SAP-C03 · Sample questions
10 real questions from the CertifyDeck SAP-C03 bank, spread across the domains of the official AWS Certified Solutions Architect - Professional skills outline, with the correct answer and a full explanation of why the other options are wrong. They are original questions written from the public syllabus, not exam dumps, and they are the same style you will find in the practice tests and mock exams.
Correct answer: A. Opt in to AWS Compute Optimizer from the management account for all member accounts, and install the CloudWatch agent on the instances to publish memory utilization.
When the management account opts in, Compute Optimizer shows findings and recommendations across the organization's accounts, and it analyzes EC2 memory utilization only when the instances publish it, for example with the CloudWatch agent. Enhanced infrastructure metrics extend the lookback period but do not create a memory metric that the instances do not publish. Per-account opt-in fails the single-place review, and a dashboard with manual decisions is the self-built analysis the team wants to avoid.
Correct answer: C. Write the prompts to JSONL files in S3, submit Amazon Bedrock batch inference jobs that read them, and use EventBridge job state change events to start the analytics load.
Batch inference takes prompts from files in S3, generates the responses asynchronously into S3 with a single job request, can notify job state changes through EventBridge, and AWS offers select models in batch mode at a lower price than on-demand inference. The Priority tier charges a premium for faster responses that nobody needs here, and a Glue job calling Converse is the self-built throttling and retry fleet the company rejects. Batch inference is not supported for provisioned models, so that combination is not possible.
Correct answer: C. Attach a resource-based policy to the central bus that allows events:PutEvents for the organization ID, and deploy to every account a rule that targets the central bus ARN with an IAM role.
An event bus policy can grant access to a whole organization by its organization ID, so new accounts can send events without any change to the central bus, and each sender rule targets the central bus ARN with an IAM role, which EventBridge requires for new cross-account bus targets and which keeps SCP boundaries in effect. Listing account IDs works but requires a central policy change for every new account. An SCP never grants permissions, so the bus would still reject the events, and per-account Lambda functions writing to Kinesis add custom code to operate.
Correct answer: B. Host the tenant models on a SageMaker AI multi-model endpoint with auto scaling, store the model artifacts in Amazon S3, and invoke it through a SageMaker AI Runtime interface VPC endpoint.
A multi-model endpoint hosts many models that use the same framework on a shared fleet, loads rarely used models from S3 on demand (accepting cold-start latency) and supports PrivateLink and auto scaling, which is the lowest-cost way to serve thousands of small models. One real-time endpoint per tenant pays for thousands of mostly idle instances. Serverless endpoints do not support VPC configuration, and nightly batch forecasts cannot answer on-demand inference calls.
Correct answer: C. Activate a Guard Hook that evaluates RDS DB instance resources in each account and Region, set its failure mode to WARN while reviewing the results, and then change it to FAIL.
CloudFormation Hooks inspect resource configurations before provisioning, and a Hook's failure mode can be WARN, which lets the operation continue with a warning, or FAIL, which stops it; the documented approach is to activate Hooks in WARN mode first to see which operations would be affected and then switch to FAIL. AWS Config proactive rules evaluate resource properties before deployment but do not prevent noncompliant resources from being deployed. Detective rules with remediation act only after the instance exists, and an SCP blocks immediately with no warning phase, so it would break deployments while the team is still learning the impact.
Correct answer: A. Add the quota increases to a Service Quotas quota request template and enable template association with the organization, and create CloudWatch alarms at 80 percent for those quotas from Service Quotas.
A Service Quotas request template, once associated with an organization that has all features enabled, automatically requests the configured quota increases whenever a new account is created in the organization, and Service Quotas can create CloudWatch alarms that notify you when usage approaches a chosen threshold. An SCP restricts permissions and cannot change quotas. Support cases and a custom Lambda function both add manual or custom work, and a weekly report or a throttle alarm arrives after the limit is already reached.
Correct answer: B. Migrate the database to Amazon RDS for MySQL in a Multi-AZ DB instance deployment, and turn on automated backups.
Replatforming to a managed service is the documented way to cut operational work: Amazon RDS manages backups, software patching, failure detection and recovery, and a Multi-AZ DB instance deployment keeps a synchronous standby in another Availability Zone with failover support, while the MySQL engine stays the same. Moving to DynamoDB is a refactor that requires the code changes the company wants to avoid. Patch Manager and AWS Backup reduce some tasks but leave the database self-managed with no managed failover, and a second EC2 instance with a Lambda function is still a home-grown failover.
Correct answer: D. Create an Availability Zone: recovery test with Regional error rate and latency success alarms and a duration above the RTO, set an S3 report destination, and run it monthly.
The Availability Zone: recovery test injects the symptoms of a power interruption in the chosen zone through AWS FIS actions and passes only if all success alarms return to OK within the Multi-AZ RTO and stay there until the test actions end; the docs advise Regional health alarms rather than per-AZ alarms and a duration longer than the RTO, and a report is generated after each run when the service has an S3 report destination. The docs describe a failure mode assessment's achievability result as directional guidance that must not be used as evidence. Alarms on the impaired zone are expected to breach, so the test would not show whether the service recovered, and a hand-built FIS experiment with manual graph comparison takes more effort and gives no pass or fail outcome.
Correct answer: C. Call PutOrganizationConfigRule for the five rules again from the delegated administrator account, this time with eu-west-1 as the Region of the API call.
The API call that deploys organization rules and conformance packs is Region specific, so the rules must be deployed again with the call made in each Region where resources must be evaluated. An aggregator only collects existing configuration and compliance data and cannot create evaluations in Regions that have no rules. Security Hub CSPM cross-Region aggregation consolidates findings into a home Region but does not make Config rules in one Region evaluate resources in another, and a delivery channel only controls where configuration data is delivered.
Correct answer: D. Create an AgentCore Identity OAuth 2.0 credential provider for the platform with the authorization code grant, keeping tokens and refresh tokens in the token vault.
The OAuth 2.0 authorization code grant (3LO) gives user-delegated access after explicit user consent, and AgentCore Identity stores the resulting access and refresh tokens in its token vault, uses the refresh token automatically when the access token expires, and keeps long-term secrets out of the agent code. The client credentials grant and a service account API key both act with an application-wide identity, which the firm forbids. Keeping refresh tokens in Secrets Manager and calling the token endpoint from the agent puts refresh tokens in the agent's hands.
Answer each question before opening the explanation, and write down why you discarded the other options. On SAP-C03, most mistakes come from choosing an option that is technically possible but does not meet every requirement in the scenario.
These 10 questions are a taste of the format, not a diagnostic. To know which domains you fail, practice by domain for free and take a timed mock exam that mirrors the official weights and duration.
Start for free, no sign-up or card. Every question includes an explanation like the ones above, and the mock exam reproduces the official SAP-C03 format.