CloudCert Path

Free DVA-C03 practice questions — 30 questions, free, no sign-up.

Try 30 free AWS Certified Developer – Associate (DVA-C03) practice questions, written from the public DVA-C03 exam guide and official AWS documentation. Every question comes with its answer and an explanation that cites its AWS source — so a wrong answer still teaches you something. Answers are shown on this page; work through each question before reading the answer.

What changed from DVA-C02? Try the free CLF-C02 quiz

Free to read, no sign-up. Based on the public DVA-C03 exam guide and official AWS documentation. These are original practice questions, not real exam questions.

The DVA-C03 exam in 60 seconds

D1 — Development with AWS Services (9 questions)

Question 1 — Lambda timeout

A Lambda function processes images uploaded to S3. Most invocations finish in 2 seconds, but large images take 8 seconds and the function times out. The function timeout is still at its default. What is the most direct fix?

  1. Increase the function's memory to the maximum value
  2. Enable provisioned concurrency for the function
  3. Increase the function's timeout setting to allow up to 8 seconds of execution (correct answer)
  4. Move the image processing to an EC2 instance
Answer: C. The default Lambda timeout is 3 seconds; raising the timeout (up to 15 minutes) directly addresses a function that needs more time. Memory affects CPU allocation, not the time limit.
Source: AWS Lambda Developer Guide — configuring function timeout.

Question 2 — SQS poison messages

An SQS standard queue feeds a Lambda consumer. One malformed message keeps failing, is retried repeatedly, and blocks processing of other messages. What should the developer configure?

  1. Increase the queue's message retention period to 14 days
  2. Increase the Lambda function's reserved concurrency
  3. Switch the queue to FIFO and enable content-based deduplication
  4. Configure a dead-letter queue with a maxReceiveCount so repeatedly failing messages are moved aside after a set number of receives (correct answer)
Answer: D. A DLQ with maxReceiveCount isolates poison messages after repeated failures so healthy messages keep flowing. Retention and concurrency do not remove the bad message.
Source: Amazon SQS Developer Guide — dead-letter queues; DVA-C03 skeleton D1 (DLQ mechanism).

Question 3 — DynamoDB Query

A table stores orders with partition key customerId and sort key orderDate. The application must list all orders for one customer in a given month, without scanning the whole table. What is the correct approach?

  1. Use Query on the table with customerId as the key condition and a condition on orderDate (correct answer)
  2. Use Scan with a filter expression on customerId
  3. Create a global secondary index with orderDate as the partition key
  4. Use BatchGetItem with the customer's ID
Answer: A. Query uses the partition key to fetch only that customer's items and can filter by the sort key range; Scan reads the whole table and is inefficient here.
Source: Amazon DynamoDB Developer Guide — Query vs Scan.

Question 4 — Event-driven decoupling

An order service must notify three independent services (email, inventory, analytics) when an order is placed, and each consumer should process the event at its own pace. Which pattern fits best?

  1. Have the order service call each service's HTTP endpoint synchronously
  2. Use a single SQS FIFO queue shared by all three consumers
  3. Store the event in DynamoDB and have each service poll the table
  4. Publish the event to an Amazon SNS topic or Amazon EventBridge event bus that fans out to the three consumers (correct answer)
Answer: D. A topic / event bus provides fan-out decoupling: one publish reaches all subscribers, each consuming independently. Synchronous calls couple the services; one shared queue would deliver each message to only one consumer.
Source: Amazon SNS / EventBridge documentation — fan-out and event-driven patterns; DVA-C03 skeleton D1.

Question 5 — API Gateway throttling

A public API built on API Gateway and Lambda is hit by a burst of traffic from a single client, causing throttling errors for other clients. What is the most appropriate first control?

  1. Disable API caching
  2. Increase the Lambda function's timeout
  3. Configure throttling limits and a usage plan with an API key for that client (correct answer)
  4. Move the API to a private VPC endpoint
Answer: C. Usage plans with per-key throttling quotas isolate abusive clients while others keep their allowance. Timeout and caching do not limit request rate.
Source: Amazon API Gateway Developer Guide — usage plans and throttling.

Question 6 — SDK pagination and retries

An application lists all objects in a large S3 bucket using the AWS SDK, but only the first 1,000 objects are returned. What should the developer do?

  1. Increase the SDK's connection timeout
  2. Call the operation once with a larger page size and no further calls
  3. Follow the pagination token returned by the operation and repeat the call until no token is returned (correct answer)
  4. Switch to a different AWS Region
Answer: C. S3 ListObjectsV2 returns results in pages with a continuation token; the SDK paginator (or manual token loop) is required to retrieve all objects.
Source: AWS SDK documentation — pagination; DVA-C03 skeleton D1 (SDK pagination).

