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.
Verified against official sources on 2026-11-17.
AWS Certified Developer – Associate (DVA-C03) checks that you can develop, test, deploy and debug applications on AWS using the SDKs, CLI and service APIs. The new version also covers AI-assisted development tools and the security of applications that call AI services. Of all the AWS exams, it's the closest to your day job.
Version change: C02 to C03
Registration for DVA-C03 opened on 2026-10-27. The last day to take DVA-C02 is 2026-11-30, and DVA-C03 is used from 2026-12-01. If you follow this roadmap, you'll take DVA-C03. Don't use DVA-C02 study material without checking it against the C03 exam guide.
At a glance
| Code | DVA-C03 |
| Level | Associate |
| Intended audience | 1 or more years developing and maintaining applications with AWS services |
| Questions | 65 (50 scored + 15 unscored) |
| Time | 130 minutes |
| Passing score | 720 (scaled 100–1000) |
| Price | USD 150 |
| Question types | Multiple choice. Multiple response (you must select every correct option to get credit) |
| Valid for | 3 years. Renewed by passing DOP-C02 |
| Out of scope | Designing architectures, designing CI/CD pipelines, administering IAM users and groups, designing RAG systems or guardrails, designing ETL pipelines, designing networks |
Official domains and weightings
| Domain | Weight | Change from C02 |
|---|---|---|
| 1. Development with AWS Services | 30% | was 32% |
| 2. Security | 26% | same. Adds Task 2.3 on AI-service security risks |
| 3. Testing and Deployment | 22% | was "Deployment", 24% |
| 4. Troubleshooting and Optimization | 22% | was 18% |
Key AWS services tested
- Compute and APIs: Lambda (concurrency, layers, destinations, event source mappings, performance tuning), API Gateway (stages, custom domains, authorizers, throttling), ECS/Fargate task definitions, EKS manifests (basics)
- Data: DynamoDB (keys, indexes, query versus scan, consistency, conditional writes, Streams, TTL), S3 (presigned URLs), ElastiCache
- Messaging: SQS (visibility timeout, DLQs, FIFO), SNS, EventBridge, Step Functions, Kinesis Data Streams
- Security: IAM roles and the credential chain, Cognito user pools and identity pools, KMS (envelope encryption), Secrets Manager, Parameter Store, AWS AI-service data controls (VPC endpoints, Bedrock Guardrails sensitive-information filters)
- Deployment: SAM, CloudFormation, CDK, CodeBuild, CodeDeploy, CodePipeline, ECR (image tags, lifecycle policies), Lambda aliases and traffic shifting
- Observability: CloudWatch Logs, Logs Insights, metrics, Embedded Metric Format, X-Ray, OpenTelemetry
- AI tools: AI-assisted coding, test generation and troubleshooting (for example, Amazon Q Developer). ⚠ VERIFY which specific AWS AI developer tools the exam names, using the in-scope services list
What you already know vs what you must learn
| You already know (C#/.NET) | You must learn |
|---|---|
async/await, HttpClient, DI, IOptions, ILogger | The AWS SDK for .NET v4: client lifetimes, the credential provider chain, retries, pagination, AWSSDK.Extensions.NETCore.Setup |
| ASP.NET Core Web API, middleware, ProblemDetails | Lambda's execution model: cold starts, the handler, static initialisation, concurrency, timeouts. API Gateway integrations |
| EF Core and SQL indexing | DynamoDB key design: partition and sort keys, GSIs, query versus scan, conditional writes for idempotency |
| xUnit and Moq, Azure DevOps or GitHub pipelines | SAM and CDK, Lambda versions and aliases, CodeDeploy canary and linear deployments |
| Application Insights | CloudWatch Logs Insights, X-Ray traces, Embedded Metric Format |
| Copilot-style AI coding | Security for AI integrations: private connectivity, PII redaction, prompt-injection risk, securing agents |
Preparation order and study plan (about 40–50 hours over 4–6 weeks, after SAA-C03)
| Week | Focus | Hours |
|---|---|---|
| 1 | Exam guide, including the C02→C03 comparison page. Module 03 (SDK, Lambda in C#, Fargate) | 10 |
| 2 | Module 04 (API Gateway, Cognito, authorizers) and DynamoDB from Module 06 | 10 |
| 3 | Module 08 (SQS, SNS, EventBridge, Step Functions, idempotency). Domain 2 security, plus Module 16's AI security section | 8–10 |
| 4 | Module 09 (SAM, CodeDeploy, traffic shifting). Module 11 (Logs Insights, X-Ray) | 8–10 |
| 5 | Free Exam Prep course, the 20-question set, this page's questions, a mock exam | 6–8 |
Mapped platform modules
- Module 03 — .NET on AWS: the SDK, Lambda in C#, containers, configuration, secrets
- Module 04 — APIs & integration: API Gateway, Cognito, JWT, throttling
- Module 06 — Databases: DynamoDB design, ElastiCache
