fix(ci): accept equal below-resolution Warp latency

* fix(heal): fence format repair during pool transitions

* fix(heal): fence format writes during transitions

Cargo.lock

@@ -12688,6 +12688,7 @@ dependencies = [
  "js-sys",
  "rand 0.10.2",
  "serde_core",
+ "sha1_smol",
  "wasm-bindgen",
 ]
 

crates/ecstore/src/core/pools.rs

@@ -1455,36 +1455,6 @@ where
     Ok(())
 }
 
-async fn run_decommission_phases<F>(
-    rx: CancellationToken,
-    regular_buckets: Vec<DecomBucketInfo>,
-    meta_buckets: Vec<DecomBucketInfo>,
-    bucket_concurrency: usize,
-    mut start_bucket: F,
-) -> Result<()>
-where
-    F: FnMut(DecomBucketInfo, CancellationToken) -> BoxFuture<'static, Result<()>>,
-{
-    decommission_cancel_signal_result(rx.is_cancelled())?;
-
-    for bucket in meta_buckets {
-        decommission_cancel_signal_result(rx.is_cancelled())?;
-        start_bucket(bucket, rx.clone()).await?;
-    }
-
-    decommission_cancel_signal_result(rx.is_cancelled())?;
-
-    if bucket_concurrency <= 1 {
-        for bucket in regular_buckets {
-            decommission_cancel_signal_result(rx.is_cancelled())?;
-            start_bucket(bucket, rx.clone()).await?;
-        }
-        return Ok(());
-    }
-
-    run_decommission_buckets_bounded(rx, regular_buckets, bucket_concurrency, start_bucket).await
-}
-
 #[cfg(test)]
 async fn wait_decommission_worker_drain(workers: &Semaphore, limit: usize) -> Result<()> {
     let permits = u32::try_from(limit)
@@ -2056,7 +2026,7 @@ impl PoolMeta {
         self.load_no_lock(pool).await
     }
 
-    async fn load_no_lock<S>(&mut self, pool: Arc<S>) -> Result<()>
+    pub(crate) async fn load_no_lock<S>(&mut self, pool: Arc<S>) -> Result<()>
     where
         S: EcstoreObjectIO,
     {
@@ -4932,6 +4902,25 @@ impl ECStore {
         Ok(())
     }
 
+    async fn decommission_buckets_concurrently(
+        self: &Arc<Self>,
+        rx: CancellationToken,
+        idx: usize,
+        pool: Arc<Sets>,
+        buckets: Vec<DecomBucketInfo>,
+        limit: usize,
+        entry_budget: Arc<Semaphore>,
+    ) -> Result<()> {
+        let store = Arc::clone(self);
+        run_decommission_buckets_bounded(rx, buckets, limit, move |bucket, rx| {
+            let store = Arc::clone(&store);
+            let pool = pool.clone();
+            let entry_budget = entry_budget.clone();
+            Box::pin(async move { store.decommission_pending_bucket(rx, idx, pool, bucket, entry_budget).await })
+        })
+        .await
+    }
+
     #[tracing::instrument(skip(self, rx))]
     async fn decommission_in_background(
         self: &Arc<Self>,
@@ -4946,15 +4935,31 @@ impl ECStore {
             pool_meta.pending_buckets(idx)
         };
         let bucket_concurrency = decommission_bucket_concurrency_limit();
+        if bucket_concurrency <= 1 {
+            for bucket in pending {
+                self.decommission_pending_bucket(rx.clone(), idx, pool.clone(), bucket, entry_budget.clone())
+                    .await?;
+            }
+            return Ok(());
+        }
+
         let (regular_buckets, meta_buckets) = split_decommission_buckets(pending);
-        let store = Arc::clone(self);
-        run_decommission_phases(rx, regular_buckets, meta_buckets, bucket_concurrency, move |bucket, rx| {
-            let store = Arc::clone(&store);
-            let pool = pool.clone();
-            let entry_budget = entry_budget.clone();
-            Box::pin(async move { store.decommission_pending_bucket(rx, idx, pool, bucket, entry_budget).await })
-        })
-        .await
+        self.decommission_buckets_concurrently(
+            rx.clone(),
+            idx,
+            pool.clone(),
+            regular_buckets,
+            bucket_concurrency,
+            entry_budget.clone(),
+        )
+        .await?;
+
+        for bucket in meta_buckets {
+            self.decommission_pending_bucket(rx.clone(), idx, pool.clone(), bucket, entry_budget.clone())
+                .await?;
+        }
+
+        Ok(())
     }
 
     #[tracing::instrument(skip(self))]
@@ -6372,8 +6377,8 @@ mod pools_tests {
         resolve_decommission_terminal_mark_after_error_result, resolve_decommission_terminal_mark_result,
         resolve_decommission_update_after_result, resolve_start_decommission_pool_meta_reload_result,
         rollback_start_decommission_pool_meta, run_decommission_buckets_bounded, run_decommission_listing_with_retry,
-        run_decommission_listing_with_retry_and_drain, run_decommission_phases, run_decommission_side_effect,
-        should_cleanup_decommission_source_entry, should_continue_decommission_queue, should_count_decommission_version_complete,
+        run_decommission_listing_with_retry_and_drain, run_decommission_side_effect, should_cleanup_decommission_source_entry,
+        should_continue_decommission_queue, should_count_decommission_version_complete,
         should_preserve_decommission_canceled_state, should_reject_decommission_cancel_as_terminal,
         should_retry_decommission_cancel_reload, should_retry_decommission_listing, should_skip_canceled_decommission_routine,
         spawn_decommission_index_cancelers, split_decommission_buckets, take_and_cancel_decommission_canceler,
@@ -6392,7 +6397,7 @@ mod pools_tests {
     use rustfs_filemeta::{MetaCacheEntries, MetadataResolutionParams};
     use rustfs_rio::Index;
     use std::sync::{
-        Arc, Mutex as StdMutex,
+        Arc,
         atomic::{AtomicBool, AtomicUsize, Ordering},
     };
     use std::time::Duration as StdDuration;
@@ -6685,66 +6690,6 @@ mod pools_tests {
         );
     }
 
-    #[tokio::test]
-    async fn test_decommission_metadata_phase_precedes_regular_failure() {
-        let events = Arc::new(StdMutex::new(Vec::new()));
-        let err = run_decommission_phases(
-            CancellationToken::new(),
-            vec![
-                DecomBucketInfo {
-                    name: "regular-fails".to_string(),
-                    ..Default::default()
-                },
-                DecomBucketInfo {
-                    name: "regular-not-started".to_string(),
-                    ..Default::default()
-                },
-            ],
-            vec![
-                DecomBucketInfo {
-                    name: crate::disk::RUSTFS_META_BUCKET.to_string(),
-                    prefix: crate::config::com::CONFIG_PREFIX.to_string(),
-                },
-                DecomBucketInfo {
-                    name: crate::disk::RUSTFS_META_BUCKET.to_string(),
-                    prefix: crate::disk::BUCKET_META_PREFIX.to_string(),
-                },
-            ],
-            1,
-            {
-                let events = Arc::clone(&events);
-                move |bucket, _rx| {
-                    let events = Arc::clone(&events);
-                    Box::pin(async move {
-                        let event = if bucket.name == crate::disk::RUSTFS_META_BUCKET {
-                            format!("meta:{}", bucket.prefix)
-                        } else {
-                            format!("regular:{}", bucket.name)
-                        };
-                        events.lock().expect("phase event lock should not be poisoned").push(event);
-                        if bucket.name == "regular-fails" {
-                            Err(Error::SlowDown)
-                        } else {
-                            Ok(())
-                        }
-                    })
-                }
-            },
-        )
-        .await
-        .expect_err("regular failure should remain fatal after metadata completes");
-
-        assert!(matches!(err, Error::SlowDown));
-        assert_eq!(
-            *events.lock().expect("phase event lock should not be poisoned"),
-            vec![
-                format!("meta:{}", crate::config::com::CONFIG_PREFIX),
-                format!("meta:{}", crate::disk::BUCKET_META_PREFIX),
-                "regular:regular-fails".to_string(),
-            ]
-        );
-    }
-
     #[tokio::test]
     async fn test_run_decommission_buckets_bounded_respects_limit() {
         let rx = CancellationToken::new();

crates/ecstore/src/core/sets.rs

@@ -988,14 +988,11 @@ impl crate::storage_api_contracts::multipart::MultipartOperations for Sets {
     }
 }
 
