Consider switching to using init containers for tile generation setup
Tile generation containers currently wait for their first new image and based on the metadata for this image parse out all missing information from the database and shapefiles. This can take a long time. It is also only ever done once. This seems like a reasonable use case for an init container: https://kubernetes.io/docs/concepts/workloads/pods/init-containers/
The problem is that currently the rabbitmq listening processes are able to buffer incoming rabbitmq events. So even though the initialization is taking a long time, we're still holding on to events that will be processed. Maybe the real problem here isn't init versus not, but just speeding up this initialization in general.