Overview

FriendFeed's web server is a relatively simple, non-blocking web server written in Python. The FriendFeed application is written using a web framework that looks a bit like web.py or Google's webapp, but with additional tools and optimizations to take advantage of the non-blocking web server and tools.

Tornado is an open source version of this web server and some of the tools we use most often at FriendFeed. The framework is distinct from most mainstream web server frameworks (and certainly most Python frameworks) because it is non-blocking and reasonably fast. Because it is non-blocking and uses epoll, it can handle 1000s of simultaneous standing connections, which means the framework is ideal for real-time web services. We built the web server specifically to handle FriendFeed's real-time features — every active user of FriendFeed maintains an open connection to the FriendFeed servers. (For more information on scaling servers to support thousands of clients, see The C10K problem.)

Here is the canonical "Hello, world" example app:

import tornado.httpserver
import tornado.ioloop
import tornado.web

class MainHandler(tornado.web.RequestHandler):
    def get(self):
        self.write("Hello, world")

application = tornado.web.Application([
    (r"/", MainHandler),
])

if __name__ == "__main__":
    http_server = tornado.httpserver.HTTPServer(application)
    http_server.listen(8888)
    tornado.ioloop.IOLoop.instance().start()

See Tornado walkthrough below for a detailed walkthrough of the tornado.web package.

We attempted to clean up the code base to reduce interdependencies between modules, so you should (theoretically) be able to use any of the modules independently in your project without using the whole package.

Download

Download the most recent version of Tornado from GitHub:

tornado-1.1.tar.gz

You can also browse the source on GitHub. To install Tornado:

tar xvzf tornado-1.1.tar.gz
cd tornado-1.1
python setup.py build
sudo python setup.py install

After installation, you should be able to run any of the demos in the demos directory included with the Tornado package.

./demos/helloworld/helloworld.py

Prerequisites

Tornado has been tested on Python 2.5 and 2.6. To use all of the features of Tornado, you need to have PycURL (version 7.18.2 or higher) and (for Python 2.5 only) simplejson installed (Python 2.6 includes JSON support in the standard library so simplejson is not needed). Complete installation instructions for Mac OS X and Ubuntu are included below for convenience.

Mac OS X 10.6 (Python 2.6)

sudo easy_install pycurl

Ubuntu Linux (Python 2.6)

sudo apt-get install python-pycurl

Ubuntu Linux (Python 2.5)

sudo apt-get install python-dev python-pycurl python-simplejson

Module index

The most important module is web, which is the web framework that includes most of the meat of the Tornado package. The other modules are tools that make web more useful. See Tornado walkthrough below for a detailed walkthrough of the web package.

Main modules

Low-level modules

Random modules

Tornado walkthrough

Request handlers and request arguments

A Tornado web application maps URLs or URL patterns to subclasses of tornado.web.RequestHandler. Those classes define get() or post() methods to handle HTTP GET or POST requests to that URL.

This code maps the root URL / to MainHandler and the URL pattern /story/([0-9]+) to StoryHandler. Regular expression groups are passed as arguments to the RequestHandler methods:

class MainHandler(tornado.web.RequestHandler):
    def get(self):
        self.write("You requested the main page")

class StoryHandler(tornado.web.RequestHandler):
    def get(self, story_id):
        self.write("You requested the story " + story_id)

application = tornado.web.Application([
    (r"/", MainHandler),
    (r"/story/([0-9]+)", StoryHandler),
])

You can get query string arguments and parse POST bodies with the get_argument() method:

class MainHandler(tornado.web.RequestHandler):
    def get(self):
        self.write('<html><body><form action="/" method="post">'
                   '<input type="text" name="message">'
                   '<input type="submit" value="Submit">'
                   '</form></body></html>')

    def post(self):
        self.set_header("Content-Type", "text/plain")
        self.write("You wrote " + self.get_argument("message"))

Uploaded files are available in self.request.files, which maps names (the name of the HTML <input type="file"> element) to a list of files. Each file is a dictionary of the form {"filename":..., "content_type":..., "body":...}.

If you want to send an error response to the client, e.g., 403 Unauthorized, you can just raise a tornado.web.HTTPError exception:

if not self.user_is_logged_in():
    raise tornado.web.HTTPError(403)

The request handler can access the object representing the current request with self.request. The HTTPRequest object includes a number of useful attributes, including:

See the class definition for HTTPRequest in httpserver for a complete list of attributes.

