Configuring and Using the Twisted.Web Server

  1. Twisted Web Development
  2. Advanced Configuration
  3. Running a Twisted Web Server
  4. Rewriting URLs
  5. Knowing When We're Not Wanted
  6. As-Is Serving

Twisted Web Development

Twisted Web serves Python objects that implement the interface IResource.

Twisted Web process

Main Concepts

The Twisted.Web server is started through the Twisted Daemonizer, as in:

% twistd web

Site Objects

Site objects serve as the glue between a port to listen for HTTP requests on, and a root Resource object.

When using twistd -n web --path /foo/bar/baz, a Site object is created with a root Resource that serves files out of the given path.

You can also create a Site instance by hand, passing it a Resource object which will serve as the root of the site:

1 2 3 4 5 6 7 8 9 10 11

from twisted.web import server, resource from twisted.internet import reactor class Simple(resource.Resource): isLeaf = True def render_GET(self, request): return "<html>Hello, world!</html>" site = server.Site(Simple()) reactor.listenTCP(8080, site)

Resource objects

Resource objects represent a single URL segment of a site. During URL parsing, getChild is called on the current Resource to produce the next Resource object.

When the leaf Resource is reached, either because there were no more URL segments or a Resource had isLeaf set to True, the leaf Resource is rendered by calling render(request). See Resource Rendering below for more about this.

During the Resource location process, the URL segments which have already been processed and those which have not yet been processed are available in request.prepath and request.postpath.

A Resource can know where it is in the URL tree by looking at request.prepath, a list of URL segment strings.

A Resource can know which path segments will be processed after it by looking at request.postpath.

If the URL ends in a slash, for example, the final URL segment will be an empty string. Resources can thus know if they were requested with or without a final slash.

Here is a simple Resource object:

1 2 3 4 5 6 7 8 9 10 11 12 13

from twisted.web.resource import Resource class Hello(Resource): isLeaf = True def getChild(self, name, request): if name == '': return self return Resource.getChild(self, name, request) def render_GET(self, request): return "Hello, world! I am located at %r." % (request.prepath,) resource = Hello()

Resource Trees

Resources can be arranged in trees using putChild. putChild puts a Resource instance into another Resource instance, making it available at the given path segment name:

1 2 3

root = Hello() root.putChild('fred', Hello()) root.putChild('bob', Hello())

If this root resource is served as the root of a Site instance, the following URLs will all be valid:


.rpy scripts

Files with the extension .rpy are python scripts which, when placed in a directory served by Twisted Web, will be executed when visited through the web.

An .rpy script must define a variable, resource, which is the Resource object that will render the request.

.rpy files are very convenient for rapid development and prototyping. Since they are executed on every web request, defining a Resource subclass in an .rpy will make viewing the results of changes to your class visible simply by refreshing the page:

1 2 3 4 5 6 7

from twisted.web.resource import Resource class MyResource(Resource): def render_GET(self, request): return "<html>Hello, world!</html>" resource = MyResource()

However, it is often a better idea to define Resource subclasses in Python modules. In order for changes in modules to be visible, you must either restart the Python process, or reload the module:

1 2 3 4 5 6

import myresource ## Comment out this line when finished debugging reload(myresource) resource = myresource.MyResource()

Creating a Twisted Web server which serves a directory is easy:

% twistd -n web --path /Users/dsp/Sites

Resource rendering

Resource rendering occurs when Twisted Web locates a leaf Resource object to handle a web request. A Resource's render method may do various things to produce output which will be sent back to the browser:

  • Return a string
  • Call request.write("stuff") as many times as desired, then call request.finish() and return server.NOT_DONE_YET (This is deceptive, since you are in fact done with the request, but is the correct way to do this)
  • Request a Deferred, return server.NOT_DONE_YET, and call request.write("stuff") and request.finish() later, in a callback on the Deferred.

