fix: close stream adaptive gaps and switch notify to stario v0.1.1

- make stream fast path honor adaptive soft payload limits end-to-end
  - split oversized fast-stream payloads into sequential frames before batching
  - use adaptive soft cap when encoding stream batch payloads
  - move timeout-like error detection into production code for adaptive tx
  - tune notify FrameReader read size explicitly to avoid throughput regression
  - drop local stario replace and depend on released b612.me/stario v0.1.1

benchmark_listen_test.go

@@ -0,0 +1,16 @@
+package notify
+
+import (
+	"os"
+	"testing"
+)
+
+const benchmarkTCPListenAddrEnv = "NOTIFY_BENCH_TCP_LISTEN_ADDR"
+
+func benchmarkTCPListenAddr(tb testing.TB) string {
+	tb.Helper()
+	if addr := os.Getenv(benchmarkTCPListenAddrEnv); addr != "" {
+		return addr
+	}
+	return "127.0.0.1:0"
+}

benchmark_tcp_proxy_test.go

@@ -0,0 +1,378 @@
+package notify
+
+import (
+	"context"
+	"errors"
+	"io"
+	"math/rand"
+	"net"
+	"os"
+	"strconv"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+const (
+	benchmarkTCPProxyEnableEnv         = "NOTIFY_BENCH_TCP_PROXY_ENABLE"
+	benchmarkTCPProxyListenAddrEnv     = "NOTIFY_BENCH_TCP_PROXY_LISTEN_ADDR"
+	benchmarkTCPProxyRateMbitEnv       = "NOTIFY_BENCH_TCP_PROXY_RATE_MBIT"
+	benchmarkTCPProxyDelayMsEnv        = "NOTIFY_BENCH_TCP_PROXY_DELAY_MS"
+	benchmarkTCPProxyLossPctEnv        = "NOTIFY_BENCH_TCP_PROXY_LOSS_PCT"
+	benchmarkTCPProxyLossPenaltyMsEnv  = "NOTIFY_BENCH_TCP_PROXY_LOSS_PENALTY_MS"
+	benchmarkTCPProxySeedEnv           = "NOTIFY_BENCH_TCP_PROXY_SEED"
+	benchmarkTCPProxyDefaultListenAddr = "127.0.0.1:0"
+)
+
+const benchmarkTCPProxyDefaultLossPenalty = 120 * time.Millisecond
+
+type benchmarkTCPProxyConfig struct {
+	Enabled     bool
+	ListenAddr  string
+	RateBytesPS float64
+	Delay       time.Duration
+	LossPct     float64
+	LossPenalty time.Duration
+	Seed        int64
+}
+
+type benchmarkTCPProxy struct {
+	listener net.Listener
+	target   string
+	cfg      benchmarkTCPProxyConfig
+
+	closed atomic.Bool
+	wg     sync.WaitGroup
+}
+
+func benchmarkTCPDialAddr(tb testing.TB, targetAddr string) string {
+	tb.Helper()
+	if targetAddr == "" {
+		return targetAddr
+	}
+	cfg := benchmarkTCPProxyConfigFromEnv(tb)
+	if !cfg.Enabled {
+		return targetAddr
+	}
+	proxy, err := startBenchmarkTCPProxy(targetAddr, cfg)
+	if err != nil {
+		tb.Fatalf("start benchmark tcp proxy failed: %v", err)
+	}
+	tb.Cleanup(proxy.stop)
+	if proxy.listener == nil || proxy.listener.Addr() == nil {
+		tb.Fatalf("benchmark tcp proxy listener is nil")
+	}
+	return proxy.listener.Addr().String()
+}
+
+func benchmarkTCPProxyConfigFromEnv(tb testing.TB) benchmarkTCPProxyConfig {
+	tb.Helper()
+	enabled, ok, err := parseBoolEnv(benchmarkTCPProxyEnableEnv)
+	if err != nil {
+		tb.Fatalf("invalid %s: %v", benchmarkTCPProxyEnableEnv, err)
+	}
+	if !ok || !enabled {
+		return benchmarkTCPProxyConfig{}
+	}
+	cfg := benchmarkTCPProxyConfig{
+		Enabled:    true,
+		ListenAddr: benchmarkTCPProxyDefaultListenAddr,
+		Seed:       1,
+	}
+	if listen := os.Getenv(benchmarkTCPProxyListenAddrEnv); listen != "" {
+		cfg.ListenAddr = listen
+	}
+	rateMbit, ok, err := parseFloatEnv(benchmarkTCPProxyRateMbitEnv)
+	if err != nil {
+		tb.Fatalf("invalid %s: %v", benchmarkTCPProxyRateMbitEnv, err)
+	}
+	if ok && rateMbit > 0 {
+		cfg.RateBytesPS = rateMbit * 1000 * 1000 / 8
+	}
+	delayMs, ok, err := parseIntEnv(benchmarkTCPProxyDelayMsEnv)
+	if err != nil {
+		tb.Fatalf("invalid %s: %v", benchmarkTCPProxyDelayMsEnv, err)
+	}
+	if ok && delayMs > 0 {
+		cfg.Delay = time.Duration(delayMs) * time.Millisecond
+	}
+	lossPct, ok, err := parseFloatEnv(benchmarkTCPProxyLossPctEnv)
+	if err != nil {
+		tb.Fatalf("invalid %s: %v", benchmarkTCPProxyLossPctEnv, err)
+	}
+	if ok {
+		if lossPct < 0 || lossPct > 100 {
+			tb.Fatalf("%s must be in [0, 100], got %.4f", benchmarkTCPProxyLossPctEnv, lossPct)
+		}
+		cfg.LossPct = lossPct
+	}
+	lossPenaltyMs, ok, err := parseIntEnv(benchmarkTCPProxyLossPenaltyMsEnv)
+	if err != nil {
+		tb.Fatalf("invalid %s: %v", benchmarkTCPProxyLossPenaltyMsEnv, err)
+	}
+	if ok && lossPenaltyMs > 0 {
+		cfg.LossPenalty = time.Duration(lossPenaltyMs) * time.Millisecond
+	}
+	if cfg.LossPct > 0 && cfg.LossPenalty <= 0 {
+		cfg.LossPenalty = benchmarkTCPProxyDefaultLossPenalty
+	}
+	seed, ok, err := parseInt64Env(benchmarkTCPProxySeedEnv)
+	if err != nil {
+		tb.Fatalf("invalid %s: %v", benchmarkTCPProxySeedEnv, err)
+	}
+	if ok {
+		cfg.Seed = seed
+	}
+	return cfg
+}
+
+func parseBoolEnv(name string) (value bool, ok bool, err error) {
+	raw := os.Getenv(name)
+	if raw == "" {
+		return false, false, nil
+	}
+	value, err = strconv.ParseBool(raw)
+	return value, true, err
+}
+
+func parseFloatEnv(name string) (value float64, ok bool, err error) {
+	raw := os.Getenv(name)
+	if raw == "" {
+		return 0, false, nil
+	}
+	value, err = strconv.ParseFloat(raw, 64)
+	return value, true, err
+}
+
+func parseIntEnv(name string) (value int, ok bool, err error) {
+	raw := os.Getenv(name)
+	if raw == "" {
+		return 0, false, nil
+	}
+	value, err = strconv.Atoi(raw)
+	return value, true, err
+}
+
+func parseInt64Env(name string) (value int64, ok bool, err error) {
+	raw := os.Getenv(name)
+	if raw == "" {
+		return 0, false, nil
+	}
+	value, err = strconv.ParseInt(raw, 10, 64)
+	return value, true, err
+}
+
+func startBenchmarkTCPProxy(targetAddr string, cfg benchmarkTCPProxyConfig) (*benchmarkTCPProxy, error) {
+	if targetAddr == "" {
+		return nil, errors.New("benchmark tcp proxy target addr is empty")
+	}
+	listenAddr := cfg.ListenAddr
+	if listenAddr == "" {
+		listenAddr = benchmarkTCPProxyDefaultListenAddr
+	}
+	listener, err := net.Listen("tcp", listenAddr)
+	if err != nil {
+		return nil, err
+	}
+	proxy := &benchmarkTCPProxy{
+		listener: listener,
+		target:   targetAddr,
+		cfg:      cfg,
+	}
+	proxy.wg.Add(1)
+	go proxy.runAcceptLoop()
+	return proxy, nil
+}
+
+func (p *benchmarkTCPProxy) runAcceptLoop() {
+	defer p.wg.Done()
+	for {
+		conn, err := p.listener.Accept()
+		if err != nil {
+			if p.closed.Load() || errors.Is(err, net.ErrClosed) {
+				return
+			}
+			continue
+		}
+		p.wg.Add(1)
+		go func(clientConn net.Conn) {
+			defer p.wg.Done()
+			p.handleConn(clientConn)
+		}(conn)
+	}
+}
+
+func (p *benchmarkTCPProxy) handleConn(clientConn net.Conn) {
+	if p == nil || clientConn == nil {
+		return
+	}
+	targetConn, err := net.Dial("tcp", p.target)
+	if err != nil {
+		_ = clientConn.Close()
+		return
+	}
+	if tcpConn, ok := clientConn.(*net.TCPConn); ok {
+		_ = tcpConn.SetNoDelay(true)
+	}
+	if tcpConn, ok := targetConn.(*net.TCPConn); ok {
+		_ = tcpConn.SetNoDelay(true)
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	errCh := make(chan error, 2)
+	leftRng := rand.New(rand.NewSource(p.cfg.Seed + 101))
+	rightRng := rand.New(rand.NewSource(p.cfg.Seed + 202))
+	go func() {
+		errCh <- benchmarkRelayStream(ctx, targetConn, clientConn, p.cfg, leftRng)
+	}()
+	go func() {
+		errCh <- benchmarkRelayStream(ctx, clientConn, targetConn, p.cfg, rightRng)
+	}()
+
+	_ = <-errCh
+	cancel()
+	_ = clientConn.Close()
+	_ = targetConn.Close()
+	<-errCh
+}
+
+func (p *benchmarkTCPProxy) stop() {
+	if p == nil {
+		return
+	}
+	if p.closed.CompareAndSwap(false, true) {
+		if p.listener != nil {
+			_ = p.listener.Close()
+		}
+	}
+	p.wg.Wait()
+}
+
+func benchmarkRelayStream(ctx context.Context, dst net.Conn, src net.Conn, cfg benchmarkTCPProxyConfig, rng *rand.Rand) error {
+	if dst == nil || src == nil {
+		return net.ErrClosed
+	}
+	chunkCh := make(chan benchmarkRelayChunk, 128)
+	readerErrCh := make(chan error, 1)
+	go benchmarkRelayReader(ctx, src, chunkCh, readerErrCh)
+
+	scheduler := benchmarkRelayScheduler{rateBytesPS: cfg.RateBytesPS}
+	for chunk := range chunkCh {
+		sendAt := scheduler.schedule(chunk.readAt, len(chunk.payload), cfg.Delay)
+		if cfg.LossPct > 0 && cfg.LossPenalty > 0 && rng != nil && rng.Float64()*100 < cfg.LossPct {
+			sendAt = sendAt.Add(cfg.LossPenalty)
+		}
+		if waitErr := benchmarkSleepUntilContext(ctx, sendAt); waitErr != nil {
+			return waitErr
+		}
+		if writeErr := writeAllToConn(dst, chunk.payload); writeErr != nil {
+			return writeErr
+		}
+	}
+	readerErr := <-readerErrCh
+	if errors.Is(readerErr, io.EOF) {
+		return nil
+	}
+	return readerErr
+}
+
+type benchmarkRelayChunk struct {
+	readAt  time.Time
+	payload []byte
+}
+
+func benchmarkRelayReader(ctx context.Context, src net.Conn, chunkCh chan<- benchmarkRelayChunk, errCh chan<- error) {
+	defer close(chunkCh)
+	if src == nil {
+		errCh <- net.ErrClosed
+		return
+	}
+	buf := make([]byte, 64*1024)
+	for {
+		n, err := src.Read(buf)
+		if n > 0 {
+			chunk := benchmarkRelayChunk{
+				readAt:  time.Now(),
+				payload: append([]byte(nil), buf[:n]...),
+			}
+			select {
+			case <-ctx.Done():
+				errCh <- ctx.Err()
+				return
+			case chunkCh <- chunk:
+			}
+		}
+		if err != nil {
+			errCh <- err
+			return
+		}
+	}
+}
+
+type benchmarkRelayScheduler struct {
+	rateBytesPS float64
+	nextAt      time.Time
+}
+
+func (s *benchmarkRelayScheduler) schedule(readAt time.Time, bytes int, delay time.Duration) time.Time {
+	if s == nil {
+		return readAt
+	}
+	sendAt := readAt
+	if delay > 0 {
+		sendAt = sendAt.Add(delay)
+	}
+	if !s.nextAt.IsZero() && s.nextAt.After(sendAt) {
+		sendAt = s.nextAt
+	}
+	if s.rateBytesPS > 0 && bytes > 0 {
+		duration := time.Duration(float64(bytes) / s.rateBytesPS * float64(time.Second))
+		if duration <= 0 {
+			duration = time.Nanosecond
+		}
+		s.nextAt = sendAt.Add(duration)
+	} else {
+		s.nextAt = sendAt
+	}
+	return sendAt
+}
+
+func benchmarkSleepUntilContext(ctx context.Context, at time.Time) error {
+	wait := time.Until(at)
+	if wait <= 0 {
+		return nil
+	}
+	return benchmarkSleepContext(ctx, wait)
+}
+
+func benchmarkSleepContext(ctx context.Context, d time.Duration) error {
+	if d <= 0 {
+		return nil
+	}
+	timer := time.NewTimer(d)
+	defer timer.Stop()
+	select {
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-timer.C:
+		return nil
+	}
+}
+
+func writeAllToConn(conn net.Conn, payload []byte) error {
+	for len(payload) > 0 {
+		n, err := conn.Write(payload)
+		if n > 0 {
+			payload = payload[n:]
+		}
+		if err != nil {
+			return err
+		}
+		if n == 0 {
+			return io.ErrNoProgress
+		}
+	}
+	return nil
+}

bulk_batch_sender.go

@@ -9,7 +9,9 @@ import (
 )
 
 const (
-	bulkBatchMaxPayloads = 16
+	bulkBatchMaxPayloads     = 64
+	bulkBatchMaxPayloadBytes = bulkFastBatchMaxPlainBytes
+	bulkBatchMaxFlushDelay   = 50 * time.Microsecond
 )
 
 const (
@@ -22,48 +24,161 @@ type bulkBatchRequestState struct {
 	value atomic.Int32
 }
 
+type bulkBatchCodec struct {
+	encodeSingle func(bulkFastFrame) ([]byte, func(), error)
+	encodeBatch  func([]bulkFastFrame) ([]byte, func(), error)
+}
+
 type bulkBatchRequest struct {
-	ctx      context.Context
-	payload  []byte
-	deadline time.Time
-	done     chan error
-	state    *bulkBatchRequestState
+	ctx             context.Context
+	frames          []bulkFastFrame
+	fastPathVersion uint8
+	payloadOwned    bool
+	deadline        time.Time
+	done            chan error
+	state           *bulkBatchRequestState
+	release         func()
+}
+
+type bulkBatchEncodedPayload struct {
+	payload []byte
+	release func()
+}
+
+func (p *bulkBatchEncodedPayload) done() {
+	if p == nil || p.release == nil {
+		return
+	}
+	p.release()
+	p.release = nil
 }
 
 type bulkBatchSender struct {
-	binding *transportBinding
-	reqCh   chan bulkBatchRequest
-	stopCh  chan struct{}
-	doneCh  chan struct{}
+	binding              *transportBinding
+	codec                bulkBatchCodec
+	writeTimeoutProvider func() time.Duration
+	reqCh                chan bulkBatchRequest
+	stopCh               chan struct{}
+	doneCh               chan struct{}
 
 	stopOnce sync.Once
+	flushMu  sync.Mutex
+	queued   atomic.Int64
 	errMu    sync.Mutex
 	err      error
 }
 
-func newBulkBatchSender(binding *transportBinding) *bulkBatchSender {
+func newBulkBatchSender(binding *transportBinding, codec bulkBatchCodec, writeTimeoutProvider func() time.Duration) *bulkBatchSender {
 	sender := &bulkBatchSender{
-		binding: binding,
-		reqCh:   make(chan bulkBatchRequest, bulkBatchMaxPayloads*4),
-		stopCh:  make(chan struct{}),
-		doneCh:  make(chan struct{}),
+		binding:              binding,
+		codec:                codec,
+		writeTimeoutProvider: writeTimeoutProvider,
+		reqCh:                make(chan bulkBatchRequest, bulkBatchMaxPayloads*4),
+		stopCh:               make(chan struct{}),
+		doneCh:               make(chan struct{}),
 	}
 	go sender.run()
 	return sender
 }
 
-func (s *bulkBatchSender) submit(ctx context.Context, payload []byte) error {
+func (s *bulkBatchSender) submitData(ctx context.Context, dataID uint64, seq uint64, fastPathVersion uint8, payload []byte) error {
+	return s.submitFramesOwned(ctx, []bulkFastFrame{{
+		Type:    bulkFastPayloadTypeData,
+		DataID:  dataID,
+		Seq:     seq,
+		Payload: payload,
+	}}, fastPathVersion, false)
+}
+
+func (s *bulkBatchSender) submitControl(ctx context.Context, frameType uint8, flags uint8, dataID uint64, seq uint64, fastPathVersion uint8, payload []byte) error {
+	return s.submitFramesOwned(ctx, []bulkFastFrame{{
+		Type:    frameType,
+		Flags:   flags,
+		DataID:  dataID,
+		Seq:     seq,
+		Payload: payload,
+	}}, fastPathVersion, false)
+}
+
+func (s *bulkBatchSender) submitWrite(ctx context.Context, dataID uint64, startSeq uint64, fastPathVersion uint8, payload []byte, chunkSize int, payloadOwned bool) (int, error) {
+	if s == nil {
+		return 0, errTransportDetached
+	}
+	if len(payload) == 0 {
+		return 0, nil
+	}
+	if chunkSize <= 0 {
+		chunkSize = defaultBulkChunkSize
+	}
+	written := 0
+	seq := startSeq
+	for written < len(payload) {
+		var batch [bulkFastBatchMaxItems]bulkFastFrame
+		frames := batch[:0]
+		batchBytes := bulkFastBatchHeaderLen
+		start := written
+		for written < len(payload) && len(frames) < bulkFastBatchMaxItems {
+			end := written + chunkSize
+			if end > len(payload) {
+				end = len(payload)
+			}
+			frame := bulkFastFrame{
+				Type:    bulkFastPayloadTypeData,
+				DataID:  dataID,
+				Seq:     seq,
+				Payload: payload[written:end],
+			}
+			frameLen := bulkFastBatchFrameLen(frame)
+			if len(frames) > 0 && batchBytes+frameLen > bulkFastBatchMaxPlainBytes {
+				break
+			}
+			frames = append(frames, frame)
+			batchBytes += frameLen
+			seq++
+			written = end
+		}
+		if len(frames) == 0 {
+			end := written + chunkSize
+			if end > len(payload) {
+				end = len(payload)
+			}
+			frames = append(frames, bulkFastFrame{
+				Type:    bulkFastPayloadTypeData,
+				DataID:  dataID,
+				Seq:     seq,
+				Payload: payload[written:end],
+			})
+			seq++
+			written = end
+		}
+		if err := s.submitFramesOwned(ctx, frames, fastPathVersion, payloadOwned); err != nil {
+			return start, err
+		}
+	}
+	return written, nil
+}
+
+func (s *bulkBatchSender) submitFrames(ctx context.Context, frames []bulkFastFrame, fastPathVersion uint8) error {
+	return s.submitFramesOwned(ctx, frames, fastPathVersion, false)
+}
+
+func (s *bulkBatchSender) submitFramesOwned(ctx context.Context, frames []bulkFastFrame, fastPathVersion uint8, payloadOwned bool) error {
 	if s == nil {
 		return errTransportDetached
 	}
 	if ctx == nil {
 		ctx = context.Background()
 	}
+	if len(frames) == 0 {
+		return nil
+	}
 	req := bulkBatchRequest{
-		ctx:     ctx,
-		payload: payload,
-		done:    make(chan error, 1),
-		state:   &bulkBatchRequestState{},
+		ctx:             ctx,
+		frames:          frames,
+		fastPathVersion: normalizeBulkFastPathVersion(fastPathVersion),
+		payloadOwned:    payloadOwned,
+		done:            make(chan error, 1),
+		state:           &bulkBatchRequestState{},
 	}
 	if deadline, ok := ctx.Deadline(); ok {
 		req.deadline = deadline
@@ -71,10 +186,25 @@ func (s *bulkBatchSender) submit(ctx context.Context, payload []byte) error {
 	if err := s.errSnapshot(); err != nil {
 		return err
 	}
+	if s.shouldDirectSubmit(req) {
+		if submitted, err := s.tryDirectSubmit(req); submitted {
+			return err
+		}
+	}
+	req = cloneQueuedBulkBatchRequest(req)
+	s.queued.Add(1)
 	select {
 	case <-ctx.Done():
+		s.queued.Add(-1)
+		if req.release != nil {
+			req.release()
+		}
 		return normalizeStreamDeadlineError(ctx.Err())
 	case <-s.stopCh:
+		s.queued.Add(-1)
+		if req.release != nil {
+			req.release()
+		}
 		return s.stoppedErr()
 	case s.reqCh <- req:
 	}
@@ -89,6 +219,55 @@ func (s *bulkBatchSender) submit(ctx context.Context, payload []byte) error {
 	}
 }
 
+func (s *bulkBatchSender) shouldDirectSubmit(req bulkBatchRequest) bool {
+	if len(req.frames) == 0 {
+		return false
+	}
+	return !bulkBatchRequestSupportsSharedSuperBatch(req)
+}
+
+func (s *bulkBatchSender) tryDirectSubmit(req bulkBatchRequest) (bool, error) {
+	if s == nil {
+		return true, errTransportDetached
+	}
+	if err := s.errSnapshot(); err != nil {
+		return true, err
+	}
+	select {
+	case <-req.ctx.Done():
+		return true, normalizeStreamDeadlineError(req.ctx.Err())
+	case <-s.stopCh:
+		return true, s.stoppedErr()
+	default:
+	}
+	if s.queued.Load() != 0 {
+		return false, nil
+	}
+	if !s.flushMu.TryLock() {
+		return false, nil
+	}
+	defer s.flushMu.Unlock()
+	if s.queued.Load() != 0 {
+		return false, nil
+	}
+	if err := s.errSnapshot(); err != nil {
+		return true, err
+	}
+	if !req.tryStart() {
+		return true, req.canceledErr()
+	}
+	if err := req.contextErr(); err != nil {
+		return true, err
+	}
+	err := s.flush([]bulkBatchRequest{req})
+	if err != nil {
+		s.setErr(err)
+		s.failPending(err)
+		return true, err
+	}
+	return true, nil
+}
+
 func (s *bulkBatchSender) run() {
 	defer close(s.doneCh)
 	for {
@@ -97,34 +276,83 @@ func (s *bulkBatchSender) run() {
 			return
 		}
 		batch := []bulkBatchRequest{req}
+		batchBytes := bulkBatchRequestApproxBytes(req)
+		timer := (*time.Timer)(nil)
+		timerCh := (<-chan time.Time)(nil)
+		if bulkBatchShouldWaitForMore(batch, batchBytes) {
+			timer = time.NewTimer(bulkBatchMaxFlushDelay)
+			timerCh = timer.C
+		}
 	drain:
-		for len(batch) < bulkBatchMaxPayloads {
+		for len(batch) < bulkBatchMaxPayloads && batchBytes < bulkBatchMaxPayloadBytes {
+			if timerCh == nil {
+				select {
+				case <-s.stopCh:
+					s.failPending(s.stoppedErr())
+					return
+				case next := <-s.reqCh:
+					batch = append(batch, next)
+					batchBytes += bulkBatchRequestApproxBytes(next)
+				default:
+					break drain
+				}
+				continue
+			}
 			select {
 			case <-s.stopCh:
+				if timer != nil {
+					timer.Stop()
+				}
 				s.failPending(s.stoppedErr())
 				return
 			case next := <-s.reqCh:
 				batch = append(batch, next)
-			default:
+				batchBytes += bulkBatchRequestApproxBytes(next)
+			case <-timerCh:
+				timerCh = nil
 				break drain
 			}
 		}
-		active, payloads := activeBulkBatchRequests(batch)
+		if timer != nil {
+			if !timer.Stop() && timerCh != nil {
+				select {
+				case <-timer.C:
+				default:
+				}
+			}
+		}
+		s.flushMu.Lock()
+		err := s.errSnapshot()
+		active := make([]bulkBatchRequest, 0, len(batch))
+		for _, item := range batch {
+			if !item.tryStart() {
+				s.finishRequest(item, item.canceledErr())
+				continue
+			}
+			if itemErr := item.contextErr(); itemErr != nil {
+				s.finishRequest(item, itemErr)
+				continue
+			}
+			active = append(active, item)
+		}
 		if len(active) == 0 {
+			s.flushMu.Unlock()
 			continue
 		}
-		deadline := bulkBatchRequestsEarliestDeadline(active)
-		err := s.flush(payloads, deadline)
+		if err == nil {
+			err = s.flush(active)
+		}
+		s.flushMu.Unlock()
 		if err != nil {
 			s.setErr(err)
 			for _, item := range active {
-				item.done <- err
+				s.finishRequest(item, err)
 			}
 			s.failPending(err)
 			return
 		}
 		for _, item := range active {
-			item.done <- err
+			s.finishRequest(item, nil)
 		}
 	}
 }
@@ -139,37 +367,6 @@ func (s *bulkBatchSender) nextRequest() (bulkBatchRequest, bool) {
 	}
 }
 
-func activeBulkBatchRequests(batch []bulkBatchRequest) ([]bulkBatchRequest, [][]byte) {
-	active := make([]bulkBatchRequest, 0, len(batch))
-	payloads := make([][]byte, 0, len(batch))
-	for _, item := range batch {
-		if !item.tryStart() {
-			item.done <- item.canceledErr()
-			continue
-		}
-		if err := item.contextErr(); err != nil {
-			item.done <- err
-			continue
-		}
-		active = append(active, item)
-		payloads = append(payloads, item.payload)
-	}
-	return active, payloads
-}
-
-func bulkBatchRequestsEarliestDeadline(batch []bulkBatchRequest) time.Time {
-	var deadline time.Time
-	for _, item := range batch {
-		if item.deadline.IsZero() {
-			continue
-		}
-		if deadline.IsZero() || item.deadline.Before(deadline) {
-			deadline = item.deadline
-		}
-	}
-	return deadline
-}
-
 func (r bulkBatchRequest) contextErr() error {
 	if r.ctx == nil {
 		return nil
@@ -203,7 +400,7 @@ func (r bulkBatchRequest) canceledErr() error {
 	return context.Canceled
 }
 
-func (s *bulkBatchSender) flush(payloads [][]byte, deadline time.Time) error {
+func (s *bulkBatchSender) flush(requests []bulkBatchRequest) error {
 	if s == nil || s.binding == nil {
 		return errTransportDetached
 	}
@@ -211,9 +408,196 @@ func (s *bulkBatchSender) flush(payloads [][]byte, deadline time.Time) error {
 	if queue == nil {
 		return errTransportFrameQueueUnavailable
 	}
-	return s.binding.withConnWriteLockDeadline(deadline, func(conn net.Conn) error {
-		return writeFramedPayloadBatchUnlocked(conn, queue, payloads)
-	})
+	payloads, err := s.encodeRequests(requests)
+	if err != nil {
+		return err
+	}
+	defer func() {
+		for index := range payloads {
+			payloads[index].done()
+		}
+	}()
+	writeTimeout := s.transportWriteTimeout()
+	for _, payload := range payloads {
+		frame := payload.payload
+		started := time.Now()
+		err := s.binding.withConnWriteLockDeadline(writeDeadlineFromTimeout(writeTimeout), func(conn net.Conn) error {
+			return writeFramedPayloadUnlocked(conn, queue, frame)
+		})
+		s.binding.observeBulkAdaptivePayloadWrite(len(frame), time.Since(started), writeTimeout, err)
+		if err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (s *bulkBatchSender) encodeRequests(requests []bulkBatchRequest) ([]bulkBatchEncodedPayload, error) {
+	if len(requests) == 0 {
+		return nil, nil
+	}
+	payloads := make([]bulkBatchEncodedPayload, 0, len(requests))
+	batch := make([]bulkFastFrame, 0, minInt(len(requests), bulkFastBatchMaxItems))
+	mixedBatchLimit := s.sharedMixedPayloadLimit()
+	batchRequestIndex := -1
+	batchDataID := uint64(0)
+	batchMixed := false
+	flushBatch := func() error {
+		if len(batch) == 0 {
+			return nil
+		}
+		payload, release, err := s.encodeBatch(batch)
+		if err != nil {
+			return err
+		}
+		payloads = append(payloads, bulkBatchEncodedPayload{payload: payload, release: release})
+		batch = batch[:0]
+		batchRequestIndex = -1
+		batchDataID = 0
+		batchMixed = false
+		return nil
+	}
+	batchBytes := bulkFastBatchHeaderLen
+	for reqIndex, req := range requests {
+		for _, frame := range req.frames {
+			if !bulkFastPathSupportsSharedBatch(req.fastPathVersion) {
+				if err := flushBatch(); err != nil {
+					return nil, err
+				}
+				batchBytes = bulkFastBatchHeaderLen
+				payload, release, err := s.encodeSingle(frame)
+				if err != nil {
+					return nil, err
+				}
+				payloads = append(payloads, bulkBatchEncodedPayload{payload: payload, release: release})
+				continue
+			}
+			frameLen := bulkFastBatchFrameLen(frame)
+			if frameLen+bulkFastBatchHeaderLen > bulkFastBatchMaxPlainBytes {
+				if err := flushBatch(); err != nil {
+					return nil, err
+				}
+				batchBytes = bulkFastBatchHeaderLen
+				payload, release, err := s.encodeSingle(frame)
+				if err != nil {
+					return nil, err
+				}
+				payloads = append(payloads, bulkBatchEncodedPayload{payload: payload, release: release})
+				continue
+			}
+			nextMixed := batchMixed
+			if len(batch) > 0 && (batchRequestIndex != reqIndex || (batchDataID != 0 && batchDataID != frame.DataID)) {
+				nextMixed = true
+			}
+			batchLimit := bulkFastBatchMaxPlainBytes
+			if nextMixed && mixedBatchLimit > 0 && mixedBatchLimit < batchLimit {
+				batchLimit = mixedBatchLimit
+			}
+			if len(batch) > 0 && (len(batch) >= bulkFastBatchMaxItems || batchBytes+frameLen > batchLimit) {
+				if err := flushBatch(); err != nil {
+					return nil, err
+				}
+				batchBytes = bulkFastBatchHeaderLen
+				nextMixed = false
+			}
+			if len(batch) == 0 {
+				batchRequestIndex = reqIndex
+				batchDataID = frame.DataID
+				batchMixed = false
+			} else if batchRequestIndex != reqIndex || (batchDataID != 0 && batchDataID != frame.DataID) {
+				batchMixed = true
+			}
+			batch = append(batch, frame)
+			batchBytes += frameLen
+		}
+	}
+	if err := flushBatch(); err != nil {
+		return nil, err
+	}
+	return payloads, nil
+}
+
+func bulkBatchRequestApproxBytes(req bulkBatchRequest) int {
+	total := 0
+	for _, frame := range req.frames {
+		total += bulkFastBatchFrameLen(frame)
+	}
+	return total
+}
+
+func bulkBatchRequestSupportsSharedSuperBatch(req bulkBatchRequest) bool {
+	if len(req.frames) == 0 || !bulkFastPathSupportsSharedBatch(req.fastPathVersion) {
+		return false
+	}
+	for _, frame := range req.frames {
+		switch frame.Type {
+		case bulkFastPayloadTypeData:
+		default:
+			return false
+		}
+	}
+	return true
+}
+
+func bulkBatchShouldWaitForMore(batch []bulkBatchRequest, batchBytes int) bool {
+	if bulkBatchMaxFlushDelay <= 0 || len(batch) == 0 {
+		return false
+	}
+	if len(batch) >= bulkBatchMaxPayloads || batchBytes >= bulkBatchMaxPayloadBytes {
+		return false
+	}
+	for _, req := range batch {
+		if !bulkBatchRequestSupportsSharedSuperBatch(req) {
+			return false
+		}
+	}
+	return true
+}
+
+func cloneQueuedBulkBatchRequest(req bulkBatchRequest) bulkBatchRequest {
+	if len(req.frames) == 0 || req.payloadOwned {
+		return req
+	}
+	clonedFrames := make([]bulkFastFrame, len(req.frames))
+	totalPayload := 0
+	for _, frame := range req.frames {
+		totalPayload += len(frame.Payload)
+	}
+	var payloadBuf []byte
+	if totalPayload > 0 {
+		payloadBuf = getBulkAsyncWritePayload(totalPayload)
+		req.release = func() {
+			putBulkAsyncWritePayload(payloadBuf)
+		}
+	}
+	offset := 0
+	for index, frame := range req.frames {
+		clonedFrames[index] = frame
+		if len(frame.Payload) == 0 {
+			clonedFrames[index].Payload = nil
+			continue
+		}
+		next := offset + len(frame.Payload)
+		clonedFrames[index].Payload = payloadBuf[offset:next]
+		copy(clonedFrames[index].Payload, frame.Payload)
+		offset = next
+	}
+	req.frames = clonedFrames
+	return req
+}
+
+func (s *bulkBatchSender) encodeSingle(frame bulkFastFrame) ([]byte, func(), error) {
+	if s == nil || s.codec.encodeSingle == nil {
+		return nil, nil, errTransportDetached
+	}
+	return s.codec.encodeSingle(frame)
+}
+
+func (s *bulkBatchSender) encodeBatch(frames []bulkFastFrame) ([]byte, func(), error) {
+	if len(frames) == 1 || s.codec.encodeBatch == nil {
+		return s.encodeSingle(frames[0])
+	}
+	return s.codec.encodeBatch(frames)
 }
 
 func (s *bulkBatchSender) stop() {
@@ -231,13 +615,23 @@ func (s *bulkBatchSender) failPending(err error) {
 	for {
 		select {
 		case item := <-s.reqCh:
-			item.done <- err
+			s.finishRequest(item, err)
 		default:
 			return
 		}
 	}
 }
 
+func (s *bulkBatchSender) finishRequest(req bulkBatchRequest, err error) {
+	if s != nil {
+		s.queued.Add(-1)
+	}
+	if req.release != nil {
+		req.release()
+	}
+	req.done <- err
+}
+
 func (s *bulkBatchSender) setErr(err error) {
 	if s == nil || err == nil {
 		return
@@ -264,3 +658,31 @@ func (s *bulkBatchSender) stoppedErr() error {
 	}
 	return errTransportDetached
 }
+
+func (s *bulkBatchSender) transportWriteDeadline() time.Time {
+	if s == nil || s.writeTimeoutProvider == nil {
+		return time.Time{}
+	}
+	return writeDeadlineFromTimeout(s.writeTimeoutProvider())
+}
+
+func (s *bulkBatchSender) transportWriteTimeout() time.Duration {
+	if s == nil || s.writeTimeoutProvider == nil {
+		return 0
+	}
+	return s.writeTimeoutProvider()
+}
+
+func (s *bulkBatchSender) sharedMixedPayloadLimit() int {
+	if s == nil || s.binding == nil {
+		return bulkAdaptiveSoftPayloadFallbackBytes
+	}
+	return s.binding.bulkAdaptiveSoftPayloadBytesSnapshot()
+}
+
+func minInt(a int, b int) int {
+	if a < b {
+		return a
+	}
+	return b
+}

bulk_benchmark_test.go

@@ -161,7 +161,7 @@ func benchmarkBulkTCPThroughput(b *testing.B, payloadSize int, dedicated bool) {
 		return nil
 	})
 
-	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+	if err := server.Listen("tcp", benchmarkTCPListenAddr(b)); err != nil {
 		b.Fatalf("server Listen failed: %v", err)
 	}
 	b.Cleanup(func() {
@@ -172,7 +172,7 @@ func benchmarkBulkTCPThroughput(b *testing.B, payloadSize int, dedicated bool) {
 	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
 		b.Fatalf("UseModernPSKClient failed: %v", err)
 	}
-	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+	if err := client.Connect("tcp", benchmarkTCPDialAddr(b, server.listener.Addr().String())); err != nil {
 		b.Fatalf("client Connect failed: %v", err)
 	}
 	b.Cleanup(func() {
@@ -258,7 +258,7 @@ func benchmarkBulkTCPThroughputConcurrent(b *testing.B, payloadSize int, concurr
 		return nil
 	})
 
-	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+	if err := server.Listen("tcp", benchmarkTCPListenAddr(b)); err != nil {
 		b.Fatalf("server Listen failed: %v", err)
 	}
 	b.Cleanup(func() {
@@ -269,7 +269,7 @@ func benchmarkBulkTCPThroughputConcurrent(b *testing.B, payloadSize int, concurr
 	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
 		b.Fatalf("UseModernPSKClient failed: %v", err)
 	}
-	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+	if err := client.Connect("tcp", benchmarkTCPDialAddr(b, server.listener.Addr().String())); err != nil {
 		b.Fatalf("client Connect failed: %v", err)
 	}
 	b.Cleanup(func() {

bulk_buffer_release_test.go

@@ -0,0 +1,119 @@
+package notify
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+)
+
+func TestBulkOwnedChunkReleaseAfterRead(t *testing.T) {
+	bulk := newBulkHandle(context.Background(), newBulkRuntime("buffer-release-read"), clientFileScope(), BulkOpenRequest{
+		BulkID: "buffer-release-read",
+		DataID: 1,
+	}, 0, nil, nil, 0, nil, nil, nil, nil, nil)
+
+	released := 0
+	if err := bulk.pushOwnedChunkWithReleaseNoReset([]byte("hello"), func() {
+		released++
+	}); err != nil {
+		t.Fatalf("pushOwnedChunkWithReleaseNoReset failed: %v", err)
+	}
+
+	buf := make([]byte, 5)
+	n, err := bulk.Read(buf)
+	if err != nil {
+		t.Fatalf("Read failed: %v", err)
+	}
+	if n != 5 || string(buf[:n]) != "hello" {
+		t.Fatalf("Read = %d %q, want 5 hello", n, string(buf[:n]))
+	}
+	if released != 1 {
+		t.Fatalf("release count = %d, want 1", released)
+	}
+}
+
+func TestBulkOwnedChunkReleaseOnReset(t *testing.T) {
+	bulk := newBulkHandle(context.Background(), newBulkRuntime("buffer-release-reset"), clientFileScope(), BulkOpenRequest{
+		BulkID: "buffer-release-reset",
+		DataID: 1,
+	}, 0, nil, nil, 0, nil, nil, nil, nil, nil)
+
+	released := 0
+	if err := bulk.pushOwnedChunkWithReleaseNoReset([]byte("hello"), func() {
+		released++
+	}); err != nil {
+		t.Fatalf("pushOwnedChunkWithReleaseNoReset failed: %v", err)
+	}
+
+	bulk.markReset(errors.New("boom"))
+	if released != 1 {
+		t.Fatalf("release count = %d, want 1", released)
+	}
+}
+
+func TestBulkReadDoesNotBlockOnAsyncWindowRelease(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	releaseStarted := make(chan struct{})
+	releaseUnblock := make(chan struct{})
+	bulk := newBulkHandle(ctx, newBulkRuntime("buffer-release-async"), clientFileScope(), BulkOpenRequest{
+		BulkID:      "buffer-release-async",
+		DataID:      1,
+		ChunkSize:   4,
+		WindowBytes: 4,
+		MaxInFlight: 1,
+	}, 0, nil, nil, 0, nil, nil, nil, nil, func(_ *bulkHandle, bytes int64, chunks int) error {
+		if bytes != 4 || chunks != 1 {
+			t.Fatalf("release = (%d,%d), want (4,1)", bytes, chunks)
+		}
+		close(releaseStarted)
+		<-releaseUnblock
+		return nil
+	})
+
+	if err := bulk.pushOwnedChunk([]byte("ping")); err != nil {
+		t.Fatalf("pushOwnedChunk failed: %v", err)
+	}
+
+	buf := make([]byte, 4)
+	doneCh := make(chan error, 1)
+	go func() {
+		n, err := bulk.Read(buf)
+		if err != nil {
+			doneCh <- err
+			return
+		}
+		if got, want := n, 4; got != want {
+			doneCh <- errors.New("unexpected read size")
+			return
+		}
+		doneCh <- nil
+	}()
+
+	select {
+	case err := <-doneCh:
+		if err != nil {
+			t.Fatalf("Read failed: %v", err)
+		}
+	case <-time.After(200 * time.Millisecond):
+		t.Fatal("Read should not block on async release sender")
+	}
+
+	select {
+	case <-releaseStarted:
+	case <-time.After(time.Second):
+		t.Fatal("window release sender did not start")
+	}
+
+	close(releaseUnblock)
+	cancel()
+	if bulk.releaseWorkerDone != nil {
+		select {
+		case <-bulk.releaseWorkerDone:
+		case <-time.After(time.Second):
+			t.Fatal("release worker did not exit")
+		}
+	}
+}

bulk_control.go

@@ -7,22 +7,25 @@ import (
 )
 
 type BulkOpenRequest struct {
-	BulkID       string
-	DataID       uint64
-	Range        BulkRange
-	Metadata     BulkMetadata
-	ReadTimeout  time.Duration
-	WriteTimeout time.Duration
-	Dedicated    bool
-	AttachToken  string
-	ChunkSize    int
-	WindowBytes  int
-	MaxInFlight  int
+	BulkID          string
+	DataID          uint64
+	FastPathVersion uint8
+	Range           BulkRange
+	Metadata        BulkMetadata
+	ReadTimeout     time.Duration
+	WriteTimeout    time.Duration
+	Dedicated       bool
+	DedicatedLaneID uint32
+	AttachToken     string
+	ChunkSize       int
+	WindowBytes     int
+	MaxInFlight     int
 }
 
 type BulkOpenResponse struct {
 	BulkID              string
 	DataID              uint64
+	FastPathVersion     uint8
 	Accepted            bool
 	Dedicated           bool
 	AttachToken         string
@@ -53,6 +56,18 @@ type BulkResetResponse struct {
 	Error    string
 }
 
+type BulkReadyRequest struct {
+	BulkID string
+	DataID uint64
+	Error  string
+}
+
+type BulkReadyResponse struct {
+	BulkID   string
+	Accepted bool
+	Error    string
+}
+
 type BulkReleaseRequest struct {
 	BulkID string
 	DataID uint64
@@ -73,6 +88,9 @@ func bindClientBulkControl(c *ClientCommon) {
 	c.SetLink(BulkResetSignalKey, func(msg *Message) {
 		c.handleInboundBulkReset(msg)
 	})
+	c.SetLink(BulkReadySignalKey, func(msg *Message) {
+		c.handleInboundBulkReady(msg)
+	})
 	c.SetLink(BulkReleaseSignalKey, func(msg *Message) {
 		c.handleInboundBulkRelease(msg)
 	})
@@ -91,14 +109,188 @@ func bindServerBulkControl(s *ServerCommon) {
 	s.SetLink(BulkResetSignalKey, func(msg *Message) {
 		s.handleInboundBulkReset(msg)
 	})
+	s.SetLink(BulkReadySignalKey, func(msg *Message) {
+		s.handleInboundBulkReady(msg)
+	})
 	s.SetLink(BulkReleaseSignalKey, func(msg *Message) {
 		s.handleInboundBulkRelease(msg)
 	})
 }
 
+func bulkAcceptInfoFromClientBulk(bulk *bulkHandle) BulkAcceptInfo {
+	if bulk == nil {
+		return BulkAcceptInfo{}
+	}
+	return BulkAcceptInfo{
+		ID:                  bulk.ID(),
+		Range:               bulk.Range(),
+		Metadata:            bulk.Metadata(),
+		Dedicated:           bulk.Dedicated(),
+		TransportGeneration: bulk.TransportGeneration(),
+		Bulk:                bulk,
+	}
+}
+
+func bulkAcceptInfoFromServerBulk(bulk *bulkHandle, logical *LogicalConn, transport *TransportConn) BulkAcceptInfo {
+	if bulk == nil {
+		return BulkAcceptInfo{}
+	}
+	if logical == nil {
+		logical = bulk.LogicalConn()
+	}
+	if transport == nil {
+		transport = bulk.TransportConn()
+	}
+	return BulkAcceptInfo{
+		ID:                  bulk.ID(),
+		Range:               bulk.Range(),
+		Metadata:            bulk.Metadata(),
+		Dedicated:           bulk.Dedicated(),
+		LogicalConn:         logical,
+		TransportConn:       transport,
+		TransportGeneration: bulk.TransportGeneration(),
+		Bulk:                bulk,
+	}
+}
+
+func dispatchBulkAccept(handler func(BulkAcceptInfo) error, bulk *bulkHandle, info BulkAcceptInfo) error {
+	if bulk == nil {
+		return errBulkNotFound
+	}
+	if !bulk.markAcceptDispatched() {
+		return nil
+	}
+	var dispatchErr error
+	defer func() {
+		bulk.finishAcceptDispatch(dispatchErr)
+	}()
+	if handler == nil {
+		dispatchErr = errBulkHandlerNotConfigured
+		bulk.markReset(dispatchErr)
+		return dispatchErr
+	}
+	if err := handler(info); err != nil {
+		dispatchErr = err
+		bulk.markReset(dispatchErr)
+		return dispatchErr
+	}
+	return nil
+}
+
+func (c *ClientCommon) clientBulkAcceptReadyNotifier(bulk *bulkHandle) func(error) {
+	return func(readyErr error) {
+		if c == nil || bulk == nil {
+			return
+		}
+		req := BulkReadyRequest{
+			BulkID: bulk.ID(),
+			DataID: bulk.dataIDSnapshot(),
+		}
+		if readyErr != nil {
+			req.Error = readyErr.Error()
+		}
+		ctx, cancel := context.WithTimeout(context.Background(), defaultBulkAcceptReadyTimeout)
+		defer cancel()
+		if _, err := sendBulkReadyClient(ctx, c, req); err != nil && bulk.Context().Err() == nil {
+			bulk.markReset(err)
+		}
+	}
+}
+
+func sendBulkReadyServer(ctx context.Context, s *ServerCommon, logical *LogicalConn, transport *TransportConn, req BulkReadyRequest) error {
+	if s == nil {
+		return errBulkServerNil
+	}
+	if transport != nil {
+		if _, err := sendBulkReadyServerTransport(ctx, s, transport, req); err == nil {
+			return nil
+		} else if !errors.Is(err, errTransportDetached) && !errors.Is(err, errBulkTransportNil) {
+			return err
+		}
+	}
+	if logical == nil {
+		return errBulkLogicalConnNil
+	}
+	_, err := sendBulkReadyServerLogical(ctx, s, logical, req)
+	return err
+}
+
+func (s *ServerCommon) serverBulkAcceptReadyNotifier(bulk *bulkHandle, logical *LogicalConn, transport *TransportConn) func(error) {
+	return func(readyErr error) {
+		if s == nil || bulk == nil {
+			return
+		}
+		req := BulkReadyRequest{
+			BulkID: bulk.ID(),
+			DataID: bulk.dataIDSnapshot(),
+		}
+		if readyErr != nil {
+			req.Error = readyErr.Error()
+		}
+		ctx, cancel := context.WithTimeout(context.Background(), defaultBulkAcceptReadyTimeout)
+		defer cancel()
+		if err := sendBulkReadyServer(ctx, s, logical, transport, req); err != nil && bulk.Context().Err() == nil {
+			bulk.markReset(err)
+		}
+	}
+}
+
+func (c *ClientCommon) startClientBulkAcceptDispatch(bulk *bulkHandle) {
+	if c == nil || bulk == nil {
+		return
+	}
+	bulk.setAcceptNotify(c.clientBulkAcceptReadyNotifier(bulk))
+	go func() {
+		_ = c.dispatchClientBulkAccept(bulk)
+	}()
+}
+
+func (s *ServerCommon) startServerBulkAcceptDispatch(bulk *bulkHandle, logical *LogicalConn, transport *TransportConn) {
+	if s == nil || bulk == nil {
+		return
+	}
+	if logical == nil {
+		logical = bulk.LogicalConn()
+	}
+	if transport == nil {
+		transport = bulk.TransportConn()
+	}
+	bulk.setAcceptNotify(s.serverBulkAcceptReadyNotifier(bulk, logical, transport))
+	go func() {
+		_ = s.dispatchServerBulkAccept(bulk, logical, transport)
+	}()
+}
+
+func (c *ClientCommon) dispatchClientBulkAccept(bulk *bulkHandle) error {
+	if c == nil {
+		return errBulkClientNil
+	}
+	runtime := c.getBulkRuntime()
+	if runtime == nil {
+		return errBulkRuntimeNil
+	}
+	return dispatchBulkAccept(runtime.handlerSnapshot(), bulk, bulkAcceptInfoFromClientBulk(bulk))
+}
+
+func (s *ServerCommon) dispatchServerBulkAccept(bulk *bulkHandle, logical *LogicalConn, transport *TransportConn) error {
+	if s == nil {
+		return errBulkServerNil
+	}
+	runtime := s.getBulkRuntime()
+	if runtime == nil {
+		return errBulkRuntimeNil
+	}
+	return dispatchBulkAccept(runtime.handlerSnapshot(), bulk, bulkAcceptInfoFromServerBulk(bulk, logical, transport))
+}
+
 func (c *ClientCommon) handleInboundBulkOpen(msg *Message) {
 	req, err := decodeBulkOpenRequest(msg)
-	resp := BulkOpenResponse{BulkID: req.BulkID, DataID: req.DataID, Dedicated: req.Dedicated}
+	resp := BulkOpenResponse{
+		BulkID:          req.BulkID,
+		DataID:          req.DataID,
+		FastPathVersion: negotiateBulkFastPathVersion(req.FastPathVersion),
+		Dedicated:       req.Dedicated,
+	}
 	if err != nil {
 		resp.Error = err.Error()
 		replyBulkControlIfNeeded(msg, resp)
@@ -133,13 +325,6 @@ func (c *ClientCommon) handleInboundBulkOpen(msg *Message) {
 		replyBulkControlIfNeeded(msg, resp)
 		return
 	}
-	handler := runtime.handlerSnapshot()
-	if handler == nil {
-		bulk.markReset(errBulkHandlerNotConfigured)
-		resp.Error = errBulkHandlerNotConfigured.Error()
-		replyBulkControlIfNeeded(msg, resp)
-		return
-	}
 	if req.Dedicated {
 		if err := c.attachDedicatedBulkSidecar(context.Background(), bulk); err != nil {
 			bulk.markReset(err)
@@ -147,17 +332,14 @@ func (c *ClientCommon) handleInboundBulkOpen(msg *Message) {
 			replyBulkControlIfNeeded(msg, resp)
 			return
 		}
+		resp.Accepted = true
+		resp.DataID = bulk.dataIDSnapshot()
+		resp.TransportGeneration = bulk.TransportGeneration()
+		replyBulkControlIfNeeded(msg, resp)
+		c.startClientBulkAcceptDispatch(bulk)
+		return
 	}
-	info := BulkAcceptInfo{
-		ID:                  bulk.ID(),
-		Range:               bulk.Range(),
-		Metadata:            bulk.Metadata(),
-		Dedicated:           bulk.Dedicated(),
-		TransportGeneration: bulk.TransportGeneration(),
-		Bulk:                bulk,
-	}
-	if err := handler(info); err != nil {
-		bulk.markReset(err)
+	if err := c.dispatchClientBulkAccept(bulk); err != nil {
 		resp.Error = err.Error()
 		replyBulkControlIfNeeded(msg, resp)
 		return
@@ -170,7 +352,12 @@ func (c *ClientCommon) handleInboundBulkOpen(msg *Message) {
 
 func (s *ServerCommon) handleInboundBulkOpen(msg *Message) {
 	req, err := decodeBulkOpenRequest(msg)
-	resp := BulkOpenResponse{BulkID: req.BulkID, DataID: req.DataID, Dedicated: req.Dedicated}
+	resp := BulkOpenResponse{
+		BulkID:          req.BulkID,
+		DataID:          req.DataID,
+		FastPathVersion: negotiateBulkFastPathVersion(req.FastPathVersion),
+		Dedicated:       req.Dedicated,
+	}
 	if err != nil {
 		resp.Error = err.Error()
 		replyBulkControlIfNeeded(msg, resp)
@@ -218,25 +405,24 @@ func (s *ServerCommon) handleInboundBulkOpen(msg *Message) {
 		replyBulkControlIfNeeded(msg, resp)
 		return
 	}
-	handler := runtime.handlerSnapshot()
-	if handler == nil {
+	s.attachServerDedicatedSidecarIfExists(logical, bulk)
+	if runtime.handlerSnapshot() == nil {
 		bulk.markReset(errBulkHandlerNotConfigured)
 		resp.Error = errBulkHandlerNotConfigured.Error()
 		replyBulkControlIfNeeded(msg, resp)
 		return
 	}
-	info := BulkAcceptInfo{
-		ID:                  bulk.ID(),
-		Range:               bulk.Range(),
-		Metadata:            bulk.Metadata(),
-		Dedicated:           bulk.Dedicated(),
-		LogicalConn:         logical,
-		TransportConn:       transport,
-		TransportGeneration: bulk.TransportGeneration(),
-		Bulk:                bulk,
+	if req.Dedicated {
+		resp.Accepted = true
+		resp.DataID = bulk.dataIDSnapshot()
+		resp.TransportGeneration = bulk.TransportGeneration()
+		replyBulkControlIfNeeded(msg, resp)
+		if bulk.dedicatedAttachedSnapshot() {
+			s.startServerBulkAcceptDispatch(bulk, logical, messageTransportConnSnapshot(msg))
+		}
+		return
 	}
-	if err := handler(info); err != nil {
-		bulk.markReset(err)
+	if err := s.dispatchServerBulkAccept(bulk, logical, transport); err != nil {
 		resp.Error = err.Error()
 		replyBulkControlIfNeeded(msg, resp)
 		return
@@ -357,6 +543,41 @@ func (c *ClientCommon) handleInboundBulkRelease(msg *Message) {
 	bulk.releaseOutboundWindow(req.Bytes, req.Chunks)
 }
 
+func (c *ClientCommon) handleInboundBulkReady(msg *Message) {
+	req, err := decodeBulkReadyRequest(msg)
+	resp := BulkReadyResponse{BulkID: req.BulkID}
+	if err != nil {
+		resp.Error = err.Error()
+		replyBulkControlIfNeeded(msg, resp)
+		return
+	}
+	runtime := c.getBulkRuntime()
+	if runtime == nil {
+		resp.Error = errBulkRuntimeNil.Error()
+		replyBulkControlIfNeeded(msg, resp)
+		return
+	}
+	bulk, ok := runtime.lookup(clientFileScope(), req.BulkID)
+	if !ok && req.DataID != 0 {
+		bulk, ok = runtime.lookupByDataID(clientFileScope(), req.DataID)
+	}
+	if !ok {
+		resp.Error = errBulkNotFound.Error()
+		replyBulkControlIfNeeded(msg, resp)
+		return
+	}
+	if resp.BulkID == "" {
+		resp.BulkID = bulk.ID()
+	}
+	readyErr := bulkReadyRemoteError(req.Error)
+	bulk.markAcceptReady(readyErr)
+	if readyErr != nil {
+		bulk.markReset(readyErr)
+	}
+	resp.Accepted = true
+	replyBulkControlIfNeeded(msg, resp)
+}
+
 func (s *ServerCommon) handleInboundBulkReset(msg *Message) {
 	req, err := decodeBulkResetRequest(msg)
 	resp := BulkResetResponse{BulkID: req.BulkID}
@@ -390,6 +611,43 @@ func (s *ServerCommon) handleInboundBulkReset(msg *Message) {
 	replyBulkControlIfNeeded(msg, resp)
 }
 
+func (s *ServerCommon) handleInboundBulkReady(msg *Message) {
+	req, err := decodeBulkReadyRequest(msg)
+	resp := BulkReadyResponse{BulkID: req.BulkID}
+	if err != nil {
+		resp.Error = err.Error()
+		replyBulkControlIfNeeded(msg, resp)
+		return
+	}
+	runtime := s.getBulkRuntime()
+	if runtime == nil {
+		resp.Error = errBulkRuntimeNil.Error()
+		replyBulkControlIfNeeded(msg, resp)
+		return
+	}
+	logical := messageLogicalConnSnapshot(msg)
+	scope := serverFileScope(logical)
+	bulk, ok := runtime.lookup(scope, req.BulkID)
+	if !ok && req.DataID != 0 {
+		bulk, ok = runtime.lookupByDataID(scope, req.DataID)
+	}
+	if !ok {
+		resp.Error = errBulkNotFound.Error()
+		replyBulkControlIfNeeded(msg, resp)
+		return
+	}
+	if resp.BulkID == "" {
+		resp.BulkID = bulk.ID()
+	}
+	readyErr := bulkReadyRemoteError(req.Error)
+	bulk.markAcceptReady(readyErr)
+	if readyErr != nil {
+		bulk.markReset(readyErr)
+	}
+	resp.Accepted = true
+	replyBulkControlIfNeeded(msg, resp)
+}
+
 func (s *ServerCommon) handleInboundBulkRelease(msg *Message) {
 	req, err := decodeBulkReleaseRequest(msg)
 	if err != nil {
@@ -625,11 +883,26 @@ func decodeBulkReleaseRequest(msg *Message) (BulkReleaseRequest, error) {
 	return req, nil
 }
 
+func decodeBulkReadyRequest(msg *Message) (BulkReadyRequest, error) {
+	var req BulkReadyRequest
+	if msg == nil {
+		return BulkReadyRequest{}, errBulkIDEmpty
+	}
+	if err := msg.Value.Orm(&req); err != nil {
+		return BulkReadyRequest{}, err
+	}
+	if req.BulkID == "" && req.DataID == 0 {
+		return BulkReadyRequest{}, errBulkIDEmpty
+	}
+	return req, nil
+}
+
 func decodeBulkOpenResponse(msg Message) (BulkOpenResponse, error) {
 	var resp BulkOpenResponse
 	if err := msg.Value.Orm(&resp); err != nil {
 		return BulkOpenResponse{}, err
 	}
+	resp.FastPathVersion = normalizeBulkFastPathVersion(resp.FastPathVersion)
 	return resp, bulkControlResultError("open", resp.Accepted, resp.Error, nil)
 }
 
@@ -649,6 +922,14 @@ func decodeBulkResetResponse(msg Message) (BulkResetResponse, error) {
 	return resp, bulkControlResultError("reset", resp.Accepted, resp.Error, nil)
 }
 
+func decodeBulkReadyResponse(msg Message) (BulkReadyResponse, error) {
+	var resp BulkReadyResponse
+	if err := msg.Value.Orm(&resp); err != nil {
+		return BulkReadyResponse{}, err
+	}
+	return resp, bulkControlResultError("ready", resp.Accepted, resp.Error, nil)
+}
+
 func bulkControlResultError(op string, accepted bool, message string, callErr error) error {
 	if callErr != nil {
 		return callErr
@@ -697,6 +978,52 @@ func bulkRemoteResetError(message string) error {
 	return errors.New(message)
 }
 
