[
https://issues.apache.org/jira/browse/KAFKA-19735?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18023119#comment-18023119
]
Lianet Magrans commented on KAFKA-19735:
----------------------------------------
Hey, in case it helps:
* I filed something related a while back here
https://issues.apache.org/jira/browse/KAFKA-16233
* why subscribe mode only? I expect that whatever improvement we want to
consider in this area would be for consumers with auto.commit enabled + group
Id (regardless of using subscribe or not). A consumer using assign can auto
commit too (if it has group ID + auto.commit enabled)
> Add automatic commit offset caching in subscribe mode
> -----------------------------------------------------
>
> Key: KAFKA-19735
> URL: https://issues.apache.org/jira/browse/KAFKA-19735
> Project: Kafka
> Issue Type: Improvement
> Reporter: kangning.li
> Assignee: kangning.li
> Priority: Minor
>
> Suppose that no new messages have been written to the target topic for a long
> time, for example, the producer has not generated new data for nearly an
> hour. On the consumer side, when the user sets enable.auto.commit=true, the
> OFFSET_COMMIT request will still be repeatedly submitted every 5 seconds.
> This not only wastes network resources but also increases the burden on
> __consumer_offset.
>
> Therefore, maybe we can add a cache for committed offsets to check if the
> currently committed offset is consistent with the most recently successfully
> committed offset. If they are the same, this offset commit can be ignored. Of
> course, all the above applies to the subscribe mode.
>
> Advantages:
> * Reduce unnecessary network requests
> * Alleviate the processing pressure on the broker side
> * Relieve the pressure of log cleaning for __consumer_offset
>
> Disadvantage:
> * There may be hidden bugs that need to be discussed and identified
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
