func (self *NsqdCoordinator) putMessagesOnSlave(coord *TopicCoordinator, logData CommitLogData, msgs []*nsqd.Message) *CoordErr { if len(msgs) == 0 { return ErrPubArgError } if logData.LogID != int64(msgs[0].ID) { return ErrPubArgError } var logMgr *TopicCommitLogMgr // this last log id should be used on slave to avoid the slave switch // override the leader's prev mpub message id. // While slave is chosen as leader, the next id should be larger than the last logid. // Because the mpub maybe already committed after the leader is down, the new leader should begin // with the last message id + 1 for next message. lastMsgLogID := int64(msgs[len(msgs)-1].ID) if logData.LastMsgLogID != lastMsgLogID { return ErrPubArgError } var queueEnd nsqd.BackendQueueEnd var topic *nsqd.Topic checkDupOnSlave := func(tc *coordData) bool { if coordLog.Level() >= levellogger.LOG_DETAIL { topicName := tc.topicInfo.Name coordLog.Debugf("pub on slave : %v, msg count: %v", topicName, len(msgs)) } logMgr = tc.logMgr if logMgr.IsCommitted(logData.LogID) { coordLog.Infof("put the already committed log id : %v", logData.LogID) return true } return false } doLocalWriteOnSlave := func(tc *coordData) *CoordErr { var localErr error var start time.Time checkCost := coordLog.Level() >= levellogger.LOG_DEBUG if self.enableBenchCost { checkCost = true } if checkCost { start = time.Now() } topicName := tc.topicInfo.Name partition := tc.topicInfo.Partition topic, localErr = self.localNsqd.GetExistingTopic(topicName, partition) if localErr != nil { coordLog.Infof("pub on slave missing topic : %v", topicName) // leave the isr and try re-sync with leader return &CoordErr{localErr.Error(), RpcErrTopicNotExist, CoordSlaveErr} } topic.Lock() var cost time.Duration if checkCost { cost = time.Now().Sub(start) if cost > time.Millisecond { coordLog.Infof("prepare write on slave local cost :%v", cost) } } queueEnd, localErr = topic.PutMessagesOnReplica(msgs, nsqd.BackendOffset(logData.MsgOffset)) if checkCost { cost2 := time.Now().Sub(start) if cost2 > time.Millisecond { coordLog.Infof("write local on slave cost :%v, %v", cost, cost2) } } topic.Unlock() if localErr != nil { logIndex, lastLogOffset, lastLog, _ := logMgr.GetLastCommitLogOffsetV2() coordLog.Errorf("put messages on slave failed: %v, slave last logid: %v, data: %v:%v, %v", localErr, logMgr.GetLastCommitLogID(), logIndex, lastLogOffset, lastLog) return &CoordErr{localErr.Error(), RpcCommonErr, CoordSlaveErr} } return nil } doLocalCommit := func() error { localErr := logMgr.AppendCommitLog(&logData, true) if localErr != nil { coordLog.Errorf("write commit log on slave failed: %v", localErr) return localErr } topic.Lock() topic.UpdateCommittedOffset(queueEnd) topic.Unlock() return nil } doLocalExit := func(err *CoordErr) { if err != nil { coordLog.Warningf("failed to batch put messages on slave: %v", err) } } return self.doWriteOpOnSlave(coord, checkDupOnSlave, doLocalWriteOnSlave, doLocalCommit, doLocalExit) }