fix: avoid hang on Waku node stop

Signed-off-by: kblinichkin <kirill.blinichkin@gmail.com>

waku/v2/node/wakunode2.go

@@ -527,42 +527,69 @@ func (w *WakuNode) Stop() {
 		return
 	}
 
-	w.bcaster.Stop()
+	// Cancel the node context before stopping protocols: once processLoop sees
+	// ctx.Done() it stops draining its channels, so a later send would block.
+	w.cancel()
+
+	stopWithTimeout(w.bcaster.Stop, "bcaster", 5*time.Second, w.log)
 
 	defer w.connectionNotif.Close()
 	defer w.addressChangesSub.Close()
 
 	w.host.Network().StopNotify(w.connectionNotif)
 
-	w.relay.Stop()
-	w.lightPush.Stop()
-	w.legacyStore.Stop()
-	w.filterFullNode.Stop()
-	w.filterLightNode.Stop()
+	// Each Stop() calls wg.Wait() internally and can block on a goroutine stuck
+	// writing to a dead connection; the timeout keeps Stop() bounded.
+	stopWithTimeout(w.relay.Stop, "relay", 10*time.Second, w.log)
+	stopWithTimeout(w.lightPush.Stop, "lightPush", 5*time.Second, w.log)
+	stopWithTimeout(w.legacyStore.Stop, "legacyStore", 5*time.Second, w.log)
+	stopWithTimeout(w.filterFullNode.Stop, "filterFullNode", 5*time.Second, w.log)
+	stopWithTimeout(w.filterLightNode.Stop, "filterLightNode", 5*time.Second, w.log)
 
 	if w.opts.enableDiscV5 {
-		w.discoveryV5.Stop()
+		stopWithTimeout(w.discoveryV5.Stop, "discoveryV5", 5*time.Second, w.log)
 	}
-	w.peerExchange.Stop()
-	w.rendezvous.Stop()
+	stopWithTimeout(w.peerExchange.Stop, "peerExchange", 5*time.Second, w.log)
+	stopWithTimeout(w.rendezvous.Stop, "rendezvous", 5*time.Second, w.log)
+	stopWithTimeout(w.peerConnector.Stop, "peerConnector", 5*time.Second, w.log)
+	stopWithTimeout(func() { _ = w.stopRlnRelay() }, "rlnRelay", 5*time.Second, w.log)
+	stopWithTimeout(w.timesource.Stop, "timesource", 5*time.Second, w.log)
 
-	w.peerConnector.Stop()
-
-	_ = w.stopRlnRelay()
-
-	w.timesource.Stop()
-
-	w.host.Close()
-
-	w.cancel()
-
-	w.wg.Wait()
+	// Bound the wait: goroutines select on ctx.Done() and should exit promptly.
+	waitDone := make(chan struct{})
+	go func() { w.wg.Wait(); close(waitDone) }()
+	select {
+	case <-waitDone:
+	case <-time.After(5 * time.Second):
+		w.log.Error("timed out waiting for node goroutines to stop")
+	}
 
 	close(w.enrChangeCh)
 
+	// host.Close() can block indefinitely draining dead sockets; bound it too.
+	hostClosed := make(chan struct{})
+	go func() { w.host.Close(); close(hostClosed) }()
+	select {
+	case <-hostClosed:
+	case <-time.After(5 * time.Second):
+		w.log.Warn("timed out waiting for host.Close(); proceeding")
+	}
+
 	w.cancel = nil
 }
 
+// stopWithTimeout runs fn and returns once it completes or timeout elapses,
+// logging a warning on timeout so one stuck component cannot hang Stop().
+func stopWithTimeout(fn func(), name string, timeout time.Duration, log *zap.Logger) {
+	done := make(chan struct{})
+	go func() { fn(); close(done) }()
+	select {
+	case <-done:
+	case <-time.After(timeout):
+		log.Warn("timed out stopping component", zap.String("component", name))
+	}
+}
+
 // Host returns the libp2p Host used by the WakuNode
 func (w *WakuNode) Host() host.Host {
 	return w.host