A business-critical relational application needs automatic recovery from an AZ failure. Read scaling is not currently required. Which baseline design best meets this requirement?

1. Use only a read replica and assume it always supplies automatic Multi-AZ failover.
2. Use one Single-AZ database with a larger instance class.
3. Use only daily manual exports and a DNS health check.
4. Use an appropriate RDS Multi-AZ deployment and test application reconnect behavior.