Deploy

CherryPy stands on its own, but as an application server, it is often located in shared or complex environments. For this reason, it is not uncommon to run CherryPy behind a reverse proxy or use other servers to host the application.

Note

CherryPy’s server has proven reliable and fast enough for years now. If the volume of traffic you receive is average, it will do well enough on its own. Nonetheless, it is common to delegate the serving of static content to more capable servers such as nginx or CDN.

Run as a daemon

CherryPy allows you to easily decouple the current process from the parent environment, using the traditional double-fork:

from cherrypy.process.plugins import Daemonizer
d = Daemonizer(cherrypy.engine)
d.subscribe()

Note

This engine plugin is only available on Unix and similar systems which provide fork().

If a startup error occurs in the forked children, the return code from the parent process will still be 0. Errors in the initial daemonizing process still return proper exit codes, but errors after the fork won’t. Therefore, if you use this plugin to daemonize, don’t use the return code as an accurate indicator of whether the process fully started. In fact, that return code only indicates if the process successfully finished the first fork.

The plugin takes optional arguments to redirect standard streams: stdin, stdout, and stderr. By default, these are all redirected to /dev/null, but you’re free to send them to log files or elsewhere.

Warning

You should be careful to not start any threads before this plugin runs. The plugin will warn if you do so, because “…the effects of calling functions that require certain resources between the call to fork() and the call to an exec function are undefined”. (ref). It is for this reason that the Server plugin runs at priority 75 (it starts worker threads), which is later than the default priority of 65 for the Daemonizer.

Run as a different user

Use this engine plugin to start your CherryPy site as root (for example, to listen on a privileged port like 80) and then reduce privileges to something more restricted.

This priority of this plugin’s “start” listener is slightly higher than the priority for server.start in order to facilitate the most common use: starting on a low port (which requires root) and then dropping to another user.

DropPrivileges(cherrypy.engine, uid=1000, gid=1000).subscribe()

PID files

The PIDFile engine plugin is pretty straightforward: it writes the process id to a file on start, and deletes the file on exit. You must provide a ‘pidfile’ argument, preferably an absolute path:

PIDFile(cherrypy.engine, '/var/run/myapp.pid').subscribe()

Systemd socket activation

Socket Activation is a systemd feature that allows to setup a system so that the systemd will sit on a port and start services ‘on demand’ (a little bit like inetd and xinetd used to do).

CherryPy has built-in socket activation support, if run from a systemd service file it will detect the LISTEN_PID environment variable to know that it should consider fd 3 to be the passed socket.

To read more about socket activation: http://0pointer.de/blog/projects/socket-activation.html

Control via Supervisord

Supervisord is a powerful process control and management tool that can perform a lot of tasks around process monitoring.

Below is a simple supervisor configuration for your CherryPy application.

[unix_http_server]
file=/tmp/supervisor.sock

[supervisord]
logfile=/tmp/supervisord.log ; (main log file;default $CWD/supervisord.log)
logfile_maxbytes=50MB        ; (max main logfile bytes b4 rotation;default 50MB)
logfile_backups=10           ; (num of main logfile rotation backups;default 10)
loglevel=info                ; (log level;default info; others: debug,warn,trace)
pidfile=/tmp/supervisord.pid ; (supervisord pidfile;default supervisord.pid)
nodaemon=false               ; (start in foreground if true;default false)
minfds=1024                  ; (min. avail startup file descriptors;default 1024)
minprocs=200                 ; (min. avail process descriptors;default 200)

[rpcinterface:supervisor]
supervisor.rpcinterface_factory = supervisor.rpcinterface:make_main_rpcinterface

[supervisorctl]
serverurl=unix:///tmp/supervisor.sock

[program:myapp]
command=python server.py
environment=PYTHONPATH=.
directory=.

This could control your server via the server.py module as the application entry point.

import cherrypy

class Root(object):
    @cherrypy.expose
    def index(self):
        return "Hello World!"


cherrypy.config.update({'server.socket_port': 8090,
                        'engine.autoreload.on': False,
                        'log.access_file': './access.log',
                        'log.error_file': './error.log'})
cherrypy.quickstart(Root())

To take the configuration (assuming it was saved in a file called supervisor.conf) into account:

$ supervisord -c supervisord.conf
$ supervisorctl update

Now, you can point your browser at http://localhost:8090/ and it will display Hello World!.

To stop supervisor, type:

$ supervisorctl shutdown

This will obviously shutdown your application.

SSL support

Note

You may want to test your server for SSL using the services from Qualys, Inc.

CherryPy can encrypt connections using SSL to create an https connection. This keeps your web traffic secure. Here’s how.

  1. Generate a private key. We’ll use openssl and follow the OpenSSL Keys HOWTO.:

$ openssl genrsa -out privkey.pem 2048

You can create either a key that requires a password to use, or one without a password. Protecting your private key with a password is much more secure, but requires that you enter the password every time you use the key. For example, you may have to enter the password when you start or restart your CherryPy server. This may or may not be feasible, depending on your setup.

If you want to require a password, add one of the -aes128, -aes192 or -aes256 switches to the command above. You should not use any of the DES, 3DES, or SEED algorithms to protect your password, as they are insecure.