+func bulkReadyRemoteError(message string) error {
+	if message == "" {
+		return nil
+	}
+	return bulkControlMessageError(message)
+}
+
+func sendBulkReadyClient(ctx context.Context, c Client, req BulkReadyRequest) (BulkReadyResponse, error) {
+	if c == nil {
+		return BulkReadyResponse{}, errBulkClientNil
+	}
+	msg, err := c.SendObjCtx(ctx, BulkReadySignalKey, req)
+	if err != nil {
+		return BulkReadyResponse{}, err
+	}
+	return decodeBulkReadyResponse(msg)
+}
+
+func sendBulkReadyServerLogical(ctx context.Context, s Server, logical *LogicalConn, req BulkReadyRequest) (BulkReadyResponse, error) {
+	if s == nil {
+		return BulkReadyResponse{}, errBulkServerNil
+	}
+	if logical == nil {
+		return BulkReadyResponse{}, errBulkLogicalConnNil
+	}
+	msg, err := s.SendObjCtxLogical(ctx, logical, BulkReadySignalKey, req)
+	if err != nil {
+		return BulkReadyResponse{}, err
+	}
+	return decodeBulkReadyResponse(msg)
+}
+
+func sendBulkReadyServerTransport(ctx context.Context, s Server, transport *TransportConn, req BulkReadyRequest) (BulkReadyResponse, error) {
+	if s == nil {
+		return BulkReadyResponse{}, errBulkServerNil
+	}
+	if transport == nil {
+		return BulkReadyResponse{}, errBulkTransportNil
+	}
+	msg, err := s.SendObjCtxTransport(ctx, transport, BulkReadySignalKey, req)
+	if err != nil {
+		return BulkReadyResponse{}, err
+	}
+	return decodeBulkReadyResponse(msg)
+}
+
 func bulkTransportGeneration(logical *LogicalConn, transport *TransportConn) uint64 {
 	return streamTransportGeneration(logical, transport)
 }

bulk_dedicated_attach_test.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/binary"
+	"errors"
 	"net"
 	"testing"
 	"time"
@@ -136,6 +137,7 @@ func TestHandleBulkAttachSystemMessageAcceptedWritesDirectReplyBeforeDedicatedHa
 		Dedicated:   true,
 		AttachToken: "attach-token",
 	}, 0, target, nil, 0, nil, nil, nil, nil, nil)
+	bulk.markAcceptHandled()
 	if err := server.getBulkRuntime().register(serverFileScope(target), bulk); err != nil {
 		t.Fatalf("register bulk runtime failed: %v", err)
 	}
@@ -215,3 +217,251 @@ func TestHandleBulkAttachSystemMessageAcceptedWritesDirectReplyBeforeDedicatedHa
 		t.Fatalf("attach sidecar logical should be removed after handoff, got %+v", got)
 	}
 }
+
+func TestHandleBulkAttachSystemMessageDoesNotExposeSharedSidecarBeforeReplyCompletes(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	UseLegacySecurityServer(server)
+
+	sidecarLeft, sidecarRight := net.Pipe()
+	defer sidecarRight.Close()
+
+	current := server.bootstrapAcceptedLogical("dedicated-attach-current-blocked", nil, sidecarLeft)
+	if current == nil {
+		t.Fatal("bootstrapAcceptedLogical(current) should return logical")
+	}
+	target := server.bootstrapAcceptedLogical("dedicated-attach-target-blocked", nil, nil)
+	if target == nil {
+		t.Fatal("bootstrapAcceptedLogical(target) should return logical")
+	}
+
+	currentBulk := newBulkHandle(context.Background(), server.getBulkRuntime(), serverFileScope(target), BulkOpenRequest{
+		BulkID:      "server-dedicated-current",
+		DataID:      17,
+		Dedicated:   true,
+		AttachToken: "attach-token",
+	}, 0, target, nil, 0, nil, nil, nil, nil, nil)
+	currentBulk.markAcceptHandled()
+	if err := server.getBulkRuntime().register(serverFileScope(target), currentBulk); err != nil {
+		t.Fatalf("register current bulk runtime failed: %v", err)
+	}
+
+	pendingBulk := newBulkHandle(context.Background(), server.getBulkRuntime(), serverFileScope(target), BulkOpenRequest{
+		BulkID:      "server-dedicated-pending",
+		DataID:      18,
+		Dedicated:   true,
+		AttachToken: "attach-token-2",
+	}, 0, target, nil, 0, nil, nil, nil, nil, nil)
+	pendingBulk.markAcceptHandled()
+	if err := server.getBulkRuntime().register(serverFileScope(target), pendingBulk); err != nil {
+		t.Fatalf("register pending bulk runtime failed: %v", err)
+	}
+
+	reqPayload, err := server.sequenceEn(bulkAttachRequest{
+		PeerID:      target.ID(),
+		BulkID:      currentBulk.ID(),
+		AttachToken: "attach-token",
+	})
+	if err != nil {
+		t.Fatalf("encode bulkAttachRequest failed: %v", err)
+	}
+	msg := Message{
+		NetType:     NET_SERVER,
+		LogicalConn: current,
+		ClientConn:  current.compatClientConn(),
+		TransferMsg: TransferMsg{
+			ID:    77,
+			Key:   systemBulkAttachKey,
+			Value: reqPayload,
+			Type:  MSG_SYS_WAIT,
+		},
+		inboundConn: sidecarLeft,
+		Time:        time.Now(),
+	}
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		_ = server.handleBulkAttachSystemMessage(msg)
+	}()
+
+	time.Sleep(50 * time.Millisecond)
+
+	if got := pendingBulk.dedicatedConnSnapshot(); got != nil {
+		t.Fatal("pending dedicated bulk should not observe shared sidecar before attach reply is fully sent")
+	}
+	if got := server.serverDedicatedSidecarSnapshot(target); got != nil {
+		t.Fatal("server dedicated sidecar should not be published before attach reply is fully sent")
+	}
+
+	type attachReplyResult struct {
+		transfer TransferMsg
+		resp     bulkAttachResponse
+		err      error
+	}
+	replyCh := make(chan attachReplyResult, 1)
+	go func() {
+		_ = sidecarRight.SetReadDeadline(time.Now().Add(time.Second))
+		replyPayload, err := readDirectSignalFramePayload(sidecarRight)
+		if err != nil {
+			replyCh <- attachReplyResult{err: err}
+			return
+		}
+		transfer, err := decodeDirectSignalPayload(server.sequenceDe, current.msgDeSnapshot(), current.secretKeySnapshot(), replyPayload)
+		if err != nil {
+			replyCh <- attachReplyResult{err: err}
+			return
+		}
+		resp, err := decodeBulkAttachResponse(server.sequenceDe, transfer.Value)
+		replyCh <- attachReplyResult{transfer: transfer, resp: resp, err: err}
+	}()
+
+	select {
+	case result := <-replyCh:
+		if result.err != nil {
+			t.Fatalf("read direct attach reply failed: %v", result.err)
+		}
+		if !result.resp.Accepted {
+			t.Fatalf("bulk attach response = %+v, want accepted", result.resp)
+		}
+	case <-time.After(2 * time.Second):
+		t.Fatal("timed out waiting for direct attach reply")
+	}
+
+	select {
+	case <-done:
+	case <-time.After(2 * time.Second):
+		t.Fatal("timed out waiting for handleBulkAttachSystemMessage to finish")
+	}
+
+	if got := pendingBulk.dedicatedConnSnapshot(); got != sidecarLeft {
+		t.Fatalf("pending dedicated bulk conn mismatch after reply: got %v want %v", got, sidecarLeft)
+	}
+	if got := server.serverDedicatedSidecarSnapshot(target); got == nil || got.conn != sidecarLeft {
+		t.Fatal("server dedicated sidecar should be published after attach reply completes")
+	}
+}
+
+func TestBulkAttachResponseErrorCarriesStructuredFields(t *testing.T) {
+	resp := toBulkAttachResponseError(&bulkAttachError{
+		Code:       bulkAttachErrorCodeTokenMismatch,
+		Retryable:  false,
+		Message:    "bulk attach token mismatch",
+		FailedSeq:  7,
+		FailedBulk: "bulk-1",
+	}, "fallback-bulk")
+
+	if resp.Accepted {
+		t.Fatalf("Accepted = %v, want false", resp.Accepted)
+	}
+	if got, want := resp.Code, string(bulkAttachErrorCodeTokenMismatch); got != want {
+		t.Fatalf("Code = %q, want %q", got, want)
+	}
+	if got, want := resp.Retryable, false; got != want {
+		t.Fatalf("Retryable = %v, want %v", got, want)
+	}
+	if got, want := resp.FailedSeq, uint64(7); got != want {
+		t.Fatalf("FailedSeq = %d, want %d", got, want)
+	}
+	if got, want := resp.FailedBulk, "bulk-1"; got != want {
+		t.Fatalf("FailedBulk = %q, want %q", got, want)
+	}
+}
+
+func TestResolveInboundDedicatedBulkRejectsAlreadyAttachedAfterDataStarted(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	UseLegacySecurityServer(server)
+
+	currentLeft, currentRight := net.Pipe()
+	defer currentRight.Close()
+	current := server.bootstrapAcceptedLogical("dedicated-attach-current", nil, currentLeft)
+	if current == nil {
+		t.Fatal("bootstrapAcceptedLogical(current) should return logical")
+	}
+	target := server.bootstrapAcceptedLogical("dedicated-attach-target", nil, nil)
+	if target == nil {
+		t.Fatal("bootstrapAcceptedLogical(target) should return logical")
+	}
+
+	bulk := newBulkHandle(context.Background(), server.getBulkRuntime(), serverFileScope(target), BulkOpenRequest{
+		BulkID:      "server-dedicated-attach-test",
+		DataID:      7,
+		Dedicated:   true,
+		AttachToken: "attach-token",
+	}, 0, target, nil, 0, nil, nil, nil, nil, nil)
+	bulk.markAcceptHandled()
+	if err := server.getBulkRuntime().register(serverFileScope(target), bulk); err != nil {
+		t.Fatalf("register bulk runtime failed: %v", err)
+	}
+	attachedLeft, attachedRight := net.Pipe()
+	defer attachedRight.Close()
+	if err := bulk.attachDedicatedConn(attachedLeft); err != nil {
+		t.Fatalf("attachDedicatedConn failed: %v", err)
+	}
+	bulk.markDedicatedDataStarted()
+
+	_, _, err := server.resolveInboundDedicatedBulk(current, bulkAttachRequest{
+		PeerID:      target.ID(),
+		BulkID:      bulk.ID(),
+		AttachToken: "attach-token",
+	})
+	if err == nil {
+		t.Fatal("resolveInboundDedicatedBulk should reject duplicate attach after data started")
+	}
+	var attachErr *bulkAttachError
+	if !errors.As(err, &attachErr) || attachErr == nil {
+		t.Fatalf("resolveInboundDedicatedBulk error type = %T, want *bulkAttachError", err)
+	}
+	if got, want := attachErr.Code, bulkAttachErrorCodeAlreadyAttached; got != want {
+		t.Fatalf("attach error code = %q, want %q", got, want)
+	}
+	if attachErr.Retryable {
+		t.Fatalf("attach error retryable = true, want false")
+	}
+}
+
+func TestResolveInboundDedicatedBulkAllowsReattachBeforeDataStarts(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	UseLegacySecurityServer(server)
+
+	currentLeft, currentRight := net.Pipe()
+	defer currentRight.Close()
+	current := server.bootstrapAcceptedLogical("dedicated-attach-current", nil, currentLeft)
+	if current == nil {
+		t.Fatal("bootstrapAcceptedLogical(current) should return logical")
+	}
+	target := server.bootstrapAcceptedLogical("dedicated-attach-target", nil, nil)
+	if target == nil {
+		t.Fatal("bootstrapAcceptedLogical(target) should return logical")
+	}
+
+	bulk := newBulkHandle(context.Background(), server.getBulkRuntime(), serverFileScope(target), BulkOpenRequest{
+		BulkID:      "server-dedicated-attach-test",
+		DataID:      7,
+		Dedicated:   true,
+		AttachToken: "attach-token",
+	}, 0, target, nil, 0, nil, nil, nil, nil, nil)
+	bulk.markAcceptHandled()
+	if err := server.getBulkRuntime().register(serverFileScope(target), bulk); err != nil {
+		t.Fatalf("register bulk runtime failed: %v", err)
+	}
+	attachedLeft, attachedRight := net.Pipe()
+	defer attachedRight.Close()
+	if err := bulk.attachDedicatedConn(attachedLeft); err != nil {
+		t.Fatalf("attachDedicatedConn failed: %v", err)
+	}
+
+	resolvedLogical, resolvedBulk, err := server.resolveInboundDedicatedBulk(current, bulkAttachRequest{
+		PeerID:      target.ID(),
+		BulkID:      bulk.ID(),
+		AttachToken: "attach-token",
+	})
+	if err != nil {
+		t.Fatalf("resolveInboundDedicatedBulk failed: %v", err)
+	}
+	if resolvedLogical != target {
+		t.Fatalf("resolved logical mismatch: got %v want %v", resolvedLogical, target)
+	}
+	if resolvedBulk != bulk {
+		t.Fatalf("resolved bulk mismatch: got %v want %v", resolvedBulk, bulk)
+	}
+}

bulk_dedicated_batch.go

@@ -12,14 +12,18 @@ import (
 )
 
 const (
-	bulkDedicatedBatchMagic         = "NBD2"
-	bulkDedicatedBatchVersion       = 1
-	bulkDedicatedBatchHeaderLen     = 20
-	bulkDedicatedBatchItemHeaderLen = 16
-	bulkDedicatedBatchMaxItems      = 32
-	bulkDedicatedBatchMaxPlainBytes = 8 * 1024 * 1024
-	bulkDedicatedSendQueueSize      = bulkDedicatedBatchMaxItems
-	bulkDedicatedReleasePayloadLen  = 12
+	bulkDedicatedBatchMagic               = "NBD2"
+	bulkDedicatedBatchVersion             = 1
+	bulkDedicatedBatchHeaderLen           = 20
+	bulkDedicatedBatchItemHeaderLen       = 16
+	bulkDedicatedSuperBatchMagic          = "NBD3"
+	bulkDedicatedSuperBatchVersion        = 1
+	bulkDedicatedSuperBatchHeaderLen      = 12
+	bulkDedicatedSuperBatchGroupHeaderLen = 12
+	bulkDedicatedBatchMaxItems            = 64
+	bulkDedicatedBatchMaxPlainBytes       = 16 * 1024 * 1024
+	bulkDedicatedSendQueueSize            = bulkDedicatedBatchMaxItems
+	bulkDedicatedReleasePayloadLen        = 12
 )
 
 const (
@@ -46,6 +50,16 @@ type bulkDedicatedSendRequest struct {
 	Payload []byte
 }
 
+type bulkDedicatedOutboundBatch struct {
+	DataID uint64
+	Items  []bulkDedicatedSendRequest
+}
+
+type bulkDedicatedInboundBatch struct {
+	DataID uint64
+	Items  []bulkDedicatedBatchItem
+}
+
 type bulkDedicatedBatchRequest struct {
 	Ctx      context.Context
 	Items    []bulkDedicatedSendRequest
@@ -165,7 +179,7 @@ func (s *bulkDedicatedSender) submitWriteBatch(ctx context.Context, items []bulk
 	if submitted, err := s.tryDirectSubmitBatch(ctx, items); submitted {
 		return err
 	}
-	queuedItems := cloneBulkDedicatedSendRequests(items)
+	queuedItems := copyBulkDedicatedSendRequests(items)
 	return s.submitBatch(ctx, queuedItems, true)
 }
 
@@ -471,6 +485,23 @@ func cloneBulkDedicatedSendRequests(items []bulkDedicatedSendRequest) []bulkDedi
 	return cloned
 }
 
+func copyBulkDedicatedSendRequests(items []bulkDedicatedSendRequest) []bulkDedicatedSendRequest {
+	if len(items) == 0 {
+		return nil
+	}
+	copied := make([]bulkDedicatedSendRequest, len(items))
+	copy(copied, items)
+	return copied
+}
+
+func bulkDedicatedSendRequestsLen(items []bulkDedicatedSendRequest) int {
+	total := 0
+	for _, item := range items {
+		total += bulkDedicatedSendRequestLen(item)
+	}
+	return total
+}
+
 func bulkDedicatedSendRequestLen(req bulkDedicatedSendRequest) int {
 	return bulkDedicatedSendRequestLenFromPayloadLen(len(req.Payload))
 }
@@ -479,6 +510,83 @@ func bulkDedicatedSendRequestLenFromPayloadLen(payloadLen int) int {
 	return bulkDedicatedBatchItemHeaderLen + payloadLen
 }
 
+func bulkDedicatedBatchesPlainLen(batches []bulkDedicatedOutboundBatch) int {
+	switch len(batches) {
+	case 0:
+		return 0
+	case 1:
+		return bulkDedicatedBatchPlainLen(batches[0].Items)
+	default:
+		total := bulkDedicatedSuperBatchHeaderLen
+		for _, batch := range batches {
+			total += bulkDedicatedSuperBatchGroupHeaderLen + bulkDedicatedSendRequestsLen(batch.Items)
+		}
+		return total
+	}
+}
+
+func encodeBulkDedicatedBatchesPlain(batches []bulkDedicatedOutboundBatch) ([]byte, error) {
+	if len(batches) == 0 {
+		return nil, errBulkFastPayloadInvalid
+	}
+	total := bulkDedicatedBatchesPlainLen(batches)
+	buf := make([]byte, total)
+	if err := writeBulkDedicatedBatchesPlain(buf, batches); err != nil {
+		return nil, err
+	}
+	return buf, nil
+}
+
+func writeBulkDedicatedBatchesPlain(buf []byte, batches []bulkDedicatedOutboundBatch) error {
+	switch len(batches) {
+	case 0:
+		return errBulkFastPayloadInvalid
+	case 1:
+		return writeBulkDedicatedBatchPlain(buf, batches[0].DataID, batches[0].Items)
+	default:
+		return writeBulkDedicatedSuperBatchPlain(buf, batches)
+	}
+}
+
+func writeBulkDedicatedSuperBatchPlain(buf []byte, batches []bulkDedicatedOutboundBatch) error {
+	if len(batches) <= 1 {
+		return errBulkFastPayloadInvalid
+	}
+	if len(buf) != bulkDedicatedBatchesPlainLen(batches) || len(buf) > bulkDedicatedBatchMaxPlainBytes {
+		return errBulkFastPayloadInvalid
+	}
+	copy(buf[:4], bulkDedicatedSuperBatchMagic)
+	buf[4] = bulkDedicatedSuperBatchVersion
+	binary.BigEndian.PutUint32(buf[8:12], uint32(len(batches)))
+	offset := bulkDedicatedSuperBatchHeaderLen
+	totalItems := 0
+	for _, batch := range batches {
+		if batch.DataID == 0 || len(batch.Items) == 0 {
+			return errBulkFastPayloadInvalid
+		}
+		totalItems += len(batch.Items)
+		if totalItems > bulkDedicatedBatchMaxItems {
+			return errBulkFastPayloadInvalid
+		}
+		binary.BigEndian.PutUint64(buf[offset:offset+8], batch.DataID)
+		binary.BigEndian.PutUint32(buf[offset+8:offset+12], uint32(len(batch.Items)))
+		offset += bulkDedicatedSuperBatchGroupHeaderLen
+		for _, item := range batch.Items {
+			buf[offset] = item.Type
+			buf[offset+1] = item.Flags
+			binary.BigEndian.PutUint64(buf[offset+4:offset+12], item.Seq)
+			binary.BigEndian.PutUint32(buf[offset+12:offset+16], uint32(len(item.Payload)))
+			offset += bulkDedicatedBatchItemHeaderLen
+			copy(buf[offset:offset+len(item.Payload)], item.Payload)
+			offset += len(item.Payload)
+		}
+	}
+	if offset != len(buf) {
+		return errBulkFastPayloadInvalid
+	}
+	return nil
+}
+
 func encodeBulkDedicatedReleasePayload(bytes int64, chunks int) ([]byte, error) {
 	if bytes <= 0 && chunks <= 0 {
 		return nil, errBulkFastPayloadInvalid
@@ -505,24 +613,26 @@ func decodeBulkDedicatedReleasePayload(payload []byte) (int64, int, error) {
 }
 
 func encodeBulkDedicatedBatchPlain(dataID uint64, items []bulkDedicatedSendRequest) ([]byte, error) {
-	if dataID == 0 || len(items) == 0 {
-		return nil, errBulkFastPayloadInvalid
-	}
-	total := bulkDedicatedBatchPlainLen(items)
-	buf := make([]byte, total)
-	if err := writeBulkDedicatedBatchPlain(buf, dataID, items); err != nil {
-		return nil, err
-	}
-	return buf, nil
+	return encodeBulkDedicatedBatchesPlain([]bulkDedicatedOutboundBatch{{
+		DataID: dataID,
+		Items:  items,
+	}})
 }
 
 func encodeBulkDedicatedBatchPayloadFast(encode transportFastPlainEncoder, secretKey []byte, dataID uint64, items []bulkDedicatedSendRequest) ([]byte, error) {
+	return encodeBulkDedicatedBatchesPayloadFast(encode, secretKey, []bulkDedicatedOutboundBatch{{
+		DataID: dataID,
+		Items:  items,
+	}})
+}
+
+func encodeBulkDedicatedBatchesPayloadFast(encode transportFastPlainEncoder, secretKey []byte, batches []bulkDedicatedOutboundBatch) ([]byte, error) {
 	if encode == nil {
 		return nil, errTransportPayloadEncryptFailed
 	}
-	plainLen := bulkDedicatedBatchPlainLen(items)
+	plainLen := bulkDedicatedBatchesPlainLen(batches)
 	return encode(secretKey, plainLen, func(dst []byte) error {
-		return writeBulkDedicatedBatchPlain(dst, dataID, items)
+		return writeBulkDedicatedBatchesPlain(dst, batches)
 	})
 }
 
@@ -606,29 +716,268 @@ func decodeBulkDedicatedBatchPlain(payload []byte) (uint64, []bulkDedicatedBatch
 	return dataID, items, true, nil
 }
 
-func decodeDedicatedBulkInboundItems(expectedDataID uint64, plain []byte) ([]bulkDedicatedBatchItem, error) {
+func decodeBulkDedicatedSuperBatchPlain(payload []byte) ([]bulkDedicatedInboundBatch, bool, error) {
+	if len(payload) < 4 || string(payload[:4]) != bulkDedicatedSuperBatchMagic {
+		return nil, false, nil
+	}
+	if len(payload) < bulkDedicatedSuperBatchHeaderLen {
+		return nil, true, errBulkFastPayloadInvalid
+	}
+	if payload[4] != bulkDedicatedSuperBatchVersion {
+		return nil, true, errBulkFastPayloadInvalid
+	}
+	groupCount := int(binary.BigEndian.Uint32(payload[8:12]))
+	if groupCount <= 0 {
+		return nil, true, errBulkFastPayloadInvalid
+	}
+	batches := make([]bulkDedicatedInboundBatch, 0, groupCount)
+	offset := bulkDedicatedSuperBatchHeaderLen
+	totalItems := 0
+	for i := 0; i < groupCount; i++ {
+		if len(payload)-offset < bulkDedicatedSuperBatchGroupHeaderLen {
+			return nil, true, errBulkFastPayloadInvalid
+		}
+		dataID := binary.BigEndian.Uint64(payload[offset : offset+8])
+		count := int(binary.BigEndian.Uint32(payload[offset+8 : offset+12]))
+		offset += bulkDedicatedSuperBatchGroupHeaderLen
+		if dataID == 0 || count <= 0 {
+			return nil, true, errBulkFastPayloadInvalid
+		}
+		totalItems += count
+		if totalItems > bulkDedicatedBatchMaxItems {
+			return nil, true, errBulkFastPayloadInvalid
+		}
+		items := make([]bulkDedicatedBatchItem, 0, count)
+		for j := 0; j < count; j++ {
+			if len(payload)-offset < bulkDedicatedBatchItemHeaderLen {
+				return nil, true, errBulkFastPayloadInvalid
+			}
+			itemType := payload[offset]
+			switch itemType {
+			case bulkFastPayloadTypeData, bulkFastPayloadTypeClose, bulkFastPayloadTypeReset, bulkFastPayloadTypeRelease:
+			default:
+				return nil, true, errBulkFastPayloadInvalid
+			}
+			flags := payload[offset+1]
+			seq := binary.BigEndian.Uint64(payload[offset+4 : offset+12])
+			dataLen := int(binary.BigEndian.Uint32(payload[offset+12 : offset+16]))
+			offset += bulkDedicatedBatchItemHeaderLen
+			if dataLen < 0 || len(payload)-offset < dataLen {
+				return nil, true, errBulkFastPayloadInvalid
+			}
+			items = append(items, bulkDedicatedBatchItem{
+				Type:    itemType,
+				Flags:   flags,
+				Seq:     seq,
+				Payload: payload[offset : offset+dataLen],
+			})
+			offset += dataLen
+		}
+		batches = append(batches, bulkDedicatedInboundBatch{
+			DataID: dataID,
+			Items:  items,
+		})
+	}
+	if offset != len(payload) {
+		return nil, true, errBulkFastPayloadInvalid
+	}
+	return batches, true, nil
+}
+
+func decodeDedicatedBulkInboundBatches(plain []byte) ([]bulkDedicatedInboundBatch, error) {
+	if batches, matched, err := decodeBulkDedicatedSuperBatchPlain(plain); matched {
+		if err != nil {
+			return nil, err
+		}
+		return batches, nil
+	}
 	if dataID, items, matched, err := decodeBulkDedicatedBatchPlain(plain); matched {
 		if err != nil {
 			return nil, err
 		}
-		if expectedDataID == 0 || dataID != expectedDataID {
-			return nil, errBulkFastPayloadInvalid
-		}
-		return items, nil
+		return []bulkDedicatedInboundBatch{{
+			DataID: dataID,
+			Items:  items,
+		}}, nil
 	}
 	frame, matched, err := decodeBulkFastFrame(plain)
 	if err != nil {
 		return nil, err
 	}
-	if !matched || expectedDataID == 0 || frame.DataID != expectedDataID {
+	if !matched || frame.DataID == 0 {
 		return nil, errBulkFastPayloadInvalid
 	}
-	return []bulkDedicatedBatchItem{{
+	return []bulkDedicatedInboundBatch{{
+		DataID: frame.DataID,
+		Items: []bulkDedicatedBatchItem{{
+			Type:    frame.Type,
+			Flags:   frame.Flags,
+			Seq:     frame.Seq,
+			Payload: frame.Payload,
+		}},
+	}}, nil
+}
+
+func decodeDedicatedBulkInboundPayload(plain []byte) (uint64, []bulkDedicatedBatchItem, error) {
+	batches, err := decodeDedicatedBulkInboundBatches(plain)
+	if err != nil {
+		return 0, nil, err
+	}
+	if len(batches) != 1 {
+		return 0, nil, errBulkFastPayloadInvalid
+	}
+	return batches[0].DataID, batches[0].Items, nil
+}
+
+func decodeDedicatedBulkInboundItems(expectedDataID uint64, plain []byte) ([]bulkDedicatedBatchItem, error) {
+	dataID, items, err := decodeDedicatedBulkInboundPayload(plain)
+	if err != nil {
+		return nil, err
+	}
+	if expectedDataID == 0 || dataID != expectedDataID {
+		return nil, errBulkFastPayloadInvalid
+	}
+	return items, nil
+}
+
+func hasBulkDedicatedMagic(payload []byte, magic string) bool {
+	return len(payload) >= 4 &&
+		payload[0] == magic[0] &&
+		payload[1] == magic[1] &&
+		payload[2] == magic[2] &&
+		payload[3] == magic[3]
+}
+
+func walkDedicatedBulkInboundPayload(plain []byte, visit func(dataID uint64, item bulkDedicatedBatchItem) error) error {
+	if visit == nil {
+		return errBulkFastPayloadInvalid
+	}
+	if hasBulkDedicatedMagic(plain, bulkDedicatedSuperBatchMagic) {
+		return walkDedicatedBulkInboundSuperBatchPlain(plain, visit)
+	}
+	if hasBulkDedicatedMagic(plain, bulkDedicatedBatchMagic) {
+		return walkDedicatedBulkInboundBatchPlain(plain, visit)
+	}
+	frame, matched, err := decodeBulkFastFrame(plain)
+	if err != nil {
+		return err
+	}
+	if !matched || frame.DataID == 0 {
+		return errBulkFastPayloadInvalid
+	}
+	return visit(frame.DataID, bulkDedicatedBatchItem{
 		Type:    frame.Type,
 		Flags:   frame.Flags,
 		Seq:     frame.Seq,
 		Payload: frame.Payload,
-	}}, nil
+	})
+}
+
+func walkDedicatedBulkInboundBatchPlain(payload []byte, visit func(dataID uint64, item bulkDedicatedBatchItem) error) error {
+	if len(payload) < bulkDedicatedBatchHeaderLen {
+		return errBulkFastPayloadInvalid
+	}
+	if payload[4] != bulkDedicatedBatchVersion {
+		return errBulkFastPayloadInvalid
+	}
+	dataID := binary.BigEndian.Uint64(payload[8:16])
+	count := int(binary.BigEndian.Uint32(payload[16:20]))
+	if dataID == 0 || count <= 0 {
+		return errBulkFastPayloadInvalid
+	}
+	offset := bulkDedicatedBatchHeaderLen
+	for i := 0; i < count; i++ {
+		if len(payload)-offset < bulkDedicatedBatchItemHeaderLen {
+			return errBulkFastPayloadInvalid
+		}
+		itemType := payload[offset]
+		switch itemType {
+		case bulkFastPayloadTypeData, bulkFastPayloadTypeClose, bulkFastPayloadTypeReset, bulkFastPayloadTypeRelease:
+		default:
+			return errBulkFastPayloadInvalid
+		}
+		flags := payload[offset+1]
+		seq := binary.BigEndian.Uint64(payload[offset+4 : offset+12])
+		dataLen := int(binary.BigEndian.Uint32(payload[offset+12 : offset+16]))
+		offset += bulkDedicatedBatchItemHeaderLen
+		if dataLen < 0 || len(payload)-offset < dataLen {
+			return errBulkFastPayloadInvalid
+		}
+		if err := visit(dataID, bulkDedicatedBatchItem{
+			Type:    itemType,
+			Flags:   flags,
+			Seq:     seq,
+			Payload: payload[offset : offset+dataLen],
+		}); err != nil {
+			return err
+		}
+		offset += dataLen
+	}
+	if offset != len(payload) {
+		return errBulkFastPayloadInvalid
+	}
+	return nil
+}
+
+func walkDedicatedBulkInboundSuperBatchPlain(payload []byte, visit func(dataID uint64, item bulkDedicatedBatchItem) error) error {
+	if len(payload) < bulkDedicatedSuperBatchHeaderLen {
+		return errBulkFastPayloadInvalid
+	}
+	if payload[4] != bulkDedicatedSuperBatchVersion {
+		return errBulkFastPayloadInvalid
+	}
+	groupCount := int(binary.BigEndian.Uint32(payload[8:12]))
+	if groupCount <= 0 {
+		return errBulkFastPayloadInvalid
+	}
+	offset := bulkDedicatedSuperBatchHeaderLen
+	totalItems := 0
+	for i := 0; i < groupCount; i++ {
+		if len(payload)-offset < bulkDedicatedSuperBatchGroupHeaderLen {
+			return errBulkFastPayloadInvalid
+		}
+		dataID := binary.BigEndian.Uint64(payload[offset : offset+8])
+		count := int(binary.BigEndian.Uint32(payload[offset+8 : offset+12]))
+		offset += bulkDedicatedSuperBatchGroupHeaderLen
+		if dataID == 0 || count <= 0 {
+			return errBulkFastPayloadInvalid
+		}
+		totalItems += count
+		if totalItems > bulkDedicatedBatchMaxItems {
+			return errBulkFastPayloadInvalid
+		}
+		for j := 0; j < count; j++ {
+			if len(payload)-offset < bulkDedicatedBatchItemHeaderLen {
+				return errBulkFastPayloadInvalid
+			}
+			itemType := payload[offset]
+			switch itemType {
+			case bulkFastPayloadTypeData, bulkFastPayloadTypeClose, bulkFastPayloadTypeReset, bulkFastPayloadTypeRelease:
+			default:
+				return errBulkFastPayloadInvalid
+			}
+			flags := payload[offset+1]
+			seq := binary.BigEndian.Uint64(payload[offset+4 : offset+12])
+			dataLen := int(binary.BigEndian.Uint32(payload[offset+12 : offset+16]))
+			offset += bulkDedicatedBatchItemHeaderLen
+			if dataLen < 0 || len(payload)-offset < dataLen {
+				return errBulkFastPayloadInvalid
+			}
+			if err := visit(dataID, bulkDedicatedBatchItem{
+				Type:    itemType,
+				Flags:   flags,
+				Seq:     seq,
+				Payload: payload[offset : offset+dataLen],
+			}); err != nil {
+				return err
+			}
+			offset += dataLen
+		}
+	}
+	if offset != len(payload) {
+		return errBulkFastPayloadInvalid
+	}
+	return nil
 }
 
 func normalizeDedicatedBulkSendError(err error) error {
@@ -643,13 +992,23 @@ func normalizeDedicatedBulkSendError(err error) error {
 }
 
 func dispatchDedicatedBulkInboundItem(bulk *bulkHandle, item bulkDedicatedBatchItem) error {
+	return dispatchDedicatedBulkInboundItemWithRelease(bulk, item, nil)
+}
+
+func dispatchDedicatedBulkInboundItemWithRelease(bulk *bulkHandle, item bulkDedicatedBatchItem, release func()) error {
 	if bulk == nil {
+		if release != nil {
+			release()
+		}
 		return io.ErrClosedPipe
 	}
 	switch item.Type {
 	case bulkFastPayloadTypeData:
-		return bulk.pushOwnedChunkNoReset(item.Payload)
+		return bulk.pushOwnedChunkWithReleaseNoReset(item.Payload, release)
 	case bulkFastPayloadTypeClose:
+		if release != nil {
+			release()
+		}
 		if item.Flags&bulkFastPayloadFlagFullClose != 0 {
 			bulk.markPeerClosed()
 			return nil
@@ -657,6 +1016,9 @@ func dispatchDedicatedBulkInboundItem(bulk *bulkHandle, item bulkDedicatedBatchI
 		bulk.markRemoteClosed()
 		return nil
 	case bulkFastPayloadTypeReset:
+		if release != nil {
+			release()
+		}
 		resetErr := errBulkReset
 		if len(item.Payload) > 0 {
 			resetErr = bulkRemoteResetError(string(item.Payload))
@@ -664,6 +1026,9 @@ func dispatchDedicatedBulkInboundItem(bulk *bulkHandle, item bulkDedicatedBatchI
 		bulk.markReset(bulkResetError(resetErr))
 		return nil
 	case bulkFastPayloadTypeRelease:
+		if release != nil {
+			release()
+		}
 		bytes, chunks, err := decodeBulkDedicatedReleasePayload(item.Payload)
 		if err != nil {
 			return err
@@ -671,6 +1036,9 @@ func dispatchDedicatedBulkInboundItem(bulk *bulkHandle, item bulkDedicatedBatchI
 		bulk.releaseOutboundWindow(bytes, chunks)
 		return nil
 	default:
+		if release != nil {
+			release()
+		}
 		return errBulkFastPayloadInvalid
 	}
 }

bulk_dedicated_lane_sender.go

@@ -0,0 +1,639 @@
+package notify
+
+import (
+	"context"
+	"net"
+	"sync"
+	"sync/atomic"
+	"time"
+)
+
+const bulkDedicatedLaneMicroBatchWait = 50 * time.Microsecond
+
+type bulkDedicatedLaneBatchRequest struct {
+	Ctx      context.Context
+	DataID   uint64
+	Items    []bulkDedicatedSendRequest
+	Deadline time.Time
+	wait     bool
+	resultCh chan error
+	state    bulkDedicatedRequestState
+	aborted  atomic.Bool
+}
+
+var bulkDedicatedLaneBatchRequestPool sync.Pool
+
+type bulkDedicatedLaneSender struct {
+	conn   net.Conn
+	encode func([]bulkDedicatedOutboundBatch) ([]byte, func(), error)
+	fail   func(error)
+
+	reqCh    chan *bulkDedicatedLaneBatchRequest
+	stopCh   chan struct{}
+	doneCh   chan struct{}
+	stopOnce sync.Once
+	flushMu  sync.Mutex
+	queued   atomic.Int64
+
+	errMu sync.Mutex
+	err   error
+}
+
+func newBulkDedicatedLaneSender(conn net.Conn, encode func([]bulkDedicatedOutboundBatch) ([]byte, func(), error), fail func(error)) *bulkDedicatedLaneSender {
+	sender := &bulkDedicatedLaneSender{
+		conn:   conn,
+		encode: encode,
+		fail:   fail,
+		reqCh:  make(chan *bulkDedicatedLaneBatchRequest, bulkDedicatedSendQueueSize),
+		stopCh: make(chan struct{}),
+		doneCh: make(chan struct{}),
+	}
+	go sender.run()
+	return sender
+}
+
+func getBulkDedicatedLaneBatchRequest() *bulkDedicatedLaneBatchRequest {
+	if pooled, ok := bulkDedicatedLaneBatchRequestPool.Get().(*bulkDedicatedLaneBatchRequest); ok && pooled != nil {
+		pooled.reset()
+		return pooled
+	}
+	req := &bulkDedicatedLaneBatchRequest{
+		resultCh: make(chan error, 1),
+	}
+	req.reset()
+	return req
+}
+
+func (r *bulkDedicatedLaneBatchRequest) reset() {
+	if r == nil {
+		return
+	}
+	r.Ctx = nil
+	r.DataID = 0
+	r.Deadline = time.Time{}
+	r.wait = false
+	r.Items = r.Items[:0]
+	r.state.value.Store(bulkDedicatedRequestQueued)
+	r.aborted.Store(false)
+	if r.resultCh != nil {
+		select {
+		case <-r.resultCh:
+		default:
+		}
+	}
+}
+
+func (r *bulkDedicatedLaneBatchRequest) prepare(ctx context.Context, dataID uint64, items []bulkDedicatedSendRequest, wait bool) {
+	if r == nil {
+		return
+	}
+	r.reset()
+	r.Ctx = ctx
+	r.DataID = dataID
+	r.wait = wait
+	if deadline, ok := ctx.Deadline(); ok {
+		r.Deadline = deadline
+	}
+	if cap(r.Items) < len(items) {
+		r.Items = make([]bulkDedicatedSendRequest, len(items))
+	} else {
+		r.Items = r.Items[:len(items)]
+	}
+	copy(r.Items, items)
+}
+
+func (r *bulkDedicatedLaneBatchRequest) recycle() {
+	if r == nil {
+		return
+	}
+	r.reset()
+	bulkDedicatedLaneBatchRequestPool.Put(r)
+}
+
+func (s *bulkDedicatedLaneSender) submitData(ctx context.Context, dataID uint64, seq uint64, payload []byte) error {
+	if s == nil {
+		return errTransportDetached
+	}
+	items := []bulkDedicatedSendRequest{{
+		Type:    bulkFastPayloadTypeData,
+		Seq:     seq,
+		Payload: append([]byte(nil), payload...),
+	}}
+	return s.submitBatch(ctx, dataID, items, false)
+}
+
+func (s *bulkDedicatedLaneSender) submitWrite(ctx context.Context, dataID uint64, startSeq uint64, payload []byte, chunkSize int) (int, error) {
+	if s == nil {
+		return 0, errTransportDetached
+	}
+	if len(payload) == 0 {
+		return 0, nil
+	}
+	if chunkSize <= 0 {
+		chunkSize = defaultBulkChunkSize
+	}
+	written := 0
+	seq := startSeq
+	for written < len(payload) {
+		var itemBuf [bulkDedicatedBatchMaxItems]bulkDedicatedSendRequest
+		items := itemBuf[:0]
+		batchBytes := bulkDedicatedBatchHeaderLen
+		start := written
+		for written < len(payload) && len(items) < bulkDedicatedBatchMaxItems {
+			end := written + chunkSize
+			if end > len(payload) {
+				end = len(payload)
+			}
+			itemLen := bulkDedicatedSendRequestLenFromPayloadLen(end - written)
+			if len(items) > 0 && batchBytes+itemLen > bulkDedicatedBatchMaxPlainBytes {
+				break
+			}
+			items = append(items, bulkDedicatedSendRequest{
+				Type:    bulkFastPayloadTypeData,
+				Seq:     seq,
+				Payload: payload[written:end],
+			})
+			batchBytes += itemLen
+			seq++
+			written = end
+		}
+		if len(items) == 0 {
+			end := written + chunkSize
+			if end > len(payload) {
+				end = len(payload)
+			}
+			items = append(items, bulkDedicatedSendRequest{
+				Type:    bulkFastPayloadTypeData,
+				Seq:     seq,
+				Payload: payload[written:end],
+			})
+			seq++
+			written = end
+		}
+		if err := s.submitWriteBatch(ctx, dataID, items); err != nil {
+			return start, err
+		}
+		start = written
+	}
+	return written, nil
+}
+
+func (s *bulkDedicatedLaneSender) submitWriteBatch(ctx context.Context, dataID uint64, items []bulkDedicatedSendRequest) error {
+	if s == nil {
+		return errTransportDetached
+	}
+	if len(items) == 0 {
+		return nil
+	}
+	if submitted, err := s.tryDirectSubmitBatch(ctx, dataID, items); submitted {
+		return err
+	}
+	queuedItems := copyBulkDedicatedSendRequests(items)
+	return s.submitBatch(ctx, dataID, queuedItems, true)
+}
+
+func (s *bulkDedicatedLaneSender) submitControl(ctx context.Context, dataID uint64, frameType uint8, flags uint8, seq uint64, payload []byte) error {
+	if s == nil {
+		return errTransportDetached
+	}
+	items := []bulkDedicatedSendRequest{{
+		Type:  frameType,
+		Flags: flags,
+		Seq:   seq,
+	}}
+	if len(payload) > 0 {
+		items[0].Payload = append([]byte(nil), payload...)
+	}
+	return s.submitBatch(ctx, dataID, items, true)
+}
+
+func (s *bulkDedicatedLaneSender) submitBatch(ctx context.Context, dataID uint64, items []bulkDedicatedSendRequest, wait bool) error {
+	if s == nil {
+		return errTransportDetached
+	}
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	if err := s.errSnapshot(); err != nil {
+		return err
+	}
+	req := getBulkDedicatedLaneBatchRequest()
+	req.prepare(ctx, dataID, items, wait)
+	s.queued.Add(1)
+	select {
+	case <-ctx.Done():
+		s.queued.Add(-1)
+		req.recycle()
+		return normalizeStreamDeadlineError(ctx.Err())
+	case <-s.stopCh:
+		s.queued.Add(-1)
+		req.recycle()
+		return s.stoppedErr()
+	case s.reqCh <- req:
+		if !wait {
+			return nil
+		}
+		return s.waitAck(req)
+	}
+}
+
+func (s *bulkDedicatedLaneSender) tryDirectSubmitBatch(ctx context.Context, dataID uint64, items []bulkDedicatedSendRequest) (bool, error) {
+	if s == nil {
+		return true, errTransportDetached
+	}
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	if len(items) == 0 {
+		return true, nil
+	}
+	if err := s.errSnapshot(); err != nil {
+		return true, err
+	}
+	select {
+	case <-ctx.Done():
+		return true, normalizeStreamDeadlineError(ctx.Err())
+	case <-s.stopCh:
+		return true, s.stoppedErr()
+	default:
+	}
+	if s.queued.Load() != 0 {
+		return false, nil
+	}
+	if !s.flushMu.TryLock() {
+		return false, nil
+	}
+	defer s.flushMu.Unlock()
+	if s.queued.Load() != 0 {
+		return false, nil
+	}
+	if err := s.errSnapshot(); err != nil {
+		return true, err
+	}
+	select {
+	case <-ctx.Done():
+		return true, normalizeStreamDeadlineError(ctx.Err())
+	case <-s.stopCh:
+		return true, s.stoppedErr()
+	default:
+	}
+	deadline, _ := ctx.Deadline()
+	if err := s.flush([]bulkDedicatedOutboundBatch{{
+		DataID: dataID,
+		Items:  items,
+	}}, deadline); err != nil {
+		err = normalizeDedicatedBulkSendError(err)
+		s.setErr(err)
+		s.failPending(err)
+		if s.fail != nil {
+			go s.fail(err)
+		}
+		return true, err
+	}
+	return true, nil
+}
+
+func (s *bulkDedicatedLaneSender) waitAck(req *bulkDedicatedLaneBatchRequest) error {
+	if s == nil {
+		return errTransportDetached
+	}
+	if req == nil {
+		return errTransportDetached
+	}
+	ctx := req.Ctx
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	select {
+	case err := <-req.resultCh:
+		req.recycle()
+		return normalizeDedicatedBulkSendError(err)
+	case <-ctx.Done():
+		if req.tryCancel() {
+			req.aborted.Store(true)
+			return normalizeStreamDeadlineError(ctx.Err())
+		}
+		err := <-req.resultCh
+		req.recycle()
+		return normalizeDedicatedBulkSendError(err)
+	}
+}
+
+func (s *bulkDedicatedLaneSender) stop() {
+	if s == nil {
+		return
+	}
+	s.stopOnce.Do(func() {
+		s.setErr(errTransportDetached)
+		close(s.stopCh)
+	})
+	<-s.doneCh
+}
+
+func (s *bulkDedicatedLaneSender) run() {
+	defer close(s.doneCh)
+	var carry *bulkDedicatedLaneBatchRequest
+	for {
+		req, ok := s.nextRequest(carry)
+		carry = nil
+		if !ok {
+			return
+		}
+		if !req.tryStart() {
+			s.finishRequest(req, req.canceledErr())
+			continue
+		}
+		if err := req.contextErr(); err != nil {
+			s.finishRequest(req, err)
+			continue
+		}
+		batchReqs := []*bulkDedicatedLaneBatchRequest{req}
+		batches := []bulkDedicatedOutboundBatch{{
+			DataID: req.DataID,
+			Items:  req.Items,
+		}}
+		batchBytes := bulkDedicatedBatchesPlainLen(batches)
+		deadline := req.Deadline
+		s.flushMu.Lock()
+		err := s.errSnapshot()
+		if err == nil {
+			carry, err = s.collectBatchRequests(&batchReqs, &batches, &batchBytes, &deadline)
+			if err == nil {
+				err = s.flush(batches, deadline)
+			}
+		}
+		s.flushMu.Unlock()
+		if err != nil {
+			err = normalizeDedicatedBulkSendError(err)
+			s.setErr(err)
+			s.finishBatchRequests(batchReqs, err)
+			if carry != nil {
+				s.finishRequest(carry, err)
+				carry = nil
+			}
+			s.failPending(err)
+			if s.fail != nil {
+				go s.fail(err)
+			}
+			return
+		}
+		s.finishBatchRequests(batchReqs, nil)
+	}
+}
+
+func appendBulkDedicatedLaneBatch(batches *[]bulkDedicatedOutboundBatch, req *bulkDedicatedLaneBatchRequest) {
+	if req == nil {
+		return
+	}
+	if len(*batches) > 0 && (*batches)[len(*batches)-1].DataID == req.DataID {
+		last := &(*batches)[len(*batches)-1]
+		last.Items = append(last.Items, req.Items...)
+		return
+	}
+	*batches = append(*batches, bulkDedicatedOutboundBatch{
+		DataID: req.DataID,
+		Items:  req.Items,
+	})
+}
+
+func bulkDedicatedLaneBatchItemCount(batches []bulkDedicatedOutboundBatch) int {
+	total := 0
+	for _, batch := range batches {
+		total += len(batch.Items)
+	}
+	return total
+}
+
+func bulkDedicatedLaneNextBatchBytes(batches []bulkDedicatedOutboundBatch, req *bulkDedicatedLaneBatchRequest, currentBytes int) int {
+	reqBytes := bulkDedicatedSendRequestsLen(req.Items)
+	if len(batches) == 0 {
+		return bulkDedicatedBatchHeaderLen + reqBytes
+	}
+	last := batches[len(batches)-1]
+	if last.DataID == req.DataID {
+		return currentBytes + reqBytes
+	}
+	if len(batches) == 1 {
+		return bulkDedicatedSuperBatchHeaderLen +
+			bulkDedicatedSuperBatchGroupHeaderLen + bulkDedicatedSendRequestsLen(batches[0].Items) +
+			bulkDedicatedSuperBatchGroupHeaderLen + reqBytes
+	}
+	return currentBytes + bulkDedicatedSuperBatchGroupHeaderLen + reqBytes
+}
+
+func (r *bulkDedicatedLaneBatchRequest) contextErr() error {
+	if r.Ctx == nil {
+		return nil
+	}
+	select {
+	case <-r.Ctx.Done():
+		return normalizeStreamDeadlineError(r.Ctx.Err())
+	default:
+		return nil
+	}
+}
+
+func (r *bulkDedicatedLaneBatchRequest) tryStart() bool {
+	if r == nil {
+		return false
+	}
+	return r.state.value.CompareAndSwap(bulkDedicatedRequestQueued, bulkDedicatedRequestStarted)
+}
+
+func (r *bulkDedicatedLaneBatchRequest) tryCancel() bool {
+	if r == nil {
+		return false
+	}
+	return r.state.value.CompareAndSwap(bulkDedicatedRequestQueued, bulkDedicatedRequestCanceled)
+}
+
+func (r *bulkDedicatedLaneBatchRequest) canceledErr() error {
+	if r == nil {
+		return context.Canceled
+	}
+	if err := r.contextErr(); err != nil {
+		return err
+	}
+	return context.Canceled
+}
+
+func (s *bulkDedicatedLaneSender) nextRequest(carry *bulkDedicatedLaneBatchRequest) (*bulkDedicatedLaneBatchRequest, bool) {
+	if carry != nil {
+		select {
+		case <-s.stopCh:
+			err := s.stoppedErr()
+			s.finishRequest(carry, err)
+			s.failPending(err)
+			return nil, false
+		default:
+			return carry, true
+		}
+	}
+	select {
+	case <-s.stopCh:
+		s.failPending(s.stoppedErr())
+		return nil, false
+	case req := <-s.reqCh:
+		return req, true
+	}
+}
+
+func (s *bulkDedicatedLaneSender) collectBatchRequests(batchReqs *[]*bulkDedicatedLaneBatchRequest, batches *[]bulkDedicatedOutboundBatch, batchBytes *int, deadline *time.Time) (*bulkDedicatedLaneBatchRequest, error) {
+	if s == nil || bulkDedicatedLaneBatchItemCount(*batches) >= bulkDedicatedBatchMaxItems || *batchBytes >= bulkDedicatedBatchMaxPlainBytes {
+		return nil, nil
+	}
+	wait := bulkDedicatedLaneMicroBatchWait
+	var (
+		timer   *time.Timer
+		timerCh <-chan time.Time
+		waited  bool
+	)
+	if wait > 0 {
+		timer = time.NewTimer(wait)
+		timerCh = timer.C
+		defer timer.Stop()
+	}
+	for {
+		if bulkDedicatedLaneBatchItemCount(*batches) >= bulkDedicatedBatchMaxItems || *batchBytes >= bulkDedicatedBatchMaxPlainBytes {
+			return nil, nil
+		}
+		var (
+			req *bulkDedicatedLaneBatchRequest
+			ok  bool
+		)
+		select {
+		case <-s.stopCh:
+			return nil, s.stoppedErr()
+		case req = <-s.reqCh:
+			ok = true
+		default:
+		}
+		if !ok {
+			if waited || timerCh == nil {
+				return nil, nil
+			}
+			waited = true
+			select {
+			case <-s.stopCh:
+				return nil, s.stoppedErr()
+			case req = <-s.reqCh:
+			case <-timerCh:
+				return nil, nil
+			}
+		}
+		if err := req.contextErr(); err != nil {
+			if req.tryCancel() {
+				s.finishRequest(req, err)
+				continue
+			}
+			s.finishRequest(req, req.canceledErr())
+			continue
+		}
+		nextItems := bulkDedicatedLaneBatchItemCount(*batches) + len(req.Items)
+		if nextItems > bulkDedicatedBatchMaxItems {
+			return req, nil
+		}
+		nextBytes := bulkDedicatedLaneNextBatchBytes(*batches, req, *batchBytes)
+		if nextBytes > bulkDedicatedBatchMaxPlainBytes {
+			return req, nil
+		}
+		if !req.tryStart() {
+			s.finishRequest(req, req.canceledErr())
+			continue
+		}
+		if err := req.contextErr(); err != nil {
+			s.finishRequest(req, err)
+			continue
+		}
+		*batchReqs = append(*batchReqs, req)
+		appendBulkDedicatedLaneBatch(batches, req)
+		*batchBytes = nextBytes
+		*deadline = earlierDeadline(*deadline, req.Deadline)
+	}
+}
+
+func earlierDeadline(current time.Time, next time.Time) time.Time {
+	if current.IsZero() {
+		return next
+	}
+	if next.IsZero() || current.Before(next) {
+		return current
+	}
+	return next
+}
+
+func (s *bulkDedicatedLaneSender) flush(batches []bulkDedicatedOutboundBatch, deadline time.Time) error {
+	if s == nil || s.conn == nil {
+		return errTransportDetached
+	}
+	payload, release, err := s.encode(batches)
+	if err != nil {
+		return err
+	}
+	if release != nil {
+		defer release()
+	}
+	return writeBulkDedicatedRecordWithDeadline(s.conn, payload, deadline)
+}
+
+func (s *bulkDedicatedLaneSender) finishRequest(req *bulkDedicatedLaneBatchRequest, err error) {
+	if s != nil {
+		s.queued.Add(-1)
+	}
+	if req == nil {
+		return
+	}
+	if req.wait && req.resultCh != nil {
+		if req.aborted.Load() {
+			req.recycle()
+			return
+		}
+		req.resultCh <- err
+		return
+	}
+	req.recycle()
+}
+
+func (s *bulkDedicatedLaneSender) finishBatchRequests(reqs []*bulkDedicatedLaneBatchRequest, err error) {
+	for _, req := range reqs {
+		s.finishRequest(req, err)
+	}
+}
+
+func (s *bulkDedicatedLaneSender) failPending(err error) {
+	for {
+		select {
+		case item := <-s.reqCh:
+			s.finishRequest(item, err)
+		default:
+			return
+		}
+	}
+}
+
+func (s *bulkDedicatedLaneSender) setErr(err error) {
+	if s == nil || err == nil {
+		return
+	}
+	s.errMu.Lock()
+	if s.err == nil {
+		s.err = err
+	}
+	s.errMu.Unlock()
+}
+
+func (s *bulkDedicatedLaneSender) errSnapshot() error {
+	if s == nil {
+		return errTransportDetached
+	}
+	s.errMu.Lock()
+	defer s.errMu.Unlock()
+	return s.err
+}
+
+func (s *bulkDedicatedLaneSender) stoppedErr() error {
+	if err := s.errSnapshot(); err != nil {
+		return err
+	}
+	return errTransportDetached
+}

bulk_dedicated_lane_sender_test.go

@@ -0,0 +1,145 @@
+package notify
+
+import (
+	"bytes"
+	"context"
+	"testing"
+	"time"
+)
+
+func TestBulkDedicatedLaneSenderCollectBatchRequestsBatchesAcrossDataIDs(t *testing.T) {
+	sender := &bulkDedicatedLaneSender{
+		reqCh:  make(chan *bulkDedicatedLaneBatchRequest, 3),
+		stopCh: make(chan struct{}),
+	}
+
+	first := &bulkDedicatedLaneBatchRequest{
+		Ctx:    context.Background(),
+		DataID: 7,
+		Items: []bulkDedicatedSendRequest{{
+			Type:    bulkFastPayloadTypeData,
+			Seq:     1,
+			Payload: []byte("a"),
+		}},
+		Deadline: time.Now().Add(2 * time.Second),
+	}
+	if !first.tryStart() {
+		t.Fatal("first request should start")
+	}
+
+	second := &bulkDedicatedLaneBatchRequest{
+		Ctx:    context.Background(),
+		DataID: 7,
+		Items: []bulkDedicatedSendRequest{{
+			Type:    bulkFastPayloadTypeData,
+			Seq:     2,
+			Payload: []byte("b"),
+		}},
+		Deadline: time.Now().Add(time.Second),
+	}
+	third := &bulkDedicatedLaneBatchRequest{
+		Ctx:    context.Background(),
+		DataID: 8,
+		Items: []bulkDedicatedSendRequest{{
+			Type:    bulkFastPayloadTypeData,
+			Seq:     3,
+			Payload: []byte("c"),
+		}},
+		Deadline: time.Now().Add(3 * time.Second),
+	}
+
+	sender.reqCh <- second
+	sender.reqCh <- third
+
+	batchReqs := []*bulkDedicatedLaneBatchRequest{first}
+	batches := []bulkDedicatedOutboundBatch{{
+		DataID: first.DataID,
+		Items:  first.Items,
+	}}
+	batchBytes := bulkDedicatedBatchesPlainLen(batches)
+	deadline := first.Deadline
+
+	carry, err := sender.collectBatchRequests(&batchReqs, &batches, &batchBytes, &deadline)
+	if err != nil {
+		t.Fatalf("collectBatchRequests error = %v", err)
+	}
+	if carry != nil {
+		t.Fatalf("carry request = %+v, want nil", carry)
+	}
+	if len(batchReqs) != 3 {
+		t.Fatalf("batched request count = %d, want 3", len(batchReqs))
+	}
+	if len(batches) != 2 {
+		t.Fatalf("batched group count = %d, want 2", len(batches))
+	}
+	if got, want := batches[0].DataID, uint64(7); got != want {
+		t.Fatalf("first batch dataID = %d, want %d", got, want)
+	}
+	if got, want := len(batches[0].Items), 2; got != want {
+		t.Fatalf("first batch item count = %d, want %d", got, want)
+	}
+	if got, want := batches[1].DataID, uint64(8); got != want {
+		t.Fatalf("second batch dataID = %d, want %d", got, want)
+	}
+	if got, want := len(batches[1].Items), 1; got != want {
+		t.Fatalf("second batch item count = %d, want %d", got, want)
+	}
+	if got, want := batches[0].Items[0].Seq, uint64(1); got != want {
+		t.Fatalf("first batch seq = %d, want %d", got, want)
+	}
+	if got, want := batches[0].Items[1].Seq, uint64(2); got != want {
+		t.Fatalf("second batch seq = %d, want %d", got, want)
+	}
+	if got, want := batches[1].Items[0].Seq, uint64(3); got != want {
+		t.Fatalf("third batch seq = %d, want %d", got, want)
+	}
+	if !deadline.Equal(second.Deadline) {
+		t.Fatalf("merged deadline = %v, want %v", deadline, second.Deadline)
+	}
+}
+
+func TestDedicatedBulkSuperBatchRoundTrip(t *testing.T) {
+	batches := []bulkDedicatedOutboundBatch{
+		{
+			DataID: 7,
+			Items: []bulkDedicatedSendRequest{
+				{Type: bulkFastPayloadTypeData, Seq: 1, Payload: []byte("alpha")},
+				{Type: bulkFastPayloadTypeData, Seq: 2, Payload: []byte("beta")},
+			},
+		},
+		{
+			DataID: 8,
+			Items: []bulkDedicatedSendRequest{
+				{Type: bulkFastPayloadTypeClose, Flags: bulkFastPayloadFlagFullClose, Seq: 3},
+				{Type: bulkFastPayloadTypeRelease, Seq: 4, Payload: []byte{1, 2, 3, 4}},
+			},
+		},
+	}
+
+	plain, err := encodeBulkDedicatedBatchesPlain(batches)
+	if err != nil {
+		t.Fatalf("encodeBulkDedicatedBatchesPlain error = %v", err)
+	}
+	got, err := decodeDedicatedBulkInboundBatches(plain)
+	if err != nil {
+		t.Fatalf("decodeDedicatedBulkInboundBatches error = %v", err)
+	}
+	if len(got) != 2 {
+		t.Fatalf("decoded batch count = %d, want 2", len(got))
+	}
+	if got[0].DataID != 7 || got[1].DataID != 8 {
+		t.Fatalf("decoded dataIDs = %d,%d, want 7,8", got[0].DataID, got[1].DataID)
+	}
+	if len(got[0].Items) != 2 || len(got[1].Items) != 2 {
+		t.Fatalf("decoded item counts = %d,%d, want 2,2", len(got[0].Items), len(got[1].Items))
+	}
+	if !bytes.Equal(got[0].Items[0].Payload, []byte("alpha")) || !bytes.Equal(got[0].Items[1].Payload, []byte("beta")) {
+		t.Fatalf("decoded first batch payloads mismatch: %+v", got[0].Items)
+	}
+	if got[1].Items[0].Type != bulkFastPayloadTypeClose || got[1].Items[0].Flags != bulkFastPayloadFlagFullClose {
+		t.Fatalf("decoded close item mismatch: %+v", got[1].Items[0])
+	}
+	if got[1].Items[1].Type != bulkFastPayloadTypeRelease || !bytes.Equal(got[1].Items[1].Payload, []byte{1, 2, 3, 4}) {
+		t.Fatalf("decoded release item mismatch: %+v", got[1].Items[1])
+	}
+}

