fix(scanner): unify unknown metadata size accounting

crates/data-usage/src/data_usage.rs

@@ -307,6 +307,38 @@ pub struct DiskUsageStatus {
     pub snapshot_exists: bool,
 }
 
+/// A bounded reconciliation record for an object whose logical size could not
+/// be trusted at the scanner boundary. The scanner persists these records in
+/// its cache; keeping the model here avoids a second, incompatible accounting
+/// representation in storage-facing crates.
+#[derive(Debug, Default, Clone, Serialize, Deserialize, PartialEq, Eq)]
+pub struct SizeReconciliationEntry {
+    /// Stable object/version identity key (not a metrics label).
+    pub key: String,
+    pub bucket: String,
+    pub object: String,
+    #[serde(default)]
+    pub version_id: Option<String>,
+    #[serde(default)]
+    pub generation: Option<String>,
+    /// Structured reason label; raw metadata values must never be stored here.
+    pub reason: String,
+    #[serde(default)]
+    pub physical_size: Option<u64>,
+    #[serde(default)]
+    pub first_seen: u64,
+    #[serde(default)]
+    pub attempts: u32,
+}
+
+/// Object scope refreshed by one scanner pass. Existing debts in this scope
+/// are removed before the pass's unresolved records are inserted.
+#[derive(Debug, Default, Clone, PartialEq, Eq)]
+pub struct SizeReconciliationScope {
+    pub bucket: String,
+    pub object: String,
+}
+
 /// Size summary for a single object or group of objects
 #[derive(Debug, Default, Clone)]
 pub struct SizeSummary {
@@ -336,6 +368,16 @@ pub struct SizeSummary {
     pub repl_target_stats: HashMap<String, ReplTargetSizeSummary>,
     /// Per-tier accounting, keyed by storage class or remote tier name
     pub tier_stats: HashMap<String, TierStats>,
+    /// Size-resolution debts observed while scanning this summary.
+    pub size_reconciliation: Vec<SizeReconciliationEntry>,
+    /// True when the per-object summary exceeded its bounded debt buffer.
+    /// Callers must retain prior ledger entries rather than treating the
+    /// partial list as a complete refresh.
+    pub size_reconciliation_truncated: bool,
+    /// Object scopes refreshed by this summary. They let the durable ledger
+    /// remove versions that resolved without allocating one key per healthy
+    /// version on the hot path.
+    pub reconciliation_scopes: Vec<SizeReconciliationScope>,
 }
 
 /// Replication target size summary
@@ -830,7 +872,8 @@ impl DataUsageEntry {
 ///
 /// The canonical wire format is written by the hand-written map-encoded
 /// `Serialize` on the scanner-side `DataUsageCacheInfo`
-/// (`crates/scanner/src/data_usage_define.rs`), which carries 16 fields.
+/// (`crates/scanner/src/data_usage_define.rs`), which carries the original 16
+/// fields plus an optional reconciliation field.
 /// This type decodes only the shared subset and is deliberately not
 /// `Serialize`: a derived (array) encoding of this 6-field subset would
 /// corrupt the cache for scanner readers, so no write path may exist here.
@@ -1774,6 +1817,51 @@ impl SizeSummary {
             entry.pending_count = entry.pending_count.saturating_add(stats.pending_count);
             entry.failed_count = entry.failed_count.saturating_add(stats.failed_count);
         }
+
+        for entry in &other.size_reconciliation {
+            self.record_size_reconciliation(entry.clone());
+        }
+        self.size_reconciliation_truncated |= other.size_reconciliation_truncated;
+        for scope in &other.reconciliation_scopes {
+            self.record_reconciliation_scope(&scope.bucket, &scope.object);
+        }
+    }
+
+    /// Add one reconciliation debt, coalescing repeated observations in the
+    /// same object summary. The scanner cache applies its own larger bound.
+    pub fn record_size_reconciliation(&mut self, entry: SizeReconciliationEntry) {
+        const MAX_SUMMARY_RECONCILIATION_ENTRIES: usize = 1024;
+        if let Some(existing) = self.size_reconciliation.iter_mut().find(|value| value.key == entry.key) {
+            existing.reason = entry.reason;
+            existing.physical_size = entry.physical_size;
+            existing.generation = entry.generation;
+            existing.version_id = entry.version_id;
+            return;
+        }
+        if self.size_reconciliation.len() < MAX_SUMMARY_RECONCILIATION_ENTRIES {
+            self.size_reconciliation.push(entry);
+        } else {
+            self.size_reconciliation_truncated = true;
+        }
+    }
+
+    /// Mark one object scope as refreshed. Duplicate scopes are suppressed so
+    /// merging summaries remains bounded and deterministic.
+    pub fn record_reconciliation_scope(&mut self, bucket: &str, object: &str) {
+        if !self
+            .reconciliation_scopes
+            .iter()
+            .any(|scope| scope.bucket == bucket && scope.object == object)
+        {
+            if self.reconciliation_scopes.len() >= 1024 {
+                self.size_reconciliation_truncated = true;
+                return;
+            }
+            self.reconciliation_scopes.push(SizeReconciliationScope {
+                bucket: bucket.to_string(),
+                object: object.to_string(),
+            });
+        }
     }
 }
 

crates/ecstore/src/bucket/bucket_target_sys.rs

@@ -3425,44 +3425,6 @@ mod tests {
         assert!(mutexes.contains_key("second"));
     }
 
-    #[tokio::test]
-    async fn update_all_targets_publishes_disable_proxy_on_target_client() {
-        // The read-proxy selector (replication_proxy::get_proxy_targets) skips
-        // targets whose TargetClient carries disable_proxy — the persisted
-        // per-target opt-out must survive client publication.
-        let sys = BucketTargetSys::default();
-        let target = |arn: &str, disable_proxy: bool| BucketTarget {
-            arn: arn.to_string(),
-            endpoint: "192.168.1.10:9000".to_string(),
-            target_bucket: "target-bucket".to_string(),
-            region: "us-east-1".to_string(),
-            disable_proxy,
-            credentials: Some(Credentials {
-                access_key: "access".to_string(),
-                secret_key: "secret".to_string(),
-                session_token: None,
-                expiration: None,
-            }),
-            ..Default::default()
-        };
-        let targets = BucketTargets {
-            targets: vec![target("arn:proxied", false), target("arn:opted-out", true)],
-        };
-
-        sys.update_all_targets("bucket", Some(&targets)).await;
-
-        let proxied = sys
-            .get_remote_target_client("bucket", "arn:proxied")
-            .await
-            .expect("client should be published");
-        assert!(!proxied.disable_proxy);
-        let opted_out = sys
-            .get_remote_target_client("bucket", "arn:opted-out")
-            .await
-            .expect("client should be published");
-        assert!(opted_out.disable_proxy, "disable_proxy must reach the published TargetClient");
-    }
-
     #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
     async fn target_updates_serialize_client_build_through_publication_per_bucket() {
         let sys = Arc::new(BucketTargetSys::default());

crates/ecstore/src/store/heal.rs

@@ -297,16 +297,10 @@ impl ECStore {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::bucket::metadata_sys;
     use crate::core::pools::{PoolDecommissionInfo, PoolStatus};
-    use crate::disk::{DeleteOptions, DiskOption, format::FormatV3, new_disk};
-    use crate::layout::endpoints::{EndpointServerPools, Endpoints, PoolEndpoints};
-    use crate::runtime::instance::InstanceContext;
-    use crate::storage_api_contracts::bucket::{BucketOperations, MakeBucketOptions};
-    use crate::storage_api_contracts::object::{ObjectIO as _, ObjectOperations};
+    use crate::disk::{DiskOption, format::FormatV3, new_disk};
+    use crate::layout::endpoints::{Endpoints, PoolEndpoints};
     use crate::store::init_format::{load_format_erasure, save_format_file};
-    use crate::store::init_local_disks_with_instance_ctx;
-    use tokio_util::sync::CancellationToken;
 
     async fn minimal_heal_pool(pool_idx: usize) -> Arc<Sets> {
         let format = FormatV3::new(1, 1);
@@ -353,51 +347,6 @@ mod tests {
         }
     }
 
-    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();
-        for pool_index in 0..2 {
-            let mut endpoints = Vec::new();
-            for disk_index in 0..4 {
-                let disk_path = temp_dir.path().join(format!("pool{pool_index}-disk{disk_index}"));
-                tokio::fs::create_dir_all(&disk_path)
-                    .await
-                    .expect("multi-pool heal test disk should be created");
-                let mut endpoint = Endpoint::try_from(disk_path.to_str().expect("disk path should be utf8"))
-                    .expect("test endpoint should parse");
-                endpoint.set_pool_index(pool_index);
-                endpoint.set_set_index(0);
-                endpoint.set_disk_index(disk_index);
-                endpoints.push(endpoint);
-            }
-            pool_endpoints.push(PoolEndpoints {
-                legacy: false,
-                set_count: 1,
-                drives_per_set: 4,
-                endpoints: Endpoints::from(endpoints),
-                cmd_line: format!("heal-owner-pool-{pool_index}"),
-                platform: "test".to_string(),
-            });
-        }
-
-        let endpoint_pools = EndpointServerPools::from(pool_endpoints);
-        let instance_ctx = Arc::new(InstanceContext::new());
-        init_local_disks_with_instance_ctx(&instance_ctx, endpoint_pools.clone())
-            .await
-            .expect("multi-pool local disks should initialize");
-        let shutdown = CancellationToken::new();
-        let store = ECStore::new_with_instance_ctx(
-            "127.0.0.1:0".parse().expect("test address should parse"),
-            endpoint_pools,
-            shutdown.clone(),
-            instance_ctx,
-        )
-        .await
-        .expect("multi-pool test store should initialize");
-        metadata_sys::init_bucket_metadata_sys(store.clone(), Vec::new()).await;
-        (temp_dir, store, shutdown)
-    }
-
     #[tokio::test]
     async fn heal_object_pool_scope_selects_only_requested_pool() {
         let store = minimal_heal_store().await;
@@ -557,229 +506,6 @@ mod tests {
         }
     }
 
-    #[tokio::test]
-    #[serial_test::serial]
-    async fn unscoped_heal_object_suspended_owner_semantics() {
-        let (_temp_dir, store, shutdown) = multi_pool_heal_store().await;
-        let bucket = format!("heal-owner-{}", Uuid::new_v4().simple());
-        let active_object = "active-owner";
-        let suspended_only_object = "suspended-only";
-        let duplicate_object = "duplicate-owner";
-        let marker_object = "marker-owner";
-        let quorum_object = "quorum-owner";
-        store
-            .make_bucket(&bucket, &MakeBucketOptions::default())
-            .await
-            .expect("bucket should be created in all pools");
-
-        let mut active_reader = PutObjReader::from_vec(b"active owner".to_vec());
-        store.pools[0]
-            .put_object(&bucket, active_object, &mut active_reader, &ObjectOptions::default())
-            .await
-            .expect("active owner object should be written");
-        let active_disks = store.pools[0].disk_set[0].disks.read().await.clone();
-        let missing_active_disk = active_disks[0].clone().expect("active disk should be online");
-        missing_active_disk
-            .delete(
-                &bucket,
-                active_object,
-                DeleteOptions {
-                    recursive: true,
-                    immediate: true,
-                    ..Default::default()
-                },
-            )
-            .await
-            .expect("active owner shard should be removed for repair");
-        assert!(
-            missing_active_disk.read_xl(&bucket, active_object, false).await.is_err(),
-            "the active owner fixture must start with one missing metadata copy"
-        );
-
-        let mut suspended_reader = PutObjReader::from_vec(b"suspended owner".to_vec());
-        store.pools[1]
-            .put_object(&bucket, suspended_only_object, &mut suspended_reader, &ObjectOptions::default())
-            .await
-            .expect("suspended owner object should be written");
-        for (pool_index, mod_time) in [1_i64, 2_i64].into_iter().enumerate() {
-            let mut duplicate_reader = PutObjReader::from_vec(format!("duplicate-pool-{pool_index}").into_bytes());
-            store.pools[pool_index]
-                .put_object(
-                    &bucket,
-                    duplicate_object,
-                    &mut duplicate_reader,
-                    &ObjectOptions {
-                        mod_time: Some(OffsetDateTime::UNIX_EPOCH + time::Duration::seconds(mod_time)),
-                        ..Default::default()
-                    },
-                )
-                .await
-                .expect("duplicate owner object should be written");
-        }
-        let duplicate_missing_disk = store.pools[0].disk_set[0].disks.read().await[0]
-            .clone()
-            .expect("duplicate active owner disk should be online");
-        duplicate_missing_disk
-            .delete(
-                &bucket,
-                duplicate_object,
-                DeleteOptions {
-                    recursive: true,
-                    immediate: true,
-                    ..Default::default()
-                },
-            )
-            .await
-            .expect("duplicate active owner shard should be removed for repair");
-        let history_version = Uuid::new_v4();
-        let mut history_reader = PutObjReader::from_vec(b"marker history".to_vec());
-        store.pools[0]
-            .put_object(
-                &bucket,
-                marker_object,
-                &mut history_reader,
-                &ObjectOptions {
-                    versioned: true,
-                    version_id: Some(history_version.to_string()),
-                    mod_time: Some(OffsetDateTime::UNIX_EPOCH + time::Duration::seconds(1)),
-                    ..Default::default()
-                },
-            )
-            .await
-            .expect("versioned marker history should be written");
-        store.pools[0]
-            .delete_object(
-                &bucket,
-                marker_object,
-                ObjectOptions {
-                    versioned: true,
-                    mod_time: Some(OffsetDateTime::UNIX_EPOCH + time::Duration::seconds(2)),
-                    ..Default::default()
-                },
-            )
-            .await
-            .expect("delete marker should be written");
-        let mut quorum_reader = PutObjReader::from_vec(b"quorum boundary".to_vec());
-        store.pools[0]
-            .put_object(&bucket, quorum_object, &mut quorum_reader, &ObjectOptions::default())
-            .await
-            .expect("quorum boundary object should be written");
-        {
-            let mut pool_meta = store.pool_meta.write().await;
-            let mut next = PoolMeta::new(&store.pools, &pool_meta);
-            next.pools[1].decommission = Some(PoolDecommissionInfo {
-                start_time: Some(OffsetDateTime::UNIX_EPOCH),
-                ..Default::default()
-            });
-            *pool_meta = next;
-        }
-
-        let (_, duplicate_owner) = store
-            .get_latest_object_info_with_idx(&bucket, duplicate_object, &ObjectOptions::default())
-            .await
-            .expect("duplicate owner should resolve");
-        assert_eq!(duplicate_owner, 1, "latest duplicate must win when all pools are eligible");
-        let (_, active_duplicate_owner) = store
-            .get_latest_object_info_with_idx(
-                &bucket,
-                duplicate_object,
-                &ObjectOptions {
-                    skip_decommissioned: true,
-                    ..Default::default()
-                },
-            )
-            .await
-            .expect("active duplicate owner should resolve");
-        assert_eq!(
-            active_duplicate_owner, 0,
-            "suspended duplicate must be excluded from active owner selection"
-        );
-        let (duplicate_result, duplicate_err) = store
-            .handle_heal_object(&bucket, duplicate_object, "", &HealOpts::default())
-            .await
-            .expect("duplicate owner heal should complete through the production path");
-        assert_eq!(duplicate_result.object, duplicate_object);
-        assert!(duplicate_err.is_none(), "active duplicate should be repaired: {duplicate_err:?}");
-        assert!(
-            duplicate_missing_disk.read_xl(&bucket, duplicate_object, false).await.is_ok(),
-            "production heal must repair the active duplicate owner rather than the suspended owner"
-        );
-        let (marker_info, marker_owner) = store
-            .get_latest_object_info_with_idx(
-                &bucket,
-                marker_object,
-                &ObjectOptions {
-                    skip_decommissioned: true,
-                    versioned: true,
-                    ..Default::default()
-                },
-            )
-            .await
-            .expect("latest delete marker should resolve");
-        assert_eq!(marker_owner, 0);
-        assert!(marker_info.delete_marker, "latest version must preserve delete-marker semantics");
-
-        let (active_result, active_err) = store
-            .handle_heal_object(&bucket, active_object, "", &HealOpts::default())
-            .await
-            .expect("unscoped active-owner heal should complete");
-        assert_eq!(active_result.object, active_object);
-        assert!(active_err.is_none(), "active owner must be selected even with a suspended pool");
-        assert!(
-            missing_active_disk.read_xl(&bucket, active_object, false).await.is_ok(),
-            "active owner heal must write the missing disk metadata: result={active_result:?}, err={active_err:?}"
-        );
-        assert!(
-            store.pools[1]
-                .get_object_info(&bucket, active_object, &ObjectOptions::default())
-                .await
-                .is_err(),
-            "the suspended pool must not be written for an active-owner object"
-        );
-
-        let (suspended_result, suspended_err) = store
-            .handle_heal_object(&bucket, suspended_only_object, "", &HealOpts::default())
-            .await
-            .expect("unscoped suspended-only heal should return a terminal result");
-        assert!(suspended_result.object.is_empty());
-        assert!(matches!(suspended_err, Some(Error::FileNotFound)));
-        assert!(
-            store.pools[1]
-                .get_object_info(&bucket, suspended_only_object, &ObjectOptions::default())
-                .await
-                .is_ok(),
-            "suspended-only data must remain untouched when unscoped heal reports absent"
-        );
-
-        let (_, explicit_err) = store
-            .handle_heal_object(
-                &bucket,
-                suspended_only_object,
-                "",
-                &HealOpts {
-                    pool: Some(1),
-                    ..Default::default()
-                },
-            )
-            .await
-            .expect("explicit suspended-owner heal should return a mapped error");
-        assert!(matches!(explicit_err, Some(Error::SlowDown)));
-
-        let original_quorum_disks = store.pools[0].disk_set[0].disks.read().await.clone();
-        let surviving_quorum_disk = original_quorum_disks[3].clone();
-        *store.pools[0].disk_set[0].disks.write().await = vec![None, None, None, surviving_quorum_disk];
-        let (_, quorum_err) = store
-            .handle_heal_object(&bucket, quorum_object, "", &HealOpts::default())
-            .await
-            .expect("quorum boundary heal should return a mapped result");
-        *store.pools[0].disk_set[0].disks.write().await = original_quorum_disks;
-        assert!(
-            matches!(quorum_err, Some(Error::ErasureReadQuorum)),
-            "quorum-boundary heal must preserve quorum error, got {quorum_err:?}"
-        );
-        shutdown.cancel();
-    }
-
     #[tokio::test]
     async fn handle_heal_format_continues_after_a_pool_error() {
         let canonical_format = FormatV3::new(1, 3);

crates/heal/src/heal/erasure_healer.rs

@@ -13,9 +13,9 @@
 // limitations under the License.
 
 use crate::heal::{
-    progress::{HealProgress, add_bytes, increment_counter},
+    progress::HealProgress,
     resume::{
-        CheckpointManager, CheckpointObjectOutcome, ReplacementTargetIdentity, ResumeManager, ResumeUtils, compose_key,
+        CheckpointManager, ReplacementTargetIdentity, ResumeManager, ResumeUtils, compose_key,
         replacement_target_identities_match,
     },
     storage::{HealStorageAPI, next_heal_listing_token},
@@ -410,9 +410,6 @@ impl ErasureSetHealer {
                 && state.successful_objects == 0
                 && state.failed_objects == 0
                 && state.skipped_objects == 0
-                && state.skipped_new_versions == 0
-                && state.skipped_ilm_expired == 0
-                && state.processed_bytes == 0
             {
                 // schedule_retry persists the authoritative resume reset before
                 // resetting the checkpoint. Reapply the checkpoint reset after
@@ -477,23 +474,6 @@ impl ErasureSetHealer {
 
         // 2. initialize progress
         self.initialize_progress(buckets, &state).await;
-        let (baseline_known, baseline_count, baseline_size, baseline_generation) = {
-            let baseline = self.progress.read().await;
-            (
-                baseline.baseline_known,
-                baseline.objects_total_count,
-                baseline.objects_total_size,
-                baseline.baseline_generation,
-            )
-        };
-        if baseline_known {
-            resume_manager
-                .set_progress_baseline(baseline_count, baseline_size, baseline_generation)
-                .await?;
-            checkpoint_manager
-                .set_progress_baseline(baseline_count, baseline_size, baseline_generation)
-                .await?;
-        }
 
         // 3. continue from checkpoint
         let current_bucket_index = checkpoint.current_bucket_index;
@@ -503,66 +483,12 @@ impl ErasureSetHealer {
         let mut successful_objects = state.successful_objects;
         let mut failed_objects = state.failed_objects;
         let mut skipped_objects = state.skipped_objects;
-        let checkpoint_has_progress = checkpoint.baseline_known
-            || checkpoint.successful_objects > 0
-            || checkpoint.failed_object_count > 0
-            || checkpoint.skipped_object_count > 0
-            || checkpoint.skipped_new_versions > 0
-            || checkpoint.skipped_ilm_expired > 0
-            || checkpoint.processed_bytes > 0
-            || checkpoint.total_objects > 0
-            || checkpoint.total_bytes > 0
-            || checkpoint.baseline_generation.is_some()
-            || checkpoint.counter_unknown;
-        let checkpoint_generation_mismatch = checkpoint.baseline_known && checkpoint.baseline_generation != baseline_generation;
-        let mut restored_counter_unknown = state.counter_unknown || checkpoint.counter_unknown;
-        if checkpoint_has_progress {
-            successful_objects = checkpoint.successful_objects;
-            failed_objects = checkpoint.failed_object_count;
-            skipped_objects = checkpoint.skipped_object_count;
-            let restored_processed_objects = successful_objects
-                .checked_add(failed_objects)
-                .and_then(|value| value.checked_add(skipped_objects))
-                .and_then(|value| value.checked_add(checkpoint.skipped_new_versions))
-                .and_then(|value| value.checked_add(checkpoint.skipped_ilm_expired));
-            let checkpoint_counter_overflow = restored_processed_objects.is_none();
-            restored_counter_unknown |= checkpoint_counter_overflow;
-            processed_objects = restored_processed_objects.unwrap_or(u64::MAX);
-            let mut progress = self.progress.write().await;
-            progress.objects_scanned = processed_objects;
-            progress.objects_healed = successful_objects;
-            progress.objects_failed = failed_objects;
-            progress.skipped_objects = skipped_objects;
-            progress.skipped_new_versions = checkpoint.skipped_new_versions;
-            progress.skipped_ilm_expired = checkpoint.skipped_ilm_expired;
-            if checkpoint.baseline_known && !checkpoint_generation_mismatch {
-                progress.objects_total_count = checkpoint.total_objects;
-                progress.objects_total_size = checkpoint.total_bytes;
-                progress.baseline_generation = checkpoint.baseline_generation;
-                progress.baseline_known = true;
-            }
-            progress.bytes_processed = checkpoint.processed_bytes;
-            progress.counter_unknown = state.counter_unknown || checkpoint.counter_unknown;
-            progress.refresh_progress_percentage();
-            if checkpoint_generation_mismatch || checkpoint_counter_overflow || progress.counter_unknown {
-                progress.mark_unknown();
-            }
-        }
-        if checkpoint_generation_mismatch {
-            restored_counter_unknown = true;
-        }
-        if restored_counter_unknown {
-            checkpoint_manager.mark_counter_unknown().await?;
-            resume_manager.mark_counter_unknown().await?;
-        }
         let mut failed_buckets = 0u64;
 