SSL Labs recommends using 2048-bit RSA keys for security (see references section at the end).

  1. Generate a certificate. We’ll use openssl and follow the OpenSSL Certificates HOWTO. Let’s start off with a self-signed certificate for testing:

$ openssl req -new -x509 -days 365 -key privkey.pem -out cert.pem

openssl will then ask you a series of questions. You can enter whatever values are applicable, or leave most fields blank. The one field you must fill in is the ‘Common Name’: enter the hostname you will use to access your site. If you are just creating a certificate to test on your own machine and you access the server by typing ‘localhost’ into your browser, enter the Common Name ‘localhost’.

  1. Decide whether you want to use python’s built-in SSL library, or the pyOpenSSL library. CherryPy supports either.

    1. Built-in. To use python’s built-in SSL, add the following line to your CherryPy config:

    cherrypy.server.ssl_module = 'builtin'

    1. pyOpenSSL. Because python did not have a built-in SSL library when CherryPy was first created, the default setting is to use pyOpenSSL. To use it you’ll need to install it (we could recommend you install cython first):

    $ pip install cython, pyOpenSSL

  2. Add the following lines in your CherryPy config to point to your certificate files:

cherrypy.server.ssl_certificate = "cert.pem"
cherrypy.server.ssl_private_key = "privkey.pem"

  1. If you have a certificate chain at hand, you can also specify it:

cherrypy.server.ssl_certificate_chain = "certchain.perm"

  1. Start your CherryPy server normally. Note that if you are debugging locally and/or using a self-signed certificate, your browser may show you security warnings.

WSGI servers

Embedding into another WSGI framework

Though CherryPy comes with a very reliable and fast enough HTTP server, you may wish to integrate your CherryPy application within a different framework. To do so, we will benefit from the WSGI interface defined in PEP 333 and PEP 3333.

Note that you should follow some basic rules when embedding CherryPy in a third-party WSGI server:

  • If you rely on the "main" channel to be published on, as it would happen within the CherryPy’s mainloop, you should find a way to publish to it within the other framework’s mainloop.

  • Start the CherryPy’s engine. This will publish to the "start" channel of the bus.

    cherrypy.engine.start()

  • Stop the CherryPy’s engine when you terminate. This will publish to the "stop" channel of the bus.

    cherrypy.engine.stop()

  • Do not call cherrypy.engine.block().

  • Disable the built-in HTTP server since it will not be used.

    cherrypy.server.unsubscribe()

  • Disable autoreload. Usually other frameworks won’t react well to it, or sometimes, provide the same feature.

    cherrypy.config.update({'engine.autoreload.on': False})

  • Disable CherryPy signals handling. This may not be needed, it depends on how the other framework handles them.

    cherrypy.engine.signals.subscribe()

  • Use the "embedded" environment configuration scheme.

    cherrypy.config.update({'environment': 'embedded'})

    Essentially this will disable the following:

    • Stdout logging

    • Autoreloader

    • Configuration checker

    • Headers logging on error

    • Tracebacks in error

    • Mismatched params error during dispatching

    • Signals (SIGHUP, SIGTERM)

Tornado

You can use tornado HTTP server as follow:

import cherrypy

class Root(object):
    @cherrypy.expose
    def index(self):
        return "Hello World!"

if __name__ == '__main__':
    import tornado
    import tornado.httpserver
    import tornado.wsgi

    # our WSGI application
    wsgiapp = cherrypy.tree.mount(Root())

    # Disable the autoreload which won't play well
    cherrypy.config.update({'engine.autoreload.on': False})

    # let's not start the CherryPy HTTP server
    cherrypy.server.unsubscribe()

    # use CherryPy's signal handling
    cherrypy.engine.signals.subscribe()

    # Prevent CherryPy logs to be propagated
    # to the Tornado logger
    cherrypy.log.error_log.propagate = False

    # Run the engine but don't block on it
    cherrypy.engine.start()

    # Run thr tornado stack
    container = tornado.wsgi.WSGIContainer(wsgiapp)
    http_server = tornado.httpserver.HTTPServer(container)
    http_server.listen(8080)
    # Publish to the CherryPy engine as if
    # we were using its mainloop
    tornado.ioloop.PeriodicCallback(lambda: cherrypy.engine.publish('main'), 100).start()
    tornado.ioloop.IOLoop.instance().start()

Twisted

You can use Twisted HTTP server as follow:

import cherrypy

from twisted.web.wsgi import WSGIResource
from twisted.internet import reactor
from twisted.internet import task

# Our CherryPy application
class Root(object):
    @cherrypy.expose
    def index(self):
        return "hello world"

# Create our WSGI app from the CherryPy application
wsgiapp = cherrypy.tree.mount(Root())

# Configure the CherryPy's app server
# Disable the autoreload which won't play well
cherrypy.config.update({'engine.autoreload.on': False})

# We will be using Twisted HTTP server so let's
# disable the CherryPy's HTTP server entirely
cherrypy.server.unsubscribe()

# If you'd rather use CherryPy's signal handler
# Uncomment the next line. I don't know how well this
# will play with Twisted however
#cherrypy.engine.signals.subscribe()

# Publish periodically onto the 'main' channel as the bus mainloop would do
task.LoopingCall(lambda: cherrypy.engine.publish('main')).start(0.1)

# Tie our app to Twisted
reactor.addSystemEventTrigger('after', 'startup', cherrypy.engine.start)
reactor.addSystemEventTrigger('before', 'shutdown', cherrypy.engine.exit)
resource = WSGIResource(reactor, reactor.getThreadPool(), wsgiapp)

Notice how we attach the bus methods to the Twisted’s own lifecycle.

