[
https://issues.apache.org/jira/browse/YUNIKORN-831?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17407865#comment-17407865
]
Chia-Ping Tsai commented on YUNIKORN-831:
-----------------------------------------
{quote}When you say the largest usage on your cluster is storage, what is that
specifically? I guess you are not referring to memory.
{quote}
ephemeral-storage
{quote}For the fix here, ideally, we should calculate a node score as a float
based on the consideration of all resource types. The fix here seems to
calculate "multiple" scores per resource type and use that for comparison. that
can solve the problem, but does not seem to be the best solution. It still does
not solve the issue I mentioned here:
[https://github.com/apache/incubator-yunikorn-core/pull/307#discussion_r693313432].
this must be fixed for the 1.0 release otherwise our sorting result will be
sub-optimal.
{quote}
you are right. it needs a better algorithm to score nodes. All this PR want to
do is to fix the score bug (by reverting to previous way) rather than offer a
better solution.
{quote}Wilfred Spiegelenburg, Sunil Govind on this, we have some previous
discussions, want to make sure we consider all people's opinion for this one.
{quote}
I'm ok to close this issue due to the better solution :)
> node sorting should check other resources of nodes instead of comparing node
> id directly
> ----------------------------------------------------------------------------------------
>
> Key: YUNIKORN-831
> URL: https://issues.apache.org/jira/browse/YUNIKORN-831
> Project: Apache YuniKorn
> Issue Type: Bug
> Reporter: Chia-Ping Tsai
> Assignee: Chia-Ping Tsai
> Priority: Blocker
> Labels: pull-request-available
> Fix For: 1.0.0
>
>
> from comment:
> https://issues.apache.org/jira/browse/YUNIKORN-807?focusedCommentId=17407164&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-17407164
>
> It seems to me a small behavior change is introduced by this patch. This
> patch scores node according to `resources.LargestUsageShare`. However, the
> largest usage in our cluster is always storage (and all nodes have same
> storage capacity). It results in the order is always based on `node id`.
> Before this patch, the order will compare remaining "usage" (vcore, memory,
> etc) when largest usage is equal.
> There are two possible solutions.
> # the nodeScore of nodeRef should keep all usages ([]float64) and the
> implementation of `Less` function should reuse `resources.CompareShares`
> # add an new config to enable users to control scope. for example, the
> following config means we should compare vcore instead of largest resource
> {code:java}
> nodesortpolicy:
> type: binpacking
> usage: vcore{code}
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
