func (s *nsqdCoordGRpcServer) PutMessages(ctx context.Context, req *pb.RpcPutMessages) (*pb.CoordErr, error) {
var coordErr pb.CoordErr
if coordLog.Level() >= levellogger.LOG_DEBUG {
s := time.Now().Unix()
defer func() {
e := time.Now().Unix()
if e-s > int64(RPC_TIMEOUT/2) {
coordLog.Infof("PutMessage rpc call used: %v", e-s)
}
}()
}
tc, err := s.nsqdCoord.checkWriteForGRpcCall(req.TopicData)
if err != nil {
coordErr.ErrMsg = err.ErrMsg
coordErr.ErrCode = int32(err.ErrCode)
coordErr.ErrType = int32(err.ErrType)
return &coordErr, nil
}
// do local pub message
var commitData CommitLogData
commitData.Epoch = EpochType(req.LogData.Epoch)
commitData.LogID = req.LogData.LogID
commitData.MsgNum = req.LogData.MsgNum
commitData.MsgCnt = req.LogData.MsgCnt
commitData.MsgSize = req.LogData.MsgSize
commitData.MsgOffset = req.LogData.MsgOffset
commitData.LastMsgLogID = req.LogData.LastMsgLogID
var msgs []*nsqd.Message
for _, pbm := range req.TopicMessage {
var msg nsqd.Message
msg.ID = nsqd.MessageID(pbm.ID)
msg.TraceID = pbm.Trace_ID
msg.Attempts = uint16(pbm.Attemps)
msg.Timestamp = pbm.Timestamp
msg.Body = pbm.Body
msgs = append(msgs, &msg)
}
err = s.nsqdCoord.putMessagesOnSlave(tc, commitData, msgs)
if err != nil {
coordErr.ErrMsg = err.ErrMsg
coordErr.ErrCode = int32(err.ErrCode)
coordErr.ErrType = int32(err.ErrType)
}
return &coordErr, nil
}
func TestChannelEmptyConsumer(t *testing.T) {
opts := nsqdNs.NewOptions()
opts.Logger = newTestLogger(t)
tcpAddr, _, nsqd, nsqdServer := mustStartNSQD(opts)
defer os.RemoveAll(opts.DataPath)
defer nsqdServer.Exit()
conn, _ := mustConnectNSQD(tcpAddr)
defer conn.Close()
topicName := "test_channel_empty" + strconv.Itoa(int(time.Now().Unix()))
topic := nsqd.GetTopicIgnPart(topicName)
channel := topic.GetChannel("channel")
client := nsqdNs.NewClientV2(0, conn, opts, nil)
client.SetReadyCount(25)
channel.AddClient(client.ID, client)
for i := 0; i < 25; i++ {
msg := nsqdNs.NewMessage(nsqdNs.MessageID(i), []byte("test"))
channel.StartInFlightTimeout(msg, 0, "", opts.MsgTimeout)
client.SendingMessage()
}
for _, cl := range channel.GetClients() {
stats := cl.Stats()
test.Equal(t, stats.InFlightCount, int64(25))
}
channel.SetConsumeOffset(channel.GetChannelEnd().Offset(), channel.GetChannelEnd().TotalMsgCnt(), true)
time.Sleep(time.Second)
for _, cl := range channel.GetClients() {
stats := cl.Stats()
test.Equal(t, stats.InFlightCount, int64(0))
}
}
func (p *protocolV2) internalPubAndTrace(client *nsqd.ClientV2, params [][]byte, traceEnable bool) ([]byte, error) {
startPub := time.Now().UnixNano()
bodyLen, topic, err := p.preparePub(client, params, p.ctx.getOpts().MaxMsgSize)
if err != nil {
return nil, err
}
if traceEnable && bodyLen <= nsqd.MsgTraceIDLength {
return nil, protocol.NewFatalClientErr(nil, "E_BAD_BODY",
fmt.Sprintf("invalid body size %d with trace id enabled", bodyLen))
}
messageBodyBuffer := topic.BufferPoolGet(int(bodyLen))
defer topic.BufferPoolPut(messageBodyBuffer)
asyncAction := shouldHandleAsync(client, params)
_, err = io.CopyN(messageBodyBuffer, client.Reader, int64(bodyLen))
if err != nil {
return nil, protocol.NewFatalClientErr(err, "E_BAD_MESSAGE", "failed to read message body")
}
messageBody := messageBodyBuffer.Bytes()[:bodyLen]
topicName := topic.GetTopicName()
partition := topic.GetTopicPart()
var traceID uint64
var realBody []byte
if traceEnable {
traceID = binary.BigEndian.Uint64(messageBody[:nsqd.MsgTraceIDLength])
