test for ensuring hydra ddp raises is raising for the wrong reason

[rsokl]

@jgbos

In the following test:

responsible-ai-toolbox/tests/test_mushin/test_lightning_hydra_ddp.py

Lines 27 to 40 in 3882320

	@pytest.mark.usefixtures("cleandir")
	def test_ddp_with_hydra_raises():
	trainer = builds(
	Trainer,
	max_epochs=1,
	accelerator="auto",
	devices="${devices}",
	strategy=builds(HydraDDP),
	fast_dev_run=True,
	)
	module = builds(TestLightningModule)
	Config = make_config(trainer=trainer, wrong_config_name=module, devices=2)
	with pytest.raises(TypeError):
	launch(Config, pl_main_task)

launch(Config, pl_main_task) raises TypeError because pl_main_task doesn't accept a single config (pyright warned me about this). I doubt this is what you meant to exercise in this test.

I am confused about what this test is doing. Config = make_config(trainer=trainer, wrong_config_name=module, devices=2) makes it seem like we are making sure that launch fails for a config with a bad field name, but the test seems like should be exercising ddp

[jgbos]

Ok, I'll take a look at this. Not sure we need this exact test, but instead need some way to raise an error that is useful to the user when using HydraDDP since it requires that the user set configs for trainer, module, and for PL 1.6 pl_testing (automatic for 1.5). Here I was just making sure it raises the error.

[jgbos]

Just to clarify, I'm not sure I can test for an error being thrown in the subprocess call in HydraDDP, so I wanted to make sure it would throw an error. Unfortunately if it throws an error in the subprocess I think the process will just hang.

[jgbos]

The tests are fixed but I need to think of way to catch the possible configuration error in the subprocess.