This free AWS Solutions Architect Associate practice exam has 19 scenario questions written to the current SAA-C03 blueprint, covering all four design domains, and every answer explains why each option is right or wrong. Choose your answers, then open “Show answer” to check. The real SAA-C03 has 65 questions in 130 minutes, and you need a scaled score of 720 to pass (facts verified November 2026 against the AWS exam page).
Want the full blueprint first? Read our SAA-C03 exam guide. Newer to AWS? Start with the free Cloud Practitioner practice exam.
Table of Contents
Domain 1: Design Secure Architectures (questions 1–6)
Design Secure Architectures is 30% of the scored exam.
1. A company wants to link its office network to a VPC. All traffic on the link must be encrypted, and access between the two sides must be limited to what is actually needed. Which solution meets both requirements?
- An AWS transit gateway attached to the VPC with open routing
- An AWS Site-to-Site VPN, with routes for the on-premises ranges and with security groups and network ACLs that allow only the required traffic
- An AWS Direct Connect connection with route tables that allow or deny traffic
- An IAM policy that allows console access only from office IP addresses
Show answer
Answer: B. The question has two requirements, so the answer must meet both. An AWS Site-to-Site VPN uses IPsec tunnels, so traffic between the office and the VPC is encrypted. Security groups and network ACLs then restrict which addresses and ports can talk to each other, which covers “only what is needed”.
Why not the others: A transit gateway connects networks but adds no encryption, and “open routing” is the opposite of limiting access. Direct Connect is a private link, but it isn’t encrypted by default, and route tables decide where packets go rather than filtering them. IAM policies control who can call AWS APIs, not which network traffic crosses a link.
Domain: Design Secure Architectures
2. EC2 instances in a company’s VPC read and write confidential objects in one S3 bucket. The security team wants the bucket to reject every request that does not come from that VPC. How should this be done?
- Create a VPC endpoint for S3 and add a bucket policy that denies requests not arriving through that endpoint.
- Attach an IAM policy to the instance role that allows access to the bucket.
- Encrypt the objects with SSE and keep the key on the instances.
- Encrypt the objects with a customer managed KMS key before upload.
Show answer
Answer: A. A VPC endpoint gives the instances a private path to S3. A bucket policy with an explicit deny on requests whose aws:SourceVpce (or aws:SourceVpc) doesn’t match then blocks everyone else, even users with valid credentials. Because it’s a resource policy on the bucket, it applies to every caller.
Why not the others: An IAM allow on one role grants that role access but doesn’t stop other identities from reaching the bucket from elsewhere. Both encryption options protect the content of the objects, but neither controls where requests come from, which is what the question asks.
Domain: Design Secure Architectures
3. Traffic to a public website has grown sharply. The company wants to understand where requests come from and what they look like, and at the same time make the site more secure. Which solution does both?
- Deploy AWS WAF with logging enabled and deliver the logs through Amazon Data Firehose to S3 for analysis.
- Deploy a Network Load Balancer with logging.
- Deploy Amazon API Gateway with logging to Kinesis Data Streams.
- Enable detailed monitoring on the web servers.
Show answer
Answer: A. AWS WAF blocks common web attacks with managed or custom rules, which covers “more secure”. With WAF logging turned on, each inspected request is recorded with details such as the source IP, country, headers, URI and which rule matched. Sending the logs to S3 through Firehose makes them easy to analyse, which covers “understand where requests come from”.
Why not the others: A Network Load Balancer works at layer 4, so it can’t inspect or filter web requests. API Gateway fronts APIs, not a whole public website, and adding it doesn’t make the site more secure on its own. Detailed monitoring gives one-minute instance metrics such as CPU, which show load but nothing about who is sending requests.
Domain: Design Secure Architectures
4. A website is served through Amazon CloudFront and must use TLS. The company wants certificates to be issued and renewed automatically with no manual steps. What is the most operationally efficient setup?
- Request the certificate in AWS Certificate Manager with email validation.
- Create the certificate with a CloudFront security policy.
- Create the certificate with origin access control.
- Request the certificate in AWS Certificate Manager with DNS validation.
Show answer
Answer: D. AWS Certificate Manager issues public certificates for CloudFront at no extra cost. With DNS validation, you add a CNAME record once. As long as that record stays in place, ACM renews the certificate automatically.
