[
https://issues.apache.org/jira/browse/FLINK-39108?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18111760#comment-18111760
]
Li Guo commented on FLINK-39108:
--------------------------------
The first checkpoint gets through for two reasons. setIsProcessingBacklog only
reschedules the periodic trigger when the new effective interval would fire
earlier and skips the disabled case, so the trigger armed by
startCheckpointScheduler survives the backlog report. And ScheduledTrigger#run
parks itself when the interval is disabled, but still triggers the current run;
that is the path the ITCase hits, because the scheduler is started at task
RUNNING after the source already reported backlog.
I re-enabled testNoCheckpointDuringBacklog locally on current master and it
still fails (2 checkpoints during backlog); with the fix the whole class
passes. PR: https://github.com/apache/flink/pull/29110. It also removes the
@Disabled (which referenced FLINK-39018, looks like a typo for this ticket).
[~roman] since you disabled the test, would you mind taking a look?
> CheckpointIntervalDuringBacklog is ignored for the first checkpoint
> -------------------------------------------------------------------
>
> Key: FLINK-39108
> URL: https://issues.apache.org/jira/browse/FLINK-39108
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Checkpointing
> Affects Versions: 1.19.0, 1.20.0, 2.2.0
> Reporter: Roman Khachatryan
> Priority: Major
> Labels: pull-request-available
>
> After adjusting
> CheckpointIntervalDuringBacklogITCase.testNoCheckpointDuringBacklog, it fails
> reliably because at least the first checkpoint is triggered regardless of the
> configured checkpoint-interval-during-backlog.
> Related:
> https://github.com/apache/flink/pull/22931#pullrequestreview-3820270489
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
