[
https://issues.apache.org/jira/browse/FLINK-32732?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17750532#comment-17750532
]
xiaogang zhou commented on FLINK-32732:
---------------------------------------
[~renqs] The main defect of the current implement is that if the earliest
records are timed out very quickly, the task can always be restarting when it
found the offset not found when it startup
> auto offset reset should be exposed to user
> -------------------------------------------
>
> Key: FLINK-32732
> URL: https://issues.apache.org/jira/browse/FLINK-32732
> Project: Flink
> Issue Type: Improvement
> Components: Connectors / Kafka
> Affects Versions: 1.16.1
> Reporter: xiaogang zhou
> Priority: Major
>
> {code:java}
> // code placeholder
> maybeOverride(
> ConsumerConfig.AUTO_OFFSET_RESET_CONFIG,
>
> startingOffsetsInitializer.getAutoOffsetResetStrategy().name().toLowerCase(),
> true); {code}
> now flink override the auto.offset.reset with the scan.startup.mode config,
> and user's explicit config does not take effect. I think maybe we should
> expose this to customer?
>
> I think after consuming kafka records from earliest to latest, the
> scan.startup.mode should no longer influence the kafka scan behave. So I
> suggest change the override to false.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
