reduce allocations on multi-disk clusters (#12311)

multi-disk clusters initialize buffer pools
per disk, this is perhaps expensive and perhaps
not useful, for a running server instance. As this
may disallow re-use of buffers across sets,

this change ensures that buffers across sets
can be re-used at drive level, this can reduce
quite a lot of memory on large drive setups.

cmd/bitrot.go

@@ -164,7 +164,6 @@ func bitrotVerify(r io.Reader, wantSize, partSize int64, algo BitrotAlgorithm, w
 
 	h := algo.New()
 	hashBuf := make([]byte, h.Size())
-	buf := make([]byte, shardSize)
 	left := wantSize
 
 	// Calculate the size of the bitrot file and compare
@@ -173,6 +172,9 @@ func bitrotVerify(r io.Reader, wantSize, partSize int64, algo BitrotAlgorithm, w
 		return errFileCorrupt
 	}
 
+	bufp := xlPoolSmall.Get().(*[]byte)
+	defer xlPoolSmall.Put(bufp)
+
 	for left > 0 {
 		// Read expected hash...
 		h.Reset()
@@ -186,13 +188,15 @@ func bitrotVerify(r io.Reader, wantSize, partSize int64, algo BitrotAlgorithm, w
 		if left < shardSize {
 			shardSize = left
 		}
-		read, err := io.CopyBuffer(h, io.LimitReader(r, shardSize), buf)
+
+		read, err := io.CopyBuffer(h, io.LimitReader(r, shardSize), *bufp)
 		if err != nil {
 			// Read's failed for object with right size, at different offsets.
-			return err
+			return errFileCorrupt
 		}
+
 		left -= read
-		if !bytes.Equal(h.Sum(nil), hashBuf) {
+		if !bytes.Equal(h.Sum(nil), hashBuf[:n]) {
 			return errFileCorrupt
 		}
 	}

cmd/erasure-healing-common.go

@@ -261,7 +261,7 @@ func disksWithAllParts(ctx context.Context, onlineDisks []StorageAPI, partsMetad
 		// Always check data, if we got it.
 		if (len(meta.Data) > 0 || meta.Size == 0) && len(meta.Parts) > 0 {
 			checksumInfo := meta.Erasure.GetChecksumInfo(meta.Parts[0].Number)
-			dataErrs[i] = bitrotVerify(bytes.NewBuffer(meta.Data),
+			dataErrs[i] = bitrotVerify(bytes.NewReader(meta.Data),
 				int64(len(meta.Data)),
 				meta.Erasure.ShardFileSize(meta.Size),
 				checksumInfo.Algorithm,

cmd/xl-storage.go

@@ -98,6 +98,27 @@ func isValidVolname(volname string) bool {
 	return true
 }
 
+var (
+	xlPoolReallyLarge = sync.Pool{
+		New: func() interface{} {
+			b := disk.AlignedBlock(blockSizeReallyLarge)
+			return &b
+		},
+	}
+	xlPoolLarge = sync.Pool{
+		New: func() interface{} {
+			b := disk.AlignedBlock(blockSizeLarge)
+			return &b
+		},
+	}
+	xlPoolSmall = sync.Pool{
+		New: func() interface{} {
+			b := disk.AlignedBlock(blockSizeSmall)
+			return &b
+		},
+	}
+)
+
 // xlStorage - implements StorageAPI interface.
 type xlStorage struct {
 	diskPath string
@@ -105,10 +126,6 @@ type xlStorage struct {
 
 	globalSync bool
 
-	poolReallyLarge sync.Pool
-	poolLarge       sync.Pool
-	poolSmall       sync.Pool
-
 	rootDisk bool
 
 	diskID string
@@ -256,24 +273,6 @@ func newXLStorage(ep Endpoint) (*xlStorage, error) {
 	p := &xlStorage{
 		diskPath:   path,
 		endpoint:   ep,
-		poolReallyLarge: sync.Pool{
-			New: func() interface{} {
-				b := disk.AlignedBlock(blockSizeReallyLarge)
-				return &b
-			},
-		},
-		poolLarge: sync.Pool{
-			New: func() interface{} {
-				b := disk.AlignedBlock(blockSizeLarge)
-				return &b
-			},
-		},
-		poolSmall: sync.Pool{
-			New: func() interface{} {
-				b := disk.AlignedBlock(blockSizeSmall)
-				return &b
-			},
-		},
 		globalSync: env.Get(config.EnvFSOSync, config.EnableOff) == config.EnableOn,
 		ctx:        GlobalContext,
 		rootDisk:   rootDisk,
@@ -1323,9 +1322,9 @@ func (o *odirectReader) Read(buf []byte) (n int, err error) {
 	}
 	if o.buf == nil {
 		if o.smallFile {
-			o.bufp = o.s.poolSmall.Get().(*[]byte)
+			o.bufp = xlPoolSmall.Get().(*[]byte)
 		} else {
-			o.bufp = o.s.poolLarge.Get().(*[]byte)
+			o.bufp = xlPoolLarge.Get().(*[]byte)
 		}
 	}
 	if o.freshRead {
@@ -1367,9 +1366,9 @@ func (o *odirectReader) Read(buf []byte) (n int, err error) {
 // Close - Release the buffer and close the file.
 func (o *odirectReader) Close() error {
 	if o.smallFile {
-		o.s.poolSmall.Put(o.bufp)
+		xlPoolSmall.Put(o.bufp)
 	} else {
-		o.s.poolLarge.Put(o.bufp)
+		xlPoolLarge.Put(o.bufp)
 	}
 	return o.f.Close()
 }
@@ -1554,11 +1553,11 @@ func (s *xlStorage) CreateFile(ctx context.Context, volume, path string, fileSiz
 	var bufp *[]byte
 	if fileSize > 0 && fileSize >= reallyLargeFileThreshold {
 		// use a larger 4MiB buffer for really large streams.
-		bufp = s.poolReallyLarge.Get().(*[]byte)
-		defer s.poolReallyLarge.Put(bufp)
+		bufp = xlPoolReallyLarge.Get().(*[]byte)
+		defer xlPoolReallyLarge.Put(bufp)
 	} else {
-		bufp = s.poolLarge.Get().(*[]byte)
-		defer s.poolLarge.Put(bufp)
+		bufp = xlPoolLarge.Get().(*[]byte)
+		defer xlPoolLarge.Put(bufp)
 	}
 
 	written, err := xioutil.CopyAligned(w, r, *bufp, fileSize)