func TestStats(t *testing.T) {
opts := nsqdNs.NewOptions()
opts.Logger = newTestLogger(t)
tcpAddr, _, nsqd, nsqdServer := mustStartNSQD(opts)
defer os.RemoveAll(opts.DataPath)
defer nsqdServer.Exit()
topicName := "test_stats" + strconv.Itoa(int(time.Now().Unix()))
topic := nsqd.GetTopicIgnPart(topicName)
msg := nsqdNs.NewMessage(0, []byte("test body"))
topic.PutMessage(msg)
conn, err := mustConnectNSQD(tcpAddr)
test.Equal(t, err, nil)
defer conn.Close()
identify(t, conn, nil, frameTypeResponse)
sub(t, conn, topicName, "ch")
stats := nsqd.GetStats(false)
t.Logf("stats: %+v", stats)
test.Equal(t, len(stats), 1)
test.Equal(t, len(stats[0].Channels), 1)
test.Equal(t, len(stats[0].Channels[0].Clients), 1)
}
func readMPUB(r io.Reader, tmp []byte, topic *nsqd.Topic, maxMessageSize int64, traceEnable bool) ([]*nsqd.Message, []*bytes.Buffer, error) {
numMessages, err := readLen(r, tmp)
if err != nil {
return nil, nil, protocol.NewFatalClientErr(err, "E_BAD_BODY", "MPUB failed to read message count")
}
if numMessages <= 0 {
return nil, nil, protocol.NewFatalClientErr(err, "E_BAD_BODY",
fmt.Sprintf("MPUB invalid message count %d", numMessages))
}
messages := make([]*nsqd.Message, 0, numMessages)
buffers := make([]*bytes.Buffer, 0, numMessages)
for i := int32(0); i < numMessages; i++ {
messageSize, err := readLen(r, tmp)
if err != nil {
return nil, buffers, protocol.NewFatalClientErr(err, "E_BAD_MESSAGE",
fmt.Sprintf("MPUB failed to read message(%d) body size", i))
}
if messageSize <= 0 {
return nil, buffers, protocol.NewFatalClientErr(nil, "E_BAD_MESSAGE",
fmt.Sprintf("MPUB invalid message(%d) body size %d", i, messageSize))
}
if int64(messageSize) > maxMessageSize {
return nil, buffers, protocol.NewFatalClientErr(nil, "E_BAD_MESSAGE",
fmt.Sprintf("MPUB message too big %d > %d", messageSize, maxMessageSize))
}
b := topic.BufferPoolGet(int(messageSize))
buffers = append(buffers, b)
_, err = io.CopyN(b, r, int64(messageSize))
if err != nil {
return nil, buffers, protocol.NewFatalClientErr(err, "E_BAD_MESSAGE", "MPUB failed to read message body")
}
msgBody := b.Bytes()[:messageSize]
traceID := uint64(0)
var realBody []byte
if traceEnable {
if messageSize <= nsqd.MsgTraceIDLength {
return nil, buffers, protocol.NewFatalClientErr(nil, "E_BAD_MESSAGE",
fmt.Sprintf("MPUB invalid message(%d) body size %d for tracing", i, messageSize))
}
traceID = binary.BigEndian.Uint64(msgBody[:nsqd.MsgTraceIDLength])
realBody = msgBody[nsqd.MsgTraceIDLength:]
} else {
realBody = msgBody
}
msg := nsqd.NewMessage(0, realBody)
msg.TraceID = traceID
messages = append(messages, msg)
topic.GetDetailStats().UpdateTopicMsgStats(int64(len(realBody)), 0)
}
return messages, buffers, nil
}
func (c *context) PutMessage(topic *nsqd.Topic,
msg []byte, traceID uint64) (nsqd.MessageID, nsqd.BackendOffset, int32, nsqd.BackendQueueEnd, error) {
if c.nsqdCoord == nil {
msg := nsqd.NewMessage(0, msg)
msg.TraceID = traceID
return topic.PutMessage(msg)
}
return c.nsqdCoord.PutMessageToCluster(topic, msg, traceID)
}
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 (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) 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
}
func (s *httpServer) doMPUB(w http.ResponseWriter, req *http.Request, ps httprouter.Params) (interface{}, error) {
startPub := time.Now().UnixNano()
if req.ContentLength > s.ctx.getOpts().MaxBodySize {
return nil, http_api.Err{413, "BODY_TOO_BIG"}
}
reqParams, topic, err := s.getExistingTopicFromQuery(req)
if err != nil {
return nil, err
}
var msgs []*nsqd.Message
var buffers []*bytes.Buffer
var exit bool
_, ok := reqParams["binary"]
if ok {
tmp := make([]byte, 4)
msgs, buffers, err = readMPUB(req.Body, tmp, topic,
s.ctx.getOpts().MaxMsgSize, false)
defer func() {
for _, b := range buffers {
topic.BufferPoolPut(b)
}
}()
if err != nil {
return nil, http_api.Err{413, err.(*protocol.FatalClientErr).Code[2:]}
}
} else {
// add 1 so that it's greater than our max when we test for it
// (LimitReader returns a "fake" EOF)
readMax := s.ctx.getOpts().MaxBodySize + 1
rdr := nsqd.NewBufioReader(io.LimitReader(req.Body, readMax))
defer nsqd.PutBufioReader(rdr)
total := 0
for !exit {
var block []byte
block, err = rdr.ReadBytes('\n')
if err != nil {
if err != io.EOF {
return nil, http_api.Err{500, "INTERNAL_ERROR"}
}
exit = true
}
total += len(block)
if int64(total) == readMax {
return nil, http_api.Err{413, "BODY_TOO_BIG"}
}
if len(block) > 0 && block[len(block)-1] == '\n' {
block = block[:len(block)-1]
}
// silently discard 0 length messages
// this maintains the behavior pre 0.2.22
if len(block) == 0 {
continue
}
if int64(len(block)) > s.ctx.getOpts().MaxMsgSize {
return nil, http_api.Err{413, "MSG_TOO_BIG"}
}
msg := nsqd.NewMessage(0, block)
msgs = append(msgs, msg)
topic.GetDetailStats().UpdateTopicMsgStats(int64(len(block)), 0)
}
}
if s.ctx.checkForMasterWrite(topic.GetTopicName(), topic.GetTopicPart()) {
_, _, _, err := s.ctx.PutMessages(topic, msgs)
//s.ctx.setHealth(err)
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()}
}
} else {
//should we forward to master of topic?
nsqd.NsqLogger().LogDebugf("should put to master: %v, from %v",
topic.GetFullName(), req.RemoteAddr)
topic.DisableForSlave()
return nil, http_api.Err{400, FailedOnNotLeader}
}
cost := time.Now().UnixNano() - startPub
topic.GetDetailStats().UpdateTopicMsgStats(0, cost/1000/int64(len(msgs)))
return "OK", nil
}