Why not the others: Email validation works, but someone has to act on the validation emails, which isn’t “no manual steps”. A CloudFront security policy only chooses the TLS protocol versions and ciphers. Origin access control restricts access to an S3 origin. Neither of those creates certificates.
Domain: Design Secure Architectures
5. Applications on EC2 must sign in to an Amazon RDS for PostgreSQL database without any stored long-term password. Which approach meets this?
- Create an IAM user per application and keep its access keys in environment variables.
- Give the instances an IAM role and have the applications request short-lived tokens through RDS IAM database authentication.
- Keep the password in a configuration file encrypted with a key stored next to it.
- Use a static database password and restrict access by the instances’ IP addresses.
Show answer
Answer: B. With IAM database authentication, the application uses its instance role’s temporary credentials to generate an authentication token, which is valid for 15 minutes. The token replaces the password, so there’s nothing long-lived to store or leak.
Why not the others: IAM user access keys are long-term credentials, wherever you keep them. An encrypted file with its key stored beside it is barely protected, and it still holds a long-term password. A static password limited by IP address is still a stored long-term password; network rules don’t change that.
Domain: Design Secure Architectures
6. Several business units share one S3 bucket, each under its own prefix, and access is enforced by a single bucket policy. Many more prefixes are coming, and that one policy is becoming unmanageable. What should the company add to keep prefix-level isolation at scale?
- An S3 access point for each business unit, with its own policy limited to that unit’s prefix
- SSE-S3 default encryption on the bucket
- Presigned URLs for all access
- A bucket policy statement that denies GetObject to everyone
Show answer
Answer: A. S3 access points are named entry points to a bucket, each with its own policy. Giving every business unit its own access point, scoped to its prefix, splits one growing bucket policy into many small, independent ones.
Why not the others: Default encryption protects data at rest but doesn’t separate who can read which prefix. Presigned URLs give temporary access to individual objects; they’re for sharing, not for managing ongoing team permissions. Denying GetObject to everyone would break legitimate access for all units.
Domain: Design Secure Architectures
Domain 2: Design Resilient Architectures (questions 7–11)
Design Resilient Architectures is 26% of the scored exam.
7. A game stores player data in a relational database in one Region. Offices on other continents run analytics on that data and want fast local reads, and the company also wants cross-Region disaster recovery at reasonable cost. Which database solution fits?
- An Amazon Aurora global database with the primary cluster in the application’s Region and secondary clusters near the offices
- An Amazon ElastiCache for Redis cluster with replicas near the offices
- An Amazon RDS Multi-AZ DB cluster in the application’s Region
- DynamoDB global tables in every Region
Show answer
Answer: A. An Aurora global database copies data from the primary Region to secondary Regions, typically with under a second of lag. The offices read from a nearby secondary cluster, and if the primary Region fails, a secondary can be promoted to take over writes. One design solves both requirements.
Why not the others: ElastiCache is a cache, not the system of record, and it doesn’t provide relational disaster recovery. A Multi-AZ DB cluster protects against an Availability Zone failure but stays in one Region. DynamoDB global tables are multi-Region, but DynamoDB isn’t relational, and the data already lives in a relational database.
More on this choice: RDS vs Aurora.
Domain: Design Resilient Architectures
8. After an employee accidentally changed rows in an Amazon RDS for MySQL table, the company wants to be able to restore the database to any moment in the past 30 days, losing at most 5 minutes of data. Which RDS feature provides this?
- Automated backups with a 30-day retention period
- Multi-AZ deployment
- Read replicas
- Manual snapshots
Show answer
Answer: A. RDS automated backups combine a daily snapshot with transaction logs that are uploaded every 5 minutes. That lets you do a point-in-time restore to any second within the retention period, which can be set up to 35 days. A 30-day retention period meets the requirement.
Why not the others: Multi-AZ standbys and read replicas copy every change, including the mistake, so they can’t take you back in time. Manual snapshots only restore to the moment each snapshot was taken, so you could lose far more than 5 minutes of data.
Domain: Design Resilient Architectures
9. A Lambda function writes messages to an SQS standard queue, and the consumer occasionally receives the same message twice. Duplicates must never reach the consumer. What is the smallest architectural change that achieves this?
- Enable long polling on the standard queue.
- Replace SQS with an Amazon MQ broker.
