func (this *WatchAppError) consumePartition(zkcluster *zk.ZkCluster, consumer sarama.Consumer,
topic string, partitionId int32, offset int64, msgCh chan<- *sarama.ConsumerMessage, wg *sync.WaitGroup) {
defer wg.Done()
p, err := consumer.ConsumePartition(topic, partitionId, offset)
if err != nil {
log.Error("%s %s/%d: offset=%d %v", zkcluster.Name(), topic, partitionId, offset, err)
return
}
defer p.Close()
for {
select {
case <-this.Stop:
return
case err := <-p.Errors():
log.Critical("cluster[%s] %s/%d: %s", zkcluster.Name(), topic, partitionId, err)
return
case msg := <-p.Messages():
if msg != nil && this.predicate(msg.Value) {
msgCh <- msg
}
}
}
}
func (this *Peek) consumePartition(zkcluster *zk.ZkCluster, kfk sarama.Client, consumer sarama.Consumer,
topic string, partitionId int32, msgCh chan *sarama.ConsumerMessage, offset int64) {
p, err := consumer.ConsumePartition(topic, partitionId, offset)
if err != nil {
this.Ui.Error(fmt.Sprintf("%s %s/%d: offset=%d %v", zkcluster.Name(), topic, partitionId, offset, err))
return
}
defer p.Close()
n := int64(0)
for {
select {
case <-this.quit:
return
case msg := <-p.Messages():
msgCh <- msg
n++
if this.lastN > 0 && n >= this.lastN {
return
}
}
}
}
func (this *Topics) clusterSummary(zkcluster *zk.ZkCluster) []topicSummary {
r := make([]topicSummary, 0, 10)
kfk, err := sarama.NewClient(zkcluster.BrokerList(), saramaConfig())
if err != nil {
this.Ui.Error(err.Error())
return nil
}
defer kfk.Close()
topicInfos, _ := kfk.Topics()
for _, t := range topicInfos {
flat := int64(0)
cum := int64(0)
alivePartitions, _ := kfk.WritablePartitions(t)
for _, partitionID := range alivePartitions {
latestOffset, _ := kfk.GetOffset(t, partitionID, sarama.OffsetNewest)
oldestOffset, _ := kfk.GetOffset(t, partitionID, sarama.OffsetOldest)
flat += (latestOffset - oldestOffset)
cum += latestOffset
}
r = append(r, topicSummary{zkcluster.ZkZone().Name(), zkcluster.Name(), t, len(alivePartitions), flat, cum})
}
return r
}
func (this *Top) clusterTopConsumers(zkcluster *zk.ZkCluster) {
var topic string
for {
total := zkcluster.TotalConsumerOffsets(this.topicPattern)
if this.topicPattern != "" {
topic = this.topicPattern
} else {
topic = "-all-"
}
key := zkcluster.Name() + ":" + topic
this.mu.Lock()
this.consumerCounters[key] = float64(total)
this.mu.Unlock()
if !this.dashboardGraph {
this.mu.Lock()
this.counters[key] = float64(total)
this.mu.Unlock()
}
time.Sleep(this.topInterval)
}
}
func (this *Mirror) makeMirror(c1, c2 *zk.ZkCluster) {
pub, err := this.makePub(c2)
swallow(err)
topics, topicsChanges, err := c1.WatchTopics()
swallow(err)
log.Printf("topics: %+v", topics)
if len(topics) == 0 {
log.Println("empty topics")
return
}
group := fmt.Sprintf("%s.%s._mirror_", c1.Name(), c2.Name())
sub, err := this.makeSub(c1, group, topics)
swallow(err)
pumpStopper := make(chan struct{})
go this.pump(sub, pub, pumpStopper)
LOOP:
for {
select {
case <-topicsChanges:
log.Println("topics changed, stopping pump...")
pumpStopper <- struct{}{} // stop pump
<-pumpStopper // await pump cleanup
// refresh c1 topics
topics, err = c1.Topics()
if err != nil {
// TODO how to handle this err?
log.Println(err)
}
log.Printf("topics: %+v", topics)
sub, err = this.makeSub(c1, group, topics)
if err != nil {
// TODO how to handle this err?
log.Println(err)
}
go this.pump(sub, pub, pumpStopper)
case <-this.quit:
log.Println("awaiting pump cleanup...")
