RMAN Incremental Backups

RMAN incremental backups back up only datafile blocks that have changed since a specified previous backup. You can make incremental backups of databases, individual tablespaces or datafiles.

The goal of an incremental backup is to back up only those data blocks that have changed since a previous backup.

The primary reasons for making incremental backups part of your strategy are:

• For use in a strategy based on incrementally updated backups, where these incremental backups are used to periodically roll forward an image copy of the database
• To reduce the amount of time needed for daily backups
• To save network bandwidth when backing up over a network
• To get adequate backup performance when the aggregate tape bandwidth available for tape write I/Os is much less than the aggregate disk bandwidth for disk read I/Os
• To be able to recover changes to objects created with the NOLOGGING option. For example, direct load inserts do not create redo log entries and their changes cannot be reproduced with media recovery. They do, however, change data blocks and so are captured by incremental backups.
• To reduce backup sizes for NOARCHIVELOG databases. Instead of making a whole database backup every time, you can make incremental backups.

As with full backups, if you are in ARCHIVELOG mode, you can make incremental backups if the database is open; if the database is in NOARCHIVELOG mode, then you can only make incremental backups after a consistent shutdown.

See Also:

Oracle Database Concepts for more information about NOLOGGING mode

One effective strategy is to make incremental backups to disk, and then back up the resulting backup sets to a media manager with BACKUP AS BACKUPSET. The incremental backups are generally smaller than full backups, which limits the space required to store them until they are moved to tape. Then, when the incremental backups on disk are backed up to tape, it is more likely that tape streaming can be sustained because all blocks of the incremental backup are copied to tape. There is no possibility of delay due to time required for RMAN to locate changed blocks in the datafiles.

4.4.1 Incremental Backup Algorithm

Each data block in a datafile contains a system change number (SCN), which is the SCN at which the most recent change was made to the block. During an incremental backup, RMAN reads the SCN of each data block in the input file and compares it to the checkpoint SCN of the parent incremental backup. If the SCN in the input data block is greater than or equal to the checkpoint SCN of the parent, then RMAN copies the block.

Note that if you enable the block change tracking feature, RMAN can refer to the change tracking file to identify changed blocks in datafiles without scanning the full contents of the datafile. Once enabled, block change tracking does not alter how you take or use incremental backups, other than offering increased performance. See "Improving Incremental Backup Performance: Change Tracking" for more details about enabling block change tracking.

4.4.1.1 Level 0 and Level 1 Incremental Backups

Incremental backups can be either level 0 or level 1. A level 0 incremental backup, which is the base for subsequent incremental backups, copies all blocks containing data, backing the datafile up into a backup set just as a full backup would. The only difference between a level 0 incremental backup and a full backup is that a full backup is never included in an incremental strategy.

A level 1 incremental backup can be either of the following types:

• A differential backup, which backs up all blocks changed after the most recent incremental backup at level 1 or 0
• A cumulative backup, which backs up all blocks changed after the most recent incremental backup at level 0

Incremental backups are differential by default.

Note:

Cumulative backups are preferable to differential backups when recovery time is more important than disk space, because during recovery each differential backup must be applied in succession. Use cumulative incremental backups instead of differential, if enough disk space is available to store cumulative incremental backups.

The size of the backup file depends solely upon the number of blocks modified and the incremental backup level.

4.4.1.2 Differential Incremental Backups

In a differential level 1 backup, RMAN backs up all blocks that have changed since the most recent cumulative or differental incremental backup, whether at level 1 or level 0. RMAN determines which level 1 backup occurred most recently and backs up all blocks modified after that backup. If no level 1 is available, RMAN copies all blocks changed since the level 0 backup.

