[
https://issues.apache.org/jira/browse/YARN-11844?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18012188#comment-18012188
]
ASF GitHub Bot commented on YARN-11844:
---------------------------------------
cnauroth opened a new pull request, #7857:
URL: https://github.com/apache/hadoop/pull/7857
### Description of PR
The NodeManager invokes an external binary (e.g. `nvidia-smi`) to discover
attached GPUs. Right now, there is a hard-coded 10-second timeout on execution
of this binary and a hard-coded max error count of 10, beyond which the
NodeManager will stop attempting discovery. This change will provide new
configuration properties to control both the timeout and the max errors, which
is useful in environments where there may be a delay in binding the GPU to the
host. Default values for the new configuration properties will be set so as to
maintain the existing behavior.
Special thanks to @jjayadeep06 for co-authoring this patch.
### How was this patch tested?
New unit test and manual testing.
### For code changes:
- [X] Does the title or this PR starts with the corresponding JIRA issue id
(e.g. 'HADOOP-17799. Your PR title ...')?
- [ ] Object storage: have the integration tests been executed and the
endpoint declared according to the connector-specific documentation?
- [ ] If adding new dependencies to the code, are these dependencies
licensed in a way that is compatible for inclusion under [ASF
2.0](http://www.apache.org/legal/resolved.html#category-a)?
- [ ] If applicable, have you updated the `LICENSE`, `LICENSE-binary`,
`NOTICE-binary` files?
> Support configuration of retry policy on GPU discovery
> ------------------------------------------------------
>
> Key: YARN-11844
> URL: https://issues.apache.org/jira/browse/YARN-11844
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: gpu, nodemanager
> Reporter: Chris Nauroth
> Assignee: Chris Nauroth
> Priority: Major
>
> The NodeManager invokes an external binary (e.g. {{nvidia-smi}}) to discover
> attached GPUs. Right now, there is a hard-coded 10-second timeout on
> execution of this binary and a hard-coded max error count of 10, beyond which
> the NodeManager will stop attempting discovery. This change will provide new
> configuration properties to control both the timeout and the max errors,
> which is useful in environments where there may be a delay in binding the GPU
> to the host. Default values for the new configuration properties will be set
> so as to maintain the existing behavior.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
