Sometimes the test "docker run multi arch image with arm64 digest on amd64 host fails" does not work running in GitHub Actions

[davidkopp]

Here two runs where the test test_docker_run_multi_arch_image_with_arm64_digest_on_amd64_host_fails did not raise an exception as expected and thus failed:

• https://github.com/green-coding-solutions/green-metrics-tool/actions/runs/17879739592
• https://github.com/green-coding-solutions/green-metrics-tool/actions/runs/17792358683

In most cases the test succeeds. I don't know what could cause that sometimes the test fails on GitHub Actions.
There seems to be an issue in the container compatibility detection logic, but I have no clue at the moment what it could be.

Are there differences in the servers used for GitHub Actions in regards of the docker orchestration?

[ArneTR]

To my knowledge the docker CLI used in GitHub Actions is the official one from Docker. So no nerdctl or similar replacement.

The machines should also be plain x86_64.

The only quirk I know from GitHub Actions is that the python cache gets polluted very often (which should not be the case here) and also that it often has network hick-ups (could that be the case? ... or should that not also lead to an error?)

So the only way to debug this properly IMHO would be to add more print() debug statements into the runner flow that can be optionally turned on. We then activate that for the pipeline and also instruct pytest to capture the STDOUT.

Can you introduce the statments that would help you debug the feature further to main?

I also tried to reproduce the error on my local linux machine but test seems to work fine ...

[davidkopp]

Interestingly, the arm64 image was successfully started on the GitHub Action machine.
From the logs:

Creating container
Checking image architecture compatibility...
Warning: Container will use emulation (image: arm64, host: amd64)
Running docker run with: docker run -it -d --name test_service -v /home/runner/work/green-metrics-tool/green-metrics-tool:/tmp/repo:ro --net GMT_default_tmp_network_2406265 alpinesha2564562b419adf48c5f3c763995d6014c123b3ce1d2e0ef2613b189779caa787192_gmt_run_tmp
Stdout: 9277c094b9d00f00be70f1c613a1a54736ce22bbd7ad15900c08ff14e8cba65b
Running commands
Started containers:  ['test_service']

I now have two questions I don't understand:

1. How is it possible that GitHub Action is sometimes actually able to run an ARM64 image using emulation?
2. Why was the ability to emulate ARM64 images not correctly detected during test collection (regarding the skipif condition)? Is there a race condition?

To understand it better, the PR #1371 adds some more debug information to the test.