Amazon FSx for Lustre is a high-performance file system optimized for fast processing of workloads such as machine learning, high-performance computing (HPC), and video processing. It integrates natively with Amazon S3, allowing you to:
Access S3 Data: FSx for Lustre can be linked to an S3 bucket, presenting S3 objects as files in the file system.
High Performance: It provides sub-millisecond latencies, high throughput, and millions of IOPS, which are ideal for ML workloads.Amazon Web Services, Inc.
Minimal Operational Overhead: Being a fully managed service, it reduces the complexity of setting up and managing high-performance file systems.
References:
Amazon FSx for Lustre - High-Performance File System Integrated with S3Amazon Web Services, Inc.
What is Amazon FSx for Lustre?

AWS Config has a built-in managed rule (access-keys-rotated) that evaluates IAM access key age. Config rules detect non-compliant resources automatically, without building custom logic.
After Config identifies old keys, EventBridge can trigger an AWS Lambda function to disable and delete the keys. Lambda provides a fully managed compute layer requiring no servers or batch environments.
Using AWS Batch (Options A, B, and D) adds unnecessary operational overhead for a simple automation task.

The requirements emphasize high availability across multiple Availability Zones, automatic recovery, and least operational overhead. The simplest way to achieve this for containerized microservices is to use a fully managed container orchestration service with serverless compute. Amazon ECS with the Fargate launch type provides exactly that.
With ECS Fargate, AWS manages the underlying compute infrastructure (no EC2 instances to provision, patch, scale, or secure at the OS level). You define task definitions and services, and ECS schedules tasks across Availability Zones based on your networking configuration (multiple subnets in different AZs). If a container fails, ECS service scheduling automatically replaces it, fulfilling the "recover from failures" requirement. Fargate also integrates with load balancing, CloudWatch logging, and scaling policies, reducing the operational burden of keeping the platform healthy.
Option A (ECS on EC2) is still managed orchestration, but it requires managing the EC2 capacity layer: AMI updates, instance patching, cluster capacity planning, and instance scaling. Option B (EKS with self-managed nodes) typically carries even more operational responsibility because you manage node groups, upgrades, and Kubernetes operational complexity in addition to the underlying instances. Option D is the highest operational overhead because it lacks managed scheduling and self-healing; you would need to build and maintain your own process manager and failover mechanisms.
Therefore, C is the best answer because it provides a managed, multi-AZ capable, self-healing container platform with minimal infrastructure management-matching the operational efficiency goal while maintaining high availability.

AWS Lake Formation natively supports fine-grained access control, including:
* Row-level security
* Cell/column-level security
These are implemented through data filters and Lake Formation permissions on governed tables and views.
This allows you to centrally define which users or groups can access which rows and which columns, including sensitive data fields, with minimal custom code or maintenance.
Why others are incorrect:
* A: IAM alone does not provide row-level or cell-level data security inside Lake Formation tables.
* C and D: Using Lambda to scrub or periodically remove sensitive data is complex, error-prone, and high overhead compared to built-in Lake Formation data filters.

AWS best practice for a highly available, resilient web application is:
* Run EC2 instances in an Auto Scaling group across multiple Availability Zones.
* Put an Application Load Balancer (ALB) in front of the Auto Scaling group.
* Use CloudWatch alarms and scaling policies for horizontal scaling based on demand.
Option D implements exactly this: a minimum of three instances spread across AZs (resilience to AZ failure), behind an ALB, with automatic horizontal scaling. This provides both elasticity and high availability with good cost efficiency (instances scale out/in based on traffic).
Option A uses only three fixed instances and relies on vertical scaling, which does not scale as well and is less resilient.
Option B overprovisions to nine instances from the start, which is unnecessarily expensive.
Option C keeps all instances in a single AZ, which does not meet resilience requirements.