Backup and Recovery - Oracle Exadata Recipes: A Problem-Solution Approach - page 230

■ the steps in this recipe assume you are using InfiniBand firmware version 1.1.3-2 or later. to restore your
switch configuration from previous firmware versions, please consult your exadata owner's manual.

Note

8-14. Recovering from Loss of Your Oracle Cluster Registry
and Voting Disks
Problem
You have lost your Oracle Cluster Registry on your compute nodes and need to restore your environment to enable
ClusterWare resources to start.

Solution
In this recipe, we will demonstrate how to recover from loss of your Oracle Cluster Registry, or OCR, and your Oracle
RAC voting disks.

Recovering Your OCR
There are some rare occurrences in which restoring your OCR may be necessary; the most common of which is if you
wish to restore your cluster resource definitions to previous values, but you could find yourself in a situation in which
you've lost the Exadata grid disks on which the mirrored versions of the OCR reside.

■ oracle protects and mirrors your oCr inside an oracle aSM disk group. While it is possible to lose a mirror of
your oCr, it is highly unlikely that events such as disk failure would leave your oCr unprotected. It is perhaps more
common for human error to lead to this event.

Note

The first step you should perform is to log in to your compute node as root and determine the location of your
Oracle Cluster Registry by looking at /etc/oracle/ocr.loc :

[root@cm01dbm01 ~]# cat /etc/oracle/ocr.loc
ocrconfig_loc=+DBFS_DG
local_only=FALSE
[root@cm01dbm01 ~]#

The above information tells you that for this Exadata installation, clusterware is expecting the OCR to be located
in an ASM disk group called DBFS_DG , which is common on recent versions of Exadata.

■ If you have physically lost the aSM disk group on which your oCr resided, CrS will fail to start. you should
disable automatic restart of oracle high availability Services by performing a crsctl disable crs as root , rebooting
your nodes, and then running crsctl start crs -excl on one of your nodes. this will start enough CrS services to
perform the remainder of the tasks in this recipe.

Note