Backend systems · concurrency
Concurrency
Control Lab
Three requests read the same record, then write in sequence. Watch a last-write-wins policy lose updates—and see optimistic locking reject stale writes before retrying them safely.
Shape the race
Each request computes its change from version 0. Change order to see which stale write wins.
Run result
Final balance
—Commit trace
CommitConflict / overwriteRecomputed retry
Why versions matter
A version check turns silent data loss into an explicit conflict. Retrying is safe only when the operation can be recomputed against fresh state; production APIs also need transaction boundaries, idempotency, bounded retries, and a clear conflict response.