Deploying WSGI Applications
The best thing about WSGI is that there are so many gateways for productive usage. If you want to switch from Apache2 with mod_wsgi to lighttpd with FastCGI it's a matter of a few minutes. If you want to distribute your application on a usb stick for presentation you can bundle everything into an executable using py2exe and the builtin wsgiref module.
The following list shows the most often used server configurations and how you can serve your WSGI applications.
The following gateways are webserver agnostic and will work with any webserver out there that suppors the underlaying interface:
- For FastCGI two well known python libraries exist. One is implemented in python and handles pretty every FastCGI configuration. It's called flup and supports also some other interfaces beside FastCGI. The other one is called python-fastcgi and is implemented in C. However the latter has a much smaller featureset.
- Simple protocol with the same benefits of FastCGI (persistent interpreters). Like FastCGI this is supported by flup.
- CGI is know to work on every major webserver out there but has the disadvantage of being slow. With Python 2.5 onwards you can use wsgiref as CGI gateway, in older Python versions you can either install wsgiref yourself or use the CGI wrapper code from the PEP 333.
The following gateways are either written especially for the Apache webserver or work best with it.
- The without a doubt best deployment platform for the apache webserver is mod_wsgi. Even though it's an apache module there are also ways to switch the underlaying interpreter into another user context. This allows you to mass host python applications like you would do with suexec/suphp.
For long time mod_python was the preferred way to deploy python applications on the apache webserver and frameworks like django still recommend this setup. However some bugs in mod_python make it hard to deploy WSGI applications, especially the undefined behavior of SCRIPT_NAME which makes routing hard.
If you want to use mod_python or have to use it, you can try the mod_python WSGI wrapper from this blog post about mod_python and wsgi.
- AJP is the Protocol used by Tomcat. flup provides ways to talk it.
Here a small example configuration for Apache and mod_wsgi:
from yourapplicaiton import YourWSGIApplication # mod_wsgi just wants an object called "application" it will use # for dispatching. pretty simple huh? application = YourWSGIApplication(configuration='probably', goes='here')
Save it as "yourapplication.wsgi" and then add this to your virtual host config:
WSGIScriptAlias / /path/to/yourapplication.wsgi/
(Note the trailing slash). Or if you want to have the application in a subfolder:
WSGIScriptAlias /foo /path/to/yourapplication.wsgi
(Without the trailing slash). Detailed usage examples for the WSGI gateways usually come with the gateways or can be found in their wikis.
Selecting the best gateway is mainly a matter of what you have available on your server and how your application is written. Some applications might not be thread safe, others work better with threads etc.
The following channels on freenode deal a lot with WSGI and you might be able to find some help there: