[
https://issues.apache.org/jira/browse/YUNIKORN-1440?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17642202#comment-17642202
]
Brandon Grams commented on YUNIKORN-1440:
-----------------------------------------
Yes, the transition to expired state appears to be working as-intended.
The issue is that
[cleanupExpiredApps|[https://github.com/apache/yunikorn-core/blob/339cc04bfe3dd9eedb386a6f040500d53bea8a46/pkg/scheduler/partition.go#L1425],]
which is responsible for periodically removing expired apps, doesn't consider
those that are stored in the completedApplications map, and so this grows
ad-infinitum.
[Here|https://github.com/apache/yunikorn-core/blob/339cc04bfe3dd9eedb386a6f040500d53bea8a46/pkg/scheduler/partition.go#L1486]
is where the apps are assigned to this map, which is executed via the
[callback|https://github.com/apache/yunikorn-core/blob/339cc04bfe3dd9eedb386a6f040500d53bea8a46/pkg/scheduler/objects/application_state.go#L197]
in the example you linked above.
> Completed applications are indefinitely retained in-memory
> ----------------------------------------------------------
>
> Key: YUNIKORN-1440
> URL: https://issues.apache.org/jira/browse/YUNIKORN-1440
> Project: Apache YuniKorn
> Issue Type: Bug
> Components: core - scheduler
> Affects Versions: 1.1.0
> Reporter: Brandon Grams
> Assignee: Brandon Grams
> Priority: Major
> Labels: pull-request-available
>
> Terminated applications are not considered in the [expired apps cleanup
> procedure|https://github.com/apache/yunikorn-core/blob/c61511fe84a1ae211046aa58f156ac2185892e8f/pkg/scheduler/partition.go#L1425],
> leading to indefinite retention of application objects and eventual memory
> pressure in the scheduler.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
