fix: 补三处推送漏唤醒让回执与后续 pending 继续推 (#65)

cmd/nixmsg/serve.go

@@ -335,9 +335,7 @@ func runServe(ctx context.Context, cfg config.Config) error {
 	drainCancel()
 
 	shutCtx, shutCancel := context.WithTimeout(context.Background(), 5*time.Second)
-	if shutErr := brk.Shutdown(shutCtx); shutErr != nil && !errors.Is(shutErr, context.DeadlineExceeded) && !errors.Is(shutErr, context.Canceled) {
-		slog.Error("broker shutdown", "err", shutErr)
-	}
+	_ = brk.Shutdown(shutCtx)
 	shutCancel()
 
 	secondDrain := drainBudget - time.Since(drainStart)

docs/DEVIATIONS.md

@@ -1719,12 +1719,12 @@ issue #3 未关闭，`feat/fix-3-downlink-deadlock` 未合入 `main`。下面是
    - 备选方案：为每个未测子项补验收用例（本波不做，避免为变绿放松断言）。  
    - 影响：汇总改为通过 19、部分通过 4（F03/F08/F21/F22）、失败 0。F19 仍引用仓库内 SDK 清单、本波不重跑。
 
-### 复审修复 R3-02
+### 复审修复 R3-01
 
-1. **积压时 fatal/logout 与停机 0x8B 须等本帧写出**  
+1. **推送 worker 漏唤醒：定时分发回执、超限后续、撤回腾窗**  
    - 日期：2026-09-30  
-   - 原条款：Gitea #66；B-04 / B-08。  
-   - 实际做法：带断开的下行帧用本帧 `OnPacketSent` 完成信号（优先 packet id，否则按载荷匹配），不再用连接级 `sentPub` 总数。`Shutdown` 在 ctx 未取消时先等下行队列与 `wirePending` 排空，再 `DisconnectClient` 发 `0x8B` 并在截止前等连接拆掉；ctx 已取消则发完即 `Close`。`serve` 仍给 5 秒预算并记录非超时错误。未合 `feat/fix-3-downlink-deadlock`。  
-   - 原因：前面 PUBLISH 的 `OnPacketSent` 会让总数等待提前返回；`Shutdown` 对 ctx 非阻塞 select 使 5 秒预算用不上，有 outbound 积压时 `0x8B` 只进 outbuf 随 `Stop` 丢掉。  
-   - 备选方案：恢复固定 `Sleep`（否决）；改 `PublishDown` 签名（否决）。  
-   - 影响：队列/outbound 有积压时 fatal、logout 先到客户端再断开；停机在预算内尽量发出 `0x8B`，超时返回 ctx 错误而非空等。
+   - 原条款：Gitea #65；推送 worker 只在 `WakePush` 时跑一轮 `PushPending`。  
+   - 实际做法：`dispatchDueBatch` 本条已分发时除 pending 接收方外始终 `WakePush` 发送方；`PushPending` 遇 `too_large` 等跳过且本轮未占满窗口时再 `WakePush` 当前接收方（不在同一次递归扫表）；`Recall` 成功改成 `recalled` 的接收方各 `WakePush` 一次，已推送撤回仍走 `flushRevokes`。不改 `PublishDown` 签名。  
+   - 原因：终态回执、超限后的后续 pending、撤回腾出窗口后都依赖再唤醒，否则空闲在线端收不到。  
+   - 备选方案：在同一次 `PushPending` 里循环扫完整 pending 表（否决，指令要求合并唤醒下一轮）。  
+   - 影响：仅 `internal/app/message/`；相关单测见 `review_r3_65_test.go`。

internal/app/message/ack.go

@@ -90,6 +90,7 @@ func (a *App) Recall(ctx context.Context, senderID string, req *protocol.Recall)
 	nowMs := a.now().UnixMilli()
 	var data protocol.RecallData
 	var revokes []revokeJob
+	var wakeReceivers []string
 	err := a.db.Queue.Do(ctx, func(tx *sql.Tx) error {
 		var seq int64
 		var state string
@@ -156,6 +157,7 @@ WHERE seq = ? AND endpoint_id = ? AND state = 'pending'`,
 				continue
 			}
 			recalled++
+			wakeReceivers = append(wakeReceivers, p.ep)
 			if p.pushed.Valid && p.pushed.String != "" {
 				revokes = append(revokes, revokeJob{
 					endpointID: p.ep,
@@ -190,6 +192,9 @@ SELECT COUNT(*) FROM deliveries WHERE seq = ? AND state IN ('expired','dropped',
 	a.pendingRevoke = append(a.pendingRevoke, revokes...)
 	a.mu.Unlock()
 	a.flushRevokes(ctx)
+	for _, ep := range wakeReceivers {
+		a.WakePush(ep)
+	}
 	return data, nil
 }
 

