drpcstream: introduce shared BufferPool for ring buffer

Add a BufferPool backed by sync.Pool that is shared across all streams
within a Manager. The ring buffer now obtains a pooled buffer on Enqueue
and copies the message into it, instead of growing a fixed slice per
slot. Dequeue hands the buffer's data to the consumer and advances the
tail immediately, and Done releases the buffer back to the pool once the
consumer is finished with it.

Keeping the Dequeue/Done contract means the consumer works with a plain
[]byte and never has to know whether the queue is backed by a pool or by
fixed buffers. Because Dequeue advances the tail right away rather than
waiting for Done, Close no longer has to block until in-flight buffers
are released.

The pool is a required parameter in the Stream constructor, created once
per Manager and passed to all streams it creates.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: shubhamdhama

drpcmanager/active_streams_test.go

@@ -22,7 +22,7 @@ func testMuxWriter(t *testing.T) *drpcwire.MuxWriter {
 }
 
 func testStream(t *testing.T, id uint64) *drpcstream.Stream {
-	return drpcstream.New(context.Background(), id, testMuxWriter(t))
+	return drpcstream.New(context.Background(), id, testMuxWriter(t), drpcstream.NewBufferPool())
 }
 
 func TestActiveStreams_AddAndGet(t *testing.T) {

drpcmanager/manager.go

@@ -61,7 +61,8 @@ type Manager struct {
 	wg sync.WaitGroup // tracks active manageStream goroutines
 
 	// streams tracks active streams.
-	streams *activeStreams
+	streams  *activeStreams
+	recvPool *drpcstream.BufferPool
 
 	pdone   drpcsignal.Chan // signals when NewServerStream has registered the new stream
 	invokes chan invokeInfo // completed invoke info from manageReader to NewServerStream
@@ -130,6 +131,7 @@ func NewWithOptions(tr drpc.Transport, kind ManagerKind, opts Options) *Manager
 	m.pendingStreams = make(map[uint64]*pendingStream)
 
 	m.streams = newActiveStreams()
+	m.recvPool = drpcstream.NewBufferPool()
 
 	// set the internal stream options
 	drpcopts.SetStreamTransport(&m.opts.Stream.Internal, m.tr)
@@ -268,7 +270,7 @@ func (m *Manager) newStream(ctx context.Context, sid uint64, kind drpc.StreamKin
 		drpcopts.SetStreamStats(&opts.Internal, cb(rpc))
 	}
 
-	stream := drpcstream.NewWithOptions(ctx, sid, m.wr, opts)
+	stream := drpcstream.NewWithOptions(ctx, sid, m.wr, m.recvPool, opts)
 
 	if err := m.streams.Add(sid, stream); err != nil {
 		return nil, err

drpcstream/buffer_pool.go

@@ -0,0 +1,42 @@
+// Copyright (C) 2026 Cockroach Labs.
+// See LICENSE for copying information.
+
+package drpcstream
+
+import "sync"
+
+// BufferPool wraps sync.Pool to provide reusable byte slices for the
+// stream receive path. Buffers obtained via Get should be returned via
+// Put when no longer needed. Forgetting to Put is safe (GC reclaims)
+// but reduces reuse.
+type BufferPool struct {
+	pool sync.Pool
+}
+
+// NewBufferPool returns a new buffer pool.
+func NewBufferPool() *BufferPool {
+	return &BufferPool{
+		pool: sync.Pool{
+			New: func() interface{} {
+				b := make([]byte, 0, 4096)
+				return &b
+			},
+		},
+	}
+}
+
+// Get returns a zero-length byte slice from the pool, retaining its
+// backing array for reuse.
+func (bp *BufferPool) Get() *[]byte {
+	p := bp.pool.Get().(*[]byte)
+	*p = (*p)[:0]
+	return p
+}
+
+// Put returns a buffer to the pool. Nil is safe to pass.
+func (bp *BufferPool) Put(b *[]byte) {
+	if b == nil {
+		return
+	}
+	bp.pool.Put(b)
+}

drpcstream/ring_buffer.go

@@ -17,9 +17,11 @@ const defaultRingBufferCapacity = 256
 // assembled packet data. It sits between manageReader (producer, calls
 // Enqueue) and the application goroutine (consumer, calls Dequeue/Done).
 //
-// Slots are pre-allocated and reused: each slot's backing array grows via
-// append to fit incoming data, then stays at its high-water mark, avoiding
-// per-message allocation in steady state.
+// Buffers are obtained from a shared BufferPool. Enqueue copies data into a
+// pooled buffer; Dequeue returns that buffer's data and advances the tail
+// immediately, and Done releases the buffer back to the pool. Keeping the
+// pool behind Dequeue/Done means the consumer does not need to know whether
+// the queue is backed by a pool or by fixed buffers.
 //
 // After Close, Dequeue drains any queued messages before returning the close
 // error. This ensures graceful shutdown (KindClose/KindCloseSend) delivers
@@ -28,43 +30,50 @@ type ringBuffer struct {
 	mu   sync.Mutex
 	cond sync.Cond
 
