My personal opinion is that once we open the floodgates we can't close them. More specifically, if someone ships an implementation to production in a way that they cannot guarantee that it can be updated, then the version used is now immutable. In practice, that means that if you ship a client device that uses software updates, because some users do not update, then that would lock us in. In that regards, I could see a world where:
1) Deploying on your server today is OK as long as you can guarantee that if the protocol changes you'll be able to disable h3 or deploy a fix in under 14 days 2) Deploying a client-side experiment today is OK as long as that experiment will automatically stop after 14 days if the client device stops hearing confirmation that the protocol is still OK to use 3) Once -invariants, -transport, -tls, and -recovery have all been published as RFCs, then deploying everywhere is OK 4) Lack of Alt-Svc doesn't change points 1-3 Thoughts? David On Thu, May 6, 2021 at 10:29 AM Lucas Pardue <[email protected]> wrote: > Hi David, > > On Thu, May 6, 2021 at 5:38 PM David Schinazi <[email protected]> > wrote: > >> >> >> On Thu, May 6, 2021 at 7:54 AM Lucas Pardue <[email protected]> >> wrote: >> >>> This email commences a formal consensus call for permitting the >>> deployment of QUIC "0x00000001" with HTTP/3 ALPN identifier "h3" *after* >>> -invariants, -transport, -tls, and -recovery have been published as RFCs >>> but *before *-http and -qpack are published as RFC. The call will end >>> on May 13. >>> >> >> Hi Lucas, since this is a formal consensus call, can you formally define >> what you meant by "deployment"? Some clarifying questions: >> - Is running a 1% experiment on clients before then OK? >> - Is running a 1% experiment on servers before then OK? >> - Is running 0x00000001+h3 on your main production server today OK if you >> don't use Alt-Svc? >> What we're describing is effectively a flag day, and it would be good to >> define its parameters clearly. >> > > These are excellent questions. So far the WG discussion has focused on the > 0x00000001+h3 question. There is ambiguity in the transport-34 and http-34 > definition of deployment and to date there wasn't much discussion of that. > The chairs would appreciate the WG to respond specifically to this point. > > Cheers > Lucas > > >