Question 7 — AI-assisted development

A developer uses Kiro / Amazon Q Developer to speed up writing a Lambda function. Which use is in line with AI-assisted development for DVA-C03?

  1. Use the assistant to generate code and unit tests from a description, then review and test the generated code before committing it (correct answer)
  2. Let the assistant choose the foundation model and design the RAG pipeline for the application
  3. Ask the assistant to design the application's governance framework for AI use
  4. Use the assistant to select which AI model to fine-tune for the workload
Answer: A. AI-assisted coding (generation, review, tests) is in scope; selecting foundation models, RAG design and AI governance design are explicitly out of scope for DVA-C03.
Source: DVA-C03 exam guide — AI-assisted development; skeleton v2 out-of-scope list.

Question 8 — Bedrock API call

An application calls a managed AI service through the Amazon Bedrock runtime API. The call must include the model identifier and the input prompt, and return generated text. Which API operation is used for a single on-demand text generation call?

  1. CreateModelCustomizationJob
  2. StartIngestionJob
  3. ListFoundationModels
  4. InvokeModel (correct answer)
Answer: D. InvokeModel performs a single synchronous inference call with a model ID and input body; the other operations manage models or ingestion, not inference.
Source: Amazon Bedrock API Reference — InvokeModel; DVA-C03 skeleton D1 (Bedrock API basics).

Question 9 — DynamoDB conditional writes

Two Lambda invocations may try to create the same user profile item at the same time. The application must ensure only the first write succeeds and the second fails cleanly. What should the developer use?

  1. A conditional write (PutItem with a condition expression such as attribute_not_exists(userId)) (correct answer)
  2. A Scan with a filter expression before every write
  3. A global secondary index on userId
  4. DynamoDB Streams to detect the duplicate after the write
Answer: A. A condition expression makes the write atomic: it succeeds only if the item does not already exist, so the second concurrent write fails with a conditional check error.
Source: Amazon DynamoDB Developer Guide — condition expressions.

D2 — Security (8 questions)

Question 10 — Cognito user pools vs identity pools

A mobile app needs users to sign in with email and password, and then access an S3 bucket with temporary AWS credentials scoped to each user. Which combination is correct?

  1. Use only a Cognito user pool; it directly issues AWS credentials for S3
  2. Use only a Cognito identity pool; it handles email/password sign-in
  3. Use a Cognito user pool for sign-in, federated into a Cognito identity pool that vends temporary credentials (correct answer)
  4. Use IAM users, one per mobile user, with long-term access keys in the app
Answer: C. User pools authenticate users; identity pools exchange that authentication for temporary, scoped AWS credentials. Embedding IAM user keys in a mobile app is not acceptable.
Source: Amazon Cognito Developer Guide — user pools vs identity pools.

Question 11 — IAM policy evaluation

An IAM policy allows s3:GetObject on a bucket, a second policy attached to the same role explicitly denies s3:GetObject on one object in that bucket, and there is no SCP involved. What happens when the role tries to read that object?

  1. Access is denied, because an explicit deny always overrides an allow (correct answer)
  2. Access is allowed, because the allow is more specific to the bucket
  3. Access is allowed, because the most recently attached policy wins
  4. Access is denied only if the request comes from outside the VPC
Answer: A. In IAM policy evaluation, an explicit deny overrides any allow, regardless of order or specificity.
Source: AWS IAM documentation — policy evaluation logic.

Question 12 — KMS envelope encryption

An application encrypts large files before storing them in S3. It uses a KMS key to encrypt a data key, then encrypts the file with that data key. What is this pattern called, and why is it used?

  1. Client-side encryption with a KMS key per file, because KMS keys can be downloaded and used to encrypt data offline
  2. Server-side encryption with S3 managed keys, because the application never handles keys
  3. Envelope encryption with a customer managed key, because KMS keys have a 4 KB limit on direct encryption (correct answer)
  4. Envelope encryption; it is used because KMS keys can directly encrypt data of any size
Answer: C. Envelope encryption: a data key encrypts the data, the KMS key encrypts the data key. KMS direct encryption is limited to 4 KB, so large data must use a data key.
Source: AWS KMS Developer Guide — envelope encryption.

Question 13 — Credential rotation

A Lambda function needs database credentials that must be rotated every 30 days without code changes. Which service and feature should the developer use?

