How it decides

This section is what fpm-tune does with a host's memory and CPU between reading the status pages and writing pm.max_children. Read it before switching a daemon to apply mode, or when a plan's number looks wrong and you want to know where it came from.

• The budget: where the memory it divides comes from, what it holds back, and how to give php-fpm a hard limit.
• Dividing the budget: the allocator: floors, demand, and what happens when the host is out of capacity.
• Measuring workers: what one worker costs, which readings count, and when a pool may be shrunk.
• Spawned children: the ffmpeg behind a worker, how it is measured, and how to declare it before it is.
• Static, dynamic, ondemand: what it writes for each pm mode, and the one suggestion it makes.
• Hysteresis: when a change is worth a reload.
• CPU per request: which of memory and CPU a pool runs out of first, and the ceiling --cpu holds it at.
• CPU measurement: appendix: how the CPU figures are measured.