perf(scanner): cut per-object allocations in the scan hot path (#6274)

crates/scanner/src/lib.rs

@@ -26,6 +26,8 @@ use http::HeaderMap;
 use rustfs_config::server_config::{Config as ServerConfig, get_global_server_config as config_get_global_server_config};
 use std::path::PathBuf;
 use std::sync::Arc;
+use std::sync::RwLock;
+use std::time::{Duration, Instant};
 use storage_api::owner::{
     ECSTORE_BUCKET_META_PREFIX, ECSTORE_RUSTFS_META_BUCKET, ECSTORE_STORAGE_FORMAT_FILE, ECSTORE_STORAGECLASS_RRS,
     ECSTORE_STORAGECLASS_STANDARD, ECSTORE_TRANSITION_COMPLETE, EcstoreBucketTargetSys, EcstoreBucketVersioningSys, EcstoreDisk,
@@ -33,7 +35,7 @@ use storage_api::owner::{
     EcstoreDiskResult, EcstoreErrorType, EcstoreEvaluator, EcstoreEvent, EcstoreLcEventSrc, EcstoreLifecycle,
     EcstoreListPathRawOptions, EcstoreNsScannerOpenRequest, EcstoreObjectOpts, EcstoreReplicationConfigurationExt,
     EcstoreReplicationScannerBridge, EcstoreResultType, EcstoreScanGuard, EcstoreSetDisks, EcstoreStorageError, EcstoreStore,
-    EcstoreTierConfig, EcstoreVersioningApi, HTTPPreconditions, HTTPRangeSpec, ObjectIO, ObjectOperations, ObjectToDelete,
+    EcstoreVersioningApi, HTTPPreconditions, HTTPRangeSpec, ObjectIO, ObjectOperations, ObjectToDelete,
     ScannerReplicationHealObject, ScannerReplicationHealResult, ScannerReplicationQueueAdmission, ecstore_apply_expiry_rule,
     ecstore_apply_transition_rule, ecstore_expiry_state_handle, ecstore_get_global_tier_config_mgr, ecstore_get_lifecycle_config,
     ecstore_get_object_lock_config, ecstore_get_replication_config, ecstore_invalidate_admin_data_usage_snapshot_cache,
@@ -363,8 +365,46 @@ pub(crate) fn resolve_scanner_server_config() -> Option<ServerConfig> {
     config_get_global_server_config()
 }
 
-pub(crate) async fn list_runtime_tiers() -> Vec<EcstoreTierConfig> {
+/// How long the scanner caches the runtime tier-name list before re-reading
-    ecstore_get_global_tier_config_mgr().read().await.list_tiers()
+/// the tier configuration manager.
+const TIER_NAME_CACHE_TTL: Duration = Duration::from_secs(30);
+
+/// Process-wide TTL cache of runtime tier names.
+///
+/// The scan hot path only needs tier *names* to seed `SizeSummary::tier_stats`
+/// per object, but every `list_tiers()` call clones each full `TierConfig`
+/// (endpoints, credentials, prefixes) from the global manager. Caching just
+/// the names keeps the per-object cost at an `Arc` clone.
+///
+/// Staleness bounds: a newly added tier starts showing up in scans at most
+/// `TIER_NAME_CACHE_TTL` later; a removed tier can leave an all-zero
+/// `TierStats` seed behind for one cache generation, which merges harmlessly
+/// by key in per-object accounting and disappears on the next refresh.
+static TIER_NAME_CACHE: RwLock<Option<(Instant, Arc<[String]>)>> = RwLock::new(None);
+
+/// Tier names currently registered in the tier configuration, cached for
+/// `TIER_NAME_CACHE_TTL`.
+pub(crate) async fn runtime_tier_names() -> Arc<[String]> {
+    {
+        let cached = TIER_NAME_CACHE.read().unwrap_or_else(|err| err.into_inner()).clone();
+        if let Some((refreshed_at, names)) = cached
+            && refreshed_at.elapsed() < TIER_NAME_CACHE_TTL
+        {
+            return names;
+        }
+    }
+
+    let tiers = ecstore_get_global_tier_config_mgr().read().await.list_tiers();
+    let names: Arc<[String]> = tiers.iter().map(|tier| tier.name.clone()).collect::<Vec<_>>().into();
+    *TIER_NAME_CACHE.write().unwrap_or_else(|err| err.into_inner()) = Some((Instant::now(), Arc::clone(&names)));
+    names
+}
+
+/// Test-only cache reset; the production cache has no invalidation hook
+/// because the TTL is its only refresh path.
+#[cfg(test)]
+fn reset_tier_name_cache_for_test() {
+    *TIER_NAME_CACHE.write().unwrap_or_else(|err| err.into_inner()) = None;
 }
 
