CherryPy is truly an open framework, you can extend and plug new functions at will either server-side or on a per-requests basis. Either way, CherryPy is made to help you build your application and support your architecture via simple patterns.

Server-wide functions

CherryPy can be considered both as a HTTP library as much as a web application framework. In that latter case, its architecture provides mechanisms to support operations across the whole server instance. This offers a powerful canvas to perform persistent operations as server-wide functions live outside the request processing itself. They are available to the whole process as long as the bus lives.

Typical use cases:

  • Keeping a pool of connection to an external server so that your need not to re-open them on each request (database connections for instance).

  • Background processing (say you need work to be done without blocking the whole request itself).

Publish/Subscribe pattern

CherryPy’s backbone consists of a bus system implementing a simple publish/subscribe messaging pattern. Simply put, in CherryPy everything is controlled via that bus. One can easily picture the bus as a sushi restaurant’s belt as in the picture below.


You can subscribe and publish to channels on a bus. A channel is bit like a unique identifier within the bus. When a message is published to a channel, the bus will dispatch the message to all subscribers for that channel.

One interesting aspect of a pubsub pattern is that it promotes decoupling between a caller and the callee. A published message will eventually generate a response but the publisher does not know where that response came from.

Thanks to that decoupling, a CherryPy application can easily access functionalities without having to hold a reference to the entity providing that functionality. Instead, the application simply publishes onto the bus and will receive the appropriate response, which is all that matter.

Typical pattern

Let’s take the following dummy application:

import cherrypy

class ECommerce(object):
    def __init__(self, db):
        self.mydb = db

    def save_kart(self, cart_data):
        cart = Cart(cart_data)

if __name__ == '__main__':
   cherrypy.quickstart(ECommerce(), '/')

The application has a reference to the database but this creates a fairly strong coupling between the database provider and the application.

Another approach to work around the coupling is by using a pubsub workflow:

import cherrypy

class ECommerce(object):
    def save_kart(self, cart_data):
        cart = Cart(cart_data)
        cherrypy.engine.publish('db-save', cart)

if __name__ == '__main__':
   cherrypy.quickstart(ECommerce(), '/')

In this example, we publish a cart instance to db-save channel. One or many subscribers can then react to that message and the application doesn’t have to know about them.


This approach is not mandatory and it’s up to you to decide how to design your entities interaction.

Implementation details

CherryPy’s bus implementation is simplistic as it registers functions to channels. Whenever a message is published to a channel, each registered function is applied with that message passed as a parameter.

The whole behaviour happens synchronously and, in that sense, if a subscriber takes too long to process a message, the remaining subscribers will be delayed.

CherryPy’s bus is not an advanced pubsub messaging broker system such as provided by zeromq or RabbitMQ. Use it with the understanding that it may have a cost.

Engine as a pubsub bus

As said earlier, CherryPy is built around a pubsub bus. All entities that the framework manages at runtime are working on top of a single bus instance, which is named the engine.

The bus implementation therefore provides a set of common channels which describe the application’s lifecycle:

   A   A         |
   |    \___     |
   |        \    |
   |         V   V

The states’ transitions trigger channels to be published to so that subscribers can react to them.

One good example is the HTTP server which will tranisition from a "STOPPED" stated to a "STARTED" state whenever a message is published to the start channel.

Built-in channels

In order to support its life-cycle, CherryPy defines a set of common channels that will be published to at various states:

  • “start”: When the bus is in the "STARTING" state

  • “main”: Periodically from the CherryPy’s mainloop

  • “stop”: When the bus is in the "STOPPING" state

  • “graceful”: When the bus requests a reload of subscribers

  • “exit”: When the bus is in the "EXITING" state

This channel will be published to by the engine automatically. Register therefore any subscribers that would need to react to the transition changes of the engine.

In addition, a few other channels are also published to during the request processing.

  • “before_request”: right before the request is processed by CherryPy

  • “after_request”: right after it has been processed

Also, from the cherrypy.process.plugins.ThreadManager plugin:

  • “acquire_thread”

  • “start_thread”

  • “stop_thread”

  • “release_thread”


In order to work with the bus, the implementation provides the following simple API:

  • The channel parameter is a string identifying the channel to which the message should be sent to

  • *args is the message and may contain any valid Python values or objects.

  • The channel parameter is a string identifying the channel the callable will be registered to.

  • callable is a Python function or method which signature must match what will be published.

  • The channel parameter is a string identifying the channel the callable was registered to.

  • callable is the Python function or method which was registered.


Plugins, simply put, are entities that play with the bus, either by publishing or subscribing to channels, usually both at the same time.


Plugins are extremely useful whenever you have functionalities:

  • Available across the whole application server

  • Associated to the application’s life-cycle

  • You want to avoid being strongly coupled to the application

Create a plugin

A typical plugin looks like this:

import cherrypy
from cherrypy.process import wspbus, plugins

class DatabasePlugin(plugins.SimplePlugin):
    def __init__(self, bus, db_klass):
        plugins.SimplePlugin.__init__(self, bus)
        self.db = db_klass()

    def start(self):
        self.bus.log('Starting up DB access')
        self.bus.subscribe("db-save", self.save_it)

    def stop(self):
        self.bus.log('Stopping down DB access')
        self.bus.unsubscribe("db-save", self.save_it)

    def save_it(self, entity):

The cherrypy.process.plugins.SimplePlugin is a helper class provided by CherryPy that will automatically subscribe your start and stop methods to the related channels.

