Thursday, February 4, 2021

Running older systems (which need cgroupv1) on systems running over cgroupv2 (systemd.unified_cgroup_hierarchy)

Run the command -- mount | grep cgroup on your host system, and if you see the all the mount entries as cgroup2 fs (instead of cgroup), then you wont be able to run run older OSs as containers on this host. If you try to force cgroup2 over cgroupv1, the following errors will occur -- 

Cannot determine cgroup we are running in: No such file or directory

Failed to allocate manager object: No such file or director

An e.g. of what happens in centos 7 on lxc.

For older systems which don't support cgroupv2, you’ll need cgroupv1 mounted in /sys/fs/cgroup/systemd on the host. There doesn't seems to be way to do this using lxc.mount.auto = ; so you’ve to use scripts (lxc.hook.mount). For this script to mount a cgroup (named X) in the guest, a cgroup named X must also be mounted on the host; this same cgroup will be made available to to the guest. Alternatively, you may mount –bind in this script from the host’s cgroupv1 mounted directory to the guest’s directory; this’s a better approach since this allows you to create cgroups inside X exclusively for the container, so the guest may not play around with other processes's cgroups.
As an e.g. –
#! /bin/bash
mount -t tmpfs -o size=1M tmpfs $LXC_ROOTFS_MOUNT/sys/fs/cgroup/
mkdir -p $LXC_ROOTFS_MOUNT/sys/fs/cgroup/systemd
#mount -t cgroup -o none,name=cgroupv1 cgroupv1 $LXC_ROOTFS_MOUNT/sys/fs/cgroup/systemd &>> /tmp/script_out.log
mount --bind /tmp/cgroup1/lxc_containers $LXC_ROOTFS_MOUNT/sys/fs/cgroup/systemd
exit 0

Can't get cgroupv1 mounted no your host? Getting "already mounted or mount point busy." -- in this case ensure the cgroup that you're mounting is not being attached to any subsystem/controllers, which is the default behavior. This's the right approach -- 

mount -t cgroup -o none,name=lxc_compat systemd /tmp/cgroup1