 pub(crate) async fn enqueue_runtime_free_version(oi: ScannerObjectInfo) {
@@ -561,6 +601,20 @@ mod tests {
     use super::*;
     use serial_test::serial;
 
+    #[tokio::test]
+    #[serial]
+    async fn runtime_tier_names_serves_cached_arc_within_ttl() {
+        reset_tier_name_cache_for_test();
+        // The tier config manager is unconfigured in unit tests, so the
+        // first call populates the cache from an empty tier list...
+        let first = runtime_tier_names().await;
+        assert!(first.is_empty());
+        // ...and a second call within the TTL must return the cached Arc
+        // (pointer-equal) without re-reading the manager.
+        let second = runtime_tier_names().await;
+        assert!(Arc::ptr_eq(&first, &second));
+    }
+
     #[test]
     #[serial]
     fn foreground_read_guard_tracks_stream_lifetime() {

crates/scanner/src/scanner_folder.rs

@@ -45,7 +45,7 @@ use rustfs_common::metrics::{
 use rustfs_common::trace_bus::{TraceEvent, TraceFunc, TraceKind, trace_emit, trace_subscriber_count};
 use rustfs_filemeta::{MetaCacheEntries, MetaCacheEntry, MetadataResolutionParams};
 use rustfs_utils::path::{SLASH_SEPARATOR, path_join_buf};
-use s3s::dto::{BucketLifecycleConfiguration, ObjectLockConfiguration};
+use s3s::dto::{BucketLifecycleConfiguration, ObjectLockConfiguration, VersioningConfiguration};
 use time::OffsetDateTime;
 use tokio::select;
 use tokio::sync::mpsc;
@@ -53,10 +53,10 @@ use tokio_util::sync::CancellationToken;
 use tracing::{debug, error, warn};
 
 use crate::{
-    BucketVersioningSys, Disk, DiskError, DiskInfoOptions, Evaluator, Event, LcEventSrc, ListPathRawOptions, ObjectOpts,
+    Disk, DiskError, DiskInfoOptions, Evaluator, Event, LcEventSrc, ListPathRawOptions, ObjectOpts, ReplicationConfig,
-    ReplicationConfig, ReplicationHealObject, ReplicationQueueAdmission, ReplicationStatusType, STORAGE_FORMAT_FILE,
+    ReplicationHealObject, ReplicationQueueAdmission, ReplicationStatusType, STORAGE_FORMAT_FILE, ScannerDiskExt as _,
-    ScannerDiskExt as _, ScannerLifecycleConfigExt as _, ScannerVersioningConfigExt as _, StorageError, apply_expiry_rule,
+    ScannerLifecycleConfigExt as _, ScannerVersioningConfigExt as _, StorageError, apply_expiry_rule, apply_transition_rule,
-    apply_transition_rule, enqueue_runtime_newer_noncurrent, is_reserved_or_invalid_bucket, list_path_raw, path2_bucket_object,
+    enqueue_runtime_newer_noncurrent, is_reserved_or_invalid_bucket, list_path_raw, path2_bucket_object,
     path2_bucket_object_with_base_path, queue_replication_heal, scanner_is_erasure,
     scanner_replication_config_for_lifecycle_eval,
 };
@@ -934,6 +934,7 @@ impl ScannerItem {
         &mut self,
         object_infos: Vec<ObjectInfo>,
         lock_retention: Option<Arc<ObjectLockConfiguration>>,
+        versioning_config: VersioningConfiguration,
         size_summary: &mut SizeSummary,
     ) {
         if object_infos.is_empty() {
@@ -958,21 +959,8 @@ impl ScannerItem {
             "Scanner lifecycle evaluation started"
         );
 
