ci: add workflow to run GPU tests on selfhosted runner#3543
ci: add workflow to run GPU tests on selfhosted runner#3543ianna merged 5 commits intoscikit-hep:mainfrom
Conversation
|
you will probably be getting this: #3480 which needs fixing |
Oh yeah, that's the one. I didn't realize it was already reported |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files🚀 New features to boost your workflow:
|
Yeah I found this some time ago on my desktop. I rarely install cudf though. |
190585d to
cf1d576
Compare
|
@ariostas - I suggest we mark the tests-cuda/test_3051_to_cuda.py::test_null PASSED [ 24%]
Error: test_strings
TypeError: StringColumn.__init__() missing 2 required positional arguments: 'size' and 'dtype'
This error occurred while calling
ak.to_cudf(
<Array ['hey', 'hi', 'hum'] type='3 * string'>
) |
Sounds good. I'll make a couple of final tweaks. I'm just trying to simplify the setup a bit. |
50443f5 to
cf86ace
Compare
cf86ace to
fa1cd5e
Compare
|
Okay, this is ready now |
I'll enable auto-merge then. Thanks! |
|
@ianna would the wiki be a good place to write some documentation about how to set up the runner? |
if it's awkward-specific, I think a maintainer-guide directory in docs would be good. |
This PR adds a workflow to run the GPU tests on the new CI node at Princeton. There are a couple of things to discuss:
pass-tests.needssince I think we should first make sure that it is reliable enough that it's not going to end up blocking PR regularly.test.ymlworkflow. This means that it will run for all PRs and pushes to main. There are some security concerns with this, so we could consider moving it to a separate workflow that needs to be triggered by aworkflow_dispatchtrigger, which only people with write permissions can do. And we could also run nightly tests to make sure things are not breaking. But I think it might also be fine to leave it it with the normal workflow for the following reasons:I don't have permissions to add a runner for this repo, so I'll need someone (probably @ianna) to help me with that. I tested it on my fork and it's working well. Here are the logs for a test run I did. There's actually one test that fails, so we should look into that.
After we settle all these things, I'll add a page in the wiki describing how to set up the selfhoster runner.