fix: discarding results do not attempt in-memory metacache writer (#11163)

Optimizations include

- do not write the metacache block if the size of the
  block is '0' and it is the first block - where listing
  is attempted for a transient prefix, this helps to
  avoid creating lots of empty metacache entries for
  `minioMetaBucket`

- avoid the entire initialization sequence of cacheCh
  , metacacheBlockWriter if we are simply going to skip
  them when discardResults is set to true.

- No need to hold write locks while writing metacache
  blocks - each block is unique, per bucket, per prefix
  and also is written by a single node.

cmd/data-crawler.go

@@ -570,8 +570,13 @@ func (f *folderScanner) scanQueuedLevels(ctx context.Context, folders []cachedFo
 							bucket:    bucket,
 							object:    entry.name,
 							versionID: "",
+							opts: &madmin.HealOpts{
+								Remove: true,
+							},
 						}, madmin.HealItemObject)
+						if !isErrObjectNotFound(err) && !isErrVersionNotFound(err) {
 							logger.LogIf(ctx, err)
+						}
 						foundObjs = foundObjs || err == nil
 						return
 					}
@@ -583,8 +588,13 @@ func (f *folderScanner) scanQueuedLevels(ctx context.Context, folders []cachedFo
 							bucket:    bucket,
 							object:    fiv.Name,
 							versionID: ver.VersionID,
+							opts: &madmin.HealOpts{
+								Remove: true,
+							},
 						}, madmin.HealItemObject)
+						if !isErrObjectNotFound(err) && !isErrVersionNotFound(err) {
 							logger.LogIf(ctx, err)
+						}
 						foundObjs = foundObjs || err == nil
 					}
 				},

cmd/erasure-object.go

@@ -674,11 +674,13 @@ func (er erasureObjects) putObject(ctx context.Context, bucket string, object st
 		return ObjectInfo{}, IncompleteBody{Bucket: bucket, Object: object}
 	}
 
+	if !opts.NoLock {
 		lk := er.NewNSLock(bucket, object)
 		if err := lk.GetLock(ctx, globalOperationTimeout); err != nil {
 			return ObjectInfo{}, err
 		}
 		defer lk.Unlock()
+	}
 
 	for i, w := range writers {
 		if w == nil {

cmd/global-heal.go

@@ -130,8 +130,10 @@ func healErasureSet(ctx context.Context, setIndex int, buckets []BucketInfo, dis
 		bucket: minioMetaBucket,
 		object: backendEncryptedFile,
 	}, madmin.HealItemMetadata); err != nil {
+		if !isErrObjectNotFound(err) && !isErrVersionNotFound(err) {
 			logger.LogIf(ctx, err)
 		}
+	}
 
 	// Heal all buckets with all objects
 	for _, bucket := range buckets {
@@ -139,8 +141,10 @@ func healErasureSet(ctx context.Context, setIndex int, buckets []BucketInfo, dis
 		if err := bgSeq.queueHealTask(healSource{
 			bucket: bucket.Name,
 		}, madmin.HealItemBucket); err != nil {
+			if !isErrObjectNotFound(err) && !isErrVersionNotFound(err) {
 				logger.LogIf(ctx, err)
 			}
+		}
 
 		var entryChs []FileInfoVersionsCh
 		var mu sync.Mutex
@@ -179,11 +183,13 @@ func healErasureSet(ctx context.Context, setIndex int, buckets []BucketInfo, dis
 					object:    version.Name,
 					versionID: version.VersionID,
 				}, madmin.HealItemObject); err != nil {
+					if !isErrObjectNotFound(err) && !isErrVersionNotFound(err) {
 						logger.LogIf(ctx, err)
 					}
 				}
 			}
 		}
+	}
 
 	return nil
 }

cmd/metacache-set.go

@@ -615,13 +615,18 @@ func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions) (entr
 	}
 
 	// Create output for our results.
-	cacheCh := make(chan metaCacheEntry, metacacheBlockSize)
+	var cacheCh chan metaCacheEntry
+	if !o.discardResult {
+		cacheCh = make(chan metaCacheEntry, metacacheBlockSize)
+	}
 