-        let versioning_config = match BucketVersioningSys::get(&self.bucket).await {
+        // `versioning_config` is resolved once per object by the caller
-            Ok(versioning_config) => versioning_config,
+        // (`get_size`) and handed in; only `prefix_enabled` is consulted here.
-            Err(_) => {
-                warn!(
-                    target: "rustfs::scanner::folder",
-                    event = EVENT_SCANNER_LIFECYCLE_ACTION,
-                    component = LOG_COMPONENT_SCANNER,
-                    subsystem = LOG_SUBSYSTEM_LIFECYCLE,
-                    bucket = %self.bucket,
-                    state = "versioning_lookup_failed_defaulting",
-                    "Scanner lifecycle action falling back to default bucket versioning"
-                );
-                Default::default()
-            }
-        };
 
         let Some(lifecycle) = self.lifecycle.as_ref() else {
             let mut cumulative_size = 0;
@@ -1402,6 +1390,11 @@ impl ScannerItem {
     fn alert_excessive_versions(&self, remaining_versions: usize, cumulative_size: i64) {
         ensure_scanner_alert_metrics_registered();
         let (too_many_versions, too_large_versions) = should_alert_excessive_versions(remaining_versions, cumulative_size);
+        // Threshold check first so healthy objects never pay for the
+        // object-path allocation below.
+        if !too_many_versions && !too_large_versions {
+            return;
+        }
         let object_path = self.object_path();
         if too_many_versions {
             global_metrics().record_scanner_source_executed(ScannerWorkSource::Alerts, 1);

crates/scanner/src/scanner_io.rs

@@ -32,7 +32,9 @@ use rustfs_data_usage::{BucketTargetUsageInfo, BucketUsageInfo};
 use rustfs_filemeta::FileMeta;
 use rustfs_lock::{LockError, NamespaceLockGuard};
 use rustfs_utils::path::path_join_buf;
-use s3s::dto::{BucketLifecycleConfiguration, ObjectLockConfiguration, ObjectLockEnabled, ReplicationConfiguration};
+use s3s::dto::{
+    BucketLifecycleConfiguration, ObjectLockConfiguration, ObjectLockEnabled, ReplicationConfiguration, VersioningConfiguration,
+};
 use sha2::{Digest as _, Sha256};
 use std::collections::{HashMap, HashSet};
 use std::future::Future;
@@ -55,7 +57,7 @@ use crate::{
     BucketTargetSys, BucketVersioningSys, Disk, DiskError, ECStore, EcstoreError as Error, EcstoreResult as Result,
     RUSTFS_META_BUCKET, ReplicationConfig, STORAGE_FORMAT_FILE, ScannerDiskExt as _, ScannerLifecycleConfigExt as _,
     ScannerReplicationConfigExt as _, ScannerVersioningConfigExt as _, SetDisks, StorageError, enqueue_runtime_free_version,
-    get_lifecycle_config, get_object_lock_config, get_replication_config, list_runtime_tiers, storageclass,
+    get_lifecycle_config, get_object_lock_config, get_replication_config, runtime_tier_names, storageclass,
 };
 
 pub(crate) const SCANNER_SKIP_FILE_ERROR: &str = "skip file";
@@ -63,6 +65,11 @@ pub(crate) const SCANNER_METADATA_CORRUPT_ERROR: &str = "scanner metadata corrup
 pub(crate) const SCANNER_METADATA_TRANSIENT_ERROR: &str = "scanner metadata transient";
 const LOG_COMPONENT_SCANNER: &str = "scanner";
 const LOG_SUBSYSTEM_IO: &str = "io";
+// Mirrors `scanner_folder.rs` so the versioning-lookup fallback warn keeps its
+// historical `rustfs::scanner::folder` lifecycle event identity after the
+// lookup moved into `get_size`.
+const LOG_SUBSYSTEM_LIFECYCLE: &str = "lifecycle";
+const EVENT_SCANNER_LIFECYCLE_ACTION: &str = "scanner_lifecycle_action";
 const EVENT_SCANNER_DISK_BUCKET_STATE: &str = "scanner_disk_bucket_state";
 const EVENT_SCANNER_DATA_USAGE_STREAM: &str = "scanner_data_usage_stream";
 const EVENT_SCANNER_CACHE_PERSIST_STATE: &str = "scanner_cache_persist_state";
@@ -3822,6 +3829,24 @@ impl ScannerIOCache for SetDisks {
     }
 }
 
