Design scalable and loosely coupled architect..

The correct answer is C. Amazon DynamoDB is a managed NoSQL database that can scale to very high request rates, while DynamoDB Accelerator (DAX) is a managed, highly available in-memory cache designed specifically for DynamoDB. DAX can provide microsecond response times for frequently accessed data and reduce read traffic to DynamoDB. Option A is incorrect because RDS PostgreSQL is relational and does not satisfy the stated NoSQL requirement. Option B is incorrect because Aurora is relational and is not designed around a DynamoDB-style key-value/document workload. Option D is incorrect because DocumentDB is a document database, but its read-replica architecture does not provide the purpose-built DynamoDB-compatible microsecond caching layer described here.

The correct answer is C. Amazon ECS with the AWS Fargate launch type provides serverless compute for containers and supports task definitions with up to 16 vCPUs and 120 GB of memory for the specified Linux platform. Fargate removes the need to provision or manage EC2 container hosts, making it suitable for a 45-minute containerized workload requiring 16 vCPUs and 32 GB of memory. Option A is incorrect because a standard Lambda invocation is limited to 15 minutes and Lambda functions have a maximum memory configuration of 10,240 MB. Option B is incorrect because EC2 Auto Scaling requires the team to operate and maintain the underlying instances, contrary to the serverless-container requirement. Option D is incorrect because the stated AWS Batch option uses a managed EC2 compute environment rather than Fargate, so it does not satisfy the requirement to avoid EC2 capacity management. If the option instead specified an AWS Batch Fargate compute environment, it would also provide serverless container execution, so the current wording of option D is important.

The correct answer is C. An Application Load Balancer operates at Layer 7 and supports HTTP/HTTPS-aware routing, including path-based routing such as /trade and /portfolio. It can terminate TLS at the load balancer and distribute requests to multiple target groups. When the targets are managed by an EC2 Auto Scaling group, the application tier can scale horizontally. Option A is incorrect because an NLB operates primarily at Layer 4 and does not provide HTTP URL path-based routing. Option B is incorrect because CloudFront and Global Accelerator do not replace the regional Layer 7 load-balancing function required here. Option D is incorrect because Classic Load Balancer does not provide the required modern path-based routing capabilities.

The correct answer is D. Horizontal scaling requires the application instances to avoid dependence on instance-local session state. If shopping-cart state remains only in one instance's RAM, requests routed to another instance can lose access to the user's session data. Externalizing session state to an appropriate shared datastore such as Amazon ElastiCache or DynamoDB allows instances to scale out and in independently. Option A is incorrect because ALB cross-zone load balancing distributes traffic across Availability Zones but does not synchronize EC2 memory. Option B is vertical scaling and does not address the requirement to exceed the capacity of a single instance. Option C is incorrect because an ALB does not store session cookies on EFS; shared file storage is also not the appropriate general-purpose mechanism for highly dynamic session state.

The correct answer is D. Amazon Redshift is a managed cloud data warehouse designed for large-scale analytical workloads. Redshift uses massively parallel processing and columnar storage and supports petabyte-scale storage. Redshift Managed Storage allows compute and storage to scale independently, while Redshift Serverless provides a serverless option. Redshift supports PostgreSQL-compatible connectivity through standard drivers, making it suitable for BI tools. Option A is incorrect because Aurora PostgreSQL is primarily an OLTP relational database and is not the purpose-built choice for petabyte-scale analytical workloads. Option B is incorrect because Athena is a serverless SQL query service over data in S3 and is not a dedicated data warehouse for this continuously intensive BI workload. Option C is incorrect because EMR is a managed big-data platform but introduces Spark/Hadoop cluster and workload-management considerations that are unnecessary for this SQL data-warehouse requirement.

The correct answer is C. AWS Step Functions provides a native optimized integration with Amazon ECS/Fargate using the Run a Job (.sync) integration pattern. The workflow starts the ECS task and waits for the task to complete before transitioning to the next state, so the final Lambda can run after the sequencing job finishes. This avoids writing a long-running Lambda polling loop and avoids periodic polling through EventBridge. Option A is incorrect because a standard Lambda invocation cannot run a three-hour polling loop and would waste compute while polling. Option B is incorrect because an SQS visibility timeout controls message redelivery visibility; it does not make SQS wait for an ECS task and then automatically advance a workflow. Option D is incorrect because periodic EventBridge/Lambda polling adds custom code and repeated compute activity when Step Functions already provides the required ECS job-wait integration.

The correct answer is B. API Gateway REST API caching supports invalidating an individual cache entry by sending Cache-Control: max-age=0. Authorization can be required for this operation by granting the caller the execute-api:InvalidateCache permission for the appropriate API resource. This refreshes the requested cache entry rather than flushing the entire stage cache. Option A is incorrect because flushing the stage cache removes all cached entries and can temporarily increase traffic to the integration endpoint. Option C is incorrect because changing the TTL to zero disables effective caching rather than selectively refreshing stale entries. Option D is incorrect because a CloudFront invalidation affects CloudFront's cache; it does not directly invalidate the separate API Gateway cache.

The correct answer is B. API Gateway Usage Plans provide tier-specific throttling and quota settings and can be associated with API keys. This allows Basic, Pro, and Enterprise customers to be assigned different request-rate and quota configurations without implementing custom usage-tracking code. Option A is incorrect because WAF rate-based rules are not a substitute for customer-tier API quotas and identifying customers by IP is unreliable. Option C is incorrect because a custom Lambda/DynamoDB rate-limiting system adds substantial operational complexity and duplicates native API Gateway functionality. Option D is incorrect because IAM roles do not directly implement numerical API Gateway rate or quota limits. Important nuance: AWS documents Usage Plan throttling and quotas as best-effort controls rather than hard security boundaries, so they should not be represented as an absolute guarantee against every excess request.

The correct answer is B. EventBridge can match EC2 API activity recorded by CloudTrail, including TerminateInstances events, and route matching events to event buses, workflows, and notification targets. A centralized cross-account event architecture can be used to aggregate security events from multiple AWS accounts. Option A relies on aggregated metrics rather than the individual API event and its rich audit context. Option C is designed primarily for resource configuration and compliance evaluation rather than immediate API-event routing. Option D is incorrect because VPC Flow Logs record network traffic metadata and do not capture EC2 control-plane API operations such as instance termination.