Hi Willy, Willy Tarreau wrote on 16.10.2017:
> Hi Pavlos! > On Mon, Oct 16, 2017 at 10:46:44AM +0200, Pavlos Parissis wrote: >> > Now after several failed attempts and with a lot of design sessions >> > with my coworkers, I've made a good progress on a totally different >> > approach which will later allow us to implement HTTP/2 on both sides, >> > as well as implement support for QUIC. I have not merged anything yet >> > because as I'm picking code from the first implementation, >> >> HTTP/2 on client and servers side, and at the same time QUIC support!! >> 1.8 version will be very hot release, thanks a lot for your hard work. > Wait wait wait... I never said that! I'm speaking about *design*. ;-) you know small finger => whole hand ;-) > Look, > "will later allow us". For 1.8, client-side only is expected. I don't know > how long it will take to support server-side, as there are some painful > shortcomings related to maintaining the server connections alive and being > able to share them (and my beloved coworkers working on threads will hate > me when they'll see that they'll have to deal with connections floating > between multiple threads due to the full mesh you can have between client > streams and servers streams). Also before supporting server-side H2, we'll > have to turn the internal HTTP into "HTTP" and not "HTTP/1" anymore. This > means we won't have native HTTP/1 anymore inthe request buffer but more > something indexed, implying a rewrite of most of the HTTP processing. For > QUIC, *some* work was started already to try to spot what will be needed. > That's also what helped a lot abandon the dead end design I initially > started with. I have also started to create a image based on the h2 branch https://gitlab.com/aleks001/haproxy18-centos It's not finished yet but when it's finished you can use the docker image then from docker hub. https://hub.docker.com/u/me2digital/ => haproxy18 > cheers, > Willy -- Best Regards Aleks