         // 4. process remaining buckets
         for (bucket_idx, bucket) in buckets.iter().enumerate().skip(current_bucket_index) {
             // check if completed
             if state.completed_buckets.contains(bucket) {
-                checkpoint_manager.complete_bucket(bucket_idx.saturating_add(1)).await?;
-                current_object_index = 0;
                 continue;
             }
 
@@ -590,42 +516,13 @@ impl ErasureSetHealer {
                 return bucket_result;
             }
 
-            // update progress
-            let progress_snapshot = self.progress.read().await;
-            let bytes_processed = progress_snapshot.bytes_processed;
-            let skipped_new_versions = progress_snapshot.skipped_new_versions;
-            let skipped_ilm_expired = progress_snapshot.skipped_ilm_expired;
-            let counter_unknown = progress_snapshot.counter_unknown;
-            drop(progress_snapshot);
-            // The checkpoint is the recovery authority for object progress.
-            // Publish its counters and fence before the resume summary so a
-            // crash between the two stores cannot make recovery select newer
-            // summary bytes with an older checkpoint ledger.
-            if counter_unknown {
-                checkpoint_manager.mark_counter_unknown().await?;
-            }
-            checkpoint_manager
-                .update_progress(successful_objects, failed_objects, skipped_objects, bytes_processed)
-                .await?;
-            checkpoint_manager
-                .set_skipped_version_counts(skipped_new_versions, skipped_ilm_expired)
-                .await?;
+            // update checkpoint position
             checkpoint_manager.update_position(bucket_idx, current_object_index).await?;
+
+            // update progress
             resume_manager
-                .update_progress_with_bytes(
-                    processed_objects,
-                    successful_objects,
-                    failed_objects,
-                    skipped_objects,
-                    bytes_processed,
-                )
+                .update_progress(processed_objects, successful_objects, failed_objects, skipped_objects)
                 .await?;
-            resume_manager
-                .set_skipped_version_counts(skipped_new_versions, skipped_ilm_expired)
-                .await?;
-            if counter_unknown {
-                resume_manager.mark_counter_unknown().await?;
-            }
 
             // check cancel status
             if self.cancel_token.is_cancelled() {
@@ -645,7 +542,6 @@ impl ErasureSetHealer {
             match bucket_result {
                 Ok(_) => {
                     resume_manager.complete_bucket(bucket).await?;
-                    checkpoint_manager.complete_bucket(bucket_idx.saturating_add(1)).await?;
                     debug!(
                         target: "rustfs::heal::erasure_healer",
                         event = EVENT_HEAL_ERASURE_BUCKET_STATE,
@@ -671,9 +567,7 @@ impl ErasureSetHealer {
                         error = %e,
                         "Erasure set bucket heal failed"
                     );
-                    // A single durable cursor and ledger cannot safely preserve
-                    // this bucket while processing a later one.
-                    break;
+                    // continue to next bucket, do not interrupt the whole process
                 }
             }
 
@@ -881,48 +775,20 @@ impl ErasureSetHealer {
 
                 // Per-version dedup identity — the single canonical key.
                 let key = compose_key(&item.name, item.version_id.as_deref());
-                if checkpoint.processed_objects.contains(&key)
-                    || checkpoint.failed_objects.contains(&key)
-                    || checkpoint.skipped_objects.contains(&key)
-                {
+                if checkpoint.processed_objects.contains(&key) || checkpoint.skipped_objects.contains(&key) {
                     continue;
                 }
 
                 if should_skip_new_version(item.mod_time_unix_nanos, started_at_secs) {
-                    let counter_ok = increment_counter(processed_objects);
+                    checkpoint_manager.add_processed_object(key).await?;
+                    *processed_objects = processed_objects.saturating_add(1);
                     completed_in_page = completed_in_page.saturating_add(1);
                     counter!("rustfs_heal_skipped_new_versions_total").increment(1);
-                    let (skipped_new, skipped_ilm, counter_unknown) = {
+                    {
                         let mut progress = self.progress.write().await;
                         progress.record_skipped_new_version();
                         progress.set_current_object(Some(format!("skipped_new: {bucket}/{}", item.name)));
-                        progress.update_object_progress(
-                            *processed_objects,
-                            *successful_objects,
-                            *failed_objects,
-                            *skipped_objects,
-                            bytes_processed,
-                        );
-                        if !counter_ok {
-                            progress.mark_unknown();
-                        }
-                        (progress.skipped_new_versions, progress.skipped_ilm_expired, progress.counter_unknown)
-                    };
-                    checkpoint_manager
-                        .record_object_outcome(
-                            key,
-                            CheckpointObjectOutcome::Processed,
-                            *successful_objects,
-                            *failed_objects,
-                            *skipped_objects,
-                            bytes_processed,
-                            skipped_new,
-                            skipped_ilm,
-                            !counter_ok || counter_unknown,
-                        )
-                        .await?;
-                    if !counter_ok || counter_unknown {
-                        resume_manager.mark_counter_unknown().await?;
+                        progress.update_progress(*processed_objects, *successful_objects, *failed_objects, bytes_processed);
                     }
                     debug!(
                         target: "rustfs::heal::erasure_healer",
@@ -954,40 +820,15 @@ impl ErasureSetHealer {
                         )
                         .await?
                 {
-                    let counter_ok = increment_counter(processed_objects);
+                    checkpoint_manager.add_processed_object(key).await?;
+                    *processed_objects = processed_objects.saturating_add(1);
                     completed_in_page = completed_in_page.saturating_add(1);
                     counter!("rustfs_heal_skipped_ilm_expired_total").increment(1);
-                    let (skipped_new, skipped_ilm, counter_unknown) = {
+                    {
                         let mut progress = self.progress.write().await;
                         progress.record_skipped_ilm_expired();
                         progress.set_current_object(Some(format!("skipped_ilm: {bucket}/{}", item.name)));
-                        progress.update_object_progress(
-                            *processed_objects,
-                            *successful_objects,
-                            *failed_objects,
-                            *skipped_objects,
-                            bytes_processed,
-                        );
-                        if !counter_ok {
-                            progress.mark_unknown();
-                        }
-                        (progress.skipped_new_versions, progress.skipped_ilm_expired, progress.counter_unknown)
-                    };
-                    checkpoint_manager
-                        .record_object_outcome(
-                            key,
-                            CheckpointObjectOutcome::Processed,
-                            *successful_objects,
-                            *failed_objects,
-                            *skipped_objects,
-                            bytes_processed,
-                            skipped_new,
-                            skipped_ilm,
-                            !counter_ok || counter_unknown,
-                        )
-                        .await?;
-                    if !counter_ok || counter_unknown {
-                        resume_manager.mark_counter_unknown().await?;
+                        progress.update_progress(*processed_objects, *successful_objects, *failed_objects, bytes_processed);
                     }
                     debug!(
                         target: "rustfs::heal::erasure_healer",
@@ -1113,11 +954,11 @@ impl ErasureSetHealer {
 
             while let Some((key, object, version_id, result)) = page_tasks.next().await {
                 let (object_size, result) = result;
-                let mut telemetry_unknown = false;
-                let checkpoint_outcome = match result {
+                match result {
                     Ok(true) => {
-                        telemetry_unknown |= !increment_counter(successful_objects);
-                        telemetry_unknown |= !add_bytes(&mut bytes_processed, object_size);
+                        *successful_objects += 1;
+                        bytes_processed = bytes_processed.saturating_add(object_size);
+                        checkpoint_manager.add_processed_object(key).await?;
                         debug!(
                             target: "rustfs::heal::erasure_healer",
                             event = EVENT_HEAL_ERASURE_OBJECT_STATE,
@@ -1130,11 +971,11 @@ impl ErasureSetHealer {
                             state = "healed",
                             "Erasure set object healed"
                         );
-                        CheckpointObjectOutcome::Processed
                     }
                     Ok(false) => {
-                        telemetry_unknown |= !increment_counter(successful_objects);
-                        telemetry_unknown |= !add_bytes(&mut bytes_processed, object_size);
+                        checkpoint_manager.add_processed_object(key).await?;
+                        *successful_objects += 1;
+                        bytes_processed = bytes_processed.saturating_add(object_size);
                         debug!(
                             target: "rustfs::heal::erasure_healer",
                             event = EVENT_HEAL_ERASURE_OBJECT_STATE,
@@ -1147,12 +988,12 @@ impl ErasureSetHealer {
                             state = "missing_treated_as_ok",
                             "Erasure set missing object treated as ok"
                         );
-                        CheckpointObjectOutcome::Processed
                     }
                     Err(err @ Error::TaskCancelled) | Err(err @ Error::TaskTimeout) => return Err(err),
                     Err(Error::TransientSkip { message }) => {
-                        telemetry_unknown |= !increment_counter(skipped_objects);
-                        telemetry_unknown |= !add_bytes(&mut bytes_processed, object_size);
+                        *skipped_objects += 1;
+                        bytes_processed = bytes_processed.saturating_add(object_size);
+                        checkpoint_manager.add_skipped_object(key).await?;
                         demote_to_debug_when!(!take_failure_log_sample(&mut transient_skip_samples_logged), warn, target: "rustfs::heal::erasure_healer", {
                             event = EVENT_HEAL_ERASURE_OBJECT_STATE,
                             component = LOG_COMPONENT_HEAL,
@@ -1165,11 +1006,11 @@ impl ErasureSetHealer {
                             error = %message,
                             "Erasure set object heal skipped due to transient error"
                         });
-                        CheckpointObjectOutcome::Skipped
                     }
                     Err(err) => {
-                        telemetry_unknown |= !increment_counter(failed_objects);
-                        telemetry_unknown |= !add_bytes(&mut bytes_processed, object_size);
+                        *failed_objects += 1;
+                        bytes_processed = bytes_processed.saturating_add(object_size);
+                        checkpoint_manager.add_failed_object(key).await?;
                         demote_to_debug_when!(!take_failure_log_sample(&mut failure_samples_logged), warn, target: "rustfs::heal::erasure_healer", {
                             event = EVENT_HEAL_ERASURE_OBJECT_STATE,
                             component = LOG_COMPONENT_HEAL,
@@ -1182,42 +1023,15 @@ impl ErasureSetHealer {
                             error = %err,
                             "Erasure set object heal failed"
                         });
-                        CheckpointObjectOutcome::Failed
                     }
-                };
+                }
 
-                telemetry_unknown |= !increment_counter(processed_objects);
+                *processed_objects += 1;
                 completed_in_page += 1;
-                let (progress_unknown, skipped_new_versions, skipped_ilm_expired) = {
+                {
                     let mut progress = self.progress.write().await;
                     progress.set_current_object(Some(format!("{bucket}/{object}")));
-                    progress.update_object_progress(
-                        *processed_objects,
-                        *successful_objects,
-                        *failed_objects,
-                        *skipped_objects,
-                        bytes_processed,
-                    );
-                    if telemetry_unknown {
-                        progress.mark_unknown();
-                    }
-                    (progress.counter_unknown, progress.skipped_new_versions, progress.skipped_ilm_expired)
-                };
-                checkpoint_manager
-                    .record_object_outcome(
-                        key,
-                        checkpoint_outcome,
-                        *successful_objects,
-                        *failed_objects,
-                        *skipped_objects,
-                        bytes_processed,
-                        skipped_new_versions,
-                        skipped_ilm_expired,
-                        telemetry_unknown || progress_unknown,
-                    )
-                    .await?;
-                if telemetry_unknown || progress_unknown {
-                    resume_manager.mark_counter_unknown().await?;
+                    progress.update_progress(*processed_objects, *successful_objects, *failed_objects, bytes_processed);
                 }
 
                 if completed_in_page.is_multiple_of(100) {
@@ -1227,22 +1041,16 @@ impl ErasureSetHealer {
 
             *current_object_index = global_obj_idx;
 
-            // Persist the checkpoint ledger and page position before exposing
-            // the next resume cursor. A crash before cursor publication keeps
-            // the page identities available for exact-once replay.
-            checkpoint_manager.advance_page(bucket_index, *current_object_index).await?;
+            // Persist the authoritative cursor FIRST (points at the next page
+            // boundary), then prune the per-version dedup sets. Both are
+            // idempotent under crash: heal_object re-heals safely.
+            let next_cursor = if is_truncated { next_token.clone() } else { None };
+            resume_manager.set_resume_cursor(next_cursor.clone()).await?;
+            checkpoint_manager.complete_page(bucket_index, *current_object_index).await?;
             // Check if there are more pages
             if !is_truncated {
                 break;
             }
-            continuation_token = next_heal_listing_token(bucket, "", next_token, is_truncated)?;
-            if continuation_token.is_none() {
-                // A truncated page without a continuation token is terminal.
-                // Retain its ledger until bucket completion is durable.
-                break;
-            }
-            resume_manager.set_resume_cursor(continuation_token.clone()).await?;
-            checkpoint_manager.prune_completed_page().await?;
 
             // Anti-loop guard: an empty page reported as truncated cannot advance
             // the cursor (there is no last identity to move past), so treat it as a
@@ -1261,6 +1069,12 @@ impl ErasureSetHealer {
                 )));
             }
             previous_page_last = page_last;
+
+            continuation_token = next_heal_listing_token(bucket, "", next_token, is_truncated)?;
+            if continuation_token.is_none() {
+                // Truncated but no continuation token: treat as end of listing.
+                break;
+            }
         }
 
         Ok(())
@@ -1269,66 +1083,10 @@ impl ErasureSetHealer {
     /// initialize progress tracking
     async fn initialize_progress(&self, _buckets: &[String], state: &crate::heal::resume::ResumeState) {
         let mut progress = self.progress.write().await;
-        let existing_baseline = (
-            progress.objects_total_count,
-            progress.objects_total_size,
-            progress.baseline_generation,
-            progress.progress_state,
-            progress.baseline_known,
-        );
-        let baseline_generation_mismatch =
-            state.baseline_known && existing_baseline.4 && state.baseline_generation != existing_baseline.2;
-        let use_persisted_baseline = state.baseline_known && !baseline_generation_mismatch;
-        progress.objects_scanned = state.processed_objects;
+        progress.objects_scanned = state.total_objects;
         progress.objects_healed = state.successful_objects;
         progress.objects_failed = state.failed_objects;
-        progress.skipped_objects = state.skipped_objects;
-        progress.skipped_new_versions = state.skipped_new_versions;
-        progress.skipped_ilm_expired = state.skipped_ilm_expired;
-        progress.bytes_processed = state.processed_bytes;
-        progress.counter_unknown = state.counter_unknown;
-        if use_persisted_baseline
-            || existing_baseline.0 > 0
-            || existing_baseline.1 > 0
-            || existing_baseline.2.is_some()
-            || existing_baseline.4
-        {
-            progress.objects_total_count = if use_persisted_baseline {
-                state.total_objects
-            } else {
-                existing_baseline.0
-            };
-            progress.objects_total_size = if use_persisted_baseline {
-                state.total_bytes
-            } else {
-                existing_baseline.1
-            };
-            progress.baseline_generation = if use_persisted_baseline {
-                state.baseline_generation
-            } else {
-                existing_baseline.2
-            };
-            progress.baseline_known = use_persisted_baseline
-                || existing_baseline.0 > 0
-                || existing_baseline.1 > 0
-                || existing_baseline.2.is_some()
-                || existing_baseline.4;
-        }
-        progress.progress_state = if use_persisted_baseline
-            || existing_baseline.0 > 0
-            || existing_baseline.1 > 0
-            || existing_baseline.2.is_some()
-            || existing_baseline.4
-        {
-            crate::heal::progress::HealProgressState::Running
-        } else {
-            crate::heal::progress::HealProgressState::Indeterminate
-        };
-        if baseline_generation_mismatch || state.counter_unknown {
-            progress.mark_unknown();
-        }
-        progress.ledger_complete = false;
-        progress.refresh_progress_percentage();
+        progress.bytes_processed = 0; // Resume state tracks object counts, not byte counters.
         progress.start_time = UNIX_EPOCH.checked_add(Duration::from_secs(state.start_time));
         progress.last_update_time = UNIX_EPOCH.checked_add(Duration::from_secs(state.last_update));
         progress.set_current_object(state.current_object.clone());
@@ -1506,8 +1264,8 @@ mod resume_loop_tests {
     };
     use crate::heal::progress::HealProgress;
     use crate::heal::resume::{
-        CheckpointManager, CheckpointObjectOutcome, RESUME_CHECKPOINT_FILE, ReplacementTargetIdentity, ResumeDeleteFailure,
-        ResumeManager, ResumeUtils, compose_key,
+        CheckpointManager, RESUME_CHECKPOINT_FILE, ReplacementTargetIdentity, ResumeDeleteFailure, ResumeManager, ResumeUtils,
+        compose_key,
     };
     use crate::heal::storage::{HealLifecycleExpiryContext, HealListItem, HealObjectInfo, HealStorageAPI};
     use crate::heal::storage_api::status::BucketInfo;
@@ -1647,7 +1405,6 @@ mod resume_loop_tests {
         list_include_lifecycle_object_info: Mutex<Vec<bool>>,
         replacement_target_identity_sequences: Mutex<VecDeque<Vec<ReplacementTargetIdentity>>>,
         fail_listing: AtomicBool,
-        fail_listing_buckets: Mutex<HashSet<String>>,
     }
 
     impl FakeStorage {
@@ -1684,9 +1441,6 @@ mod resume_loop_tests {
         fn fail_listing(&self) {
             self.fail_listing.store(true, Ordering::SeqCst);
         }
-        fn fail_bucket_listing(&self, bucket: &str) {
-            self.fail_listing_buckets.lock().unwrap().insert(bucket.to_string());
-        }
     }
 
     #[async_trait::async_trait]
@@ -1777,7 +1531,7 @@ mod resume_loop_tests {
         }
         async fn list_objects_for_heal_page(
             &self,
-            bucket: &str,
+            _bucket: &str,
             _prefix: &str,
             continuation_token: Option<&str>,
             include_lifecycle_object_info: bool,
@@ -1786,7 +1540,7 @@ mod resume_loop_tests {
                 .lock()
                 .unwrap()
                 .push(include_lifecycle_object_info);
-            if self.fail_listing.load(Ordering::SeqCst) || self.fail_listing_buckets.lock().unwrap().contains(bucket) {
+            if self.fail_listing.load(Ordering::SeqCst) {
                 return Err(Error::other("injected listing failure"));
             }
             let key = continuation_token.map(str::to_string);
@@ -2164,49 +1918,6 @@ mod resume_loop_tests {
         assert!(state.completed_buckets.is_empty(), "the failed bucket must remain resumable");
     }
 
