www.awsgood.com
You won't be asked what Auto Scaling does. You'll be handed a paragraph that looks like a ticket: an order service on a single m5.xlarge, one RDS instance, 40k orders a day, an RTO of 15 minutes, and a CFO who wants the bill to stay flat. Then four answers where three are technically correct and one matches the constraint.
That's the shape of the whole exam guide, and it's why reading the domain list alone doesn't help much.
| Domain | Weight | Rough share of a 65-question exam |
|---|---|---|
| 1. Design Secure Architectures | 30% | ~19 |
| 2. Design Resilient Architectures | 26% | ~17 |
| 3. Design High-Performing Architectures | 24% | ~16 |
| 4. Design Cost-Optimized Architectures | 20% | ~13 |
Domain 2 splits into two task statements. The first is design scalable and loosely coupled architectures — SQS, SNS, EventBridge, Step Functions, API Gateway, container and instance scaling, Aurora replicas, DynamoDB capacity modes, ElastiCache. The second is design highly available and fault-tolerant architectures — Multi-AZ, Route 53 routing and health checks, backup and cross-region replication, and the RTO/RPO vocabulary.
Notice how much of that overlaps with Domain 3 and Domain 4. A queue-based decoupling answer is often also a cost answer. That overlap is where the "MOST cost-effective" and "LEAST operational overhead" qualifiers earn their keep.
Every disaster recovery question is really asking you to place the workload on this ladder and then justify it against budget. Memorize the ordering, not the marketing names.
| Strategy | Typical RTO | Typical RPO | Standing cost | Wording that points here |
|---|---|---|---|---|
| Backup and restore | 4–24 hours | Hours | Lowest — snapshots in S3 only | "cost is the primary concern", "can tolerate a day of lost data" |
| Pilot light | 30–120 minutes | Minutes | Low — data replicated, compute switched off | "core data must be replicated", "capacity launched only during a disaster" |
| Warm standby | 5–30 minutes | Seconds | Medium — scaled-down full stack always running | "must serve traffic within 15 minutes", "scaled-down copy in a second Region" |
| Multi-site active/active | Near zero | Near zero | Highest — full stack in two or more Regions | "no downtime", "users routed to both Regions simultaneously" |
The trap I see people fall into: a question says "pilot light" in the answer options but the requirement is "serving traffic within 10 minutes". Pilot light means the compute doesn't exist yet — you still need to launch ASGs, warm AMIs, and wait for health checks. Warm standby is the correct pick there.
Read the number. RTO in minutes with a scaled-down stack already running is warm standby, every time.
When the scenario says a consumer occasionally fails and messages vanish, the fix is a queue with a dead-letter queue, not a larger instance. Two parameters get tested repeatedly.
Visibility timeout. If your consumer takes 45 seconds and the queue's visibility timeout is the 30-second default, the message becomes visible again mid-processing and a second consumer picks it up. No error is thrown — you just get duplicate charges. Set the timeout to at least 6× your p99 processing time and add a DLQ with maxReceiveCount around 3 to 5.
Fan-out. SNS to multiple SQS queues when several consumers each need every message; a single shared queue when they're competing workers. EventBridge when the routing is content-based or you need SaaS and AWS service sources. Step Functions when the orchestration has branches, retries, and human approval steps — not for simple buffering.
# Confirm the redrive policy actually survived your last IaC deploy
aws sqs get-queue-attributes
--queue-url https://sqs.us-east-1.amazonaws.com/123456789012/orders
--attribute-names VisibilityTimeout RedrivePolicy
--output json
If `RedrivePolicy` comes back empty, your template dropped it and you're one bad deploy away from an unbounded retry loop.
## When Multi-AZ is the wrong answer
This is the section I wish I'd read before my first attempt. Multi-AZ is a single-Region high-availability control. It is not a DR strategy, and it does not solve most of what the exam calls "resilient".
- **Read scaling.** The Multi-AZ standby in RDS serves no reads. If the scenario is about read throughput, the answer is read replicas or Aurora replicas.
- **Regional failure.** Multi-AZ does not survive losing `us-east-1`. Cross-region read replicas, S3 Cross-Region Replication, or AWS Backup cross-region copies do.
- **Bad deploys and dropped tables.** Multi-AZ fails over to a standby that has the same broken schema. That's what point-in-time recovery and backup retention are for.
- **Tiny internal tools.** A staging dashboard with an RTO of 8 hours does not need a standby. Nightly snapshots plus an IaC rebuild is cheaper and the exam rewards you for saying so.
We got burned by the first one in a real migration: an analytics job was pointed at the primary endpoint and the team assumed Multi-AZ meant "it can't slow down production". It can, and did.
## The failover drill where the DNS TTL wasn't the problem
We ran a DR game day on 2026-09-14 in `us-east-1` and forced an RDS failover with `aws rds reboot-db-instance --db-instance-identifier orders-prod --force-failover`. The database was accepting connections again at 68 seconds by the RDS event timestamp. The application threw `SQLSTATE[HY000] [2002] Connection timed out` for the next 6 minutes and paged us.
