Some datanommer queries are just too long

[abompard]

Some datanommer queries are too long to be practical, for example querying the number of messages with topic org.fedoraproject.prod.buildsys.build.state.change for user foobar in the entire datanommer DB takes about 2 minutes. It's unrealistic to run this on each incoming buildsys.state.change message. As a result, fedbadges is unable to cope with the amount of messages coming in.

I think fedbadges (or maybe datanommer.models) should develop a form of cached values. Fedbadges could cache the number of messages with a given topic and a given username, and this would go a looong way to make it faster. It would increase this number when a message comes in, and optionally rebuild the cache periodically. This cache could be stored in a Redis or memcached instance (in the same Openshift project or shared).