The following command performs a level 1 differential incremental backup of the database:

RMAN> BACKUP INCREMENTAL LEVEL 1 DATABASE;

If no level 0 backup is available, then the behavior depends upon the compatibility mode setting. If compatibility is >=10.0.0, RMAN copies all blocks changed since the file was created, and stores the results as a level 1 backup. In other words, the SCN at the time the incremental backup is taken is the file creation SCN. If compatibility <10.0.0, RMAN generates a level 0 backup of the file contents at the time of the backup, to be consistent with the behavior in previous releases.

Figure 4-1 Differential Incremental Backups (Default)

Description of "Figure 4-1 Differential Incremental Backups (Default)"

In the example shown in Figure 4-1, the following occurs:

• Sunday

An incremental level 0 backup backs up all blocks that have ever been in use in this database.

• Monday - Saturday

On each day from Monday through Saturday, a differential incremental level 1 backup backs up all blocks that have changed since the most recent incremental backup at level 1 or 0. So, the Monday backup copies blocks changed since Sunday level 0 backup, the Tuesday backup copies blocks changed since the Monday level 1 backup, and so forth.

• The cycle is repeated for the next week.

4.4.1.3 Cumulative Incremental Backups

In a cumulative level 1 backup, RMAN backs up all the blocks used since the most recent level 0 incremental backup. Cumulative incremental backups reduce the work needed for a restore by ensuring that you only need one incremental backup from any particular level. Cumulative backups require more space and time than differential backups, however, because they duplicate the work done by previous backups at the same level.

The following command performs a cumulative level 1 incremental backup of the database:

BACKUP INCREMENTAL LEVEL 1 CUMULATIVE DATABASE; # blocks changed since level 0

Figure 4-2 Cumulative Incremental Backups

Description of "Figure 4-2 Cumulative Incremental Backups"

In the example shown in Figure 4-2, the following occurs:

• Sunday

An incremental level 0 backup backs up all blocks that have ever been in use in this database.

• Monday - Saturday

A cumulative incremental level 1 backup copies all blocks changed since the most recent level 0 backup. Because the most recent level 0 backup was created on Sunday, the level 1 backup on each day Monday through Saturday backs up all blocks changed since the Sunday backup.

• The cycle is repeated for the next week.

4.4.1.4 Basic Incremental Backup Strategy

Choose a backup scheme according to an acceptable MTTR (mean time to recover). For example, you can implement a three-level backup scheme so that a full or level 0 backup is taken monthly, a cumulative level 1 is taken weekly, and a differential level 1 is taken daily. In this scheme, you never have to apply more than a day's worth of redo for complete recovery.

When deciding how often to take full or level 0 backups, a good rule of thumb is to take a new level 0 whenever 50% or more of the data has changed. If the rate of change to your database is predictable, then you can observe the size of your incremental backups to determine when a new level 0 is appropriate. The following query displays the number of blocks written to a backup set for each datafile with at least 50% of its blocks backed up:

SELECT FILE#, INCREMENTAL_LEVEL, COMPLETION_TIME, BLOCKS, DATAFILE_BLOCKS

  FROM V$BACKUP_DATAFILE

  WHERE INCREMENTAL_LEVEL > 0

  AND BLOCKS / DATAFILE_BLOCKS > .5

  ORDER BY COMPLETION_TIME;

Compare the number of blocks in differential or cumulative backups to a base level 0 backup. For example, if you only create level 1 cumulative backups, then when the most recent level 1 backup is about half of the size of the base level 0 backup, take a new level 0.

4.4.2 Making Incremental Backups: BACKUP INCREMENTAL

After starting RMAN, run the BACKUP INCREMENTAL command at the RMAN prompt. This example makes a level 0 incremental backup of the database:

BACKUP INCREMENTAL LEVEL 0 DATABASE;

This example makes a differential level 1 backup of the SYSTEM tablespace and datafile tools01.dbf. It will only back up those data blocks changed since the most recent level 1 or level 0 backup:

BACKUP INCREMENTAL LEVEL 1

  TABLESPACE SYSTEM

  DATAFILE 'ora_home/oradata/trgt/tools01.dbf';

This example makes a cumulative level 1 backup of the tablespace users, backing up all blocks changed since the most recent level 0 backup.

