Multiprocess mode for migration gauges #498

When using the prometheus client in the multiprocess mode, by default, Gauges export a separate timeseries per process. Therefore, migrations Gauges are exported as follows:

django_migrations_unapplied_total{connection="default",pid="26847"} 0.0
django_migrations_unapplied_total{connection="default",pid="26846"} 0.0
django_migrations_unapplied_total{connection="default",pid="26845"} 0.0
django_migrations_unapplied_total{connection="default",pid="26844"} 0.0
django_migrations_unapplied_total{connection="default",pid="26843"} 0.0
django_migrations_unapplied_total{connection="default",pid="26842"} 0.0

In a typical Django deployment with multiple processes running in parallel and periodical restarts of individual processes on top of that (e.g. by limiting the number of requests before restarting the worker in gunicorn), this quickly leads to unmanageably large /metrics/ response.

The proper way seems to be calculating the maximum value across all processes, by utilizing the corresponding multiprocess_mode param as documented here.