[
https://issues.apache.org/jira/browse/YUNIKORN-468?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17257908#comment-17257908
]
Wilfred Spiegelenburg commented on YUNIKORN-468:
------------------------------------------------
{quote}Would YUNIKORN-466 fix this automatically? I think so.
{quote}
No that change would just make sure that the cluster size shrinks/grows
correctly with the change.
{quote}Can we do something similar (more or less) or re-use
node.GetAllAllocations foreach block of
drf_preemption_policy#trySurgicalPreemptionOnNode till difference (old capacity
- new capacity) equals total pre-empted resources (sum of pre-empted resource
gained in loop traversal)?
{quote}
Yes but we would need to define the rules on what we preempt in that case
(FIFO, LIFO or priority etc).
Overall yes we can do a preemption cycle not sure if we can re-use the current
pre-emption logic that is why I logged this issue.
> Node resource updates should trigger cleanup
> --------------------------------------------
>
> Key: YUNIKORN-468
> URL: https://issues.apache.org/jira/browse/YUNIKORN-468
> Project: Apache YuniKorn
> Issue Type: Improvement
> Components: core - common
> Reporter: Wilfred Spiegelenburg
> Priority: Minor
>
> Updating a node can set the overall size of a node and the resources
> allocated by a different scheduler (co-existence). The core just updates the
> node size and currently stops scheduling new resources even when the node is
> over subscribed.
> We should add a clean up in case that the node was shrunk and uses more than
> the resources available. the clean up should pre-empt enough resources to get
> it below the node size again.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
