That is a workaround and precisely the point the author makes. It increases operational complexity and creates a divide between records in the vector DB and the relational DB.