Server

ws4py comes with a few server implementations built around the main WebSocket class.

CherryPy

ws4py provides an extension to CherryPy 3 to enable WebSocket from the framework layer. It is based on the CherryPy plugin and tool mechanisms.

The WebSocket tool plays at the request level on every request received by the server. Its goal is to perform the WebSocket handshake and, if it succeeds, to create the WebSocket instance (well a subclass you will be implementing) and push it to the plugin.

The WebSocket plugin works at the CherryPy system level and has a single instance throughout. Its goal is to track websocket instances created by the tool and free their resources when connections are closed.

Here is a simple example of an echo server:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
import cherrypy
from ws4py.server.cherrypyserver import WebSocketPlugin, WebSocketTool
from ws4py.websocket import EchoWebSocket

cherrypy.config.update({'server.socket_port': 9000})
WebSocketPlugin(cherrypy.engine).subscribe()
cherrypy.tools.websocket = WebSocketTool()

class Root(object):
    @cherrypy.expose
    def index(self):
        return 'some HTML with a websocket javascript connection'

    @cherrypy.expose
    def ws(self):
        # you can access the class instance through
        handler = cherrypy.request.ws_handler

cherrypy.quickstart(Root(), '/', config={'/ws': {'tools.websocket.on': True,
                                                 'tools.websocket.handler_cls': EchoWebSocket}})

Note how we specify the class which should be instanciated by the server on each connection. The great aspect of the tool mechanism is that you can specify a different class on a per-path basis.

gevent

gevent is a coroutine, called greenlets, implementation for very concurrent applications. ws4py offers a server implementation for this library on top of the WSGI protocol. Using it is as simple as:

1
2
3
4
5
6
7
from gevent import monkey; monkey.patch_all()
from ws4py.websocket import EchoWebSocket
from ws4py.server.geventserver import WSGIServer
from ws4py.server.wsgiutils import WebSocketWSGIApplication

server = WSGIServer(('localhost', 9000), WebSocketWSGIApplication(handler_cls=EchoWebSocket))
server.serve_forever()

First we patch all the standard modules so that the stdlib runs well with as gevent. Then we simply create a WSGI server and specify the class which will be instanciated internally each time a connection is successful.

wsgiref

wsgiref is a built-in WSGI package that provides various classes and helpers to develop against WSGI. Mostly it provides a basic WSGI server that can be usedfor testing or simple demos. ws4py provides support for websocket on wsgiref for testing purpose as well. It’s not meant to be used in production, since it can only initiate web socket connections one at a time, as a result of being single threaded. However, once accepted, ws4py takes over, which is multithreaded by default.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
from wsgiref.simple_server import make_server
from ws4py.websocket import EchoWebSocket
from ws4py.server.wsgirefserver import WSGIServer, WebSocketWSGIRequestHandler
from ws4py.server.wsgiutils import WebSocketWSGIApplication

server = make_server('', 9000, server_class=WSGIServer,
                     handler_class=WebSocketWSGIRequestHandler,
                     app=WebSocketWSGIApplication(handler_cls=EchoWebSocket))
server.initialize_websockets_manager()
server.serve_forever()

asyncio

asyncio is the implementation of PEP 3156, the new asynchronous framework for concurrent applications.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
from ws4py.async_websocket import EchoWebSocket

loop = asyncio.get_event_loop()

def start_server():
    proto_factory = lambda: WebSocketProtocol(EchoWebSocket)
    return loop.create_server(proto_factory, '', 9007)

s = loop.run_until_complete(start_server())
print('serving on', s.sockets[0].getsockname())
loop.run_forever()

Warning

The provided HTTP server used for the handshake is clearly not production ready. However, once the handshake is performed, the rest of the code runs the same stack as the other server implementations. It should be easy to replace the HTTP interface with any asyncio aware HTTP framework.