 	// Create filter for results.
 	filterCh := make(chan metaCacheEntry, 100)
 	filteredResults := o.gatherResults(filterCh)
 	closeChannels := func() {
+		if !o.discardResult {
 			close(cacheCh)
+		}
 		close(filterCh)
 	}
 
@@ -657,19 +662,26 @@ func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions) (entr
 		}()
 
 		const retryDelay = 200 * time.Millisecond
-		const maxTries = 10
+		const maxTries = 5
 
-		// Write results to disk.
-		bw := newMetacacheBlockWriter(cacheCh, func(b *metacacheBlock) error {
-			if o.discardResult {
+		var bw *metacacheBlockWriter
 		// Don't save single object listings.
+		if !o.discardResult {
+			// Write results to disk.
+			bw = newMetacacheBlockWriter(cacheCh, func(b *metacacheBlock) error {
+				// if the block is 0 bytes and its a first block skip it.
+				// skip only this for Transient caches.
+				if len(b.data) == 0 && b.n == 0 && o.Transient {
 					return nil
 				}
 				o.debugln("listPath: saving block", b.n, "to", o.objectPath(b.n))
-			r, err := hash.NewReader(bytes.NewBuffer(b.data), int64(len(b.data)), "", "", int64(len(b.data)), false)
+				r, err := hash.NewReader(bytes.NewReader(b.data), int64(len(b.data)), "", "", int64(len(b.data)), false)
 				logger.LogIf(ctx, err)
 				custom := b.headerKV()
-			_, err = er.putObject(ctx, minioMetaBucket, o.objectPath(b.n), NewPutObjReader(r, nil, nil), ObjectOptions{UserDefined: custom})
+				_, err = er.putObject(ctx, minioMetaBucket, o.objectPath(b.n), NewPutObjReader(r, nil, nil), ObjectOptions{
+					UserDefined: custom,
+					NoLock:      true, // No need to hold namespace lock, each prefix caches uniquely.
+				})
 				if err != nil {
 					metaMu.Lock()
 					if meta.error != "" {
@@ -705,6 +717,7 @@ func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions) (entr
 				}
 				return nil
 			})
+		}
 
 		// How to resolve results.
 		resolver := metadataResolutionParams{
@@ -721,14 +734,18 @@ func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions) (entr
 			filterPrefix: o.FilterPrefix,
 			minDisks:     listingQuorum,
 			agreed: func(entry metaCacheEntry) {
+				if !o.discardResult {
 					cacheCh <- entry
+				}
 				filterCh <- entry
 			},
 			partial: func(entries metaCacheEntries, nAgreed int, errs []error) {
 				// Results Disagree :-(
 				entry, ok := entries.resolve(&resolver)
 				if ok {
+					if !o.discardResult {
 						cacheCh <- *entry
+					}
 					filterCh <- *entry
 				}
 			},
@@ -749,6 +766,7 @@ func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions) (entr
 		metaMu.Unlock()
 
 		closeChannels()
+		if !o.discardResult {
 			if err := bw.Close(); err != nil {
 				metaMu.Lock()
 				meta.error = err.Error()
@@ -756,6 +774,7 @@ func (er *erasureObjects) listPath(ctx context.Context, o listPathOptions) (entr
 				meta, err = o.updateMetacacheListing(meta, rpc)
 				metaMu.Unlock()
 			}
+		}
 	}()
 
 	return filteredResults()

cmd/object-api-interface.go

@@ -51,6 +51,7 @@ type ObjectOptions struct {
 	DeleteMarkerReplicationStatus string                 // Is only set in DELETE operations
 	VersionPurgeStatus            VersionPurgeStatusType // Is only set in DELETE operations for delete marker version to be permanently deleted.
 	TransitionStatus              string                 // status of the transition
+	NoLock                        bool                   // indicates to lower layers if the caller is expecting to hold locks.
 }
 
 // BucketOptions represents bucket options for ObjectLayer bucket operations