[ https://issues.apache.org/jira/browse/KAFKA-15446?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17762997#comment-17762997 ]
许胜斌 commented on KAFKA-15446: ----------------------------- What should we do now? I have restarted Node1 and Node2, but the problem has not been resolved. I'm not too daring to restart node0 now because for partitions with a leader of node0, only the latest data is available on node0. Once I restart node0, it will cause these topics to be unable to be written for a period of time. Finally, can restarting node0 solve the problem? > Upgrading from 2.0 to 2.8, with replica out of sync exceeding 12 hours > ---------------------------------------------------------------------- > > Key: KAFKA-15446 > URL: https://issues.apache.org/jira/browse/KAFKA-15446 > Project: Kafka > Issue Type: Bug > Components: replication > Affects Versions: 2.8.2 > Environment: centos7、java8 > Reporter: 许胜斌 > Priority: Major > Labels: replica > Attachments: image-2023-09-08-16-37-12-364.png > > > !image-2023-09-08-16-37-12-364.png! > There are three brokers in the cluster. When the leader of the partition is > node 0, it cannot be synchronized to nodes 1 and 2. This problem has lasted > for more than ten hours, and the log.dir of the corresponding partition on > nodes 1 and 2 has not been updated for a long time, indicating that data > replication has stopped. > However, when the leader of the partition is node 1 or node 2, it can be > synchronized to other nodes. > the error log is: > [2023-09-08 16:35:05,238] WARN [ReplicaFetcher replicaId=2, leaderId=0, > fetcherId=0] Reset fetch offset for partition msg_for_dispatche-0 from > 3636534258 to current leader's start offset 14558984559 > (kafka.server.ReplicaFetcherThread) > [2023-09-08 16:35:05,238] INFO The cleaning for partition msg_for_dispatche-0 > is aborted and paused (kafka.log.LogManager) > [2023-09-08 16:35:05,238] INFO [Log partition=msg_for_dispatche-0, > dir=/usr/local/kafka/kafka-logs] Deleting segments as part of log truncation: > LogSegment(baseOffset=3636534258, size=0, lastModifiedTime=1694162105000, > largestRecordTimestamp=None) (kafka.log.Log) > [2023-09-08 16:35:05,241] INFO [Log partition=msg_for_dispatche-0, > dir=/usr/local/kafka/kafka-logs] Loading producer state till offset > 14558984559 with message format version 2 (kafka.log.Log) > [2023-09-08 16:35:05,241] INFO Cleaning for partition msg_for_dispatche-0 is > resumed (kafka.log.LogManager) -- This message was sent by Atlassian Jira (v8.20.10#820010)