Overriding RequestHandler methods

In addition to get()/post()/etc, certain other methods in RequestHandler are designed to be overridden by subclasses when necessary. On every request, the following sequence of calls takes place:

  1. A new RequestHandler object is created on each request
  2. initialize() is called with keyword arguments from the Application configuration. (the initialize method is new in Tornado 1.1; in older versions subclasses would override __init__ instead). initialize should typically just save the arguments passed into member variables; it may not produce any output or call methods like send_error.
  3. prepare() is called. This is most useful in a base class shared by all of your handler subclasses, as prepare is called no matter which HTTP method is used. prepare may produce output; if it calls finish (or send_error, etc), processing stops here.
  4. One of the HTTP methods is called: get(), post(), put(), etc. If the URL regular expression contains capturing groups, they are passed as arguments to this method.

Here is an example demonstrating the initialize() method:

class ProfileHandler(RequestHandler):
    def initialize(self, database):
        self.database = database

    def get(self, username):
        ...

app = Application([
    (r'/user/(.*)', ProfileHandler, dict(database=database)),
    ])

Other methods designed for overriding include:

Templates

You can use any template language supported by Python, but Tornado ships with its own templating language that is a lot faster and more flexible than many of the most popular templating systems out there. See the template module documentation for complete documentation.

A Tornado template is just HTML (or any other text-based format) with Python control sequences and expressions embedded within the markup:

<html>
   <head>
      <title>{{ title }}</title>

   </head>
   <body>
     <ul>
       {% for item in items %}
         <li>{{ escape(item) }}</li>
       {% end %}
     </ul>

   </body>
 </html>

If you saved this template as "template.html" and put it in the same directory as your Python file, you could render this template with:

class MainHandler(tornado.web.RequestHandler):
    def get(self):
        items = ["Item 1", "Item 2", "Item 3"]
        self.render("template.html", title="My title", items=items)

Tornado templates support control statements and expressions. Control statements are surronded by {% and %}, e.g., {% if len(items) > 2 %}. Expressions are surrounded by {{ and }}, e.g., {{ items[0] }}.

Control statements more or less map exactly to Python statements. We support if, for, while, and try, all of which are terminated with {% end %}. We also support template inheritance using the extends and block statements, which are described in detail in the documentation for the template module.

Expressions can be any Python expression, including function calls. Template code is executed in a namespace that includes the following objects and functions (Note that this list applies to templates rendered using RequestHandler.render and render_string. If you're using the template module directly outside of a RequestHandler many of these entries are not present).

When you are building a real application, you are going to want to use all of the features of Tornado templates, especially template inheritance. Read all about those features in the template module section (some features, including UIModules are implemented in the web module)

Under the hood, Tornado templates are translated directly to Python. The expressions you include in your template are copied verbatim into a Python function representing your template. We don't try to prevent anything in the template language; we created it explicitly to provide the flexibility that other, stricter templating systems prevent. Consequently, if you write random stuff inside of your template expressions, you will get random Python errors when you execute the template.

Cookies and secure cookies

You can set cookies in the user's browser with the set_cookie method:

class MainHandler(tornado.web.RequestHandler):
    def get(self):
        if not self.get_cookie("mycookie"):
            self.set_cookie("mycookie", "myvalue")
            self.write("Your cookie was not set yet!")
        else:
            self.write("Your cookie was set!")

Cookies are easily forged by malicious clients. If you need to set cookies to, e.g., save the user ID of the currently logged in user, you need to sign your cookies to prevent forgery. Tornado supports this out of the box with the set_secure_cookie and get_secure_cookie methods. To use these methods, you need to specify a secret key named cookie_secret when you create your application. You can pass in application settings as keyword arguments to your application:

application = tornado.web.Application([
    (r"/", MainHandler),
], cookie_secret="61oETzKXQAGaYdkL5gEmGeJJFuYh7EQnp2XdTP1o/Vo=")

Signed cookies contain the encoded value of the cookie in addition to a timestamp and an HMAC signature. If the cookie is old or if the signature doesn't match, get_secure_cookie will return None just as if the cookie isn't set. The secure version of the example above:

class MainHandler(tornado.web.RequestHandler):
    def get(self):
        if not self.get_secure_cookie("mycookie"):
            self.set_secure_cookie("mycookie", "myvalue")
            self.write("Your cookie was not set yet!")
        else:
            self.write("Your cookie was set!")

User authentication

The currently authenticated user is available in every request handler as self.current_user, and in every template as current_user. By default, current_user is None.

