Amazon Web Services SAA-C03 - AWS Certified Solutions Architect - Associate (SAA-C03)

For steady, predictable EC2 usage with potential changes in instance families over time, AWS recommends Savings Plans. Compute Savings Plans “apply to any EC2 instance regardless of region, instance family, operating system, or tenancy,” and also apply to AWS Fargate and AWS Lambda, delivering the most flexibility over a multi-year horizon. A 3-year term provides the highest discount among Savings Plans for long-lived workloads. EC2 Instance Savings Plans are limited to a chosen instance family in a region; as needs evolve (e.g., size or family changes), discounts may not fully apply. Spot Instances are not appropriate for long, interruption-sensitive jobs because Spot capacity can be reclaimed with short notice. Therefore, a Compute Savings Plan (3-year) best matches cost optimization with flexibility for growth and changes.

The requirements specify SSH access, no internet exposure, and connection through the AWS Management Console. The AWS-native solution designed specifically for this use case is EC2 Instance Connect Endpoint (EICE).

Option B correctly implements this approach. EC2 Instance Connect Endpoint enables secure SSH access to EC2 instances in private subnets without requiring a bastion host, public IP address, or inbound internet access. Developers authenticate through the AWS Management Console, and the connection is established over HTTPS (port 443), which is why the security group must allow inbound traffic on port 443.

The AmazonEC2InstanceConnect IAM managed policy grants developers permission to push temporary SSH keys to the instance, ensuring short-lived, auditable access that aligns with AWS security best practices. This approach significantly reduces attack surface and operational complexity.

Option A and D incorrectly attempt to use AWS Systems Manager for SSH access on port 22. Systems Manager Session Manager does not use SSH and operates over port 443 without opening inbound ports. Option C is incorrect because EC2 Instance Connect Endpoint does not accept inbound connections on port 22; SSH traffic is tunneled through the endpoint using HTTPS.

Therefore, B is the correct solution because it provides secure, console-based SSH access to private EC2 instances with minimal infrastructure and maximum security.

Amazon S3 is the most cost-effective storage service for large and growing collections of image files, and Amazon CloudFront is the standard way to deliver those objects globally with low latency. AWS documentation states that CloudFront can use an S3 bucket as an origin and cache the objects at edge locations around the world. This reduces latency for viewers and reduces repeated load on the origin. EBS, EFS, and FSx are not the best fit for large-scale internet-facing object delivery because they are not purpose-built global object distribution services. Therefore, S3 plus CloudFront is the correct and most economical architecture for global image delivery at scale.

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

Amazon API Gateway supports multiple API types: REST, HTTP, WebSocket, and gRPC. HTTP APIs are a newer, lightweight option that support JWT authorizers natively, enabling secure, scalable authentication for APIs with JSON Web Tokens.

HTTP APIs can integrate with ALBs as a backend target, providing direct connectivity and simplified API management with JWT authentication built in.

REST APIs support JWT but are more feature-rich and complex, often used for legacy or more complex use cases, and have higher costs and latency. WebSocket APIs are for real-time, bidirectional communication, which is not requested here. gRPC APIs support RPC calls but are less common for public HTTP-based APIs with JWT auth.

Therefore, HTTP API with JWT authorizers is the best fit for this use case.

The correct security-group design is to allow inbound database access from the EC2 instances’ security group rather than from individual IP addresses. AWS security groups can reference another security group as the source, which is the scalable and recommended approach for application-to-database communication in dynamic environments such as Auto Scaling groups. That way, newly launched instances automatically inherit access without requiring IP-based rule updates. Option A is brittle because private IPs change as instances scale. Option C is not how security-group rules are defined, and option D reverses the traffic direction. Therefore, allowing the database security group to accept traffic from the web-servers security group is the right architecture.

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

Key Requirements:

Handle traffic spikes efficiently and reduce latency caused by cold starts.

Optimize costs during low traffic periods.

Analysis of Options:

Option A:

Provisioned Concurrency:Reduces cold start latency by pre-warming Lambda environments for the required number of concurrent executions.

AWS Application Auto Scaling:Automatically adjusts provisioned concurrency based on demand, ensuring cost optimization by scaling down during low traffic.

Correct Approach:Provides a balance between performance during traffic spikes and cost optimization during idle periods.

Option B:

Using EC2 instances with Auto Scaling introduces unnecessary complexity for a serverless architecture. It requires additional management and does not address the issue of cold starts for Lambda.

Incorrect Approach:Contradicts the serverless design philosophy and increases operational overhead.

Option C:

Setting a fixed concurrency level ensures performance during spikes but does not optimize costs during low traffic. This approach would maintain provisioned instances unnecessarily.

Incorrect Approach:Lacks cost optimization.

Option D:

Using EventBridge Scheduler for periodic invocations may reduce cold starts but does not dynamically scale based on traffic demand. It also leads to unnecessary invocations during idle times.

Incorrect Approach:Suboptimal for high traffic fluctuations and cost control.

