Opscraft
AWS core services · Lesson 7 of 8 · 3 min read

Resilience, disaster recovery and cost

High availability

  • Spread across at least 2–3 AZs: ASGs/node groups, NAT per AZ, Multi-AZ databases, ALB in all AZs, PodTopologySpread in EKS.
  • Remove single points: one NAT gateway, single-AZ RDS, one EC2 instance with state on instance store.
  • Use managed services' built-in HA (S3, DynamoDB, Aurora) where possible.

Disaster recovery strategies

Strategy RTO / RPO Cost How
Backup & restore Hours / hours Lowest AWS Backup, snapshots copied cross-region, IaC to rebuild
Pilot light Tens of minutes Low Core data replicated (Aurora Global, S3 CRR); compute off, scaled up on failover
Warm standby Minutes Medium Scaled-down full stack running in DR region
Multi-site active/active Near zero Highest Both regions serve traffic (Route 53 latency/failover, DynamoDB global tables)

RTO = how long to recover; RPO = how much data you can lose. Test DR regularly (game days); IaC + GitOps make the rebuild repeatable.

Enlarge diagramBackup & restorelowest costPilot lightdata replicated, computeoffWarm standbysmall copy runningActive/activeboth regions serveRTO: hoursRTO: tens of minutesRTO: minutesRTO: near zerocheaper, slower ←→ faster, costlierRTO = how long you can be down. RPO = how much data you can lose.
Figure 1. The four DR strategies: the faster you want to recover, the more you keep running (and pay for) in the second region.
Swipe sideways to see all of it, or tap to open it full screen.

Cost optimisation

  • Visibility: Cost Explorer, Cost and Usage Report + Athena/QuickSight, cost allocation tags (team, service, env) enforced via tag policies/SCPs, Budgets with alerts, Cost Anomaly Detection.
  • Compute: rightsizing (Compute Optimizer), Savings Plans for baseline, Spot for stateless/CI (Karpenter consolidation), Graviton, schedule non-prod off hours.
  • Network (often the surprise): NAT gateway data processing (use VPC endpoints for S3/ECR/DynamoDB), cross-AZ traffic, data transfer out; keep chatty services in-AZ where safe.
  • Storage: S3 lifecycle and Intelligent-Tiering, delete unattached EBS volumes and old snapshots, gp2 → gp3 migration (cheaper, independent IOPS), ECR lifecycle policies, CloudWatch Logs retention (default is never expire!).
Enlarge diagramComputeNetworkStorageLogsrightsize, Savings Plans, Spot, Graviton, stop non-prod at nightNAT data charges → VPC endpoints; avoid cross-AZ chatterS3 lifecycle, delete unattached EBS and old snapshots, gp2 → gp3CloudWatch ingestion and retention
Figure 2. Where AWS bills usually hide, and the first fix for each.
Swipe sideways to see all of it, or tap to open it full screen.