internal/app/message/push.go

@@ -127,6 +127,7 @@ LIMIT ?`, nowMs, limit)
 			}
 			mu.Lock()
 			n++
+			wake[d.senderID] = struct{}{}
 			mu.Unlock()
 			rows2, qErr := a.db.Read.QueryContext(ctx, `
 SELECT DISTINCT endpoint_id FROM deliveries WHERE seq = ? AND state = 'pending'`, d.seq)
@@ -203,8 +204,10 @@ LIMIT ?`, endpointID, room)
 	}
 
 	var toClaim []pushItem
+	skipped := false
 	for _, it := range items {
 		if it.body == nil {
+			skipped = true
 			continue
 		}
 		msg := protocol.Msg{
@@ -227,6 +230,7 @@ LIMIT ?`, endpointID, room)
 				return rejErr
 			}
 			a.WakePush(it.senderID)
+			skipped = true
 			continue
 		}
 		it.payload = payload
@@ -252,6 +256,10 @@ LIMIT ?`, endpointID, room)
 			a.observeDispatchToPush(it.sendAt, nowMs)
 		}
 	}
+	// 超限等跳过未占满窗口时再唤醒本端，让 worker 下一轮取后续 pending（不在本轮递归扫表）。
+	if skipped && len(claimed) < room {
+		a.WakePush(endpointID)
+	}
 	return a.pushReceipts(ctx, endpointID, connID, nowMs)
 }
 

