The bottle run() function, when called without any parameters, starts a local development server on port 8080. You can access and test your application via http://localhost:8080/ if you are on the same host.

To get your application available to the outside world, specify the IP of the interface the server should listen to (e.g. run(host='')) or let the server listen to all interfaces at once (e.g. run(host='')). The listening port can be changed in a similar way, but you need root or admin rights to choose a port below 1024. Port 80 is the standard for HTTP servers:

run(host='', port=80) # Listen to HTTP requests on all interfaces

Server Options

The built-in default server is based on wsgiref WSGIServer. This non-threading HTTP server is perfectly fine for development and early production, but may become a performance bottleneck when server load increases. There are three ways to eliminate this bottleneck:

  • Use a different server that is either multi-threaded or asynchronous.
  • Start multiple server processes and spread the load with a load-balancer.
  • Do both.

Multi-threaded servers are the ‘classic’ way to do it. They are very robust, reasonably fast and easy to manage. As a drawback, they can only handle a limited number of connections at the same time and utilize only one CPU core due to the “Global Interpreter Lock” (GIL). This does not hurt most applications, they are waiting for network IO most of the time anyway, but may slow down CPU intensive tasks (e.g. image processing).

Asynchronous servers are very fast, can handle a virtually unlimited number of concurrent connections and are easy to manage, but can get a bit tricky. To take full advantage of their potential, you need to design your application accordingly and understand the concepts of the specific server.

Multi-processing (forking) servers are not limited by the GIL and utilize more than one CPU core, but make communication between server instances more expensive. You need a database or external message query to share state between processes, or design your application so that it does not need any shared state. The setup is also a bit more complicated, but there are good tutorials available.

Switching the Server Backend

The easiest way to increase performance is to install a multi-threaded server library like paste or cherrypy and tell Bottle to use that instead of the single-threaded server:'paste')

Bottle ships with a lot of ready-to-use adapters for the most common WSGI servers and automates the setup process. Here is an incomplete list:

Name Homepage Description
cgi   Run as CGI script
flup flup Run as FastCGI process
gae gae Helper for Google App Engine deployments
wsgiref wsgiref Single-threaded default server
cherrypy cherrypy Multi-threaded and very stable
paste paste Multi-threaded, stable, tried and tested
rocket rocket Multi-threaded
waitress waitress Multi-threaded, poweres Pyramid
gunicorn gunicorn Pre-forked, partly written in C
eventlet eventlet Asynchronous framework with WSGI support.
gevent gevent Asynchronous (greenlets)
diesel diesel Asynchronous (greenlets)
fapws3 fapws3 Asynchronous (network side only), written in C
tornado tornado Asynchronous, powers some parts of Facebook
twisted twisted Asynchronous, well tested but… twisted
meinheld meinheld Asynchronous, partly written in C
bjoern bjoern Asynchronous, very fast and written in C
auto   Automatically selects an available server adapter

The full list is available through server_names.

If there is no adapter for your favorite server or if you need more control over the server setup, you may want to start the server manually. Refer to the server documentation on how to run WSGI applications. Here is an example for paste:

application = bottle.default_app()
from paste import httpserver
httpserver.serve(application, host='', port=80)

Apache mod_wsgi

Instead of running your own HTTP server from within Bottle, you can attach Bottle applications to an Apache server using mod_wsgi.

All you need is an app.wsgi file that provides an application object. This object is used by mod_wsgi to start your application and should be a WSGI-compatible Python callable.

File /var/www/yourapp/app.wsgi:

# Change working directory so relative paths (and template lookup) work again

import bottle
# ... build or import your bottle application here ...
# Do NOT use with mod_wsgi
application = bottle.default_app()

The Apache configuration may look like this:

<VirtualHost *>

    WSGIDaemonProcess yourapp user=www-data group=www-data processes=1 threads=5
    WSGIScriptAlias / /var/www/yourapp/app.wsgi

    <Directory /var/www/yourapp>
        WSGIProcessGroup yourapp
        WSGIApplicationGroup %{GLOBAL}
        Order deny,allow
        Allow from all

Google AppEngine

New in version 0.9.

The gae server adapter is used to run applications on Google App Engine. It works similar to the cgi adapter in that it does not start a new HTTP server, but prepares and optimizes your application for Google App Engine and makes sure it conforms to their API:'gae') # No need for a host or port setting.

It is always a good idea to let GAE serve static files directly. Here is example for a working app.yaml:

application: myapp
version: 1
runtime: python
api_version: 1

- url: /static
  static_dir: static

- url: /.*

Load Balancer (Manual Setup)

A single Python process can utilize only one CPU at a time, even if there are more CPU cores available. The trick is to balance the load between multiple independent Python processes to utilize all of your CPU cores.

Instead of a single Bottle application server, you start one instance for each CPU core available using different local port (localhost:8080, 8081, 8082, …). You can choose any server adapter you want, even asynchronous ones. Then a high performance load balancer acts as a reverse proxy and forwards each new requests to a random port, spreading the load between all available back-ends. This way you can use all of your CPU cores and even spread out the load between different physical servers.

One of the fastest load balancers available is Pound but most common web servers have a proxy-module that can do the work just fine.

Pound example:

    Port    80

            Port    8080
            Port    8081

Apache example:

<Proxy balancer://mycluster>
ProxyPass / balancer://mycluster

Lighttpd example:

server.modules += ( "mod_proxy" )
proxy.server = (
    "" => (
        "wsgi1" => ( "host" => "", "port" => 8080 ),
        "wsgi2" => ( "host" => "", "port" => 8081 )

Good old CGI

A CGI server starts a new process for each request. This adds a lot of overhead but is sometimes the only option, especially on cheap hosting packages. The cgi server adapter does not actually start a CGI server, but transforms your bottle application into a valid CGI application:'cgi')