realBody = messageBody[nsqd.MsgTraceIDLength:]
} else {
realBody = messageBody
}
if p.ctx.checkForMasterWrite(topicName, partition) {
id := nsqd.MessageID(0)
offset := nsqd.BackendOffset(0)
rawSize := int32(0)
if asyncAction {
err = internalPubAsync(client.PubTimeout, messageBodyBuffer, topic)
} else {
id, offset, rawSize, _, err = p.ctx.PutMessage(topic, realBody, traceID)
}
//p.ctx.setHealth(err)
if err != nil {
topic.GetDetailStats().UpdatePubClientStats(client.RemoteAddr().String(), client.UserAgent, "tcp", 1, true)
nsqd.NsqLogger().LogErrorf("topic %v put message failed: %v", topic.GetFullName(), err)
if clusterErr, ok := err.(*consistence.CommonCoordErr); ok {
if !clusterErr.IsLocalErr() {
return nil, protocol.NewClientErr(err, FailedOnNotWritable, "")
}
}
return nil, protocol.NewClientErr(err, "E_PUB_FAILED", err.Error())
}
topic.GetDetailStats().UpdatePubClientStats(client.RemoteAddr().String(), client.UserAgent, "tcp", 1, false)
cost := time.Now().UnixNano() - startPub
topic.GetDetailStats().UpdateTopicMsgStats(int64(len(realBody)), cost/1000)
if !traceEnable {
return okBytes, nil
}
return getTracedReponse(messageBodyBuffer, id, traceID, offset, rawSize)
} else {
topic.GetDetailStats().UpdatePubClientStats(client.RemoteAddr().String(), client.UserAgent, "tcp", 1, true)
//forward to master of topic
nsqd.NsqLogger().LogDebugf("should put to master: %v, from %v",
topic.GetFullName(), client.RemoteAddr)
topic.DisableForSlave()
return nil, protocol.NewClientErr(err, FailedOnNotLeader, "")
}
}
func (self *NsqdCoordinator) PutMessagesToCluster(topic *nsqd.Topic,
msgs []*nsqd.Message) (nsqd.MessageID, nsqd.BackendOffset, int32, error) {
var commitLog CommitLogData
topicName := topic.GetTopicName()
partition := topic.GetTopicPart()
coord, checkErr := self.getTopicCoord(topicName, partition)
if checkErr != nil {
return nsqd.MessageID(commitLog.LogID), nsqd.BackendOffset(commitLog.MsgOffset), commitLog.MsgSize, checkErr.ToErrorType()
}
var queueEnd nsqd.BackendQueueEnd
var logMgr *TopicCommitLogMgr
doLocalWrite := func(d *coordData) *CoordErr {
topic.Lock()
logMgr = d.logMgr
id, offset, writeBytes, totalCnt, qe, localErr := topic.PutMessagesNoLock(msgs)
queueEnd = qe
topic.Unlock()
if localErr != nil {
coordLog.Warningf("put batch messages 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 = int64(msgs[len(msgs)-1].ID)
commitLog.MsgOffset = int64(offset)
commitLog.MsgSize = writeBytes
// This MsgCnt is the total count until now (include the current written batch message count)
commitLog.MsgCnt = totalCnt
commitLog.MsgNum = int32(len(msgs))
return nil
}
doLocalExit := func(err *CoordErr) {
if err != nil {
coordLog.Infof("topic %v PutMessagesToCluster error: %v", topic.GetFullName(), 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.ResetBackendEndNoLock(nsqd.BackendOffset(commitLog.MsgOffset), commitLog.MsgCnt-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.PutMessages(&tcData.topicLeaderSession, &tcData.topicInfo, commitLog, msgs)
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 nsqd.MessageID(commitLog.LogID), nsqd.BackendOffset(commitLog.MsgOffset), commitLog.MsgSize, err
}
func (s *httpServer) internalPUB(w http.ResponseWriter, req *http.Request, ps httprouter.Params, enableTrace bool) (interface{}, error) {
startPub := time.Now().UnixNano()
// do not support chunked for http pub, use tcp pub instead.
