[
https://issues.apache.org/jira/browse/YARN-4048?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15096669#comment-15096669
]
Naganarasimha G R commented on YARN-4048:
-----------------------------------------
Hi [~geynard],
Sorry to hear you too faced the same problem,
what we did is we have a configuration to optionally opt for cpuset based
approach if we face issues like this. And suppose its 16 core machine and we
configured 75% of cpu can be used for YARN, then we try to configure 12 (we try
to round it off to lower value if the configurations doesnt match) cores for
YARN in the CPU cgroup subsystem. So yarn's containers will be ensured to run
only on the first 12 cores of the system and remaining 4 will be at the
system's disposal for other processes. *This approach ensures CPU is isolated
with other processes but not among the yarn's containers.*
> Linux kernel panic under strict CPU limits
> ------------------------------------------
>
> Key: YARN-4048
> URL: https://issues.apache.org/jira/browse/YARN-4048
> Project: Hadoop YARN
> Issue Type: Bug
> Components: nodemanager
> Affects Versions: 2.7.1
> Reporter: Chengbing Liu
> Priority: Critical
> Attachments: panic.png
>
>
> With YARN-2440 and YARN-2531, we have seen some kernel panics happening under
> heavy pressure. Even with YARN-2809, it still panics.
> We are using CentOS 6.5, hadoop 2.5.0-cdh5.2.0 with the above patches. I
> guess the latest version also has the same issue.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
