FrankenPHP Worker Runtime vs Classic Symfony: Filters Not Applying

[Selanas]

Hi everyone,

I need some help understanding a pretty weird problem I’m encountering with FrankenPHP’s worker mode and the FrankenPHP Runtime.

Premises:
Both environments run the exact same codebase, built into the same container image. The environment configurations are the same (except for the runtime runner). The Caddy server is the same, and so is the Caddy config. Everything is identical except for the running context. No cache issues, no gateway troubles—I’ve triple-checked all of that.

The first environment runs locally with the classic Symfony index.php. The second runs in worker mode using the FrankenPHP Runtime.

Problem:
When I run the exact same request, only on the worker mode run, my filters seem to be ignored. I go from 1 query to 63 queries (just some bad eager loading; the exact number isn’t what’s important here). The filters are ignored, and I’m left with a simple getCollection result.

All parameters are correctly passed in the trace, and I’ve tried both dev and prod mode. As soon as I redeploy without the worker mode, everything works fine again.

I’ve been stuck on this for hours, and I could really use a more distant/expert perspective on the matter. Any ideas to help me understand this problem?

Extra informations:

• The project is built around API Platform.
• The architecture follows Domain-Driven Design.
• We’ve re-implement the API Platform’s out-of-the-box filters with a custom providers using #[AutowireIterator(tag: 'api_platform.doctrine.orm.query_extension.collection')].
• A switch selects the correct kernel using $context['APP_ID']. (should'nt have any inpact at this point i think)
• No mercure hub or Vulcain at this point of the project

[withinboredom]

This seems suspiciously like some part of the request is from another request -- perhaps a dependency being injected isn't being reset after a request?