bulk_dedicated_sidecar.go

@@ -0,0 +1,451 @@
+package notify
+
+import (
+	"context"
+	"net"
+	"sync"
+)
+
+type bulkDedicatedSidecar struct {
+	laneID    uint32
+	conn      net.Conn
+	closeOnce sync.Once
+
+	senderMu sync.Mutex
+	sender   *bulkDedicatedLaneSender
+}
+
+type bulkDedicatedLane struct {
+	id           uint32
+	activeBulks  int
+	sidecar      *bulkDedicatedSidecar
+	attachFlight *bulkDedicatedAttachFlight
+}
+
+type bulkDedicatedAttachFlight struct {
+	done chan struct{}
+	once sync.Once
+	err  error
+}
+
+func normalizeBulkDedicatedLaneID(laneID uint32) uint32 {
+	if laneID == 0 {
+		return 1
+	}
+	return laneID
+}
+
+func newBulkDedicatedSidecar(conn net.Conn, laneID uint32) *bulkDedicatedSidecar {
+	if conn == nil {
+		return nil
+	}
+	return &bulkDedicatedSidecar{
+		laneID: normalizeBulkDedicatedLaneID(laneID),
+		conn:   conn,
+	}
+}
+
+func newBulkDedicatedAttachFlight() *bulkDedicatedAttachFlight {
+	return &bulkDedicatedAttachFlight{
+		done: make(chan struct{}),
+	}
+}
+
+func (f *bulkDedicatedAttachFlight) finish(err error) {
+	if f == nil {
+		return
+	}
+	f.once.Do(func() {
+		f.err = err
+		close(f.done)
+	})
+}
+
+func (f *bulkDedicatedAttachFlight) wait(ctx context.Context) error {
+	if f == nil {
+		return nil
+	}
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	select {
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-f.done:
+		return f.err
+	}
+}
+
+func (s *bulkDedicatedSidecar) close() {
+	if s == nil {
+		return
+	}
+	s.closeOnce.Do(func() {
+		if sender := s.laneSenderSnapshot(); sender != nil {
+			sender.stop()
+		}
+		if s.conn != nil {
+			_ = s.conn.Close()
+		}
+	})
+}
+
+func (s *bulkDedicatedSidecar) laneSenderSnapshot() *bulkDedicatedLaneSender {
+	if s == nil {
+		return nil
+	}
+	s.senderMu.Lock()
+	defer s.senderMu.Unlock()
+	return s.sender
+}
+
+func (s *bulkDedicatedSidecar) laneSenderWithFactory(factory func(net.Conn) *bulkDedicatedLaneSender) *bulkDedicatedLaneSender {
+	if s == nil || factory == nil {
+		return nil
+	}
+	s.senderMu.Lock()
+	defer s.senderMu.Unlock()
+	if s.sender != nil {
+		return s.sender
+	}
+	if s.conn == nil {
+		return nil
+	}
+	s.sender = factory(s.conn)
+	return s.sender
+}
+
+func (c *ClientCommon) clientDedicatedSidecarSnapshot() *bulkDedicatedSidecar {
+	if c == nil {
+		return nil
+	}
+	c.bulkDedicatedSidecarMu.Lock()
+	defer c.bulkDedicatedSidecarMu.Unlock()
+	return firstClientDedicatedSidecarLocked(c.bulkDedicatedLanes)
+}
+
+func firstClientDedicatedSidecarLocked(lanes map[uint32]*bulkDedicatedLane) *bulkDedicatedSidecar {
+	var (
+		selected *bulkDedicatedSidecar
+		bestID   uint32
+	)
+	for laneID, lane := range lanes {
+		if lane == nil || lane.sidecar == nil {
+			continue
+		}
+		if selected == nil || laneID < bestID {
+			selected = lane.sidecar
+			bestID = laneID
+		}
+	}
+	return selected
+}
+
+func (c *ClientCommon) reserveBulkDedicatedLane() uint32 {
+	if c == nil {
+		return normalizeBulkDedicatedLaneID(0)
+	}
+	c.bulkDedicatedSidecarMu.Lock()
+	defer c.bulkDedicatedSidecarMu.Unlock()
+	if c.bulkDedicatedLanes == nil {
+		c.bulkDedicatedLanes = make(map[uint32]*bulkDedicatedLane)
+	}
+	limit := c.bulkDedicatedLaneLimitSnapshot()
+	var best *bulkDedicatedLane
+	for _, lane := range c.bulkDedicatedLanes {
+		if lane == nil {
+			continue
+		}
+		if best == nil || lane.activeBulks < best.activeBulks || (lane.activeBulks == best.activeBulks && lane.id < best.id) {
+			best = lane
+		}
+	}
+	if best == nil || ((limit <= 0 || len(c.bulkDedicatedLanes) < limit) && best.activeBulks > 0) {
+		c.bulkDedicatedNextLaneID++
+		laneID := normalizeBulkDedicatedLaneID(c.bulkDedicatedNextLaneID)
+		best = &bulkDedicatedLane{id: laneID}
+		c.bulkDedicatedLanes[laneID] = best
+	}
+	best.activeBulks++
+	return best.id
+}
+
+func (c *ClientCommon) releaseBulkDedicatedLane(laneID uint32) {
+	if c == nil {
+		return
+	}
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	c.bulkDedicatedSidecarMu.Lock()
+	defer c.bulkDedicatedSidecarMu.Unlock()
+	lane := c.bulkDedicatedLanes[laneID]
+	if lane == nil {
+		return
+	}
+	if lane.activeBulks > 0 {
+		lane.activeBulks--
+	}
+	if lane.activeBulks == 0 && lane.sidecar == nil && lane.attachFlight == nil {
+		delete(c.bulkDedicatedLanes, laneID)
+	}
+}
+
+func (c *ClientCommon) clientDedicatedSidecarSnapshotForLane(laneID uint32) *bulkDedicatedSidecar {
+	if c == nil {
+		return nil
+	}
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	c.bulkDedicatedSidecarMu.Lock()
+	defer c.bulkDedicatedSidecarMu.Unlock()
+	if lane := c.bulkDedicatedLanes[laneID]; lane != nil {
+		return lane.sidecar
+	}
+	return nil
+}
+
+func (c *ClientCommon) beginClientDedicatedSidecarAttach(laneID uint32) (*bulkDedicatedSidecar, *bulkDedicatedAttachFlight, bool) {
+	if c == nil {
+		return nil, nil, false
+	}
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	c.bulkDedicatedSidecarMu.Lock()
+	defer c.bulkDedicatedSidecarMu.Unlock()
+	if c.bulkDedicatedLanes == nil {
+		c.bulkDedicatedLanes = make(map[uint32]*bulkDedicatedLane)
+	}
+	lane := c.bulkDedicatedLanes[laneID]
+	if lane == nil {
+		lane = &bulkDedicatedLane{id: laneID}
+		c.bulkDedicatedLanes[laneID] = lane
+	}
+	if lane.sidecar != nil {
+		return lane.sidecar, nil, false
+	}
+	if lane.attachFlight != nil {
+		return nil, lane.attachFlight, false
+	}
+	flight := newBulkDedicatedAttachFlight()
+	lane.attachFlight = flight
+	return nil, flight, true
+}
+
+func (c *ClientCommon) finishClientDedicatedSidecarAttach(laneID uint32, flight *bulkDedicatedAttachFlight, err error) {
+	if c == nil || flight == nil {
+		return
+	}
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	c.bulkDedicatedSidecarMu.Lock()
+	if lane := c.bulkDedicatedLanes[laneID]; lane != nil && lane.attachFlight == flight {
+		lane.attachFlight = nil
+		if lane.activeBulks == 0 && lane.sidecar == nil {
+			delete(c.bulkDedicatedLanes, laneID)
+		}
+	}
+	c.bulkDedicatedSidecarMu.Unlock()
+	flight.finish(err)
+}
+
+func (c *ClientCommon) installClientDedicatedSidecar(laneID uint32, sidecar *bulkDedicatedSidecar) (*bulkDedicatedSidecar, bool) {
+	if c == nil || sidecar == nil {
+		return nil, false
+	}
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	c.bulkDedicatedSidecarMu.Lock()
+	defer c.bulkDedicatedSidecarMu.Unlock()
+	if c.bulkDedicatedLanes == nil {
+		c.bulkDedicatedLanes = make(map[uint32]*bulkDedicatedLane)
+	}
+	lane := c.bulkDedicatedLanes[laneID]
+	if lane == nil {
+		lane = &bulkDedicatedLane{id: laneID}
+		c.bulkDedicatedLanes[laneID] = lane
+	}
+	if lane.sidecar != nil {
+		return lane.sidecar, false
+	}
+	lane.sidecar = sidecar
+	return sidecar, true
+}
+
+func (c *ClientCommon) clearClientDedicatedSidecar(laneID uint32, sidecar *bulkDedicatedSidecar) bool {
+	if c == nil || sidecar == nil {
+		return false
+	}
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	c.bulkDedicatedSidecarMu.Lock()
+	defer c.bulkDedicatedSidecarMu.Unlock()
+	lane := c.bulkDedicatedLanes[laneID]
+	if lane == nil || lane.sidecar != sidecar {
+		return false
+	}
+	lane.sidecar = nil
+	if lane.activeBulks == 0 && lane.attachFlight == nil {
+		delete(c.bulkDedicatedLanes, laneID)
+	}
+	return true
+}
+
+func (c *ClientCommon) closeClientDedicatedSidecar() {
+	if c == nil {
+		return
+	}
+	c.bulkDedicatedSidecarMu.Lock()
+	lanes := c.bulkDedicatedLanes
+	c.bulkDedicatedLanes = make(map[uint32]*bulkDedicatedLane)
+	c.bulkDedicatedSidecarMu.Unlock()
+	for _, lane := range lanes {
+		if lane == nil {
+			continue
+		}
+		if lane.sidecar != nil {
+			lane.sidecar.close()
+		}
+		if lane.attachFlight != nil {
+			lane.attachFlight.finish(errServiceShutdown)
+		}
+	}
+}
+
+func (c *ClientCommon) handleClientDedicatedSidecarFailure(sidecar *bulkDedicatedSidecar, err error) {
+	if c == nil || sidecar == nil {
+		return
+	}
+	if !c.clearClientDedicatedSidecar(sidecar.laneID, sidecar) {
+		return
+	}
+	runtime := c.getBulkRuntime()
+	if runtime != nil && sidecar.conn != nil {
+		runtime.handleDedicatedReadErrorByConn(clientFileScope(), sidecar.conn, err)
+	}
+	sidecar.close()
+}
+
+func (s *ServerCommon) serverDedicatedSidecarSnapshot(logical *LogicalConn) *bulkDedicatedSidecar {
+	if s == nil || logical == nil {
+		return nil
+	}
+	s.bulkDedicatedSidecarMu.Lock()
+	defer s.bulkDedicatedSidecarMu.Unlock()
+	return firstServerDedicatedSidecarLocked(s.bulkDedicatedSidecars[logical])
+}
+
+func firstServerDedicatedSidecarLocked(lanes map[uint32]*bulkDedicatedSidecar) *bulkDedicatedSidecar {
+	var (
+		selected *bulkDedicatedSidecar
+		bestID   uint32
+	)
+	for laneID, sidecar := range lanes {
+		if sidecar == nil {
+			continue
+		}
+		if selected == nil || laneID < bestID {
+			selected = sidecar
+			bestID = laneID
+		}
+	}
+	return selected
+}
+
+func (s *ServerCommon) serverDedicatedSidecarSnapshotForLane(logical *LogicalConn, laneID uint32) *bulkDedicatedSidecar {
+	if s == nil || logical == nil {
+		return nil
+	}
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	s.bulkDedicatedSidecarMu.Lock()
+	defer s.bulkDedicatedSidecarMu.Unlock()
+	if lanes := s.bulkDedicatedSidecars[logical]; lanes != nil {
+		return lanes[laneID]
+	}
+	return nil
+}
+
+func (s *ServerCommon) installServerDedicatedSidecar(logical *LogicalConn, laneID uint32, sidecar *bulkDedicatedSidecar) *bulkDedicatedSidecar {
+	if s == nil || logical == nil || sidecar == nil {
+		return nil
+	}
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	s.bulkDedicatedSidecarMu.Lock()
+	defer s.bulkDedicatedSidecarMu.Unlock()
+	lanes := s.bulkDedicatedSidecars[logical]
+	if lanes == nil {
+		lanes = make(map[uint32]*bulkDedicatedSidecar)
+		s.bulkDedicatedSidecars[logical] = lanes
+	}
+	prev := lanes[laneID]
+	lanes[laneID] = sidecar
+	return prev
+}
+
+func (s *ServerCommon) clearServerDedicatedSidecar(logical *LogicalConn, laneID uint32, sidecar *bulkDedicatedSidecar) bool {
+	if s == nil || logical == nil || sidecar == nil {
+		return false
+	}
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	s.bulkDedicatedSidecarMu.Lock()
+	defer s.bulkDedicatedSidecarMu.Unlock()
+	lanes := s.bulkDedicatedSidecars[logical]
+	if lanes == nil || lanes[laneID] != sidecar {
+		return false
+	}
+	delete(lanes, laneID)
+	if len(lanes) == 0 {
+		delete(s.bulkDedicatedSidecars, logical)
+	}
+	return true
+}
+
+func (s *ServerCommon) closeServerDedicatedSidecar(logical *LogicalConn) {
+	if s == nil || logical == nil {
+		return
+	}
+	s.bulkDedicatedSidecarMu.Lock()
+	lanes := s.bulkDedicatedSidecars[logical]
+	delete(s.bulkDedicatedSidecars, logical)
+	s.bulkDedicatedSidecarMu.Unlock()
+	for _, sidecar := range lanes {
+		if sidecar != nil {
+			sidecar.close()
+		}
+	}
+}
+
+func (s *ServerCommon) closeAllServerDedicatedSidecars() {
+	if s == nil {
+		return
+	}
+	s.bulkDedicatedSidecarMu.Lock()
+	lanesByLogical := s.bulkDedicatedSidecars
+	s.bulkDedicatedSidecars = make(map[*LogicalConn]map[uint32]*bulkDedicatedSidecar)
+	s.bulkDedicatedSidecarMu.Unlock()
+	for _, lanes := range lanesByLogical {
+		for _, sidecar := range lanes {
+			if sidecar != nil {
+				sidecar.close()
+			}
+		}
+	}
+}
+
+func (s *ServerCommon) handleServerDedicatedSidecarFailure(logical *LogicalConn, sidecar *bulkDedicatedSidecar, err error) {
+	if s == nil || logical == nil || sidecar == nil {
+		return
+	}
+	if !s.clearServerDedicatedSidecar(logical, sidecar.laneID, sidecar) {
+		return
+	}
+	runtime := s.getBulkRuntime()
+	if runtime != nil && sidecar.conn != nil {
+		runtime.handleDedicatedReadErrorByConn(serverFileScope(logical), sidecar.conn, err)
+	}
+	sidecar.close()
+}
+
+func (s *ServerCommon) attachServerDedicatedSidecarIfExists(logical *LogicalConn, bulk *bulkHandle) {
+	if s == nil || logical == nil || bulk == nil || !bulk.Dedicated() {
+		return
+	}
+	sidecar := s.serverDedicatedSidecarSnapshotForLane(logical, bulk.dedicatedLaneIDSnapshot())
+	if sidecar == nil || sidecar.conn == nil {
+		return
+	}
+	_ = bulk.attachDedicatedConnShared(sidecar.conn)
+}

bulk_dispatcher.go

@@ -12,6 +12,10 @@ import (
 const bulkDispatchRejectTimeout = 300 * time.Millisecond
 
 func (c *ClientCommon) dispatchFastBulkFrame(frame bulkFastFrame) {
+	c.dispatchFastBulkFrameWithOwner(frame, nil)
+}
+
+func (c *ClientCommon) dispatchFastBulkFrameWithOwner(frame bulkFastFrame, owner *bulkReadPayloadOwner) {
 	if frame.DataID == 0 {
 		return
 	}
@@ -38,7 +42,13 @@ func (c *ClientCommon) dispatchFastBulkFrame(frame bulkFastFrame) {
 	}
 	switch frame.Type {
 	case bulkFastPayloadTypeData:
-		if err := bulk.pushOwnedChunk(frame.Payload); err != nil {
+		var err error
+		if owner != nil {
+			err = bulk.pushChunkWithOwnershipOptionsAndRelease(frame.Payload, true, true, owner.retainChunk())
+		} else {
+			err = bulk.pushOwnedChunk(frame.Payload)
+		}
+		if err != nil {
 			if c.showError || c.debugMode {
 				fmt.Println("client bulk push chunk error", err)
 			}
@@ -58,14 +68,28 @@ func (c *ClientCommon) dispatchFastBulkFrame(frame bulkFastFrame) {
 			resetErr = bulkRemoteResetError(string(frame.Payload))
 		}
 		bulk.markReset(bulkResetError(resetErr))
+	case bulkFastPayloadTypeRelease:
+		bytes, chunks, err := decodeBulkDedicatedReleasePayload(frame.Payload)
+		if err != nil {
+			if c.showError || c.debugMode {
+				fmt.Println("client bulk release decode error", err)
+			}
+			c.bestEffortRejectInboundBulkData(bulk.ID(), frame.DataID, err.Error())
+			return
+		}
+		bulk.releaseOutboundWindow(bytes, chunks)
 	}
 }
 
 func (c *ClientCommon) dispatchFastBulkData(frame bulkFastDataFrame) {
-	c.dispatchFastBulkFrame(frame)
+	c.dispatchFastBulkFrameWithOwner(frame, nil)
 }
 
 func (s *ServerCommon) dispatchFastBulkFrame(logical *LogicalConn, transport *TransportConn, conn net.Conn, frame bulkFastFrame) {
+	s.dispatchFastBulkFrameWithOwner(logical, transport, conn, frame, nil)
+}
+
+func (s *ServerCommon) dispatchFastBulkFrameWithOwner(logical *LogicalConn, transport *TransportConn, conn net.Conn, frame bulkFastFrame, owner *bulkReadPayloadOwner) {
 	if logical == nil || frame.DataID == 0 {
 		return
 	}
@@ -91,7 +115,13 @@ func (s *ServerCommon) dispatchFastBulkFrame(logical *LogicalConn, transport *Tr
 	}
 	switch frame.Type {
 	case bulkFastPayloadTypeData:
-		if err := bulk.pushOwnedChunk(frame.Payload); err != nil {
+		var err error
+		if owner != nil {
+			err = bulk.pushChunkWithOwnershipOptionsAndRelease(frame.Payload, true, true, owner.retainChunk())
+		} else {
+			err = bulk.pushOwnedChunk(frame.Payload)
+		}
+		if err != nil {
 			if s.showError || s.debugMode {
 				fmt.Println("server bulk push chunk error", err)
 			}
@@ -111,11 +141,82 @@ func (s *ServerCommon) dispatchFastBulkFrame(logical *LogicalConn, transport *Tr
 			resetErr = bulkRemoteResetError(string(frame.Payload))
 		}
 		bulk.markReset(bulkResetError(resetErr))
+	case bulkFastPayloadTypeRelease:
+		bytes, chunks, err := decodeBulkDedicatedReleasePayload(frame.Payload)
+		if err != nil {
+			if s.showError || s.debugMode {
+				fmt.Println("server bulk release decode error", err)
+			}
+			s.bestEffortRejectInboundBulkData(logical, transport, conn, bulk.ID(), frame.DataID, err.Error())
+			return
+		}
+		bulk.releaseOutboundWindow(bytes, chunks)
 	}
 }
 
 func (s *ServerCommon) dispatchFastBulkData(logical *LogicalConn, transport *TransportConn, conn net.Conn, frame bulkFastDataFrame) {
-	s.dispatchFastBulkFrame(logical, transport, conn, frame)
+	s.dispatchFastBulkFrameWithOwner(logical, transport, conn, frame, nil)
+}
+
+func (c *ClientCommon) tryDispatchBorrowedBulkTransportPayload(payload []byte) bool {
+	if c == nil || len(payload) == 0 {
+		return false
+	}
+	plain, plainRelease, err := decryptTransportPayloadCodecOwnedPooled(c.modernPSKRuntime, c.msgDe, c.SecretKey, payload)
+	if err != nil {
+		if c.showError || c.debugMode {
+			fmt.Println("client decode transport payload error", err)
+		}
+		return true
+	}
+	owner := newBulkReadPayloadOwner(plainRelease)
+	matched, walkErr := walkBulkFastFrames(plain, func(frame bulkFastFrame) error {
+		c.dispatchFastBulkFrameWithOwner(frame, owner)
+		return nil
+	})
+	if owner != nil {
+		owner.done()
+	}
+	if !matched {
+		return false
+	}
+	if walkErr != nil && (c.showError || c.debugMode) {
+		fmt.Println("client decode bulk fast payload error", walkErr)
+	}
+	return true
+}
+
+func (s *ServerCommon) tryDispatchBorrowedBulkTransportPayload(source interface{}, payload []byte) bool {
+	if s == nil || len(payload) == 0 {
+		return false
+	}
+	logical, transport := s.resolveInboundSource(source)
+	if logical == nil {
+		return false
+	}
+	plain, plainRelease, err := decryptTransportPayloadCodecOwnedPooled(logical.modernPSKRuntimeSnapshot(), logical.msgDeSnapshot(), logical.secretKeySnapshot(), payload)
+	if err != nil {
+		if s.showError || s.debugMode {
+			fmt.Println("server decode transport payload error", err)
+		}
+		return true
+	}
+	conn := serverInboundConn(source)
+	owner := newBulkReadPayloadOwner(plainRelease)
+	matched, walkErr := walkBulkFastFrames(plain, func(frame bulkFastFrame) error {
+		s.dispatchFastBulkFrameWithOwner(logical, transport, conn, frame, owner)
+		return nil
+	})
+	if owner != nil {
+		owner.done()
+	}
+	if !matched {
+		return false
+	}
+	if walkErr != nil && (s.showError || s.debugMode) {
+		fmt.Println("server decode bulk fast payload error", walkErr)
+	}
+	return true
 }
 
 func (c *ClientCommon) bestEffortRejectInboundBulkData(bulkID string, dataID uint64, message string) {

bulk_e2e_benchmark_test.go

@@ -308,7 +308,11 @@ func newBulkBenchmarkClient(tb testing.TB, network string, server bulkBenchmarkS
 			tb.Fatalf("UseSignalReliabilityClient failed: %v", err)
 		}
 	}
-	if err := client.Connect(network, server.addr); err != nil {
+	dialAddr := server.addr
+	if network == "tcp" {
+		dialAddr = benchmarkTCPDialAddr(tb, dialAddr)
+	}
+	if err := client.Connect(network, dialAddr); err != nil {
 		tb.Fatalf("client Connect failed: %v", err)
 	}
 	tb.Cleanup(func() {
@@ -333,6 +337,9 @@ func bulkBenchmarkListenAddr(tb testing.TB, network string) string {
 	case "unix":
 		return filepath.Join(tb.TempDir(), "notify-bulk.sock")
 	case "udp", "tcp":
+		if network == "tcp" {
+			return benchmarkTCPListenAddr(tb)
+		}
 		return "127.0.0.1:0"
 	default:
 		tb.Fatalf("unsupported benchmark network %q", network)

bulk_fastpath.go

@@ -120,32 +120,145 @@ func decodeBulkFastDataFrame(payload []byte) (bulkFastDataFrame, bool, error) {
 }
 
 func (c *ClientCommon) encodeFastBulkDataPayload(dataID uint64, seq uint64, chunk []byte) ([]byte, error) {
-	if c != nil && c.fastBulkEncode != nil {
-		return c.fastBulkEncode(c.SecretKey, dataID, seq, chunk)
-	}
-	scratch := getBulkFastFrameScratch(len(chunk))
-	defer putBulkFastFrameScratch(scratch)
-	frame := scratch[:bulkFastPayloadHeaderLen+len(chunk)]
-	if err := encodeBulkFastDataFrameHeader(frame, dataID, seq, len(chunk)); err != nil {
-		return nil, err
-	}
-	copy(frame[bulkFastPayloadHeaderLen:], chunk)
-	return c.encryptTransportPayload(frame)
+	return c.encodeBulkFastPayload(bulkFastFrame{
+		Type:    bulkFastPayloadTypeData,
+		DataID:  dataID,
+		Seq:     seq,
+		Payload: chunk,
+	})
 }
 
-func (c *ClientCommon) sendFastBulkData(ctx context.Context, dataID uint64, seq uint64, chunk []byte) error {
+func (c *ClientCommon) encodeBulkFastPayload(frame bulkFastFrame) ([]byte, error) {
+	if c == nil {
+		return nil, errBulkClientNil
+	}
+	if c.fastPlainEncode != nil {
+		return encodeBulkFastFramePayloadFast(c.fastPlainEncode, c.SecretKey, frame)
+	}
+	plain, err := encodeBulkFastFramePayload(frame)
+	if err != nil {
+		return nil, err
+	}
+	return c.encryptTransportPayload(plain)
+}
+
+func (c *ClientCommon) encodeBulkFastBatchPayload(frames []bulkFastFrame) ([]byte, error) {
+	if c == nil {
+		return nil, errBulkClientNil
+	}
+	if c.fastPlainEncode != nil {
+		return encodeBulkFastBatchPayloadFast(c.fastPlainEncode, c.SecretKey, frames)
+	}
+	plain, err := encodeBulkFastBatchPlain(frames)
+	if err != nil {
+		return nil, err
+	}
+	return c.encryptTransportPayload(plain)
+}
+
+func (c *ClientCommon) encodeBulkFastPayloadPooled(frame bulkFastFrame) ([]byte, func(), error) {
+	if c == nil {
+		return nil, nil, errBulkClientNil
+	}
+	if runtime := c.modernPSKRuntime; runtime != nil {
+		return encodeBulkFastFramePayloadPooled(runtime, frame)
+	}
+	if c.fastPlainEncode != nil {
+		payload, err := encodeBulkFastFramePayloadFast(c.fastPlainEncode, c.SecretKey, frame)
+		return payload, nil, err
+	}
+	plain, err := encodeBulkFastFramePayload(frame)
+	if err != nil {
+		return nil, nil, err
+	}
+	payload, err := c.encryptTransportPayload(plain)
+	return payload, nil, err
+}
+
+func (c *ClientCommon) encodeBulkFastBatchPayloadPooled(frames []bulkFastFrame) ([]byte, func(), error) {
+	if c == nil {
+		return nil, nil, errBulkClientNil
+	}
+	if runtime := c.modernPSKRuntime; runtime != nil {
+		return encodeBulkFastBatchPayloadPooled(runtime, frames)
+	}
+	if c.fastPlainEncode != nil {
+		payload, err := encodeBulkFastBatchPayloadFast(c.fastPlainEncode, c.SecretKey, frames)
+		return payload, nil, err
+	}
+	plain, err := encodeBulkFastBatchPlain(frames)
+	if err != nil {
+		return nil, nil, err
+	}
+	payload, err := c.encryptTransportPayload(plain)
+	return payload, nil, err
+}
+
+func (c *ClientCommon) sendFastBulkData(ctx context.Context, dataID uint64, seq uint64, chunk []byte, fastPathVersion uint8) error {
+	binding := c.clientTransportBindingSnapshot()
+	if binding == nil {
+		return net.ErrClosed
+	}
+	if sender := binding.clientBulkBatchSenderSnapshot(c); sender != nil {
+		return sender.submitData(ctx, dataID, seq, fastPathVersion, chunk)
+	}
 	payload, err := c.encodeFastBulkDataPayload(dataID, seq, chunk)
 	if err != nil {
 		return err
 	}
+	return c.writePayloadToTransport(payload)
+}
+
+func (c *ClientCommon) sendFastBulkWrite(ctx context.Context, dataID uint64, startSeq uint64, chunkSize int, fastPathVersion uint8, payload []byte, payloadOwned bool) (int, error) {
+	if len(payload) == 0 {
+		return 0, nil
+	}
+	binding := c.clientTransportBindingSnapshot()
+	if binding == nil {
+		return 0, net.ErrClosed
+	}
+	if sender := binding.clientBulkBatchSenderSnapshot(c); sender != nil {
+		return sender.submitWrite(ctx, dataID, startSeq, fastPathVersion, payload, chunkSize, payloadOwned)
+	}
+	if chunkSize <= 0 {
+		chunkSize = defaultBulkChunkSize
+	}
+	written := 0
+	seq := startSeq
+	for written < len(payload) {
+		end := written + chunkSize
+		if end > len(payload) {
+			end = len(payload)
+		}
+		if err := c.sendFastBulkData(ctx, dataID, seq, payload[written:end], fastPathVersion); err != nil {
+			return written, err
+		}
+		seq++
+		written = end
+	}
+	return written, nil
+}
+
+func (c *ClientCommon) sendFastBulkControl(ctx context.Context, frameType uint8, flags uint8, dataID uint64, seq uint64, fastPathVersion uint8, payload []byte) error {
+	frame := bulkFastFrame{
+		Type:    frameType,
+		Flags:   flags,
+		DataID:  dataID,
+		Seq:     seq,
+		Payload: payload,
+	}
 	binding := c.clientTransportBindingSnapshot()
 	if binding == nil {
 		return net.ErrClosed
 	}
-	if sender := binding.bulkBatchSenderSnapshot(); sender != nil {
-		return sender.submit(ctx, payload)
+	if sender := binding.clientBulkBatchSenderSnapshot(c); sender != nil {
+		return sender.submitControl(ctx, frameType, flags, dataID, seq, fastPathVersion, payload)
 	}
-	return c.writePayloadToTransport(payload)
+	encoded, err := c.encodeBulkFastPayload(frame)
+	if err != nil {
+		return err
+	}
+	return c.writePayloadToTransport(encoded)
 }
 
 func (c *ClientCommon) encodeBulkFastControlPayload(frameType uint8, flags uint8, dataID uint64, seq uint64, payload []byte) ([]byte, error) {
@@ -157,22 +270,81 @@ func (c *ClientCommon) encodeBulkFastControlPayload(frameType uint8, flags uint8
 }
 
 func (s *ServerCommon) encodeFastBulkDataPayloadLogical(logical *LogicalConn, dataID uint64, seq uint64, chunk []byte) ([]byte, error) {
-	if logical != nil {
-		if fastBulkEncode := logical.fastBulkEncodeSnapshot(); fastBulkEncode != nil {
-			return fastBulkEncode(logical.secretKeySnapshot(), dataID, seq, chunk)
-		}
-	}
-	scratch := getBulkFastFrameScratch(len(chunk))
-	defer putBulkFastFrameScratch(scratch)
-	frame := scratch[:bulkFastPayloadHeaderLen+len(chunk)]
-	if err := encodeBulkFastDataFrameHeader(frame, dataID, seq, len(chunk)); err != nil {
-		return nil, err
-	}
-	copy(frame[bulkFastPayloadHeaderLen:], chunk)
-	return s.encryptTransportPayloadLogical(logical, frame)
+	return s.encodeBulkFastPayloadLogical(logical, bulkFastFrame{
+		Type:    bulkFastPayloadTypeData,
+		DataID:  dataID,
+		Seq:     seq,
+		Payload: chunk,
+	})
 }
 
-func (s *ServerCommon) sendFastBulkDataTransport(ctx context.Context, logical *LogicalConn, transport *TransportConn, dataID uint64, seq uint64, chunk []byte) error {
+func (s *ServerCommon) encodeBulkFastPayloadLogical(logical *LogicalConn, frame bulkFastFrame) ([]byte, error) {
+	if logical == nil {
+		return nil, errTransportDetached
+	}
+	if fastPlainEncode := logical.fastPlainEncodeSnapshot(); fastPlainEncode != nil {
+		return encodeBulkFastFramePayloadFast(fastPlainEncode, logical.secretKeySnapshot(), frame)
+	}
+	plain, err := encodeBulkFastFramePayload(frame)
+	if err != nil {
+		return nil, err
+	}
+	return s.encryptTransportPayloadLogical(logical, plain)
+}
+
+func (s *ServerCommon) encodeBulkFastBatchPayloadLogical(logical *LogicalConn, frames []bulkFastFrame) ([]byte, error) {
+	if logical == nil {
+		return nil, errTransportDetached
+	}
+	if fastPlainEncode := logical.fastPlainEncodeSnapshot(); fastPlainEncode != nil {
+		return encodeBulkFastBatchPayloadFast(fastPlainEncode, logical.secretKeySnapshot(), frames)
+	}
+	plain, err := encodeBulkFastBatchPlain(frames)
+	if err != nil {
+		return nil, err
+	}
+	return s.encryptTransportPayloadLogical(logical, plain)
+}
+
+func (s *ServerCommon) encodeBulkFastPayloadLogicalPooled(logical *LogicalConn, frame bulkFastFrame) ([]byte, func(), error) {
+	if logical == nil {
+		return nil, nil, errTransportDetached
+	}
+	if runtime := logical.modernPSKRuntimeSnapshot(); runtime != nil {
+		return encodeBulkFastFramePayloadPooled(runtime, frame)
+	}
+	if fastPlainEncode := logical.fastPlainEncodeSnapshot(); fastPlainEncode != nil {
+		payload, err := encodeBulkFastFramePayloadFast(fastPlainEncode, logical.secretKeySnapshot(), frame)
+		return payload, nil, err
+	}
+	plain, err := encodeBulkFastFramePayload(frame)
+	if err != nil {
+		return nil, nil, err
+	}
+	payload, err := s.encryptTransportPayloadLogical(logical, plain)
+	return payload, nil, err
+}
+
+func (s *ServerCommon) encodeBulkFastBatchPayloadLogicalPooled(logical *LogicalConn, frames []bulkFastFrame) ([]byte, func(), error) {
+	if logical == nil {
+		return nil, nil, errTransportDetached
+	}
+	if runtime := logical.modernPSKRuntimeSnapshot(); runtime != nil {
+		return encodeBulkFastBatchPayloadPooled(runtime, frames)
+	}
+	if fastPlainEncode := logical.fastPlainEncodeSnapshot(); fastPlainEncode != nil {
+		payload, err := encodeBulkFastBatchPayloadFast(fastPlainEncode, logical.secretKeySnapshot(), frames)
+		return payload, nil, err
+	}
+	plain, err := encodeBulkFastBatchPlain(frames)
+	if err != nil {
+		return nil, nil, err
+	}
+	payload, err := s.encryptTransportPayloadLogical(logical, plain)
+	return payload, nil, err
+}
+
+func (s *ServerCommon) sendFastBulkDataTransport(ctx context.Context, logical *LogicalConn, transport *TransportConn, dataID uint64, seq uint64, chunk []byte, fastPathVersion uint8) error {
 	if err := s.ensureServerTransportSendReady(transport); err != nil {
 		return err
 	}
@@ -182,18 +354,87 @@ func (s *ServerCommon) sendFastBulkDataTransport(ctx context.Context, logical *L
 	if logical == nil {
 		return errTransportDetached
 	}
+	if binding := logical.transportBindingSnapshot(); binding != nil {
+		if binding.queueSnapshot() != nil {
+			if sender := binding.serverBulkBatchSenderSnapshot(logical); sender != nil {
+				return sender.submitData(ctx, dataID, seq, fastPathVersion, chunk)
+			}
+		}
+	}
 	payload, err := s.encodeFastBulkDataPayloadLogical(logical, dataID, seq, chunk)
 	if err != nil {
 		return err
 	}
+	return s.writeEnvelopePayload(logical, transport, nil, payload)
+}
+
+func (s *ServerCommon) sendFastBulkWriteTransport(ctx context.Context, logical *LogicalConn, transport *TransportConn, dataID uint64, startSeq uint64, chunkSize int, fastPathVersion uint8, payload []byte, payloadOwned bool) (int, error) {
+	if len(payload) == 0 {
+		return 0, nil
+	}
+	if err := s.ensureServerTransportSendReady(transport); err != nil {
+		return 0, err
+	}
+	if logical == nil && transport != nil {
+		logical = transport.logicalConnSnapshot()
+	}
+	if logical == nil {
+		return 0, errTransportDetached
+	}
 	if binding := logical.transportBindingSnapshot(); binding != nil {
 		if binding.queueSnapshot() != nil {
-			if sender := binding.bulkBatchSenderSnapshot(); sender != nil {
-				return sender.submit(ctx, payload)
+			if sender := binding.serverBulkBatchSenderSnapshot(logical); sender != nil {
+				return sender.submitWrite(ctx, dataID, startSeq, fastPathVersion, payload, chunkSize, payloadOwned)
 			}
 		}
 	}
-	return s.writeEnvelopePayload(logical, transport, nil, payload)
+	if chunkSize <= 0 {
+		chunkSize = defaultBulkChunkSize
+	}
+	written := 0
+	seq := startSeq
+	for written < len(payload) {
+		end := written + chunkSize
+		if end > len(payload) {
+			end = len(payload)
+		}
+		if err := s.sendFastBulkDataTransport(ctx, logical, transport, dataID, seq, payload[written:end], fastPathVersion); err != nil {
+			return written, err
+		}
+		seq++
+		written = end
+	}
+	return written, nil
+}
+
+func (s *ServerCommon) sendFastBulkControlTransport(ctx context.Context, logical *LogicalConn, transport *TransportConn, frameType uint8, flags uint8, dataID uint64, seq uint64, fastPathVersion uint8, payload []byte) error {
+	if err := s.ensureServerTransportSendReady(transport); err != nil {
+		return err
+	}
+	if logical == nil && transport != nil {
+		logical = transport.logicalConnSnapshot()
+	}
+	if logical == nil {
+		return errTransportDetached
+	}
+	if binding := logical.transportBindingSnapshot(); binding != nil {
+		if binding.queueSnapshot() != nil {
+			if sender := binding.serverBulkBatchSenderSnapshot(logical); sender != nil {
+				return sender.submitControl(ctx, frameType, flags, dataID, seq, fastPathVersion, payload)
+			}
+		}
+	}
+	encoded, err := s.encodeBulkFastPayloadLogical(logical, bulkFastFrame{
+		Type:    frameType,
+		Flags:   flags,
+		DataID:  dataID,
+		Seq:     seq,
+		Payload: payload,
+	})
+	if err != nil {
+		return err
+	}
+	return s.writeEnvelopePayload(logical, transport, nil, encoded)
 }
 
 func (s *ServerCommon) encodeBulkFastControlPayloadLogical(logical *LogicalConn, frameType uint8, flags uint8, dataID uint64, seq uint64, payload []byte) ([]byte, error) {
@@ -224,18 +465,22 @@ func (c *ClientCommon) dispatchInboundTransportPayload(payload []byte, now time.
 	if err != nil {
 		return err
 	}
-	if frame, matched, err := decodeBulkFastFrame(plain); matched {
+	if frames, matched, err := decodeBulkFastFrames(plain); matched {
 		if err != nil {
 			return err
 		}
-		c.dispatchFastBulkFrame(frame)
+		for _, frame := range frames {
+			c.dispatchFastBulkFrame(frame)
+		}
 		return nil
 	}
-	if frame, matched, err := decodeStreamFastDataFrame(plain); matched {
+	if frames, matched, err := decodeStreamFastDataFrames(plain); matched {
 		if err != nil {
 			return err
 		}
-		c.dispatchFastStreamData(frame)
+		for _, frame := range frames {
+			c.dispatchFastStreamData(frame)
+		}
 		return nil
 	}
 	env, err := c.decodeEnvelopePlain(plain)
@@ -257,18 +502,22 @@ func (s *ServerCommon) dispatchInboundTransportPayload(logical *LogicalConn, tra
 	if err != nil {
 		return err
 	}
-	if frame, matched, err := decodeBulkFastFrame(plain); matched {
+	if frames, matched, err := decodeBulkFastFrames(plain); matched {
 		if err != nil {
 			return err
 		}
-		s.dispatchFastBulkFrame(logical, transport, conn, frame)
+		for _, frame := range frames {
+			s.dispatchFastBulkFrame(logical, transport, conn, frame)
+		}
 		return nil
 	}
-	if frame, matched, err := decodeStreamFastDataFrame(plain); matched {
+	if frames, matched, err := decodeStreamFastDataFrames(plain); matched {
 		if err != nil {
 			return err
 		}
-		s.dispatchFastStreamData(logical, transport, conn, frame)
+		for _, frame := range frames {
+			s.dispatchFastStreamData(logical, transport, conn, frame)
+		}
 		return nil
 	}
 	env, err := s.decodeEnvelopePlain(plain)

bulk_runtime.go

@@ -2,7 +2,7 @@ package notify
 
 import (
 	"fmt"
-	"strconv"
+	"net"
 	"strings"
 	"sync"
 	"sync/atomic"
@@ -16,14 +16,14 @@ type bulkRuntime struct {
 	mu      sync.RWMutex
 	handler func(BulkAcceptInfo) error
 	bulks   map[string]*bulkHandle
-	data    map[string]*bulkHandle
+	data    map[string]map[uint64]*bulkHandle
 }
 
 func newBulkRuntime(rolePrefix string) *bulkRuntime {
 	return &bulkRuntime{
 		rolePrefix: rolePrefix,
 		bulks:      make(map[string]*bulkHandle),
-		data:       make(map[string]*bulkHandle),
+		data:       make(map[string]map[uint64]*bulkHandle),
 	}
 }
 
@@ -66,8 +66,8 @@ func (r *bulkRuntime) register(scope string, bulk *bulkHandle) error {
 	if bulk == nil || bulk.id == "" {
 		return errBulkIDEmpty
 	}
+	scope = normalizeFileScope(scope)
 	key := bulkRuntimeKey(scope, bulk.id)
-	dataKey := bulkRuntimeDataKey(scope, bulk.dataID)
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	if _, ok := r.bulks[key]; ok {
@@ -76,11 +76,16 @@ func (r *bulkRuntime) register(scope string, bulk *bulkHandle) error {
 	if bulk.dataID == 0 {
 		return errBulkDataIDEmpty
 	}
-	if _, ok := r.data[dataKey]; ok {
+	dataScope := r.data[scope]
+	if dataScope == nil {
+		dataScope = make(map[uint64]*bulkHandle)
+		r.data[scope] = dataScope
+	}
+	if _, ok := dataScope[bulk.dataID]; ok {
 		return errBulkAlreadyExists
 	}
 	r.bulks[key] = bulk
-	r.data[dataKey] = bulk
+	dataScope[bulk.dataID] = bulk
 	return nil
 }
 
@@ -99,10 +104,14 @@ func (r *bulkRuntime) lookupByDataID(scope string, dataID uint64) (*bulkHandle,
 	if r == nil || dataID == 0 {
 		return nil, false
 	}
-	key := bulkRuntimeDataKey(scope, dataID)
+	scope = normalizeFileScope(scope)
 	r.mu.RLock()
 	defer r.mu.RUnlock()
-	bulk, ok := r.data[key]
+	dataScope := r.data[scope]
+	if dataScope == nil {
+		return nil, false
+	}
+	bulk, ok := dataScope[dataID]
 	return bulk, ok
 }
 
@@ -110,11 +119,17 @@ func (r *bulkRuntime) remove(scope string, bulkID string) {
 	if r == nil || bulkID == "" {
 		return
 	}
+	scope = normalizeFileScope(scope)
 	key := bulkRuntimeKey(scope, bulkID)
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	if bulk := r.bulks[key]; bulk != nil && bulk.dataID != 0 {
-		delete(r.data, bulkRuntimeDataKey(scope, bulk.dataID))
+		if dataScope := r.data[scope]; dataScope != nil {
+			delete(dataScope, bulk.dataID)
+			if len(dataScope) == 0 {
+				delete(r.data, scope)
+			}
+		}
 	}
 	delete(r.bulks, key)
 }
@@ -149,6 +164,140 @@ func (r *bulkRuntime) closeMatching(match func(string) bool, err error) {
 	}
 }
 
+func (r *bulkRuntime) resetDedicatedByConn(scope string, conn net.Conn, err error) {
+	if r == nil || conn == nil {
+		return
+	}
+	scope = normalizeFileScope(scope)
+	resetErr := bulkRuntimeCloseError(err)
+	prefix := scope + "\x00"
+	r.mu.RLock()
+	bulks := make([]*bulkHandle, 0, len(r.bulks))
+	for key, bulk := range r.bulks {
+		if bulk == nil {
+			continue
+		}
+		if !strings.HasPrefix(key, prefix) {
+			continue
+		}
+		if !bulk.Dedicated() {
+			continue
+		}
+		if bulk.dedicatedConnSnapshot() != conn {
+			continue
+		}
+		bulks = append(bulks, bulk)
+	}
+	r.mu.RUnlock()
+	for _, bulk := range bulks {
+		bulk.markReset(resetErr)
+	}
+}
+
+func (r *bulkRuntime) handleDedicatedReadErrorByConn(scope string, conn net.Conn, err error) {
+	if r == nil || conn == nil {
+		return
+	}
+	scope = normalizeFileScope(scope)
+	prefix := scope + "\x00"
+	r.mu.RLock()
+	bulks := make([]*bulkHandle, 0, len(r.bulks))
+	for key, bulk := range r.bulks {
+		if bulk == nil {
+			continue
+		}
+		if !strings.HasPrefix(key, prefix) {
+			continue
+		}
+		if !bulk.Dedicated() {
+			continue
+		}
+		if bulk.dedicatedConnSnapshot() != conn {
+			continue
+		}
+		bulks = append(bulks, bulk)
+	}
+	r.mu.RUnlock()
+	for _, bulk := range bulks {
+		handleDedicatedBulkReadError(bulk, err)
+	}
+}
+
+func (r *bulkRuntime) attachSharedDedicatedConn(scope string, laneID uint32, conn net.Conn) {
+	if r == nil || conn == nil {
+		return
+	}
+	scope = normalizeFileScope(scope)
+	laneID = normalizeBulkDedicatedLaneID(laneID)
+	prefix := scope + "\x00"
+	r.mu.RLock()
+	bulks := make([]*bulkHandle, 0, len(r.bulks))
+	for key, bulk := range r.bulks {
+		if bulk == nil {
+			continue
+		}
+		if !strings.HasPrefix(key, prefix) {
+			continue
+		}
+		if !bulk.Dedicated() {
+			continue
+		}
+		if bulk.dedicatedLaneIDSnapshot() != laneID {
+			continue
+		}
+		bulks = append(bulks, bulk)
+	}
+	r.mu.RUnlock()
+	for _, bulk := range bulks {
+		current := bulk.dedicatedConnSnapshot()
+		switch {
+		case current == conn:
+			_ = bulk.attachDedicatedConnShared(conn)
+		case current == nil:
+			_ = bulk.attachDedicatedConnShared(conn)
+		default:
+			oldConn, oldSender, err := bulk.replaceDedicatedConnShared(conn)
+			if err != nil {
+				bulk.markReset(err)
+				continue
+			}
+			if oldSender != nil {
+				oldSender.stop()
+			}
+			if oldConn != nil {
+				_ = oldConn.Close()
+			}
+		}
+	}
+}
+
+func (r *bulkRuntime) dedicatedBulksForConn(scope string, conn net.Conn) []*bulkHandle {
+	if r == nil || conn == nil {
+		return nil
+	}
+	scope = normalizeFileScope(scope)
+	prefix := scope + "\x00"
+	r.mu.RLock()
+	bulks := make([]*bulkHandle, 0, len(r.bulks))
+	for key, bulk := range r.bulks {
+		if bulk == nil {
+			continue
+		}
+		if !strings.HasPrefix(key, prefix) {
+			continue
+		}
+		if !bulk.Dedicated() {
+			continue
+		}
+		if bulk.dedicatedConnSnapshot() != conn {
+			continue
+		}
+		bulks = append(bulks, bulk)
+	}
+	r.mu.RUnlock()
+	return bulks
+}
+
 func (r *bulkRuntime) snapshots() []BulkSnapshot {
 	if r == nil {
 		return nil
@@ -170,10 +319,6 @@ func bulkRuntimeKey(scope string, bulkID string) string {
 	return normalizeFileScope(scope) + "\x00" + bulkID
 }
 
-func bulkRuntimeDataKey(scope string, dataID uint64) string {
-	return normalizeFileScope(scope) + "\x01" + strconv.FormatUint(dataID, 10)
-}
-
 func bulkRuntimeCloseError(err error) error {
 	if err != nil {
 		return err

bulk_shared_batch.go

@@ -0,0 +1,228 @@
+package notify
+
+import (
+	"encoding/binary"
+)
+
+const (
+	bulkFastPathVersionV1      = 1
+	bulkFastPathVersionV2      = 2
+	bulkFastPathVersionCurrent = bulkFastPathVersionV2
+)
+
+const (
+	bulkFastBatchMagic         = "NBF2"
+	bulkFastBatchVersion       = 1
+	bulkFastBatchHeaderLen     = 12
+	bulkFastBatchItemHeaderLen = 24
+	bulkFastBatchMaxItems      = 64
+	bulkFastBatchMaxPlainBytes = 8 * 1024 * 1024
+)
+
+func normalizeBulkFastPathVersion(version uint8) uint8 {
+	if version < bulkFastPathVersionV1 {
+		return bulkFastPathVersionV1
+	}
+	if version > bulkFastPathVersionCurrent {
+		return bulkFastPathVersionCurrent
+	}
+	return version
+}
+
+func negotiateBulkFastPathVersion(version uint8) uint8 {
+	return normalizeBulkFastPathVersion(version)
+}
+
+func bulkFastPathSupportsSharedBatch(version uint8) bool {
+	return normalizeBulkFastPathVersion(version) >= bulkFastPathVersionV2
+}
+
+func bulkFastBatchFrameLen(frame bulkFastFrame) int {
+	return bulkFastBatchItemHeaderLen + len(frame.Payload)
+}
+
+func bulkFastBatchPlainLen(frames []bulkFastFrame) int {
+	total := bulkFastBatchHeaderLen
+	for _, frame := range frames {
+		total += bulkFastBatchFrameLen(frame)
+	}
+	return total
+}
+
+func encodeBulkFastFramePayload(frame bulkFastFrame) ([]byte, error) {
+	return encodeBulkFastControlFrame(frame.Type, frame.Flags, frame.DataID, frame.Seq, frame.Payload)
+}
+
+func encodeBulkFastFramePayloadFast(encode transportFastPlainEncoder, secretKey []byte, frame bulkFastFrame) ([]byte, error) {
+	if encode == nil {
+		return nil, errTransportPayloadEncryptFailed
+	}
+	plainLen := bulkFastPayloadHeaderLen + len(frame.Payload)
+	return encode(secretKey, plainLen, func(dst []byte) error {
+		if err := encodeBulkFastFrameHeader(dst, frame.Type, frame.Flags, frame.DataID, frame.Seq, len(frame.Payload)); err != nil {
+			return err
+		}
+		copy(dst[bulkFastPayloadHeaderLen:], frame.Payload)
+		return nil
+	})
+}
+
+func encodeBulkFastFramePayloadPooled(runtime *modernPSKCodecRuntime, frame bulkFastFrame) ([]byte, func(), error) {
+	if runtime == nil {
+		return nil, nil, errTransportPayloadEncryptFailed
+	}
+	plainLen := bulkFastPayloadHeaderLen + len(frame.Payload)
+	return runtime.sealFilledPayloadPooled(plainLen, func(dst []byte) error {
+		if err := encodeBulkFastFrameHeader(dst, frame.Type, frame.Flags, frame.DataID, frame.Seq, len(frame.Payload)); err != nil {
+			return err
+		}
+		copy(dst[bulkFastPayloadHeaderLen:], frame.Payload)
+		return nil
+	})
+}
+
+func encodeBulkFastBatchPlain(frames []bulkFastFrame) ([]byte, error) {
+	if len(frames) == 0 {
+		return nil, errBulkFastPayloadInvalid
+	}
+	buf := make([]byte, bulkFastBatchPlainLen(frames))
+	if err := writeBulkFastBatchPlain(buf, frames); err != nil {
+		return nil, err
+	}
+	return buf, nil
+}
+
+func encodeBulkFastBatchPayloadFast(encode transportFastPlainEncoder, secretKey []byte, frames []bulkFastFrame) ([]byte, error) {
+	if encode == nil {
+		return nil, errTransportPayloadEncryptFailed
+	}
+	plainLen := bulkFastBatchPlainLen(frames)
+	return encode(secretKey, plainLen, func(dst []byte) error {
+		return writeBulkFastBatchPlain(dst, frames)
+	})
+}
+
+func encodeBulkFastBatchPayloadPooled(runtime *modernPSKCodecRuntime, frames []bulkFastFrame) ([]byte, func(), error) {
+	if runtime == nil {
+		return nil, nil, errTransportPayloadEncryptFailed
+	}
+	return runtime.sealFilledPayloadPooled(bulkFastBatchPlainLen(frames), func(dst []byte) error {
+		return writeBulkFastBatchPlain(dst, frames)
+	})
+}
+
+func writeBulkFastBatchPlain(dst []byte, frames []bulkFastFrame) error {
+	if len(frames) == 0 || len(dst) != bulkFastBatchPlainLen(frames) {
+		return errBulkFastPayloadInvalid
+	}
+	copy(dst[:4], bulkFastBatchMagic)
+	dst[4] = bulkFastBatchVersion
+	binary.BigEndian.PutUint32(dst[8:12], uint32(len(frames)))
+	offset := bulkFastBatchHeaderLen
+	for _, frame := range frames {
+		if frame.DataID == 0 {
+			return errBulkFastPayloadInvalid
+		}
+		dst[offset] = frame.Type
+		dst[offset+1] = frame.Flags
+		binary.BigEndian.PutUint64(dst[offset+4:offset+12], frame.DataID)
+		binary.BigEndian.PutUint64(dst[offset+12:offset+20], frame.Seq)
+		binary.BigEndian.PutUint32(dst[offset+20:offset+24], uint32(len(frame.Payload)))
+		offset += bulkFastBatchItemHeaderLen
+		copy(dst[offset:offset+len(frame.Payload)], frame.Payload)
+		offset += len(frame.Payload)
+	}
+	return nil
+}
+
+func walkBulkFastBatchPlain(payload []byte, fn func(bulkFastFrame) error) (bool, error) {
+	if len(payload) < 4 || string(payload[:4]) != bulkFastBatchMagic {
+		return false, nil
+	}
+	if len(payload) < bulkFastBatchHeaderLen {
+		return true, errBulkFastPayloadInvalid
+	}
+	if payload[4] != bulkFastBatchVersion {
+		return true, errBulkFastPayloadInvalid
+	}
+	count := int(binary.BigEndian.Uint32(payload[8:12]))
+	if count <= 0 {
+		return true, errBulkFastPayloadInvalid
+	}
+	offset := bulkFastBatchHeaderLen
+	for index := 0; index < count; index++ {
+		if len(payload)-offset < bulkFastBatchItemHeaderLen {
+			return true, errBulkFastPayloadInvalid
+		}
+		frameType := payload[offset]
+		switch frameType {
+		case bulkFastPayloadTypeData, bulkFastPayloadTypeClose, bulkFastPayloadTypeReset, bulkFastPayloadTypeRelease:
+		default:
+			return true, errBulkFastPayloadInvalid
+		}
+		flags := payload[offset+1]
+		dataID := binary.BigEndian.Uint64(payload[offset+4 : offset+12])
+		seq := binary.BigEndian.Uint64(payload[offset+12 : offset+20])
+		payloadLen := int(binary.BigEndian.Uint32(payload[offset+20 : offset+24]))
+		offset += bulkFastBatchItemHeaderLen
+		if dataID == 0 || payloadLen < 0 || len(payload)-offset < payloadLen {
+			return true, errBulkFastPayloadInvalid
+		}
+		if fn != nil {
+			if err := fn(bulkFastFrame{
+				Type:    frameType,
+				Flags:   flags,
+				DataID:  dataID,
+				Seq:     seq,
+				Payload: payload[offset : offset+payloadLen],
+			}); err != nil {
+				return true, err
+			}
+		}
+		offset += payloadLen
+	}
+	if offset != len(payload) {
+		return true, errBulkFastPayloadInvalid
+	}
+	return true, nil
+}
+
+func decodeBulkFastBatchPlain(payload []byte) ([]bulkFastFrame, bool, error) {
+	frames := make([]bulkFastFrame, 0, 1)
+	matched, err := walkBulkFastBatchPlain(payload, func(frame bulkFastFrame) error {
+		frames = append(frames, frame)
+		return nil
+	})
+	if !matched || err != nil {
+		return nil, matched, err
+	}
+	return frames, true, nil
+}
+
+func walkBulkFastFrames(payload []byte, fn func(bulkFastFrame) error) (bool, error) {
+	if matched, err := walkBulkFastBatchPlain(payload, fn); matched {
+		return true, err
+	}
+	frame, matched, err := decodeBulkFastFrame(payload)
+	if !matched || err != nil {
+		return matched, err
+	}
+	if fn != nil {
+		if err := fn(frame); err != nil {
+			return true, err
+		}
+	}
+	return true, nil
+}
+
+func decodeBulkFastFrames(payload []byte) ([]bulkFastFrame, bool, error) {
+	frames := make([]bulkFastFrame, 0, 1)
+	matched, err := walkBulkFastFrames(payload, func(frame bulkFastFrame) error {
+		frames = append(frames, frame)
+		return nil
+	})
+	if !matched || err != nil {
+		return nil, matched, err
+	}
+	return frames, true, nil
+}