- Replace the queue with an SQS FIFO queue that has content-based deduplication enabled.
- Enable content-based deduplication on the standard queue.
Show answer
Answer: C. Standard queues deliver at least once, so occasional duplicates are expected. FIFO queues add deduplication and exactly-once processing. You can’t convert an existing standard queue, so you create a new FIFO queue and point the function at it. That’s still a small change compared with switching messaging systems.
Why not the others: Long polling reduces empty responses and cost; it does nothing about duplicates. Content-based deduplication is a FIFO-only setting, so it can’t be turned on for a standard queue. Moving to Amazon MQ is a much bigger change than the question allows.
Domain: Design Resilient Architectures
10. An application backed by one DynamoDB table is expanding to a second Region. Users in each Region need fast local reads and writes against the same data. What is the MOST operationally efficient way to provide this?
- Create a second table and replicate with a custom application built on the Kinesis Client Library.
- Create a second table and copy changes with a Lambda function triggered by DynamoDB Streams.
- Add a global secondary index and point the second Region’s application at it.
- Convert the existing table into a global table by adding a replica in the second Region.
Show answer
Answer: D. DynamoDB global tables replicate a table across Regions, and every replica accepts reads and writes. Adding a replica Region to the existing table is a managed operation, so there’s no replication code to build or run.
Why not the others: Both custom replication designs, with the Kinesis Client Library or with Streams and Lambda, can work, but you have to build, monitor and handle conflicts yourself. That’s the opposite of “most operationally efficient”. A global secondary index is just another way to query the same table in the same Region; despite the name, it isn’t a cross-Region copy.
Domain: Design Resilient Architectures
11. Containerized video processing microservices must run across several Availability Zones and restart automatically when they fail. Which platform involves the LEAST operational overhead?
- Amazon EKS with self-managed nodes
- Amazon ECS with the EC2 launch type
- Containers started by hand on EC2 instances
- Amazon ECS with the Fargate launch type
Show answer
Answer: D. An ECS service keeps the desired number of tasks running and spreads them across Availability Zones, replacing any that fail. With Fargate, there are also no instances to patch, scale or replace, which makes this the lowest-overhead option.
Why not the others: EKS with self-managed nodes and ECS on EC2 both meet the availability requirement, but you manage the underlying servers. Containers started by hand have no orchestrator, so nothing restarts them when they fail.
More on this choice: ECS vs EKS.
Domain: Design Resilient Architectures
Domain 3: Design High-Performing Architectures (questions 12–15)
Design High-Performing Architectures is 24% of the scored exam.
12. On the last Monday of each month, thousands of users run reports on an Amazon Redshift cluster and queries queue up. The company wants faster queries during that surge without taking the cluster offline. What should it enable?
- Concurrency scaling on the relevant workload management queues
- Redshift Spectrum before each surge
- An elastic resize before each surge
- A classic resize before each surge
Show answer
Answer: A. With concurrency scaling, Redshift automatically adds temporary cluster capacity when queries start to queue, and removes it when the surge is over. Users keep querying throughout, so there’s no downtime.
Why not the others: Both resize options change the cluster itself. An elastic resize makes the cluster briefly unavailable, and a classic resize keeps it read-only for much longer, which conflicts with “without taking the cluster offline”. Redshift Spectrum lets you query data in S3; it doesn’t add capacity for queries against the cluster.
Domain: Design High-Performing Architectures
13. Monitoring stations on several continents each produce about 500 GB of data a day and have fast internet connections. All of it should arrive in one S3 bucket as quickly as possible with minimal operational complexity. Which approach is best?
- Enable S3 Transfer Acceleration on the bucket and upload directly with multipart uploads.
- Upload to a bucket in the nearest Region and replicate to the central bucket, then clean up.
- Ship a Snowball Edge device from every site every day.
- Upload to EC2 instances in the nearest Region and copy EBS snapshots across Regions.
Show answer
Answer: A. S3 Transfer Acceleration sends each upload to the nearest edge location, then over the AWS network to the bucket’s Region, which is faster than the public internet over long distances. Multipart upload sends large files in parallel parts. It’s one bucket and one setting, with no extra infrastructure.
Why not the others: Regional buckets plus replication would work, but they add buckets, replication rules and cleanup, which isn’t minimal complexity. Shipping Snowball devices every day is slow and labour-intensive when the sites already have fast connections. Staging data on EC2 and copying snapshots adds servers and steps for no gain.
