The issue described is a classic case of read-heavy reporting workloads interfering with transactional application traffic. Although the RDS instance is configured for Multi-AZ, Multi-AZ deployments are designed for high availability and failover, not for scaling read performance. All reads and writes still occur on the primary instance, which explains why reporting queries negatively affect customer-facing performance.

Option C, adding an Amazon RDS read replica, is the most effective and AWS-recommended solution. Read replicas asynchronously replicate data from the primary instance and provide a separate endpoint for read-only workloads such as reporting, analytics, and dashboards. By redirecting the finance team’s reporting queries to the replica endpoint, the primary database is freed to handle transactional workloads, significantly improving overall application performance.

Option A (RDS Proxy) helps manage database connections efficiently, especially for bursty workloads, but it does not offload query execution or isolate heavy read traffic. Option B is invalid because RDS Multi-AZ does not support spreading a single DB instance across three AZs, and even if it did, it would not reduce read contention. Option D increases instance resources but does not solve the root cause; reporting queries would still compete with production traffic and would also increase cost unnecessarily.

Therefore, C is the optimal solution because it isolates reporting workloads, scales read capacity, improves application responsiveness, and aligns with AWS best practices for high-performance database architectures.

AWS Step Functions provides a low-code, fully managed workflow service with a visual interface to orchestrate AWS Glue jobs in sequence. Step Functions integrates natively with AWS Glue and can be triggered by Amazon EventBridge based on events or schedules.

AWS Documentation Extract:

“AWS Step Functions makes it easy to coordinate multiple AWS services into serverless workflows so you can build and update apps quickly. Step Functions provides visual workflows and integrates with AWS Glue for ETL orchestration.”

(Source: AWS Step Functions documentation)

A: Manual scripting and dashboards do not provide a low-code or integrated visual workflow.

C: QuickSight is for BI, not workflow visualization.

D: ECS adds unnecessary complexity.

EC2 Image Builder is the AWS-managed service specifically designed to automate the creation, patching, hardening, and testing of AMIs.

For this scenario, best practice is to:

Use EC2 Image Builder pipelines to generate new golden AMIs whenever features are deployed or on a schedule.

Include build components such as update-linux to automatically apply OS security patches and updates during image creation.

Deploy the new AMIs through existing Auto Scaling groups, ensuring that every newly launched instance is already patched and compliant.

This approach:

Prevents OS vulnerabilities from being introduced at deployment time.

Scales across dozens of applications because AMI pipelines are reusable and automated.

Minimizes manual effort and reduces drift between instances.

Amazon Inspector (Option A) identifies vulnerabilities but does not itself patch AMIs.

AWS Backup (Option B) is for data protection, not vulnerability management.

Patch Manager (Option C) patches running instances, but the requirement is to ensure vulnerabilities are not introduced with new AMIs; golden image pipelines with EC2 Image Builder are the recommended solution.

Requirement Analysis: Need secure, direct connectivity from an on-premises client to an RDS cluster, accessible only when in the office.

VPC with Private Subnets: Ensures the RDS cluster is not publicly accessible, enhancing security.

Site-to-Site VPN: Provides secure, encrypted connection between on-premises office and AWS VPC.

Implementation:

Create a VPC with two private subnets.

Launch the RDS cluster in the private subnets.

Set up a Site-to-Site VPN connection with a customer gateway in the office.

Conclusion: This setup ensures secure and direct connectivity with minimal exposure, meeting the requirement for secure access from the office.

References

AWS Site-to-Site VPN:AWS Site-to-Site VPN Documentation

Amazon RDS:Amazon RDS Documentation

Why Option D is Correct:

GPU Access: Only EC2 instances in the GPU family (e.g., P2, P3) can provide GPU resources.

ECS Orchestration: Simplifies container deployment and management.

Why Other Options Are Not Ideal:

Option A: Lambda does not support GPU-based runtimes.

Option B: AWS Fargate does not support GPU-based workloads.

Option C: ECR is a container registry, not an orchestration or execution service.

AWS References:

Amazon ECS with GPU Instances:AWS Documentation - ECS GPU Instances

Correct Approach:

AWS Security Groups:

Security groups operate at the instance level, making them the ideal tool for controlling access to specific resources such as an Amazon RDS database.

