Amazon Web Services Updated DVA-C02 Exam Questions and Answers by florrie

When a Lambda function is configured to run inside a VPC, it receives ENIs in the selected subnets and uses those subnets’ routing to reach other networks. If the function is placed in private subnets (common when it needs to access an RDS database in private subnets), it typically does not have a direct route to the internet. As a result, outbound calls to public external APIs will fail unless the VPC provides a controlled egress path.

The standard AWS networking pattern for private-subnet internet access is a NAT gateway (or NAT instance) deployed in a public subnet , with a route from the private subnet(s) to the NAT gateway (0.0.0.0/0 → NAT). The NAT gateway then uses an internet gateway attached to the VPC to reach the public internet while keeping the Lambda function (and the RDS database) in private address space. This resolves the connectivity issue while maintaining the security posture of private subnets.

Option A is not the solution: Lambda automatically provisions ENIs for VPC-enabled functions; manually provisioning an ENI does not provide internet access. Option C (security group outbound rules) is necessary but not sufficient—security groups control allowed traffic, but without a proper route to the internet (via NAT), the packets still cannot reach external endpoints. Option D (gateway VPC endpoint) is for AWS services like S3 and DynamoDB (gateway endpoints), not for arbitrary public internet APIs, and “in a public subnet” is not how gateway endpoints are used; they are associated with route tables.

Therefore, the correct solution is B : add a NAT gateway in a public subnet and update private subnet route tables so the new Lambda function can reach external public APIs while still accessing the private RDS database.

Requirement Summary:

NLP Lambda function with a large pre-trained model

Lambda layer became 8.7 GB → Exceeds AWS limits

Function returns RequestEntityTooLargeException

Need: High-performing, portable, low initialization time

Important AWS Limits:

Lambda Layers size limit (combined across all layers): 250 MB (unzipped)

Deployment package size (unzipped): 250 MB

Lambda container image support allows up to 10 GB image size

Evaluate Options:

A: Store model in S3 and load during execution

Leads to cold start latency every time

Model loading from S3 is slower and not suitable for real-time NLP

Not optimal for performance

B: Use EFS mounted to Lambda

⚠️ Valid for large models, but adds latency during cold start as model loads from EFS

Requires EFS setup, VPC, and has added network I/O overhead

Still slower than bundling in container image

C: Split into five Lambda layers

Still violates the total layer size limit of 250 MB (unzipped)

You cannot exceed that even with multiple layers

D: Use Docker container image

Allows bundling up to 10 GB of dependencies and models

High portability and performance

Avoids latency of downloading models at runtime

Ideal for scientific/NLP models

Lambda container image support: https://docs.aws.amazon.com/lambda/latest/dg/images-create.html

Lambda limits: https://docs.aws.amazon.com/lambda/latest/dg/gettingstarted-limits.html

Using large models with Lambda: https://aws.amazon.com/blogs/machine-learning/deploying-large-machine-learning-models-on-aws-lambda-with-container-images/

The requirement has two parts: (1) authenticate users with a third-party SAML IdP , and (2) after authentication, allow those users to access AWS services like Amazon S3 and DynamoDB. The AWS pattern for this is to federate the external identity into AWS and then exchange that identity for temporary AWS credentials .

Amazon Cognito identity pools are designed to provide AWS credentials to authenticated users (via AWS STS) so the users can call AWS services directly or through an application backend. Identity pools support federation with external identity providers, including SAML 2.0 . When a user authenticates with the third-party SAML IdP, the identity pool can accept the SAML assertion, map the user to an IAM role, and return temporary, scoped credentials (AccessKeyId/SecretAccessKey/SessionToken) associated with that role. Those credentials can be restricted using IAM policies to only the required S3 buckets, DynamoDB tables, and actions, satisfying least privilege.

Option A is incorrect because a Cognito user pool is primarily a user directory and OIDC/OAuth provider for application authentication; while user pools can integrate with SAML IdPs for federation, user pools alone do not directly issue AWS service credentials. The key requirement here is access to S3/DynamoDB, which is the role of an identity pool . Option C is not appropriate because IAM is not intended to provide end-user sign-up/sign-in for external users; it is for AWS identities and permissions. Option D is unrelated: CloudFront signed URLs control access to CloudFront-distributed content and do not authenticate users with SAML or provide AWS credentials.

Therefore, B meets both requirements: federate SAML authentication through a Cognito identity pool and provide temporary AWS credentials for accessing S3 and DynamoDB.