_eq!(persisted.current_object_index, 20);
+    temp_dir.close().expect("remove stale manager test directory");
+}
+
+#[tokio::test]
+async fn resumable_selector_isolates_future_and_corrupt_checkpoints() {
+    let (temp_dir, disk) = schema_test_disk().await;
+    let future_task = ResumeUtils::generate_task_id();
+    let corrupt_task = ResumeUtils::generate_task_id();
+    for task_id in [&future_task, &corrupt_task] {
+        ResumeManager::new(
+            disk.clone(),
+            task_id.to_string(),
+            "erasure_set".to_string(),
+            "pool_0_set_0".to_string(),
+            vec!["bucket".to_string()],
+        )
+        .await
+        .expect("create resumable state fixture");
+    }
+
+    let future_path = format!("{BUCKET_META_PREFIX}/{future_task}_{RESUME_CHECKPOINT_FILE}");
+    let mut future = ResumeCheckpoint::new(future_task.clone());
+    future.schema_version = CURRENT_CHECKPOINT_SCHEMA + 1;
+    let future_bytes = serde_json::to_vec(&future).expect("serialize future checkpoint fixture");
+    disk.write_all(RUSTFS_META_BUCKET, &future_path, future_bytes.clone().into())
+        .await
+        .expect("write future checkpoint fixture");
+    let corrupt_path = format!("{BUCKET_META_PREFIX}/{corrupt_task}_{RESUME_CHECKPOINT_FILE}");
+    let corrupt_bytes = b"{truncated";
+    disk.write_all(RUSTFS_META_BUCKET, &corrupt_path, corrupt_bytes.as_slice().into())
+        .await
+        .expect("write corrupt checkpoint fixture");
+
+    assert!(CheckpointManager::is_resumable(&disk, &future_task).await.is_err());
+    assert!(CheckpointManager::is_resumable(&disk, &corrupt_task).await.is_err());
+    assert!(ResumeUtils::get_resumable_tasks(&disk).await.is_err());
+    for (task_id, path, bytes) in [
+        (&future_task, future_path, future_bytes),
+        (&corrupt_task, corrupt_path, corrupt_bytes.to_vec()),
+    ] {
+        assert_eq!(
+            disk.read_all(RUSTFS_META_BUCKET, &path)
+                .await
+                .expect("read isolated checkpoint bytes"),
+            bytes
+        );
+        let blocked_path = format!("{BUCKET_META_PREFIX}/{task_id}_{RESUME_CHECKPOINT_BLOCKED_FILE}");
+        assert!(
+            !disk
+                .read_all(RUSTFS_META_BUCKET, &blocked_path)
+                .await
+                .expect("read checkpoint blocked marker")
+                .is_empty()
+        );
+    }
+    temp_dir.close().expect("remove selector isolation test directory");
+}
+
 #[test]
 fn test_persist_throttle_batches_until_threshold() {
     let mut throttle = PersistThrottle::new();

crates/heal/src/heal/resume/utils.rs

@@ -21,7 +21,7 @@ use uuid::Uuid;
 use super::super::{BUCKET_META_PREFIX, DiskError, DiskStore, HealDiskExt as _, RUSTFS_META_BUCKET};
 use super::replacement::{ReplacementPhase, ReplacementRecoveryRecord};
 use super::{
-    EVENT_HEAL_RESUME_STATE, LOG_COMPONENT_HEAL, LOG_SUBSYSTEM_RESUME, REPLACEMENT_COMPLETION_PROOF_FILE,
+    CheckpointManager, EVENT_HEAL_RESUME_STATE, LOG_COMPONENT_HEAL, LOG_SUBSYSTEM_RESUME, REPLACEMENT_COMPLETION_PROOF_FILE,
     REPLACEMENT_INTENT_FILE, RESUME_STATE_FILE, ResumeManager, ResumeStateFile, is_replacement_intent, path_to_str,
     replacement_recovery_corruption_for_state_load, replacement_recovery_dir, validate_resume_task_id,
 };
@@ -67,6 +67,7 @@ impl ResumeUtils {
                 // Extract task ID from filename: {task_id}_ahm_resume_state.json
                 if let Some(task_id) = entry.strip_suffix(&format!("_{RESUME_STATE_FILE}"))
                     && validate_resume_task_id(task_id).is_ok()
+                    && CheckpointManager::is_resumable(disk, task_id).await?
                 {
                     task_ids.push(task_id.to_string());
                 }

