Merge pull request #9424 from yyforyongyu/fix-gossip-ann

multi: fix inconsistent state in gossip syncer

discovery/gossiper.go

@@ -813,6 +813,8 @@ func (d *AuthenticatedGossiper) stop() {
 func (d *AuthenticatedGossiper) ProcessRemoteAnnouncement(msg lnwire.Message,
 	peer lnpeer.Peer) chan error {
 
+	log.Debugf("Processing remote msg %T from peer=%x", msg, peer.PubKey())
+
 	errChan := make(chan error, 1)
 
 	// For messages in the known set of channel series queries, we'll
@@ -835,9 +837,13 @@ func (d *AuthenticatedGossiper) ProcessRemoteAnnouncement(msg lnwire.Message,
 
 		// If we've found the message target, then we'll dispatch the
 		// message directly to it.
-		syncer.ProcessQueryMsg(m, peer.QuitSignal())
+		err := syncer.ProcessQueryMsg(m, peer.QuitSignal())
+		if err != nil {
+			log.Errorf("Process query msg from peer %x got %v",
+				peer.PubKey(), err)
+		}
 
-		errChan <- nil
+		errChan <- err
 		return errChan
 
 	// If a peer is updating its current update horizon, then we'll dispatch
@@ -2382,7 +2388,8 @@ func (d *AuthenticatedGossiper) handleNodeAnnouncement(nMsg *networkMsg,
 	timestamp := time.Unix(int64(nodeAnn.Timestamp), 0)
 
 	log.Debugf("Processing NodeAnnouncement: peer=%v, timestamp=%v, "+
-		"node=%x", nMsg.peer, timestamp, nodeAnn.NodeID)
+		"node=%x, source=%x", nMsg.peer, timestamp, nodeAnn.NodeID,
+		nMsg.source.SerializeCompressed())
 
 	// We'll quickly ask the router if it already has a newer update for
 	// this node so we can skip validating signatures if not required.
@@ -2441,7 +2448,8 @@ func (d *AuthenticatedGossiper) handleNodeAnnouncement(nMsg *networkMsg,
 	// TODO(roasbeef): get rid of the above
 
 	log.Debugf("Processed NodeAnnouncement: peer=%v, timestamp=%v, "+
-		"node=%x", nMsg.peer, timestamp, nodeAnn.NodeID)
+		"node=%x, source=%x", nMsg.peer, timestamp, nodeAnn.NodeID,
+		nMsg.source.SerializeCompressed())
 
 	return announcements, true
 }
@@ -3045,9 +3053,9 @@ func (d *AuthenticatedGossiper) handleChanUpdate(nMsg *networkMsg,
 		edgeToUpdate = e2
 	}
 
-	log.Debugf("Validating ChannelUpdate: channel=%v, from node=%x, has "+
-		"edge=%v", chanInfo.ChannelID, pubKey.SerializeCompressed(),
-		edgeToUpdate != nil)
+	log.Debugf("Validating ChannelUpdate: channel=%v, for node=%x, has "+
+		"edge policy=%v", chanInfo.ChannelID,
+		pubKey.SerializeCompressed(), edgeToUpdate != nil)
 
 	// Validate the channel announcement with the expected public key and
 	// channel capacity. In the case of an invalid channel update, we'll

discovery/sync_manager.go

@@ -529,8 +529,8 @@ func (m *SyncManager) createGossipSyncer(peer lnpeer.Peer) *GossipSyncer {
 	s.setSyncState(chansSynced)
 	s.setSyncType(PassiveSync)
 
-	log.Debugf("Created new GossipSyncer[state=%s type=%s] for peer=%v",
-		s.syncState(), s.SyncType(), peer)
+	log.Debugf("Created new GossipSyncer[state=%s type=%s] for peer=%x",
+		s.syncState(), s.SyncType(), peer.PubKey())
 
 	return s
 }

discovery/sync_manager_test.go

@@ -28,7 +28,7 @@ func randPeer(t *testing.T, quit chan struct{}) *mockPeer {
 func peerWithPubkey(pk *btcec.PublicKey, quit chan struct{}) *mockPeer {
 	return &mockPeer{
 		pk:       pk,
-		sentMsgs: make(chan lnwire.Message),
+		sentMsgs: make(chan lnwire.Message, 1),
 		quit:     quit,
 	}
 }
@@ -483,7 +483,9 @@ func TestSyncManagerWaitUntilInitialHistoricalSync(t *testing.T) {
 		// transition it to chansSynced to ensure the remaining syncers
 		// aren't started as active.
 		if i == 0 {
-			assertSyncerStatus(t, s, syncingChans, PassiveSync)
+			assertSyncerStatus(
+				t, s, waitingQueryRangeReply, PassiveSync,
+			)
 			continue
 		}
 

