What the exam covers
Domain 1Design Secure Architectures30%
- 1.1Secure access to AWS resources
- 1.2Secure workloads and applications
- 1.3Data security controls
Domain 2Design Resilient Architectures26%
- 2.1Scalable and loosely coupled architectures
- 2.2Highly available and fault-tolerant architectures
Domain 3Design High-Performing Architectures24%
- 3.1High-performing storage
- 3.2High-performing and elastic compute
- 3.3High-performing databases
- 3.4High-performing networks
- 3.5Data ingestion and transformation
Domain 4Design Cost-Optimized Architectures20%
- 4.1Cost-optimized storage
- 4.2Cost-optimized compute
- 4.3Cost-optimized databases
- 4.4Cost-optimized networks
Sample questions
A web application on Amazon EC2 fetches images from URLs that users supply. A penetration test shows that an attacker can make the application request the instance metadata endpoint and read the temporary credentials of the instance's IAM role, a server-side request forgery attack. The application still needs its IAM role. Which change BEST reduces this risk?
- Move the instances to a private subnet behind a NAT gateway.
- Require Instance Metadata Service Version 2 (IMDSv2) on the instances so that metadata requests need a session token.
- Replace the IAM role with IAM user access keys stored in a file on the instances.
- Add a network ACL rule that denies inbound traffic from the internet on port 80.
Answer
B. IMDSv2 requires a session token obtained with a PUT request and a custom header, which typical server-side request forgery flaws cannot produce, so the role credentials can no longer be read that way. Moving the instances to a private subnet does not help because the metadata endpoint is reached from inside the instance itself. Long-term access keys on disk would be a larger risk than the role.
An IAM role has an identity-based policy that allows s3:* and dynamodb:*. The role also has a permissions boundary that allows s3:* and sqs:*. No other policies apply. Which actions can the role perform? (Choose two.)
- sqs:SendMessage
- s3:GetObject
- dynamodb:PutItem
- s3:PutObject
- iam:CreateRole
Answer
B, D. A permissions boundary limits the role to the intersection of its identity-based policy and the boundary, which here is only S3. DynamoDB is allowed by the identity policy but not the boundary, and SQS is allowed by the boundary but never granted by the identity policy.
A company connects its data center to AWS with a single AWS Direct Connect connection. The company wants a backup path so that connectivity continues if the Direct Connect connection fails. The backup only needs to handle reduced traffic for short periods. Which solution is MOST cost-effective?
- Order a second dedicated Direct Connect connection at a different Direct Connect location
- Create VPC peering between the VPC and the data center as a backup
- Configure an AWS Site-to-Site VPN connection to the same virtual private gateway as a backup path
- Deploy AWS Global Accelerator in front of the Direct Connect connection
Answer
C. A Site-to-Site VPN over the internet is a low-cost backup for Direct Connect, and BGP can prefer the Direct Connect path while it is up. A second Direct Connect connection gives more resilience but costs much more than needed for occasional reduced traffic. VPC peering cannot connect to on-premises networks, and Global Accelerator does not back up Direct Connect.
A company uses a customer managed AWS KMS key in Account A to encrypt data in an Amazon S3 bucket. An application running under an IAM role in Account B must decrypt objects from the bucket. Which steps are required to allow the application to use the key? (Choose two.)
- Attach an IAM policy to the role in Account B that allows kms:Decrypt on the key ARN
- Enable automatic key rotation on the key in Account A
- Create a KMS key in Account B with the same key ID
- Update the key policy in Account A to allow Account B to use the key for decryption
- Create an alias for the key in Account B that points to the key in Account A
Answer
D, A. Cross-account use of a KMS key requires both the key policy in the owning account to trust the other account and an IAM policy in the other account that grants the principal the needed KMS actions. Aliases cannot point to keys in other accounts. Rotation does not affect permissions, and a key in Account B has different key material and cannot decrypt the data.
A company has 50 developers in one AWS account who need identical permissions. The permissions change often as new services are adopted, and developers join and leave the team every month. Which approach requires the LEAST operational overhead?
- Create an IAM group, attach a customer managed policy to it, and add or remove developers from the group.
