Pre-fork worker model
A server design where a master process starts a fixed pool of worker processes ahead of time, and each worker handles one request at a time. Apache prefork, PHP-FPM, and Gunicorn work this way. It isolates requests well, but every worker costs memory, and a request that waits holds its worker.
Runtimes
In Quorum#
PHP-FPM and Python's Gunicorn run a pool of workers that you size from 1 to 400 on the App Server card; PHP-FPM starts with 5. Each worker takes one request, so slow database calls fill the pool, and each worker costs RAM (120 MB in Gunicorn), so a big pool can swap. PHP-FPM's Static mode keeps the whole pool resident. Dynamic and On demand modes save memory but make 2% or 10% of requests wait 20 or 50 ms for a fork.
Read next#
Process per request, Thread pool, GIL (global interpreter lock), Little's law