fix:snapsync network abnormality leads to status error

Qitmeer · Dec 3, 2024 · b3a013e · b3a013e
1 parent 5873a46
commit b3a013e
Show file tree

Hide file tree

Showing 4 changed files with 23 additions and 12 deletions.
diff --git a/meerevm/meer/meerchain.go b/meerevm/meer/meerchain.go
@@ -695,6 +695,9 @@ func (b *MeerChain) SyncTo(target common.Hash) error {
 		if system.InterruptRequested(b.consensus.Interrupt()) {
 			return errors.New("System interrupt")
 		}
+		if b.IsShutdown() {
+			return errors.New("MeerChain is shutdown")
+		}
 	}
 }
 

diff --git a/p2p/synch/handshake.go b/p2p/synch/handshake.go
@@ -70,7 +70,6 @@ func (ps *PeerSync) processConnected(msg *ConnectedMsg) {
 func (ps *PeerSync) immediatelyConnected(pe *peers.Peer) {
 	pe.HSlock.Lock()
 	defer pe.HSlock.Unlock()
-
 	ps.Connection(pe)
 }
 

diff --git a/p2p/synch/peersync.go b/p2p/synch/peersync.go
@@ -172,13 +172,10 @@ func (ps *PeerSync) handleStallSample() {
 		return
 	}
 	lbid := ps.Chain().BlockDAG().GetLastBlockID()
-	if ps.lastBlockID <= 0 {
-		ps.lastBlockID = lbid
-		return
-	}
 	if ps.lastBlockID != lbid {
 		return
 	}
+	ps.lastBlockID = lbid
 	ps.TryAgainUpdateSyncPeer(true)
 }
 
@@ -734,11 +731,12 @@ func (ps *PeerSync) getProcessID() string {
 
 func NewPeerSync(sy *Sync) *PeerSync {
 	peerSync := &PeerSync{
-		sy:        sy,
-		msgChan:   make(chan interface{}),
-		quit:      make(chan struct{}),
-		pause:     false,
-		interrupt: make(chan struct{}),
+		sy:          sy,
+		msgChan:     make(chan interface{}),
+		quit:        make(chan struct{}),
+		pause:       false,
+		interrupt:   make(chan struct{}),
+		lastBlockID: meerdag.MaxId,
 	}
 
 	return peerSync

diff --git a/p2p/synch/snapsync.go b/p2p/synch/snapsync.go
@@ -85,8 +85,19 @@ cleanup:
 	}
 	add := 0
 	ps.snapStatus.locker.Lock()
+	tryReq := 10
+	var ret *pb.SnapSyncRsp
+
 	for !ps.snapStatus.isPointCompleted() {
-		ret, err := ps.syncSnapStatus(bestPeer)
+		for i := 0; i < tryReq; i++ {
+			ret, err = ps.syncSnapStatus(bestPeer)
+			if err != nil {
+				log.Warn("Snap-sync waiting for next try", "err", err.Error(), "cur", i, "max", tryReq)
+				time.Sleep(time.Second * 5)
+				continue
+			}
+			break
+		}
 		if err != nil {
 			log.Warn("Snap-sync", "err", err.Error())
 			break
@@ -106,8 +117,8 @@ cleanup:
 			break
 		}
 		ps.snapStatus.setSyncPoint(latest)
-		log.Trace("Snap-sync status update point", "point", latest.GetHash().String())
 		add += len(sds)
+		log.Trace("Snap-sync", "point", latest.GetHash().String(), "data_num", len(sds), "total", add)
 	}
 	ps.snapStatus.locker.Unlock()
 	err = ps.Chain().MeerChain().SyncTo(ps.snapStatus.GetSyncPoint().GetState().GetEVMHash())