bulk_shared_batch_test.go

@@ -0,0 +1,458 @@
+package notify
+
+import (
+	"context"
+	"testing"
+	"time"
+)
+
+func TestBulkFastBatchPlainRoundTrip(t *testing.T) {
+	releasePayload, err := encodeBulkDedicatedReleasePayload(4096, 2)
+	if err != nil {
+		t.Fatalf("encodeBulkDedicatedReleasePayload failed: %v", err)
+	}
+	frames := []bulkFastFrame{
+		{
+			Type:    bulkFastPayloadTypeData,
+			DataID:  11,
+			Seq:     7,
+			Payload: []byte("alpha"),
+		},
+		{
+			Type:    bulkFastPayloadTypeRelease,
+			DataID:  12,
+			Seq:     0,
+			Payload: releasePayload,
+		},
+	}
+	wire, err := encodeBulkFastBatchPlain(frames)
+	if err != nil {
+		t.Fatalf("encodeBulkFastBatchPlain failed: %v", err)
+	}
+	decoded, matched, err := decodeBulkFastBatchPlain(wire)
+	if err != nil {
+		t.Fatalf("decodeBulkFastBatchPlain failed: %v", err)
+	}
+	if !matched {
+		t.Fatal("decodeBulkFastBatchPlain should match encoded batch")
+	}
+	if got, want := len(decoded), len(frames); got != want {
+		t.Fatalf("decoded frame count = %d, want %d", got, want)
+	}
+	for index := range frames {
+		if got, want := decoded[index].Type, frames[index].Type; got != want {
+			t.Fatalf("frame %d type = %d, want %d", index, got, want)
+		}
+		if got, want := decoded[index].DataID, frames[index].DataID; got != want {
+			t.Fatalf("frame %d dataID = %d, want %d", index, got, want)
+		}
+		if got, want := decoded[index].Seq, frames[index].Seq; got != want {
+			t.Fatalf("frame %d seq = %d, want %d", index, got, want)
+		}
+		if got, want := string(decoded[index].Payload), string(frames[index].Payload); got != want {
+			t.Fatalf("frame %d payload = %q, want %q", index, got, want)
+		}
+	}
+}
+
+func TestBulkBatchSenderEncodeRequestsCoalescesSharedFastV2Frames(t *testing.T) {
+	var (
+		singleCalls int
+		batchCalls  [][]bulkFastFrame
+	)
+	sender := &bulkBatchSender{
+		codec: bulkBatchCodec{
+			encodeSingle: func(frame bulkFastFrame) ([]byte, func(), error) {
+				singleCalls++
+				return []byte("single"), nil, nil
+			},
+			encodeBatch: func(frames []bulkFastFrame) ([]byte, func(), error) {
+				cloned := make([]bulkFastFrame, len(frames))
+				copy(cloned, frames)
+				batchCalls = append(batchCalls, cloned)
+				return []byte("batch"), nil, nil
+			},
+		},
+	}
+	payloads, err := sender.encodeRequests([]bulkBatchRequest{
+		{
+			frames: []bulkFastFrame{{
+				Type:    bulkFastPayloadTypeData,
+				DataID:  101,
+				Seq:     1,
+				Payload: []byte("a"),
+			}},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+		{
+			frames: []bulkFastFrame{{
+				Type:    bulkFastPayloadTypeRelease,
+				DataID:  101,
+				Seq:     0,
+				Payload: []byte("rel"),
+			}},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("encodeRequests failed: %v", err)
+	}
+	if got, want := len(payloads), 1; got != want {
+		t.Fatalf("payload count = %d, want %d", got, want)
+	}
+	if singleCalls != 0 {
+		t.Fatalf("single encode calls = %d, want 0", singleCalls)
+	}
+	if got, want := len(batchCalls), 1; got != want {
+		t.Fatalf("batch encode calls = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls[0]), 2; got != want {
+		t.Fatalf("batched item count = %d, want %d", got, want)
+	}
+}
+
+func TestBulkBatchSenderEncodeRequestsCoalescesAcrossRequestsAndBulks(t *testing.T) {
+	var (
+		singleCalls int
+		batchCalls  [][]bulkFastFrame
+	)
+	sender := &bulkBatchSender{
+		codec: bulkBatchCodec{
+			encodeSingle: func(frame bulkFastFrame) ([]byte, func(), error) {
+				singleCalls++
+				return []byte("single"), nil, nil
+			},
+			encodeBatch: func(frames []bulkFastFrame) ([]byte, func(), error) {
+				cloned := make([]bulkFastFrame, len(frames))
+				copy(cloned, frames)
+				batchCalls = append(batchCalls, cloned)
+				return []byte("batch"), nil, nil
+			},
+		},
+	}
+	payloads, err := sender.encodeRequests([]bulkBatchRequest{
+		{
+			frames: []bulkFastFrame{{
+				Type:    bulkFastPayloadTypeData,
+				DataID:  101,
+				Seq:     1,
+				Payload: []byte("bulk-a"),
+			}},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+		{
+			frames: []bulkFastFrame{{
+				Type:    bulkFastPayloadTypeData,
+				DataID:  202,
+				Seq:     1,
+				Payload: []byte("bulk-b"),
+			}},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("encodeRequests failed: %v", err)
+	}
+	if got, want := len(payloads), 1; got != want {
+		t.Fatalf("payload count = %d, want %d", got, want)
+	}
+	if singleCalls != 0 {
+		t.Fatalf("single encode calls = %d, want 0", singleCalls)
+	}
+	if got, want := len(batchCalls), 1; got != want {
+		t.Fatalf("batch encode calls = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls[0]), 2; got != want {
+		t.Fatalf("batched item count = %d, want %d", got, want)
+	}
+	if got, want := batchCalls[0][0].DataID, uint64(101); got != want {
+		t.Fatalf("first batched dataID = %d, want %d", got, want)
+	}
+	if got, want := batchCalls[0][1].DataID, uint64(202); got != want {
+		t.Fatalf("second batched dataID = %d, want %d", got, want)
+	}
+}
+
+func TestBulkBatchSenderEncodeRequestsSplitsLargeCrossBulkSuperBatch(t *testing.T) {
+	var batchCalls [][]bulkFastFrame
+	sender := &bulkBatchSender{
+		codec: bulkBatchCodec{
+			encodeSingle: func(frame bulkFastFrame) ([]byte, func(), error) {
+				return []byte("single"), nil, nil
+			},
+			encodeBatch: func(frames []bulkFastFrame) ([]byte, func(), error) {
+				cloned := make([]bulkFastFrame, len(frames))
+				copy(cloned, frames)
+				batchCalls = append(batchCalls, cloned)
+				return []byte("batch"), nil, nil
+			},
+		},
+	}
+	payload := make([]byte, 768*1024)
+	payloads, err := sender.encodeRequests([]bulkBatchRequest{
+		{
+			frames: []bulkFastFrame{
+				{
+					Type:    bulkFastPayloadTypeData,
+					DataID:  101,
+					Seq:     1,
+					Payload: payload,
+				},
+				{
+					Type:    bulkFastPayloadTypeData,
+					DataID:  101,
+					Seq:     2,
+					Payload: payload,
+				},
+			},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+		{
+			frames: []bulkFastFrame{
+				{
+					Type:    bulkFastPayloadTypeData,
+					DataID:  202,
+					Seq:     1,
+					Payload: payload,
+				},
+				{
+					Type:    bulkFastPayloadTypeData,
+					DataID:  202,
+					Seq:     2,
+					Payload: payload,
+				},
+			},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("encodeRequests failed: %v", err)
+	}
+	if got, want := len(payloads), 2; got != want {
+		t.Fatalf("payload count = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls), 2; got != want {
+		t.Fatalf("batch encode calls = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls[0]), 2; got != want {
+		t.Fatalf("first payload frame count = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls[1]), 2; got != want {
+		t.Fatalf("second payload frame count = %d, want %d", got, want)
+	}
+	if got, want := batchCalls[0][0].DataID, uint64(101); got != want {
+		t.Fatalf("first payload dataID = %d, want %d", got, want)
+	}
+	if got, want := batchCalls[1][0].DataID, uint64(202); got != want {
+		t.Fatalf("second payload dataID = %d, want %d", got, want)
+	}
+}
+
+func TestBulkReleaseFastRoundTripTransport(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	if err := UseModernPSKServer(server, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKServer failed: %v", err)
+	}
+
+	acceptCh := make(chan BulkAcceptInfo, 1)
+	server.SetBulkHandler(func(info BulkAcceptInfo) error {
+		acceptCh <- info
+		return nil
+	})
+
+	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+		t.Fatalf("server Listen failed: %v", err)
+	}
+	defer func() {
+		_ = server.Stop()
+	}()
+
+	client := NewClient().(*ClientCommon)
+	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKClient failed: %v", err)
+	}
+	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+		t.Fatalf("client Connect failed: %v", err)
+	}
+	defer func() {
+		_ = client.Stop()
+	}()
+
+	const chunkSize = 64 * 1024
+	bulk, err := client.OpenBulk(context.Background(), BulkOpenOptions{
+		Range:       BulkRange{Offset: 0, Length: chunkSize},
+		ChunkSize:   chunkSize,
+		WindowBytes: chunkSize,
+		MaxInFlight: 1,
+	})
+	if err != nil {
+		t.Fatalf("client OpenBulk failed: %v", err)
+	}
+
+	accepted := waitAcceptedBulk(t, acceptCh, 2*time.Second)
+	clientHandle := bulk.(*bulkHandle)
+	serverHandle := accepted.Bulk.(*bulkHandle)
+	if got, want := clientHandle.FastPathVersion(), uint8(bulkFastPathVersionV2); got != want {
+		t.Fatalf("client fast path version = %d, want %d", got, want)
+	}
+	if got, want := serverHandle.FastPathVersion(), uint8(bulkFastPathVersionV2); got != want {
+		t.Fatalf("server fast path version = %d, want %d", got, want)
+	}
+
+	clientHandle.mu.Lock()
+	clientHandle.outboundAvailBytes = 0
+	clientHandle.outboundInFlight = 1
+	clientHandle.mu.Unlock()
+
+	releaseFn := serverBulkReleaseSender(server, accepted.LogicalConn, accepted.TransportConn)
+	if err := releaseFn(serverHandle, chunkSize, 1); err != nil {
+		t.Fatalf("server bulk release sender failed: %v", err)
+	}
+
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		clientHandle.mu.Lock()
+		avail := clientHandle.outboundAvailBytes
+		inFlight := clientHandle.outboundInFlight
+		clientHandle.mu.Unlock()
+		if avail == chunkSize && inFlight == 0 {
+			return
+		}
+		time.Sleep(10 * time.Millisecond)
+	}
+
+	clientHandle.mu.Lock()
+	avail := clientHandle.outboundAvailBytes
+	inFlight := clientHandle.outboundInFlight
+	clientHandle.mu.Unlock()
+	t.Fatalf("client outbound window not released by fast path: avail=%d inFlight=%d", avail, inFlight)
+}
+
+func TestBulkBatchSenderEncodeRequestsResetsBatchBytesAfterFlushBoundary(t *testing.T) {
+	var (
+		singleCalls int
+		batchCalls  [][]bulkFastFrame
+	)
+	sender := &bulkBatchSender{
+		codec: bulkBatchCodec{
+			encodeSingle: func(frame bulkFastFrame) ([]byte, func(), error) {
+				singleCalls++
+				return []byte("single"), nil, nil
+			},
+			encodeBatch: func(frames []bulkFastFrame) ([]byte, func(), error) {
+				cloned := make([]bulkFastFrame, len(frames))
+				copy(cloned, frames)
+				batchCalls = append(batchCalls, cloned)
+				return []byte("batch"), nil, nil
+			},
+		},
+	}
+	largePayload := make([]byte, bulkFastBatchMaxPlainBytes-bulkFastBatchHeaderLen-bulkFastBatchItemHeaderLen-128)
+	payloads, err := sender.encodeRequests([]bulkBatchRequest{
+		{
+			frames: []bulkFastFrame{{
+				Type:    bulkFastPayloadTypeData,
+				DataID:  101,
+				Seq:     1,
+				Payload: largePayload,
+			}},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+		{
+			frames: []bulkFastFrame{{
+				Type:    bulkFastPayloadTypeData,
+				DataID:  101,
+				Seq:     2,
+				Payload: []byte("sep"),
+			}},
+			fastPathVersion: bulkFastPathVersionV1,
+		},
+		{
+			frames: []bulkFastFrame{{
+				Type:    bulkFastPayloadTypeData,
+				DataID:  202,
+				Seq:     1,
+				Payload: []byte("a"),
+			}},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+		{
+			frames: []bulkFastFrame{{
+				Type:    bulkFastPayloadTypeData,
+				DataID:  202,
+				Seq:     2,
+				Payload: []byte("b"),
+			}},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("encodeRequests failed: %v", err)
+	}
+	if got, want := len(payloads), 3; got != want {
+		t.Fatalf("payload count = %d, want %d", got, want)
+	}
+	if got, want := singleCalls, 2; got != want {
+		t.Fatalf("single encode calls = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls), 1; got != want {
+		t.Fatalf("batch encode calls = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls[0]), 2; got != want {
+		t.Fatalf("post-flush batched frame count = %d, want %d", got, want)
+	}
+}
+
+func TestBulkBatchSenderEncodeRequestsUsesBindingAdaptiveSoftLimit(t *testing.T) {
+	binding := &transportBinding{}
+	binding.observeBulkAdaptivePayloadWrite(8*1024*1024, 640*time.Millisecond, 0, nil)
+	var batchCalls [][]bulkFastFrame
+	sender := &bulkBatchSender{
+		binding: binding,
+		codec: bulkBatchCodec{
+			encodeSingle: func(frame bulkFastFrame) ([]byte, func(), error) {
+				return []byte("single"), nil, nil
+			},
+			encodeBatch: func(frames []bulkFastFrame) ([]byte, func(), error) {
+				cloned := make([]bulkFastFrame, len(frames))
+				copy(cloned, frames)
+				batchCalls = append(batchCalls, cloned)
+				return []byte("batch"), nil, nil
+			},
+		},
+	}
+	payload := make([]byte, 96*1024)
+	payloads, err := sender.encodeRequests([]bulkBatchRequest{
+		{
+			frames: []bulkFastFrame{
+				{Type: bulkFastPayloadTypeData, DataID: 101, Seq: 1, Payload: payload},
+				{Type: bulkFastPayloadTypeData, DataID: 101, Seq: 2, Payload: payload},
+			},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+		{
+			frames: []bulkFastFrame{
+				{Type: bulkFastPayloadTypeData, DataID: 202, Seq: 1, Payload: payload},
+				{Type: bulkFastPayloadTypeData, DataID: 202, Seq: 2, Payload: payload},
+			},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("encodeRequests failed: %v", err)
+	}
+	if got, want := len(payloads), 2; got != want {
+		t.Fatalf("payload count = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls), 2; got != want {
+		t.Fatalf("batch encode calls = %d, want %d", got, want)
+	}
+	for index, want := range []uint64{101, 202} {
+		if got := batchCalls[index][0].DataID; got != want {
+			t.Fatalf("payload %d first dataID = %d, want %d", index, got, want)
+		}
+		if got, wantItems := len(batchCalls[index]), 2; got != wantItems {
+			t.Fatalf("payload %d item count = %d, want %d", index, got, wantItems)
+		}
+	}
+}

bulk_snapshot.go

@@ -7,49 +7,56 @@ import (
 )
 
 type BulkSnapshot struct {
-	ID                        string
-	DataID                    uint64
-	Scope                     string
-	Range                     BulkRange
-	Metadata                  BulkMetadata
-	BindingOwner              string
-	BindingAlive              bool
-	BindingCurrent            bool
-	BindingReason             string
-	BindingError              string
-	Dedicated                 bool
-	DedicatedAttached         bool
-	SessionEpoch              uint64
-	LogicalClientID           string
-	TransportGeneration       uint64
-	TransportAttached         bool
-	TransportHasRuntimeConn   bool
-	TransportCurrent          bool
-	TransportDetachReason     string
-	TransportDetachKind       string
-	TransportDetachGeneration uint64
-	TransportDetachError      string
-	TransportDetachedAt       time.Time
-	ReattachEligible          bool
-	LocalClosed               bool
-	LocalReadClosed           bool
-	RemoteClosed              bool
-	PeerReadClosed            bool
-	BufferedChunks            int
-	BufferedBytes             int
-	ReadTimeout               time.Duration
-	WriteTimeout              time.Duration
-	ChunkSize                 int
-	WindowBytes               int
-	MaxInFlight               int
-	BytesRead                 int64
-	BytesWritten              int64
-	ReadCalls                 int64
-	WriteCalls                int64
-	OpenedAt                  time.Time
-	LastReadAt                time.Time
-	LastWriteAt               time.Time
-	ResetError                string
+	ID                                  string
+	DataID                              uint64
+	FastPathVersion                     uint8
+	Scope                               string
+	Range                               BulkRange
+	Metadata                            BulkMetadata
+	BindingOwner                        string
+	BindingAlive                        bool
+	BindingCurrent                      bool
+	BindingReason                       string
+	BindingError                        string
+	BindingBulkAdaptiveSoftPayloadBytes int
+	Dedicated                           bool
+	DedicatedLaneID                     uint32
+	DedicatedAttached                   bool
+	DedicatedAttachState                string
+	DedicatedAttachAttempts             uint32
+	DedicatedAttachLastCode             string
+	DedicatedDataStarted                bool
+	SessionEpoch                        uint64
+	LogicalClientID                     string
+	TransportGeneration                 uint64
+	TransportAttached                   bool
+	TransportHasRuntimeConn             bool
+	TransportCurrent                    bool
+	TransportDetachReason               string
+	TransportDetachKind                 string
+	TransportDetachGeneration           uint64
+	TransportDetachError                string
+	TransportDetachedAt                 time.Time
+	ReattachEligible                    bool
+	LocalClosed                         bool
+	LocalReadClosed                     bool
+	RemoteClosed                        bool
+	PeerReadClosed                      bool
+	BufferedChunks                      int
+	BufferedBytes                       int
+	ReadTimeout                         time.Duration
+	WriteTimeout                        time.Duration
+	ChunkSize                           int
+	WindowBytes                         int
+	MaxInFlight                         int
+	BytesRead                           int64
+	BytesWritten                        int64
+	ReadCalls                           int64
+	WriteCalls                          int64
+	OpenedAt                            time.Time
+	LastReadAt                          time.Time
+	LastWriteAt                         time.Time
+	ResetError                          string
 }
 
 type clientBulkSnapshotReader interface {

bulk_stack_benchmark_test.go

@@ -88,7 +88,7 @@ func BenchmarkDedicatedWireLocalhostThroughput(b *testing.B) {
 func benchmarkDedicatedWireLocalhostThroughput(b *testing.B, key []byte, encode transportFastPlainEncoder, payloadSize int) {
 	b.Helper()
 
-	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	listener, err := net.Listen("tcp", benchmarkTCPListenAddr(b))
 	if err != nil {
 		b.Fatalf("net.Listen failed: %v", err)
 	}
@@ -107,7 +107,7 @@ func benchmarkDedicatedWireLocalhostThroughput(b *testing.B, key []byte, encode
 		acceptCh <- conn
 	}()
 
-	clientConn, err := net.Dial("tcp", listener.Addr().String())
+	clientConn, err := net.Dial("tcp", benchmarkTCPDialAddr(b, listener.Addr().String()))
 	if err != nil {
 		b.Fatalf("net.Dial failed: %v", err)
 	}

bulk_test.go

@@ -109,6 +109,9 @@ func TestBulkOpenRoundTripTCP(t *testing.T) {
 	if !clientSnapshots[0].BindingAlive || !clientSnapshots[0].BindingCurrent || !clientSnapshots[0].TransportAttached || !clientSnapshots[0].TransportCurrent {
 		t.Fatalf("client bulk binding snapshot mismatch: %+v", clientSnapshots[0])
 	}
+	if got, want := clientSnapshots[0].BindingBulkAdaptiveSoftPayloadBytes, bulkAdaptiveSoftPayloadStartBytes; got != want {
+		t.Fatalf("client bulk BindingBulkAdaptiveSoftPayloadBytes = %d, want %d", got, want)
+	}
 	serverSnapshots, err := GetServerBulkSnapshots(server)
 	if err != nil {
 		t.Fatalf("GetServerBulkSnapshots failed: %v", err)
@@ -119,6 +122,9 @@ func TestBulkOpenRoundTripTCP(t *testing.T) {
 	if got, want := serverSnapshots[0].BindingOwner, "server-transport"; got != want {
 		t.Fatalf("server bulk BindingOwner = %q, want %q", got, want)
 	}
+	if got, want := serverSnapshots[0].BindingBulkAdaptiveSoftPayloadBytes, bulkAdaptiveSoftPayloadStartBytes; got != want {
+		t.Fatalf("server bulk BindingBulkAdaptiveSoftPayloadBytes = %d, want %d", got, want)
+	}
 	if !serverSnapshots[0].BindingAlive || !serverSnapshots[0].BindingCurrent || !serverSnapshots[0].TransportAttached || !serverSnapshots[0].TransportCurrent {
 		t.Fatalf("server bulk binding snapshot mismatch: %+v", serverSnapshots[0])
 	}
@@ -135,6 +141,314 @@ func TestBulkOpenRoundTripTCP(t *testing.T) {
 	waitForBulkContextDone(t, bulk.Context(), 2*time.Second)
 }
 
+func TestDedicatedBulkOpenUnblocksSynchronousReadHandler(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	if err := UseModernPSKServer(server, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKServer failed: %v", err)
+	}
+
+	readCh := make(chan string, 1)
+	server.SetBulkHandler(func(info BulkAcceptInfo) error {
+		defer func() {
+			_ = info.Bulk.Close()
+		}()
+		buf := make([]byte, 5)
+		if _, err := io.ReadFull(info.Bulk, buf); err != nil {
+			return err
+		}
+		readCh <- string(buf)
+		return nil
+	})
+
+	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+		t.Fatalf("server Listen failed: %v", err)
+	}
+	defer func() {
+		_ = server.Stop()
+	}()
+
+	client := NewClient().(*ClientCommon)
+	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKClient failed: %v", err)
+	}
+	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+		t.Fatalf("client Connect failed: %v", err)
+	}
+	defer func() {
+		_ = client.Stop()
+	}()
+
+	type openResult struct {
+		bulk Bulk
+		err  error
+	}
+	openCh := make(chan openResult, 1)
+	go func() {
+		bulk, err := client.OpenDedicatedBulk(context.Background(), BulkOpenOptions{
+			ID: "sync-read-handler",
+			Range: BulkRange{
+				Offset: 0,
+				Length: 5,
+			},
+		})
+		openCh <- openResult{bulk: bulk, err: err}
+	}()
+
+	var bulk Bulk
+	select {
+	case result := <-openCh:
+		if result.err != nil {
+			t.Fatalf("client OpenDedicatedBulk failed: %v", result.err)
+		}
+		bulk = result.bulk
+	case <-time.After(2 * time.Second):
+		t.Fatal("client OpenDedicatedBulk timed out while remote handler was synchronously reading")
+	}
+	defer func() {
+		if bulk != nil {
+			_ = bulk.Close()
+		}
+	}()
+
+	if _, err := bulk.Write([]byte("hello")); err != nil {
+		t.Fatalf("client dedicated bulk Write failed: %v", err)
+	}
+	select {
+	case got := <-readCh:
+		if got != "hello" {
+			t.Fatalf("server handler read %q, want %q", got, "hello")
+		}
+	case <-time.After(2 * time.Second):
+		t.Fatal("timed out waiting for synchronous handler read")
+	}
+}
+
+func TestDedicatedBulkOpenUnblocksOnBlockingFirstWrite(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	if err := UseModernPSKServer(server, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKServer failed: %v", err)
+	}
+
+	payload := strings.Repeat("w", 16)
+	writeDone := make(chan error, 1)
+	server.SetBulkHandler(func(info BulkAcceptInfo) error {
+		_, err := io.WriteString(info.Bulk, payload)
+		if err == nil {
+			err = info.Bulk.CloseWrite()
+		}
+		writeDone <- err
+		return err
+	})
+
+	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+		t.Fatalf("server Listen failed: %v", err)
+	}
+	defer func() {
+		_ = server.Stop()
+	}()
+
+	client := NewClient().(*ClientCommon)
+	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKClient failed: %v", err)
+	}
+	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+		t.Fatalf("client Connect failed: %v", err)
+	}
+	defer func() {
+		_ = client.Stop()
+	}()
+
+	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
+	defer cancel()
+	bulk, err := client.OpenDedicatedBulk(ctx, BulkOpenOptions{
+		ID: "blocking-write-ready",
+		Range: BulkRange{
+			Offset: 0,
+			Length: int64(len(payload)),
+		},
+		ChunkSize:   4,
+		WindowBytes: 4,
+		MaxInFlight: 1,
+	})
+	if err != nil {
+		t.Fatalf("client OpenDedicatedBulk failed: %v", err)
+	}
+
+	readBulkExactly(t, bulk, payload, 2*time.Second)
+	waitForBulkReadEOF(t, bulk, 2*time.Second)
+
+	select {
+	case err := <-writeDone:
+		if err != nil {
+			t.Fatalf("server handler write failed: %v", err)
+		}
+	case <-time.After(2 * time.Second):
+		t.Fatal("timed out waiting for blocking first write to finish")
+	}
+
+	if err := bulk.Close(); err != nil {
+		t.Fatalf("client dedicated bulk Close failed: %v", err)
+	}
+	waitForBulkContextDone(t, bulk.Context(), 2*time.Second)
+}
+
+func TestServerOpenBulkLogicalDedicatedUnblocksOnBlockingFirstRead(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	if err := UseModernPSKServer(server, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKServer failed: %v", err)
+	}
+
+	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+		t.Fatalf("server Listen failed: %v", err)
+	}
+	defer func() {
+		_ = server.Stop()
+	}()
+
+	client := NewClient().(*ClientCommon)
+	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKClient failed: %v", err)
+	}
+	readCh := make(chan string, 1)
+	client.SetBulkHandler(func(info BulkAcceptInfo) error {
+		defer func() {
+			_ = info.Bulk.Close()
+		}()
+		buf := make([]byte, 5)
+		if _, err := io.ReadFull(info.Bulk, buf); err != nil {
+			return err
+		}
+		readCh <- string(buf)
+		return nil
+	})
+	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+		t.Fatalf("client Connect failed: %v", err)
+	}
+	defer func() {
+		_ = client.Stop()
+	}()
+
+	logical := waitForTransferControlLogicalConn(t, server, 2*time.Second)
+	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
+	defer cancel()
+	bulk, err := server.OpenBulkLogical(ctx, logical, BulkOpenOptions{
+		ID: "server-blocking-read-ready",
+		Range: BulkRange{
+			Offset: 0,
+			Length: 5,
+		},
+		Dedicated: true,
+	})
+	if err != nil {
+		t.Fatalf("server OpenBulkLogical dedicated failed: %v", err)
+	}
+
+	if _, err := bulk.Write([]byte("hello")); err != nil {
+		t.Fatalf("server dedicated bulk Write failed: %v", err)
+	}
+
+	select {
+	case got := <-readCh:
+		if got != "hello" {
+			t.Fatalf("client handler read %q, want %q", got, "hello")
+		}
+	case <-time.After(2 * time.Second):
+		t.Fatal("timed out waiting for blocking first read to finish")
+	}
+
+	if err := bulk.Close(); err != nil {
+		t.Fatalf("server dedicated bulk Close failed: %v", err)
+	}
+	waitForBulkContextDone(t, bulk.Context(), 2*time.Second)
+}
+
+func TestDedicatedBulkOpenReturnsHandlerFailureAfterAccepted(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	if err := UseModernPSKServer(server, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKServer failed: %v", err)
+	}
+
+	server.SetBulkHandler(func(info BulkAcceptInfo) error {
+		time.Sleep(80 * time.Millisecond)
+		return errors.New("dedicated handler failed after accept")
+	})
+
+	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+		t.Fatalf("server Listen failed: %v", err)
+	}
+	defer func() {
+		_ = server.Stop()
+	}()
+
+	client := NewClient().(*ClientCommon)
+	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKClient failed: %v", err)
+	}
+	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+		t.Fatalf("client Connect failed: %v", err)
+	}
+	defer func() {
+		_ = client.Stop()
+	}()
+
+	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
+	defer cancel()
+	_, err := client.OpenDedicatedBulk(ctx, BulkOpenOptions{
+		ID: "accepted-then-fail",
+		Range: BulkRange{
+			Offset: 0,
+			Length: 1,
+		},
+	})
+	if err == nil || !strings.Contains(err.Error(), "dedicated handler failed after accept") {
+		t.Fatalf("client OpenDedicatedBulk error = %v, want dedicated handler failure after accept", err)
+	}
+}
+
+func TestServerOpenBulkLogicalDedicatedReturnsHandlerFailureAfterAccepted(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	if err := UseModernPSKServer(server, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKServer failed: %v", err)
+	}
+
+	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+		t.Fatalf("server Listen failed: %v", err)
+	}
+	defer func() {
+		_ = server.Stop()
+	}()
+
+	client := NewClient().(*ClientCommon)
+	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKClient failed: %v", err)
+	}
+	client.SetBulkHandler(func(info BulkAcceptInfo) error {
+		time.Sleep(80 * time.Millisecond)
+		return errors.New("client dedicated handler failed after accept")
+	})
+	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+		t.Fatalf("client Connect failed: %v", err)
+	}
+	defer func() {
+		_ = client.Stop()
+	}()
+
+	logical := waitForTransferControlLogicalConn(t, server, 2*time.Second)
+	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
+	defer cancel()
+	_, err := server.OpenBulkLogical(ctx, logical, BulkOpenOptions{
+		ID: "server-accepted-then-fail",
+		Range: BulkRange{
+			Offset: 0,
+			Length: 1,
+		},
+		Dedicated: true,
+	})
+	if err == nil || !strings.Contains(err.Error(), "client dedicated handler failed after accept") {
+		t.Fatalf("server OpenBulkLogical dedicated error = %v, want client dedicated handler failure after accept", err)
+	}
+}
+
 func TestBulkOpenRoundTripServerLogicalTCP(t *testing.T) {
 	server := NewServer().(*ServerCommon)
 	if err := UseModernPSKServer(server, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
@@ -532,15 +846,18 @@ func TestDedicatedBulkWritePrefersClosedPipeOverContextCanceled(t *testing.T) {
 		MaxInFlight: 4,
 	}, 0, nil, nil, 0, nil, nil, func(context.Context, *bulkHandle, []byte) error {
 		return nil
-	}, func(ctx context.Context, bulk *bulkHandle, payload []byte) (int, error) {
+	}, func(ctx context.Context, bulk *bulkHandle, startSeq uint64, payload []byte, payloadOwned bool) (int, error) {
 		bulk.markPeerClosed()
 		<-ctx.Done()
 		return 0, ctx.Err()
 	}, nil)
 
 	_, err := bulk.Write([]byte("abcdefgh"))
-	if !errors.Is(err, io.ErrClosedPipe) {
-		t.Fatalf("bulk Write error = %v, want %v", err, io.ErrClosedPipe)
+	if err != nil && !errors.Is(err, io.ErrClosedPipe) {
+		t.Fatalf("bulk Write error = %v, want nil or %v", err, io.ErrClosedPipe)
+	}
+	if err := bulk.waitPendingAsyncWrites(context.Background()); err != nil && !errors.Is(err, io.ErrClosedPipe) {
+		t.Fatalf("bulk waitPendingAsyncWrites error = %v, want nil or %v", err, io.ErrClosedPipe)
 	}
 }
 

client.go

@@ -35,6 +35,7 @@ type ClientCommon struct {
 	fastStreamEncode           transportFastStreamEncoder
 	fastBulkEncode             transportFastBulkEncoder
 	fastPlainEncode            transportFastPlainEncoder
+	modernPSKRuntime           *modernPSKCodecRuntime
 	handshakeRsaPubKey         []byte
 	SecretKey                  []byte
 	noFinSyncMsgMaxKeepSeconds int
@@ -55,6 +56,26 @@ type ClientCommon struct {
 	streamRuntime              *streamRuntime
 	recordRuntime              *recordRuntime
 	bulkRuntime                *bulkRuntime
+	bulkDefaultOpenMode        BulkOpenMode
+	bulkNetworkProfile         BulkNetworkProfile
+	bulkOpenTuning             BulkOpenTuning
+	bulkDedicatedAttachLimit   int
+	bulkDedicatedAttachSem     chan struct{}
+	bulkDedicatedAttachRetry   int
+	bulkDedicatedAttachBackoff time.Duration
+	bulkDedicatedDialTimeout   time.Duration
+	bulkDedicatedHelloTimeout  time.Duration
+	bulkDedicatedActiveLimit   int
+	bulkDedicatedActive        atomic.Int32
+	bulkDedicatedActiveWait    chan struct{}
+	bulkDedicatedLaneLimit     int
+	bulkDedicatedSidecarMu     sync.Mutex
+	bulkDedicatedLanes         map[uint32]*bulkDedicatedLane
+	bulkDedicatedNextLaneID    uint32
+	bulkAttachAttemptCount     atomic.Int64
+	bulkAttachRetryCount       atomic.Int64
+	bulkAttachSuccessCount     atomic.Int64
+	bulkAttachFallbackCount    atomic.Int64
 	connectionRetryState       *connectionRetryState
 	securityReadyCheck         bool
 	debugMode                  bool
@@ -63,21 +84,32 @@ type ClientCommon struct {
 func NewClient() Client {
 	transport := defaultModernPSKTransportBundle()
 	var client = ClientCommon{
-		maxReadTimeout:     0,
-		maxWriteTimeout:    0,
-		peerIdentity:       newClientPeerIdentity(),
-		sequenceEn:         encode,
-		sequenceDe:         Decode,
-		keyExchangeFn:      aesRsaHello,
-		SecretKey:          nil,
-		handshakeRsaPubKey: defaultRsaPubKey,
-		msgEn:              transport.msgEn,
-		msgDe:              transport.msgDe,
-		fastStreamEncode:   transport.fastStreamEncode,
-		fastBulkEncode:     transport.fastBulkEncode,
-		fastPlainEncode:    transport.fastPlainEncode,
-		skipKeyExchange:    true,
-		securityReadyCheck: true,
+		maxReadTimeout:             0,
+		maxWriteTimeout:            0,
+		peerIdentity:               newClientPeerIdentity(),
+		sequenceEn:                 encode,
+		sequenceDe:                 Decode,
+		keyExchangeFn:              aesRsaHello,
+		SecretKey:                  nil,
+		handshakeRsaPubKey:         defaultRsaPubKey,
+		msgEn:                      transport.msgEn,
+		msgDe:                      transport.msgDe,
+		fastStreamEncode:           transport.fastStreamEncode,
+		fastBulkEncode:             transport.fastBulkEncode,
+		fastPlainEncode:            transport.fastPlainEncode,
+		skipKeyExchange:            true,
+		securityReadyCheck:         true,
+		bulkDefaultOpenMode:        BulkOpenModeShared,
+		bulkNetworkProfile:         BulkNetworkProfileDefault,
+		bulkOpenTuning:             defaultBulkOpenTuning(),
+		bulkDedicatedAttachLimit:   defaultBulkDedicatedAttachLimit,
+		bulkDedicatedAttachRetry:   defaultBulkDedicatedAttachRetry,
+		bulkDedicatedAttachBackoff: defaultBulkDedicatedAttachBackoff,
+		bulkDedicatedDialTimeout:   defaultBulkDedicatedDialTimeout,
+		bulkDedicatedHelloTimeout:  defaultBulkDedicatedHelloTimeout,
+		bulkDedicatedActiveLimit:   defaultBulkDedicatedActiveLimit,
+		bulkDedicatedActiveWait:    make(chan struct{}),
+		bulkDedicatedLaneLimit:     defaultBulkDedicatedLaneLimit,
 	}
 	client.alive.Store(false)
 	client.useHeartBeat = true
@@ -93,6 +125,10 @@ func NewClient() Client {
 	client.streamRuntime = newStreamRuntime("cstrm")
 	client.recordRuntime = newRecordRuntime()
 	client.bulkRuntime = newBulkRuntime("cblk")
+	client.bulkDedicatedLanes = make(map[uint32]*bulkDedicatedLane)
+	if client.bulkDedicatedAttachLimit > 0 {
+		client.bulkDedicatedAttachSem = make(chan struct{}, client.bulkDedicatedAttachLimit)
+	}
 	client.connectionRetryState = newConnectionRetryState()
 	client.onFileEvent = normalizeFileEventCallback(nil)
 	client.fileEventObserver = normalizeFileEventCallback(nil)
@@ -103,3 +139,12 @@ func NewClient() Client {
 	client.getTransferState().setBuiltinHandler(client.builtinFileTransferHandler)
 	return &client
 }
+
+func (c *ClientCommon) maxWriteTimeoutSnapshot() time.Duration {
+	if c == nil {
+		return 0
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.maxWriteTimeout
+}

client_bulk.go

@@ -1,6 +1,9 @@
 package notify
 
-import "context"
+import (
+	"context"
+	"errors"
+)
 
 func (c *ClientCommon) SetBulkHandler(fn func(BulkAcceptInfo) error) {
 	runtime := c.getBulkRuntime()
@@ -10,10 +13,67 @@ func (c *ClientCommon) SetBulkHandler(fn func(BulkAcceptInfo) error) {
 	runtime.setHandler(fn)
 }
 
+func (c *ClientCommon) OpenSharedBulk(ctx context.Context, opt BulkOpenOptions) (Bulk, error) {
+	opt.Mode = BulkOpenModeShared
+	opt.Dedicated = false
+	return c.OpenBulk(ctx, opt)
+}
+
+func (c *ClientCommon) OpenDedicatedBulk(ctx context.Context, opt BulkOpenOptions) (Bulk, error) {
+	opt.Mode = BulkOpenModeDedicated
+	opt.Dedicated = true
+	return c.OpenBulk(ctx, opt)
+}
+
 func (c *ClientCommon) OpenBulk(ctx context.Context, opt BulkOpenOptions) (Bulk, error) {
+	if normalizeBulkOpenMode(opt.Mode) == BulkOpenModeDefault && !opt.Dedicated {
+		opt.Mode = c.bulkDefaultOpenModeSnapshot()
+	}
+	opt = normalizeBulkOpenOptions(opt)
+	switch opt.Mode {
+	case BulkOpenModeDedicated:
+		opt.Dedicated = true
+		return c.openBulkWithDedicatedMode(ctx, opt)
+	case BulkOpenModeAuto:
+		// Auto mode prefers dedicated path and falls back to shared if dedicated fails.
+		if err := clientDedicatedBulkSupportError(c); err == nil {
+			dedicatedOpt := opt
+			dedicatedOpt.Mode = BulkOpenModeDedicated
+			dedicatedOpt.Dedicated = true
+			bulk, dedicatedErr := c.openBulkWithDedicatedMode(ctx, dedicatedOpt)
+			if dedicatedErr == nil {
+				return bulk, nil
+			}
+			sharedOpt := opt
+			sharedOpt.Mode = BulkOpenModeShared
+			sharedOpt.Dedicated = false
+			sharedBulk, sharedErr := c.openBulkWithDedicatedMode(ctx, sharedOpt)
+			if sharedErr == nil {
+				c.bulkAttachFallbackCount.Add(1)
+				return sharedBulk, nil
+			}
+			return nil, errors.Join(dedicatedErr, sharedErr)
+		}
+		opt.Mode = BulkOpenModeShared
+		opt.Dedicated = false
+		c.bulkAttachFallbackCount.Add(1)
+		return c.openBulkWithDedicatedMode(ctx, opt)
+	case BulkOpenModeShared, BulkOpenModeDefault:
+		opt.Mode = BulkOpenModeShared
+		opt.Dedicated = false
+		return c.openBulkWithDedicatedMode(ctx, opt)
+	default:
+		opt.Mode = BulkOpenModeShared
+		opt.Dedicated = false
+		return c.openBulkWithDedicatedMode(ctx, opt)
+	}
+}
+
+func (c *ClientCommon) openBulkWithDedicatedMode(ctx context.Context, opt BulkOpenOptions) (Bulk, error) {
 	if c == nil {
 		return nil, errBulkClientNil
 	}
+	opt = applyBulkOpenTuningDefaults(opt, c.bulkOpenTuningSnapshot())
 	runtime := c.getBulkRuntime()
 	if runtime == nil {
 		return nil, errBulkRuntimeNil
@@ -33,6 +93,66 @@ func (c *ClientCommon) OpenBulk(ctx context.Context, opt BulkOpenOptions) (Bulk,
 	if _, exists := runtime.lookup(clientFileScope(), req.BulkID); exists {
 		return nil, errBulkAlreadyExists
 	}
+	if req.Dedicated {
+		req.DedicatedLaneID = c.reserveBulkDedicatedLane()
+		if req.DataID == 0 {
+			req.DataID = runtime.nextDataID()
+		}
+		if req.AttachToken == "" {
+			req.AttachToken = newBulkAttachToken()
+		}
+		bulk := newBulkHandle(c.clientStopContextSnapshot(), runtime, clientFileScope(), req, c.currentClientSessionEpoch(), nil, nil, 0, clientBulkCloseSender(c), clientBulkResetSender(c), clientBulkDataSender(c, c.currentClientSessionEpoch()), clientBulkWriteSender(c, c.currentClientSessionEpoch()), clientBulkReleaseSender(c))
+		bulk.setClientSnapshotOwner(c)
+		bulk.markAcceptHandled()
+		if err := runtime.register(clientFileScope(), bulk); err != nil {
+			c.releaseBulkDedicatedLane(req.DedicatedLaneID)
+			return nil, err
+		}
+		resp, err := sendBulkOpenClient(ctx, c, req)
+		if err != nil {
+			bulk.markReset(err)
+			return nil, err
+		}
+		if resp.DataID != 0 && resp.DataID != req.DataID {
+			err = errBulkAlreadyExists
+			_, _ = sendBulkResetClient(context.Background(), c, BulkResetRequest{
+				BulkID: req.BulkID,
+				DataID: req.DataID,
+				Error:  "bulk dedicated data id mismatch",
+			})
+			bulk.markReset(err)
+			return nil, err
+		}
+		if resp.TransportGeneration != 0 {
+			bulk.transportGeneration = resp.TransportGeneration
+		}
+		if resp.FastPathVersion != 0 {
+			bulk.fastPathVersion = normalizeBulkFastPathVersion(resp.FastPathVersion)
+		}
+		if resp.AttachToken != "" {
+			req.AttachToken = resp.AttachToken
+			bulk.setDedicatedAttachToken(resp.AttachToken)
+		}
+		if err := c.attachDedicatedBulkSidecar(ctx, bulk); err != nil {
+			_, _ = sendBulkResetClient(context.Background(), c, BulkResetRequest{
+				BulkID: req.BulkID,
+				DataID: req.DataID,
+				Error:  err.Error(),
+			})
+			bulk.markReset(err)
+			return nil, err
+		}
+		if err := bulk.waitAcceptReady(ctx); err != nil {
+			_, _ = sendBulkResetClient(context.Background(), c, BulkResetRequest{
+				BulkID: req.BulkID,
+				DataID: req.DataID,
+				Error:  err.Error(),
+			})
+			bulk.markReset(err)
+			return nil, err
+		}
+		return bulk, nil
+	}
 	resp, err := sendBulkOpenClient(ctx, c, req)
 	if err != nil {
 		return nil, err
@@ -40,6 +160,9 @@ func (c *ClientCommon) OpenBulk(ctx context.Context, opt BulkOpenOptions) (Bulk,
 	if resp.DataID != 0 {
 		req.DataID = resp.DataID
 	}
+	if resp.FastPathVersion != 0 {
+		req.FastPathVersion = resp.FastPathVersion
+	}
 	req.Dedicated = resp.Dedicated
 	if resp.AttachToken != "" {
 		req.AttachToken = resp.AttachToken
@@ -49,7 +172,9 @@ func (c *ClientCommon) OpenBulk(ctx context.Context, opt BulkOpenOptions) (Bulk,
 	}
 	bulk := newBulkHandle(c.clientStopContextSnapshot(), runtime, clientFileScope(), req, c.currentClientSessionEpoch(), nil, nil, resp.TransportGeneration, clientBulkCloseSender(c), clientBulkResetSender(c), clientBulkDataSender(c, c.currentClientSessionEpoch()), clientBulkWriteSender(c, c.currentClientSessionEpoch()), clientBulkReleaseSender(c))
 	bulk.setClientSnapshotOwner(c)
+	bulk.markAcceptHandled()
 	if err := runtime.register(clientFileScope(), bulk); err != nil {
+		c.releaseBulkDedicatedLane(req.DedicatedLaneID)
 		_, _ = sendBulkResetClient(context.Background(), c, BulkResetRequest{
 			BulkID: req.BulkID,
 			DataID: req.DataID,
@@ -78,15 +203,16 @@ func clientBulkRequest(runtime *bulkRuntime, opt BulkOpenOptions) BulkOpenReques
 		id = runtime.nextID()
 	}
 	return normalizeBulkOpenRequest(BulkOpenRequest{
-		BulkID:       id,
-		Range:        opt.Range,
-		Metadata:     cloneBulkMetadata(opt.Metadata),
-		ReadTimeout:  opt.ReadTimeout,
-		WriteTimeout: opt.WriteTimeout,
-		Dedicated:    opt.Dedicated,
-		ChunkSize:    opt.ChunkSize,
-		WindowBytes:  opt.WindowBytes,
-		MaxInFlight:  opt.MaxInFlight,
+		BulkID:          id,
+		FastPathVersion: bulkFastPathVersionCurrent,
+		Range:           opt.Range,
+		Metadata:        cloneBulkMetadata(opt.Metadata),
+		ReadTimeout:     opt.ReadTimeout,
+		WriteTimeout:    opt.WriteTimeout,
+		Dedicated:       opt.Dedicated,
+		ChunkSize:       opt.ChunkSize,
+		WindowBytes:     opt.WindowBytes,
+		MaxInFlight:     opt.MaxInFlight,
 	})
 }
 
@@ -148,12 +274,12 @@ func clientBulkDataSender(c *ClientCommon, epoch uint64) bulkDataSender {
 		if dataID == 0 {
 			return errBulkDataPathNotReady
 		}
-		return c.sendFastBulkData(ctx, dataID, bulk.nextOutboundDataSeq(), chunk)
+		return c.sendFastBulkData(ctx, dataID, bulk.nextOutboundDataSeq(), chunk, bulk.fastPathVersionSnapshot())
 	}
 }
 
 func clientBulkWriteSender(c *ClientCommon, epoch uint64) bulkWriteSender {
-	return func(ctx context.Context, bulk *bulkHandle, payload []byte) (int, error) {
+	return func(ctx context.Context, bulk *bulkHandle, startSeq uint64, payload []byte, payloadOwned bool) (int, error) {
 		if c == nil {
 			return 0, errBulkClientNil
 		}
@@ -168,12 +294,19 @@ func clientBulkWriteSender(c *ClientCommon, epoch uint64) bulkWriteSender {
 			if err := bulk.waitDedicatedReady(ctx); err != nil {
 				return 0, err
 			}
-			return c.sendDedicatedBulkWrite(ctx, bulk, payload)
+			return c.sendDedicatedBulkWrite(ctx, bulk, startSeq, payload)
 		}
 		if epoch != 0 && !c.isClientSessionEpochCurrent(epoch) {
 			return 0, errTransportDetached
 		}
-		return 0, nil
+		if bulk == nil {
+			return 0, errBulkRuntimeNil
+		}
+		dataID := bulk.dataIDSnapshot()
+		if dataID == 0 {
+			return 0, errBulkDataPathNotReady
+		}
+		return c.sendFastBulkWrite(ctx, dataID, startSeq, bulk.chunkSize, bulk.fastPathVersionSnapshot(), payload, payloadOwned)
 	}
 }
 
@@ -185,8 +318,20 @@ func clientBulkReleaseSender(c *ClientCommon) bulkReleaseSender {
 		if bytes <= 0 && chunks <= 0 {
 			return nil
 		}
+		ctx, cancel, err := bulk.newWriteContext(bulk.Context(), bulk.writeTimeout)
+		if err != nil {
+			return err
+		}
+		defer cancel()
 		if bulk.Dedicated() {
-			return c.sendDedicatedBulkRelease(context.Background(), bulk, bytes, chunks)
+			return c.sendDedicatedBulkRelease(ctx, bulk, bytes, chunks)
+		}
+		if bulk.fastPathVersionSnapshot() >= bulkFastPathVersionV2 {
+			payload, err := encodeBulkDedicatedReleasePayload(bytes, chunks)
+			if err != nil {
+				return err
+			}
+			return c.sendFastBulkControl(ctx, bulkFastPayloadTypeRelease, 0, bulk.dataIDSnapshot(), 0, bulk.fastPathVersionSnapshot(), payload)
 		}
 		return sendBulkReleaseClient(c, BulkReleaseRequest{
 			BulkID: bulk.ID(),

client_bulk_config.go

@@ -0,0 +1,364 @@
+package notify
+
+import "time"
+
+func defaultBulkOpenTuning() BulkOpenTuning {
+	return normalizeBulkOpenTuning(BulkOpenTuning{
+		ChunkSize:   defaultBulkChunkSize,
+		WindowBytes: defaultBulkOpenWindowBytes,
+		MaxInFlight: defaultBulkOpenMaxInFlight,
+	})
+}
+
+func normalizeBulkOpenTuning(tuning BulkOpenTuning) BulkOpenTuning {
+	if tuning.ChunkSize <= 0 {
+		tuning.ChunkSize = defaultBulkChunkSize
+	}
+	if tuning.WindowBytes <= 0 {
+		tuning.WindowBytes = defaultBulkOpenWindowBytes
+	}
+	if tuning.WindowBytes < tuning.ChunkSize {
+		tuning.WindowBytes = tuning.ChunkSize
+	}
+	if tuning.MaxInFlight <= 0 {
+		tuning.MaxInFlight = defaultBulkOpenMaxInFlight
+	}
+	return tuning
+}
+
+func applyBulkOpenTuningDefaults(opt BulkOpenOptions, tuning BulkOpenTuning) BulkOpenOptions {
+	tuning = normalizeBulkOpenTuning(tuning)
+	if opt.ChunkSize <= 0 {
+		opt.ChunkSize = tuning.ChunkSize
+	}
+	if opt.WindowBytes <= 0 {
+		opt.WindowBytes = tuning.WindowBytes
+	}
+	if opt.MaxInFlight <= 0 {
+		opt.MaxInFlight = tuning.MaxInFlight
+	}
+	return opt
+}
+
+func normalizeBulkNetworkProfile(profile BulkNetworkProfile) BulkNetworkProfile {
+	switch profile {
+	case BulkNetworkProfileDefault, BulkNetworkProfileLAN, BulkNetworkProfileWAN, BulkNetworkProfileConstrained:
+		return profile
+	default:
+		return BulkNetworkProfileDefault
+	}
+}
+
+func bulkNetworkProfileName(profile BulkNetworkProfile) string {
+	switch normalizeBulkNetworkProfile(profile) {
+	case BulkNetworkProfileLAN:
+		return "lan"
+	case BulkNetworkProfileWAN:
+		return "wan"
+	case BulkNetworkProfileConstrained:
+		return "constrained"
+	case BulkNetworkProfileDefault:
+		fallthrough
+	default:
+		return "default"
+	}
+}
+
+func bulkNetworkProfilePreset(profile BulkNetworkProfile) (BulkOpenMode, BulkDedicatedAttachConfig, BulkOpenTuning) {
+	switch normalizeBulkNetworkProfile(profile) {
+	case BulkNetworkProfileLAN:
+		return BulkOpenModeAuto, BulkDedicatedAttachConfig{
+				AttachLimit:  defaultBulkDedicatedAttachLimit,
+				ActiveLimit:  defaultBulkDedicatedActiveLimit,
+				LaneLimit:    4,
+				Retry:        defaultBulkDedicatedAttachRetry,
+				Backoff:      defaultBulkDedicatedAttachBackoff,
+				DialTimeout:  defaultBulkDedicatedDialTimeout,
+				HelloTimeout: defaultBulkDedicatedHelloTimeout,
+			}, BulkOpenTuning{
+				ChunkSize:   1024 * 1024,
+				WindowBytes: 32 * 1024 * 1024,
+				MaxInFlight: 64,
+			}
+	case BulkNetworkProfileWAN:
+		return BulkOpenModeAuto, BulkDedicatedAttachConfig{
+				AttachLimit:  2,
+				ActiveLimit:  defaultBulkDedicatedActiveLimit,
+				LaneLimit:    2,
+				Retry:        4,
+				Backoff:      250 * time.Millisecond,
+				DialTimeout:  15 * time.Second,
+				HelloTimeout: 20 * time.Second,
+			}, BulkOpenTuning{
+				ChunkSize:   512 * 1024,
+				WindowBytes: 8 * 1024 * 1024,
+				MaxInFlight: 16,
+			}
+	case BulkNetworkProfileConstrained:
+		return BulkOpenModeShared, BulkDedicatedAttachConfig{
+				AttachLimit:  1,
+				ActiveLimit:  2,
+				LaneLimit:    1,
+				Retry:        5,
+				Backoff:      400 * time.Millisecond,
+				DialTimeout:  20 * time.Second,
+				HelloTimeout: 30 * time.Second,
+			}, BulkOpenTuning{
+				ChunkSize:   128 * 1024,
+				WindowBytes: 1 * 1024 * 1024,
+				MaxInFlight: 8,
+			}
+	default:
+		return BulkOpenModeShared, BulkDedicatedAttachConfig{
+			AttachLimit:  defaultBulkDedicatedAttachLimit,
+			ActiveLimit:  defaultBulkDedicatedActiveLimit,
+			LaneLimit:    defaultBulkDedicatedLaneLimit,
+			Retry:        defaultBulkDedicatedAttachRetry,
+			Backoff:      defaultBulkDedicatedAttachBackoff,
+			DialTimeout:  defaultBulkDedicatedDialTimeout,
+			HelloTimeout: defaultBulkDedicatedHelloTimeout,
+		}, defaultBulkOpenTuning()
+	}
+}
+
+func normalizeBulkDedicatedAttachConfig(cfg BulkDedicatedAttachConfig) BulkDedicatedAttachConfig {
+	if cfg.AttachLimit < 0 {
+		cfg.AttachLimit = 0
+	}
+	if cfg.ActiveLimit < 0 {
+		cfg.ActiveLimit = 0
+	}
+	if cfg.LaneLimit < 0 {
+		cfg.LaneLimit = 0
+	}
+	if cfg.Retry < 0 {
+		cfg.Retry = 0
+	}
+	if cfg.Backoff <= 0 {
+		cfg.Backoff = defaultBulkDedicatedAttachBackoff
+	}
+	if cfg.DialTimeout <= 0 {
+		cfg.DialTimeout = defaultBulkDedicatedDialTimeout
+	}
+	if cfg.HelloTimeout <= 0 {
+		cfg.HelloTimeout = defaultBulkDedicatedHelloTimeout
+	}
+	return cfg
+}
+
+func (c *ClientCommon) setBulkDedicatedAttachSemaphoreLocked(limit int) {
+	if limit <= 0 {
+		c.bulkDedicatedAttachSem = nil
+		return
+	}
+	c.bulkDedicatedAttachSem = make(chan struct{}, limit)
+}
+
+func (c *ClientCommon) applyBulkDedicatedAttachConfigLocked(cfg BulkDedicatedAttachConfig) {
+	cfg = normalizeBulkDedicatedAttachConfig(cfg)
+	c.bulkDedicatedAttachLimit = cfg.AttachLimit
+	c.setBulkDedicatedAttachSemaphoreLocked(cfg.AttachLimit)
+	c.bulkDedicatedActiveLimit = cfg.ActiveLimit
+	c.notifyBulkDedicatedActiveWaitersLocked()
+	c.bulkDedicatedLaneLimit = cfg.LaneLimit
+	c.bulkDedicatedAttachRetry = cfg.Retry
+	c.bulkDedicatedAttachBackoff = cfg.Backoff
+	c.bulkDedicatedDialTimeout = cfg.DialTimeout
+	c.bulkDedicatedHelloTimeout = cfg.HelloTimeout
+}
+
+func (c *ClientCommon) ensureBulkDedicatedActiveWaitLocked() chan struct{} {
+	if c == nil {
+		return nil
+	}
+	if c.bulkDedicatedActiveWait == nil {
+		c.bulkDedicatedActiveWait = make(chan struct{})
+	}
+	return c.bulkDedicatedActiveWait
+}
+
+func (c *ClientCommon) notifyBulkDedicatedActiveWaitersLocked() {
+	if c == nil {
+		return
+	}
+	waitCh := c.ensureBulkDedicatedActiveWaitLocked()
+	close(waitCh)
+	c.bulkDedicatedActiveWait = make(chan struct{})
+}
+
+func (c *ClientCommon) applyBulkOpenTuningLocked(tuning BulkOpenTuning) {
+	c.bulkOpenTuning = normalizeBulkOpenTuning(tuning)
+}
+
+func (c *ClientCommon) bulkDefaultOpenModeSnapshot() BulkOpenMode {
+	if c == nil {
+		return BulkOpenModeShared
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	mode := normalizeBulkOpenMode(c.bulkDefaultOpenMode)
+	if mode == BulkOpenModeDefault {
+		mode = BulkOpenModeShared
+	}
+	return mode
+}
+
+func (c *ClientCommon) bulkDedicatedAttachSemaphoreSnapshot() chan struct{} {
+	if c == nil {
+		return nil
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.bulkDedicatedAttachSem
+}
+
+func (c *ClientCommon) bulkDedicatedActiveLimitSnapshot() int {
+	if c == nil {
+		return 0
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if c.bulkDedicatedActiveLimit < 0 {
+		return 0
+	}
+	return c.bulkDedicatedActiveLimit
+}
+
+func (c *ClientCommon) bulkDedicatedActiveWaitSnapshot() chan struct{} {
+	if c == nil {
+		return nil
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return c.ensureBulkDedicatedActiveWaitLocked()
+}
+
+func (c *ClientCommon) bulkDedicatedLaneLimitSnapshot() int {
+	if c == nil {
+		return defaultBulkDedicatedLaneLimit
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	limit := c.bulkDedicatedLaneLimit
+	if limit < 0 {
+		return 0
+	}
+	if limit == 0 {
+		return 0
+	}
+	return limit
+}
+
+func (c *ClientCommon) SetBulkDefaultOpenMode(mode BulkOpenMode) {
+	if c == nil {
+		return
+	}
+	mode = normalizeBulkOpenMode(mode)
+	if mode == BulkOpenModeDefault {
+		mode = BulkOpenModeShared
+	}
+	c.mu.Lock()
+	c.bulkDefaultOpenMode = mode
+	c.mu.Unlock()
+}
+
+func (c *ClientCommon) BulkDefaultOpenMode() BulkOpenMode {
+	return c.bulkDefaultOpenModeSnapshot()
+}
+
+func (c *ClientCommon) bulkOpenTuningSnapshot() BulkOpenTuning {
+	if c == nil {
+		return defaultBulkOpenTuning()
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return normalizeBulkOpenTuning(c.bulkOpenTuning)
+}
+
+func (c *ClientCommon) SetBulkOpenTuning(tuning BulkOpenTuning) {
+	if c == nil {
+		return
+	}
+	c.mu.Lock()
+	c.applyBulkOpenTuningLocked(tuning)
+	c.mu.Unlock()
+}
+
+func (c *ClientCommon) BulkOpenTuning() BulkOpenTuning {
+	return c.bulkOpenTuningSnapshot()
+}
+
+func (c *ClientCommon) SetBulkDedicatedAttachConfig(cfg BulkDedicatedAttachConfig) {
+	if c == nil {
+		return
+	}
+	c.mu.Lock()
+	c.applyBulkDedicatedAttachConfigLocked(cfg)
+	c.mu.Unlock()
+}
+
+func (c *ClientCommon) BulkDedicatedAttachConfig() BulkDedicatedAttachConfig {
+	if c == nil {
+		return normalizeBulkDedicatedAttachConfig(BulkDedicatedAttachConfig{})
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return normalizeBulkDedicatedAttachConfig(BulkDedicatedAttachConfig{
+		AttachLimit:  c.bulkDedicatedAttachLimit,
+		ActiveLimit:  c.bulkDedicatedActiveLimit,
+		LaneLimit:    c.bulkDedicatedLaneLimit,
+		Retry:        c.bulkDedicatedAttachRetry,
+		Backoff:      c.bulkDedicatedAttachBackoff,
+		DialTimeout:  c.bulkDedicatedDialTimeout,
+		HelloTimeout: c.bulkDedicatedHelloTimeout,
+	})
+}
+
+func (c *ClientCommon) SetBulkNetworkProfile(profile BulkNetworkProfile) {
+	if c == nil {
+		return
+	}
+	profile = normalizeBulkNetworkProfile(profile)
+	mode, cfg, tuning := bulkNetworkProfilePreset(profile)
+	c.mu.Lock()
+	c.bulkNetworkProfile = profile
+	c.bulkDefaultOpenMode = mode
+	c.applyBulkDedicatedAttachConfigLocked(cfg)
+	c.applyBulkOpenTuningLocked(tuning)
+	c.mu.Unlock()
+}
+
+func (c *ClientCommon) BulkNetworkProfile() BulkNetworkProfile {
+	if c == nil {
+		return BulkNetworkProfileDefault
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	return normalizeBulkNetworkProfile(c.bulkNetworkProfile)
+}
+
+func (s *ServerCommon) applyBulkOpenTuningLocked(tuning BulkOpenTuning) {
+	s.bulkOpenTuning = normalizeBulkOpenTuning(tuning)
+}
+
+func (s *ServerCommon) bulkOpenTuningSnapshot() BulkOpenTuning {
+	if s == nil {
+		return defaultBulkOpenTuning()
+	}
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+	return normalizeBulkOpenTuning(s.bulkOpenTuning)
+}
+
+func (s *ServerCommon) SetBulkOpenTuning(tuning BulkOpenTuning) {
+	if s == nil {
+		return
+	}
+	s.mu.Lock()
+	s.applyBulkOpenTuningLocked(tuning)
+	s.mu.Unlock()
+}
+
+func (s *ServerCommon) BulkOpenTuning() BulkOpenTuning {
+	return s.bulkOpenTuningSnapshot()
+}

client_config.go

@@ -75,6 +75,7 @@ func (c *ClientCommon) SetMsgEn(fn func([]byte, []byte) []byte) {
 	c.fastStreamEncode = nil
 	c.fastBulkEncode = nil
 	c.fastPlainEncode = nil
+	c.modernPSKRuntime = nil
 	c.securityReadyCheck = false
 }
 
