Test downstream tests with pytest-xdist

[user27182]

No description provided.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Please upload report for BASE (main@08dcaaf). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #173   +/-   ##
=======================================
  Coverage        ?   98.03%           
=======================================
  Files           ?        2           
  Lines           ?      204           
  Branches        ?       29           
=======================================
  Hits            ?      200           
  Misses          ?        2           
  Partials        ?        2           

☔ 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.

[akaszynski]

Thanks for fixing the merge.

[akaszynski]

Compared with:
https://github.com/pyvista/pytest-pyvista/actions/runs/16125715470/job/45502024651

I'm seeing around a 3 minute improvement in test times:
https://github.com/pyvista/pytest-pyvista/actions/runs/16127303206/job/45507174163?pr=173

Specifics:
5m 15s for this branch for Unit Testing step in Downstream tests vs 8m 29s without pytest-xdist.

[user27182]

Yes we saw substantial reduction in times downstream too in pyvista/pyvista#7624. For the limited tests in this repo, we seem to not have any issues with the parallelism. But the full CI suite downstream was seeing concurrency issues across test_plotting and test_renderer, and we appear to have race conditions with reading mesh files with all the fixtures, predominantly with core testing... which led to some refactoring of fixtures but currently hit a snag with pyvista/pyvista#7631 since test coverage is not reporting properly with all the fixtures...

[user27182]

It's also a bit of an unknown if features like disallow_unused_cache from #169 will work without a chance of breaking, since that option uses a single global variable to store test states... not sure how that's managed in parallel.

[MatthewFlamm]

pytest-mpl has this workflow in a PR, although it hasn't been implemented yet. It would be useful to keep an eye on the progress there.

matplotlib/pytest-mpl#242