To implement user authentication in your application, you need to override the get_current_user() method in your request handlers to determine the current user based on, e.g., the value of a cookie. Here is an example that lets users log into the application simply by specifying a nickname, which is then saved in a cookie:

class BaseHandler(tornado.web.RequestHandler):
    def get_current_user(self):
        return self.get_secure_cookie("user")

class MainHandler(BaseHandler):
    def get(self):
        if not self.current_user:
            self.redirect("/login")
            return
        name = tornado.escape.xhtml_escape(self.current_user)
        self.write("Hello, " + name)

class LoginHandler(BaseHandler):
    def get(self):
        self.write('<html><body><form action="/login" method="post">'
                   'Name: <input type="text" name="name">'
                   '<input type="submit" value="Sign in">'
                   '</form></body></html>')

    def post(self):
        self.set_secure_cookie("user", self.get_argument("name"))
        self.redirect("/")

application = tornado.web.Application([
    (r"/", MainHandler),
    (r"/login", LoginHandler),
], cookie_secret="61oETzKXQAGaYdkL5gEmGeJJFuYh7EQnp2XdTP1o/Vo=")

You can require that the user be logged in using the Python decorator tornado.web.authenticated. If a request goes to a method with this decorator, and the user is not logged in, they will be redirected to login_url (another application setting). The example above could be rewritten:

class MainHandler(BaseHandler):
    @tornado.web.authenticated
    def get(self):
        name = tornado.escape.xhtml_escape(self.current_user)
        self.write("Hello, " + name)

settings = {
    "cookie_secret": "61oETzKXQAGaYdkL5gEmGeJJFuYh7EQnp2XdTP1o/Vo=",
    "login_url": "/login",
}
application = tornado.web.Application([
    (r"/", MainHandler),
    (r"/login", LoginHandler),
], **settings)

If you decorate post() methods with the authenticated decorator, and the user is not logged in, the server will send a 403 response.

Tornado comes with built-in support for third-party authentication schemes like Google OAuth. See the auth module for more details. Check out the Tornado Blog example application for a complete example that uses authentication (and stores user data in a MySQL database).

Cross-site request forgery protection

Cross-site request forgery, or XSRF, is a common problem for personalized web applications. See the Wikipedia article for more information on how XSRF works.

The generally accepted solution to prevent XSRF is to cookie every user with an unpredictable value and include that value as an additional argument with every form submission on your site. If the cookie and the value in the form submission do not match, then the request is likely forged.

Tornado comes with built-in XSRF protection. To include it in your site, include the application setting xsrf_cookies:

settings = {
    "cookie_secret": "61oETzKXQAGaYdkL5gEmGeJJFuYh7EQnp2XdTP1o/Vo=",
    "login_url": "/login",
    "xsrf_cookies": True,
}
application = tornado.web.Application([
    (r"/", MainHandler),
    (r"/login", LoginHandler),
], **settings)

If xsrf_cookies is set, the Tornado web application will set the _xsrf cookie for all users and reject all POST requests that do not contain a correct _xsrf value. If you turn this setting on, you need to instrument all forms that submit via POST to contain this field. You can do this with the special function xsrf_form_html(), available in all templates:

<form action="/login" method="post">
  {{ xsrf_form_html() }}
  <div>Username: <input type="text" name="username"/></div>
  <div>Password: <input type="password" name="password"/></div>
  <div><input type="submit" value="Sign in"/></div>

</form>

If you submit AJAX POST requests, you will also need to instrument your JavaScript to include the _xsrf value with each request. This is the jQuery function we use at FriendFeed for AJAX POST requests that automatically adds the _xsrf value to all requests:

function getCookie(name) {
    var r = document.cookie.match("\b" + name + "=([^;]*)\b");
    return r ? r[1] : undefined;
}

jQuery.postJSON = function(url, args, callback) {
    args._xsrf = getCookie("_xsrf");
    $.ajax({url: url, data: $.param(args), dataType: "text", type: "POST",
        success: function(response) {
        callback(eval("(" + response + ")"));
    }});
};

Static files and aggressive file caching

You can serve static files from Tornado by specifying the static_path setting in your application:

settings = {
    "static_path": os.path.join(os.path.dirname(__file__), "static"),
    "cookie_secret": "61oETzKXQAGaYdkL5gEmGeJJFuYh7EQnp2XdTP1o/Vo=",
    "login_url": "/login",
    "xsrf_cookies": True,
}
application = tornado.web.Application([
    (r"/", MainHandler),
    (r"/login", LoginHandler),
], **settings)

This setting will automatically make all requests that start with /static/ serve from that static directory, e.g., http://localhost:8888/static/foo.png will serve the file foo.png from the specified static directory. We also automatically serve /robots.txt and /favicon.ico from the static directory (even though they don't start with the /static/ prefix).

