Pulp

Problem 1: Checking memory consumption of tasks running on workers is a multi-step process (hunt down PID, disambiguate it from all the others, etc).

Problem 2: It's difficult to measure the memory consumption total across multiple tasks and multiple workers.

Systemd allows you to opt-in to memory tracking, and it tracks across forks as well. So the memory consumption of a worker unit is the consumption of the entire group of processes.

Likewise the memory consumption of the "slice" is the memory consumption of the group of worker units, which is neat.