By default, security groups deny all incoming traffic. You can allow access by explicitly specifying another security group.

Associating an RDS database security group with the EC2 instances ' security group ensures only the specified EC2 instances can access the RDS database.

Incorrect Options Analysis:

Option A: Using CIDR blocks for IP-based access is less secure and more difficult to manage. Additionally, Auto Scaling groups dynamically allocate IP addresses, making this approach impractical.

Option B: Network ACLs (NACLs) operate at the subnet level and are stateless. While NACLs can deny or allow traffic, they are not suited to application-specific access control.

Option D: Similar to Option B, using a NACL with CIDR ranges for EC2 IPs is difficult to manage and not application-specific.

To address the high CPU utilization on the EC2 instance and the degraded performance of Amazon RDS for read requests, the solution involves two key actions: resizing the EC2 instance and leveraging Amazon RDS read replicas.

Resizing the EC2 Instance: The first step is to resize the EC2 instance to a type with more CPU capacity to handle the higher computational demands during peak usage times. This helps to alleviate the immediate pressure on the CPU.

Auto Scaling Group with a Size of 1: Although the application can only run on a single EC2 instance due to its monolithic nature, creating an Auto Scaling group with a minimum and maximum size of 1 ensures that the instance is automatically restarted or replaced in case of failure, maintaining high availability.

RDS Read Replica: Configuring an RDS read replica allows the application to offload read requests to a separate instance, thus reducing the load on the primary RDS instance. This improves the performance of read operations, which were previously bottlenecked due to the high CPU usage on the EC2 instance.

Why Not Other Options?:

Option B: Redirecting all traffic to the RDS read replica is not recommended because replicas are meant for read traffic only, not for write operations. This could lead to data consistency issues.

Option C: Increasing the RDS instance type capacity helps, but it doesn’t address the high CPU usage on the EC2 instance, nor does it provide a solution for scaling reads.

Option D: While resizing both the EC2 and RDS instances increases their capacities, it doesn’t address the specific need to offload read traffic from the primary RDS instance.

AWS References:

Amazon RDS Read Replicas- Explains how to create and use read replicas to offload read traffic from the primary database instance.

Resizing Your EC2 Instance- Guidance on resizing EC2 instances to meet workload demands.

Why Option C is Correct:

S3 Access Points: Provide scalable management of access to large datasets with specific permissions for individual prefixes.

Dynamic Prefixes: Access points simplify managing access to a growing number of prefixes without relying solely on a single bucket policy.

Fine-Grained Control: Resource-based permissions on access points enforce prefix-level isolation effectively.

Why Other Options Are Not Ideal:

Option A: Using deny/allow bucket policies introduces complexity and is less scalable for dynamic prefixes.

Option B: Encryption ensures data security but does not address access management.

Option D: Pre-signed URLs are temporary and not suitable for managing access at scale.

AWS References:

Amazon S3 Access Points:AWS Documentation - S3 Access Points

Comprehensive and Detailed 250 to 300 words of Explanation (AWS documentation-based, no links):

The requirements are straightforward: expose the Lambda microservice through an HTTPS endpoint and authenticate calls using IAM. Lambda function URLs are a built-in feature that provides a dedicated HTTPS endpoint for a Lambda function without requiring API Gateway, ALB, or CloudFront. When configured with the authentication type AWS_IAM, the endpoint requires requests to be signed with AWS Signature Version 4 and authorized by IAM policies. This directly satisfies the “must use IAM to authenticate calls” requirement with the least architectural complexity.

Option A can also secure an endpoint with IAM, but it proposes using a Lambda authorizer, which is typically used for custom authorizers (JWT/OAuth/Cognito/external identity). For IAM authentication in API Gateway, you generally use IAM authorization on the method, not an authorizer function. Also, API Gateway REST APIs introduce additional service configuration and per-request costs when a simpler managed option exists that meets the requirements.

Options C and D are not appropriate. Lambda@Edge and CloudFront Functions run at CloudFront edge locations with different programming and deployment models; they are designed for CDN request/response manipulation, not as the primary mechanism to expose a regional Lambda microservice endpoint with IAM authentication. CloudFront Functions in particular is for lightweight JavaScript at the edge and does not provide a native “AWS_IAM authentication type” for invoking an origin Lambda as a microservice endpoint.