crates/kms/tests/vault_ha_failover_live.rs

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

rustfs/src/admin/handlers/rebalance.rs

@@ -177,15 +177,12 @@ async fn rollback_cluster_rebalance_start(
                 terminal_reload_attempt_at: Some(terminal_reload_attempt_at),
                 terminal_reload_failures: terminal_reload_failures.clone(),
             };
-            store
-                .record_rebalance_stop_propagation(rebalance_id, record)
-                .await
-                .map_err(|err| {
-                    format!(
-                        "cluster rebalance rollback for {rebalance_id} partial; failed to persist stop propagation: {err}; {}",
-                        rebalance_rollback_failure_message(rebalance_id, &stop_failures, &terminal_reload_failures)
-                    )
-                })?;
+            store.record_rebalance_stop_propagation(record).await.map_err(|err| {
+                format!(
+                    "cluster rebalance rollback for {rebalance_id} partial; failed to persist stop propagation: {err}; {}",
+                    rebalance_rollback_failure_message(rebalance_id, &stop_failures, &terminal_reload_failures)
+                )
+            })?;
             return Err(rebalance_rollback_failure_message(
                 rebalance_id,
                 &stop_failures,
@@ -200,7 +197,7 @@ async fn rollback_cluster_rebalance_start(
         .await
         .map_err(|err| format!("local stop_rebalance rollback for {rebalance_id} failed: {err}"))?;
     store
-        .save_rebalance_stats_for_id(usize::MAX, RebalSaveOpt::StoppedAt, rebalance_id)
+        .save_rebalance_stats(usize::MAX, RebalSaveOpt::StoppedAt)
         .await
         .map_err(|err| format!("local rollback stop metadata save for {rebalance_id} failed: {err}"))?;
     Ok(())
@@ -682,7 +679,7 @@ impl Operation for RebalanceStart {
                                     terminal_reload_attempt_at: Some(terminal_reload_attempt_at),
                                     terminal_reload_failures: terminal_reload_failures.clone(),
                                 };
-                                store.record_rebalance_stop_propagation(&id, record).await.map_err(|err| {
+                                store.record_rebalance_stop_propagation(record).await.map_err(|err| {
                                     rebalance_internal_error(format!(
                                         "failed to persist rebalance local-start rollback propagation metadata: {err}"
                                     ))
@@ -872,38 +869,6 @@ impl Operation for RebalanceStatus {
     }
 }
 
-async fn rebalance_stop_target_id(store: &Arc<ECStore>) -> S3Result<Option<String>> {
-    store
-        .prepare_rebalance_stop()
-        .await
-        .map_err(|e| s3_error!(InternalError, "failed to prepare rebalance metadata for stop: {}", e))
-}
-
-async fn stop_rebalance_admission_first(
-    store: &Arc<ECStore>,
-    notification_sys: Option<&NotificationSys>,
-    expected_rebalance_id: &str,
-) -> S3Result<Vec<String>> {
-    // prepare_rebalance_stop already closed admission for this exact run.
-
-    if let Some(notification_sys) = notification_sys {
-        return notification_sys
-            .stop_rebalance_failures(Some(expected_rebalance_id))
-            .await
-            .map_err(|e| s3_error!(InternalError, "failed to stop rebalance via notification system: {}", e));
-    }
-
-    store
-        .stop_rebalance_for_id(Some(expected_rebalance_id))
-        .await
-        .map_err(|e| s3_error!(InternalError, "failed to stop rebalance: {}", e))?;
-    store
-        .save_rebalance_stats_for_id(usize::MAX, RebalSaveOpt::StoppedAt, expected_rebalance_id)
-        .await
-        .map_err(|e| s3_error!(InternalError, "failed to persist rebalance stop metadata: {}", e))?;
-    Ok(Vec::new())
-}
-
 // RebalanceStop
 pub struct RebalanceStop {}
 
@@ -951,15 +916,36 @@ impl Operation for RebalanceStop {
             return Err(s3_error!(InternalError, "object layer is not initialized"));
         };
 
-        let Some(expected_rebalance_id) = rebalance_stop_target_id(&store).await? else {
+        store
+            .load_rebalance_meta()
+            .await
+            .map_err(|e| s3_error!(InternalError, "failed to load rebalance metadata before stop: {}", e))?;
+        let expected_rebalance_id = store.current_rebalance_id().await;
+
+        if !store.is_rebalance_conflicting_with_decommission().await {
             log_rebalance_request_rejected("stop", "rebalance_not_started", &request_id, &actor, &remote_addr);
             return Err(s3_error!(NoSuchResource, "pool rebalance is not started"));
-        };
+        }
 
