fix(scanner): defer usage publication during pool recovery (#6333)

* fix(scanner): defer usage publication during pool recovery

* fix(scanner): preserve metrics when publication is deferred

* fix(scanner): route test types through storage boundary

* fix(scanner): keep cache floor deferred during movement

crates/ecstore/src/store/mod.rs

@@ -343,6 +343,23 @@ impl ECStore {
         let (decommission, rebalance) = tokio::join!(self.is_decommission_running(), self.is_rebalance_started());
         decommission || rebalance
     }
+
+    /// Returns whether scanner metadata may still be hidden by a local
+    /// data-movement state. Terminal failed/canceled decommission entries
+    /// remain suspended until an operator clears or retries them, so they are
+    /// a publication barrier even after the worker has stopped.
+    pub async fn scanner_data_usage_publication_blocked(&self) -> bool {
+        if self.scanner_data_movement_active().await {
+            return true;
+        }
+
+        let pool_meta = self.pool_meta.read().await;
+        pool_meta.pools.iter().any(|pool| {
+            pool.decommission
+                .as_ref()
+                .is_some_and(|info| !info.queued && (info.failed || info.canceled))
+        })
+    }
 }
 
 // impl Clone for ECStore {
@@ -875,6 +892,7 @@ impl crate::storage_api_contracts::admin::StorageAdminApi for ECStore {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::core::pools::{PoolDecommissionInfo, PoolStatus};
     use crate::layout::endpoints::{Endpoints, PoolEndpoints, SetupType};
     use crate::runtime::global::reset_local_disk_test_state;
     use crate::runtime::sources::{clear_local_disk_id_map_for_test, local_disk_path_by_id};
@@ -911,6 +929,72 @@ mod tests {
         })
     }
 
+    #[tokio::test]
+    async fn scanner_data_usage_publication_blocks_active_and_unqueued_terminal_decommission() {
+        let store = build_store_with_ctx(Arc::new(InstanceContext::new()));
+        let cases = [
+            (
+                "active",
+                PoolDecommissionInfo {
+                    start_time: Some(OffsetDateTime::now_utc()),
+                    ..Default::default()
+                },
+                true,
+            ),
+            (
+                "failed",
+                PoolDecommissionInfo {
+                    failed: true,
+                    ..Default::default()
+                },
+                true,
+            ),
+            (
+                "canceled",
+                PoolDecommissionInfo {
+                    canceled: true,
+                    ..Default::default()
+                },
+                true,
+            ),
+            (
+                "queued_failed",
+                PoolDecommissionInfo {
+                    failed: true,
+                    queued: true,
+                    ..Default::default()
+                },
+                false,
+            ),
+            (
+                "complete",
+                PoolDecommissionInfo {
+                    complete: true,
+                    ..Default::default()
+                },
+                false,
+            ),
+            ("idle", PoolDecommissionInfo::default(), false),
+        ];
+
+        for (name, decommission, expected) in cases {
+            *store.pool_meta.write().await = PoolMeta {
+                pools: vec![PoolStatus {
+                    id: 0,
+                    cmd_line: format!("scanner-publication-{name}"),
+                    last_update: OffsetDateTime::now_utc(),
+                    decommission: Some(decommission),
+                }],
+                ..Default::default()
+            };
+            assert_eq!(
+                store.scanner_data_usage_publication_blocked().await,
+                expected,
+                "unexpected scanner publication barrier state for {name}"
+            );
+        }
+    }
+
     // The object graph is the isolation carrier: two ECStore instances holding
     // distinct contexts report independent erasure state through their real
     // `&self` accessors — no cross-contamination.

crates/scanner/src/scanner.rs

@@ -1081,18 +1081,6 @@ async fn run_data_scanner_cycle(
         }
     };
     let (sender, receiver) = mpsc::channel::<DataUsageInfo>(1);
-    let storeapi_clone = storeapi.clone();
-    let ctx_clone = ctx.clone();
-    let mut usage_persist_task = AbortOnDropHandle::new(tokio::spawn(async move {
-        store_data_usage_in_backend_with_outcome_for_epoch_and_baseline(
-            ctx_clone,
-            storeapi_clone,
-            receiver,
-            Some(leader_epoch),
-            Some(usage_persist_baseline),
-        )
-        .await
-    }));
 
