New Relic is an excellent web application performance reporting tool. After a great experience using it on Heroku, I went to enable it for an existing self-hosted application. While Django 1.4 has an excellent WSGI configuration build-in, Django 1.3 makes it a little trickier. To complicate matters even more, I was using debian's gunicorn package, which not not make changing the Django executable easy.

New Relic has excellent documentation for downloading, installing and configuring their tools. Here are my crib notes:

view plainprint?

    pip install newrelic django-newrelic-extensions  
    # sign up on newrelic.com, get $LICENSE_KEY under "Account Settings"  
    mkdir /etc/newrelic  
    newrelic-admin generate-config $LICENSE_KEY /etc/newrelic/newrelic.ini  
    newrelic-admin validate-config /etc/newrelic/newrelic.ini 

You should now be able to log in to new relic and see an application called "Python Agent Test". This means that the test worked; new relic can communicate with the hosted servers. You can run gunicorn manually from the command line with new relic as follows:

view plainprint?

    cd /opt/my-django-project  
    export NEW_RELIC_CONFIG_FILE=/etc/newrelic/newrelic.ini  
    newrelic-admin run-program python manage.py run_gunicorn  

Notice that new relic replaces the python executable with its own command, "newrelic-admin run-program". I'm using Ubuntu, which uses the Debian gunicorn package, which stores its configuration in /etc/gunicorn.d/django. Here is my current config file:

view plainprint?

    #/etc/gunicorn.d/django  
    CONFIG = {  
        'mode': 'django',  
        'working_dir': '/opt/my-django-project',  
        'user': 'django',  
        'group': 'django',  
      
        # stupid hack to get new relic to load in Django 1.3  
        # once we move to 1.4 we should use straight WSGI  
        'python': '/opt/my-django-project/conf/newrelic.sh',  
        'environment': {  
            'NEW_RELIC_CONFIG_FILE': '/etc/newrelic/newrelic.ini',  
            },  
      
        'args': (  
            '--bind=0.0.0.0:80',  
            '--workers=4',   
            '--backlog=0',  
            '--worker-class=sync',  # gevent doesn't work smoothly with new relic yet  
            '--preload',  
        ),  
    }  

Notice the hack to inject the new relic runner script. I have over-ridden the python executable path to point to a third script, newrelic.sh.

view plainprint?

    #!/usr/bin/env bash  
    /usr/local/bin/newrelic-admin run-program python $*