fix(notify): 根治低带宽控制面阻塞与传输写入卡死

- 为控制消息增加优先级、公平调度、队列字节预算和自适应批处理
- 支持可取消的写门等待，收紧 shared/dedicated bulk、stream 和 Reply 写入边界
- 修复 bulk reset/close、连接 handoff 和安全 profile 切换时序
- 保留旧取消与超时错误契约，新增阶段化 TransportSendError
- 增加 ReplyCtx、ReplyObjCtx、写超时配置及黑洞连接和竞态回归测试

bulk.go

@@ -11,6 +11,10 @@ import (
 	"time"
 )
 
+// A zero public WriteTimeout preserves the legacy option contract, but a
+// physical bulk write still needs a finite bound so a black-hole peer cannot
+// hold the async worker (and CloseWrite) forever. The default is constant so
+// concurrent bulk operations cannot observe a process-wide timeout change.
 const (
 	BulkOpenSignalKey    = "notify.bulk.open"
 	BulkCloseSignalKey   = "notify.bulk.close"
@@ -26,6 +30,8 @@ const (
 	defaultBulkControlReadTimeout  = 0
 	defaultBulkControlWriteTimeout = 0
 	defaultBulkAcceptReadyTimeout  = 10 * time.Second
+	defaultBulkResetNotifyTimeout  = 30 * time.Second
+	defaultBulkDataWriteTimeout    = 2 * time.Minute
 )
 
 type BulkMetadata map[string]string
@@ -1278,9 +1284,13 @@ func (b *bulkHandle) close(full bool) error {
 			return nil
 		}
 		closeFn := b.closeFn
+		writeTimeout := b.writeTimeout
 		b.mu.Unlock()
 		if closeFn != nil && !b.dedicatedWriteHalfClosedSnapshot() {
-			if err := closeFn(context.Background(), b, true); err != nil && !errors.Is(err, errBulkNotFound) && !b.canIgnoreDedicatedCloseSendError(err) {
+			closeCtx, cancel := b.closeContext(writeTimeout)
+			err := closeFn(closeCtx, b, true)
+			cancel()
+			if err != nil && !errors.Is(err, errBulkNotFound) && !b.canIgnoreDedicatedCloseSendError(err) {
 				return err
 			}
 		}
@@ -1302,12 +1312,19 @@ func (b *bulkHandle) close(full bool) error {
 		return nil
 	}
 	closeFn := b.closeFn
+	writeTimeout := b.writeTimeout
 	b.mu.Unlock()
-	if err := b.waitPendingAsyncWrites(context.Background()); err != nil {
+	drainCtx, drainCancel := b.closeContext(writeTimeout)
+	err := b.waitPendingAsyncWrites(drainCtx)
+	drainCancel()
+	if err != nil {
 		return err
 	}
 	if closeFn != nil {
-		if err := closeFn(context.Background(), b, full); err != nil && !errors.Is(err, errBulkNotFound) && !b.canIgnoreDedicatedCloseSendError(err) {
+		closeCtx, cancel := b.closeContext(writeTimeout)
+		err := closeFn(closeCtx, b, full)
+		cancel()
+		if err != nil && !errors.Is(err, errBulkNotFound) && !b.canIgnoreDedicatedCloseSendError(err) {
 			return err
 		}
 	}
@@ -1348,13 +1365,29 @@ func (b *bulkHandle) Reset(err error) error {
 	}
 	resetFn := b.resetFn
 	b.mu.Unlock()
-	if resetFn != nil {
-		if sendErr := resetFn(context.Background(), b, bulkResetMessage(resetErr)); sendErr != nil {
-			return sendErr
-		}
+	if !b.applyResetState(resetErr) {
+		return b.resetErrSnapshot()
+	}
+	defer b.finalize()
+	if resetFn == nil {
+		return nil
+	}
+	timeout := defaultBulkResetNotifyTimeout
+	if b.writeTimeout > 0 && b.writeTimeout < timeout {
+		timeout = b.writeTimeout
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), timeout)
+	defer cancel()
+	done := make(chan error, 1)
+	go func() {
+		done <- resetFn(ctx, b, bulkResetMessage(resetErr))
+	}()
+	select {
+	case sendErr := <-done:
+		return sendErr
+	case <-ctx.Done():
+		return ctx.Err()
 	}
-	b.markReset(resetErr)
-	return nil
 }
 
