Flaky Tests

[fitzthum]

Anyone who has submitted a PR recently has probably noticed that the image-rs tests take a few tries. I think this is just because we aren't always able to pull the image. This seems to happen most frequently on s390x but maybe it is elsewhere too. It's not a huge inconvenience but maybe there is a way we can fix this?

cc @BbolroC

[fitzthum]

For example https://github.com/confidential-containers/guest-components/actions/runs/13429528372/job/37546759608?pr=916

[BbolroC]

Thanks for the notice. I will figure it out soon.

[BbolroC]

@fitzthum @Xynnn007 The test failure doesn’t seem flaky—it’s been failing consistently since #902 was merged. Interestingly, that PR didn’t trigger the CI job we’re looking at.

[mythi]

Interestingly, that PR didn’t trigger the CI job we’re looking at.

CDH workflow triggers were not updated when the pull functionality got moved there: changes to image-rs/* do not trigger CDH tests even if CDH now has a dependency to image-rs.

[Xynnn007]

CDH workflow triggers were not updated when the pull functionality got moved there: changes to image-rs/* do not trigger CDH tests even if CDH now has a dependency to image-rs.

That is the case.

[BbolroC]

Interestingly, that PR didn’t trigger the CI job we’re looking at.

CDH workflow triggers were not updated when the pull functionality got moved there: changes to image-rs/* do not trigger CDH tests even if CDH now has a dependency to image-rs.

Then, I think it would be great time to extend our dependency graph. 😉

[BbolroC]

Update:

The tests are passing after cleaning up /run/image-rs/ before running the test. Shout-out to @Xynnn007 ❤

https://github.com/confidential-containers/guest-components/actions/runs/13429528372?pr=916

[fitzthum]

Ok, we fixed the tests that were failing every time, but there are also some flaky jobs.

For example, https://github.com/confidential-containers/guest-components/actions/runs/13515374662/job/37762988211#step:20:474

We're pulling a container from the aliyun container service here. I wonder if we want to copy these containers to another registry? We could also keep that registry but skip that test on bare metal runners that might have more trouble connecting to it.

[mythi]

I wonder if we want to copy these containers to another registry?

+1