[
https://issues.apache.org/jira/browse/YARN-3304?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14384992#comment-14384992
]
Karthik Kambatla commented on YARN-3304:
----------------------------------------
Just uploaded the v5 patch. The changes are primarily to define
ResourceCalculatorProcessTree#UNAVAILABLE and use that elsewhere in the code,
and javadoc changes.
[~adhoot] - ProcfsProcessTree#getCumulativeCpuTime is different compared to
getCumulativeMem methods in that class, and even getCumulativeCpuTime in
WindowsProcfsProcessTree. It seems to continue adding up usage across multiple
calls to getCumulativeCpuTime, while rest of them freshly compute on every
call. Is this intentional? If not, we might want to fix that too.
> ResourceCalculatorProcessTree#getCpuUsagePercent default return value is
> inconsistent with other getters
> --------------------------------------------------------------------------------------------------------
>
> Key: YARN-3304
> URL: https://issues.apache.org/jira/browse/YARN-3304
> Project: Hadoop YARN
> Issue Type: Bug
> Components: nodemanager
> Reporter: Junping Du
> Assignee: Junping Du
> Priority: Blocker
> Attachments: YARN-3304-v2.patch, YARN-3304-v3.patch,
> YARN-3304-v4-boolean-way.patch, YARN-3304-v4-negative-way-MR.patch,
> YARN-3304-v4-negtive-value-way.patch, YARN-3304.patch, yarn-3304-5.patch
>
>
> Per discussions in YARN-3296, getCpuUsagePercent() will return -1 for
> unavailable case while other resource metrics are return 0 in the same case
> which sounds inconsistent.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