@@ -89,6 +90,7 @@ func (c *ClientCommon) SetMsgDe(fn func([]byte, []byte) []byte) {
 	c.fastStreamEncode = nil
 	c.fastBulkEncode = nil
 	c.fastPlainEncode = nil
+	c.modernPSKRuntime = nil
 	c.securityReadyCheck = false
 }
 
@@ -108,6 +110,13 @@ func (c *ClientCommon) GetSecretKey() []byte {
 // Prefer UseModernPSKClient or UseLegacySecurityClient.
 func (c *ClientCommon) SetSecretKey(key []byte) {
 	c.SecretKey = key
+	if len(key) == 0 {
+		c.modernPSKRuntime = nil
+	} else if runtime, err := newModernPSKCodecRuntime(key, defaultModernPSKAAD); err == nil {
+		c.modernPSKRuntime = runtime
+	} else {
+		c.modernPSKRuntime = nil
+	}
 	c.securityReadyCheck = len(key) == 0
 	c.skipKeyExchange = true
 }

client_conn.go

@@ -72,6 +72,23 @@ func (c *ClientConn) readTUMessageLoop(rt *clientConnSessionRuntime) {
 	conn := rt.tuConn
 	generation := rt.transportGeneration
 	defer closeClientConnSessionRuntimeTransportDone(rt)
+	if conn != nil && !isPacketTransportConn(conn) {
+		reader := newTransportFrameReader(conn, nil)
+		for {
+			select {
+			case <-sessionStopChan(stopCtx):
+				if c.shouldCloseClientConnTransportOnStop(conn) {
+					_ = conn.Close()
+				}
+				return
+			default:
+			}
+			payload, release, err := c.readTUTransportPayloadPooled(conn, reader)
+			if !c.handleTUTransportPayloadReadResultWithSessionPooled(stopCtx, conn, generation, payload, release, err) {
+				return
+			}
+		}
+	}
 	buf := streamReadBuffer()
 	for {
 		select {

client_conn_attachment.go

@@ -13,6 +13,7 @@ type clientConnAttachmentState struct {
 	fastStreamEncode transportFastStreamEncoder
 	fastBulkEncode   transportFastBulkEncoder
 	fastPlainEncode  transportFastPlainEncoder
+	modernPSKRuntime *modernPSKCodecRuntime
 	handshakeRsaKey  []byte
 	secretKey        []byte
 	lastHeartBeat    int64
@@ -154,6 +155,7 @@ func (c *ClientConn) applyClientConnAttachmentProfile(maxReadTimeout time.Durati
 		state.maxWriteTimeout = maxWriteTimeout
 		state.msgEn = msgEn
 		state.msgDe = msgDe
+		state.modernPSKRuntime = nil
 		state.handshakeRsaKey = cloneClientConnAttachmentBytes(handshakeRsaKey)
 		state.secretKey = cloneClientConnAttachmentBytes(secretKey)
 	})
@@ -208,6 +210,7 @@ func (c *ClientConn) setClientConnMsgEn(fn func([]byte, []byte) []byte) {
 		state.fastStreamEncode = nil
 		state.fastBulkEncode = nil
 		state.fastPlainEncode = nil
+		state.modernPSKRuntime = nil
 	})
 }
 
@@ -224,6 +227,7 @@ func (c *ClientConn) setClientConnMsgDe(fn func([]byte, []byte) []byte) {
 		state.fastStreamEncode = nil
 		state.fastBulkEncode = nil
 		state.fastPlainEncode = nil
+		state.modernPSKRuntime = nil
 	})
 }
 
@@ -286,6 +290,64 @@ func (c *ClientConn) setClientConnSecretKey(key []byte) {
 	})
 }
 
+func (c *LogicalConn) attachmentStateRaw() *clientConnAttachmentState {
+	if c == nil {
+		return nil
+	}
+	if state := c.attachment.Load(); state != nil {
+		if client := c.compatClientConn(); client != nil {
+			client.attachment.Store(state)
+		}
+		return state
+	}
+	if client := c.compatClientConn(); client != nil {
+		if state := client.attachment.Load(); state != nil {
+			if c.attachment.CompareAndSwap(nil, state) {
+				client.attachment.Store(state)
+				return state
+			}
+			return c.attachmentStateRaw()
+		}
+	}
+	return nil
+}
+
+func (c *LogicalConn) modernPSKRuntimeSnapshot() *modernPSKCodecRuntime {
+	if state := c.attachmentStateRaw(); state != nil {
+		return state.modernPSKRuntime
+	}
+	return nil
+}
+
+func (c *LogicalConn) setModernPSKRuntime(runtime *modernPSKCodecRuntime) {
+	c.updateAttachmentState(func(state *clientConnAttachmentState) {
+		state.modernPSKRuntime = runtime
+	})
+}
+
+func (c *ClientConn) clientConnAttachmentStateRaw() *clientConnAttachmentState {
+	if c == nil {
+		return nil
+	}
+	if logical := c.logicalView.Load(); logical != nil {
+		return logical.attachmentStateRaw()
+	}
+	return c.attachment.Load()
+}
+
+func (c *ClientConn) clientConnModernPSKRuntimeSnapshot() *modernPSKCodecRuntime {
+	if state := c.clientConnAttachmentStateRaw(); state != nil {
+		return state.modernPSKRuntime
+	}
+	return nil
+}
+
+func (c *ClientConn) setClientConnModernPSKRuntime(runtime *modernPSKCodecRuntime) {
+	c.updateClientConnAttachmentState(func(state *clientConnAttachmentState) {
+		state.modernPSKRuntime = runtime
+	})
+}
+
 func (c *ClientConn) clientConnLastHeartbeatUnixSnapshot() int64 {
 	if c == nil {
 		return 0

client_conn_transport.go

@@ -1,6 +1,7 @@
 package notify
 
 import (
+	"b612.me/stario"
 	"context"
 	"net"
 	"os"
@@ -15,6 +16,10 @@ type serverInboundSourcePusher interface {
 	pushMessageSource([]byte, interface{})
 }
 
+type serverInboundSourceFastPusher interface {
+	pushTransportPayloadSourceFast([]byte, func(), interface{}) bool
+}
+
 func (c *LogicalConn) readTUMessage() {
 	rt := c.clientConnSessionRuntimeSnapshot()
 	if rt == nil {
@@ -37,6 +42,23 @@ func (c *LogicalConn) readTUMessageLoop(rt *clientConnSessionRuntime) {
 	conn := rt.tuConn
 	generation := rt.transportGeneration
 	defer closeClientConnSessionRuntimeTransportDone(rt)
+	if conn != nil && !isPacketTransportConn(conn) {
+		reader := newTransportFrameReader(conn, nil)
+		for {
+			select {
+			case <-sessionStopChan(stopCtx):
+				if c.shouldCloseTransportOnStop(conn) {
+					_ = conn.Close()
+				}
+				return
+			default:
+			}
+			payload, release, err := c.readTUTransportPayloadPooled(conn, reader)
+			if !c.handleTUTransportPayloadReadResultWithSessionPooled(stopCtx, conn, generation, payload, release, err) {
+				return
+			}
+		}
+	}
 	buf := streamReadBuffer()
 	for {
 		select {
@@ -54,6 +76,55 @@ func (c *LogicalConn) readTUMessageLoop(rt *clientConnSessionRuntime) {
 	}
 }
 
+func (c *LogicalConn) readTUTransportPayloadPooled(conn net.Conn, reader *stario.FrameReader) ([]byte, func(), error) {
+	if reader == nil {
+		return nil, nil, net.ErrClosed
+	}
+	if conn == nil {
+		return nil, nil, net.ErrClosed
+	}
+	if timeout := c.clientConnMaxReadTimeoutSnapshot(); timeout > 0 {
+		_ = conn.SetReadDeadline(time.Now().Add(timeout))
+	}
+	return reader.NextPooled()
+}
+
+func (c *LogicalConn) handleTUTransportPayloadReadResultWithSessionPooled(stopCtx context.Context, conn net.Conn, generation uint64, payload []byte, release func(), err error) bool {
+	if transportReadShouldStop(stopCtx) || !c.ownsTransportRead(conn, generation) {
+		if release != nil {
+			release()
+		}
+		if c.shouldCloseTransportOnStop(conn) {
+			_ = conn.Close()
+		}
+		return false
+	}
+	if err == os.ErrDeadlineExceeded {
+		return true
+	}
+	if err != nil {
+		if release != nil {
+			release()
+		}
+		select {
+		case <-sessionStopChan(stopCtx):
+			if c.shouldCloseTransportOnStop(conn) {
+				_ = conn.Close()
+			}
+			return false
+		default:
+		}
+		if detacher, ok := c.Server().(serverLogicalTransportDetacher); ok && c.shouldPreserveLogicalPeerOnTransportLoss() {
+			detacher.detachLogicalSessionTransport(c, "read error", err)
+			return false
+		}
+		c.stopServerOwnedSession("read error", err)
+		return false
+	}
+	c.pushServerOwnedTransportPayload(payload, release, conn, generation)
+	return true
+}
+
 func (c *LogicalConn) readFromTUTransportConnWithBuffer(conn net.Conn, data []byte) (int, []byte, error) {
 	if len(data) == 0 {
 		data = streamReadBuffer()
@@ -140,6 +211,29 @@ func (c *LogicalConn) pushServerOwnedTransportMessage(data []byte, conn net.Conn
 	server.pushMessage(data, c.clientConnIDSnapshot())
 }
 
+func (c *LogicalConn) pushServerOwnedTransportPayload(payload []byte, release func(), conn net.Conn, generation uint64) {
+	if c == nil || len(payload) == 0 {
+		if release != nil {
+			release()
+		}
+		return
+	}
+	server := c.Server()
+	if server == nil {
+		if release != nil {
+			release()
+		}
+		return
+	}
+	if pusher, ok := server.(serverInboundSourceFastPusher); ok {
+		pusher.pushTransportPayloadSourceFast(payload, release, newServerInboundSource(c, conn, nil, generation))
+		return
+	}
+	if release != nil {
+		release()
+	}
+}
+
 func (c *LogicalConn) shouldCloseTransportOnStop(conn net.Conn) bool {
 	if c == nil || conn == nil {
 		return false
@@ -185,10 +279,65 @@ func (c *ClientConn) readFromTUTransportConnWithBuffer(conn net.Conn, data []byt
 	return num, data, err
 }
 
+func (c *ClientConn) readTUTransportPayloadPooled(conn net.Conn, reader *stario.FrameReader) ([]byte, func(), error) {
+	if logical := c.LogicalConn(); logical != nil {
+		return logical.readTUTransportPayloadPooled(conn, reader)
+	}
+	if reader == nil {
+		return nil, nil, net.ErrClosed
+	}
+	if conn == nil {
+		return nil, nil, net.ErrClosed
+	}
+	if timeout := c.clientConnMaxReadTimeoutSnapshot(); timeout > 0 {
+		_ = conn.SetReadDeadline(time.Now().Add(timeout))
+	}
+	return reader.NextPooled()
+}
+
 func (c *ClientConn) handleTUTransportReadResult(num int, data []byte, err error) bool {
 	return c.handleTUTransportReadResultWithSession(c.clientConnTransportStopContextSnapshot(), c.clientConnTransportSnapshot(), c.clientConnTransportGenerationSnapshot(), num, data, err)
 }
 
+func (c *ClientConn) handleTUTransportPayloadReadResultWithSessionPooled(stopCtx context.Context, conn net.Conn, generation uint64, payload []byte, release func(), err error) bool {
+	if logical := c.LogicalConn(); logical != nil {
+		return logical.handleTUTransportPayloadReadResultWithSessionPooled(stopCtx, conn, generation, payload, release, err)
+	}
+	if transportReadShouldStop(stopCtx) || !c.ownsTransportRead(conn, generation) {
+		if release != nil {
+			release()
+		}
+		if c.shouldCloseClientConnTransportOnStop(conn) {
+			_ = conn.Close()
+		}
+		return false
+	}
+	if err == os.ErrDeadlineExceeded {
+		return true
+	}
+	if err != nil {
+		if release != nil {
+			release()
+		}
+		select {
+		case <-sessionStopChan(stopCtx):
+			if c.shouldCloseClientConnTransportOnStop(conn) {
+				_ = conn.Close()
+			}
+			return false
+		default:
+		}
+		if detacher, ok := c.server.(serverLogicalTransportDetacher); ok && c.shouldPreserveLogicalPeerOnTransportLoss() {
+			detacher.detachLogicalSessionTransport(logicalConnFromClient(c), "read error", err)
+			return false
+		}
+		c.stopServerOwnedSession("read error", err)
+		return false
+	}
+	c.pushServerOwnedTransportPayload(payload, release, conn, generation)
+	return true
+}
+
 func (c *ClientConn) handleTUTransportReadResultWithSession(stopCtx context.Context, conn net.Conn, generation uint64, num int, data []byte, err error) bool {
 	if logical := c.LogicalConn(); logical != nil {
 		return logical.handleTUTransportReadResultWithSession(stopCtx, conn, generation, num, data, err)
@@ -255,6 +404,26 @@ func (c *ClientConn) pushServerOwnedTransportMessage(data []byte, conn net.Conn,
 	c.server.pushMessage(data, c.clientConnIDSnapshot())
 }
 
+func (c *ClientConn) pushServerOwnedTransportPayload(payload []byte, release func(), conn net.Conn, generation uint64) {
+	if logical := c.LogicalConn(); logical != nil {
+		logical.pushServerOwnedTransportPayload(payload, release, conn, generation)
+		return
+	}
+	if c == nil || c.server == nil || len(payload) == 0 {
+		if release != nil {
+			release()
+		}
+		return
+	}
+	if pusher, ok := c.server.(serverInboundSourceFastPusher); ok {
+		pusher.pushTransportPayloadSourceFast(payload, release, newServerInboundSource(logicalConnFromClient(c), conn, nil, generation))
+		return
+	}
+	if release != nil {
+		release()
+	}
+}
+
 func (c *ClientConn) shouldCloseClientConnTransportOnStop(conn net.Conn) bool {
 	if logical := c.LogicalConn(); logical != nil {
 		return logical.shouldCloseTransportOnStop(conn)

client_runtime.go

@@ -433,6 +433,21 @@ func (c *ClientCommon) readMessageLoop(stopCtx context.Context, conn net.Conn, q
 	}
 	binding := newTransportBinding(conn, queue)
 	dispatcher := c.clientInboundDispatcherSnapshot()
+	if conn != nil && queue != nil && !isPacketTransportConn(conn) {
+		reader := newTransportFrameReader(conn, queue)
+		for {
+			select {
+			case <-stopCtx.Done():
+				c.closeClientTransportBinding(binding)
+				return
+			default:
+			}
+			payload, release, err := c.readTransportPayloadPooled(conn, reader)
+			if !c.handleTransportPayloadReadResultWithSession(stopCtx, binding, payload, release, err, epoch, dispatcher) {
+				return
+			}
+		}
+	}
 	buf := streamReadBuffer()
 	for {
 		select {

client_session_runtime_test.go

@@ -320,7 +320,7 @@ func TestSetClientSessionRuntimeStopsOldBindingWorkersOnReattach(t *testing.T) {
 	defer oldLeft.Close()
 	defer oldRight.Close()
 	oldBinding := newTransportBinding(oldLeft, queue)
-	oldSender := oldBinding.bulkBatchSenderSnapshot()
+	oldSender := oldBinding.clientBulkBatchSenderSnapshot(client)
 
 	client.setClientSessionRuntime(&clientSessionRuntime{
 		transport: oldBinding,
@@ -345,7 +345,7 @@ func TestSetClientSessionRuntimeStopsOldBindingWorkersOnReattach(t *testing.T) {
 		epoch:     2,
 	})
 
-	err := oldSender.submit(context.Background(), []byte("payload"))
+	err := oldSender.submitData(context.Background(), 1, 1, bulkFastPathVersionV1, []byte("payload"))
 	if err != errTransportDetached {
 		t.Fatalf("old sender submit after reattach = %v, want %v", err, errTransportDetached)
 	}

client_stream.go

@@ -35,6 +35,11 @@ func (c *ClientCommon) OpenStream(ctx context.Context, opt StreamOpenOptions) (S
 	if resp.DataID != 0 {
 		req.DataID = resp.DataID
 	}
+	if resp.FastPathVersion != 0 {
+		req.FastPathVersion = resp.FastPathVersion
+	} else {
+		req.FastPathVersion = streamFastPathVersionV1
+	}
 	req.Metadata = mergeStreamMetadata(req.Metadata, resp.Metadata)
 	stream := newStreamHandle(c.clientStopContextSnapshot(), runtime, clientFileScope(), req, c.currentClientSessionEpoch(), nil, nil, resp.TransportGeneration, clientStreamCloseSender(c), clientStreamResetSender(c), clientStreamDataSender(c, c.currentClientSessionEpoch()), runtime.configSnapshot())
 	stream.setClientSnapshotOwner(c)
@@ -66,11 +71,12 @@ func clientStreamRequest(runtime *streamRuntime, opt StreamOpenOptions) StreamOp
 		id = runtime.nextID()
 	}
 	return normalizeStreamOpenRequest(StreamOpenRequest{
-		StreamID:     id,
-		Channel:      opt.Channel,
-		Metadata:     cloneStreamMetadata(opt.Metadata),
-		ReadTimeout:  opt.ReadTimeout,
-		WriteTimeout: opt.WriteTimeout,
+		StreamID:        id,
+		FastPathVersion: streamFastPathVersionCurrent,
+		Channel:         opt.Channel,
+		Metadata:        cloneStreamMetadata(opt.Metadata),
+		ReadTimeout:     opt.ReadTimeout,
+		WriteTimeout:    opt.WriteTimeout,
 	})
 }
 
@@ -110,7 +116,7 @@ func clientStreamDataSender(c *ClientCommon, epoch uint64) streamDataSender {
 			}
 		}
 		if dataID := stream.dataIDSnapshot(); dataID != 0 {
-			return c.sendFastStreamData(dataID, stream.nextOutboundDataSeq(), chunk)
+			return c.sendFastStreamData(ctx, stream, chunk)
 		}
 		return c.sendEnvelope(newStreamDataEnvelope(stream.ID(), chunk))
 	}

client_transport.go

@@ -130,24 +130,85 @@ func (c *ClientCommon) handleTransportReadResultWithSessionDispatcher(stopCtx co
 	return true
 }
 
+func (c *ClientCommon) readTransportPayloadPooled(conn net.Conn, reader *stario.FrameReader) ([]byte, func(), error) {
+	if reader == nil {
+		return nil, nil, net.ErrClosed
+	}
+	if conn == nil {
+		return nil, nil, net.ErrClosed
+	}
+	if c.maxReadTimeout.Seconds() != 0 {
+		_ = conn.SetReadDeadline(time.Now().Add(c.maxReadTimeout))
+	}
+	return reader.NextPooled()
+}
+
+func (c *ClientCommon) handleTransportPayloadReadResultWithSession(stopCtx context.Context, binding *transportBinding, payload []byte, release func(), err error, epoch uint64, dispatcher *inboundDispatcher) bool {
+	if err == os.ErrDeadlineExceeded {
+		return true
+	}
+	if err != nil {
+		if release != nil {
+			release()
+		}
+		if c.showError || c.debugMode {
+			fmt.Println("client read error", err)
+		}
+		select {
+		case <-sessionStopChan(stopCtx):
+			c.closeClientTransportBinding(binding)
+			return false
+		default:
+		}
+		c.stopClientSessionIfCurrent(epoch, "client read error", err)
+		return false
+	}
+	c.dispatchTransportPayloadFast(payload, release, dispatcher)
+	return true
+}
+
+func (c *ClientCommon) dispatchTransportPayloadFast(payload []byte, release func(), dispatcher *inboundDispatcher) {
+	if len(payload) == 0 {
+		if release != nil {
+			release()
+		}
+		return
+	}
+	if c.tryDispatchBorrowedBulkTransportPayload(payload) {
+		if release != nil {
+			release()
+		}
+		return
+	}
+	owned := append([]byte(nil), payload...)
+	if release != nil {
+		release()
+	}
+	if dispatcher == nil {
+		now := time.Now()
+		if err := c.dispatchInboundTransportPayload(owned, now); err != nil && (c.showError || c.debugMode) {
+			fmt.Println("client decode envelope error", err)
+		}
+		return
+	}
+	c.wg.Add(1)
+	if !dispatcher.Dispatch(clientInboundDispatchSource(), func() {
+		defer c.wg.Done()
+		now := time.Now()
+		if err := c.dispatchInboundTransportPayload(owned, now); err != nil && (c.showError || c.debugMode) {
+			fmt.Println("client decode envelope error", err)
+		}
+	}) {
+		c.wg.Done()
+	}
+}
+
 func (c *ClientCommon) pushMessageFast(queue *stario.StarQueue, data []byte, dispatcher *inboundDispatcher) bool {
 	if queue == nil || dispatcher == nil || len(data) == 0 {
 		return false
 	}
-	if err := queue.ParseMessageOwned(data, "b612", func(msg stario.MsgQueue) error {
-		payload := msg.Msg
-		c.wg.Add(1)
-		if !dispatcher.Dispatch(clientInboundDispatchSource(), func() {
-			defer c.wg.Done()
-			now := time.Now()
-			if err := c.dispatchInboundTransportPayload(payload, now); err != nil {
-				if c.showError || c.debugMode {
-					fmt.Println("client decode envelope error", err)
-				}
-			}
-		}) {
-			c.wg.Done()
-		}
+	if err := queue.ParseMessageView(data, "b612", func(frame stario.FrameView) error {
+		c.dispatchTransportPayloadFast(frame.Payload, nil, dispatcher)
 		return nil
 	}); err != nil && (c.showError || c.debugMode) {
 		fmt.Println("client parse inbound frame error", err)

clienttype.go

@@ -18,6 +18,14 @@ type Client interface {
 	SetStreamConfig(StreamConfig)
 	SetTransferResumeStore(TransferResumeStore)
 	RecoverTransferSnapshots(context.Context) error
+	SetBulkNetworkProfile(BulkNetworkProfile)
+	BulkNetworkProfile() BulkNetworkProfile
+	SetBulkDefaultOpenMode(BulkOpenMode)
+	BulkDefaultOpenMode() BulkOpenMode
+	SetBulkOpenTuning(BulkOpenTuning)
+	BulkOpenTuning() BulkOpenTuning
+	SetBulkDedicatedAttachConfig(BulkDedicatedAttachConfig)
+	BulkDedicatedAttachConfig() BulkDedicatedAttachConfig
 	SetFileReceiveDir(dir string) error
 	send(msg TransferMsg) (WaitMsg, error)
 	sendEnvelope(env Envelope) error
@@ -77,6 +85,8 @@ type Client interface {
 	OpenStream(ctx context.Context, opt StreamOpenOptions) (Stream, error)
 	OpenRecordStream(ctx context.Context, opt RecordOpenOptions) (RecordStream, error)
 	OpenBulk(ctx context.Context, opt BulkOpenOptions) (Bulk, error)
+	OpenSharedBulk(ctx context.Context, opt BulkOpenOptions) (Bulk, error)
+	OpenDedicatedBulk(ctx context.Context, opt BulkOpenOptions) (Bulk, error)
 	SendTransfer(ctx context.Context, opt TransferSendOptions) (TransferHandle, error)
 	SendFile(ctx context.Context, filePath string) error
 }

default.go

@@ -126,6 +126,8 @@ func init() {
 	RegisterName("b612.me/notify.BulkCloseResponse", BulkCloseResponse{})
 	RegisterName("b612.me/notify.BulkResetRequest", BulkResetRequest{})
 	RegisterName("b612.me/notify.BulkResetResponse", BulkResetResponse{})
+	RegisterName("b612.me/notify.BulkReadyRequest", BulkReadyRequest{})
+	RegisterName("b612.me/notify.BulkReadyResponse", BulkReadyResponse{})
 	RegisterName("b612.me/notify.BulkReleaseRequest", BulkReleaseRequest{})
 	RegisterName("b612.me/notify.bulkAttachRequest", bulkAttachRequest{})
 	RegisterName("b612.me/notify.bulkAttachResponse", bulkAttachResponse{})

diagnostics_snapshot.go

@@ -59,6 +59,11 @@ type DiagnosticsSummary struct {
 	LogicalCount          int
 	CurrentTransportCount int
 
+	BulkAttachAttempts int64
+	BulkAttachRetries  int64
+	BulkAttachSuccess  int64
+	BulkAutoFallbacks  int64
+
 	StreamCount       int
 	ActiveStreamCount int
 	StaleStreamCount  int
@@ -236,7 +241,11 @@ func serverCurrentTransportRuntimeSnapshots(s Server) ([]TransportConnRuntimeSna
 
 func summarizeClientDiagnosticsSnapshot(snapshot ClientDiagnosticsSnapshot) DiagnosticsSummary {
 	summary := DiagnosticsSummary{
-		LogicalCount: diagnosticsLogicalCountFromClientRuntime(snapshot.Runtime),
+		LogicalCount:       diagnosticsLogicalCountFromClientRuntime(snapshot.Runtime),
+		BulkAttachAttempts: snapshot.Runtime.BulkAttachAttempts,
+		BulkAttachRetries:  snapshot.Runtime.BulkAttachRetries,
+		BulkAttachSuccess:  snapshot.Runtime.BulkAttachSuccess,
+		BulkAutoFallbacks:  snapshot.Runtime.BulkAutoFallbacks,
 	}
 	if snapshot.Runtime.TransportAttached {
 		summary.CurrentTransportCount = 1

go.mod

@@ -4,7 +4,7 @@ go 1.24.0
 
 require (
 	b612.me/starcrypto v1.0.2
-	b612.me/stario v0.1.0
+	b612.me/stario v0.1.1
 	github.com/Microsoft/go-winio v0.6.2
 )
 

go.sum

@@ -1,7 +1,7 @@
 b612.me/starcrypto v1.0.2 h1:6f8YHNMHZPwxDSRxY2OJeMP4ExKa/cakLIO04f0gLhE=
 b612.me/starcrypto v1.0.2/go.mod h1:I7oYTmQgnVPj5S5yKwoTyqkItq1HgF9XdJT/v3qs5QE=
-b612.me/stario v0.1.0 h1:V1uA7fLYzgTadOXpnyPaFC3z0MAKFIM/RKXzZUDXvL4=
-b612.me/stario v0.1.0/go.mod h1:7kjE69oFqNrca0P72L5+ZbTV09QGJ2N3bBY3qeFXOGc=
+b612.me/stario v0.1.1 h1:WIQy5DdK2Tkk+PIRORaVb76f4KY+64UvWChXNI7hSVY=
+b612.me/stario v0.1.1/go.mod h1:qMMqjaMhKmdfFn5T0oleL5L4FpFWEHsuIrT878hyo7I=
 github.com/Microsoft/go-winio v0.6.2 h1:F2VQgta7ecxGYO8k3ZZz3RS8fVIXVxONVUPlNERoyfY=
 github.com/Microsoft/go-winio v0.6.2/go.mod h1:yd8OoFMLzJbo9gZq8j5qaps8bJ9aShtEA8Ipt1oGCvU=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=

logical_conn.go

@@ -409,6 +409,7 @@ func (c *LogicalConn) applyAttachmentProfile(maxReadTimeout time.Duration, maxWr
 		state.fastStreamEncode = fastStreamEncode
 		state.fastBulkEncode = fastBulkEncode
 		state.fastPlainEncode = fastPlainEncode
+		state.modernPSKRuntime = nil
 		state.handshakeRsaKey = cloneClientConnAttachmentBytes(handshakeRsaKey)
 		state.secretKey = cloneClientConnAttachmentBytes(secretKey)
 	})
@@ -553,6 +554,12 @@ func (c *LogicalConn) runtimeSnapshot() ClientConnRuntimeSnapshot {
 		snapshot.HasRuntimeConn = c.transportSnapshot() != nil
 		snapshot.HasRuntimeStopCtx = rt.stopCtx != nil
 	}
+	if binding := c.transportBindingSnapshot(); binding != nil {
+		snapshot.TransportBulkAdaptiveSoftPayloadBytes = binding.bulkAdaptiveSoftPayloadBytesSnapshot()
+		snapshot.TransportStreamAdaptiveSoftPayloadBytes = binding.streamAdaptiveSoftPayloadBytesSnapshot()
+		snapshot.TransportStreamAdaptiveWaitThresholdBytes = binding.streamAdaptiveWaitThresholdBytesSnapshot()
+		snapshot.TransportStreamAdaptiveFlushDelay = binding.streamAdaptiveFlushDelaySnapshot()
+	}
 	if detach := c.transportDetachSnapshot(); detach != nil {
 		snapshot.TransportDetachReason = detach.Reason
 		snapshot.TransportDetachKind = classifyClientConnTransportDetachReason(detach.Reason)
@@ -816,6 +823,7 @@ func (c *LogicalConn) applyClientConnAttachmentProfile(maxReadTimeout time.Durat
 		state.maxWriteTimeout = maxWriteTimeout
 		state.msgEn = msgEn
 		state.msgDe = msgDe
+		state.modernPSKRuntime = nil
 		state.handshakeRsaKey = cloneClientConnAttachmentBytes(handshakeRsaKey)
 		state.secretKey = cloneClientConnAttachmentBytes(secretKey)
 	})

raw_tcp_benchmark_test.go

@@ -41,7 +41,7 @@ func BenchmarkRawTCPLocalhostThroughput(b *testing.B) {
 func benchmarkRawTCPLocalhostThroughput(b *testing.B, payloadSize int) {
 	b.Helper()
 
-	listener, err := net.Listen("tcp", "127.0.0.1:0")
+	listener, err := net.Listen("tcp", benchmarkTCPListenAddr(b))
 	if err != nil {
 		b.Fatalf("net.Listen failed: %v", err)
 	}
@@ -60,7 +60,7 @@ func benchmarkRawTCPLocalhostThroughput(b *testing.B, payloadSize int) {
 		acceptCh <- conn
 	}()
 
-	clientConn, err := net.Dial("tcp", listener.Addr().String())
+	clientConn, err := net.Dial("tcp", benchmarkTCPDialAddr(b, listener.Addr().String()))
 	if err != nil {
 		b.Fatalf("net.Dial failed: %v", err)
 	}

release_p0_test.go

@@ -5,6 +5,7 @@ import (
 	"errors"
 	"net"
 	"strings"
+	"sync/atomic"
 	"testing"
 	"time"
 )
@@ -14,6 +15,29 @@ type releaseP0TestAddr string
 func (a releaseP0TestAddr) Network() string { return "tcp" }
 func (a releaseP0TestAddr) String() string  { return string(a) }
 
+type closeInspectConn struct {
+	closeFn func()
+	closed  atomic.Bool
+}
+
+func (c *closeInspectConn) Read([]byte) (int, error)         { return 0, net.ErrClosed }
+func (c *closeInspectConn) Write(p []byte) (int, error)      { return len(p), nil }
+func (c *closeInspectConn) LocalAddr() net.Addr              { return releaseP0TestAddr("local") }
+func (c *closeInspectConn) RemoteAddr() net.Addr             { return releaseP0TestAddr("remote") }
+func (c *closeInspectConn) SetDeadline(time.Time) error      { return nil }
+func (c *closeInspectConn) SetReadDeadline(time.Time) error  { return nil }
+func (c *closeInspectConn) SetWriteDeadline(time.Time) error { return nil }
+
+func (c *closeInspectConn) Close() error {
+	if c == nil {
+		return nil
+	}
+	if c.closed.CompareAndSwap(false, true) && c.closeFn != nil {
+		c.closeFn()
+	}
+	return nil
+}
+
 func TestGetLogicalConnRuntimeSnapshotWithoutCompatClient(t *testing.T) {
 	server := NewServer().(*ServerCommon)
 	logical := &LogicalConn{server: server}
@@ -101,6 +125,116 @@ func TestHandleDedicatedBulkReadErrorPreservesUnderlyingCause(t *testing.T) {
 	}
 }
 
+func TestHandleClientDedicatedSidecarFailureMarksBulkBeforeClosingConn(t *testing.T) {
+	client := NewClient().(*ClientCommon)
+	runtime := client.getBulkRuntime()
+	if runtime == nil {
+		t.Fatal("client bulk runtime should not be nil")
+	}
+	bulk := newBulkHandle(context.Background(), runtime, clientFileScope(), BulkOpenRequest{
+		BulkID:    "sidecar-order",
+		DataID:    7,
+		Dedicated: true,
+		Range: BulkRange{
+			Length: 1,
+		},
+	}, 0, nil, nil, 0, nil, nil, nil, nil, nil)
+	var closeObservedErr error
+	conn := &closeInspectConn{
+		closeFn: func() {
+			closeObservedErr = bulk.resetErrSnapshot()
+		},
+	}
+	if err := bulk.attachDedicatedConnShared(conn); err != nil {
+		t.Fatalf("attachDedicatedConnShared failed: %v", err)
+	}
+	if err := runtime.register(clientFileScope(), bulk); err != nil {
+		t.Fatalf("register bulk failed: %v", err)
+	}
+	sidecar := newBulkDedicatedSidecar(conn, 1)
+	client.installClientDedicatedSidecar(1, sidecar)
+
+	client.handleClientDedicatedSidecarFailure(sidecar, errors.New("boom sidecar"))
+
+	if !errors.Is(closeObservedErr, errTransportDetached) {
+		t.Fatalf("closeObservedErr = %v, want transport detached", closeObservedErr)
+	}
+	if !strings.Contains(closeObservedErr.Error(), "dedicated bulk read error") || !strings.Contains(closeObservedErr.Error(), "boom sidecar") {
+		t.Fatalf("closeObservedErr detail = %q, want dedicated read error and cause", closeObservedErr.Error())
+	}
+}
+
+func TestCleanupClientSessionResourcesMarksBulkBeforeClosingSidecar(t *testing.T) {
+	client := NewClient().(*ClientCommon)
+	runtime := client.getBulkRuntime()
+	if runtime == nil {
+		t.Fatal("client bulk runtime should not be nil")
+	}
+	bulk := newBulkHandle(context.Background(), runtime, clientFileScope(), BulkOpenRequest{
+		BulkID:    "cleanup-order",
+		DataID:    9,
+		Dedicated: true,
+		Range: BulkRange{
+			Length: 1,
+		},
+	}, 0, nil, nil, 0, nil, nil, nil, nil, nil)
+	var closeObservedErr error
+	conn := &closeInspectConn{
+		closeFn: func() {
+			closeObservedErr = bulk.resetErrSnapshot()
+		},
+	}
+	if err := bulk.attachDedicatedConnShared(conn); err != nil {
+		t.Fatalf("attachDedicatedConnShared failed: %v", err)
+	}
+	if err := runtime.register(clientFileScope(), bulk); err != nil {
+		t.Fatalf("register bulk failed: %v", err)
+	}
+	client.installClientDedicatedSidecar(1, newBulkDedicatedSidecar(conn, 1))
+
+	client.cleanupClientSessionResources()
+
+	if !errors.Is(closeObservedErr, errServiceShutdown) {
+		t.Fatalf("closeObservedErr = %v, want %v", closeObservedErr, errServiceShutdown)
+	}
+}
+
+func TestBestEffortRejectInboundDedicatedDataUsesDedicatedResetRecord(t *testing.T) {
+	server := NewServer().(*ServerCommon)
+	UseLegacySecurityServer(server)
+
+	logical := server.bootstrapAcceptedLogical("dedicated-reject", nil, nil)
+	if logical == nil {
+		t.Fatal("bootstrapAcceptedLogical should return logical")
+	}
+	conn := newBulkAttachScriptConn(nil)
+
+	server.bestEffortRejectInboundDedicatedData(logical, conn, 42, "unknown data id")
+
+	recordConn := newBulkAttachScriptConn(conn.writtenBytes())
+	payload, err := readBulkDedicatedRecord(recordConn)
+	if err != nil {
+		t.Fatalf("readBulkDedicatedRecord failed: %v", err)
+	}
+	plain, err := server.decryptTransportPayloadLogical(logical, payload)
+	if err != nil {
+		t.Fatalf("decryptTransportPayloadLogical failed: %v", err)
+	}
+	items, err := decodeDedicatedBulkInboundItems(42, plain)
+	if err != nil {
+		t.Fatalf("decodeDedicatedBulkInboundItems failed: %v", err)
+	}
+	if len(items) != 1 {
+		t.Fatalf("decoded items = %d, want 1", len(items))
+	}
+	if items[0].Type != bulkFastPayloadTypeReset {
+		t.Fatalf("reset item type = %d, want %d", items[0].Type, bulkFastPayloadTypeReset)
+	}
+	if got, want := string(items[0].Payload), "unknown data id"; got != want {
+		t.Fatalf("reset payload = %q, want %q", got, want)
+	}
+}
+
 func TestRegisterAcceptedLogicalWithoutCompatClient(t *testing.T) {
 	server := NewServer().(*ServerCommon)
 	UseLegacySecurityServer(server)

security_psk.go

@@ -40,6 +40,8 @@ type modernPSKTransportBundle struct {
 	fastPlainEncode  transportFastPlainEncoder
 }
 
+var modernPSKPayloadPool sync.Pool
+
 // ModernPSKOptions configures the modern PSK transport profile.
 //
 // The current profile derives a 32-byte transport key with Argon2id and uses
@@ -81,6 +83,10 @@ func UseModernPSKClient(c Client, sharedSecret []byte, opts *ModernPSKOptions) e
 		return err
 	}
 	transport := buildModernPSKTransportBundle(aad)
+	runtime, err := newModernPSKCodecRuntime(key, aad)
+	if err != nil {
+		return err
+	}
 	c.SetSecretKey(key)
 	c.SetMsgEn(transport.msgEn)
 	c.SetMsgDe(transport.msgDe)
@@ -88,6 +94,7 @@ func UseModernPSKClient(c Client, sharedSecret []byte, opts *ModernPSKOptions) e
 		client.fastStreamEncode = transport.fastStreamEncode
 		client.fastBulkEncode = transport.fastBulkEncode
 		client.fastPlainEncode = transport.fastPlainEncode
+		client.modernPSKRuntime = runtime
 	}
 	c.SetSkipExchangeKey(true)
 	return nil
@@ -104,6 +111,10 @@ func UseModernPSKServer(s Server, sharedSecret []byte, opts *ModernPSKOptions) e
 		return err
 	}
 	transport := buildModernPSKTransportBundle(aad)
+	runtime, err := newModernPSKCodecRuntime(key, aad)
+	if err != nil {
+		return err
+	}
 	s.SetSecretKey(key)
 	s.SetDefaultCommEncode(transport.msgEn)
 	s.SetDefaultCommDecode(transport.msgDe)
@@ -111,6 +122,7 @@ func UseModernPSKServer(s Server, sharedSecret []byte, opts *ModernPSKOptions) e
 		server.defaultFastStreamEncode = transport.fastStreamEncode
 		server.defaultFastBulkEncode = transport.fastBulkEncode
 		server.defaultFastPlainEncode = transport.fastPlainEncode
+		server.defaultModernPSKRuntime = runtime
 	}
 	return nil
 }
@@ -127,6 +139,7 @@ func UseLegacySecurityClient(c Client) {
 		client.fastStreamEncode = nil
 		client.fastBulkEncode = nil
 		client.fastPlainEncode = nil
+		client.modernPSKRuntime = nil
 	}
 	c.SetSkipExchangeKey(false)
 	c.SetRsaPubKey(bytes.Clone(defaultRsaPubKey))
@@ -144,6 +157,7 @@ func UseLegacySecurityServer(s Server) {
 		server.defaultFastStreamEncode = nil
 		server.defaultFastBulkEncode = nil
 		server.defaultFastPlainEncode = nil
+		server.defaultModernPSKRuntime = nil
 	}
 	s.SetRsaPrivKey(bytes.Clone(defaultRsaKey))
 }
@@ -185,14 +199,14 @@ func buildModernPSKCodecs(aad []byte) (func([]byte, []byte) []byte, func([]byte,
 
 func buildModernPSKTransportBundle(aad []byte) modernPSKTransportBundle {
 	aadCopy := bytes.Clone(aad)
-	cache := &modernPSKCodecCache{}
+	cache := newModernPSKCodecCache(aadCopy)
 	msgEn := func(key []byte, plain []byte) []byte {
 		runtime, err := cache.runtimeForKey(key)
 		if err != nil {
 			log.Print(err)
 			return nil
 		}
-		out, err := runtime.sealPlainPayload(aadCopy, plain)
+		out, err := runtime.sealPlainPayload(plain)
 		if err != nil {
 			log.Print(err)
 			return nil
@@ -214,9 +228,7 @@ func buildModernPSKTransportBundle(aad []byte) modernPSKTransportBundle {
 			log.Print(err)
 			return nil
 		}
-		nonce := encrypted[len(modernPSKMagic):headerLen]
-		ciphertext := encrypted[headerLen:]
-		plain, err := runtime.aead.Open(make([]byte, 0, len(ciphertext)), nonce, ciphertext, aadCopy)
+		plain, err := runtime.openPayload(encrypted)
 		if err != nil {
 			log.Print(err)
 			return nil
@@ -228,21 +240,21 @@ func buildModernPSKTransportBundle(aad []byte) modernPSKTransportBundle {
 		if err != nil {
 			return nil, err
 		}
-		return runtime.sealStreamFastPayload(aadCopy, dataID, seq, payload)
+		return runtime.sealStreamFastPayload(dataID, seq, payload)
 	}
 	fastBulkEncode := func(key []byte, dataID uint64, seq uint64, payload []byte) ([]byte, error) {
 		runtime, err := cache.runtimeForKey(key)
 		if err != nil {
 			return nil, err
 		}
-		return runtime.sealBulkFastPayload(aadCopy, dataID, seq, payload)
+		return runtime.sealBulkFastPayload(dataID, seq, payload)
 	}
 	fastPlainEncode := func(key []byte, plainLen int, fill func([]byte) error) ([]byte, error) {
 		runtime, err := cache.runtimeForKey(key)
 		if err != nil {
 			return nil, err
 		}
-		return runtime.sealFilledPayload(aadCopy, plainLen, fill)
+		return runtime.sealFilledPayload(plainLen, fill)
 	}
 	return modernPSKTransportBundle{
 		msgEn:            msgEn,
@@ -269,16 +281,23 @@ func (s *ServerCommon) validateSecurityConfiguration() error {
 
 type modernPSKCodecCache struct {
 	mu      sync.Mutex
+	aad     []byte
 	key     []byte
 	runtime *modernPSKCodecRuntime
 }
 
 type modernPSKCodecRuntime struct {
 	aead   cipher.AEAD
+	key    []byte
+	aad    []byte
 	prefix [modernPSKNonceSize - 8]byte
 	seq    atomic.Uint64
 }
 
+func newModernPSKCodecCache(aad []byte) *modernPSKCodecCache {
+	return &modernPSKCodecCache{aad: bytes.Clone(aad)}
+}
+
 func (c *modernPSKCodecCache) runtimeForKey(key []byte) (*modernPSKCodecRuntime, error) {
 	if c == nil {
 		return nil, errModernPSKSecretEmpty
@@ -288,7 +307,7 @@ func (c *modernPSKCodecCache) runtimeForKey(key []byte) (*modernPSKCodecRuntime,
 	if c.runtime != nil && bytes.Equal(c.key, key) {
 		return c.runtime, nil
 	}
-	runtime, err := newModernPSKCodecRuntime(key)
+	runtime, err := newModernPSKCodecRuntime(key, c.aad)
 	if err != nil {
 		return nil, err
 	}
@@ -297,7 +316,7 @@ func (c *modernPSKCodecCache) runtimeForKey(key []byte) (*modernPSKCodecRuntime,
 	return runtime, nil
 }
 
-func newModernPSKCodecRuntime(key []byte) (*modernPSKCodecRuntime, error) {
+func newModernPSKCodecRuntime(key []byte, aad []byte) (*modernPSKCodecRuntime, error) {
 	if len(key) == 0 {
 		return nil, errModernPSKSecretEmpty
 	}
@@ -311,6 +330,8 @@ func newModernPSKCodecRuntime(key []byte) (*modernPSKCodecRuntime, error) {
 	}
 	runtime := &modernPSKCodecRuntime{
 		aead: aead,
+		key:  bytes.Clone(key),
+		aad:  bytes.Clone(aad),
 	}
 	if _, err := cryptorand.Read(runtime.prefix[:]); err != nil {
 		return nil, err
@@ -318,6 +339,13 @@ func newModernPSKCodecRuntime(key []byte) (*modernPSKCodecRuntime, error) {
 	return runtime, nil
 }
 
+func (r *modernPSKCodecRuntime) fork() (*modernPSKCodecRuntime, error) {
+	if r == nil {
+		return nil, errModernPSKSecretEmpty
+	}
+	return newModernPSKCodecRuntime(r.key, r.aad)
+}
+
 func (r *modernPSKCodecRuntime) nextNonce() [modernPSKNonceSize]byte {
 	var nonce [modernPSKNonceSize]byte
 	if r == nil {
@@ -328,8 +356,8 @@ func (r *modernPSKCodecRuntime) nextNonce() [modernPSKNonceSize]byte {
 	return nonce
 }
 
-func (r *modernPSKCodecRuntime) sealStreamFastPayload(aad []byte, dataID uint64, seq uint64, payload []byte) ([]byte, error) {
-	return r.sealFilledPayload(aad, streamFastPayloadHeaderLen+len(payload), func(frame []byte) error {
+func (r *modernPSKCodecRuntime) sealStreamFastPayload(dataID uint64, seq uint64, payload []byte) ([]byte, error) {
+	return r.sealFilledPayload(streamFastPayloadHeaderLen+len(payload), func(frame []byte) error {
 		if err := encodeStreamFastDataFrameHeader(frame, dataID, seq, len(payload)); err != nil {
 			return err
 		}
@@ -338,11 +366,11 @@ func (r *modernPSKCodecRuntime) sealStreamFastPayload(aad []byte, dataID uint64,
 	})
 }
 
-func (r *modernPSKCodecRuntime) sealBulkFastPayload(aad []byte, dataID uint64, seq uint64, payload []byte) ([]byte, error) {
+func (r *modernPSKCodecRuntime) sealBulkFastPayload(dataID uint64, seq uint64, payload []byte) ([]byte, error) {
 	if r == nil {
 		return nil, errTransportPayloadEncryptFailed
 	}
-	return r.sealFilledPayload(aad, bulkFastPayloadHeaderLen+len(payload), func(frame []byte) error {
+	return r.sealFilledPayload(bulkFastPayloadHeaderLen+len(payload), func(frame []byte) error {
 		if err := encodeBulkFastDataFrameHeader(frame, dataID, seq, len(payload)); err != nil {
 			return err
 		}
@@ -351,14 +379,14 @@ func (r *modernPSKCodecRuntime) sealBulkFastPayload(aad []byte, dataID uint64, s
 	})
 }
 
-func (r *modernPSKCodecRuntime) sealPlainPayload(aad []byte, plain []byte) ([]byte, error) {
-	return r.sealFilledPayload(aad, len(plain), func(dst []byte) error {
+func (r *modernPSKCodecRuntime) sealPlainPayload(plain []byte) ([]byte, error) {
+	return r.sealFilledPayload(len(plain), func(dst []byte) error {
 		copy(dst, plain)
 		return nil
 	})
 }
 
-func (r *modernPSKCodecRuntime) sealFilledPayload(aad []byte, plainLen int, fill func([]byte) error) ([]byte, error) {
+func (r *modernPSKCodecRuntime) sealFilledPayload(plainLen int, fill func([]byte) error) ([]byte, error) {
 	if r == nil {
 		return nil, errTransportPayloadEncryptFailed
 	}
@@ -368,6 +396,35 @@ func (r *modernPSKCodecRuntime) sealFilledPayload(aad []byte, plainLen int, fill
 	nonce := r.nextNonce()
 	headerLen := len(modernPSKMagic) + modernPSKNonceSize
 	out := make([]byte, headerLen+plainLen+r.aead.Overhead())
+	sealed, err := r.sealInto(out, headerLen, nonce, plainLen, fill)
+	if err != nil {
+		return nil, err
+	}
+	return out[:headerLen+len(sealed)], nil
+}
+
+func (r *modernPSKCodecRuntime) sealFilledPayloadPooled(plainLen int, fill func([]byte) error) ([]byte, func(), error) {
+	if r == nil {
+		return nil, nil, errTransportPayloadEncryptFailed
+	}
+	if plainLen < 0 {
+		return nil, nil, errTransportPayloadEncryptFailed
+	}
+	nonce := r.nextNonce()
+	headerLen := len(modernPSKMagic) + modernPSKNonceSize
+	totalLen := headerLen + plainLen + r.aead.Overhead()
+	out := getModernPSKPayloadBuffer(totalLen)
+	sealed, err := r.sealInto(out, headerLen, nonce, plainLen, fill)
+	if err != nil {
+		putModernPSKPayloadBuffer(out)
+		return nil, nil, err
+	}
+	return out[:headerLen+len(sealed)], func() {
+		putModernPSKPayloadBuffer(out)
+	}, nil
+}
+
+func (r *modernPSKCodecRuntime) sealInto(out []byte, headerLen int, nonce [modernPSKNonceSize]byte, plainLen int, fill func([]byte) error) ([]byte, error) {
 	copy(out[:len(modernPSKMagic)], modernPSKMagic)
 	copy(out[len(modernPSKMagic):headerLen], nonce[:])
 	frame := out[headerLen : headerLen+plainLen]
@@ -376,6 +433,98 @@ func (r *modernPSKCodecRuntime) sealFilledPayload(aad []byte, plainLen int, fill
 			return nil, err
 		}
 	}
-	sealed := r.aead.Seal(frame[:0], nonce[:], frame, aad)
-	return out[:headerLen+len(sealed)], nil
+	return r.aead.Seal(frame[:0], nonce[:], frame, r.aad), nil
+}
+
+func (r *modernPSKCodecRuntime) openPayload(encrypted []byte) ([]byte, error) {
+	if r == nil {
+		return nil, errTransportPayloadDecryptFailed
+	}
+	headerLen := len(modernPSKMagic) + modernPSKNonceSize
+	if len(encrypted) < headerLen {
+		return nil, errModernPSKPayload
+	}
+	if !bytes.Equal(encrypted[:len(modernPSKMagic)], modernPSKMagic) {
+		return nil, errModernPSKPayload
+	}
+	nonce := encrypted[len(modernPSKMagic):headerLen]
+	ciphertext := encrypted[headerLen:]
+	return r.aead.Open(make([]byte, 0, len(ciphertext)), nonce, ciphertext, r.aad)
+}
+
+func (r *modernPSKCodecRuntime) openPayloadPooled(encrypted []byte, release func()) ([]byte, func(), error) {
+	if r == nil {
+		if release != nil {
+			release()
+		}
+		return nil, nil, errTransportPayloadDecryptFailed
+	}
+	headerLen := len(modernPSKMagic) + modernPSKNonceSize
+	if len(encrypted) < headerLen {
+		if release != nil {
+			release()
+		}
+		return nil, nil, errModernPSKPayload
+	}
+	if !bytes.Equal(encrypted[:len(modernPSKMagic)], modernPSKMagic) {
+		if release != nil {
+			release()
+		}
+		return nil, nil, errModernPSKPayload
+	}
+	nonce := encrypted[len(modernPSKMagic):headerLen]
+	ciphertext := encrypted[headerLen:]
+	plain, err := r.aead.Open(ciphertext[:0], nonce, ciphertext, r.aad)
+	if err != nil {
+		if release != nil {
+			release()
+		}
+		return nil, nil, err
+	}
+	return plain, release, nil
+}
+
+func (r *modernPSKCodecRuntime) openPayloadOwnedPooled(encrypted []byte) ([]byte, func(), error) {
+	if r == nil {
+		return nil, nil, errTransportPayloadDecryptFailed
+	}
+	headerLen := len(modernPSKMagic) + modernPSKNonceSize
+	if len(encrypted) < headerLen {
+		return nil, nil, errModernPSKPayload
+	}
+	if !bytes.Equal(encrypted[:len(modernPSKMagic)], modernPSKMagic) {
+		return nil, nil, errModernPSKPayload
+	}
+	nonce := encrypted[len(modernPSKMagic):headerLen]
+	ciphertext := encrypted[headerLen:]
+	plainLen := len(ciphertext) - r.aead.Overhead()
+	if plainLen < 0 {
+		return nil, nil, errModernPSKPayload
+	}
+	out := getModernPSKPayloadBuffer(plainLen)
+	plain, err := r.aead.Open(out[:0], nonce, ciphertext, r.aad)
+	if err != nil {
+		putModernPSKPayloadBuffer(out)
+		return nil, nil, err
+	}
+	return plain, func() {
+		putModernPSKPayloadBuffer(out)
+	}, nil
+}
+
+func getModernPSKPayloadBuffer(size int) []byte {
+	if size <= 0 {
+		return nil
+	}
+	if pooled, ok := modernPSKPayloadPool.Get().([]byte); ok && cap(pooled) >= size {
+		return pooled[:size]
+	}
+	return make([]byte, size)
+}
+
+func putModernPSKPayloadBuffer(buf []byte) {
+	if cap(buf) == 0 || cap(buf) > 32*1024*1024 {
+		return
+	}
+	modernPSKPayloadPool.Put(buf[:0])
 }

security_psk_test.go

@@ -83,6 +83,12 @@ func TestDefaultConstructorsUseModernTransportAfterSetSecretKey(t *testing.T) {
 	sharedKey := []byte("0123456789abcdef0123456789abcdef")
 	client.SetSecretKey(sharedKey)
 	server.SetSecretKey(sharedKey)
+	if client.modernPSKRuntime == nil {
+		t.Fatal("client modernPSKRuntime should be installed after SetSecretKey")
+	}
+	if server.defaultModernPSKRuntime == nil {
+		t.Fatal("server defaultModernPSKRuntime should be installed after SetSecretKey")
+	}
 
 	plain := []byte("notify default modern transport")
 	wire := client.msgEn(client.SecretKey, plain)
@@ -92,6 +98,29 @@ func TestDefaultConstructorsUseModernTransportAfterSetSecretKey(t *testing.T) {
 	}
 }
 
