It looks like this run was canceled, not that it failed. Is that right?

Ya, not sure what that was about. I'll see if I can restart it.

Hmm, all the jobs ended early again. Two again are marked as failed but show canceled, and the rest are canceled. Not sure what's going on with that.

I can run py.test locally and have it pass, but I can't test against the whole matrix while I'm on plane wifi.

You need to add a fail-fast: false to force everything to run. Otherwise GHA will cancel all tests as soon as the first one fails.

@abkfenris Should I make that change to the github workflow in this PR?

Ya, we can try that, but I'm not certain that it will actually fix the underlying issue. I tried running things again yesterday and it still is showing two jobs as failed, but the logs of all jobs say canceled.

https://github.com/xpublish-community/xpublish/actions/runs/5894225786?pr=225

It sounds like there may be some resource contention issues, though we've had the same workflow run successfully on other PRs.

I can sometimes get a single job to pass https://github.com/xpublish-community/xpublish/actions/runs/5894225786/job/16109245196?pr=225