Oracle Installer Process

You also need to know that the installer process will run a SQL script in $ORACLE_HOME/rdbms/admin/catSNMP.sql that creates an SNMP agent role called SNMPAGENT to access the v$ tables. To manage the RAC system with the help of the OEM console and the management server requires a certain amount of discipline at the RAC nodes and instances. The steps to configure such requirements will be covered briefly with the help of an example.

This example assumes a cluster with two node clusters, with node names RAC1 and RAC2.

db_name=V920  # database name
service_name=V920
db_domain=acme.com
Instance on host RAC1 : V9201
Instance on host RAC2 : V9202

Configuration Steps
Ensure that the /var/opt/oracle/oratab file exists and it has an entry for the database (not for the instance) on each of the nodes.

V920:/u01/app/oracle/product/920:N

Check that both nodes are properly configured. The command lsnodes gives the node name on each of the nodes.

Command> lsnodes l -n

Ensure that the global service daemon is running on each of the nodes.
Configure the listener.ora file using port 1545 (as an example) on both the nodes.

LISTENER_N1 =
(DESCRIPTION =
(ADDRESS = (PROTOCOL = TCP)(HOST = RAC1)
(PORT = 1545)) )
SID_LIST_LISTENER_N1 =
(SID_LIST =
(SID_DESC =
(ORACLE_HOME = /uo1/app/oracle/product/920)
(SID_NAME = V9201) ))

Since the listener is not running on the default port of 1521, the local_listener parameter in the init.ora of the instances will need to be set so that dynamic service registration will work properly.

local_listener = "(address=(port=1545)(protocol=tcp)(host=rac1))"

On the second node change the references of listener_N1 to listener_N2 and change the host=rac1 to host=rac2.
Configure tnsnames.ora on all nodes.
An example of the tnsnames.ora file is shown below.

V10g.ACME.COM=
(DESCRIPTION=
(LOAD_BALANCE=ON) (FAILOVER=ON)
(ADDRESS_LIST=
(ADDRESS=(PROTOCOL=TCP)(HOST=RAC1.ACME.COM)
(PORT=1545))
(ADDRESS=(PROTOCOL=TCP)(HOST=RAC2.ACME.COM)
(PORT=1545)) )
(CONNECT_DATA=
(SERVICE_NAME=V10g.ACME.COM) ))

V10g1.ACME.COM=
(DESCRIPTION=
(ADDRESS=(PROTOCOL=TCP)(HOST=RAC1.ACME.COM)
(PORT=1545))
(CONNECT_DATA=
(SERVICE_NAME=V10g.ACME.COM)
(INSTANCE_NAME=V10g1) ) )

V10g2.ACME.COM=
(DESCRIPTION=
(ADDRESS=(PROTOCOL=TCP)(HOST=RAC2.ACME.COM)
(PORT=1545))
(CONNECT_DATA=
(SERVICE_NAME=V10g.ACME.COM)
(INSTANCE_NAME=V10g2) ) )

Note that the intelligent agent on the node will use the tnsnames.ora file. The service_names parameter is the combination of the init.ora parameters

< service_names> .< db_domain>

with the dot "." separating the two values.
The tnsnames.ora file is the same on all nodes, except the order of the addresses in the address_list section of the V10g.acme.com alias is different.
Test SQL*Net connections - if the listener.ora file, local_listener parameter, tnsnames.ora file, and the sqlnet.ora file are properly configured it should be possible to successfully connect to the V10g, V10g1, and V10g2 aliases from all nodes.

From the first node test the following connections

> sqlplus system/manager@v10g
> sqlplus system/manager@v10g1

From the second node test the following connections:

> sqlplus system/manager@v10g
> sqlplus system/manager@v10g2

Start the intelligent agent on both nodes. To start, use the command agentctl start. Check the $ORACLE_HOME/network/agents/services.ora file for correct discovery. The contents of this file help to understand the correctness of the configuration.
If any problems are noticed, set a higher trace level and check the message. To set up trace level 16, add the following line in the $ORACLE_HOME/network/admin/snmp_rw.ora file:

DBSNMP.TRACE_LEVEL=16

From the Enterprise Manager Console, discover the nodes rac1 and rac2. This will create one entry as V10g.acme.com under the Databases folder with a subfolder called Instances. The Instances folder will contain a subfolder named Cluster Database Instances.
Within this folder, the two instances V10g1.acme.com and V10g2.acme.com should be visible.

An important point to note is that in the case where ORACLE_HOME is shared by multiple RAC nodes (Cluster File system implementations or NFS situations), it is not possible to start the Intelligent Agent on both nodes simultaneously, since the $ORACLE_HOME/network/agent/*.q files would be overwritten each time the agent is started.
Therefore, Oracle recommends three options to avoid this. Follow one of these options to handle the shared Oracle Home situation. Install the Intelligent Agent for each node in its own Oracle Home location, distinct from the shared Oracle home location. Copy or link the tnsnames.ora and listener.ora to the Intelligent Agent's Oracle home location from the shared Oracle home. Then, start the Intelligent Agent using Intelligent Agent's Oracle home.
For RAC systems like HP TruCluster and PolyServe Matrix Server-based Linux RAC, it is possible to use the Context Dependent Symbolic Link (CDSL) facility to separate the $ORACLE_HOME/network directory from the shared Oracle Home installation without having to physically install the Intelligent Agent on each node in the cluster. This employs using a specific keyword in the filename (or a symbolic link) that distinguishes the name of the current member node of the cluster.
Beginning with version 9.2.x, the Intelligent Agent can store the binary state files (*.q) into an alternate directory rather than use the hard-coded $ORACLE_HOME/network/agent directory. This functionality is very similar to the CDSL method listed in option 2 above.
To force the agent to read and write the binary state files to/from an alternate location, the environment variable ora_oemagent_dir must be set prior to starting the agent. For example, the following would force the agent to read and write its state files to the /usr/oracle/agent directory:

export ORA_OEMAGENT_DIR=/usr/oracle/agent
agentctl start

For more details and updates, please refer to the Oracle MOSC Note (158295.1).