To improve performance, it is generally a good idea for browsers to cache static resources aggressively so browsers won't send unnecessary If-Modified-Since or Etag requests that might block the rendering of the page. Tornado supports this out of the box with static content versioning.

To use this feature, use the static_url() method in your templates rather than typing the URL of the static file directly in your HTML:

<html>
   <head>
      <title>FriendFeed - {{ _("Home") }}</title>
   </head>

   <body>
     <div><img src="{{ static_url("images/logo.png") }}"/></div>
   </body>
 </html>

The static_url() function will translate that relative path to a URI that looks like /static/images/logo.png?v=aae54. The v argument is a hash of the content in logo.png, and its presence makes the Tornado server send cache headers to the user's browser that will make the browser cache the content indefinitely.

Since the v argument is based on the content of the file, if you update a file and restart your server, it will start sending a new v value, so the user's browser will automatically fetch the new file. If the file's contents don't change, the browser will continue to use a locally cached copy without ever checking for updates on the server, significantly improving rendering performance.

In production, you probably want to serve static files from a more optimized static file server like nginx. You can configure most any web server to support these caching semantics. Here is the nginx configuration we use at FriendFeed:

location /static/ {
    root /var/friendfeed/static;
    if ($query_string) {
        expires max;
    }
 }

Localization

The locale of the current user (whether they are logged in or not) is always available as self.locale in the request handler and as locale in templates. The name of the locale (e.g., en_US) is available as locale.name, and you can translate strings with the locale.translate method. Templates also have the global function call _() available for string translation. The translate function has two forms:

_("Translate this string")

which translates the string directly based on the current locale, and

_("A person liked this", "%(num)d people liked this", len(people)) % {"num": len(people)}

which translates a string that can be singular or plural based on the value of the third argument. In the example above, a translation of the first string will be returned if len(people) is 1, or a translation of the second string will be returned otherwise.

The most common pattern for translations is to use Python named placeholders for variables (the %(num)d in the example above) since placeholders can move around on translation.

Here is a properly localized template:

<html>

   <head>
      <title>FriendFeed - {{ _("Sign in") }}</title>
   </head>
   <body>
     <form action="{{ request.path }}" method="post">

       <div>{{ _("Username") }} <input type="text" name="username"/></div>
       <div>{{ _("Password") }} <input type="password" name="password"/></div>
       <div><input type="submit" value="{{ _("Sign in") }}"/></div>

       {{ xsrf_form_html() }}
     </form>
   </body>
 </html>

By default, we detect the user's locale using the Accept-Language header sent by the user's browser. We choose en_US if we can't find an appropriate Accept-Language value. If you let user's set their locale as a preference, you can override this default locale selection by overriding get_user_locale in your request handler:

class BaseHandler(tornado.web.RequestHandler):
    def get_current_user(self):
        user_id = self.get_secure_cookie("user")
        if not user_id: return None
        return self.backend.get_user_by_id(user_id)

    def get_user_locale(self):
        if "locale" not in self.current_user.prefs:
            # Use the Accept-Language header
            return None
        return self.current_user.prefs["locale"]

If get_user_locale returns None, we fall back on the Accept-Language header.

You can load all the translations for your application using the tornado.locale.load_translations method. It takes in the name of the directory which should contain CSV files named after the locales whose translations they contain, e.g., es_GT.csv or fr_CA.csv. The method loads all the translations from those CSV files and infers the list of supported locales based on the presence of each CSV file. You typically call this method once in the main() method of your server:

def main():
    tornado.locale.load_translations(
        os.path.join(os.path.dirname(__file__), "translations"))
    start_server()

You can get the list of supported locales in your application with tornado.locale.get_supported_locales(). The user's locale is chosen to be the closest match based on the supported locales. For example, if the user's locale is es_GT, and the es locale is supported, self.locale will be es for that request. We fall back on en_US if no close match can be found.

See the locale module documentation for detailed information on the CSV format and other localization methods.

UI modules

Tornado supports UI modules to make it easy to support standard, reusable UI widgets across your application. UI modules are like special functional calls to render components of your page, and they can come packaged with their own CSS and JavaScript.