The Resource class, which is usually what one's Resource classes subclass, has a convenient default implementation of render. It will call a method named self.render_METHOD where METHOD is whatever HTTP method was used to request this resource. Examples: request_GET, request_POST, request_HEAD, and so on. It is recommended that you have your resource classes subclass Resource and implement render_METHOD methods as opposed to render itself. Note that for certain resources, request_POST = request_GET may be desirable in case one wants to process arguments passed to the resource regardless of whether they used GET (?foo=bar&baz=quux, and so forth) or POST.


HTTP is a stateless protocol; every request-response is treated as an individual unit, distinguishable from any other request only by the URL requested. With the advent of Cookies in the mid nineties, dynamic web servers gained the ability to distinguish between requests coming from different browser sessions by sending a Cookie to a browser. The browser then sends this cookie whenever it makes a request to a web server, allowing the server to track which requests come from which browser session.

Twisted Web provides an abstraction of this browser-tracking behavior called the Session object. Calling request.getSession() checks to see if a session cookie has been set; if not, it creates a unique session id, creates a Session object, stores it in the Site, and returns it. If a session object already exists, the same session object is returned. In this way, you can store data specific to the session in the session object.

Advanced Configuration

Non-trivial configurations of Twisted Web are achieved with Python configuration files. This is a Python snippet which builds up a variable called application. Usually, a twisted.application.internet.TCPServer instance will be used to make the application listen on a TCP port (80, in case direct web serving is desired), with the listener being a twisted.web.server.Site. The resulting file can then be run with twistd -y. Alternatively a reactor object can be used directly to make a runnable script.

The Site will wrap a Resource object -- the root.

1 2 3 4 5 6 7 8 9

from twisted.application import internet, service from twisted.web import static, server root = static.File("/var/www/htdocs") application = service.Application('web') site = server.Site(root) sc = service.IServiceCollection(application) i = internet.TCPServer(80, site) i.setServiceParent(sc)

Most advanced configurations will be in the form of tweaking the root resource object.

Adding Children

Usually, the root's children will be based on the filesystem's contents. It is possible to override the filesystem by explicit putChild methods.

Here are two examples. The first one adds a /doc child to serve the documentation of the installed packages, while the second one adds a cgi-bin directory for CGI scripts.

1 2 3 4 5 6 7

from twisted.internet import reactor from twisted.web import static, server root = static.File("/var/www/htdocs") root.putChild("doc", static.File("/usr/share/doc")) reactor.listenTCP(80, server.Site(root))

1 2 3 4 5 6 7

from twisted.internet import reactor from twisted.web import static, server, twcgi root = static.File("/var/www/htdocs") root.putChild("cgi-bin", twcgi.CGIDirectory("/var/www/cgi-bin")) reactor.listenTCP(80, server.Site(root))

Modifying File Resources

File resources, be they root object or children thereof, have two important attributes that often need to be modified: indexNames and processors. indexNames determines which files are treated as index files -- served up when a directory is rendered. processors determine how certain file extensions are treated.

Here is an example for both, creating a site where all .rpy extensions are Resource Scripts, and which renders directories by searching for a index.rpy file.

1 2 3 4 5 6 7 8 9 10 11

from twisted.application import internet, service from twisted.web import static, server, script root = static.File("/var/www/htdocs") root.indexNames=['index.rpy'] root.processors = {'.rpy': script.ResourceScript} application = service.Application('web') sc = service.IServiceCollection(application) site = server.Site(root) i = internet.TCPServer(80, site) i.setServiceParent(sc)

File objects also have a method called ignoreExt. This method can be used to give extension-less URLs to users, so that implementation is hidden. Here is an example:

1 2 3 4 5 6 7 8 9 10 11

from twisted.application import internet, service from twisted.web import static, server, script root = static.File("/var/www/htdocs") root.ignoreExt(".rpy") root.processors = {'.rpy': script.ResourceScript} application = service.Application('web') sc = service.IServiceCollection(application) site = server.Site(root) i = internet.TCPServer(80, site) i.setServiceParent(sc)

Now, a URL such as /foo might be served from a Resource Script called foo.rpy, if no file by the name of foo exists.