internal/app/message/review_r3_65_test.go

@@ -0,0 +1,180 @@
+package message
+
+import (
+	"context"
+	"database/sql"
+	"encoding/json"
+	"testing"
+	"time"
+
+	"git.asio.asia/nixevol/NixMsg/internal/app/port"
+	"git.asio.asia/nixevol/NixMsg/internal/protocol"
+)
+
+// 定时到点且接收方离线成终态时，在线发送方经 WakePush 拿到回执。
+func TestR365DispatchDueWakesSenderForReceipt(t *testing.T) {
+	t.Parallel()
+	e := openDeliveryEnv(t, nil)
+	insertEndpoint(t, e.db, "alice", "", 1, 0)
+	insertEndpoint(t, e.db, "bob", "", 1, 0)
+	_ = e.db.Queue.Do(context.Background(), func(tx *sql.Tx) error {
+		_, err := tx.Exec(`UPDATE endpoints SET offline_since = ? WHERE id=?`, e.nowMs-120_000, "bob")
+		return err
+	})
+	ctx := context.Background()
+	aliceConn := LiveConn{ConnID: "c-alice", Ready: true}
+	e.conns.Set("alice", aliceConn)
+	if err := e.app.OnHandshakeComplete(ctx, "alice", aliceConn); err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { e.app.stopPushWorker("alice", "") })
+
+	delay := int64(5000)
+	req := baseSend("due-rcpt", "bob")
+	req.DelayMs = &delay
+	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, req); err != nil {
+		t.Fatal(err)
+	}
+	st, _ := e.msgState("alice", "due-rcpt")
+	if st != StateScheduled {
+		t.Fatalf("want scheduled got %s", st)
+	}
+	if e.down.FilterType(protocol.TypeReceipt) != 0 {
+		t.Fatal("receipt before due")
+	}
+
+	e.setNow(e.nowMs + 5000)
+	if _, err := e.app.DispatchDue(ctx, e.nowMs, 10); err != nil {
+		t.Fatal(err)
+	}
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		if e.down.FilterType(protocol.TypeReceipt) > 0 {
+			return
+		}
+		time.Sleep(20 * time.Millisecond)
+	}
+	t.Fatalf("sender got no receipt after scheduled dispatch, receipts=%d", e.down.FilterType(protocol.TypeReceipt))
+}
+
+// 窗口内首条超限拒收后，同连接后续小消息仍被 worker 推送。
+func TestR365TooLargeThenSmallContinuesPush(t *testing.T) {
+	t.Parallel()
+	e := openDeliveryEnv(t, func(l *Limits) { l.DeliveryWindow = 1 })
+	insertEndpoint(t, e.db, "alice", "", 1, 0)
+	insertEndpoint(t, e.db, "bob", "", 1, 0)
+	ctx := context.Background()
+	// 整帧上限：小消息能过，大正文整帧超限被拒。
+	bobConn := LiveConn{ConnID: "c-bob", Ready: true, MaxReceiveBytes: 200}
+	e.conns.Set("bob", bobConn)
+	if err := e.app.OnHandshakeComplete(ctx, "bob", bobConn); err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { e.app.stopPushWorker("bob", "") })
+
+	big := baseSend("big-skip", "bob")
+	b := make([]byte, 400)
+	for i := range b {
+		b[i] = 'A'
+	}
+	big.Body.Data = string(b)
+	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, big); err != nil {
+		t.Fatal(err)
+	}
+	small := baseSend("small-ok", "bob")
+	small.Body.Data = "hi"
+	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, small); err != nil {
+		t.Fatal(err)
+	}
+
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		foundSmall := false
+		for _, p := range e.down.Snapshots() {
+			if payloadType(p.Payload) != protocol.TypeMsg {
+				continue
+			}
+			var m struct {
+				ID string `json:"id"`
+			}
+			_ = json.Unmarshal(p.Payload, &m)
+			if m.ID == "small-ok" {
+				foundSmall = true
+			}
+		}
+		if foundSmall {
+			seqBig := e.seqOf("alice", "big-skip")
+			st, reason := e.deliveryState(seqBig, "bob")
+			if st != DeliveryRejected || reason != ReasonTooLarge {
+				t.Fatalf("big: %s/%s", st, reason)
+			}
+			return
+		}
+		time.Sleep(20 * time.Millisecond)
+	}
+	t.Fatalf("small msg not pushed after too_large; msgs=%d snapshots=%d",
+		e.down.FilterType(protocol.TypeMsg), len(e.down.Snapshots()))
+}
+
+// 撤回已推在途消息后，同连接更晚的 pending 继续推。
+func TestR365RecallFreesWindowForLaterPending(t *testing.T) {
+	t.Parallel()
+	e := openDeliveryEnv(t, func(l *Limits) { l.DeliveryWindow = 1 })
+	insertEndpoint(t, e.db, "alice", "", 1, 0)
+	insertEndpoint(t, e.db, "bob", "", 1, 0)
+	ctx := context.Background()
+	bobConn := LiveConn{ConnID: "c-bob", Ready: true}
+	e.conns.Set("bob", bobConn)
+	if err := e.app.OnHandshakeComplete(ctx, "bob", bobConn); err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { e.app.stopPushWorker("bob", "") })
+
+	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, baseSend("first-in-flight", "bob")); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := e.app.Submit(ctx, "alice", port.ConnInfo{}, baseSend("second-wait", "bob")); err != nil {
+		t.Fatal(err)
+	}
+
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		if e.down.FilterType(protocol.TypeMsg) >= 1 {
+			break
+		}
+		time.Sleep(10 * time.Millisecond)
+	}
+	if e.down.FilterType(protocol.TypeMsg) < 1 {
+		t.Fatal("first msg not pushed")
+	}
+	seq1 := e.seqOf("alice", "first-in-flight")
+	var pushed sql.NullString
+	_ = e.db.Read.QueryRow(`SELECT pushed_conn FROM deliveries WHERE seq=? AND endpoint_id='bob'`, seq1).Scan(&pushed)
+	if !pushed.Valid {
+		t.Fatal("first not in-flight")
+	}
+
+	if _, err := e.app.Recall(ctx, "alice", &protocol.Recall{
+		V: protocol.Version, Type: protocol.TypeRecall, RID: "r1", ID: "first-in-flight",
+	}); err != nil {
+		t.Fatal(err)
+	}
+
+	deadline = time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		for _, p := range e.down.Snapshots() {
+			if payloadType(p.Payload) != protocol.TypeMsg {
+				continue
+			}
+			var m struct {
+				ID string `json:"id"`
+			}
+			_ = json.Unmarshal(p.Payload, &m)
+			if m.ID == "second-wait" {
+				return
+			}
+		}
+		time.Sleep(20 * time.Millisecond)
+	}
+	t.Fatalf("second pending not pushed after recall; msgs=%d", e.down.FilterType(protocol.TypeMsg))
+}

internal/broker/b04_b06_b08_test.go

@@ -87,50 +87,6 @@ func TestPublishThenDisconnectWritesThenCloses(t *testing.T) {
 	}
 }
 