BACKUP INCREMENTAL LEVEL = 1 CUMULATIVE

  TABLESPACE users;

4.4.3 Incrementally Updated Backups: Rolling Forward Image Copy Backups

Oracle's Incrementally Updated Backups feature lets you avoid the overhead of taking full image copy backups of datafiles, while providing the same recovery advantages as image copy backups.

At the beginning of a backup strategy, RMAN creates an image copy backup of the datafile. Then, at regular intervals, such as daily, level 1 incremental backups are taken, and applied to the image copy backup, rolling it forward to the point in time when the level 1 incremental was created.

During restore and recovery of the database, RMAN can restore from this incrementally updated copy and then apply changes from the redo log, with the same results as restoring the database from a full backup taken at the SCN of the most recently applied incremental level 1 backup.

A backup strategy based on incrementally updated backups can help minimize time required for media recovery of your database. For example, if you run scripts to implement this strategy daily, then at recovery time, you never have more than one day of redo to apply.

4.4.3.1 Incrementally Updated Backups: A Basic Example

To create incremental backups for use in an incrementally updated backups strategy, you must use the BACKUP... FOR RECOVER OF COPY WITH TAG form of the BACKUP command. How the command works is best understood in the context of an example script that would implement the strategy.

This script, run on a regular basis, is all that is required to implement a strategy based on incrementally updated backups:

RUN {

   RECOVER COPY OF DATABASE WITH TAG 'incr_update';

   BACKUP INCREMENTAL LEVEL 1 FOR RECOVER OF COPY WITH TAG 'incr_update'

       DATABASE;

   }

The syntax used in the script does not, however, make it clear how the strategy works. To understand the script and the strategy, it is necessary to understand the effects of these two commands when no datafile copies or incremental backups exist.

• The BACKUP INCREMENTAL LEVEL 1... FOR RECOVER OF COPY WITH TAG... command does not actually always create a level 1 incremental backup. If there is no level 0 image copy backup of an particular datafile, then executing this command creates an image copy backup of the datafile on disk with the specified tag instead of creating the level 1 backup.

Note:

Even when the BACKUP INCREMENTAL LEVEL 1 ... FOR RECOVER OF COPY command is used with DEVICE TYPE SBT to create a backup on tape, the first time it is used it creates the image copy on disk, and does not write any backup on tape. Subsequent incremental level 1 backups can be created on tape once the image copy is on disk.

Thus, the first time the script runs, it creates the image copy of the datafile needed to begin the cycle of incremental updates. In the second run and all subsequent runs, it produces level 1 incremental backups of the datafile.

• The RECOVER COPY OF DATABASE WITH TAG... command causes RMAN to apply any available incremental level 1 backups to a set of datafile copies with the specified tag.

If there is no incremental backup or no datafile copy, the command generates a message but does not generate an error.

The first time the script runs, this command has no effect, because there is neither a datafile copy nor a level 1 incremental backup.

The second time the script runs, there is a datafile copy (created by the first BACKUP command), but no incremental level 1 backup, so again, the command has no effect.

On the third run and all subsequent runs, there is a datafile copy and a level 1 incremental from the previous run, so the level 1 incremental is applied to the datafile copy, bringing the datafile copy up to the checkpoint SCN of the level 1 incremental.

Note also the following details about how this example works:

• Each time a datafile is added to the database, an image copy of the new datafile is created the next time the script runs. The time after that, the first level 1 incremental for that datafile is created, and on all subsequent runs the new datafile is processed like any other datafile.
• Tags must be used to identify the incremental level 0 datafile copies created for use in this strategy, so that they do not interfere with other backup strategies you implement. If you have multiple incremental backup strategies in effect, RMAN cannot unambiguously create incremental level 1 backups unless you tag level 0 backups.