- Create a group for each developer and nest those groups inside a parent group that has the policy.
- Create an IAM role whose trust policy names the developers' group as the principal.
- Attach an inline policy to each developer's IAM user and update every user when permissions change.
Answer
A. A group with a managed policy lets administrators change permissions in one place and manage membership simply by adding or removing users. Per-user inline policies multiply the work, IAM does not support nested groups, and a group cannot be a principal in a trust policy.
A company needs an encrypted connection between its on-premises data center and a VPC within a few days. Traffic volumes are moderate, and the company already has a reliable internet connection. What should a solutions architect recommend?
- Create a VPC peering connection between the VPC and the on-premises network
- Create an AWS Site-to-Site VPN connection between a virtual private gateway and the on-premises customer gateway device
- Order an AWS Direct Connect dedicated connection without any additional encryption
- Expose the VPC resources through an internet-facing Network Load Balancer with TLS listeners
Answer
B. Site-to-Site VPN uses IPsec tunnels over the existing internet connection and can be set up quickly. A Direct Connect connection usually takes weeks to provision and is not encrypted by default. VPC peering only connects VPCs, and a public NLB exposes the resources to the internet.
A partner portal uses AWS Transfer Family to receive PGP-encrypted files into Amazon S3. Every uploaded file must be decrypted, tagged, and copied to a processing folder, and the company wants these steps to run automatically after each upload using the managed service's built-in features. Which solution meets these requirements?
- Configure the S3 bucket with a lifecycle rule that moves files to the processing folder.
- Enable S3 Versioning on the bucket and process each new version manually.
- Run a cron job on an EC2 instance that lists the bucket every few minutes.
- Attach a Transfer Family managed workflow to the server with steps that run after each file upload completes.
Answer
D. Transfer Family managed workflows run post-upload processing steps such as copy, tag, decrypt, and custom Lambda steps automatically for each file. Lifecycle rules only transition or expire objects and cannot move them to another folder or process them.
A startup is launching a mobile app backend with a REST API. Traffic is unpredictable, ranging from near zero overnight to sudden spikes after marketing campaigns. The team has no operations staff and wants to pay nothing for idle servers. Which architecture is MOST suitable?
- Amazon API Gateway with AWS Lambda functions and an Amazon DynamoDB table in on-demand capacity mode.
- An Application Load Balancer with an Auto Scaling group of EC2 instances and Amazon RDS Multi-AZ.
- Amazon EC2 instances in a cluster placement group with an Amazon Aurora provisioned cluster.
- Amazon ECS on EC2 instances with a fixed capacity of four container instances and Amazon RDS.
Answer
A. API Gateway, Lambda, and DynamoDB on-demand are serverless, scale automatically with demand, and charge per request, so there are no idle servers to pay for or manage. EC2-based designs keep at least some instances running and require patching and capacity management.
A company records all customer support calls and stores the audio files in Amazon S3. The company wants to know automatically whether customers were satisfied or frustrated on each call so that supervisors can follow up. The company has no machine learning expertise and does not want to train or host models. Which combination of AWS services should a solutions architect use? (Choose two.)
- Amazon Polly to convert the call recordings to text.
- Amazon Rekognition to detect the sentiment of the audio files.
- Amazon Textract to extract the customer's words from the audio files.
- Amazon Comprehend to detect the sentiment of the resulting text.
- Amazon Transcribe to convert the call recordings to text.
Answer
E, D. Amazon Transcribe is the managed speech-to-text service, and Amazon Comprehend analyzes text for sentiment, so together they produce a sentiment result for each call with no models to train. Amazon Polly works in the opposite direction, turning text into speech, and Rekognition and Textract analyze images and documents rather than audio.
A video platform needs to pass job requests between services. Each request includes a video file that can be several hundred megabytes, which is far larger than an Amazon SQS message can hold. The company wants to keep using SQS for decoupling. What should a solutions architect recommend?
- Use an Amazon SNS topic instead, because SNS messages have no size limit.
- Compress each file with gzip and send it as the SQS message body.
- Store each file in Amazon S3 and send an SQS message that contains a reference to the S3 object.
- Split each file into many SQS messages and reassemble them in the consumer.