Therefore, B is the cleanest and most secure fit: a native HTTPS endpoint backed by Lambda, protected with IAM-based SigV4 authentication.

After 180 days, the images become infrequently accessed but still must remain immediately available, which fits S3 Standard-IA better than One Zone-IA because the question requires high availability and redundancy throughout the lifecycle. AWS documents that Standard-IA is for long-lived, infrequently accessed data with millisecond access and multi-AZ resilience, while One Zone-IA stores data in a single AZ. After 360 days, the company still needs instant access, so S3 Glacier Instant Retrieval is the right archival class. AWS states that Glacier Instant Retrieval provides real-time access with lower storage cost than Standard-IA for archive data that still must be immediately retrievable. Glacier Flexible Retrieval would not meet the instant-access requirement.

============

For cost-effectiveness and high availability in Amazon EMR workloads, the best approach is to configure atransient cluster(which runs for the duration of the job and then terminates) withOn-Demand Instancesfor the primary and core nodes, andSpot Instancesfor the task nodes. Here ' s why:

Primary and core nodes on On-Demand Instances: These nodes are critical because they manage the cluster and store data on HDFS. Running them on On-Demand Instances ensures stability and that no data is lost, as Spot Instances can be interrupted.

Task nodes on Spot Instances: Task nodes handle additional processing and can be used with Spot Instances to reduce costs. Spot Instances are much cheaper but can be interrupted, which is fine for non-critical tasks as the framework can handle retries.

Atransient clusteris more cost-effective than a long-running cluster for workloads that only run for 6 hours a day. Transient clusters automatically terminate after the workload completes, saving costs by not keeping the cluster running when it ' s not needed.

Option A: A long-running cluster may result in unnecessary costs when the cluster isn ' t being used.

Option C: Running core nodes on Spot Instances risks data loss if the Spot Instances are interrupted, violating the requirement for zero data loss.

Option D: Running both core and task nodes on Spot Instances is highly risky for data-critical workloads.

AWS References:

Amazon EMR Cluster Management

Using Spot Instances in EMR

Spot Instances are the best pricing model because the jobscan tolerate interruptions, and the company wants the most cost-effective option. AWS recommends theprice-capacity-optimizedstrategy for Spot because it selects pools that balancelow pricewithlower interruption risk. Using multiple instance types in the Auto Scaling group also helps the company stay flexible enough to adopt newer generations over time. Reserved Instances do not fit as well because they reduce flexibility and do not align with the requirement to keep shifting toward the latest instance generations each year. Capacity-optimized alone minimizes interruption risk, but AWS recommends price-capacity-optimized when you want both lower interruption rates and better prices. Therefore, Spot with price-capacity-optimized allocation is the strongest answer.

============

The problem is scale-out latency: when traffic spikes, new instances take time to launch, initialize, and pass health checks. This can temporarily degrade user experience even if Auto Scaling eventually adds enough capacity. The goal is to improve responsiveness to sudden spikes cost-effectively.

A key Auto Scaling feature for this situation is an EC2 Auto Scaling warm pool. A warm pool maintains a set of pre-initialized instances (stopped or running, depending on configuration) that Auto Scaling can rapidly transition into service. This reduces the time needed for instance launch and initialization, enabling faster scale-out during unexpected surges.

Option B is best because it supports the application’s peak needs by ensuring Auto Scaling can scale up to the required maximum capacity and keeps warm capacity ready. While a warm pool has some cost (especially if instances are kept running), it is typically more cost-effective than permanently running peak capacity all the time. It improves user experience by minimizing cold start delays during spikes.

Option A is counterproductive: lowering minimum capacity increases the likelihood of scale-out delays. Option C is illogical as written (“increase maximum to meet normal demand”); maximum should reflect peak potential, and a warm pool without sufficient max headroom will not solve the spike problem. Option D uses scheduled scaling, but the spike is unexpected, and increasing instance sizes increases cost and does not address launch delay; it may also reduce horizontal scaling flexibility.

Therefore, B is the most cost-effective way to improve scale-out performance and protect user experience during unexpected bursts.

