[
https://issues.apache.org/jira/browse/FLINK-9598?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16545318#comment-16545318
]
Yun Tang commented on FLINK-9598:
---------------------------------
Hi [~premsantosh], the "Minimum Pause Between Checkpoints" is actually the
initial delay between successful checkpoints, you can find the logical in
CheckpointCoordinator#_triggerCheckpoint_() method, in which after
expired-checkpoint cleaner detects some checkpoint expired, it will trigger
another checkpoint ASAP through CheckpointCoordinator#_triggerQueuedRequests_()
method, no matter Flink-1.3.2 or latest Flink-1.5.1
I think a user usually wants to get a successful checkpoint as quickly as
possible again, and the running checkpoint would not stall your application
running in general as the sub-tasks only start snapshot when checkpoint barrier
comes, not all sub-tasks are executing snapshot process.
In my point of view, it would be better to redefine some of the javadocs e.g.
attribute _minPauseBetweenCheckpointsNanos_ in CheckpointCoordinator. What's
your opinion [~yanghua], if you don't have time to do these trivial works, I'd
like to take some time to redefine all related javadocs.
> [Checkpoints] The config Minimum Pause Between Checkpoints doesn't work when
> there's a checkpoint failure
> ---------------------------------------------------------------------------------------------------------
>
> Key: FLINK-9598
> URL: https://issues.apache.org/jira/browse/FLINK-9598
> Project: Flink
> Issue Type: Bug
> Affects Versions: 1.3.2
> Reporter: Prem Santosh
> Assignee: vinoyang
> Priority: Major
> Attachments: Screen Shot 2018-06-20 at 7.44.10 AM.png
>
>
> We have set the config Minimum Pause Between Checkpoints to be 10 min but
> noticed that when a checkpoint fails (because it timesout before it
> completes) the application immediately starts taking the next checkpoint.
> This basically stalls the application's progress since its always taking
> checkpoints.
> [^Screen Shot 2018-06-20 at 7.44.10 AM.png] is a screenshot of this issue.
> Details:
> * Running Flink-1.3.2 on EMR
> * checkpoint timeout duration: 40 min
> * minimum pause between checkpoints: 10 min
> There is also a [relevant
> thread|http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/Having-a-backoff-while-experiencing-checkpointing-failures-td20618.html]
> that I found on the Flink users group.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
