Thanks for posting it here.

I just wrote one about RAM too - https://join.piefed.social/2026/10/11/reducing-piefed-ram-usage/

Would it not be better if most of the worker counts were dynamic?

Celery can do that and I used to use that mode.

Thing is you need to size your VPS to accommodate peak load anyway so you might as well just run all the processes all the time.

But that assumes nothing else runs on the VPS, no?
(And also that resource sharing on the VPS provider side is irrelevant)

I have to imagine it’s pretty common for a self-hoster to be using proxmox, where lower idle demand can be a benefit to other containers, and in the other case where piefed is on dedicated hardware lower idle resource use can mean lower power consumption

One of the reasons I switched to a fixed number of processes is because discarding and then spinning up new processes uses more CPU than just leaving them idle. And Lemmy sends traffic in bursts with long periods of silence in between so load is constantly spiking and processes are constantly being closed down and then recreated.

There's probably some celery config to control this and make it wait longer before closing down a process, which would keep them going during the silent periods... But then why not just have a fixed number of processes, heh.