Since the company already usesIAM Identity Center, the lowest-overhead and most consistent design is to manage access throughpermission setsassigned to groups. AWS documentation explains that permission sets are the central mechanism IAM Identity Center uses to grant least-privilege access to AWS accounts and resources. This approach fits team-based access for development, testing, and operations without having to manage individual IAM users in each account. It also aligns with AWS guidance to apply least privilege by creating restrictive permission sets instead of broad standing permissions. The other options either introduce more manual identity management or create unnecessary account complexity for the scenario described. (AWS Documentation)

============

To restrict S3 access to a specific VPC, AWS documentation specifies using a Deny policy with aws:SourceVpc for all unmatched VPCs:

Deny if aws:SourceVpc != vpc-11aabb22

This ensures only that VPC can reach the bucket via a VPC endpoint.

Allow policies (Options B and C) are incorrectly structured or use the wrong condition logic.

Option D limits by account, not VPC, and does not enforce VPC-level access.

=====================================================

AWS Transfer Family is a fully managed service that provides SFTP, FTPS, and FTP access directly to Amazon S3. It allows organizations to support legacy data transfer protocols without managing infrastructure. This approach gives the vendors a familiar SFTP interface, with files landing directly in S3, and requires minimal operational effort compared to managing EC2 servers or using gateways.

Reference Extract from AWS Documentation / Study Guide:

" AWS Transfer Family enables secure transfer of files directly into and out of Amazon S3 using SFTP, FTPS, and FTP, and is fully managed by AWS, significantly reducing operational overhead. "

Source: AWS Certified Solutions Architect – Official Study Guide, Data Migration and Transfer section.

The most cost-effective and scalable solution for generating thumbnails when photos are uploaded to an S3 bucket is to useS3 event notificationsto trigger anAWS Lambda function. This approach avoids the need for a long-running EC2 instance or EMR cluster, making it highly cost-effective because Lambda only charges for the time it takes to process each event.

S3 Event Notifications: Automatically triggers the Lambda function when a new photo is uploaded to the S3 bucket.

AWS Lambda: A serverless compute service that scales automatically and only charges for execution time, which makes it the most economical choice when dealing with periodic events like photo uploads.

The Lambda function can generate the thumbnail and upload it to a second S3 bucket, fulfilling the requirement efficiently.

Option AandOption B(EMR or EC2 with scheduled scripts)**: These are less cost-effective as they involve continuously running infrastructure, which incurs unnecessary costs.

Option D (S3 Storage Lens): S3 Storage Lens is a tool for storage analytics and is not designed for event-based photo processing.

AWS References:

Amazon S3 Event Notifications

AWS Lambda Pricing

Amazon FSx for ONTAP supports both NFS and SMB protocols and includes automated tiering between SSD and capacity pool storage, optimizing cost and performance. It is ideal for mixed operating systems and varied access patterns with minimal administrative overhead.

Comprehensive and Detailed 250 to 300 words of Explanation (AWS documentation-based, no links):

Routing requests to the “appropriate microservices” implies Layer 7 (HTTP/HTTPS) routing such as host-based or path-based rules (for example, /customers to one service and /orders to another). In EKS, the most direct and cost-efficient way to provide this style of routing is to use Kubernetes Ingress with the AWS Load Balancer Controller to provision an Application Load Balancer (ALB). An ALB natively supports HTTP/HTTPS request routing features (paths, hosts, headers), which map cleanly to microservices patterns and reduce the need for additional infrastructure components.

A Network Load Balancer (NLB) (Option A) is primarily Layer 4 and is best for TCP/UDP/TLS pass-through and ultra-high performance. It does not provide the same native HTTP path-based routing behavior required to steer requests between microservices based on URL structure, so it often forces you to add another routing layer (like an in-cluster ingress gateway), which can increase cost and operational complexity.

Option C is not an appropriate request routing mechanism; Lambda is compute, not an ingress/routing layer for EKS. Option D (API Gateway) can front services, but for simple microservice routing into EKS it typically adds per-request costs and additional configuration compared with using an ALB that is purpose-built for HTTP load balancing and routing. API Gateway is a great fit for API management features (throttling, usage plans, keys, transformations), but those are not requirements here.

Therefore, B is the most cost-effective and architecturally aligned solution for HTTP routing across EKS-hosted microservices using managed AWS-native ingress integration.