[ https://issues.apache.org/jira/browse/KAFKA-19519?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Lan Ding updated KAFKA-19519: ----------------------------- Description: 在 KAFKA-19427 中,有一个用例是,当有一个消费者组订阅了大量的主题/分区时,当该组被重新平衡时,然后协调器代理将该组的分配存储给__consumer_offsets,它将抛出错误 RecordTooLargeException 。 目前,要解决这个问题,唯一的解决方案是增加代理级别的“message.max.bytes”配置。但此更改的副作用是它可能允许所有主题而不覆盖主题级别 {{max.message.bytes}} 配置,现在将允许更高的消息大小。 我们可以引入一个新的配置来驱动组协调器使用的值(例如 group.coordinator.append.max.bytes),而不是依赖代理 message.max.bytes。这将用于在创建主题时在主题级别设置最大字节数。 was: In KAFKA-19427, there's a use case that when there is a consumer group subscribes huge amount of topics/partitions, when this group is rebalanced, and then the coordinator broker stores the assignment of this group to __consumer_offsets, it will throw an error RecordTooLargeException . Currently, to resolve this issue, the only solution is to increase the broker-level `message.max.bytes` config. But the side effect of this change is it potentially allows all topics without override the topic level {{max.message.bytes}} config, will now allow higher message size. We could introduce a new config to drive the value used by the group coordinator - e.g group.coordinator.append.max.bytes - instead of relying on the broker message.max.bytes. This would be used to set the max bytes at the topic level when the topic is created. > Introduce a new config for group coordinator max record size > ------------------------------------------------------------ > > Key: KAFKA-19519 > URL: https://issues.apache.org/jira/browse/KAFKA-19519 > Project: Kafka > Issue Type: Improvement > Reporter: Luke Chen > Priority: Major > > 在 KAFKA-19427 > 中,有一个用例是,当有一个消费者组订阅了大量的主题/分区时,当该组被重新平衡时,然后协调器代理将该组的分配存储给__consumer_offsets,它将抛出错误 > RecordTooLargeException 。 > > 目前,要解决这个问题,唯一的解决方案是增加代理级别的“message.max.bytes”配置。但此更改的副作用是它可能允许所有主题而不覆盖主题级别 > {{max.message.bytes}} 配置,现在将允许更高的消息大小。 > > 我们可以引入一个新的配置来驱动组协调器使用的值(例如 group.coordinator.append.max.bytes),而不是依赖代理 > message.max.bytes。这将用于在创建主题时在主题级别设置最大字节数。 -- This message was sent by Atlassian Jira (v8.20.10#820010)