because 95% of the tables you need for this are relational (queue, completed_job, users, etc, etc) and for the few things that are unstructured, you can represent them well in PG with JSONB (which is what we do for inputs and outputs).