"Two users edit the same record at the same time. What happens?" This comes up in backend and system design interviews constantly, often disguised as a ticket-booking or inventory question. The answer interviewers want is a comparison of two concurrency strategies and when you'd choose each.
Alice and Bob both read a product with stock = 5. Alice sells 2 and writes 3. Bob sells 1 and writes 4. Final stock: 4, though 3 units left the warehouse. Nothing crashed; one update silently overwrote the other. Name this problem first: it shows you know what you're protecting against.
Assume conflict is likely and take a lock before reading:
BEGIN; SELECT stock FROM products WHERE id = 42 FOR UPDATE; UPDATE products SET stock = stock - 2 WHERE id = 42; COMMIT;
Bob's SELECT ... FOR UPDATE waits until Alice commits. Correct, but every writer queues behind the lock, and holding locks across slow work (or user think-time) wrecks throughput and invites deadlocks.
Assume conflict is rare. Add a version column, read without locking, and make the write conditional:
UPDATE products SET stock = 3, version = version + 1 WHERE id = 42 AND version = 7;
If zero rows were updated, someone else got there first: reload and retry, or tell the user. No locks are held between read and write, so it scales well and suits long-lived edits like forms or documents.
If-Match) carry the optimistic idea over the network."For inventory I'd start with an atomic conditional update, SET stock = stock - 2 WHERE stock >= 2, which avoids both explicit locks and version checks for this simple case. I'd move to versioned optimistic locking for edits users make over minutes, and to a per-item queue if a single hot item caused heavy retries." It shows you know the third option many candidates miss.
Disclosure: I work on an AI interview assistant at Linkjob AI. This explanation stands on its own.