     let done_cycle = Metrics::time(Metric::ScanCycle);
     let cycle_budget = ScannerCycleBudget::new(ctx, cycle_budget_config);
@@ -1107,47 +1095,78 @@ async fn run_data_scanner_cycle(
             scan_mode,
         )
         .await;
+    let publication_defer_reason = match &scan_result {
+        Ok(result) => final_data_usage_publication_defer_reason(storeapi.as_ref(), result.status).await,
+        Err(_) => Some(ScannerCycleDeferReason::ActivityBaselineUnavailable),
+    };
     let budget_elapsed = cycle_budget.budget_elapsed() && !ctx.is_cancelled();
-    let usage_persist_outcome = match wait_for_data_usage_persist_task(ctx, &mut usage_persist_task, usage_persist_timeout).await
-    {
-        DataUsagePersistTaskResult::Completed(outcome) => outcome,
-        DataUsagePersistTaskResult::JoinFailed(err) => {
-            error!(
-                target: "rustfs::scanner",
-                event = EVENT_SCANNER_PERSIST_STATE,
-                component = LOG_COMPONENT_SCANNER,
-                subsystem = LOG_SUBSYSTEM_RUNTIME,
-                cycle = cycle_info.current,
-                state = "usage_persist_task_failed",
-                error = %err,
-                "Scanner data usage persistence task failed"
-            );
-            DataUsagePersistOutcome::Failed
+    let usage_persist_outcome = match publication_defer_reason {
+        Some(reason) => {
+            drop(receiver);
+            DataUsagePersistOutcome::Deferred(reason)
         }
-        DataUsagePersistTaskResult::Cancelled => {
-            debug!(
-                target: "rustfs::scanner",
-                event = EVENT_SCANNER_PERSIST_STATE,
-                component = LOG_COMPONENT_SCANNER,
-                subsystem = LOG_SUBSYSTEM_RUNTIME,
-                cycle = cycle_info.current,
-                state = "usage_persist_task_cancelled",
-                "Scanner data usage persistence task cancelled"
-            );
-            DataUsagePersistOutcome::Failed
-        }
-        DataUsagePersistTaskResult::TimedOut => {
-            error!(
-                target: "rustfs::scanner",
-                event = EVENT_SCANNER_PERSIST_STATE,
-                component = LOG_COMPONENT_SCANNER,
-                subsystem = LOG_SUBSYSTEM_RUNTIME,
-                cycle = cycle_info.current,
-                timeout = ?usage_persist_timeout,
-                state = "usage_persist_task_timed_out",
-                "Scanner data usage persistence task timed out"
-            );
-            DataUsagePersistOutcome::Failed
+        None => {
+            // ScannerIO emits its complete or observational update only after
+            // all set workers finish. Persist after the final activity fence;
+            // this also avoids blocking the scanner on a denied publication.
+            let storeapi_clone = storeapi.clone();
+            let ctx_clone = ctx.clone();
+            let route_probe_store = storeapi.clone();
+            let mut usage_persist_task = AbortOnDropHandle::new(tokio::spawn(async move {
+                store_data_usage_in_backend_with_outcome_for_epoch_and_baseline_and_route_probe(
+                    ctx_clone,
+                    storeapi_clone,
+                    receiver,
+                    Some(leader_epoch),
+                    Some(usage_persist_baseline),
+                    move || {
+                        let storeapi = route_probe_store.clone();
+                        async move { storeapi.scanner_data_usage_publication_blocked().await }
+                    },
+                )
+                .await
+            }));
+            match wait_for_data_usage_persist_task(ctx, &mut usage_persist_task, usage_persist_timeout).await {
+                DataUsagePersistTaskResult::Completed(outcome) => outcome,
+                DataUsagePersistTaskResult::JoinFailed(err) => {
+                    error!(
+                        target: "rustfs::scanner",
+                        event = EVENT_SCANNER_PERSIST_STATE,
+                        component = LOG_COMPONENT_SCANNER,
+                        subsystem = LOG_SUBSYSTEM_RUNTIME,
+                        cycle = cycle_info.current,
+                        state = "usage_persist_task_failed",
+                        error = %err,
+                        "Scanner data usage persistence task failed"
+                    );
+                    DataUsagePersistOutcome::Failed
+                }
+                DataUsagePersistTaskResult::Cancelled => {
+                    debug!(
+                        target: "rustfs::scanner",
+                        event = EVENT_SCANNER_PERSIST_STATE,
+                        component = LOG_COMPONENT_SCANNER,
+                        subsystem = LOG_SUBSYSTEM_RUNTIME,
+                        cycle = cycle_info.current,
+                        state = "usage_persist_task_cancelled",
+                        "Scanner data usage persistence task cancelled"
+                    );
+                    DataUsagePersistOutcome::Failed
+                }
+                DataUsagePersistTaskResult::TimedOut => {
+                    error!(
+                        target: "rustfs::scanner",
+                        event = EVENT_SCANNER_PERSIST_STATE,
+                        component = LOG_COMPONENT_SCANNER,
+                        subsystem = LOG_SUBSYSTEM_RUNTIME,
+                        cycle = cycle_info.current,
+                        timeout = ?usage_persist_timeout,
+                        state = "usage_persist_task_timed_out",
+                        "Scanner data usage persistence task timed out"
+                    );
+                    DataUsagePersistOutcome::Failed
+                }
+            }
         }
     };
     let unresolved_heal_work = global_metrics().current_scan_cycle_has_unresolved_heal_work();
@@ -1191,33 +1210,51 @@ async fn run_data_scanner_cycle(
         mark_scan_cycle_idle(cycle_info, &mut cycle_metrics_guard).await;
         return ScannerCycleOutcome::Failed;
     }
-    if let Some(required_cycle) = scan_cycle_result.required_cycle_floor() {
-        warn!(
-            target: "rustfs::scanner",
-            event = EVENT_SCANNER_CYCLE_STATE,
-            component = LOG_COMPONENT_SCANNER,
-            subsystem = LOG_SUBSYSTEM_RUNTIME,
-            cycle = cycle_info.current,
-            required_cycle,
-            state = "cache_cycle_ahead",
-            "Scanner cycle is recovering to a newer durable cache generation"
-        );
-        emit_scan_cycle_partial_with_source(cycle_start.elapsed(), ScanCyclePartialReason::Unknown, None);
-        return if persist_required_scanner_cycle_floor(
-            ctx,
-            storeapi.clone(),
-            cycle_info,
-            cycle_revision,
-            leader_epoch,
-            required_cycle,
-            &mut cycle_metrics_guard,
-        )
-        .await
-        {
-            ScannerCycleOutcome::Partial
-        } else {
-            ScannerCycleOutcome::Failed
-        };
+    match scanner_cycle_pre_commit_outcome(scan_cycle_result.required_cycle_floor(), &usage_persist_outcome) {
+        Some(ScannerCyclePreCommitOutcome::RecoverCacheCycle(required_cycle)) => {
+            warn!(
+                target: "rustfs::scanner",
+                event = EVENT_SCANNER_CYCLE_STATE,
+                component = LOG_COMPONENT_SCANNER,
+                subsystem = LOG_SUBSYSTEM_RUNTIME,
+                cycle = cycle_info.current,
+                required_cycle,
+                state = "cache_cycle_ahead",
+                "Scanner cycle is recovering to a newer durable cache generation"
+            );
+            emit_scan_cycle_partial_with_source(cycle_start.elapsed(), ScanCyclePartialReason::Unknown, None);
+            return if persist_required_scanner_cycle_floor(
+                ctx,
+                storeapi.clone(),
+                cycle_info,
+                cycle_revision,
+                leader_epoch,
+                required_cycle,
+                &mut cycle_metrics_guard,
+            )
+            .await
+            {
+                ScannerCycleOutcome::Partial
+            } else {
+                ScannerCycleOutcome::Failed
+            };
+        }
+        Some(ScannerCyclePreCommitOutcome::Deferred(reason)) => {
+            info!(
+                target: "rustfs::scanner",
+                event = EVENT_SCANNER_CYCLE_STATE,
+                component = LOG_COMPONENT_SCANNER,
+                subsystem = LOG_SUBSYSTEM_RUNTIME,
+                cycle = cycle_info.current,
+                reason = reason.as_str(),
+                state = "deferred",
+                "Scanner cycle deferred before data usage publication"
+            );
+            emit_scan_cycle_deferred(cycle_start.elapsed());
+            mark_scan_cycle_idle(cycle_info, &mut cycle_metrics_guard).await;
+            return ScannerCycleOutcome::Deferred(reason);
+        }
+        None => {}
     }
     if usage_persist_outcome == DataUsagePersistOutcome::Failed {
         error!(
@@ -2000,6 +2037,56 @@ impl Drop for ScannerScanModeGuard {
     }
 }
 
