Re: [quicwg/base-drafts] Limit RCID state (#3547)

[martinduke <[e-mail]>]

OK -- there has been an enormous amount of discussion. The current state of the PR is that all RCIDs will eventually be sent, but if there are more than active_connection_id_limit, some may be delayed by an RTT or more. If it gets out of hand the RCID sender can throw a connection error.

I believe this reflects what @martinthomson articulated a few dozen messages ago, and in the thread above I have pseudocode for a lightweight implementation of this (i.e. essentially no additional resources for an arbitrarily large backlog of RCIDs).

The advantage of this approach is that it is lightweight and is very subtle change to behavior that will not will not affect interoperability at all or break the existing contract of a Retire Prior To.

There are two alternatives that have some traction:
1) Just bite the bullet and do #3553. This breaks interoperability but is a clean architectural change that addresses the root of the problem. I would be very comfortable with this, and prefer it to other things that break the existing semantics.
2) @kazuho's "leaky" approach that allows endpoints to simply not send some RCIDs at all if there are too many pending. I used to prefer this, but given my existence proof above I'm not as excited about it anymore. It breaks the contract that is in draft-27 instead of merely bending it.

I hope that covers the decision space. I would like to put this issue out of its misery.

-- 
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/pull/3547#issuecomment-607547785