Re: [quicwg/base-drafts] Acknowledging *_CLOSE (#733)

[MikkelFJ <[e-mail]>]

This is a two-phase commit problem. I am also concerned with state lingering until timeout. Explicit confirmation is complex though. Like:

1. (A) generates close token (CT) and sends PREPARE_CLOSE(CT) periodically.
2. (B) receives (CT) and responds with CORFIRM_CLOSE(CT) periodically.
3. (A) receives CONFIRM_CLOSE(CT) and stops sending PREPARE_CLOSE(CT) sends CLOSE(CT) a few times then destroys context including CT. If CONFIRM close is not received within timeout, the context is destroyed.
4. (B) receives CLOSE(CT) or times out and destroys context.

There ought to be general application timeout as well so a lost PREPARE_CLOSE will also close the connection, but this timeout would be long.

I'm not really sure a CT token is needed but it might make it possible to drop most app context early.

The timeout can be short like 4RTT or so, because there is an expected response. There is no ACK involved in this and the timeouts are different because ACK can be delayed by an endpoint but CONFIRM_CLOSE and CLOSE must be ASAP.

In a proper two-phase commit context the state should be serialized to disk in order to handle crashes, but we do this with timeout instead.

-- 
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/733#issuecomment-322697055