Improve delete performance by reducing the number of calls (#9092)

- Remove the requirement to honor storage class for deletes
- Improve `posix.DeleteFileBulk` code to Stat the volumeDir
  only once per call, rather than for all object paths.

cmd/posix.go

@@ -1340,9 +1340,40 @@ func (s *posix) DeleteFile(volume, path string) (err error) {
 }
 
 func (s *posix) DeleteFileBulk(volume string, paths []string) (errs []error, err error) {
+	atomic.AddInt32(&s.activeIOCount, 1)
+	defer func() {
+		atomic.AddInt32(&s.activeIOCount, -1)
+	}()
+
+	volumeDir, err := s.getVolDir(volume)
+	if err != nil {
+		return nil, err
+	}
+
+	// Stat a volume entry.
+	_, err = os.Stat(volumeDir)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return nil, errVolumeNotFound
+		} else if os.IsPermission(err) {
+			return nil, errVolumeAccessDenied
+		} else if isSysErrIO(err) {
+			return nil, errFaultyDisk
+		}
+		return nil, err
+	}
+
 	errs = make([]error, len(paths))
+	// Following code is needed so that we retain SlashSeparator
+	// suffix if any in path argument.
 	for idx, path := range paths {
-		errs[idx] = s.DeleteFile(volume, path)
+		filePath := pathJoin(volumeDir, path)
+		errs[idx] = checkPathLength(filePath)
+		if errs[idx] != nil {
+			continue
+		}
+		// Delete file and delete parent directory as well if its empty.
+		errs[idx] = deleteFile(volumeDir, filePath)
 	}
 	return
 }

cmd/xl-v1-object.go

@@ -747,32 +747,37 @@ func (xl xlObjects) deleteObject(ctx context.Context, bucket, object string, wri
 // object.
 func (xl xlObjects) doDeleteObjects(ctx context.Context, bucket string, objects []string, errs []error, writeQuorums []int, isDirs []bool) ([]error, error) {
 	var tmpObjs = make([]string, len(objects))
-	disks := xl.getDisks()
 	if bucket == minioMetaTmpBucket {
 		copy(tmpObjs, objects)
 	} else {
-		for i, object := range objects {
-			if errs[i] != nil {
+		for idx := range objects {
+			if errs[idx] != nil {
 				continue
 			}
+			tmpObjs[idx] = mustGetUUID()
 			var err error
-			tmpObjs[i] = mustGetUUID()
-			// Rename the current object while requiring write quorum, but also consider
-			// that a non found object in a given disk as a success since it already
-			// confirms that the object doesn't have a part in that disk (already removed)
-			if isDirs[i] {
-				disks, err = rename(ctx, disks, bucket, object, minioMetaTmpBucket, tmpObjs[i], true, writeQuorums[i],
+			// Rename the current object while requiring
+			// write quorum, but also consider that a non
+			// found object in a given disk as a success
+			// since it already confirms that the object
+			// doesn't have a part in that disk (already removed)
+			if isDirs[idx] {
+				_, err = rename(ctx, xl.getDisks(), bucket, objects[idx],
+					minioMetaTmpBucket, tmpObjs[idx], true, writeQuorums[idx],
 					[]error{errFileNotFound, errFileAccessDenied})
 			} else {
-				disks, err = rename(ctx, disks, bucket, object, minioMetaTmpBucket, tmpObjs[i], true, writeQuorums[i],
+				_, err = rename(ctx, xl.getDisks(), bucket, objects[idx],
+					minioMetaTmpBucket, tmpObjs[idx], true, writeQuorums[idx],
 					[]error{errFileNotFound})
 			}
 			if err != nil {
-				errs[i] = err
+				errs[idx] = err
 			}
 		}
 	}
 
+	disks := xl.getDisks()
+
 	// Initialize list of errors.
 	var opErrs = make([]error, len(disks))
 	var delObjErrs = make([][]error, len(disks))
@@ -840,23 +845,17 @@ func (xl xlObjects) deleteObjects(ctx context.Context, bucket string, objects []
 		}
 	}
 
-	for i, object := range objects {
+	for i := range objects {
 		if errs[i] != nil {
 			continue
 		}
-		if isObjectDirs[i] {
-			writeQuorums[i] = len(xl.getDisks())/2 + 1
-		} else {
-			var err error
-			// Read metadata associated with the object from all disks.
-			partsMetadata, readXLErrs := readAllXLMetadata(ctx, xl.getDisks(), bucket, object)
-			// get Quorum for this object
-			_, writeQuorums[i], err = objectQuorumFromMeta(ctx, xl, partsMetadata, readXLErrs)
-			if err != nil {
-				errs[i] = toObjectErr(err, bucket, object)
-				continue
-			}
-		}
+		// Assume (N/2 + 1) quorums for all objects
+		// this is a theoretical assumption such that
+		// for delete's we do not need to honor storage
+		// class for objects which have reduced quorum
+		// storage class only needs to be honored for
+		// Read() requests alone which we already do.
+		writeQuorums[i] = len(xl.getDisks())/2 + 1
 	}
 
 	return xl.doDeleteObjects(ctx, bucket, objects, errs, writeQuorums, isObjectDirs)