 func (b *bulkHandle) Snapshot() BulkSnapshot {
@@ -1396,18 +1429,34 @@ func (b *bulkHandle) markReset(err error) {
 	if b == nil {
 		return
 	}
-	resetErr := bulkResetError(err)
-	b.mu.Lock()
-	if b.resetErr == nil {
-		b.resetErr = resetErr
-		b.clearBufferedDataLocked()
-		b.closeWriteStateLocked()
+	b.applyResetState(bulkResetError(err))
+	b.finalize()
+}
+
+func (b *bulkHandle) applyResetState(resetErr error) bool {
+	if b == nil {
+		return false
 	}
+	b.mu.Lock()
+	if b.resetErr != nil {
+		b.notifyFlowLocked()
+		b.mu.Unlock()
+		return false
+	}
+	b.resetErr = bulkResetError(resetErr)
+	b.clearBufferedDataLocked()
+	b.closeWriteStateLocked()
 	b.notifyFlowLocked()
 	b.mu.Unlock()
-	b.markAcceptReady(resetErr)
+	b.markAcceptReady(b.resetErrSnapshot())
 	b.notifyReadable()
-	b.finalize()
+	if b.cancel != nil {
+		b.cancel()
+	}
+	if b.writeCtxCancel != nil {
+		b.writeCtxCancel()
+	}
+	return true
 }
 
 func (b *bulkHandle) pushChunk(chunk []byte) error {
@@ -1818,7 +1867,12 @@ func (b *bulkHandle) finalize() {
 	if b.writeCtxCancel != nil {
 		b.writeCtxCancel()
 	}
-	if sender := b.clearDedicatedSender(); sender != nil {
+	sender := b.clearDedicatedSender()
+	conn, owned := b.clearDedicatedConn()
+	if conn != nil && owned {
+		_ = conn.Close()
+	}
+	if sender != nil {
 		sender.stop()
 	}
 	if b.client != nil && b.releaseDedicatedActiveReserved() {
@@ -1827,9 +1881,6 @@ func (b *bulkHandle) finalize() {
 	if b.client != nil {
 		b.client.releaseBulkDedicatedLane(b.dedicatedLaneIDSnapshot())
 	}
-	if conn, owned := b.clearDedicatedConn(); conn != nil && owned {
-		_ = conn.Close()
-	}
 	if b.runtime != nil {
 		b.runtime.remove(b.runtimeScope, b.id)
 	}
@@ -2296,6 +2347,33 @@ func bulkWriteContext(parent context.Context, timeout time.Duration) (context.Co
 	return ctx, cancel, nil
 }
 
+func bulkCloseContext(parent context.Context, timeout time.Duration) (context.Context, context.CancelFunc) {
+	if parent == nil {
+		parent = context.Background()
+	}
+	if timeout <= 0 {
+		return context.WithTimeout(parent, defaultBulkDataWriteTimeout)
+	}
+	return context.WithTimeout(parent, timeout)
+}
+
+func (b *bulkHandle) closeContext(timeout time.Duration) (context.Context, context.CancelFunc) {
+	if b == nil {
+		return bulkCloseContext(nil, timeout)
+	}
+	parent := b.Context()
+	b.mu.Lock()
+	gracefulPeerClose := b.remoteClosed || b.peerReadClosed
+	b.mu.Unlock()
+	if gracefulPeerClose {
+		// A peer half/full close cancels the bulk context as part of normal EOF
+		// delivery. Keep the final close notification alive so the peer can
+		// complete the protocol handshake, while retaining its write timeout.
+		parent = context.Background()
+	}
+	return bulkCloseContext(parent, timeout)
+}
+
 func normalizeBulkOpenRequest(req BulkOpenRequest) BulkOpenRequest {
 	req.Range = normalizeBulkRange(req.Range)
 	req.Metadata = cloneBulkMetadata(req.Metadata)