AWS Certified Solutions Architect – Associate (SAA-C03)
The core AWS architecture exam. It tests designing secure, resilient, high-performing and cost-optimised solutions.
Verified against official sources on 2026-11-17.
AWS Certified Solutions Architect – Associate (SAA-C03) checks that you can design AWS solutions based on the Well-Architected Framework. You choose the right services for a set of requirements, and you improve existing designs. Every later exam in your roadmap assumes you know this material.
At a glance
| Code | SAA-C03 |
| Level | Associate |
| Intended audience | At least 1 year of hands-on experience designing cloud solutions on AWS |
| Questions | 65 (50 scored + 15 unscored) |
| Time | 130 minutes |
| Passing score | 720 (scaled 100–1000) |
| Price | USD 150 |
| Question types | Multiple choice, multiple response |
| Valid for | 3 years. Renewed by passing SAP-C03 |
Official domains and weightings
| Domain | Weight |
|---|---|
| 1. Design Secure Architectures | 30% |
| 2. Design Resilient Architectures | 26% |
| 3. Design High-Performing Architectures | 24% |
| 4. Design Cost-Optimized Architectures | 20% |
Key AWS services tested
- Networking: VPC (subnets, route tables, NAT gateways, security groups, NACLs), VPC endpoints and PrivateLink, Transit Gateway, Direct Connect, Site-to-Site VPN, Route 53, CloudFront, Global Accelerator, Elastic Load Balancing
- Security: IAM (policies, roles, cross-account access), IAM Identity Center, Organizations and SCPs, KMS, Secrets Manager, ACM, WAF, Shield, Cognito, GuardDuty, Macie
- Compute: EC2 and Auto Scaling, Lambda, ECS/Fargate, EKS, Elastic Beanstalk
- Storage: S3 (storage classes, lifecycle, replication, Object Lock), EBS, EFS, FSx (including FSx for Windows File Server), Storage Gateway, AWS Backup
- Databases: RDS (Multi-AZ, read replicas), Aurora (Global Database, Serverless), DynamoDB (global tables, DAX), ElastiCache, Redshift
- Integration: SQS (standard and FIFO), SNS, EventBridge, Step Functions, Kinesis, API Gateway
- Migration: DMS, Application Migration Service, DataSync, Snow Family
What you already know vs what you must learn
| You already know | You must learn |
|---|---|
| Layered application design, caching, queues (MSMQ, Service Bus, RabbitMQ) | VPC design: public, private and isolated subnets, NAT, endpoints, and how traffic actually flows |
| SQL Server high availability (Always On, failover clusters) | RDS Multi-AZ versus read replicas versus Aurora Global Database, and RTO and RPO for each |
| OAuth, JWT, TLS | IAM policy logic, roles instead of keys, KMS key policies |
| Azure App Service, Functions, Service Bus (if you've used Azure) | The AWS equivalents and how they differ (the SQS visibility timeout, SNS fan-out, EventBridge rules) |
| Cost of servers in a data centre | Usage-based cost thinking: S3 classes, data transfer charges, Savings Plans, Spot |
Preparation order and study plan (about 60–80 hours over 6–8 weeks)
| Week | Focus | Hours |
|---|---|---|
| 1 | Exam guide. Module 01 refresher. IAM in depth (Module 05) | 8–10 |
| 2–3 | Module 02 networking (build the secure VPC lab). Domain 1 | 15–20 |
| 4 | Module 06 databases and Module 08 messaging. Domain 2 | 10–12 |
| 5 | Module 12 HA and DR, plus compute options from Module 03. Domain 3 | 10–12 |
| 6 | Module 15 FinOps (S3 classes, purchase options). Domain 4 | 6–8 |
| 7–8 | Free Skill Builder Exam Prep and the 20-question set. This page's questions. Review weak domains. Mock exam (Phase 8) | 10–15 |
Mapped platform modules
- Module 02 — Networking: the most important module for Domain 1
- Module 05 — Security: IAM, KMS, WAF
- Module 03 — .NET on AWS: choosing between compute options
- Module 06 — Databases: RDS, Aurora, DynamoDB, ElastiCache
- Module 08 — Messaging & microservices: decoupling with SQS, SNS and EventBridge
- Module 12 — HA & DR: Multi-AZ, DR strategies, Auto Scaling
- Module 15 — Governance & FinOps: cost optimisation
Practical skills to build
- Build a 3-tier VPC: ALB in public subnets, Fargate tasks in private subnets, RDS in isolated subnets, with a gateway endpoint for S3
- Deploy RDS with Multi-AZ, force a failover, and watch the DNS endpoint switch
- Write an S3 lifecycle policy that moves statements to cheaper classes and expires them after the retention period