<-pumpStopper
log.Printf("total transferred: %s %smsgs",
gofmt.ByteSize(this.transferBytes),
gofmt.Comma(this.transferN))
break LOOP
}
}
pub.Close()
}
func (this *TopBroker) clusterTopProducers(zkcluster *zk.ZkCluster) {
kfk, err := sarama.NewClient(zkcluster.BrokerList(), sarama.NewConfig())
if err != nil {
return
}
defer kfk.Close()
for {
hostOffsets := make(map[string]int64)
topics, err := kfk.Topics()
swallow(err)
<-this.signalsCh[zkcluster.Name()]
for _, topic := range topics {
if !patternMatched(topic, this.topic) {
continue
}
partions, err := kfk.WritablePartitions(topic)
swallow(err)
for _, partitionID := range partions {
leader, err := kfk.Leader(topic, partitionID)
swallow(err)
latestOffset, err := kfk.GetOffset(topic, partitionID,
sarama.OffsetNewest)
swallow(err)
host, _, err := net.SplitHostPort(leader.Addr())
swallow(err)
if this.shortIp {
host = shortIp(host)
}
if _, present := hostOffsets[host]; !present {
hostOffsets[host] = 0
}
hostOffsets[host] += latestOffset
}
}
this.hostOffsetCh <- hostOffsets
kfk.RefreshMetadata(topics...)
}
}
func (this *LsZk) printCluster(zkcluster *zk.ZkCluster) {
this.Ui.Output(color.Green(zkcluster.Name()))
children, err := zkcluster.ListChildren(this.recursive)
if err != nil {
this.Ui.Error(fmt.Sprintf("%s%s", strings.Repeat(" ", 4), err))
return
}
for _, c := range children {
this.Ui.Output(fmt.Sprintf("%s%s", strings.Repeat(" ", 4), c))
if strings.HasSuffix(c, "brokers") {
this.Ui.Output(fmt.Sprintf("%s%s/ids", strings.Repeat(" ", 4), c))
this.Ui.Output(fmt.Sprintf("%s%s/topics", strings.Repeat(" ", 4), c))
}
}
}
func (this *Lags) printConsumersLagTable(zkcluster *zk.ZkCluster) {
lines := make([]string, 0)
header := "ConsumerGroup|Topic/Partition|Produced|Consumed|Lag|Committed|Uptime"
lines = append(lines, header)
// sort by group name
consumersByGroup := zkcluster.ConsumersByGroup(this.groupPattern)
sortedGroups := make([]string, 0, len(consumersByGroup))
for group, _ := range consumersByGroup {
sortedGroups = append(sortedGroups, group)
}
sort.Strings(sortedGroups)
for _, group := range sortedGroups {
if !patternMatched(group, this.groupPattern) {
continue
}
sortedTopicAndPartitionIds := make([]string, 0, len(consumersByGroup[group]))
consumers := make(map[string]zk.ConsumerMeta)
for _, t := range consumersByGroup[group] {
key := fmt.Sprintf("%s:%s", t.Topic, t.PartitionId)
sortedTopicAndPartitionIds = append(sortedTopicAndPartitionIds, key)
consumers[key] = t
}
sort.Strings(sortedTopicAndPartitionIds)
for _, topicAndPartitionId := range sortedTopicAndPartitionIds {
consumer := consumers[topicAndPartitionId]
if !patternMatched(consumer.Topic, this.topicPattern) {
continue
}
if !consumer.Online {
continue
}
if this.problematicMode && consumer.Lag <= int64(this.lagThreshold) {
continue
}
if consumer.ConsumerZnode == nil {
this.Ui.Warn(fmt.Sprintf("%+v has no znode", consumer))
continue
}
lines = append(lines,
fmt.Sprintf("%s|%s/%s|%s|%s|%s|%s|%s",
group,
consumer.Topic, consumer.PartitionId,
gofmt.Comma(consumer.ProducerOffset),
gofmt.Comma(consumer.ConsumerOffset),
gofmt.Comma(consumer.Lag),
gofmt.PrettySince(consumer.Mtime.Time()),
gofmt.PrettySince(consumer.ConsumerZnode.Uptime())))
}
}
if len(lines) > 1 {
this.Ui.Info(fmt.Sprintf("%s ▾", zkcluster.Name()))
this.Ui.Output(columnize.SimpleFormat(lines))
}
}
func (this *UnderReplicated) displayUnderReplicatedPartitionsOfCluster(zkcluster *zk.ZkCluster) []string {
brokerList := zkcluster.BrokerList()
if len(brokerList) == 0 {
this.Ui.Warn(fmt.Sprintf("%s empty brokers", zkcluster.Name()))
return nil
}
kfk, err := sarama.NewClient(brokerList, saramaConfig())
if err != nil {
this.Ui.Error(fmt.Sprintf("%s %+v %s", zkcluster.Name(), brokerList, err.Error()))
