[lxc-users] Cgroup saga continues - lxc-autostart woes.

[Ben Green]

On 21/11/2020 21:54, Serge E. Hallyn wrote:
> I've never used lxc-autostart, but looking at the manpage, do you
> have lxc.start.auto set in the containers you want to start?  What
> do the configs look like?
>
> Do the containers actually start but in the wrong cgroup?  Or do
> they just not start?

They start, but the tasks part of some cgroup filesystems are 
unpopulated with PID numbers. This is particularly troublesome for the 
memory cgroup, for example: 
/sys/fs/cgroup/memory/forcontainers/lxc/example_server/tasks would 
contain nothing. This means the contains ends up with the memory limits 
assigned to the host.

I can get the memory limits to work again, but copying the tasks pids 
from /sys/fs/cgroup/cpuset/forcontainers/lxc/example_server/tasks into 
/sys/fs/cgroup/memory/forcontainers/lxc/example_server/tasks. Then the 
server reports the required memory allowance again.

Other cgroups file systems are affected too, 'devices' and 'blkio' for 
example. We aren't doing anything with those though, so I'm 
concentrating 'memory' which I really need to make use of.

Cheers,

Ben