-#[async_trait::async_trait]
-impl crate::storage_api_contracts::heal::HealOperations for Sets {
-    type Error = Error;
-    type HealResultItem = HealResultItem;
-    type HealOptions = HealOpts;
-
-    #[tracing::instrument(skip(self))]
-    async fn heal_format(&self, dry_run: bool) -> Result<(HealResultItem, Option<Error>)> {
+impl Sets {
+    pub(crate) async fn heal_format_with_fence<F>(&self, dry_run: bool, fence_lost: F) -> Result<(HealResultItem, Option<Error>)>
+    where
+        F: Fn() -> bool + Send + Sync,
+    {
         let (disks, init_errs) = init_storage_disks_with_errors(
             &self.endpoints.endpoints,
             &DiskOption {
@@ -1068,6 +1065,9 @@ impl crate::storage_api_contracts::heal::HealOperations for Sets {
             // Save new formats `format.json` on unformatted disks.
             for (index, (fm, disk)) in tmp_new_formats.iter_mut().zip(disks.iter()).enumerate() {
                 if fm.is_some() && disk.is_some() {
+                    if fence_lost() {
+                        return Ok((res, Some(StorageError::SlowDown)));
+                    }
                     if let Err(err) = save_format_file(disk, fm).await {
                         if let Some(disk) = disk.as_ref() {
                             let _ = disk.close().await;
@@ -1101,6 +1101,18 @@ impl crate::storage_api_contracts::heal::HealOperations for Sets {
         }
         Ok((res, None))
     }
+}
+
+#[async_trait::async_trait]
+impl crate::storage_api_contracts::heal::HealOperations for Sets {
+    type Error = Error;
+    type HealResultItem = HealResultItem;
+    type HealOptions = HealOpts;
+
+    #[tracing::instrument(skip(self))]
+    async fn heal_format(&self, dry_run: bool) -> Result<(HealResultItem, Option<Error>)> {
+        self.heal_format_with_fence(dry_run, || false).await
+    }
     #[tracing::instrument(skip(self))]
     async fn heal_bucket(&self, bucket: &str, opts: &HealOpts) -> Result<HealResultItem> {
         let mut result = HealResultItem {

crates/ecstore/src/store/heal.rs

@@ -13,7 +13,12 @@
 // limitations under the License.
 
 use super::*;
+use crate::core::pools::POOL_META_NAME;
+use crate::services::rebalance::{REBAL_META_NAME, RebalStatus};
+use crate::set_disk::get_lock_acquire_timeout;
 use crate::storage_api_contracts::heal::HealOperations as _;
+use crate::storage_api_contracts::namespace::NamespaceLocking as _;
+use rustfs_lock::NamespaceLockGuard;
 use tracing::trace;
 
 const LOG_COMPONENT_ECSTORE: &str = "ecstore";
@@ -30,7 +35,119 @@ fn invalid_heal_pool_index(pool_idx: usize, pool_count: usize) -> Error {
     )
 }
 
+#[derive(Debug, Clone, Copy)]
+enum HealFormatPoolSkip {
+    Completed,
+    Retryable,
+}
+
+fn classify_heal_format_pool(
+    pool_idx: usize,
+    pool_cmd_line: &str,
+    pool_meta: &PoolMeta,
+    rebalance_meta: Option<&RebalanceMeta>,
+) -> Option<HealFormatPoolSkip> {
+    let Some(pool) = pool_meta.pools.get(pool_idx) else {
+        return Some(HealFormatPoolSkip::Retryable);
+    };
+
+    if pool.id != pool_idx || pool_cmd_line.is_empty() || pool.cmd_line.is_empty() || pool.cmd_line != pool_cmd_line {
+        return Some(HealFormatPoolSkip::Retryable);
+    }
+
+    if let Some(decommission) = pool.decommission.as_ref() {
+        if decommission.complete {
+            return Some(HealFormatPoolSkip::Completed);
+        }
+        if decommission.failed || decommission.canceled || decommission.queued || pool_meta.is_suspended(pool_idx) {
+            return Some(HealFormatPoolSkip::Retryable);
+        }
+    }
+
+    if let Some(meta) = rebalance_meta {
+        let Some(pool_stats) = meta.pool_stats.get(pool_idx) else {
+            return Some(HealFormatPoolSkip::Retryable);
+        };
+        if pool_stats.info.stopping || (pool_stats.participating && pool_stats.info.status == RebalStatus::Started) {
+            return Some(HealFormatPoolSkip::Retryable);
+        }
+    }
+
+    None
+}
+
+fn heal_format_pool_skip_error(skip: HealFormatPoolSkip) -> Error {
+    match skip {
+        HealFormatPoolSkip::Completed => StorageError::NoHealRequired,
+        HealFormatPoolSkip::Retryable => StorageError::SlowDown,
+    }
+}
+
+fn heal_format_fence_lost_error() -> Error {
+    StorageError::SlowDown
+}
+
 impl ECStore {
+    async fn acquire_heal_format_fence(
+        &self,
+    ) -> Result<(NamespaceLockGuard, NamespaceLockGuard, PoolMeta, Option<RebalanceMeta>)> {
+        let metadata_pool = self
+            .pools
+            .first()
+            .cloned()
+            .ok_or_else(|| Error::other("heal format requires at least one storage pool"))?;
+
+        // Metadata fence order is part of the decommission/rebalance protocol:
+        // pool.bin must always be acquired before rebalance.bin.
+        let pool_lock = metadata_pool.new_ns_lock(RUSTFS_META_BUCKET, POOL_META_NAME).await?;
+        let pool_guard = pool_lock.get_write_lock(get_lock_acquire_timeout()).await?;
+        let rebalance_lock = metadata_pool.new_ns_lock(RUSTFS_META_BUCKET, REBAL_META_NAME).await?;
+        let rebalance_guard = rebalance_lock.get_write_lock(get_lock_acquire_timeout()).await?;
+
+        if pool_guard.is_lock_lost() || rebalance_guard.is_lock_lost() {
+            return Err(heal_format_fence_lost_error());
+        }
+
+        let mut pool_meta = PoolMeta::default();
+        pool_meta.load_no_lock(metadata_pool.clone()).await?;
+        if pool_meta.pools.len() != self.pools.len()
+            || pool_meta.pools.iter().enumerate().any(|(pool_idx, pool)| {
+                pool.id != pool_idx || pool.cmd_line.is_empty() || pool.cmd_line != self.pools[pool_idx].endpoints.cmd_line
+            })
+        {
+            return Err(heal_format_fence_lost_error());
+        }
+
+        let mut rebalance_meta = RebalanceMeta::new();
+        let rebalance_meta = match rebalance_meta
+            .load_with_opts(
+                metadata_pool,
+                ObjectOptions {
+                    no_lock: true,
+                    ..Default::default()
+                },
+            )
+            .await
+        {
+            Ok(()) => Some(rebalance_meta),
+            Err(Error::ConfigNotFound) => None,
+            Err(err) => return Err(err),
+        };
+
+        if rebalance_meta
+            .as_ref()
+            .is_some_and(|meta| meta.pool_stats.len() != self.pools.len())
+        {
+            return Err(heal_format_fence_lost_error());
+        }
+
+        if pool_guard.is_lock_lost() || rebalance_guard.is_lock_lost() {
+            return Err(heal_format_fence_lost_error());
+        }
+
+        Ok((pool_guard, rebalance_guard, pool_meta, rebalance_meta))
+    }
+
     fn get_pools_for_heal_object(&self, opts: &HealOpts) -> Result<Vec<Arc<Sets>>> {
         match opts.pool {
             Some(pool_idx) => Ok(vec![
@@ -52,9 +169,26 @@ impl ECStore {
         };
 
