A lost update occurs when two transactions read the same value, calculate different updates, and only one write survives. The discarded write can leave shared data incorrect. section 6.9.7
Bank account balance
Alice’s balance starts at $0. T1 adds $100 and T2 adds $200. If both read $0 before writing, T2 may leave a final balance of $200, losing T1’s addition instead of producing the expected $300. section 6.9.7
Snapshot isolation uses a consistent snapshot from the transaction’s start and can detect conflicting writes to the same data, typically rolling one transaction back. Repeatable reads also prevent lost updates, but through a different mechanism involving locks on read data. section 6.9.7
When choosing isolation guarantees:
- Treat lost updates as a data-correctness risk when concurrent operations are expected to accumulate rather than replace one another. section 6.9.7
- Compare stronger guarantees such as serializability against their potential performance cost in high-concurrency systems. section 6.9.9