AWS Solution Architect References:

AWS Lambda Provisioned Concurrency

AWS Application Auto Scaling with Lambda

SAML 2.0-based federation allows organizations to integrate their on-premises Active Directory with AWS, enabling Single Sign-On (SSO) for AWS Management Console and AWS CLI/API access. This lets users continue using their existing AD credentials, removing the need to manage separate identities for AWS and on-premises systems. Mapping roles to AD groups provides granular access control and seamless management.

Reference Extract from AWS Documentation / Study Guide:

" AWS supports SAML 2.0 for federated access, enabling integration with Active Directory or other identity providers to provide single sign-on to AWS resources without creating separate IAM users. "

Source: AWS Certified Solutions Architect – Official Study Guide, Identity and Access Management section.

AWS best practice is to use IAM roles with instance profiles for EC2 instances so that applications obtain temporary credentials automatically and do not need to store access keys.

To honor the principle of least privilege, each microservice should have an IAM role that grants only the specific permissions it needs.

Therefore, creating individual IAM roles per microservice and attaching them via instance profiles (Option D) both minimizes long-term credential management and applies least privilege cleanly.

Why others are not correct:

A: Using IAM users with access keys in code is insecure and high-overhead (key rotation, secret management).

B: A single broad role violates least privilege because every microservice gets more permissions than it needs.

C: Separate accounts per microservice is extreme over-segmentation and significantly increases operational complexity.

For near-real-time streaming workloads that are highly latency-sensitive, the network layer must provide ultra-low latency and high throughput between compute nodes. Elastic Fabric Adapter (EFA) is specifically designed to meet these requirements. EFA enables EC2 instances to communicate using a high-performance network interface that bypasses traditional TCP/IP networking, providing lower and more consistent latency. This is especially valuable for tightly coupled workloads that require fast inter-node communication.

Option B is the correct choice because EFA supports custom networking stacks and is optimized for applications such as real-time data processing, distributed computing, and high-performance workloads. EFA integrates directly with supported EC2 instance types and allows applications to take advantage of enhanced networking capabilities without requiring complex multi-VPC architectures.

Option A introduces additional network hops and latency due to VPC peering and is not suitable for ultra-low-latency workloads. Option C (spread placement groups) is designed to increase fault tolerance by placing instances on separate hardware, but it does not optimize for low-latency communication and may actually increase network distance between instances. Option D improves storage performance, not network performance, and does not affect inter-instance latency.

Therefore, B is the best solution because Elastic Fabric Adapter is the AWS-recommended approach for achieving the lowest possible network latency between EC2 instances in performance-critical streaming architectures.

Using Service Control Policies (SCPs) to enforce permissions boundaries ensures that all created roles are constrained, regardless of attached policies. This is the most secure, scalable, and AWS-recommended preventive control.

Encryption at Rest: AWS Key Management Service (AWS KMS) provides centralized control over the cryptographic keys used to protect data. By using AWS KMS with Amazon S3, you can manage encryption keys and define policies to control access to them.

Encryption in Transit: By enforcing the aws:SecureTransport condition in the S3 bucket policy, you ensure that all data is transmitted over HTTPS, protecting data in transit.

Access Control: IAM policies allow you to define fine-grained permissions for users and roles, ensuring that only authorized entities can access the customer records.

Monitoring: AWS CloudTrail provides a record of actions taken by a user, role, or AWS service in Amazon S3, enabling you to monitor access to the records.

Key Requirements:

Serverless architecture.

Handle traffic bursts with high availability.

Process orders asynchronouslyin the order they are received.

Analysis of Options:

Option A:Amazon SNS delivers messages to subscribers. However, SNS does not ensure ordering, making it unsuitable for FIFO (First In, First Out) requirements.

Option B:Amazon SQS FIFO queues support ordering and ensure messages are delivered exactly once. AWS Lambda functions can be triggered by SQS to process messages asynchronously and efficiently. This satisfies all requirements.

Option C:Amazon SQS standard queues do not guarantee message order and have " at-least-once " delivery, making them unsuitable for the FIFO requirement.

Option D:Similar to Option A, SNS does not ensure message ordering, and using AWS Batch adds complexity without directly addressing the requirements.

AWS References:

Amazon SQS FIFO Queues

AWS Lambda and SQS Integration

AWS documentation states that when Lambda processes events, a dead-letter queue (DLQ) using Amazon SQS is the correct mechanism to capture failed invocations for later reprocessing.

SQS provides durable buffering and decoupling, allowing records to be retried safely without loss.

Kinesis Firehose (Option B) does not support SQL replays or arbitrary replay control. SNS (Option D) is not a durable store. OpenSearch (Option C) is not intended for durable retry queues.

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

Aurora blue/green deployments are specifically designed for:

Safely testing schema changes and database upgrades in an isolated environment (green) without impacting production (blue).

Performing a fast, low-downtime switchover once testing is complete and validated.

