Example #1
0
func (m *master) fullSync() error {
	log.Info("begin full sync")

	if err := m.conn.Send("fullsync"); err != nil {
		return err
	}

	m.state.Set(replSyncState)

	dumpPath := path.Join(m.app.cfg.DataDir, "master.dump")
	f, err := os.OpenFile(dumpPath, os.O_CREATE|os.O_WRONLY, 0644)
	if err != nil {
		return err
	}

	defer os.Remove(dumpPath)

	err = m.conn.ReceiveBulkTo(f)
	f.Close()
	if err != nil {
		log.Errorf("read dump data error %s", err.Error())
		return err
	}

	if _, err = m.app.ldb.LoadDumpFile(dumpPath); err != nil {
		log.Errorf("load dump file error %s", err.Error())
		return err
	}

	return nil
}
Example #2
0
func (app *App) publishNewLog(l *rpl.Log) {
	if !app.cfg.Replication.Sync {
		//no sync replication, we will do async
		return
	}

	app.info.Replication.PubLogNum.Add(1)

	app.slock.Lock()

	slaveNum := len(app.slaves)

	total := (slaveNum + 1) / 2
	if app.cfg.Replication.WaitMaxSlaveAcks > 0 {
		total = num.MinInt(total, app.cfg.Replication.WaitMaxSlaveAcks)
	}

	n := 0
	logId := l.ID
	for _, s := range app.slaves {
		lastLogID := s.lastLogID.Get()
		if lastLogID == logId {
			//slave has already owned this log
			n++
		} else if lastLogID > logId {
			log.Errorf("invalid slave %s, lastlogid %d > %d", s.slaveListeningAddr, lastLogID, logId)
		}
	}

	app.slock.Unlock()

	if n >= total {
		//at least total slaves have owned this log
		return
	}

	startTime := time.Now()
	done := make(chan struct{}, 1)
	go func() {
		n := 0
		for i := 0; i < slaveNum; i++ {
			id := <-app.slaveSyncAck
			if id < logId {
				log.Infof("some slave may close with last logid %d < %d", id, logId)
			} else {
				n++
				if n >= total {
					break
				}
			}
		}
		done <- struct{}{}
	}()

	select {
	case <-done:
	case <-time.After(time.Duration(app.cfg.Replication.WaitSyncTime) * time.Millisecond):
		log.Info("replication wait timeout")
	}

	stopTime := time.Now()
	app.info.Replication.PubLogAckNum.Add(1)
	app.info.Replication.PubLogTotalAckTime.Add(stopTime.Sub(startTime))
}