Project: sefer nolad

Service: sefer-nolad-bethaorchim

Environment: production

Hi,

We run a single-replica Node service that renders PDFs using headless Chrome. Each render spawns a Chrome process tree that peaks at roughly 200 tasks, so we can only run three renders concurrently before approaching the cgroup PID ceiling.

What we measure from inside the container:

/sys/fs/cgroup pids.max 1000

pids.current at idle 24

pids.current per render ~200 peak

The constraint is purely the task count. At the same time, the machine itself is almost idle:

CPU ~3% of 24 vCPU

Memory ~4% of 24 GB

So we are paying for capacity we cannot use, and the only thing standing between us and roughly eight concurrent renders is pids.max.

Two questions:

Is pids.max configurable per service, and can it be raised for us? We could not find it documented for any plan.

If it is fixed, is it fixed per replica? We want to confirm that horizontal scaling gives each replica its own budget rather than splitting one.

We are not asking for more CPU or memory. Only the task-count ceiling.

Thanks.