Crash Recovery
- Crash recovery mechanisms guard against
- System crashes
- Media failures
- A log or journal records all changes before they modify the database
- The log is assumed to survive system crashes and media failures
- After a crash the recovery manager follows the ARIES stretegy
- Analysis
- of changes that have not been written and active transactions
- Redo
- actions that were not written to the database
- Undo
- transactions that were not completed
- The log is also updated during the recovery to guard against repeated crashes
- A checkpoint is periodically recorded to the log to reduce the recovery overhead
- A checkpointing strategy is also used to take backup copies of live databases