3.1 KiB
Undoing & Redoing Transactions
Deferred Update Protocol
Updates not written until after transaction has reached commit point. If transaction fails before committing, no modification is made to database, hence no undoing is required. May be necessary to redo updates of committed transactions as effect may not have reached database.
Procedure
Redo transaction is both <T1 start> and <T1 commit> are present in the log
Example
Failure in different places:
In case(a), no commit is present, so no redo or undo is required.
case(b), T0 commits, but T1 does not, hence T0 is redone.
case(b), T0 and T1 commits, so both are redone.
Immediate Update Protocol
Updates applied as they occur. Following a failure, redo updates of committed transactions. Undo effects of transactions uncommitted. If no commit record, transacation was active and must be undone. Undo operations performed in reverse order in log. In recovery, use log to undo or redo.
Protocol
Undo
- If <Ti,start> is in the log but <Ti,commit> is not then restore the value of all data items updated by Ti to their old values, going backwards from the last log record for Ti. Redo
- If <Ti,start> and <Ti,commit> are both in the log then set the value of all data items updated by Ti to the new values, going forward from the first log record for Ti
Example
Checkpoints
After failure, may not know how far back to search log. Checkpoints limit log searching, made automatically by DBMS Creation of checkpoints schedules at predetermined intervals. Definition: Checkpoints are points of sync between database and log file. All buffers are written to secondary storage.
Procedure
- Write to secondary storage
- All log records
- All modified buffer blocks to database.
- Checkpoint record to log file which contains identities of transactions active at time of checkpoint
- Resume processing.
Example 1 ( Immediate )
- Starts at t0, fails at tf.
- T1 and T6 have to be undone
- In absence of other info, recovery manager has to redo T2, 3, 4, and 5.
Example 2 ( serial with checkpoint, Immediate )
- Dotted lines = checkpoints
- T1 is safe since updates are written to disk
- T4 needs to be undone
- T2 and 3 need to be redone
Example 3 ( concurrent with checkpoint, Immediate )
- T2 and 3 safe since written to disk
- T1 and 6 need to be undone.
- Since checkpoint, redo T4 and 5.
Purpose of Recovery Log File
- Scan log backwards
- Create undo / redo lists
- Undo List: Transactions active at time of crash
- Perform undo(T) for every transaction in list
- Stop when reaching <T, start> for every transaction in list.
- Redo List: Transactions committed after last checkpoint
- Redo for each transaction in list.
Example of Log Use ( Immediate )
