Transaction Isolation Levels & MVCC

When multiple concurrent transactions execute simultaneously against a database, interleaving operations without isolation causes four classic data corruption anomalies:

1. Dirty Read: Transaction 1 modifies a row without committing. Transaction 2 reads the uncommitted row. Transaction 1 then aborts and rolls back. Transaction 2 has acted on "dirty" phantom data that never legally existed in the database.
2. Non-Repeatable Read (Fuzzy Read): Transaction 1 reads a row (e.g., balance = 100). Transaction 2 updates that row (balance =150) and commits. Transaction 1 reads the exact same row again within the same transaction and sees different data ($150).
3. Phantom Read: Transaction 1 runs a range query: SELECT * FROM users WHERE age > 30 (returns 5 rows). Transaction 2 inserts a brand new user with age 35 and commits. Transaction 1 runs the exact same query again and sees 6 rows (a phantom row appeared!).
4. Write Skew Anomaly: Two concurrent transactions read overlapping data sets, evaluate an invariant (e.g., "On-call doctor rule: at least 1 doctor must be on duty"), modify disjoint rows (Doctor A goes off duty while Doctor B concurrently goes off duty), and both commit, violating the global rule (0 doctors left on duty!).