+async fn final_data_usage_publication_defer_reason(
+    storeapi: &ECStore,
+    status: ScannerCycleStatus,
+) -> Option<ScannerCycleDeferReason> {
+    match status {
+        ScannerCycleStatus::Complete | ScannerCycleStatus::Superseded => {
+            if storeapi.scanner_data_usage_publication_blocked().await {
+                return Some(ScannerCycleDeferReason::DataMovement);
+            }
+            if status == ScannerCycleStatus::Complete {
+                let distributed = storeapi.setup_is_dist_erasure().await;
+                match probe_scanner_activity(storeapi, distributed).await {
+                    Ok(snapshot) if scanner_activity_allows_usage_publication(&snapshot) => None,
+                    Ok(_) => Some(ScannerCycleDeferReason::DataMovement),
+                    Err(_) => Some(ScannerCycleDeferReason::ActivityBaselineUnavailable),
+                }
+            } else {
+                // A superseded cycle is explicitly observational and cannot
+                // replace the authoritative snapshot. It may still be
+                // persisted as a convergence baseline for the next cycle.
+                None
+            }
+        }
+        ScannerCycleStatus::Deferred(reason) => Some(reason),
+        // Incomplete cycles do not publish a usage snapshot. Keep the
+        // decision permissive so existing partial-cycle handling remains
+        // unchanged if a future scanner path emits a bookkeeping update.
+        ScannerCycleStatus::Incomplete => None,
+    }
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+enum ScannerCyclePreCommitOutcome {
+    RecoverCacheCycle(u64),
+    Deferred(ScannerCycleDeferReason),
+}
+
+fn scanner_cycle_pre_commit_outcome(
+    required_cycle_floor: Option<u64>,
+    usage_persist_outcome: &DataUsagePersistOutcome,
+) -> Option<ScannerCyclePreCommitOutcome> {
+    // Keep the publication barrier fail-closed: `.bloomcycle.bin` uses the
+    // same routed writer and its floor must remain pending while data movement
+    // hides the source pool.
+    match usage_persist_outcome {
+        DataUsagePersistOutcome::Deferred(reason) => Some(ScannerCyclePreCommitOutcome::Deferred(*reason)),
+        _ => required_cycle_floor.map(ScannerCyclePreCommitOutcome::RecoverCacheCycle),
+    }
+}
+
 fn scanner_cycle_completion_outcome(
     scan_status: ScannerCycleStatus,
     usage_persist_outcome: DataUsagePersistOutcome,
@@ -2007,6 +2094,7 @@ fn scanner_cycle_completion_outcome(
     has_failed_dirty_usage: bool,
 ) -> ScannerCycleOutcome {
     match (scan_status, usage_persist_outcome) {
+        (_, DataUsagePersistOutcome::Deferred(reason)) => ScannerCycleOutcome::Deferred(reason),
         (_, DataUsagePersistOutcome::Failed) => ScannerCycleOutcome::Failed,
         (ScannerCycleStatus::Deferred(reason), DataUsagePersistOutcome::NoUpdate)
             if !has_dirty_usage && !has_failed_dirty_usage =>