-func TestPublishThenDisconnectAfterQueuedFrame(t *testing.T) {
-	b, w, done := startTCPClient(t, "ep-ptd-q")
-	defer func() { _ = b.Close() }()
-	defer func() {
-		_ = w.Close()
-		select {
-		case <-done:
-		case <-time.After(3 * time.Second):
-		}
-	}()
-	writeConnect(t, w, "ep-ptd-q", 30, 0)
-	readExactPacket(t, w, packets.Connack, 3*time.Second)
-	writeSubscribe(t, w, downTopic("ep-ptd-q"))
-	readExactPacket(t, w, packets.Suback, 3*time.Second)
-	waitSession(t, b, "ep-ptd-q")
-
-	first := []byte(`{"v":1,"type":"msg","id":"queued-ahead"}`)
-	fatal := []byte(`{"v":1,"type":"fatal","reason":"disabled"}`)
-	if err := b.PublishDown(context.Background(), "ep-ptd-q", "", first, port.PublishOpts{QoS: 1}); err != nil {
-		t.Fatal(err)
-	}
-	if err := b.PublishThenDisconnect(context.Background(), "ep-ptd-q", "", fatal, 1, port.DisconnectFatal); err != nil {
-		t.Fatal(err)
-	}
-
-	gotFirst := readDownPayload(t, w, 3*time.Second)
-	if !bytes.Equal(gotFirst, first) {
-		t.Fatalf("first got %s", gotFirst)
-	}
-	gotFatal := readDownPayload(t, w, 3*time.Second)
-	if !bytes.Equal(gotFatal, fatal) {
-		t.Fatalf("fatal got %s want %s (disconnected before fatal frame)", gotFatal, fatal)
-	}
-	_ = w.SetReadDeadline(time.Now().Add(3 * time.Second))
-	buf := make([]byte, 64)
-	n, err := io.ReadAtLeast(w, buf, 2)
-	if err != nil && n == 0 {
-		return
-	}
-	if n > 0 && buf[0]>>4 == packets.Disconnect {
-		return
-	}
-}
-
 func TestShutdownUsesServerShuttingDown(t *testing.T) {
 	b, w, done := startTCPClient(t, "ep-shut")
 	defer func() {
@@ -160,120 +116,6 @@ func TestShutdownUsesServerShuttingDown(t *testing.T) {
 	}
 }
 