return nil
}
defer kfk.Close()
topics, err := kfk.Topics()
swallow(err)
if len(topics) == 0 {
return nil
}
lines := make([]string, 0, 10)
for _, topic := range topics {
// get partitions and check if some dead
alivePartitions, err := kfk.WritablePartitions(topic)
if err != nil {
this.Ui.Error(fmt.Sprintf("%s topic[%s] cannot fetch writable partitions: %v", zkcluster.Name(), topic, err))
continue
}
partions, err := kfk.Partitions(topic)
if err != nil {
this.Ui.Error(fmt.Sprintf("%s topic[%s] cannot fetch partitions: %v", zkcluster.Name(), topic, err))
continue
}
if len(alivePartitions) != len(partions) {
this.Ui.Error(fmt.Sprintf("%s topic[%s] has %s partitions: %+v/%+v", zkcluster.Name(),
topic, color.Red("dead"), alivePartitions, partions))
}
for _, partitionID := range alivePartitions {
replicas, err := kfk.Replicas(topic, partitionID)
if err != nil {
this.Ui.Error(fmt.Sprintf("%s topic[%s] P:%d: %v", zkcluster.Name(), topic, partitionID, err))
continue
}
isr, isrMtime, partitionCtime := zkcluster.Isr(topic, partitionID)
underReplicated := false
if len(isr) != len(replicas) {
underReplicated = true
}
if underReplicated {
leader, err := kfk.Leader(topic, partitionID)
swallow(err)
latestOffset, err := kfk.GetOffset(topic, partitionID, sarama.OffsetNewest)
swallow(err)
oldestOffset, err := kfk.GetOffset(topic, partitionID, sarama.OffsetOldest)
swallow(err)
lines = append(lines, fmt.Sprintf("\t%s Partition:%d/%s Leader:%d Replicas:%+v Isr:%+v/%s Offset:%d-%d Num:%d",
topic, partitionID,
gofmt.PrettySince(partitionCtime),
leader.ID(), replicas, isr,
gofmt.PrettySince(isrMtime),
oldestOffset, latestOffset, latestOffset-oldestOffset))
}
}
}
return lines
}
func (this *Mirror) runMirror(c1, c2 *zk.ZkCluster, limit int64) {
this.startedAt = time.Now()
log.Info("start [%s/%s] -> [%s/%s] with bandwidth %sbps",
c1.ZkZone().Name(), c1.Name(),
c2.ZkZone().Name(), c2.Name(),
gofmt.Comma(limit*8))
pub, err := this.makePub(c2)
if err != nil {
panic(err)
}
log.Trace("pub[%s/%s] created", c2.ZkZone().Name(), c2.Name())
go func(pub sarama.AsyncProducer, c *zk.ZkCluster) {
for {
select {
case <-this.quit:
return
case err := <-pub.Errors():
// messages will only be returned here after all retry attempts are exhausted.
//
// e,g
// Failed to produce message to topic xx: write tcp src->kfk: i/o timeout
// kafka: broker not connected
log.Error("pub[%s/%s] %v", c.ZkZone().Name(), c.Name(), err)
}
}
}(pub, c2)
group := this.groupName(c1, c2)
ever := true
round := 0
for ever {
round++
topics, topicsChanges, err := c1.WatchTopics()
if err != nil {
log.Error("#%d [%s/%s]watch topics: %v", round, c1.ZkZone().Name(), c1.Name(), err)
time.Sleep(time.Second * 10)
continue
}
topics = this.realTopics(topics)
sub, err := this.makeSub(c1, group, topics)
if err != nil {
log.Error("#%d [%s/%s] %v", round, c1.ZkZone().Name(), c1.Name(), err)
time.Sleep(time.Second * 10)
continue
}
log.Info("#%d starting pump [%s/%s] -> [%s/%s] %d topics with group %s for %+v", round,
c1.ZkZone().Name(), c1.Name(),
c2.ZkZone().Name(), c2.Name(), len(topics), group, topics)
pumpStopper := make(chan struct{})
pumpStopped := make(chan struct{})
go this.pump(sub, pub, pumpStopper, pumpStopped)
select {
case <-topicsChanges:
// TODO log the diff the topics
log.Warn("#%d [%s/%s] topics changed, stopping pump...", round, c1.Name(), c2.Name())
pumpStopper <- struct{}{} // stop pump
<-pumpStopped // await pump cleanup
case <-this.quit:
log.Info("#%d awaiting pump cleanup...", round)
<-pumpStopped
ever = false
case <-pumpStopped:
// pump encounters problems, just retry
}
}
log.Info("total transferred: %s %smsgs",
gofmt.ByteSize(this.transferBytes),
gofmt.Comma(this.transferN))
log.Info("closing pub...")
