func (c *context) FinishMessage(ch *nsqd.Channel, clientID int64, clientAddr string, msgID nsqd.MessageID) error {
if c.nsqdCoord == nil {
_, _, _, err := ch.FinishMessage(clientID, clientAddr, msgID)
if err == nil {
ch.ContinueConsumeForOrder()
}
return err
}
return c.nsqdCoord.FinishMessageToCluster(ch, clientID, clientAddr, msgID)
}
func (c *context) SetChannelOffset(ch *nsqd.Channel, startFrom *ConsumeOffset, force bool) (int64, int64, error) {
var l *consistence.CommitLogData
var queueOffset int64
cnt := int64(0)
var err error
if startFrom.OffsetType == offsetTimestampType {
if c.nsqdCoord != nil {
l, queueOffset, cnt, err = c.nsqdCoord.SearchLogByMsgTimestamp(ch.GetTopicName(), ch.GetTopicPart(), startFrom.OffsetValue)
} else {
err = errors.New("Not supported while coordinator disabled")
}
} else if startFrom.OffsetType == offsetSpecialType {
if startFrom.OffsetValue == -1 {
e := ch.GetChannelEnd()
queueOffset = int64(e.Offset())
cnt = e.TotalMsgCnt()
} else {
nsqd.NsqLogger().Logf("not known special offset :%v", startFrom)
err = errors.New("not supported offset type")
}
} else if startFrom.OffsetType == offsetVirtualQueueType {
queueOffset = startFrom.OffsetValue
cnt = 0
if c.nsqdCoord != nil {
l, queueOffset, cnt, err = c.nsqdCoord.SearchLogByMsgOffset(ch.GetTopicName(), ch.GetTopicPart(), queueOffset)
} else {
err = errors.New("Not supported while coordinator disabled")
}
} else if startFrom.OffsetType == offsetMsgCountType {
if c.nsqdCoord != nil {
l, queueOffset, cnt, err = c.nsqdCoord.SearchLogByMsgCnt(ch.GetTopicName(), ch.GetTopicPart(), startFrom.OffsetValue)
} else {
err = errors.New("Not supported while coordinator disabled")
}
} else {
nsqd.NsqLogger().Logf("not supported offset type:%v", startFrom)
err = errors.New("not supported offset type")
}
if err != nil {
nsqd.NsqLogger().Logf("failed to search the consume offset: %v, err:%v", startFrom, err)
return 0, 0, err
}
nsqd.NsqLogger().Logf("%v searched log : %v, offset: %v:%v", startFrom, l, queueOffset, cnt)
if c.nsqdCoord == nil {
err = ch.SetConsumeOffset(nsqd.BackendOffset(queueOffset), cnt, force)
if err != nil {
if err != nsqd.ErrSetConsumeOffsetNotFirstClient {
nsqd.NsqLogger().Logf("failed to set the consume offset: %v, err:%v", startFrom, err)
return 0, 0, err
}
nsqd.NsqLogger().Logf("the consume offset: %v can only be set by the first client", startFrom)
}
} else {
err = c.nsqdCoord.SetChannelConsumeOffsetToCluster(ch, queueOffset, cnt, force)
if err != nil {
if coordErr, ok := err.(*consistence.CommonCoordErr); ok {
if coordErr.IsEqual(consistence.ErrLocalSetChannelOffsetNotFirstClient) {
nsqd.NsqLogger().Logf("the consume offset: %v can only be set by the first client", startFrom)
return queueOffset, cnt, nil
}
}
nsqd.NsqLogger().Logf("failed to set the consume offset: %v (%v:%v), err: %v ", startFrom, queueOffset, cnt, err)
return 0, 0, err
}
}
return queueOffset, cnt, nil
}
func (p *protocolV2) messagePump(client *nsqd.ClientV2, startedChan chan bool,
stoppedChan chan bool) {
var err error
var buf bytes.Buffer
var clientMsgChan chan *nsqd.Message
var subChannel *nsqd.Channel
// NOTE: `flusherChan` is used to bound message latency for
// the pathological case of a channel on a low volume topic
// with >1 clients having >1 RDY counts
var flusherChan <-chan time.Time
var sampleRate int32
subEventChan := client.SubEventChan
identifyEventChan := client.IdentifyEventChan
outputBufferTicker := time.NewTicker(client.OutputBufferTimeout)
heartbeatTicker := time.NewTicker(client.HeartbeatInterval)
heartbeatChan := heartbeatTicker.C
heartbeatFailedCnt := 0
msgTimeout := client.MsgTimeout
// v2 opportunistically buffers data to clients to reduce write system calls
// we force flush in two cases:
// 1. when the client is not ready to receive messages
// 2. we're buffered and the channel has nothing left to send us
// (ie. we would block in this loop anyway)
//
flushed := true
// signal to the goroutine that started the messagePump
// that we've started up
close(startedChan)
for {
if subChannel == nil || !client.IsReadyForMessages() {
// the client is not ready to receive messages...
