A traditional RDS Multi-AZ DB instance fails over during an outage. Existing clients hold broken connections and cache the old address indefinitely. Which application behavior should be implemented?

1. Reconnect to the traditional standby as a second independent writable database before failover.
2. Assume synchronous replication guarantees no connection can ever break.
3. Hard-code the former primary private IP permanently.
4. Use bounded reconnection and retry behavior that honors endpoint DNS changes and transaction uncertainty.