What the exam asks
- Scale reads with read replicas or Aurora Replicas and reader endpoints, knowing that a Multi-AZ instance standby cannot serve reads.
- Choose Aurora over RDS for faster failover, up to 15 low-lag replicas, auto scaling replicas and Global Database.
- Use Aurora Serverless v2 for variable load and RDS Proxy for connection storms.
- Choose storage and instance class (gp3 or Provisioned IOPS, memory-optimised classes) from CloudWatch symptoms.
Core ideas
Four ways to add database instances, and what each one is for
| Feature | Replication | Readable? | Main purpose | Failover |
|---|---|---|---|---|
| RDS Multi-AZ DB instance | Synchronous to one standby in another AZ | No, the standby is passive | High availability | Automatic, typically 1–2 minutes. The endpoint’s DNS moves to the standby |
| RDS Multi-AZ DB cluster (MySQL, PostgreSQL) | Semi-synchronous to two readable standbys in 3 AZs | Yes, through the reader endpoint | HA plus some read scaling, lower write latency | Automatic, typically under a minute |
| RDS read replica | Asynchronous (replica lag possible) | Yes, through its own endpoint | Read scaling, reporting, cross-Region reads, DR | Manual promotion to a standalone database |
| Aurora Replica | Shares the cluster storage volume, lag usually well under 100 ms | Yes, through the reader endpoint | Read scaling and automatic failover target | Automatic promotion, usually much faster than RDS |
Rule of thumb: A read replica can itself be made Multi-AZ, and an RDS read replica can be in another Region.