[lxc-users] lxc 2.0.7: sysvinit on the host breaks systemd based containers
Serge E. Hallyn
serge at hallyn.com
Thu May 4 19:00:50 UTC 2017
Quoting Harald Dunkel (harri at afaics.de):
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA256
>
> On 04/03/17 07:03, Harald Dunkel wrote:
> > Hi folks,
> >
> > using sysvinit-core on the host the systemd based containers get stuck in /sbin/init. lxc-attach shows:
> >
> > root at lxcclient:~# ps -ef UID PID PPID C STIME TTY TIME CMD root 1 0 0 11:49 ? 00:00:00 /sbin/init root 24 0 0 12:05 pts/0 00:00:00 /bin/bash root 26 24 0 12:05 pts/0 00:00:00 ps -ef
> >
> > This host uses lxc 2.0.7, as it can be found in Jessie backports. See below for a list of installed packages. The container was created on the same host.
> >
> > I am bound to sysvinit-core on my HA systems. For lxc 1.1.5 there was no such problem, so I wonder if there is a workaround on the host?
> >
>
> PS: If I drop /cgroup from /etc/fstab on the host and reboot, then
> the problem seems to be gone.
>
> Does this ring a bell?
Sounds like just systemd refusing to boot because all cgroups are comounted?
Are you sure that reverting to 1.1.5 fixes it, and it's not a newer systemd
breaking it?
One thought that comes to mind - perhaps in the past you used the lxcfs cgroup
bind mounts, and now you try to use the cgroup namespaces. The lxcfs cgroup
mounts always created a filesystem per cgroup controller.
It would help to ask for more debugging information from systemd,
lxc.init_cmd = /sbin/init log_target=console log_level=debug
as well as looking at /sys/fs/cgroup in the container while systemd is
hung.
More information about the lxc-users
mailing list