Answer
C. The claim-check pattern, which the Amazon SQS Extended Client Library implements, stores the large payload in S3 and passes a pointer in the message. Splitting files across many messages is complex and fragile, and SNS also has a message size limit.
A real-time chat application must push new messages from the server to connected browser clients instantly, and clients must also send messages to the server over the same connection. The company wants a managed, serverless API front end. Which solution meets these requirements?
- Use an Amazon SQS queue that browser clients poll directly for new messages.
- Use an Amazon API Gateway WebSocket API with AWS Lambda integrations for the connect, disconnect, and message routes.
- Use an Amazon API Gateway REST API and have clients poll it every second for new messages.
- Use an Amazon API Gateway HTTP API with response caching enabled.
Answer
B. WebSocket APIs keep a persistent two-way connection so the backend can push messages to clients and clients can send messages over the same connection. Polling a REST API or SQS adds latency and cost and is not a true server push.
Amazon EC2 instances in a VPC must resolve hostnames in the corp.example.internal domain, which is hosted on DNS servers in the company's data center. The data center is connected to the VPC through AWS Direct Connect, and the on-premises DNS records change frequently. What should a solutions architect do?
- Change the VPC's DHCP options set to use AmazonProvidedDNS only.
- Create a private hosted zone for corp.example.internal and copy the on-premises records into it.
- Create a Route 53 Resolver inbound endpoint in the VPC and point the on-premises DNS servers at it.
- Create a Route 53 Resolver outbound endpoint and a forwarding rule for corp.example.internal that targets the on-premises DNS servers, and associate the rule with the VPC.
Answer
D. An outbound endpoint with a forwarding rule sends queries for the on-premises domain from the VPC to the data center's DNS servers, so answers are always current. An inbound endpoint handles the opposite direction, letting on-premises systems resolve AWS names. Copying records into a private hosted zone would go stale as records change, and AmazonProvidedDNS alone cannot answer for the on-premises domain.
A company's application stores data in an Amazon DynamoDB table in us-east-1 and serves users in North America, Europe and Asia. Users in Europe and Asia experience high latency for both reads and writes. The company wants each user to read and write data in a nearby AWS Region, with changes replicated to all Regions automatically. What should a solutions architect do?
- Deploy a DynamoDB Accelerator (DAX) cluster in us-east-1 in front of the table.
- Put an Amazon CloudFront distribution in front of the application and cache all API responses.
- Configure DynamoDB global tables with replicas in Regions near each user group, and deploy the application in those Regions.
- Export the table to Amazon S3 every day and import it into tables in the other Regions.
Answer
C. Global tables are multi-active: each Region has a full replica that accepts reads and writes, and DynamoDB replicates changes between Regions automatically. DAX in us-east-1 would not reduce network latency for distant users and doesn't help writes. Daily exports create stale, separate copies, and CloudFront caching doesn't help writes.
A company runs a packaged application on a Microsoft SQL Server database. The vendor requires administrator access to the database host's operating system to install its monitoring agent and custom drivers and to change instance-level SQL Server settings. The company wants to move the database to AWS and offload backups and routine infrastructure management as far as possible while meeting the vendor's requirements. Which solution should a solutions architect choose?
- Amazon RDS for SQL Server
- Amazon Aurora PostgreSQL-Compatible Edition
- Amazon DynamoDB
- Amazon RDS Custom for SQL Server
Answer
D. Amazon RDS Custom automates backups and infrastructure tasks while giving privileged access to the underlying operating system and database environment, which the vendor's agent, drivers and settings require. Standard Amazon RDS for SQL Server does not allow operating system access, and Aurora PostgreSQL or DynamoDB would mean changing the database engine of a packaged application. Note: the Oracle variant, RDS Custom for Oracle, reaches end of support on March 31, 2027, so it is not a choice for new work.
A company's Auto Scaling group should add one instance when average CPU is between 60% and 75%, three instances when it is between 75% and 90%, and five instances when it is above 90%. Which type of scaling policy is designed for this behavior?