-    #[tokio::test]
-    async fn bucket_failure_stops_before_a_later_bucket_checkpoint() {
-        let env = make_env().await;
-        let task_id = ResumeUtils::generate_task_id();
-        let buckets = vec!["a".to_string(), "b".to_string()];
-        let resume = ResumeManager::new(
-            env.healer.disk.clone(),
-            task_id.clone(),
-            "erasure_set".to_string(),
-            "pool_0_set_0".to_string(),
-            buckets.clone(),
-        )
-        .await
-        .unwrap();
-        let checkpoint = CheckpointManager::new(env.healer.disk.clone(), task_id.clone())
-            .await
-            .unwrap();
-        env.storage.fail_bucket_listing("a");
-        for _ in 0..3 {
-            assert!(resume.schedule_retry().await.unwrap());
-        }
-
-        env.healer
-            .execute_heal_with_resume(&buckets, "pool_0_set_0", &resume, &checkpoint)
-            .await
-            .expect_err("the first bucket failure must keep the pass incomplete");
-        let persisted = checkpoint.get_checkpoint().await;
-        assert_eq!(persisted.current_bucket_index, 0);
-        assert!(resume.get_state().await.completed_buckets.is_empty());
-
-        let resumed = ResumeManager::load_from_disk(env.healer.disk.clone(), &task_id)
-            .await
-            .unwrap();
-        let checkpoint = CheckpointManager::load_from_disk(env.healer.disk.clone(), &task_id)
-            .await
-            .unwrap();
-        env.healer
-            .execute_heal_with_resume(&buckets, "pool_0_set_0", &resumed, &checkpoint)
-            .await
-            .expect_err("recovery must retry the earlier failed bucket");
-        assert!(!resumed.get_state().await.completed);
-    }
-
     #[tokio::test]
     async fn completed_resume_state_is_not_selected_for_a_new_heal() {
         let env = make_env().await;
@@ -2396,175 +2107,8 @@ mod resume_loop_tests {
         let mut names: Vec<String> = env.storage.calls().into_iter().map(|(n, _)| n).collect();
         names.sort();
         assert_eq!(names, vec!["a", "b", "c", "d"], "every object exactly once, none dropped/doubled");
-        // Keep the final page cursor until the outer loop durably completes the
-        // bucket, so a crash can replay only this page against its identities.
-        assert_eq!(env.resume.resume_cursor().await, Some("t1".to_string()));
-    }
-
-    #[tokio::test]
-    async fn persisted_failure_waits_for_the_bounded_retry_after_page_replay() {
-        let env = make_env().await;
-        env.storage.set_page(
-            None,
-            Page {
-                items: vec![item("object", Some("v1"), false)],
-                next: None,
-                truncated: false,
-            },
-        );
-        env.checkpoint
-            .record_object_outcome(
-                compose_key("object", Some("v1")),
-                CheckpointObjectOutcome::Failed,
-                0,
-                1,
-                0,
-                0,
-                0,
-                0,
-                false,
-            )
-            .await
-            .unwrap();
-        env.checkpoint.advance_page(0, 1).await.unwrap();
-
-        let resumed = ResumeManager::load_from_disk(env.healer.disk.clone(), &env.task_id)
-            .await
-            .unwrap();
-        let checkpoint = CheckpointManager::load_from_disk(env.healer.disk.clone(), &env.task_id)
-            .await
-            .unwrap();
-
-        env.healer
-            .execute_heal_with_resume(&["b".to_string()], "pool_0_set_0", &resumed, &checkpoint)
-            .await
-            .expect_err("the persisted failure must schedule a bounded retry");
-        assert!(
-            env.storage.calls().is_empty(),
-            "the failed identity must not be repeated in the same pass"
-        );
-
-        env.healer
-            .execute_heal_with_resume(&["b".to_string()], "pool_0_set_0", &resumed, &checkpoint)
-            .await
-            .expect("the bounded retry must heal the object");
-        assert_eq!(env.storage.calls(), vec![("object".to_string(), Some("v1".to_string()))]);
-        let state = resumed.get_state().await;
-        assert_eq!(state.successful_objects, 1);
-        assert_eq!(state.failed_objects, 0);
-    }
-
-    #[tokio::test]
-    async fn final_page_crash_replays_only_the_retained_page_identities() {
-        let env = make_env().await;
-        env.storage.set_page(
-            None,
-            Page {
-                items: vec![item("first", Some("v1"), false)],
-                next: Some("final-page".to_string()),
-                truncated: true,
-            },
-        );
-        env.storage.set_page(
-            Some("final-page"),
-            Page {
-                items: vec![item("last", Some("v1"), false)],
-                next: None,
-                truncated: false,
-            },
-        );
-
-        let (processed, successful, failed, skipped, result) = run(&env).await;
-        result.expect("the bucket pass must finish before the simulated crash");
-        assert_eq!((processed, successful, failed, skipped), (2, 2, 0, 0));
-
-        let resumed = ResumeManager::load_from_disk(env.healer.disk.clone(), &env.task_id)
-            .await
-            .unwrap();
-        let checkpoint = CheckpointManager::load_from_disk(env.healer.disk.clone(), &env.task_id)
-            .await
-            .unwrap();
-        env.healer
-            .execute_heal_with_resume(&["b".to_string()], "pool_0_set_0", &resumed, &checkpoint)
-            .await
-            .expect("the retained final-page ledger must make recovery exact");
-
-        assert_eq!(
-            env.storage.calls(),
-            vec![
-                ("first".to_string(), Some("v1".to_string())),
-                ("last".to_string(), Some("v1".to_string()))
-            ]
-        );
-        let state = resumed.get_state().await;
-        assert_eq!(state.successful_objects, 2);
-        assert_eq!(state.processed_objects, 2);
-    }
-
-    #[tokio::test]
-    async fn truncated_page_without_token_retains_its_replay_ledger() {
-        let env = make_env().await;
-        env.storage.set_page(
-            None,
-            Page {
-                items: vec![item("object", Some("v1"), false)],
-                next: None,
-                truncated: true,
-            },
-        );
-
-        let (processed, successful, failed, skipped, result) = run(&env).await;
-        result.expect("the tokenless truncated page is a terminal page");
-        assert_eq!((processed, successful, failed, skipped), (1, 1, 0, 0));
-
-        let resumed = ResumeManager::load_from_disk(env.healer.disk.clone(), &env.task_id)
-            .await
-            .unwrap();
-        let checkpoint = CheckpointManager::load_from_disk(env.healer.disk.clone(), &env.task_id)
-            .await
-            .unwrap();
-        env.healer
-            .execute_heal_with_resume(&["b".to_string()], "pool_0_set_0", &resumed, &checkpoint)
-            .await
-            .expect("terminal-page recovery must not replay a durable identity");
-
-        assert_eq!(env.storage.calls(), vec![("object".to_string(), Some("v1".to_string()))]);
-        let state = resumed.get_state().await;
-        assert_eq!(state.successful_objects, 1);
-        assert_eq!(state.processed_objects, 1);
-    }
-
-    #[tokio::test]
-    async fn completed_bucket_reconciles_its_final_page_checkpoint_after_crash() {
-        let env = make_env().await;
-        env.storage.set_page(
-            None,
-            Page {
-                items: vec![item("object", Some("v1"), false)],
-                next: None,
-                truncated: false,
-            },
-        );
-
-        let (_, _, _, _, result) = run(&env).await;
-        result.expect("the bucket pass must finish before the simulated crash");
-        env.resume.complete_bucket("b").await.unwrap();
-
-        let resumed = ResumeManager::load_from_disk(env.healer.disk.clone(), &env.task_id)
-            .await
-            .unwrap();
-        let checkpoint = CheckpointManager::load_from_disk(env.healer.disk.clone(), &env.task_id)
-            .await
-            .unwrap();
-        env.healer
-            .execute_heal_with_resume(&["b".to_string()], "pool_0_set_0", &resumed, &checkpoint)
-            .await
-            .expect("recovery must finish the checkpoint transition without replaying the bucket");
-
-        assert_eq!(env.storage.calls(), vec![("object".to_string(), Some("v1".to_string()))]);
-        let checkpoint = checkpoint.get_checkpoint().await;
-        assert_eq!(checkpoint.current_bucket_index, 1);
-        assert!(checkpoint.processed_objects.is_empty());
+        // Final page not truncated => cursor cleared.
+        assert_eq!(env.resume.resume_cursor().await, None);
     }
 
     #[tokio::test]

crates/heal/src/heal/manager.rs

@@ -2012,44 +2012,16 @@ impl HealManager {
         }
 
         let mut snapshot = HealProgress::default();
-        let mut has_object_sweep = false;
-        let mut all_object_baselines_known = true;
-        let mut counter_overflow = false;
-        let mut stage_current = 0_u64;
-        let mut stage_total = 0_u64;
         for task in active_tasks {
             let progress = task.get_progress().await;
-            let object_sweep = matches!(progress.kind, crate::heal::progress::HealProgressKind::ObjectSweep);
-            has_object_sweep |= object_sweep;
-            if object_sweep {
-                all_object_baselines_known &= progress.baseline_known;
-            }
-            counter_overflow |=
-                progress.counter_unknown || matches!(progress.progress_state, crate::heal::progress::HealProgressState::Unknown);
-            match stage_current.checked_add(progress.stage_current) {
-                Some(sum) => stage_current = sum,
-                None => counter_overflow = true,
-            }
-            match stage_total.checked_add(progress.stage_total) {
-                Some(sum) => stage_total = sum,
-                None => counter_overflow = true,
-            }
-            for (target, value) in [
-                (&mut snapshot.objects_scanned, progress.objects_scanned),
-                (&mut snapshot.objects_healed, progress.objects_healed),
-                (&mut snapshot.objects_failed, progress.objects_failed),
-                (&mut snapshot.skipped_objects, progress.skipped_objects),
-                (&mut snapshot.skipped_new_versions, progress.skipped_new_versions),
-                (&mut snapshot.skipped_ilm_expired, progress.skipped_ilm_expired),
-                (&mut snapshot.objects_total_count, progress.objects_total_count),
-                (&mut snapshot.objects_total_size, progress.objects_total_size),
-                (&mut snapshot.bytes_processed, progress.bytes_processed),
-            ] {
-                match target.checked_add(value) {
-                    Some(sum) => *target = sum,
-                    None => counter_overflow = true,
-                }
-            }
+            snapshot.objects_scanned = snapshot.objects_scanned.saturating_add(progress.objects_scanned);
+            snapshot.objects_healed = snapshot.objects_healed.saturating_add(progress.objects_healed);
+            snapshot.objects_failed = snapshot.objects_failed.saturating_add(progress.objects_failed);
+            snapshot.skipped_new_versions = snapshot.skipped_new_versions.saturating_add(progress.skipped_new_versions);
+            snapshot.skipped_ilm_expired = snapshot.skipped_ilm_expired.saturating_add(progress.skipped_ilm_expired);
+            snapshot.objects_total_count = snapshot.objects_total_count.saturating_add(progress.objects_total_count);
+            snapshot.objects_total_size = snapshot.objects_total_size.saturating_add(progress.objects_total_size);
+            snapshot.bytes_processed = snapshot.bytes_processed.saturating_add(progress.bytes_processed);
             snapshot.start_time = match (snapshot.start_time, progress.start_time) {
                 (Some(current), Some(next)) => Some(current.min(next)),
                 (None, next) => next,
@@ -2064,36 +2036,7 @@ impl HealManager {
                 snapshot.current_object = progress.current_object;
             }
         }
-        snapshot.kind = if has_object_sweep {
-            crate::heal::progress::HealProgressKind::ObjectSweep
-        } else {
-            crate::heal::progress::HealProgressKind::Stage
-        };
-        snapshot.stage_current = stage_current;
-        snapshot.stage_total = stage_total;
-        snapshot.baseline_known = has_object_sweep && all_object_baselines_known;
-        snapshot.progress_state = if counter_overflow {
-            crate::heal::progress::HealProgressState::Unknown
-        } else if has_object_sweep && !all_object_baselines_known {
-            crate::heal::progress::HealProgressState::Indeterminate
-        } else if has_object_sweep {
-            crate::heal::progress::HealProgressState::Running
-        } else if stage_total == 0 {
-            crate::heal::progress::HealProgressState::Indeterminate
-        } else {
-            crate::heal::progress::HealProgressState::Running
-        };
-        if counter_overflow {
-            snapshot.progress_percentage = 0.0;
-        } else if !has_object_sweep {
-            snapshot.progress_percentage = if stage_total == 0 {
-                0.0
-            } else {
-                ((stage_current as f64 / stage_total as f64) * 100.0).min(99.999)
-            };
-        } else {
-            snapshot.refresh_progress_percentage();
-        }
+        snapshot.refresh_progress_percentage();
         snapshot.refresh_estimated_completion_time();
         Some(snapshot)
     }

crates/heal/src/heal/progress.rs

@@ -15,70 +15,15 @@
 use serde::{Deserialize, Serialize};
 use std::time::{Duration, SystemTime};
 
-pub(crate) fn increment_counter(counter: &mut u64) -> bool {
-    match counter.checked_add(1) {
-        Some(next) => {
-            *counter = next;
-            true
-        }
-        None => {
-            *counter = u64::MAX;
-            false
-        }
-    }
-}
-
-pub(crate) fn add_bytes(total: &mut u64, amount: u64) -> bool {
-    match total.checked_add(amount) {
-        Some(next) => {
-            *total = next;
-            true
-        }
-        None => {
-            *total = u64::MAX;
-            false
-        }
-    }
-}
-
-#[derive(Debug, Default, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
-#[serde(rename_all = "camelCase")]
-pub enum HealProgressKind {
-    #[default]
-    Unknown,
-    Stage,
-    ObjectSweep,
-}
-
-/// Whether the object ledger can produce a meaningful percentage.
-///
-/// A zero-valued baseline is not a completed scan: it means that no complete
-/// usage snapshot was available.  Keep this state explicit so callers do not
-/// mistake the legacy `0.0` wire value for a measured zero-percent result.
-#[derive(Debug, Default, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
-#[serde(rename_all = "camelCase")]
-pub enum HealProgressState {
-    #[default]
-    Unknown,
-    Indeterminate,
-    Running,
-    Completed,
-}
-
 #[derive(Debug, Default, Clone, Serialize, Deserialize)]
 #[serde(rename_all = "camelCase")]
 pub struct HealProgress {
-    #[serde(default)]
-    pub kind: HealProgressKind,
     /// Objects scanned
     pub objects_scanned: u64,
     /// Objects healed
     pub objects_healed: u64,
     /// Objects failed
     pub objects_failed: u64,
-    /// Versions deferred for a later retry pass.
-    #[serde(default)]
-    pub skipped_objects: u64,
     /// Versions skipped because they were written after this heal started
     pub skipped_new_versions: u64,
     /// Versions skipped because lifecycle already selected them for expiry
@@ -99,38 +44,11 @@ pub struct HealProgress {
     pub last_update_time: Option<SystemTime>,
     /// Estimated completion time
     pub estimated_completion_time: Option<SystemTime>,
-    /// Current stage number. Stage updates are intentionally independent from
-    /// the object ledger below.
-    #[serde(default)]
-    pub stage_current: u64,
-    /// Number of stages in the current task.
-    #[serde(default)]
-    pub stage_total: u64,
-    /// Explicitly distinguishes a missing usage baseline from measured 0%.
-    #[serde(default)]
-    pub progress_state: HealProgressState,
-    /// True only after the task's durable completion ledger was committed.
-    #[serde(default)]
-    pub ledger_complete: bool,
-    /// Generation of the usage snapshot used for the baseline, if available.
-    #[serde(default)]
-    pub baseline_generation: Option<u64>,
-    /// Whether the baseline was explicitly observed.  This is separate from
-    /// the counters so a known empty scope (0 objects, 0 bytes) is not
-    /// confused with a legacy snapshot that omitted the baseline fields.
-    #[serde(default)]
-    pub baseline_known: bool,
-    /// Internal telemetry fence set when an aggregate counter overflows or
-    /// becomes inconsistent.  It prevents a later refresh from fabricating a
-    /// percentage from the poisoned values.
-    #[serde(default)]
-    pub counter_unknown: bool,
 }
 
 impl HealProgress {
     pub fn new() -> Self {
         Self {
-            kind: HealProgressKind::Unknown,
             start_time: Some(SystemTime::now()),
             last_update_time: Some(SystemTime::now()),
             ..Default::default()
@@ -138,87 +56,12 @@ impl HealProgress {
     }
 
     pub fn update_progress(&mut self, scanned: u64, healed: u64, failed: u64, bytes: u64) {
-        self.update_object_sweep_progress(scanned, healed, failed, bytes);
-    }
-
-    pub fn update_object_sweep_progress(&mut self, scanned: u64, healed: u64, failed: u64, bytes: u64) {
-        self.kind = HealProgressKind::ObjectSweep;
         self.objects_scanned = scanned;
         self.objects_healed = healed;
         self.objects_failed = failed;
         self.bytes_processed = bytes;
         self.last_update_time = Some(SystemTime::now());
 
-        let explicit_skipped = match self.skipped_new_versions.checked_add(self.skipped_ilm_expired) {
-            Some(value) => value,
-            None => {
-                self.mark_unknown();
-                0
-            }
-        };
-        let skipped = healed
-            .checked_add(failed)
-            .and_then(|value| value.checked_add(explicit_skipped))
-            .and_then(|value| scanned.checked_sub(value))
-            .unwrap_or(0);
-        self.update_object_progress(scanned, healed, failed, skipped, bytes);
-    }
-
-    /// Update task stage progress without modifying object counters.
-    pub fn update_stage(&mut self, current: u64, total: u64) {
-        let object_sweep_active = matches!(self.kind, HealProgressKind::ObjectSweep);
-        if !object_sweep_active {
-            self.kind = HealProgressKind::Stage;
-        }
-        self.ledger_complete = false;
-        self.stage_current = current.min(total);
-        self.stage_total = total;
-        if object_sweep_active {
-            self.last_update_time = Some(SystemTime::now());
-            self.refresh_progress_percentage();
-            return;
-        }
-        self.progress_state = if total == 0 {
-            HealProgressState::Indeterminate
-        } else {
-            HealProgressState::Running
-        };
-        self.progress_percentage = if total == 0 {
-            0.0
-        } else {
-            (current as f64 / total as f64 * 100.0).min(100.0)
-        };
-        self.last_update_time = Some(SystemTime::now());
-    }
-
-    /// Update the disjoint object ledger. `scanned` is the number of terminal
-    /// object outcomes and must equal healed + failed + deferred skipped plus
-    /// the two terminal skip classes. Overflow is a corrupt/unknown counter
-    /// state, not a reason to abort a completed heal.
-    pub fn update_object_progress(&mut self, scanned: u64, healed: u64, failed: u64, skipped: u64, bytes: u64) {
-        self.kind = HealProgressKind::ObjectSweep;
-        // `skipped` is the transient/deferred class.  The two explicit skip
-        // counters are terminal classifications too, so include them in the
-        // same ledger without making callers maintain a second aggregate.
-        let outcomes = healed
-            .checked_add(failed)
-            .and_then(|value| value.checked_add(skipped))
-            .and_then(|value| value.checked_add(self.skipped_new_versions))
-            .and_then(|value| value.checked_add(self.skipped_ilm_expired));
-        self.objects_scanned = scanned;
-        self.objects_healed = healed;
-        self.objects_failed = failed;
-        self.skipped_objects = skipped;
-        self.bytes_processed = bytes;
-        self.last_update_time = Some(SystemTime::now());
-        self.ledger_complete = false;
-        if outcomes != Some(scanned) {
-            // Telemetry corruption must not abort a heal.  Preserve the
-            // counters for diagnostics, but do not derive a percentage from a
-            // double-counted or overflowing ledger.
-            self.mark_unknown();
-            return;
-        }
         self.refresh_progress_percentage();
         self.refresh_estimated_completion_time();
     }
@@ -226,88 +69,50 @@ impl HealProgress {
     pub fn set_total_baseline(&mut self, objects_total_count: u64, objects_total_size: u64) {
         self.objects_total_count = objects_total_count;
         self.objects_total_size = objects_total_size;
-        self.baseline_known = true;
         self.last_update_time = Some(SystemTime::now());
         self.refresh_progress_percentage();
         self.refresh_estimated_completion_time();
     }
 
-    pub fn set_total_baseline_with_generation(&mut self, objects_total_count: u64, objects_total_size: u64, generation: u64) {
-        self.baseline_generation = Some(generation);
-        self.set_total_baseline(objects_total_count, objects_total_size);
-    }
-
     pub fn record_skipped_new_version(&mut self) {
-        let Some(next) = self.skipped_new_versions.checked_add(1) else {
-            self.mark_unknown();
-            return;
-        };
-        self.skipped_new_versions = next;
+        self.skipped_new_versions = self.skipped_new_versions.saturating_add(1);
         self.last_update_time = Some(SystemTime::now());
         self.refresh_progress_percentage();
         self.refresh_estimated_completion_time();
     }
 
     pub fn record_skipped_ilm_expired(&mut self) {
-        let Some(next) = self.skipped_ilm_expired.checked_add(1) else {
-            self.mark_unknown();
-            return;
-        };
-        self.skipped_ilm_expired = next;
+        self.skipped_ilm_expired = self.skipped_ilm_expired.saturating_add(1);
         self.last_update_time = Some(SystemTime::now());
         self.refresh_progress_percentage();
         self.refresh_estimated_completion_time();
     }
 
-    fn completed_for_baseline(&self) -> Option<u64> {
+    fn completed_for_baseline(&self) -> u64 {
         self.objects_healed
-            .checked_add(self.objects_failed)?
-            .checked_add(self.skipped_objects)?
-            .checked_add(self.skipped_new_versions)?
-            .checked_add(self.skipped_ilm_expired)
+            .saturating_add(self.objects_failed)
+            .saturating_add(self.skipped_new_versions)
+            .saturating_add(self.skipped_ilm_expired)
     }
 
     pub(crate) fn refresh_progress_percentage(&mut self) {
-        if self.ledger_complete {
-            self.progress_state = HealProgressState::Completed;
-            self.progress_percentage = 100.0;
-            return;
-        }
-        if self.counter_unknown {
-            self.progress_state = HealProgressState::Unknown;
-            self.progress_percentage = 0.0;
-            return;
-        }
-        if !self.baseline_known {
-            self.progress_state = HealProgressState::Indeterminate;
-            self.progress_percentage = 0.0;
-            self.estimated_completion_time = None;
-            return;
-        }
         if self.objects_total_size > 0 {
             self.progress_percentage = ((self.bytes_processed as f64 / self.objects_total_size as f64) * 100.0).min(100.0);
-            self.progress_percentage = self.progress_percentage.min(99.999);
-            self.progress_state = HealProgressState::Running;
             return;
         }
         if self.objects_total_count > 0 {
-            let Some(completed) = self.completed_for_baseline() else {
-                self.progress_state = HealProgressState::Unknown;
-                self.progress_percentage = 0.0;
-                return;
-            };
+            let completed = self.completed_for_baseline();
             self.progress_percentage = ((completed as f64 / self.objects_total_count as f64) * 100.0).min(100.0);
-            self.progress_percentage = self.progress_percentage.min(99.999);
-            self.progress_state = HealProgressState::Running;
             return;
         }
-        if self.baseline_known {
-            self.progress_state = HealProgressState::Running;
-            self.progress_percentage = 0.0;
-            return;
+
+        let total = self
+            .objects_scanned
+            .saturating_add(self.objects_healed)
+            .saturating_add(self.objects_failed);
+        if total > 0 {
+            self.progress_percentage = (self.objects_healed as f64 / total as f64) * 100.0;
         }
-        self.progress_state = HealProgressState::Indeterminate;
-        self.progress_percentage = 0.0;
     }
 
     pub fn set_current_object(&mut self, object: Option<String>) {
@@ -320,11 +125,7 @@ impl HealProgress {
             self.estimated_completion_time = None;
             return;
         };
-        if self.is_completed()
-            || self.progress_percentage <= 0.0
-            || self.progress_percentage >= 100.0
-            || self.bytes_processed == 0
-        {
+        if self.is_completed() || !(0.0..100.0).contains(&self.progress_percentage) || self.bytes_processed == 0 {
             self.estimated_completion_time = None;
             return;
         }
@@ -341,39 +142,18 @@ impl HealProgress {
     }
 
     pub fn is_completed(&self) -> bool {
-        self.ledger_complete
-    }
-
-    /// Mark telemetry unknown while allowing the underlying heal operation to
-    /// continue.  This is used for corrupt/overflowing counters at the
-    /// observability boundary; it must never turn a successful heal into an
-    /// execution error.
-    pub fn mark_unknown(&mut self) {
-        self.counter_unknown = true;
-        self.progress_state = HealProgressState::Unknown;
-        self.ledger_complete = false;
-        self.progress_percentage = 0.0;
-        self.estimated_completion_time = None;
-        self.last_update_time = Some(SystemTime::now());
-    }
-
-    /// Mark the object ledger terminal only after the enclosing task has
-    /// committed all durable resume state and cleanup fences.
-    pub fn mark_completed(&mut self) {
-        let telemetry_unknown = self.counter_unknown || self.progress_state == HealProgressState::Unknown;
-        self.ledger_complete = true;
-        if !telemetry_unknown {
-            self.progress_state = HealProgressState::Completed;
+        if self.progress_percentage >= 100.0 {
+            return true;
         }
-        self.progress_percentage = 100.0;
-        self.last_update_time = Some(SystemTime::now());
-        self.estimated_completion_time = None;
+        if self.objects_total_count > 0 || self.objects_total_size > 0 {
+            return false;
+        }
+
+        self.objects_scanned > 0 && self.objects_healed.saturating_add(self.objects_failed) >= self.objects_scanned
     }
 
     pub fn get_success_rate(&self) -> f64 {
-        let Some(total) = self.objects_healed.checked_add(self.objects_failed) else {
-            return 0.0;
-        };
+        let total = self.objects_healed + self.objects_failed;
         if total > 0 {
             (self.objects_healed as f64 / total as f64) * 100.0
         } else {
@@ -450,7 +230,6 @@ mod tests {
         assert_eq!(progress.objects_scanned, 0);
         assert_eq!(progress.objects_healed, 0);
         assert_eq!(progress.objects_failed, 0);
-        assert_eq!(progress.skipped_objects, 0);
         assert_eq!(progress.skipped_new_versions, 0);
         assert_eq!(progress.skipped_ilm_expired, 0);
         assert_eq!(progress.objects_total_count, 0);
@@ -471,8 +250,10 @@ mod tests {
         assert_eq!(progress.objects_healed, 8);
         assert_eq!(progress.objects_failed, 2);
         assert_eq!(progress.bytes_processed, 1024);
-        assert_eq!(progress.progress_state, HealProgressState::Indeterminate);
-        assert_eq!(progress.progress_percentage, 0.0);
+        // Progress percentage should be calculated based on healed/total
+        // total = scanned + healed + failed = 10 + 8 + 2 = 20
+        // healed/total = 8/20 = 0.4 = 40%
+        assert!((progress.progress_percentage - 40.0).abs() < 0.001);
         assert!(progress.last_update_time.is_some());
     }
 