My first configuration was wrong in a subtle way: I'd set retries at 3 attempts × 2 seconds and assumed the connection pool would recover on its own. It didn't. The pool was holding 50 dead sockets, the client library handed them out without validating, and each failed borrow burned the whole retry budget in 6 seconds before giving up.
Diagnosis order, which is the part worth copying: check the RDS event log first, then DNS, then the client.
```bash
aws rds describe-events --source-type db-instance --duration 60
--query 'Events[].{At:Date,Msg:Message}' --output table
dig +short orders-prod.abc123.us-east-1.rds.amazonaws.com
The event log showed the failover finished cleanly. `dig` returned the new private IP within seconds. So the database and DNS were fine — the stale sockets were ours. Fix: max connection lifetime of 60 seconds, TCP keepalive on, exponential backoff with jitter up to 30 seconds, and RDS Proxy in front of the app for the connection multiplexing. A second forced failover produced zero failed user requests and a p99 of 1.4 seconds.
For the exam, remember the managed version of that lesson: RDS Proxy exists to absorb failover churn and connection storms. When a scenario mentions "many short-lived connections" plus "frequent failovers", that's the answer.
## Nine days beats three months of scrolling
Budget 12–20 hours total if you already run AWS day to day. I'd spend roughly 4 of those on Domain 2 alone, because its concepts reappear inside the other three domains.
My split: read the official exam guide once and highlight every verb, then do one hands-on lab per DR rung in a throwaway account, then two full practice sets with review. Skip video courses past 1.5× speed — the exam is written in AWS's own phrasing, so docs and the sample questions beat lecture hours.
```bash
# Audit your own account against Domain 2 while you study
aws rds describe-db-instances
--query 'DBInstances[].{Id:DBInstanceIdentifier,MultiAZ:MultiAZ,AZ:AvailabilityZone,Standby:SecondaryAvailabilityZone}'
--output table
aws elbv2 describe-target-health --target-group-arn arn:aws:elasticloadbalancing:us-east-1:123456789012:targetgroup/app/abc123
--query 'TargetHealthDescriptions[].TargetHealth.State' --output text | sort | uniq -c
That second command printing `3 healthy` and `1 unhealthy` is a better HA lesson than any slide — someone is running a target group with a threshold so loose it never goes unhealthy, or so tight it flaps.
## Registration, cost, and which exam to take first
SAA-C03 costs USD 150 in the US (regional pricing differs), runs 130 minutes, and is delivered through Pearson VUE either at a test centre or online proctored. If English isn't your first language you can request extra time when you schedule. Results appear on screen at the end — no waiting period.
Take SAA-C03 before SAP-C02 unless you already own the associate-level badge and have 3+ years of multi-account work behind you. Doing the associate first isn't a formality; the resilience and cost domains are where the pro exam builds from.
You can reschedule free up to 48 hours before your slot through the certification portal, and a failed attempt needs a 14-calendar-day wait plus a fresh fee. Book three weeks out, not three months — a close date is the only study tool that reliably works.
## Exam technique: two minutes a question, then move
Sixty-five questions in 130 minutes is 2 minutes each, and Domain 2 scenarios eat closer to 3. Flag anything that takes more than 150 seconds and come back.
Read the final sentence of the stem first — that's the actual question, and AWS buries it under two sentences of context. Then eliminate answers that solve a different domain: a cheaper instance type in a resilience question, a Multi-AZ deployment in a cost question. Qualifiers decide the tiebreak, and "LEAST operational overhead" almost always means managed service over self-managed.
For multiple-response items, AWS doesn't publish partial-credit rules, so treat them as all-or-nothing and pick exactly the stated number of answers. Guessing a fifth option when it asks for two is a guaranteed miss.
If you're running these architectures for real and want a second pair of eyes on the DR plan before you book the exam, that's the kind of review the awsgood.com team does alongside managed operations work.
## Sources
- [AWS Certified Solutions Architect – Associate (official page and exam guide)](https://aws.amazon.com/certification/certified-solutions-architect-associate/)
- [Multi-AZ DB instance deployments for Amazon RDS](https://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/Concepts.MultiAZ.html)
- [Disaster recovery options in the cloud (AWS whitepaper)](https://docs.aws.amazon.com/whitepapers/latest/disaster-recovery-workloads-on-aws/disaster-recovery-options-in-the-cloud.html)
- [Amazon SQS visibility timeout and dead-letter queues](https://docs.aws.amazon.com/AWSSimpleQueueService/latest/SQSDeveloperGuide/sqs-visibility-timeout.html)
## FAQ: booking and retake questions
**Can I take SAA-C03 online from home?** Yes, through Pearson VUE's online proctored option. You'll need a webcam, a room with no other people, and a clean desk. The check-in process takes 15–20 minutes, so join early.
**What happens if I fail?** You wait 14 calendar days and pay the full fee again. Use the score report's domain breakdown — it shows which of the four domains dragged you under 720, and it's usually Domain 2 or 4 for hands-on engineers.
**Is the exam available in other languages?** Yes, SAA-C03 is offered in several languages including Japanese, Korean, and Mandarin. Extra time for non-native English speakers is requested at scheduling, not on exam day.