hmm... yeah... if the RDS instance is configured to reject unrecognized hosts and there's some lag between a new SpiceDB node being scheduled and the list of RDS-allowed hosts being updated, i could see that being the result