- Put SQS between an API and a worker, and scale the worker on queue depth
- Write a bucket policy that allows access only through a specific VPC endpoint
Exam-prep checklist
- I can draw traffic flows for public, private and isolated subnets, including the NAT gateway and endpoints
- I know when to use security groups and when to use NACLs (stateful versus stateless)
- I can choose between RDS Multi-AZ, read replicas, Aurora Global Database and DynamoDB global tables, given an RTO and RPO
- I know the 4 DR strategies (backup and restore, pilot light, warm standby, multi-site) and how cost and RTO trade off
- I know every S3 storage class, its retrieval time and its minimum storage duration
- I can choose between SQS standard, SQS FIFO, SNS, EventBridge and Kinesis
- I know which purchase option fits each workload pattern
- I can explain when to use ALB, NLB and Gateway Load Balancer
- I scored well above 720-equivalent on the free practice set and a mock exam
Official links
Practice questions
These 5 questions are original, banking-flavoured scenarios. They aren't real exam questions.
A payments API runs on ECS Fargate tasks in private subnets with no internet access. The tasks need to read and write payment files in an S3 bucket in the same Region. The security team requires that this traffic doesn't go over the internet, and the finance team wants the lowest cost. What should the architect do?
- A. Add a NAT gateway and route S3 traffic through it
- B. Create an S3 gateway VPC endpoint, add it to the private subnets' route tables, and restrict the bucket policy with the
aws:SourceVpcecondition - C. Create an S3 interface VPC endpoint in each Availability Zone
- D. Move the tasks to public subnets with public IP addresses and restrict access with security groups
Correct answer: B
Explanation: A gateway endpoint for S3 adds a route (using a prefix list) so that S3 traffic stays on the AWS network. Gateway endpoints for S3 and DynamoDB have no hourly or data-processing charge. Restricting the bucket policy to that endpoint (aws:SourceVpce) means the bucket can't be reached from anywhere else. This is a defence-in-depth control that bank security reviewers look for.
Why the others are wrong:
- A: A NAT gateway sends traffic to S3's public endpoints, and you pay hourly and per-GB processing charges. It works, but it's more expensive and less private.
- C: Interface endpoints for S3 are private, but they have hourly and per-GB charges. Use them when you need access from on-premises or from another Region, not as the cheapest same-Region option.
- D: Putting tasks in public subnets increases exposure and goes against the "private by default" principle.
A bank's customer statements service uses Amazon RDS for SQL Server. If the primary database host or its Availability Zone fails, the service must recover automatically within minutes and without losing committed transactions. What meets this requirement?
- A. Create a read replica in another AZ and promote it manually during an outage
- B. Enable a Multi-AZ deployment
- C. Take automated backups and restore to a new instance in another AZ
- D. Move to a larger instance class with Provisioned IOPS storage
Correct answer: B
Explanation: A Multi-AZ deployment keeps a standby in another AZ that is updated with synchronous replication. If a failure happens, RDS fails over automatically and moves the DNS endpoint to the standby, usually within minutes, with no committed data lost. Your .NET connection strings don't change. Make sure your app has retry logic, for example EF Core's EnableRetryOnFailure.
Why the others are wrong:
- A: Read replicas use asynchronous replication, so recent transactions can be lost. Promotion is a manual step, and replicas are meant for read scaling.
- C: Restoring from a backup takes longer and loses data written since the last backup (point-in-time recovery has a gap of a few minutes).
- D: A bigger instance improves performance, not availability. It's still a single point of failure.
Card transactions arrive from an API and must be processed by a downstream ledger service. Transactions for the same account must be processed in order and must not be processed twice. Transactions for different accounts can be processed in parallel. What design fits best?