Domain: Design High-Performing Architectures
14. A serverless REST API will serve customers all over the world, and nobody knows yet how much traffic to expect. Latency should be low wherever the caller is. Which design is best?
- An edge-optimized API Gateway REST API with Lambda functions tuned through memory settings and provisioned concurrency
- A Regional API Gateway REST API with Lambda reserved concurrency
- Network Load Balancers in each Region invoking Lambda functions
- A Regional API Gateway REST API with an HTTP integration
Show answer
Answer: A. An edge-optimized endpoint routes callers through the nearest CloudFront edge location, which cuts latency for a global audience. Lambda scales with unknown traffic. Raising a function’s memory also gives it more CPU, and provisioned concurrency keeps functions warm so callers avoid cold starts.
Why not the others: A Regional endpoint serves everyone from one Region, so distant callers see more latency. Reserved concurrency caps how far a function can scale; it doesn’t make it faster. A Network Load Balancer can’t invoke Lambda functions (an Application Load Balancer can). An HTTP integration isn’t serverless, and it doesn’t address global latency.
Domain: Design High-Performing Architectures
15. A ticketing application’s DynamoDB table has fixed provisioned capacity. Traffic is normally light but occasionally spikes far above that capacity, and during spikes requests are throttled and users see errors. What should a solutions architect change?
- Add a read replica to the table.
- Switch the table to on-demand capacity mode.
- Change the application to strongly consistent reads.
- Buy DynamoDB reserved capacity.
Show answer
Answer: B. In on-demand capacity mode, DynamoDB serves whatever request rate arrives, and you pay per request. That fits light traffic with sudden, unpredictable spikes. Fixed provisioned capacity is the cause of the throttling.
Why not the others: DynamoDB doesn’t have read replicas the way RDS does. Strongly consistent reads use twice the read capacity of eventually consistent reads, so they’d make throttling worse. Reserved capacity is a discount on provisioned throughput; it doesn’t raise the limit.
Domain: Design High-Performing Architectures
Domain 4: Design Cost-Optimized Architectures (questions 16–19)
Design Cost-Optimized Architectures is 20% of the scored exam.
16. An on-premises NFS file store is to be backed up to Amazon S3 through an NFS interface. Backup files should be archived 5 days after upload, and waiting a few days to retrieve an archived file is acceptable. What is the MOST cost-effective design?
- An S3 File Gateway with a lifecycle rule that moves files to S3 Standard-IA after 5 days
- A Tape Gateway with a lifecycle rule to S3 Standard-IA
- A Volume Gateway with a lifecycle rule to S3 Glacier Deep Archive
- An S3 File Gateway backed by a bucket with a lifecycle rule that moves files to S3 Glacier Deep Archive after 5 days
Show answer
Answer: D. An S3 File Gateway presents an NFS share on premises and stores each file as an S3 object. A lifecycle rule can then move those objects to S3 Glacier Deep Archive, the lowest-cost storage class. Retrieval can take up to 48 hours, which the company has said is acceptable.
Why not the others: Standard-IA keeps files instantly available, but it costs much more to store than Deep Archive, and fast retrieval isn’t needed here. Tape Gateway presents virtual tapes for backup software, and Volume Gateway presents block storage volumes. Neither offers the NFS file interface the question requires.
Domain: Design Cost-Optimized Architectures
17. Web servers on EC2 are busy at two predictable times each day and nearly idle otherwise. The company wants fault tolerance at the lowest cost. What should it do?
- Buy Reserved Instances for peak capacity.
- Stop instances with a cron job when demand is low.
- Resize the instances larger at peak with a script.
- Run the servers in an Auto Scaling group that scales with demand.
Show answer
Answer: D. An EC2 Auto Scaling group adds instances for the two peaks and removes them afterwards, so you only pay for capacity while it’s needed. Because the peaks are predictable, scheduled scaling fits well. The group also replaces unhealthy instances and spreads them across Availability Zones, which gives fault tolerance.
Why not the others: Reserving peak capacity means paying for servers that sit idle most of the day. A cron job that stops instances saves money but doesn’t replace failed servers, so there’s no fault tolerance. Resizing to a larger instance type requires stopping the instance, which interrupts service at the worst moment.