When the start and stop channels are published on, those methods are called accordingly.

Notice then how our plugin subscribes to the db-save channel so that the bus can dispatch messages to the plugin.

Enable a plugin

To enable the plugin, it has to be registered to the the bus as follows:

DatabasePlugin(cherrypy.engine, SQLiteDB).subscribe()

The SQLiteDB here is a fake class that is used as our database provider.

Disable a plugin

You can also unregister a plugin as follows:


This is often used when you want to prevent the default HTTP server from being started by CherryPy, for instance if you run on top of a different HTTP server (WSGI capable):


Let’s see an example using this default application:

import cherrypy

class Root(object):
    def index(self):
        return "hello world"

if __name__ == '__main__':

For instance, this is what you would see when running this application:

[27/Apr/2014:13:04:07] ENGINE Listening for SIGHUP.
[27/Apr/2014:13:04:07] ENGINE Listening for SIGTERM.
[27/Apr/2014:13:04:07] ENGINE Listening for SIGUSR1.
[27/Apr/2014:13:04:07] ENGINE Bus STARTING
[27/Apr/2014:13:04:07] ENGINE Started monitor thread 'Autoreloader'.
[27/Apr/2014:13:04:08] ENGINE Serving on
[27/Apr/2014:13:04:08] ENGINE Bus STARTED

Now let’s unsubscribe the HTTP server:

import cherrypy

class Root(object):
    def index(self):
        return "hello world"

if __name__ == '__main__':

This is what we get:

[27/Apr/2014:13:08:06] ENGINE Listening for SIGHUP.
[27/Apr/2014:13:08:06] ENGINE Listening for SIGTERM.
[27/Apr/2014:13:08:06] ENGINE Listening for SIGUSR1.
[27/Apr/2014:13:08:06] ENGINE Bus STARTING
[27/Apr/2014:13:08:06] ENGINE Started monitor thread 'Autoreloader'.
[27/Apr/2014:13:08:06] ENGINE Bus STARTED

As you can see, the server is not started. The missing:

[27/Apr/2014:13:04:08] ENGINE Serving on

Per-request functions

One of the most common task in a web application development is to tailor the request’s processing to the runtime context.

Within CherryPy, this is performed via what are called Tools. If you are familiar with Django or WSGI middlewares, CherryPy tools are similar in spirit. They add functions that are applied during the request/response processing.

Hook point

A hook point is a point during the request/response processing.

Here is a quick rundown of the “hook points” that you can hang your tools on:

  • “on_start_resource” - The earliest hook; the Request-Line and request headers have been processed and a dispatcher has set request.handler and request.config.

  • “before_request_body” - Tools that are hooked up here run right before the request body would be processed.

  • “before_handler” - Right before the request.handler (the exposed callable that was found by the dispatcher) is called.

  • “before_finalize” - This hook is called right after the page handler has been processed and before CherryPy formats the final response object. It helps you for example to check for what could have been returned by your page handler and change some headers if needed.

  • “on_end_resource” - Processing is complete - the response is ready to be returned. This doesn’t always mean that the request.handler (the exposed page handler) has executed! It may be a generator. If your tool absolutely needs to run after the page handler has produced the response body, you need to either use on_end_request instead, or wrap the response.body in a generator which applies your tool as the response body is being generated.

  • “before_error_response” - Called right before an error response (status code, body) is set.

  • “after_error_response” - Called right after the error response (status code, body) is set and just before the error response is finalized.

  • “on_end_request” - The request/response conversation is over, all data has been written to the client, nothing more to see here, move along.


A tool is a simple callable object (function, method, object implementing a __call__ method) that is attached to a hook point.

Below is a simple tool that is attached to the before_finalize hook point, hence after the page handler was called:

def logit():

Tools can also be created and assigned manually. The decorator registration is equivalent to:

cherrypy.tools.logit = cherrypy.Tool('before_finalize', logit)

Using that tool is as simple as follows:

class Root(object):
    def index(self):
        return "hello world"

Obviously the tool may be declared the other usual ways.


The name of the tool, technically the attribute set to cherrypy.tools, does not have to match the name of the callable. However, it is that name that will be used in the configuration to refer to that tool.

Stateful tools

The tools mechanism is really flexible and enables rich per-request functionalities.

Straight tools as shown in the previous section are usually good enough. However, if your workflow requires some sort of state during the request processing, you will probably want a class-based approach:

import time

import cherrypy

class TimingTool(cherrypy.Tool):
    def __init__(self):
        cherrypy.Tool.__init__(self, 'before_handler',

    def _setup(self):

    def start_timer(self):
        cherrypy.request._time = time.time()

    def end_timer(self):
        duration = time.time() - cherrypy.request._time
        cherrypy.log("Page handler took %.4f" % duration)

cherrypy.tools.timeit = TimingTool()

This tool computes the time taken by the page handler for a given request. It stores the time at which the handler is about to get called and logs the time difference right after the handler returned its result.

