Hi Chris,
>From the logs it seems your controller machine id=2 has some soft failure,
such as a long GC which casued it to session timeout from ZK. Then when it
is shutting down it hits a known bug (related to KAFKA-1311) in 0.8.1 which
will cause it to hang on controller shutdown. We are currently releasing a
new version 0.8.1.1 with the fix to this bug.
Thanks,
Guozhang
On Thu, Apr 3, 2014 at 5:11 AM, Krzysztof Ociepa <ociepa.krzysz...@gmail.com
> wrote:
> Hi Team,
>
> I test Kafka 0.8.1 and have problems with missing brokers. What I did:
>
> 1. Installed Zookeeper 3.3.6 servers on 3 dedicated machines
> 2. Installed Kafka 0.8.1 brokers on 3 machines (different from Zookeeper
> ones)
> 3. Configured brokers as above:
> # Replication configurations
> num.replica.fetchers=2
> replica.fetch.max.bytes=1048576
> replica.fetch.wait.max.ms=500
> replica.high.watermark.checkpoint.interval.ms=5000
> replica.socket.timeout.ms=30000
> replica.socket.receive.buffer.bytes=65536
> replica.lag.time.max.ms=10000
> replica.lag.max.messages=4000
> controller.socket.timeout.ms=30000
> controller.message.queue.size=10
> controlled.shutdown.enable=true
> auto.leader.rebalance.enable=true
> leader.imbalance.check.interval.seconds=1200
>
> # ZK configuration
> zookeeper.connection.timeout.ms=6000
> zookeeper.sync.time.ms=3000
>
> 4. Started Zookeeper and brokers. Created topics. In Zookeeper client I
> saw:
> ls /brokers/ids
> [3, 2, 1]
>
> 5. Ran producers and consumers with very low traffic. All worked fine.
> 6. Increased traffic to ~100 messages per second and the controller
> (ID=2 / machine 2) failed with broker on the same machine (I've seen few
> times this scenario on my cluster).
>
> My logs as following.
> Machine 2:
>
> Controller ID=2:
> [2014-04-02 10:38:37,152] INFO
> [Controller-2-to-broker-3-send-thread], Starting
> (kafka.controller.RequestSendThread)
> [2014-04-02 10:42:21,131] INFO [SessionExpirationListener on 2], ZK
> expired; shut down all controller components and try to re-elect
> (kafka.controller.KafkaController$SessionExpirationListener)
> [2014-04-02 10:42:21,176] INFO [delete-topics-thread], Shutting down
> (kafka.controller.TopicDeletionManager$DeleteTopicsThread)
>
> Server ID=2:
> [2014-04-02 10:38:39,113] INFO Partition [topic1,1] on broker 2:
> Expanding ISR for partition [topic1,1] from 2,1 to 2,1,3
> (kafka.cluster.Partition)
> [2014-04-02 10:38:39,129] INFO Partition [topic2,1] on broker 2:
> Expanding ISR for partition [topic2,1] from 2,1 to 2,1,3
> (kafka.cluster.Partition)
> [2014-04-02 10:38:39,488] INFO Closing socket connection to
> /11.111.1.19. (kafka.network.Processor)
> [2014-04-02 10:38:45,343] INFO Closing socket connection to
> /11.111.1.19. (kafka.network.Processor)
> [2014-04-02 10:39:21,842] INFO Closing socket connection to
> /11.111.1.110. (kafka.network.Processor)
> [2014-04-02 10:41:24,605] INFO Partition [topic3,1] on broker 2:
> Shrinking ISR for partition [topic3,1] from 2,1,3 to 2,1
> (kafka.cluster.Partition)
> [2014-04-02 10:41:26,748] INFO Partition [topic3,1] on broker 2:
> Expanding ISR for partition [topic3,1] from 2,1 to 2,1,3
> (kafka.cluster.Partition)
> [2014-04-02 10:42:22,727] INFO Closing socket connection to
> /11.111.1.21. (kafka.network.Processor)
> [2014-04-02 10:42:23,181] INFO Closing socket connection to
> /11.111.1.21. (kafka.network.Processor)
> [2014-04-02 10:42:24,351] INFO Closing socket connection to
> /11.111.1.25. (kafka.network.Processor)
> [2014-04-02 10:42:24,501] INFO Closing socket connection to
> /11.111.1.25. (kafka.network.Processor)
> [2014-04-02 10:42:27,915] ERROR [ReplicaFetcherThread-0-1], Error
> for partition [topic2,4] to broker 1:class
> kafka.common.NotLeaderForPartitionException
> (kafka.server.ReplicaFetcherThread)
> ...