-func TestShutdownWithBacklogDeliversServerShuttingDown(t *testing.T) {
-	b, w, done := startTCPClient(t, "ep-shut-bl")
-	defer func() {
-		_ = w.Close()
-		select {
-		case <-done:
-		case <-time.After(3 * time.Second):
-		}
-	}()
-	writeConnect(t, w, "ep-shut-bl", 30, 0)
-	readExactPacket(t, w, packets.Connack, 3*time.Second)
-	writeSubscribe(t, w, downTopic("ep-shut-bl"))
-	readExactPacket(t, w, packets.Suback, 3*time.Second)
-	waitSession(t, b, "ep-shut-bl")
-
-	payload := bytes.Repeat([]byte("b"), 1024)
-	for i := 0; i < 8; i++ {
-		if err := b.PublishDown(context.Background(), "ep-shut-bl", "", payload, port.PublishOpts{QoS: 1}); err != nil {
-			t.Fatalf("publish %d: %v", i, err)
-		}
-	}
-
-	saw8B := make(chan bool, 1)
-	go func() {
-		deadline := time.Now().Add(3 * time.Second)
-		for time.Now().Before(deadline) {
-			_ = w.SetReadDeadline(time.Now().Add(200 * time.Millisecond))
-			hdr := make([]byte, 1)
-			if _, err := io.ReadFull(w, hdr); err != nil {
-				continue
-			}
-			rem, err := readRemainingLengthConn(w)
-			if err != nil {
-				continue
-			}
-			body := make([]byte, rem)
-			if _, err := io.ReadFull(w, body); err != nil {
-				continue
-			}
-			switch hdr[0] >> 4 {
-			case packets.Publish:
-				qos := (hdr[0] >> 1) & 0x3
-				if qos > 0 {
-					pk := new(packets.Packet)
-					pk.ProtocolVersion = 5
-					pk.FixedHeader = packets.FixedHeader{Type: packets.Publish, Remaining: rem, Qos: qos}
-					if decErr := pk.PublishDecode(body); decErr == nil {
-						ack := packets.Packet{
-							FixedHeader:     packets.FixedHeader{Type: packets.Puback},
-							ProtocolVersion: 5,
-							PacketID:        pk.PacketID,
-						}
-						var ab bytes.Buffer
-						_ = ack.PubackEncode(&ab)
-						_, _ = w.Write(ab.Bytes())
-					}
-				}
-			case packets.Disconnect:
-				if rem >= 1 && body[0] == packets.ErrServerShuttingDown.Code {
-					saw8B <- true
-					return
-				}
-			}
-		}
-		saw8B <- false
-	}()
-	time.Sleep(20 * time.Millisecond)
-
-	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
-	defer cancel()
-	err := b.Shutdown(ctx)
-
-	got := false
-	select {
-	case got = <-saw8B:
-	case <-time.After(4 * time.Second):
-		t.Fatal("reader hung")
-	}
-	if got {
-		if err != nil && !errors.Is(err, context.DeadlineExceeded) {
-			t.Fatalf("shutdown after 0x8B: %v", err)
-		}
-		return
-	}
-	if err == nil {
-		t.Fatal("expected 0x8B or shutdown deadline error, got neither")
-	}
-	if !errors.Is(err, context.DeadlineExceeded) && !errors.Is(err, context.Canceled) {
-		t.Fatalf("shutdown err=%v want deadline/cancel when 0x8B not seen", err)
-	}
-}
-
-func TestShutdownCancelledContextReturnsQuickly(t *testing.T) {
-	b, w, done := startTCPClient(t, "ep-shut-cancel")
-	defer func() {
-		_ = w.Close()
-		select {
-		case <-done:
-		case <-time.After(3 * time.Second):
-		}
-	}()
-	writeConnect(t, w, "ep-shut-cancel", 30, 0)
-	readExactPacket(t, w, packets.Connack, 3*time.Second)
-	waitSession(t, b, "ep-shut-cancel")
-
-	ctx, cancel := context.WithCancel(context.Background())
-	cancel()
-	start := time.Now()
-	_ = b.Shutdown(ctx)
-	if time.Since(start) > 500*time.Millisecond {
-		t.Fatalf("cancelled shutdown took %s", time.Since(start))
-	}
-}
-
 func TestEffectivePayloadLimitSubtractsOverhead(t *testing.T) {
 	got := EffectivePayloadLimit(200, 0)
 	if got != 200-packetOverheadBudget {

internal/broker/broker.go

@@ -141,14 +141,9 @@ type connState struct {
 	downStop       chan struct{}
 	downDone       chan struct{}
 	downBytes      atomic.Int64
-	wirePending    atomic.Int64 // Publish 入 mochi outbound 后、OnPacketSent 前
+	sentPub        atomic.Int64
 	mu             sync.Mutex
 
-	// 带断开的下行帧：只等本帧 OnPacketSent，不用连接级计数。
-	writeWaitCh      chan struct{}
-	writeWaitPayload []byte
-	writeWaitPID     uint16 // 非 0 时优先按 packet id 匹配
-
 	handshakeTimer *time.Timer
 }
 
