Falcon doesn't buffer the response body. Are you using Rack 3? Can you print out the middleware stack? Can you run with falcon verbose mode to see how the output body is being handled?

@ioquatix ah I think I see the issue. It looks like request that establishes the stream (which creates an infinite loop to hold the connection) prevents the server from handling anymore requests.

This PR is not using Rack 3 @ioquatix (I'm working on it in #1857) – if we're talking Rack 3, would it be easier to get our tests and example working with Puma/Falcon if we adopted "callable body" (rack/rack#1745, puma/puma#2740)?

I need to play with this myself, and catch up on everything discussed, rack/rack#1600 also has a fair bit of discussion, still have a big bunch unread emails from rack/rack from last year and summer 2021 😅

@dentarg you should consult the upgrade guide if you want to see a summary of all changes that affect applications/middlewares): https://github.com/rack/rack/blob/main/UPGRADE-GUIDE.md

Yes you should rely on callable bodies as this is the correct way to implement streaming with Rack. Everything else is a hack and may have unintended buffering.