@@ -481,8 +262,7 @@ mod tests {
         let mut progress = HealProgress::new();
         progress.start_time = Some(SystemTime::now() - Duration::from_secs(10));
 
-        progress.set_total_baseline(100, 16384);
-        progress.update_progress(25, 25, 0, 4096);
+        progress.update_progress(100, 25, 0, 4096);
 
         let eta = progress
             .estimated_completion_time
@@ -495,7 +275,7 @@ mod tests {
         let mut progress = HealProgress::new();
         progress.set_total_baseline(10, 8192);
 
-        progress.update_progress(25, 25, 0, 4096);
+        progress.update_progress(100, 25, 0, 4096);
 
         assert!((progress.progress_percentage - 50.0).abs() < 0.001);
     }
@@ -505,7 +285,7 @@ mod tests {
         let mut progress = HealProgress::new();
         progress.set_total_baseline(10, 0);
 
-        progress.update_progress(5, 3, 2, 0);
+        progress.update_progress(100, 3, 2, 0);
 
         assert!((progress.progress_percentage - 50.0).abs() < 0.001);
     }
@@ -515,7 +295,7 @@ mod tests {
         let mut progress = HealProgress::new();
         progress.set_total_baseline(10, 0);
 
-        progress.update_progress(5, 3, 2, 0);
+        progress.update_progress(100, 3, 2, 0);
         progress.record_skipped_new_version();
 
         assert_eq!(progress.skipped_new_versions, 1);
@@ -556,8 +336,7 @@ mod tests {
     fn test_heal_progress_update_progress_all_healed() {
         let mut progress = HealProgress::new();
         // When scanned=0, healed=10, failed=0: total=10, progress = 10/10 = 100%
-        progress.update_progress(10, 10, 0, 2048);
-        progress.mark_completed();
+        progress.update_progress(0, 10, 0, 2048);
 
         // All healed, should be 100%
         assert!((progress.progress_percentage - 100.0).abs() < 0.001);
@@ -615,7 +394,6 @@ mod tests {
         assert_eq!(json["objectsScanned"], 10);
         assert_eq!(json["objectsHealed"], 8);
         assert_eq!(json["objectsFailed"], 2);
-        assert_eq!(json["skippedObjects"], 0);
         assert_eq!(json["skippedNewVersions"], 0);
         assert_eq!(json["skippedIlmExpired"], 0);
         assert_eq!(json["bytesProcessed"], 1024);
@@ -627,7 +405,6 @@ mod tests {
     fn test_heal_progress_is_completed_by_percentage() {
         let mut progress = HealProgress::new();
         progress.update_progress(10, 10, 0, 1024);
-        progress.mark_completed();
 
         assert!(progress.is_completed());
     }
@@ -638,7 +415,7 @@ mod tests {
         progress.objects_scanned = 10;
         progress.objects_healed = 8;
         progress.objects_failed = 2;
-        progress.mark_completed();
+        // healed + failed = 8 + 2 = 10 >= scanned = 10
         assert!(progress.is_completed());
     }
 
@@ -678,66 +455,6 @@ mod tests {
         assert!((progress.get_success_rate() - 100.0).abs() < 0.001);
     }
 
-    #[test]
-    fn single_object_progress_reaches_terminal_100() {
-        let mut progress = HealProgress::new();
-        progress.update_object_progress(1, 1, 0, 0, 128);
-        assert!(!progress.is_completed());
-        progress.mark_completed();
-        assert!(progress.is_completed());
-        assert_eq!(progress.progress_percentage, 100.0);
-    }
-
-    #[test]
-    fn progress_without_baseline_is_indeterminate() {
-        let mut progress = HealProgress::new();
-        progress.update_object_progress(1, 1, 0, 0, 128);
-        assert_eq!(progress.progress_state, HealProgressState::Indeterminate);
-        assert_eq!(progress.progress_percentage, 0.0);
-        assert!(progress.estimated_completion_time.is_none());
-    }
-
-    #[test]
-    fn progress_retry_is_exactly_once() {
-        let mut progress = HealProgress::new();
-        progress.set_total_baseline(1, 128);
-        progress.update_object_progress(1, 1, 0, 0, 128);
-        progress.update_object_progress(1, 1, 0, 0, 128);
-        assert_eq!(progress.objects_scanned, 1);
-        assert_eq!(progress.objects_healed, 1);
-        assert_eq!(progress.bytes_processed, 128);
-    }
-
-    #[test]
-    fn progress_never_triggers_cleanup_before_terminal_ledger_empty() {
-        let mut progress = HealProgress::new();
-        progress.progress_percentage = 100.0;
-        assert!(!progress.is_completed());
-        progress.mark_completed();
-        assert!(progress.is_completed());
-    }
-
-    #[test]
-    fn progress_counter_overflow_is_marked_unknown_without_aborting_completed_heal() {
-        let mut progress = HealProgress::new();
-        progress.update_object_progress(u64::MAX, u64::MAX, 1, 0, 0);
-        assert_eq!(progress.progress_state, HealProgressState::Unknown);
-        progress.mark_completed();
-        assert!(progress.is_completed());
-        assert_eq!(progress.progress_state, HealProgressState::Unknown);
-    }
-
-    #[test]
-    fn stage_updates_do_not_double_count_object_outcomes() {
-        let mut progress = HealProgress::new();
-        progress.update_object_progress(2, 1, 0, 1, 256);
-        progress.update_stage(3, 4);
-        assert_eq!(progress.kind, HealProgressKind::ObjectSweep);
-        assert_eq!(progress.objects_scanned, 2);
-        assert_eq!(progress.objects_healed, 1);
-        assert_eq!(progress.skipped_objects, 1);
-    }
-
     #[test]
     fn test_heal_statistics_new() {
         let stats = HealStatistics::new();

crates/heal/src/heal/resume.rs

@@ -31,7 +31,7 @@ mod checkpoint;
 mod replacement;
 mod utils;
 
-pub use checkpoint::{CheckpointManager, CheckpointObjectOutcome, ResumeCheckpoint};
+pub use checkpoint::{CheckpointManager, ResumeCheckpoint};
 pub(crate) use replacement::replacement_target_identities_match;
 use replacement::replacement_targets_match_identities;
 pub use replacement::{
@@ -340,12 +340,6 @@ pub struct ResumeState {
     pub failed_objects: u64,
     /// skipped objects
     pub skipped_objects: u64,
-    /// Terminal versions skipped because they were newer than the heal start.
-    #[serde(default)]
-    pub skipped_new_versions: u64,
-    /// Terminal versions handed to lifecycle expiry.
-    #[serde(default)]
-    pub skipped_ilm_expired: u64,
     /// current bucket
     pub current_bucket: Option<String>,
     /// current object
@@ -360,24 +354,6 @@ pub struct ResumeState {
     pub retry_count: u32,
     /// max retries
     pub max_retries: u32,
-    /// Bytes accounted by the object ledger; additive for old snapshots.
-    #[serde(default)]
-    pub processed_bytes: u64,
-    /// Total bytes from a complete usage snapshot, when available.
-    #[serde(default)]
-    pub total_bytes: u64,
-    /// Generation of the usage snapshot used for the baseline.
-    #[serde(default)]
-    pub baseline_generation: Option<u64>,
-    /// Whether the usage baseline is known. Missing in old snapshots means
-    /// indeterminate rather than a measured zero baseline.
-    #[serde(default)]
-    pub baseline_known: bool,
-    /// Persistent telemetry fence for counter/byte overflow or corruption.
-    /// It must survive a restart so a saturated snapshot is never presented as
-    /// a measured percentage on the next resume.
-    #[serde(default)]
-    pub counter_unknown: bool,
 }
 
 impl ResumeState {
@@ -401,8 +377,6 @@ impl ResumeState {
             successful_objects: 0,
             failed_objects: 0,
             skipped_objects: 0,
-            skipped_new_versions: 0,
-            skipped_ilm_expired: 0,
             current_bucket: None,
             current_object: None,
             completed_buckets: Vec::new(),
@@ -410,11 +384,6 @@ impl ResumeState {
             error_message: None,
             retry_count: 0,
             max_retries: 3,
-            processed_bytes: 0,
-            total_bytes: 0,
-            baseline_generation: None,
-            baseline_known: false,
-            counter_unknown: false,
         }
     }
 
@@ -443,39 +412,6 @@ impl ResumeState {
         self.last_update = SystemTime::now().duration_since(UNIX_EPOCH).unwrap_or_default().as_secs();
     }
 
-    pub fn update_progress_with_bytes(
-        &mut self,
-        processed: u64,
-        successful: u64,
-        failed: u64,
-        skipped: u64,
-        processed_bytes: u64,
-    ) {
-        self.update_progress(processed, successful, failed, skipped);
-        self.processed_bytes = processed_bytes;
-    }
-
-    pub fn set_skipped_version_counts(&mut self, new_versions: u64, ilm_expired: u64) {
-        self.skipped_new_versions = new_versions;
-        self.skipped_ilm_expired = ilm_expired;
-        self.last_update = SystemTime::now().duration_since(UNIX_EPOCH).unwrap_or_default().as_secs();
-    }
-
-    pub fn set_progress_baseline(&mut self, total_objects: u64, total_bytes: u64, generation: Option<u64>) {
-        self.total_objects = total_objects;
-        self.total_bytes = total_bytes;
-        self.baseline_generation = generation;
-        // This method is called only after a complete usage snapshot has been
-        // validated.  A complete but empty snapshot is still a known baseline.
-        self.baseline_known = true;
-        self.last_update = SystemTime::now().duration_since(UNIX_EPOCH).unwrap_or_default().as_secs();
-    }
-
-    pub fn mark_counter_unknown(&mut self) {
-        self.counter_unknown = true;
-        self.last_update = SystemTime::now().duration_since(UNIX_EPOCH).unwrap_or_default().as_secs();
-    }
-
     pub fn set_current_item(&mut self, bucket: Option<String>, object: Option<String>) {
         self.current_bucket = bucket;
         self.current_object = object;
@@ -501,7 +437,6 @@ impl ResumeState {
         if let Some(pos) = self.pending_buckets.iter().position(|b| b == bucket) {
             self.pending_buckets.remove(pos);
         }
-        self.resume_cursor = None;
         self.last_update = SystemTime::now().duration_since(UNIX_EPOCH).unwrap_or_default().as_secs();
     }
 
@@ -519,10 +454,6 @@ impl ResumeState {
         self.successful_objects = 0;
         self.failed_objects = 0;
         self.skipped_objects = 0;
-        self.skipped_new_versions = 0;
-        self.skipped_ilm_expired = 0;
-        self.processed_bytes = 0;
-        self.counter_unknown = false;
         self.completed = false;
         // A retry re-scans every bucket from the beginning, so the version
         // cursor must be cleared too — otherwise the retry would resume mid-scan.
@@ -545,28 +476,14 @@ impl ResumeState {
     }
 
     pub fn get_progress_percentage(&self) -> f64 {
-        if self.completed {
-            return 100.0;
-        }
-        if self.counter_unknown {
-            return 0.0;
-        }
-        if !self.baseline_known {
-            return 0.0;
-        }
-        if self.total_bytes > 0 {
-            return ((self.processed_bytes as f64 / self.total_bytes as f64) * 100.0).min(99.999);
-        }
         if self.total_objects == 0 {
             return 0.0;
         }
-        ((self.processed_objects as f64 / self.total_objects as f64) * 100.0).min(99.999)
+        (self.processed_objects as f64 / self.total_objects as f64) * 100.0
     }
 
     pub fn get_success_rate(&self) -> f64 {
-        let Some(total) = self.successful_objects.checked_add(self.failed_objects) else {
-            return 0.0;
-        };
+        let total = self.successful_objects + self.failed_objects;
         if total == 0 {
             return 0.0;
         }
@@ -837,14 +754,6 @@ impl ResumeManager {
             state.successful_objects = 0;
             state.failed_objects = 0;
             state.skipped_objects = 0;
-            state.skipped_new_versions = 0;
-            state.skipped_ilm_expired = 0;
-            state.processed_bytes = 0;
-            state.total_objects = 0;
-            state.total_bytes = 0;
-            state.baseline_generation = None;
-            state.baseline_known = false;
-            state.counter_unknown = false;
             state.completed = false;
             state.completed_buckets.clear();
             state.schema_version = CURRENT_RESUME_SCHEMA;
@@ -929,41 +838,6 @@ impl ResumeManager {
         self.save_state_throttled().await
     }
 
-    pub async fn update_progress_with_bytes(
-        &self,
-        processed: u64,
-        successful: u64,
-        failed: u64,
-        skipped: u64,
-        processed_bytes: u64,
-    ) -> Result<()> {
-        let mut state = self.state.write().await;
-        state.update_progress_with_bytes(processed, successful, failed, skipped, processed_bytes);
-        drop(state);
-        self.save_state_throttled().await
-    }
-
-    pub async fn set_progress_baseline(&self, total_objects: u64, total_bytes: u64, generation: Option<u64>) -> Result<()> {
-        let mut state = self.state.write().await;
-        state.set_progress_baseline(total_objects, total_bytes, generation);
-        drop(state);
-        self.save_state_throttled().await
-    }
-
-    pub async fn mark_counter_unknown(&self) -> Result<()> {
-        let mut state = self.state.write().await;
-        state.mark_counter_unknown();
-        drop(state);
-        self.save_state().await
-    }
-
-    pub async fn set_skipped_version_counts(&self, new_versions: u64, ilm_expired: u64) -> Result<()> {
-        let mut state = self.state.write().await;
-        state.set_skipped_version_counts(new_versions, ilm_expired);
-        drop(state);
-        self.save_state_throttled().await
-    }
-
     /// Set current item. Called once per healed object, so persistence is
     /// throttled: the in-memory state always updates, but the snapshot is only
     /// written every `PERSIST_EVERY_MUTATIONS` calls or `PERSIST_INTERVAL`.
@@ -1008,7 +882,7 @@ impl ResumeManager {
         let mut state = self.state.write().await;
         state.complete_bucket(bucket);
         drop(state);
-        self.save_state().await
+        self.save_state_throttled().await
     }
 
     /// mark task completed

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

@@ -34,13 +34,6 @@ const EVENT_HEAL_CHECKPOINT_STATE: &str = "heal_checkpoint_state";
 /// to the new `compose_key` identities, so a stale checkpoint is discarded.
 pub(super) const CURRENT_CHECKPOINT_SCHEMA: u32 = 5;
 
-#[derive(Debug, Clone, Copy)]
-pub enum CheckpointObjectOutcome {
-    Processed,
-    Failed,
-    Skipped,
-}
-
 /// resume checkpoint
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct ResumeCheckpoint {
@@ -64,30 +57,6 @@ pub struct ResumeCheckpoint {
     pub failed_objects: HashSet<String>,
     /// skipped objects
     pub skipped_objects: HashSet<String>,
-    /// Aggregate object ledger counters restored alongside the dedup sets.
-    #[serde(default)]
-    pub successful_objects: u64,
-    #[serde(default)]
-    pub failed_object_count: u64,
-    #[serde(default)]
-    pub skipped_object_count: u64,
-    #[serde(default)]
-    pub skipped_new_versions: u64,
-    #[serde(default)]
-    pub skipped_ilm_expired: u64,
-    #[serde(default)]
-    pub processed_bytes: u64,
-    #[serde(default)]
-    pub total_objects: u64,
-    #[serde(default)]
-    pub total_bytes: u64,
-    #[serde(default)]
-    pub baseline_generation: Option<u64>,
-    #[serde(default)]
-    pub baseline_known: bool,
-    /// Persistent telemetry fence for counter/byte overflow or corruption.
-    #[serde(default)]
-    pub counter_unknown: bool,
 }
 
 impl ResumeCheckpoint {
@@ -101,17 +70,6 @@ impl ResumeCheckpoint {
             processed_objects: HashSet::new(),
             failed_objects: HashSet::new(),
             skipped_objects: HashSet::new(),
-            successful_objects: 0,
-            failed_object_count: 0,
-            skipped_object_count: 0,
-            skipped_new_versions: 0,
-            skipped_ilm_expired: 0,
-            processed_bytes: 0,
-            total_objects: 0,
-            total_bytes: 0,
-            baseline_generation: None,
-            baseline_known: false,
-            counter_unknown: false,
         }
     }
 
@@ -133,34 +91,6 @@ impl ResumeCheckpoint {
         self.skipped_objects.insert(object);
     }
 
-    pub fn update_progress(&mut self, successful: u64, failed: u64, skipped: u64, bytes: u64) {
-        self.successful_objects = successful;
-        self.failed_object_count = failed;
-        self.skipped_object_count = skipped;
-        self.processed_bytes = bytes;
-    }
-
-    pub fn set_progress_baseline(&mut self, total_objects: u64, total_bytes: u64, generation: Option<u64>) {
-        self.total_objects = total_objects;
-        self.total_bytes = total_bytes;
-        self.baseline_generation = generation;
-        // The caller has already validated that this is a complete snapshot;
-        // preserve the distinction between a known empty scope and an old
-        // checkpoint that omitted all baseline fields.
-        self.baseline_known = true;
-    }
-
-    pub fn mark_counter_unknown(&mut self) {
-        self.counter_unknown = true;
-        self.checkpoint_time = SystemTime::now().duration_since(UNIX_EPOCH).unwrap_or_default().as_secs();
-    }
-
-    pub fn set_skipped_version_counts(&mut self, new_versions: u64, ilm_expired: u64) {
-        self.skipped_new_versions = new_versions;
-        self.skipped_ilm_expired = ilm_expired;
-        self.checkpoint_time = SystemTime::now().duration_since(UNIX_EPOCH).unwrap_or_default().as_secs();
-    }
-
     /// Advance past a fully-processed page: objects below `object_index` are
     /// skipped by position on resume, so the per-object sets no longer need
     /// their entries and would otherwise grow with the whole bucket.
@@ -177,17 +107,6 @@ impl ResumeCheckpoint {
         self.update_position(0, 0);
         self.processed_objects.clear();
         self.skipped_objects.clear();
-        self.successful_objects = 0;
-        self.failed_object_count = 0;
-        self.skipped_object_count = 0;
-        self.skipped_new_versions = 0;
-        self.skipped_ilm_expired = 0;
-        self.processed_bytes = 0;
-        self.total_objects = 0;
-        self.total_bytes = 0;
-        self.baseline_generation = None;
-        self.baseline_known = false;
-        self.counter_unknown = false;
         self.failed_objects.clear();
     }
 }
@@ -266,17 +185,6 @@ impl CheckpointManager {
             checkpoint.processed_objects.clear();
             checkpoint.failed_objects.clear();
             checkpoint.skipped_objects.clear();
-            checkpoint.successful_objects = 0;
-            checkpoint.failed_object_count = 0;
-            checkpoint.skipped_object_count = 0;
-            checkpoint.skipped_new_versions = 0;
-            checkpoint.skipped_ilm_expired = 0;
-            checkpoint.processed_bytes = 0;
-            checkpoint.total_objects = 0;
-            checkpoint.total_bytes = 0;
-            checkpoint.baseline_generation = None;
-            checkpoint.baseline_known = false;
-            checkpoint.counter_unknown = false;
             checkpoint.current_bucket_index = 0;
             checkpoint.current_object_index = 0;
             checkpoint.schema_version = CURRENT_CHECKPOINT_SCHEMA;
@@ -317,7 +225,7 @@ impl CheckpointManager {
         self.save_checkpoint_throttled().await
     }
 
-    /// Persist a completed page position while retaining its identities.
+    /// Advance past a completed page and prune the per-object sets, then persist.
     pub async fn complete_page(&self, bucket_index: usize, object_index: usize) -> Result<()> {
         let mut checkpoint = self.checkpoint.write().await;
         checkpoint.complete_page(bucket_index, object_index);
@@ -325,35 +233,6 @@ impl CheckpointManager {
         self.save_checkpoint_throttled().await
     }
 
-    /// Persist the page position while retaining identities until the resume
-    /// cursor is durable.
-    pub async fn advance_page(&self, bucket_index: usize, object_index: usize) -> Result<()> {
-        let mut checkpoint = self.checkpoint.write().await;
-        checkpoint.update_position(bucket_index, object_index);
-        drop(checkpoint);
-        self.save_checkpoint().await
-    }
-
-    /// Remove the previous page's dedup identities only after its resume cursor
-    /// has been durably exposed.
-    pub async fn prune_completed_page(&self) -> Result<()> {
-        let mut checkpoint = self.checkpoint.write().await;
-        checkpoint.processed_objects.clear();
-        checkpoint.skipped_objects.clear();
-        checkpoint.failed_objects.clear();
-        drop(checkpoint);
-        self.save_checkpoint().await
-    }
-
-    /// Advance to the next bucket and clear the final page identities after the
-    /// resume state has durably recorded the completed bucket.
-    pub async fn complete_bucket(&self, next_bucket_index: usize) -> Result<()> {
-        let mut checkpoint = self.checkpoint.write().await;
-        checkpoint.complete_page(next_bucket_index, 0);
-        drop(checkpoint);
-        self.save_checkpoint().await
-    }
-
     /// Reset the checkpoint to the start of the scan for a retry, then persist.
     pub async fn reset_for_retry(&self) -> Result<()> {
         let mut checkpoint = self.checkpoint.write().await;
@@ -388,62 +267,6 @@ impl CheckpointManager {
         self.save_checkpoint_if_due().await
     }
 
