[
https://issues.apache.org/jira/browse/YARN-9100?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16907951#comment-16907951
]
Szilard Nemeth commented on YARN-9100:
--------------------------------------
Patch looks good in overall, I only found one minor bit of weirdness in the
test code (TestGpuResourceAllocator):
In TestGpuResourceAllocator#assertAllocatedGpu:
We have:
{code:java}
assertEquals(1, allocation.getAllowedGPUs().size());
assertEquals(0, allocation.getDeniedGPUs().size());
Set<GpuDevice> allowedGPUs = allocation.getAllowedGPUs();
assertEquals(1, allowedGPUs.size());
GpuDevice allocatedGpu = (GpuDevice) allowedGPUs.toArray()[0];
assertEquals(expectedGpu, allocatedGpu);
assertAssignmentInStateStore(expectedGpu, container);
{code}
I think the code block of
{code:java}
Set<GpuDevice> allowedGPUs = allocation.getAllowedGPUs();
assertEquals(1, allowedGPUs.size());
{code}
is superfluous, as the size of allowed GPUs list is already checked above for
the allocation.
Please fix this minor bit and I think we are good to go!
Thanks!
> Add tests for GpuResourceAllocator and do minor code cleanup
> ------------------------------------------------------------
>
> Key: YARN-9100
> URL: https://issues.apache.org/jira/browse/YARN-9100
> Project: Hadoop YARN
> Issue Type: Improvement
> Reporter: Szilard Nemeth
> Assignee: Peter Bacsko
> Priority: Major
> Attachments: YARN-9100-004.patch, YARN-9100-005.patch,
> YARN-9100-006.patch, YARN-9100-007.patch, YARN-9100-008.patch,
> YARN-9100.001.patch, YARN-9100.002.patch, YARN-9100.003.patch
>
>
> Add tests for GpuResourceAllocator and do minor code cleanup
> - Improved log and exception messages
> - Added some new debug logs
> - Some methods are named like *Copy, these are returning copies of internal
> data structures. The word "copy" is just a noise in their name, so they have
> been renamed. Additionally, the copied data structures modified to be
> immutable.
> - The waiting loop in method assignGpus were decoupled into a new class,
> RetryCommand.
> Some more words about the new class RetryCommand:
> There are some similar waiting loops in the code in: AMRMClient,
> AMRMClientAsync and even in GenericTestUtils (see waitFor method).
> RetryCommand could be a future replacement of these duplicated code, as it
> gives a solution to this waiting loop problem in a generic way.
> The only downside of the usage of RetryCommand in GpuResourceAllocator
> (startGpuAssignmentLoop) is the ugly exception handling part, but that's
> solely because how Java deals with checked exceptions vs. lambdas. If there's
> a cleaner way to solve the exception handling, I'm open for any suggestions.
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
