You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@yunikorn.apache.org by "Manikandan R (Jira)" <ji...@apache.org> on 2020/12/24 16:00:00 UTC
[jira] [Commented] (YUNIKORN-497) handle placeholder pod recovery
[ https://issues.apache.org/jira/browse/YUNIKORN-497?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17254582#comment-17254582 ]
Manikandan R commented on YUNIKORN-497:
---------------------------------------
Bit confused with "The Allocation, which is the result the scheduler passes back does not change." in the doc as later section discuss about "Allocation" interface changes in detail. After going through this JIRA, it becomes clear now. Probably, you will need to change this bit in doc.
> handle placeholder pod recovery
> -------------------------------
>
> Key: YUNIKORN-497
> URL: https://issues.apache.org/jira/browse/YUNIKORN-497
> Project: Apache YuniKorn
> Issue Type: Sub-task
> Components: core - scheduler, shim - kubernetes
> Reporter: Wilfred Spiegelenburg
> Priority: Major
>
> When recovering a node the existing allocations are communicated via an si.Allocation message. The message after the changes for the interface does not have the placeholder information. That means recovery of placeholder would generate a normal allocation.
> The si.Allocation object needs the same fields a the si.AllocationAsk ( [document|https://docs.google.com/document/d/1YpLZLqYsp5IaeEdl_AU9wZ-f16CfaaoRhZdalDP4RCg/edit?usp=sharing] is updated)
> The shim must add the fields when it generates the existing allocation list. Recovery on the core side needs to reinstate the fields into the allocation that gets created.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@yunikorn.apache.org
For additional commands, e-mail: issues-help@yunikorn.apache.org