-    /// Atomically persist an object's dedup identity with its aggregate result.
-    pub async fn record_object_outcome(
-        &self,
-        object: String,
-        outcome: CheckpointObjectOutcome,
-        successful: u64,
-        failed: u64,
-        skipped: u64,
-        bytes: u64,
-        skipped_new_versions: u64,
-        skipped_ilm_expired: u64,
-        counter_unknown: bool,
-    ) -> Result<()> {
-        let mut checkpoint = self.checkpoint.write().await;
-        match outcome {
-            CheckpointObjectOutcome::Processed => checkpoint.add_processed_object(object),
-            CheckpointObjectOutcome::Failed => checkpoint.add_failed_object(object),
-            CheckpointObjectOutcome::Skipped => checkpoint.add_skipped_object(object),
-        }
-        checkpoint.update_progress(successful, failed, skipped, bytes);
-        checkpoint.set_skipped_version_counts(skipped_new_versions, skipped_ilm_expired);
-        if counter_unknown {
-            checkpoint.mark_counter_unknown();
-        }
-        drop(checkpoint);
-        self.save_checkpoint_if_due().await
-    }
-
-    pub async fn update_progress(&self, successful: u64, failed: u64, skipped: u64, bytes: u64) -> Result<()> {
-        let mut checkpoint = self.checkpoint.write().await;
-        checkpoint.update_progress(successful, failed, skipped, bytes);
-        drop(checkpoint);
-        self.save_checkpoint_if_due().await
-    }
-
-    pub async fn set_progress_baseline(&self, total_objects: u64, total_bytes: u64, generation: Option<u64>) -> Result<()> {
-        let mut checkpoint = self.checkpoint.write().await;
-        checkpoint.set_progress_baseline(total_objects, total_bytes, generation);
-        drop(checkpoint);
-        self.save_checkpoint_throttled().await
-    }
-
-    pub async fn mark_counter_unknown(&self) -> Result<()> {
-        let mut checkpoint = self.checkpoint.write().await;
-        checkpoint.mark_counter_unknown();
-        drop(checkpoint);
-        self.save_checkpoint().await
-    }
-
-    pub async fn set_skipped_version_counts(&self, new_versions: u64, ilm_expired: u64) -> Result<()> {
-        let mut checkpoint = self.checkpoint.write().await;
-        checkpoint.set_skipped_version_counts(new_versions, ilm_expired);
-        drop(checkpoint);
-        self.save_checkpoint_throttled().await
-    }
-
     async fn save_checkpoint_if_due(&self) -> Result<()> {
         let should_save = self.throttle.lock().map(|mut throttle| throttle.record()).unwrap_or(true);
         if !should_save {

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

@@ -1296,7 +1296,6 @@ async fn test_resume_state_progress() {
     assert_eq!(progress, 0.0); // total_objects is 0
 
     state.total_objects = 100;
-    state.baseline_known = true;
     let progress = state.get_progress_percentage();
     assert_eq!(progress, 10.0);
 }
@@ -1476,40 +1475,6 @@ fn test_checkpoint_object_sets_dedupe_and_prune() {
     assert!(checkpoint.failed_objects.is_empty());
 }
 
-#[tokio::test]
-async fn checkpoint_page_commit_keeps_ledger_until_cursor_is_durable() {
-    let (_temp_dir, disk) = schema_test_disk().await;
-    let task_id = ResumeUtils::generate_task_id();
-    let checkpoint = CheckpointManager::new(disk.clone(), task_id.clone()).await.unwrap();
-
-    checkpoint
-        .record_object_outcome(
-            "bucket/object:v1".to_string(),
-            CheckpointObjectOutcome::Processed,
-            1,
-            0,
-            0,
-            128,
-            0,
-            0,
-            false,
-        )
-        .await
-        .unwrap();
-    checkpoint.advance_page(0, 1).await.unwrap();
-
-    let reloaded = CheckpointManager::load_from_disk(disk.clone(), &task_id).await.unwrap();
-    let snapshot = reloaded.get_checkpoint().await;
-    assert_eq!(snapshot.current_object_index, 1);
-    assert_eq!(snapshot.successful_objects, 1);
-    assert_eq!(snapshot.processed_bytes, 128);
-    assert!(snapshot.processed_objects.contains("bucket/object:v1"));
-
-    checkpoint.prune_completed_page().await.unwrap();
-    let reloaded = CheckpointManager::load_from_disk(disk, &task_id).await.unwrap();
-    assert!(reloaded.get_checkpoint().await.processed_objects.is_empty());
-}
-
 #[test]
 fn test_checkpoint_loads_legacy_vec_format() {
     // Checkpoints written before the HashSet migration stored the object
@@ -1674,120 +1639,6 @@ async fn current_normal_resume_schema_preserves_progress() {
     temp_dir.close().expect("remove schema test directory");
 }
 
-#[test]
-fn progress_checkpoint_restores_bytes_and_generation() {
-    let mut checkpoint = ResumeCheckpoint::new("progress-checkpoint".to_string());
-    checkpoint.set_progress_baseline(9, 4096, Some(77));
-    checkpoint.update_progress(4, 1, 2, 2048);
-    checkpoint.set_skipped_version_counts(3, 1);
-    checkpoint.mark_counter_unknown();
-
-    let restored: ResumeCheckpoint =
-        serde_json::from_slice(&serde_json::to_vec(&checkpoint).expect("serialize checkpoint")).expect("deserialize checkpoint");
-    assert_eq!(restored.processed_bytes, 2048);
-    assert_eq!(restored.total_objects, 9);
-    assert_eq!(restored.total_bytes, 4096);
-    assert_eq!(restored.baseline_generation, Some(77));
-    assert!(restored.baseline_known);
-    assert_eq!(restored.skipped_new_versions, 3);
-    assert_eq!(restored.skipped_ilm_expired, 1);
-    assert!(restored.counter_unknown);
-}
-
-#[test]
-fn old_progress_schema_migrates_missing_fields_to_unknown() {
-    let state = ResumeState::new(
-        "legacy-progress".to_string(),
-        "erasure_set".to_string(),
-        "pool_0_set_0".to_string(),
-        Vec::new(),
-    );
-    let mut value = serde_json::to_value(state).expect("serialize legacy-compatible state");
-    let object = value.as_object_mut().expect("state must be an object");
-    for field in [
-        "processed_bytes",
-        "total_bytes",
-        "baseline_generation",
-        "baseline_known",
-        "skipped_new_versions",
-        "skipped_ilm_expired",
-    ] {
-        object.remove(field);
-    }
-    object.insert("total_objects".to_string(), serde_json::json!(10));
-    object.insert("processed_objects".to_string(), serde_json::json!(5));
-    let restored: ResumeState = serde_json::from_value(value).expect("deserialize old progress state");
-    assert_eq!(restored.processed_bytes, 0);
-    assert_eq!(restored.total_bytes, 0);
-    assert_eq!(restored.baseline_generation, None);
-    assert!(!restored.baseline_known, "missing baseline must remain unknown");
-    assert_eq!(restored.get_progress_percentage(), 0.0);
-    assert_eq!(restored.skipped_new_versions, 0);
-    assert_eq!(restored.skipped_ilm_expired, 0);
-}
-
-#[test]
-fn progress_counter_unknown_survives_resume_round_trip() {
-    let mut state = ResumeState::new(
-        "overflow-progress".to_string(),
-        "erasure_set".to_string(),
-        "pool_0_set_0".to_string(),
-        Vec::new(),
-    );
-    state.mark_counter_unknown();
-
-    let restored: ResumeState =
-        serde_json::from_slice(&serde_json::to_vec(&state).expect("serialize resume state")).expect("deserialize resume state");
-    assert!(restored.counter_unknown);
-}
-
-#[tokio::test]
-async fn checkpoint_progress_survives_a_torn_resume_summary_write() {
-    let (_temp_dir, disk) = schema_test_disk().await;
-    let task_id = ResumeUtils::generate_task_id();
-    let _resume = ResumeManager::new(
-        disk.clone(),
-        task_id.clone(),
-        "erasure_set".to_string(),
-        "pool_0_set_0".to_string(),
-        vec!["bucket".to_string()],
-    )
-    .await
-    .expect("resume state should persist");
-    let checkpoint = CheckpointManager::new(disk.clone(), task_id.clone())
-        .await
-        .expect("checkpoint should persist");
-
-    // This is the ordering used by the erasure-set loop: the checkpoint is
-    // durable before the summary write.  Stop here to model a crash in the
-    // inter-store window and verify that the recovery authority retains the
-    // telemetry fence and bytes.
-    checkpoint
-        .update_progress(3, 0, 0, 1024)
-        .await
-        .expect("checkpoint progress should persist");
-    checkpoint.mark_counter_unknown().await.expect("unknown fence should persist");
-    checkpoint
-        .update_position(0, 3)
-        .await
-        .expect("checkpoint position should persist");
-
-    let restored_checkpoint = CheckpointManager::load_from_disk(disk.clone(), &task_id)
-        .await
-        .expect("checkpoint should reload")
-        .get_checkpoint()
-        .await;
-    let restored_resume = ResumeManager::load_from_disk(disk, &task_id)
-        .await
-        .expect("resume summary should reload")
-        .get_state()
-        .await;
-    assert!(restored_checkpoint.counter_unknown);
-    assert_eq!(restored_checkpoint.processed_bytes, 1024);
-    assert_eq!(restored_checkpoint.current_object_index, 3);
-    assert!(!restored_resume.counter_unknown, "summary is intentionally the torn/older store");
-}
-
 #[tokio::test]
 async fn future_resume_and_checkpoint_schemas_are_rejected() {
     let (temp_dir, disk) = schema_test_disk().await;

crates/heal/src/heal/storage.rs

@@ -19,8 +19,6 @@ use base64::engine::general_purpose::URL_SAFE_NO_PAD;
 use rustfs_common::heal_channel::{HealOpts, HealScanMode};
 use rustfs_madmin::heal_commands::HealResultItem;
 use serde::{Deserialize, Serialize};
-use std::collections::hash_map::DefaultHasher;
-use std::hash::{Hash, Hasher};
 use std::sync::Arc;
 use tracing::{debug, error, warn};
 
@@ -36,9 +34,6 @@ pub use super::{HealObjectInfo, HealObjectOptions, HealPutObjReader};
 pub struct HealBucketUsageBaseline {
     pub objects_count: u64,
     pub bytes: u64,
-    /// Stable identity of the validated usage snapshot and selected scope.
-    /// `None` is retained for test/legacy providers that cannot expose one.
-    pub generation: Option<u64>,
 }
 
 pub struct HealLifecycleExpiryContext {
@@ -790,30 +785,11 @@ impl HealStorageAPI for ECStoreHealStorage {
         let mut baseline = HealBucketUsageBaseline::default();
         for bucket in buckets {
             if let Some(usage) = info.buckets_usage.get(bucket) {
-                baseline.objects_count = match baseline.objects_count.checked_add(usage.objects_count) {
-                    Some(total) => total,
-                    // A corrupt/overflowing usage snapshot is not a usable
-                    // denominator.  Leave progress indeterminate instead of
-                    // turning saturation into a plausible percentage.
-                    None => return Ok(None),
-                };
-                baseline.bytes = match baseline.bytes.checked_add(usage.size) {
-                    Some(total) => total,
-                    None => return Ok(None),
-                };
+                baseline.objects_count = baseline.objects_count.saturating_add(usage.objects_count);
+                baseline.bytes = baseline.bytes.saturating_add(usage.size);
             }
         }
 
-        let identity = info.snapshot_identity();
-        let mut hasher = DefaultHasher::new();
-        identity.last_update.hash(&mut hasher);
-        identity.scanner_cycle.hash(&mut hasher);
-        identity.scanner_epoch.hash(&mut hasher);
-        let mut scope = buckets.to_vec();
-        scope.sort_unstable();
-        scope.hash(&mut hasher);
-        baseline.generation = Some(hasher.finish());
-
         Ok(Some(baseline))
     }
 

crates/heal/src/heal/task.rs

@@ -649,7 +649,7 @@ impl HealTask {
 
         let mut progress = self.progress.write().await;
         progress.set_current_object(Some(format!("skipped: {bucket}/{object}")));
-        progress.update_stage(1, 1);
+        progress.update_progress(0, 1, 0, 0);
         Ok(())
     }
 
@@ -733,7 +733,7 @@ impl HealTask {
             "Heal object skipped for data usage cache after transient error"
         );
         let mut progress = self.progress.write().await;
-        progress.update_stage(3, 3);
+        progress.update_progress(3, 3, 0, 0);
         true
     }
 
@@ -757,7 +757,7 @@ impl HealTask {
         );
         let mut progress = self.progress.write().await;
         progress.set_current_object(Some(format!("skipped: {bucket}/{object}")));
-        progress.update_stage(4, 4);
+        progress.update_progress(4, 4, 0, 0);
         true
     }
 
@@ -831,10 +831,6 @@ impl HealTask {
 
         match &result {
             Ok(_) => {
-                // A stage can reach its final step before the durable resume
-                // ledger and cleanup fences commit. Publish terminal 100 only
-                // after the enclosing operation has returned success.
-                self.progress.write().await.mark_completed();
                 let mut status = self.status.write().await;
                 *status = HealTaskStatus::Completed;
                 demote_to_debug_when!(self.heal_type.is_per_object(), info, target: "rustfs::heal::task", {

crates/heal/src/heal/task/heal_bucket.rs

@@ -13,7 +13,6 @@
 // limitations under the License.
 /// bucket/cluster/prefix heal: the recursive bucket-objects sweep and the erasure-set usage baseline
 use super::*;
-use crate::heal::progress::{add_bytes, increment_counter};
 
 impl HealTask {
     pub(super) async fn heal_bucket(&self, bucket: &str) -> Result<()> {
@@ -33,7 +32,7 @@ impl HealTask {
         {
             let mut progress = self.progress.write().await;
             progress.set_current_object(Some(format!("bucket: {bucket}")));
-            progress.update_stage(0, 3);
+            progress.update_progress(0, 3, 0, 0);
         }
 
         // Step 1: Check if bucket exists
@@ -67,7 +66,7 @@ impl HealTask {
 
         {
             let mut progress = self.progress.write().await;
-            progress.update_stage(1, 3);
+            progress.update_progress(1, 3, 0, 0);
         }
 
         // Step 2: Perform bucket heal using ecstore
@@ -123,7 +122,7 @@ impl HealTask {
 
                 if !self.options.recursive {
                     let mut progress = self.progress.write().await;
-                    progress.update_stage(3, 3);
+                    progress.update_progress(3, 3, 0, 0);
                 }
                 Ok(())
             }
@@ -143,7 +142,7 @@ impl HealTask {
                 );
                 {
                     let mut progress = self.progress.write().await;
-                    progress.update_stage(3, 3);
+                    progress.update_progress(3, 3, 0, 0);
                 }
                 Err(Error::TaskExecutionFailed {
                     message: format!("Failed to heal bucket {bucket}: {e}"),
@@ -246,7 +245,6 @@ impl HealTask {
         let mut scanned = 0u64;
         let mut healed = 0u64;
         let mut failed = 0u64;
-        let mut skipped = 0u64;
         let mut retryable_failed = 0u64;
         let mut permanent_failed = 0u64;
         let mut bytes = 0u64;
@@ -288,14 +286,16 @@ impl HealTask {
                 let mut retry = Vec::with_capacity(pending.len());
                 for item in pending {
                     self.check_control_flags().await?;
-                    let mut telemetry_unknown = false;
                     let object = item.name.as_str();
+                    if retry_attempt == 0 {
+                        scanned = scanned.saturating_add(1);
+                    }
                     {
                         let mut progress = self.progress.write().await;
                         progress.set_current_object(Some(format!("{bucket}/{object}")));
+                        progress.update_progress(scanned, healed, failed, bytes);
                     }
 
-                    let mut terminal_outcome = true;
                     let error = match self
                         .await_with_control(
                             self.storage
@@ -304,13 +304,13 @@ impl HealTask {
                         .await
                     {
                         Ok((result, None)) => {
-                            telemetry_unknown |= !increment_counter(&mut healed);
-                            telemetry_unknown |= !add_bytes(&mut bytes, u64::try_from(result.object_size).unwrap_or(u64::MAX));
+                            healed = healed.saturating_add(1);
+                            bytes = bytes.saturating_add(u64::try_from(result.object_size).unwrap_or_default());
                             self.record_result_item(result).await;
                             None
                         }
                         Ok((_, Some(err))) if is_missing_object_dir_heal_result(object, &err) => {
-                            telemetry_unknown |= !increment_counter(&mut healed);
+                            healed = healed.saturating_add(1);
                             debug!(
                                 target: "rustfs::heal::task",
                                 event = EVENT_HEAL_BUCKET_RESULT,
@@ -329,7 +329,6 @@ impl HealTask {
 
                     if let Some(err) = error {
                         if Self::should_skip_data_usage_cache_heal_error(bucket, object, &err) {
-                            telemetry_unknown |= !increment_counter(&mut skipped);
                             warn!(
                                 target: "rustfs::heal::task",
                                 event = EVENT_HEAL_BUCKET_RESULT,
@@ -343,7 +342,6 @@ impl HealTask {
                                 "Heal bucket object repair skipped due to transient metadata error"
                             );
                         } else if err.is_recoverable_heal() && retry_attempt < MAX_BUCKET_OBJECT_HEAL_RETRIES {
-                            terminal_outcome = false;
                             debug!(
                                 target: "rustfs::heal::task",
                                 event = EVENT_HEAL_BUCKET_RESULT,
@@ -359,7 +357,7 @@ impl HealTask {
                             );
                             retry.push(item);
                         } else {
-                            telemetry_unknown |= !increment_counter(&mut failed);
+                            failed = failed.saturating_add(1);
                             if err.is_recoverable_heal() {
                                 retryable_failed = retryable_failed.saturating_add(1);
                             } else {
@@ -385,19 +383,8 @@ impl HealTask {
                         }
                     }
 
-                    if terminal_outcome {
-                        telemetry_unknown |= !increment_counter(&mut scanned);
-                    }
-
-                    if !terminal_outcome {
-                        continue;
-                    }
-
                     let mut progress = self.progress.write().await;
-                    progress.update_object_progress(scanned, healed, failed, skipped, bytes);
-                    if telemetry_unknown {
-                        progress.mark_unknown();
-                    }
+                    progress.update_progress(scanned, healed, failed, bytes);
                 }
                 pending = retry;
                 retry_attempt = retry_attempt.saturating_add(1);
@@ -444,7 +431,7 @@ impl HealTask {
         Ok(())
     }
 
-    pub(super) async fn apply_erasure_set_usage_baseline(&self, buckets: &[String], set_disk_id: &str) -> Result<()> {
+    pub(super) async fn apply_erasure_set_usage_baseline(&self, buckets: &[String]) -> Result<()> {
         let baseline = match self
             .await_with_control(self.storage.erasure_set_usage_baseline(buckets))
             .await
@@ -455,26 +442,9 @@ impl HealTask {
             Err(_) => return Ok(()),
         };
 
-        let HealBucketUsageBaseline {
-            objects_count,
-            bytes,
-            generation,
-        } = baseline;
-        let generation = generation.map(|snapshot_generation| {
-            use std::hash::{Hash, Hasher};
-            let mut hasher = std::collections::hash_map::DefaultHasher::new();
-            snapshot_generation.hash(&mut hasher);
-            set_disk_id.hash(&mut hasher);
-            self.options.pool_index.hash(&mut hasher);
-            self.options.set_index.hash(&mut hasher);
-            hasher.finish()
-        });
+        let HealBucketUsageBaseline { objects_count, bytes } = baseline;
         let mut progress = self.progress.write().await;
-        if let Some(generation) = generation {
-            progress.set_total_baseline_with_generation(objects_count, bytes, generation);
-        } else {
-            progress.set_total_baseline(objects_count, bytes);
-        }
+        progress.set_total_baseline(objects_count, bytes);
         Ok(())
     }
 }

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

@@ -32,7 +32,7 @@ impl HealTask {
         {
             let mut progress = self.progress.write().await;
             progress.set_current_object(Some(format!("erasure_set: {} ({} buckets)", set_disk_id, buckets.len())));
-            progress.update_stage(0, 4);
+            progress.update_progress(0, 4, 0, 0);
         }
 
         let is_auto_replacement = matches!(self.source, HealRequestSource::AutoHeal) && !self.heal_endpoints.is_empty();
@@ -158,7 +158,7 @@ impl HealTask {
             None
         };
 
-        self.apply_erasure_set_usage_baseline(&buckets, &set_disk_id).await?;
+        self.apply_erasure_set_usage_baseline(&buckets).await?;
 
         let healing_marker = format!("{set_disk_id}:{}", self.id);
         if let Some((disk, resume_manager, _)) = replacement_resume.as_ref() {
@@ -244,7 +244,7 @@ impl HealTask {
                         );
                         {
                             let mut progress = self.progress.write().await;
-                            progress.update_stage(4, 4);
+                            progress.update_progress(4, 4, 0, 0);
                         }
                         return Err(Error::TaskExecutionFailed {
                             message: format!("Failed to heal disk format for {set_disk_id}: {e}"),
@@ -297,7 +297,7 @@ impl HealTask {
                 );
                 {
                     let mut progress = self.progress.write().await;
-                    progress.update_stage(4, 4);
+                    progress.update_progress(4, 4, 0, 0);
                 }
                 return Err(Error::TaskExecutionFailed {
                     message: format!("Failed to heal disk format for {set_disk_id}: {e}"),
@@ -307,7 +307,7 @@ impl HealTask {
 
         {
             let mut progress = self.progress.write().await;
-            progress.update_stage(1, 4);
+            progress.update_progress(1, 4, 0, 0);
         }
 
         // The rebuilt disks are formatted now: mark them as healing so
@@ -336,7 +336,7 @@ impl HealTask {
 
         {
             let mut progress = self.progress.write().await;
-            progress.update_stage(2, 4);
+            progress.update_progress(2, 4, 0, 0);
         }
 
         // Step 3: Heal bucket structure
@@ -420,7 +420,7 @@ impl HealTask {
 
         {
             let mut progress = self.progress.write().await;
-            progress.update_stage(3, 4);
+            progress.update_progress(3, 4, 0, 0);
         }
 
         // Step 4: Execute erasure set heal with resume
@@ -463,7 +463,9 @@ impl HealTask {
         };
 
         {
-            self.progress.write().await.update_stage(4, 4);
+            let mut progress = self.progress.write().await;
+            let bytes_processed = progress.bytes_processed;
+            progress.update_progress(4, 4, 0, bytes_processed);
         }
 
         match result {

crates/heal/src/heal/task/heal_metadata.rs

@@ -32,7 +32,7 @@ impl HealTask {
         {
             let mut progress = self.progress.write().await;
             progress.set_current_object(Some(format!("metadata: {bucket}/{object}")));
-            progress.update_stage(0, 3);
+            progress.update_progress(0, 3, 0, 0);
         }
 
         // Step 1: Check if object exists
@@ -74,7 +74,7 @@ impl HealTask {
 
         {
             let mut progress = self.progress.write().await;
-            progress.update_stage(1, 3);
+            progress.update_progress(1, 3, 0, 0);
         }
 
         // Step 2: Perform metadata heal using ecstore
@@ -122,7 +122,7 @@ impl HealTask {
                     );
                     {
                         let mut progress = self.progress.write().await;
-                        progress.update_stage(3, 3);
+                        progress.update_progress(3, 3, 0, 0);
                     }
                     return Err(Error::TaskExecutionFailed {
                         message: format!("Failed to heal metadata {bucket}/{object}: {e}"),
@@ -145,7 +145,7 @@ impl HealTask {
 
                 {
                     let mut progress = self.progress.write().await;
-                    progress.update_stage(3, 3);
+                    progress.update_progress(3, 3, 0, 0);
                 }
                 self.record_result_item(result).await;
                 Ok(())
@@ -167,7 +167,7 @@ impl HealTask {
                 );
                 {
                     let mut progress = self.progress.write().await;
-                    progress.update_stage(3, 3);
+                    progress.update_progress(3, 3, 0, 0);
                 }
                 Err(Error::TaskExecutionFailed {
                     message: format!("Failed to heal metadata {bucket}/{object}: {e}"),
@@ -194,7 +194,7 @@ impl HealTask {
         {
             let mut progress = self.progress.write().await;
             progress.set_current_object(Some(format!("ec_decode: {bucket}/{object}")));
-            progress.update_stage(0, 3);
+            progress.update_progress(0, 3, 0, 0);
         }
 
