Fix test execution sometimes hanging (OSOE-129)

[Piedone]

Sometimes, for seemingly no good reason, the Tests step hangs and times out if we let it run. Let's fix this.

• See e.g.:
  • This build, logs:
    logs_364.zip. Happened on Ubuntu.
  • This one happened on Windows. Logs:
    logs_334.zip
  • Another one, ironically happened when setting the workflow timeout as a stopgap to prevent them from running for 6 hours.
  • See others under canceled Actions (note that cancellation can be by the user's request too, a run is only interesting if it took one hour and was canceled by the system).
• It seems that the test run doesn't even start dotnet test. Maybe a command before that is hanging.

[Piedone]

This might have been fixed by accident after moving the test-executing code to PS. We need to see if new such occasions happen.

[Piedone]

Nope, here's a recent one.

[Piedone]

And here.

[Piedone]

Hmm, it seems that if we set maxParallelThreads = MaxParallelTests = 2 then it reliably gets stuck. See the previous ones, and here and here. This coincides with the Azure D2v2 VM used by GitHub-hosted runners having 2 vCores. However, when the issues was originally opened, we didn't have such config. Also, this run as with a parallelism of 4. Maybe the multiples of vCores cause an issue?

[Piedone]

Doesn't seem so. Here and here a run with a parallelism of 3 got stuck. Maybe it's the fifth test run that changes parallelism that always hangs.

[Piedone]

Perhaps now it's fixed? No new timeouts since June.

[sarahelsaig]

I had a non-repeatable timeout in SNOW-119 just 2 days ago.

[Piedone]

Can you point to the build? It doesn't show up under https://github.com/Lombiq/Open-Source-Orchard-Core-Extensions/actions?query=is%3Acancelled, not I see any issues under https://github.com/Lombiq/Open-Source-Orchard-Core-Extensions/actions/runs/2848622770.

[sarahelsaig]

Ah, I didn't realize this issue was Actions-specific. Then disregard my previous comment. The timeout was in TeamCity.

[Piedone]

I see.

[Piedone]

There are still no such timeouts. I guess it solved itself, then. Good bug!