Disaster Recovery

RPO and RTO, the four DR strategies from backup-and-restore to multi-site active-active, testing DR, and data-level protection against corruption and ransomware.

What it is

Disaster recovery (DR) is the ability to restore service and data after a major disruption: region outage, data corruption, ransomware, accidental deletion or a catastrophic bad deployment. It is part of business continuity.

  • RPO (Recovery Point Objective): the maximum acceptable data loss, measured in time.
  • RTO (Recovery Time Objective): the maximum acceptable time to restore service.

The four strategies

Strategy Description RPO RTO Cost
Backup and restore Backups copied to another region; rebuild infra with IaC when needed Hours Hours to a day Lowest
Pilot light Core data replicated continuously; minimal compute off or scaled to zero Minutes Tens of minutes to hours Low
Warm standby Scaled-down but running copy in another region Seconds to minutes Minutes Medium
Multi-site active-active Full capacity in several regions serving traffic Near zero Near zero (seconds) Highest

These four strategies are the AWS Well-Architected reliability guidance categories; Azure and Google Cloud describe equivalent patterns.

flowchart LR
  subgraph Primary region
    APP1[App tier] --> DB1[(Primary DB)]
    S3a[(Object storage)]
  end
  subgraph DR region: warm standby
    APP2[App tier<br/>scaled down] -.-> DB2[(Replica DB)]
    S3b[(Replicated storage)]
  end
  DB1 -- async replication --> DB2
  S3a -- cross-region replication --> S3b
  DNS[Global DNS / LB<br/>health-based failover] --> APP1
  DNS -. on failover .-> APP2
  BK[(Immutable backups<br/>separate account)] 
  DB1 -- snapshots --> BK

Corruption and ransomware are different from outages

Replication copies corruption instantly. You also need point-in-time recovery, immutable backups (object lock/WORM, vaulted backups in a separate account or project with separate credentials), and tested restore procedures. Many organizations follow a 3-2-1(-1-0) style rule: three copies, two media, one offsite, one immutable/offline, zero restore errors verified.

Real-world example

A retailer sets tiers: checkout and payments (RPO ≈ seconds, RTO 15 minutes) run warm standby in a second region with database replication and pre-provisioned capacity at 30%; the product catalog (RPO 1 hour, RTO 4 hours) uses pilot light; internal reporting (RPO 24 hours, RTO 2 days) uses backup and restore. Failover is exercised twice a year with a game day.

Advantages and disadvantages

More aggressive DR Less aggressive DR
Lower business impact from disasters Lower cost
Often required by regulators for critical services Simpler operations
Higher cost, complexity, more consistency issues Longer outages, more data loss

Technology options

AWS Backup, Elastic Disaster Recovery, Aurora Global Database, DynamoDB global tables, S3 replication and Object Lock; Azure Site Recovery, Azure Backup, geo-replication for Azure SQL/Cosmos DB; Google Cloud Backup and DR, cross-region Cloud SQL replicas, Spanner multi-region; Velero for Kubernetes; Terraform/OpenTofu for rebuilding infrastructure.

Architect-level considerations

[!ARCHITECT] Derive RPO/RTO per capability from business impact, then choose the cheapest strategy that meets them. Include dependencies: a warm-standby app is useless if its identity provider, secrets or third-party integrations are single-region.

[!EA] Require a DR tier, a runbook and a test date for every Tier 0/1 system in the architecture review.

[!CTO] An untested DR plan is a hope, not a plan. Fund regular game days and measure actual RPO/RTO achieved.

Interview questions and answers

What is the difference between high availability and disaster recovery?

HA handles component and AZ failures automatically within a region with little or no data loss. DR handles larger disasters (region loss, corruption, ransomware) and may involve declared failover, some data loss, and restoration from backups.

How do you test DR without impacting customers?

Regular restore tests into isolated environments, automated backup verification, scheduled failover exercises during low traffic, partial failovers (one cell or a percentage of traffic), and chaos experiments on dependencies.

What does RPO = 0 across regions imply?

Synchronous cross-region commits (consensus-based multi-region databases like Spanner or Aurora DSQL, or synchronous replication), with higher write latency, and careful handling of what happens when the other region is unreachable.