Re: [quicwg/base-drafts] H3 GOAWAY should be symmetric and cover bidi and uni streams (#2632)

[Lucas Pardue <[e-mail]>]

> I think this argument makes sense, but this feels increasingly like a transport feature and not an HTTP feature.

I don't have a strong opinion here. For reference, the original issue about moving GOAWAY out of transport is https://github.com/quicwg/base-drafts/issues/696

For the separate frames topic:
HTTP/3 currently lacks explanation about how GOAWAY works with MAX_STREAMS (see #2629). So my suggestion pre-empts that without much thought put into the final fix. Therefore it might be wrong. Some weak arguments: 

Before sending a GOAWAY, there is no explicit limit set by the server. The stream ID limit is implicit and real but more importantly, you don't invoke the requirement `a connection SHOULD send an initial GOAWAY frame with the last Stream ID set to the current value of QUIC’s MAX_STREAM_ID and SHOULD NOT increase the MAX_STREAM_ID thereafter.`. Imagination required on how we fix this language up for MAX_STREAMS.

>From Alan's problem statement description, I imagined GOAWAY to operate more like a drain state trigger. So having separate frames would allow the draining of unidirectional traffic and bidirectional traffic independently. E.g. a client declares "I'm intending to leave soon, and I don't want to accept new things from you (pushes) but I want the ability to create new requests, so please continue to give me bidi stream credits".

On a more practical level, a frame that includes both fields when only one type wants to be drained will likely end up with a set to max_int. So we'd waste bytes to express "don't do anything". Modelling this by making fields optional adds complexity for little gain. 

Finally, we a separate GOAWAY for unidirectional streams opens the door for it to be an extension frame. However, I think this would be the wrong thing to do as it would make managing extensions very difficult.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/2632#issuecomment-484520163