Domain: Design Cost-Optimized Architectures
18. A company wants to be told when its actual AWS spending, or the spending forecast for the month, goes above a set amount. It does not want to build anything custom. What should it use?
- AWS Trusted Advisor with CloudWatch data exported for machine learning
- A monthly review of the bill by the finance team
- A budget in AWS Budgets with alerts on actual and forecasted cost sent to an SNS topic
- AWS Cost Explorer with a Lambda function that estimates costs
Show answer
Answer: C. AWS Budgets lets you set a cost threshold and alert on actual spend, forecasted spend, or both. Alerts go by email or to an SNS topic, with no code.
Why not the others: Trusted Advisor with exported data and Cost Explorer with a Lambda function both need custom work, which the company has ruled out. Cost Explorer can forecast costs, but it isn’t the service that sends threshold alerts. A monthly manual review is too late and too slow to warn anyone in time.
Domain: Design Cost-Optimized Architectures
19. A finance app keeps user details in DynamoDB. When a user asks for their data to be removed, it must be gone within one month. Which approach has the LEAST operational overhead?
- Write expiring items to a DynamoDB stream and delete them with a Lambda function.
- Use AWS Config to detect TTL expiry and invoke a Lambda function.
- Enable DynamoDB TTL on an expiry attribute, and have a Lambda function set that attribute on the user’s items when a deletion request arrives.
- Run a Lambda function from an EventBridge rule that fires when TTL expires.
Show answer
Answer: C. DynamoDB Time to Live (TTL) deletes items automatically after the timestamp in a chosen attribute has passed, typically within a few days, and TTL deletes don’t consume write capacity. All the application has to do is stamp the user’s items with an expiry time when the request arrives, which comfortably meets the one-month deadline.
Why not the others: Deleting items through a stream and a Lambda function rebuilds what TTL already does for free. AWS Config tracks resource configuration, not individual table items. EventBridge has no event that fires when a TTL timestamp passes, so that rule would never run.
Domain: Design Cost-Optimized Architectures
Your score by domain
Count your correct answers in each domain.
| Domain | Questions | Your score |
|---|---|---|
| Design Secure Architectures | 1–6 | __ / 6 |
| Design Resilient Architectures | 7–11 | __ / 5 |
| Design High-Performing Architectures | 12–15 | __ / 4 |
| Design Cost-Optimized Architectures | 16–19 | __ / 4 |
A short set shows where you’re weak; it can’t predict your real exam result. For every miss, find the word in the question that should have pointed you to the right answer, such as “least operational overhead”, “most cost-effective” or “without downtime”. SAA-C03 is mostly about reading those constraints.
Ready for a full, timed exam?
The real SAA-C03 gives you 130 minutes for 65 scenario questions. Our SAA-C03 practice exams match that format, with every option explained.
Free mock exam
Take SAA-C03 Mock Exam 1, free
A full-length, timed mock in the real exam format, with every answer explained so you know exactly where you stand.
65
questions
130
minutes
720
to pass
Free account required.
If you score 80% or higher on every mock exam in the course, sit the real exam within 30 days of buying and don’t pass, we’ll refund the course in full. Terms apply.
Weak on a topic? Read these next
- SAA-C03 exam guide: domains, scoring and a 6-week study plan.
- RDS vs Aurora: relational databases and the exam cues for each.
- ECS vs EKS: containers, Kubernetes and Fargate.
AWS Solutions Architect Associate practice exam FAQ
Are these real SAA-C03 exam questions?
No. AWS doesn’t publish live exam questions, and sharing them breaks the AWS Certification agreement. These are written to the official SAA-C03 exam guide to test the same skills.
How many questions are on the real SAA-C03 exam?
65 in 130 minutes. 50 are scored and 15 are unscored, and you can’t tell which is which.
What is the SAA-C03 passing score?
720 on a scaled score from 100 to 1000. Scoring is compensatory, so there’s no pass mark per domain.
Is this practice exam free?
Yes, with no sign-up. Mock Exam 1 in our course is also free; it needs a free account.
AWSBoy is not affiliated with Amazon Web Services. AWS and related marks are trademarks of Amazon.com, Inc. or its affiliates.
Sources (checked November 2026): AWS Certified Solutions Architect – Associate exam page and SAA-C03 exam guide · AWS service documentation linked in each answer.




