Question 1

A Vietnamese fintech startup is planning to expand across Southeast Asia. They currently use AWS but are considering migrating to GCP or staying. How would you analyze and recommend?

Accepted Answer

Analyze via: (1) Data residency and compliance requirements per SEA country, (2) Latency comparison from nearest AWS/GCP regions (Singapore, Jakarta), (3) Pricing and committed use discounts, (4) Managed service comparison (RDS vs. Cloud SQL, EKS vs. GKE), (5) Migration cost and risk. Strong answers recognize this is not a purely technical question - switching costs (team retraining, IaC refactoring) often outweigh savings. GCP excels in AI/ML (Vertex AI, BigQuery); AWS has a stronger ecosystem and support presence in SEA.

Question 2

You are setting up Infrastructure as Code for a new 5-person backend team in Vietnam, mostly with Python backgrounds. Would you choose Terraform or AWS CDK and why? How would you structure the repository?

Accepted Answer

Consider: (1) Team background and learning curve, (2) Terraform: HCL declarative, broad provider ecosystem, multi-cloud; CDK: Python-native, type-safe, L2/L3 constructs, (3) Module/stack structure (monorepo vs. polyrepo), (4) State management (remote state S3 + DynamoDB locking), (5) CI/CD integration. For a Python team, CDK has an onboarding advantage. Bonus for mentioning: Terragrunt for DRY Terraform, CDK Pipelines for self-mutating pipelines, handling secrets (not hardcoded in IaC), and drift detection.

Question 3

The company cloud bill increased 40% last quarter with no significant traffic growth. The CEO demands a 25% reduction in 30 days. Where do you start?

Accepted Answer

Structured approach: (1) Cost Explorer analysis by service, team, environment - identify top 3 cost drivers, (2) Quick wins: EC2/RDS rightsizing, delete unattached EBS/EIPs, old snapshots, (3) Reserved Instances or Savings Plans for stable workloads, (4) Spot Instances for batch/non-critical, (5) Architecture review: NAT Gateway egress, data transfer costs. Look for candidates who mention: tagging strategy for cost allocation, proactive Budget Alerts, AWS Compute Optimizer, and S3 Intelligent Tiering. Bonus if they ask about SLA tolerance for downtime during optimization.

Question 4

You need to deploy a new microservice version to Kubernetes production with zero downtime, but the new version has a breaking database schema change. How do you plan this?

Accepted Answer

Expand/contract pattern: (1) Deploy migration adding new column (no deletion yet), (2) Deploy v2 alongside v1 (reads new column, falls back to old), (3) Rolling update or Blue-Green via Argo Rollouts, (4) Backfill data, (5) Once v1 is fully drained - drop old column and backward-compat code. Bonus for: PodDisruptionBudget to ensure minimum availability; readiness probes to prevent traffic routing to unready pods; feature flags for quick rollback; database migration tooling (Flyway/Liquibase).

Question 5

Design a serverless architecture for a Vietnamese e-commerce order processing application handling 10,000 orders/day with spikes during major sales events (11/11, Black Friday) reaching 50,000 orders/hour. What serverless limitations must you account for?

Accepted Answer

Design: (1) API Gateway → Lambda → SQS (buffer) → Lambda processor → DynamoDB, (2) Concurrency limits: Lambda default 1,000, requires limit increase, (3) SQS batch size and visibility timeout matching processing time, (4) Dead Letter Queue for failed messages, (5) Idempotency keys to prevent duplicate processing. Limitations to mention: Lambda cold start with Java/C# (use Provisioned Concurrency or switch to Node/Python); Lambda 15-minute timeout; 6MB payload limit via API Gateway; VPC Lambdas have higher latency.

Question 6

You just audited a company AWS account and found multiple IAM users with AdministratorAccess and access keys created 2 years ago. How do you remediate without disrupting operations?

Accepted Answer

Action plan: (1) Inventory via IAM Access Analyzer and CloudTrail - see who actually uses what, (2) Enforce MFA immediately for all human users, (3) Create permission boundaries, (4) Rotate access keys (create new key, update applications, disable old after 30 days, delete), (5) Migrate to IAM Roles + Instance Profiles for EC2/Lambda, (6) Set up AWS Organizations SCPs. Strong candidates mention: principle of least privilege, not using the root account for daily operations, enabling CloudTrail and GuardDuty, and how to communicate change management to affected teams.

Question 7

Design a disaster recovery plan for a Vietnamese digital banking system with RTO = 4 hours and RPO = 15 minutes. The system runs on AWS ap-southeast-1 (Singapore). Cost is a significant factor.

Accepted Answer

Warm Standby strategy: (1) Cross-region replication to ap-southeast-3 (Jakarta) or ap-east-1 (Hong Kong), (2) RDS automated backups + cross-region read replica, (3) S3 Cross-Region Replication, (4) Route 53 health checks + failover routing policy, (5) Infrastructure pre-provisioned at reduced scale (warm), (6) Runbook and quarterly DR drills. With RTO 4h and RPO 15 min, Pilot Light (RTO 1-2h but RPO ~1h) may not meet RPO; Warm Standby (RTO 10-30 min, RPO 1-5 min with read replica) is the best cost/performance fit. Multi-site Active-Active is better but doubles cost.

Question 8

Your company wants to serve users in Vietnam, Thailand, and Indonesia with latency under 100ms. Design a multi-region architecture and explain how you handle distributed data consistency.

Accepted Answer

Design: (1) CloudFront with origins in Singapore + Jakarta, (2) Application tier in ap-southeast-1 (SG) and ap-southeast-3 (Jakarta), (3) Data strategy: classify data by read/write pattern - static data via CDN, session data Redis ElastiCache per-region, transactional data via Aurora Global Database or DynamoDB Global Tables, (4) Eventual vs. strong consistency trade-off per use case. Look for: practical understanding of CAP theorem; awareness that DynamoDB Global Tables have eventual consistency with ~1 second replication lag; using sticky routing (user always routed to the same region) to reduce conflicts.