crates/scanner/src/scanner/tests.rs

@@ -153,6 +153,7 @@ struct MemoryConfigStore {
     objects: Mutex<HashMap<String, Vec<u8>>>,
     revisions: Mutex<HashMap<String, u64>>,
     fail_put_number: Mutex<HashMap<String, usize>>,
+    object_not_found_put_number: Mutex<HashMap<String, usize>>,
     error_after_commit_put_number: Mutex<HashMap<String, usize>>,
     interleaving_puts: Mutex<HashMap<String, (usize, Vec<u8>)>>,
     cancel_after_interleaving_puts: Mutex<HashMap<String, CancellationToken>>,
@@ -224,6 +225,9 @@ impl crate::storage_api::scanner_io::ObjectIO for MemoryConfigStore {
         if self.fail_put_number.lock().await.get(&key) == Some(&put_count) {
             return Err(EcstoreError::other("injected put failure"));
         }
+        if self.object_not_found_put_number.lock().await.get(&key) == Some(&put_count) {
+            return Err(EcstoreError::ObjectNotFound(bucket.to_string(), object.to_string()));
+        }
 
         let interleaving_data = {
             let mut interleaving_puts = self.interleaving_puts.lock().await;
@@ -1431,6 +1435,170 @@ async fn test_store_data_usage_in_backend_preserves_newer_snapshot() {
     assert_eq!(outcome, DataUsagePersistOutcome::Current);
 }
 
