Puma cluster worker using huge amounts of memory

I am running Katello as well. There are about 300 clients. According to the documentation, 8/48 should be overkill for this setup. The following plugins are installed:

foreman-tasks
foreman_hooks
foreman_openscap
foreman_puppet
foreman_remote_execution
katello

I included the lines from top just to show the VIRT/RES of each rails (puma worker) process. After another hour, swap is also almost exhausted, and the VIRT/RES of each rails process has increased by 1.5-2x.