Data tier resilience
The exam loves Multi-AZ vs read replica vs global trade-offs. Know sync vs async replication, failover time, and read scaling per service.
Amazon RDS
- Multi-AZ deployment — synchronous standby in another AZ; automatic failover to standby (same DNS endpoint); for HA, not read scaling.
- Read replicas — asynchronous copies for read scaling and DR; promote replica to standalone DB for recovery.
- Backup — automated backups to S3 (point-in-time recovery within retention); manual snapshots until deleted.
- Maintenance window — Multi-AZ failover during patch for minimal downtime.
Exam tip: Need more read capacity → read replicas. Need AZ failure protection → Multi-AZ.
Amazon Aurora
- Shared storage across AZs (6 copies, auto repair).
- Aurora Replicas — low-lag read endpoints; one can be failover target (tier 0/1 priority).
- Aurora Multi-AZ — writer + replicas across AZs; faster failover than standard RDS (often ~30 seconds exam ballpark).
- Aurora Global Database — one primary Region, up to 5 secondary read-only Regions; < 1 second cross-Region replication lag for DR; promote secondary for RTO.
- Aurora Serverless v2 — scales ACUs for variable workloads.
Amazon DynamoDB
- On-demand vs provisioned capacity; auto scaling on provisioned.
- Global tables — multi-Region, multi-active replication; last writer wins conflict resolution; for low-latency global reads/writes and Region-level DR.
- Point-in-time recovery (PITR) — continuous backup 35 days.
- On-demand backup — snapshots for long-term retention.
- DAX — in-memory cache for microsecond read latency (not a HA mechanism).
Streams + Lambda — react to changes for CQRS/event-driven patterns.
When to pick what
| Requirement | Often choose |
|---|---|
| Relational OLTP, HA in Region | RDS Multi-AZ or Aurora |
| Read-heavy relational | Aurora/RDS read replicas + connection pooling (RDS Proxy) |
| Massive scale key-value, ms latency | DynamoDB |
| Global active-active NoSQL | DynamoDB global tables |
| Session/cache offload | ElastiCache (Redis cluster mode for shard HA) |
RDS Proxy
Connection pooling for Lambda/serverless → RDS/Aurora; reduces connection storms and supports IAM auth and failover absorption.
Exam traps
- Read replica in another Region helps DR but replication is async — RPO > 0.
- Multi-AZ RDS standby is not readable (except Aurora replicas are).
- DynamoDB global tables need same table name and enable streams-based replication setup.
Official reference
SAA-C03 exam guide — fault-tolerant data stores.