CI: vfkit broken on macos tahoe runners

[Luap99]

I think vfkit v0.6.2 might be broken at least in our CI env.

  → Enter [It] no settings should change if no flags - /Users/MacM1-5-worker/ci/task-6413995885723648/pkg/machine/e2e/set_test.go:96 @ 01/07/26 09:22:16.64
  /Users/MacM1-5-worker/ci/task-6413995885723648/bin/darwin/podman machine init --disk-size 11 --image /private/tmp/ci/podman-machine.aarch64.applehv.raw a509bf48caa4
  Machine init complete
  To start your machine run:

  	podman machine start a509bf48caa4

  /Users/MacM1-5-worker/ci/task-6413995885723648/bin/darwin/podman machine set a509bf48caa4
  /Users/MacM1-5-worker/ci/task-6413995885723648/bin/darwin/podman machine start a509bf48caa4
  Starting machine "a509bf48caa4"
  Error: vfkit exited unexpectedly with exit code 1

  [FAILED] Expected
      <int>: 125
  to match exit code:
      <int>: 0
  In [It] at: /Users/MacM1-5-worker/ci/task-6413995885723648/pkg/machine/e2e/set_test.go:112 @ 01/07/26 09:22:24.151

  Full Stack Trace
    github.com/containers/podman/v6/pkg/machine/e2e_test.init.func18.3()
    	/Users/MacM1-5-worker/ci/task-6413995885723648/pkg/machine/e2e/set_test.go:112 +0x3fc

https://api.cirrus-ci.com/v1/artifact/task/6413995885723648/html/machine-applehv-podman-darwin-rootless-host.log.html#t--podman-machine-set-set-machine-cpus--disk--memory--1

I guess someone needs to run that with log level debug to get the error message?

cc @baude @ashley-cui @cfergeau

[cfergeau]

Thanks for the heads up! I’ve created a v0.6.2 tag yesterday, but haven’t fully finished the release yet, it’s good to have early feedback!
I’ve just tested podman 5.7.1 with the unsigned binary from https://github.com/crc-org/vfkit/actions/runs/20755542863 and podman machine start worked fine, so it’s not totally broken. I’ll take a closer look at the output from your tests.

[cfergeau]

I cannot reproduce these failures locally on my M1, not sure what fails in CI, I’ll try again tomorrow.

[Luap99]

To be clear I am not confident that vfkit is the issue I just noticed the new bump so I thought it is related. Without a reproducer it might still be related to something else. I know I bumped the our macos runners to tahoe but that was on monday but maybe it is related to that since we didn't have other runs since then.

[Luap99]

Ok I finally checked on the failing macos worker it runs on vfkit 0.6.1 unless I am missing something about the CI setup.

So the version bump was a red herring to me, but it does seem to fail on our worker with tahoe so I guess we need to try to reproduce on that, maybe something releated to the aws image we pull in there

ProductName:		macOS
ProductVersion:		26.2
BuildVersion:		25C56

[timcoding1988]

i cannot reproduce this in my mac with.

➜  ~ podman -v
podman version 5.7.1
➜  ~ vfkit -v
vfkit version: v0.6.1
➜  ~ sw_vers
ProductName:		macOS
ProductVersion:		26.2
BuildVersion:		25C56

this could be ami image issue? @Luap99

[baude]

@cfergeau did you guys hit anything like this in your testign with tahoe?

[cfergeau]

@cfergeau did you guys hit anything like this in your testing with tahoe?

Testing on Tahoe is not as extensive as we’d like as the only macos runners with virtualization support are limited to macos 15. I should take a look at what you are using for your macOS e2e tests, maybe vfkit could use something similar.

There are no known regressions that I know of when moving from 15 to 26, or from vfkit 0.6.x to 0.6.2, but I’m trying to reproduce this issue to get a better idea of what’s going on. I’m running the e2e tests with podman 5.7.1 for now, but no failure. This report was against the main branch though, so I’ll be testing this next.