+#[tokio::test]
+async fn test_usage_save_object_not_found_defers_only_with_a_fresh_route_barrier() {
+    for (route_blocked, expected) in [
+        (true, DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement)),
+        (false, DataUsagePersistOutcome::Failed),
+    ] {
+        let store = Arc::new(MemoryConfigStore::default());
+        let key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_OBJ_NAME_PATH.as_str());
+        let baseline = complete_usage_with_bucket_count(Some(std::time::SystemTime::UNIX_EPOCH + Duration::from_secs(10)), 1);
+        let baseline_data = serde_json::to_vec(&baseline).expect("baseline usage snapshot should encode");
+        store.objects.lock().await.insert(key.clone(), baseline_data.clone());
+        store.revisions.lock().await.insert(key.clone(), 1);
+        store.object_not_found_put_number.lock().await.insert(key.clone(), 1);
+
+        let (sender, receiver) = mpsc::channel(1);
+        sender
+            .send(complete_usage_with_bucket_count(
+                Some(std::time::SystemTime::UNIX_EPOCH + Duration::from_secs(20)),
+                2,
+            ))
+            .await
+            .expect("new usage snapshot should enqueue");
+        drop(sender);
+        let probe_calls = Arc::new(std::sync::atomic::AtomicUsize::new(0));
+        let route_probe_calls = probe_calls.clone();
+
+        let outcome = store_data_usage_in_backend_with_outcome_for_epoch_and_baseline_and_route_probe(
+            CancellationToken::new(),
+            store.clone(),
+            receiver,
+            None,
+            Some(DataUsagePersistBaseline {
+                data: Some(Bytes::from(baseline_data.clone())),
+                revision: DataUsageCacheRevision::Etag("memory-1".to_string()),
+            }),
+            move || {
+                let probe_calls = route_probe_calls.clone();
+                async move {
+                    let call = probe_calls.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
+                    route_blocked && call > 1
+                }
+            },
+        )
+        .await;
+
+        assert_eq!(outcome, expected);
+        assert_eq!(
+            probe_calls.load(std::sync::atomic::Ordering::SeqCst),
+            3,
+            "ObjectNotFound must be followed by a fresh route-barrier probe"
+        );
+        assert_eq!(
+            store.objects.lock().await.get(&key),
+            Some(&baseline_data),
+            "a route failure must not replace the authoritative baseline"
+        );
+    }
+}
+
+#[tokio::test]
+async fn test_usage_save_route_barrier_prevents_missing_snapshot_creation() {
+    for observational in [false, true] {
+        let store = Arc::new(MemoryConfigStore::default());
+        let target_path = if observational {
+            DATA_USAGE_OBSERVED_OBJ_NAME_PATH.as_str()
+        } else {
+            DATA_USAGE_OBJ_NAME_PATH.as_str()
+        };
+        let target_key = memory_config_key(RUSTFS_META_BUCKET, target_path);
+        let mut incoming = complete_usage_with_bucket_count(Some(std::time::SystemTime::UNIX_EPOCH + Duration::from_secs(20)), 1);
+        incoming.usage_snapshot_converged = Some(!observational);
+        let (sender, receiver) = mpsc::channel(1);
+        sender.send(incoming).await.expect("usage snapshot should enqueue");
+        drop(sender);
+
+        let outcome = store_data_usage_in_backend_with_outcome_for_epoch_and_baseline_and_route_probe(
+            CancellationToken::new(),
+            store.clone(),
+            receiver,
+            None,
+            Some(DataUsagePersistBaseline {
+                data: None,
+                revision: DataUsageCacheRevision::Missing,
+            }),
+            || async { true },
+        )
+        .await;
+
+        assert_eq!(outcome, DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement));
+        assert!(!store.objects.lock().await.contains_key(&target_key));
+        assert_eq!(
+            store.put_counts.lock().await.get(&target_key),
+            None,
+            "the final pool-state fence must run before the first PUT"
+        );
+    }
+}
+
+#[tokio::test]
+#[serial]
+async fn test_usage_route_barrier_precedes_durable_reconciliation() {
+    let store = Arc::new(MemoryConfigStore::default());
+    let key = memory_config_key(RUSTFS_META_BUCKET, DATA_USAGE_OBJ_NAME_PATH.as_str());
+    let snapshot = complete_usage_with_bucket_count(Some(std::time::SystemTime::UNIX_EPOCH + Duration::from_secs(20)), 1);
+    let snapshot_data = serde_json::to_vec(&snapshot).expect("usage snapshot should encode");
+    let (sender, receiver) = mpsc::channel(1);
+    sender.send(snapshot).await.expect("usage snapshot should enqueue");
+    drop(sender);
+
+    let outcome = store_data_usage_in_backend_with_outcome_for_epoch_and_baseline_and_route_probe(
+        CancellationToken::new(),
+        store.clone(),
+        receiver,
+        None,
+        Some(DataUsagePersistBaseline {
+            data: Some(Bytes::from(snapshot_data)),
+            revision: DataUsageCacheRevision::Etag("memory-1".to_string()),
+        }),
+        || async { true },
+    )
+    .await;
+
+    assert_eq!(outcome, DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement));
+    assert_eq!(store.put_counts.lock().await.get(&key), None);
+}
+
+#[tokio::test]
+#[serial]
+async fn test_deferred_usage_save_keeps_last_real_save_metric() {
+    let metrics = global_metrics();
+    metrics.record_scanner_usage_save_result(ScannerUsageSaveResult::Success);
+    let before = metrics.report().await.usage_freshness;
+
+    let store = Arc::new(MemoryConfigStore::default());
+    let (sender, receiver) = mpsc::channel(1);
+    sender
+        .send(complete_usage_with_bucket_count(
+            Some(std::time::SystemTime::UNIX_EPOCH + Duration::from_secs(20)),
+            1,
+        ))
+        .await
+        .expect("usage snapshot should enqueue");
+    drop(sender);
+
+    let outcome = store_data_usage_in_backend_with_outcome_for_epoch_and_baseline_and_route_probe(
+        CancellationToken::new(),
+        store,
+        receiver,
+        None,
+        Some(DataUsagePersistBaseline {
+            data: None,
+            revision: DataUsageCacheRevision::Missing,
+        }),
+        || async { true },
+    )
+    .await;
+
+    assert_eq!(outcome, DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement));
+    let after = metrics.report().await.usage_freshness;
+    assert_eq!(after.last_usage_save_result, before.last_usage_save_result);
+    assert_eq!(after.last_usage_save_result_code, before.last_usage_save_result_code);
+    assert_eq!(after.last_usage_save_unix_secs, before.last_usage_save_unix_secs);
+}
+
 #[tokio::test]
 async fn test_store_data_usage_in_backend_fences_interleaving_newer_writer() {
     let store = Arc::new(MemoryConfigStore::default());
@@ -2325,6 +2493,15 @@ async fn test_store_data_usage_in_backend_reports_missing_snapshot() {
 
 #[test]
 fn test_scanner_cycle_completion_prioritizes_persist_failure() {
+    assert_eq!(
+        scanner_cycle_completion_outcome(
+            ScannerCycleStatus::Complete,
+            DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement),
+            true,
+            false,
+        ),
+        ScannerCycleOutcome::Deferred(ScannerCycleDeferReason::DataMovement)
+    );
     assert_eq!(
         scanner_cycle_completion_outcome(
             ScannerCycleStatus::Deferred(ScannerCycleDeferReason::ActivityBaselineUnavailable),
@@ -2421,6 +2598,33 @@ fn test_scanner_cycle_completion_prioritizes_persist_failure() {
     );
 }
 
