Re: [hybi] #4: handshake does not work properly with HTTP reverse proxy.

On Tue, Jul 20, 2010 at 2:06 AM, Willy Tarreau <w@1wt.eu> wrote:

> On Tue, Jul 20, 2010 at 04:40:33PM +1000, Greg Wilkins wrote:
> > On 20 July 2010 14:43, Willy Tarreau <w@1wt.eu> wrote:
> > > Also, I would really insist that we add a "Connection: close" in the
> > > initial HTTP handshake, whatever it looks like in the end, so that we
> > > protect the servers against any form of request injection after the
> > > first one.
> >
> > I think that if we fail to come up with a HTTP compliant handshake, then
> a
> > Connection:close would be a good get out of goal free card.
> >
> > However, it does prevent one possible good use case of using the HTTP
> > connection as a HTTP connection in the case that websocket is rejected.
> > Until websocket is widely deployed, many sites will want to fall back to
> a
> > HTTP transport if websockets is rejected.  If the connection can be kept
> > open, then it may be used by the browser for subsequent HTTP requests,
> > saving a round trip.  If a failed handshake response can contain other
> useful
> > content, then you may be able to save 2 round trips in the establishment
> of
> > HTTP transports.
> >
> > This is already an issue with cometd-2, now that we support websocket as
> > an optional transport.  When it is enabled, there is noticeable
> additional
> > establishing a comet web page (eg logging into a chat room), while
> websocket
> > hand shake is tried and fails, before the next handshake is tried.   This
> is a
> > discouragement to enabling websocket support in the framework, which is
> > then does not encourage infrastructure providers to support websocket.
>
> Agreed. Also, keeping the connection open is required with some
> authentication
> schemes such as NTLM which authenticate the connection instead of the
> request.
> While this will not be a problem on the Internet, it could make it
> difficult
> to integrate websocket into enterprise applications. This is another reason
> not
> to send any unadvertised data in the handshake by the way !
>

+1.


>
> Regards,
> Willy
>
> _______________________________________________
> hybi mailing list
> hybi@ietf.org
> https://www.ietf.org/mailman/listinfo/hybi
>