Mem, vmem, and swappiness with cgroups

Hi all,

I’m confused by how memory & swap are handled when the cgroups hook is enabled. In my config, I have the “memory” subsystem enabled, with the “swappiness” set to 0 and “soft_limit” to false. The “memsw” subsystem is disabled. Also, I have “$enforce mem” set in the MoM’s configs.

My expectations are that (i) the jobs will run only in RAM and (ii) any attempt to use more RAM than specified in the “mem” job resource would result in the job being killed. Instead, I see that such jobs continue to run, with all the excessive allocated memory put in the swap. What am I missing?

Using OpenPBS 20.0.1 with the groups hook from the git master.