+#[test]
+fn scanner_cycle_cache_floor_stays_pending_during_deferred_usage_publication() {
+    for reason in [
+        ScannerCycleDeferReason::DataMovement,
+        ScannerCycleDeferReason::ActivityBaselineUnavailable,
+    ] {
+        let deferred = DataUsagePersistOutcome::Deferred(reason);
+        assert_eq!(
+            scanner_cycle_pre_commit_outcome(Some(19), &deferred),
+            Some(ScannerCyclePreCommitOutcome::Deferred(reason)),
+            "a blocked publication must not persist the routed scanner cycle floor"
+        );
+        assert_eq!(
+            scanner_cycle_pre_commit_outcome(None, &deferred),
+            Some(ScannerCyclePreCommitOutcome::Deferred(reason))
+        );
+    }
+    assert_eq!(
+        scanner_cycle_pre_commit_outcome(Some(19), &DataUsagePersistOutcome::Saved),
+        Some(ScannerCyclePreCommitOutcome::RecoverCacheCycle(19))
+    );
+    assert_eq!(
+        scanner_cycle_pre_commit_outcome(Some(19), &DataUsagePersistOutcome::Failed),
+        Some(ScannerCyclePreCommitOutcome::RecoverCacheCycle(19))
+    );
+}
+
 #[test]
 #[serial]
 fn finalizing_a_saved_cycle_acknowledges_its_exact_dirty_snapshot() {
@@ -2448,6 +2652,23 @@ fn finalizing_a_saved_cycle_acknowledges_its_exact_dirty_snapshot() {
     assert!(!crate::scanner_io::dirty_usage_buckets_pending());
 }
 
+#[test]
+#[serial]
+fn finalizing_a_deferred_usage_save_keeps_dirty_work_pending() {
+    crate::scanner_io::clear_dirty_usage_bucket("photos");
+    crate::scanner_io::record_dirty_usage_bucket("photos");
+    let dirty_snapshot = crate::scanner_io::dirty_usage_buckets_for_tests();
+    let deferred = crate::scanner_io::ScannerCycleResult::new(ScannerCycleStatus::Complete, Some(dirty_snapshot));
+
+    let (outcome, _, acknowledgements) =
+        finalize_scanner_cycle_result(deferred, DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement));
+
+    assert_eq!(outcome, ScannerCycleOutcome::Deferred(ScannerCycleDeferReason::DataMovement));
+    assert!(acknowledgements.is_empty());
+    assert!(crate::scanner_io::dirty_usage_buckets_pending());
+    crate::scanner_io::clear_dirty_usage_bucket("photos");
+}
+
 #[tokio::test]
 async fn scanner_cycle_keeps_remote_pending_acknowledgement() {
     let pending = remote_dirty_usage_acknowledgement_pending(7, 1, std::future::ready(Ok::<bool, std::io::Error>(true))).await;

crates/scanner/src/scanner/usage_store.rs

@@ -22,6 +22,10 @@ pub(super) enum DataUsagePersistOutcome {
     AlreadyDurable,
     PriorCycleDurable,
     Saved,
+    /// The metadata route is temporarily unavailable (for example while a
+    /// terminal decommission state keeps the source pool suspended).  The
+    /// caller must retry without acknowledging dirty usage.
+    Deferred(ScannerCycleDeferReason),
     Failed,
 }
 
@@ -92,10 +96,33 @@ pub(super) async fn store_data_usage_in_backend_with_outcome_for_epoch(
 pub(super) async fn store_data_usage_in_backend_with_outcome_for_epoch_and_baseline(
     ctx: CancellationToken,
     storeapi: Arc<impl ScannerObjectIO + ScannerConfigObjectDelete>,
-    mut receiver: mpsc::Receiver<DataUsageInfo>,
+    receiver: mpsc::Receiver<DataUsageInfo>,
     leader_epoch: Option<u64>,
     initial_baseline: Option<DataUsagePersistBaseline>,
 ) -> DataUsagePersistOutcome {
+    store_data_usage_in_backend_with_outcome_for_epoch_and_baseline_and_route_probe(
+        ctx,
+        storeapi,
+        receiver,
+        leader_epoch,
+        initial_baseline,
+        || async { false },
+    )
+    .await
+}
+
+pub(super) async fn store_data_usage_in_backend_with_outcome_for_epoch_and_baseline_and_route_probe<F, Fut>(
+    ctx: CancellationToken,
+    storeapi: Arc<impl ScannerObjectIO + ScannerConfigObjectDelete>,
+    mut receiver: mpsc::Receiver<DataUsageInfo>,
+    leader_epoch: Option<u64>,
+    initial_baseline: Option<DataUsagePersistBaseline>,
+    route_probe: F,
+) -> DataUsagePersistOutcome
+where
+    F: Fn() -> Fut + Send + Sync,
+    Fut: Future<Output = bool> + Send,
+{
     let mut outcome = DataUsagePersistOutcome::NoUpdate;
     let mut next_baseline = initial_baseline;
 