The import bits is that the cherrypy.Tool constructor allows you to register to a hook point but, to attach the same tool to a different hook point, you must use the cherrypy.request.hooks.attach method. The cherrypy.Tool._setup method is automatically called by CherryPy when the tool is applied to the request.

Next, let’s see how to use our tool:

class Root(object):
    def index(self):
        return "hello world"

Tools ordering

Since you can register many tools at the same hookpoint, you may wonder in which order they will be applied.

CherryPy offers a deterministic, yet so simple, mechanism to do so. Simply set the priority attribute to a value from 1 to 100, lower values providing greater priority.

If you set the same priority for several tools, they will be called in the order you declare them in your configuration.


All of the builtin CherryPy tools are collected into a Toolbox called cherrypy.tools. It responds to config entries in the "tools" namespace. You can add your own Tools to this Toolbox as described above.

You can also make your own Toolboxes if you need more modularity. For example, you might create multiple Tools for working with JSON, or you might publish a set of Tools covering authentication and authorization from which everyone could benefit (hint, hint). Creating a new Toolbox is as simple as:

import cherrypy

# Create a new Toolbox.
newauthtools = cherrypy._cptools.Toolbox("newauth")

# Add a Tool to our new Toolbox.
def check_access(default=False):
    if not getattr(cherrypy.request, "userid", default):
        raise cherrypy.HTTPError(401)

Then, in your application, use it just like you would use cherrypy.tools, with the additional step of registering your toolbox with your app. Note that doing so automatically registers the "newauth" config namespace; you can see the config entries in action below:

import cherrypy

class Root(object):
    def default(self):
        return "Hello"

conf = {
   '/demo': {
       'newauth.check_access.on': True,
       'newauth.check_access.default': True,

app = cherrypy.tree.mount(Root(), config=conf)

Request parameters manipulation

HTTP uses strings to carry data between two endpoints. However your application may make better use of richer object types. As it wouldn’t be really readable, nor a good idea regarding maintenance, to let each page handler deserialize data, it’s a common pattern to delegate this functions to tools.

For instance, let’s assume you have a user id in the query-string and some user data stored into a database. You could retrieve the data, create an object and pass it on to the page handler instead of the user id.

import cherrypy

class UserManager(cherrypy.Tool):
    def __init__(self):
        cherrypy.Tool.__init__(self, 'before_handler',
                               self.load, priority=10)

    def load(self):
        req = cherrypy.request

        # let's assume we have a db session
        # attached to the request somehow
        db = req.db

        # retrieve the user id and remove it
        # from the request parameters
        user_id = req.params.pop('user_id')
        req.params['user'] = db.get(int(user_id))

cherrypy.tools.user = UserManager()

class Root(object):
    def index(self, user):
        return "hello %s" % user.name

In other words, CherryPy give you the power to:

  • inject data, that wasn’t part of the initial request, into the page handler

  • remove data as well

  • convert data into a different, more useful, object to remove that burden from the page handler itself

Tailored dispatchers

Dispatching is the art of locating the appropriate page handler for a given request. Usually, dispatching is based on the request’s URL, the query-string and, sometimes, the request’s method (GET, POST, etc.).

Based on this, CherryPy comes with various dispatchers already.

In some cases however, you will need a little more. Here is an example of dispatcher that will always ensure the incoming URL leads to a lower-case page handler.

import random
import string

import cherrypy
from cherrypy._cpdispatch import Dispatcher

class StringGenerator(object):
   def generate(self, length=8):
       return ''.join(random.sample(string.hexdigits, int(length)))

class ForceLowerDispatcher(Dispatcher):
    def __call__(self, path_info):
        return Dispatcher.__call__(self, path_info.lower())

if __name__ == '__main__':
    conf = {
        '/': {
            'request.dispatch': ForceLowerDispatcher(),
    cherrypy.quickstart(StringGenerator(), '/', conf)

Once you run this snippet, go to:

In both cases, you will be led to the generate page handler. Without our home-made dispatcher, the second one would fail and return a 404 error (RFC 7231#section-6.5.4).

Tool or dispatcher?

In the previous example, why not simply use a tool? Well, the sooner a tool can be called is always after the page handler has been found. In our example, it would be already too late as the default dispatcher would have not even found a match for /GENerAte.

A dispatcher exists mostly to determine the best page handler to serve the requested resource.

On the other hand, tools are there to adapt the request’s processing to the runtime context of the application and the request’s content.

Usually, you will have to write a dispatcher only if you have a very specific use case to locate the most adequate page handler. Otherwise, the default ones will likely suffice.

Request body processors

Since its 3.2 release, CherryPy provides a really elegant and powerful mechanism to deal with a request’s body based on its mimetype. Refer to the cherrypy._cpreqbody module to understand how to implement your own processors.