[ 
https://issues.apache.org/jira/browse/YARN-9419?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17052306#comment-17052306
 ] 

Adam Antal commented on YARN-9419:
----------------------------------

Thanks for the patch [~gandras]!

Minor nits:
- {{#validateConfig}} is a bit too broad name, I suggest to be a bit more 
specific here, like {{validateContainerExecutor}} or similar.
- For obtaining the configured container-executor, we can use the following 
piece of code:
{code:java}
Class containerExecutorClass = 
conf.getClass(YarnConfiguration.NM_CONTAINER_EXECUTOR,
            DefaultContainerExecutor.class, ContainerExecutor.class);
{code}
similarly as in {{NodeManager#createContainerExecutor}}. Then we can simply 
call .equals on it, and compare it with DCE.
- I liked the idea of putting {{DEFAULT_NM_CONTAINER_EXECUTOR}} to 
{{YarnConfiguration}}. If you do this, please add the same to 
{{NodeManager#createContainerExecutor}} as well, because this is where the CE 
is instantiated in the NodeManager.

> Log a warning if GPU isolation is enabled but LinuxContainerExecutor is 
> disabled
> --------------------------------------------------------------------------------
>
>                 Key: YARN-9419
>                 URL: https://issues.apache.org/jira/browse/YARN-9419
>             Project: Hadoop YARN
>          Issue Type: Bug
>            Reporter: Szilard Nemeth
>            Assignee: Andras Gyori
>            Priority: Major
>         Attachments: YARN-9419.001.patch
>
>
> A WARN log should be added at least (logged once on startup) that notifies 
> the user about a potentially offending configuration: GPU isolation is 
> enabled but LCE is disabled.
> I think this is a dangerous, yet valid configuration: As LCE is the only 
> container executor that utilizes cgroups, no real HW-isolation happens if LCE 
> is disabled. 
> Let's suppose we have 2 GPU devices in 1 node:
>  # NM reports 2 devices (as a Resource) to RM
>  # RM assigns GPU#1 to container#2 that requests 1 GPU device
>  # When container#2 is also requesting 1 GPU device, RM is going to assign 
> either GPU#1 or GPU#2, so there's no guarantee that GPU#2 will be assigned. 
> If GPU#1 is assigned to a second container, nasty things could happen.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org

Reply via email to