Save that code into a module named cptw.py and run it as follows:

$ twistd -n web --port 8080 --wsgi cptw.wsgiapp

uwsgi

You can use uwsgi HTTP server as follow:

import cherrypy

# Our CherryPy application
class Root(object):
    @cherrypy.expose
    def index(self):
        return "hello world"

cherrypy.config.update({'engine.autoreload.on': False})
cherrypy.server.unsubscribe()
cherrypy.engine.start()

wsgiapp = cherrypy.tree.mount(Root())

Save this into a Python module called mymod.py and run it as follows:

$ uwsgi --socket 127.0.0.1:8080 --protocol=http --wsgi-file mymod.py --callable wsgiapp

Virtual Hosting

CherryPy has support for virtual-hosting. It does so through a dispatchers that locate the appropriate resource based on the requested domain.

Below is a simple example for it:

import cherrypy

class Root(object):
    def __init__(self):
        self.app1 = App1()
        self.app2 = App2()

class App1(object):
    @cherrypy.expose
    def index(self):
        return "Hello world from app1"

class App2(object):
    @cherrypy.expose
    def index(self):
        return "Hello world from app2"

if __name__ == '__main__':
    hostmap = {
        'company.com:8080': '/app1',
        'home.net:8080': '/app2',
    }

    config = {
        'request.dispatch': cherrypy.dispatch.VirtualHost(**hostmap)
    }

    cherrypy.quickstart(Root(), '/', {'/': config})

In this example, we declare two domains and their ports:

  • company.com:8080

  • home.net:8080

Thanks to the cherrypy.dispatch.VirtualHost dispatcher, we tell CherryPy which application to dispatch to when a request arrives. The dispatcher looks up the requested domain and call the according application.

Note

To test this example, simply add the following rules to your hosts file:

127.0.0.1       company.com
127.0.0.1       home.net

Reverse-proxying

Apache

Nginx

nginx is a fast and modern HTTP server with a small footprint. It is a popular choice as a reverse proxy to application servers such as CherryPy.

This section will not cover the whole range of features nginx provides. Instead, it will simply provide you with a basic configuration that can be a good starting point.

 1upstream apps {
 2   server 127.0.0.1:8080;
 3   server 127.0.0.1:8081;
 4}
 5
 6gzip_http_version 1.0;
 7gzip_proxied      any;
 8gzip_min_length   500;
 9gzip_disable      "MSIE [1-6]\.";
10gzip_types        text/plain text/xml text/css
11                  text/javascript
12                  application/javascript;
13
14server {
15   listen 80;
16   server_name  www.example.com;
17
18   access_log  /app/logs/www.example.com.log combined;
19   error_log  /app/logs/www.example.com.log;
20
21   location ^~ /static/  {
22      root /app/static/;
23   }
24
25   location / {
26      proxy_pass         http://apps;
27      proxy_redirect     off;
28      proxy_set_header   Host $host;
29      proxy_set_header   X-Real-IP $remote_addr;
30      proxy_set_header   X-Forwarded-For $proxy_add_x_forwarded_for;
31      proxy_set_header   X-Forwarded-Host $server_name;
32   }
33}

Edit this configuration to match your own paths. Then, save this configuration into a file under /etc/nginx/conf.d/ (assuming Ubuntu). The filename is irrelevant. Then run the following commands:

$ sudo service nginx stop
$ sudo service nginx start

Hopefully, this will be enough to forward requests hitting the nginx frontend to your CherryPy application. The upstream block defines the addresses of your CherryPy instances.

It shows that you can load-balance between two application servers. Refer to the nginx documentation to understand how this achieved.

upstream apps {
   server 127.0.0.1:8080;
   server 127.0.0.1:8081;
}

Later on, this block is used to define the reverse proxy section.

Now, let’s see our application:

import cherrypy

class Root(object):
    @cherrypy.expose
    def index(self):
        return "hello world"

if __name__ == '__main__':
    cherrypy.config.update({
        'server.socket_port': 8080,
        'tools.proxy.on': True,
        'tools.proxy.base': 'http://www.example.com'
    })
    cherrypy.quickstart(Root())

If you run two instances of this code, one on each port defined in the nginx section, you will be able to reach both of them via the load-balancing done by nginx.

Notice how we define the proxy tool. It is not mandatory and used only so that the CherryPy request knows about the true client’s address. Otherwise, it would know only about the nginx’s own address. This is most visible in the logs.

The base attribute should match the server_name section of the nginx configuration.