Virtual Hosts

Virtual hosting is done via a special resource, that should be used as the root resource -- NameVirtualHost. NameVirtualHost has an attribute named default, which holds the default website. If a different root for some other name is desired, the addHost method should be called.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

from twisted.application import internet, service from twisted.web import static, server, vhost, script root = vhost.NameVirtualHost() # Add a default -- htdocs root.default=static.File("/var/www/htdocs") # Add a simple virtual host -- root.addHost("", static.File("/var/www/foo")) # Add a simple virtual host -- root.addHost("", static.File("/var/www/bar")) # The "baz" people want to use Resource Scripts in their web site baz = static.File("/var/www/baz") baz.processors = {'.rpy': script.ResourceScript} baz.ignoreExt('.rpy') root.addHost('baz', baz) application = service.Application('web') sc = service.IServiceCollection(application) site = server.Site(root) i = internet.TCPServer(80, site) i.setServiceParent(sc)

Advanced Techniques

Since the configuration is a Python snippet, it is possible to use the full power of Python. Here are some simple examples:

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

# No need for configuration of virtual hosts -- just make sure # a directory /var/vhosts/<vhost name> exists: from twisted.web import vhost, static, server from twisted.application import internet, service root = vhost.NameVirtualHost() root.default = static.File("/var/www/htdocs") for dir in os.listdir("/var/vhosts"): root.addHost(dir, static.File(os.path.join("/var/vhosts", dir))) application = service.Application('web') sc = service.IServiceCollection(application) site = server.Site(root) i = internet.TCPServer(80, site) i.setServiceParent(sc)

1 2 3 4 5 6 7 8 9 10 11 12

# Determine ports we listen on based on a file with numbers: from twisted.web import vhost, static, server from twisted.application import internet, service root = static.File("/var/www/htdocs") site = server.Site(root) application = service.Application('web') serviceCollection = service.IServiceCollection(application) for num in map(int, open("/etc/web/ports").read().split()): serviceCollection.addCollection(internet.TCPServer(num, site))

Running a Twisted Web Server

In many cases, you'll end up repeating common usage patterns of twisted.web. In those cases you'll probably want to use Twisted's pre-configured web server setup.

The easiest way to run a Twisted Web server is with the Twisted Daemonizer. For example, this command will run a web server which serves static files from a particular directory:

% twistd web --path /path/to/web/content

If you just want to serve content from your own home directory, the following will do:

% twistd web --path ~/public_html/

You can stop the server at any time by going back to the directory you started it in and running the command:

% kill `cat`

Some other configuration options are available as well:

  • --port: Specify the port for the web server to listen on. This defaults to 8080.
  • --logfile: Specify the path to the log file.

The full set of options that are available can be seen with:

% twistd web --help

Serving Flat HTML

Twisted.Web serves flat HTML files just as it does any other flat file.

Resource Scripts

A Resource script is a Python file ending with the extension .rpy, which is required to create an instance of a (subclass of a) twisted.web.resource.Resource.

Resource scripts have 3 special variables:

A very simple Resource Script might look like:

1 2 3 4 5 6

from twisted.web import resource class MyGreatResource(resource.Resource): def render_GET(self, request): return "<html>foo</html>" resource = MyGreatResource()

A slightly more complicated resource script, which accesses some persistent data, might look like:

1 2 3 4 5 6 7 8 9 10 11 12 13 14

from twisted.web import resource from SillyWeb import Counter counter = registry.getComponent(Counter) if not counter: registry.setComponent(Counter, Counter()) counter = registry.getComponent(Counter) class MyResource(resource.Resource): def render_GET(self, request): counter.increment() return "you are visitor %d" % counter.getValue() resource = MyResource()

This is assuming you have the SillyWeb.Counter module, implemented something like the following:

1 2 3 4 5 6 7 8 9 10

class Counter: def __init__(self): self.value = 0 def increment(self): self.value += 1 def getValue(self): return self.value

Web UIs

The Nevow framework, available as part of the Quotient project, is an advanced system for giving Web UIs to your application. Nevow uses Twisted Web but is not itself part of Twisted.