+/// Seed [`SizeSummary::tier_stats`] from the cached tier-name list.
+///
+/// Preserves the original seeding semantics: with no tiers configured the map
+/// stays completely empty (STANDARD/RRS are not seeded either); otherwise the
+/// standard storage classes are seeded alongside every configured tier so
+/// per-object accounting always finds its tier key.
+fn tier_stats_template(tier_names: &[String]) -> HashMap<String, TierStats> {
+    let mut tier_stats = HashMap::with_capacity(tier_names.len() + 2);
+    for tier_name in tier_names {
+        tier_stats.insert(tier_name.clone(), TierStats::default());
+    }
+    if !tier_stats.is_empty() {
+        tier_stats.insert(storageclass::STANDARD.to_string(), TierStats::default());
+        tier_stats.insert(storageclass::RRS.to_string(), TierStats::default());
+    }
+    tier_stats
+}
+
 #[async_trait::async_trait]
 impl ScannerIODisk for Disk {
     async fn get_size(&self, mut item: ScannerItem) -> Result<SizeSummary> {
@@ -3861,10 +3886,26 @@ impl ScannerIODisk for Disk {
             }
         };
 
-        let versioned = BucketVersioningSys::get(&item.bucket)
+        // Single versioning lookup per object, shared with `apply_actions`
-            .await
+        // (which used to query it a second time). On failure keep the
-            .map(|v| v.versioned(&item.object_path()))
+        // historical fallback: default configuration (versioned = false) plus
-            .unwrap_or(false);
+        // the warn that `apply_actions` used to emit.
+        let versioning_config = match BucketVersioningSys::get(&item.bucket).await {
+            Ok(versioning_config) => versioning_config,
+            Err(_) => {
+                warn!(
+                    target: "rustfs::scanner::folder",
+                    event = EVENT_SCANNER_LIFECYCLE_ACTION,
+                    component = LOG_COMPONENT_SCANNER,
+                    subsystem = LOG_SUBSYSTEM_LIFECYCLE,
+                    bucket = %item.bucket,
+                    state = "versioning_lookup_failed_defaulting",
+                    "Scanner lifecycle action falling back to default bucket versioning"
+                );
+                VersioningConfiguration::default()
+            }
+        };
+        let versioned = versioning_config.versioned(&item.object_path());
 
         let object_infos = fivs
             .versions
@@ -3879,19 +3920,10 @@ impl ScannerIODisk for Disk {
 
         let mut size_summary = SizeSummary::default();
 
-        let tiers = list_runtime_tiers().await;
+        // Tier names come from the process-wide TTL cache; seeding from them
-
+        // replaces the per-object clone of every full TierConfig.
-        for tier in tiers.iter() {
+        let tier_names = runtime_tier_names().await;
-            size_summary.tier_stats.insert(tier.name.clone(), TierStats::default());
+        size_summary.tier_stats = tier_stats_template(&tier_names);
-        }
-        if !size_summary.tier_stats.is_empty() {
-            size_summary
-                .tier_stats
-                .insert(storageclass::STANDARD.to_string(), TierStats::default());
-            size_summary
-                .tier_stats
-                .insert(storageclass::RRS.to_string(), TierStats::default());
-        }
 
         let lock_config = object_lock_config_for_scanner_item(&item).await;
 
@@ -3901,7 +3933,8 @@ impl ScannerIODisk for Disk {
         // `object_infos`.
         global_metrics().record_scanner_versions_scanned(object_infos.len() as u64);
 
-        item.apply_actions(object_infos, lock_config, &mut size_summary).await;
+        item.apply_actions(object_infos, lock_config, versioning_config, &mut size_summary)
+            .await;
 
         if !free_version_infos.is_empty() {
             for oi in free_version_infos {
@@ -4968,6 +5001,23 @@ mod tests {
         assert!(is_xl_meta_path("/data/bucket/object/xl.meta"));
     }
 
