I have several RHEL7 machines that are part of a GPFS cluster and when the memory is depleted the GPFS daemon and drivers crash bringing down the filesystems that they manage.

There are 3 cgroup memory hierarchies that I want to limit: user.slice, kubepods.slice and lsf (for the LSF job queue). I would like to put a 100G limit on the total memory consumption of these 3 groups. For that I would have to either create a new root cgroup and put them inside this group or choose one of them to be the root. The problem is that the user.slice and lsf cgroups seem to be pretty much hardcoded.

Does anyone know of a way to reorganize these hierarchies? The documentation is pretty obscure or missing.