Spreadable Web Servers

One of the most interesting applications of Twisted.Web is the distributed webserver; multiple servers can all answer requests on the same port, using the twisted.spread package for spreadable computing. In two different directories, run the commands:

% twistd web --user
% twistd web --personal [other options, if you desire]

Once you're running both of these instances, go to http://localhost:8080/your_username.twistd/ -- you will see the front page from the server you created with the --personal option. What's happening here is that the request you've sent is being relayed from the central (User) server to your own (Personal) server, over a PB connection. This technique can be highly useful for small community sites; using the code that makes this demo work, you can connect one HTTP port to multiple resources running with different permissions on the same machine, on different local machines, or even over the internet to a remote site.

By default, a personal server listens on a UNIX socket in the owner's home directory. The --port option can be used to make it listen on a different address, such as a TCP or SSL server or on a UNIX server in a different location. If you use this option to make a personal server listen on a different address, the central (User) server won't be able to find it, but a custom server which uses the same APIs as the central server might. Another use of the --port option is to make the UNIX server robust against system crashes. If the server crashes and the UNIX socket is left on the filesystem, the personal server will not be able to restart until it is removed. However, if --port unix:/home/username/.twistd-web-pb:wantPID=1 is supplied when creating the personal server, then a lockfile will be used to keep track of whether the server socket is in use and automatically delete it when it is not.

Serving PHP/Perl/CGI

Everything related to CGI is located in the twisted.web.twcgi, and it's here you'll find the classes that you need to subclass in order to support the language of your (or somebody elses) taste. You'll also need to create your own kind of resource if you are using a non-unix operating system (such as Windows), or if the default resources has wrong pathnames to the parsers.

The following snippet is a .rpy that serves perl-files. Look at twisted.web.twcgi for more examples regarding twisted.web and CGI.

1 2 3 4 5 6 7 8 9

from twisted.web import static, twcgi class PerlScript(twcgi.FilteredScript): filter = '/usr/bin/perl' # Points to the perl parser resource = static.File("/perlsite") # Points to the perl website resource.processors = {".pl": PerlScript} # Files that end with .pl will be # processed by PerlScript resource.indexNames = ['']

Serving WSGI Applications

WSGI is the Web Server Gateway Interface. It is a specification for web servers and application servers to communicate with Python web applications. All modern Python web frameworks support the WSGI interface.

The easiest way to get started with WSGI application is to use the twistd command:

% twistd -n web --wsgi=helloworld.application

This assumes that you have a WSGI application called application in your helloworld module/package, which might look like this:

1 2 3 4

def application(environ, start_response): """Basic WSGI Application""" start_response('200 OK', [('Content-type','text/plain')]) return ['Hello World!']

The above setup will be suitable for many applications where all that is needed is to server the WSGI application at the site's root. However, for greater control, Twisted provides support for using WSGI applications as resources twisted.web.wsgi.WSGIResource.

Here is an example of a WSGI application being served as the root resource for a site, in the following tac file:

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

from twisted.web import server from twisted.web.wsgi import WSGIResource from twisted.python.threadpool import ThreadPool from twisted.internet import reactor from twisted.application import service, strports # Create and start a thread pool, wsgiThreadPool = ThreadPool() wsgiThreadPool.start() # ensuring that it will be stopped when the reactor shuts down reactor.addSystemEventTrigger('after', 'shutdown', wsgiThreadPool.stop) def application(environ, start_response): """A basic WSGI application""" start_response('200 OK', [('Content-type','text/plain')]) return ['Hello World!'] # Create the WSGI resource wsgiAppAsResource = WSGIResource(reactor, wsgiThreadPool, application) # Hooks for twistd application = service.Application('Twisted.web.wsgi Hello World Example') server = strports.service('tcp:8080', server.Site(wsgiAppAsResource)) server.setServiceParent(application)

This can then be run like any other .tac file:

% twistd -ny myapp.tac

Because of the synchronous nature of WSGI, each application call (for each request) is called within a thread, and the result is written back to the web server. For this, a twisted.python.threadpool.ThreadPool instance is used.

