Re: [quicwg/base-drafts] unrecoverable loss pattern leads to deadlock (#2863)

[Jana Iyengar <[e-mail]>]

What @kazuho points out is a problem. Keeping HS keys forever prevents the deadlock, but it does so by admitting that HS packets can be sent forever. This means after migration as well. That is something we don't consider in the spec. 

@ianswett: I don't think you're right about this:
> I'm arguing the client should never have unacknowledged Handshake data when it migrates, because it waits for handshake confirmation to migrate and at that point, all Handshake data should be discarded.
Handshake data is discarded when the handshake keys are dropped, not when confirmation is reached (see [section 5.3 in recovery draft](https://quicwg.org/base-drafts/draft-ietf-quic-recovery.html#handshakes-and-new-paths). Now with handshake keys forever, the client can keep retransmitting forever. And as @kazuho notes, the server can surely send keep retransmitting as well since if it hasn't reached confirmation yet.

There are a few ways to curb this... but I'm strongly inclined towards an explicit signal from both sides indicating handshake confirmation, after which migration can be initiated.

-- 
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/2863#issuecomment-544852798