@@ -113,6 +140,19 @@ pub(super) async fn store_data_usage_in_backend_with_outcome_for_epoch_and_basel
         } else {
             DATA_USAGE_OBJ_NAME_PATH.as_str()
         };
+        if route_probe().await {
+            debug!(
+                target: "rustfs::scanner",
+                event = EVENT_SCANNER_PERSIST_STATE,
+                component = LOG_COMPONENT_SCANNER,
+                subsystem = LOG_SUBSYSTEM_RUNTIME,
+                path = %target_path,
+                state = "publication_blocked_before_reconcile",
+                "Scanner data usage publication deferred by the pool-state fence"
+            );
+            outcome = DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement);
+            break;
+        }
 
         if observational && data_usage_info.usage_snapshot_authoritative_baseline.is_none() {
             let authoritative_data = match next_baseline.as_ref() {
@@ -275,6 +315,18 @@ pub(super) async fn store_data_usage_in_backend_with_outcome_for_epoch_and_basel
             if ctx.is_cancelled() {
                 break 'updates;
             }
+            if route_probe().await {
+                debug!(
+                    target: "rustfs::scanner",
+                    event = EVENT_SCANNER_PERSIST_STATE,
+                    component = LOG_COMPONENT_SCANNER,
+                    subsystem = LOG_SUBSYSTEM_RUNTIME,
+                    path = %target_path,
+                    state = "publication_blocked_before_save",
+                    "Scanner data usage publication deferred by the final pool-state fence"
+                );
+                break DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement);
+            }
 
             let done_save = Metrics::time(Metric::SaveUsage);
             let save_result = save_config_shared_with_preconditions(
@@ -313,6 +365,33 @@ pub(super) async fn store_data_usage_in_backend_with_outcome_for_epoch_and_basel
                         "Scanner data usage CAS conflict will be reconciled"
                     );
                 }
+                Err(e @ EcstoreError::ObjectNotFound(_, _)) => {
+                    let route_blocked = route_probe().await;
+                    if route_blocked {
+                        warn!(
+                            target: "rustfs::scanner",
+                            event = EVENT_SCANNER_PERSIST_STATE,
+                            component = LOG_COMPONENT_SCANNER,
+                            subsystem = LOG_SUBSYSTEM_RUNTIME,
+                            path = %target_path,
+                            state = "publication_deferred",
+                            error = %e,
+                            "Scanner data usage route is blocked by data movement; retrying later"
+                        );
+                        break DataUsagePersistOutcome::Deferred(ScannerCycleDeferReason::DataMovement);
+                    }
+                    error!(
+                        target: "rustfs::scanner",
+                        event = EVENT_SCANNER_PERSIST_STATE,
+                        component = LOG_COMPONENT_SCANNER,
+                        subsystem = LOG_SUBSYSTEM_RUNTIME,
+                        path = %target_path,
+                        state = "save_failed",
+                        error = %e,
+                        "Scanner data usage save failed"
+                    );
+                    break DataUsagePersistOutcome::Failed;
+                }
                 Err(e) => {
                     error!(
                         target: "rustfs::scanner",
@@ -370,6 +449,13 @@ pub(super) async fn store_data_usage_in_backend_with_outcome_for_epoch_and_basel
                 outcome = DataUsagePersistOutcome::Failed;
                 continue;
             }
+            DataUsagePersistOutcome::Deferred(reason) => {
+                // A deferred publication is an intentional retryable state, not a
+                // failed save. Keep the last real save result so admin freshness
+                // reporting does not turn a pool-recovery fence into a false error.
+                outcome = DataUsagePersistOutcome::Deferred(reason);
+                break 'updates;
+            }
             DataUsagePersistOutcome::Saved => {
                 if observational {
                     invalidate_admin_data_usage_snapshot_cache().await;

crates/scanner/src/scanner_io/io_cycle.rs

@@ -49,6 +49,25 @@ impl ScannerIOCycle for ECStore {
     ) -> Result<ScannerCycleResult> {
         let child_token = ctx.child_token();
 
+        // Check the local pool metadata before listing buckets. A failed or
+        // canceled decommission remains suspended after its worker exits, so
+        // starting a scan in that state could build a snapshot that cannot be
+        // routed to the authoritative metadata object.
+        if self.scanner_data_usage_publication_blocked().await {
+            debug!(
+                target: "rustfs::scanner::io",
+                event = EVENT_SCANNER_SET_STATE,
+                component = LOG_COMPONENT_SCANNER,
+                subsystem = LOG_SUBSYSTEM_IO,
+                state = "cycle_data_usage_route_blocked",
+                "Scanner cycle deferred while data usage metadata remains hidden by data movement"
+            );
+            return Ok(ScannerCycleResult::new(
+                ScannerCycleStatus::Deferred(ScannerCycleDeferReason::DataMovement),
+                None,
+            ));
+        }
+
         let distributed = self.setup_is_dist_erasure().await;
         let activity_before = match scanner_activity_preflight(crate::scanner::probe_scanner_activity(self, distributed).await) {
             ScannerActivityPreflight::Ready(snapshot) => snapshot,

