Staff Interview Questions at Amazon

Design a database schema for an e-commerce platform that handles flash sales. During a flash sale, 10,000 users try to buy the last 100 units simultaneously. Your schema must prevent overselling. Show the schema and the critical query.

Design a hotel booking system. Rooms must not be double-booked. Searches should show real-time availability. Handle 10,000 concurrent booking attempts for popular dates.

What is the Saga pattern in microservices? When would you choose choreography over orchestration?

What is a sidecar proxy and how does it work in service mesh architecture?

What is a service mesh and what problem does it solve that a regular load balancer doesn't?

Design a distributed job scheduler like AWS Batch. Jobs have dependencies (job B starts only after job A completes). The system must handle 100K jobs per day.

What is the N-API and why is it important for Node.js native addons? When would you write a native addon?

What is event sourcing and how does it differ from CRUD? What are the trade-offs?

What is the Java memory model and happens-before relationship? Why does it matter for concurrent code?

Design a system where microservice A must call microservices B, C, and D. If any one fails, the entire operation must rollback. B and C have already succeeded when D fails. How do you handle this?

You need Kafka to guarantee exactly-once message processing from producer to consumer, including writing to a database. Explain what you need to configure at each layer.

Design a distributed task scheduler for 10 million scheduled tasks (delayed, recurring, one-time).

Design a system to send 1 billion push notifications per day.

Design an API gateway for 100K req/sec: authenticates, rate limits per user, routes to 20 microservices, adds observability.

Design a distributed counter for 1 million increments per second across 100 servers tracking daily active users.

Explain Python metaclasses. When would you actually use one in production code?

Design a distributed lock service. How does it ensure mutual exclusion across 10 servers?

Given a directed graph representing service dependencies, detect all strongly connected components and explain how this maps to identifying circular dependencies in a microservices architecture.

What is the Strangler Fig pattern for migrating from a monolith to microservices?

Design a video streaming service like Netflix. Focus on: video storage, encoding, CDN delivery.

Your company wants to migrate from a monolith to microservices in 6 months. You're asked to lead this. What's the first thing you tell management?

Your microservices communicate via synchronous REST calls. Service A calls B, B calls C, C calls D. D is slow and times out, causing a cascade failure that takes down A. How do you redesign this to be resilient?

Design a distributed rate limiter that works across 50 geographically distributed servers without a centralized bottleneck. It must allow 1000 requests per user per minute globally.

Your company's monolith handles 10,000 requests/second. The checkout service is the bottleneck. How do you extract it into a microservice without downtime and without breaking existing functionality?

Design WhatsApp's message delivery system. Messages must be delivered exactly once, in order, and the sender must know when the recipient read the message. Handle 50 billion messages per day.

Design Instagram's newsfeed. 500 million users, each follows up to 2000 accounts. Feed must load in under 200ms. How do you architect this?

What is React's Fiber architecture and why was it introduced? How does it enable concurrent features?

Your microservices use HTTP REST. Under peak load, Service A waits synchronously for Service B which waits for Service C. Adding a new feature requires adding Service D to the chain. What's the systemic problem?

Explain the difference between choreography and orchestration in saga pattern. When would you choose each?

Explain the Outbox Pattern. When is it necessary and how does it solve the dual-write problem?

Explain database partitioning vs sharding. When would you use each?