Fewer workers, if the RAM is the limit. Not more.
pm.max_children against the memory you paid for, a slow-request log, opcache actually on. Raising the pool on a 2 GB VPS is how it starts swapping.
Idle workers, max children reached, request time
If max children is never reached, the pool is not the wait. Adding workers does nothing. The time is in the script or the database.
A WooCommerce request is not a 32 MB process. The ceiling is RAM minus MariaDB, divided by the real size. The rest wait, instead of swapping.
Off, or full, so every request compiles. We turn it on and size it. We do not disable it to “fix” a stale file. A reload does that.
A dead socket is support. A slow query is MariaDB.
Optimization assumes the pool is running and the site answers. If it 502s, or one query holds every worker, we switch jobs and say so.
€55 an hour. The first hour is the measurement.
Minimum 1 hour. Further changes are agreed. A bigger VPS is recommended only if the measured worker size cannot fit the traffic you already have.
Site slow under load, fine when you are alone on it?
Tell us the RAM and the PHP version. No root password in the first message.