High Availability and Disaster Recovery
Module 3 · Design Resilient Architectures

High Availability and Disaster Recovery

Design highly available and fault-tolerant architectures: Availability Zones and Regions, Multi-AZ patterns, database replication options, Route 53 routing and failover, backups, and the four disaster recovery strategies.

Availability Zones and Regions

A Region is a geographic area (such as us-east-1). Each Region has multiple Availability Zones (AZs): one or more data centers with independent power, cooling, and networking, connected by low-latency links.

  • High availability (HA) means the system keeps working when a component or an AZ fails. On AWS, that almost always means Multi-AZ.
  • Fault tolerance goes a step further: the failure causes no user-visible interruption, usually because redundant capacity is already running.
  • Disaster recovery (DR) is how you recover from a larger event, such as the loss of a whole Region, a bad deployment, or deleted data.

Read the question carefully. If it says “highly available,” Multi-AZ is usually enough. Multi-Region is the answer only when the question mentions Regional outages, global users, or a very low RTO/RPO across Regions.


The Multi-AZ Pattern

Route 53RegionApplication Load Balancer (both AZs)Availability Zone AAvailability Zone BPublic subnet: NAT gatewayPublic subnet: NAT gatewayEC2 × 2 (Auto Scaling group)EC2 × 2 (Auto Scaling group)RDS primaryreads + writesRDS standbyautomatic failover targetsyncreplication

The pattern: spread every tier across at least two AZs, size each AZ so the others can carry the load if one fails, and use services that are already Multi-AZ (S3, DynamoDB, SQS, Lambda, EFS) wherever you can.


Database Resilience Options

OptionWhat it doesReadable?Use it for
RDS Multi-AZ (instance)Synchronous standby in another AZ; automatic failover, usually in a minute or twoNo, standby onlyHigh availability
RDS Multi-AZ DB clusterOne writer and two readable standbys in three AZsYesHA plus read capacity, faster failover
RDS read replicasAsynchronous copies, in-Region or cross-Region; can be promotedYesScaling reads; cross-Region DR
AuroraStorage keeps six copies across three AZs; up to 15 low-lag replicas; fast failoverYesHA and read scaling with less to manage
Aurora Global DatabaseReplicates to other Regions, typically with under a second of lagYes, in secondary RegionsLow-RPO cross-Region DR and local global reads
DynamoDB global tablesMulti-Region, multi-active replicationYes, read and write in every RegionGlobal apps that write in many Regions

The two features people mix up most often, side by side:

Multi-AZ: availabilityAppPrimaryAZ AStandbyAZ BsyncStandby serves no trafficAutomatic failover, same endpointRead replicas: read scalingAppPrimarywritesReplicasame RegionReplicaother RegionasyncReplicas serve reads; they can lagPromotion is a manual stepreads + writeswritesreads

Remember the difference: Multi-AZ is for availability, and read replicas are for read scaling (plus cross-Region DR). Exam answers that use a read replica for automatic failover, or a classic Multi-AZ standby to serve reads, are wrong.


Route 53: Routing and Failover

Routing policyUse it when
SimpleOne resource, no health checks
FailoverActive-passive: send traffic to the secondary when the primary’s health check fails
WeightedSplit traffic by percentage (canary or blue/green releases)
LatencySend users to the Region with the lowest latency for them
GeolocationRoute by the user’s country or continent (compliance, localized content)
GeoproximityRoute by distance, with a bias to shift traffic between Regions
Multivalue answerReturn several healthy IPs for simple client-side load spreading
IP-basedRoute by the client’s IP range

Alias records point a domain name (including the zone apex, like example.com) at AWS resources such as an ALB, CloudFront distribution, or S3 website, at no query charge.


Backups and the Four DR Strategies

AWS Backup centralizes backup policies across EBS, RDS, Aurora, DynamoDB, EFS, FSx, S3, and more, including cross-Region and cross-account copies. Copies in a separate account protect you from someone deleting both the data and its backups.

DR strategies, from cheapest and slowest to most expensive and fastest:

StrategyWhat runs in the DR RegionRTO / RPO
Backup and restoreNothing but backupsHours
Pilot lightData is replicated and live; servers are off or minimalTens of minutes
Warm standbyA scaled-down but fully working copy of productionMinutes
Multi-site active/activeFull production in more than one Region, serving trafficNear zero

Choose the cheapest strategy that meets the RTO and RPO in the question.

⚠️ Common Pitfall: Backups You Have Never Restored

Common mistake: the design says “nightly snapshots” and stops there.

An untested backup is a hope, not a plan. Real DR plans include restore tests, documented runbooks, and the service quotas needed in the recovery Region. Exam questions also test the time dimension: a nightly backup can’t meet a 15-minute RPO.


Key Takeaway

Multi-AZ is the default for high availability; go multi-Region only when Regional failure or global users are part of the requirement. Know which database feature gives failover (Multi-AZ) and which gives read scaling (read replicas), how Route 53 failover works with health checks, and pick the cheapest DR strategy that meets the RTO and RPO.

🧪 Try It: Design for the RTO and RPO

For each scenario, pick a DR strategy and the main services. Then compare with the suggested answers.

  1. An internal reporting app. RTO 24 hours, RPO 24 hours.
  2. An e-commerce checkout. RTO 15 minutes, RPO 1 minute.
  3. A global trading platform. RTO and RPO near zero.

Suggested answers: (1) backup and restore with AWS Backup cross-Region copies; (2) warm standby with Aurora Global Database and a scaled-down ASG plus Route 53 failover; (3) multi-site active/active with DynamoDB global tables or Aurora Global Database, latency-based routing, and full capacity in each Region.

🧠 Knowledge Check

1. A company needs its RDS for MySQL database to fail over automatically if an AZ fails. What should they enable?

2. Users report slow reads from an RDS database during reporting hours. Writes are fine. What is the MOST appropriate fix?

3. Which Route 53 routing policy sends all traffic to a primary site and switches to a secondary when a health check fails?

4. A workload needs an RTO of a few minutes after a Regional outage and the company wants to keep costs down. Which DR strategy fits BEST?