System Design: Scalable, Highly Available Inventory Management

Context

Design an inventory management system for a high-traffic e-commerce platform that tracks stock across multiple warehouses and supports real-time purchasing flows. The system must remain correct under high concurrency and flash-sale spikes.

Functional Requirements

• Track inventory by product across multiple warehouses.
• Operations (all idempotent):
  1. Reserve: hold stock for an order with expiry (TTL).
  2. Release: free a reservation (manual or TTL expiry).
  3. Purchase: convert a reservation to a committed order and decrement available stock.
• Prevent overselling under concurrent requests.
• Emit order/inventory events with at-least-once delivery.
• Read endpoints:
  • Get product availability (global total and per-warehouse breakdown).

Non-Functional Requirements

• Flash-sale workload: ~50k RPS reads, ~5k RPS writes.
• High availability and horizontal scalability.
• Idempotent APIs; at-least-once events; exactly-once semantics at service boundaries where feasible.

Specify

• Overall architecture (services, databases, caches, message buses).
• Data model (items, warehouses, reservations, versioning).
• Consistency model (strong vs eventual) with justification.
• Concurrency control and race-condition avoidance strategies (optimistic concurrency, conditional updates, idempotency keys, selective locks/quorum if needed).
• Scaling strategies (read replicas, sharding, write queues, CQRS).
• Cache design and invalidation.
• Handling of partial failures and retries.
• Exactly-once semantics at boundaries.
• Backpressure and rate limiting.
• Observability (metrics, tracing).
• Disaster recovery and multi-region considerations.