Software Engineer Interview Questions & Model Answers

For a distributed rate limiter handling millions of requests per second, I recommend the Sliding Window Counter algorithm backed by Redis. To handle distributed clock drift and concurrent race conditions, execute Lua scripts atomically in Redis. If high availability is prioritized over strict limit compliance, implement local in-memory token buckets with periodic asynchronous synchronization to a central Redis cluster.

Pessimistic locking explicitly locks the target database row (`SELECT ... FOR UPDATE`) upon read, blocking concurrent transactions until the transaction commits. It is ideal for heavy write collision scenarios like inventory reservations. Optimistic locking avoids database-level locks; instead, it checks a record version number before updating. If the version changed, the transaction aborts and retries. Optimistic locking is vastly superior for high-read, low-contention web applications.

In a Node.js microservice handling 4,000 requests/sec, our servers were experiencing gradual memory exhaustion and crashing every 36 hours. Instead of randomly changing code, I took V8 heap snapshots in staging and analyzed memory allocation timelines in Chrome DevTools. I discovered an uncleaned event listener on our WebSocket connection handler holding references to disconnected socket objects. After refactoring the listener lifecycle and running a 12-hour soak test, memory consumption stabilized at a flat 320MB, eliminating all production pod restarts.