         let notification_sys = current_notification_system();
         let stop_attempt_at = OffsetDateTime::now_utc();
-        let stop_failures =
-            stop_rebalance_admission_first(&store, notification_sys.as_deref(), expected_rebalance_id.as_str()).await?;
+        let mut stop_failures = Vec::new();
+        if let Some(notification_sys) = notification_sys.as_ref() {
+            stop_failures = notification_sys
+                .stop_rebalance_failures(expected_rebalance_id.as_deref())
+                .await
+                .map_err(|e| s3_error!(InternalError, "failed to stop rebalance via notification system: {}", e))?;
+        } else {
+            store
+                .stop_rebalance_for_id(expected_rebalance_id.as_deref())
+                .await
+                .map_err(|e| s3_error!(InternalError, "failed to stop rebalance: {}", e))?;
+
+            store
+                .save_rebalance_stats(usize::MAX, RebalSaveOpt::StoppedAt)
+                .await
+                .map_err(|e| s3_error!(InternalError, "failed to persist rebalance stop metadata: {}", e))?;
+        }
 
         info!(
             event = EVENT_ADMIN_REBALANCE_STATE,
@@ -1021,7 +1007,7 @@ impl Operation for RebalanceStop {
                 terminal_reload_failures: terminal_reload_failures.clone(),
             };
             store
-                .record_rebalance_stop_propagation(expected_rebalance_id.as_str(), record)
+                .record_rebalance_stop_propagation(record)
                 .await
                 .map_err(|e| s3_error!(InternalError, "failed to persist rebalance stop propagation metadata: {}", e))?;
 
@@ -1095,218 +1081,15 @@ mod rebalance_handler_tests {
         RebalPoolProgress, RebalanceAdminStatus, RebalancePoolStatus, RebalanceStartStep, RebalanceStopPropagationStatus,
         build_rebalance_admin_status, build_rebalance_pool_statuses, build_rebalance_stop_propagation_status,
         rebalance_pool_used, rebalance_query_present, rebalance_remaining_buckets, rebalance_rollback_failure_message,
-        rebalance_rollback_stop_failure_message, rebalance_start_rollback_error, rebalance_start_steps, rebalance_stop_target_id,
-        rebalance_used_pct, rollback_result_label, stop_rebalance_admission_first,
+        rebalance_rollback_stop_failure_message, rebalance_start_rollback_error, rebalance_start_steps, rebalance_used_pct,
+        rollback_result_label,
     };
     use crate::admin::storage_api::rebalance::{
-        DiskStat, RebalSaveOpt, RebalStatus, RebalanceCleanupWarningEntry, RebalanceCleanupWarnings, RebalanceInfo,
-        RebalanceMeta, RebalanceStats, RebalanceStopPropagationRecord, encode_rebalance_stop_propagation_record,
+        DiskStat, RebalStatus, RebalanceCleanupWarningEntry, RebalanceCleanupWarnings, RebalanceInfo, RebalanceMeta,
+        RebalanceStats, RebalanceStopPropagationRecord, encode_rebalance_stop_propagation_record,
     };
     use time::OffsetDateTime;
 