crates/scanner/src/scanner_io/tests.rs

@@ -17,7 +17,9 @@ use super::io_disk::tier_stats_template;
 use super::*;
 use crate::scanner_budget::ScannerCycleBudgetConfig;
 use crate::scanner_folder::ScannerItem;
-use crate::storage_api::owner::{EcstoreRebalStatus, EcstoreRebalanceInfo, EcstoreRebalanceMeta, EcstoreRebalanceStats};
+use crate::storage_api::owner::{
+    EcstorePoolDecommissionInfo, EcstoreRebalStatus, EcstoreRebalanceInfo, EcstoreRebalanceMeta, EcstoreRebalanceStats,
+};
 use crate::storage_api::scan::{BucketOperations as _, DeleteBucketOptions, MakeBucketOptions, ObjectIO as _};
 use crate::{
     DiskOption, ECStore, Endpoint, EndpointServerPools, Endpoints, InstanceContext, PoolEndpoints, ScannerObjectOptions,
@@ -182,6 +184,39 @@ async fn scanner_cycle_is_deferred_while_rebalance_is_active() {
     assert!(receiver.recv().await.is_none(), "rebalance-deferred cycle must not publish usage");
 }
 
+#[tokio::test]
+#[serial]
+async fn scanner_cycle_is_deferred_while_terminal_decommission_is_blocked() {
+    let (_temp_dir, store) = setup_two_pool_scanner_store().await;
+    for decommission in [
+        EcstorePoolDecommissionInfo {
+            failed: true,
+            ..Default::default()
+        },
+        EcstorePoolDecommissionInfo {
+            canceled: true,
+            ..Default::default()
+        },
+    ] {
+        store.pool_meta.write().await.pools[0].decommission = Some(decommission);
+        assert!(store.scanner_data_usage_publication_blocked().await);
+
+        let ctx = CancellationToken::new();
+        let budget = ScannerCycleBudget::new(&ctx, ScannerCycleBudgetConfig::default());
+        let (updates, mut receiver) = mpsc::channel(1);
+        let result = tokio::time::timeout(
+            Duration::from_secs(30),
+            ScannerIOCycle::nsscanner_with_status(store.as_ref(), ctx, budget, updates, 1, 1, HealScanMode::Normal),
+        )
+        .await
+        .expect("terminal-decommission-deferred scanner cycle should finish")
+        .expect("terminal-decommission-deferred scanner cycle should succeed");
+
+        assert_eq!(result.status, ScannerCycleStatus::Deferred(ScannerCycleDeferReason::DataMovement));
+        assert!(receiver.recv().await.is_none(), "blocked cycle must not publish usage");
+    }
+}
+
 #[tokio::test]
 async fn data_usage_publish_fails_when_receiver_is_closed() {
     let (updates, receiver) = mpsc::channel(1);
@@ -236,6 +271,10 @@ async fn multi_pool_scanner_cycle_publishes_combined_usage() {
     assert_eq!(bucket_usage.size, 11);
     assert_eq!(usage.objects_total_count, 2);
     assert_eq!(usage.objects_total_size, 11);
+    assert!(
+        receiver.recv().await.is_none(),
+        "a scanner cycle must publish at most one terminal usage snapshot"
+    );
 }
 
 #[tokio::test]

crates/scanner/src/storage_api.rs

@@ -47,6 +47,8 @@ pub(crate) use rustfs_ecstore::api::bucket::versioning_sys::BucketVersioningSys
 pub(crate) use rustfs_ecstore::api::cache::{
     ListPathRawOptions as EcstoreListPathRawOptions, list_path_raw as ecstore_list_path_raw,
 };
+#[cfg(test)]
+pub(crate) use rustfs_ecstore::api::capacity::PoolDecommissionInfo as EcstorePoolDecommissionInfo;
 pub(crate) use rustfs_ecstore::api::capacity::{
     is_reserved_or_invalid_bucket as ecstore_is_reserved_or_invalid_bucket, path2_bucket_object as ecstore_path2_bucket_object,
     path2_bucket_object_with_base_path as ecstore_path2_bucket_object_with_base_path,
@@ -127,9 +129,9 @@ pub(crate) mod owner {
     #[cfg(test)]
     pub(crate) use super::{
         EcstoreDiskOption, EcstoreDiskStore, EcstoreEndpoint, EcstoreEndpointServerPools, EcstoreEndpoints,
-        EcstoreInstanceContext, EcstorePoolEndpoints, EcstoreRebalStatus, EcstoreRebalanceInfo, EcstoreRebalanceMeta,
-        EcstoreRebalanceStats, ecstore_config_init, ecstore_init_bucket_metadata_sys, ecstore_init_local_disks_with_instance_ctx,
-        ecstore_new_disk,
+        EcstoreInstanceContext, EcstorePoolDecommissionInfo, EcstorePoolEndpoints, EcstoreRebalStatus, EcstoreRebalanceInfo,
+        EcstoreRebalanceMeta, EcstoreRebalanceStats, ecstore_config_init, ecstore_init_bucket_metadata_sys,
+        ecstore_init_local_disks_with_instance_ctx, ecstore_new_disk,
     };
 }