[jira] [Commented] (KAFKA-12485) Speed up Consumer#committed by returning cached offsets for owned partitions

["A. Sophie Blee-Goldman (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/KAFKA-12485?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17302980#comment-17302980 ] 

A. Sophie Blee-Goldman commented on KAFKA-12485:
------------------------------------------------

Side note: if we get this optimization in then we should be able to skip the last(?) followup for KIP-572: handling TimeoutExceptions thrown from the mainConsumer#committed call when reinitializing offsets for corrupted tasks in TaskManager#closeAndRevive. cc [~mjsax]

> Speed up Consumer#committed by returning cached offsets for owned partitions
> ----------------------------------------------------------------------------
>
>                 Key: KAFKA-12485
>                 URL: https://issues.apache.org/jira/browse/KAFKA-12485
>             Project: Kafka
>          Issue Type: Improvement
>          Components: consumer
>            Reporter: A. Sophie Blee-Goldman
>            Priority: Major
>              Labels: newbie++
>
> All of the KafkaConsumer#committed APIs will currently make a remote blocking call to the server to fetch the committed offsets. This is typically used to reset the offsets after a crash or restart, or to fetch offsets for other consumers in the group. However some users may wish to invoke this API on partitions which are currently owned by the Consumer, in which case the remote call is unnecessary since it should be able to just keep track of what it has itself committed.
> We should consider optimizing these APIs to just return the cached offsets in place of the remote call when passed in only partitions that are currently owned. This is similar to what we do in Consumer#position, although there we have a guarantee that the partitions are owned by the Consumer whereas in #committed we do not



--
This message was sent by Atlassian Jira
(v8.3.4#803005)