         // Step 1: Check if object exists
@@ -236,7 +236,7 @@ impl HealTask {
 
         {
             let mut progress = self.progress.write().await;
-            progress.update_stage(1, 3);
+            progress.update_progress(1, 3, 0, 0);
         }
 
         // Step 2: Perform EC decode heal using ecstore
@@ -284,7 +284,7 @@ impl HealTask {
                     );
                     {
                         let mut progress = self.progress.write().await;
-                        progress.update_stage(3, 3);
+                        progress.update_progress(3, 3, 0, 0);
                     }
                     return Err(Error::TaskExecutionFailed {
                         message: format!("Failed to heal EC decode {bucket}/{object}: {e}"),
@@ -309,7 +309,7 @@ impl HealTask {
 
                 {
                     let mut progress = self.progress.write().await;
-                    progress.update_object_progress(1, 1, 0, 0, object_size);
+                    progress.update_progress(3, 3, 0, object_size);
                 }
                 self.record_result_item(result).await;
                 Ok(())
@@ -331,7 +331,7 @@ impl HealTask {
                 );
                 {
                     let mut progress = self.progress.write().await;
-                    progress.update_stage(3, 3);
+                    progress.update_progress(3, 3, 0, 0);
                 }
                 Err(Error::TaskExecutionFailed {
                     message: format!("Failed to heal EC decode {bucket}/{object}: {e}"),

crates/heal/src/heal/task/heal_object.rs

@@ -36,7 +36,7 @@ impl HealTask {
         {
             let mut progress = self.progress.write().await;
             progress.set_current_object(Some(format!("{bucket}/{object}")));
-            progress.update_stage(0, 4);
+            progress.update_progress(0, 4, 0, 0);
         }
 
         // Step 1: Check if object exists and get metadata
@@ -132,7 +132,7 @@ impl HealTask {
 
         {
             let mut progress = self.progress.write().await;
-            progress.update_stage(1, 3);
+            progress.update_progress(1, 3, 0, 0);
         }
 
         // Step 2: directly call ecstore to perform heal
@@ -187,7 +187,7 @@ impl HealTask {
                         );
                         {
                             let mut progress = self.progress.write().await;
-                            progress.update_stage(3, 3);
+                            progress.update_progress(3, 3, 0, 0);
                         }
                         return Ok(());
                     }
@@ -207,7 +207,7 @@ impl HealTask {
 
                     {
                         let mut progress = self.progress.write().await;
-                        progress.update_stage(3, 3);
+                        progress.update_progress(3, 3, 0, 0);
                     }
 
                     if Self::should_return_typed_heal_error(&e) {
@@ -249,7 +249,7 @@ impl HealTask {
 
                 {
                     let mut progress = self.progress.write().await;
-                    progress.update_object_progress(1, 1, 0, 0, object_size);
+                    progress.update_progress(3, 3, 0, object_size);
                 }
                 self.record_result_item(result).await;
                 Ok(())
@@ -275,7 +275,7 @@ impl HealTask {
                     );
                     {
                         let mut progress = self.progress.write().await;
-                        progress.update_stage(3, 3);
+                        progress.update_progress(3, 3, 0, 0);
                     }
                     return Ok(());
                 }
@@ -295,7 +295,7 @@ impl HealTask {
 
                 {
                     let mut progress = self.progress.write().await;
-                    progress.update_stage(3, 3);
+                    progress.update_progress(3, 3, 0, 0);
                 }
 
                 if Self::should_return_typed_heal_error(&e) {
@@ -414,7 +414,7 @@ impl HealTask {
 
                 {
                     let mut progress = self.progress.write().await;
-                    progress.update_object_progress(1, 1, 0, 0, object_size);
+                    progress.update_progress(4, 4, 0, object_size);
                 }
                 self.record_result_item(result).await;
                 Ok(())

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

@@ -2096,7 +2096,6 @@ async fn erasure_set_heal_applies_usage_baseline_to_progress() {
         usage_baseline: Mutex::new(Some(HealBucketUsageBaseline {
             objects_count: 10,
             bytes: 8,
-            generation: Some(1),
         })),
         ..Default::default()
     });
@@ -2120,8 +2119,6 @@ async fn erasure_set_heal_applies_usage_baseline_to_progress() {
     let progress = task.get_progress().await;
     assert_eq!(progress.objects_total_count, 10);
     assert_eq!(progress.objects_total_size, 8);
-    assert!(progress.baseline_generation.is_some());
-    assert!(progress.baseline_known);
     assert_eq!(progress.bytes_processed, 2);
     assert!((progress.progress_percentage - 25.0).abs() < 0.001);
 }

crates/replication/src/config.rs

@@ -60,9 +60,7 @@ pub const REPLICATION_READ_ONLY_HISTORICAL_FIELDS: &[&str] = &[
     "Destination.ReplicationTime",
 ];
 
-// v2: disableProxy moved from unsupported to writable (per-target read-proxy
-// opt-out is accepted by set-remote-target and the `proxy` update op).
-pub const REMOTE_TARGET_CAPABILITY_CONTRACT_VERSION: u32 = 2;
+pub const REMOTE_TARGET_CAPABILITY_CONTRACT_VERSION: u32 = 1;
 
 pub const REMOTE_TARGET_WRITABLE_FIELDS: &[&str] = &[
     "sourcebucket",
@@ -85,12 +83,9 @@ pub const REMOTE_TARGET_WRITABLE_FIELDS: &[&str] = &[
     // madmin default of 60s); the per-target health-check interval is not
     // yet applied — the heartbeat keeps its global env-configured interval.
     "healthCheckDuration",
-    // Per-target read-proxy opt-out, consumed by the proxy-target selector
-    // (contract v2; previously only importable via MinIO bucket-targets.json).
-    "disableProxy",
 ];
 
-pub const REMOTE_TARGET_UNSUPPORTED_FIELDS: &[&str] = &["edge", "edgeSyncBeforeExpiry"];
+pub const REMOTE_TARGET_UNSUPPORTED_FIELDS: &[&str] = &["disableProxy", "edge", "edgeSyncBeforeExpiry"];
 
 #[derive(Debug, Clone, Serialize, Deserialize, Default)]
 pub struct ObjectOpts {

crates/scanner/src/data_usage_define.rs

@@ -29,7 +29,8 @@ use rustfs_config::ENV_SCANNER_CACHE_SAVE_TIMEOUT_SECS;
 pub use rustfs_data_usage::{
     AllTierStats, BucketTargetUsageInfo, BucketUsageInfo, DATA_USAGE_OBJECT_NAME, DATA_USAGE_OBSERVED_OBJECT_NAME,
     DataUsageEntry, DataUsageHash, DataUsageHashMap, DataUsageInfo, LEGACY_DATA_USAGE_OBJECT_NAME, PrefixUsageEntry,
-    PrefixUsageQuery, PrefixUsageSummary, ReplTargetSizeSummary, SizeSummary, TierStats, hash_path, prefix_usage_in_cache,
+    PrefixUsageQuery, PrefixUsageSummary, ReplTargetSizeSummary, SizeReconciliationEntry, SizeReconciliationScope, SizeSummary,
+    TierStats, hash_path, prefix_usage_in_cache,
 };
 use rustfs_utils::path::{SLASH_SEPARATOR, path_join_buf};
 use tokio::time::{Duration, Instant, sleep, timeout};
@@ -192,6 +193,10 @@ const MAX_DATA_USAGE_CACHE_DEPTH: usize = 1024;
 pub trait ScannerSizeSummaryExt {
     /// Fold one object's contribution into the summary, including its tier.
     fn actions_accounting(&mut self, oi: &ObjectInfo, size: i64, actual_size: i64);
+    /// Fold counters and physical tier usage for an object whose metadata is
+    /// valid but whose logical size is currently unavailable. Logical totals
+    /// stay unchanged.
+    fn actions_accounting_unknown(&mut self, oi: &ObjectInfo);
 }
 
 impl ScannerSizeSummaryExt for SizeSummary {
@@ -225,6 +230,34 @@ impl ScannerSizeSummaryExt for SizeSummary {
             });
         }
     }
+
+    fn actions_accounting_unknown(&mut self, oi: &ObjectInfo) {
+        if oi.delete_marker {
+            self.delete_markers = self.delete_markers.saturating_add(1);
+            return;
+        }
+
+        if oi.version_id.is_some_and(|v| !v.is_nil()) {
+            self.versions = self.versions.saturating_add(1);
+        }
+
+        if oi.transitioned_object.free_version {
+            return;
+        }
+
+        let tier = if oi.transitioned_object.status == TRANSITION_COMPLETE {
+            oi.transitioned_object.tier.clone()
+        } else {
+            oi.storage_class.clone().unwrap_or_else(|| storageclass::STANDARD.to_string())
+        };
+        if let Some(tier_stats) = self.tier_stats.get_mut(&tier) {
+            *tier_stats = tier_stats.add(&TierStats {
+                total_size: u64::try_from(oi.size).unwrap_or(0),
+                num_versions: 1,
+                num_objects: u64::from(oi.is_latest),
+            });
+        }
+    }
 }
 
 // ===== Cache-related data structures =====
@@ -344,6 +377,10 @@ pub struct DataUsageCacheInfo {
     pub scan_plan_digest: Option<DataUsageScanPlanDigest>,
     #[serde(default)]
     pub cache_key_format: u16,
+    /// Bounded durable debts for versions whose logical size was not trusted.
+    /// The map key is an identity key, never a user-controlled metric label.
+    #[serde(default)]
+    pub size_reconciliation: HashMap<String, SizeReconciliationEntry>,
 }
 
 impl Serialize for DataUsageCacheInfo {
@@ -353,7 +390,8 @@ impl Serialize for DataUsageCacheInfo {
     {
         // Keep this metadata map-encoded so older readers can ignore fields
         // appended by newer scanner versions during rolling upgrades.
-        let mut state = serializer.serialize_map(Some(16))?;
+        let field_count = 16 + usize::from(!self.size_reconciliation.is_empty());
+        let mut state = serializer.serialize_map(Some(field_count))?;
         state.serialize_entry("name", &self.name)?;
         state.serialize_entry("next_cycle", &self.next_cycle)?;
         state.serialize_entry("leader_epoch", &self.leader_epoch)?;
@@ -370,6 +408,9 @@ impl Serialize for DataUsageCacheInfo {
         state.serialize_entry("snapshot_complete", &self.snapshot_complete)?;
         state.serialize_entry("scan_plan_digest", &self.scan_plan_digest)?;
         state.serialize_entry("cache_key_format", &self.cache_key_format)?;
+        if !self.size_reconciliation.is_empty() {
+            state.serialize_entry("size_reconciliation", &self.size_reconciliation)?;
+        }
         state.end()
     }
 }
@@ -428,14 +469,18 @@ impl DataUsageCache {
                     self.checked_flatten(name).is_some()
                 });
         if !reusable {
-            let pending_heals = if self.info.name == name {
-                std::mem::take(&mut self.info.pending_heals)
+            let (pending_heals, size_reconciliation) = if self.info.name == name {
+                (
+                    std::mem::take(&mut self.info.pending_heals),
+                    std::mem::take(&mut self.info.size_reconciliation),
+                )
             } else {
-                Vec::new()
+                (Vec::new(), HashMap::new())
             };
             *self = Self::default();
             self.info.name = name.to_string();
             self.info.pending_heals = pending_heals;
+            self.info.size_reconciliation = size_reconciliation;
         }
 
         self.info.next_cycle = next_cycle;

crates/scanner/src/data_usage_define/tests.rs

@@ -673,6 +673,34 @@ fn size_summary_actions_accounting_accumulates_tier_stats() {
     );
 }
 
+#[test]
+fn size_summary_unknown_accounting_keeps_physical_tier_and_version_only() {
+    let mut summary = SizeSummary::default();
+    summary
+        .tier_stats
+        .insert(storageclass::STANDARD.to_string(), TierStats::default());
+    let object = ObjectInfo {
+        size: 12,
+        storage_class: Some(storageclass::STANDARD.to_string()),
+        version_id: Some(uuid::Uuid::new_v4()),
+        is_latest: true,
+        ..Default::default()
+    };
+
+    summary.actions_accounting_unknown(&object);
+
+    assert_eq!(summary.total_size, 0, "unknown logical size must not become zero or physical bytes");
+    assert_eq!(summary.versions, 1);
+    assert_eq!(
+        summary.tier_stats.get(storageclass::STANDARD),
+        Some(&TierStats {
+            total_size: 12,
+            num_versions: 1,
+            num_objects: 1,
+        })
+    );
+}
+
 #[test]
 fn test_data_usage_entry_merge_sums_failed_objects() {
     let mut left = DataUsageEntry {
@@ -1079,6 +1107,16 @@ fn data_usage_cache_prepare_for_scan_preserves_pending_heal_only_progress() {
             scan_plan_digest: Some(TEST_PLAN_DIGEST),
             cache_key_format: DATA_USAGE_CACHE_KEY_FORMAT,
             pending_heals: vec![pending_heal.clone()],
+            size_reconciliation: HashMap::from([(
+                "size-key".to_string(),
+                SizeReconciliationEntry {
+                    key: "size-key".to_string(),
+                    bucket: "bucket".to_string(),
+                    object: "prefix/object".to_string(),
+                    reason: "invalid_declared_size".to_string(),
+                    ..Default::default()
+                },
+            )]),
             ..Default::default()
         },
         ..Default::default()
@@ -1088,6 +1126,7 @@ fn data_usage_cache_prepare_for_scan_preserves_pending_heal_only_progress() {
 
     assert_eq!(outcome, DataUsageCachePrepareOutcome::Reused);
     assert_eq!(cache.info.pending_heals, vec![pending_heal]);
+    assert!(cache.info.size_reconciliation.contains_key("size-key"));
     assert!(cache.cache.is_empty());
     assert!(!cache.info.snapshot_complete);
 }

crates/scanner/src/scanner_folder.rs

@@ -20,8 +20,9 @@ use std::time::{Duration, Instant, SystemTime};
 
 use crate::ReplTargetSizeSummary;
 use crate::data_usage_define::{
-    DATA_USAGE_SCAN_CHECKPOINT_VERSION, DataUsageCache, DataUsageEntry, DataUsageHash, DataUsageHashMap, DataUsageScanCheckpoint,
-    DataUsageScanCheckpointReason, PendingScannerHeal, PendingScannerHealKind, ScannerSizeSummaryExt, SizeSummary, hash_path,
+    DATA_USAGE_SCAN_CHECKPOINT_VERSION, DataUsageCache, DataUsageCacheInfo, DataUsageEntry, DataUsageHash, DataUsageHashMap,
+    DataUsageScanCheckpoint, DataUsageScanCheckpointReason, PendingScannerHeal, PendingScannerHealKind, ScannerSizeSummaryExt,
+    SizeReconciliationEntry, SizeSummary, hash_path,
 };
 use crate::error::ScannerError;
 use crate::runtime_config::{
@@ -97,6 +98,9 @@ const METRIC_SCANNER_EXCESS_FOLDERS_TOTAL: &str = "rustfs_scanner_excess_folders
 const METRIC_SCANNER_PENDING_HEAL_PRUNE_TOTAL: &str = "rustfs_scanner_pending_heal_prune_total";
 const METRIC_SCANNER_PENDING_HEAL_MALFORMED_TOTAL: &str = "rustfs_scanner_pending_heal_malformed_total";
 const MAX_PENDING_SCANNER_HEAL_RETRIES_PER_BUCKET: usize = 128;
+const MAX_SIZE_RECONCILIATION_ENTRIES_PER_BUCKET: usize = 10_000;
+const MAX_SIZE_RECONCILIATION_BYTES_PER_BUCKET: usize = 8 * 1024 * 1024;
+const MAX_SIZE_RECONCILIATION_AGE_SECS: u64 = 7 * 24 * 60 * 60;
 
 // --- scanner excess alerts as S3 notification events (rustfs/backlog#1868) --
 //
@@ -364,7 +368,7 @@ impl PendingScannerAccounting<'_> {
     fn apply(self, size_summary: &mut SizeSummary, cumulative_size: &mut i64, queued: bool) {
         let size = if queued { self.expired_size } else { self.retained_size };
         size_summary.actions_accounting(self.object, size, self.retained_size);
-        *cumulative_size += size;
+        *cumulative_size = cumulative_size.saturating_add(size);
     }
 }
 
@@ -675,6 +679,54 @@ pub struct FolderScanner {
     list_path_raw_options_observer: Option<mpsc::UnboundedSender<ListPathRawTimeoutSnapshot>>,
 }
 
+fn size_reconciliation_entry_bytes(entry: &SizeReconciliationEntry) -> usize {
+    entry.key.len()
+        + entry.bucket.len()
+        + entry.object.len()
+        + entry.version_id.as_deref().map_or(0, str::len)
+        + entry.generation.as_deref().map_or(0, str::len)
+        + entry.reason.len()
+        + std::mem::size_of::<u64>()
+        + std::mem::size_of::<u32>()
+}
+
+fn prune_size_reconciliation(info: &mut DataUsageCacheInfo, now: u64) {
+    info.size_reconciliation.retain(|key, entry| {
+        if entry.first_seen == 0 || entry.first_seen > now {
+            entry.first_seen = now;
+        }
+        key == &entry.key
+            && entry.key.len() <= 4096
+            && entry.bucket.len() <= 512
+            && entry.object.len() <= 512
+            && entry.version_id.as_deref().is_none_or(|value| value.len() <= 64)
+            && entry.generation.as_deref().is_none_or(|value| value.len() <= 64)
+            && entry.reason.len() <= 64
+            && now.saturating_sub(entry.first_seen) <= MAX_SIZE_RECONCILIATION_AGE_SECS
+    });
+
+    while info.size_reconciliation.len() > MAX_SIZE_RECONCILIATION_ENTRIES_PER_BUCKET
+        || info
+            .size_reconciliation
+            .values()
+            .map(size_reconciliation_entry_bytes)
+            .sum::<usize>()
+            > MAX_SIZE_RECONCILIATION_BYTES_PER_BUCKET
+    {
+        let oldest = info
+            .size_reconciliation
+            .iter()
+            .min_by(|(left_key, left), (right_key, right)| {
+                left.first_seen.cmp(&right.first_seen).then_with(|| left_key.cmp(right_key))
+            })
+            .map(|(key, _)| key.clone());
+        let Some(oldest) = oldest else {
+            break;
+        };
+        info.size_reconciliation.remove(&oldest);
+    }
+}
+
 impl FolderScanner {
     fn now_secs() -> u64 {
         SystemTime::now()
@@ -748,6 +800,55 @@ impl FolderScanner {
         }
     }
 
+    /// Apply the per-object size-resolution ledger updates in one place. The
+    /// scanner cache is the durable boundary; both working copies are updated
+    /// so an incremental publication cannot lose a debt or its resolution.
+    fn apply_size_reconciliation(&mut self, summary: &SizeSummary) {
+        let now = Self::now_secs();
+        // Keep an unresolved identity in place while refreshing its object
+        // scope. This lets repeated observations increment `attempts`; only
+        // debts absent from the current pass are considered resolved.
+        let current_keys = summary
+            .size_reconciliation
+            .iter()
+            .map(|entry| entry.key.clone())
+            .collect::<HashSet<_>>();
+        for info in [&mut self.new_cache.info, &mut self.update_cache.info] {
+            prune_size_reconciliation(info, now);
+
+            if !summary.size_reconciliation_truncated {
+                for scope in &summary.reconciliation_scopes {
+                    let scope_bucket = item_actions::bounded_reconciliation_field(&scope.bucket);
+                    let scope_object = item_actions::bounded_reconciliation_field(&scope.object);
+                    info.size_reconciliation.retain(|key, entry| {
+                        entry.bucket != scope_bucket || entry.object != scope_object || current_keys.contains(key)
+                    });
+                }
+            }
+
+            for incoming in &summary.size_reconciliation {
+                if let Some(existing) = info.size_reconciliation.get_mut(&incoming.key) {
+                    existing.reason = incoming.reason.clone();
+                    existing.physical_size = incoming.physical_size;
+                    existing.generation = incoming.generation.clone();
+                    existing.version_id = incoming.version_id.clone();
+                    existing.attempts = existing.attempts.saturating_add(1);
+                    continue;
+                }
+
+                if size_reconciliation_entry_bytes(incoming) > MAX_SIZE_RECONCILIATION_BYTES_PER_BUCKET {
+                    continue;
+                }
+
+                let mut entry = incoming.clone();
+                entry.first_seen = now;
+                entry.attempts = 1;
+                info.size_reconciliation.insert(entry.key.clone(), entry);
+            }
+            prune_size_reconciliation(info, now);
+        }
+    }
+
     fn record_scan_resume_hint(&mut self, folder: &str) {
         self.new_cache.info.scan_resume_after = Some(folder.to_string());
         self.update_cache.info.scan_resume_after = Some(folder.to_string());
@@ -1426,6 +1527,7 @@ impl FolderScanner {
                 abandoned_children.remove(&path_join_buf(&[&item.bucket, &item.object_path()]));
 
                 apply_scanner_size_summary(into, &sz);
+                self.apply_size_reconciliation(&sz);
                 into.objects += 1;
                 object_count += 1;
                 self.budget.record_object_scanned();
@@ -2194,6 +2296,10 @@ pub async fn scan_data_folder(
         list_path_raw_options_observer: None,
     };
 
+    let now = FolderScanner::now_secs();
+    prune_size_reconciliation(&mut scanner.new_cache.info, now);
+    prune_size_reconciliation(&mut scanner.update_cache.info, now);
+
     // Check if context is cancelled
     if ctx.is_cancelled() {
         return Err(ScannerError::Other("Operation cancelled".to_string()));
@@ -2217,7 +2323,9 @@ pub async fn scan_data_folder(
             new_cache.force_compact(DATA_SCANNER_COMPACT_AT_CHILDREN);
             new_cache.info.last_update = Some(SystemTime::now());
             new_cache.info.next_cycle = cache.info.next_cycle;
-            let unresolved_objects = root.failed_objects > 0 || !new_cache.info.failed_objects.is_empty();
+            let unresolved_objects = root.failed_objects > 0
+                || !new_cache.info.failed_objects.is_empty()
+                || !new_cache.info.size_reconciliation.is_empty();
             new_cache.info.snapshot_complete = !unresolved_objects;
             let had_scan_checkpoint = cache.info.scan_checkpoint.is_some() || new_cache.info.scan_checkpoint.is_some();
             new_cache.info.scan_resume_after = None;
@@ -2245,7 +2353,7 @@ pub async fn scan_data_folder(
                 if root_has_progress {
                     new_cache.replace_hashed(&root_hash, &None, &root);
                 }
-                if partial_cache_is_useful(&root, pending_heals_changed) {
+                if partial_cache_is_useful(&root, pending_heals_changed) || !new_cache.info.size_reconciliation.is_empty() {
                     if new_cache.root().is_some() {
                         new_cache.force_compact(DATA_SCANNER_COMPACT_AT_CHILDREN);
                     }

crates/scanner/src/scanner_folder/item_actions.rs

@@ -13,6 +13,7 @@
 // limitations under the License.
 /// Per-object scan actions: ScannerItem, the get-size failure policy, and the heal/ILM admission helpers.
 use super::*;
+use sha2::{Digest as _, Sha256};
 
