2 + 2 = 6 and how we fix it: lost updates in Postgres

▤ Summary

This article explains how lost updates happen in Postgres and why transaction isolation levels matter. It compares READ COMMITTED, REPEATABLE READ, and SERIALIZABLE, then shows how each one affects concurrency anomalies and error handling. It walks through practical fixes such as atomic SQL updates, SELECT FOR UPDATE, advisory locks, and optimistic locking with version columns. It also highlights the throughput tradeoffs between correctness and performance in real workloads. The piece ends with a concrete account-transfer example that shows how to avoid race conditions and deadlocks.

▦ Linked Companies