fix: issues with handling delete markers in metacache (#11150)

Additional cases handled

- fix address situations where healing is not
  triggered on failed writes and deletes.

- consider object exists during listing when
  metadata can be successfully decoded.

cmd/erasure-multipart.go

@@ -846,22 +846,22 @@ func (er erasureObjects) CompleteMultipartUpload(ctx context.Context, bucket str
 	}
 
 	// Check if there is any offline disk and add it to the MRF list
-	for i, disk := range onlineDisks {
-		if disk == nil || storageDisks[i] == nil {
+	for _, disk := range onlineDisks {
+		if disk != nil && disk.IsOnline() {
+			continue
+		}
 		er.addPartial(bucket, object, fi.VersionID)
 		break
 	}
-	}
 
 	for i := 0; i < len(onlineDisks); i++ {
-		if onlineDisks[i] == nil {
-			continue
-		}
+		if onlineDisks[i] != nil && onlineDisks[i].IsOnline() {
 			// Object info is the same in all disks, so we can pick
 			// the first meta from online disk
 			fi = partsMetadata[i]
 			break
 		}
+	}
 
 	// Success, return object info.
 	return fi.ToObjectInfo(bucket, object), nil

cmd/erasure-object.go

@@ -727,21 +727,21 @@ func (er erasureObjects) putObject(ctx context.Context, bucket string, object st
 	// Whether a disk was initially or becomes offline
 	// during this upload, send it to the MRF list.
 	for i := 0; i < len(onlineDisks); i++ {
-		if onlineDisks[i] == nil || storageDisks[i] == nil {
+		if onlineDisks[i] != nil && onlineDisks[i].IsOnline() {
+			continue
+		}
 		er.addPartial(bucket, object, fi.VersionID)
 		break
 	}
-	}
 
 	for i := 0; i < len(onlineDisks); i++ {
-		if onlineDisks[i] == nil {
-			continue
-		}
+		if onlineDisks[i] != nil && onlineDisks[i].IsOnline() {
 			// Object info is the same in all disks, so we can pick
 			// the first meta from online disk
 			fi = partsMetadata[i]
 			break
 		}
+	}
 
 	return fi.ToObjectInfo(bucket, object), nil
 }
@@ -922,18 +922,17 @@ func (er erasureObjects) DeleteObjects(ctx context.Context, bucket string, objec
 	for _, version := range versions {
 		// Check if there is any offline disk and add it to the MRF list
 		for _, disk := range storageDisks {
-			if disk == nil {
-				// ignore delete markers for quorum
-				if version.Deleted {
+			if disk != nil && disk.IsOnline() {
+				// Skip attempted heal on online disks.
 				continue
 			}
+
 			// all other direct versionId references we should
 			// ensure no dangling file is left over.
 			er.addPartial(bucket, version.Name, version.VersionID)
 			break
 		}
 	}
-	}
 
 	return dobjects, errs
 }
@@ -1042,11 +1041,12 @@ func (er erasureObjects) DeleteObject(ctx context.Context, bucket, object string
 	}
 
 	for _, disk := range storageDisks {
-		if disk == nil {
+		if disk != nil && disk.IsOnline() {
+			continue
+		}
 		er.addPartial(bucket, object, opts.VersionID)
 		break
 	}
-	}
 
 	return ObjectInfo{
 		Bucket:             bucket,

cmd/metacache-entries.go

@@ -73,11 +73,13 @@ func (e *metaCacheEntry) matches(other *metaCacheEntry, bucket string) bool {
 	if len(e.metadata) != len(other.metadata) || e.name != other.name {
 		return false
 	}
+
 	eFi, eErr := e.fileInfo(bucket)
-	oFi, oErr := e.fileInfo(bucket)
+	oFi, oErr := other.fileInfo(bucket)
 	if eErr != nil || oErr != nil {
 		return eErr == oErr
 	}
+
 	return eFi.ModTime.Equal(oFi.ModTime) && eFi.Size == oFi.Size && eFi.VersionID == oFi.VersionID
 }
 
@@ -213,11 +215,12 @@ func (m metaCacheEntries) resolve(r *metadataResolutionParams) (selected *metaCa
 		}
 
 		// Get new entry metadata
-		objExists++
 		fiv, err := entry.fileInfo(r.bucket)
 		if err != nil {
 			continue
 		}
+
+		objExists++
 		if selFIV == nil {
 			selected = entry
 			selFIV = fiv
@@ -227,14 +230,8 @@ func (m metaCacheEntries) resolve(r *metadataResolutionParams) (selected *metaCa
 		if selected.matches(entry, r.bucket) {
 			continue
 		}
-
-		// Select latest modtime.
-		if fiv.ModTime.After(selFIV.ModTime) {
-			selected = entry
-			selFIV = fiv
-			continue
-		}
 	}
+
 	// If directory, we need quorum.
 	if dirExists > 0 && dirExists < r.dirQuorum {
 		return nil, false

cmd/metacache-server-pool.go

@@ -147,13 +147,11 @@ func (z *erasureServerPools) listPath(ctx context.Context, o listPathOptions) (e
 	var wg sync.WaitGroup
 	var errs []error
 	allAtEOF := true
-	asked := 0
 	mu.Lock()
 	// Ask all sets and merge entries.
 	for _, zone := range z.serverPools {
 		for _, set := range zone.sets {
 			wg.Add(1)
-			asked++
 			go func(i int, set *erasureObjects) {
 				defer wg.Done()
 				e, err := set.listPath(ctx, o)

cmd/metacache-set.go

@@ -168,12 +168,10 @@ func (o *listPathOptions) gatherResults(in <-chan metaCacheEntry) func() (metaCa
 				o.debugln("not in dir", o.Prefix, o.Separator)
 				continue
 			}
-			if !o.InclDeleted && entry.isObject() {
-				if entry.isLatestDeletemarker() {
-					o.debugln("latest delete")
+			if !o.InclDeleted && entry.isObject() && entry.isLatestDeletemarker() {
+				o.debugln("latest is delete marker")
 				continue
 			}
-			}
 			if o.Limit > 0 && results.len() >= o.Limit {
 				// We have enough and we have more.
 				// Do not return io.EOF

cmd/metacache-stream.go

@@ -519,7 +519,7 @@ func (r *metacacheReader) readN(n int, inclDeleted, inclDirs bool, prefix string
 		if !inclDirs && meta.isDir() {
 			continue
 		}
-		if meta.isDir() && !inclDeleted && meta.isLatestDeletemarker() {
+		if !inclDeleted && meta.isLatestDeletemarker() {
 			continue
 		}
 		res = append(res, meta)