+func TestCustomCodecOverridesClearModernRuntime(t *testing.T) {
+	client := NewClient().(*ClientCommon)
+	server := NewServer().(*ServerCommon)
+	sharedKey := []byte("0123456789abcdef0123456789abcdef")
+	client.SetSecretKey(sharedKey)
+	server.SetSecretKey(sharedKey)
+	if client.modernPSKRuntime == nil || server.defaultModernPSKRuntime == nil {
+		t.Fatal("modern runtimes should be installed before override")
+	}
+
+	client.SetMsgEn(defaultMsgEn)
+	client.SetMsgDe(defaultMsgDe)
+	server.SetDefaultCommEncode(defaultMsgEn)
+	server.SetDefaultCommDecode(defaultMsgDe)
+
+	if client.modernPSKRuntime != nil {
+		t.Fatal("client modernPSKRuntime should be cleared after custom codec override")
+	}
+	if server.defaultModernPSKRuntime != nil {
+		t.Fatal("server defaultModernPSKRuntime should be cleared after custom codec override")
+	}
+}
+
 func TestDefaultConstructorsDecodeSignalEnvelopeWithModernTransport(t *testing.T) {
 	client := NewClient().(*ClientCommon)
 	server := NewServer().(*ServerCommon)

server.go

@@ -33,6 +33,7 @@ type ServerCommon struct {
 	defaultFastStreamEncode    transportFastStreamEncoder
 	defaultFastBulkEncode      transportFastBulkEncoder
 	defaultFastPlainEncode     transportFastPlainEncoder
+	defaultModernPSKRuntime    *modernPSKCodecRuntime
 	linkFns                    map[string]func(message *Message)
 	defaultFns                 func(message *Message)
 	noFinSyncMsgMaxKeepSeconds int64
@@ -47,6 +48,9 @@ type ServerCommon struct {
 	streamRuntime              *streamRuntime
 	recordRuntime              *recordRuntime
 	bulkRuntime                *bulkRuntime
+	bulkOpenTuning             BulkOpenTuning
+	bulkDedicatedSidecarMu     sync.Mutex
+	bulkDedicatedSidecars      map[*LogicalConn]map[uint32]*bulkDedicatedSidecar
 	connectionRetryState       *connectionRetryState
 	detachedClientKeepSeconds  int64
 	securityReadyCheck         bool
@@ -81,6 +85,8 @@ func NewServer() Server {
 	server.streamRuntime = newStreamRuntime("sstrm")
 	server.recordRuntime = newRecordRuntime()
 	server.bulkRuntime = newBulkRuntime("sblk")
+	server.bulkOpenTuning = defaultBulkOpenTuning()
+	server.bulkDedicatedSidecars = make(map[*LogicalConn]map[uint32]*bulkDedicatedSidecar)
 	server.connectionRetryState = newConnectionRetryState()
 	server.onFileEvent = normalizeFileEventCallback(nil)
 	server.fileEventObserver = normalizeFileEventCallback(nil)

server_bulk.go

@@ -1,6 +1,9 @@
 package notify
 
-import "context"
+import (
+	"context"
+	"errors"
+)
 
 func (s *ServerCommon) SetBulkHandler(fn func(BulkAcceptInfo) error) {
 	runtime := s.getBulkRuntime()
@@ -11,9 +14,48 @@ func (s *ServerCommon) SetBulkHandler(fn func(BulkAcceptInfo) error) {
 }
 
 func (s *ServerCommon) OpenBulkLogical(ctx context.Context, logical *LogicalConn, opt BulkOpenOptions) (Bulk, error) {
+	opt = normalizeBulkOpenOptions(opt)
+	switch opt.Mode {
+	case BulkOpenModeDedicated:
+		opt.Dedicated = true
+		return s.openBulkLogicalWithMode(ctx, logical, opt)
+	case BulkOpenModeAuto:
+		if err := logicalDedicatedBulkSupportError(logical); err == nil {
+			dedicatedOpt := opt
+			dedicatedOpt.Mode = BulkOpenModeDedicated
+			dedicatedOpt.Dedicated = true
+			bulk, dedicatedErr := s.openBulkLogicalWithMode(ctx, logical, dedicatedOpt)
+			if dedicatedErr == nil {
+				return bulk, nil
+			}
+			sharedOpt := opt
+			sharedOpt.Mode = BulkOpenModeShared
+			sharedOpt.Dedicated = false
+			sharedBulk, sharedErr := s.openBulkLogicalWithMode(ctx, logical, sharedOpt)
+			if sharedErr == nil {
+				return sharedBulk, nil
+			}
+			return nil, errors.Join(dedicatedErr, sharedErr)
+		}
+		opt.Mode = BulkOpenModeShared
+		opt.Dedicated = false
+		return s.openBulkLogicalWithMode(ctx, logical, opt)
+	case BulkOpenModeShared, BulkOpenModeDefault:
+		opt.Mode = BulkOpenModeShared
+		opt.Dedicated = false
+		return s.openBulkLogicalWithMode(ctx, logical, opt)
+	default:
+		opt.Mode = BulkOpenModeShared
+		opt.Dedicated = false
+		return s.openBulkLogicalWithMode(ctx, logical, opt)
+	}
+}
+
+func (s *ServerCommon) openBulkLogicalWithMode(ctx context.Context, logical *LogicalConn, opt BulkOpenOptions) (Bulk, error) {
 	if s == nil {
 		return nil, errBulkServerNil
 	}
+	opt = applyBulkOpenTuningDefaults(opt, s.bulkOpenTuningSnapshot())
 	if logical == nil {
 		return nil, errBulkLogicalConnNil
 	}
@@ -42,17 +84,44 @@ func (s *ServerCommon) OpenBulkLogical(ctx context.Context, logical *LogicalConn
 			req.AttachToken = newBulkAttachToken()
 		}
 		bulk := newBulkHandle(logical.stopContextSnapshot(), runtime, scope, req, 0, logical, logical.CurrentTransportConn(), logical.transportGenerationSnapshot(), serverBulkCloseSender(s, logical, nil), serverBulkResetSender(s, logical, nil), serverBulkDataSender(s, logical.CurrentTransportConn()), serverBulkWriteSender(s, logical, logical.CurrentTransportConn()), serverBulkReleaseSender(s, logical, logical.CurrentTransportConn()))
+		bulk.markAcceptHandled()
 		if err := runtime.register(scope, bulk); err != nil {
 			return nil, err
 		}
+		s.attachServerDedicatedSidecarIfExists(logical, bulk)
 		resp, err := sendBulkOpenServerLogical(ctx, s, logical, req)
 		if err != nil {
-			runtime.remove(scope, bulk.ID())
+			bulk.markReset(err)
+			return nil, err
+		}
+		if resp.DataID != 0 && resp.DataID != req.DataID {
+			err = errBulkAlreadyExists
+			_, _ = sendBulkResetServerLogical(context.Background(), s, logical, BulkResetRequest{
+				BulkID: req.BulkID,
+				DataID: req.DataID,
+				Error:  "bulk dedicated data id mismatch",
+			})
+			bulk.markReset(err)
 			return nil, err
 		}
 		if resp.TransportGeneration != 0 {
 			bulk.transportGeneration = resp.TransportGeneration
 		}
+		if resp.FastPathVersion != 0 {
+			bulk.fastPathVersion = normalizeBulkFastPathVersion(resp.FastPathVersion)
+		}
+		if resp.AttachToken != "" {
+			bulk.setDedicatedAttachToken(resp.AttachToken)
+		}
+		if err := bulk.waitAcceptReady(ctx); err != nil {
+			_, _ = sendBulkResetServerLogical(context.Background(), s, logical, BulkResetRequest{
+				BulkID: req.BulkID,
+				DataID: req.DataID,
+				Error:  err.Error(),
+			})
+			bulk.markReset(err)
+			return nil, err
+		}
 		return bulk, nil
 	}
 	resp, err := sendBulkOpenServerLogical(ctx, s, logical, req)
@@ -62,6 +131,9 @@ func (s *ServerCommon) OpenBulkLogical(ctx context.Context, logical *LogicalConn
 	if resp.DataID != 0 {
 		req.DataID = resp.DataID
 	}
+	if resp.FastPathVersion != 0 {
+		req.FastPathVersion = resp.FastPathVersion
+	}
 	req.Dedicated = resp.Dedicated
 	if resp.AttachToken != "" {
 		req.AttachToken = resp.AttachToken
@@ -71,6 +143,7 @@ func (s *ServerCommon) OpenBulkLogical(ctx context.Context, logical *LogicalConn
 	}
 	transport := logical.CurrentTransportConn()
 	bulk := newBulkHandle(logical.stopContextSnapshot(), runtime, scope, req, 0, logical, transport, resp.TransportGeneration, serverBulkCloseSender(s, logical, nil), serverBulkResetSender(s, logical, nil), serverBulkDataSender(s, transport), serverBulkWriteSender(s, logical, transport), serverBulkReleaseSender(s, logical, transport))
+	bulk.markAcceptHandled()
 	if err := runtime.register(scope, bulk); err != nil {
 		_, _ = sendBulkResetServerLogical(context.Background(), s, logical, BulkResetRequest{
 			BulkID: req.BulkID,
@@ -79,13 +152,53 @@ func (s *ServerCommon) OpenBulkLogical(ctx context.Context, logical *LogicalConn
 		})
 		return nil, err
 	}
+	s.attachServerDedicatedSidecarIfExists(logical, bulk)
 	return bulk, nil
 }
 
 func (s *ServerCommon) OpenBulkTransport(ctx context.Context, transport *TransportConn, opt BulkOpenOptions) (Bulk, error) {
+	opt = normalizeBulkOpenOptions(opt)
+	switch opt.Mode {
+	case BulkOpenModeDedicated:
+		opt.Dedicated = true
+		return s.openBulkTransportWithMode(ctx, transport, opt)
+	case BulkOpenModeAuto:
+		if err := transportDedicatedBulkSupportError(transport); err == nil {
+			dedicatedOpt := opt
+			dedicatedOpt.Mode = BulkOpenModeDedicated
+			dedicatedOpt.Dedicated = true
+			bulk, dedicatedErr := s.openBulkTransportWithMode(ctx, transport, dedicatedOpt)
+			if dedicatedErr == nil {
+				return bulk, nil
+			}
+			sharedOpt := opt
+			sharedOpt.Mode = BulkOpenModeShared
+			sharedOpt.Dedicated = false
+			sharedBulk, sharedErr := s.openBulkTransportWithMode(ctx, transport, sharedOpt)
+			if sharedErr == nil {
+				return sharedBulk, nil
+			}
+			return nil, errors.Join(dedicatedErr, sharedErr)
+		}
+		opt.Mode = BulkOpenModeShared
+		opt.Dedicated = false
+		return s.openBulkTransportWithMode(ctx, transport, opt)
+	case BulkOpenModeShared, BulkOpenModeDefault:
+		opt.Mode = BulkOpenModeShared
+		opt.Dedicated = false
+		return s.openBulkTransportWithMode(ctx, transport, opt)
+	default:
+		opt.Mode = BulkOpenModeShared
+		opt.Dedicated = false
+		return s.openBulkTransportWithMode(ctx, transport, opt)
+	}
+}
+
+func (s *ServerCommon) openBulkTransportWithMode(ctx context.Context, transport *TransportConn, opt BulkOpenOptions) (Bulk, error) {
 	if s == nil {
 		return nil, errBulkServerNil
 	}
+	opt = applyBulkOpenTuningDefaults(opt, s.bulkOpenTuningSnapshot())
 	if transport == nil {
 		return nil, errBulkTransportNil
 	}
@@ -118,17 +231,44 @@ func (s *ServerCommon) OpenBulkTransport(ctx context.Context, transport *Transpo
 			req.AttachToken = newBulkAttachToken()
 		}
 		bulk := newBulkHandle(logical.stopContextSnapshot(), runtime, scope, req, 0, logical, transport, transport.TransportGeneration(), serverBulkCloseSender(s, logical, transport), serverBulkResetSender(s, logical, transport), serverBulkDataSender(s, transport), serverBulkWriteSender(s, logical, transport), serverBulkReleaseSender(s, logical, transport))
+		bulk.markAcceptHandled()
 		if err := runtime.register(scope, bulk); err != nil {
 			return nil, err
 		}
+		s.attachServerDedicatedSidecarIfExists(logical, bulk)
 		resp, err := sendBulkOpenServerTransport(ctx, s, transport, req)
 		if err != nil {
-			runtime.remove(scope, bulk.ID())
+			bulk.markReset(err)
+			return nil, err
+		}
+		if resp.DataID != 0 && resp.DataID != req.DataID {
+			err = errBulkAlreadyExists
+			_, _ = sendBulkResetServerTransport(context.Background(), s, transport, BulkResetRequest{
+				BulkID: req.BulkID,
+				DataID: req.DataID,
+				Error:  "bulk dedicated data id mismatch",
+			})
+			bulk.markReset(err)
 			return nil, err
 		}
 		if resp.TransportGeneration != 0 {
 			bulk.transportGeneration = resp.TransportGeneration
 		}
+		if resp.FastPathVersion != 0 {
+			bulk.fastPathVersion = normalizeBulkFastPathVersion(resp.FastPathVersion)
+		}
+		if resp.AttachToken != "" {
+			bulk.setDedicatedAttachToken(resp.AttachToken)
+		}
+		if err := bulk.waitAcceptReady(ctx); err != nil {
+			_, _ = sendBulkResetServerTransport(context.Background(), s, transport, BulkResetRequest{
+				BulkID: req.BulkID,
+				DataID: req.DataID,
+				Error:  err.Error(),
+			})
+			bulk.markReset(err)
+			return nil, err
+		}
 		return bulk, nil
 	}
 	resp, err := sendBulkOpenServerTransport(ctx, s, transport, req)
@@ -138,6 +278,9 @@ func (s *ServerCommon) OpenBulkTransport(ctx context.Context, transport *Transpo
 	if resp.DataID != 0 {
 		req.DataID = resp.DataID
 	}
+	if resp.FastPathVersion != 0 {
+		req.FastPathVersion = resp.FastPathVersion
+	}
 	req.Dedicated = resp.Dedicated
 	if resp.AttachToken != "" {
 		req.AttachToken = resp.AttachToken
@@ -146,6 +289,7 @@ func (s *ServerCommon) OpenBulkTransport(ctx context.Context, transport *Transpo
 		return nil, errBulkDataIDEmpty
 	}
 	bulk := newBulkHandle(logical.stopContextSnapshot(), runtime, scope, req, 0, logical, transport, resp.TransportGeneration, serverBulkCloseSender(s, logical, transport), serverBulkResetSender(s, logical, transport), serverBulkDataSender(s, transport), serverBulkWriteSender(s, logical, transport), serverBulkReleaseSender(s, logical, transport))
+	bulk.markAcceptHandled()
 	if err := runtime.register(scope, bulk); err != nil {
 		_, _ = sendBulkResetServerTransport(context.Background(), s, transport, BulkResetRequest{
 			BulkID: req.BulkID,
@@ -154,6 +298,7 @@ func (s *ServerCommon) OpenBulkTransport(ctx context.Context, transport *Transpo
 		})
 		return nil, err
 	}
+	s.attachServerDedicatedSidecarIfExists(logical, bulk)
 	return bulk, nil
 }
 
@@ -164,15 +309,16 @@ func serverBulkRequest(runtime *bulkRuntime, opt BulkOpenOptions) BulkOpenReques
 		id = runtime.nextID()
 	}
 	return normalizeBulkOpenRequest(BulkOpenRequest{
-		BulkID:       id,
-		Range:        opt.Range,
-		Metadata:     cloneBulkMetadata(opt.Metadata),
-		ReadTimeout:  opt.ReadTimeout,
-		WriteTimeout: opt.WriteTimeout,
-		Dedicated:    opt.Dedicated,
-		ChunkSize:    opt.ChunkSize,
-		WindowBytes:  opt.WindowBytes,
-		MaxInFlight:  opt.MaxInFlight,
+		BulkID:          id,
+		FastPathVersion: bulkFastPathVersionCurrent,
+		Range:           opt.Range,
+		Metadata:        cloneBulkMetadata(opt.Metadata),
+		ReadTimeout:     opt.ReadTimeout,
+		WriteTimeout:    opt.WriteTimeout,
+		Dedicated:       opt.Dedicated,
+		ChunkSize:       opt.ChunkSize,
+		WindowBytes:     opt.WindowBytes,
+		MaxInFlight:     opt.MaxInFlight,
 	})
 }
 
@@ -247,12 +393,12 @@ func serverBulkDataSender(s *ServerCommon, transport *TransportConn) bulkDataSen
 		if dataID == 0 {
 			return errBulkDataPathNotReady
 		}
-		return s.sendFastBulkDataTransport(ctx, bulk.LogicalConn(), transport, dataID, bulk.nextOutboundDataSeq(), chunk)
+		return s.sendFastBulkDataTransport(ctx, bulk.LogicalConn(), transport, dataID, bulk.nextOutboundDataSeq(), chunk, bulk.fastPathVersionSnapshot())
 	}
 }
 
 func serverBulkWriteSender(s *ServerCommon, logical *LogicalConn, transport *TransportConn) bulkWriteSender {
-	return func(ctx context.Context, bulk *bulkHandle, payload []byte) (int, error) {
+	return func(ctx context.Context, bulk *bulkHandle, startSeq uint64, payload []byte, payloadOwned bool) (int, error) {
 		if s == nil {
 			return 0, errBulkServerNil
 		}
@@ -267,7 +413,7 @@ func serverBulkWriteSender(s *ServerCommon, logical *LogicalConn, transport *Tra
 			if err := bulk.waitDedicatedReady(ctx); err != nil {
 				return 0, err
 			}
-			return s.sendDedicatedBulkWrite(ctx, bulk.LogicalConn(), bulk, payload)
+			return s.sendDedicatedBulkWrite(ctx, bulk.LogicalConn(), bulk, startSeq, payload)
 		}
 		if transport == nil {
 			return 0, errBulkTransportNil
@@ -275,7 +421,14 @@ func serverBulkWriteSender(s *ServerCommon, logical *LogicalConn, transport *Tra
 		if !transport.IsCurrent() {
 			return 0, errTransportDetached
 		}
-		return 0, nil
+		if bulk == nil {
+			return 0, errBulkRuntimeNil
+		}
+		dataID := bulk.dataIDSnapshot()
+		if dataID == 0 {
+			return 0, errBulkDataPathNotReady
+		}
+		return s.sendFastBulkWriteTransport(ctx, bulk.LogicalConn(), transport, dataID, startSeq, bulk.chunkSize, bulk.fastPathVersionSnapshot(), payload, payloadOwned)
 	}
 }
 
@@ -287,8 +440,20 @@ func serverBulkReleaseSender(s *ServerCommon, logical *LogicalConn, transport *T
 		if bytes <= 0 && chunks <= 0 {
 			return nil
 		}
+		ctx, cancel, err := bulk.newWriteContext(bulk.Context(), bulk.writeTimeout)
+		if err != nil {
+			return err
+		}
+		defer cancel()
 		if bulk.Dedicated() {
-			return s.sendDedicatedBulkRelease(context.Background(), logical, bulk, bytes, chunks)
+			return s.sendDedicatedBulkRelease(ctx, logical, bulk, bytes, chunks)
+		}
+		if transport != nil && transport.IsCurrent() && bulk.fastPathVersionSnapshot() >= bulkFastPathVersionV2 {
+			payload, err := encodeBulkDedicatedReleasePayload(bytes, chunks)
+			if err != nil {
+				return err
+			}
+			return s.sendFastBulkControlTransport(ctx, logical, transport, bulkFastPayloadTypeRelease, 0, bulk.dataIDSnapshot(), 0, bulk.fastPathVersionSnapshot(), payload)
 		}
 		req := BulkReleaseRequest{
 			BulkID: bulk.ID(),

server_config.go

@@ -35,6 +35,7 @@ func (s *ServerCommon) SetDefaultCommEncode(fn func([]byte, []byte) []byte) {
 	s.defaultFastStreamEncode = nil
 	s.defaultFastBulkEncode = nil
 	s.defaultFastPlainEncode = nil
+	s.defaultModernPSKRuntime = nil
 	s.securityReadyCheck = false
 }
 
@@ -45,6 +46,7 @@ func (s *ServerCommon) SetDefaultCommDecode(fn func([]byte, []byte) []byte) {
 	s.defaultFastStreamEncode = nil
 	s.defaultFastBulkEncode = nil
 	s.defaultFastPlainEncode = nil
+	s.defaultModernPSKRuntime = nil
 	s.securityReadyCheck = false
 }
 
@@ -97,6 +99,13 @@ func (s *ServerCommon) GetSecretKey() []byte {
 // Prefer UseModernPSKServer or UseLegacySecurityServer.
 func (s *ServerCommon) SetSecretKey(key []byte) {
 	s.SecretKey = key
+	if len(key) == 0 {
+		s.defaultModernPSKRuntime = nil
+	} else if runtime, err := newModernPSKCodecRuntime(key, defaultModernPSKAAD); err == nil {
+		s.defaultModernPSKRuntime = runtime
+	} else {
+		s.defaultModernPSKRuntime = nil
+	}
 	s.securityReadyCheck = len(key) == 0
 }
 

server_inbound_source.go

@@ -59,26 +59,8 @@ func (s *ServerCommon) pushMessageSourceFast(queue *stario.StarQueue, data []byt
 	if queue == nil || dispatcher == nil || len(data) == 0 {
 		return false
 	}
-	if err := queue.ParseMessageOwned(data, source, func(msg stario.MsgQueue) error {
-		payload := msg.Msg
-		source := msg.Conn
-		s.wg.Add(1)
-		if !dispatcher.Dispatch(serverInboundDispatchSource(source), func() {
-			defer s.wg.Done()
-			logical, transport := s.resolveInboundSource(source)
-			if logical == nil {
-				return
-			}
-			now := time.Now()
-			inboundConn := serverInboundConn(source)
-			if err := s.dispatchInboundTransportPayload(logical, transport, inboundConn, payload, now); err != nil {
-				if s.showError || s.debugMode {
-					fmt.Println("server decode envelope error", err)
-				}
-			}
-		}) {
-			s.wg.Done()
-		}
+	if err := queue.ParseMessageView(data, source, func(frame stario.FrameView) error {
+		s.pushTransportPayloadSourceFast(frame.Payload, nil, frame.Conn)
 		return nil
 	}); err != nil && (s.showError || s.debugMode) {
 		fmt.Println("server parse inbound frame error", err)
@@ -86,6 +68,59 @@ func (s *ServerCommon) pushMessageSourceFast(queue *stario.StarQueue, data []byt
 	return true
 }
 
+func (s *ServerCommon) pushTransportPayloadSourceFast(payload []byte, release func(), source interface{}) bool {
+	dispatcher := s.serverInboundDispatcherSnapshot()
+	if len(payload) == 0 {
+		if release != nil {
+			release()
+		}
+		return false
+	}
+	if dispatcher == nil {
+		queue := s.serverQueueSnapshot()
+		if queue == nil {
+			if release != nil {
+				release()
+			}
+			return false
+		}
+		frame := queue.BuildMessage(payload)
+		if release != nil {
+			release()
+		}
+		if err := queue.ParseMessage(frame, source); err != nil && (s.showError || s.debugMode) {
+			fmt.Println("server enqueue inbound frame error", err)
+		}
+		return true
+	}
+	if s.tryDispatchBorrowedBulkTransportPayload(source, payload) {
+		if release != nil {
+			release()
+		}
+		return true
+	}
+	owned := append([]byte(nil), payload...)
+	if release != nil {
+		release()
+	}
+	s.wg.Add(1)
+	if !dispatcher.Dispatch(serverInboundDispatchSource(source), func() {
+		defer s.wg.Done()
+		logical, transport := s.resolveInboundSource(source)
+		if logical == nil {
+			return
+		}
+		now := time.Now()
+		inboundConn := serverInboundConn(source)
+		if err := s.dispatchInboundTransportPayload(logical, transport, inboundConn, owned, now); err != nil && (s.showError || s.debugMode) {
+			fmt.Println("server decode envelope error", err)
+		}
+	}) {
+		s.wg.Done()
+	}
+	return true
+}
+
 func serverInboundConn(source interface{}) net.Conn {
 	switch data := source.(type) {
 	case net.Conn:

server_listen.go

@@ -130,6 +130,7 @@ func (s *ServerCommon) removeLogical(logical *LogicalConn) {
 	s.getFileAckPool().closeScopeFamily(scope)
 	s.getSignalAckPool().closeScopeFamily(scope)
 	s.getReceivedSignalCache().closeScope(scope)
+	s.closeServerDedicatedSidecar(logical)
 	s.getPeerRegistry().removeLogical(logical)
 }
 

server_session.go

@@ -57,6 +57,7 @@ func (s *ServerCommon) detachClientSessionTransport(client *ClientConn, reason s
 	if runtime := s.getBulkRuntime(); runtime != nil {
 		runtime.closeScope(serverFileScope(client), errTransportDetached)
 	}
+	s.closeServerDedicatedSidecar(logicalConnFromClient(client))
 	client.detachServerOwnedTransport()
 }
 
@@ -80,6 +81,7 @@ func (s *ServerCommon) detachLogicalSessionTransport(logical *LogicalConn, reaso
 	if runtime := s.getBulkRuntime(); runtime != nil {
 		runtime.closeScope(serverFileScope(logical), errTransportDetached)
 	}
+	s.closeServerDedicatedSidecar(logical)
 	logical.detachServerOwnedTransport()
 }
 
@@ -108,6 +110,7 @@ func (s *ServerCommon) registerAcceptedLogical(logical *LogicalConn) *LogicalCon
 	}
 	logical.setServer(s)
 	logical.applyAttachmentProfile(s.maxReadTimeout, s.maxWriteTimeout, s.defaultMsgEn, s.defaultMsgDe, s.defaultFastStreamEncode, s.defaultFastBulkEncode, s.defaultFastPlainEncode, s.handshakeRsaKey, s.SecretKey)
+	logical.setModernPSKRuntime(s.defaultModernPSKRuntime)
 	logical.markHeartbeatNow()
 	return s.getPeerRegistry().registerLogical(logical)
 }

server_stream.go

@@ -33,6 +33,11 @@ func (s *ServerCommon) OpenStreamLogical(ctx context.Context, logical *LogicalCo
 	if resp.DataID != 0 {
 		req.DataID = resp.DataID
 	}
+	if resp.FastPathVersion != 0 {
+		req.FastPathVersion = resp.FastPathVersion
+	} else {
+		req.FastPathVersion = streamFastPathVersionV1
+	}
 	req.Metadata = mergeStreamMetadata(req.Metadata, resp.Metadata)
 	transport := logical.CurrentTransportConn()
 	stream := newStreamHandle(logical.stopContextSnapshot(), runtime, scope, req, 0, logical, transport, resp.TransportGeneration, serverStreamCloseSender(s, logical, nil), serverStreamResetSender(s, logical, nil), serverStreamDataSender(s, transport), runtime.configSnapshot())
@@ -73,6 +78,11 @@ func (s *ServerCommon) OpenStreamTransport(ctx context.Context, transport *Trans
 	if resp.DataID != 0 {
 		req.DataID = resp.DataID
 	}
+	if resp.FastPathVersion != 0 {
+		req.FastPathVersion = resp.FastPathVersion
+	} else {
+		req.FastPathVersion = streamFastPathVersionV1
+	}
 	req.Metadata = mergeStreamMetadata(req.Metadata, resp.Metadata)
 	stream := newStreamHandle(logical.stopContextSnapshot(), runtime, scope, req, 0, logical, transport, resp.TransportGeneration, serverStreamCloseSender(s, logical, transport), serverStreamResetSender(s, logical, transport), serverStreamDataSender(s, transport), runtime.configSnapshot())
 	if err := runtime.register(scope, stream); err != nil {
@@ -91,11 +101,12 @@ func serverStreamRequest(runtime *streamRuntime, opt StreamOpenOptions) StreamOp
 		id = runtime.nextID()
 	}
 	return normalizeStreamOpenRequest(StreamOpenRequest{
-		StreamID:     id,
-		Channel:      opt.Channel,
-		Metadata:     cloneStreamMetadata(opt.Metadata),
-		ReadTimeout:  opt.ReadTimeout,
-		WriteTimeout: opt.WriteTimeout,
+		StreamID:        id,
+		FastPathVersion: streamFastPathVersionCurrent,
+		Channel:         opt.Channel,
+		Metadata:        cloneStreamMetadata(opt.Metadata),
+		ReadTimeout:     opt.ReadTimeout,
+		WriteTimeout:    opt.WriteTimeout,
 	})
 }
 
@@ -148,7 +159,7 @@ func serverStreamDataSender(s *ServerCommon, transport *TransportConn) streamDat
 			}
 		}
 		if dataID := stream.dataIDSnapshot(); dataID != 0 {
-			return s.sendFastStreamDataTransport(stream.LogicalConn(), transport, dataID, stream.nextOutboundDataSeq(), chunk)
+			return s.sendFastStreamDataTransport(ctx, stream.LogicalConn(), transport, stream, chunk)
 		}
 		return s.sendEnvelopeTransport(transport, newStreamDataEnvelope(stream.ID(), chunk))
 	}

servertype.go

@@ -24,6 +24,8 @@ type Server interface {
 	SetStreamConfig(StreamConfig)
 	SetTransferResumeStore(TransferResumeStore)
 	RecoverTransferSnapshots(context.Context) error
+	SetBulkOpenTuning(BulkOpenTuning)
+	BulkOpenTuning() BulkOpenTuning
 	SetFileReceiveDir(dir string) error
 	send(c *ClientConn, msg TransferMsg) (WaitMsg, error)
 	sendEnvelope(c *ClientConn, env Envelope) error

session_runtime_snapshot.go

@@ -6,18 +6,34 @@ import (
 )
 
 type ClientRuntimeSnapshot struct {
-	OwnerState        string
-	Alive             bool
-	SessionEpoch      uint64
-	TransportAttached bool
-	HasRuntimeConn    bool
-	HasRuntimeQueue   bool
-	HasRuntimeStopCtx bool
-	ConnectSource     string
-	ConnectNetwork    string
-	ConnectAddress    string
-	CanReconnect      bool
-	Retry             ConnectionRetrySnapshot
+	OwnerState                                string
+	Alive                                     bool
+	SessionEpoch                              uint64
+	TransportAttached                         bool
+	HasRuntimeConn                            bool
+	HasRuntimeQueue                           bool
+	HasRuntimeStopCtx                         bool
+	ConnectSource                             string
+	ConnectNetwork                            string
+	ConnectAddress                            string
+	CanReconnect                              bool
+	BulkNetworkProfile                        string
+	BulkDefaultMode                           string
+	BulkChunkSize                             int
+	BulkWindowBytes                           int
+	BulkMaxInFlight                           int
+	BulkAttachLimit                           int
+	BulkActiveLimit                           int
+	BulkLaneLimit                             int
+	BulkAttachAttempts                        int64
+	BulkAttachRetries                         int64
+	BulkAttachSuccess                         int64
+	BulkAutoFallbacks                         int64
+	TransportBulkAdaptiveSoftPayloadBytes     int
+	TransportStreamAdaptiveSoftPayloadBytes   int
+	TransportStreamAdaptiveWaitThresholdBytes int
+	TransportStreamAdaptiveFlushDelay         time.Duration
+	Retry                                     ConnectionRetrySnapshot
 }
 
 type ServerRuntimeSnapshot struct {
@@ -33,36 +49,43 @@ type ServerRuntimeSnapshot struct {
 	HasRuntimeUDPListener           bool
 	HasRuntimeQueue                 bool
 	HasRuntimeStopCtx               bool
+	BulkChunkSize                   int
+	BulkWindowBytes                 int
+	BulkMaxInFlight                 int
 	Retry                           ConnectionRetrySnapshot
 }
 
 type ClientConnRuntimeSnapshot struct {
-	ClientID                  string
-	RemoteAddress             string
-	Alive                     bool
-	Reason                    string
-	Error                     string
-	IdentityBound             bool
-	UsesStreamTransport       bool
-	TransportGeneration       uint64
-	TransportAttached         bool
-	HasRuntimeConn            bool
-	HasRuntimeStopCtx         bool
-	TransportAttachCount      uint64
-	TransportDetachCount      uint64
-	LastTransportAttachAt     time.Time
-	DetachedClientKeepSec     int64
-	LastHeartbeatAt           time.Time
-	TransportDetachReason     string
-	TransportDetachKind       string
-	TransportDetachGeneration uint64
-	TransportDetachError      string
-	TransportDetachedAt       time.Time
-	TransportDetachHasExpiry  bool
-	TransportDetachExpiry     time.Time
-	TransportDetachRemaining  time.Duration
-	TransportDetachExpired    bool
-	ReattachEligible          bool
+	ClientID                                  string
+	RemoteAddress                             string
+	Alive                                     bool
+	Reason                                    string
+	Error                                     string
+	IdentityBound                             bool
+	UsesStreamTransport                       bool
+	TransportGeneration                       uint64
+	TransportAttached                         bool
+	HasRuntimeConn                            bool
+	HasRuntimeStopCtx                         bool
+	TransportAttachCount                      uint64
+	TransportDetachCount                      uint64
+	LastTransportAttachAt                     time.Time
+	DetachedClientKeepSec                     int64
+	LastHeartbeatAt                           time.Time
+	TransportDetachReason                     string
+	TransportDetachKind                       string
+	TransportDetachGeneration                 uint64
+	TransportDetachError                      string
+	TransportDetachedAt                       time.Time
+	TransportDetachHasExpiry                  bool
+	TransportDetachExpiry                     time.Time
+	TransportDetachRemaining                  time.Duration
+	TransportDetachExpired                    bool
+	ReattachEligible                          bool
+	TransportBulkAdaptiveSoftPayloadBytes     int
+	TransportStreamAdaptiveSoftPayloadBytes   int
+	TransportStreamAdaptiveWaitThresholdBytes int
+	TransportStreamAdaptiveFlushDelay         time.Duration
 }
 
 func (c *ClientCommon) clientRuntimeSnapshot() ClientRuntimeSnapshot {
@@ -85,6 +108,26 @@ func (c *ClientCommon) clientRuntimeSnapshot() ClientRuntimeSnapshot {
 		snapshot.ConnectAddress = source.addr
 		snapshot.CanReconnect = source.canReconnect()
 	}
+	snapshot.BulkNetworkProfile = bulkNetworkProfileName(c.BulkNetworkProfile())
+	snapshot.BulkDefaultMode = bulkOpenModeName(c.BulkDefaultOpenMode())
+	tuning := c.BulkOpenTuning()
+	cfg := c.BulkDedicatedAttachConfig()
+	snapshot.BulkChunkSize = tuning.ChunkSize
+	snapshot.BulkWindowBytes = tuning.WindowBytes
+	snapshot.BulkMaxInFlight = tuning.MaxInFlight
+	snapshot.BulkAttachLimit = cfg.AttachLimit
+	snapshot.BulkActiveLimit = cfg.ActiveLimit
+	snapshot.BulkLaneLimit = cfg.LaneLimit
+	snapshot.BulkAttachAttempts = c.bulkAttachAttemptCount.Load()
+	snapshot.BulkAttachRetries = c.bulkAttachRetryCount.Load()
+	snapshot.BulkAttachSuccess = c.bulkAttachSuccessCount.Load()
+	snapshot.BulkAutoFallbacks = c.bulkAttachFallbackCount.Load()
+	if binding := c.clientTransportBindingSnapshot(); binding != nil {
+		snapshot.TransportBulkAdaptiveSoftPayloadBytes = binding.bulkAdaptiveSoftPayloadBytesSnapshot()
+		snapshot.TransportStreamAdaptiveSoftPayloadBytes = binding.streamAdaptiveSoftPayloadBytesSnapshot()
+		snapshot.TransportStreamAdaptiveWaitThresholdBytes = binding.streamAdaptiveWaitThresholdBytesSnapshot()
+		snapshot.TransportStreamAdaptiveFlushDelay = binding.streamAdaptiveFlushDelaySnapshot()
+	}
 	snapshot.Retry = c.connectionRetrySnapshot()
 	return snapshot
 }
@@ -118,6 +161,10 @@ func (s *ServerCommon) serverRuntimeSnapshot() ServerRuntimeSnapshot {
 		snapshot.HasRuntimeQueue = rt.queue != nil
 		snapshot.HasRuntimeStopCtx = rt.stopCtx != nil
 	}
+	tuning := s.BulkOpenTuning()
+	snapshot.BulkChunkSize = tuning.ChunkSize
+	snapshot.BulkWindowBytes = tuning.WindowBytes
+	snapshot.BulkMaxInFlight = tuning.MaxInFlight
 	snapshot.Retry = s.connectionRetrySnapshot()
 	return snapshot
 }
@@ -153,6 +200,12 @@ func (c *ClientConn) clientConnRuntimeSnapshot() ClientConnRuntimeSnapshot {
 		snapshot.HasRuntimeConn = c.clientConnTransportSnapshot() != nil
 		snapshot.HasRuntimeStopCtx = rt.stopCtx != nil
 	}
+	if binding := c.clientConnTransportBindingSnapshot(); binding != nil {
+		snapshot.TransportBulkAdaptiveSoftPayloadBytes = binding.bulkAdaptiveSoftPayloadBytesSnapshot()
+		snapshot.TransportStreamAdaptiveSoftPayloadBytes = binding.streamAdaptiveSoftPayloadBytesSnapshot()
+		snapshot.TransportStreamAdaptiveWaitThresholdBytes = binding.streamAdaptiveWaitThresholdBytesSnapshot()
+		snapshot.TransportStreamAdaptiveFlushDelay = binding.streamAdaptiveFlushDelaySnapshot()
+	}
 	if detach := c.clientConnTransportDetachSnapshot(); detach != nil {
 		snapshot.TransportDetachReason = detach.Reason
 		snapshot.TransportDetachKind = c.clientConnTransportDetachKindSnapshot()

session_runtime_snapshot_test.go

@@ -37,6 +37,45 @@ func TestGetClientRuntimeSnapshotDefaults(t *testing.T) {
 	if snapshot.ConnectSource != "" || snapshot.ConnectNetwork != "" || snapshot.ConnectAddress != "" || snapshot.CanReconnect {
 		t.Fatalf("unexpected default connect source snapshot: %+v", snapshot)
 	}
+	if got, want := snapshot.BulkNetworkProfile, "default"; got != want {
+		t.Fatalf("BulkNetworkProfile mismatch: got %q want %q", got, want)
+	}
+	if got, want := snapshot.BulkDefaultMode, "shared"; got != want {
+		t.Fatalf("BulkDefaultMode mismatch: got %q want %q", got, want)
+	}
+	if got, want := snapshot.BulkChunkSize, defaultBulkChunkSize; got != want {
+		t.Fatalf("BulkChunkSize mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkWindowBytes, defaultBulkOpenWindowBytes; got != want {
+		t.Fatalf("BulkWindowBytes mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkMaxInFlight, defaultBulkOpenMaxInFlight; got != want {
+		t.Fatalf("BulkMaxInFlight mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkAttachLimit, defaultBulkDedicatedAttachLimit; got != want {
+		t.Fatalf("BulkAttachLimit mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkActiveLimit, defaultBulkDedicatedActiveLimit; got != want {
+		t.Fatalf("BulkActiveLimit mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkLaneLimit, defaultBulkDedicatedLaneLimit; got != want {
+		t.Fatalf("BulkLaneLimit mismatch: got %d want %d", got, want)
+	}
+	if snapshot.BulkAttachAttempts != 0 || snapshot.BulkAttachRetries != 0 || snapshot.BulkAttachSuccess != 0 || snapshot.BulkAutoFallbacks != 0 {
+		t.Fatalf("unexpected default bulk attach counters: %+v", snapshot)
+	}
+	if snapshot.TransportBulkAdaptiveSoftPayloadBytes != 0 {
+		t.Fatalf("TransportBulkAdaptiveSoftPayloadBytes mismatch: got %d want 0", snapshot.TransportBulkAdaptiveSoftPayloadBytes)
+	}
+	if snapshot.TransportStreamAdaptiveSoftPayloadBytes != 0 {
+		t.Fatalf("TransportStreamAdaptiveSoftPayloadBytes mismatch: got %d want 0", snapshot.TransportStreamAdaptiveSoftPayloadBytes)
+	}
+	if snapshot.TransportStreamAdaptiveWaitThresholdBytes != 0 {
+		t.Fatalf("TransportStreamAdaptiveWaitThresholdBytes mismatch: got %d want 0", snapshot.TransportStreamAdaptiveWaitThresholdBytes)
+	}
+	if snapshot.TransportStreamAdaptiveFlushDelay != 0 {
+		t.Fatalf("TransportStreamAdaptiveFlushDelay mismatch: got %s want 0", snapshot.TransportStreamAdaptiveFlushDelay)
+	}
 	if snapshot.Retry != (ConnectionRetrySnapshot{}) {
 		t.Fatalf("Retry snapshot mismatch: %+v", snapshot.Retry)
 	}
@@ -78,11 +117,144 @@ func TestGetServerRuntimeSnapshotDefaults(t *testing.T) {
 	if !snapshot.HasRuntimeStopCtx {
 		t.Fatalf("HasRuntimeStopCtx mismatch: got %v want true", snapshot.HasRuntimeStopCtx)
 	}
+	if got, want := snapshot.BulkChunkSize, defaultBulkChunkSize; got != want {
+		t.Fatalf("BulkChunkSize mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkWindowBytes, defaultBulkOpenWindowBytes; got != want {
+		t.Fatalf("BulkWindowBytes mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkMaxInFlight, defaultBulkOpenMaxInFlight; got != want {
+		t.Fatalf("BulkMaxInFlight mismatch: got %d want %d", got, want)
+	}
 	if snapshot.Retry != (ConnectionRetrySnapshot{}) {
 		t.Fatalf("Retry snapshot mismatch: %+v", snapshot.Retry)
 	}
 }
 
+func TestGetClientRuntimeSnapshotIncludesBulkAttachStats(t *testing.T) {
+	client := NewClient().(*ClientCommon)
+	client.SetBulkNetworkProfile(BulkNetworkProfileWAN)
+	client.SetBulkDefaultOpenMode(BulkOpenModeAuto)
+	client.bulkAttachAttemptCount.Store(11)
+	client.bulkAttachRetryCount.Store(5)
+	client.bulkAttachSuccessCount.Store(6)
+	client.bulkAttachFallbackCount.Store(3)
+
+	snapshot, err := GetClientRuntimeSnapshot(client)
+	if err != nil {
+		t.Fatalf("GetClientRuntimeSnapshot failed: %v", err)
+	}
+	if got, want := snapshot.BulkNetworkProfile, "wan"; got != want {
+		t.Fatalf("BulkNetworkProfile mismatch: got %q want %q", got, want)
+	}
+	if got, want := snapshot.BulkDefaultMode, "auto"; got != want {
+		t.Fatalf("BulkDefaultMode mismatch: got %q want %q", got, want)
+	}
+	if got, want := snapshot.BulkChunkSize, 512*1024; got != want {
+		t.Fatalf("BulkChunkSize mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkWindowBytes, 8*1024*1024; got != want {
+		t.Fatalf("BulkWindowBytes mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkMaxInFlight, 16; got != want {
+		t.Fatalf("BulkMaxInFlight mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkAttachLimit, 2; got != want {
+		t.Fatalf("BulkAttachLimit mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkActiveLimit, 4096; got != want {
+		t.Fatalf("BulkActiveLimit mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkLaneLimit, 2; got != want {
+		t.Fatalf("BulkLaneLimit mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkAttachAttempts, int64(11); got != want {
+		t.Fatalf("BulkAttachAttempts mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkAttachRetries, int64(5); got != want {
+		t.Fatalf("BulkAttachRetries mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkAttachSuccess, int64(6); got != want {
+		t.Fatalf("BulkAttachSuccess mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.BulkAutoFallbacks, int64(3); got != want {
+		t.Fatalf("BulkAutoFallbacks mismatch: got %d want %d", got, want)
+	}
+}
+
+func TestGetClientRuntimeSnapshotIncludesAdaptiveBulkSoftPayload(t *testing.T) {
+	client := NewClient().(*ClientCommon)
+	stopCtx, stopFn := context.WithCancel(context.Background())
+	defer stopFn()
+	queue := stario.NewQueueCtx(stopCtx, 4, math.MaxUint32)
+	left, right := net.Pipe()
+	defer left.Close()
+	defer right.Close()
+
+	binding := newTransportBinding(left, queue)
+	binding.observeBulkAdaptivePayloadWrite(8*1024*1024, 640*time.Millisecond, 0, nil)
+	client.setClientSessionRuntime(&clientSessionRuntime{
+		transport: binding,
+		stopCtx:   stopCtx,
+		stopFn:    stopFn,
+		queue:     queue,
+		epoch:     1,
+	})
+	client.markSessionStarted()
+
+	snapshot, err := GetClientRuntimeSnapshot(client)
+	if err != nil {
+		t.Fatalf("GetClientRuntimeSnapshot failed: %v", err)
+	}
+	if got, want := snapshot.TransportBulkAdaptiveSoftPayloadBytes, bulkAdaptiveSoftPayloadMinBytes; got != want {
+		t.Fatalf("TransportBulkAdaptiveSoftPayloadBytes mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.TransportStreamAdaptiveSoftPayloadBytes, streamAdaptiveSoftPayloadStartBytes; got != want {
+		t.Fatalf("TransportStreamAdaptiveSoftPayloadBytes mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.TransportStreamAdaptiveWaitThresholdBytes, streamBatchWaitThreshold; got != want {
+		t.Fatalf("TransportStreamAdaptiveWaitThresholdBytes mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.TransportStreamAdaptiveFlushDelay, streamBatchMaxFlushDelay; got != want {
+		t.Fatalf("TransportStreamAdaptiveFlushDelay mismatch: got %s want %s", got, want)
+	}
+}
+
+func TestGetClientRuntimeSnapshotIncludesAdaptiveStreamTuning(t *testing.T) {
+	client := NewClient().(*ClientCommon)
+	stopCtx, stopFn := context.WithCancel(context.Background())
+	defer stopFn()
+	queue := stario.NewQueueCtx(stopCtx, 4, math.MaxUint32)
+	left, right := net.Pipe()
+	defer left.Close()
+	defer right.Close()
+
+	binding := newTransportBinding(left, queue)
+	binding.observeStreamAdaptivePayloadWrite(2*1024*1024, 640*time.Millisecond, 0, nil)
+	client.setClientSessionRuntime(&clientSessionRuntime{
+		transport: binding,
+		stopCtx:   stopCtx,
+		stopFn:    stopFn,
+		queue:     queue,
+		epoch:     1,
+	})
+	client.markSessionStarted()
+
+	snapshot, err := GetClientRuntimeSnapshot(client)
+	if err != nil {
+		t.Fatalf("GetClientRuntimeSnapshot failed: %v", err)
+	}
+	if got, want := snapshot.TransportStreamAdaptiveSoftPayloadBytes, streamAdaptiveSoftPayloadMinBytes; got != want {
+		t.Fatalf("TransportStreamAdaptiveSoftPayloadBytes mismatch: got %d want %d", got, want)
+	}
+	if got, want := snapshot.TransportStreamAdaptiveWaitThresholdBytes, streamAdaptiveWaitThresholdMinBytes; got != want {
+		t.Fatalf("TransportStreamAdaptiveWaitThresholdBytes mismatch: got %d want %d", got, want)
+	}
+	if got := snapshot.TransportStreamAdaptiveFlushDelay; got != 0 {
+		t.Fatalf("TransportStreamAdaptiveFlushDelay mismatch: got %s want 0", got)
+	}
+}
+
 func TestGetRuntimeSnapshotRejectsNil(t *testing.T) {
 	if _, err := GetClientRuntimeSnapshot(nil); !errors.Is(err, errClientRuntimeSnapshotNil) {
 		t.Fatalf("GetClientRuntimeSnapshot nil error = %v, want %v", err, errClientRuntimeSnapshotNil)

session_state.go

@@ -140,6 +140,7 @@ func (c *ClientCommon) cleanupClientSessionResources() {
 	if runtime := c.getBulkRuntime(); runtime != nil {
 		runtime.closeAll(errServiceShutdown)
 	}
+	c.closeClientDedicatedSidecar()
 }
 
 func (s *ServerCommon) cleanupServerSessionResources() {
@@ -158,4 +159,5 @@ func (s *ServerCommon) cleanupServerSessionResources() {
 	if runtime := s.getBulkRuntime(); runtime != nil {
 		runtime.closeAll(errServiceShutdown)
 	}
+	s.closeAllServerDedicatedSidecars()
 }

signal_benchmark_test.go

@@ -76,7 +76,7 @@ func startSignalRoundTripServerForBenchmark(b *testing.B) (*ServerCommon, string
 	server.SetLink("signal-roundtrip", func(msg *Message) {
 		_ = msg.Reply([]byte("ack:" + string(msg.Value)))
 	})
-	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+	if err := server.Listen("tcp", benchmarkTCPListenAddr(b)); err != nil {
 		if benchmarkListenPermissionDenied(err) {
 			b.Skipf("tcp benchmark requires local listen permission: %v", err)
 		}
@@ -102,7 +102,7 @@ func newSignalRoundTripBenchmarkClient(b *testing.B, addr string) *ClientCommon
 	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
 		b.Fatalf("UseModernPSKClient failed: %v", err)
 	}
-	if err := client.Connect("tcp", addr); err != nil {
+	if err := client.Connect("tcp", benchmarkTCPDialAddr(b, addr)); err != nil {
 		b.Fatalf("client Connect failed: %v", err)
 	}
 	return client

snapshot_binding.go

@@ -3,20 +3,24 @@ package notify
 import "time"
 
 type snapshotBindingDiagnostics struct {
-	BindingOwner              string
-	BindingAlive              bool
-	BindingCurrent            bool
-	BindingReason             string
-	BindingError              string
-	TransportAttached         bool
-	TransportHasRuntimeConn   bool
-	TransportCurrent          bool
-	TransportDetachReason     string
-	TransportDetachKind       string
-	TransportDetachError      string
-	TransportDetachGeneration uint64
-	TransportDetachedAt       time.Time
-	ReattachEligible          bool
+	BindingOwner                            string
+	BindingAlive                            bool
+	BindingCurrent                          bool
+	BindingReason                           string
+	BindingError                            string
+	BindingBulkAdaptiveSoftPayloadBytes     int
+	BindingStreamAdaptiveSoftPayloadBytes   int
+	BindingStreamAdaptiveWaitThresholdBytes int
+	BindingStreamAdaptiveFlushDelay         time.Duration
+	TransportAttached                       bool
+	TransportHasRuntimeConn                 bool
+	TransportCurrent                        bool
+	TransportDetachReason                   string
+	TransportDetachKind                     string
+	TransportDetachError                    string
+	TransportDetachGeneration               uint64
+	TransportDetachedAt                     time.Time
+	ReattachEligible                        bool
 }
 
 func snapshotBindingDiagnosticsFromClient(c *ClientCommon, sessionEpoch uint64) snapshotBindingDiagnostics {
@@ -36,6 +40,12 @@ func snapshotBindingDiagnosticsFromClient(c *ClientCommon, sessionEpoch uint64)
 	diag.TransportAttached = c.clientTransportAttachedSnapshot()
 	diag.TransportHasRuntimeConn = c.clientTransportConnSnapshot() != nil
 	diag.TransportCurrent = diag.BindingCurrent && diag.TransportAttached
+	if binding := c.clientTransportBindingSnapshot(); binding != nil {
+		diag.BindingBulkAdaptiveSoftPayloadBytes = binding.bulkAdaptiveSoftPayloadBytesSnapshot()
+		diag.BindingStreamAdaptiveSoftPayloadBytes = binding.streamAdaptiveSoftPayloadBytesSnapshot()
+		diag.BindingStreamAdaptiveWaitThresholdBytes = binding.streamAdaptiveWaitThresholdBytesSnapshot()
+		diag.BindingStreamAdaptiveFlushDelay = binding.streamAdaptiveFlushDelaySnapshot()
+	}
 	return diag
 }
 
@@ -72,5 +82,11 @@ func snapshotBindingDiagnosticsFromLogical(logical *LogicalConn, transport *Tran
 		diag.TransportCurrent = runtime.TransportAttached
 	}
 	diag.BindingCurrent = diag.BindingAlive && diag.TransportCurrent
+	if binding := logical.transportBindingSnapshot(); binding != nil {
+		diag.BindingBulkAdaptiveSoftPayloadBytes = binding.bulkAdaptiveSoftPayloadBytesSnapshot()
+		diag.BindingStreamAdaptiveSoftPayloadBytes = binding.streamAdaptiveSoftPayloadBytesSnapshot()
+		diag.BindingStreamAdaptiveWaitThresholdBytes = binding.streamAdaptiveWaitThresholdBytesSnapshot()
+		diag.BindingStreamAdaptiveFlushDelay = binding.streamAdaptiveFlushDelaySnapshot()
+	}
 	return diag
 }

stream.go

@@ -7,6 +7,7 @@ import (
 	"net"
 	"os"
 	"sync"
+	"sync/atomic"
 	"time"
 )
 
@@ -93,7 +94,8 @@ type streamHandle struct {
 	runtimeScope        string
 	id                  string
 	dataID              uint64
-	outboundSeq         uint64
+	fastPathVersion     uint8
+	outboundSeq         atomic.Uint64
 	channel             StreamChannel
 	metadata            StreamMetadata
 	sessionEpoch        uint64
@@ -132,6 +134,9 @@ type streamHandle struct {
 	writeDeadlineOverride bool
 	readDeadlineNotify    chan struct{}
 	writeDeadlineNotify   chan struct{}
+	writeWaitSeq          uint64
+	writeWaitCancel       context.CancelFunc
+	writeWaitChanged      chan struct{}
 	bytesRead             int64
 	bytesWritten          int64
 	readCalls             int64
@@ -157,6 +162,7 @@ func newStreamHandle(parent context.Context, runtime *streamRuntime, runtimeScop
 		runtimeScope:        runtimeScope,
 		id:                  req.StreamID,
 		dataID:              req.DataID,
+		fastPathVersion:     normalizeStreamFastPathVersion(req.FastPathVersion),
 		channel:             normalizeStreamChannel(req.Channel),
 		metadata:            cloneStreamMetadata(req.Metadata),
 		sessionEpoch:        sessionEpoch,
@@ -224,13 +230,25 @@ func (s *streamHandle) dataIDSnapshot() uint64 {
 }
 
 func (s *streamHandle) nextOutboundDataSeq() uint64 {
+	return s.reserveOutboundDataSeqs(1)
+}
+
+func (s *streamHandle) reserveOutboundDataSeqs(count int) uint64 {
 	if s == nil {
 		return 0
 	}
-	s.mu.Lock()
-	defer s.mu.Unlock()
-	s.outboundSeq++
-	return s.outboundSeq
+	if count <= 0 {
+		count = 1
+	}
+	end := s.outboundSeq.Add(uint64(count))
+	return end - uint64(count) + 1
+}
+
+func (s *streamHandle) fastPathVersionSnapshot() uint8 {
+	if s == nil {
+		return streamFastPathVersionV1
+	}
+	return normalizeStreamFastPathVersion(s.fastPathVersion)
 }
 
 func (s *streamHandle) Channel() StreamChannel {
@@ -377,6 +395,7 @@ func (s *streamHandle) Write(p []byte) (int, error) {
 	sendDataFn := s.sendDataFn
 	chunkSize := s.chunkSize
 	writeTimeout := s.writeTimeout
+	writeDeadlineOverride := s.writeDeadlineOverride
 	streamCtx := s.ctx
 	runtime := s.runtime
 	s.mu.Unlock()
@@ -399,6 +418,20 @@ func (s *streamHandle) Write(p []byte) (int, error) {
 			end = len(p)
 		}
 		chunk := p[written:end]
+		if !writeDeadlineOverride && writeTimeout <= 0 {
+			if tryAcquireStreamOutboundBudget(runtime, len(chunk)) {
+				err := sendDataFn(streamCtx, s, chunk)
+				releaseStreamOutboundBudget(runtime, len(chunk))
+				if err != nil {
+					if written > 0 {
+						s.recordWrite(written, time.Now())
+					}
+					return written, s.normalizeWriteError(err)
+				}
+				written = end
+				continue
+			}
+		}
 		sendCtx, cancel, deadlineChanged, err := s.newWriteContext(streamCtx, writeTimeout)
 		if err != nil {
 			if written > 0 {
@@ -464,7 +497,15 @@ func (s *streamHandle) SetWriteDeadline(deadline time.Time) error {
 	s.writeDeadline = deadline
 	s.writeDeadlineOverride = true
 	signalStreamDeadlineChangeLocked(&s.writeDeadlineNotify)
+	waitCancel := s.writeWaitCancel
+	if s.writeWaitChanged != nil {
+		close(s.writeWaitChanged)
+		s.writeWaitChanged = nil
+	}
 	s.mu.Unlock()
+	if waitCancel != nil {
+		waitCancel()
+	}
 	return nil
 }
 
@@ -535,7 +576,6 @@ func (s *streamHandle) newWriteContext(parent context.Context, writeTimeout time
 	}
 	s.mu.Lock()
 	deadline := s.effectiveWriteDeadlineLocked(time.Now(), writeTimeout)
-	deadlineNotify := s.writeDeadlineNotify
 	s.mu.Unlock()
 	if !deadline.IsZero() && !deadline.After(time.Now()) {
 		return nil, func() {}, nil, os.ErrDeadlineExceeded
@@ -548,19 +588,20 @@ func (s *streamHandle) newWriteContext(parent context.Context, writeTimeout time
 		baseCtx, baseCancel = context.WithCancel(parent)
 	}
 	changed := make(chan struct{})
-	done := make(chan struct{})
-	go func() {
-		defer close(done)
-		select {
-		case <-baseCtx.Done():
-		case <-deadlineNotify:
-			close(changed)
-			baseCancel()
-		}
-	}()
+	s.mu.Lock()
+	s.writeWaitSeq++
+	waitSeq := s.writeWaitSeq
+	s.writeWaitCancel = baseCancel
+	s.writeWaitChanged = changed
+	s.mu.Unlock()
 	cancel := func() {
 		baseCancel()
-		<-done
+		s.mu.Lock()
+		if s.writeWaitSeq == waitSeq {
+			s.writeWaitCancel = nil
+			s.writeWaitChanged = nil
+		}
+		s.mu.Unlock()
 	}
 	return baseCtx, cancel, changed, nil
 }
@@ -814,7 +855,11 @@ func (s *streamHandle) pushChunkWithOwnership(chunk []byte, owned bool) error {
 		s.finalize()
 		return err
 	}
-	s.readQueue = append(s.readQueue, stored)
+	if len(s.readBuf) == 0 && len(s.readQueue) == 0 {
+		s.readBuf = stored
+	} else {
+		s.readQueue = append(s.readQueue, stored)
+	}
 	s.bufferedBytes += len(stored)
 	s.notifyReadableLocked()
 	s.mu.Unlock()
@@ -917,6 +962,10 @@ func (s *streamHandle) snapshot() StreamSnapshot {
 	snapshot.BindingCurrent = diag.BindingCurrent
 	snapshot.BindingReason = diag.BindingReason
 	snapshot.BindingError = diag.BindingError
+	snapshot.BindingBulkAdaptiveSoftPayloadBytes = diag.BindingBulkAdaptiveSoftPayloadBytes
+	snapshot.BindingStreamAdaptiveSoftPayloadBytes = diag.BindingStreamAdaptiveSoftPayloadBytes
+	snapshot.BindingStreamAdaptiveWaitThresholdBytes = diag.BindingStreamAdaptiveWaitThresholdBytes
+	snapshot.BindingStreamAdaptiveFlushDelay = diag.BindingStreamAdaptiveFlushDelay
 	snapshot.TransportAttached = diag.TransportAttached
 	snapshot.TransportHasRuntimeConn = diag.TransportHasRuntimeConn
 	snapshot.TransportCurrent = diag.TransportCurrent
@@ -1057,8 +1106,23 @@ func acquireStreamOutboundBudget(runtime *streamRuntime, ctx context.Context, si
 	return runtime.acquireOutbound(ctx, size)
 }
 
