[
https://issues.apache.org/jira/browse/YUNIKORN-222?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wilfred Spiegelenburg resolved YUNIKORN-222.
--------------------------------------------
Fix Version/s: 0.9
Resolution: Fixed
If the application allocation is not confirmed before the ask is updated the
state is changed from accepted to running (via waiting) instead of accepted to
starting. This breaks the stateaware scheduling policy of a queue.
Not affected by this issue is the removal of an ask, just the update of the
repeats is affected.
The ask repeat update state change check was racing with the allocation
confirmation from the cache. The ask repeat update could be triggered before
the confirmation. This only affects applications in the accepted state. An
application in accepted state was left with no pending ask, no confirmed
allocation and always an inflight allocation. This made the triggering the
state change ambiguous: there was no set of values that always showed the
correct state.
Changes committed
> Application state transition race on ask update
> -----------------------------------------------
>
> Key: YUNIKORN-222
> URL: https://issues.apache.org/jira/browse/YUNIKORN-222
> Project: Apache YuniKorn
> Issue Type: Bug
> Components: core - scheduler
> Reporter: Wilfred Spiegelenburg
> Assignee: Wilfred Spiegelenburg
> Priority: Major
> Labels: pull-request-available
> Fix For: 0.9
>
>
> There is a race between the ask being updated and the allocation being
> confirmed on an application. This causes a premature state change that can
> cause stateaware scheduling to break from YUNIKORN-99.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