         let mut count_no_heal = 0;
+        let mut count_completed = 0;
         let mut first_error = None;
-        for pool in self.pools.iter() {
-            let (mut result, err) = pool.heal_format(dry_run).await?;
+        for (pool_idx, pool) in self.pools.iter().enumerate() {
+            let (pool_guard, rebalance_guard, pool_meta, rebalance_meta) = self.acquire_heal_format_fence().await?;
+            if pool_guard.is_lock_lost() || rebalance_guard.is_lock_lost() {
+                first_error.get_or_insert(heal_format_fence_lost_error());
+                break;
+            }
+            if let Some(skip) = classify_heal_format_pool(pool_idx, &pool.endpoints.cmd_line, &pool_meta, rebalance_meta.as_ref())
+            {
+                if matches!(skip, HealFormatPoolSkip::Completed) {
+                    count_completed += 1;
+                } else {
+                    first_error.get_or_insert(heal_format_pool_skip_error(skip));
+                }
+                continue;
+            }
+
+            let fence_lost = || pool_guard.is_lock_lost() || rebalance_guard.is_lock_lost();
+            let (mut result, err) = pool.heal_format_with_fence(dry_run, fence_lost).await?;
             if let Some(err) = err {
                 match err {
                     StorageError::NoHealRequired => {
@@ -69,11 +203,18 @@ impl ECStore {
             r.set_count += result.set_count;
             r.before.drives.append(&mut result.before.drives);
             r.after.drives.append(&mut result.after.drives);
+
+            // A lease can be lost after the final write; fail closed before
+            // reporting the pool as successfully healed.
+            if pool_guard.is_lock_lost() || rebalance_guard.is_lock_lost() {
+                first_error.get_or_insert(heal_format_fence_lost_error());
+                break;
+            }
         }
         if let Some(err) = first_error {
             return Ok((r, Some(err)));
         }
-        if count_no_heal == self.pools.len() {
+        if count_no_heal + count_completed == self.pools.len() {
             info!(
                 event = EVENT_HEAL_FORMAT_COMPLETED,
                 component = LOG_COMPONENT_ECSTORE,
@@ -302,6 +443,7 @@ mod tests {
     use crate::disk::{DeleteOptions, DiskOption, format::FormatV3, new_disk};
     use crate::layout::endpoints::{EndpointServerPools, Endpoints, PoolEndpoints};
     use crate::runtime::instance::InstanceContext;
+    use crate::services::rebalance::{RebalanceInfo, RebalanceStats};
     use crate::storage_api_contracts::bucket::{BucketOperations, MakeBucketOptions};
     use crate::storage_api_contracts::object::{ObjectIO as _, ObjectOperations};
     use crate::store::init_format::{load_format_erasure, save_format_file};
@@ -353,6 +495,164 @@ mod tests {
         }
     }
 
+    fn pool_meta_with_decommission(info: PoolDecommissionInfo) -> PoolMeta {
+        PoolMeta {
+            pools: vec![PoolStatus {
+                id: 0,
+                cmd_line: "pool-0".to_string(),
+                last_update: OffsetDateTime::UNIX_EPOCH,
+                decommission: Some(info),
+            }],
+            ..Default::default()
+        }
+    }
+
+    #[test]
+    fn heal_format_pool_state_barriers_are_classified() {
+        let active = pool_meta_with_decommission(PoolDecommissionInfo {
+            start_time: Some(OffsetDateTime::UNIX_EPOCH),
+            ..Default::default()
+        });
+        assert!(matches!(
+            classify_heal_format_pool(0, "pool-0", &active, None),
+            Some(HealFormatPoolSkip::Retryable)
+        ));
+
+        for info in [
+            PoolDecommissionInfo {
+                failed: true,
+                ..Default::default()
+            },
+            PoolDecommissionInfo {
+                canceled: true,
+                ..Default::default()
+            },
+        ] {
+            assert!(matches!(
+                classify_heal_format_pool(0, "pool-0", &pool_meta_with_decommission(info), None),
+                Some(HealFormatPoolSkip::Retryable)
+            ));
+        }
+
+        let completed = pool_meta_with_decommission(PoolDecommissionInfo {
+            complete: true,
+            ..Default::default()
+        });
+        assert!(matches!(
+            classify_heal_format_pool(0, "pool-0", &completed, None),
+            Some(HealFormatPoolSkip::Completed)
+        ));
+    }
+
+    #[test]
+    fn heal_format_pool_rebalance_barriers_and_identity_are_fail_closed() {
+        let identity_meta = pool_meta_with_decommission(PoolDecommissionInfo::default());
+        let rebalance = RebalanceMeta {
+            pool_stats: vec![RebalanceStats {
+                participating: true,
+                info: RebalanceInfo {
+                    status: RebalStatus::Started,
+                    ..Default::default()
+                },
+                ..Default::default()
+            }],
+            ..Default::default()
+        };
+        assert!(matches!(
+            classify_heal_format_pool(0, "pool-0", &identity_meta, Some(&rebalance)),
+            Some(HealFormatPoolSkip::Retryable)
+        ));
+
+        let stopping = RebalanceMeta {
+            pool_stats: vec![RebalanceStats {
+                info: RebalanceInfo {
+                    stopping: true,
+                    ..Default::default()
+                },
+                ..Default::default()
+            }],
+            ..Default::default()
+        };
+        assert!(matches!(
+            classify_heal_format_pool(0, "pool-0", &identity_meta, Some(&stopping)),
+            Some(HealFormatPoolSkip::Retryable)
+        ));
+
+        let identity = pool_meta_with_decommission(PoolDecommissionInfo::default());
+        assert!(matches!(
+            classify_heal_format_pool(0, "pool-new", &identity, None),
+            Some(HealFormatPoolSkip::Retryable)
+        ));
+
+        let identity_without_decommission = PoolMeta {
+            pools: vec![PoolStatus {
+                id: 0,
+                cmd_line: "pool-0".to_string(),
+                last_update: OffsetDateTime::UNIX_EPOCH,
+                decommission: None,
+            }],
+            ..Default::default()
+        };
+        assert!(matches!(
+            classify_heal_format_pool(0, "pool-new", &identity_without_decommission, None),
+            Some(HealFormatPoolSkip::Retryable)
+        ));
+
+        assert!(matches!(
+            classify_heal_format_pool(0, "", &identity_meta, None),
+            Some(HealFormatPoolSkip::Retryable)
+        ));
+
+        assert!(matches!(
+            classify_heal_format_pool(0, "pool-0", &PoolMeta::default(), None),
+            Some(HealFormatPoolSkip::Retryable)
+        ));
+
+        let stopped = RebalanceMeta {
+            stopped_at: Some(OffsetDateTime::UNIX_EPOCH),
+            pool_stats: vec![RebalanceStats {
+                participating: true,
+                info: RebalanceInfo {
+                    status: RebalStatus::Stopped,
+                    ..Default::default()
+                },
+                ..Default::default()
+            }],
+            ..Default::default()
+        };
+        assert!(classify_heal_format_pool(0, "pool-0", &identity_meta, Some(&stopped)).is_none());
+
+        let stopping_after_stop = RebalanceMeta {
+            stopped_at: Some(OffsetDateTime::UNIX_EPOCH),
+            pool_stats: vec![RebalanceStats {
+                participating: true,
+                info: RebalanceInfo {
+                    status: RebalStatus::Started,
+                    stopping: true,
+                    ..Default::default()
+                },
+                ..Default::default()
+            }],
+            ..Default::default()
+        };
+        assert!(matches!(
+            classify_heal_format_pool(0, "pool-0", &identity_meta, Some(&stopping_after_stop)),
+            Some(HealFormatPoolSkip::Retryable)
+        ));
+    }
+
+    #[test]
+    fn skipped_heal_format_pool_is_never_reported_as_success() {
+        assert!(matches!(
+            heal_format_pool_skip_error(HealFormatPoolSkip::Retryable),
+            StorageError::SlowDown
+        ));
+        assert!(matches!(
+            heal_format_pool_skip_error(HealFormatPoolSkip::Completed),
+            StorageError::NoHealRequired
+        ));
+    }
+
     async fn multi_pool_heal_store() -> (tempfile::TempDir, Arc<ECStore>, CancellationToken) {
         let temp_dir = tempfile::tempdir().expect("multi-pool heal test directory should be created");
         let mut pool_endpoints = Vec::new();
@@ -889,6 +1189,18 @@ mod tests {
             bucket_fence_registry: std::sync::Arc::default(),
         };
 