-    fn started_rebalance_meta(id: &str) -> RebalanceMeta {
-        RebalanceMeta {
-            id: id.to_string(),
-            pool_stats: vec![RebalanceStats {
-                participating: true,
-                info: RebalanceInfo {
-                    start_time: Some(OffsetDateTime::now_utc()),
-                    status: RebalStatus::Started,
-                    ..Default::default()
-                },
-                ..Default::default()
-            }],
-            ..Default::default()
-        }
-    }
-
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn real_admin_stop_cancels_paused_entry_before_waiting_for_activation_gate() {
-        const REBALANCE_ID: &str = "admin-stop-paused-entry";
-        let mut fixture =
-            crate::admin::storage_api::ecstore_rebalance::test_util::PausedRebalanceEntryTestFixture::new(REBALANCE_ID).await;
-        fixture.wait_until_entry_paused().await;
-
-        let stop_store = fixture.store();
-        let mut stop_task = tokio::spawn(async move {
-            let expected_rebalance_id = rebalance_stop_target_id(&stop_store)
-                .await
-                .expect("admin stop target resolution should succeed")
-                .expect("the active rebalance should remain stoppable");
-            stop_rebalance_admission_first(&stop_store, None, expected_rebalance_id.as_str()).await
-        });
-        fixture.wait_until_admission_cancelled().await;
-        fixture.wait_until_stop_waiting_for_entry().await;
-        assert!(!stop_task.is_finished(), "admin stop must wait for the paused entry guard to drain");
-
-        fixture.release_entry();
-        fixture.assert_entry_cancelled().await;
-        let stop_failures = tokio::time::timeout(std::time::Duration::from_secs(5), &mut stop_task)
-            .await
-            .expect("admin stop should finish after the entry guard drains")
-            .expect("admin stop task should not panic")
-            .expect("admin stop should persist the terminal state");
-        assert!(stop_failures.is_empty());
-        assert!(!fixture.store().is_rebalance_conflicting_with_decommission().await);
-    }
-
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn real_admin_stop_accepts_same_run_terminalization_after_prepare() {
-        const REBALANCE_ID: &str = "admin-stop-terminal-after-prepare";
-        const REPLACEMENT_ID: &str = "admin-stop-replacement";
-        let (_temp_dirs, store) =
-            crate::admin::storage_api::ecstore_rebalance::test_util::test_store_with_persisted_rebalance_meta(
-                started_rebalance_meta(REBALANCE_ID),
-            )
-            .await;
-        let terminal_barrier = std::sync::Arc::new(tokio::sync::Barrier::new(2));
-        let worker_barrier = std::sync::Arc::clone(&terminal_barrier);
-        let worker_store = std::sync::Arc::clone(&store);
-        let terminal_task = tokio::spawn(async move {
-            worker_barrier.wait().await;
-            {
-                let mut rebalance_meta = worker_store.rebalance_meta.write().await;
-                let meta = rebalance_meta
-                    .as_mut()
-                    .expect("the prepared rebalance metadata should remain installed");
-                assert_eq!(meta.id, REBALANCE_ID);
-                let pool = meta
-                    .pool_stats
-                    .first_mut()
-                    .expect("the prepared rebalance should have a pool");
-                pool.info.status = RebalStatus::Stopped;
-                pool.info.end_time = Some(OffsetDateTime::now_utc());
-            }
-            worker_store
-                .save_rebalance_stats_for_id(0, RebalSaveOpt::Stats, REBALANCE_ID)
-                .await
-        });
-
-        let expected_rebalance_id = rebalance_stop_target_id(&store)
-            .await
-            .expect("admin stop target resolution should succeed")
-            .expect("the active rebalance should remain stoppable");
-        assert_eq!(expected_rebalance_id, REBALANCE_ID);
-        let cancel = store
-            .rebalance_meta
-            .read()
-            .await
-            .as_ref()
-            .and_then(|meta| meta.cancel.clone())
-            .expect("prepare should install the admission cancellation token");
-        assert!(cancel.is_cancelled());
-
-        terminal_barrier.wait().await;
-        tokio::time::timeout(std::time::Duration::from_secs(30), terminal_task)
-            .await
-            .expect("worker terminalization should finish after the barrier opens")
-            .expect("worker terminalization task should not panic")
-            .expect("worker terminalization should persist");
-        assert!(!store.is_rebalance_conflicting_with_decommission().await);
-
-        *store.rebalance_meta.write().await = None;
-        store
-            .load_rebalance_meta()
-            .await
-            .expect("the worker terminal state should reload before the final stop");
-        {
-            let terminal = store.rebalance_meta.read().await;
-            let terminal = terminal.as_ref().expect("the worker terminal state should remain persisted");
-            assert_eq!(terminal.id, REBALANCE_ID);
-            assert_eq!(terminal.pool_stats[0].info.status, RebalStatus::Stopped);
-        }
-
-        let stop_failures = stop_rebalance_admission_first(&store, None, expected_rebalance_id.as_str())
-            .await
-            .expect("same-run terminalization after prepare should be a successful stop");
-        assert!(stop_failures.is_empty());
-
-        *store.rebalance_meta.write().await = Some(started_rebalance_meta(REPLACEMENT_ID));
-        let error = stop_rebalance_admission_first(&store, None, expected_rebalance_id.as_str())
-            .await
-            .expect_err("the prepared stop must not mutate a replacement run");
-        assert!(error.to_string().contains(REBALANCE_ID));
-        let replacement = store.rebalance_meta.read().await;
-        let replacement = replacement.as_ref().expect("the replacement run should remain installed");
-        assert_eq!(replacement.id, REPLACEMENT_ID);
-        assert_eq!(replacement.pool_stats[0].info.status, RebalStatus::Started);
-        assert!(replacement.cancel.is_none());
-        assert!(replacement.stopped_at.is_none());
-    }
-
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn real_admin_stop_loads_persisted_active_rebalance_from_cold_memory() {
-        const REBALANCE_ID: &str = "admin-stop-cold-memory";
-        let (_temp_dirs, store) =
-            crate::admin::storage_api::ecstore_rebalance::test_util::test_store_with_persisted_rebalance_meta(
-                started_rebalance_meta(REBALANCE_ID),
-            )
-            .await;
-        *store.rebalance_meta.write().await = None;
-        assert!(store.current_rebalance_id().await.is_none());
-
-        let expected_rebalance_id = rebalance_stop_target_id(&store)
-            .await
-            .expect("admin stop should load persisted rebalance metadata")
-            .expect("persisted active rebalance should be stoppable");
-        assert_eq!(expected_rebalance_id, REBALANCE_ID);
-
-        let stop_failures = stop_rebalance_admission_first(&store, None, expected_rebalance_id.as_str())
-            .await
-            .expect("admin stop should persist the terminal state after a cold load");
-        assert!(stop_failures.is_empty());
-        assert!(!store.is_rebalance_conflicting_with_decommission().await);
-
-        *store.rebalance_meta.write().await = None;
-        store
-            .load_rebalance_meta()
-            .await
-            .expect("the persisted terminal rebalance metadata should remain readable");
-        assert_eq!(store.current_rebalance_id().await.as_deref(), Some(REBALANCE_ID));
-        assert!(!store.is_rebalance_conflicting_with_decommission().await);
-    }
-
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn real_admin_stop_refreshes_persisted_active_over_stale_inactive_memory() {
-        const PERSISTED_REBALANCE_ID: &str = "admin-stop-persisted-active";
-        const STALE_REBALANCE_ID: &str = "admin-stop-stale-terminal";
-        let (_temp_dirs, store) =
-            crate::admin::storage_api::ecstore_rebalance::test_util::test_store_with_persisted_rebalance_meta(
-                started_rebalance_meta(PERSISTED_REBALANCE_ID),
-            )
-            .await;
-        *store.rebalance_meta.write().await = Some(RebalanceMeta {
-            id: STALE_REBALANCE_ID.to_string(),
-            stopped_at: Some(OffsetDateTime::now_utc()),
-            ..Default::default()
-        });
-        assert_eq!(store.current_rebalance_id().await.as_deref(), Some(STALE_REBALANCE_ID));
-        assert!(!store.is_rebalance_conflicting_with_decommission().await);
-
-        let expected_rebalance_id = rebalance_stop_target_id(&store)
-            .await
-            .expect("admin stop should refresh stale inactive local metadata")
-            .expect("persisted active rebalance should replace the stale local terminal state");
-        assert_eq!(expected_rebalance_id, PERSISTED_REBALANCE_ID);
-
-        let stop_failures = stop_rebalance_admission_first(&store, None, expected_rebalance_id.as_str())
-            .await
-            .expect("admin stop should persist the refreshed run's terminal state");
-        assert!(stop_failures.is_empty());
-
-        *store.rebalance_meta.write().await = None;
-        store
-            .load_rebalance_meta()
-            .await
-            .expect("the refreshed run's persisted terminal metadata should remain readable");
-        assert_eq!(store.current_rebalance_id().await.as_deref(), Some(PERSISTED_REBALANCE_ID));
-        assert!(!store.is_rebalance_conflicting_with_decommission().await);
-    }
-
     #[test]
     fn test_calculate_rebalance_progress_running() {
         let start = OffsetDateTime::from_unix_timestamp(1_000).unwrap();