  1. Store the password in an environment variable and update the function monthly
  2. Hard-code the credentials and redeploy the function every 30 days
  3. Store the credentials in Systems Manager Parameter Store as a SecureString and rotate them manually
  4. Store the credentials in AWS Secrets Manager and enable automatic rotation with a Lambda rotation function (correct answer)
Answer: D. Secrets Manager supports automatic rotation on a schedule, and the application retrieves the current secret at runtime, so no code change is needed.
Source: AWS Secrets Manager documentation — rotation.

Question 14 — Presigned URLs

A developer must let an unauthenticated user download one private S3 object for the next 15 minutes, without making the object public or sharing AWS credentials. What is the correct approach?

  1. Generate a presigned URL for the object with a 15-minute expiry (correct answer)
  2. Make the bucket public and remove the policy after 15 minutes
  3. Create an IAM user for the external user and email the access key
  4. Copy the object to a public bucket and delete it after 15 minutes
Answer: A. A presigned URL grants time-limited access to a single private object using the signer's credentials, without exposing them or changing the bucket policy.
Source: Amazon S3 documentation — presigned URLs; DVA-C03 guide skill 2.1.4.

Question 15 — AI service networking and privacy

An application sends customer data to Amazon Bedrock. The security team requires that the traffic never cross the public internet and that the data is not used to train models. Which design meets both requirements?

  1. Call Bedrock over the public endpoint and enable logging to CloudTrail
  2. Use an IAM policy that denies bedrock:InvokeModel from outside the corporate network
  3. Encrypt the prompt with a KMS key before sending it
  4. Access Bedrock through an AWS PrivateLink VPC endpoint, and rely on Bedrock's policy that inputs/outputs are not used for training (correct answer)
Answer: D. A VPC endpoint (PrivateLink) keeps traffic inside the AWS network, and Bedrock does not use customer inputs or outputs to train models. Encryption alone does not change the network path.
Source: Amazon Bedrock documentation — VPC endpoints and data privacy; DVA-C03 guide Task 2.3.

Question 16 — Sensitive data in AI interactions

Application logs capture the full prompts sent to an AI service, including customer email addresses. The team must reduce exposure of personal data in logs. What is the most appropriate action?

  1. Disable all logging for the application
  2. Store the logs in a public S3 bucket so they are easy to audit
  3. Detect and mask or redact PII/PHI in log output before it is written, and limit who can read the logs (correct answer)
  4. Send the raw prompts to a second AI service to summarise them
Answer: C. Sensitive-information detection and redaction in logs is part of AI-service security (Task 2.3): remove PII before it is persisted and restrict log access. Disabling logs removes troubleshooting ability.
Source: DVA-C03 exam guide Task 2.3 (sensitive information detection, protection of sensitive content in monitoring/logs).

Question 17 — SigV4 signed requests

A developer writes code that calls an AWS service API directly over HTTPS, without the AWS SDK. What must each request include to prove the caller's identity?

  1. The AWS account root password in an HTTP header
  2. A long-term access key pasted into the request URL as a query parameter
  3. A Signature Version 4 (SigV4) signature computed from the access key, request details and timestamp (correct answer)
  4. A Cognito user pool JWT token in the Authorization header
Answer: C. SigV4 signs the request (method, URI, headers, payload hash, timestamp) with the secret access key; the service verifies the signature without the secret ever being sent.
Source: AWS documentation — signing AWS API requests with SigV4.

D3 — Testing and Deployment (7 questions)

Question 18 — Blue/green vs canary

A team deploys a new API version to production. They want to shift 10% of traffic to the new version, monitor errors, then gradually increase the share, rolling back automatically if errors spike. Which deployment strategy is this?

  1. All-at-once deployment
  2. Blue/green deployment with a full traffic switch
  3. Rolling deployment to all instances simultaneously
  4. Canary deployment (correct answer)
Answer: D. A canary shifts a small percentage of traffic to the new version first and expands it based on health; blue/green switches all traffic at once after the new environment is ready.
Source: AWS CodeDeploy documentation — deployment configurations; DVA-C03 guide skill 3.3.6.

Question 19 — IaC change sets

A developer has a CloudFormation template change for production. Before applying it, they want to see exactly which resources will be added, modified or replaced, without changing anything. What should they use?

  1. A change set (correct answer)
  2. A nested stack with no parameters
  3. A stack policy that denies all updates
  4. A CloudFormation drift detection report only
Answer: A. A change set previews the exact impact of a template/stack update before execution; drift detection reports differences between actual and expected state, not planned changes.
Source: AWS CloudFormation documentation — change sets.

