Building Disaster Recovery Serviceguard Solutions Using Metrocluster with 3PAR Remote Copy

B Configuring Oracle RAC in SADTA
Overview of Metrocluster for RAC
The Oracle RAC database can be deployed in a Metrocluster environment for disaster tolerance
using SADTA. This configuration is referred to as Metrocluster for RAC. In this architecture, a
disaster tolerant RAC database is configured as two RAC databases that are replicas of each
other; one at each site of the Metrocluster. At any given time in a Metrocluster, the RAC database
at only one site is up and actively services clients while the other RAC database, which is a replica
on the remote site, remains passive. The active RAC database data I/O is continuously replicated
(synchronously or asynchronously) to the remote site using 3PAR Remote Copy.
When the active RAC database fails, or when the site hosting the active RAC database is lost in
a disaster, Metrocluster automatically initiates a site failover for the RAC database. A Metrocluster
site failover activates the passive database configuration at the remote site by starting it using the
replicated data in that site. After a successful site failover, the redundant RAC passive database
becomes the new active RAC database.
While configuring replicas of the RAC databases within a Metrocluster, the two sites in the
Serviceguard cluster must be configured by grouping cluster nodes based on the sites they are
located in. If you have CVM or CFS configured in your environment, create an SGeRAC-enabled
Oracle Clusterware sub-cluster and a Serviceguard Storage Management Suite CFS sub-cluster at
each site. Only nodes within a Serviceguard site can be members of these clusters.
For every site aware disaster tolerant RAC database, a RAC database must be configured at each
site using the Oracle Clusterware sub-cluster of the site. The database at each site uses the CFS
sub-cluster file systems created over the local disk of a replicated disk pair.
The RAC database processes, the disk groups, and file systems at each site are configured in a
stack of inter-dependent MNP packages. The RAC database processes are packaged using the
SGeRAC Toolkit (delivered as part of the SGeRAC product). The CVM DG MNP and CFS MP MNP
packages are created using SG SMS commands for the disk groups and cluster file systems that
are used to store the database.
For more information on using SG SMS commands to create disk groups and cluster file systems,
see the documents available at http://www.hp.com/go/hpux-serviceguard-docs ->HP Serviceguard
Storage Management Suite <version>. Also, see the Serviceguard, SGeRAC, and SMS Compatibility
and Feature Matrix and the latest release notes for your version of Serviceguard for up-to-date
information about support for CVM and CFS, available at: http://www.hp.com/go/
hpux-serviceguard-docs.
For SADTA, a Site Controller Package must be configured to provide robust site failover semantics
for a site aware disaster tolerant RAC database. The Site Controller Package starts the configured
local RAC MNP stack packages on the site where it is started. The Site Controller Package monitors
the started RAC MNP stack packages. When these packages fail, the Site Controller Package fails
over to the remote site. As part of its startup on the remote site node during failover, the Site
Controller Package prepares the replicated data storage and runs the passive RAC MNP stack
packages in the remote site ensuring disaster tolerance for the database.
A disaster tolerant RAC database has two identical but independent RAC databases configured
over the replicated storage in a Metrocluster, therefore packages of both sites RAC MNP stacks
must not be up and running simultaneously. If the packages of the redundant stack at both sites
are running simultaneously, it leads to data corruption. SADTA provides a Site Safety Latch
mechanism at the site nodes that prevents inadvertent simultaneous direct startup of the RAC MNP
stack packages at both sites.
Multiple site aware disaster tolerant RAC databases can be configured in a Metrocluster. Figure 11
shows one such configuration with two site aware disaster tolerant RAC databases: hrdb and
salesdb.
74 Configuring Oracle RAC in SADTA