RDS & Aurora

RDS & Aurora Quick Notes

Amazon RDS

• Engines: MySQL, MariaDB, PostgreSQL, Oracle, SQL Server, …
• You manage: Less OS/patching than EC2 self-managed; still pick size, storage, Multi-AZ.
• Storage autoscaling: Grow disk with least ops when workload is expanding.

Multi-AZ vs Read Replicas

Multi-AZ, Read replica

Replication: Synchronous, Asynchronous

Purpose: HA / failover, Read scale

Scope: ≥2 AZs, same Region, Same AZ, cross-AZ, or cross-Region

Readable?: Standby is not a read endpoint, Yes

• Failover DNS: Same endpoint/CNAME — AWS retargets it to the standby.
• Cross-Region replica charges: Data transfer applies across Regions (same-Region replica replication framed as no transfer charge on exams).
• Encryption inheritance: Encrypted primary → replicas are encrypted (cannot mix).

Encrypt an Existing Unencrypted DB

1. Snapshot
2. Copy snapshot with encryption
3. Restore from encrypted snapshot
4. Cut over (cannot flip encryption on in place)

Share Encrypted DB with Another Account

• Create encrypted snapshot → share snapshot → grant access to the KMS key.

RDS Custom

• Managed Multi-AZ-style HA plus DBA access to customize DB and OS.
• Fit: Legacy Oracle needing OS/engine customizations standard RDS forbids.

Amazon Aurora

• Shared storage layer replicated across AZs; compute nodes attach to it.
• Endpoints: Cluster (writer), reader, instance.
• Aurora Replicas: Scale reads + failover targets; use Aurora Auto Scaling when lag spikes.
• Serverless: Capacity in ACUs — spiky / unpredictable workloads.
• I/O-Optimized: Better price/performance when I/O charges dominate.
• Global Database: Cross-Region primary + secondaries — RPO ~seconds, RTO ~1 minute. Can convert an existing cluster by adding a secondary Region.

Aurora Failover Priority

1. Lowest tier number (0–15) wins
2. If tied → largest instance size wins

Example: tier-1 (16 TB) vs tier-1 (32 TB) → promote the 32 TB replica.

In Transit

• Enable SSL/TLS to the DB (separate from storage encryption).
• RDS Proxy: Connection pooling for Lambda / bursty apps (avoids connection storms).

Self-Managed Escape Hatch

• Need massive PostgreSQL with full OS control → EC2 + io1/io2 EBS (not RDS).