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

[Kazuho Oku <[e-mail]>]

@martinthomson 
> I know that you have characterized this as "leaky", but I'm not sure that I understand exactly what is being leaked. Nothing is lost in #3547, except that the endpoint requesting retirement has to tolerate a potentially extended period without old connection IDs being retired.

In the approach we take in #3547, there is a possibility that the issuer of the CID might not receive RCID frames that the peer sends, when the issuer is issuing new CIDs at high rate and the receiver is also consuming them at a hight rate. That is the consequence of endpoints limiting the amount of inflight RCID frames it tracks.

I think we can provide advice so that the "leak" would not be an issue in practice, and assuming that would happen, I am happy with #3547 being the solution.

I think #3553 is cleaner, but I acknowledge that it is more disruptive than #3547. Part of my intent behind creating #3553 is to see how a clean fix (in terms of state machinery) would look like, so that it can be compared against #3547.

> In case this helps, I have no big preference between #3547 and #3550. There is a certain elegance to #3550 that might tip me toward that, but it is slightly more disruptive.

As stated in https://github.com/quicwg/base-drafts/pull/3550#issuecomment-605540981, #3550 by itself only partly fixes the problem. I am not against adopting #3550 along with #3547 though.

#3548 is also a partial fix, I also think that it is a non-starter.

-- 
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-605771367