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

[martinduke <[e-mail]>]

@martinduke commented on this pull request.



> @@ -1069,6 +1069,15 @@ to cease using the connection IDs when requested can result in connection
 failures, as the issuing endpoint might be unable to continue using the
 connection IDs with the active connection.
 
+An endpoint SHOULD limit the number of outstanding RETIRE_CONNECTION_ID frames
+to bound the necessary state. In order to allow a peer to retire all previously
+issued connection IDs, the limit on the number of outstanding
+RETIRE_CONNECTION_IDs SHOULD be at least the active_connection_id_limit. An

@kazuho. I initially read "outstanding" as meaning "intended to retire", just like you, but further up in the thread @martinthomson said that's not what he meant. And anyway, I soon saw that that interpretation is equally easy to implement and more compliant with the existing spec.

First, assume the endpoint never retires a CID voluntarily. I just have to store the active CIDs. The retired CIDs require only two state variables:
- highest retire prior to (highest_rpt)
- lowest RCID seqnum unacked (lowest_rua)

Each packet records the highest seqnum retired because it always retires everything below that, as you'll see below.

`process_ncid():
    if (retire_prior_to <= highest_rpt)
        quit;
    if (retire_prior_to > lowest_rua + 100)
       abort;
    if (highest_rpt == lowest_rua)
        highest_rpt = retire_prior_to
        send_rcid_pkt()
    else
        highest_rpt = retire_prior_to

send_rcid_pkt():
    limit = MIN(highest_rpt, lowest_rua + 2 * active_cid_limit)
    for i = lowest_rua : limit
        send_rcid(i)
    pkt.max_seq_num = limit

process_ack(pkt):
    lowest_rua = MAX(lowest_rua, pkt.max_seq_num + 1) 
    if (lowest_rua < highest_rpt) 
        send_rcid_pkt()`

If I have a backlog of CIDs to retire, I will clear 2 * active_cid_limit of them each RTT assuming no losses. This number is essentially arbitrary but should definitely fit in a single packet. 100 is an arbitrarily large number just in case the client is being malicious. Again, you don't want it to be so small that unlucky packet losses cause a connection error, but beyond that any number is fine.

If we are retiring CIDs voluntarily, we'll have to keep some extra state. We can put them in a special data structure that send_rcid_pkt also checks, but will definitely be finite as it is controlled locally. If I retire them in sequence, nothing stops me from increasing highest_rpt even though I never received it in an NCID frame, and therefore using the mechanism above.

***

The nice thing about this design is that we are fully compliant with the draft-27 language in every way, except that the RCID frames are not always sent *immediately*. #3553 is a cleaner design but this one has essentially no impact on the peer at all if it's implementing -27.

-- 
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#discussion_r401340815