Process per request
A server model in which each request in flight occupies a whole worker process until it finishes, as in Apache prefork with mod_php, PHP-FPM, and Gunicorn sync workers. It isolates requests well, but every waiting request holds a process and its memory, so the pool size and the RAM cap how many requests run at once.
Runtimes
In Quorum#
Apache mod_php, the starting runtime, runs a process per request in flight, up to 150, each 60 MB because it embeds PHP, so RAM runs out early. PHP-FPM and Gunicorn run a fixed pool that you size (PHP-FPM ships with 5 workers); a request holds its worker while it waits on the Database, and a pool too big for the box swaps until the kernel kills it. Each process also holds its own database connection, so a big pool can hit Postgres's 100-connection limit.
Read next#
Pre-fork worker model, Shared nothing, Little's law, Connection pooling