[
https://issues.apache.org/jira/browse/YARN-6672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16522509#comment-16522509
]
Íñigo Goiri commented on YARN-6672:
-----------------------------------
Thanks [~haibochen] for [^YARN-6672-YARN-1011.02.patch].
A few comments:
* Use logger style in ContainerScheduler#228.
* Add links in {{ContainerSchedulerOverallocationPreemptionEvent}} javadoc
description.
* Make the fields in {{SnapshotBasedOverAllocationPreemptionPolicy}} final.
* In {{SnapshotBasedOverAllocationPreemptionPolicy}}, I would always return the
newInstance for ResourceUtilization. So it should be a matter of sanitzing both
vcoreOverLimit and memoryOverLimit.
* I would add a unit test for SnapshotBasedOverAllocationPreemptionPolicy with
the 4/5 cases (both OK, bad in CPU, bad in memory, bad for both, and a couple
negative cases).
* For the unit tests in TestContainerSchedulerWithOverAllocation, I would try
to do the new unit tests with and without the feature enabled. This would
require some refactor.
* Instead of {{2.0f/2}} in testPreemptionUponHighCPUUtilization, we should have
some constant/extracted variable.
> Add NM preemption of opportunistic containers when utilization goes high
> ------------------------------------------------------------------------
>
> Key: YARN-6672
> URL: https://issues.apache.org/jira/browse/YARN-6672
> Project: Hadoop YARN
> Issue Type: Sub-task
> Affects Versions: 3.0.0-alpha3
> Reporter: Haibo Chen
> Assignee: Haibo Chen
> Priority: Major
> Attachments: YARN-6672-YARN-1011.00.patch,
> YARN-6672-YARN-1011.01.patch, YARN-6672-YARN-1011.02.patch
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
