func (self *NsqdCoordinator) FinishMessageToCluster(channel *nsqd.Channel, clientID int64, clientAddr string, msgID nsqd.MessageID) error {
	topicName := channel.GetTopicName()
	partition := channel.GetTopicPart()
	coord, checkErr := self.getTopicCoord(topicName, partition)
	if checkErr != nil {
		return checkErr.ToErrorType()
	}

	var syncOffset ChannelConsumerOffset
	changed := false
	var confirmed nsqd.BackendQueueEnd
	if channel.IsOrdered() {
		if !coord.GetData().IsISRReadyForWrite() {
			coordLog.Warningf("topic(%v) finish message ordered failed since no enough ISR", topicName)
			coordErrStats.incWriteErr(ErrWriteQuorumFailed)
			return ErrWriteQuorumFailed.ToErrorType()
		}

		confirmed = channel.GetConfirmed()
	}
	// TODO: maybe use channel to aggregate all the sync of message to reduce the rpc call.

	doLocalWrite := func(d *coordData) *CoordErr {
		offset, cnt, tmpChanged, localErr := channel.FinishMessage(clientID, clientAddr, msgID)
		if localErr != nil {
			coordLog.Infof("channel %v finish local msg %v error: %v", channel.GetName(), msgID, localErr)
			changed = false
			return &CoordErr{localErr.Error(), RpcNoErr, CoordLocalErr}
		}
		changed = tmpChanged
		syncOffset.VOffset = int64(offset)
		syncOffset.VCnt = cnt
		return nil
	}
	doLocalExit := func(err *CoordErr) {}
	doLocalCommit := func() error {
		channel.ContinueConsumeForOrder()
		return nil
	}
	doLocalRollback := func() {
		if channel.IsOrdered() && confirmed != nil {
			coordLog.Warningf("rollback channel confirm to : %v", confirmed)
			// reset read to last confirmed
			channel.SetConsumeOffset(confirmed.Offset(), confirmed.TotalMsgCnt(), true)
		}
	}
	doRefresh := func(d *coordData) *CoordErr {
		return nil
	}
	doSlaveSync := func(c *NsqdRpcClient, nodeID string, tcData *coordData) *CoordErr {
		if !changed || channel.IsEphemeral() {
			return nil
		}
		var rpcErr *CoordErr
		if channel.IsOrdered() {
			// if ordered, we need make sure all the consume offset is synced to all replicas
			rpcErr = c.UpdateChannelOffset(&tcData.topicLeaderSession, &tcData.topicInfo, channel.GetName(), syncOffset)
		} else {
			c.NotifyUpdateChannelOffset(&tcData.topicLeaderSession, &tcData.topicInfo, channel.GetName(), syncOffset)
		}
		if rpcErr != nil {
			coordLog.Infof("sync channel(%v) offset to replica %v failed: %v, offset: %v", channel.GetName(),
				nodeID, rpcErr, syncOffset)
		}
		return rpcErr
	}
	handleSyncResult := func(successNum int, tcData *coordData) bool {
		// we can ignore the error if this channel is not ordered. (just sync next time)
		if successNum == len(tcData.topicInfo.ISR) || !channel.IsOrdered() {
			return true
		}
		return false
	}
	clusterErr := self.doSyncOpToCluster(false, coord, doLocalWrite, doLocalExit, doLocalCommit, doLocalRollback,
		doRefresh, doSlaveSync, handleSyncResult)
	if clusterErr != nil {
		return clusterErr.ToErrorType()
	}
	return nil
}