[
https://issues.apache.org/jira/browse/KAFKA-2877?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15025225#comment-15025225
]
Jason Gustafson commented on KAFKA-2877:
----------------------------------------
It looks like when we updated the new consumer for persistence in KAFKA-2017,
we forgot to add handling for the request timeout error code. This code is
possible when group metadata is written to the log if the write timeout expires
before enough replicas have acknowledged it. I've added a simple patch which
checks for this error.
> Messages sometimes not delivered by new consumer after Kafka restart
> ---------------------------------------------------------------------
>
> Key: KAFKA-2877
> URL: https://issues.apache.org/jira/browse/KAFKA-2877
> Project: Kafka
> Issue Type: Bug
> Components: consumer
> Affects Versions: 0.9.0.0
> Reporter: Rajini Sivaram
> Assignee: Jason Gustafson
> Priority: Critical
>
> After a Kafka restart, our health check consumer which subscribes to five
> topics with one partition each, was receiving messages from four out of the
> five topics. This has happened twice, the second time today was on 0.9.0.0
> RC3.
> Some of the system test failures in
> http://jenkins.confluent.io/job/kafka_system_tests_branch_builder/220/ when
> the replication test was modified to use SSL/SASL clients and the throughput
> of the producer was reduced, also show a similar problem. Many of the
> replication tests fail intermittently when new consumer is used in order to
> run clients with SSL/SASL.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