discovery/syncer.go

@@ -475,6 +475,39 @@ func (g *GossipSyncer) Stop() {
 	})
 }
 
+// handleSyncingChans handles the state syncingChans for the GossipSyncer. When
+// in this state, we will send a QueryChannelRange msg to our peer and advance
+// the syncer's state to waitingQueryRangeReply.
+func (g *GossipSyncer) handleSyncingChans() {
+	// Prepare the query msg.
+	queryRangeMsg, err := g.genChanRangeQuery(g.genHistoricalChanRangeQuery)
+	if err != nil {
+		log.Errorf("Unable to gen chan range query: %v", err)
+		return
+	}
+
+	// Acquire a lock so the following state transition is atomic.
+	//
+	// NOTE: We must lock the following steps as it's possible we get an
+	// immediate response (ReplyChannelRange) after sending the query msg.
+	// The response is handled in ProcessQueryMsg, which requires the
+	// current state to be waitingQueryRangeReply.
+	g.Lock()
+	defer g.Unlock()
+
+	// Send the msg to the remote peer, which is non-blocking as
+	// `sendToPeer` only queues the msg in Brontide.
+	err = g.cfg.sendToPeer(queryRangeMsg)
+	if err != nil {
+		log.Errorf("Unable to send chan range query: %v", err)
+		return
+	}
+
+	// With the message sent successfully, we'll transition into the next
+	// state where we wait for their reply.
+	g.setSyncState(waitingQueryRangeReply)
+}
+
 // channelGraphSyncer is the main goroutine responsible for ensuring that we
 // properly channel graph state with the remote peer, and also that we only
 // send them messages which actually pass their defined update horizon.
@@ -495,27 +528,7 @@ func (g *GossipSyncer) channelGraphSyncer() {
 		// understand, as we'll as responding to any other queries by
 		// them.
 		case syncingChans:
-			// If we're in this state, then we'll send the remote
-			// peer our opening QueryChannelRange message.
-			queryRangeMsg, err := g.genChanRangeQuery(
-				g.genHistoricalChanRangeQuery,
-			)
-			if err != nil {
-				log.Errorf("Unable to gen chan range "+
-					"query: %v", err)
-				return
-			}
-
-			err = g.cfg.sendToPeer(queryRangeMsg)
-			if err != nil {
-				log.Errorf("Unable to send chan range "+
-					"query: %v", err)
-				return
-			}
-
-			// With the message sent successfully, we'll transition
-			// into the next state where we wait for their reply.
-			g.setSyncState(waitingQueryRangeReply)
+			g.handleSyncingChans()
 
 		// In this state, we've sent out our initial channel range
 		// query and are waiting for the final response from the remote
@@ -1342,9 +1355,9 @@ func (g *GossipSyncer) ApplyGossipFilter(filter *lnwire.GossipTimestampRange) er
 		return err
 	}
 
-	log.Infof("GossipSyncer(%x): applying new update horizon: start=%v, "+
-		"end=%v, backlog_size=%v", g.cfg.peerPub[:], startTime, endTime,
-		len(newUpdatestoSend))
+	log.Infof("GossipSyncer(%x): applying new remote update horizon: "+
+		"start=%v, end=%v, backlog_size=%v", g.cfg.peerPub[:],
+		startTime, endTime, len(newUpdatestoSend))
 
 	// If we don't have any to send, then we can return early.
 	if len(newUpdatestoSend) == 0 {
@@ -1515,12 +1528,15 @@ func (g *GossipSyncer) ProcessQueryMsg(msg lnwire.Message, peerQuit <-chan struc
 	// Reply messages should only be expected in states where we're waiting
 	// for a reply.
 	case *lnwire.ReplyChannelRange, *lnwire.ReplyShortChanIDsEnd:
+		g.Lock()
 		syncState := g.syncState()
+		g.Unlock()
+
 		if syncState != waitingQueryRangeReply &&
 			syncState != waitingQueryChanReply {
 
-			return fmt.Errorf("received unexpected query reply "+
-				"message %T", msg)
+			return fmt.Errorf("unexpected msg %T received in "+
+				"state %v", msg, syncState)
 		}
 		msgChan = g.gossipMsgs