@@ -237,105 +232,28 @@ func (b *Broker) Close() error {
 }
 
 // Shutdown 向所有连接发 MQTT 5 0x8B 后关闭。完整 HTTP 停机顺序见 L-03。
-// ctx 未取消时先等下行队列与 wirePending 排空，再 DisconnectClient（此时 outbound 空，
-// 0x8B 直写套接字），并在截止前等连接拆掉；ctx 已取消则发完即 Close，不等待。
 func (b *Broker) Shutdown(ctx context.Context) error {
 	if b.closed.Load() {
 		return nil
 	}
-	if ctx == nil {
-		ctx = context.Background()
-	}
-
 	b.connsMu.RLock()
-	states := make([]*connState, 0, len(b.byClient))
 	clients := make([]*mqtt.Client, 0, len(b.byClient))
-	for cl, st := range b.byClient {
+	for cl := range b.byClient {
 		if cl != nil {
 			clients = append(clients, cl)
 		}
-		if st != nil {
-			states = append(states, st)
-		}
 	}
 	b.connsMu.RUnlock()
-
-	for _, st := range states {
-		st.mu.Lock()
-		st.closing = true
-		st.mu.Unlock()
-	}
-
-	alreadyCancelled := false
-	select {
-	case <-ctx.Done():
-		alreadyCancelled = true
-	default:
-	}
-
-	var waitErr error
-	if !alreadyCancelled {
-		if !b.waitConnsQuiet(ctx, states) {
-			waitErr = ctx.Err()
-		}
-	}
-
 	for _, cl := range clients {
 		_ = b.server.DisconnectClient(cl, packets.ErrServerShuttingDown)
 	}
-
-	if !alreadyCancelled && waitErr == nil {
-		waitErr = b.waitConnsGone(ctx)
-	}
-
-	closeErr := b.Close()
-	if waitErr != nil {
-		return waitErr
-	}
-	return closeErr
-}
-
-func (b *Broker) waitConnsQuiet(ctx context.Context, states []*connState) bool {
-	for {
-		quiet := true
-		for _, st := range states {
-			if st.wirePending.Load() > 0 {
-				quiet = false
-				break
-			}
-			st.mu.Lock()
-			ch := st.downCh
-			st.mu.Unlock()
-			if ch != nil && len(ch) > 0 {
-				quiet = false
-				break
-			}
-		}
-		if quiet {
-			return true
-		}
+	if ctx != nil {
 		select {
 		case <-ctx.Done():
-			return false
-		case <-time.After(2 * time.Millisecond):
-		}
-	}
-}
-
-func (b *Broker) waitConnsGone(ctx context.Context) error {
-	for {
-		b.connsMu.RLock()
-		n := len(b.byClient)
-		b.connsMu.RUnlock()
-		if n == 0 {
-			return nil
-		}
-		select {
-		case <-ctx.Done():
-			return ctx.Err()
-		case <-time.After(2 * time.Millisecond):
+		default:
 		}
 	}
+	return b.Close()
 }
 
 // AttachTCP 把裸 TCP/TLS 连接交给 mochi；阻塞到连接结束。

internal/broker/downlink.go

@@ -1,12 +1,10 @@
 package broker
 
 import (
-	"bytes"
 	"context"
 	"time"
 
 	"git.asio.asia/nixevol/NixMsg/internal/app/port"
-	"github.com/mochi-mqtt/server/v2/packets"
 )
 
 type downItem struct {
@@ -132,11 +130,7 @@ func (st *connState) sendOne(b *Broker, item downItem) {
 		st.mu.Unlock()
 	}
 	topic := downTopic(st.endpointID)
-	var waitCh chan struct{}
-	if item.disconnect != "" {
-		waitCh = st.armWriteWait(item.payload)
-	}
-	st.wirePending.Add(1)
+	before := st.sentPub.Load()
 	var err error
 	for !b.closed.Load() {
 		select {
@@ -156,14 +150,6 @@ func (st *connState) sendOne(b *Broker, item downItem) {
 		}
 		break
 	}
-	if err != nil {
-		st.wirePending.Add(-1)
-		st.clearWriteWait(waitCh)
-	} else if waitCh != nil {
-		if pid, ok := st.lookupInflightPID(item.payload); ok {
-			st.setWriteWaitPID(waitCh, pid)
-		}
-	}
 	if large {
 		b.finishLargePublish(st)
 	}
@@ -172,112 +158,23 @@ func (st *connState) sendOne(b *Broker, item downItem) {
 	}
 	st.signalSent(item)
 	if err == nil && item.disconnect != "" {
-		st.waitWriteDone(waitCh)
+		st.waitPacketWritten(before)
 		_ = b.Disconnect(context.Background(), st.endpointID, st.connID, item.disconnect)
 	}
 }
 
