Databases Reference
In-Depth Information
If you wanted to locate the snapshot controlfile in /commonfs, you would have given:
RMAN> configure snapshot controlfile name to '/commonfs/snapcf_PRODB.ctl';
The technique is the same. The difference lies in whether you begin the target name with a + or with a /,
indicating an ASM group or a mount point respectively.
How It Works
When an RMAN job executes, it has to read the information from the controlfile repeatedly. But the controlfile is a
constantly updated file - a moving target. Therefore RMAN takes a snapshot of the controlfile before starting the job.
This “copy” of the controlfile is known as snapshot controlfile. It is used only for RMAN; nothing else. After the backup
job is complete, RMAN updates the actual controlfile from the snapshot controlfile. Therefore the controlfile is locked
only twice - one during the creation of snapshot controlfile and later during the update from the snapshot to the real
controlfile.
Since the snapshot controlfile is pretty critical in the backup operation, it's important that (a) it's on a highly
performant disk and (b) the RMAN sessions can access it. If you let the RMAN process run on all the nodes of the
cluster, then this file must be visible to all the nodes. The default location of the file is on Oracle Home. You can
confirm that by using the following command:
RMAN> show snapshot controlfile name;
RMAN configuration parameters for database with db_unique_name PRODB are:
CONFIGURE SNAPSHOT CONTROLFILE NAME TO '/ndsint/oradb/db2/dbs/snapcf_PRODB1.f'; # default
The default location, as you can see above, is
$ORACLE_HOME/dbs
directory and the name is
snapcf_<InstanceName>.f
. Normally, Oracle Home is a local file system; so other RAC nodes can't access it.
You should change it to a new location that is common to all the nodes.
Parallelizing a Backup Across All Nodes in a Cluster
(11.2 and Higher)
Problem
In your Oracle 11.2 RAC database you have more than one node, and you want to leverage the full power of all the
nodes during the backup rather than taxing only one node.
Solution
RMAN is an operating system program that is invoked from the command line. Operating systems are specific to a
server (or a node), and hence the program can run on only one node. The RMAN command creates the channels to be
used for backup, and all the channels are created on the same node where the RMAN command is invoked—making
it heavily used while underutilizing the rest of the nodes. Your objective is to spread the channels across all the nodes
of the cluster regardless of the node on which the RMAN command was invoked.
The solution changes beginning in Oracle Database 11.2. That version introduces the Single Client Access Name
(SCAN) feature. In this recipe, we will explain the approach with the SCAN concept. The next recipe shows the pre-11.2
method.
