Thread-Safe Buy/Sell Without Overselling

A phone-screen coding round: implement buy and sell functions keyed by client, symbol, and quantity while ensuring concurrent requests can never oversell inventory. The follow-up extends the design to multiple buy/sell systems connected to multiple exchanges. Requirements Implement two functions to buy and sell shares of a stock. Each request identifies a client, stock symbol, and quantity. Under concurrent access from multiple threads, the implementation must guarantee the stock is never oversold (position constraints always hold). The expected solution uses locking to protect the shared inventory/position state. Follow-up: discuss how the design changes when k buy/sell systems connect to k exchanges. Notes In one process, keep the position for each (client, symbol) key and perform the sell validation and decrement in the same critical section. A global lock is simplest; per-key or striped locks reduce contention while preserving the invariant. Across k systems, process-local locks no longer coordinate. Route each position key through one authoritative writer or use shared transactional state with an atomic conditional update. Preserve per-key ordering, make requests idempotent, and reconcile rejected or partial exchange fills before releasing reservations. The interviewer probes the concurrency model rather than algorithmic complexity. Preparation Implement a thread-safe inventory with a lock (and discuss lock granularity / atomic alternatives); reason aloud about the race window you're closing. Extend the design to two service instances: explain why local locks fail, choose single-writer partitioning or a transactional store, and walk through duplicate, out-of-order, rejected, and partially filled requests.