[
https://issues.apache.org/jira/browse/FLINK-18054?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17129076#comment-17129076
]
Aljoscha Krettek commented on FLINK-18054:
------------------------------------------
I agree that this is a problem. [~kkl0u] and I discussed various problems that
come with this in the past. The most obvious one is that an environment can
spawn multiple jobs, so there needs to be a way of tying a savepoint to a job
and not to an environment, possible by specifying it on the {{execute()}}
method.
This is a general problem of the mismatch between a Flink Application and a
Flink Job: an application can spawn multiple jobs (multiple execute calls),
restarting that application from a savepoint then becomes impossible because
savepoints are tied to jobs and not applications.
> Environment should provide API to resume from savepoint
> -------------------------------------------------------
>
> Key: FLINK-18054
> URL: https://issues.apache.org/jira/browse/FLINK-18054
> Project: Flink
> Issue Type: Improvement
> Components: API / DataStream, Test Infrastructure
> Affects Versions: 1.10.1
> Reporter: Sergii Mikhtoniuk
> Priority: Minor
>
> Flink's savepoint API lacks symmetry.
> While you can stop the job with savepoint as:
> {code}
> job.stopWithSavepoint(advanceToEndOfTime, checkpointDir)
> {code}
> ... there is no API to resume from one.
> This make it very hard to test stop-resume functionality as it forces you to
> run a proper cluster and use CLI.
> Workaround:
> To use savepoints in my "unit" tests I'm resorting to subclassing local
> environment and calling {{streamGraphGenerator.setSavepointRestoreSettings}}.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
