iobufpool uses *[]byte instead of []byte to reduce allocations

internal/iobufpool/iobufpool.go

@@ -1,4 +1,7 @@
 // Package iobufpool implements a global segregated-fit pool of buffers for IO.
+//
+// It uses *[]byte instead of []byte to avoid the sync.Pool allocation with Put. Unfortunately, using a pointer to avoid
+// an allocation is purposely not documented. https://github.com/golang/go/issues/16323
 package iobufpool
 
 import "sync"
@@ -10,17 +13,27 @@ var pools [18]*sync.Pool
 func init() {
 	for i := range pools {
 		bufLen := 1 << (minPoolExpOf2 + i)
-		pools[i] = &sync.Pool{New: func() any { return make([]byte, bufLen) }}
+		pools[i] = &sync.Pool{
+			New: func() any {
+				buf := make([]byte, bufLen)
+				return &buf
+			},
+		}
 	}
 }
 
 // Get gets a []byte of len size with cap <= size*2.
-func Get(size int) []byte {
+func Get(size int) *[]byte {
 	i := getPoolIdx(size)
 	if i >= len(pools) {
-		return make([]byte, size)
+		buf := make([]byte, size)
+		return &buf
 	}
-	return pools[i].Get().([]byte)[:size]
+
+	ptrBuf := (pools[i].Get().(*[]byte))
+	*ptrBuf = (*ptrBuf)[:size]
+
+	return ptrBuf
 }
 
 func getPoolIdx(size int) int {
@@ -36,8 +49,8 @@ func getPoolIdx(size int) int {
 }
 
 // Put returns buf to the pool.
-func Put(buf []byte) {
-	i := putPoolIdx(cap(buf))
+func Put(buf *[]byte) {
+	i := putPoolIdx(cap(*buf))
 	if i < 0 {
 		return
 	}

internal/nbconn/bufferqueue.go

@@ -8,11 +8,11 @@ const minBufferQueueLen = 8
 
 type bufferQueue struct {
 	lock  sync.Mutex
-	queue [][]byte
+	queue []*[]byte
 	r, w  int
 }
 
-func (bq *bufferQueue) pushBack(buf []byte) {
+func (bq *bufferQueue) pushBack(buf *[]byte) {
 	bq.lock.Lock()
 	defer bq.lock.Unlock()
 
@@ -23,7 +23,7 @@ func (bq *bufferQueue) pushBack(buf []byte) {
 	bq.w++
 }
 
-func (bq *bufferQueue) pushFront(buf []byte) {
+func (bq *bufferQueue) pushFront(buf *[]byte) {
 	bq.lock.Lock()
 	defer bq.lock.Unlock()
 
@@ -35,7 +35,7 @@ func (bq *bufferQueue) pushFront(buf []byte) {
 	bq.w++
 }
 
-func (bq *bufferQueue) popFront() []byte {
+func (bq *bufferQueue) popFront() *[]byte {
 	bq.lock.Lock()
 	defer bq.lock.Unlock()
 
@@ -51,7 +51,7 @@ func (bq *bufferQueue) popFront() []byte {
 		bq.r = 0
 		bq.w = 0
 		if len(bq.queue) > minBufferQueueLen {
-			bq.queue = make([][]byte, minBufferQueueLen)
+			bq.queue = make([]*[]byte, minBufferQueueLen)
 		}
 	}
 
@@ -64,7 +64,7 @@ func (bq *bufferQueue) growQueue() {
 		desiredLen = minBufferQueueLen
 	}
 
-	newQueue := make([][]byte, desiredLen)
+	newQueue := make([]*[]byte, desiredLen)
 	copy(newQueue, bq.queue)
 	bq.queue = newQueue
 }

internal/nbconn/nbconn.go

@@ -129,9 +129,9 @@ func (c *NetConn) Read(b []byte) (n int, err error) {
 		if buf == nil {
 			break
 		}
-		copiedN := copy(b[n:], buf)
-		if copiedN < len(buf) {
-			buf = buf[copiedN:]
+		copiedN := copy(b[n:], *buf)
+		if copiedN < len(*buf) {
+			*buf = (*buf)[copiedN:]
 			c.readQueue.pushFront(buf)
 		} else {
 			iobufpool.Put(buf)
@@ -168,7 +168,7 @@ func (c *NetConn) Write(b []byte) (n int, err error) {
 	}
 
 	buf := iobufpool.Get(len(b))
-	copy(buf, b)
+	copy(*buf, b)
 	c.writeQueue.pushBack(buf)
 	return len(b), nil
 }
@@ -286,14 +286,14 @@ func (c *NetConn) flush() error {
 	}()
 
