add Delayed sending

consensus/consensus_msg_sender.go

@@ -56,7 +56,7 @@ func (sender *MessageSender) Reset(blockNum uint64) {
 }
 
 // SendWithRetry sends message with retry logic.
-func (sender *MessageSender) SendWithRetry(blockNum uint64, msgType msg_pb.MessageType, groups []nodeconfig.GroupID, p2pMsg []byte, immediate bool) error {
+func (sender *MessageSender) SendWithRetry(blockNum uint64, msgType msg_pb.MessageType, groups []nodeconfig.GroupID, p2pMsg []byte) error {
 	if sender.retryTimes != 0 {
 		msgRetry := MessageRetry{blockNum: blockNum, groups: groups, p2pMsg: p2pMsg, msgType: msgType, retryCount: 0}
 		atomic.StoreUint32(&msgRetry.isActive, 1)
@@ -65,10 +65,19 @@ func (sender *MessageSender) SendWithRetry(blockNum uint64, msgType msg_pb.Messa
 			sender.Retry(&msgRetry)
 		}()
 	}
-	if immediate {
 	return sender.host.SendMessageToGroups(groups, p2pMsg)
+}
+
+// DelayedSendWithRetry is similar to SendWithRetry but without the initial message sending but only retries.
+func (sender *MessageSender) DelayedSendWithRetry(blockNum uint64, msgType msg_pb.MessageType, groups []nodeconfig.GroupID, p2pMsg []byte) {
+	if sender.retryTimes != 0 {
+		msgRetry := MessageRetry{blockNum: blockNum, groups: groups, p2pMsg: p2pMsg, msgType: msgType, retryCount: 0}
+		atomic.StoreUint32(&msgRetry.isActive, 1)
+		sender.messagesToRetry.Store(msgType, &msgRetry)
+		go func() {
+			sender.Retry(&msgRetry)
+		}()
 	}
-	return nil
 }
 
 // SendWithoutRetry sends message without retry logic.

consensus/consensus_v2.go

@@ -161,17 +161,14 @@ func (consensus *Consensus) finalCommit() {
 	// Note: leader already sent 67% commit in preCommit. The 100% commit won't be sent immediately
 	// to save network traffic. It will only be sent in retry if consensus doesn't move forward.
 	// Or if the leader is changed for next block, the 100% committed sig will be sent to the next leader immediately.
-	sendImmediately := false
 	if !consensus.IsLeader() || block.IsLastBlockInEpoch() {
-		sendImmediately = true
-	}
+		// send immediately
 		if err := consensus.msgSender.SendWithRetry(
 			block.NumberU64(),
 			msg_pb.MessageType_COMMITTED, []nodeconfig.GroupID{
 				nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID)),
 			},
-		p2p.ConstructMessage(msgToSend),
-		sendImmediately); err != nil {
+			p2p.ConstructMessage(msgToSend)); err != nil {
 			consensus.getLogger().Warn().Err(err).Msg("[finalCommit] Cannot send committed message")
 		} else {
 			consensus.getLogger().Info().
@@ -179,6 +176,19 @@ func (consensus *Consensus) finalCommit() {
 				Uint64("blockNum", consensus.blockNum).
 				Msg("[finalCommit] Sent Committed Message")
 		}
+	} else {
+		// delayed send
+		consensus.msgSender.DelayedSendWithRetry(
+			block.NumberU64(),
+			msg_pb.MessageType_COMMITTED, []nodeconfig.GroupID{
+				nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID)),
+			},
+			p2p.ConstructMessage(msgToSend))
+		consensus.getLogger().Info().
+			Hex("blockHash", curBlockHash[:]).
+			Uint64("blockNum", consensus.blockNum).
+			Msg("[finalCommit] Queued Committed Message")
+	}
 
 	// Dump new block into level db
 	// In current code, we add signatures in block in tryCatchup, the block dump to explorer does not contains signatures
@@ -550,8 +560,7 @@ func (consensus *Consensus) preCommitAndPropose(blk *types.Block) error {
 		msg_pb.MessageType_COMMITTED, []nodeconfig.GroupID{
 			nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID)),
 		},
-		p2p.ConstructMessage(msgToSend),
-		true); err != nil {
+		p2p.ConstructMessage(msgToSend)); err != nil {
 		consensus.getLogger().Warn().Err(err).Msg("[preCommitAndPropose] Cannot send committed message")
 	} else {
 		consensus.getLogger().Info().

consensus/leader.go

@@ -80,7 +80,7 @@ func (consensus *Consensus) announce(block *types.Block) {
 	if err := consensus.msgSender.SendWithRetry(
 		consensus.blockNum, msg_pb.MessageType_ANNOUNCE, []nodeconfig.GroupID{
 			nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID)),
-		}, p2p.ConstructMessage(msgToSend), true); err != nil {
+		}, p2p.ConstructMessage(msgToSend)); err != nil {
 		consensus.getLogger().Warn().
 			Str("groupID", string(nodeconfig.NewGroupIDByShardID(
 				nodeconfig.ShardID(consensus.ShardID),

consensus/threshold.go

@@ -74,7 +74,6 @@ func (consensus *Consensus) didReachPrepareQuorum() error {
 			nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID)),
 		},
 		p2p.ConstructMessage(msgToSend),
-		true,
 	); err != nil {
 		consensus.getLogger().Warn().Msg("[OnPrepare] Cannot send prepared message")
 	} else {

consensus/view_change.go

@@ -265,7 +265,6 @@ func (consensus *Consensus) startViewChange() {
 			[]nodeconfig.GroupID{
 				nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID))},
 			p2p.ConstructMessage(msgToSend),
-			true,
 		); err != nil {
 			consensus.getLogger().Err(err).
 				Msg("[startViewChange] could not send out the ViewChange message")
@@ -295,7 +294,6 @@ func (consensus *Consensus) startNewView(viewID uint64, newLeaderPriKey *bls.Pri
 		[]nodeconfig.GroupID{
 			nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID))},
 		p2p.ConstructMessage(msgToSend),
-		true,
 	); err != nil {
 		return errors.New("failed to send out the NewView message")
 	}

node/node_newblock.go

@@ -44,7 +44,7 @@ func (node *Node) WaitForConsensusReadyV2(readySignal chan consensus.ProposalTyp
 					Msg("Consensus new block proposal: STOPPED!")
 				return
 			case proposalType := <-readySignal:
-				retryCount := 3
+				retryCount := 0
 				for node.Consensus != nil && node.Consensus.IsLeader() {
 					time.Sleep(SleepPeriod)
 					utils.Logger().Info().
@@ -97,9 +97,10 @@ func (node *Node) WaitForConsensusReadyV2(readySignal chan consensus.ProposalTyp
 						node.BlockChannel <- newBlock
 						break
 					} else {
-						utils.Logger().Err(err).Msg("!!!!!!!!!Failed Proposing New Block!!!!!!!!!")
-						retryCount--
-						if retryCount == 0 {
+						retryCount++
+						utils.Logger().Err(err).Int("retryCount", retryCount).
+							Msg("!!!!!!!!!Failed Proposing New Block!!!!!!!!!")
+						if retryCount > 3 {
 							// break to avoid repeated failures
 							break
 						}