You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@kafka.apache.org by "Guozhang Wang (JIRA)" <ji...@apache.org> on 2014/06/03 23:10:01 UTC

[jira] [Created] (KAFKA-1483) Split Brain about Leader Partitions

Guozhang Wang created KAFKA-1483:
------------------------------------

             Summary: Split Brain about Leader Partitions
                 Key: KAFKA-1483
                 URL: https://issues.apache.org/jira/browse/KAFKA-1483
             Project: Kafka
          Issue Type: Bug
            Reporter: Guozhang Wang
            Assignee: Jun Rao
             Fix For: 0.9.0


Today in the server there are two places storing the leader partition info:

1) leaderPartitions list in the ReplicaManager.
2) leaderBrokerIdOpt in the Partition.

1) is used as the ground truth to decide if the server is the current leader for serving requests; 2) is used as the ground truth for reporting leader counts metrics, etc and for the background Shrinking-ISR thread to decide which partition to check. There is a risk that these two ground truth caches are not consistent, and we'd better only make one of them as the ground truth.



--
This message was sent by Atlassian JIRA
(v6.2#6252)