Merge pull request #3448 from LeoHChen/viewchange_fix_out_of_sync

Viewchange fix out of sync

consensus/config.go

@@ -6,6 +6,10 @@ import "time"
 const (
 	// default timeout configuration is shorten to 45 seconds as the consensus is 5s
 	viewChangeTimeout = 45
+	// viewChangeSlot means every 90 seconds, the view change ID will be advanced.
+	// so that the nodes init view change process within the 90 seconds range will
+	// be have the same view change ID
+	viewChangeSlot = 90
 	// The duration of viewChangeTimeout for each view change
 	viewChangeDuration time.Duration = viewChangeTimeout * time.Second
 

consensus/quorum/one-node-staked-vote.go

@@ -146,6 +146,9 @@ func (v *stakedVoteWeight) IsQuorumAchieved(p Phase) bool {
 // IsQuorumAchivedByMask ..
 func (v *stakedVoteWeight) IsQuorumAchievedByMask(mask *bls_cosi.Mask) bool {
 	threshold := v.QuorumThreshold()
+	if mask == nil {
+		return false
+	}
 	currentTotalPower := v.computeTotalPowerByMask(mask)
 	if currentTotalPower == nil {
 		return false

consensus/view_change.go

@@ -133,7 +133,7 @@ func (consensus *Consensus) getNextViewID() (uint64, time.Duration) {
 		return consensus.fallbackNextViewID()
 	}
 	// diff only increases
-	diff := uint64((curTimestamp - blockTimestamp) / viewChangeTimeout)
+	diff := uint64((curTimestamp - blockTimestamp) / viewChangeSlot)
 	nextViewID := diff + lastBlockViewID
 
 	consensus.getLogger().Info().
@@ -169,6 +169,8 @@ func (consensus *Consensus) getNextLeaderKey(viewID uint64) *bls.PublicKeyWrappe
 			consensus.getLogger().Error().Msg("[getNextLeaderKey] Failed to get current header from blockchain")
 			lastLeaderPubKey = consensus.LeaderPubKey
 		} else {
+			lastBlockViewID := curHeader.ViewID().Uint64()
+			gap = int(viewID - lastBlockViewID)
 			// this is the truth of the leader based on blockchain blocks
 			lastLeaderPubKey, err = consensus.getLeaderPubKeyFromCoinbase(curHeader)
 			if err != nil || lastLeaderPubKey == nil {
@@ -351,7 +353,7 @@ func (consensus *Consensus) onViewChange(msg *msg_pb.Message) {
 		return
 	}
 
-	if consensus.Decider.IsQuorumAchieved(quorum.ViewChange) {
+	if consensus.Decider.IsQuorumAchievedByMask(consensus.vc.GetViewIDBitmap(recvMsg.ViewID)) {
 		consensus.getLogger().Info().
 			Int64("have", consensus.Decider.SignersCount(quorum.ViewChange)).
 			Int64("need", consensus.Decider.TwoThirdsSignersCount()).

consensus/view_change_construct.go

@@ -202,6 +202,10 @@ func (vc *viewChange) VerifyNewViewMsg(recvMsg *FBFTMessage) (*types.Block, erro
 	binary.LittleEndian.PutUint64(viewIDBytes, recvMsg.ViewID)
 
 	if !m3Sig.VerifyHash(m3Mask.AggregatePublic, viewIDBytes) {
+		vc.getLogger().Warn().
+			Bytes("viewIDBytes", viewIDBytes).
+			Interface("AggregatePublic", m3Mask.AggregatePublic).
+			Msg("m3Sig.VerifyHash Failed")
 		return nil, errors.New("[VerifyNewViewMsg] Unable to Verify Aggregated Signature of M3 (ViewID) payload")
 	}
 
@@ -209,6 +213,10 @@ func (vc *viewChange) VerifyNewViewMsg(recvMsg *FBFTMessage) (*types.Block, erro
 	if recvMsg.M2AggSig != nil {
 		m2Sig := recvMsg.M2AggSig
 		if !m2Sig.VerifyHash(m2Mask.AggregatePublic, NIL) {
+			vc.getLogger().Warn().
+				Bytes("NIL", NIL).
+				Interface("AggregatePublic", m2Mask.AggregatePublic).
+				Msg("m2Sig.VerifyHash Failed")
 			return nil, errors.New("[VerifyNewViewMsg] Unable to Verify Aggregated Signature of M2 (NIL) payload")
 		}
 	}