背景
某線上日誌收集服務報警，開啟域名報502錯誤碼。 收集服務由2臺netty HA伺服器組成,netty伺服器將客戶端投遞來的protobuf日誌解析併傳送到kafka，開啟其中一個應用的日誌，發現如下報錯：
org.apache.kafka.common.errors.TimeoutException: Expiring 1 record(s)
在排除了netty服務的錯誤之後，去檢視kafka的日誌。 發現報錯，排查過程如下；
配置資訊
| 系統 | kafka版本 | broker數量 |
|---|---|---|
| CentOS7.4 | 2.1.0 | 3 |
線上有三臺Kafka Broker，id分別為0、1、2，伺服器只部署了Kafka服務。
問題
執行緒是否存活
首先jps檢視Kafka執行緒是否存活，三臺機器都沒問題，kafka依然在執行。
GC問題
檢視kafkaServer-gc.log.1.current的日誌，gc日誌沒發現異常。
Broker 0/server.log
[2019-08-02 15:17:03,699] WARN Attempting to send response via channel for which there is no open connection, connection id 172.21.3.14:9092-172.21.3.11:54311-107706 (kafka.network.Processor)
[2019-08-02 15:19:12,490] INFO [GroupMetadataManager brokerId=0] Removed 0 expired offsets in 0 milliseconds. (kafka.coordinator.group.GroupMetadataManager)
[2019-08-02 15:26:54,405] INFO [ReplicaFetcher replicaId=0, leaderId=1, fetcherId=0] Error sending fetch request (sessionId=1112819217, epoch=1897450) to node 1: java.io.IOException: Connection to 1 was disconnected before the response was read. (org.apache.kafka.clients.FetchSessionHandler)
[2019-08-02 15:26:54,411] WARN [ReplicaFetcher replicaId=0, leaderId=1, fetcherId=0] Error in response for fetch request (type=FetchRequest, replicaId=0, maxWait=500, minBytes=1, maxBytes=10485760, fetchData={galaxy_client-7=(offset=15680912, logStartOffset=14755985, maxBytes=1048576, currentLeaderEpoch=Optional[9])}, isolationLevel=READ_UNCOMMITTED, toForget=, metadata=(sessionId=1112819217, epoch=1897450)) (kafka.server.ReplicaFetcherThread)
java.io.IOException: Connection to 1 was disconnected before the response was read
at org.apache.kafka.clients.NetworkClientUtils.sendAndReceive(NetworkClientUtils.java:97)
at kafka.server.ReplicaFetcherBlockingSend.sendRequest(ReplicaFetcherBlockingSend.scala:97)
at kafka.server.ReplicaFetcherThread.fetchFromLeader(ReplicaFetcherThread.scala:190)
at kafka.server.AbstractFetcherThread.kafka$server$AbstractFetcherThread$$processFetchRequest(AbstractFetcherThread.scala:241)
at kafka.server.AbstractFetcherThread$$anonfun$maybeFetch$1.apply(AbstractFetcherThread.scala:130)
at kafka.server.AbstractFetcherThread$$anonfun$maybeFetch$1.apply(AbstractFetcherThread.scala:129)
at scala.Option.foreach(Option.scala:257)
at kafka.server.AbstractFetcherThread.maybeFetch(AbstractFetcherThread.scala:129)
at kafka.server.AbstractFetcherThread.doWork(AbstractFetcherThread.scala:111)
at kafka.utils.ShutdownableThread.run(ShutdownableThread.scala:82)
[2019-08-02 15:27:26,433] INFO [ReplicaFetcher replicaId=0, leaderId=1, fetcherId=0] Error sending fetch request (sessionId=1112819217, epoch=INITIAL) to node 1: java.io.IOException: Connection to 1 was disconnected before the response was read. (org.apache.kafka.clients.FetchSessionHandler)
Broker 1/server.log
[2019-08-02 15:26:53,751] WARN [GroupCoordinator 1]: Failed to write empty metadata for group StoreToHiveV2: The group is rebalancing, so a rejoin is needed. (kafka.coordinator.group.GroupCoordinator)
[2019-08-02 15:26:55,515] WARN [GroupCoordinator 1]: Failed to write empty metadata for group LPmkt: The group is rebalancing, so a rejoin is needed. (kafka.coordinator.group.GroupCoordinator)
[2019-08-02 15:26:56,124] WARN [GroupCoordinator 1]: Failed to write empty metadata for group Store2CarbonClientServer: The group is rebalancing, so a rejoin is needed. (kafka.coordinator.group.GroupCoordinator)
[2019-08-02 15:26:56,575] WARN [GroupCoordinator 1]: Failed to write empty metadata for group StoreToHive: The group is rebalancing, so a rejoin is needed. (kafka.coordinator.group.GroupCoordinator)
Broker 2/server.log
[2019-08-02 15:26:54,514] WARN [ReplicaFetcher replicaId=2, leaderId=1, fetcherId=0] Error in response for fetch request (type=FetchRequest, replicaId=2, maxWait=500, minBytes=1, maxBytes=10485760, fetchData={galaxy_client-7=(offset=15680912, logStartOffset=14755985, maxBytes=1048576, currentLeaderEpoch=Optional[9])}, isolationLevel=READ_UNCOMMITTED, toForget=, metadata=(sessionId=1550207796, epoch=1890003)) (kafka.server.ReplicaFetcherThread)
java.io.IOException: Connection to 1 was disconnected before the response was read
at org.apache.kafka.clients.NetworkClientUtils.sendAndReceive(NetworkClientUtils.java:97)
at kafka.server.ReplicaFetcherBlockingSend.sendRequest(ReplicaFetcherBlockingSend.scala:97)
at kafka.server.ReplicaFetcherThread.fetchFromLeader(ReplicaFetcherThread.scala:190)
at kafka.server.AbstractFetcherThread.kafka$server$AbstractFetcherThread$$processFetchRequest(AbstractFetcherThread.scala:241)
at kafka.server.AbstractFetcherThread$$anonfun$maybeFetch$1.apply(AbstractFetcherThread.scala:130)
at kafka.server.AbstractFetcherThread$$anonfun$maybeFetch$1.apply(AbstractFetcherThread.scala:129)
at scala.Option.foreach(Option.scala:257)
at kafka.server.AbstractFetcherThread.maybeFetch(AbstractFetcherThread.scala:129)
at kafka.server.AbstractFetcherThread.doWork(AbstractFetcherThread.scala:111)
at kafka.utils.ShutdownableThread.run(ShutdownableThread.scala:82)
[2019-08-02 15:27:26,569] INFO [ReplicaFetcher replicaId=2, leaderId=1, fetcherId=0] Error sending fetch request (sessionId=1550207796, epoch=INITIAL) to node 1: java.io.IOException: Connection to 1 was disconnected before the response was read. (org.apache.kafka.clients.FetchSessionHandler)
問題排查
由於是線上應用，基本沒時間排查問題，所以立即重啟了三臺Kafka，重啟後得到緩解。
在服務正常之後，在網上搜了很多答案，有的人說是網路問題，close wait過多導致broker網路不通，從而Kafka產生腦裂問題。 在各個伺服器上執行命令：
netstat -n | awk '/^tcp/ {++S[$NF]} END {for(a in S) print a, S[a]}'
結果
ESTABLISHED 172
TIME_WAIT 32
也沒發現異常，等下次再出現問題時，要執行這個命令看看伺服器的狀態。
社群bug？
https://issues.apache.org/jira/browse/KAFKA-6582
下方有人評論：We are running 2.1.1 in production (and four other environments) since March without this issue showing again.
所以找時間升級Kafka版本吧~