 /// Cached folder information for scanning
 #[derive(Clone, Debug)]
@@ -32,6 +33,259 @@ pub(super) enum GetSizeFailureAction {
     HealMetadata { object: String },
 }
 
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub(super) enum SizeResolutionReason {
+    CompressedSizeUnknown,
+    InvalidPhysicalSize,
+    UnsupportedCompression,
+    InvalidObjectSize,
+    InvalidPartSize,
+    InvalidDeclaredSize,
+    SizeOverflowOrMismatch,
+}
+
+impl SizeResolutionReason {
+    fn as_str(self) -> &'static str {
+        match self {
+            Self::CompressedSizeUnknown => "compressed_size_unknown",
+            Self::InvalidPhysicalSize => "invalid_physical_size",
+            Self::UnsupportedCompression => "unsupported_compression",
+            Self::InvalidObjectSize => "invalid_object_size",
+            Self::InvalidPartSize => "invalid_part_size",
+            Self::InvalidDeclaredSize => "invalid_declared_size",
+            Self::SizeOverflowOrMismatch => "size_overflow_or_mismatch",
+        }
+    }
+}
+
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub(super) enum SizeResolution {
+    Known { logical: i64, physical: i64 },
+    Unknown { physical: i64, reason: SizeResolutionReason },
+    Corrupt { physical: i64, reason: SizeResolutionReason },
+}
+
+impl SizeResolution {
+    fn known_size(&self) -> Option<i64> {
+        match self {
+            Self::Known { logical, .. } => Some(*logical),
+            Self::Unknown { .. } | Self::Corrupt { .. } => None,
+        }
+    }
+}
+
+fn size_reconciliation_key(oi: &ObjectInfo, reason: SizeResolutionReason) -> String {
+    let version = oi
+        .version_id
+        .filter(|version| !version.is_nil())
+        .map(|version| version.to_string())
+        .unwrap_or_default();
+    let generation = oi
+        .data_dir
+        .filter(|generation| !generation.is_nil())
+        .map(|generation| generation.to_string())
+        .unwrap_or_default();
+    // Length-prefix each component so an object key containing the separator
+    // cannot alias another identity. S3 keys are bounded in normal operation;
+    // oversized persisted values use a digest so a corrupt metadata record
+    // cannot grow the ledger without bound.
+    fn component(value: &str) -> String {
+        const MAX_COMPONENT_LEN: usize = 512;
+        if value.len() <= MAX_COMPONENT_LEN {
+            return format!("{}:{}", value.len(), value);
+        }
+        let digest = Sha256::digest(value.as_bytes());
+        let digest = hex_simd::encode_to_string(digest, hex_simd::AsciiCase::Lower);
+        format!("hash:{}:{}", value.len(), digest)
+    }
+    format!(
+        "{}|{}|{}|{}|{}",
+        component(&oi.bucket),
+        component(&oi.name),
+        component(&version),
+        component(&generation),
+        component(reason.as_str())
+    )
+}
+
+pub(super) fn bounded_reconciliation_field(value: &str) -> String {
+    const MAX_FIELD_LEN: usize = 512;
+    if value.len() <= MAX_FIELD_LEN {
+        return value.to_string();
+    }
+    let digest = hex_simd::encode_to_string(Sha256::digest(value.as_bytes()), hex_simd::AsciiCase::Lower);
+    let prefix_len = MAX_FIELD_LEN - 65;
+    let prefix = value
+        .char_indices()
+        .take_while(|(offset, ch)| offset.saturating_add(ch.len_utf8()) <= prefix_len)
+        .map(|(_, ch)| ch)
+        .collect::<String>();
+    format!("{}~{}", prefix, digest)
+}
+
+fn record_size_resolution(summary: &mut SizeSummary, oi: &ObjectInfo, resolution: &SizeResolution) {
+    match resolution {
+        SizeResolution::Known { .. } => {}
+        SizeResolution::Unknown { physical, reason } | SizeResolution::Corrupt { physical, reason } => {
+            summary.record_size_reconciliation(SizeReconciliationEntry {
+                key: size_reconciliation_key(oi, *reason),
+                bucket: bounded_reconciliation_field(&oi.bucket),
+                object: bounded_reconciliation_field(&oi.name),
+                version_id: oi
+                    .version_id
+                    .filter(|version| !version.is_nil())
+                    .map(|version| version.to_string()),
+                generation: oi
+                    .data_dir
+                    .filter(|generation| !generation.is_nil())
+                    .map(|generation| generation.to_string()),
+                reason: reason.as_str().to_string(),
+                physical_size: u64::try_from(*physical).ok(),
+                first_seen: 0,
+                attempts: 0,
+            });
+        }
+    }
+}
+
+/// Resolve the size metadata once at the scanner trust boundary. A compressed
+/// -1 sentinel is valid legacy metadata, but it cannot participate in normal
+/// logical-size accounting or size-filtered lifecycle rules.
+pub(super) fn resolve_size(oi: &ObjectInfo) -> SizeResolution {
+    let physical = oi.size;
+    if physical < 0 {
+        return SizeResolution::Corrupt {
+            physical,
+            reason: SizeResolutionReason::InvalidPhysicalSize,
+        };
+    }
+
+    let compressed = match oi.compression_read_plan() {
+        Ok((_, _, compressed)) => compressed,
+        Err(_) => {
+            return SizeResolution::Corrupt {
+                physical,
+                reason: SizeResolutionReason::UnsupportedCompression,
+            };
+        }
+    };
+
+    if oi.actual_size < -1 || (oi.actual_size == -1 && !compressed) {
+        return SizeResolution::Corrupt {
+            physical,
+            reason: SizeResolutionReason::InvalidObjectSize,
+        };
+    }
+
+    // Match ObjectInfo::get_actual_size: a positive in-memory value is the
+    // authoritative decoded size. Stale declared/part metadata must not turn
+    // an otherwise valid object into a false corruption report.
+    if oi.actual_size > 0 {
+        return SizeResolution::Known {
+            logical: oi.actual_size,
+            physical,
+        };
+    }
+
+    if oi
+        .parts
+        .iter()
+        .any(|part| part.actual_size < -1 || (part.actual_size < 0 && !compressed))
+    {
+        return SizeResolution::Corrupt {
+            physical,
+            reason: SizeResolutionReason::InvalidPartSize,
+        };
+    }
+
+    let declared = rustfs_utils::http::get_str(&oi.user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE);
+    let declared = match declared {
+        Some(value) if value.is_empty() => {
+            return SizeResolution::Corrupt {
+                physical,
+                reason: SizeResolutionReason::InvalidDeclaredSize,
+            };
+        }
+        Some(value) => match value.parse::<i64>() {
+            Ok(value) if value >= 0 => Some(value),
+            _ => {
+                return SizeResolution::Corrupt {
+                    physical,
+                    reason: SizeResolutionReason::InvalidDeclaredSize,
+                };
+            }
+        },
+        None => None,
+    };
+
+    let logical = match oi.get_actual_size() {
+        Ok(size) if size == -1 && compressed && declared.is_none() => {
+            return SizeResolution::Unknown {
+                physical,
+                reason: SizeResolutionReason::CompressedSizeUnknown,
+            };
+        }
+        Ok(size) if size >= 0 => size,
+        Ok(_) | Err(_) => {
+            return SizeResolution::Corrupt {
+                physical,
+                reason: SizeResolutionReason::SizeOverflowOrMismatch,
+            };
+        }
+    };
+
+    if compressed && logical == 0 && physical != 0 && oi.parts.is_empty() && declared.is_none() {
+        return SizeResolution::Corrupt {
+            physical,
+            reason: SizeResolutionReason::SizeOverflowOrMismatch,
+        };
+    }
+
+    SizeResolution::Known { logical, physical }
+}
+
+fn resolve_sizes(object_infos: &[ObjectInfo]) -> Vec<SizeResolution> {
+    object_infos.iter().map(resolve_size).collect()
+}
+
+fn lifecycle_rule_has_size_filter(lifecycle: &BucketLifecycleConfiguration, rule_id: &str) -> bool {
+    let filter_has_size = |filter: &s3s::dto::LifecycleRuleFilter| {
+        filter.object_size_greater_than.is_some()
+            || filter.object_size_less_than.is_some()
+            || filter
+                .and
+                .as_ref()
+                .is_some_and(|and| and.object_size_greater_than.is_some() || and.object_size_less_than.is_some())
+    };
+    lifecycle
+        .rules
+        .iter()
+        .find(|rule| rule.id.as_deref().unwrap_or_default() == rule_id)
+        .and_then(|rule| rule.filter.as_ref())
+        .is_some_and(filter_has_size)
+}
+
+fn lifecycle_event_allowed(resolution: &SizeResolution, event: &Event, lifecycle: &BucketLifecycleConfiguration) -> bool {
+    match resolution {
+        // Corrupt metadata cannot safely authorize a destructive action, even
+        // when the evaluator happened to produce a time-only event.
+        SizeResolution::Corrupt { .. } => false,
+        // A valid-but-unknown logical size may still execute lifecycle
+        // actions whose rule is independent of object-size predicates. The
+        // evaluator has already selected the rule; only that rule's filter
+        // can make the missing logical value action-critical.
+        SizeResolution::Unknown { .. } => !lifecycle_rule_has_size_filter(lifecycle, &event.rule_id),
+        SizeResolution::Known { .. } => true,
+    }
+}
+
+/// A successful newer-noncurrent batch consumes both known and unresolved
+/// versions from the retained-version alert count. The two accounting paths
+/// are separate because only known sizes can contribute byte totals.
+fn remaining_versions_after_queued_noncurrent(remaining_versions: usize, known_count: usize, unknown_count: usize) -> usize {
+    remaining_versions.saturating_sub(known_count.saturating_add(unknown_count))
+}
+
 /// How the corrupt-metadata branch records the repair after attempting an
 /// MRF intent (backlog#1894 axis A).
 #[derive(Debug, PartialEq, Eq)]
@@ -319,34 +573,45 @@ impl ScannerItem {
             "Scanner lifecycle evaluation started"
         );
 
+        let resolved_sizes = resolve_sizes(&object_infos);
+        if let Some(first) = object_infos.first() {
+            size_summary.record_reconciliation_scope(
+                &bounded_reconciliation_field(&first.bucket),
+                &bounded_reconciliation_field(&first.name),
+            );
+        }
+        for (oi, resolution) in object_infos.iter().zip(resolved_sizes.iter()) {
+            record_size_resolution(size_summary, oi, resolution);
+        }
+        let has_corrupt_size = resolved_sizes
+            .iter()
+            .any(|resolution| matches!(resolution, SizeResolution::Corrupt { .. }));
+
         // `versioning_config` is resolved once per object by the caller
         // (`get_size`) and handed in; only `prefix_enabled` is consulted here.
 
-        let Some(lifecycle) = self.lifecycle.as_ref() else {
-            let mut cumulative_size = 0;
-            for oi in object_infos.iter() {
-                let actual_size = match oi.get_actual_size() {
-                    Ok(size) => size,
-                    Err(_) => {
-                        warn!(
-                            target: "rustfs::scanner::folder",
-                            event = EVENT_SCANNER_LIFECYCLE_ACTION,
-                            component = LOG_COMPONENT_SCANNER,
-                            subsystem = LOG_SUBSYSTEM_LIFECYCLE,
-                            bucket = %self.bucket,
-                            object = %oi.name,
-                            state = "size_lookup_failed",
-                            "Scanner lifecycle action used fallback size"
-                        );
+        let Some(lifecycle) = self.lifecycle.clone() else {
+            let mut cumulative_size: i64 = 0;
+            for (oi, resolved_size) in object_infos.iter().zip(resolved_sizes.iter()) {
+                let accounting_size = match resolved_size {
+                    SizeResolution::Known { logical, .. } => *logical,
+                    // A valid compressed legacy sentinel has no logical size,
+                    // but heal and replication still need to run. The
+                    // physical size is only an input to those operations; it
+                    // is not folded into the logical total below.
+                    SizeResolution::Unknown { physical, .. } => {
+                        self.heal_actions(oi, *physical, size_summary).await;
+                        size_summary.actions_accounting_unknown(oi);
                         continue;
                     }
+                    SizeResolution::Corrupt { .. } => continue,
                 };
 
-                let size = self.heal_actions(oi, actual_size, size_summary).await;
+                let size = self.heal_actions(oi, accounting_size, size_summary).await;
 
-                size_summary.actions_accounting(oi, size, actual_size);
+                size_summary.actions_accounting(oi, size, accounting_size);
 
-                cumulative_size += size;
+                cumulative_size = cumulative_size.saturating_add(size);
             }
 
             self.alert_excessive_versions(object_infos.len(), cumulative_size);
@@ -400,25 +665,108 @@ impl ScannerItem {
         let mut to_delete_objs: Vec<ObjectToDelete> = Vec::new();
         let mut noncurrent_events: Vec<Event> = Vec::new();
         let mut noncurrent_accounting: Vec<PendingScannerAccounting<'_>> = Vec::new();
+        let mut noncurrent_unknown: Vec<&ObjectInfo> = Vec::new();
         let mut cumulative_size = 0;
         let mut remaining_versions = object_infos.len();
         'eventLoop: {
             for (i, event) in events.iter().enumerate() {
                 let oi = &object_infos[i];
-                let actual_size = match oi.get_actual_size() {
-                    Ok(size) => size,
-                    Err(_) => {
-                        warn!(
-                            target: "rustfs::scanner::folder",
-                            event = EVENT_SCANNER_LIFECYCLE_ACTION,
-                            component = LOG_COMPONENT_SCANNER,
-                            subsystem = LOG_SUBSYSTEM_LIFECYCLE,
-                            bucket = %self.bucket,
-                            object = %oi.name,
-                            state = "size_lookup_failed",
-                            "Scanner lifecycle action used fallback size"
-                        );
-                        0
+                let known_size = resolved_sizes[i].known_size();
+                if has_corrupt_size
+                    && matches!(
+                        event.action,
+                        IlmAction::DeleteAllVersionsAction | IlmAction::DelMarkerDeleteAllVersionsAction
+                    )
+                {
+                    // An all-version delete would also remove a corrupt
+                    // sibling that could not be reconciled safely.
+                    continue;
+                }
+                if !lifecycle_event_allowed(&resolved_sizes[i], event, &lifecycle) {
+                    // An unknown logical size must not make an otherwise
+                    // non-destructive scan disappear from heal/physical-tier
+                    // accounting. Size-filtered or deferred events remain
+                    // pending, so retain the version-only physical counters.
+                    if let SizeResolution::Unknown { physical, .. } = &resolved_sizes[i] {
+                        self.heal_actions(oi, *physical, size_summary).await;
+                        size_summary.actions_accounting_unknown(oi);
+                    }
+                    continue;
+                }
+                let actual_size = match known_size {
+                    Some(size) => size,
+                    None => {
+                        match event.action {
+                            IlmAction::DeleteAction
+                            | IlmAction::DeleteRestoredAction
+                            | IlmAction::DeleteRestoredVersionAction
+                            | IlmAction::DeleteAllVersionsAction
+                            | IlmAction::DelMarkerDeleteAllVersionsAction => {
+                                let done_ilm = Metrics::time_ilm(event.action);
+                                let trace_started_at = trace_start_instant();
+                                let queued = apply_expiry_rule(event, &LcEventSrc::Scanner, oi).await;
+                                emit_scanner_ilm_action_trace(&self.bucket, &oi.name, event.action, 1, queued, trace_started_at);
+                                if record_scanner_ilm_action_if_queued(global_metrics(), event.action, 1, queued) {
+                                    done_ilm(1)();
+                                    if event.action == IlmAction::DeleteAllVersionsAction
+                                        || event.action == IlmAction::DelMarkerDeleteAllVersionsAction
+                                    {
+                                        remaining_versions = 0;
+                                    }
+                                } else if matches!(
+                                    event.action,
+                                    IlmAction::DeleteAction
+                                        | IlmAction::DeleteRestoredAction
+                                        | IlmAction::DeleteRestoredVersionAction
+                                ) {
+                                    size_summary.actions_accounting_unknown(oi);
+                                } else {
+                                    size_summary.actions_accounting_unknown(oi);
+                                    for (j, retained) in object_infos.iter().enumerate().skip(i + 1) {
+                                        match &resolved_sizes[j] {
+                                            SizeResolution::Known { logical, .. } => PendingScannerAccounting {
+                                                object: retained,
+                                                retained_size: *logical,
+                                                expired_size: 0,
+                                            }
+                                            .apply(size_summary, &mut cumulative_size, false),
+                                            SizeResolution::Unknown { .. } => {
+                                                size_summary.actions_accounting_unknown(retained);
+                                            }
+                                            SizeResolution::Corrupt { .. } => {}
+                                        }
+                                    }
+                                }
+                            }
+                            IlmAction::DeleteVersionAction => {
+                                if let Some(opt) = object_opts.get(i) {
+                                    to_delete_objs.push(ObjectToDelete {
+                                        object_name: opt.name.clone(),
+                                        version_id: opt.version_id,
+                                        ..Default::default()
+                                    });
+                                    noncurrent_events.push(event.clone());
+                                    noncurrent_unknown.push(oi);
+                                }
+                            }
+                            IlmAction::TransitionAction | IlmAction::TransitionVersionAction => {
+                                let trace_started_at = trace_start_instant();
+                                let queued = apply_transition_rule(event, &LcEventSrc::Scanner, oi).await;
+                                emit_scanner_ilm_action_trace(&self.bucket, &oi.name, event.action, 1, queued, trace_started_at);
+                                if record_scanner_ilm_action_if_queued(global_metrics(), event.action, 1, queued) {
+                                    let done_ilm = Metrics::time_ilm(event.action);
+                                    done_ilm(1)();
+                                }
+                                size_summary.actions_accounting_unknown(oi);
+                            }
+                            IlmAction::NoneAction | IlmAction::ActionCount => {
+                                if let SizeResolution::Unknown { physical, .. } = &resolved_sizes[i] {
+                                    self.heal_actions(oi, *physical, size_summary).await;
+                                }
+                                size_summary.actions_accounting_unknown(oi);
+                            }
+                        }
+                        continue;
                     }
                 };
 
@@ -446,36 +794,24 @@ impl ScannerItem {
                             done_ilm(1)();
                             remaining_versions = 0;
                         } else {
-                            PendingScannerAccounting {
-                                object: oi,
-                                retained_size: actual_size,
-                                expired_size: 0,
-                            }
-                            .apply(size_summary, &mut cumulative_size, false);
-                            for retained in object_infos.iter().skip(i + 1) {
-                                let retained_size = match retained.get_actual_size() {
-                                    Ok(size) => size,
-                                    Err(_) => {
-                                        warn!(
-                                            target: "rustfs::scanner::folder",
-                                            event = EVENT_SCANNER_LIFECYCLE_ACTION,
-                                            component = LOG_COMPONENT_SCANNER,
-                                            subsystem = LOG_SUBSYSTEM_LIFECYCLE,
-                                            bucket = %self.bucket,
-                                            object = %retained.name,
-                                            state = "size_lookup_failed",
-                                            "Scanner lifecycle action used fallback size"
-                                        );
-                                        0
-                                    }
-                                };
+                            if let Some(actual_size) = known_size {
                                 PendingScannerAccounting {
-                                    object: retained,
-                                    retained_size,
+                                    object: oi,
+                                    retained_size: actual_size,
                                     expired_size: 0,
                                 }
                                 .apply(size_summary, &mut cumulative_size, false);
                             }
+                            for (j, retained) in object_infos.iter().enumerate().skip(i + 1) {
+                                if let Some(retained_size) = resolved_sizes[j].known_size() {
+                                    PendingScannerAccounting {
+                                        object: retained,
+                                        retained_size,
+                                        expired_size: 0,
+                                    }
+                                    .apply(size_summary, &mut cumulative_size, false);
+                                }
+                            }
                         }
                         break 'eventLoop;
                     }
@@ -511,11 +847,13 @@ impl ScannerItem {
                                 version_id: opt.version_id,
                                 ..Default::default()
                             });
-                            noncurrent_accounting.push(PendingScannerAccounting {
-                                object: oi,
-                                retained_size: actual_size,
-                                expired_size: 0,
-                            });
+                            if let Some(actual_size) = known_size {
+                                noncurrent_accounting.push(PendingScannerAccounting {
+                                    object: oi,
+                                    retained_size: actual_size,
+                                    expired_size: 0,
+                                });
+                            }
                             account_now = false;
                         }
                         noncurrent_events.push(event.clone());
@@ -548,7 +886,7 @@ impl ScannerItem {
 
