[
https://issues.apache.org/jira/browse/FLINK-17998?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17134348#comment-17134348
]
Daniel Laszlo Magyar commented on FLINK-17998:
----------------------------------------------
I've just opened the PR [https://github.com/apache/flink/pull/12636].
Apart from the decisions mentioned in the description, I'd like to highlight
that the question arose whether the archives should be cleaned up only if
{{historyserver.archive.clean-expired-jobs}} is explicitly set to {{true}} or
not ( it is {{false}} by default).
After discussing this with [~gyfora], from the user experience point of view it
seemed to be desirable to make these config options independent and remove the
least recently modified archives even if
{{historyserver.archive.clean-expired-jobs}} is unset/set to {{false. }}The
proposed implementation reflects this decision but is open for debate of course.
> Support limiting job history size
> ---------------------------------
>
> Key: FLINK-17998
> URL: https://issues.apache.org/jira/browse/FLINK-17998
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / Configuration
> Reporter: Daniel Laszlo Magyar
> Assignee: Daniel Laszlo Magyar
> Priority: Minor
> Labels: pull-request-available
>
> Flink history server stores information for an unlimited number of completed
> jobs, which is inconvenient to dig through on long-running instances and
> might cause performance degradation in the long run.
> Spark for example addresses this issue by limiting the number of jobs to keep
> by a [configuration
> parameter|https://spark.apache.org/docs/2.2.0/monitoring.html#spark-configuration-options]
> {{spark.history.retainedApplications}} with a default of 50 jobs to keep.
> Introducing a similar configuration option to the Flink history server would
> improve the usability and performability of long-running history server
> instances.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
