[ https://issues.apache.org/jira/browse/KAFKA-10890?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
GeoffreyStark updated KAFKA-10890: ---------------------------------- Attachment: server.properties > Broker just stated Ignoring LeaderAndIsr request from controller > ----------------------------------------------------------------- > > Key: KAFKA-10890 > URL: https://issues.apache.org/jira/browse/KAFKA-10890 > Project: Kafka > Issue Type: Bug > Components: core > Affects Versions: 2.0.0 > Environment: kfk 2.0 > 74 brokers > 3 replica-factors > Reporter: GeoffreyStark > Priority: Major > Attachments: image-2020-12-28-16-59-03-492.png, > image-2020-12-29-18-00-22-344.png, jstack-1013broker-1228-1312, > kafka元数据混乱.docx, server.properties > > > In auto.leader.rebalance.enabled=true > Taking a Broker 1013 (with many leader partitions) offline; > Then, after the leaders of these subdivisions have been elected from other > ISRs, > In theory, 1013 should once again become the leader of those partitions > Indeed, if you look at the describe command and the information of the > corresponding partition in ZK, you can see that the leader has changed back > to 1013. > But if you look at the Controller log and the 1013 log at this point, you'll > see that there are some warning messages, > > > It looks like the controller failed to send a LeaderAndIsr request to a 1013 > node after it restarted, Then the 1013 node‘s log has been > > {code:java} > "Ignoring LeaderAndIsr request from controller 1017 with the correlation id > 33 epoch 6 fro partition sp since its associated leader 5 is not who the > current leader epoch 5 (state. Change. Logger). " > {code} > > > After a while, > the producer reports an error. > {code:java} > "The server is not The leader for that topically -- partition.. Going to > request metadata update now" > {code} > -- This message was sent by Atlassian Jira (v8.3.4#803005)