txWS is the best twisted-based framework for websocket: it does not assume that you want to combine it with a web server, and allows you to wrap your existing protocols into websocket
AutoBahn is an absolutely top notch implementation of WebSockets for twisted. It's probably one of the most complete and robust implementations on any platform, and their test suite is currently the gold standard against which other implementors test their own implementations.
I wrote the websockets implementation for mongrel2, and just as an FYI, it only supports the more recent websockets standard (they made some incompatible changes around draft version 7). The good news is that the only browser that ever had the old version on by default was chrome (firefox and opera defaulted it off for security reasons), so there aren't too many browsers out in the wild using the old version.
If you're interested in a easy-to-use hosted solution, you could try Pusher ( http://pusher.com ). It would be the fastest way to get up and running and you wouldn't have to worry about the details.
Love caniuse.com but unfortunately there's lot missing from this support matrix.
More important than which browsers support the API is which underlying protocol version the sockets speak. A lot of major browsers support the API but very few speak the same protocol.
I believe today's announcement is a finalized protocol standard which means that there's finally going to be a lingua franca for websockets.
I agree about socket.io. It works quiet well for most of what I need. The only issue I have with is is I can't figure out if it now supports sending and receiving ArrayBuffers? Does anyone know anything about this?
My use case involves sending model data to display in WebGL. Sending it any other way would be inefficient.
socket.io is a great library, agreed. However, a nice thing about standards is you can program to the specification and provide support on other platforms. Working on a project that involves node, a browser client, and a couple of mobile platforms -- socket.io is great for the server->browser websocket or comet case, but for the mobile case investigating the proper library (or hand rolling a compatible WS/comet implementation) is needed.
The step towards standardization will make this type of cross "platform" compatibility easier in the long run.
Many proxies deployed today won't understand the upgrade header and in the case of an explicit proxy config will ignore the CONNECT operation if performed on 80 (rather than 443).
Yep, they undid the crazy 8 bytes of body that wasn't declared by content-length, back in hixie-76. The current protocols play nicely with HTTP aware proxies now.
They introduced plenty of other crazy things.. but at least it works with proxies.
At least with hixie-76, connections would fail to connect if the intermediate proxies are not understanding WebSockets. Now, your connection will succeed, both client and server will believe they can successfully communicate, but communication will not work and die a slow timeout. Not very reliable, is it ?
I look forward to pushing this to the absolute limit. Chat? Can we do something cool with voice? What about websites-as-desktop-software that started when Ajax got cool, but paused when developers ran into the limits of Ajax? How far can that now go? GPS services?
Well kind of. Last summer I did a road trip to newfoundland (canada) with a laptop, a gps and a webcam onboard [1]. The idea was to have people (friends and family) to follow our trip through a website to which I pushed webcam frames and gps coordinate in real time using websocket (socket.io) over my smarthphone. It was fun!
ITYM "WebSockets is now officially IETF Proposed Standard" or something like that.
Publication as an RFC doesn't signify any status
of the protocol standards-wise, almost any protocol
can be documented in an informational, experimental or
historical RFC.
You can already do peer-to-peer UDP communication in a browser with Flash RTMFP protocol. If I remember right it was shipped with Flash 10, which got released 4 years ago.
I don't understand why they didn't just provide an api to normal IP layer sockets, it would have given a lot more flexibility. I'm thinking the sort of lower level networking access you get in node.js
If people really wanted to send data over this weird semi-HTTP websocket protocol then someone could have implemented it as a js library on top of proper socket calls.
I'm guessing the main issue was cross-domain security, but I would have liked to see this solved using some sort of policy file that can be requested from the domain before connecting, like Flash does.
Except, it's implemented on top of TCP. So you still have the head-of-line blocking problem. In a real-time application like games, if some congestion causes a few packets to be dropped, you don't care about them any more (assuming your protocol gives absolute positions of everyone involved); but in TCP, they will be retransmitted, and later packets queued until it can catch up. As websockets are based on TCP, you can't avoid this. Even if you have a "fire and forget" message based layer built on top of TCP, the head-of-line blocking will still kill your performance, causing people with bad connections to get persistent lag rather than a few dropped frames.
That's true, but in practice networked games built on websockets are doing fine right now, and will until browsers get stable capabilities for more intensive 3D gaming (likely in around 2 years time). I guess I meant they won't be needed yet.
Well, I'm working on a web-based real-time multi-player game, and I'd say that UDP would be very welcome by me. When I'm on a good connection, TCP is fine and things work perfectly. But as soon as my connection becomes just a tiny bit unreliable and packets start dropping, it becomes completely unplayable, as TCP has to retransmit those dropped packets before any new packets are processed.
Since all we get is TCP right now though, you can take advantage of that at least and easily do things like send deltas instead of absolute positions, and even things that depend on reliable order of transmission (which you can't rely on if it were UDP), in order to reduce bandwidth use.
WSGI doesn't support it (http://librelist.com/browser//flask/2010/9/1/flask-and-webso...). gevent is an option (http://blog.pythonisito.com/2011/07/gevent-zeromq-websockets...), but would something like Mongrel2 be better (http://mongrel2.org/)?
UPDATE: There's also Juggernaut (http://flask.pocoo.org/snippets/80/), which uses node.js under the hood.