A transaction groups related database operations into a logical unit. It helps prevent partial updates from leaving data in an invalid state.

Example: bank transfer

• Deduct money from Account A.
• Add the same amount to Account B.
• If a required operation fails, roll back the transaction.
• Production implementations must also validate balances, affected rows, and business rules.

ACID

• Atomicity: all transaction operations succeed or the transaction is rolled back.
• Consistency: defined constraints and business invariants are preserved.
• Isolation: concurrent transactions are controlled according to the isolation level.
• Durability: committed changes are designed to survive failures within the database's guarantees.

Isolation levels and anomalies

• Common levels: Read Uncommitted, Read Committed, Repeatable Read, Serializable.
• Dirty read: reading uncommitted changes.
• Non-repeatable read: rereading a row and seeing a different committed value.
• Phantom read: repeating a query and seeing additional or missing rows due to concurrent changes.
• Exact behavior varies by database system.