Does the Frankenphp symfony Runtime class run only once in worker mode?

[aszenz]

I have a custom symfony runtime that accesses HTTP headers, when extending this runtime from Frankenphp's own symfony runtime I find that HTTP headers are missing.

I guess it makes sense that the runtime is only run once and lacks access to http headers since those are per request but wanted to confirm from other contributors.

Related question: If I cannot access HTTP headers in the Runtime, what's the next best place to access them in a Symfony app, I want to access them as soon as possible in the request processing cycle.

[AlliBalliBaba]

How are you creating a new request? Headers should be present in $_SERVER as with a regular PHP request. Otherwise they can also be accessed through apache_request_headers() or frankenphp_request_headers().

Is your question if this class is just run once?

[aszenz]

Yes that's the question is the runner class run only once. Basically the worker script

[AlliBalliBaba]

The worker script is run in a loop. When it's done, it just starts again. During each run, the worker script handles an amount of requests through calling frankenphp_handle_request in another loop. The amount of requests is usually limited to reduce the impact of potential memory leaks.

-> worker script starts
-> worker script encounters `frankenphp_handle_request` and waits for a request
-> worker script receives a request and handles it
-> worker script encounters `frankenphp_handle_request` and waits for a request
.....
.....
-> worker script has reach max amount of requests
-> worker script finishes
-> worker script starts
-> worker script encounters `frankenphp_handle_request` and waits for a request
.....

[aszenz]

Cool, so basically when it starts it has no idea of any request.

How many requests are typically handled by a single worker before it gets restarted?

Also are requests processed in threads/coroutines in a single worker?

[AlliBalliBaba]

I'd say ~1000 requests is usually fine. If your app is well set up and you only use popular extensions you can theoretically also do infinite requests, but at some point you might also want to recycle active connections to external services.

The worker will not restart automatically though, you have to set the limit in the runtime (like here in do(...) ).

Requests are processed in threads. Each worker has many threads (can be configured in the Caddyfile).
Requests are dispatched to threads in order to minimize connections. So if the first thread is busy, check the second thread. If the second thread is also busy, check the third thread and so on . You can also configure it to dynamically spawn threads if all threads are busy.

Currently no coroutines (at least on the PHP side) since most native PHP IO is still blocking.