> [2014-04-02 10:42:34,575] ERROR Conditional update of path
> /brokers/topics/topic1/partitions/1/state with data
>
> {"controller_epoch":7,"leader":2,"version":1,"leader_epoch":61,"isr":[2]}
> and expected version 119 failed due to
> org.apache.zookeeper.KeeperException$BadVersionException:
> KeeperErrorCode = BadVersion for
> /brokers/topics/topic1/partitions/1/state (kafka.utils.ZkUtils$)
> [2014-04-02 10:42:34,575] INFO Partition [topic1,1] on broker 2:
> Cached zkVersion [119] not equal to that in zookeeper, skip updating
> ISR (kafka.cluster.Partition)
>
> State-change ID=2:
> [2014-04-02 10:38:37,134] TRACE Controller 2 epoch 7 received
> response correlationId 313 for a request sent to broker
> id:2,host:11.111.1.23,port:9092 (state.change.logger)
> [2014-04-02 10:38:37,696] TRACE Controller 2 epoch 7 received
> response correlationId 313 for a request sent to broker
> id:1,host:11.111.1.21,port:9092 (state.change.logger)
> [2014-04-02 10:38:38,682] TRACE Controller 2 epoch 7 received
> response correlationId 312 for a request sent to broker
> id:3,host:11.111.1.25,port:9092 (state.change.logger)
> [2014-04-02 10:38:38,909] TRACE Controller 2 epoch 7 received
> response correlationId 313 for a request sent to broker
> id:3,host:11.111.1.25,port:9092 (state.change.logger)
> [2014-04-02 10:38:38,918] TRACE Controller 2 epoch 7 received
> response correlationId 313 for a request sent to broker
> id:3,host:11.111.1.25,port:9092 (state.change.logger)
>
> Machine 1:
>
> Controller ID=1:
> [2014-04-02 10:42:20,342] INFO [Controller 1]: Broker 1 starting
> become controller state transition (kafka.controller.KafkaController)
> [2014-04-02 10:42:20,372] INFO [Controller 1]: Controller 1
> incremented epoch to 8 (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,176] DEBUG [Channel manager on controller 1]:
> Controller 1 trying to connect to broker 1
> (kafka.controller.ControllerChannelManager)
> [2014-04-02 10:42:21,238] INFO
> [Controller-1-to-broker-1-send-thread], Controller 1 connected to
> id:1,host:11.111.1.21,port:9092 for sending state change requests
> (kafka.controller.RequestSendThread)
> [2014-04-02 10:42:21,239] DEBUG [Channel manager on controller 1]:
> Controller 1 trying to connect to broker 3
> (kafka.controller.ControllerChannelManager)
> [2014-04-02 10:42:21,460] INFO
> [Controller-1-to-broker-3-send-thread], Controller 1 connected to
> id:3,host:11.111.1.25,port:9092 for sending state change requests
> (kafka.controller.RequestSendThread)
> [2014-04-02 10:42:21,483] INFO
> [Controller-1-to-broker-3-send-thread], Starting
> (kafka.controller.RequestSendThread)
> [2014-04-02 10:42:21,483] INFO
> [Controller-1-to-broker-1-send-thread], Starting
> (kafka.controller.RequestSendThread)
> [2014-04-02 10:42:21,506] INFO [Controller 1]: Partitions undergoing
> preferred replica election: (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,515] INFO [Controller 1]: Partitions that
> completed preferred replica election:
> (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,516] INFO [Controller 1]: Resuming preferred
> replica election for partitions: (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,653] INFO [Controller 1]: Partitions being
> reassigned: Map() (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,655] INFO [Controller 1]: Partitions already
> reassigned: List() (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,656] INFO [Controller 1]: Resuming reassignment
> of partitions: Map() (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,742] INFO [Controller 1]: List of topics to be
> deleted: (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,743] INFO [Controller 1]: List of topics
> ineligible for deletion: topic1,topic2,topic3
> (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,848] INFO [Controller 1]: Currently active
> brokers in the cluster: Set(1, 3) (kafka.controller.KafkaController)
> [2014-04-02 10:42:21,849] INFO [Controller 1]: Currently shutting
> brokers in the cluster: Set() (kafka.controller.KafkaController)
>
> After failover, Controller is on machine 1 and broker ID=2 disappeared
> in Zookeeper:
> ls /brokers/ids
> [3, 1]
>
> Only restart broker ID=2 helps in order to back. My questionis:
> Why broker is not able to back and works?
> (I see problem with the bad version in Zookeeper but I would like to
> better understand it and prevent in future)
>
> Thanks,
> Chris
>
>
>
--
-- Guozhang