+    #[test]
+    fn tier_stats_template_seeds_tiers_and_standard_classes() {
+        let template = tier_stats_template(&["WARM".to_string(), "COLD".to_string()]);
+
+        assert_eq!(template.len(), 4);
+        for tier in ["WARM", "COLD", storageclass::STANDARD, storageclass::RRS] {
+            assert_eq!(template.get(tier), Some(&TierStats::default()), "missing seed for tier {tier}");
+        }
+    }
+
+    #[test]
+    fn tier_stats_template_stays_empty_without_tiers() {
+        let template = tier_stats_template(&[]);
+
+        assert!(template.is_empty());
+    }
+
     #[tokio::test]
     async fn get_size_treats_missing_metadata_as_skip_file() {
         let temp_dir = std::env::temp_dir().join(format!("rustfs-scanner-missing-meta-{}", Uuid::new_v4()));

crates/scanner/src/storage_api.rs

@@ -99,7 +99,6 @@ pub(crate) use rustfs_ecstore::api::set_disk::SetDisks as EcstoreSetDisks;
 pub(crate) use rustfs_ecstore::api::storage::ECStore as EcstoreStore;
 #[cfg(test)]
 pub(crate) use rustfs_ecstore::api::storage::init_local_disks_with_instance_ctx as ecstore_init_local_disks_with_instance_ctx;
-pub(crate) use rustfs_ecstore::api::tier::tier_config::TierConfig as EcstoreTierConfig;
 use rustfs_storage_api as storage_contracts;
 
 pub(crate) mod owner {
@@ -114,15 +113,15 @@ pub(crate) mod owner {
         EcstoreDiskLocation, EcstoreDiskResult, EcstoreErrorType, EcstoreEvaluator, EcstoreEvent, EcstoreEventArgs,
         EcstoreLcEventSrc, EcstoreLifecycle, EcstoreListPathRawOptions, EcstoreNsScannerOpenRequest, EcstoreObjectOpts,
         EcstoreReplicationConfigurationExt, EcstoreReplicationScannerBridge, EcstoreResultType, EcstoreScanGuard,
-        EcstoreSetDisks, EcstoreStorageError, EcstoreStore, EcstoreTierConfig, EcstoreVersioningApi,
+        EcstoreSetDisks, EcstoreStorageError, EcstoreStore, EcstoreVersioningApi, ScannerReplicationHealObject,
-        ScannerReplicationHealObject, ScannerReplicationHealResult, ScannerReplicationQueueAdmission, ecstore_apply_expiry_rule,
+        ScannerReplicationHealResult, ScannerReplicationQueueAdmission, ecstore_apply_expiry_rule, ecstore_apply_transition_rule,
-        ecstore_apply_transition_rule, ecstore_expiry_state_handle, ecstore_get_global_tier_config_mgr,
+        ecstore_expiry_state_handle, ecstore_get_global_tier_config_mgr, ecstore_get_lifecycle_config,
-        ecstore_get_lifecycle_config, ecstore_get_object_lock_config, ecstore_get_replication_config,
+        ecstore_get_object_lock_config, ecstore_get_replication_config, ecstore_invalidate_admin_data_usage_snapshot_cache,
-        ecstore_invalidate_admin_data_usage_snapshot_cache, ecstore_invalidate_data_usage_snapshot_cache, ecstore_is_erasure,
+        ecstore_invalidate_data_usage_snapshot_cache, ecstore_is_erasure, ecstore_is_erasure_sd,
-        ecstore_is_erasure_sd, ecstore_is_reserved_or_invalid_bucket, ecstore_list_path_raw,
+        ecstore_is_reserved_or_invalid_bucket, ecstore_list_path_raw, ecstore_object_opts_from_object_info,
-        ecstore_object_opts_from_object_info, ecstore_path2_bucket_object, ecstore_path2_bucket_object_with_base_path,
+        ecstore_path2_bucket_object, ecstore_path2_bucket_object_with_base_path, ecstore_read_config,
-        ecstore_read_config, ecstore_replace_bucket_usage_memory_from_info, ecstore_resolve_object_store_handle,
+        ecstore_replace_bucket_usage_memory_from_info, ecstore_resolve_object_store_handle, ecstore_save_config,
-        ecstore_save_config, ecstore_send_event, scanner_replication_config_for_lifecycle_eval,
+        ecstore_send_event, scanner_replication_config_for_lifecycle_eval,
     };
 
     #[cfg(test)]