打算在一个项目同时使用两个consumer消费两个topic,在配置文件中配置了consumer的默认groupid,未给两个consumer指定各自的groupid,于是两个consumer都使用同一个groupid

# 指定默认消费者group id
spring.kafka.consumer.group-id=test-message-group

但在断点调试过程中发现两个consumer偶尔正常工作,偶尔却在不断的rebanlance,并且伴随着心跳发送失败。特别在频繁读取数据或者断点调试时间比较长的时候频繁出现。而在一个consumer无法成功rebanlance时,无法消费数据。

异常状态的输出日志如下,一个consumer无法完成rebanlance,另一个conumser则无法发送心跳

[Consumer clientId=consumer-1, groupId=test-message-group] Attempt to heartbeat failed since group is rebalancing
[Consumer clientId=consumer-1, groupId=test-message-group] Attempt to heartbeat failed since group is rebalancing
[Consumer clientId=consumer-1, groupId=test-message-group] Attempt to heartbeat failed since group is rebalancing
[Consumer clientId=consumer-2, groupId=test-message-group] (Re-)joining group
[Consumer clientId=consumer-2, groupId=test-message-group] (Re-)joining group
[Consumer clientId=consumer-1, groupId=test-message-group] Attempt to heartbeat failed since group is rebalancing
[Consumer clientId=consumer-1, groupId=test-message-group] Attempt to heartbeat failed since group is rebalancing

两个consumer正常状态下的输出日志如下,均显示为 Successfully joined group with generation XXX

[Consumer clientId=consumer-1, groupId=test-message-group] (Re-)joining group
[Consumer clientId=consumer-2, groupId=test-message-group] Successfully joined group with generation 125
[Consumer clientId=consumer-1, groupId=test-message-group] Successfully joined group with generation 125
[Consumer clientId=consumer-1, groupId=test-message-group] Setting newly assigned partitions: HolderMsg-0, HolderMsg-1, HolderMsg-2
[Consumer clientId=consumer-2, groupId=test-message-group] Setting newly assigned partitions: TcMsg-2, TcMsg-0, TcMsg-1
[Consumer clientId=consumer-1, groupId=test-message-group] Setting offset for partition HolderMsg-0 to the committed offset FetchPosition{offset=7, offsetEpoch=Optional[0], currentLeader=LeaderAndEpoch{leader=192.168.202.128:9092 (id: 0 rack: null), epoch=0}}
[Consumer clientId=consumer-2, groupId=test-message-group] Setting offset for partition TcMsg-2 to the committed offset FetchPosition{offset=4, offsetEpoch=Optional[0], currentLeader=LeaderAndEpoch{leader=192.168.202.128:9092 (id: 0 rack: null), epoch=0}}
[Consumer clientId=consumer-1, groupId=test-message-group] Setting offset for partition HolderMsg-1 to the committed offset FetchPosition{offset=5, offsetEpoch=Optional[0], currentLeader=LeaderAndEpoch{leader=192.168.202.128:9092 (id: 0 rack: null), epoch=0}}
[Consumer clientId=consumer-2, groupId=test-message-group] Setting offset for partition TcMsg-0 to the committed offset FetchPosition{offset=2, offsetEpoch=Optional[0], currentLeader=LeaderAndEpoch{leader=192.168.202.128:9092 (id: 0 rack: null), epoch=0}}
[Consumer clientId=consumer-1, groupId=test-message-group] Setting offset for partition HolderMsg-2 to the committed offset FetchPosition{offset=7, offsetEpoch=Optional[0], currentLeader=LeaderAndEpoch{leader=192.168.202.128:9092 (id: 0 rack: null), epoch=0}}
[Consumer clientId=consumer-2, groupId=test-message-group] Setting offset for partition TcMsg-1 to the committed offset FetchPosition{offset=3, offsetEpoch=Optional[0], currentLeader=LeaderAndEpoch{leader=192.168.202.128:9092 (id: 0 rack: null), epoch=0}}

 

在反复的调试过程中,我意识到这两个consumer反复重新加入group中,而他们的group是同一个,猜测是两个consumer使用同一个groupid导致,相互影响。在分别为两个consumer指定了单独的groupid之后,异常的情况没有再出现。

基于这种现象,查询了一些资料,在一篇blog中找到了比较详细的讲解

https://olnrao.wordpress.com/2015/05/15/apache-kafka-case-of-mysterious-rebalances/

文章中提到,消费者在zookeeper中注册中,消费者注册标识符(Consumer Identifiers Registry)是保存在zookeeper的/consumers/[group_id]/ids/[consumer_connector_id]的路径下,这些消费者注册节点形成一棵树,当有消费者加入或离开时,树上所有的消费者都会被通知到,从而进行rebanlance。

消费者在zookeeper注册的路径与topic并没有关系,反而与groupid绑定,这是因为同一个consumer可以消费不同的topic。如果不同的consumer使用同一个groupid消费不同的topic,而任何一个topic的consumer出现加入或离开等变化时,所有groupid组里的consumer都会发生rebanlance。从而可能导致上面调试时出现的问题。

所以kafka 不同的consumer需要使用不同的group id,以减小相互之间的影响。

posted on 2020-02-12 01:44  zhaoshizi  阅读(23252)  评论(1编辑  收藏  举报