- A scheduled scaling action for each CPU range
- A predictive scaling policy based on CPU utilization
- A target tracking scaling policy with a CPU target of 60%
- A step scaling policy with step adjustments based on the size of the alarm breach
Answer
D. Step scaling policies let you define different capacity adjustments for different ranges of an alarm breach, which matches the requirement exactly. Target tracking decides the adjustment itself to hold a target value and does not let you specify fixed step sizes, and scheduled or predictive scaling are based on time or forecasts rather than current breach size.
A property management company is installing thousands of security cameras across its buildings. The company needs to ingest the live video from every camera securely into AWS, store it durably with time-based indexing, and make it available for live and recorded playback and for video analytics applications. The company does not want to manage any ingestion servers. Which AWS service should a solutions architect use?
- Amazon Data Firehose
- Amazon Kinesis Video Streams
- Amazon Kinesis Data Streams
- AWS DataSync
Answer
B. Amazon Kinesis Video Streams is built to ingest live video from many devices, store and index it by time, and serve it for playback and analytics without servers to manage. Kinesis Data Streams and Amazon Data Firehose handle small data records such as logs and events and do not provide video storage, indexing or playback.
A company has many AWS accounts and wants to be alerted automatically when spending for any service rises unusually, for example because of a misconfigured job. The company does not want to set and maintain fixed spending thresholds for each service. Which solution meets these requirements?
- Set up AWS Cost Anomaly Detection with monitors and alert subscriptions.
- Use Cost Explorer to review daily costs manually for each account.
- Create a separate AWS Budgets budget with a fixed threshold for every service.
- Create Amazon CloudWatch alarms on CPU utilization for all instances.
Answer
A. Cost Anomaly Detection uses machine learning to learn normal spending patterns and alerts on unusual increases without fixed thresholds. Budgets would work only with thresholds that someone must set and keep updated, which the company wants to avoid.
A media team generates lower-resolution video renditions, each several hundred megabytes, from original footage that is stored safely elsewhere. The renditions are read only a few times a month but must be available in milliseconds when requested, and any lost rendition can be regenerated in minutes. Which S3 storage class is the MOST cost-effective for the renditions?
- S3 Standard-Infrequent Access
- S3 Standard
- S3 Glacier Flexible Retrieval
- S3 One Zone-Infrequent Access
Answer
D. One Zone-IA costs less than Standard-IA because it stores data in a single Availability Zone, which is acceptable for data that can be recreated, and it still returns objects in milliseconds. Standard-IA would also work but pays extra for multi-AZ resilience the team does not need. Glacier Flexible Retrieval cannot serve objects in milliseconds.
A company is creating an Amazon RDS for PostgreSQL DB instance for a new product. Data growth is hard to predict. In the past the team allocated very large volumes to be safe and paid for storage that stayed empty for years, and the team knows that allocated RDS storage cannot be reduced later. The company wants to pay only for the storage it needs without risking running out of space. What should a solutions architect do?
- Store the database files on an instance store volume to avoid storage charges.
- Allocate storage for the current need and enable RDS storage autoscaling with a maximum storage threshold.
- Allocate the largest storage size that the instance class supports when the DB instance is created.
- Allocate a small volume and create a read replica each time the volume is nearly full.
Answer
B. RDS storage autoscaling increases the allocated storage automatically when free space runs low, up to a maximum that the company sets, so the database can start small and grow with the data. Allocating the maximum size up front repeats the over-provisioning problem, and read replicas copy the data rather than adding space to the primary.
A company is reviewing its Amazon EBS spending. It finds that many volumes are older gp2 volumes and that dozens of volumes are not attached to any instance because their instances were terminated. Which actions will reduce EBS costs? (Choose two.)
- Change the gp2 volumes to gp3 by using Elastic Volumes.
- Change the unattached volumes to io2 so that they can be attached faster later.
- Enable fast snapshot restore on all existing snapshots.
- Enable EBS encryption by default for the account.
- Take a final snapshot of each unattached volume and then delete the volume.
Answer
A, E. gp3 offers a lower price per GiB than gp2 with configurable performance, and unattached volumes are billed for their full provisioned size even though nothing uses them, so a final snapshot preserves the data at lower cost. Encryption by default does not change price, io2 costs more, and fast snapshot restore adds an hourly charge.