clientMsgChan = nil
flusherChan = nil
// force flush
client.LockWrite()
err = client.Flush()
client.UnlockWrite()
if err != nil {
goto exit
}
flushed = true
} else if flushed {
// last iteration we flushed...
// do not select on the flusher ticker channel
clientMsgChan = subChannel.GetClientMsgChan()
flusherChan = nil
} else {
// we're buffered (if there isn't any more data we should flush)...
// select on the flusher ticker channel, too
clientMsgChan = subChannel.GetClientMsgChan()
flusherChan = outputBufferTicker.C
}
select {
case <-client.ExitChan:
goto exit
case <-flusherChan:
// if this case wins, we're either starved
// or we won the race between other channels...
// in either case, force flush
client.LockWrite()
err = client.Flush()
client.UnlockWrite()
if err != nil {
goto exit
}
flushed = true
case <-client.ReadyStateChan:
case subChannel = <-subEventChan:
// you can't SUB anymore
nsqd.NsqLogger().Logf("client %v sub to channel: %v", client.ID,
subChannel.GetName())
subEventChan = nil
case identifyData := <-identifyEventChan:
// you can't IDENTIFY anymore
identifyEventChan = nil
outputBufferTicker.Stop()
if identifyData.OutputBufferTimeout > 0 {
outputBufferTicker = time.NewTicker(identifyData.OutputBufferTimeout)
}
heartbeatTicker.Stop()
heartbeatChan = nil
if identifyData.HeartbeatInterval > 0 {
heartbeatTicker = time.NewTicker(identifyData.HeartbeatInterval)
heartbeatChan = heartbeatTicker.C
}
if identifyData.SampleRate > 0 {
sampleRate = identifyData.SampleRate
}
msgTimeout = identifyData.MsgTimeout
case <-heartbeatChan:
if subChannel != nil && client.IsReadyForMessages() {
// try wake up the channel
subChannel.TryWakeupRead()
}
err = Send(client, frameTypeResponse, heartbeatBytes)
nsqd.NsqLogger().LogDebugf("PROTOCOL(V2): [%s] send heartbeat", client)
if err != nil {
heartbeatFailedCnt++
nsqd.NsqLogger().LogWarningf("PROTOCOL(V2): [%s] send heartbeat failed %v times, %v", client, heartbeatFailedCnt, err)
if heartbeatFailedCnt > 2 {
goto exit
}
} else {
heartbeatFailedCnt = 0
}
case msg, ok := <-clientMsgChan:
if !ok {
goto exit
}
if sampleRate > 0 && rand.Int31n(100) > sampleRate {
// FIN automatically, all message will not wait to confirm if not sending,
// and the reader keep moving forward.
offset, _, _, _ := subChannel.ConfirmBackendQueue(msg)
// TODO: sync to replica nodes.
_ = offset
continue
}
// avoid re-send some confirmed message,
// this may happen while the channel reader is reset to old position
// due to some retry or leader change.