- Module 08 — Messaging & microservices: SQS, SNS, EventBridge, idempotency
- Module 09 — DevOps & CI/CD: SAM, CodeDeploy, Lambda aliases
- Module 11 — Monitoring: CloudWatch, X-Ray, OpenTelemetry for .NET
- Module 16 — GenAI: calling Bedrock from C# securely (new Task 2.3)
Practical skills to build
- Build an ASP.NET Core minimal API on Lambda behind API Gateway with a Cognito JWT authorizer, using SAM
- Write a DynamoDB idempotency table for
POST /paymentsusing a conditionalPutItem - Read a secret with Secrets Manager caching, and never log it
- Deploy a new Lambda version with a 10% canary and automatic rollback on a CloudWatch alarm
- Use Logs Insights to find the slowest requests by correlation ID, and follow one through X-Ray
- Build a container image in CodeBuild, push it to ECR with an immutable tag, and deploy it to Fargate
Exam-prep checklist
- I can explain the SDK credential provider chain and why ECS task roles and Lambda execution roles beat access keys
- I know DynamoDB's read and write capacity maths, GSI versus LSI, and when a scan is unavoidable
- I can set the SQS visibility timeout correctly for a Lambda consumer, and configure a DLQ and redrive
- I know the Lambda concurrency controls (reserved and provisioned) and what each one solves
- I know API Gateway's authorizer types (IAM, Cognito, Lambda) and when to use HTTP APIs versus REST APIs
- I can use Lambda aliases and CodeDeploy deployment preferences in SAM
- I can explain envelope encryption and when to use the KMS
GenerateDataKeyoperation - I know the new AI-security task: private connectivity to AI services, PII filtering, prompt-injection protection, human approval for sensitive agent actions
- I've used presigned URLs for time-limited S3 access
Official links
- Exam guide (DVA-C03)
- Comparison of DVA-C02 and DVA-C03
- Exam product page
- Skill Builder Exam Prep plan (DVA-C03)
Practice questions
These 5 questions are original, banking-flavoured scenarios. They aren't real exam questions.
A C# Lambda function connects to an Aurora PostgreSQL database that stores loan applications. Security policy says the password must not be stored in code or in plaintext configuration, and must be rotated automatically every 30 days. What should the developer do?
- A. Put the password in an encrypted Lambda environment variable that uses the default AWS managed key
- B. Store the credentials in AWS Secrets Manager with rotation enabled, give the execution role
secretsmanager:GetSecretValueon that secret only, and cache the value in the function - C. Put the password in
appsettings.jsonand deploy it with the function package - D. Store the password in a private S3 object and read it at startup
Correct answer: B
Explanation: Secrets Manager stores secrets encrypted with KMS, supports managed automatic rotation for Aurora and RDS, and records access in CloudTrail. A narrowly scoped execution role gives least privilege. Caching, either with the Secrets Manager caching library or the AWS Parameters and Secrets Lambda Extension, avoids calling the API on every invocation. Your code must cope with the password changing during rotation: when a connection fails, refresh the secret and retry. IAM database authentication is another option that removes passwords altogether.
Why the others are wrong:
- A: Environment variables are encrypted at rest, but anyone with
lambda:GetFunctionConfigurationcan read them, and there's no built-in rotation. - C: Secrets in deployment packages end up in source control, build logs and artifacts.
- D: S3 provides no rotation workflow and adds custom code. Secrets Manager is the purpose-built service.
Mobile clients retry POST /payments when the network times out, so the same payment request can arrive twice. Each request carries an Idempotency-Key header. The API stores payments in DynamoDB. How should the developer prevent duplicate payments reliably under concurrent requests?
- A. Scan the table for the key before writing
- B. Use
PutItemon an idempotency item with the condition expressionattribute_not_exists(pk), wherepkis the idempotency key, and return the stored result if the condition fails - C. Do a strongly consistent
GetItemfor the key, and callPutItemif it isn't found - D. Set a TTL on payment items so that duplicates expire
Correct answer: B
Explanation: A conditional write is atomic: if two requests race, exactly one write succeeds and the other gets a ConditionalCheckFailedException. The API can catch that exception and return the original response. To write the idempotency item and the payment item atomically together, use TransactWriteItems. Powertools for AWS Lambda (.NET) has an idempotency utility built on the same idea.
