fix: 停机先停接受并等待循环再断开 MQTT

cmd/nixmsg/serve.go

@@ -43,7 +43,20 @@ func cmdServe(_ []string) error {
 	setupJSONLogger(cfg.Log)
 	ctx, stop := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM)
 	defer stop()
-	return runServe(ctx, cfg)
+	return runServe(withServeStop(ctx, stop), cfg)
+}
+
+type serveStopKey struct{}
+
+func withServeStop(ctx context.Context, stop context.CancelFunc) context.Context {
+	return context.WithValue(ctx, serveStopKey{}, stop)
+}
+
+func invokeServeStop(ctx context.Context) {
+	stop, _ := ctx.Value(serveStopKey{}).(context.CancelFunc)
+	if stop != nil {
+		stop()
+	}
 }
 
 func runServe(ctx context.Context, cfg config.Config) error {
@@ -297,27 +310,59 @@ func runServe(ctx context.Context, cfg config.Config) error {
 		}
 	}
 
-	loopCtx, loopCancel := context.WithCancel(ctx)
+	loopCtx, loopCancel := context.WithCancel(context.Background())
 	defer loopCancel()
-	go messageLoops(loopCtx, msgApp, db, hashPool, metricsReg)
+	loopsDone := make(chan struct{})
+	go func() {
+		defer close(loopsDone)
+		messageLoops(loopCtx, msgApp, db, hashPool, metricsReg)
+	}()
 
 	<-ctx.Done()
+	invokeServeStop(ctx)
+	shutdownDeadline := time.Now().Add(30 * time.Second)
+	_ = lnSrv.StopAccept()
 	loopCancel()
-	// B-08：先对 MQTT 连接发 0x8B。HTTP Shutdown 与监听器完整停机顺序见 L-03。
-	shutCtx, shutCancel := context.WithTimeout(context.Background(), 5*time.Second)
-	_ = brk.Shutdown(shutCtx)
-	shutCancel()
-	_ = lnSrv.Close()
-	drainCtx, drainCancel := context.WithTimeout(context.Background(), 10*time.Second)
-	defer drainCancel()
+	<-loopsDone
+
+	drainBudget := 10 * time.Second
+	drainStart := time.Now()
+	drainCtx, drainCancel := context.WithTimeout(context.Background(), drainBudget)
 	if drainErr := db.Queue.Drain(drainCtx); drainErr != nil && !errors.Is(drainErr, context.DeadlineExceeded) {
 		slog.Error("write queue drain", "err", drainErr)
 	}
+	drainCancel()
+
+	shutCtx, shutCancel := context.WithTimeout(context.Background(), 5*time.Second)
+	_ = brk.Shutdown(shutCtx)
+	shutCancel()
+
+	secondDrain := drainBudget - time.Since(drainStart)
+	if secondDrain < time.Second {
+		secondDrain = time.Until(shutdownDeadline)
+	}
+	if secondDrain < time.Second {
+		secondDrain = time.Second
+	}
+	drain2, drain2Cancel := context.WithTimeout(context.Background(), secondDrain)
+	if drainErr := db.Queue.Drain(drain2); drainErr != nil && !errors.Is(drainErr, context.DeadlineExceeded) {
+		slog.Error("write queue drain", "err", drainErr)
+	}
+	drain2Cancel()
+
+	waitRemain := time.Until(shutdownDeadline)
+	if waitRemain < 2*time.Second {
+		waitRemain = 2 * time.Second
+	}
+	waitCtx, waitCancel := context.WithTimeout(context.Background(), waitRemain)
+	_ = lnSrv.Wait(waitCtx)
+	waitCancel()
 	return nil
 }
 
 func messageLoops(ctx context.Context, msgApp *message.App, db *store.DB, hashPool auth.HashPool, met *metrics.Registry) {
 	msgApp.StartLoops(ctx)
+	defer msgApp.WaitLoops()
 	t := time.NewTicker(15 * time.Second)
 	defer t.Stop()
 	sample := func() {