Re: [hybi] Proposal: HTTP upgrade process

[Vladimir Katardjiev <[e-mail]>]

On 17 aug 2010, at 13.11, Willy Tarreau wrote:

> On Tue, Aug 17, 2010 at 12:18:58PM +0200, Vladimir Katardjiev wrote:
>> Hm. It seems I accidentally sent out the wrong version of the mail. I didn't intend to sound as if I was arguing for the eight bytes themselves. What I thought I wrote in addition was that the intention was for the 8 bytes to be functionally part of the WS stream, so figure this:
>> 
>> 1) HTTP Upgrade request
>> 2) 101 response
>> 3) 8 bytes random req
>> 4) WS OK response
> 
> Yes, that's it. And the 8 bytes random could even be a PING message.
> 
>> Since this would be 2 RTTs (1-2 + 3-4), in -76, 1&3 and 2&4 were baked into each-other.
> 
> no, not necessarily 2 RTTs if the sender sends the 1+3 at the same time,
> then the responder can receive 1, send 2, then receive 3 and send 4.
> 
> Please check the various cases of HTTP vs round trips in my earlier e-mail here :
> 
>   http://www.ietf.org/mail-archive/web/hybi/current/msg03238.html
> 
>> I don't necessarily agree with that; I'd prefer separation of concerns, but I was mainly intending to explain why those eight bytes found their way into the text in the first place.
> 
> It was to ensure that intermediaries were able to forward them. But in my
> opinion, it does not bring that many advantages over sending them as headers.

When it was under discussion, it was quite the opposite actually. The random eight bytes were intended to _break_ intermediaries that didn't follow HTTP Upgrade correctly. That's why there's no point to those bytes if they are sent in a header -- they might as well not be sent at all.

It has to be something that comes where HTTP would expect there to be a "body", otherwise it won't adequately test whether or not the established connection can support websocket messages. At least that was the intention. I still feel that it's a worthwhile goal to ensure that the established connection is usable before we throw it over to the application.

Vladimir