rustfs/src/admin/storage_api.rs

@@ -70,9 +70,7 @@ mod ecstore_notification {
 }
 
 #[allow(unused_imports)]
-pub(crate) mod ecstore_rebalance {
-    #[cfg(test)]
-    pub(crate) use crate::storage::storage_api::ecstore_rebalance::test_util;
+mod ecstore_rebalance {
     pub(crate) use crate::storage::storage_api::ecstore_rebalance::{
         DiskStat, RebalSaveOpt, RebalStatus, RebalanceCleanupWarningEntry, RebalanceCleanupWarnings, RebalanceInfo,
         RebalanceMeta, RebalanceStats, RebalanceStopPropagationRecord, decode_rebalance_stop_propagation_record,

rustfs/src/storage/storage_api.rs

@@ -501,8 +501,6 @@ pub(crate) mod ecstore_notification {
 
 #[allow(unused_imports)]
 pub(crate) mod ecstore_rebalance {
-    #[cfg(test)]
-    pub(crate) use rustfs_ecstore::api::rebalance::test_util;
     pub(crate) use rustfs_ecstore::api::rebalance::{
         DiskStat, RebalSaveOpt, RebalStatus, RebalanceCleanupWarningEntry, RebalanceCleanupWarnings, RebalanceInfo,
         RebalanceMeta, RebalanceStats, RebalanceStopPropagationRecord, decode_rebalance_stop_propagation_record,

scripts/test/vault_ha_kms_live.sh

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