[ 
https://issues.apache.org/jira/browse/KAFKA-9909?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17104937#comment-17104937
 ] 

Guozhang Wang commented on KAFKA-9909:
--------------------------------------

[~gopikrishna.chaga...@target.com] We do have future plans to enable 
checkpointing (potentially to remote storage as well), but to your use cases, I 
think it still boils down to whether the error is retriable: if the message 
itself is ill-formatted (like shown in your example code) then simply 
restarting the process and hoping it would self-heal would not make sense. 
Instead some manual intervention is still needed, sth like: after checkpointed 
at offset 1, restart at offset 3 (i.e. skipping offset 2 since we know it is a 
poison pill), by doing that the record at offset 2 would NOT be processed and 
users may optionally require the producer to re-send the record etc.

> Kafka Streams : offset control to Streams API
> ---------------------------------------------
>
>                 Key: KAFKA-9909
>                 URL: https://issues.apache.org/jira/browse/KAFKA-9909
>             Project: Kafka
>          Issue Type: Improvement
>          Components: streams
>    Affects Versions: 2.5.0
>         Environment: All
>            Reporter: Gopikrishna
>            Priority: Minor
>              Labels: Offset, commit
>
> Hello team, really inspired the way streams api is running today. I would 
> like to have a feature to be flexible regarding the offset. when we write the 
> processor api, processor context object can be used to commit the offset. 
> this is not effective. but streams are controlling the offset. the moment the 
> process method executed or scheduled window completed, the offset is 
> committed automatically by streams internally. 
> Like traditional kafka consumer, its better the context object should have 
> complete control over the offset whether to commit or not. This will give 
> more control to the api to handle failovers and especially when message 
> cannot be processed, context should not commit the offset. Appreciate this 
> can be implemented. 
>  
> h4. enable.auto.commit is by default false, but streams are committing 
> automatically the offset. 



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

Reply via email to