test: skip oomd test on a unified container on a hybrid host

[yuwata]

Fixes #20593
Fixes #20655.

[yuwata]

(I have no idea what we should do in oomd on such a case.)

[yuwata]

cc @anitazha and @DaanDeMeyer

[DaanDeMeyer]

Slightly out of my league as well. Let's wait for anita.

[anitazha]

I need to check in more detail when I'm not AFK but memory.current is provided when the memory controller is enabled. So I think the correct message would be something like "cgroup2 memory controller is not enabled". And the actual daemon should also include a condition check for the memory controller.

[yuwata]

Ah, checking the existence of memory controller is better. Thanks. Will update.

[yuwata]

@anitazha Updated. PTAL.

[bluca]

test-oomd still unhappy

[yuwata]

Still not sure if this is a correct way, but at least, now semaphoreci is green. PTAL.

[yuwata]

What's the difference between cg_kernel_controllers() and cg_mask_supported()??

[anitazha]

What's the difference between cg_kernel_controllers() and cg_mask_supported()??

It looks like cg_kernel_controllers() parses /proc/cgroup which lists all the controllers (cgroup v1 and v2) compiled in the kernel (including information about how many cgroups are using them and whether they were disabled via boot parameter, etc.). But cg_mask_supported() will figure out which hierarchy systemd is using and list the available controllers based on the current hierarchy. For unified (cgroup v2) this will be /sys/fs/cgroups/cgroup.controllers.

Thanks for fixing this Yu the current approach lgtm. Should be good to merge after removing the calls to cg_kernel_controllers()

[yuwata]

@anitazha Thank you for your review and the explanation. cg_kernel_controllers() are dropped. Upgrading the green label.

[bluca]

LGTM, thanks for fixing this, we can go back to seeing all green ticks yay