Re: [quicwg/base-drafts] active_connection_id_limit interacts poorly with Retire Prior To (#3193)

[Eric Kinnear <[e-mail]>]

Proxying from the PR in regards to enforcing MUST error if the limit is exceeded: 
@MikeBishop:
> "MUST error" requires tracking. It's possible that an implementation might decide to drop some CIDs and reduce the number it keeps without closing the connection. "SHOULD/MAY error" provides the same deterrent effect for the issuer's misbehavior without placing a requirement on the receiver.

@marten-seemann: 
> We're requiring other limit checks (e.g. for STREAM_LIMIT_ERROR and for FLOW_CONTROL_ERROR), so I'd prefer to keep this a requirement as well.
>
>I don't think dropping should be regarded as acceptable behavior (although you can't prevent implementations from doing it), since a CID issued by an endpoint consumes state on that endpoint as well.

@janaiyengar: 
> I'm in agreement with @MikeBishop -- I would suggest changing this to a SHOULD error. This is different than the flow control errors, since an endpoint does not have to store all connection IDs that have been received.

@kazuho: 
> I might suggest having a design discussion in the issue (i.e. #3193) rather than here. There has been certain amount of discussion in that issue, and it seems to me that we have preference for MUST NOT.

@erickinnear:
> I'm not sure that I'm against the SHOULD, but I think the idea behind enforcing this was that an endpoint would indeed now need to store every CID it receives, and if it's not willing to store X amount, then it shouldn't have advertised a limit of X.
>
> And even if you still didn't store them all, you technically don't need to store them to keep track of how many you've been issued, which you'll probably need in order to cope with getting Retire Prior To anyways.

-- 
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/3193#issuecomment-554934384