 	for buf := c.writeQueue.popFront(); buf != nil; buf = c.writeQueue.popFront() {
-		remainingBuf := buf
+		remainingBuf := *buf
 		for len(remainingBuf) > 0 {
 			n, err := c.nonblockingWrite(remainingBuf)
 			remainingBuf = remainingBuf[n:]
 			if err != nil {
 				if !errors.Is(err, ErrWouldBlock) {
-					buf = buf[:len(remainingBuf)]
-					copy(buf, remainingBuf)
+					*buf = (*buf)[:len(remainingBuf)]
+					copy(*buf, remainingBuf)
 					c.writeQueue.pushFront(buf)
 					return err
 				}
@@ -321,9 +321,9 @@ func (c *NetConn) flush() error {
 func (c *NetConn) BufferReadUntilBlock() error {
 	for {
 		buf := iobufpool.Get(8 * 1024)
-		n, err := c.nonblockingRead(buf)
+		n, err := c.nonblockingRead(*buf)
 		if n > 0 {
-			buf = buf[:n]
+			*buf = (*buf)[:n]
 			c.readQueue.pushBack(buf)
 		} else if n == 0 {
 			iobufpool.Put(buf)

pgconn/pgconn.go

@@ -1175,20 +1175,20 @@ func (pgConn *PgConn) CopyFrom(ctx context.Context, r io.Reader, sql string) (Co
 
 	buf := iobufpool.Get(65536)
 	defer iobufpool.Put(buf)
-	buf[0] = 'd'
+	(*buf)[0] = 'd'
 
 	var readErr, pgErr error
 	for pgErr == nil {
 		// Read chunk from r.
 		var n int
-		n, readErr = r.Read(buf[5:cap(buf)])
+		n, readErr = r.Read((*buf)[5:cap(*buf)])
 
 		// Send chunk to PostgreSQL.
 		if n > 0 {
-			buf = buf[0 : n+5]
-			pgio.SetInt32(buf[1:], int32(n+4))
+			*buf = (*buf)[0 : n+5]
+			pgio.SetInt32((*buf)[1:], int32(n+4))
 
-			writeErr := pgConn.frontend.SendUnbufferedEncodedCopyData(buf)
+			writeErr := pgConn.frontend.SendUnbufferedEncodedCopyData(*buf)
 			if writeErr != nil {
 				pgConn.asyncClose()
 				return CommandTag{}, err

pgproto3/chunkreader.go

@@ -14,7 +14,7 @@ import (
 type chunkReader struct {
 	r io.Reader
 
-	buf    []byte
+	buf    *[]byte
 	rp, wp int // buf read position and write position
 
 	minBufSize int
@@ -45,7 +45,7 @@ func newChunkReader(r io.Reader, minBufSize int) *chunkReader {
 func (r *chunkReader) Next(n int) (buf []byte, err error) {
 	// Reset the buffer if it is empty
 	if r.rp == r.wp {
-		if len(r.buf) != r.minBufSize {
+		if len(*r.buf) != r.minBufSize {
 			iobufpool.Put(r.buf)
 			r.buf = iobufpool.Get(r.minBufSize)
 		}
@@ -55,36 +55,36 @@ func (r *chunkReader) Next(n int) (buf []byte, err error) {
 
 	// n bytes already in buf
 	if (r.wp - r.rp) >= n {
-		buf = r.buf[r.rp : r.rp+n : r.rp+n]
+		buf = (*r.buf)[r.rp : r.rp+n : r.rp+n]
 		r.rp += n
 		return buf, err
 	}
 
 	// buf is smaller than requested number of bytes
-	if len(r.buf) < n {
+	if len(*r.buf) < n {
 		bigBuf := iobufpool.Get(n)
-		r.wp = copy(bigBuf, r.buf[r.rp:r.wp])
+		r.wp = copy((*bigBuf), (*r.buf)[r.rp:r.wp])
 		r.rp = 0
 		iobufpool.Put(r.buf)
 		r.buf = bigBuf
 	}
 
 	// buf is large enough, but need to shift filled area to start to make enough contiguous space
 	minReadCount := n - (r.wp - r.rp)
-	if (len(r.buf) - r.wp) < minReadCount {
-		r.wp = copy(r.buf, r.buf[r.rp:r.wp])
+	if (len(*r.buf) - r.wp) < minReadCount {
+		r.wp = copy((*r.buf), (*r.buf)[r.rp:r.wp])
 		r.rp = 0
 	}
 
 	// Read at least the required number of bytes from the underlying io.Reader
-	readBytesCount, err := io.ReadAtLeast(r.r, r.buf[r.wp:], minReadCount)
+	readBytesCount, err := io.ReadAtLeast(r.r, (*r.buf)[r.wp:], minReadCount)
 	r.wp += readBytesCount
 	// fmt.Println("read", n)
 	if err != nil {
 		return nil, err
 	}
 
-	buf = r.buf[r.rp : r.rp+n : r.rp+n]
+	buf = (*r.buf)[r.rp : r.rp+n : r.rp+n]
 	r.rp += n
 	return buf, nil
 }