Our use-case for the read-replica is to keep user-facing requests fast by directing less important queries to the read replica. For the most part this means tasks like sending order confirmation emails should prefer to use the read only replica. These tasks generally happen as Celery tasks on dedicated worker processes.