Oracle does this by writing all the database blocks
changed by those committed transactions to the datafiles on disk. This operation is
called a checkpoint. Completed checkpoints are recorded in the control file, datafile
headers, and redo log.
Rollforward
At any point in time, the online redo logs will be ahead of the datafiles by a certain
amount of time or number of committed transactions. Instance recovery closes this
gap and ensures that the datafiles reflect all committed transactions up to the time
the instance crashed. Oracle performs instance recovery by rolling forward through
the online redo log and replaying all the changes from the last completed checkpoint
to the time of instance failure. This operation is called the rollforward phase of
instance recovery.
While implementing rollforward recovery, Oracle reads the necessary database
blocks into the System Global Area and reproduces the changes that were originally
applied to the blocks. This process includes reproducing the undo or rollback information,
in addition to the data changes.
Pages:
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570