Design an Inventory Reservation System

To design this system, I would first clarify that we need to support flash sales where thousands of users hit the checkout button simultaneously. The core challenge is preventing overselling while minimizing latency. I propose a hybrid approach using a fast in-memory store like Redis for the reservation logic and a durable database for final state. For the reservation API, the client sends a request with a product ID and quantity. We generate a unique reservation token. Before decrementing stock, we check if sufficient inventory exists. To handle concurrency, we use Redis's Lua scripting to atomically decrement the reserved count only if it doesn't exceed the total. This prevents race conditions better than simple read-modify-write cycles. Crucially, every reservation must have a Time-To-Live (TTL). When a user initiates checkout, we set a 10-minute TTL on the reservation key in Redis. If the user completes payment within this window, we confirm the order and delete the key. If they don't, Redis automatically expires the key, triggering a callback or event to increment the available stock back to its original level. For durability, we write the reservation attempt to an asynchronous queue. A consumer service processes the queue to update the primary database, ensuring we have an audit trail even if the cache fails. Finally, we add a fallback mechanism where a scheduled job scans for stale reservations older than the TTL but not yet released, acting as a safety net for any edge cases.