Using VHostMonster

It is common to use one server (for example, Apache) on a site with multiple names which then uses reverse proxy (in Apache, via mod_proxy) to different internal web servers, possibly on different machines. However, naive configuration causes miscommunication: the internal server firmly believes it is running on internal-name:port, and will generate URLs to that effect, which will be completely wrong when received by the client.

While Apache has the ProxyPassReverse directive, it is really a hack and is nowhere near comprehensive enough. Instead, the recommended practice in case the internal web server is Twisted.Web is to use VHostMonster.

From the Twisted side, using VHostMonster is easy: just drop a file named (for example) vhost.rpy containing the following:

1 2

from twisted.web import vhost resource = vhost.VHostMonsterResource()

Of course, an equivalent .trp can also be used. Make sure the web server is configured with the correct processors for the rpy or trp extensions (the web server twistd web --path generates by default is so configured).

From the Apache side, instead of using the following ProxyPass directive:

<VirtualHost ip-addr>
ProxyPass / http://localhost:8538/

Use the following directive:

<VirtualHost ip-addr>
ProxyPass / http://localhost:8538/vhost.rpy/http/

Here is an example for Twisted.Web's reverse proxy:

1 2 3 4 5 6 7 8 9 10 11 12

from twisted.application import internet, service from twisted.web import proxy, server, vhost vhostName = '' reverseProxy = proxy.ReverseProxyResource('internal', 8538, '/vhost.rpy/http/'+vhostName+'/') root = vhost.NameVirtualHost() root.addHost(vhostName, reverseProxy) site = server.Site(root) application = service.Application('web-proxy') sc = service.IServiceCollection(application) i = internet.TCPServer(80, site) i.setServiceParent(sc)

Rewriting URLs

Sometimes it is convenient to modify the content of the Request object before passing it on. Because this is most often used to rewrite either the URL, the similarity to Apache's mod_rewrite has inspired the twisted.web.rewrite module. Using this module is done via wrapping a resource with a twisted.web.rewrite.RewriterResource which then has rewrite rules. Rewrite rules are functions which accept a request object, and possible modify it. After all rewrite rules run, the child resolution chain continues as if the wrapped resource, rather than the RewriterResource, was the child.

Here is an example, using the only rule currently supplied by Twisted itself:


default_root = rewrite.RewriterResource(default, rewrite.tildeToUsers)

This causes the URL /~foo/bar.html to be treated like /users/foo/bar.html. If done after setting default's users child to a distrib.UserDirectory, it gives a configuration similar to the classical configuration of web server, common since the first NCSA servers.

Knowing When We're Not Wanted

Sometimes it is useful to know when the other side has broken the connection. Here is an example which does that:

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

from twisted.web.resource import Resource from twisted.web import server from twisted.internet import reactor from twisted.python.util import println class ExampleResource(Resource): def render_GET(self, request): request.write("hello world") d = request.notifyFinish() d.addCallback(lambda _: println("finished normally")) d.addErrback(println, "error") reactor.callLater(10, request.finish) return server.NOT_DONE_YET resource = ExampleResource()

This will allow us to run statistics on the log-file to see how many users are frustrated after merely 10 seconds.

As-Is Serving

Sometimes, you want to be able to send headers and status directly. While you can do this with a ResourceScript, an easier way is to use ASISProcessor. Use it by, for example, adding it as a processor for the .asis extension. Here is a sample file:

HTTP/1.0 200 OK
Content-Type: text/html

Hello world


Version: 10.1.0