-	buf   [][]byte // ring of byte slices
-	head  int      // next write position (producer)
-	tail  int      // next read position (consumer)
-	count int      // number of occupied slots
+	pool  *BufferPool // shared pool buffers are obtained from
+	buf   []*[]byte   // ring of pooled buffer pointers
+	head  int         // next write position (producer)
+	tail  int         // next read position (consumer)
+	count int         // number of occupied slots
 
-	held bool  // true between Dequeue and Done
-	err  error // terminal error, set by Close
+	held *[]byte // buffer from the last Dequeue, released by Done
+	err  error   // terminal error, set by Close
 }
 
-func (rb *ringBuffer) init() {
+func (rb *ringBuffer) init(pool *BufferPool) {
 	rb.cond.L = &rb.mu
-	rb.buf = make([][]byte, defaultRingBufferCapacity)
+	rb.pool = pool
+	rb.buf = make([]*[]byte, defaultRingBufferCapacity)
 }
 
-// Enqueue copies data into the next write slot. If the buffer is full, it
-// blocks until a slot is freed or the buffer is closed. If the buffer is
-// closed, Enqueue returns silently without enqueuing.
+// Enqueue copies data into a pooled buffer and places it in the next write
+// slot. If the buffer is full, it blocks until a slot is freed or the buffer
+// is closed. If the buffer is closed, Enqueue returns silently.
 func (rb *ringBuffer) Enqueue(data []byte) {
+	b := rb.pool.Get()
+	*b = append(*b, data...)
+
 	rb.mu.Lock()
 	defer rb.mu.Unlock()
 
 	for rb.count == len(rb.buf) && rb.err == nil {
 		rb.cond.Wait()
 	}
 	if rb.err != nil {
+		rb.pool.Put(b)
 		return
 	}
 
-	rb.buf[rb.head] = append(rb.buf[rb.head][:0], data...)
+	rb.buf[rb.head] = b
 	rb.head = (rb.head + 1) % len(rb.buf)
 	rb.count++
 	rb.cond.Broadcast()
 }
 
-// Dequeue returns the data from the next read slot. If the buffer is empty,
-// it blocks until data is available or the buffer is closed. The returned
-// slice is valid until Done is called.
+// Dequeue returns the data from the next buffered message and advances the
+// tail. The returned slice is valid until Done is called, which releases the
+// underlying buffer back to the pool. Done must be called exactly once after
+// each successful Dequeue.
 func (rb *ringBuffer) Dequeue() ([]byte, error) {
 	rb.mu.Lock()
 	defer rb.mu.Unlock()
@@ -76,37 +85,31 @@ func (rb *ringBuffer) Dequeue() ([]byte, error) {
 		return nil, rb.err
 	}
 
-	rb.held = true
-	return rb.buf[rb.tail], nil
-}
-
-// Done advances the read pointer, making the slot available for reuse.
-// It must be called exactly once after each successful Dequeue.
-//
-// TODO(shubham): remove this method once a shared buffer pool is introduced.
-// With a pool, Dequeue will advance the tail immediately and the caller will
-// return the buffer to the pool directly.
-func (rb *ringBuffer) Done() {
-	rb.mu.Lock()
-	defer rb.mu.Unlock()
-
+	b := rb.buf[rb.tail]
+	rb.buf[rb.tail] = nil
 	rb.tail = (rb.tail + 1) % len(rb.buf)
 	rb.count--
-	rb.held = false
+	rb.held = b
 	rb.cond.Broadcast()
+
+	return *b, nil
+}
+
+// Done releases the buffer from the most recent Dequeue back to the pool,
+// invalidating the slice that Dequeue returned. It must be called exactly
+// once after each successful Dequeue. Because the queue is single-consumer,
+// Done is only ever called from the same goroutine as Dequeue.
+func (rb *ringBuffer) Done() {
+	rb.pool.Put(rb.held)
+	rb.held = nil
 }
 
 // Close marks the buffer as closed with the given error. All blocked Enqueue
-// and Dequeue calls are woken and will return. Close waits for any in-progress
-// Dequeue/Done pair to complete before setting the error. Subsequent calls are
-// no-ops.
+// and Dequeue calls are woken and will return. Subsequent calls are no-ops.
 func (rb *ringBuffer) Close(err error) {
 	rb.mu.Lock()
 	defer rb.mu.Unlock()
 
-	for rb.held {
-		rb.cond.Wait()
-	}
 	if rb.err != nil {
 		return
 	}