Question 20 — Container image builds

A Dockerfile for a Lambda container image installs dependencies, then copies application code, in a single stage, producing a large image. Which practice most effectively reduces the final image size?

  1. Use a multi-stage build so the final image contains only the runtime and the built artefacts (correct answer)
  2. Add more layers to the Dockerfile
  3. Push the image to ECR with immutable tags
  4. Increase the Lambda function's memory
Answer: A. Multi-stage builds discard build-time tools and intermediate files from the final image. Tag immutability is a versioning control, not a size optimisation.
Source: Docker / AWS documentation — multi-stage builds; DVA-C03 guide skill 3.1.2 (container image building).

Question 21 — ECS deployments

An ECS service runs 4 tasks behind a load balancer. A new task definition is deployed and one new task fails its health check. What does ECS do by default with a rolling update?

  1. It stops the entire service immediately
  2. It deletes the task definition
  3. It deletes the load balancer and recreates it
  4. It keeps serving traffic from the healthy old tasks and replaces the failed new task, respecting the deployment configuration's minimum healthy percent (correct answer)
Answer: D. A rolling update replaces tasks gradually; the service maintains the minimum healthy percent, so healthy old tasks keep serving while failed new tasks are replaced.
Source: Amazon ECS Developer Guide — service deployments.

Question 22 — Testing strategy

A developer writes unit tests for a function that calls DynamoDB. The tests must run in CI without AWS credentials or network access, and must run in milliseconds. What is the appropriate approach?

  1. Run the tests against the production DynamoDB table
  2. Skip unit tests and rely only on manual testing in the AWS console
  3. Mock the DynamoDB client in the unit tests so no real AWS call is made (correct answer)
  4. Create a real DynamoDB table for each test run and delete it afterwards
Answer: C. Mocking the AWS client isolates the unit under test, needs no credentials or network, and runs fast; integration tests (not unit tests) are the place for real AWS resources.
Source: AWS testing guidance — unit vs integration testing; DVA-C03 guide Task 3.x (testing in AWS environments).

Question 23 — CodeArtifact versioning

A team publishes internal libraries to CodeArtifact. A build must always resolve the same library version it was tested with, even after a newer version is published. What should the build reference?

  1. The latest tag of the package
  2. A specific pinned version of the package (correct answer)
  3. The package's upstream public repository directly
  4. The newest version available at build time
Answer: B. Pinning an exact version makes builds reproducible; floating tags such as latest can silently change what a build resolves.
Source: AWS CodeArtifact documentation — package versions.

Question 24 — CodePipeline failure rollback

A CodePipeline deployment stage fails in CodeDeploy. The team wants the pipeline to stop, keep the previous good version running, and surface the failure. Which behaviour achieves this?

  1. Configure the pipeline to ignore the failed stage and continue to the next stage
  2. Let the failed deployment trigger CodeDeploy's automatic rollback to the last known good revision, and let the pipeline stage report failure (correct answer)
  3. Delete the pipeline and recreate it after each failure
  4. Increase the CodeBuild timeout so the deployment eventually succeeds
Answer: B. CodeDeploy can automatically roll back to the previous revision on failure, and the pipeline stage fails visibly so the team is alerted; ignoring the failure would leave a broken or partial deployment in place.
Source: AWS CodeDeploy documentation — automatic rollbacks.

D4 — Troubleshooting and Optimization (6 questions)

Question 25 — Troubleshooting Lambda timeouts

A Lambda function intermittently times out, but only when it calls a downstream API. CloudWatch Logs show the function starts, then nothing until the timeout. What is the most useful next diagnostic step?

  1. Increase the function's memory and redeploy immediately
  2. Check the downstream API's latency and the function's VPC / network path, and add X-Ray tracing to see where time is spent (correct answer)
  3. Delete the function's CloudWatch log group
  4. Change the function's runtime to a newer version
Answer: B. The log gap points to time spent waiting on the downstream call; tracing and network/latency checks locate the wait instead of guessing at a fix.
Source: AWS Lambda / X-Ray documentation — tracing and troubleshooting.

Question 26 — DynamoDB throttling

An application using DynamoDB provisioned capacity sees ProvisionedThroughputExceededException during a traffic spike on one partition key. What is the most effective immediate mitigation?

  1. Enable DynamoDB Streams on the table
  2. Switch the table to on-demand capacity mode, or increase provisioned capacity / use DAX caching for hot reads (correct answer)
  3. Delete and recreate the table with the same key schema
  4. Enable point-in-time recovery on the table
Answer: B. On-demand mode absorbs spikes without pre-provisioning, and DAX caches hot reads; a hot partition key is a capacity-distribution problem, not a backup problem.
Source: Amazon DynamoDB Developer Guide — throughput exceeded errors, on-demand mode, DAX.

