LXC - Host CPU usage shown in htop and other applications

[ferdl9999]

Hello everyone,

since I have updated from PVE 6 to PVE 7 and PVE 8 I notice that on LXC containers in for example htop or other applications that are reading out the CPU utilization, the host CPU usage is shown and not the LXC container CPU usage. Even when there is no load at all on the LXC container, there is utilization on the CPU cores shown in htop for example. However, via Proxmox API and Proxmox GUI the "correct" CPU usage values are shown. When using PVE 6, this was not the case - the correct CPU usage values were shown here in htop and other applications.

Does anyone else see this behavior too? Is there any solution for it?
Unfortunately, I do not find any information regarding this in the internet (only related to wrong load / host load shown).

Thank you for your help!

 

[Dunuin]

Does anyone else see this behavior too? Is there any solution for it?

Yes, that's normal. With PVE 6 there was a workaround (adding the "-l" flag to lxcfs.service) but I think I remember some posts complaining that this isn't working anymore. Probably you used that workaround, because if I remember right, PVE6 was showing the hosts CPU utilization too.

The problem here is htop and so on are reading the CPU utilization at the wrong location and not at the location where the LXC is offering the correct values.

See for example here: https://forum.proxmox.com/threads/lxc-loadavg.44463/#post-239269

 

[ferdl9999]

I think with adding the "-l" flag you only "fixed" the issue of the shown load avg that was identical to the host load avg. But I think the utilization of the CPU cores was per default "correct". But I am not sure about that, however.

I have a lot of users asking why in htop or other applications there is some higher CPU core usage shown even if they have basically nothing causing CPU load running on their LXC container.
Is there any workaround for this behavior in PVE 7 / PVE 8? I have a lot of nodes still running PVE 6 because of this behavior, a lot of endusers (on the PVE7 / PVE8 nodes) are frequently asking about these incorrect CPU usage values. Maybe doesn't sound like such a big problem for other members here, but for me it is unfortunately. Thanks in advance for your help!

// EDIT: Just tested using the "-l" flag on lxcfs.service on the newest PVE 8 and it fixed the incorrect CPU load avg, but not the core utilization

 

[ferdl9999]

Update: Found this issue: https://github.com/lxc/lxcfs/issues/538

Adding systemd.unified_cgroup_hierarchy=0 to the kernel boot arguments fixed the issue for me. But is there any better solution? Does it have any negative effects?