pub.Close()
}
func (this *Mirror) groupName(c1, c2 *zk.ZkCluster) string {
return fmt.Sprintf("_mirror_.%s.%s.%s.%s", c1.ZkZone().Name(), c1.Name(), c2.ZkZone().Name(), c2.Name())
}
func (this *Topics) displayTopicsOfCluster(zkcluster *zk.ZkCluster) {
echoBuffer := func(lines []string) {
for _, l := range lines {
this.Ui.Output(l)
}
}
linesInTopicMode := make([]string, 0)
if this.verbose {
linesInTopicMode = this.echoOrBuffer(zkcluster.Name(), linesInTopicMode)
}
// get all alive brokers within this cluster
brokers := zkcluster.Brokers()
if len(brokers) == 0 {
linesInTopicMode = this.echoOrBuffer(fmt.Sprintf("%4s%s", " ",
color.Red("%s empty brokers", zkcluster.Name())), linesInTopicMode)
echoBuffer(linesInTopicMode)
return
}
if this.verbose {
sortedBrokerIds := make([]string, 0, len(brokers))
for brokerId, _ := range brokers {
sortedBrokerIds = append(sortedBrokerIds, brokerId)
}
sort.Strings(sortedBrokerIds)
for _, brokerId := range sortedBrokerIds {
if this.ipInNumber {
linesInTopicMode = this.echoOrBuffer(fmt.Sprintf("%4s%s %s", " ",
color.Green(brokerId), brokers[brokerId]), linesInTopicMode)
} else {
linesInTopicMode = this.echoOrBuffer(fmt.Sprintf("%4s%s %s", " ",
color.Green(brokerId), brokers[brokerId].NamedString()), linesInTopicMode)
}
}
}
kfk, err := sarama.NewClient(zkcluster.BrokerList(), saramaConfig())
if err != nil {
if this.verbose {
linesInTopicMode = this.echoOrBuffer(color.Yellow("%5s%+v %s", " ",
zkcluster.BrokerList(), err.Error()), linesInTopicMode)
}
return
}
defer kfk.Close()
topics, err := kfk.Topics()
swallow(err)
if len(topics) == 0 {
if this.topicPattern == "" && this.verbose {
linesInTopicMode = this.echoOrBuffer(fmt.Sprintf("%5s%s", " ",
color.Magenta("no topics")), linesInTopicMode)
echoBuffer(linesInTopicMode)
}
return
}
sortedTopics := make([]string, 0, len(topics))
for _, t := range topics {
sortedTopics = append(sortedTopics, t)
}
sort.Strings(sortedTopics)
topicsCtime := zkcluster.TopicsCtime()
hasTopicMatched := false
for _, topic := range sortedTopics {
if !patternMatched(topic, this.topicPattern) {
continue
}
if this.since > 0 && time.Since(topicsCtime[topic]) > this.since {
continue
}
this.topicN++
hasTopicMatched = true
if this.verbose {
linesInTopicMode = this.echoOrBuffer(strings.Repeat(" ", 4)+color.Cyan(topic), linesInTopicMode)
}
// get partitions and check if some dead
alivePartitions, err := kfk.WritablePartitions(topic)
swallow(err)
partions, err := kfk.Partitions(topic)
swallow(err)
if len(alivePartitions) != len(partions) {
linesInTopicMode = this.echoOrBuffer(fmt.Sprintf("%30s %s %s P: %s/%+v",
zkcluster.Name(), color.Cyan("%-50s", topic), color.Red("partial dead"), color.Green("%+v", alivePartitions), partions), linesInTopicMode)
}
replicas, err := kfk.Replicas(topic, partions[0])
if err != nil {
this.Ui.Error(fmt.Sprintf("%s/%d %v", topic, partions[0], err))
}
this.partitionN += len(partions)
if !this.verbose {
linesInTopicMode = this.echoOrBuffer(fmt.Sprintf("%30s %s %3dP %dR %s",
zkcluster.Name(),
color.Cyan("%-50s", topic),
len(partions), len(replicas),
gofmt.PrettySince(topicsCtime[topic])), linesInTopicMode)
continue
}
for _, partitionID := range alivePartitions {
