[
https://issues.apache.org/jira/browse/FLINK-23317?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17388851#comment-17388851
]
Nicolaus Weidner commented on FLINK-23317:
------------------------------------------
After some more discussion with [~dmvk], there seems to be a problem with this
suggestion:
On subsume, we want to discard the oldest checkpoint. If it is not in memory
anymore, it has to be refetched, which will add overhead (I do not have a
feeling for how much).
Note also that for incremental checkpoints, we currently fetch all checkpoints
on restore so that we can register shared state (which is used for reference
counting and cleanup of unreferenced state). I don't currently see a way to
avoid that.
[~sewen] Did you have some solution in mind for the cleanup process when
suggesting to keep only one checkpoint in
https://issues.apache.org/jira/browse/FLINK-22483?focusedCommentId=17333103&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-17333103
?
> Only keep the latest checkpoint in CompletedCheckpointStore
> -----------------------------------------------------------
>
> Key: FLINK-23317
> URL: https://issues.apache.org/jira/browse/FLINK-23317
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Coordination
> Reporter: David Morávek
> Assignee: Nicolaus Weidner
> Priority: Minor
>
> Issue based on the discussion from FLINK-22483
> We can lower the memory footprint of CompletedCheckpointStore by keeping only
> the latest checkpoint / savepoint, that will be used for recovery. We need to
> respect `preferCheckpoint`, but the sentiment of this stays the same.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
