Development server and production

This section describes how to create new projects, manage the development server and deploy in production.

Cyclone ships with the cyclone command line tool to help on creating projects, and running your applications.

To create a new “Hello world” application:

$ cyclone app -n >

If you need a full featured project, with Redis and MySQL support, as well as sign up and password reset forms, check out the signup project template:

$ cyclone app --project=foobar --appskel=signup

Add --git to the command line if you want it to be initialized as a git repository.

For details on creating new projects see:


The cyclone command line is a wrapper to help on the development process.

Servers can be started like this:

$ cyclone run

As long as there’s at least one class that inherits from cyclone.web.Application in the Python file - like the above. You can also specify it using the absolute Python module path:

$ cyclone run --app=hello.Application

Always set debug=True in cyclone.web.Application.settings to get more detailed log messages, for development.

For more complex setups, like choosing different pid and log files, as well as daemonizing your server, use twistd:

$ twistd --pidfile=/var/run/ --logfile=/var/log/cyclone.log \
         --reactor=epoll cyclone --port 8888 --listen

Cyclone project templates ship with Debian init scripts for starting the server in production. For a single instance, or for one instance per CPU core, by setting the CPU affinity.

Faster DBs and Nginx

Always consider using Unix Sockets in production. They are considerably faster as they use less operating system resources, like all the syn/ack on regular INET/TCP connections.

Unix Sockets may be used for database communication, as well as for Cyclone servers behind Nginx.


import cyclone.redis
from twisted.internet import defer
from twisted.enterprise import adbapi

class DatabaseMixin(object):
    redis = cyclone.redis.lazyUnixConnectionPool(path="/tmp/redis.sock")
    mysql = adbapi.ConnectionPool("MySQLdb", db="dummy",

class MainHandler(web.RequestHandler, DatabaseMixin):
    def get(self):
        rs1 = yield self.mysql.runQuery("SELECT 1")
        rs2 = yield self.redis.get("foo")

And start one instance of this server per CPU core on the system, listening on Unix Sockets:

$ twistd --pidfile=/tmp/ -n cyclone -u /tmp/cyclone1.sock
$ twistd --pidfile=/tmp/ -n cyclone -u /tmp/cyclone2.sock

Now make Nginx connect on Cyclone via Unix Socket, with this configuration:

upstream backend {
  server unix:/tmp/cyclone1.sock;
  server unix:/tmp/cyclone2.sock;

server {
  listen      80;
  server_name localhost;

  location / {
    proxy_pass        http://backend;
    proxy_redirect    off;
    proxy_set_header  Host             $host;
    proxy_set_header  X-Real-IP        $remote_addr;
    proxy_set_header  X-Forwarded-For  $proxy_add_x_forwarded_for;

When Cyclone is reverse proxied by Nginx, you must set xheaders=True in cyclone.web.Application.settings so it uses X-Real-IP and X-Forwarded-For HTTP headers.

PyPy is better

Cyclone runs much better on PyPy than it does on CPython. Turns out that Twisted is actually faster, and therefore Cyclone too.

Besides using Unix Sockets for databases and Nginx, it is highly recommended that you run Cyclone on PyPy in production.

It has been tested on PyPy 1.8, 1.9 and 2.0-beta1 and runs well. There are some limitations though, like PyPy 1.8 does not support OpenSSL.

If you’re not familiar with setting up PyPy, try our bash script that selects the right PyPy version for your architecture and install it on /opt, with symbolic links to /usr/local/bin.

Make sure you have a compiler and openssl-dev. On Debian and Ubuntu systems, install these packages:

$ sudo apt-get install build-essential libssl-dev

Then run the installer script:

Table Of Contents

Previous topic

Deferreds and inline callbacks

Next topic — Command-line line tool

This Page