Re: [quicwg/base-drafts] Required state for retaining unacked RETIRE_CONNECTION_ID frames is unbound (#3509)

[martinduke <[e-mail]>]

What a mess we've made! Is there interest in doing a Zoom about this? Maybe 0900 Tokyo/1100 Sydney/1700 Seattle/2000 Boston on Tue/Wed?

The root cause is that the flow control mechanism isn't all that well designed. IMO #3553 is the actual correct fix to this problem, but as people say, it's disruptive.

#3548 uses ACKs as an implicit signal, which seems like a problem. #3553 has a different implicit signal, and I don't think it solves the case where the client is withholding acks for unsolicited RCIDs.

I really think we should just use #3547 or #3553 depending on our appetite for changing the wire image. #3547 is great for the RCID sender, and if the receiver won't miss any RCIDs unless it's doing pathological stuff or the ack loss patterns are tremendously unlucky.

#3553, OTOH, is a comprehensive fix to the problem.




-- 
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/3509#issuecomment-606120826