[
https://issues.apache.org/jira/browse/YUNIKORN-63?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17071407#comment-17071407
]
Wilfred Spiegelenburg commented on YUNIKORN-63:
---
Tracking why this is being added [from the pr
comment|https://github.com/apache/incubator-yunikorn-k8shim/pull/91#issuecomment-606200513]:
{quote}{quote}See my comment in the jira: why this change from not needing it
in apache/incubator-yunikorn-core#108 to needing it now?
{quote}
This is found while testing the recovery on a real cluster. When it tries to
report the occupied resources to the scheduler-core, the node is not yet
registered. Therefore, the latest patch ensures that when we register the node,
we need to have the initial occupied resources counted correctly.
My previous fix (only updating the node capacity) might have the same issue,
but I guess I missed that during the local test (not covering the recovery
test).
{quote}
> Report occupied resources while registering a node
> --
>
> Key: YUNIKORN-63
> URL: https://issues.apache.org/jira/browse/YUNIKORN-63
> Project: Apache YuniKorn
> Issue Type: Improvement
> Components: core - scheduler, scheduler-interface, shim - kubernetes
>Reporter: Weiwei Yang
>Assignee: Weiwei Yang
>Priority: Major
> Labels: pull-request-available
> Time Spent: 1h
> Remaining Estimate: 0h
>
> We need to support this while registering a node, otherwise, it could be
> race-condition that the shim reports an update but the node is not yet
> registered.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
-
To unsubscribe, e-mail: issues-unsubscr...@yunikorn.apache.org
For additional commands, e-mail: issues-h...@yunikorn.apache.org