AWS Solutions Architect Interview Questions

Architecture Patterns

1. What are the 6 pillars of the AWS Well-Architected Framework?

1. Operational Excellence: DevOps principles, monitoring, automation.
2. Security: IAM, data protection, detecting anomalies.
3. Reliability: Recovery planning, handling failure, scaling.
4. Performance Efficiency: Right sizing, using serverless/managed services.
5. Cost Optimization: Analyzing spend, managed services, spot instances.
6. Sustainability: Minimizing carbon footprint, maximizing utilization.

2. Design a High Availability (HA) structure for a database.

Use RDS Multi-AZ. * Master DB in AZ-A. * Standby Replica in AZ-B. * Synchronous replication ensures zero data loss. * Automatic DNS failover if Master dies. For Read Scalability, add Read Replicas (Async).

3. When to use DynamoDB vs RDS?

• DynamoDB: Non-relational (NoSQL). Flexible schema. Massive scale (TB/PB). Single-digit ms latency. Stateless apps (Session store, Cart, Gaming).
• RDS: Relational. Structured schema. ACID compliance required. Complex joins/queries. Legacy apps.

Storage & Cost

4. Explain Storage Tiering in S3.

• Standard: Hot data.
• Intelligent-Tiering: Auto-moves data based on access patterns (Unknown patterns).
• Standard-IA: Accessed less than once a month.
• One Zone-IA: Cheaper, but data lost if AZ fails.
• Glacier Instant Retrieval: Ms access, for archive.
• Glacier Flexible Retrieval: Minutes/Hours.
• Glacier Deep Archive: 12-48 hours. Cheapest.

5. Strategy for Cost Optimization?

• Compute: RIs/Savings Plans for steady state, Spot for batch.
• Storage: Lifecycle policies to Glacier.
• Network: Use VPC Endpoints (avoid NAT Gateway charges). CloudFront for egress (cheaper than EC2 egress).
• Governance: AWS Budgets, Cost Explorer, Trusted Advisor.

Intermediate/Advanced

6. What is the "Strangler Fig" pattern in migration?

Incremental migration strategy. * Put a Proxy/ELB in front of the legacy monolith. * Build new microservices for specific features. * Route traffic for those features to new services. * Gradually "strangle" the monolith until it does nothing. Decommission it.

7. How to handle "Session State" in a stateless architecture?

Do not store sessions in EC2 memory (Sticky sessions limit scaling). Store session data in an external fast store: ElastiCache (Redis) or DynamoDB.

8. Explain a "Fan-out" architecture using SNS/SQS.

User uploads photo -> S3 event -> SNS Topic. SNS Topic subscribers: * SQS Queue A -> Lambda (Resize Image). * SQS Queue B -> Lambda (Update Search Index). * SQS Queue C -> Lambda (Send Email). Allows parallel, decoupled processing of a single event.

9. How do you design for Disaster Recovery with RPO < 15 mins and RTO < 1 hour?

Strategy: Warm Standby. * Data: Cross-Region Replication (CRR) for S3 and Aurora Global Database (Latency < 1s). * Compute: Minimal fleet running in DR region (Auto Scaling Group min=1). * Failover: Route 53 Health Checks fail over to DR region. ASG scales up.

10. Compare Interface VPC Endpoint vs Gateway Endpoint.

• Gateway: S3/DynamoDB only. Route Table entry. Free. Public IP routing logic (but stays on backbone).
• Interface: Most AWS services. ENI with Private IP. Costs money ($/hr). Supports Security Groups.

11. What is Global Accelerator vs CloudFront?

• CloudFront: Content Delivery Network (CDN). Caches HTTP/HTTPS content (static/dynamic). Best for websites/video.
• Global Accelerator: TCP/UDP proxy using AWS Backbone. No caching. Provides static IP. Best for non-HTTP protocols (Gaming, MQTT, VoIP) or where caching isn't useful but low latency implementation is required.

12. How to ensure idempotency in API design?

Client sends a unique idempotency-key in the header. Server checks (in DynamoDB/Redis) if this key was processed. * If yes: Return saved result (don't re-process payment). * If no: Process and save result + key.

13. What involves "Event Sourcing"?

Instead of storing just the current state (Bank Balance: $100), store the sequence of events (Deposit $50, Withdraw $20). * Benefit: Audit trail, replayability, time travel debugging. * AWS: Kinesis Data Streams or EventBridge or DynamoDB Streams.

14. Design a Multi-Tenant SaaS solution (Isolation vs Pool).

• Silo (Isolation): Separate Account or VPC per tenant. High security, easy billing. High cost/management.
• Pool (Services): Shared Resources. TenantID in DynamoDB PK. High efficiency. Harder security (must ensure code filters by TenantID).

15. How to secure a Lambda function accessing RDS?

• Network: Lambda in VPC.
• Access: SG allows Lambda to RDS port.
• Auth: Use IAM Database Authentication (Token) instead of Password.
• Secrets: If password needed, fetch from Secrets Manager.

16. What is "CloudFront Origin Access Control (OAC)"?

Security feature that restricts S3 bucket access only to CloudFront. S3 Bucket Policy explicitly allows the CloudFront Service Principal. Users cannot bypass CDN to access S3 directly. (OAC replaces legacy OAI).

17. Architecture for real-time leaderboards?

Redis (ElastiCache) Sorted Sets. Extremely fast (in-memory) ranking and sorting. DynamoDB is good, but Redis Sorted Sets are O(log(N)) for updates/retrieval of rank.

18. How to handle 504 Gateway Timeout from API Gateway?

504 means the upstream (Lambda/HTTP) didn't respond within 29 seconds (API Gateway hard limit). * Fix: Optimize code. * Async Pattern: If job takes > 29s, switch to async. API returns "202 Accepted". Lambda processes in background and updates DB/sends webhook.

19. Explain AWS Outposts.

Hybrid cloud service. AWS ships a physical rack of servers to your on-premise datacenter. It runs AWS infrastructure (EC2, EBS, RDS) but connects back to Region. Use for ultra-low latency requirements (Factory floor, Hospital).

20. When to use Kinesis Data Firehose vs Data Streams?

• Streams: You write custom code (Lambda/EC2) to analyze data in real-time (< 1s latency). Complex.
• Firehose: You just want to "load" data into S3/Redshift/Splunk. Managed delivery. Zero code. Latency ~60s.