leader, err := kfk.Leader(topic, partitionID)
swallow(err)
replicas, err := kfk.Replicas(topic, partitionID)
if err != nil {
this.Ui.Error(fmt.Sprintf("%s/%d %v", topic, partitionID, err))
}
isr, isrMtime, partitionCtime := zkcluster.Isr(topic, partitionID)
isrMtimeSince := gofmt.PrettySince(isrMtime)
if time.Since(isrMtime).Hours() < 24 {
// ever out of sync last 24h
isrMtimeSince = color.Magenta(isrMtimeSince)
}
underReplicated := false
if len(isr) != len(replicas) {
underReplicated = true
}
latestOffset, err := kfk.GetOffset(topic, partitionID,
sarama.OffsetNewest)
swallow(err)
oldestOffset, err := kfk.GetOffset(topic, partitionID,
sarama.OffsetOldest)
swallow(err)
if this.count > 0 && (latestOffset-oldestOffset) < this.count {
continue
}
this.totalMsgs += latestOffset - oldestOffset
this.totalOffsets += latestOffset
if !underReplicated {
linesInTopicMode = this.echoOrBuffer(fmt.Sprintf("%8d Leader:%s Replicas:%+v Isr:%+v Offset:%16s - %-16s Num:%-15s %s-%s",
partitionID,
color.Green("%d", leader.ID()), replicas, isr,
gofmt.Comma(oldestOffset), gofmt.Comma(latestOffset), gofmt.Comma(latestOffset-oldestOffset),
gofmt.PrettySince(partitionCtime), isrMtimeSince), linesInTopicMode)
} else {
// use red for alert
linesInTopicMode = this.echoOrBuffer(fmt.Sprintf("%8d Leader:%s Replicas:%+v Isr:%s Offset:%16s - %-16s Num:%-15s %s-%s",
partitionID,
color.Green("%d", leader.ID()), replicas, color.Red("%+v", isr),
gofmt.Comma(oldestOffset), gofmt.Comma(latestOffset), gofmt.Comma(latestOffset-oldestOffset),
gofmt.PrettySince(partitionCtime), isrMtimeSince), linesInTopicMode)
}
}
}
if this.topicPattern != "" {
if hasTopicMatched {
echoBuffer(linesInTopicMode)
}
} else {
echoBuffer(linesInTopicMode)
}
}
func (this *Top) clusterTopProducers(zkcluster *zk.ZkCluster) {
cluster := zkcluster.Name()
brokerList := zkcluster.BrokerList()
if len(brokerList) == 0 {
return
}
kfk, err := sarama.NewClient(brokerList, sarama.NewConfig())
if err != nil {
return
}
defer kfk.Close()
for {
topics, err := kfk.Topics()
if err != nil || len(topics) == 0 {
return
}
for _, topic := range topics {
if !patternMatched(topic, this.topicPattern) {
continue
}
msgs := int64(0)
alivePartitions, err := kfk.WritablePartitions(topic)
swallow(err)
for _, partitionID := range alivePartitions {
latestOffset, err := kfk.GetOffset(topic, partitionID,
sarama.OffsetNewest)
if err != nil {
// this broker is down
continue
}
msgs += latestOffset
}
this.mu.Lock()
if _, present := this.brokers[cluster+":"+topic]; !present {
// calculate the broker leading partitions
leadingPartitions := make(map[string]int) // broker:lead partitions n
for _, pid := range alivePartitions {
leader, err := kfk.Leader(topic, pid)
swallow(err)
leadingPartitions[leader.Addr()]++
}
brokers := make([]string, 0)
for addr, n := range leadingPartitions {
brokers = append(brokers, fmt.Sprintf("%d@%s", n, addr))
}
this.brokers[cluster+":"+topic] = this.discardPortOfBrokerAddr(brokers)
}
this.counters[cluster+":"+topic] = float64(msgs)
this.partitions[cluster+":"+topic] = len(alivePartitions)
this.mu.Unlock()
}
time.Sleep(time.Second)
kfk.RefreshMetadata(topics...)
}
}