For example, if you are implementing a blog, and you want to have blog entries appear on both the blog home page and on each blog entry page, you can make an Entry module to render them on both pages. First, create a Python module for your UI modules, e.g., uimodules.py:

class Entry(tornado.web.UIModule):
    def render(self, entry, show_comments=False):
        return self.render_string(
            "module-entry.html", show_comments=show_comments)

Tell Tornado to use uimodules.py using the ui_modules setting in your application:

class HomeHandler(tornado.web.RequestHandler):
    def get(self):
        entries = self.db.query("SELECT * FROM entries ORDER BY date DESC")
        self.render("home.html", entries=entries)

class EntryHandler(tornado.web.RequestHandler):
    def get(self, entry_id):
        entry = self.db.get("SELECT * FROM entries WHERE id = %s", entry_id)
        if not entry: raise tornado.web.HTTPError(404)
        self.render("entry.html", entry=entry)

settings = {
    "ui_modules": uimodules,
}
application = tornado.web.Application([
    (r"/", HomeHandler),
    (r"/entry/([0-9]+)", EntryHandler),
], **settings)

Within home.html, you reference the Entry module rather than printing the HTML directly:

{% for entry in entries %}
  {{ modules.Entry(entry) }}
{% end %}

Within entry.html, you reference the Entry module with the show_comments argument to show the expanded form of the entry:

{{ modules.Entry(entry, show_comments=True) }}

Modules can include custom CSS and JavaScript functions by overriding the embedded_css, embedded_javascript, javascript_files, or css_files methods:

class Entry(tornado.web.UIModule):
    def embedded_css(self):
        return ".entry { margin-bottom: 1em; }"

    def render(self, entry, show_comments=False):
        return self.render_string(
            "module-entry.html", show_comments=show_comments)

Module CSS and JavaScript will be included once no matter how many times a module is used on a page. CSS is always included in the <head> of the page, and JavaScript is always included just before the </body> tag at the end of the page.

Non-blocking, asynchronous requests

When a request handler is executed, the request is automatically finished. Since Tornado uses a non-blocking I/O style, you can override this default behavior if you want a request to remain open after the main request handler method returns using the tornado.web.asynchronous decorator.

When you use this decorator, it is your responsibility to call self.finish() to finish the HTTP request, or the user's browser will simply hang:

class MainHandler(tornado.web.RequestHandler):
    @tornado.web.asynchronous
    def get(self):
        self.write("Hello, world")
        self.finish()

Here is a real example that makes a call to the FriendFeed API using Tornado's built-in asynchronous HTTP client:

class MainHandler(tornado.web.RequestHandler):
    @tornado.web.asynchronous
    def get(self):
        http = tornado.httpclient.AsyncHTTPClient()
        http.fetch("http://friendfeed-api.com/v2/feed/bret",
                   callback=self.on_response)

    def on_response(self, response):
        if response.error: raise tornado.web.HTTPError(500)
        json = tornado.escape.json_decode(response.body)
        self.write("Fetched " + str(len(json["entries"])) + " entries "
                   "from the FriendFeed API")
        self.finish()

When get() returns, the request has not finished. When the HTTP client eventually calls on_response(), the request is still open, and the response is finally flushed to the client with the call to self.finish().

For a more advanced asynchronous example, take a look at the chat example application, which implements an AJAX chat room using long polling. Users of long polling may want to override on_connection_close() to clean up after the client closes the connection (but see that method's docstring for caveats).

Third party authentication

Tornado's auth module implements the authentication and authorization protocols for a number of the most popular sites on the web, including Google/Gmail, Facebook, Twitter, Yahoo, and FriendFeed. The module includes methods to log users in via these sites and, where applicable, methods to authorize access to the service so you can, e.g., download a user's address book or publish a Twitter message on their behalf.

Here is an example handler that uses Google for authentication, saving the Google credentials in a cookie for later access:

class GoogleHandler(tornado.web.RequestHandler, tornado.auth.GoogleMixin):
    @tornado.web.asynchronous
    def get(self):
        if self.get_argument("openid.mode", None):
            self.get_authenticated_user(self._on_auth)
            return
        self.authenticate_redirect()

    def _on_auth(self, user):
        if not user:
            self.authenticate_redirect()
            return
        # Save the user with, e.g., set_secure_cookie()

See the auth module documentation for more details.

Performance

Web application performance is generally bound by architecture, not frontend performance. That said, Tornado is pretty fast relative to most popular Python web frameworks.