+        let err = store
+            .handle_heal_format(false)
+            .await
+            .expect_err("missing pool metadata must fail closed before format writes");
+        assert!(matches!(err, StorageError::SlowDown));
+
+        let pool_meta = PoolMeta::new(&store.pools, &PoolMeta::default());
+        pool_meta
+            .save(store.pools.clone())
+            .await
+            .expect("pool metadata should be persisted before format heal");
+
         let (result, err) = store
             .handle_heal_format(false)
             .await
@@ -902,5 +1214,22 @@ mod tests {
             .await
             .expect("the later pool should be healed despite the first pool error");
         assert_eq!(healed.erasure.this, recoverable_format.erasure.sets[0][2]);
+
+        let mut completed_meta = PoolMeta::new(&store.pools, &PoolMeta::default());
+        for status in &mut completed_meta.pools {
+            status.decommission = Some(PoolDecommissionInfo {
+                complete: true,
+                ..Default::default()
+            });
+        }
+        completed_meta
+            .save(store.pools.clone())
+            .await
+            .expect("completed pool metadata should be persisted");
+        let (_, err) = store
+            .handle_heal_format(false)
+            .await
+            .expect("completed pools should be reported as a no-op");
+        assert!(matches!(err, Some(StorageError::NoHealRequired)));
     }
 }

crates/ecstore/src/store/object.rs

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

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

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

crates/heal/src/heal/task/heal_erasure_set.rs

@@ -231,6 +231,10 @@ impl HealTask {
                             "Heal erasure set format repair skipped because no format heal was required"
                         );
                     } else {
+                        let error = e;
+                        if error.is_recoverable_heal() {
+                            return Err(error);
+                        }
                         error!(
                             target: "rustfs::heal::task",
                             event = EVENT_HEAL_ERASURE_SET_RESULT,
@@ -239,7 +243,7 @@ impl HealTask {
                             task_id = %self.id,
                             set_disk_id,
                             result = "format_failed",
-                            error = %e,
+                            error = %error,
                             "Heal erasure set failed"
                         );
                         {
@@ -247,7 +251,7 @@ impl HealTask {
                             progress.update_progress(4, 4, 0, 0);
                         }
                         return Err(Error::TaskExecutionFailed {
-                            message: format!("Failed to heal disk format for {set_disk_id}: {e}"),
+                            message: format!("Failed to heal disk format for {set_disk_id}: {error}"),
                         });
                     }
                 } else {
@@ -284,6 +288,9 @@ impl HealTask {
             Err(Error::TaskCancelled) => return Err(Error::TaskCancelled),
             Err(Error::TaskTimeout) => return Err(Error::TaskTimeout),
             Err(e) => {
+                if e.is_recoverable_heal() {
+                    return Err(e);
+                }
                 error!(
                     target: "rustfs::heal::task",
                     event = EVENT_HEAL_ERASURE_SET_RESULT,

crates/heal/src/heal/task/tests.rs

@@ -547,6 +547,7 @@ struct MockStorage {
     heal_object_outcome: Mutex<Option<MockHealObjectOutcome>>,
     heal_object_outcomes: Mutex<HashMap<String, VecDeque<MockHealObjectOutcome>>>,
     format_no_heal_required: Mutex<bool>,
+    format_error: Mutex<Option<Error>>,
     global_format_calls: Mutex<u32>,
     replacement_format_calls: Mutex<Vec<(usize, usize, Vec<String>)>>,
     replacement_targets_ready: Mutex<bool>,
@@ -867,6 +868,9 @@ impl HealStorageAPI for MockStorage {
 
     async fn heal_format(&self, _dry_run: bool) -> Result<(HealResultItem, Option<Error>)> {
         *self.global_format_calls.lock().unwrap() += 1;
+        if let Some(error) = self.format_error.lock().unwrap().take() {
+            return Err(error);
+        }
         let no_heal_required = *self.format_no_heal_required.lock().unwrap();
         if no_heal_required {
             Ok((HealResultItem::default(), Some(Error::Storage(EcstoreError::NoHealRequired))))
@@ -2052,6 +2056,30 @@ async fn test_erasure_set_heal_continues_after_format_no_heal_required() {
     );
 }
 
+#[tokio::test]
+async fn erasure_set_format_slowdown_is_propagated() {
+    let storage = Arc::new(MockStorage {
+        format_error: Mutex::new(Some(Error::Storage(EcstoreError::SlowDown))),
+        ..Default::default()
+    });
+    let request = HealRequest::new(
+        HealType::ErasureSet {
+            buckets: Vec::new(),
+            set_disk_id: "pool_0_set_0".to_string(),
+        },
+        HealOptions::default(),
+        HealPriority::Normal,
+    );
+    let task = HealTask::from_request(request, storage);
+
+    let error = task
+        .execute()
+        .await
+        .expect_err("format SlowDown must remain recoverable for the task manager");
+
+    assert!(matches!(error, Error::Storage(EcstoreError::SlowDown)));
+}
+
 #[tokio::test]
 async fn erasure_set_bucket_prepass_failure_stops_before_object_heal() {
     let temp = TempDir::new().expect("temporary directory should be created");

crates/test-utils/src/lib.rs

@@ -245,6 +245,18 @@ impl TestECStoreEnvBuilder {
             .await
             .expect("build test ECStore");
 
+        // The production bootstrap only persists pool.bin from the elected
+        // first cluster node.  Test stores intentionally have no cluster
+        // election, but heal-format still requires that durable fence before
+        // it can write any disk format.  Materialize the validated topology
+        // here so the shared fixture models a ready single-node store.
+        let mut pool_meta = ecstore.pool_meta.read().await.clone();
+        pool_meta.dont_save = false;
+        pool_meta
+            .save(ecstore.pools.clone())
+            .await
+            .expect("persist test pool metadata");
+
         if self.init_bucket_metadata {
             let buckets_list = ecstore
                 .list_bucket(&BucketOptions {

rustfs/Cargo.toml

@@ -322,7 +322,7 @@ thiserror = { workspace = true }
 tracing.workspace = true
 url = { workspace = true }
 urlencoding = { workspace = true }
-uuid = { workspace = true, features = ["v4", "fast-rng", "macro-diagnostics"] }
+uuid = { workspace = true, features = ["v4", "v5", "fast-rng", "macro-diagnostics"] }
 zip = { workspace = true }
 libc = { workspace = true }
 rand = { workspace = true, features = ["serde"] }
@@ -345,7 +345,7 @@ libsystemd.workspace = true
 libmimalloc-sys.workspace = true
 
 [dev-dependencies]
-uuid = { workspace = true, features = ["v4", "fast-rng", "macro-diagnostics"] }
+uuid = { workspace = true, features = ["v4", "v5", "fast-rng", "macro-diagnostics"] }
 serial_test = { workspace = true }
 tempfile = { workspace = true }
 aws-config = { workspace = true }

rustfs/src/admin/handlers/site_replication.rs

@@ -41,7 +41,7 @@ use crate::admin::storage_api::config::save_admin_config;
 use crate::admin::storage_api::contract::bucket::{
     BucketOperations, BucketOptions, DeleteBucketOptions, MakeBucketOptions, SRBucketDeleteOp,
 };
-use crate::admin::storage_api::error::Error as StorageError;
+use crate::admin::storage_api::error::{Error as StorageError, is_err_bucket_not_found};
 use crate::admin::storage_api::runtime::ECStore;
 use crate::admin::utils::{encode_compatible_admin_payload, read_compatible_admin_body};
 use crate::auth::constant_time_eq;
@@ -55,6 +55,7 @@ use crate::storage::storage_api::{
 use base64::Engine;
 use base64::engine::general_purpose::STANDARD as BASE64_STANDARD;
 use base64::engine::general_purpose::URL_SAFE_NO_PAD;
+use futures::StreamExt;
 use hmac::{Hmac, Mac};
 use http::header::{CONTENT_TYPE, HOST};
 use http::{HeaderMap, HeaderValue, Uri};
@@ -2096,6 +2097,18 @@ async fn remote_add_preflight_info(site: &PeerSite) -> S3Result<SiteReplicationA
             format!("invalid site replication metainfo from `{}`: {e}", site.endpoint),
         )
     })?;
+    if info.deployment_id.is_empty() {
+        // The peer will be tracked under a locally derived fallback ID
+        // (deployment_id_for_endpoint) instead of its real deployment ID.
+        warn!(
+            event = EVENT_ADMIN_SITE_REPLICATION_STATE,
+            component = LOG_COMPONENT_ADMIN,
+            subsystem = LOG_SUBSYSTEM_SITE_REPLICATION,
+            result = "peer_deployment_id_missing",
+            peer_endpoint = %site.endpoint,
+            "admin site replication state"
+        );
+    }
 
     let idp_body = send_peer_admin_get_request_with_client(
         &client,
@@ -2206,20 +2219,30 @@ fn site_replication_bootstrap_token(uri: &Uri) -> Option<String> {
     query_pairs(uri).get("bootstrapToken").cloned()
 }
 