- A. An SQS standard queue with several consumers
- B. An SQS FIFO queue with
MessageGroupIdset to the account ID and content-based or explicit deduplication - C. An SNS standard topic with an HTTPS subscription to the ledger service
- D. Synchronous REST calls from the API to the ledger service, with retries
Correct answer: B
Explanation: FIFO queues keep messages in strict order within a message group and remove duplicates sent within the 5-minute deduplication window. Using the account ID as the group ID gives you ordering per account, while different accounts are still processed in parallel across groups. The consumer should still be idempotent, for example with a DynamoDB idempotency record, because processing failures can cause messages to be redelivered.
Why the others are wrong:
- A: Standard queues deliver at least once with best-effort ordering, so duplicates and out-of-order messages can occur.
- C: SNS standard topics don't guarantee ordering or deduplication, and they don't buffer messages if the subscriber is down for long.
- D: Synchronous calls tie the API's availability to the ledger's and give no buffering. Retries without idempotency can create duplicates.
Monthly PDF statements are read often in the first 30 days, then less than once a quarter. When a customer does ask for an old statement, it must load in milliseconds. Statements must be kept for 7 years and then deleted. What's the most cost-effective approach?
- A. Keep everything in S3 Standard
- B. Use a lifecycle rule to move objects to S3 Glacier Instant Retrieval after 30 days and expire them after 7 years
- C. Use a lifecycle rule to move objects to S3 Glacier Deep Archive after 30 days
- D. Store the statements on Amazon EFS with lifecycle management
Correct answer: B
Explanation: Glacier Instant Retrieval is designed for long-lived data that's read about once a quarter but still needs millisecond access. Its storage cost is lower than S3 Standard-IA's, in exchange for higher retrieval charges, which suits rare reads. A lifecycle expiration action deletes objects after 7 years. If the retention rule says the records can't be changed during that period, add S3 Object Lock. The bank's records team must confirm the actual retention period.
Why the others are wrong:
- A: S3 Standard is the most expensive class for data that's rarely read.
- C: Deep Archive retrieval takes hours, which fails the millisecond requirement.
- D: EFS is a shared file system for compute. Per GB it costs much more than S3 for archive storage.
Standard-IA would also meet the millisecond requirement, but for data read less than once a quarter, Glacier Instant Retrieval costs less to store.
An internet-facing Application Load Balancer fronts a .NET loan-application API. Penetration testers found it accepts SQL-injection payloads in query strings, and one IP address sent 50,000 requests in 5 minutes. Which two actions address both findings at the edge?
- A. Attach an AWS WAF web ACL with an AWS Managed Rules SQL-injection rule group
- B. Add a WAF rate-based rule that blocks IPs exceeding a request threshold
- C. Add a deny rule for the attacking IP to the ALB's security group
- D. Add NACL rules that block SQL keywords
- E. Rely on AWS Shield Standard
Correct answers: A and B
Explanation: AWS WAF inspects HTTP requests (layer 7). Managed rule groups detect common SQL-injection patterns, and rate-based rules automatically block sources that go over a threshold. Fix the code too: always use parameterised queries in EF Core or Dapper. WAF is an extra layer, not a replacement for secure code.
Why the others are wrong:
- C: Security groups support allow rules only, and blocking one IP doesn't stop the next one.
- D: NACLs filter on IP address, port and protocol. They can't inspect HTTP content.
- E: Shield Standard protects against network- and transport-layer DDoS attacks. It doesn't inspect requests for SQL injection or apply rate limits per IP.
Independent learning resource — not affiliated with, endorsed or sponsored by Amazon Web Services. AWS and related marks are trademarks of Amazon.com, Inc. or its affiliates. All banking data in examples is fictional. Sample architectures are for learning and are not automatically compliant with GDPR, PCI DSS or any banking regulation.
AWS Certified Cloud Practitioner (CLF-C02)
The foundational AWS exam. It covers cloud concepts, security, core services and billing, with no coding.
AWS Certified Developer – Associate (DVA-C03)
The developer exam. It tests building, securing, testing, deploying and debugging applications on AWS, now including AI-assisted development.