Each request makes a new connection to the underlying PostgreSQL database.  We
should be using some sort of persistent connection pooling.