The incremental level 1 backups to apply to those image copies are selected based upon the checkpoint SCNs of the image copy datafiles and the available incremental level 1 backups. (The tag used on the image copy being recovered is not a factor in the selection of the incremental level backups.)

In practice, you would schedule the example script to run once each day, possibly at midnight. On a typical night (that is, after the first two nights), when the script completed the following files would be available for a point-in-time recovery:

• An image copy of the database, as of the checkpoint SCN of the preceding run of the script, 24 hours earlier
• An incremental backup for the changes since the checkpoint SCN of preceding run
• Archived redo logs including all changes between the checkpoint SCN of the image copy and the current time

If, at some point during the following 24 hours, you need to restore and recover your database from this backup, for either complete or point-in-time recovery, you can restore the datafiles from the incrementally updated datafile copies, and apply changes from the most recent incremental level 1 and the redo logs to reach the desired SCN. At most, you will have 24 hours of redo to apply, which limits how long point-in-time recovery will take.

See Also:

Oracle Database 2 Day DBA to see how this technique is used in the Oracle-suggested backup strategy in Enterprise Manager.

4.4.3.2 Incrementally Updated Backups: A One Week Example

The basic example can be extended to provide fast recoverability to a window greater than 24 hours. Alter the RECOVER COPY... WITH TAG to perform incomplete recovery of the datafile copies to the point in time in the past where you want your window of recoverability to begin. This example shows how to maintain a seven day window:

RUN {

   RECOVER COPY OF DATABASE WITH TAG 'incr_update'

       UNTIL TIME 'SYSDATE - 7';

   BACKUP INCREMENTAL LEVEL 1 FOR RECOVER OF COPY WITH TAG 'incr_update'

       DATABASE;

   }

The effect of the script is as follows:

• On the first night the RECOVER COPY... UNTIL TIME statement has no effect, and the BACKUP INCREMENTAL... FOR RECOVER OF COPY statement creates the incremental level 0 copy.
• On the second through seventh nights, the RECOVER COPY... UNTIL TIME statement has no effect because TIME 'SYSDATE - 7' is still a time in the future. The BACKUP INCREMENTAL... FOR RECOVER OF COPY statement creates differetial incremental level 1 backups containing the block changes for the previous day.
• On the eighth and all subsequent nights night, the RECOVER COPY... UNTIL TIME statement applies the level 1 incremental from seven days ago to the copy of the database. The BACKUP INCREMENTAL... FOR RECOVER OF COPY statement creates an incremental backup containing the changes for the previous day.

As with the basic example, you have fast recoverability to any point in time between the SCN of the datafile copies and the present, using block changes from the incremental backups and individual changes from the redo logs. Because you have the daily level 1 incrementals, you still never need to apply more than one day of redo.

4.4.4 Improving Incremental Backup Performance: Change Tracking

RMAN's change tracking feature for incremental backups improves incremental backup performance by recording changed blocks in each datafile in a change tracking file. If change tracking is enabled, RMAN uses the change tracking file to identify changed blocks for incremental backup, thus avoiding the need to scan every block in the datafile.

After enabling change tracking, the first level 0 incremental backup still has to scan the entire datafile, as the change tracking file does not yet reflect the status of the blocks. Subsequent incremental backup that use this level 0 as parent will take advantage of the change tracking file.

Using change tracking in no way changes the commands used to perform incremental backups, and the change tracking files themselves generally require little maintenance after initial configuration.

Change tracking is disabled by default, because it does introduce some minimal performance overhead on your database during normal operations. However, the benefits of avoiding full datafile scans during backup are considerable, especially if only a small percentage of data blocks are changed between backups. If your backup strategy involves incremental backups, then you should enable change tracking.

One change tracking file is created for the whole database. By default, the change tracking file is created as an Oracle managed file in DB_CREATE_FILE_DEST. You can also specify the name of the block change tracking file, placing it in any location you choose.

