Recap

Logging is essential for database recovery, ensuring that transactions maintain Atomicity, Consistency, Isolation, and Durability (ACID) properties.

Motivation

• Crash Recovery: Ensure transaction durability and database consistency even in system failures.
• Concurrency Control: Maintain correctness when multiple transactions are executing.
• Database Performance: Balance between write operations and efficient recovery.

Failure Classification

A DBMS must be able to handle different types of failures:

1. Transaction Failures

• Logical Errors: Integrity constraint violations, application errors, etc.
• Internal State Errors: Deadlocks, resource limits, forced rollbacks.

2. System Failures

• Software Failures: Unhandled exceptions, DBMS crashes.
• Hardware Failures: Power loss, server crashes.
• Fail-Stop Assumption: Assumes non-volatile storage remains intact after a system crash.

3. Storage Media Failures

• Non-Repairable Disk Failures: Head crashes, complete data corruption.
• Solution: DBMS cannot recover from this alone; needs backup restoration.