Possibly relevant to the issue: on all affected VMs, I run a self-built kernel 3.14.59 with most cgroup options disabled: CONFIG_CGROUPS=y # CONFIG_CGROUP_DEBUG is not set # CONFIG_CGROUP_FREEZER is not set # CONFIG_CGROUP_DEVICE is not set # CONFIG_CGROUP_CPUACCT is not set # CONFIG_CGROUP_PERF is not set # CONFIG_CGROUP_SCHED is not set # CONFIG_CGROUP_NET_PRIO is not set CONFIG_CGROUP_NET_CLASSID=y That was not an issue in the past. Normal cgroup grouping of processes works, as can be seen with "systemctl status" on normal process groups.