erasure: avoid io.Copy in hotpaths to reduce allocation (#11213)

cmd/erasure-decode.go

@@ -186,7 +186,6 @@ func (p *parallelReader) Read(dst [][]byte) ([][]byte, error) {
 		readerIndex++
 	}
 	wg.Wait()
-
 	if p.canDecode(newBuf) {
 		p.offset += p.shardSize
 		if healRequired != 0 {

cmd/erasure-utils.go

@@ -17,7 +17,6 @@
 package cmd
 
 import (
-	"bytes"
 	"context"
 	"io"
 
@@ -77,20 +76,25 @@ func writeDataBlocks(ctx context.Context, dst io.Writer, enBlocks [][]byte, data
 			// from subsequent blocks.
 			offset = 0
 		}
+
 		// We have written all the blocks, write the last remaining block.
 		if write < int64(len(block)) {
-			n, err := io.Copy(dst, bytes.NewReader(block[:write]))
+			n, err := dst.Write(block[:write])
 			if err != nil {
 				if err != io.ErrClosedPipe {
 					logger.LogIf(ctx, err)
 				}
 				return 0, err
 			}
-			totalWritten += n
+			if int64(n) != write {
+				return 0, io.ErrShortWrite
+			}
+			totalWritten += int64(n)
 			break
 		}
+
 		// Copy the block.
-		n, err := io.Copy(dst, bytes.NewReader(block))
+		n, err := dst.Write(block)
 		if err != nil {
 			// The writer will be closed incase of range queries, which will emit ErrClosedPipe.
 			if err != io.ErrClosedPipe {
@@ -99,11 +103,15 @@ func writeDataBlocks(ctx context.Context, dst io.Writer, enBlocks [][]byte, data
 			return 0, err
 		}
 
+		if n != len(block) {
+			return 0, io.ErrShortWrite
+		}
+
 		// Decrement output size.
-		write -= n
+		write -= int64(n)
 
 		// Increment written.
-		totalWritten += n
+		totalWritten += int64(n)
 	}
 
 	// Success.

cmd/object-handlers.go

@@ -432,6 +432,7 @@ func (api objectAPIHandlers) GetObjectHandler(w http.ResponseWriter, r *http.Req
 		return
 	}
 	defer gr.Close()
+
 	objInfo := gr.ObjInfo
 
 	// filter object lock metadata if permission does not permit

cmd/server-main.go

@@ -319,7 +319,13 @@ func initAllSubsystems(ctx context.Context, newObject ObjectLayer) (err error) {
 	}
 
 	if globalIsErasure {
-		logger.Info(fmt.Sprintf("Verifying %d buckets are consistent across drives...", len(buckets)))
+		if len(buckets) > 0 {
+			if len(buckets) == 1 {
+				logger.Info(fmt.Sprintf("Verifying if %d bucket is consistent across drives...", len(buckets)))
+			} else {
+				logger.Info(fmt.Sprintf("Verifying if %d buckets are consistent across drives...", len(buckets)))
+			}
+		}
 		for _, bucket := range buckets {
 			if _, err = newObject.HealBucket(ctx, bucket.Name, madmin.HealOpts{}); err != nil {
 				return fmt.Errorf("Unable to list buckets to heal: %w", err)

cmd/xl-storage.go

@@ -64,6 +64,9 @@ const (
 	// Size of each buffer.
 	readAheadBufSize = 1 << 20
 
+	// Small file threshold below which the metadata accompanies the data.
+	smallFileThreshold = 32 * humanize.KiByte
+
 	// XL metadata file carries per object metadata.
 	xlStorageFormatFile = "xl.meta"
 )
@@ -1174,7 +1177,7 @@ func (s *xlStorage) ReadVersion(ctx context.Context, volume, path, versionID str
 		// - object has not yet transitioned
 		// - object size lesser than 32KiB
 		// - object has maximum of 1 parts
-		if fi.TransitionStatus == "" && fi.DataDir != "" && fi.Size < 32*humanize.KiByte && len(fi.Parts) == 1 {
+		if fi.TransitionStatus == "" && fi.DataDir != "" && fi.Size < smallFileThreshold && len(fi.Parts) == 1 {
 			fi.Data, err = s.readAllData(volumeDir, pathJoin(volumeDir, path, fi.DataDir, fmt.Sprintf("part.%d", fi.Parts[0].Number)))
 			if err != nil {
 				return fi, err