In this pattern:

The blue environment is the current production database.

The green environment is a synchronized copy used for testing changes.

You apply schema changes to the green environment, run tests, and when ready, perform a managed switchover that minimizes downtime and risk.

Why others are not ideal:

A: A separate staging cluster allows testing but does not provide automated, low-downtime synchronization and switchover.

B: Aurora read replicas are for read scaling; schema changes on replicas are not supported in the way described, and promotion alone doesn’t solve controlled testing and replication of changes.

D: Moving to DynamoDB changes database engines and data models entirely, and does not match the requirement to keep using Aurora MySQL.

The requirement issecure at-least-once processing, not exactly-once processing. Amazon SQSstandard queuesprovide at-least-once delivery, which matches the requirement and is generally more cost-effective than FIFO queues. When Lambda is configured with an encrypted SQS queue that usesSSE-KMS, AWS requires the Lambda execution role to have thekms:Decryptpermission so Lambda can poll and process messages successfully. FIFO queues add ordering and deduplication features that are not required here, so they are not the most economical choice. Option A is therefore the best fit because it combines the correct delivery model, queue encryption with AWS KMS, and the necessary IAM permission for Lambda.

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

AWS Fargate with Spot capacity is the most cost-effective and serverless container option for short-duration jobs that are flexible on start time. Since the job is not latency-critical and runs nightly, it is a good candidate for Fargate Spot, which can offer up to 70% cost savings over On-Demand pricing.

The job runs for 1 hour, which exceeds the AWS Lambda maximum execution time (15 minutes). Therefore, Lambda is not suitable. EC2-based solutions involve higher operational overhead and cost, making Fargate Spot the best low-cost, low-maintenance solution.

Big data workloads that need to process terabytes of data in memory requirememory-optimized instancesfor the core and task nodes to ensure sufficient memory for processing data efficiently.

Primary Node:Ageneral purpose instanceis suitable because it manages cluster operations, including coordination and monitoring, and does not process data directly.

Core and Task Nodes:These nodes handle data storage and processing.Memory-optimized instancesare ideal because they provide high memory-to-CPU ratios, which is critical for in-memory big data workloads.

Why Other Options Are Incorrect:

Option A:Storage optimized and compute optimized instances are not suitable for workloads that rely heavily on in-memory processing.

Option B:A memory-optimized primary node is unnecessary because the primary node does not process data.

Option D:General purpose instances for all nodes will not provide sufficient memory for processing terabytes of data in memory.

AWS Documentation References:

Amazon EMR Instance Types

Memory-Optimized Instances

This solution offers the most scalable and cost-effective approach with minimal changes to the existing web portal and no code modifications.

Amazon EC2 On-Demand Instances in an Auto Scaling Group: Running the web portal on EC2 On-Demand instances ensures 100% uptime and scalability. The Auto Scaling group will maintain the desired number of instances, automatically scaling up or down as needed, ensuring high availability for the employee web portal.

AWS Lambda for Document Extraction: Lambda is a serverless compute service that allows you to run code in response to events without provisioning or managing servers. By using Lambda to run the document extraction program, you can trigger the function whenever an employee uploads a document. This approach is cost-effective since you only pay for the compute time used by the Lambda function.

No Code Changes Required: This solution integrates with the existing infrastructure with minimal implementation effort and does not require any modifications to the web portal ' s code.

Why Not Other Options?:

Option B (Spot Instances): Spot Instances are not suitable for workloads requiring 100% uptime, as they can be terminated by AWS with short notice.

Option C (Savings Plan): A Savings Plan could reduce costs but does not address the requirement for running the document extraction program efficiently or without code changes.

Option D (S3 with API Gateway and Lambda): This would require significant changes to the existing web portal setup, including moving the portal to S3 and reconfiguring its architecture, which contradicts the requirement of minimal implementation effort and no code changes.

AWS References:

Amazon EC2 Auto Scaling- Information on how to use Auto Scaling for EC2 instances.

AWS Lambda- Overview of AWS Lambda and its use cases.

This is a classic use case for Amazon Kinesis Data Streams + OpenSearch + QuickSight:

Kinesis Data Streams: For real-time ingestion and processing

OpenSearch Service: For fast full-text search, indexing, and analysis

QuickSight: For rich dashboard visualizations

This stack is fully managed, scalable, and native to AWS, minimizing operational overhead.

AWS Lambda function URLs are specifically designed to give a Lambda function a simple direct HTTP(S) endpoint. AWS documentation describes function URLs as a straightforward way to invoke a Lambda function through an HTTP request, and AWS even provides a webhook tutorial built around Lambda function URLs. That makes this the most operationally efficient option because it avoids placing an ALB in front of the function or building an unnecessary messaging layer with SNS or SQS. The requirement is simply to expose the Lambda function so a third party can call it through a webhook. Function URLs are the lightest-weight AWS-native feature for that use case and provide the fastest path from webhook sender to function execution. (AWS Documentation)

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