Database recovery

Database recovery is a DBMS function that, in the event of logical and physical failures, brings the database to an up-to-date and consistent state.

In the event of a logical failure or a rollback signal of one transaction , the change log is scanned in the opposite direction, and all records of the canceled transaction are extracted from the log up to the mark of the start of the transaction. According to the extracted information, actions are performed that cancel the transaction. This process is called rollback.

In the event of a physical failure, if neither the change log nor the database itself is corrupted, the rollforward process is performed. The log is scanned forward, starting from the previous control point. All entries are retrieved from the journal until the end of the journal. Information extracted from the journal is entered into data blocks of external memory, in which the change number mark is less than that recorded in the journal. If a failure occurs during the sweep process, the log scan will start again from the beginning, but the recovery will actually continue from the point where it was interrupted.

In the event of a physical failure, if the change log is available, but the database itself is damaged, then the process of restoring the database from the backup should be performed. After recovery, the database will be in the state at the time of the backup. To restore the database at the time of failure, you must run all changes using the change log .

In the event of a physical failure, if the change log is not available, but the database itself is not damaged, recovery is possible only at the time of the previous checkpoint.

In the event of a physical failure, if both the change log and the database itself are damaged, then restoration is possible only at the time of the backup.