-fn bootstrap_bucket_make_op_path(bucket: &SRBucketInfo) -> String {
+/// Query for a peer `make-with-versioning` bucket op. `versioningEnabled`
+/// always travels so the outbound query matches MinIO's site-replication
+/// make-bucket wire contract: MinIO's own create-bucket hook sends
+/// `versioningEnabled=true` on this op. RustFS's inbound handler
+/// force-enables versioning either way.
+fn make_with_versioning_bucket_op_path(bucket: &str, created_at: Option<&str>, lock_enabled: bool) -> String {
     let mut query = form_urlencoded::Serializer::new(String::new());
-    query.append_pair("bucket", &bucket.bucket);
-    query.append_pair("operation", "make-with-versioning");
-    if let Some(created_at) = bucket
-        .created_at
-        .and_then(|value| value.format(&time::format_description::well_known::Rfc3339).ok())
-    {
-        query.append_pair("createdAt", &created_at);
+    query.append_pair("bucket", bucket);
+    query.append_pair("operation", SITE_REPLICATION_BUCKET_OP_MAKE_WITH_VERSIONING);
+    query.append_pair("versioningEnabled", "true");
+    if let Some(created_at) = created_at {
+        query.append_pair("createdAt", created_at);
     }
-    if bucket.object_lock_config.is_some() {
+    if lock_enabled {
         query.append_pair("lockEnabled", "true");
     }
-    format!("/rustfs/admin/v3/site-replication/peer/bucket-ops?{}", query.finish())
+    format!("{SITE_REPLICATION_PEER_BUCKET_OPS_PATH}?{}", query.finish())
+}
+
+fn bootstrap_bucket_make_op_path(bucket: &SRBucketInfo) -> String {
+    let created_at = bucket
+        .created_at
+        .and_then(|value| value.format(&time::format_description::well_known::Rfc3339).ok());
+    make_with_versioning_bucket_op_path(&bucket.bucket, created_at.as_deref(), bucket.object_lock_config.is_some())
 }
 
 fn bootstrap_bucket_meta_item(bucket: &SRBucketInfo, item_type: &str, updated_at: Option<OffsetDateTime>) -> SRBucketMeta {
@@ -4246,16 +4269,7 @@ async fn broadcast_site_replication_make_bucket(
         .format(&time::format_description::well_known::Rfc3339)
         .unwrap_or_default();
 
-    let path = {
-        let mut query = form_urlencoded::Serializer::new(String::new());
-        query.append_pair("bucket", bucket);
-        query.append_pair("operation", "make-with-versioning");
-        query.append_pair("createdAt", &created_at);
-        if lock_enabled {
-            query.append_pair("lockEnabled", "true");
-        }
-        format!("/rustfs/admin/v3/site-replication/peer/bucket-ops?{}", query.finish())
-    };
+    let path = make_with_versioning_bucket_op_path(bucket, Some(&created_at), lock_enabled);
     let path = if let Some(token) = bootstrap_token {
         with_site_replication_bootstrap_token(&path, token)
     } else {
@@ -10206,13 +10220,25 @@ impl Operation for SiteReplicationStatusHandler {
     }
 }
 
+/// `POST /v3/site-replication/devnull` — peer link-check upload drain.
+/// MinIO streams multi-megabyte probe bodies here during site netperf link
+/// checks and expects an unbounded discard (its handler copies to io.Discard);
+/// buffering through the 1MB admin body cap turned any larger probe into a
+/// 400 and a false link failure. Stream and discard instead — no size cap.
+async fn drain_site_replication_devnull(mut input: Body) -> S3Result<()> {
+    while let Some(chunk) = input.next().await {
+        chunk.map_err(|e| s3_error!(InvalidRequest, "failed to read devnull stream: {}", e))?;
+    }
+    Ok(())
+}
+
 pub struct SiteReplicationDevNullHandler {}
 
 #[async_trait::async_trait]
 impl Operation for SiteReplicationDevNullHandler {
     async fn call(&self, req: S3Request<Body>, _params: Params<'_, '_>) -> S3Result<S3Response<(StatusCode, Body)>> {
         validate_site_replication_admin_request(&req, AdminAction::SiteReplicationOperationAction).await?;
-        let _ = read_plain_admin_body(req.input).await?;
+        drain_site_replication_devnull(req.input).await?;
         Ok(empty_response(StatusCode::NO_CONTENT))
     }
 }
@@ -10471,6 +10497,19 @@ impl Operation for SRPeerJoinHandler {
     }
 }
 
+/// Outcome of a peer-driven `purge-deleted-bucket` replay. A bucket that is
+/// already gone means the purge raced an earlier replay or a local delete —
+/// that is success — but any other failure must reach the sender like the
+/// sibling delete branches do: swallowing it answered 200 while the bucket
+/// survived on this site.
+fn purge_deleted_bucket_result(result: Result<(), StorageError>) -> S3Result<()> {
+    match result {
+        Ok(()) => Ok(()),
+        Err(err) if is_err_bucket_not_found(&err) => Ok(()),
+        Err(err) => Err(ApiError::from(err).into()),
+    }
+}
+
 pub struct SRPeerBucketOpsHandler {}
 
 #[async_trait::async_trait]
@@ -10570,16 +10609,18 @@ impl Operation for SRPeerBucketOpsHandler {
                     .map_err(ApiError::from)?;
             }
             "purge-deleted-bucket" => {
-                let _ = store
-                    .delete_bucket(
-                        &bucket,
-                        &DeleteBucketOptions {
-                            force: true,
-                            srdelete_op: SRBucketDeleteOp::Purge,
-                            ..Default::default()
-                        },
-                    )
-                    .await;
+                purge_deleted_bucket_result(
+                    store
+                        .delete_bucket(
+                            &bucket,
+                            &DeleteBucketOptions {
+                                force: true,
+                                srdelete_op: SRBucketDeleteOp::Purge,
+                                ..Default::default()
+                            },
+                        )
+                        .await,
+                )?;
             }
             _ => return Err(s3_error!(InvalidRequest, "unsupported site replication bucket operation")),
         }
@@ -13925,6 +13966,54 @@ mod tests {
         assert!(!query_flag(&uri, "missing"));
     }
 
