kafka protocol_error ai_generated true

org.apache.kafka.common.errors.LeaderEpochStaleException: 请求中的领导纪元比当前领导纪元旧

org.apache.kafka.common.errors.LeaderEpochStaleException: The leader epoch in the request is older than the current leader epoch

ID: kafka/leader-epoch-stale

其他格式: JSON · Markdown 中文 · English
78%修复率
83%置信度
1证据数
2023-04-22首次发现

版本兼容性

版本状态引入弃用备注
Kafka 2.8.0 active — — —
Kafka 3.0.0 active — — —
Kafka 3.4.0 active — — —
Kafka 3.6.0 active — — —

根因分析

跟随者或客户端发送了带有过时领导纪元的请求，表明它尚未了解最近的领导选举，导致代理拒绝该请求。

English

A follower or client sent a request with a stale leader epoch, indicating it has not yet learned about a recent leader election, causing the broker to reject the request.

generic

官方文档

https://kafka.apache.org/documentation/#replication_config

解决方案

  1. Ensure followers can quickly learn about leader changes by reducing the metadata.max.age.ms on the broker:

metadata.max.age.ms=5000  # 5 seconds

Also check that network connectivity between followers and the controller is stable; monitor controller election logs.
  2. If the issue is persistent, manually trigger a preferred leader election to stabilize the leader epoch:

kafka-leader-election --bootstrap-server localhost:9092 --election-type preferred --partition 0 --topic my_topic

This forces all replicas to acknowledge the current leader.

无效尝试

常见但无效的做法:

  1. Restart the follower broker that sent the stale request 75% 失败

    Restarting may reset the follower's state, but the underlying replication lag or network issue causing the stale epoch will persist.

  2. Increase the replica.fetch.wait.max.ms on the follower 85% 失败

    This setting controls how long a follower waits for fetch responses, but does not update the leader epoch; the follower needs to actively learn the new epoch via metadata requests.