Question 27 — API Gateway 502 errors

Users report intermittent HTTP 502 errors from an API Gateway REST API backed by Lambda. CloudWatch shows the Lambda function itself rarely errors. What is the most likely cause to check first?

  1. The API Gateway stage has no usage plan
  2. The Lambda function returned a malformed response (not matching the expected proxy integration format) or the integration timed out (correct answer)
  3. The API key has expired
  4. The S3 bucket holding the API's static assets is private
Answer: B. A 502 from API Gateway with a Lambda proxy integration most often means the integration returned an invalid response format or timed out; the Lambda function succeeding does not guarantee a well-formed proxy response.
Source: Amazon API Gateway documentation — troubleshooting 502 errors.

Question 28 — Cold-start optimisation

A latency-sensitive Lambda function suffers long cold starts after periods of inactivity. Which option most directly reduces cold-start latency for steady traffic?

  1. Provisioned concurrency, which keeps execution environments initialised and ready (correct answer)
  2. Increasing the function's timeout
  3. Enabling SQS as an event source
  4. Reducing the function's reserved concurrency to zero
Answer: A. Provisioned concurrency pre-initialises environments, so invocations skip the init phase; timeout and event source do not affect cold-start time.
Source: AWS Lambda Developer Guide — provisioned concurrency.

Question 29 — Troubleshooting SQS poison messages

Messages in an SQS queue are being processed multiple times and the same message reappears after each failure. The consumer's visibility timeout is 30 seconds, but processing takes 90 seconds. What is the cause?

  1. The queue's retention period is too short
  2. The visibility timeout is shorter than the processing time, so the message becomes visible again and is redelivered while still being processed (correct answer)
  3. The queue is FIFO instead of standard
  4. The DLQ maxReceiveCount is set too high
Answer: B. When processing exceeds the visibility timeout, SQS makes the message visible again and another consumer (or the same one) receives it, causing duplicates; the fix is to raise the visibility timeout above the processing time (or extend it with ChangeMessageVisibility).
Source: Amazon SQS Developer Guide — visibility timeout.

Question 30 — Verifying AI diagnostic suggestions

An AI assistant suggests a fix for a production latency issue, based on a summary of recent logs. Before applying it, what should the developer do?

  1. Apply the suggestion immediately, since it is based on the logs
  2. Validate the suggestion against the raw CloudWatch Logs / metrics and X-Ray traces, reproduce the evidence, and test the change in a non-production environment first (correct answer)
  3. Ask the assistant to generate a second, different suggestion and apply both
  4. Disable the AI assistant so it cannot suggest further changes
Answer: B. AI-assisted diagnosis must be verified with primary evidence (logs, metrics, traces) and tested before production; applying an unverified suggestion risks making the incident worse.
Source: DVA-C03 exam guide Task 4.x — AI-assisted diagnosis (validate AI suggestions with log/metric evidence before release); skeleton v2 D4.

Want all 300 questions?

The full Developer Associate Practice Exam Pack (300 original questions, timed 65-question mock exams, per-question explanations) is not ready yet. It starts when either of these happens: the waitlist below reaches 50 people, or the free CLF-C02 demo on this site reaches 300 unique visitors.

Joining opens your email app with a pre-addressed message — nothing is sent automatically. We will only use your email to tell you when the full pack is ready, and you can ask us to delete it at any time. We do not send the sample questions by email — they are on this page. See the Privacy Policy.

FAQ

Are these the real exam questions?

No. Every question is written by us from the public DVA-C03 exam guide and official AWS documentation. Using real exam questions would be a violation of AWS exam policy; ours are original practice questions built to the same syllabus.

Do I need to create an account or give you my email?

No. The 30 sample questions are on this page and need no sign-up. The email field above is only for the waitlist for the full 300-question pack.

When does DVA-C03 start?

1 December 2026. The last day to sit DVA-C02 is 30 November 2026.

How is DVA-C03 different from DVA-C02?

See DVA-C03 vs DVA-C02 — what changed: AppSync is out, container image building and presigned URLs moved domains, AI-assisted development and AI-services security (Task 2.3) are new, and Troubleshooting grew from 18% to 22% of the exam.

Who do I contact for help?

support@cloudcertpath.com — we reply within 3 business days.

Questions on this page were checked against the public DVA-C03 exam guide. Exam facts can change — confirm them with AWS Training and Certification before you book. We are not affiliated with, endorsed by, or sponsored by Amazon Web Services, Inc.