[
https://issues.apache.org/jira/browse/FLINK-12058?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16804518#comment-16804518
]
vinoyang commented on FLINK-12058:
----------------------------------
Hi [~till.rohrmann] thanks for opening this issue. I discussed this problem
under another issue FLINK-10966 with [~StephanEwen] . It depends on the
implementation of the actor style task. He and [~pnowojski] are designing this
feature. And as you know, this issue is also related to FLINK-10724. I have
opened a PR for FLINK-10724, but it still has a little problem to be fixed.
Hope for you guys can help to review that PR. cc [~azagrebin]
> Cancel checkpoint operations belonging to a discarded/aborted checkpoint
> ------------------------------------------------------------------------
>
> Key: FLINK-12058
> URL: https://issues.apache.org/jira/browse/FLINK-12058
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Checkpointing
> Affects Versions: 1.7.2, 1.8.0
> Reporter: Till Rohrmann
> Assignee: vinoyang
> Priority: Major
>
> In order to save CPU cycles and reduce disk and network I/O, we should try to
> cancel local checkpoint operations belonging to discarded aborted or subsumed
> checkpoints. For example, if a {{Task}} declines a checkpoint, the
> {{CheckpointCoordinator}} will discard this checkpoint. However, other
> checkpointing operations belonging to this checkpoint won't be necessarily
> notified and canceled.
> The notification mechanism could piggy back on the existing
> {{CancelCheckpointMarker}} or be a separate signal sent to all participating
> {{Tasks}}.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
