Re: [quicwg/base-drafts] Which DCID do Handshake retransmissions use? (#3348)

[Kazuho Oku <[e-mail]>]

@martinthomson I think that that is a good way at looking into the issue. I agree that the endpoints can start using the new CIDs during handshake, if the peer sends NCID frames.

The only problem that I can think of is related to retirement (see https://github.com/quicwg/base-drafts/issues/3348#issuecomment-574493824) When the server provides TP.preferred_address, and also decides to retire the CIDs immediately after sending all the handshake transcript, there's going to be a very small time window in which it has to send *two* CIDs (one for the current path and another for the path using the preferred address). Unless the QUIC stack allows bundling of two NCID frames when building a QUIC packet, the client might end up not having sufficient amount of CIDs, in which case it might terminate the connection. 

It is my view that this is very different from the ordinary case of retiring CIDs, in which case there would be enough time window to guarantee that a handful number of new generation CIDs are delivered before the older ones are delivered (cc @nibanks).

That said, I now tend to think that the most we might have to do is the following two:
* clarify that CIDs received using NCID frames can be used for Handshake packets
* have a cautionary text stating that a server needs to make sure that the client has enough number of new generation CIDs when it retires an older generation of CIDs

-- 
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/3348#issuecomment-574962583