[lxc-users] Strange CPUSET CGROUP behaviour

Ben Green ben at bristolwireless.net
Thu Sep 3 14:27:37 UTC 2020


Hi all,

We are starting to use LXC in production here, after solving our final
problem (inconsistent and disappearing cgroup space - Thread name
"Disappearing Cgroups" on this list) completely with the help of Serge
E. Hallyn.

Now something odd has come up. I have one server on the Debian kernel
4.19.0-9-amd64 and another on 4.19.0-10-amd64. On the 4.19.0-9-amd64
kernel we cat this:

# cat /sys/fs/cgroup/cpuset/forcontainers/tasks
1216
5065
12372
etc.

But on the other one with the 4.19.0-10-amd64 we get this:

#  cat /sys/fs/cgroup/cpuset/forcontainers/tasks

No output! Also on the 4.19.0-10-amd64 machine it is not possible to add
to the list of tasks:

# echo $$ >> /sys/fs/cgroup/cpuset/forcontainers/tasks
-bash: echo: write error: No space left on device

However, we've had this happen before, and thought that adding
'swapaccount=1' to the grub command line arguments had fixed it. Seems
it hadn't. I'm not even sure therefore that this is consistent between
kernels. It appears it might be entirely random.

Has anyone else experienced this phenomena? Any clues as to what might
be happening?

It seems that the cgroup has a messed up mount, but it looks fine to me:

  cgroup on /sys/fs/cgroup/cpuset type cgroup
(rw,nosuid,nodev,noexec,relatime,cpuset)

Remove and remake the /sys/fs/cgroup/cpuset/forcontainers directory and
the problem persists.

-- 
Cheers,
Ben Green


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <http://lists.linuxcontainers.org/pipermail/lxc-users/attachments/20200903/ab39962f/attachment.sig>


More information about the lxc-users mailing list