Database Recovery · Visual IT Encyclopedia

1. Definition

A restoration technique that performs log-based REDO and UNDO to bring a database back to a consistent state after a failure

Log scanned: everything from the start up to the failure

REDO

Target: Transactions that reached COMMIT before the failure (T1, T2, T4)
How: Write the new values recorded in the log again, forward from the first record
Guarantees: Durability: committed results survive the failure

Log scanned: everything from the start up to the failure

UNDO

Target: Transactions still running when the failure happened (T3, T5)
How: Write back the old values recorded in the log, backward from the last record
Guarantees: Atomicity: no half-finished changes stay in the database

Log scanned: everything from the start up to the failure

REDO

UNDO

REDO

UNDO

Writes to the DB file: May write even before COMMIT
After the failure: Committed T1, T2, T4 are redone; T3, T5, still running, are undone

Log scanned: from the checkpoint up to the failure

Skipped

REDO

UNDO

REDO

UNDO

Checkpoint: A mark made after writing every change so far to the DB file, listing the transactions still running
After the failure: T1, finished before the checkpoint, is skipped; T2, T4 are redone; T3, T5 are undone
Dashed range: T3 started before the checkpoint, so the log is read further back to its start record to undo it