Unfortunately it looks like in #1752 I accidentally introduced a bug. Apologies for not catching it.
Bug description: If a client sends a 0-RTT datagram to the server after sending the datagram with its Initial packet, and it is received and processed by the endpoint before the corresponding Incoming is accept-ed, the 0-RTT packet will not find any associated connection and thus be dropped. As a consequence, this seems to cause request/response pairs on connections where this occurs to actually take 3 round trips to go through successfully.
I am working on a fix, and I will see if I can get a PR opened for it tomorrow.
djc
commented
2 months ago
Unfortunately it looks like in #1752 I accidentally introduced a bug. Apologies for not catching it.
Bug description: If a client sends a 0-RTT datagram to the server after sending the datagram with its Initial packet, and it is received and processed by the endpoint before the corresponding
Incomingisaccept-ed, the 0-RTT packet will not find any associated connection and thus be dropped. As a consequence, this seems to cause request/response pairs on connections where this occurs to actually take 3 round trips to go through successfully.I am working on a fix, and I will see if I can get a PR opened for it tomorrow.