[
https://issues.apache.org/jira/browse/KAFKA-10575?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17553806#comment-17553806
]
Guozhang Wang commented on KAFKA-10575:
---------------------------------------
I took another look at this ticket while working on KAFKA-10199 in parallel,
and here are some updates:
1. I could confirm that today we only call `onRestoreEnd` for case 1 above, and
for case 2/3 from [~ableegoldman] we do not (in fact case 3) is just a special
case of case 2) since we would first transit to CLOSED anyways).
2. On a second thought, there may be different group of users who were
anticipating the semantics of such callbacks, for example:
a) The original complaint that drives this ticket, is based on the anticipation
that each `onRestoreStart` would always be paired with an `onRestoreEnd`. This
is not actually the case because of case 2/3 above.
b) Others may anticipate that `onRestoreEnd` is only triggered when the
restoration is actually completed. In fact this is what we explicitly stated in
the javadocs.
So, if we just call `onRestoreEnd` on case 2/3) above, we may make users in a)
happier but we would break compatibilities of users in b). In addition, since
given the same topic-partition, and store names, there might be multiple
restoration process happening at the same time e.g. when there are standby
replicas, it's not very straight-forward trying to pair each one of
`onRestoreStart` with a unique `onRestoreEnd`.
With those thoughts, I'm now leaning towards not just calling `onRestoreEnd`
for case 2/3), but instead introduce a new API e.g. `onRestorePaused` for case
2/3), plus also document clearly that not every `onRestoreStart` would be
paired exactly with an `onRestoreEnd/Paused` to reduce user's unrealistic
anticipations.
Thoughts?
> StateRestoreListener#onRestoreEnd should always be triggered
> ------------------------------------------------------------
>
> Key: KAFKA-10575
> URL: https://issues.apache.org/jira/browse/KAFKA-10575
> Project: Kafka
> Issue Type: Improvement
> Components: streams
> Reporter: Guozhang Wang
> Assignee: highluck
> Priority: Major
>
> Today we only trigger `StateRestoreListener#onRestoreEnd` when we complete
> the restoration of an active task and transit it to the running state.
> However the restoration can also be stopped when the restoring task gets
> closed (because it gets migrated to another client, for example). We should
> also trigger the callback indicating its progress when the restoration
> stopped in any scenarios.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