                 if account_now {
                     size_summary.actions_accounting(oi, size, actual_size);
-                    cumulative_size += size;
+                    cumulative_size = cumulative_size.saturating_add(size);
                 }
             }
         }
@@ -576,11 +914,20 @@ impl ScannerItem {
             }
             if record_scanner_ilm_action_if_queued(global_metrics(), action, count, queued) {
                 done_ilm(count)();
-                remaining_versions = remaining_versions.saturating_sub(noncurrent_accounting.len());
+                remaining_versions = remaining_versions_after_queued_noncurrent(
+                    remaining_versions,
+                    noncurrent_accounting.len(),
+                    noncurrent_unknown.len(),
+                );
             }
             for pending in noncurrent_accounting {
                 pending.apply(size_summary, &mut cumulative_size, queued);
             }
+            if !queued {
+                for object in noncurrent_unknown {
+                    size_summary.actions_accounting_unknown(object);
+                }
+            }
         }
         self.alert_excessive_versions(remaining_versions, cumulative_size);
     }
@@ -929,4 +1276,361 @@ mod tests {
         assert_eq!(item.object_name, "object");
         assert_eq!(item.object_path(), "object");
     }
+
+    #[test]
+    fn size_resolution_rejects_negative_overflow_and_unknown_compression() {
+        let compressed = |actual_size: i64, declared: Option<&str>| {
+            let mut user_defined = HashMap::new();
+            rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+            if let Some(declared) = declared {
+                rustfs_utils::http::insert_str(&mut user_defined, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, declared.to_string());
+            }
+            ObjectInfo {
+                size: 12,
+                actual_size,
+                user_defined: Arc::new(user_defined),
+                ..Default::default()
+            }
+        };
+
+        let normal = ObjectInfo {
+            size: 12,
+            actual_size: 10,
+            ..Default::default()
+        };
+        assert_eq!(
+            resolve_size(&normal),
+            SizeResolution::Known {
+                logical: 10,
+                physical: 12
+            }
+        );
+
+        let stale_declared_metadata = ObjectInfo {
+            size: 12,
+            actual_size: 10,
+            user_defined: Arc::new(HashMap::from([("x-rustfs-internal-actual-size".to_string(), "not-a-size".to_string())])),
+            parts: Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
+                actual_size: -2,
+                ..Default::default()
+            }]),
+            ..Default::default()
+        };
+        assert_eq!(
+            resolve_size(&stale_declared_metadata),
+            SizeResolution::Known {
+                logical: 10,
+                physical: 12
+            }
+        );
+
+        assert_eq!(
+            resolve_size(&compressed(0, Some("9"))),
+            SizeResolution::Known {
+                logical: 9,
+                physical: 12
+            }
+        );
+        assert_eq!(
+            resolve_size(&compressed(-1, None)),
+            SizeResolution::Unknown {
+                physical: 12,
+                reason: SizeResolutionReason::CompressedSizeUnknown,
+            }
+        );
+        assert!(matches!(
+            resolve_size(&compressed(0, Some("not-a-size"))),
+            SizeResolution::Corrupt {
+                reason: SizeResolutionReason::InvalidDeclaredSize,
+                ..
+            }
+        ));
+        assert!(matches!(
+            resolve_size(&ObjectInfo {
+                size: 12,
+                actual_size: -2,
+                ..Default::default()
+            }),
+            SizeResolution::Corrupt { .. }
+        ));
+        assert!(matches!(resolve_size(&compressed(0, Some("-1"))), SizeResolution::Corrupt { .. }));
+        assert!(matches!(resolve_size(&compressed(0, Some(""))), SizeResolution::Corrupt { .. }));
+
+        let unsupported = {
+            let mut object = compressed(0, None);
+            let mut metadata = (*object.user_defined).clone();
+            rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "unsupported".to_string());
+            object.user_defined = Arc::new(metadata);
+            object
+        };
+        assert!(matches!(resolve_size(&unsupported), SizeResolution::Corrupt { .. }));
+
+        let invalid_part = {
+            let mut object = compressed(0, None);
+            object.parts = Arc::new(vec![rustfs_filemeta::ObjectPartInfo {
+                size: 12,
+                actual_size: -2,
+                ..Default::default()
+            }]);
+            object
+        };
+        assert!(matches!(resolve_size(&invalid_part), SizeResolution::Corrupt { .. }));
+
+        let overflow = {
+            let mut object = compressed(0, None);
+            object.parts = Arc::new(vec![
+                rustfs_filemeta::ObjectPartInfo {
+                    size: 1,
+                    actual_size: i64::MAX,
+                    ..Default::default()
+                },
+                rustfs_filemeta::ObjectPartInfo {
+                    size: 1,
+                    actual_size: 1,
+                    ..Default::default()
+                },
+            ]);
+            object
+        };
+        assert!(matches!(resolve_size(&overflow), SizeResolution::Corrupt { .. }));
+
+        let mismatch = compressed(0, None);
+        assert!(matches!(resolve_size(&mismatch), SizeResolution::Corrupt { .. }));
+        assert_eq!(
+            resolve_size(&ObjectInfo {
+                size: 0,
+                actual_size: 0,
+                ..Default::default()
+            }),
+            SizeResolution::Known { logical: 0, physical: 0 }
+        );
+    }
+
+    #[test]
+    fn size_resolution_records_and_replays_one_identity() {
+        let version_id = uuid::Uuid::new_v4();
+        let generation = uuid::Uuid::new_v4();
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "not-a-number".to_string());
+        let corrupt = ObjectInfo {
+            bucket: "bucket".to_string(),
+            name: "object".to_string(),
+            size: 12,
+            version_id: Some(version_id),
+            data_dir: Some(generation),
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+
+        let mut summary = SizeSummary::default();
+        let resolution = resolve_size(&corrupt);
+        record_size_resolution(&mut summary, &corrupt, &resolution);
+        record_size_resolution(&mut summary, &corrupt, &resolution);
+        assert_eq!(summary.size_reconciliation.len(), 1);
+        assert_eq!(summary.size_reconciliation[0].reason, "invalid_declared_size");
+        assert_eq!(summary.size_reconciliation[0].physical_size, Some(12));
+
+        let known = ObjectInfo {
+            actual_size: 12,
+            user_defined: Arc::new(HashMap::new()),
+            ..corrupt.clone()
+        };
+        record_size_resolution(&mut summary, &known, &resolve_size(&known));
+        summary.record_reconciliation_scope(&known.bucket, &known.name);
+        assert_eq!(summary.reconciliation_scopes.len(), 1);
+        assert_eq!(summary.reconciliation_scopes[0].bucket, "bucket");
+    }
+
+    #[test]
+    fn malformed_size_has_same_ilm_accounting() {
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_ACTUAL_SIZE, "invalid".to_string());
+        let object = ObjectInfo {
+            bucket: "bucket".to_string(),
+            name: "object".to_string(),
+            size: 12,
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+        let resolution = resolve_size(&object);
+        let mut without_ilm = SizeSummary::default();
+        let mut with_ilm = SizeSummary::default();
+        record_size_resolution(&mut without_ilm, &object, &resolution);
+        record_size_resolution(&mut with_ilm, &object, &resolution);
+        assert_eq!(without_ilm.size_reconciliation, with_ilm.size_reconciliation);
+        assert_eq!(without_ilm.total_size, 0);
+        assert_eq!(with_ilm.total_size, 0);
+        assert!(without_ilm.tier_stats.is_empty());
+        assert!(with_ilm.tier_stats.is_empty());
+    }
+
+    #[test]
+    fn size_resolution_parses_once_per_version() {
+        let objects = vec![
+            ObjectInfo {
+                bucket: "bucket".to_string(),
+                name: "one".to_string(),
+                size: 1,
+                actual_size: 1,
+                ..Default::default()
+            },
+            ObjectInfo {
+                bucket: "bucket".to_string(),
+                name: "two".to_string(),
+                size: 2,
+                actual_size: -2,
+                ..Default::default()
+            },
+        ];
+        let resolutions = resolve_sizes(&objects);
+        assert_eq!(resolutions.len(), objects.len());
+        assert!(matches!(resolutions[0], SizeResolution::Known { logical: 1, .. }));
+        assert!(matches!(resolutions[1], SizeResolution::Corrupt { .. }));
+    }
+
+    #[test]
+    fn queued_unknown_noncurrent_versions_are_removed_from_alert_count() {
+        assert_eq!(remaining_versions_after_queued_noncurrent(3, 1, 2), 0);
+        assert_eq!(remaining_versions_after_queued_noncurrent(7, 2, 1), 4);
+        assert_eq!(remaining_versions_after_queued_noncurrent(usize::MAX, usize::MAX, usize::MAX), 0);
+    }
+
+    #[test]
+    fn malformed_size_blocks_size_dependent_transition_but_allows_time_only_expiry() {
+        let size_filtered = BucketLifecycleConfiguration {
+            rules: vec![s3s::dto::LifecycleRule {
+                status: s3s::dto::ExpirationStatus::from_static(s3s::dto::ExpirationStatus::ENABLED),
+                expiration: None,
+                abort_incomplete_multipart_upload: None,
+                del_marker_expiration: None,
+                id: Some("size".to_string()),
+                filter: Some(s3s::dto::LifecycleRuleFilter {
+                    object_size_greater_than: Some(1),
+                    ..Default::default()
+                }),
+                noncurrent_version_expiration: None,
+                noncurrent_version_transitions: None,
+                prefix: None,
+                transitions: None,
+            }],
+            ..Default::default()
+        };
+        let unknown = SizeResolution::Unknown {
+            physical: 12,
+            reason: SizeResolutionReason::CompressedSizeUnknown,
+        };
+        let size_event = Event {
+            action: IlmAction::DeleteAction,
+            rule_id: "size".to_string(),
+            ..Default::default()
+        };
+        assert!(!lifecycle_event_allowed(&unknown, &size_event, &size_filtered));
+        assert!(!lifecycle_event_allowed(
+            &unknown,
+            &Event {
+                action: IlmAction::TransitionAction,
+                rule_id: "size".to_string(),
+                ..Default::default()
+            },
+            &size_filtered
+        ));
+        let mixed_filters = BucketLifecycleConfiguration {
+            rules: vec![
+                size_filtered.rules[0].clone(),
+                s3s::dto::LifecycleRule {
+                    status: s3s::dto::ExpirationStatus::from_static(s3s::dto::ExpirationStatus::ENABLED),
+                    expiration: None,
+                    abort_incomplete_multipart_upload: None,
+                    del_marker_expiration: None,
+                    id: Some("time".to_string()),
+                    filter: None,
+                    noncurrent_version_expiration: None,
+                    noncurrent_version_transitions: None,
+                    prefix: None,
+                    transitions: None,
+                },
+            ],
+            ..Default::default()
+        };
+        assert!(lifecycle_event_allowed(
+            &unknown,
+            &Event {
+                action: IlmAction::DeleteAction,
+                rule_id: "time".to_string(),
+                ..Default::default()
+            },
+            &mixed_filters
+        ));
+        assert!(lifecycle_event_allowed(
+            &unknown,
+            &Event {
+                action: IlmAction::TransitionAction,
+                ..Default::default()
+            },
+            &BucketLifecycleConfiguration::default()
+        ));
+        assert!(!lifecycle_event_allowed(
+            &SizeResolution::Corrupt {
+                physical: 12,
+                reason: SizeResolutionReason::InvalidDeclaredSize,
+            },
+            &Event {
+                action: IlmAction::DeleteAction,
+                ..Default::default()
+            },
+            &BucketLifecycleConfiguration::default()
+        ));
+        assert!(lifecycle_event_allowed(
+            &SizeResolution::Known {
+                logical: 10,
+                physical: 12,
+            },
+            &Event {
+                action: IlmAction::DeleteAllVersionsAction,
+                ..Default::default()
+            },
+            &BucketLifecycleConfiguration::default()
+        ));
+        assert!(lifecycle_event_allowed(
+            &unknown,
+            &Event {
+                action: IlmAction::DeleteAction,
+                rule_id: "time-only".to_string(),
+                ..Default::default()
+            },
+            &BucketLifecycleConfiguration::default()
+        ));
+    }
+
+    #[tokio::test]
+    async fn long_object_size_reconciliation_scope_uses_bounded_identity() {
+        let object_name = "o".repeat(600);
+        let mut item = scanner_item_with_prefix("");
+        item.object_name = object_name.clone();
+
+        let mut metadata = HashMap::new();
+        rustfs_utils::http::insert_str(&mut metadata, rustfs_utils::http::SUFFIX_COMPRESSION, "zstd".to_string());
+        let object = ObjectInfo {
+            bucket: item.bucket.clone(),
+            name: object_name.clone(),
+            size: 12,
+            actual_size: -1,
+            version_id: Some(uuid::Uuid::new_v4()),
+            user_defined: Arc::new(metadata),
+            ..Default::default()
+        };
+        let mut summary = SizeSummary::default();
+        item.apply_actions(vec![object], None, VersioningConfiguration::default(), &mut summary)
+            .await;
+
+        let bounded_bucket = bounded_reconciliation_field(&item.bucket);
+        let bounded_object = bounded_reconciliation_field(&object_name);
+        assert_eq!(summary.reconciliation_scopes[0].bucket, bounded_bucket);
+        assert_eq!(summary.reconciliation_scopes[0].object, bounded_object);
+        assert_eq!(summary.size_reconciliation[0].object, bounded_object);
+        assert_eq!(summary.versions, 1);
+        assert_eq!(summary.total_size, 0);
+    }
 }

crates/scanner/src/scanner_folder/tests.rs

@@ -388,6 +388,66 @@ async fn test_record_failed_ttl_zero_noop() {
     assert!(!scanner.should_skip_failed("path2"));
 }
 
+#[tokio::test]
+async fn malformed_size_reconciliation_replays_after_restart() {
+    let (mut scanner, temp_dir) = build_test_scanner().await;
+    let _guard = TestGuard::new(60, 100, &mut scanner, temp_dir);
+
+    let entry = SizeReconciliationEntry {
+        key: "1:b|6:object|0:|0:".to_string(),
+        bucket: "b".to_string(),
+        object: "object".to_string(),
+        reason: "invalid_declared_size".to_string(),
+        physical_size: Some(12),
+        ..Default::default()
+    };
+    let mut summary = SizeSummary::default();
+    summary.record_size_reconciliation(entry.clone());
+    summary.record_reconciliation_scope("b", "object");
+    scanner.apply_size_reconciliation(&summary);
+    scanner.apply_size_reconciliation(&summary);
+
+    assert_eq!(scanner.new_cache.info.size_reconciliation.len(), 1);
+    assert_eq!(scanner.update_cache.info.size_reconciliation.len(), 1);
+    assert_eq!(scanner.new_cache.info.size_reconciliation[&entry.key].attempts, 2);
+
+    let encoded = rmp_serde::to_vec_named(&scanner.new_cache.info).expect("size ledger should encode");
+    let decoded: crate::data_usage_define::DataUsageCacheInfo =
+        rmp_serde::from_slice(&encoded).expect("size ledger should decode");
+    assert_eq!(decoded.size_reconciliation.len(), 1);
+    assert_eq!(decoded.size_reconciliation[&entry.key].reason, "invalid_declared_size");
+
+    let mut resolved = SizeSummary::default();
+    resolved.record_reconciliation_scope("b", "object");
+    scanner.apply_size_reconciliation(&resolved);
+    assert!(scanner.new_cache.info.size_reconciliation.is_empty());
+    assert!(scanner.update_cache.info.size_reconciliation.is_empty());
+}
+
+#[tokio::test]
+async fn malformed_size_reconciliation_clears_bounded_long_object_scope() {
+    let (mut scanner, temp_dir) = build_test_scanner().await;
+    let _guard = TestGuard::new(60, 100, &mut scanner, temp_dir);
+    let long_object = "o".repeat(600);
+    let bounded_object = item_actions::bounded_reconciliation_field(&long_object);
+    let entry = SizeReconciliationEntry {
+        key: "long-object-key".to_string(),
+        bucket: "b".to_string(),
+        object: bounded_object,
+        reason: "invalid_declared_size".to_string(),
+        ..Default::default()
+    };
+    let mut summary = SizeSummary::default();
+    summary.record_size_reconciliation(entry);
+    scanner.apply_size_reconciliation(&summary);
+    assert_eq!(scanner.new_cache.info.size_reconciliation.len(), 1);
+
+    let mut resolved = SizeSummary::default();
+    resolved.record_reconciliation_scope("b", &long_object);
+    scanner.apply_size_reconciliation(&resolved);
+    assert!(scanner.new_cache.info.size_reconciliation.is_empty());
+}
+
 #[test]
 fn test_classify_get_size_failure_marks_metadata_heal_object_path() {
     let temp_dir = std::env::temp_dir();

rustfs/src/admin/handlers/replication.rs

@@ -73,8 +73,6 @@ enum TargetUpdateOp {
     /// Connection group: credentials plus endpoint, target bucket, and TLS settings.
     Credentials,
     Sync,
-    /// Per-target read-proxy opt-out (`disableProxy`).
-    Proxy,
     Bandwidth,
     Path,
 }
@@ -83,13 +81,12 @@ fn parse_remote_target_update_ops(queries: &HashMap<String, String>) -> S3Result
     const SUPPORTED_OPS: &[(&str, TargetUpdateOp)] = &[
         ("creds", TargetUpdateOp::Credentials),
         ("sync", TargetUpdateOp::Sync),
-        ("proxy", TargetUpdateOp::Proxy),
         ("bandwidth", TargetUpdateOp::Bandwidth),
         ("path", TargetUpdateOp::Path),
     ];
     // Present in the MinIO wire contract, but they drive target fields this
     // version rejects as unsupported — fail loudly instead of silently ignoring.
-    const UNSUPPORTED_OPS: &[&str] = &["healthcheck", "edge", "edgeSyncBeforeExpiry"];
+    const UNSUPPORTED_OPS: &[&str] = &["proxy", "healthcheck", "edge", "edgeSyncBeforeExpiry"];
 
     for key in UNSUPPORTED_OPS {
         if queries.get(*key).is_some_and(|value| value == "true") {
@@ -315,10 +312,11 @@ impl RemoteTargetRequest {
             ));
         }
 
-        for (unsupported, configured) in REMOTE_TARGET_UNSUPPORTED_FIELDS
-            .iter()
-            .copied()
-            .zip([self.edge, self.edge_sync_before_expiry])
+        for (unsupported, configured) in
+            REMOTE_TARGET_UNSUPPORTED_FIELDS
+                .iter()
+                .copied()
+                .zip([self.disable_proxy, self.edge, self.edge_sync_before_expiry])
         {
             if configured {
                 return Err(s3_error!(
@@ -704,7 +702,6 @@ impl Operation for SetRemoteTargetHandler {
                         target.deployment_id = remote_target.deployment_id.clone();
                     }
                     TargetUpdateOp::Sync => target.replication_sync = remote_target.replication_sync,
-                    TargetUpdateOp::Proxy => target.disable_proxy = remote_target.disable_proxy,
                     TargetUpdateOp::Bandwidth => target.bandwidth_limit = remote_target.bandwidth_limit,
                     TargetUpdateOp::Path => target.path = remote_target.path.clone(),
                 }
@@ -1523,7 +1520,6 @@ mod tests {
             ("update", "true"),
             ("creds", "true"),
             ("sync", "true"),
-            ("proxy", "true"),
             ("bandwidth", "true"),
             ("path", "true"),
         ]))
@@ -1533,7 +1529,6 @@ mod tests {
             vec![
                 TargetUpdateOp::Credentials,
                 TargetUpdateOp::Sync,
-                TargetUpdateOp::Proxy,
                 TargetUpdateOp::Bandwidth,
                 TargetUpdateOp::Path
             ]
@@ -2075,6 +2070,7 @@ mod tests {
             ("credentials.session_token", serde_json::json!("session-token")),
             ("credentials.expiration", serde_json::json!("2026-01-01T00:00:00Z")),
             ("api", serde_json::json!("s3v2")),
+            ("disableProxy", serde_json::json!(true)),
             ("edge", serde_json::json!(true)),
             ("edgeSyncBeforeExpiry", serde_json::json!(true)),
         ] {
@@ -2304,44 +2300,6 @@ mod tests {
         assert!(!REMOTE_TARGET_UNSUPPORTED_FIELDS.contains(&"healthCheckDuration"));
     }
 
-    #[test]
-    fn remote_target_disable_proxy_is_declared_writable_edge_stays_unsupported() {
-        assert!(REMOTE_TARGET_WRITABLE_FIELDS.contains(&"disableProxy"));
-        assert!(!REMOTE_TARGET_UNSUPPORTED_FIELDS.contains(&"disableProxy"));
-        // edge sync has no implementation behind it — it must stay rejected.
-        assert!(REMOTE_TARGET_UNSUPPORTED_FIELDS.contains(&"edge"));
-        assert!(REMOTE_TARGET_UNSUPPORTED_FIELDS.contains(&"edgeSyncBeforeExpiry"));
-    }
-
-    #[test]
-    fn remote_target_create_accepts_disable_proxy() {
-        let mut request = valid_remote_target_request();
-        request["disableProxy"] = serde_json::json!(true);
-
-        let target = serde_json::from_value::<RemoteTargetRequest>(request)
-            .expect("request should deserialize")
-            .into_bucket_target()
-            .expect("disableProxy is a supported per-target read-proxy opt-out");
-
-        assert!(target.disable_proxy);
-    }
-
-    #[test]
-    fn update_body_with_proxy_op_toggles_disable_proxy_without_credentials() {
-        // Mirrors the other partial-update groups: a proxy-only update body may
-        // omit the connection fields entirely.
-        let body = serde_json::json!({
-            "arn": "arn:rustfs:replication:us-east-1:dep:target",
-            "type": "replication",
-            "disableProxy": true
-        });
-        let request: RemoteTargetRequest = serde_json::from_value(body).expect("partial update body should deserialize");
-        let target = request
-            .into_update_bucket_target(&[TargetUpdateOp::Proxy])
-            .expect("proxy-only update must not require credentials");
-        assert!(target.disable_proxy);
-    }
-
     #[test]
     fn remote_target_capability_fields_do_not_overlap() {
         for field in REMOTE_TARGET_UNSUPPORTED_FIELDS {

rustfs/src/admin/handlers/system.rs

@@ -1262,9 +1262,7 @@ mod tests {
         assert_eq!(response.summary.manual_transition_jobs.state, CapabilityState::Supported);
         assert_eq!(response.replication.contract_version, 1);
         assert_eq!(response.replication.bucket_replication.contract_version, 1);
-        // v2: disableProxy moved from unsupported to writable (per-target
-        // read-proxy opt-out reached the admin API).
-        assert_eq!(response.replication.remote_targets.contract_version, 2);
+        assert_eq!(response.replication.remote_targets.contract_version, 1);
         assert_eq!(response.replication.bucket_replication.status.state, CapabilityState::Supported);
         assert_eq!(response.replication.remote_targets.status.state, CapabilityState::Supported);
         assert_eq!(
@@ -1295,15 +1293,7 @@ mod tests {
                 .remote_targets
                 .fields
                 .iter()
-                .any(|field| field.name == "disableProxy" && field.state == super::ReplicationFieldState::Supported)
-        );
-        assert!(
-            response
-                .replication
-                .remote_targets
-                .fields
-                .iter()
-                .any(|field| field.name == "edge" && field.state == super::ReplicationFieldState::Unsupported)
+                .any(|field| field.name == "disableProxy" && field.state == super::ReplicationFieldState::Unsupported)
         );
         assert!(
             response
@@ -1374,7 +1364,7 @@ mod tests {
         assert_eq!(value["summary"]["manual_transition_jobs"]["state"], "supported");
         assert_eq!(value["replication"]["contract_version"], 1);
         assert_eq!(value["replication"]["bucket_replication"]["contract_version"], 1);
-        assert_eq!(value["replication"]["remote_targets"]["contract_version"], 2);
+        assert_eq!(value["replication"]["remote_targets"]["contract_version"], 1);
         assert_eq!(value["replication"]["bucket_replication"]["status"]["state"], "supported");
         assert_eq!(value["replication"]["remote_targets"]["status"]["state"], "supported");
         assert_eq!(
@@ -1393,14 +1383,7 @@ mod tests {
                 .as_array()
                 .expect("remote target fields should be an array")
                 .iter()
-                .any(|field| field["name"] == "disableProxy" && field["state"] == "supported")
-        );
-        assert!(
-            value["replication"]["remote_targets"]["fields"]
-                .as_array()
-                .expect("remote target fields should be an array")
-                .iter()
-                .any(|field| field["name"] == "edge" && field["state"] == "unsupported")
+                .any(|field| field["name"] == "disableProxy" && field["state"] == "unsupported")
         );
         assert!(
             value["replication"]["remote_targets"]["fields"]