We currently have a handful of Posit Workbenches running through containers on a SuperdomeFlex server (960 cores, 7.5 TB RAM). On top of these containers, we also have users submitting calculations through the SLURM job scheduler.
When there is heavy usage of containers and SLURM, we experience notable slowdowns since SLURM cannot "see" the resources that the containers are using. I spoke with SchedMD about how to carve out resources from what SLURM manages, and they noted that the first step was to set cgroups for the containers? Does anyone here know how we can go about setting those cgroups so we can carve the resources these containers use from those that SLURM manages?