+    /// A5 red-light: a `purge-deleted-bucket` replay must report success when
+    /// the bucket is already gone, and must propagate every other failure —
+    /// the swallowed error answered 200 while the bucket survived.
+    #[test]
+    fn test_purge_deleted_bucket_result_tolerates_only_missing_bucket() {
+        assert!(purge_deleted_bucket_result(Ok(())).is_ok());
+        assert!(purge_deleted_bucket_result(Err(StorageError::BucketNotFound("photos".to_string()))).is_ok());
+        assert!(purge_deleted_bucket_result(Err(StorageError::VolumeNotFound)).is_ok());
+        let err = purge_deleted_bucket_result(Err(StorageError::StorageFull))
+            .expect_err("non-not-found delete failures must propagate");
+        assert_ne!(*err.code(), S3ErrorCode::NoSuchBucket);
+    }
+
+    /// C5 red-light: the site-replication devnull drain must accept bodies
+    /// beyond the 1MB admin body cap — MinIO's link check streams large
+    /// probe bodies and treats a 400 as a broken link.
+    #[tokio::test]
+    async fn test_site_replication_devnull_drains_body_beyond_admin_cap() {
+        let body = Body::from(vec![0u8; MAX_ADMIN_REQUEST_BODY_SIZE + 1]);
+        drain_site_replication_devnull(body)
+            .await
+            .expect("devnull must drain bodies larger than the admin body cap");
+    }
+
+    /// A3 red-light: `versioningEnabled` must travel on every outbound
+    /// make-with-versioning bucket op so the query matches MinIO's
+    /// site-replication make-bucket wire contract (MinIO's own hook sends
+    /// `versioningEnabled=true` on this op).
+    #[test]
+    fn test_make_with_versioning_op_paths_send_versioning_enabled() {
+        let bucket = SRBucketInfo {
+            bucket: "photos".to_string(),
+            created_at: Some(OffsetDateTime::UNIX_EPOCH),
+            object_lock_config: Some(BASE64_STANDARD.encode("<ObjectLockConfiguration/>")),
+            ..Default::default()
+        };
+        let bootstrap = bootstrap_bucket_make_op_path(&bucket);
+        assert!(bootstrap.contains("operation=make-with-versioning"), "{bootstrap}");
+        assert!(bootstrap.contains("versioningEnabled=true"), "{bootstrap}");
+        assert!(bootstrap.contains("createdAt="), "{bootstrap}");
+        assert!(bootstrap.contains("lockEnabled=true"), "{bootstrap}");
+
+        // The broadcast path (create-bucket hook) shares the same builder.
+        let broadcast = make_with_versioning_bucket_op_path("photos", Some("1970-01-01T00:00:00Z"), false);
+        assert!(broadcast.contains("versioningEnabled=true"), "{broadcast}");
+        assert!(!broadcast.contains("lockEnabled"), "{broadcast}");
+    }
+
     #[tokio::test]
     #[serial]
     async fn test_add_bootstrap_scope_only_allows_expected_bucket_setup_until_guard_drops() {

rustfs/src/admin/site_replication_identity.rs

@@ -13,9 +13,9 @@
 // limitations under the License.
 
 use rustfs_madmin::{PeerInfo, SyncStatus};
-use std::collections::{BTreeMap, hash_map::DefaultHasher};
-use std::hash::{Hash, Hasher};
+use std::collections::BTreeMap;
 use url::Url;
+use uuid::Uuid;
 
 fn has_http_scheme(endpoint: &str) -> bool {
     endpoint.get(..7).is_some_and(|prefix| prefix.eq_ignore_ascii_case("http://"))
@@ -66,10 +66,12 @@ pub fn site_identity_key(endpoint: &str) -> String {
         .unwrap_or_else(|| trimmed.to_ascii_lowercase())
 }
 
+/// Fallback deployment ID for a peer that reported none. UUIDv5 over the
+/// canonical endpoint: the ID is persisted in site-replication state and
+/// broadcast to peers, so it must be identical across Rust toolchains
+/// (`DefaultHasher` is not) and across spellings of the same endpoint.
 pub fn deployment_id_for_endpoint(endpoint: &str) -> String {
-    let mut hasher = DefaultHasher::new();
-    endpoint.hash(&mut hasher);
-    format!("{:016x}", hasher.finish())
+    Uuid::new_v5(&Uuid::NAMESPACE_URL, canonical_endpoint(endpoint).as_bytes()).to_string()
 }
 
 pub fn same_identity_endpoint(left: &str, right: &str) -> bool {
@@ -174,6 +176,23 @@ mod tests {
         }
     }
 
+    /// B8 red-light: the fallback deployment ID must be a toolchain-stable
+    /// UUIDv5 over the canonical endpoint — `DefaultHasher` output is not
+    /// guaranteed stable across Rust releases, yet the ID is persisted in
+    /// site-replication state and broadcast to peers.
+    #[test]
+    fn deployment_id_for_endpoint_is_stable_uuid_v5_over_canonical_endpoint() {
+        let endpoint = "https://node-a.example.com:9000";
+        let id = deployment_id_for_endpoint(endpoint);
+        let parsed = uuid::Uuid::parse_str(&id).expect("fallback deployment ID must be a UUID");
+        assert_eq!(parsed.get_version_num(), 5, "fallback deployment ID must be UUIDv5");
+        // Deterministic for the same endpoint and for spelling variants that
+        // share a canonical form; distinct endpoints stay distinct.
+        assert_eq!(id, deployment_id_for_endpoint(endpoint));
+        assert_eq!(id, deployment_id_for_endpoint(" HTTPS://Node-A.Example.Com:9000/ "));
+        assert_ne!(id, deployment_id_for_endpoint("https://node-b.example.com:9000"));
+    }
+
     #[test]
     fn canonical_endpoint_accepts_case_insensitive_scheme() {
         assert_eq!(

rustfs/src/admin/storage_api.rs

@@ -51,7 +51,7 @@ mod ecstore_disk {
 }
 
 mod ecstore_error {
-    pub(crate) use crate::storage::storage_api::ecstore_error::StorageError;
+    pub(crate) use crate::storage::storage_api::ecstore_error::{StorageError, is_err_bucket_not_found};
 }
 
 #[allow(unused_imports)]
@@ -919,6 +919,7 @@ pub(crate) mod contract {
 }
 
 pub(crate) mod error {
+    pub(crate) use super::ecstore_error::is_err_bucket_not_found;
     pub(crate) use super::{Error, StorageError};
 }
 

scripts/run_hotpath_warp_abba.sh

@@ -34,8 +34,8 @@ CONCURRENCY=8
 DURATION="60s"
 ROUNDS=3
 COOLDOWN_SECS=20
-DATASET_SETUP_DURATION="10s"
 HEALTH_TIMEOUT_SECS=180
+DATASET_OBJECTS_PER_WORKER=8
 FAIL_PCT=10
 WARN_PCT=5
 ALLOW_REGRESSION=false
@@ -100,9 +100,6 @@ Benchmark:
   --duration <dur>          warp duration per cell (default 60s).
   --rounds <n>              rounds per cell; must be >= 3 (default 3).
   --cooldown <n>            cooldown seconds between rounds/sizes (default 20).
-  --dataset-setup-duration <dur>
-                            isolated Warp PUT warm-up for get/mixed legs
-                            (default 10s; not included in the measurement).
   --concurrency <n>         warp concurrency (default 8).
   --warp-bin <path>         warp binary (default warp).
 
@@ -173,7 +170,6 @@ while [[ $# -gt 0 ]]; do
     --duration) DURATION="$2"; shift 2 ;;
     --rounds) ROUNDS="$2"; shift 2 ;;
     --cooldown) COOLDOWN_SECS="$2"; shift 2 ;;
-    --dataset-setup-duration) DATASET_SETUP_DURATION="$2"; shift 2 ;;
     --health-timeout) HEALTH_TIMEOUT_SECS="$2"; shift 2 ;;
     --fail-pct) FAIL_PCT="$2"; shift 2 ;;
     --warn-pct) WARN_PCT="$2"; shift 2 ;;
@@ -366,29 +362,18 @@ measure() {
     --duration "$DURATION" --rounds "$ROUNDS" --cooldown-secs "$COOLDOWN_SECS"
     --out-dir "$cell"
   )
-  [[ "$mode" == "put" ]] || args+=(--extra-args "--noclear")
+  if [[ "$mode" != "put" ]]; then
+    # Warp defaults to 2,500 setup objects per round. At 10 MiB that writes
+    # 25 GiB before every 12-second measurement, so the matrix cannot finish
+    # inside the workflow budget. Eight objects per worker keeps preparation
+    # bounded while retaining a multi-object working set for relative A/B.
+    args+=(--extra-args "--objects $((CONCURRENCY * DATASET_OBJECTS_PER_WORKER)) --noclear")
+  fi
   [[ -n "$baseline_csv" ]] && args+=(--baseline-csv "$baseline_csv")
   run "$ENHANCED_BENCH" "${args[@]}" >&2
   echo "$cell"
 }
 