if req.ContentLength > s.ctx.getOpts().MaxMsgSize {
return nil, http_api.Err{413, "MSG_TOO_BIG"}
} else if req.ContentLength <= 0 {
return nil, http_api.Err{406, "MSG_EMPTY"}
}
// add 1 so that it's greater than our max when we test for it
// (LimitReader returns a "fake" EOF)
params, topic, err := s.getExistingTopicFromQuery(req)
if err != nil {
nsqd.NsqLogger().Logf("get topic err: %v", err)
return nil, http_api.Err{404, E_TOPIC_NOT_EXIST}
}
readMax := req.ContentLength + 1
b := topic.BufferPoolGet(int(req.ContentLength))
defer topic.BufferPoolPut(b)
asyncAction := !enableTrace
n, err := io.CopyN(b, io.LimitReader(req.Body, readMax), int64(req.ContentLength))
body := b.Bytes()[:req.ContentLength]
if err != nil {
nsqd.NsqLogger().Logf("read request body error: %v", err)
body = body[:n]
if err == io.EOF || err == io.ErrUnexpectedEOF {
// we ignore EOF, maybe the ContentLength is not match?
nsqd.NsqLogger().LogWarningf("read request body eof: %v, ContentLength: %v,return length %v.",
err, req.ContentLength, n)
} else {
return nil, http_api.Err{500, "INTERNAL_ERROR"}
}
}
if len(body) == 0 {
return nil, http_api.Err{406, "MSG_EMPTY"}
}
if s.ctx.checkForMasterWrite(topic.GetTopicName(), topic.GetTopicPart()) {
var err error
traceIDStr := params.Get("trace_id")
traceID, err := strconv.ParseUint(traceIDStr, 10, 0)
if enableTrace && err != nil {
nsqd.NsqLogger().Logf("trace id invalid %v, %v",
traceIDStr, err)
return nil, http_api.Err{400, "INVALID_TRACE_ID"}
}
id := nsqd.MessageID(0)
offset := nsqd.BackendOffset(0)
rawSize := int32(0)
if asyncAction {
err = internalPubAsync(nil, b, topic)
} else {
id, offset, rawSize, _, err = s.ctx.PutMessage(topic, body, traceID)
}
if err != nil {
nsqd.NsqLogger().LogErrorf("topic %v put message failed: %v", topic.GetFullName(), err)
if clusterErr, ok := err.(*consistence.CommonCoordErr); ok {
if !clusterErr.IsLocalErr() {
return nil, http_api.Err{400, FailedOnNotWritable}
}
}
return nil, http_api.Err{503, err.Error()}
}
cost := time.Now().UnixNano() - startPub
topic.GetDetailStats().UpdateTopicMsgStats(int64(len(body)), cost/1000)
if enableTrace {
return struct {
Status string `json:"status"`
ID uint64 `json:"id"`
TraceID string `json:"trace_id"`
QueueOffset uint64 `json:"queue_offset"`
DataRawSize uint32 `json:"rawsize"`
}{"OK", uint64(id), traceIDStr, uint64(offset), uint32(rawSize)}, nil
} else {
return "OK", nil
}
} else {
nsqd.NsqLogger().LogDebugf("should put to master: %v, from %v",
topic.GetFullName(), req.RemoteAddr)
topic.DisableForSlave()
return nil, http_api.Err{400, FailedOnNotLeader}
}
}