If you see these settings in a PHP-FPM config. file (real life example):
pm.max_children = 100
pm.start_servers = 30
pm.min_spare_servers = 30
pm.max_spare_servers = 100
What does it actually mean?
1) There will never be less than 30 processes (because it starts with 30, and minimum spare is 30)
2) There will always be at least 30 idle processes, except when more than 70 are in use.
3) Idle processes will never be killed off, unless they hit pm.max_requests (because the number of processes cannot possibly exceed the maximum spare)
Too many idle processes is bad, because they use up virtual memory. It is especially bad on a server with no swap, because then they can’t be shunted out of RAM to make room for active processes.
Also, remember that the settings are per-pool. On this particular server, there were three pools with identical settings. The result was that there were never less than 90 PHP processes, even if all were idle.
So please be mindful of the law of unintended consequences, especially when creating additional PHP FPM pools.
Credits to my ex colleague Danny 🙂