Should Neqo support the HTTP DATAGRAM Capsule?

[mxinden]

3.5. The DATAGRAM Capsule

This document defines the DATAGRAM (0x00) Capsule Type. This Capsule allows HTTP Datagrams to be sent on a stream using the Capsule Protocol. This is particularly useful when HTTP is running over a transport that does not support the QUIC DATAGRAM frame.

https://www.rfc-editor.org/rfc/rfc9297.html#section-3.5

One use-case would be proxying an HTTP/3 connection over a MASQUE connect-udp HTTP/3 proxy connection (#2796), where the proxy connection does not support the QUIC datagram extension.

Asked differently, do we expect wide deployments of MASQUE connect-udp proxy servers without the QUIC datagram extension?

Note that this is not relevant for WebTransport datagrams, as WebTransport requires WebTransport datagrams to be sent via QUIC datagrams:

WebTransport over HTTP/3 also requires support for QUIC datagrams. To indicate support, both the client and the server MUST send a max_datagram_frame_size transport parameter with a value greater than 0 (see Section 3 of [QUIC-DATAGRAM]).

Any WebTransport requests sent by the client without enabling QUIC and HTTP datagrams MUST be treated as malformed by the server, as described in Section 4.1.2 of [HTTP3].

https://www.ietf.org/archive/id/draft-ietf-webtrans-http3-13.html#section-3.1

[larseggert]

Asked differently, do we expect wide deployments of MASQUE connect-udp proxy servers without the QUIC datagram extension?

Do we know what Fastly supports? I'd prioritize it accordingly.

[martinthomson]

Given where things are going with respect to discussion of intermediating MASQUE and the capsule protocol stuff, I would say the answer is yes. The concern here is that some intermediaries might relay the HTTP/2 version without rewriting capsules.

It should be relatively easy to parse and pass along in much the same way as a frame. The only trick I can see is in managing the weirdness you get as a result (like datagrams that span multiple packets).