Modifying Physical Cluster Resources in a Compaq Parallel Database Cluster
4
ECG001/1298
•
When an existing array is removed from the cluster’s shared storage, if the array was not the
last one brought online, the order of the drives will change for all cluster nodes.
Preventing the Issue
It is highly recommended that a power-on sequence for the shared storage arrays be determined
during the initial configuration of the cluster. Thereafter, this power-on sequence must be
followed exactly to ensure that each node sees the disk in a proper order in the Windows NT Disk
Administrator utility. When adding shared storage (a physical drive or an array), make it the last
drive (or array) to be brought online. In the case of adding an array, make sure it is the last array
to be powered-on; and add it to the power-on sequence as such.
On each node use disk administrator to verify that the disks are brought on-line in the expected
order before starting the Oracle instance. Also, verify that the PGMS service is set to start
manually.
Simplifying Detection of the Issue
A method that simplifies detection of the issue is to create logical partitions that are not all the
same size. With different-sized logical partitions, disk administrator can be used to see the size of
the partitions, which can be quickly and easily compared against the sizes written in the suggested
Oracle worksheet. If the drive order in the worksheet does not compare with the drive order
shown in disk administrator, the drives were brought online out of sequence.
For example, the shared disks associated with the database could have a small unused partition as
the first partition on the disk. Assuming there are three disks associated with the database, a
sample configuration might be:
•
Harddisk1 Partition1 is sized to 11 MB
•
Harddisk2 Partition1 is sized to 12 MB
•
Harddisk3 Partition1 is sized to 13 MB
When the cluster is powered on, if disk administrator shows harddisk1, partition1 as 12MB or
13MB, the drives were brought on-line out of sequence. The cluster should be shut down and
restarted correctly.
What Oracle Component is Affected?
How does the drive ordering issue manifest itself in Oracle8 Server? The focal point of this issue
resides in the Oracle symbolic link table files; frequently identified as ORALINKx.TBL. These
files create symbolic links, which are used by Oracle8 Server to map specific Oracle database
files to specific hard drive partitions. A link is required for each database file. A two node cluster
requires nine database files and one cluster file, and therefore requires ten links. The first
symbolic link table file is used by a two-node cluster. Each additional node requires two more
database files, two more links, and one more symbolic link table file .
Oracle suggests using a worksheet to define the symbolic links. For more information about the
worksheet, refer to the
Oracle Parallel Server Getting Started Release 8.0.5 for Windows NT
manual
. An example of such a worksheet follows. The example assumes the cluster consists of
two nodes – Node1 has two internal disk drives and Node2 has one internal disk drive.
Additionally, it assumes the shared storage consists of two fibre channel arrays - the first array
has three RAID logical drives and the second array has two RAID logical drives. Therefore,
