[
https://issues.apache.org/jira/browse/YUNIKORN-2171?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17788610#comment-17788610
]
Wilfred Spiegelenburg commented on YUNIKORN-2171:
-------------------------------------------------
Affected versions go all the way back to 0.10 that is when the deadlock fix for
node removals was added in YUNIKORN-551.
The messages logged would be in order:
{code:java}
INFO core.scheduler.partition scheduler/partition.go:647 removing node
from partition
INFO core.scheduler.queue objects/queue.go:1266 updating root queue
max resources
INFO core.scheduler.partition scheduler/partition.go:634 Updated
available resources from removed node
WARN core.scheduler.application objects/applicaion.go:1508 queue
update failed unexpectedly
INFO core.scheduler.partition scheduler/partition.go:781 allocation
removed from node
{code}
Since this is a race which also needs specific circumstances for the queues I
have not been able to reproduce this in a test case. The only way I can
reproduce issue is by manually putting the system in a state that has half
removed the node and kick of scheduling.
> race condition in node removal and scheduling cycle
> ---------------------------------------------------
>
> Key: YUNIKORN-2171
> URL: https://issues.apache.org/jira/browse/YUNIKORN-2171
> Project: Apache YuniKorn
> Issue Type: Bug
> Components: core - scheduler
> Affects Versions: 1.0.0, 1.1.0, 1.2.0, 1.3.0, 1.4.0
> Reporter: Wilfred Spiegelenburg
> Assignee: Wilfred Spiegelenburg
> Priority: Critical
>
> When a node gets removed the partition resources and thus the root max
> resources are decreased. The node removal locks the partition, removes the
> node and releases the partition lock before proceeding. Cleanup of the
> allocations happens after that. This means that for a short period of time
> the root queue max resources are already decreased while the usage is not.
> The scheduling cycle could be running during the node removal. The queue
> headroom calculation uses the queue max resources and usage to calculate the
> difference. The whole hierarchy is traversed for this.
> If the headroom is limited by the root queue then we could have a race
> between the removal of the node allocations and scheduling:
> * scheduling starts and queue headroom is calculated
> * node is removed, queue max is lowered
> * scheduling finds new allocation
> * new allocation gets added to the queue updating usage
> * root queue is over its max already or would go over max: scheduling fails
> * node allocation removal proceeds and corrects the queue usage
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
