Preference Server Failover Session State Replication and Failover Performance Recommendation

14-26 Oracle Fusion Middleware High Availability Guide Figure 14–7 Discoverer Topology for Multiple Instances The following sections discuss specific high-availability considerations relevant for Discoverer. After making configuration changes to provide high availability, you must restart the Oracle HTTP Server and the managed servers in which the Discoverer Java EE applications are deployed.

14.5.2.1 Preference Server Failover

Failure detection and restart of the Discoverer Java EE application in a cluster is managed by WebLogic Server. The Preference Server is a singleton process and runs only one instance in the entire cluster. If the Preference Server process goes down, OPMN brings it up again automatically. If OPMN too is down, the administrator must either start the Preference Server manually or move the files containing the preferences reg_key.dc and pref.txt to another machine where the Preference Server is configured and then start the Preference Server on that machine. If the Preference Server is started in another node in the cluster, the Discoverer Java EE application must be configured to recognize the new Preference Server.

14.5.2.2 Session State Replication and Failover

When a server or machine that is handling requests pertaining to a particular Discoverer HTTP session is down, subsequent request are diverted to other managed servers in the cluster. The session state can be replicated in the new server. The necessary information to achieve this replication is available in the HTTP request as GETPOST. Note: Discoverer requires a Session Server C++ process to be available to complete any request. Therefore, in a failover situation, the response time would be marginally higher because a C++ process must be spawned and brought to the required state. Configuring High Availability for Oracle Portal, Forms, Reports, and Discoverer 14-27

14.5.2.3 Performance Recommendation

The Discoverer Java EE Servlet spawns a Session Server process for each HTTP request that it receives. If multiple Discoverer instances managed server and the associated oracle instance exist in a single machine, the load on the CPU might become very high because all Discoverer instances might spawn C++ processes simultaneously. Adding Discoverer instances on different machines yields better performance than adding them on the same machine.

14.5.2.4 Propagation of Configuration Changes Across the Cluster