Note:

In a Real Applications Clusters (RAC) environment, the change tracking file must be located on shared storage accessible from all nodes in the cluster.

Oracle saves enough change-tracking information to enable incremental backups to be taken using any of the 8 most recent incremental backups as its parent.

Although RMAN does not support backup and recovery of the change-tracking file itself, if the whole database or a subset needs to be restored and recovered, then recovery has no user-visible effect on change tracking. After the restore and recovery, the change tracking file is cleared, and starts recording block changes again. The next incremental backup after any recovery is able to use change-tracking data.

4.4.4.1 Enabling and Disabling Change Tracking

You can enable or disable change tracking when the database is either open or mounted. To alter the change tracking setting, you must use SQL*Plus to connect to the target database with administrator privileges.

To store the change tracking file in the database area, set DB_CREATE_FILE_DEST in the target database. Then issue the following SQL statement to enable change tracking:

SQL> ALTER DATABASE ENABLE BLOCK CHANGE TRACKING;

You can also create the change tracking file in a location you choose yourself, using the following SQL statement:

SQL> ALTER DATABASE ENABLE BLOCK CHANGE TRACKING

USING FILE '/mydir/rman_change_track.f' REUSE;

The REUSE option tells Oracle to overwrite any existing file with the specified name.

To disable change tracking, use this SQL statement:

SQL> ALTER DATABASE DISABLE BLOCK CHANGE TRACKING;

If the change tracking file was stored in the database area, then it is deleted when you disable change tracking.

4.4.4.2 Checking Whether Change Tracking is Enabled

From SQL*Plus, you can query V$BLOCK_CHANGE_TRACKING.STATUS to determine whether change tracking is enabled, and if it is, query V$BLOCK_CHANGE_TRACKING.FILENAME to display the filename.

4.4.4.3 Moving the Change Tracking File

If you need to move the change tracking file, the ALTER DATABASE RENAME FILE command updates the control file to refer to the new location. The process outlined in this section describes how to change the location of the change tracking file while preserving its contents.

To relocate the change tracking file:

1. If necessary, determine the current name of the change tracking file:

2.  SELECT filename

3.  FROM V$BLOCK_CHANGE_TRACKING;

4.   

5. Shut down the database. For example:

6.  SHUTDOWN IMMEDIATE

7.   

8. Using host operating system commands, move the change tracking file to its new location.
9. Mount the database and move the change tracking file to a location that has more space. For example:

10.ALTER DATABASE RENAME FILE     'ora_home/dbs/change_trk.f' TO '/new_disk/change_trk.f';

11. 

12. Open the database:

13.ALTER DATABASE OPEN;

14. 

If you cannot shut down the database, then you must disable change tracking and re-enable it at the new location, as in the following example:

ALTER DATABASE DISABLE BLOCK CHANGE TRACKING;

ALTER DATABASE ENABLE BLOCK CHANGE TRACKING USING FILE 'new_location';

If you choose this method, you will lose the contents of the change tracking file. Until the next time you complete a level 0 incremental backup, RMAN will have to scan the entire file.

4.4.4.4 Estimating Size of the Change Tracking File on Disk

The size of the change tracking file is proportional to the size of the database and the number of enabled threads of redo. The size is not related to the frequency of updates to the database. Typically, the space required for block change tracking is approximately 1/30,000 the size of the data blocks to be tracked. Note, however, the following two factors that may cause the file to be larger than this estimate suggests:

• To avoid overhead of allocating space as your database grows, the change tracking file size starts at 10MB, and new space is allocated in 10MB incremenents. Thus, for any database up to approximately 300GB the file size is no smaller than 10MB, for up to approximately 600GB the file size is no smaller than 20MB, and so on.
• For each datafile, a minimum of 320K of space is allocated in the change tracking file, regardless of the size of the file. Thus, if you have a large number of relatively small datafiles, the change tracking file is larger than for databases with a smaller number of larger datafiles containing the same data.