PyMongo and mod_wsgi

To run your application under mod_wsgi, follow these guidelines:

  • Run mod_wsgi in daemon mode with the WSGIDaemonProcess directive.

  • Assign each application to a separate daemon with WSGIProcessGroup.

  • Use WSGIApplicationGroup %{GLOBAL} to ensure your application is running in the daemon’s main Python interpreter, not a sub interpreter.

For example, this mod_wsgi configuration ensures an application runs in the main interpreter:

<VirtualHost *>
    WSGIDaemonProcess my_process
    WSGIScriptAlias /my_app /path/to/app.wsgi
    WSGIProcessGroup my_process
    WSGIApplicationGroup %{GLOBAL}
</VirtualHost>

If you have multiple applications that use PyMongo, put each in a separate daemon, still in the global application group:

<VirtualHost *>
    WSGIDaemonProcess my_process
    WSGIScriptAlias /my_app /path/to/app.wsgi
    <Location /my_app>
        WSGIProcessGroup my_process
    </Location>

    WSGIDaemonProcess my_other_process
    WSGIScriptAlias /my_other_app /path/to/other_app.wsgi
    <Location /my_other_app>
        WSGIProcessGroup my_other_process
    </Location>

    WSGIApplicationGroup %{GLOBAL}
</VirtualHost>

Background: mod_wsgi can run in “embedded” mode when only WSGIScriptAlias is set, or “daemon” mode with WSGIDaemonProcess. In daemon mode, mod_wsgi can run your application in the Python main interpreter, or in sub interpreters. The correct way to run a PyMongo application is in daemon mode, using the main interpreter.

Python C extensions in general have issues running in multiple Python sub interpreters. These difficulties are explained in the documentation for Py_NewInterpreter and in the Multiple Python Sub Interpreters section of the mod_wsgi documentation.

Beginning with PyMongo 2.7, the C extension for BSON detects when it is running in a sub interpreter and activates a workaround, which adds a small cost to BSON decoding. To avoid this cost, use WSGIApplicationGroup %{GLOBAL} to ensure your application runs in the main interpreter.

Since your program runs in the main interpreter it should not share its process with any other applications, lest they interfere with each other’s state. Each application should have its own daemon process, as shown in the example above.