New config to adjust timeouts between NGINX and PHP-FPM (fix PHP zombie process)

Description

Hi,

I have a question that I haven't found the answer online. I made this graph to illustrate the situation. When there are many requests waiting in the backlog queue (because the workers are busy), a time desynchronization happens between NGINX and PHP-FPM.

The PHP-FPM continues getting the requests from the backlog queue even though NGINX has already answered the client with 504 response.

This is a waste of resources, as the PHP result will never reach NGINX, as NGINX has already responded to the request with 504 to the client

When this happens the new requests receives a timeout because they go to the end of the queue and are processed with delay

If there is a configuration that takes into account the time that the process spends in the backlog queue, these "zombie" processes do not even need to be processed if they have already reached the timeout limit

Issue に記載されている PHP-FPM のバックログ処理と NGINX のタイムアウトとの相互作用の調査から始めてください。要求された設定動作が実現可能かどうかを判断し、クライアントタイムアウトをすでに超過したリクエストをどのように処理するかを定義してください。合意済みの設定設計と、検証済みのタイムアウトシナリオが決まれば完了です。