Avoid exports in AppConfig when running management commands

[olzhasar-reef]

Currently, migrations are being exported inside the ready() method of the django_prometheus.apps.AppConfig. This works fine for a default threaded prometheus mode, but leads to malformed metrics when using prometheus in multiprocess mode.

Consider the following typical deployment entrypoint:

# entrypoint.sh

./manage.py migrate  # - here metrics are exported for the pid
gunicorn some.app.wsgi - # metrics are exported again for different pid(s)

This leads to problems like this:

django_migrations_unapplied_total{connection="default",pid="16241"} 0.0                         
django_migrations_unapplied_total{connection="default",pid="16234"} 1.0                         
django_migrations_unapplied_total{connection="default",pid="16251"} 0.0                         
django_migrations_unapplied_total{connection="default",pid="16239"} 1.0                         
django_migrations_unapplied_total{connection="default",pid="16248"} 0.0                         
django_migrations_unapplied_total{connection="default",pid="16258"} 0.0                         
django_migrations_unapplied_total{connection="default",pid="16249"} 0.0                                                                                                                         
django_migrations_unapplied_total{connection="default",pid="16246"} 0.0  

This PR introduces a simple check to avoid exporting metrics in manage.py commands

[olzhasar-reef]

Is there anything that could be done to move this forward?

[olzhasar-reef]

Closing in favor of #499