Backup & Restore
RTO: hours–days. RPO: minutes–hours. Cost: 💲. Snapshots, S3, CRR to a “cold” Region. Rebuild on demand.
SAA-C03 · High Availability
Exam domain: Design Resilient Architectures (26%)
The exam is obsessed with the AZ ≠ Region distinction. Internalize:
- AZ — physically isolated data-center cluster inside a Region. ms latency to peers.
- Region — geographic area with ≥ 3 AZs. Hundreds of ms across Regions.
Four DR strategies (from cheap to hot)
Section titled “Four DR strategies (from cheap to hot)”Pilot Light
RTO: 10s of minutes. RPO: minutes. Cost: 💲💲. Core (DB replicas, minimal EC2) always running in DR Region. Scale up on failover.
Warm Standby
RTO: minutes. RPO: seconds. Cost: 💲💲💲. Scaled-down but functional copy of prod. Scale out on failover.
Multi-site active/active
RTO: seconds. RPO: near-zero. Cost: 💲💲💲💲. Both regions serve traffic. Requires app-level conflict handling.
Auto Scaling Group anatomy
Section titled “Auto Scaling Group anatomy”flowchart LR cw[CloudWatch alarm] -->|scale-out| asg[Auto Scaling Group] asg --> lt[Launch template] lt --> ec2a[EC2 in AZ-a] lt --> ec2b[EC2 in AZ-b] ec2a & ec2b --> tg[Target Group] tg --- alb[Application Load Balancer]
Scaling policies
Section titled “Scaling policies”- Target tracking — “keep CPU at 50%”. The default answer.
- Step scaling — thresholds → step sizes. Use when metric jumps in bursts.
- Scheduled — known traffic patterns (business hours).
- Predictive — ML-driven forecast. Combine with target tracking, not alone.
Health checks
Section titled “Health checks”Global services vs regional services
Section titled “Global services vs regional services”| Service | Scope | Notes |
|---|---|---|
| Route 53 | Global | DNS + health checks + weighted/latency routing |
| CloudFront | Global | Edge cache in 400+ PoPs |
| IAM | Global | One user, all Regions |
| S3 | Regional | Bucket lives in one Region |
| DynamoDB | Regional | Global Tables = multi-Region active-active |
| RDS | Regional | Cross-region read replicas for DR |
| Aurora | Regional | Aurora Global Database for < 1s RPO |
Route 53 routing policies
Section titled “Route 53 routing policies”Pick one
- Simple — one record, one answer.
- Weighted — split traffic by percentage (canary, A/B).
- Latency — send users to the Region with lowest RTT.
- Failover — active/passive with health check.
- Geolocation — route by user’s country.
- Multi-value — up to 8 healthy records, poor-man’s LB.