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

[Mike Bishop <[e-mail]>]

#3129 makes a subtle but major change to the semantics, and I think we need to carefully consider whether that's something we want.  We were conflating these two models during interim discussion, and I don't think we're clear on our direction yet.

Instead of "Clients MUST NOT send new requests on the connection after receiving GOAWAY", we now have "Endpoints MUST NOT initiate new requests or pushes on the connection **with an identifier greater than or equal to the smallest identifier received in a GOAWAY frame.**"  In other words, GOAWAY(`max_varint`) has changed from meaning "I'll service your in-flight requests, but don't make any new ones" to a no-op, since the client couldn't have initiated such a request anyway.

I understand that this might be a desirable pattern in the server-to-client direction, where certain use patterns consider the set of pushes part of the fulfillment of the client's request, and the client actively wants the server to continue initiating pushes until the request is complete.  This doesn't seem a wise semantic to bring to the client-to-server direction.

In HTTP/2, the recipient of the GOAWAY was prohibited from initiating any more activity (requests or pushes) regardless of the value in the frame.  My first preference is to preserve that, and actually have the same semantics as HTTP/2.

-- 
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-545581410