SAA2 — Design Resilient Architectures

Why this module exists

Domain 2 is 26% of scored content — the second-largest domain. And it is the one where the exam's "choose the best answer" framing hurts most, because resilience is a spectrum with a price tag. Four options can all be more resilient than doing nothing, and only one matches the recovery objective and budget the stem actually stated.

The domain has only two task statements, and they are doing different jobs:

2.1 — Design scalable and loosely coupled architectures. This is about not failing: absorbing load, isolating components so one slow thing doesn't take down the next thing.
2.2 — Design highly available and/or fault-tolerant architectures. This is about failing well: surviving the loss of an instance, an Availability Zone, or a Region.

Most people study 2.2 and skim 2.1. That's backwards — 2.1 has sixteen "Knowledge of" items to 2.2's twelve, and decoupling questions are more common than DR questions.

The one idea to hold onto

Resilience is bought in units of "what can I lose?" — and every unit has a price.

   WHAT YOU CAN LOSE          WHAT IT COSTS             THE CONTROL
   ────────────────           ────────────              ───────────
   one instance          →    another instance     →    Auto Scaling + ELB health checks
   one Availability Zone →    capacity in 2+ AZs   →    multi-AZ subnets, Multi-AZ RDS
   one Region            →    a second footprint   →    backup/restore → pilot light →
                                                        warm standby → active/active
   one dependency        →    a queue and a retry  →    SQS, DLQ, idempotent consumers

The exam's job is to make you pick the cheapest row that meets the stated requirement. Reaching for multi-Region active/active when the stem says "development environment" is as wrong as reaching for a single instance when it says "must survive a Regional outage".

⚠️ The tell is always RTO and RPO. If the stem gives you numbers, they select the row. If it gives you "minimal downtime" with no number, the tie-break is operational overhead and cost — Domain 1's rule, applied here.

What you'll be able to do

Choose between SQS, SNS, EventBridge and Step Functions on the shape of the requirement, and say what each one actually guarantees.
Explain horizontal vs vertical scaling and why stateless workloads are a precondition for one of them.
Pick ALB, NLB or Gateway Load Balancer from the OSI layer the requirement implies.
Choose between ECS, EKS, Fargate and Lambda without hand-waving about "serverless".
Name the four DR strategies in cost order, state what distinguishes pilot light from warm standby, and map an RTO/RPO pair onto one of them.
Explain why AWS says to use data plane operations in a failover path, and name a control-plane dependency that would undermine a DR design.
Read a failing distributed system: X-Ray trace map, health checks, service quotas, connection exhaustion.

Lessons

#	Lesson	Task statement	Read	Listen
1	Decoupling — queues, topics, events and workflows	2.1	30 min	11 min
2	Scaling — horizontal, vertical, and why state is the constraint	2.1	26 min	9 min
3	Load balancing and multi-tier architecture	2.1	26 min	11 min
4	Containers and serverless — ECS, EKS, Fargate, Lambda	2.1	28 min	12 min
5	High availability — surviving an instance, an AZ, a dependency	2.2	30 min	11 min
6	Disaster recovery — the four strategies and the data plane rule	2.2	32 min	13 min

Then: Cheat sheet · Lab · Quiz · Interview · Runbook

Where this sits in the exam

From the SAA-C03 exam guide — Content Domain 2 (verified 2026-09-22), Design Resilient Architectures — 26% of scored content:

Task statement	This module
2.1 Design scalable and loosely coupled architectures	lessons 1–4
2.2 Design highly available and/or fault-tolerant architectures	lessons 5–6

A note on sources

This module cites the HTML exam guide at docs.aws.amazon.com/aws-certification/latest/solutions-architect-associate-03/, not the d1.awsstatic.com PDF. The HTML is maintained and is demonstrably ahead of that PDF — see ../PLAN.md for the diff.

Two places where a page I fetched did not contain something examinable are flagged inline with ⚠️ and the page to read. The largest is the ALB/NLB/Gateway Load Balancer feature comparison: the Elastic Load Balancing "What is" page links to per-type user guides rather than carrying a comparison table, so lesson 3 builds the comparison from the exam guide's own wording plus the per-type guides, and says so.

Facts verified 2026-09-22 against the pages cited in each lesson.

Keeps playing into the following modules — 208 min from here to the end of certification prep.