Practice Test #1

DynamoDB Streams can emit item-level changes, but DynamoDB is not the best primary ingestion bus for millions of concurrent events that must be shared broadly in near real time. Also, DynamoDB does not provide a native “rule” mechanism to remove/mask PII on write; you would need application logic or a Lambda-based pipeline. This option also couples ingestion directly to the database, reducing flexibility for multiple analytics consumers. Trap: DynamoDB Streams sounds like fan-out, but this design lacks a native write-time PII masking rule and makes DynamoDB the ingestion bus.

Kinesis Data Firehose is optimized for loading streaming data into storage/analytics destinations (commonly S3, Redshift, OpenSearch) with optional Lambda transformation. However, Firehose is not designed for multiple independent near-real-time consumers; it delivers to destinations rather than acting as a shared event stream. Additionally, DynamoDB is not a typical/native Firehose destination, making this architecture mismatched to the stated DynamoDB requirement. Trap: Firehose sounds streaming-friendly, but it is a delivery service, not the shared stream for multiple near-real-time consumers.

Storing batched events as S3 files and processing them with Lambda is a batch-oriented pattern that increases end-to-end latency and is less suitable for “near-real-time” requirements. It also adds operational complexity around file sizing, partitioning, and reprocessing. While S3 is excellent for durable data lakes, it is not the right primary mechanism for real-time event fan-out to multiple analytics applications. Trap: S3 is durable, but batching files is too slow and operationally heavy for near-real-time event fan-out.

AWS CloudFormation templates provide infrastructure as code and parameterized deployments, so they can standardize LMS stacks. However, CloudFormation alone does not provide a governed self-service catalog experience, portfolio distribution, product version lifecycle management, or built-in constraints for many independent institutions. Trap: Create AWS CloudFormation templates for the educational institutions sounds related, but the scenario is really testing centralized governance with self-service deployment.

, Patch Manager, State Manager, Automation runbooks, Parameter Store) and can help configure and maintain deployed resources. It is not intended as a primary mechanism for distributing standardized deployable "products" with centralized governance and self-service provisioning across many institutions. Trap: Create AWS Systems Manager templates for the educational institutions sounds related, but the scenario is really testing centralized governance with self-service deployment.

AWS Config records configuration changes, evaluates resources against rules, and supports compliance reporting and remediation workflows. It helps enforce and audit regional compliance requirements after resources exist, but it does not provide a self-service deployment mechanism or centralized version-controlled product distribution. Trap: Create AWS Config items for the educational institutions sounds related, but the scenario is really testing centralized governance with self-service deployment.

SageMaker JumpStart still means selecting, deploying, and operating an inference endpoint, which is unnecessary for standard inappropriate-image detection. Trap: A prebuilt model sounds close, but the scenario explicitly avoids custom model development and operational ML endpoints.

CloudFront Functions are not suitable for invoking AI services, and Textract extracts text rather than detecting inappropriate visual content. Trap: It mentions medical images, but OCR does not solve image moderation.

Rekognition Video is for video analysis, and an S3 trigger runs after upload rather than providing an upload-time blocking decision. Trap: It uses the Rekognition name, but the media type and timing do not match the requirement.

Global tables help with regional availability, but corrupted writes replicate to the other Region. Trap: Multi-Region sounds resilient, but it is not a rollback mechanism for logical corruption.

Daily exports cannot meet a 15-minute RPO, and Glacier Deep Archive retrieval cannot reliably meet a 1-hour RTO. Trap: Archival storage is cheap, but it is too slow and too infrequent for this recovery objective.

Customers cannot take EBS snapshots of DynamoDB storage volumes because DynamoDB is a managed service. Trap: The 15-minute snapshot cadence matches the RPO wording, but the mechanism is not available for DynamoDB.

NAT instances are self-managed and should be in public subnets, not private subnets, to provide internet egress. Trap: Multiple instances sound highly available, but they add patching, scaling, and failover work.

An internet gateway attaches to a VPC, not a subnet, and direct IGW routing would not preserve private-only access. Trap: It mentions internet access, but it breaks the security requirement.

An egress-only internet gateway is for IPv6 outbound-only traffic and does not replace NAT gateways for IPv4 private subnet egress. Trap: The outbound-only name is attractive, but it is the wrong gateway type for typical IPv4 internet access.

Aurora Standard can be cheaper for low or moderate I/O, but per-I/O charges can become expensive during a high-traffic launch. Trap: Standard sounds cheaper, but the workload is explicitly high load and cost-sensitive as I/O scales.

Provisioned IOPS is not an Aurora cluster storage configuration for Aurora Serverless v2. Trap: It is a familiar RDS/EBS performance answer, but it does not apply to Aurora cluster storage here.

General Purpose SSD is also an EBS volume type, not the storage configuration choice for Aurora cluster storage. Trap: The name sounds cost-effective, but it belongs to non-Aurora EBS-backed storage decisions.

SQL Server on EC2 gives OS access, but database mirroring across Regions is self-managed and creates substantial operational overhead. Trap: It satisfies control, but fails the managed operations requirement.

Standard RDS for SQL Server with cross-Region backups can help recovery, but it does not provide full Windows OS access. Trap: Backups sound like DR, but the OS access requirement is decisive.

Standard RDS Multi-AZ is limited to Availability Zones in one Region and does not allow full access to the Windows operating system. Trap: Multi-AZ improves availability, but it is not cross-Region DR and lacks OS control.

Security groups do not understand geography, and maintaining all US IP ranges manually would be unreliable and operationally heavy. Trap: It sounds possible because countries map to IPs, but security groups only accept CIDR rules.

The ALB security group cannot filter by geographic location; it only allows or denies traffic by protocol, port, and source. Trap: It places the control at the ALB but chooses the wrong ALB feature.

Network ACLs are stateless IP and port filters and cannot evaluate a request source country. Trap: NACLs can block networks, but they are not geographic or application-aware controls.

Standard RDS for Oracle does not provide privileged operating system access, and replacing database features with Lambda changes the application design. Trap: RDS for Oracle may support Oracle workloads, but the host-access requirement is the deciding constraint.

Migrating to PostgreSQL requires a heterogeneous conversion and removes the existing Oracle feature compatibility. Trap: PostgreSQL can reduce license costs, but the rewrite risk violates the preserve-features requirement.

DynamoDB is not a drop-in replacement for Oracle stored procedures or third-party Oracle features. Trap: DMS can move some data, but it does not make a NoSQL target support Oracle-specific behavior.