Add ability to customize VM specs in LXD VM GPU Passthrough test (Bugfix)

[MitchellAugustin]

Description

Some modern GPUs have BAR sizes that exceed the default QEMU pci64-hole-size allowance, or require additional host memory / CPU allocations to work properly in guest VMs when passed through.

This series enables users of the gpgpu provider to apply custom environment options to override these defaults, as-needed.

Users who do not specify these optional environment variables will see no change in behavior.

Resolved issues

Options required for passthrough to work correctly with large-BAR GPUs can now be applied by the user

Documentation

I have added a new unit test for the LXD.set_config() function, and for the additional arguments I added to LXDVM

Tests

This PR was verified on a DGX B200 (buzzsaw) with the following environment variables:

[environment]
QEMU_OPTS=" -global q35-pcihost.pci-hole64-size=8192G"
VM_RAM_MB=8192
VM_CPUS=8

This section was appended to the default ./test-gpgpu launcher. I then verified that the VM and container GPU passthrough tests both passed as expected when run with these parameters on that platform.

To verify that there were no regressions for other platforms, I loaded my branch on a platform with a consumer-grade GPU (luma), then ran the unmodified ./test-gpgpu and confirmed that all tests passed (as they had when this platform was certified originally) and no custom options were applied.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 50.63%. Comparing base (2dab338) to head (1e06546).
⚠️ Report is 115 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2007      +/-   ##
==========================================
+ Coverage   50.60%   50.63%   +0.02%     
==========================================
  Files         384      384              
  Lines       41180    41203      +23     
  Branches     7636     7639       +3     
==========================================
+ Hits        20841    20864      +23     
  Misses      19594    19594              
  Partials      745      745              

Flag	Coverage Δ
checkbox-support	63.95% <100.00%> (+0.01%)	⬆️
provider-gpgpu	93.06% <100.00%> (+0.44%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[MitchellAugustin]

I'm not sure why codecov claims my tests are missing coverage for the arguments it highlighted in yellow. I tested to confirm that both the test I wrote in de7fa54, as well as the autogenerated test_parse_args() test are hitting those conditionals. (I did this by adding a write to a file on my local branch that saves the passed value within that conditional, and confirmed that they were as expected when I ran python3 manage.py test.)

Since it seems the coverage was actually already the same before de7fa54, just with mock values, let me know if you want me to remove de7fa54 or keep it as-is.

With that said, this should be ready for review.

[MitchellAugustin]

Here are the passing tests on Buzzsaw:

• VM tests
• Container tests (as a control)

Independent of my changes, the VM tests will hang if they run after the container tests due to the driver still being in use. (@pedro-avalos and I have discussed addressing this in a separate PR.)

[pedro-avalos]

Thank you so much for all the research and effort you put into this. I've left some suggestions mainly for the arguments and argument-parsing.

[MitchellAugustin]

Thanks for the review @pedro-avalos . I applied all suggestions and will retest shortly and update you here once the B200 is available.

[pedro-avalos]

Great, thanks for adding those changes!

[MitchellAugustin]

@pedro-avalos I cleaned up and signed all of the commits in the tree, and verified that everything is still working as described previously. Let me know if you need any more adjustments.

[MitchellAugustin]

Thanks @pedro-avalos !