コード例 #1
0
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
}
コード例 #2
0
func (self *NsqdCoordinator) PutMessageToCluster(topic *nsqd.Topic,
	body []byte, traceID uint64) (nsqd.MessageID, nsqd.BackendOffset, int32, nsqd.BackendQueueEnd, error) {
	var commitLog CommitLogData
	var queueEnd nsqd.BackendQueueEnd
	msg := nsqd.NewMessage(0, body)
	msg.TraceID = traceID

	topicName := topic.GetTopicName()
	partition := topic.GetTopicPart()
	coord, checkErr := self.getTopicCoord(topicName, partition)
	if checkErr != nil {
		return msg.ID, nsqd.BackendOffset(commitLog.MsgOffset), commitLog.MsgSize, queueEnd, checkErr.ToErrorType()
	}

	var logMgr *TopicCommitLogMgr

	doLocalWrite := func(d *coordData) *CoordErr {
		logMgr = d.logMgr
		topic.Lock()
		id, offset, writeBytes, qe, localErr := topic.PutMessageNoLock(msg)
		queueEnd = qe
		topic.Unlock()
		if localErr != nil {
			coordLog.Warningf("put message to local failed: %v", localErr)
			return &CoordErr{localErr.Error(), RpcNoErr, CoordLocalErr}
		}
		commitLog.LogID = int64(id)
		// epoch should not be changed.
		// leader epoch change means leadership change, leadership change
		// need disable write which should hold the write lock.
		// However, we are holding write lock while doing the cluster write replication.
		commitLog.Epoch = d.GetTopicEpochForWrite()
		commitLog.LastMsgLogID = commitLog.LogID
		commitLog.MsgOffset = int64(offset)
		commitLog.MsgSize = writeBytes
		commitLog.MsgCnt = queueEnd.TotalMsgCnt()
		commitLog.MsgNum = 1

		return nil
	}
	doLocalExit := func(err *CoordErr) {
		if err != nil {
			coordLog.Infof("topic %v PutMessageToCluster msg %v error: %v", topic.GetFullName(), msg, err)
			if coord.IsWriteDisabled() {
				topic.DisableForSlave()
			}
		}
	}
	doLocalCommit := func() error {
		localErr := logMgr.AppendCommitLog(&commitLog, false)
		if localErr != nil {
			coordLog.Errorf("topic : %v, Generator %v failed write commit log : %v, logmgr: %v, %v",
				topic.GetFullName(), topic.GetMsgGenerator(), localErr, logMgr.pLogID, logMgr.nLogID)
		}
		topic.Lock()
		topic.UpdateCommittedOffset(queueEnd)
		topic.Unlock()
		return localErr
	}
	doLocalRollback := func() {
		coordLog.Warningf("failed write begin rollback : %v, %v", topic.GetFullName(), commitLog)
		topic.Lock()
		topic.RollbackNoLock(nsqd.BackendOffset(commitLog.MsgOffset), 1)
		topic.Unlock()
	}
	doRefresh := func(d *coordData) *CoordErr {
		logMgr = d.logMgr
		if d.GetTopicEpochForWrite() != commitLog.Epoch {
			coordLog.Warningf("write epoch changed during write: %v, %v", d.GetTopicEpochForWrite(), commitLog)
			return ErrEpochMismatch
		}
		self.requestNotifyNewTopicInfo(d.topicInfo.Name, d.topicInfo.Partition)
		return nil
	}
	doSlaveSync := func(c *NsqdRpcClient, nodeID string, tcData *coordData) *CoordErr {
		// should retry if failed, and the slave should keep the last success write to avoid the duplicated
		putErr := c.PutMessage(&tcData.topicLeaderSession, &tcData.topicInfo, commitLog, msg)
		if putErr != nil {
			coordLog.Infof("sync write to replica %v failed: %v. put offset:%v, logmgr: %v, %v",
				nodeID, putErr, commitLog, logMgr.pLogID, logMgr.nLogID)
		}
		return putErr
	}
	handleSyncResult := func(successNum int, tcData *coordData) bool {
		if successNum == len(tcData.topicInfo.ISR) {
			return true
		}
		return false
	}

	clusterErr := self.doSyncOpToCluster(true, coord, doLocalWrite, doLocalExit, doLocalCommit, doLocalRollback,
		doRefresh, doSlaveSync, handleSyncResult)

	var err error
	if clusterErr != nil {
		err = clusterErr.ToErrorType()
	}
	return msg.ID, nsqd.BackendOffset(commitLog.MsgOffset), commitLog.MsgSize, queueEnd, err
}