[
https://issues.apache.org/jira/browse/KAFKA-7132?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16532980#comment-16532980
]
Guozhang Wang commented on KAFKA-7132:
--------------------------------------
[~Yohan123] There are two things we should consider here, [~enether] has
mentioned one, that is to guarantee offset ordering for consumption. Another
thing is to guarantee at-least-once semantics by default. Resuming from the
last committed offset would likely introduce duplicated records to be
processed, but would also avoid data loss. Restarting from the latest offset
(I'm not sure what do you mean by "it recovers at a later offset", so I'd
assume you meant to say when consumer resumes, the log has grown to offset 120)
would cause you to lose the data from 100 - 120, while using a separate
consumer to cover the gap would violate ordering guarantees.
> Consider adding faster form of rebalancing
> ------------------------------------------
>
> Key: KAFKA-7132
> URL: https://issues.apache.org/jira/browse/KAFKA-7132
> Project: Kafka
> Issue Type: Improvement
> Components: consumer
> Reporter: Richard Yu
> Priority: Critical
> Labels: performance
>
> Currently, when a consumer falls out of a consumer group, it will restart
> processing from the last checkpointed offset. However, this design could
> result in a lag which some users could not afford to let happen. For example,
> lets say a consumer crashed at offset 100, with the last checkpointed offset
> being at 70. When it recovers at a later offset (say, 120), it will be behind
> by an offset range of 50 (120 - 70). This is because the consumer restarted
> at 70, forcing it to reprocess old data. To avoid this from happening, one
> option would be to allow the current consumer to start processing not from
> the last checkpointed offset (which is 70 in the example), but from 120 where
> it recovers. Meanwhile, a new KafkaConsumer will be instantiated and start
> reading from offset 70 in concurrency with the old process, and will be
> terminated once it reaches 120. In this manner, a considerable amount of lag
> can be avoided, particularly since the old consumer could proceed as if
> nothing had happened.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
