Running tests is beginning to become extremely time consuming

[LukeWood]

Does anyone have any ideas to improve this?

[sebastian-sz]

I think we could consider running pytest in parallel, using pytest-xdist.

Baseline: Current master branch + default command pytest -v keras_cv/ takes 472.09s (0:07:52) on my machine.

Running 4 processes with pytest-xdist slices this time more than 2x:

pip install pytest-xdist
pytest -v keras_cv -n 4  # 189.70s (0:03:09)

This argument can be probably tuned, depending on the machine.

This could be a starting point.

[bhack]

We could also evaluate to cluster tests in different jobs as jobs in GitHub Action runs in parallel.

We could also check if there are few tests with a very specific execution overhead.