Merge pull request #4933 from halseth/router-delay-zombie-pruning

routing: delay initial zombie prune by 30 sec

routing/router.go

@@ -43,6 +43,12 @@ const (
 	// if a channel should be pruned or not.
 	DefaultChannelPruneExpiry = time.Duration(time.Hour * 24 * 14)
 
+	// DefaultFirstTimePruneDelay is the time we'll wait after startup
+	// before attempting to prune the graph for zombie channels. We don't
+	// do it immediately after startup to allow lnd to start up without
+	// getting blocked by this job.
+	DefaultFirstTimePruneDelay = 30 * time.Second
+
 	// defaultStatInterval governs how often the router will log non-empty
 	// stats related to processing new channels, updates, or node
 	// announcements.
@@ -306,6 +312,12 @@ type Config struct {
 	// should examine the channel graph to garbage collect zombie channels.
 	GraphPruneInterval time.Duration
 
+	// FirstTimePruneDelay is the time we'll wait after startup before
+	// attempting to prune the graph for zombie channels. We don't do it
+	// immediately after startup to allow lnd to start up without getting
+	// blocked by this job.
+	FirstTimePruneDelay time.Duration
+
 	// QueryBandwidth is a method that allows the router to query the lower
 	// link layer to determine the up to date available bandwidth at a
 	// prospective link to be traversed. If the  link isn't available, then
@@ -485,11 +497,21 @@ func (r *ChannelRouter) Start() error {
 
 	// If AssumeChannelValid is present, then we won't rely on pruning
 	// channels from the graph based on their spentness, but whether they
-	// are considered zombies or not.
+	// are considered zombies or not. We will start zombie pruning after a
+	// small delay, to avoid slowing down startup of lnd.
 	if r.cfg.AssumeChannelValid {
-		if err := r.pruneZombieChans(); err != nil {
-			return err
+		time.AfterFunc(r.cfg.FirstTimePruneDelay, func() {
+			select {
+			case <-r.quit:
+				return
+			default:
 			}
+
+			log.Info("Initial zombie prune starting")
+			if err := r.pruneZombieChans(); err != nil {
+				log.Errorf("Unable to prune zombies: %v", err)
+			}
+		})
 	} else {
 		// Otherwise, we'll use our filtered chain view to prune
 		// channels as soon as they are detected as spent on-chain.
@@ -875,6 +897,9 @@ func (r *ChannelRouter) pruneZombieChans() error {
 	}
 
 	log.Infof("Pruning %v zombie channels", len(chansToPrune))
+	if len(chansToPrune) == 0 {
+		return nil
+	}
 
 	// With the set of zombie-like channels obtained, we'll do another pass
 	// to delete them from the channel graph.

routing/router_test.go

@@ -1541,6 +1541,9 @@ func TestWakeUpOnStaleBranch(t *testing.T) {
 		Control:            makeMockControlTower(),
 		ChannelPruneExpiry: time.Hour * 24,
 		GraphPruneInterval: time.Hour * 2,
+
+		// We'll set the delay to zero to prune immediately.
+		FirstTimePruneDelay: 0,
 	})
 	if err != nil {
 		t.Fatalf("unable to create router %v", err)
@@ -2157,6 +2160,9 @@ func testPruneChannelGraphDoubleDisabled(t *testing.T, assumeValid bool) {
 	if !assumeValid {
 		assertChannelsPruned(t, ctx.graph, testChannels)
 	} else {
+		// Sleep to allow the pruning to finish.
+		time.Sleep(200 * time.Millisecond)
+
 		prunedChannel := testChannels[len(testChannels)-1].ChannelID
 		assertChannelsPruned(t, ctx.graph, testChannels, prunedChannel)
 	}

server.go

@@ -777,7 +777,8 @@ func newServer(cfg *Config, listenAddrs []net.Addr,
 		MissionControl:      s.missionControl,
 		SessionSource:       paymentSessionSource,
 		ChannelPruneExpiry:  routing.DefaultChannelPruneExpiry,
-		GraphPruneInterval: time.Duration(time.Hour),
+		GraphPruneInterval:  time.Hour,
+		FirstTimePruneDelay: routing.DefaultFirstTimePruneDelay,
 		QueryBandwidth:      queryBandwidth,
 		AssumeChannelValid:  cfg.Routing.AssumeChannelValid,
 		NextPaymentID:       sequencer.NextID,