-prepare_dataset() {
-  local leg="$1" workload="$2" mode="$3" size="$4" sync_label="$5" bucket="$6"
-  [[ "$mode" != "put" ]] || return 0
-
-  local setup_cell="$OUT_DIR/$workload/$sync_label/$leg/dataset-setup"
-  local args=(
-    --tool warp --warp-bin "$WARP_BIN" --warp-mode put
-    --endpoint "$ADDRESS" --access-key "$ACCESS_KEY" --secret-key "$SECRET_KEY"
-    --region "$REGION" --bucket "$bucket" --sizes "$size" --concurrency "$CONCURRENCY"
-    --duration "$DATASET_SETUP_DURATION" --rounds 1 --cooldown-secs 0
-    --extra-args "--noclear"
-    --out-dir "$setup_cell"
-  )
-  log "preparing isolated dataset: $sync_label/$workload/$leg bucket=$bucket"
-  run "$ENHANCED_BENCH" "${args[@]}" >&2
-}
-
 write_schedule_header() {
   echo "sync_label,drive_sync,workload,mode,size,leg,phase,binary,out_dir,bucket,dataset_setup" >"$OUT_DIR/abba_schedule.csv"
 }
@@ -399,7 +384,7 @@ append_schedule() {
   phase="$(phase_for_leg "$leg")"
   bin="$(binary_for_leg "$leg")"
   local dataset_setup="none"
-  [[ "$mode" == "put" ]] || dataset_setup="warp-put"
+  [[ "$mode" == "put" ]] || dataset_setup="warp-native-bounded"
   echo "$sync_label,$drive_sync,$workload,$mode,$size,$leg,$phase,$bin,$OUT_DIR/$workload/$sync_label/$leg,$bucket,$dataset_setup" >>"$OUT_DIR/abba_schedule.csv"
 }
 
@@ -481,7 +466,8 @@ dataset_namespace=$DATASET_NAMESPACE
 local_run_data_root=$RUN_DATA_ROOT
 bucket_isolation=per-leg
 bucket_prefix=rustfs-abba-$DATASET_NAMESPACE
-dataset_setup=get-and-mixed-via-warp-put
+dataset_setup=get-and-mixed-via-bounded-warp-native
+dataset_objects=$((CONCURRENCY * DATASET_OBJECTS_PER_WORKER))
 endpoint=$ADDRESS
 warp_version=$("$WARP_BIN" --version 2>/dev/null | head -n1 || echo unknown)
 EOF
@@ -502,7 +488,6 @@ for ds_spec in "${DRIVE_SYNC_MATRIX[@]}"; do
       log "=== $sync_label $workload leg $leg ($(phase_for_leg "$leg")) ==="
       bucket="$(bucket_for_leg "$sync_label" "$workload" "$leg")"
       bring_up "$leg" "$drive_sync" "$workload" "$mode" "$size" "$sync_label" "$bucket"
-      prepare_dataset "$leg" "$workload" "$mode" "$size" "$sync_label" "$bucket"
       append_schedule "$sync_label" "$drive_sync" "$workload" "$mode" "$size" "$leg" "$bucket"
 
       baseline_csv=""

scripts/run_object_batch_bench_enhanced.sh

@@ -672,7 +672,7 @@ extract_report_line() {
   local regex="$1"
   local file="$2"
   awk -v regex="$regex" '
-    /^Report:/ {
+    /^(Report|Operation):/ {
       in_report = 1
       next
     }
@@ -703,9 +703,9 @@ normalize_duration_metric() {
 extract_metrics() {
   local log_file="$1"
 
-  local average_line reqs_line throughput reqps latency req_p90 req_p99 reqps_num
+  local average_line request_line throughput reqps latency req_p90 req_p99 reqps_num
   average_line="$(extract_report_line '^[[:space:]]*[*][[:space:]]+Average:' "$log_file")"
-  reqs_line="$(extract_report_line '^[[:space:]]*[*][[:space:]]+Reqs:' "$log_file")"
+  request_line="$(extract_report_line '^[[:space:]]*[*][[:space:]]+(Reqs:[[:space:]]+)?Avg:' "$log_file")"
 
   if [[ -n "$average_line" ]]; then
     throughput="$(echo "$average_line" | sed -E 's/^.*Average:[[:space:]]*//; s/,[[:space:]]*.*$//')"
@@ -715,20 +715,16 @@ extract_metrics() {
     reqps="$(extract_first '[0-9]+(\.[0-9]+)?[[:space:]]*(obj/s|req/s|ops/s|requests/s)' "$log_file")"
   fi
 
-  if [[ -n "$reqs_line" ]]; then
-    latency="$(echo "$reqs_line" | rg -o 'Avg:[[:space:]]+[0-9]+(\.[0-9]+)?(ms|us|µs|s)' | sed -E 's/^Avg:[[:space:]]+//')"
-    req_p90="$(echo "$reqs_line" | rg -o '90%:[[:space:]]+[0-9]+(\.[0-9]+)?(ms|us|µs|s)' | sed -E 's/^90%:[[:space:]]+//')"
-    req_p99="$(echo "$reqs_line" | rg -o '99%:[[:space:]]+[0-9]+(\.[0-9]+)?(ms|us|µs|s)' | sed -E 's/^99%:[[:space:]]+//')"
+  if [[ -n "$request_line" ]]; then
+    latency="$(echo "$request_line" | rg -o 'Avg:[[:space:]]+[0-9]+(\.[0-9]+)?(ms|us|µs|s)' | sed -E 's/^Avg:[[:space:]]+//')"
+    req_p90="$(echo "$request_line" | rg -o '90%:[[:space:]]+[0-9]+(\.[0-9]+)?(ms|us|µs|s)' | sed -E 's/^90%:[[:space:]]+//')"
+    req_p99="$(echo "$request_line" | rg -o '99%:[[:space:]]+[0-9]+(\.[0-9]+)?(ms|us|µs|s)' | sed -E 's/^99%:[[:space:]]+//')"
   else
     latency="$(rg -o 'Reqs:[[:space:]]+Avg:[[:space:]]+[0-9]+(\.[0-9]+)?(ms|us|µs|s)' "$log_file" | head -n1 | sed -E 's/^Reqs:[[:space:]]+Avg:[[:space:]]+//')"
     req_p90="$(rg -o '90%:[[:space:]]+[0-9]+(\.[0-9]+)?(ms|us|µs|s)' "$log_file" | head -n1 | sed -E 's/^90%:[[:space:]]+//')"
     req_p99="$(rg -o '99%:[[:space:]]+[0-9]+(\.[0-9]+)?(ms|us|µs|s)' "$log_file" | head -n1 | sed -E 's/^99%:[[:space:]]+//')"
   fi
 
-  if [[ -z "$latency" ]]; then
-    latency="$(extract_first '[0-9]+(\.[0-9]+)?[[:space:]]*(ms|us|µs|s)' "$log_file")"
-  fi
-
   throughput="$(trim "${throughput:-N/A}")"
   reqps="$(trim "${reqps:-N/A}")"
   latency="$(trim "${latency:-N/A}")"
@@ -1128,6 +1124,8 @@ run_one_attempt() {
       "--concurrent" "$CONCURRENCY"
       "--duration" "$DURATION"
       "--region" "$REGION"
+      "--no-color"
+      "--analyze.v"
     )
     if [[ "$INSECURE" == "true" ]]; then
       cmd+=("--insecure")
@@ -1212,6 +1210,12 @@ run_one_attempt() {
     req_p99_ms="$(to_ms "$req_p99_human")"
   fi
 