We ran a few remedial load tests on a simple "Hello, world" application in each of the most popular Python web frameworks (Django, web.py, and CherryPy) to get the baseline performance of each relative to Tornado. We used Apache/mod_wsgi for Django and web.py and ran CherryPy as a standalone server, which was our impression of how each framework is typically run in production environments. We ran 4 single-threaded Tornado frontends behind an nginx reverse proxy, which is how we recommend running Tornado in production (our load test machine had four cores, and we recommend 1 frontend per core).

We load tested each with Apache Benchmark (ab) on the a separate machine with the command

ab -n 100000 -c 25 http://10.0.1.x/

The results (requests per second) on a 2.4GHz AMD Opteron processor with 4 cores:

In our tests, Tornado consistently had 4X the throughput of the next fastest framework, and even a single standalone Tornado frontend got 33% more throughput even though it only used one of the four cores.

Not very scientific, but at a high level, it should give you a sense that we have cared about performance as we built Tornado, and it shouldn't add too much latency to your apps relative to most Python web development frameworks.

Running Tornado in production

At FriendFeed, we use nginx as a load balancer and static file server. We run multiple instances of the Tornado web server on multiple frontend machines. We typically run one Tornado frontend per core on the machine (sometimes more depending on utilization).

When running behind a load balancer like nginx, it is recommended to pass xheaders=True to the HTTPServer constructor. This will tell Tornado to use headers like X-Real-IP to get the user's IP address instead of attributing all traffic to the balancer's IP address.

This is a barebones nginx config file that is structurally similar to the one we use at FriendFeed. It assumes nginx and the Tornado servers are running on the same machine, and the four Tornado servers are running on ports 8000 - 8003:

user nginx;
worker_processes 1;

error_log /var/log/nginx/error.log;
pid /var/run/nginx.pid;

events {
    worker_connections 1024;
    use epoll;
}

http {
    # Enumerate all the Tornado servers here
    upstream frontends {
        server 127.0.0.1:8000;
        server 127.0.0.1:8001;
        server 127.0.0.1:8002;
        server 127.0.0.1:8003;
    }

    include /etc/nginx/mime.types;
    default_type application/octet-stream;

    access_log /var/log/nginx/access.log;

    keepalive_timeout 65;
    proxy_read_timeout 200;
    sendfile on;
    tcp_nopush on;
    tcp_nodelay on;
    gzip on;
    gzip_min_length 1000;
    gzip_proxied any;
    gzip_types text/plain text/html text/css text/xml
               application/x-javascript application/xml
               application/atom+xml text/javascript;

    # Only retry if there was a communication error, not a timeout
    # on the Tornado server (to avoid propagating "queries of death"
    # to all frontends)
    proxy_next_upstream error;

    server {
        listen 80;

        # Allow file uploads
        client_max_body_size 50M;

        location ^~ /static/ {
            root /var/www;
            if ($query_string) {
                expires max;
            }
        }
        location = /favicon.ico {
            rewrite (.*) /static/favicon.ico;
        }
        location = /robots.txt {
            rewrite (.*) /static/robots.txt;
        }

        location / {
            proxy_pass_header Server;
            proxy_set_header Host $http_host;
            proxy_redirect false;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Scheme $scheme;
            proxy_pass http://frontends;
        }
    }
}

WSGI and Google AppEngine

Tornado comes with limited support for WSGI. However, since WSGI does not support non-blocking requests, you cannot use any of the asynchronous/non-blocking features of Tornado in your application if you choose to use WSGI instead of Tornado's HTTP server. Some of the features that are not available in WSGI applications: @tornado.web.asynchronous, the httpclient module, and the auth module.

You can create a valid WSGI application from your Tornado request handlers by using WSGIApplication in the wsgi module instead of using tornado.web.Application. Here is an example that uses the built-in WSGI CGIHandler to make a valid Google AppEngine application:

import tornado.web
import tornado.wsgi
import wsgiref.handlers

class MainHandler(tornado.web.RequestHandler):
    def get(self):
        self.write("Hello, world")

if __name__ == "__main__":
    application = tornado.wsgi.WSGIApplication([
        (r"/", MainHandler),
    ])
    wsgiref.handlers.CGIHandler().run(application)

See the appengine example application for a full-featured AppEngine app built on Tornado.

Caveats and support

Because FriendFeed and other large users of Tornado run behind nginx or Apache proxies, Tornado's HTTP server currently does not attempt to handle multi-line headers and some types of malformed input.

You can discuss Tornado and report bugs on the Tornado developer mailing list.