+func tryAcquireStreamOutboundBudget(runtime *streamRuntime, size int) bool {
+	if runtime == nil {
+		return true
+	}
+	return runtime.tryAcquireOutbound(size)
+}
+
+func releaseStreamOutboundBudget(runtime *streamRuntime, size int) {
+	if runtime == nil {
+		return
+	}
+	runtime.releaseOutbound(size)
+}
+
 func normalizeStreamOpenRequest(req StreamOpenRequest) StreamOpenRequest {
 	req.Channel = normalizeStreamChannel(req.Channel)
+	req.FastPathVersion = normalizeStreamFastPathVersion(req.FastPathVersion)
 	req.Metadata = cloneStreamMetadata(req.Metadata)
 	return req
 }

stream_batch_codec.go

@@ -0,0 +1,6 @@
+package notify
+
+type streamBatchCodec struct {
+	encodeSingle func(streamFastDataFrame) ([]byte, error)
+	encodeBatch  func([]streamFastDataFrame) ([]byte, error)
+}

stream_batch_sender.go

@@ -0,0 +1,582 @@
+package notify
+
+import (
+	"context"
+	"net"
+	"sync"
+	"sync/atomic"
+	"time"
+)
+
+const (
+	streamBatchMaxPayloads     = 64
+	streamBatchMaxPayloadBytes = 2 * 1024 * 1024
+	streamBatchMaxFlushDelay   = 50 * time.Microsecond
+	streamBatchWaitThreshold   = 128 * 1024
+)
+
+const (
+	streamBatchRequestQueued int32 = iota
+	streamBatchRequestStarted
+	streamBatchRequestCanceled
+)
+
+type streamBatchRequestState struct {
+	value atomic.Int32
+}
+
+type streamBatchRequest struct {
+	ctx             context.Context
+	frame           streamFastDataFrame
+	hasFrame        bool
+	encodedPayload  []byte
+	hasEncoded      bool
+	frames          []streamFastDataFrame
+	fastPathVersion uint8
+	deadline        time.Time
+	done            chan error
+	state           *streamBatchRequestState
+}
+
+type streamBatchSender struct {
+	binding              *transportBinding
+	codec                streamBatchCodec
+	writeTimeoutProvider func() time.Duration
+	reqCh                chan streamBatchRequest
+	stopCh               chan struct{}
+	doneCh               chan struct{}
+
+	stopOnce sync.Once
+	flushMu  sync.Mutex
+	queued   atomic.Int64
+	errMu    sync.Mutex
+	err      error
+}
+
+func newStreamBatchSender(binding *transportBinding, codec streamBatchCodec, writeTimeoutProvider func() time.Duration) *streamBatchSender {
+	sender := &streamBatchSender{
+		binding:              binding,
+		codec:                codec,
+		writeTimeoutProvider: writeTimeoutProvider,
+		reqCh:                make(chan streamBatchRequest, streamBatchMaxPayloads*4),
+		stopCh:               make(chan struct{}),
+		doneCh:               make(chan struct{}),
+	}
+	go sender.run()
+	return sender
+}
+
+func (s *streamBatchSender) submitData(ctx context.Context, dataID uint64, seq uint64, fastPathVersion uint8, payload []byte) error {
+	if s == nil {
+		return errTransportDetached
+	}
+	if len(payload) == 0 {
+		return nil
+	}
+	return s.submitRequest(streamBatchRequest{
+		ctx: ctx,
+		frame: streamFastDataFrame{
+			DataID:  dataID,
+			Seq:     seq,
+			Payload: payload,
+		},
+		hasFrame:        true,
+		fastPathVersion: normalizeStreamFastPathVersion(fastPathVersion),
+	})
+}
+
+func (s *streamBatchSender) submitEncoded(ctx context.Context, fastPathVersion uint8, payload []byte) error {
+	if s == nil {
+		return errTransportDetached
+	}
+	if len(payload) == 0 {
+		return nil
+	}
+	return s.submitRequest(streamBatchRequest{
+		ctx:             ctx,
+		encodedPayload:  payload,
+		hasEncoded:      true,
+		fastPathVersion: normalizeStreamFastPathVersion(fastPathVersion),
+	})
+}
+
+func (s *streamBatchSender) submitFrames(ctx context.Context, fastPathVersion uint8, frames []streamFastDataFrame) error {
+	if s == nil {
+		return errTransportDetached
+	}
+	if len(frames) == 0 {
+		return nil
+	}
+	queuedFrames := append([]streamFastDataFrame(nil), frames...)
+	req := streamBatchRequest{
+		ctx:             ctx,
+		frames:          queuedFrames,
+		fastPathVersion: normalizeStreamFastPathVersion(fastPathVersion),
+	}
+	if len(queuedFrames) == 1 {
+		req.frame = queuedFrames[0]
+		req.frames = nil
+		req.hasFrame = true
+	}
+	return s.submitRequest(req)
+}
+
+func (s *streamBatchSender) submitRequest(req streamBatchRequest) error {
+	if s == nil {
+		return errTransportDetached
+	}
+	if req.ctx == nil {
+		req.ctx = context.Background()
+	}
+	if !req.hasFrame && !req.hasEncoded && len(req.frames) == 0 {
+		return nil
+	}
+	req.fastPathVersion = normalizeStreamFastPathVersion(req.fastPathVersion)
+	req.done = make(chan error, 1)
+	req.state = &streamBatchRequestState{}
+	if deadline, ok := req.ctx.Deadline(); ok {
+		req.deadline = deadline
+	}
+	if err := s.errSnapshot(); err != nil {
+		return err
+	}
+	if s.shouldDirectSubmit(req) {
+		if submitted, err := s.tryDirectSubmit(req); submitted {
+			return err
+		}
+	}
+	s.queued.Add(1)
+	select {
+	case <-req.ctx.Done():
+		s.queued.Add(-1)
+		return normalizeStreamDeadlineError(req.ctx.Err())
+	case <-s.stopCh:
+		s.queued.Add(-1)
+		return s.stoppedErr()
+	case s.reqCh <- req:
+	}
+	select {
+	case err := <-req.done:
+		return err
+	case <-req.ctx.Done():
+		if req.tryCancel() {
+			return normalizeStreamDeadlineError(req.ctx.Err())
+		}
+		return <-req.done
+	}
+}
+
+func (s *streamBatchSender) shouldDirectSubmit(req streamBatchRequest) bool {
+	if req.hasEncoded {
+		return false
+	}
+	if !req.hasFrame && len(req.frames) == 0 {
+		return false
+	}
+	return !streamFastPathSupportsBatch(req.fastPathVersion)
+}
+
+func (s *streamBatchSender) tryDirectSubmit(req streamBatchRequest) (bool, error) {
+	if s == nil {
+		return true, errTransportDetached
+	}
+	if err := s.errSnapshot(); err != nil {
+		return true, err
+	}
+	select {
+	case <-req.ctx.Done():
+		return true, normalizeStreamDeadlineError(req.ctx.Err())
+	case <-s.stopCh:
+		return true, s.stoppedErr()
+	default:
+	}
+	if s.queued.Load() != 0 {
+		return false, nil
+	}
+	if !s.flushMu.TryLock() {
+		return false, nil
+	}
+	defer s.flushMu.Unlock()
+	if s.queued.Load() != 0 {
+		return false, nil
+	}
+	if err := s.errSnapshot(); err != nil {
+		return true, err
+	}
+	if !req.tryStart() {
+		return true, req.canceledErr()
+	}
+	if err := req.contextErr(); err != nil {
+		return true, err
+	}
+	if err := s.flush([]streamBatchRequest{req}); err != nil {
+		s.setErr(err)
+		s.failPending(err)
+		return true, err
+	}
+	return true, nil
+}
+
+func (s *streamBatchSender) run() {
+	defer close(s.doneCh)
+	for {
+		req, ok := s.nextRequest()
+		if !ok {
+			return
+		}
+		batch := []streamBatchRequest{req}
+		batchBytes := streamBatchRequestApproxBytes(req)
+		softPayloadLimit := s.batchSoftPayloadLimit()
+		waitThreshold := s.batchWaitThreshold()
+		flushDelay := s.batchFlushDelay()
+		timer := (*time.Timer)(nil)
+		timerCh := (<-chan time.Time)(nil)
+		if flushDelay > 0 && batchBytes < waitThreshold && batchBytes < softPayloadLimit && len(batch) < streamBatchMaxPayloads {
+			timer = time.NewTimer(flushDelay)
+			timerCh = timer.C
+		}
+	drain:
+		for len(batch) < streamBatchMaxPayloads && batchBytes < softPayloadLimit {
+			if timerCh == nil {
+				select {
+				case <-s.stopCh:
+					s.failPending(s.stoppedErr())
+					return
+				case next := <-s.reqCh:
+					batch = append(batch, next)
+					batchBytes += streamBatchRequestApproxBytes(next)
+				default:
+					break drain
+				}
+				continue
+			}
+			select {
+			case <-s.stopCh:
+				if timer != nil {
+					timer.Stop()
+				}
+				s.failPending(s.stoppedErr())
+				return
+			case next := <-s.reqCh:
+				batch = append(batch, next)
+				batchBytes += streamBatchRequestApproxBytes(next)
+			case <-timerCh:
+				timerCh = nil
+				break drain
+			}
+		}
+		if timer != nil {
+			if !timer.Stop() && timerCh != nil {
+				select {
+				case <-timer.C:
+				default:
+				}
+			}
+		}
+		s.flushMu.Lock()
+		err := s.errSnapshot()
+		active := make([]streamBatchRequest, 0, len(batch))
+		for _, item := range batch {
+			if !item.tryStart() {
+				s.finishRequest(item, item.canceledErr())
+				continue
+			}
+			if itemErr := item.contextErr(); itemErr != nil {
+				s.finishRequest(item, itemErr)
+				continue
+			}
+			active = append(active, item)
+		}
+		if len(active) == 0 {
+			s.flushMu.Unlock()
+			continue
+		}
+		if err == nil {
+			err = s.flush(active)
+		}
+		s.flushMu.Unlock()
+		if err != nil {
+			s.setErr(err)
+			for _, item := range active {
+				s.finishRequest(item, err)
+			}
+			s.failPending(err)
+			return
+		}
+		for _, item := range active {
+			s.finishRequest(item, nil)
+		}
+	}
+}
+
+func (s *streamBatchSender) nextRequest() (streamBatchRequest, bool) {
+	select {
+	case <-s.stopCh:
+		s.failPending(s.stoppedErr())
+		return streamBatchRequest{}, false
+	case req := <-s.reqCh:
+		return req, true
+	}
+}
+
+func (s *streamBatchSender) flush(requests []streamBatchRequest) error {
+	if s == nil || s.binding == nil {
+		return errTransportDetached
+	}
+	queue := s.binding.queueSnapshot()
+	if queue == nil {
+		return errTransportFrameQueueUnavailable
+	}
+	payloads, err := s.encodeRequests(requests)
+	if err != nil {
+		return err
+	}
+	writeTimeout := s.transportWriteTimeout()
+	payloadBytes := 0
+	for _, payload := range payloads {
+		payloadBytes += len(payload)
+	}
+	started := time.Now()
+	err = s.binding.withConnWriteLockDeadline(writeDeadlineFromTimeout(writeTimeout), func(conn net.Conn) error {
+		return writeFramedPayloadBatchUnlocked(conn, queue, payloads)
+	})
+	s.binding.observeStreamAdaptivePayloadWrite(payloadBytes, time.Since(started), writeTimeout, err)
+	return err
+}
+
+func (s *streamBatchSender) transportWriteTimeout() time.Duration {
+	if s == nil || s.writeTimeoutProvider == nil {
+		return 0
+	}
+	return s.writeTimeoutProvider()
+}
+
+func (s *streamBatchSender) batchSoftPayloadLimit() int {
+	if s == nil || s.binding == nil {
+		return streamAdaptiveSoftPayloadFallbackBytes
+	}
+	return s.binding.streamAdaptiveSoftPayloadBytesSnapshot()
+}
+
+func (s *streamBatchSender) batchWaitThreshold() int {
+	if s == nil || s.binding == nil {
+		return streamBatchWaitThreshold
+	}
+	return s.binding.streamAdaptiveWaitThresholdBytesSnapshot()
+}
+
+func (s *streamBatchSender) batchFlushDelay() time.Duration {
+	if s == nil || s.binding == nil {
+		return streamBatchMaxFlushDelay
+	}
+	return s.binding.streamAdaptiveFlushDelaySnapshot()
+}
+
+func (s *streamBatchSender) batchPlainPayloadLimit() int {
+	limit := s.batchSoftPayloadLimit()
+	if limit <= streamFastBatchHeaderLen {
+		return streamFastBatchHeaderLen + 1
+	}
+	return minInt(limit, streamFastBatchMaxPlainBytes)
+}
+
+func (s *streamBatchSender) encodeRequests(requests []streamBatchRequest) ([][]byte, error) {
+	if len(requests) == 0 {
+		return nil, nil
+	}
+	payloads := make([][]byte, 0, len(requests))
+	batchPlainLimit := s.batchPlainPayloadLimit()
+	var batch []streamFastDataFrame
+	flushBatch := func() error {
+		if len(batch) == 0 {
+			return nil
+		}
+		payload, err := s.encodeBatch(batch)
+		if err != nil {
+			return err
+		}
+		payloads = append(payloads, payload)
+		batch = batch[:0]
+		return nil
+	}
+	batchBytes := streamFastBatchHeaderLen
+	appendFrame := func(frame streamFastDataFrame, fastPathVersion uint8) error {
+		if !streamFastPathSupportsBatch(fastPathVersion) {
+			if err := flushBatch(); err != nil {
+				return err
+			}
+			batchBytes = streamFastBatchHeaderLen
+			payload, err := s.encodeSingle(frame)
+			if err != nil {
+				return err
+			}
+			payloads = append(payloads, payload)
+			return nil
+		}
+		frameLen := streamFastBatchFrameLen(frame)
+		if frameLen+streamFastBatchHeaderLen > batchPlainLimit {
+			if err := flushBatch(); err != nil {
+				return err
+			}
+			batchBytes = streamFastBatchHeaderLen
+			payload, err := s.encodeSingle(frame)
+			if err != nil {
+				return err
+			}
+			payloads = append(payloads, payload)
+			return nil
+		}
+		if len(batch) > 0 && (len(batch) >= streamFastBatchMaxItems || batchBytes+frameLen > batchPlainLimit) {
+			if err := flushBatch(); err != nil {
+				return err
+			}
+			batchBytes = streamFastBatchHeaderLen
+		}
+		if batch == nil {
+			batch = make([]streamFastDataFrame, 0, minInt(len(requests), streamFastBatchMaxItems))
+		}
+		batch = append(batch, frame)
+		batchBytes += frameLen
+		return nil
+	}
+	for _, req := range requests {
+		if req.hasFrame {
+			if err := appendFrame(req.frame, req.fastPathVersion); err != nil {
+				return nil, err
+			}
+		}
+		for _, frame := range req.frames {
+			if err := appendFrame(frame, req.fastPathVersion); err != nil {
+				return nil, err
+			}
+		}
+		if req.hasEncoded {
+			if err := flushBatch(); err != nil {
+				return nil, err
+			}
+			batchBytes = streamFastBatchHeaderLen
+			payloads = append(payloads, req.encodedPayload)
+		}
+	}
+	if err := flushBatch(); err != nil {
+		return nil, err
+	}
+	return payloads, nil
+}
+
+func streamBatchRequestApproxBytes(req streamBatchRequest) int {
+	total := 0
+	if req.hasFrame {
+		total += streamFastBatchFrameLen(req.frame)
+	}
+	for _, frame := range req.frames {
+		total += streamFastBatchFrameLen(frame)
+	}
+	if req.hasEncoded {
+		total += len(req.encodedPayload)
+	}
+	return total
+}
+
+func (s *streamBatchSender) encodeSingle(frame streamFastDataFrame) ([]byte, error) {
+	if s == nil || s.codec.encodeSingle == nil {
+		return nil, errTransportDetached
+	}
+	return s.codec.encodeSingle(frame)
+}
+
+func (s *streamBatchSender) encodeBatch(frames []streamFastDataFrame) ([]byte, error) {
+	if len(frames) == 1 || s.codec.encodeBatch == nil {
+		return s.encodeSingle(frames[0])
+	}
+	return s.codec.encodeBatch(frames)
+}
+
+func (s *streamBatchSender) finishRequest(req streamBatchRequest, err error) {
+	if s != nil {
+		s.queued.Add(-1)
+	}
+	req.done <- err
+}
+
+func (s *streamBatchSender) stop() {
+	if s == nil {
+		return
+	}
+	s.stopOnce.Do(func() {
+		s.setErr(errTransportDetached)
+		close(s.stopCh)
+	})
+	<-s.doneCh
+}
+
+func (s *streamBatchSender) failPending(err error) {
+	for {
+		select {
+		case item := <-s.reqCh:
+			s.finishRequest(item, err)
+		default:
+			return
+		}
+	}
+}
+
+func (s *streamBatchSender) setErr(err error) {
+	if s == nil || err == nil {
+		return
+	}
+	s.errMu.Lock()
+	if s.err == nil {
+		s.err = err
+	}
+	s.errMu.Unlock()
+}
+
+func (s *streamBatchSender) errSnapshot() error {
+	if s == nil {
+		return errTransportDetached
+	}
+	s.errMu.Lock()
+	defer s.errMu.Unlock()
+	return s.err
+}
+
+func (s *streamBatchSender) stoppedErr() error {
+	if err := s.errSnapshot(); err != nil {
+		return err
+	}
+	return errTransportDetached
+}
+
+func (r streamBatchRequest) contextErr() error {
+	if r.ctx == nil {
+		return nil
+	}
+	select {
+	case <-r.ctx.Done():
+		return normalizeStreamDeadlineError(r.ctx.Err())
+	default:
+		return nil
+	}
+}
+
+func (r streamBatchRequest) tryStart() bool {
+	if r.state == nil {
+		return true
+	}
+	return r.state.value.CompareAndSwap(streamBatchRequestQueued, streamBatchRequestStarted)
+}
+
+func (r streamBatchRequest) tryCancel() bool {
+	if r.state == nil {
+		return false
+	}
+	return r.state.value.CompareAndSwap(streamBatchRequestQueued, streamBatchRequestCanceled)
+}
+
+func (r streamBatchRequest) canceledErr() error {
+	if err := r.contextErr(); err != nil {
+		return err
+	}
+	return context.Canceled
+}

stream_benchmark_test.go

@@ -128,7 +128,7 @@ func benchmarkStreamTCPThroughput(b *testing.B, payloadSize int, cfg StreamConfi
 		return nil
 	})
 
-	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+	if err := server.Listen("tcp", benchmarkTCPListenAddr(b)); err != nil {
 		b.Fatalf("server Listen failed: %v", err)
 	}
 	b.Cleanup(func() {
@@ -140,7 +140,7 @@ func benchmarkStreamTCPThroughput(b *testing.B, payloadSize int, cfg StreamConfi
 	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
 		b.Fatalf("UseModernPSKClient failed: %v", err)
 	}
-	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+	if err := client.Connect("tcp", benchmarkTCPDialAddr(b, server.listener.Addr().String())); err != nil {
 		b.Fatalf("client Connect failed: %v", err)
 	}
 	b.Cleanup(func() {
@@ -216,7 +216,7 @@ func benchmarkStreamTCPThroughputConcurrent(b *testing.B, payloadSize int, concu
 		return nil
 	})
 
-	if err := server.Listen("tcp", "127.0.0.1:0"); err != nil {
+	if err := server.Listen("tcp", benchmarkTCPListenAddr(b)); err != nil {
 		b.Fatalf("server Listen failed: %v", err)
 	}
 	b.Cleanup(func() {
@@ -228,7 +228,7 @@ func benchmarkStreamTCPThroughputConcurrent(b *testing.B, payloadSize int, concu
 	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
 		b.Fatalf("UseModernPSKClient failed: %v", err)
 	}
-	if err := client.Connect("tcp", server.listener.Addr().String()); err != nil {
+	if err := client.Connect("tcp", benchmarkTCPDialAddr(b, server.listener.Addr().String())); err != nil {
 		b.Fatalf("client Connect failed: %v", err)
 	}
 	b.Cleanup(func() {

stream_control.go

@@ -7,17 +7,19 @@ import (
 )
 
 type StreamOpenRequest struct {
-	StreamID     string
-	DataID       uint64
-	Channel      StreamChannel
-	Metadata     StreamMetadata
-	ReadTimeout  time.Duration
-	WriteTimeout time.Duration
+	StreamID        string
+	DataID          uint64
+	FastPathVersion uint8
+	Channel         StreamChannel
+	Metadata        StreamMetadata
+	ReadTimeout     time.Duration
+	WriteTimeout    time.Duration
 }
 
 type StreamOpenResponse struct {
 	StreamID            string
 	DataID              uint64
+	FastPathVersion     uint8
 	Accepted            bool
 	TransportGeneration uint64
 	Metadata            StreamMetadata
@@ -92,6 +94,8 @@ func (c *ClientCommon) handleInboundStreamOpen(msg *Message) {
 		return
 	}
 	scope := clientFileScope()
+	req.FastPathVersion = negotiateStreamFastPathVersion(req.FastPathVersion)
+	resp.FastPathVersion = req.FastPathVersion
 	if req.DataID == 0 {
 		req.DataID = runtime.nextDataID()
 		resp.DataID = req.DataID
@@ -178,6 +182,8 @@ func (s *ServerCommon) handleInboundStreamOpen(msg *Message) {
 	}
 	transport := messageTransportConnSnapshot(msg)
 	scope := serverFileScope(logical)
+	req.FastPathVersion = negotiateStreamFastPathVersion(req.FastPathVersion)
+	resp.FastPathVersion = req.FastPathVersion
 	if req.DataID == 0 {
 		req.DataID = runtime.nextDataID()
 		resp.DataID = req.DataID

stream_fastpath.go

@@ -1,8 +1,10 @@
 package notify
 
 import (
+	"context"
 	"encoding/binary"
 	"errors"
+	"io"
 )
 
 var (
@@ -15,6 +17,7 @@ const (
 	streamFastPayloadVersion   = 1
 	streamFastPayloadTypeData  = 1
 	streamFastPayloadHeaderLen = 28
+	streamFastBatchDirectLimit = 512 * 1024
 )
 
 type streamFastDataFrame struct {
@@ -24,6 +27,56 @@ type streamFastDataFrame struct {
 	Payload []byte
 }
 
+func streamAdaptiveFramePayloadLimit(binding *transportBinding) int {
+	if binding == nil {
+		return 0
+	}
+	limit := binding.streamAdaptiveSoftPayloadBytesSnapshot() - streamFastPayloadHeaderLen
+	if limit <= 0 {
+		return 1
+	}
+	maxPayload := streamFastBatchMaxPlainBytes - streamFastPayloadHeaderLen
+	if limit > maxPayload {
+		return maxPayload
+	}
+	return limit
+}
+
+func streamFastSplitFrameCount(size int, maxPayload int) int {
+	if size <= 0 || maxPayload <= 0 {
+		return 1
+	}
+	return (size + maxPayload - 1) / maxPayload
+}
+
+func buildStreamFastSplitFrames(dataID uint64, startSeq uint64, chunk []byte, maxPayload int) []streamFastDataFrame {
+	if len(chunk) == 0 {
+		return nil
+	}
+	if maxPayload <= 0 || len(chunk) <= maxPayload {
+		return []streamFastDataFrame{{
+			DataID:  dataID,
+			Seq:     startSeq,
+			Payload: chunk,
+		}}
+	}
+	frames := make([]streamFastDataFrame, 0, streamFastSplitFrameCount(len(chunk), maxPayload))
+	seq := startSeq
+	for offset := 0; offset < len(chunk); offset += maxPayload {
+		end := offset + maxPayload
+		if end > len(chunk) {
+			end = len(chunk)
+		}
+		frames = append(frames, streamFastDataFrame{
+			DataID:  dataID,
+			Seq:     seq,
+			Payload: chunk[offset:end],
+		})
+		seq++
+	}
+	return frames
+}
+
 func encodeStreamFastDataFrameHeader(dst []byte, dataID uint64, seq uint64, payloadLen int) error {
 	if dataID == 0 {
 		return errStreamFastDataIDEmpty
@@ -51,6 +104,31 @@ func encodeStreamFastDataFrame(dataID uint64, seq uint64, payload []byte) ([]byt
 	return frame, nil
 }
 
+func encodeStreamFastFramePayload(frame streamFastDataFrame) ([]byte, error) {
+	framePayload := make([]byte, streamFastPayloadHeaderLen+len(frame.Payload))
+	if err := encodeStreamFastDataFrameHeader(framePayload, frame.DataID, frame.Seq, len(frame.Payload)); err != nil {
+		return nil, err
+	}
+	framePayload[6] = frame.Flags
+	copy(framePayload[streamFastPayloadHeaderLen:], frame.Payload)
+	return framePayload, nil
+}
+
+func encodeStreamFastFramePayloadFast(encode transportFastPlainEncoder, secretKey []byte, frame streamFastDataFrame) ([]byte, error) {
+	if encode == nil {
+		return nil, errTransportPayloadEncryptFailed
+	}
+	plainLen := streamFastPayloadHeaderLen + len(frame.Payload)
+	return encode(secretKey, plainLen, func(dst []byte) error {
+		if err := encodeStreamFastDataFrameHeader(dst, frame.DataID, frame.Seq, len(frame.Payload)); err != nil {
+			return err
+		}
+		dst[6] = frame.Flags
+		copy(dst[streamFastPayloadHeaderLen:], frame.Payload)
+		return nil
+	})
+}
+
 func decodeStreamFastDataFrame(payload []byte) (streamFastDataFrame, bool, error) {
 	if len(payload) < 4 || string(payload[:4]) != streamFastPayloadMagic {
 		return streamFastDataFrame{}, false, nil
@@ -77,18 +155,66 @@ func decodeStreamFastDataFrame(payload []byte) (streamFastDataFrame, bool, error
 	}, true, nil
 }
 
-func (c *ClientCommon) encodeFastStreamDataPayload(dataID uint64, seq uint64, chunk []byte) ([]byte, error) {
-	if c != nil && c.fastStreamEncode != nil {
-		return c.fastStreamEncode(c.SecretKey, dataID, seq, chunk)
+func (c *ClientCommon) encodeFastStreamPayload(frame streamFastDataFrame) ([]byte, error) {
+	if c != nil && c.fastStreamEncode != nil && frame.Flags == 0 {
+		return c.fastStreamEncode(c.SecretKey, frame.DataID, frame.Seq, frame.Payload)
 	}
-	plain, err := encodeStreamFastDataFrame(dataID, seq, chunk)
+	if c != nil && c.fastPlainEncode != nil {
+		return encodeStreamFastFramePayloadFast(c.fastPlainEncode, c.SecretKey, frame)
+	}
+	plain, err := encodeStreamFastFramePayload(frame)
 	if err != nil {
 		return nil, err
 	}
 	return c.encryptTransportPayload(plain)
 }
 
-func (c *ClientCommon) sendFastStreamData(dataID uint64, seq uint64, chunk []byte) error {
+func (c *ClientCommon) encodeFastStreamDataPayload(dataID uint64, seq uint64, chunk []byte) ([]byte, error) {
+	return c.encodeFastStreamPayload(streamFastDataFrame{
+		DataID:  dataID,
+		Seq:     seq,
+		Payload: chunk,
+	})
+}
+
+func (c *ClientCommon) encodeFastStreamBatchPayload(frames []streamFastDataFrame) ([]byte, error) {
+	if c == nil {
+		return nil, errStreamClientNil
+	}
+	if c.fastPlainEncode != nil {
+		return encodeStreamFastBatchPayloadFast(c.fastPlainEncode, c.SecretKey, frames)
+	}
+	plain, err := encodeStreamFastBatchPlain(frames)
+	if err != nil {
+		return nil, err
+	}
+	return c.encryptTransportPayload(plain)
+}
+
+func (c *ClientCommon) sendFastStreamData(ctx context.Context, stream *streamHandle, chunk []byte) error {
+	if stream == nil {
+		return io.ErrClosedPipe
+	}
+	dataID := stream.dataIDSnapshot()
+	fastPathVersion := stream.fastPathVersionSnapshot()
+	if binding := c.clientTransportBindingSnapshot(); binding != nil && streamFastPathSupportsBatch(fastPathVersion) {
+		if sender := binding.clientStreamBatchSenderSnapshot(c); sender != nil {
+			if maxPayload := streamAdaptiveFramePayloadLimit(binding); maxPayload > 0 && len(chunk) > maxPayload {
+				startSeq := stream.reserveOutboundDataSeqs(streamFastSplitFrameCount(len(chunk), maxPayload))
+				return sender.submitFrames(ctx, fastPathVersion, buildStreamFastSplitFrames(dataID, startSeq, chunk, maxPayload))
+			}
+			seq := stream.reserveOutboundDataSeqs(1)
+			if len(chunk) < streamFastBatchDirectLimit {
+				return sender.submitData(ctx, dataID, seq, fastPathVersion, chunk)
+			}
+			payload, err := c.encodeFastStreamDataPayload(dataID, seq, chunk)
+			if err != nil {
+				return err
+			}
+			return sender.submitEncoded(ctx, fastPathVersion, payload)
+		}
+	}
+	seq := stream.reserveOutboundDataSeqs(1)
 	payload, err := c.encodeFastStreamDataPayload(dataID, seq, chunk)
 	if err != nil {
 		return err
@@ -96,29 +222,78 @@ func (c *ClientCommon) sendFastStreamData(dataID uint64, seq uint64, chunk []byt
 	return c.writePayloadToTransport(payload)
 }
 
-func (s *ServerCommon) encodeFastStreamDataPayloadLogical(logical *LogicalConn, dataID uint64, seq uint64, chunk []byte) ([]byte, error) {
-	if logical != nil {
-		if fastStreamEncode := logical.fastStreamEncodeSnapshot(); fastStreamEncode != nil {
-			return fastStreamEncode(logical.secretKeySnapshot(), dataID, seq, chunk)
-		}
+func (s *ServerCommon) encodeFastStreamPayloadLogical(logical *LogicalConn, frame streamFastDataFrame) ([]byte, error) {
+	if logical == nil {
+		return nil, errTransportDetached
 	}
-	plain, err := encodeStreamFastDataFrame(dataID, seq, chunk)
+	if fastStreamEncode := logical.fastStreamEncodeSnapshot(); fastStreamEncode != nil && frame.Flags == 0 {
+		return fastStreamEncode(logical.secretKeySnapshot(), frame.DataID, frame.Seq, frame.Payload)
+	}
+	if fastPlainEncode := logical.fastPlainEncodeSnapshot(); fastPlainEncode != nil {
+		return encodeStreamFastFramePayloadFast(fastPlainEncode, logical.secretKeySnapshot(), frame)
+	}
+	plain, err := encodeStreamFastFramePayload(frame)
 	if err != nil {
 		return nil, err
 	}
 	return s.encryptTransportPayloadLogical(logical, plain)
 }
 
-func (s *ServerCommon) sendFastStreamDataTransport(logical *LogicalConn, transport *TransportConn, dataID uint64, seq uint64, chunk []byte) error {
+func (s *ServerCommon) encodeFastStreamDataPayloadLogical(logical *LogicalConn, dataID uint64, seq uint64, chunk []byte) ([]byte, error) {
+	return s.encodeFastStreamPayloadLogical(logical, streamFastDataFrame{
+		DataID:  dataID,
+		Seq:     seq,
+		Payload: chunk,
+	})
+}
+
+func (s *ServerCommon) encodeFastStreamBatchPayloadLogical(logical *LogicalConn, frames []streamFastDataFrame) ([]byte, error) {
+	if logical == nil {
+		return nil, errTransportDetached
+	}
+	if fastPlainEncode := logical.fastPlainEncodeSnapshot(); fastPlainEncode != nil {
+		return encodeStreamFastBatchPayloadFast(fastPlainEncode, logical.secretKeySnapshot(), frames)
+	}
+	plain, err := encodeStreamFastBatchPlain(frames)
+	if err != nil {
+		return nil, err
+	}
+	return s.encryptTransportPayloadLogical(logical, plain)
+}
+
+func (s *ServerCommon) sendFastStreamDataTransport(ctx context.Context, logical *LogicalConn, transport *TransportConn, stream *streamHandle, chunk []byte) error {
 	if err := s.ensureServerTransportSendReady(transport); err != nil {
 		return err
 	}
+	if stream == nil {
+		return io.ErrClosedPipe
+	}
 	if logical == nil && transport != nil {
 		logical = transport.logicalConnSnapshot()
 	}
 	if logical == nil {
 		return errTransportDetached
 	}
+	dataID := stream.dataIDSnapshot()
+	fastPathVersion := stream.fastPathVersionSnapshot()
+	if binding := logical.transportBindingSnapshot(); binding != nil && binding.queueSnapshot() != nil && streamFastPathSupportsBatch(fastPathVersion) {
+		if sender := binding.serverStreamBatchSenderSnapshot(logical); sender != nil {
+			if maxPayload := streamAdaptiveFramePayloadLimit(binding); maxPayload > 0 && len(chunk) > maxPayload {
+				startSeq := stream.reserveOutboundDataSeqs(streamFastSplitFrameCount(len(chunk), maxPayload))
+				return sender.submitFrames(ctx, fastPathVersion, buildStreamFastSplitFrames(dataID, startSeq, chunk, maxPayload))
+			}
+			seq := stream.reserveOutboundDataSeqs(1)
+			if len(chunk) < streamFastBatchDirectLimit {
+				return sender.submitData(ctx, dataID, seq, fastPathVersion, chunk)
+			}
+			payload, err := s.encodeFastStreamDataPayloadLogical(logical, dataID, seq, chunk)
+			if err != nil {
+				return err
+			}
+			return sender.submitEncoded(ctx, fastPathVersion, payload)
+		}
+	}
+	seq := stream.reserveOutboundDataSeqs(1)
 	payload, err := s.encodeFastStreamDataPayloadLogical(logical, dataID, seq, chunk)
 	if err != nil {
 		return err

stream_fastpath_test.go

@@ -4,6 +4,8 @@ import (
 	"b612.me/stario"
 	"context"
 	"math"
+	"sync"
+	"sync/atomic"
 	"testing"
 	"time"
 )
@@ -31,6 +33,46 @@ func TestStreamFastDataFrameRoundTrip(t *testing.T) {
 	}
 }
 
+func TestStreamFastBatchPlainRoundTrip(t *testing.T) {
+	frames := []streamFastDataFrame{
+		{
+			DataID:  11,
+			Seq:     7,
+			Payload: []byte("alpha"),
+		},
+		{
+			DataID:  12,
+			Seq:     8,
+			Payload: []byte("beta"),
+		},
+	}
+	wire, err := encodeStreamFastBatchPlain(frames)
+	if err != nil {
+		t.Fatalf("encodeStreamFastBatchPlain failed: %v", err)
+	}
+	decoded, matched, err := decodeStreamFastBatchPlain(wire)
+	if err != nil {
+		t.Fatalf("decodeStreamFastBatchPlain failed: %v", err)
+	}
+	if !matched {
+		t.Fatal("decodeStreamFastBatchPlain should match encoded batch")
+	}
+	if got, want := len(decoded), len(frames); got != want {
+		t.Fatalf("decoded frame count = %d, want %d", got, want)
+	}
+	for index := range frames {
+		if got, want := decoded[index].DataID, frames[index].DataID; got != want {
+			t.Fatalf("frame %d data id = %d, want %d", index, got, want)
+		}
+		if got, want := decoded[index].Seq, frames[index].Seq; got != want {
+			t.Fatalf("frame %d seq = %d, want %d", index, got, want)
+		}
+		if got, want := string(decoded[index].Payload), string(frames[index].Payload); got != want {
+			t.Fatalf("frame %d payload = %q, want %q", index, got, want)
+		}
+	}
+}
+
 func TestClientDispatchInboundTransportPayloadFastStream(t *testing.T) {
 	client := NewClient().(*ClientCommon)
 	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
@@ -60,6 +102,165 @@ func TestClientDispatchInboundTransportPayloadFastStream(t *testing.T) {
 	readStreamExactly(t, stream, "fast-payload", 2*time.Second)
 }
 
+func TestClientDispatchInboundTransportPayloadFastStreamBatch(t *testing.T) {
+	client := NewClient().(*ClientCommon)
+	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
+		t.Fatalf("UseModernPSKClient failed: %v", err)
+	}
+	runtime := client.getStreamRuntime()
+	if runtime == nil {
+		t.Fatal("client stream runtime should not be nil")
+	}
+	streamA := newStreamHandle(context.Background(), runtime, clientFileScope(), StreamOpenRequest{
+		StreamID:        "fast-client-a",
+		DataID:          23,
+		FastPathVersion: streamFastPathVersionCurrent,
+		Channel:         StreamDataChannel,
+	}, 0, nil, nil, 0, nil, nil, nil, runtime.configSnapshot())
+	streamB := newStreamHandle(context.Background(), runtime, clientFileScope(), StreamOpenRequest{
+		StreamID:        "fast-client-b",
+		DataID:          24,
+		FastPathVersion: streamFastPathVersionCurrent,
+		Channel:         StreamDataChannel,
+	}, 0, nil, nil, 0, nil, nil, nil, runtime.configSnapshot())
+	if err := runtime.register(clientFileScope(), streamA); err != nil {
+		t.Fatalf("register streamA failed: %v", err)
+	}
+	if err := runtime.register(clientFileScope(), streamB); err != nil {
+		t.Fatalf("register streamB failed: %v", err)
+	}
+
+	payload, err := client.encodeFastStreamBatchPayload([]streamFastDataFrame{
+		{DataID: 23, Seq: 1, Payload: []byte("fast-a")},
+		{DataID: 24, Seq: 2, Payload: []byte("fast-b")},
+	})
+	if err != nil {
+		t.Fatalf("encodeFastStreamBatchPayload failed: %v", err)
+	}
+	if err := client.dispatchInboundTransportPayload(payload, time.Now()); err != nil {
+		t.Fatalf("dispatchInboundTransportPayload failed: %v", err)
+	}
+
+	readStreamExactly(t, streamA, "fast-a", 2*time.Second)
+	readStreamExactly(t, streamB, "fast-b", 2*time.Second)
+}
+
+func TestStreamBatchSenderEncodeRequestsCoalescesFastV2Frames(t *testing.T) {
+	var (
+		singleCalls int
+		batchCalls  [][]streamFastDataFrame
+	)
+	sender := &streamBatchSender{
+		codec: streamBatchCodec{
+			encodeSingle: func(frame streamFastDataFrame) ([]byte, error) {
+				singleCalls++
+				return []byte("single"), nil
+			},
+			encodeBatch: func(frames []streamFastDataFrame) ([]byte, error) {
+				cloned := make([]streamFastDataFrame, len(frames))
+				copy(cloned, frames)
+				batchCalls = append(batchCalls, cloned)
+				return []byte("batch"), nil
+			},
+		},
+	}
+	payloads, err := sender.encodeRequests([]streamBatchRequest{
+		{
+			frames: []streamFastDataFrame{{
+				DataID:  101,
+				Seq:     1,
+				Payload: []byte("a"),
+			}},
+			fastPathVersion: streamFastPathVersionV2,
+		},
+		{
+			frames: []streamFastDataFrame{{
+				DataID:  102,
+				Seq:     2,
+				Payload: []byte("b"),
+			}},
+			fastPathVersion: streamFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("encodeRequests failed: %v", err)
+	}
+	if got, want := len(payloads), 1; got != want {
+		t.Fatalf("payload count = %d, want %d", got, want)
+	}
+	if singleCalls != 0 {
+		t.Fatalf("single encode calls = %d, want 0", singleCalls)
+	}
+	if got, want := len(batchCalls), 1; got != want {
+		t.Fatalf("batch encode calls = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls[0]), 2; got != want {
+		t.Fatalf("batched frame count = %d, want %d", got, want)
+	}
+}
+
+func TestStreamBatchSenderEncodeRequestsFlushesBeforePreEncodedPayload(t *testing.T) {
+	var (
+		singleCalls int
+		batchCalls  [][]streamFastDataFrame
+	)
+	sender := &streamBatchSender{
+		codec: streamBatchCodec{
+			encodeSingle: func(frame streamFastDataFrame) ([]byte, error) {
+				singleCalls++
+				return append([]byte("single-"), frame.Payload...), nil
+			},
+			encodeBatch: func(frames []streamFastDataFrame) ([]byte, error) {
+				cloned := make([]streamFastDataFrame, len(frames))
+				copy(cloned, frames)
+				batchCalls = append(batchCalls, cloned)
+				return []byte("batch-2"), nil
+			},
+		},
+	}
+	payloads, err := sender.encodeRequests([]streamBatchRequest{
+		{
+			frames: []streamFastDataFrame{
+				{DataID: 101, Seq: 1, Payload: []byte("a")},
+				{DataID: 102, Seq: 2, Payload: []byte("b")},
+			},
+			fastPathVersion: streamFastPathVersionV2,
+		},
+		{
+			encodedPayload:  []byte("raw"),
+			hasEncoded:      true,
+			fastPathVersion: streamFastPathVersionV2,
+		},
+		{
+			frames: []streamFastDataFrame{
+				{DataID: 103, Seq: 3, Payload: []byte("c")},
+			},
+			fastPathVersion: streamFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("encodeRequests failed: %v", err)
+	}
+	if got, want := len(payloads), 3; got != want {
+		t.Fatalf("payload count = %d, want %d", got, want)
+	}
+	if got, want := string(payloads[0]), "batch-2"; got != want {
+		t.Fatalf("first payload = %q, want %q", got, want)
+	}
+	if got, want := string(payloads[1]), "raw"; got != want {
+		t.Fatalf("second payload = %q, want %q", got, want)
+	}
+	if got, want := string(payloads[2]), "single-c"; got != want {
+		t.Fatalf("third payload = %q, want %q", got, want)
+	}
+	if got, want := len(batchCalls), 1; got != want {
+		t.Fatalf("batch encode calls = %d, want %d", got, want)
+	}
+	if singleCalls != 1 {
+		t.Fatalf("single encode calls = %d, want %d", singleCalls, 1)
+	}
+}
+
 func TestClientPushMessageFastDispatchesDirectWithRuntimeDispatcher(t *testing.T) {
 	client := NewClient().(*ClientCommon)
 	if err := UseModernPSKClient(client, integrationSharedSecret, integrationModernPSKOptions()); err != nil {
@@ -110,3 +311,216 @@ func TestClientPushMessageFastDispatchesDirectWithRuntimeDispatcher(t *testing.T
 	default:
 	}
 }
+
+func TestStreamBatchSenderEncodeRequestsResetsBatchBytesAfterFlushBoundary(t *testing.T) {
+	var (
+		singleCalls int
+		batchCalls  [][]streamFastDataFrame
+	)
+	sender := &streamBatchSender{
+		codec: streamBatchCodec{
+			encodeSingle: func(frame streamFastDataFrame) ([]byte, error) {
+				singleCalls++
+				return []byte("single"), nil
+			},
+			encodeBatch: func(frames []streamFastDataFrame) ([]byte, error) {
+				cloned := make([]streamFastDataFrame, len(frames))
+				copy(cloned, frames)
+				batchCalls = append(batchCalls, cloned)
+				return []byte("batch"), nil
+			},
+		},
+	}
+	largePayload := make([]byte, streamFastBatchMaxPlainBytes-streamFastBatchHeaderLen-streamFastBatchItemHeaderLen-128)
+	payloads, err := sender.encodeRequests([]streamBatchRequest{
+		{
+			frames: []streamFastDataFrame{{
+				DataID:  101,
+				Seq:     1,
+				Payload: largePayload,
+			}},
+			fastPathVersion: streamFastPathVersionV2,
+		},
+		{
+			encodedPayload:  []byte("raw"),
+			hasEncoded:      true,
+			fastPathVersion: streamFastPathVersionV2,
+		},
+		{
+			frames: []streamFastDataFrame{{
+				DataID:  202,
+				Seq:     1,
+				Payload: []byte("a"),
+			}},
+			fastPathVersion: streamFastPathVersionV2,
+		},
+		{
+			frames: []streamFastDataFrame{{
+				DataID:  202,
+				Seq:     2,
+				Payload: []byte("b"),
+			}},
+			fastPathVersion: streamFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("encodeRequests failed: %v", err)
+	}
+	if got, want := len(payloads), 3; got != want {
+		t.Fatalf("payload count = %d, want %d", got, want)
+	}
+	if got, want := singleCalls, 1; got != want {
+		t.Fatalf("single encode calls = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls), 1; got != want {
+		t.Fatalf("batch encode calls = %d, want %d", got, want)
+	}
+	if got, want := len(batchCalls[0]), 2; got != want {
+		t.Fatalf("post-flush batched frame count = %d, want %d", got, want)
+	}
+}
+
+func TestStreamBatchSenderEncodeRequestsUsesAdaptiveSoftLimit(t *testing.T) {
+	binding := &transportBinding{}
+	binding.observeStreamAdaptivePayloadWrite(2*1024*1024, 640*time.Millisecond, 0, nil)
+
+	var (
+		singleCalls int
+		batchCalls  int
+	)
+	sender := &streamBatchSender{
+		binding: binding,
+		codec: streamBatchCodec{
+			encodeSingle: func(frame streamFastDataFrame) ([]byte, error) {
+				singleCalls++
+				return []byte("single"), nil
+			},
+			encodeBatch: func(frames []streamFastDataFrame) ([]byte, error) {
+				batchCalls++
+				return []byte("batch"), nil
+			},
+		},
+	}
+	payload := make([]byte, 160*1024)
+	payloads, err := sender.encodeRequests([]streamBatchRequest{
+		{
+			frames: []streamFastDataFrame{{
+				DataID:  101,
+				Seq:     1,
+				Payload: payload,
+			}},
+			fastPathVersion: streamFastPathVersionV2,
+		},
+		{
+			frames: []streamFastDataFrame{{
+				DataID:  102,
+				Seq:     2,
+				Payload: payload,
+			}},
+			fastPathVersion: streamFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("encodeRequests failed: %v", err)
+	}
+	if got, want := len(payloads), 2; got != want {
+		t.Fatalf("payload count = %d, want %d", got, want)
+	}
+	if got, want := singleCalls, 2; got != want {
+		t.Fatalf("single encode calls = %d, want %d", got, want)
+	}
+	if batchCalls != 0 {
+		t.Fatalf("batch encode calls = %d, want 0", batchCalls)
+	}
+}
+
+func TestClientSendFastStreamDataSplitsLargeChunkWhenAdaptiveSoftLimitShrinks(t *testing.T) {
+	binding := newTransportBinding(&delayedWriteConn{}, stario.NewQueue())
+	binding.observeStreamAdaptivePayloadWrite(2*1024*1024, 640*time.Millisecond, 0, nil)
+
+	var (
+		mu           sync.Mutex
+		singleFrames []streamFastDataFrame
+		batchFrames  [][]streamFastDataFrame
+	)
+	binding.streamSender = newStreamBatchSender(binding, streamBatchCodec{
+		encodeSingle: func(frame streamFastDataFrame) ([]byte, error) {
+			mu.Lock()
+			singleFrames = append(singleFrames, streamFastDataFrame{
+				DataID:  frame.DataID,
+				Seq:     frame.Seq,
+				Payload: append([]byte(nil), frame.Payload...),
+			})
+			mu.Unlock()
+			return []byte{1}, nil
+		},
+		encodeBatch: func(frames []streamFastDataFrame) ([]byte, error) {
+			cloned := make([]streamFastDataFrame, len(frames))
+			for index := range frames {
+				cloned[index] = streamFastDataFrame{
+					DataID:  frames[index].DataID,
+					Seq:     frames[index].Seq,
+					Payload: append([]byte(nil), frames[index].Payload...),
+				}
+			}
+			mu.Lock()
+			batchFrames = append(batchFrames, cloned)
+			mu.Unlock()
+			return []byte{2}, nil
+		},
+	}, nil)
+	defer binding.stopBackgroundWorkers()
+
+	client := NewClient().(*ClientCommon)
+	stopCtx, stopFn := context.WithCancel(context.Background())
+	defer stopFn()
+	client.setClientSessionRuntime(&clientSessionRuntime{
+		transport:             binding,
+		transportAttached:     true,
+		stopCtx:               stopCtx,
+		stopFn:                stopFn,
+		queue:                 binding.queueSnapshot(),
+		inboundDispatcher:     newInboundDispatcher(),
+		suppressGoodByeOnStop: &atomic.Bool{},
+	})
+
+	stream := &streamHandle{
+		dataID:          41,
+		fastPathVersion: streamFastPathVersionV2,
+	}
+	chunk := make([]byte, streamFastBatchDirectLimit+128*1024)
+	for index := range chunk {
+		chunk[index] = byte(index)
+	}
+
+	if err := client.sendFastStreamData(context.Background(), stream, chunk); err != nil {
+		t.Fatalf("sendFastStreamData failed: %v", err)
+	}
+
+	expectedFrames := streamFastSplitFrameCount(len(chunk), streamAdaptiveFramePayloadLimit(binding))
+	if got, want := int(stream.outboundSeq.Load()), expectedFrames; got != want {
+		t.Fatalf("reserved seq count = %d, want %d", got, want)
+	}
+
+	mu.Lock()
+	defer mu.Unlock()
+	if len(batchFrames) != 0 {
+		t.Fatalf("batch encode calls = %d, want 0", len(batchFrames))
+	}
+	if got, want := len(singleFrames), expectedFrames; got != want {
+		t.Fatalf("single frame count = %d, want %d", got, want)
+	}
+	rebuilt := make([]byte, 0, len(chunk))
+	for index, frame := range singleFrames {
+		if got, want := frame.DataID, uint64(41); got != want {
+			t.Fatalf("frame %d data id = %d, want %d", index, got, want)
+		}
+		if got, want := frame.Seq, uint64(index+1); got != want {
+			t.Fatalf("frame %d seq = %d, want %d", index, got, want)
+		}
+		rebuilt = append(rebuilt, frame.Payload...)
+	}
+	if string(rebuilt) != string(chunk) {
+		t.Fatal("rebuilt payload does not match original chunk")
+	}
+}

stream_flow.go

@@ -3,15 +3,19 @@ package notify
 import (
 	"context"
 	"sync"
+	"sync/atomic"
 )
 
 type streamFlowController struct {
-	mu             sync.Mutex
-	queue          []*streamFlowRequest
-	inFlightBytes  int
-	inFlightChunks int
-	windowBytes    int
-	maxChunks      int
+	mu sync.Mutex
+
+	queue []*streamFlowRequest
+
+	inFlightBytes  atomic.Int64
+	inFlightChunks atomic.Int64
+	windowBytes    atomic.Int64
+	maxChunks      atomic.Int64
+	waiters        atomic.Int32
 }
 
 type streamFlowRequest struct {
@@ -22,10 +26,10 @@ type streamFlowRequest struct {
 
 func newStreamFlowController(cfg streamConfig) *streamFlowController {
 	cfg = normalizeStreamConfig(cfg)
-	return &streamFlowController{
-		windowBytes: cfg.OutboundWindowBytes,
-		maxChunks:   cfg.OutboundMaxInFlightChunks,
-	}
+	controller := &streamFlowController{}
+	controller.windowBytes.Store(int64(cfg.OutboundWindowBytes))
+	controller.maxChunks.Store(int64(cfg.OutboundMaxInFlightChunks))
+	return controller
 }
 
 func (c *streamFlowController) applyConfig(cfg streamConfig) {
@@ -33,9 +37,12 @@ func (c *streamFlowController) applyConfig(cfg streamConfig) {
 		return
 	}
 	cfg = normalizeStreamConfig(cfg)
+	c.windowBytes.Store(int64(cfg.OutboundWindowBytes))
+	c.maxChunks.Store(int64(cfg.OutboundMaxInFlightChunks))
+	if c.waiters.Load() == 0 {
+		return
+	}
 	c.mu.Lock()
-	c.windowBytes = cfg.OutboundWindowBytes
-	c.maxChunks = cfg.OutboundMaxInFlightChunks
 	c.drainLocked()
 	c.mu.Unlock()
 }
@@ -47,58 +54,32 @@ func (c *streamFlowController) acquire(ctx context.Context, size int) (func(), e
 	if ctx == nil {
 		ctx = context.Background()
 	}
+	if c.tryAcquire(size) {
+		return c.releaseFunc(size), nil
+	}
 	req := &streamFlowRequest{
 		size:  size,
 		ready: make(chan struct{}),
 	}
 
 	c.mu.Lock()
+	if c.tryAcquireLocked(size) {
+		c.mu.Unlock()
+		return c.releaseFunc(size), nil
+	}
 	c.queue = append(c.queue, req)
+	c.waiters.Add(1)
 	c.drainLocked()
 	c.mu.Unlock()
 
 	select {
 	case <-req.ready:
-		released := false
-		return func() {
-			c.mu.Lock()
-			if released {
-				c.mu.Unlock()
-				return
-			}
-			released = true
-			c.inFlightBytes -= size
-			if c.inFlightBytes < 0 {
-				c.inFlightBytes = 0
-			}
-			if c.inFlightChunks > 0 {
-				c.inFlightChunks--
-			}
-			c.drainLocked()
-			c.mu.Unlock()
-		}, nil
+		return c.releaseFunc(size), nil
 	case <-ctx.Done():
 		c.mu.Lock()
 		if req.admitted {
 			c.mu.Unlock()
-			released := false
-			return func() {
-				c.mu.Lock()
-				if released {
-					c.mu.Unlock()
-					return
-				}
-				released = true
-				c.inFlightBytes -= size
-				if c.inFlightBytes < 0 {
-					c.inFlightBytes = 0
-				}
-				if c.inFlightChunks > 0 {
-					c.inFlightChunks--
-				}
-				c.drainLocked()
-				c.mu.Unlock()
-			}, nil
+			return c.releaseFunc(size), nil
 		}
 		c.removeLocked(req)
 		c.drainLocked()
@@ -107,6 +88,128 @@ func (c *streamFlowController) acquire(ctx context.Context, size int) (func(), e
 	}
 }
 
+func (c *streamFlowController) tryAcquire(size int) bool {
+	if c == nil || size <= 0 {
+		return true
+	}
+	if c.waiters.Load() != 0 {
+		return false
+	}
+	return c.tryAcquireCAS(size)
+}
+
+func (c *streamFlowController) tryAcquireLocked(size int) bool {
+	if c == nil || size <= 0 {
+		return true
+	}
+	if len(c.queue) != 0 {
+		return false
+	}
+	return c.tryAcquireCAS(size)
+}
+
+func (c *streamFlowController) tryAcquireCAS(size int) bool {
+	if c == nil || size <= 0 {
+		return true
+	}
+	size64 := int64(size)
+	for {
+		window := c.windowBytes.Load()
+		maxChunks := c.maxChunks.Load()
+		inFlightBytes := c.inFlightBytes.Load()
+		inFlightChunks := c.inFlightChunks.Load()
+
+		if maxChunks > 0 && inFlightChunks >= maxChunks {
+			return false
+		}
+		if window > 0 && inFlightBytes+size64 > window {
+			if !(inFlightBytes == 0 && inFlightChunks == 0) {
+				return false
+			}
+		}
+		if !c.inFlightBytes.CompareAndSwap(inFlightBytes, inFlightBytes+size64) {
+			continue
+		}
+		if c.addChunksCAS(1, maxChunks) {
+			return true
+		}
+		c.subBytesCAS(size64)
+		return false
+	}
+}
+
+func (c *streamFlowController) addChunksCAS(delta int64, maxChunks int64) bool {
+	if c == nil || delta <= 0 {
+		return true
+	}
+	for {
+		current := c.inFlightChunks.Load()
+		if maxChunks > 0 && current+delta > maxChunks {
+			return false
+		}
+		if c.inFlightChunks.CompareAndSwap(current, current+delta) {
+			return true
+		}
+	}
+}
+
+func (c *streamFlowController) subBytesCAS(delta int64) {
+	if c == nil || delta <= 0 {
+		return
+	}
+	for {
+		current := c.inFlightBytes.Load()
+		next := current - delta
+		if next < 0 {
+			next = 0
+		}
+		if c.inFlightBytes.CompareAndSwap(current, next) {
+			return
+		}
+	}
+}
+
+func (c *streamFlowController) subChunksCAS(delta int64) {
+	if c == nil || delta <= 0 {
+		return
+	}
+	for {
+		current := c.inFlightChunks.Load()
+		next := current - delta
+		if next < 0 {
+			next = 0
+		}
+		if c.inFlightChunks.CompareAndSwap(current, next) {
+			return
+		}
+	}
+}
+
+func (c *streamFlowController) releaseFunc(size int) func() {
+	released := false
+	return func() {
+		if released {
+			return
+		}
+		released = true
+		c.release(size)
+	}
+}
+
+func (c *streamFlowController) release(size int) {
+	if c == nil || size <= 0 {
+		return
+	}
+	c.subBytesCAS(int64(size))
+	c.subChunksCAS(1)
+	if c.waiters.Load() == 0 {
+		return
+	}
+	c.mu.Lock()
+	c.drainLocked()
+	c.mu.Unlock()
+}
+
 func (c *streamFlowController) removeLocked(req *streamFlowRequest) {
 	if c == nil || req == nil {
 		return
@@ -118,6 +221,7 @@ func (c *streamFlowController) removeLocked(req *streamFlowRequest) {
 		copy(c.queue[i:], c.queue[i+1:])
 		c.queue[len(c.queue)-1] = nil
 		c.queue = c.queue[:len(c.queue)-1]
+		c.waiters.Add(-1)
 		return
 	}
 }
@@ -132,18 +236,17 @@ func (c *streamFlowController) drainLocked() {
 			c.queue = c.queue[1:]
 			continue
 		}
-		if c.maxChunks > 0 && c.inFlightChunks >= c.maxChunks {
+		if !c.canAdmitLocked(req.size) {
 			return
 		}
-		if !c.canAdmitLocked(req.size) {
+		if !c.tryAcquireCAS(req.size) {
 			return
 		}
 		copy(c.queue[0:], c.queue[1:])
 		c.queue[len(c.queue)-1] = nil
 		c.queue = c.queue[:len(c.queue)-1]
+		c.waiters.Add(-1)
 		req.admitted = true
-		c.inFlightBytes += req.size
-		c.inFlightChunks++
 		close(req.ready)
 	}
 }
@@ -155,11 +258,18 @@ func (c *streamFlowController) canAdmitLocked(size int) bool {
 	if size <= 0 {
 		return true
 	}
-	if c.windowBytes <= 0 {
+	window := c.windowBytes.Load()
+	chunks := c.inFlightChunks.Load()
+	bytes := c.inFlightBytes.Load()
+	maxChunks := c.maxChunks.Load()
+	if maxChunks > 0 && chunks >= maxChunks {
+		return false
+	}
+	if window <= 0 {
 		return true
 	}
-	if c.inFlightBytes+size <= c.windowBytes {
+	if bytes+int64(size) <= window {
 		return true
 	}
-	return c.inFlightBytes == 0 && c.inFlightChunks == 0
+	return bytes == 0 && chunks == 0
 }

stream_flow_test.go

@@ -105,3 +105,51 @@ func TestStreamFlowControllerAdmitsRequestsFIFO(t *testing.T) {
 		t.Fatalf("second admitted request = %d, want 3", second)
 	}
 }
+
+func TestStreamFlowControllerTryAcquireDoesNotBypassQueuedWaiter(t *testing.T) {
+	controller := newStreamFlowController(streamConfig{
+		ChunkSize:                 4,
+		InboundQueueLimit:         1,
+		InboundBufferedBytesLimit: 4,
+		OutboundWindowBytes:       4,
+		OutboundMaxInFlightChunks: 1,
+	})
+
+	releaseFirst, err := controller.acquire(context.Background(), 4)
+	if err != nil {
+		t.Fatalf("first acquire failed: %v", err)
+	}
+	defer releaseFirst()
+
+	waiterReady := make(chan struct{})
+	waiterAcquired := make(chan func(), 1)
+	go func() {
+		close(waiterReady)
+		releaseSecond, err := controller.acquire(context.Background(), 4)
+		if err != nil {
+			t.Errorf("waiter acquire failed: %v", err)
+			return
+		}
+		waiterAcquired <- releaseSecond
+	}()
+
+	<-waiterReady
+	time.Sleep(20 * time.Millisecond)
+
+	if controller.tryAcquire(4) {
+		t.Fatal("tryAcquire should not bypass queued waiter")
+	}
+
+	releaseFirst()
+
+	var releaseSecond func()
+	select {
+	case releaseSecond = <-waiterAcquired:
+	case <-time.After(time.Second):
+		t.Fatal("timed out waiting for queued waiter to acquire")
+	}
+	if releaseSecond == nil {
+		t.Fatal("queued waiter returned nil release func")
+	}
+	releaseSecond()
+}

stream_runtime.go

@@ -3,7 +3,6 @@ package notify
 import (
 	"context"
 	"fmt"
-	"strconv"
 	"strings"
 	"sync"
 	"sync/atomic"
@@ -17,7 +16,7 @@ type streamRuntime struct {
 	mu      sync.RWMutex
 	handler func(StreamAcceptInfo) error
 	streams map[string]*streamHandle
-	data    map[string]*streamHandle
+	data    map[string]map[uint64]*streamHandle
 	cfg     streamConfig
 	flow    *streamFlowController
 }
@@ -27,7 +26,7 @@ func newStreamRuntime(rolePrefix string) *streamRuntime {
 	return &streamRuntime{
 		rolePrefix: rolePrefix,
 		streams:    make(map[string]*streamHandle),
-		data:       make(map[string]*streamHandle),
+		data:       make(map[string]map[uint64]*streamHandle),
 		cfg:        cfg,
 		flow:       newStreamFlowController(cfg),
 	}
@@ -72,18 +71,23 @@ func (r *streamRuntime) register(scope string, stream *streamHandle) error {
 	if stream == nil || stream.id == "" {
 		return errStreamIDEmpty
 	}
+	scope = normalizeFileScope(scope)
 	key := streamRuntimeKey(scope, stream.id)
-	dataKey := streamRuntimeDataKey(scope, stream.dataID)
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	if _, ok := r.streams[key]; ok {
 		return errStreamAlreadyExists
 	}
 	if stream.dataID != 0 {
-		if _, ok := r.data[dataKey]; ok {
+		dataScope := r.data[scope]
+		if dataScope == nil {
+			dataScope = make(map[uint64]*streamHandle)
+			r.data[scope] = dataScope
+		}
+		if _, ok := dataScope[stream.dataID]; ok {
 			return errStreamAlreadyExists
 		}
-		r.data[dataKey] = stream
+		dataScope[stream.dataID] = stream
 	}
 	r.streams[key] = stream
 	return nil
@@ -104,10 +108,14 @@ func (r *streamRuntime) lookupByDataID(scope string, dataID uint64) (*streamHand
 	if r == nil || dataID == 0 {
 		return nil, false
 	}
-	key := streamRuntimeDataKey(scope, dataID)
+	scope = normalizeFileScope(scope)
 	r.mu.RLock()
 	defer r.mu.RUnlock()
-	stream, ok := r.data[key]
+	dataScope := r.data[scope]
+	if dataScope == nil {
+		return nil, false
+	}
+	stream, ok := dataScope[dataID]
 	return stream, ok
 }
 