+  if [[ "$DRY_RUN" != "true" && "$TOOL" == "warp" && "$status" == "ok" ]] \
+    && rg -q '^[[:space:]]*(Total[[:space:]]+)?Errors:[[:space:]]+[1-9][0-9]*[.]?([[:space:]]|$)' "$log_file"; then
+    status="failed"
+    exit_code=1
+  fi
+
   if [[ "$DRY_RUN" != "true" && "$status" == "ok" ]]; then
     if [[ "$throughput_bps" == "N/A" && "$reqps" == "N/A" ]]; then
       status="failed"
@@ -1318,10 +1322,21 @@ compare_baseline() {
 
       dr="N/A"; dl="N/A"; dt="N/A"; dp90="N/A"; dp99="N/A"; ne="N/A"; be="N/A"; de="N/A"
       if (br!="N/A" && n_req!="N/A" && br+0!=0) dr=sprintf("%.2f", ((n_req-br)/br)*100)
-      if (bl!="N/A" && n_lat!="N/A" && bl+0!=0) dl=sprintf("%.2f", ((n_lat-bl)/bl)*100)
+      if (bl!="N/A" && n_lat!="N/A") {
+        if (bl+0!=0) dl=sprintf("%.2f", ((n_lat-bl)/bl)*100)
+        else if (n_lat+0==0) dl="0.00"
+      }
       if (bt!="N/A" && n_thr!="N/A" && bt+0!=0) dt=sprintf("%.2f", ((n_thr-bt)/bt)*100)
-      if (bp90!="N/A" && n_p90!="N/A" && bp90+0!=0) dp90=sprintf("%.2f", ((n_p90-bp90)/bp90)*100)
-      if (bp99!="N/A" && n_p99!="N/A" && bp99+0!=0) dp99=sprintf("%.2f", ((n_p99-bp99)/bp99)*100)
+      # Warp v1 rounds sub-millisecond latency to 0s. Two zero readings are
+      # the same below-resolution bucket; a nonzero candidate remains invalid.
+      if (bp90!="N/A" && n_p90!="N/A") {
+        if (bp90+0!=0) dp90=sprintf("%.2f", ((n_p90-bp90)/bp90)*100)
+        else if (n_p90+0==0) dp90="0.00"
+      }
+      if (bp99!="N/A" && n_p99!="N/A") {
+        if (bp99+0!=0) dp99=sprintf("%.2f", ((n_p99-bp99)/bp99)*100)
+        else if (n_p99+0==0) dp99="0.00"
+      }
       if (n_ok!="N/A" && n_fail!="N/A" && n_ok+n_fail>0) ne=sprintf("%.2f", (n_fail/(n_ok+n_fail))*100)
       if (bok!="N/A" && bfail!="N/A" && bok+bfail>0) be=sprintf("%.2f", (bfail/(bok+bfail))*100)
       if (ne!="N/A" && be!="N/A") de=sprintf("%.2f", ne-be)

scripts/test_hotpath_warp_abba.sh

@@ -58,8 +58,13 @@ rg -qx 'evidence_mode=dry-run' "$OUT_DIR/manifest.env"
 rg -qx 'formal_evidence=false' "$OUT_DIR/manifest.env"
 rg -qx 'performance_conclusion=not_measured_dry_run' "$OUT_DIR/manifest.env"
 rg -qx 'bucket_isolation=per-leg' "$OUT_DIR/manifest.env"
-rg -qx 'dataset_setup=get-and-mixed-via-warp-put' "$OUT_DIR/manifest.env"
-[[ "$(rg -c -- '--extra-args --noclear' "$TRACE_FILE")" == "64" ]]
+rg -qx 'dataset_setup=get-and-mixed-via-bounded-warp-native' "$OUT_DIR/manifest.env"
+rg -qx 'dataset_objects=64' "$OUT_DIR/manifest.env"
+[[ "$(rg -c -- '--extra-args --objects\\ 64\\ --noclear' "$TRACE_FILE")" == "32" ]]
+if rg -q -- 'dataset-setup' "$TRACE_FILE"; then
+  echo "unexpected redundant dataset setup command" >&2
+  exit 1
+fi
 ! rg -q -- 'rustfs-bench' "$TRACE_FILE"
 
 if "$RUNNER" \

scripts/test_object_batch_bench_enhanced.sh

@@ -74,13 +74,28 @@ FAKE_WARP="${TMP_DIR}/fake-warp"
 cat >"$FAKE_WARP" <<'EOF'
 #!/usr/bin/env bash
 set -euo pipefail
+[[ " $* " == *" --analyze.v "* ]]
+[[ " $* " == *" --no-color "* ]]
+if [[ "${FAKE_WARP_ZERO_LATENCY:-0}" == "1" ]]; then
+  cat <<'LOG'
+Operation: GET. Concurrency: 8. Ran: 7s
+Requests considered: 1000:
+ * Average: 160.00 MiB/s, 40960.00 obj/s
+ * Avg: 0s, 50%: 0s, 90%: 0s, 99%: 0s, Fastest: 0s, Slowest: 1ms, StdDev: 0s
+LOG
+  exit 0
+fi
 cat <<'LOG'
  -       PUT Average: 161 Obj/s, 5.0MiB/s; Current 161 Obj/s, 5.0MiB/s.
-Report: GET. Concurrency: 64. Ran: 7s
+Operation: GET. Concurrency: 64. Ran: 7s
+Requests considered: 1000:
  * Average: 653.90 MiB/s, 20925.58 obj/s
- * Reqs: Avg: 3.5ms, 50%: 2.0ms, 90%: 3.6ms, 99%: 24.1ms, Fastest: 0.2ms, Slowest: 607.7ms, StdDev: 20.6ms
+ * Avg: 3.5ms, 50%: 2.0ms, 90%: 3.6ms, 99%: 24.1ms, Fastest: 0.2ms, Slowest: 607.7ms, StdDev: 20.6ms
 Throughput, split into 7 x 1s:
 LOG
+if [[ "${FAKE_WARP_ERRORS:-0}" == "1" ]]; then
+  echo 'Total Errors: 1.'
+fi
 EOF
 chmod +x "$FAKE_WARP"
 
@@ -103,6 +118,32 @@ chmod +x "$FAKE_WARP"
 
 rg -q '^32767B,warp,1,1,128,ok,0,[^,]+,[^,]+,653.90 MiB/s,685663846.400000,20925.58,3.5 ms,3.500000,[^,]+,3.6 ms,3.600000,24.1 ms,24.100000$' "${TMP_DIR}/fake-warp-run/round_results.csv"
 
+cat >"${TMP_DIR}/warp-no-details.log" <<'EOF'
+warp: Starting benchmark in 3s...
+Operation: PUT. Concurrency: 8
+ * Average: 2.76 MiB/s, 707.03 obj/s
+EOF
+"$RUNNER" --extract-metrics-from-log "${TMP_DIR}/warp-no-details.log" >"${TMP_DIR}/warp-no-details.csv"
+rg -qx '2.76 MiB/s,2894069.760000,707.03,N/A,N/A,N/A,N/A,N/A,N/A' "${TMP_DIR}/warp-no-details.csv"
+
+if FAKE_WARP_ERRORS=1 "$RUNNER" \
+  --tool warp \
+  --endpoint http://127.0.0.1:9000 \
+  --access-key test-access \
+  --secret-key test-secret \
+  --sizes 32767B \
+  --rounds 1 \
+  --retry-per-round 1 \
+  --retry-sleep-secs 1 \
+  --cooldown-secs 0 \
+  --duration 1s \
+  --out-dir "${TMP_DIR}/fake-warp-errors" \
+  --warp-bin "$FAKE_WARP" >/dev/null 2>&1; then
+  echo "expected Warp request errors to fail the benchmark" >&2
+  exit 1
+fi
+rg -q ',failed,1,' "${TMP_DIR}/fake-warp-errors/round_results.csv"
+
 "$RUNNER" \
   --tool warp \
   --endpoint http://127.0.0.1:9000 \
@@ -132,4 +173,28 @@ awk -F',' '
   END { exit found ? 0 : 1 }
 ' "${TMP_DIR}/fake-warp-candidate/baseline_compare.csv"
 
+for leg in baseline candidate; do
+  zero_args=(
+    --tool warp
+    --endpoint http://127.0.0.1:9000
+    --access-key test-access
+    --secret-key test-secret
+    --sizes 4KiB
+    --rounds 1
+    --retry-per-round 1
+    --cooldown-secs 0
+    --duration 1s
+    --out-dir "${TMP_DIR}/fake-warp-zero-${leg}"
+    --warp-bin "$FAKE_WARP"
+  )
+  if [[ "$leg" == "candidate" ]]; then
+    zero_args+=(--baseline-csv "${TMP_DIR}/fake-warp-zero-baseline/median_summary.csv")
+  fi
+  FAKE_WARP_ZERO_LATENCY=1 "$RUNNER" "${zero_args[@]}" >/dev/null 2>&1
+done
+
+"${SCRIPT_DIR}/hotpath_warp_ab_gate.sh" \
+  --compare-csv "${TMP_DIR}/fake-warp-zero-candidate/baseline_compare.csv" \
+  --require-tail-error >/dev/null
+
 echo "object batch benchmark enhanced tests passed"