Fixing flaky CI

[JukkaL]

GitHub actions have been flaky recently. Often some jobs immediately fail. Example: https://github.com/python/mypy/actions/runs/13202245866/job/36856826752?pr=18630

I wonder if this is related to the recent switch to Linux ARM runners. Any ideas?

[sterliakov]

It is related to ARM runners, see my report here (closed as duplicate of discussion, so a bit difficult to navigate) and follow the link to https://github.com/orgs/community/discussions/148648#discussioncomment-11866202. They propose using 22.04 version for now, but I don't see any confirmation in thread that it works.

[cdce8p]

It is related to ARM runners

Yes, unfortunately they haven't been able to resolve it until know. The first reports of it are almost four weeks old at this point.

The latest I've read is that they suggest to try out the ubuntu-22.04-arm variants as @sterliakov mentioned. Probably because they've deployed a fix there first to see how it goes.

There are a few options now:

1. Just revert back to ubuntu-latest for now. (Revert most of Use new Github actions ARM runners for tests #18483)
2. Switch to ubuntu-22.04-arm and see how that goes.
3. Keep the current ubuntu-24.04-arm runners and hope they fix the issue soon. Usually all that's needed is to restart the failing job as it's just the action/checkout step which failed. This also applies to https://github.com/mypyc/mypy_mypyc-wheels where we use the arm runners to speed up the aarch builds.
   Looking at the workflow runs would suggest that it has already become more stable in the last few days. https://github.com/mypyc/mypy_mypyc-wheels/actions

I personally hope they fix it soon for good in which case we don't have to do anything.

[JukkaL]

Let's wait a few more weeks (until end of Feb), and if the situation doesn't improve, let's try switching to ubuntu-22.04-arm first, and if it doesn't help, we may want to revert to non-ARM runners. Reverting to 22.04 shouldn't have much concrete impact.

[cdce8p]

I've seen multiple workflow runs fail the few days. Reportedly ubuntu-22.04-arm should be fixed. Opened #18676 to see how it looks. We can always switch back later.

[sterliakov]

Just for the record, they now say in the linked discussion that the issue is resolved for 24.04 as well.