Redo Log
The redo log records after-images of modifications; crash recovery replays committed-but-not-flushed changes to guarantee durability.
Purpose
To understand the redo phase of crash recovery and analyze database durability guarantees and performance.
Mechanism
On commit, the redo log must be fsynced to disk (durability). Data pages may flush lazily (no-force). After a crash, scan the redo log, replay committed transactions (redo), and roll back uncommitted ones via undo.
Use cases
- Crash recovery design
- Durability guarantee
- WAL implementation