[
https://issues.apache.org/jira/browse/KAFKA-2857?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15326653#comment-15326653
]
Jason Gustafson commented on KAFKA-2857:
----------------------------------------
[~imandhan] The __consumer_offsets topic is a normal topic, so it can happen
that all the partition leaders and replicas are down, in which case we'll
return an error saying the group coordinator is not available. The cluster has
to be pretty big to hit this, but it can happen. It seems like our choices are
basically to either let the command retry on this error, or try to give the
user a more helpful message. I'd probably favor the latter. Since the most
likely scenario for this case is the one mentioned in the description (where
the topic hasn't been created yet), I think it would make sense to mention it
explicitly in the message and let the user retry.
> ConsumerGroupCommand throws GroupCoordinatorNotAvailableException when
> describing a non-existent group before the offset topic is created
> -----------------------------------------------------------------------------------------------------------------------------------------
>
> Key: KAFKA-2857
> URL: https://issues.apache.org/jira/browse/KAFKA-2857
> Project: Kafka
> Issue Type: Bug
> Components: tools
> Reporter: Ismael Juma
> Assignee: Ishita Mandhan
> Priority: Minor
>
> If we describe a non-existing group before the offset topic is created, like
> the following:
> {code}
> bin/kafka-consumer-groups.sh --bootstrap-server localhost:9092 --new-consumer
> --describe --group gggg
> {code}
> We get the following error:
> {code}
> Error while executing consumer group command The group coordinator is not
> available.
> org.apache.kafka.common.errors.GroupCoordinatorNotAvailableException: The
> group coordinator is not available.
> {code}
> The exception is thrown in the `adminClient.describeConsumerGroup` call. We
> can't interpret this exception as meaning that the group doesn't exist
> because it could also be thrown f all replicas for a offset topic partition
> are down (as explained by Jun).
> Jun also suggested that we should distinguish if a coordinator is not
> available from the case where a coordinator doesn't exist.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
