Fedora rawhide: podman checkpoint e2e and system test failures

[lsm5]

Starting with crun, can move to criu if required. This is seen on Fedora rawhide only, both aarch64 and x86_64.

Summarizing 3 Failures:
  [FAIL] Podman checkpoint [It] podman checkpoint container with established tcp connections
  /var/tmp/tmp.pxsEr2Kr3j/podman-5.6.0-dev/test/e2e/checkpoint_test.go:367
  [FAIL] Podman checkpoint [It] podman checkpoint and restore container with different port mappings
  /var/tmp/tmp.pxsEr2Kr3j/podman-5.6.0-dev/test/e2e/checkpoint_test.go:1089
  [FAIL] Podman checkpoint [It] podman checkpoint and restore containers with --print-stats
  /var/tmp/tmp.pxsEr2Kr3j/podman-5.6.0-dev/test/e2e/checkpoint_test.go:1383

Links:

• fedora-rawhide-aarch64 log
• fedora-rawhide-x86_64 log
• container-selinux PR

EDIT: didn't notice earlier, 1 system test failed too: not ok 3 [520] podman checkpoint --export, with volumes . You'll find it in the lower log on the same link.

Note to self: enable podman checkpoint e2e tests on crun PRs as well. Currently, we only have system tests running.

[Luap99]

If I had to guess this is a new kernel <-> criu interaction issue that we regularly get with new kernels since 6.16 merge window builds are hitting rawhide now. I don't see a crun or criu update that leaves the kernel as only culprit to me.

It would likely be good to reproduce manually to get the output from dump.log so we know the exact error.
cc @adrianreber @rst0git

[rst0git]

@Luap99 This seems to be related to

• Debug rawhide failures checkpoint-restore/criu#2625
• Support MADV_GUARD_INSTALL checkpoint-restore/criu#2626

[eriksjolund]

I see different behaviour when running

podman --runtime crun container checkpoint --export /tmp/export.tar.zst mycontainer --leave-running 

The container stops running on Fedora 42 (kernel: 6.15.4-200.fc42.x86_64) but when I try it on
Fedora CoreOS (kernel: 6.14.11-300.fc42.aarch64) the container keeps on running (as it should).

For details, see
podman-container-tools/podman#26576

[Luap99]

@rst0git Just saying 6.16 is released and this is still broken. We have skipped all the broken checkpoints tests by now for our CI podman-container-tools/podman#26289.

Anyway I guess we can close the crun issue as this is not a crun problem.