Why the others are wrong:
- A: A scan reads the whole table, which is slow and expensive, and there's still a race between the check and the write.
- C: Even a strongly consistent read followed by a write leaves a time-of-check to time-of-use gap. Two requests can both see "not found".
- D: TTL deletes old items in the background. It doesn't stop duplicates from being created.
A team deploys a fee-calculation Lambda function with AWS SAM. They want new versions to receive 10% of traffic for 10 minutes, then 100%, and to roll back automatically if an error-rate alarm fires. What should they configure?
- A. Deploy straight to
$LATESTand watch the CloudWatch dashboards - B. Set
AutoPublishAliasand aDeploymentPreferenceof typeCanary10Percent10Minuteswith the CloudWatch alarm listed underAlarms - C. Create two functions and split traffic with Route 53 weighted records
- D. Use API Gateway stage variables and change them by hand after 10 minutes
Correct answer: B
Explanation: With AutoPublishAlias, SAM publishes a new version on each deployment and points an alias at it. DeploymentPreference makes CodeDeploy shift alias traffic gradually (canary or linear). If any listed alarm goes into the ALARM state during the deployment, CodeDeploy moves traffic back automatically. You can also add pre-traffic and post-traffic hook functions for smoke tests.
Why the others are wrong:
- A:
$LATESThas no traffic shifting and no automatic rollback. - C: Route 53 weights DNS answers for endpoints. It isn't how you shift Lambda invocations, and it adds DNS caching delays.
- D: Changing stage variables by hand isn't gradual, isn't triggered by alarms, and depends on people.
An SQS queue triggers a .NET Lambda function that posts transactions to a ledger. The function's timeout is 60 seconds and it often runs for about 45 seconds. The queue's visibility timeout is 30 seconds. The ledger team reports duplicate postings. What's the most likely cause and fix?
- A. The batch size is too small. Increase it to 10
- B. The visibility timeout is shorter than the processing time, so messages become visible again and are processed twice. Set the visibility timeout to at least six times the function timeout, and make the handler idempotent
- C. Long polling is turned off. Set
ReceiveMessageWaitTimeSecondsto 20 - D. The queue should be FIFO, because standard queues always deliver duplicates
Correct answer: B
Explanation: While a message is being processed, it's hidden for the length of the visibility timeout. If processing takes longer than that, the message reappears and another invocation picks it up. AWS recommends a visibility timeout of at least six times the function timeout for Lambda event sources, to allow for retries and batching. Because SQS is still at-least-once, also make the ledger write idempotent and configure a DLQ with a sensible maxReceiveCount.
Why the others are wrong:
- A: Bigger batches make each invocation longer, which makes the problem worse.
- C: Long polling reduces empty receives and cost. It doesn't change visibility.
- D: Standard queues occasionally deliver duplicates. The main cause here is the timeout mismatch, which would also cause reprocessing with FIFO queues.
A bank's .NET support application sends chat transcripts to Amazon Bedrock for summarisation. Requirements: (1) traffic to Bedrock must not go over the public internet, and (2) card numbers and other PII must be masked in the model's responses. Which two actions meet these requirements?
- A. Create an interface VPC endpoint (AWS PrivateLink) for the Bedrock runtime and call it from the application's private subnets
- B. Apply an Amazon Bedrock guardrail with sensitive-information filters set to mask PII types such as card numbers
- C. Turn off CloudTrail so that prompts aren't logged
- D. Fine-tune a model on past transcripts so it learns not to output card numbers
- E. Call the public Bedrock endpoint over TLS 1.2
Correct answers: A and B
Explanation: An interface endpoint keeps calls to the Bedrock API on the AWS network, reachable from private subnets, and you can lock it down with an endpoint policy. Guardrails can detect sensitive information in inputs and outputs and mask or block it. Ideally, also redact card numbers before sending the transcript, because data you don't send can't leak. These controls support data protection, but the bank's privacy and model-risk teams still have to approve the use case.
Why the others are wrong:
- C: CloudTrail is your audit trail, and turning it off goes against banking audit requirements. To keep sensitive content out of logs, use CloudWatch Logs data protection policies or log less data instead.
- D: Fine-tuning on transcripts that contain PII increases the risk and gives no guarantee about outputs.
- E: TLS encrypts the traffic, but it still goes over the public internet, which fails requirement 1.
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 Solutions Architect – Associate (SAA-C03)
The core AWS architecture exam. It tests designing secure, resilient, high-performing and cost-optimised solutions.
AWS Certified Security – Specialty (SCS-C03)
The most important exam for banking. It covers detection, incident response, infrastructure security, IAM, data protection and governance.