New BEAM There, Done That episode relevant to anyone building on the BEAM, not just Phoenix users.
The most Erlang-relevant discussion: each connection gets its own process, and when an error occurs anywhere in request handling the resulting trace is entirely BEAM-native - no opaque C boundary. You can read it top to bottom.
Mat also covers the HTTP parser security surface in terms this community will recognise: resource consumption attacks, connections held open indefinitely, oversized inputs. Protocol-level problems independent of language that foundational libraries have to solve correctly.
The episode closes on what it looks like to maintain infrastructure at this level when AI-generated contributions start arriving - there’s an HTTP/3 PR open on the repo and the reasoning behind leaving it unmerged is worth hearing.
