Databases · Flashcard
What is write skew?
- ATwo transactions read overlapping data and then write different rows, leaving a state no serial order allows
- BTwo transactions write the very same row, and the commit that lands second silently discards the first value
- COne transaction reads a row twice and sees two different values because another commit landed in between
- DOne transaction reads rows another has written but not committed, and keeps that result after the rollback
Why this is the answer
In write skew each transaction checks a condition, finds it satisfied, and writes somewhere the other one did not look — individually legal, jointly impossible, which is why snapshot isolation alone cannot catch it. Overwriting the same row is a lost update, seeing a changed value on re-read is a non-repeatable read, and reading uncommitted data is a dirty read, which PostgreSQL never permits.
Official docs