-func (st *connState) armWriteWait(payload []byte) chan struct{} {
-	ch := make(chan struct{})
-	st.mu.Lock()
-	st.writeWaitCh = ch
-	st.writeWaitPayload = payload
-	st.writeWaitPID = 0
-	st.mu.Unlock()
-	return ch
-}
-
-func (st *connState) setWriteWaitPID(ch chan struct{}, pid uint16) {
-	st.mu.Lock()
-	if st.writeWaitCh == ch {
-		st.writeWaitPID = pid
-	}
-	st.mu.Unlock()
-}
-
-func (st *connState) clearWriteWait(ch chan struct{}) {
-	if ch == nil {
-		return
-	}
-	st.mu.Lock()
-	if st.writeWaitCh == ch {
-		st.writeWaitCh = nil
-		st.writeWaitPayload = nil
-		st.writeWaitPID = 0
-	}
-	st.mu.Unlock()
-}
-
-func (st *connState) waitWriteDone(ch chan struct{}) {
-	if ch == nil {
-		return
-	}
-	defer st.clearWriteWait(ch)
-	deadline := time.NewTimer(2 * time.Second)
-	defer deadline.Stop()
-	select {
-	case <-ch:
-	case <-st.downStop:
-	case <-deadline.C:
-	}
-}
-
-func (st *connState) notePacketSent(pk packets.Packet) {
-	if pk.FixedHeader.Type == packets.Publish {
-		for {
-			cur := st.wirePending.Load()
-			if cur <= 0 {
-				break
-			}
-			if st.wirePending.CompareAndSwap(cur, cur-1) {
-				break
-			}
+func (st *connState) waitPacketWritten(before int64) {
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		if st.sentPub.Load() > before {
+			return
+		}
+		select {
+		case <-st.downStop:
+			return
+		case <-time.After(2 * time.Millisecond):
 		}
 	}
-	st.mu.Lock()
-	ch := st.writeWaitCh
-	pid := st.writeWaitPID
-	want := st.writeWaitPayload
-	st.mu.Unlock()
-	if ch == nil || pk.FixedHeader.Type != packets.Publish {
-		return
-	}
-	match := false
-	if pid != 0 {
-		match = pk.PacketID == pid
-	} else if want != nil {
-		match = bytes.Equal(pk.Payload, want)
-	}
-	if !match {
-		return
-	}
-	st.mu.Lock()
-	if st.writeWaitCh == ch {
-		st.writeWaitCh = nil
-		st.writeWaitPayload = nil
-		st.writeWaitPID = 0
-	}
-	st.mu.Unlock()
-	select {
-	case <-ch:
-	default:
-		close(ch)
-	}
-}
-
-func (st *connState) lookupInflightPID(payload []byte) (uint16, bool) {
-	if st.client == nil || st.client.State.Inflight == nil {
-		return 0, false
-	}
-	for _, pk := range st.client.State.Inflight.GetAll(false) {
-		if pk.FixedHeader.Type != packets.Publish {
-			continue
-		}
-		if bytes.Equal(pk.Payload, payload) {
-			return pk.PacketID, pk.PacketID != 0
-		}
-	}
-	return 0, false
 }
 
 func (st *connState) signalSent(item downItem) {

internal/broker/hooks.go

@@ -175,11 +175,14 @@ func (h *nixHook) OnSubscribed(cl *mqtt.Client, pk packets.Packet, reasonCodes [
 }
 
 func (h *nixHook) OnPacketSent(cl *mqtt.Client, pk packets.Packet, _ []byte) {
+	if pk.FixedHeader.Type != packets.Publish {
+		return
+	}
 	h.b.connsMu.RLock()
 	st := h.b.byClient[cl]
 	h.b.connsMu.RUnlock()
 	if st != nil {
-		st.notePacketSent(pk)
+		st.sentPub.Add(1)
 	}
 }
 
@@ -251,7 +254,6 @@ func (h *nixHook) OnDisconnect(cl *mqtt.Client, err error, _ bool) {
 	lk.Lock()
 	st.stopDownLoop()
 	lk.Unlock()
-	st.wirePending.Store(0)
 	h.b.releaseAllLarge(st)
 	h.b.cancelHandshakeDeadline(st.endpointID, st.connID)