@@ -115,11 +123,17 @@ func (r *streamRuntime) remove(scope string, streamID string) {
 	if r == nil || streamID == "" {
 		return
 	}
+	scope = normalizeFileScope(scope)
 	key := streamRuntimeKey(scope, streamID)
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	if stream := r.streams[key]; stream != nil && stream.dataID != 0 {
-		delete(r.data, streamRuntimeDataKey(scope, stream.dataID))
+		if dataScope := r.data[scope]; dataScope != nil {
+			delete(dataScope, stream.dataID)
+			if len(dataScope) == 0 {
+				delete(r.data, scope)
+			}
+		}
 	}
 	delete(r.streams, key)
 }
@@ -131,6 +145,20 @@ func (r *streamRuntime) acquireOutbound(ctx context.Context, size int) (func(),
 	return r.flow.acquire(ctx, size)
 }
 
+func (r *streamRuntime) tryAcquireOutbound(size int) bool {
+	if r == nil || r.flow == nil {
+		return true
+	}
+	return r.flow.tryAcquire(size)
+}
+
+func (r *streamRuntime) releaseOutbound(size int) {
+	if r == nil || r.flow == nil {
+		return
+	}
+	r.flow.release(size)
+}
+
 func (r *streamRuntime) snapshots() []StreamSnapshot {
 	if r == nil {
 		return nil
@@ -182,10 +210,6 @@ func streamRuntimeKey(scope string, streamID string) string {
 	return normalizeFileScope(scope) + "\x00" + streamID
 }
 
-func streamRuntimeDataKey(scope string, dataID uint64) string {
-	return normalizeFileScope(scope) + "\x01" + strconv.FormatUint(dataID, 10)
-}
-
 func (c *ClientCommon) getStreamRuntime() *streamRuntime {
 	if c == nil {
 		return nil

stream_shared_batch.go

@@ -0,0 +1,145 @@
+package notify
+
+import "encoding/binary"
+
+const (
+	streamFastPathVersionV1      = 1
+	streamFastPathVersionV2      = 2
+	streamFastPathVersionCurrent = streamFastPathVersionV2
+)
+
+const (
+	streamFastBatchMagic         = "NSB1"
+	streamFastBatchVersion       = 1
+	streamFastBatchHeaderLen     = 12
+	streamFastBatchItemHeaderLen = 24
+	streamFastBatchMaxItems      = 64
+	streamFastBatchMaxPlainBytes = 8 * 1024 * 1024
+)
+
+func normalizeStreamFastPathVersion(version uint8) uint8 {
+	if version < streamFastPathVersionV1 {
+		return streamFastPathVersionV1
+	}
+	if version > streamFastPathVersionCurrent {
+		return streamFastPathVersionCurrent
+	}
+	return version
+}
+
+func negotiateStreamFastPathVersion(version uint8) uint8 {
+	return normalizeStreamFastPathVersion(version)
+}
+
+func streamFastPathSupportsBatch(version uint8) bool {
+	return normalizeStreamFastPathVersion(version) >= streamFastPathVersionV2
+}
+
+func streamFastBatchFrameLen(frame streamFastDataFrame) int {
+	return streamFastBatchItemHeaderLen + len(frame.Payload)
+}
+
+func streamFastBatchPlainLen(frames []streamFastDataFrame) int {
+	total := streamFastBatchHeaderLen
+	for _, frame := range frames {
+		total += streamFastBatchFrameLen(frame)
+	}
+	return total
+}
+
+func encodeStreamFastBatchPlain(frames []streamFastDataFrame) ([]byte, error) {
+	if len(frames) == 0 {
+		return nil, errStreamFastPayloadInvalid
+	}
+	buf := make([]byte, streamFastBatchPlainLen(frames))
+	if err := writeStreamFastBatchPlain(buf, frames); err != nil {
+		return nil, err
+	}
+	return buf, nil
+}
+
+func encodeStreamFastBatchPayloadFast(encode transportFastPlainEncoder, secretKey []byte, frames []streamFastDataFrame) ([]byte, error) {
+	if encode == nil {
+		return nil, errTransportPayloadEncryptFailed
+	}
+	plainLen := streamFastBatchPlainLen(frames)
+	return encode(secretKey, plainLen, func(dst []byte) error {
+		return writeStreamFastBatchPlain(dst, frames)
+	})
+}
+
+func writeStreamFastBatchPlain(dst []byte, frames []streamFastDataFrame) error {
+	if len(frames) == 0 || len(dst) != streamFastBatchPlainLen(frames) {
+		return errStreamFastPayloadInvalid
+	}
+	copy(dst[:4], streamFastBatchMagic)
+	dst[4] = streamFastBatchVersion
+	binary.BigEndian.PutUint32(dst[8:12], uint32(len(frames)))
+	offset := streamFastBatchHeaderLen
+	for _, frame := range frames {
+		if frame.DataID == 0 {
+			return errStreamFastPayloadInvalid
+		}
+		dst[offset] = frame.Flags
+		binary.BigEndian.PutUint64(dst[offset+4:offset+12], frame.DataID)
+		binary.BigEndian.PutUint64(dst[offset+12:offset+20], frame.Seq)
+		binary.BigEndian.PutUint32(dst[offset+20:offset+24], uint32(len(frame.Payload)))
+		offset += streamFastBatchItemHeaderLen
+		copy(dst[offset:offset+len(frame.Payload)], frame.Payload)
+		offset += len(frame.Payload)
+	}
+	return nil
+}
+
+func decodeStreamFastBatchPlain(payload []byte) ([]streamFastDataFrame, bool, error) {
+	if len(payload) < 4 || string(payload[:4]) != streamFastBatchMagic {
+		return nil, false, nil
+	}
+	if len(payload) < streamFastBatchHeaderLen {
+		return nil, true, errStreamFastPayloadInvalid
+	}
+	if payload[4] != streamFastBatchVersion {
+		return nil, true, errStreamFastPayloadInvalid
+	}
+	count := int(binary.BigEndian.Uint32(payload[8:12]))
+	if count <= 0 {
+		return nil, true, errStreamFastPayloadInvalid
+	}
+	frames := make([]streamFastDataFrame, 0, count)
+	offset := streamFastBatchHeaderLen
+	for index := 0; index < count; index++ {
+		if len(payload)-offset < streamFastBatchItemHeaderLen {
+			return nil, true, errStreamFastPayloadInvalid
+		}
+		flags := payload[offset]
+		dataID := binary.BigEndian.Uint64(payload[offset+4 : offset+12])
+		seq := binary.BigEndian.Uint64(payload[offset+12 : offset+20])
+		payloadLen := int(binary.BigEndian.Uint32(payload[offset+20 : offset+24]))
+		offset += streamFastBatchItemHeaderLen
+		if dataID == 0 || payloadLen < 0 || len(payload)-offset < payloadLen {
+			return nil, true, errStreamFastPayloadInvalid
+		}
+		frames = append(frames, streamFastDataFrame{
+			Flags:   flags,
+			DataID:  dataID,
+			Seq:     seq,
+			Payload: payload[offset : offset+payloadLen],
+		})
+		offset += payloadLen
+	}
+	if offset != len(payload) {
+		return nil, true, errStreamFastPayloadInvalid
+	}
+	return frames, true, nil
+}
+
+func decodeStreamFastDataFrames(payload []byte) ([]streamFastDataFrame, bool, error) {
+	if frames, matched, err := decodeStreamFastBatchPlain(payload); matched {
+		return frames, true, err
+	}
+	frame, matched, err := decodeStreamFastDataFrame(payload)
+	if !matched || err != nil {
+		return nil, matched, err
+	}
+	return []streamFastDataFrame{frame}, true, nil
+}

stream_snapshot.go

@@ -7,48 +7,52 @@ import (
 )
 
 type StreamSnapshot struct {
-	ID                        string
-	DataID                    uint64
-	Scope                     string
-	Channel                   StreamChannel
-	Metadata                  StreamMetadata
-	BindingOwner              string
-	BindingAlive              bool
-	BindingCurrent            bool
-	BindingReason             string
-	BindingError              string
-	SessionEpoch              uint64
-	LogicalClientID           string
-	LocalAddress              string
-	RemoteAddress             string
-	TransportGeneration       uint64
-	TransportAttached         bool
-	TransportHasRuntimeConn   bool
-	TransportCurrent          bool
-	TransportDetachReason     string
-	TransportDetachKind       string
-	TransportDetachGeneration uint64
-	TransportDetachError      string
-	TransportDetachedAt       time.Time
-	ReattachEligible          bool
-	LocalClosed               bool
-	LocalReadClosed           bool
-	RemoteClosed              bool
-	PeerReadClosed            bool
-	BufferedChunks            int
-	BufferedBytes             int
-	ReadTimeout               time.Duration
-	WriteTimeout              time.Duration
-	BytesRead                 int64
-	BytesWritten              int64
-	ReadCalls                 int64
-	WriteCalls                int64
-	OpenedAt                  time.Time
-	LastReadAt                time.Time
-	LastWriteAt               time.Time
-	ReadDeadline              time.Time
-	WriteDeadline             time.Time
-	ResetError                string
+	ID                                      string
+	DataID                                  uint64
+	Scope                                   string
+	Channel                                 StreamChannel
+	Metadata                                StreamMetadata
+	BindingOwner                            string
+	BindingAlive                            bool
+	BindingCurrent                          bool
+	BindingReason                           string
+	BindingError                            string
+	BindingBulkAdaptiveSoftPayloadBytes     int
+	BindingStreamAdaptiveSoftPayloadBytes   int
+	BindingStreamAdaptiveWaitThresholdBytes int
+	BindingStreamAdaptiveFlushDelay         time.Duration
+	SessionEpoch                            uint64
+	LogicalClientID                         string
+	LocalAddress                            string
+	RemoteAddress                           string
+	TransportGeneration                     uint64
+	TransportAttached                       bool
+	TransportHasRuntimeConn                 bool
+	TransportCurrent                        bool
+	TransportDetachReason                   string
+	TransportDetachKind                     string
+	TransportDetachGeneration               uint64
+	TransportDetachError                    string
+	TransportDetachedAt                     time.Time
+	ReattachEligible                        bool
+	LocalClosed                             bool
+	LocalReadClosed                         bool
+	RemoteClosed                            bool
+	PeerReadClosed                          bool
+	BufferedChunks                          int
+	BufferedBytes                           int
+	ReadTimeout                             time.Duration
+	WriteTimeout                            time.Duration
+	BytesRead                               int64
+	BytesWritten                            int64
+	ReadCalls                               int64
+	WriteCalls                              int64
+	OpenedAt                                time.Time
+	LastReadAt                              time.Time
+	LastWriteAt                             time.Time
+	ReadDeadline                            time.Time
+	WriteDeadline                           time.Time
+	ResetError                              string
 }
 
 type clientStreamSnapshotReader interface {

transport_binding.go

@@ -15,9 +15,14 @@ type transportBinding struct {
 	queue   *stario.StarQueue
 	writeMu sync.Mutex
 
+	adaptiveTx adaptiveTxState
+
 	controlMu     sync.Mutex
 	controlSender *controlBatchSender
 
+	streamMu     sync.Mutex
+	streamSender *streamBatchSender
+
 	bulkMu     sync.Mutex
 	bulkSender *bulkBatchSender
 }
@@ -71,7 +76,7 @@ func (b *transportBinding) withConnWriteLockDeadline(deadline time.Time, fn func
 	return fn(conn)
 }
 
-func (b *transportBinding) bulkBatchSenderSnapshot() *bulkBatchSender {
+func (b *transportBinding) bulkBatchSenderSnapshotWithCodec(codec bulkBatchCodec, writeTimeout func() time.Duration) *bulkBatchSender {
 	if b == nil {
 		return nil
 	}
@@ -80,10 +85,39 @@ func (b *transportBinding) bulkBatchSenderSnapshot() *bulkBatchSender {
 	if b.bulkSender != nil {
 		return b.bulkSender
 	}
-	b.bulkSender = newBulkBatchSender(b)
+	b.bulkSender = newBulkBatchSender(b, codec, writeTimeout)
 	return b.bulkSender
 }
 
+func (b *transportBinding) clientBulkBatchSenderSnapshot(c *ClientCommon) *bulkBatchSender {
+	if b == nil || c == nil {
+		return nil
+	}
+	return b.bulkBatchSenderSnapshotWithCodec(bulkBatchCodec{
+		encodeSingle: c.encodeBulkFastPayloadPooled,
+		encodeBatch:  c.encodeBulkFastBatchPayloadPooled,
+	}, c.maxWriteTimeoutSnapshot)
+}
+
+func (b *transportBinding) serverBulkBatchSenderSnapshot(logical *LogicalConn) *bulkBatchSender {
+	if b == nil || logical == nil {
+		return nil
+	}
+	server := logical.Server()
+	common, ok := server.(*ServerCommon)
+	if !ok || common == nil {
+		return nil
+	}
+	return b.bulkBatchSenderSnapshotWithCodec(bulkBatchCodec{
+		encodeSingle: func(frame bulkFastFrame) ([]byte, func(), error) {
+			return common.encodeBulkFastPayloadLogicalPooled(logical, frame)
+		},
+		encodeBatch: func(frames []bulkFastFrame) ([]byte, func(), error) {
+			return common.encodeBulkFastBatchPayloadLogicalPooled(logical, frames)
+		},
+	}, logical.maxWriteTimeoutSnapshot)
+}
+
 func (b *transportBinding) controlBatchSenderSnapshot() *controlBatchSender {
 	if b == nil {
 		return nil
@@ -97,6 +131,48 @@ func (b *transportBinding) controlBatchSenderSnapshot() *controlBatchSender {
 	return b.controlSender
 }
 
+func (b *transportBinding) streamBatchSenderSnapshotWithCodec(codec streamBatchCodec, writeTimeout func() time.Duration) *streamBatchSender {
+	if b == nil {
+		return nil
+	}
+	b.streamMu.Lock()
+	defer b.streamMu.Unlock()
+	if b.streamSender != nil {
+		return b.streamSender
+	}
+	b.streamSender = newStreamBatchSender(b, codec, writeTimeout)
+	return b.streamSender
+}
+
+func (b *transportBinding) clientStreamBatchSenderSnapshot(c *ClientCommon) *streamBatchSender {
+	if b == nil || c == nil {
+		return nil
+	}
+	return b.streamBatchSenderSnapshotWithCodec(streamBatchCodec{
+		encodeSingle: c.encodeFastStreamPayload,
+		encodeBatch:  c.encodeFastStreamBatchPayload,
+	}, c.maxWriteTimeoutSnapshot)
+}
+
+func (b *transportBinding) serverStreamBatchSenderSnapshot(logical *LogicalConn) *streamBatchSender {
+	if b == nil || logical == nil {
+		return nil
+	}
+	server := logical.Server()
+	common, ok := server.(*ServerCommon)
+	if !ok || common == nil {
+		return nil
+	}
+	return b.streamBatchSenderSnapshotWithCodec(streamBatchCodec{
+		encodeSingle: func(frame streamFastDataFrame) ([]byte, error) {
+			return common.encodeFastStreamPayloadLogical(logical, frame)
+		},
+		encodeBatch: func(frames []streamFastDataFrame) ([]byte, error) {
+			return common.encodeFastStreamBatchPayloadLogical(logical, frames)
+		},
+	}, logical.maxWriteTimeoutSnapshot)
+}
+
 func (b *transportBinding) stopBackgroundWorkers() {
 	if b == nil {
 		return
@@ -104,12 +180,18 @@ func (b *transportBinding) stopBackgroundWorkers() {
 	b.controlMu.Lock()
 	controlSender := b.controlSender
 	b.controlMu.Unlock()
+	b.streamMu.Lock()
+	streamSender := b.streamSender
+	b.streamMu.Unlock()
 	b.bulkMu.Lock()
 	bulkSender := b.bulkSender
 	b.bulkMu.Unlock()
 	if controlSender != nil {
 		controlSender.stop()
 	}
+	if streamSender != nil {
+		streamSender.stop()
+	}
 	if bulkSender != nil {
 		bulkSender.stop()
 	}

transport_binding_adaptive.go

@@ -0,0 +1,383 @@
+package notify
+
+import (
+	"sync"
+	"time"
+)
+
+const (
+	bulkAdaptiveSoftPayloadMinBytes       = 256 * 1024
+	bulkAdaptiveSoftPayloadFallbackBytes  = 2 * 1024 * 1024
+	bulkAdaptiveSoftPayloadStartBytes     = bulkFastBatchMaxPlainBytes
+	bulkAdaptiveSoftPayloadTargetFlush    = 4 * time.Millisecond
+	bulkAdaptiveSoftPayloadSlowFlush      = 16 * time.Millisecond
+	bulkAdaptiveSoftPayloadMinSampleBytes = 256 * 1024
+	bulkAdaptiveSoftPayloadGrowSuccesses  = 8
+
+	streamAdaptiveSoftPayloadMinBytes       = 256 * 1024
+	streamAdaptiveSoftPayloadFallbackBytes  = streamBatchMaxPayloadBytes
+	streamAdaptiveSoftPayloadStartBytes     = streamBatchMaxPayloadBytes
+	streamAdaptiveSoftPayloadTargetFlush    = 4 * time.Millisecond
+	streamAdaptiveSoftPayloadSlowFlush      = 16 * time.Millisecond
+	streamAdaptiveSoftPayloadMinSampleBytes = 64 * 1024
+	streamAdaptiveSoftPayloadGrowSuccesses  = 8
+
+	streamAdaptiveWaitThresholdMinBytes = 32 * 1024
+	streamAdaptiveFlushDelayMid         = 25 * time.Microsecond
+)
+
+var bulkAdaptiveSoftPayloadSteps = [...]int{
+	256 * 1024,
+	512 * 1024,
+	1024 * 1024,
+	2 * 1024 * 1024,
+	4 * 1024 * 1024,
+	bulkFastBatchMaxPlainBytes,
+}
+
+var streamAdaptiveSoftPayloadSteps = [...]int{
+	256 * 1024,
+	512 * 1024,
+	1024 * 1024,
+	streamBatchMaxPayloadBytes,
+}
+
+type adaptiveTxState struct {
+	mu sync.Mutex
+
+	bulkSoftPayloadBytes int
+	bulkGoodputBytesPerS float64
+	bulkGrowStreak       int
+
+	streamSoftPayloadBytes int
+	streamGoodputBytesPerS float64
+	streamGrowStreak       int
+}
+
+func (b *transportBinding) bulkAdaptiveSoftPayloadBytesSnapshot() int {
+	if b == nil {
+		return bulkAdaptiveSoftPayloadFallbackBytes
+	}
+	return b.adaptiveTx.bulkSoftPayloadBytesSnapshot()
+}
+
+func (b *transportBinding) observeBulkAdaptivePayloadWrite(payloadBytes int, elapsed time.Duration, timeout time.Duration, err error) {
+	if b == nil {
+		return
+	}
+	b.adaptiveTx.observeBulkPayloadWrite(payloadBytes, elapsed, timeout, err)
+}
+
+func (b *transportBinding) streamAdaptiveSoftPayloadBytesSnapshot() int {
+	if b == nil {
+		return streamAdaptiveSoftPayloadFallbackBytes
+	}
+	return b.adaptiveTx.streamSoftPayloadBytesSnapshot()
+}
+
+func (b *transportBinding) streamAdaptiveWaitThresholdBytesSnapshot() int {
+	if b == nil {
+		return streamBatchWaitThreshold
+	}
+	return b.adaptiveTx.streamWaitThresholdBytesSnapshot()
+}
+
+func (b *transportBinding) streamAdaptiveFlushDelaySnapshot() time.Duration {
+	if b == nil {
+		return streamBatchMaxFlushDelay
+	}
+	return b.adaptiveTx.streamFlushDelaySnapshot()
+}
+
+func (b *transportBinding) observeStreamAdaptivePayloadWrite(payloadBytes int, elapsed time.Duration, timeout time.Duration, err error) {
+	if b == nil {
+		return
+	}
+	b.adaptiveTx.observeStreamPayloadWrite(payloadBytes, elapsed, timeout, err)
+}
+
+func (s *adaptiveTxState) bulkSoftPayloadBytesSnapshot() int {
+	if s == nil {
+		return bulkAdaptiveSoftPayloadStartBytes
+	}
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.bulkSoftPayloadBytesLocked()
+}
+
+func (s *adaptiveTxState) observeBulkPayloadWrite(payloadBytes int, elapsed time.Duration, timeout time.Duration, err error) {
+	if s == nil || payloadBytes <= 0 {
+		return
+	}
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	current := s.bulkSoftPayloadBytesLocked()
+	target, hasSample := s.observeBulkGoodputLocked(payloadBytes, elapsed)
+	nearTimeout := timeout > 0 && elapsed >= (timeout*3)/4
+	if isTimeoutLikeError(err) || nearTimeout {
+		s.bulkGrowStreak = 0
+		if hasSample && target < current {
+			s.bulkSoftPayloadBytes = target
+			return
+		}
+		s.bulkSoftPayloadBytes = previousBulkAdaptiveSoftPayloadStep(current)
+		return
+	}
+	if err != nil {
+		s.bulkGrowStreak = 0
+		return
+	}
+	if !hasSample {
+		return
+	}
+	if elapsed >= bulkAdaptiveSoftPayloadSlowFlush {
+		s.bulkGrowStreak = 0
+		if target < current {
+			s.bulkSoftPayloadBytes = target
+			return
+		}
+		s.bulkSoftPayloadBytes = previousBulkAdaptiveSoftPayloadStep(current)
+		return
+	}
+	if target > current {
+		s.bulkGrowStreak++
+		if s.bulkGrowStreak >= bulkAdaptiveSoftPayloadGrowSuccesses {
+			s.bulkSoftPayloadBytes = nextBulkAdaptiveSoftPayloadStep(current, target)
+			s.bulkGrowStreak = 0
+		}
+		return
+	}
+	if target < current && elapsed >= bulkAdaptiveSoftPayloadTargetFlush*2 {
+		s.bulkSoftPayloadBytes = target
+		s.bulkGrowStreak = 0
+		return
+	}
+	s.bulkGrowStreak = 0
+}
+
+func (s *adaptiveTxState) bulkSoftPayloadBytesLocked() int {
+	if s.bulkSoftPayloadBytes == 0 {
+		s.bulkSoftPayloadBytes = bulkAdaptiveSoftPayloadStartBytes
+	}
+	return normalizeBulkAdaptiveSoftPayloadBytes(s.bulkSoftPayloadBytes)
+}
+
+func (s *adaptiveTxState) streamSoftPayloadBytesSnapshot() int {
+	if s == nil {
+		return streamAdaptiveSoftPayloadStartBytes
+	}
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.streamSoftPayloadBytesLocked()
+}
+
+func (s *adaptiveTxState) streamWaitThresholdBytesSnapshot() int {
+	if s == nil {
+		return streamBatchWaitThreshold
+	}
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return streamAdaptiveWaitThresholdBytesForSoftPayload(s.streamSoftPayloadBytesLocked())
+}
+
+func (s *adaptiveTxState) streamFlushDelaySnapshot() time.Duration {
+	if s == nil {
+		return streamBatchMaxFlushDelay
+	}
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return streamAdaptiveFlushDelayForSoftPayload(s.streamSoftPayloadBytesLocked())
+}
+
+func (s *adaptiveTxState) observeStreamPayloadWrite(payloadBytes int, elapsed time.Duration, timeout time.Duration, err error) {
+	if s == nil || payloadBytes <= 0 {
+		return
+	}
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	current := s.streamSoftPayloadBytesLocked()
+	target, hasSample := s.observeStreamGoodputLocked(payloadBytes, elapsed)
+	nearTimeout := timeout > 0 && elapsed >= (timeout*3)/4
+	if isTimeoutLikeError(err) || nearTimeout {
+		s.streamGrowStreak = 0
+		if hasSample && target < current {
+			s.streamSoftPayloadBytes = target
+			return
+		}
+		s.streamSoftPayloadBytes = previousStreamAdaptiveSoftPayloadStep(current)
+		return
+	}
+	if err != nil {
+		s.streamGrowStreak = 0
+		return
+	}
+	if !hasSample {
+		return
+	}
+	if elapsed >= streamAdaptiveSoftPayloadSlowFlush {
+		s.streamGrowStreak = 0
+		if target < current {
+			s.streamSoftPayloadBytes = target
+			return
+		}
+		s.streamSoftPayloadBytes = previousStreamAdaptiveSoftPayloadStep(current)
+		return
+	}
+	if target > current {
+		s.streamGrowStreak++
+		if s.streamGrowStreak >= streamAdaptiveSoftPayloadGrowSuccesses {
+			s.streamSoftPayloadBytes = nextStreamAdaptiveSoftPayloadStep(current, target)
+			s.streamGrowStreak = 0
+		}
+		return
+	}
+	if target < current && elapsed >= streamAdaptiveSoftPayloadTargetFlush*2 {
+		s.streamSoftPayloadBytes = target
+		s.streamGrowStreak = 0
+		return
+	}
+	s.streamGrowStreak = 0
+}
+
+func (s *adaptiveTxState) streamSoftPayloadBytesLocked() int {
+	if s.streamSoftPayloadBytes == 0 {
+		s.streamSoftPayloadBytes = streamAdaptiveSoftPayloadStartBytes
+	}
+	return normalizeStreamAdaptiveSoftPayloadBytes(s.streamSoftPayloadBytes)
+}
+
+func (s *adaptiveTxState) observeBulkGoodputLocked(payloadBytes int, elapsed time.Duration) (int, bool) {
+	if payloadBytes < bulkAdaptiveSoftPayloadMinSampleBytes || elapsed <= 0 {
+		return 0, false
+	}
+	sample := float64(payloadBytes) / elapsed.Seconds()
+	if sample <= 0 {
+		return 0, false
+	}
+	if s.bulkGoodputBytesPerS <= 0 {
+		s.bulkGoodputBytesPerS = sample
+	} else {
+		const alpha = 0.25
+		s.bulkGoodputBytesPerS = s.bulkGoodputBytesPerS*(1-alpha) + sample*alpha
+	}
+	target := int(s.bulkGoodputBytesPerS * bulkAdaptiveSoftPayloadTargetFlush.Seconds())
+	return normalizeBulkAdaptiveSoftPayloadBytes(target), true
+}
+
+func (s *adaptiveTxState) observeStreamGoodputLocked(payloadBytes int, elapsed time.Duration) (int, bool) {
+	if payloadBytes < streamAdaptiveSoftPayloadMinSampleBytes || elapsed <= 0 {
+		return 0, false
+	}
+	sample := float64(payloadBytes) / elapsed.Seconds()
+	if sample <= 0 {
+		return 0, false
+	}
+	if s.streamGoodputBytesPerS <= 0 {
+		s.streamGoodputBytesPerS = sample
+	} else {
+		const alpha = 0.25
+		s.streamGoodputBytesPerS = s.streamGoodputBytesPerS*(1-alpha) + sample*alpha
+	}
+	target := int(s.streamGoodputBytesPerS * streamAdaptiveSoftPayloadTargetFlush.Seconds())
+	return normalizeStreamAdaptiveSoftPayloadBytes(target), true
+}
+
+func normalizeBulkAdaptiveSoftPayloadBytes(size int) int {
+	if size <= bulkAdaptiveSoftPayloadMinBytes {
+		return bulkAdaptiveSoftPayloadMinBytes
+	}
+	for _, step := range bulkAdaptiveSoftPayloadSteps {
+		if size <= step {
+			return step
+		}
+	}
+	return bulkAdaptiveSoftPayloadStartBytes
+}
+
+func previousBulkAdaptiveSoftPayloadStep(current int) int {
+	current = normalizeBulkAdaptiveSoftPayloadBytes(current)
+	for index := len(bulkAdaptiveSoftPayloadSteps) - 1; index >= 0; index-- {
+		step := bulkAdaptiveSoftPayloadSteps[index]
+		if current > step {
+			return step
+		}
+	}
+	return bulkAdaptiveSoftPayloadMinBytes
+}
+
+func nextBulkAdaptiveSoftPayloadStep(current int, target int) int {
+	current = normalizeBulkAdaptiveSoftPayloadBytes(current)
+	target = normalizeBulkAdaptiveSoftPayloadBytes(target)
+	for _, step := range bulkAdaptiveSoftPayloadSteps {
+		if step > current {
+			if step > target {
+				return target
+			}
+			return step
+		}
+	}
+	return bulkAdaptiveSoftPayloadStartBytes
+}
+
+func normalizeStreamAdaptiveSoftPayloadBytes(size int) int {
+	if size <= streamAdaptiveSoftPayloadMinBytes {
+		return streamAdaptiveSoftPayloadMinBytes
+	}
+	for _, step := range streamAdaptiveSoftPayloadSteps {
+		if size <= step {
+			return step
+		}
+	}
+	return streamAdaptiveSoftPayloadStartBytes
+}
+
+func previousStreamAdaptiveSoftPayloadStep(current int) int {
+	current = normalizeStreamAdaptiveSoftPayloadBytes(current)
+	for index := len(streamAdaptiveSoftPayloadSteps) - 1; index >= 0; index-- {
+		step := streamAdaptiveSoftPayloadSteps[index]
+		if current > step {
+			return step
+		}
+	}
+	return streamAdaptiveSoftPayloadMinBytes
+}
+
+func nextStreamAdaptiveSoftPayloadStep(current int, target int) int {
+	current = normalizeStreamAdaptiveSoftPayloadBytes(current)
+	target = normalizeStreamAdaptiveSoftPayloadBytes(target)
+	for _, step := range streamAdaptiveSoftPayloadSteps {
+		if step > current {
+			if step > target {
+				return target
+			}
+			return step
+		}
+	}
+	return streamAdaptiveSoftPayloadStartBytes
+}
+
+func streamAdaptiveWaitThresholdBytesForSoftPayload(size int) int {
+	size = normalizeStreamAdaptiveSoftPayloadBytes(size)
+	threshold := size / 16
+	if threshold < streamAdaptiveWaitThresholdMinBytes {
+		return streamAdaptiveWaitThresholdMinBytes
+	}
+	if threshold > streamBatchWaitThreshold {
+		return streamBatchWaitThreshold
+	}
+	return threshold
+}
+
+func streamAdaptiveFlushDelayForSoftPayload(size int) time.Duration {
+	size = normalizeStreamAdaptiveSoftPayloadBytes(size)
+	switch {
+	case size >= streamAdaptiveSoftPayloadStartBytes:
+		return streamBatchMaxFlushDelay
+	case size >= 1024*1024:
+		return streamAdaptiveFlushDelayMid
+	default:
+		return 0
+	}
+}

transport_binding_adaptive_test.go

@@ -0,0 +1,160 @@
+package notify
+
+import (
+	"b612.me/stario"
+	"net"
+	"testing"
+	"time"
+)
+
+func TestTransportBindingAdaptiveBulkSoftPayloadStartsAggressive(t *testing.T) {
+	binding := &transportBinding{}
+	if got, want := binding.bulkAdaptiveSoftPayloadBytesSnapshot(), bulkAdaptiveSoftPayloadStartBytes; got != want {
+		t.Fatalf("adaptive bulk soft payload = %d, want %d", got, want)
+	}
+}
+
+func TestTransportBindingAdaptiveBulkSoftPayloadShrinksAfterSlowWrite(t *testing.T) {
+	binding := &transportBinding{}
+	binding.observeBulkAdaptivePayloadWrite(8*1024*1024, 640*time.Millisecond, 0, nil)
+	if got, want := binding.bulkAdaptiveSoftPayloadBytesSnapshot(), bulkAdaptiveSoftPayloadMinBytes; got != want {
+		t.Fatalf("adaptive bulk soft payload = %d, want %d", got, want)
+	}
+}
+
+func TestTransportBindingAdaptiveBulkSoftPayloadRecoversAfterGoodWrites(t *testing.T) {
+	binding := &transportBinding{}
+	binding.observeBulkAdaptivePayloadWrite(8*1024*1024, 640*time.Millisecond, 0, nil)
+	samples := bulkAdaptiveSoftPayloadGrowSuccesses * (len(bulkAdaptiveSoftPayloadSteps) - 1)
+	for i := 0; i < samples; i++ {
+		binding.observeBulkAdaptivePayloadWrite(8*1024*1024, 6*time.Millisecond, 0, nil)
+	}
+	if got, want := binding.bulkAdaptiveSoftPayloadBytesSnapshot(), bulkAdaptiveSoftPayloadStartBytes; got != want {
+		t.Fatalf("adaptive bulk soft payload = %d, want %d", got, want)
+	}
+}
+
+type delayedWriteConn struct {
+	delay time.Duration
+}
+
+func (c *delayedWriteConn) Read([]byte) (int, error)         { return 0, net.ErrClosed }
+func (c *delayedWriteConn) Write(p []byte) (int, error)      { time.Sleep(c.delay); return len(p), nil }
+func (c *delayedWriteConn) Close() error                     { return nil }
+func (c *delayedWriteConn) LocalAddr() net.Addr              { return nil }
+func (c *delayedWriteConn) RemoteAddr() net.Addr             { return nil }
+func (c *delayedWriteConn) SetDeadline(time.Time) error      { return nil }
+func (c *delayedWriteConn) SetReadDeadline(time.Time) error  { return nil }
+func (c *delayedWriteConn) SetWriteDeadline(time.Time) error { return nil }
+
+func TestBulkBatchSenderFlushAdaptsToSlowWrites(t *testing.T) {
+	binding := newTransportBinding(&delayedWriteConn{delay: 20 * time.Millisecond}, stario.NewQueue())
+	sender := newTestBulkBatchSender(binding)
+	defer sender.stop()
+
+	payload := make([]byte, 512*1024)
+	err := sender.flush([]bulkBatchRequest{
+		{
+			frames: []bulkFastFrame{
+				{Type: bulkFastPayloadTypeData, DataID: 101, Seq: 1, Payload: payload},
+				{Type: bulkFastPayloadTypeData, DataID: 101, Seq: 2, Payload: payload},
+			},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+		{
+			frames: []bulkFastFrame{
+				{Type: bulkFastPayloadTypeData, DataID: 202, Seq: 1, Payload: payload},
+				{Type: bulkFastPayloadTypeData, DataID: 202, Seq: 2, Payload: payload},
+			},
+			fastPathVersion: bulkFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("flush failed: %v", err)
+	}
+	if got := binding.bulkAdaptiveSoftPayloadBytesSnapshot(); got >= bulkAdaptiveSoftPayloadStartBytes {
+		t.Fatalf("adaptive bulk soft payload = %d, want smaller than %d after slow write", got, bulkAdaptiveSoftPayloadStartBytes)
+	}
+}
+
+func TestTransportBindingAdaptiveStreamStartsAggressive(t *testing.T) {
+	binding := &transportBinding{}
+	if got, want := binding.streamAdaptiveSoftPayloadBytesSnapshot(), streamAdaptiveSoftPayloadStartBytes; got != want {
+		t.Fatalf("adaptive stream soft payload = %d, want %d", got, want)
+	}
+	if got, want := binding.streamAdaptiveWaitThresholdBytesSnapshot(), streamBatchWaitThreshold; got != want {
+		t.Fatalf("adaptive stream wait threshold = %d, want %d", got, want)
+	}
+	if got, want := binding.streamAdaptiveFlushDelaySnapshot(), streamBatchMaxFlushDelay; got != want {
+		t.Fatalf("adaptive stream flush delay = %s, want %s", got, want)
+	}
+}
+
+func TestTransportBindingAdaptiveStreamShrinksAfterSlowWrite(t *testing.T) {
+	binding := &transportBinding{}
+	binding.observeStreamAdaptivePayloadWrite(2*1024*1024, 640*time.Millisecond, 0, nil)
+	if got, want := binding.streamAdaptiveSoftPayloadBytesSnapshot(), streamAdaptiveSoftPayloadMinBytes; got != want {
+		t.Fatalf("adaptive stream soft payload = %d, want %d", got, want)
+	}
+	if got, want := binding.streamAdaptiveWaitThresholdBytesSnapshot(), streamAdaptiveWaitThresholdMinBytes; got != want {
+		t.Fatalf("adaptive stream wait threshold = %d, want %d", got, want)
+	}
+	if got := binding.streamAdaptiveFlushDelaySnapshot(); got != 0 {
+		t.Fatalf("adaptive stream flush delay = %s, want 0", got)
+	}
+}
+
+func TestTransportBindingAdaptiveStreamRecoversAfterGoodWrites(t *testing.T) {
+	binding := &transportBinding{}
+	binding.observeStreamAdaptivePayloadWrite(2*1024*1024, 640*time.Millisecond, 0, nil)
+	samples := streamAdaptiveSoftPayloadGrowSuccesses * (len(streamAdaptiveSoftPayloadSteps) - 1)
+	for i := 0; i < samples; i++ {
+		binding.observeStreamAdaptivePayloadWrite(2*1024*1024, 6*time.Millisecond, 0, nil)
+	}
+	if got, want := binding.streamAdaptiveSoftPayloadBytesSnapshot(), streamAdaptiveSoftPayloadStartBytes; got != want {
+		t.Fatalf("adaptive stream soft payload = %d, want %d", got, want)
+	}
+	if got, want := binding.streamAdaptiveWaitThresholdBytesSnapshot(), streamBatchWaitThreshold; got != want {
+		t.Fatalf("adaptive stream wait threshold = %d, want %d", got, want)
+	}
+	if got, want := binding.streamAdaptiveFlushDelaySnapshot(), streamBatchMaxFlushDelay; got != want {
+		t.Fatalf("adaptive stream flush delay = %s, want %s", got, want)
+	}
+}
+
+func TestStreamBatchSenderFlushAdaptsToSlowWrites(t *testing.T) {
+	binding := newTransportBinding(&delayedWriteConn{delay: 20 * time.Millisecond}, stario.NewQueue())
+	sender := newTestStreamBatchSender(binding, nil)
+	defer sender.stop()
+
+	payload := make([]byte, 512*1024)
+	err := sender.flush([]streamBatchRequest{
+		{
+			frame: streamFastDataFrame{
+				DataID:  101,
+				Seq:     1,
+				Payload: payload,
+			},
+			hasFrame:        true,
+			fastPathVersion: streamFastPathVersionV2,
+		},
+		{
+			frame: streamFastDataFrame{
+				DataID:  202,
+				Seq:     1,
+				Payload: payload,
+			},
+			hasFrame:        true,
+			fastPathVersion: streamFastPathVersionV2,
+		},
+	})
+	if err != nil {
+		t.Fatalf("flush failed: %v", err)
+	}
+	if got := binding.streamAdaptiveSoftPayloadBytesSnapshot(); got >= streamAdaptiveSoftPayloadStartBytes {
+		t.Fatalf("adaptive stream soft payload = %d, want smaller than %d after slow write", got, streamAdaptiveSoftPayloadStartBytes)
+	}
+	if got := binding.streamAdaptiveWaitThresholdBytesSnapshot(); got >= streamBatchWaitThreshold {
+		t.Fatalf("adaptive stream wait threshold = %d, want smaller than %d after slow write", got, streamBatchWaitThreshold)
+	}
+}

transport_codec.go

@@ -9,12 +9,93 @@ var (
 	errTransportPayloadDecryptFailed = errors.New("transport payload decrypt failed")
 )
 
+func encryptTransportPayloadCodec(runtime *modernPSKCodecRuntime, msgEn func([]byte, []byte) []byte, secretKey []byte, data []byte) ([]byte, error) {
+	if runtime != nil {
+		encoded, err := runtime.sealPlainPayload(data)
+		if err != nil {
+			return nil, errTransportPayloadEncryptFailed
+		}
+		return encoded, nil
+	}
+	if msgEn == nil {
+		return nil, errTransportPayloadEncryptFailed
+	}
+	encoded := msgEn(secretKey, data)
+	if encoded == nil && len(data) != 0 {
+		return nil, errTransportPayloadEncryptFailed
+	}
+	return encoded, nil
+}
+
+func decryptTransportPayloadCodec(runtime *modernPSKCodecRuntime, msgDe func([]byte, []byte) []byte, secretKey []byte, data []byte) ([]byte, error) {
+	if runtime != nil {
+		plain, err := runtime.openPayload(data)
+		if err != nil {
+			return nil, errTransportPayloadDecryptFailed
+		}
+		return plain, nil
+	}
+	if msgDe == nil {
+		return nil, errTransportPayloadDecryptFailed
+	}
+	plain := msgDe(secretKey, data)
+	if plain == nil && len(data) != 0 {
+		return nil, errTransportPayloadDecryptFailed
+	}
+	return plain, nil
+}
+
+func decryptTransportPayloadCodecPooled(runtime *modernPSKCodecRuntime, msgDe func([]byte, []byte) []byte, secretKey []byte, data []byte, release func()) ([]byte, func(), error) {
+	if runtime != nil {
+		plain, plainRelease, err := runtime.openPayloadPooled(data, release)
+		if err != nil {
+			return nil, nil, errTransportPayloadDecryptFailed
+		}
+		return plain, plainRelease, nil
+	}
+	if msgDe == nil {
+		if release != nil {
+			release()
+		}
+		return nil, nil, errTransportPayloadDecryptFailed
+	}
+	plain := msgDe(secretKey, data)
+	if release != nil {
+		release()
+	}
+	if plain == nil && len(data) != 0 {
+		return nil, nil, errTransportPayloadDecryptFailed
+	}
+	return plain, nil, nil
+}
+
+func decryptTransportPayloadCodecOwnedPooled(runtime *modernPSKCodecRuntime, msgDe func([]byte, []byte) []byte, secretKey []byte, data []byte) ([]byte, func(), error) {
+	if runtime != nil {
+		plain, plainRelease, err := runtime.openPayloadOwnedPooled(data)
+		if err != nil {
+			return nil, nil, errTransportPayloadDecryptFailed
+		}
+		return plain, plainRelease, nil
+	}
+	if msgDe == nil {
+		return nil, nil, errTransportPayloadDecryptFailed
+	}
+	plain := msgDe(secretKey, data)
+	if plain == nil && len(data) != 0 {
+		return nil, nil, errTransportPayloadDecryptFailed
+	}
+	return plain, nil, nil
+}
+
 func (c *ClientCommon) encodeTransferMsg(msg TransferMsg) ([]byte, error) {
 	data, err := c.sequenceEn(msg)
 	if err != nil {
 		return nil, err
 	}
-	data = c.msgEn(c.SecretKey, data)
+	data, err = c.encryptTransportPayload(data)
+	if err != nil {
+		return nil, err
+	}
 	queue := c.clientQueueSnapshot()
 	if queue == nil {
 		return nil, errClientSessionQueueUnavailable
@@ -23,7 +104,11 @@ func (c *ClientCommon) encodeTransferMsg(msg TransferMsg) ([]byte, error) {
 }
 
 func (c *ClientCommon) decodeTransferMsg(data []byte) (TransferMsg, error) {
-	msg, err := c.sequenceDe(c.msgDe(c.SecretKey, data))
+	plain, err := c.decryptTransportPayload(data)
+	if err != nil {
+		return TransferMsg{}, err
+	}
+	msg, err := c.sequenceDe(plain)
 	if err != nil {
 		return TransferMsg{}, err
 	}
@@ -41,7 +126,10 @@ func (s *ServerCommon) encodeTransferMsg(c *ClientConn, msg TransferMsg) ([]byte
 	}
 	msgEn := c.clientConnMsgEnSnapshot()
 	secretKey := c.clientConnSecretKeySnapshot()
-	data = msgEn(secretKey, data)
+	data, err = encryptTransportPayloadCodec(c.clientConnModernPSKRuntimeSnapshot(), msgEn, secretKey, data)
+	if err != nil {
+		return nil, err
+	}
 	queue := s.serverQueueSnapshot()
 	if queue == nil {
 		return nil, errServerSessionQueueUnavailable
@@ -52,7 +140,11 @@ func (s *ServerCommon) encodeTransferMsg(c *ClientConn, msg TransferMsg) ([]byte
 func (s *ServerCommon) decodeTransferMsg(c *ClientConn, data []byte) (TransferMsg, error) {
 	msgDe := c.clientConnMsgDeSnapshot()
 	secretKey := c.clientConnSecretKeySnapshot()
-	msg, err := s.sequenceDe(msgDe(secretKey, data))
+	plain, err := decryptTransportPayloadCodec(c.clientConnModernPSKRuntimeSnapshot(), msgDe, secretKey, data)
+	if err != nil {
+		return TransferMsg{}, err
+	}
+	msg, err := s.sequenceDe(plain)
 	if err != nil {
 		return TransferMsg{}, err
 	}
@@ -80,11 +172,7 @@ func (c *ClientCommon) encodeEnvelopePlain(env Envelope) ([]byte, error) {
 }
 
 func (c *ClientCommon) encryptTransportPayload(data []byte) ([]byte, error) {
-	encoded := c.msgEn(c.SecretKey, data)
-	if encoded == nil && len(data) != 0 {
-		return nil, errTransportPayloadEncryptFailed
-	}
-	return encoded, nil
+	return encryptTransportPayloadCodec(c.modernPSKRuntime, c.msgEn, c.SecretKey, data)
 }
 
 func (c *ClientCommon) encodeEnvelope(env Envelope) ([]byte, error) {
@@ -108,11 +196,7 @@ func (c *ClientCommon) decodeEnvelope(data []byte) (Envelope, error) {
 }
 
 func (c *ClientCommon) decryptTransportPayload(data []byte) ([]byte, error) {
-	plain := c.msgDe(c.SecretKey, data)
-	if plain == nil && len(data) != 0 {
-		return nil, errTransportPayloadDecryptFailed
-	}
-	return plain, nil
+	return decryptTransportPayloadCodec(c.modernPSKRuntime, c.msgDe, c.SecretKey, data)
 }
 
 func (c *ClientCommon) decodeEnvelopePlain(data []byte) (Envelope, error) {
@@ -167,11 +251,7 @@ func (s *ServerCommon) encryptTransportPayloadLogical(logical *LogicalConn, data
 	if msgEn == nil {
 		return nil, errTransportDetached
 	}
-	encoded := msgEn(secretKey, data)
-	if encoded == nil && len(data) != 0 {
-		return nil, errTransportPayloadEncryptFailed
-	}
-	return encoded, nil
+	return encryptTransportPayloadCodec(logical.modernPSKRuntimeSnapshot(), msgEn, secretKey, data)
 }
 
 func (s *ServerCommon) encodeEnvelopeLogical(logical *LogicalConn, env Envelope) ([]byte, error) {
@@ -210,11 +290,7 @@ func (s *ServerCommon) decryptTransportPayloadLogical(logical *LogicalConn, data
 	if msgDe == nil {
 		return nil, errTransportDetached
 	}
-	plain := msgDe(secretKey, data)
-	if plain == nil && len(data) != 0 {
-		return nil, errTransportPayloadDecryptFailed
-	}
-	return plain, nil
+	return decryptTransportPayloadCodec(logical.modernPSKRuntimeSnapshot(), msgDe, secretKey, data)
 }
 
 func (s *ServerCommon) decodeEnvelopePlain(data []byte) (Envelope, error) {

transport_conn.go

@@ -1,6 +1,7 @@
 package notify
 
 import (
+	"b612.me/stario"
 	"context"
 	"errors"
 	"net"
@@ -16,7 +17,7 @@ type TransportConn struct {
 }
 
 const (
-	transportStreamReadBufferSize = 256 * 1024
+	transportStreamReadBufferSize = 1024 * 1024
 	transportPacketReadBufferSize = 64 * 1024
 )
 
@@ -28,6 +29,17 @@ func packetReadBuffer() []byte {
 	return make([]byte, transportPacketReadBufferSize)
 }
 
+func newTransportFrameReader(conn net.Conn, queue *stario.StarQueue) *stario.FrameReader {
+	reader := stario.NewFrameReader(conn, queue)
+	if reader == nil {
+		return nil
+	}
+	if transportStreamReadBufferSize > stario.DefaultFrameReaderBufferSize {
+		reader.SetReadBufferSize(transportStreamReadBufferSize)
+	}
+	return reader
+}
+
 type TransportConnRuntimeSnapshot struct {
 	ClientID                  string
 	RemoteAddress             string

transport_write_test.go

@@ -62,18 +62,72 @@ func TestWriteFullToConnSerializesConcurrentWriters(t *testing.T) {
 	}
 }
 
+func newTestBulkBatchSender(binding *transportBinding) *bulkBatchSender {
+	return newTestBulkBatchSenderWithWriteTimeout(binding, nil)
+}
+
+func newTestBulkBatchSenderWithWriteTimeout(binding *transportBinding, writeTimeout func() time.Duration) *bulkBatchSender {
+	return newBulkBatchSender(binding, bulkBatchCodec{
+		encodeSingle: func(frame bulkFastFrame) ([]byte, func(), error) {
+			return append([]byte(nil), frame.Payload...), nil, nil
+		},
+		encodeBatch: func(frames []bulkFastFrame) ([]byte, func(), error) {
+			payload, err := encodeBulkFastBatchPlain(frames)
+			return payload, nil, err
+		},
+	}, writeTimeout)
+}
+
+func newTestStreamBatchSender(binding *transportBinding, writeTimeout func() time.Duration) *streamBatchSender {
+	return newStreamBatchSender(binding, streamBatchCodec{
+		encodeSingle: func(frame streamFastDataFrame) ([]byte, error) {
+			return encodeStreamFastFramePayload(frame)
+		},
+		encodeBatch: func(frames []streamFastDataFrame) ([]byte, error) {
+			return encodeStreamFastBatchPlain(frames)
+		},
+	}, writeTimeout)
+}
+
 func TestBulkBatchSenderRespectsWriteDeadlineWhenReceiverStalls(t *testing.T) {
 	left, right := net.Pipe()
 	defer left.Close()
 	defer right.Close()
 
-	sender := newBulkBatchSender(newTransportBinding(left, stario.NewQueue()))
-	ctx, cancel := context.WithTimeout(context.Background(), 50*time.Millisecond)
-	defer cancel()
+	sender := newTestBulkBatchSenderWithWriteTimeout(newTransportBinding(left, stario.NewQueue()), func() time.Duration {
+		return 50 * time.Millisecond
+	})
 
 	errCh := make(chan error, 1)
 	go func() {
-		errCh <- sender.submit(ctx, []byte("payload"))
+		errCh <- sender.submitData(context.Background(), 1, 1, bulkFastPathVersionV1, []byte("payload"))
+	}()
+
+	select {
+	case err := <-errCh:
+		if err == nil {
+			t.Fatal("sender.submit should fail when receiver stalls")
+		}
+		if !isTimeoutLikeError(err) {
+			t.Fatalf("sender.submit error = %v, want timeout-like error", err)
+		}
+	case <-time.After(time.Second):
+		t.Fatal("sender.submit should not hang when receiver stalls")
+	}
+}
+
+func TestStreamBatchSenderRespectsBindingWriteDeadlineWhenReceiverStalls(t *testing.T) {
+	left, right := net.Pipe()
+	defer left.Close()
+	defer right.Close()
+
+	sender := newTestStreamBatchSender(newTransportBinding(left, stario.NewQueue()), func() time.Duration {
+		return 50 * time.Millisecond
+	})
+
+	errCh := make(chan error, 1)
+	go func() {
+		errCh <- sender.submitData(context.Background(), 1, 1, streamFastPathVersionV2, []byte("payload"))
 	}()
 
 	select {
@@ -258,12 +312,12 @@ func TestWriteNetBuffersFullUnlockedUsesUnwrappedVectoredConn(t *testing.T) {
 func TestBulkBatchSenderSkipsQueuedCanceledRequest(t *testing.T) {
 	conn := newBlockingPacketWriteConn()
 	binding := newTransportBinding(conn, stario.NewQueue())
-	sender := newBulkBatchSender(binding)
+	sender := newTestBulkBatchSender(binding)
 	defer sender.stop()
 
 	firstErrCh := make(chan error, 1)
 	go func() {
-		firstErrCh <- sender.submit(context.Background(), []byte("first"))
+		firstErrCh <- sender.submitData(context.Background(), 1, 1, bulkFastPathVersionV1, []byte("first"))
 	}()
 
 	select {
@@ -275,7 +329,7 @@ func TestBulkBatchSenderSkipsQueuedCanceledRequest(t *testing.T) {
 	ctx, cancel := context.WithCancel(context.Background())
 	secondErrCh := make(chan error, 1)
 	go func() {
-		secondErrCh <- sender.submit(ctx, []byte("second"))
+		secondErrCh <- sender.submitData(ctx, 1, 2, bulkFastPathVersionV1, []byte("second"))
 	}()
 	time.Sleep(20 * time.Millisecond)
 	cancel()
@@ -306,16 +360,48 @@ func TestBulkBatchSenderSkipsQueuedCanceledRequest(t *testing.T) {
 	}
 }
 
+func TestBulkBatchSenderDoesNotDirectSubmitShareableV2Data(t *testing.T) {
+	sender := &bulkBatchSender{}
+	req := bulkBatchRequest{
+		frames: []bulkFastFrame{{
+			Type:    bulkFastPayloadTypeData,
+			DataID:  1,
+			Seq:     1,
+			Payload: make([]byte, 256*1024),
+		}},
+		fastPathVersion: bulkFastPathVersionV2,
+	}
+	if sender.shouldDirectSubmit(req) {
+		t.Fatal("shareable v2 shared bulk data should queue for super-batch instead of direct submit")
+	}
+}
+
+func TestBulkBatchSenderDirectSubmitsUnbatchableRequest(t *testing.T) {
+	sender := &bulkBatchSender{}
+	req := bulkBatchRequest{
+		frames: []bulkFastFrame{{
+			Type:    bulkFastPayloadTypeRelease,
+			DataID:  1,
+			Seq:     0,
+			Payload: []byte("rel"),
+		}},
+		fastPathVersion: bulkFastPathVersionV2,
+	}
+	if !sender.shouldDirectSubmit(req) {
+		t.Fatal("unbatchable shared bulk control request should still direct submit")
+	}
+}
+
 func TestBulkBatchSenderReturnsFlushResultAfterStartedContextCancel(t *testing.T) {
 	conn := newBlockingPacketWriteConn()
 	binding := newTransportBinding(conn, stario.NewQueue())
-	sender := newBulkBatchSender(binding)
+	sender := newTestBulkBatchSender(binding)
 	defer sender.stop()
 
 	ctx, cancel := context.WithCancel(context.Background())
 	errCh := make(chan error, 1)
 	go func() {
-		errCh <- sender.submit(ctx, []byte("payload"))
+		errCh <- sender.submitData(ctx, 1, 1, bulkFastPathVersionV1, []byte("payload"))
 	}()
 
 	select {
@@ -346,10 +432,18 @@ func TestBulkBatchSenderReturnsFlushResultAfterStartedContextCancel(t *testing.T
 
 func TestTransportBindingStopBackgroundWorkersStopsSharedSender(t *testing.T) {
 	binding := newTransportBinding(newBlockingPacketWriteConn(), stario.NewQueue())
-	sender := binding.bulkBatchSenderSnapshot()
+	sender := binding.bulkBatchSenderSnapshotWithCodec(bulkBatchCodec{
+		encodeSingle: func(frame bulkFastFrame) ([]byte, func(), error) {
+			return append([]byte(nil), frame.Payload...), nil, nil
+		},
+		encodeBatch: func(frames []bulkFastFrame) ([]byte, func(), error) {
+			payload, err := encodeBulkFastBatchPlain(frames)
+			return payload, nil, err
+		},
+	}, nil)
 	binding.stopBackgroundWorkers()
 
-	err := sender.submit(context.Background(), []byte("payload"))
+	err := sender.submitData(context.Background(), 1, 1, bulkFastPathVersionV1, []byte("payload"))
 	if !errors.Is(err, errTransportDetached) {
 		t.Fatalf("sender.submit after stop = %v, want %v", err, errTransportDetached)
 	}