if subChannel.IsConfirmed(msg) {
continue
}
subChannel.StartInFlightTimeout(msg, client.ID, client.String(), msgTimeout)
client.SendingMessage()
err = SendMessage(client, msg, &buf, subChannel.IsOrdered())
if err != nil {
goto exit
}
flushed = false
}
}
exit:
nsqd.NsqLogger().LogDebugf("PROTOCOL(V2): [%s] exiting messagePump", client)
heartbeatTicker.Stop()
outputBufferTicker.Stop()
if err != nil {
nsqd.NsqLogger().Logf("PROTOCOL(V2): [%s] messagePump error - %s", client, err)
}
close(stoppedChan)
}
func (self *NsqdCoordinator) updateChannelOffsetOnSlave(tc *coordData, channelName string, offset ChannelConsumerOffset) *CoordErr {
topicName := tc.topicInfo.Name
partition := tc.topicInfo.Partition
if !tc.IsMineISR(self.myNode.GetID()) {
return ErrTopicWriteOnNonISR
}
if coordLog.Level() >= levellogger.LOG_DETAIL {
coordLog.Debugf("got update channel(%v) offset on slave : %v", channelName, offset)
}
coord, coordErr := self.getTopicCoord(topicName, partition)
if coordErr != nil {
return ErrMissingTopicCoord
}
topic, localErr := self.localNsqd.GetExistingTopic(topicName, partition)
if localErr != nil {
coordLog.Warningf("slave missing topic : %v", topicName)
// TODO: leave the isr and try re-sync with leader
return &CoordErr{localErr.Error(), RpcCommonErr, CoordSlaveErr}
}
if topic.GetTopicPart() != partition {
coordLog.Errorf("topic on slave has different partition : %v vs %v", topic.GetTopicPart(), partition)
return ErrLocalMissingTopic
}
var ch *nsqd.Channel
ch, localErr = topic.GetExistingChannel(channelName)
// if a new channel on slave, we should set the consume offset by force
if localErr != nil {
offset.AllowBackward = true
ch = topic.GetChannel(channelName)
coordLog.Infof("slave init the channel : %v, %v, offset: %v", topic.GetTopicName(), channelName, ch.GetConfirmed())
}
if ch.IsEphemeral() {
coordLog.Errorf("ephemeral channel %v should not be synced on slave", channelName)
}
currentEnd := ch.GetChannelEnd()
if nsqd.BackendOffset(offset.VOffset) > currentEnd.Offset() {
coordLog.Debugf("update channel(%v) consume offset exceed end %v on slave : %v", channelName, offset, currentEnd)
// cache the offset (using map?) to reduce the slave channel flush.
coord.consumeMgr.Lock()
cur, ok := coord.consumeMgr.channelConsumeOffset[channelName]
if !ok || cur.VOffset < offset.VOffset {
coord.consumeMgr.channelConsumeOffset[channelName] = offset
}
coord.consumeMgr.Unlock()
if offset.Flush {
topic.ForceFlush()
currentEnd = ch.GetChannelEnd()
if nsqd.BackendOffset(offset.VOffset) > currentEnd.Offset() {
offset.VOffset = int64(currentEnd.Offset())
offset.VCnt = currentEnd.TotalMsgCnt()
}
} else {
return nil
}
}
err := ch.ConfirmBackendQueueOnSlave(nsqd.BackendOffset(offset.VOffset), offset.VCnt, offset.AllowBackward)
if err != nil {
coordLog.Warningf("update local channel(%v) offset %v failed: %v, current channel end: %v, topic end: %v",
channelName, offset, err, currentEnd, topic.TotalDataSize())
if err == nsqd.ErrExiting {
return &CoordErr{err.Error(), RpcNoErr, CoordTmpErr}
}
return &CoordErr{err.Error(), RpcCommonErr, CoordSlaveErr}
}
return nil
}
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
}
func (self *NsqdCoordinator) SetChannelConsumeOffsetToCluster(ch *nsqd.Channel, queueOffset int64, cnt int64, force bool) error {
topicName := ch.GetTopicName()
partition := ch.GetTopicPart()
coord, checkErr := self.getTopicCoord(topicName, partition)
if checkErr != nil {
return checkErr.ToErrorType()
}
var syncOffset ChannelConsumerOffset
syncOffset.AllowBackward = true
syncOffset.VCnt = cnt
syncOffset.VOffset = queueOffset
doLocalWrite := func(d *coordData) *CoordErr {
err := ch.SetConsumeOffset(nsqd.BackendOffset(queueOffset), cnt, force)
if err != nil {
if err != nsqd.ErrSetConsumeOffsetNotFirstClient {
coordLog.Infof("failed to set the consume offset: %v, err:%v", queueOffset, err)
return &CoordErr{err.Error(), RpcNoErr, CoordLocalErr}
}
coordLog.Debugf("the consume offset: %v can only be set by the first client", queueOffset)
return ErrLocalSetChannelOffsetNotFirstClient
}
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 {
if ch.IsEphemeral() {
return nil
}
rpcErr := c.UpdateChannelOffset(&tcData.topicLeaderSession, &tcData.topicInfo, ch.GetName(), syncOffset)
if rpcErr != nil {
coordLog.Infof("sync channel(%v) offset to replica %v failed: %v, offset: %v", ch.GetName(),
nodeID, rpcErr, syncOffset)
}
return rpcErr
}
handleSyncResult := func(successNum int, tcData *coordData) bool {
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
}