func (self *NsqdCoordinator) DeleteChannel(topic *nsqd.Topic, channelName string) error {
topicName := topic.GetTopicName()
partition := topic.GetTopicPart()
coord, checkErr := self.getTopicCoord(topicName, partition)
if checkErr != nil {
return checkErr.ToErrorType()
}
doLocalWrite := func(d *coordData) *CoordErr {
localErr := topic.DeleteExistingChannel(channelName)
if localErr != nil {
coordLog.Infof("deleteing local channel %v error: %v", channelName, localErr)
}
return nil
}
doLocalExit := func(err *CoordErr) {}
doLocalCommit := func() error {
return nil
}
doLocalRollback := func() {
}
doRefresh := func(d *coordData) *CoordErr {
return nil
}
doSlaveSync := func(c *NsqdRpcClient, nodeID string, tcData *coordData) *CoordErr {
rpcErr := c.DeleteChannel(&tcData.topicLeaderSession, &tcData.topicInfo, channelName)
if rpcErr != nil {
coordLog.Infof("delete channel(%v) to replica %v failed: %v", channelName,
nodeID, rpcErr)
}
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) {
return true
}
return false
}
clusterErr := self.doSyncOpToCluster(false, coord, doLocalWrite, doLocalExit, doLocalCommit, doLocalRollback,
doRefresh, doSlaveSync, handleSyncResult)
if clusterErr != nil {
return clusterErr.ToErrorType()
}
return nil
}
func (c *context) internalPubLoop(topic *nsqd.Topic) {
messages := make([]*nsqd.Message, 0, 100)
pubInfoList := make([]*nsqd.PubInfo, 0, 100)
topicName := topic.GetTopicName()
partition := topic.GetTopicPart()
nsqd.NsqLogger().Logf("start pub loop for topic: %v ", topic.GetFullName())
defer func() {
done := false
for !done {
select {
case info := <-topic.GetWaitChan():
pubInfoList = append(pubInfoList, info)
default:
done = true
}
}
nsqd.NsqLogger().Logf("quit pub loop for topic: %v, left: %v ", topic.GetFullName(), len(pubInfoList))
for _, info := range pubInfoList {
info.Err = nsqd.ErrExiting
close(info.Done)
}
}()
quitChan := topic.QuitChan()
infoChan := topic.GetWaitChan()
for {
select {
case <-quitChan:
return
case info := <-infoChan:
if info.MsgBody.Len() <= 0 {
nsqd.NsqLogger().Logf("empty msg body")
}
messages = append(messages, nsqd.NewMessage(0, info.MsgBody.Bytes()))
pubInfoList = append(pubInfoList, info)
// TODO: avoid too much in a batch
default:
if len(pubInfoList) == 0 {
nsqd.NsqLogger().LogDebugf("topic %v pub loop waiting for message", topic.GetFullName())
select {
case <-quitChan:
return
case info := <-infoChan:
if info.MsgBody.Len() <= 0 {
nsqd.NsqLogger().Logf("empty msg body")
}
messages = append(messages, nsqd.NewMessage(0, info.MsgBody.Bytes()))
pubInfoList = append(pubInfoList, info)
}
continue
}
if len(pubInfoList) > 1 {
nsqd.NsqLogger().LogDebugf("pub loop batch number: %v", len(pubInfoList))
}
var retErr error
if c.checkForMasterWrite(topicName, partition) {
_, _, _, err := c.PutMessages(topic, messages)
if err != nil {
nsqd.NsqLogger().LogErrorf("topic %v put messages %v failed: %v", topic.GetFullName(), len(messages), err)
retErr = err
}
} else {
topic.DisableForSlave()
nsqd.NsqLogger().LogDebugf("should put to master: %v",
topic.GetFullName())
retErr = consistence.ErrNotTopicLeader.ToErrorType()
}
for _, info := range pubInfoList {
info.Err = retErr
close(info.Done)
}
pubInfoList = pubInfoList[:0]
messages = messages[:0]
}
}
}
func (self *NsqdCoordinator) putMessageOnSlave(coord *TopicCoordinator, logData CommitLogData, msg *nsqd.Message) *CoordErr {
var logMgr *TopicCommitLogMgr
var topic *nsqd.Topic
var queueEnd nsqd.BackendQueueEnd
checkDupOnSlave := func(tc *coordData) bool {
if coordLog.Level() >= levellogger.LOG_DETAIL {
topicName := tc.topicInfo.Name
coordLog.Debugf("pub on slave : %v, msg %v", topicName, msg.ID)
}
logMgr = tc.logMgr
if logMgr.IsCommitted(logData.LogID) {
coordLog.Infof("pub the already committed log id : %v", logData.LogID)
return true
}
return false
}
doLocalWriteOnSlave := func(tc *coordData) *CoordErr {
var localErr error
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}
}
if topic.GetTopicPart() != partition {
coordLog.Errorf("topic on slave has different partition : %v vs %v", topic.GetTopicPart(), partition)
return &CoordErr{ErrLocalTopicPartitionMismatch.String(), RpcErrTopicNotExist, CoordSlaveErr}
}
topic.Lock()
queueEnd, localErr = topic.PutMessageOnReplica(msg, nsqd.BackendOffset(logData.MsgOffset))
topic.Unlock()
if localErr != nil {
coordLog.Errorf("put message on slave failed: %v", localErr)
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.Infof("slave put message %v error: %v", logData, err)
}
}
return self.doWriteOpOnSlave(coord, checkDupOnSlave, doLocalWriteOnSlave, doLocalCommit, doLocalExit)
}
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
}