Re: HTTP/2 and Websockets

On 27 November 2014 07:53:01 GMT+08:00, Greg Wilkins <gregw@intalio.com> wrote:
>On 26 November 2014 at 08:25, Roberto Peon <grmocg@gmail.com> wrote:
>
>> Sadly, since it was something that a few of us in the WG were pushing
>for,
>> as it would have allowed intermediaries without knowledge of
>protocols to
>> do what was right for them.
>> That is why I truly hope that the WS extension happens (and
>quickly)-- I'd
>> love to see loadbalancers relieved of having to introspect into the
>data
>> further than the HTTP2 frame layer in order to figure out how to deal
>with
>> it reasonably.
>>
>
>+1
>
>Non-constructive comments from me are that this is simply way too late
>in
>the process for this to be considered and I suspect that with the
>framing

Hmmmm it's late in the process to consider it because everybody put off working on it!

It's not a new observation this http2 stuff will be carrying ws traffic and clearly Hirano-san has been trying to move it on for a long while.

>layer we have given ourselves, all possible solutions will now be
>pretty
>ugly.
>
>It comes down to a choice between:
>
>a) defining new frame types for WS.   Which is going to require very
>explicit support from intermediaries (imagine if we had to do that in
>TCP
>for every new protocol!).
>
>b) reusing the existing HEADER/DATA frames, but they miss some useful
>semantics and have been so conflated with HTTP semantics that we are
>back
>to websockets pretending to be HTTP - which is exactly the sort of
>protocol
>mis-use that we were chartered to resolve!

No... for http2 it's helpful to think about this as a generic stream upgrade to opaque data in DATA issue, not particularly websockets.

-Andy

>Currently I can't pick between these two ugly ducklings.       We long
>ago
>missed the opportunity of coming up with a truly multiple protocol web
>framing layer that would support diverse semantics such as HTTP,
>websocket
>and whatever the future may bring.    So failing that, our best
>solution is
>to pick whatever hack we have to sooner rather than later so that we
>can at
>least hammer in special case handling for websockets into
>intermediaries
>and hope that no new semantics emerge.
>
>Will try to find the time in the next week to experiment with some of
>the
>options, so we can be a bit more constructive about which we prefer.
>
>regards

Received on Thursday, 27 November 2014 00:03:23 UTC