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

[Kazuho Oku <[e-mail]>]

@ianswett 
> The other option I can imagine is to make the RETIRE_CONNECTION_ID frame cumulative, which provides a strong incentive to use connection IDs in order. Though we may have to make an exception for the connection ID in the received packet if we did that.

While I won't say that's impossible, I'd argue that the design is going to be complex.

When CIDs are used in a legitimate manner, they are not going to be retired one by one. A CID is retired when a path using that CID is abandoned.

Consider the case where an endpoint has 4 active CIDs (CID0 to CID3), and two paths. One path could be using CID0, the other path could be using CID3. When the endpoint abandon the path that is using CID3, it has to retire CID3 but it cannot retire others.

@MikeBishop The amount of CIDs that can be sent on wire at a time is restricted by the CWND, therefore there is no guarantee that you can send RCID frames for all CIDs that you want to retire at once. In other words, I think the approach at best collapses into what @martinthomson has proposed (i.e. leaky behavior).

-- 
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-604076404