Configuring OPS/RAC net service for failover
There are three files to be configured: listener.ora, tnsnames.ora, and sqlnet.ora. You can use netasst to configure them graphically, or you can just edit these files manually if you are confident with OPS/RAC. We directly edited these files, since it is faster to do so. You must manually edit the files to implement a new listener for failover and netasst does not allow using the same port number for another listener.
Implementing failover does not allow the use of static service configuration parameters in the listener.ora file. Therefore, in order to implement OPS/RAC failover, an additional listener must be specified on each node in the listener.ora file without service registration information. Edit the sqlnet.ora file to ensure that tnsnames.ora is used to resolve a net service name.
Client load balance
When you set load_balance=on in tnsnames.ora, OPS/RAC will balance the load over the list of listener addresses by picking one at random.
There are two different kinds of OPS/RAC failovers: connection-time failover (CTF) and transparent application failover (TAF). For TAF, there are several different configurations using TYPE and METHOD parameters.
Connection-time failover (CTF)
Connection-time failover refers to a client attempting to connect to a second listener when the attempt to connect to the first listener fails. To implement this, create a new net service name with the database name as its identifier, such as opswas, instead of opshp1 or opshp2. Put the address information of all nodes in the cluster in this newly created global net service name.
Use one of two options:
Setting this means: try each address in order until one succeeds.
Setting this means: try each address randomly until one succeeds.
WebSphere works with either setting.
Example 12-8 is a sample tnsnames.ora file using these OPS/RAC connection-time failover options.
Example 12-8 Sample OPS/RAC CTF tnsnames.ora file
WASOPSCTF.somecorp.com=
(description=
(load_balance=on)
(failover=on)
(address=
(protocol=tcp)
(host=idops1)
(port=1521))
(address=
(protocol=tcp)
(host=idops2)
(port=1521))
(connect_data=
(service_name=WASOPSCTF.somecorp.com))))
Transparent application failover (TAF)
Transparent application failover enables an application to automatically reconnect to a database if the connection is broken. Active transactions roll back, and the new database connection is identical to the original one no matter how the connection is lost.
Implementing transparent application failover requires you to manually configure tnsnames.ora with the specification of a primary node, a backup node, failover type, and failover method. You can also combine OPS/RAC CTF with OPS/RAC TAF. As shown in Table 12-1, there are 12 configurations (2 CTF x 3 TYPE x 2 METHOD) for the OPS/RAC TAF implementations.
Table 12-1
| CTF | TYPE | METHOD |
|---|---|---|
| FAILOVER=ON | SESSION | BASIC |
| FAILOVER=ON LOAD_BALANCE=ON | SELECT | PRECONNECT |
| NONE |
Combinations of OPS/RAC/TAF failover configuration implementations
The parameter TYPE specifies the type of failover. This is a required parameter for the TAF failover. There are three options: SESSION, SELECT, and NONE. Selection of SESSION means that OPS/RAC fails over the session. A new session will be created automatically for use on the backup if a user's connection is lost. This type of failover does not attempt to recover selects after a failover. Choosing SELECT means that OPS/RAC allows users with open cursors to continue fetching on them after a failover. Although the performance for this type of TAF during a failover is good, it has an overhead during normal select operations. Selection of NONE means that no failover function is used. This is the default.
The parameter METHOD specifies how fast a failover occurs from the primary node to the backup node. There are two options: BASIC and PRECONNECT. Selection of BASIC means that OPS/RAC establishes connection at failover time. This option requires almost no work on the backup server until failover time. Selection of PRECONNECT means that OPS/RAC pre-establishes connections all the time. This is good for failover performance, but bad for normal operation, since it requires the backup instance to support all connections from every supported instance.
Example 12-9 shows a sample tnsnames.ora for OPS/RAC/TAF configuration with CTF.
Example 12-9 Sample OPS/RAC TAF with CTF tnsnames.ora file
WASOPSTAF.somecorp.com=
(description=
(load_balance=on)
(failover=on)
(address=
(protocol=tcp)
(host=opshp1)
(port=1521)
(address=
(protocol=tcp)
(host=opshp2)
(port=1521)
(connect_data=
(service_name=WASOPSTAF.somecorp.com)
(failover_mode=
(type=session)
(method=preconnect)
(retries=0)
(delay=10))))
Alternatively, Example 12-10 shows a sample tnsnames.ora for the OPS/RAC/TAF with primary and secondary servers configured.
Example 12-10 Sample OPS/RAC TAF with primary/secondary servers tnsnames.ora file
WASOPSTAF.somecorp.com=
(description=
(address=
(protocol=tcp)
(host=opshp1)
(port=1521))
(connect_data=
(service_name=WASOPSTAF.somecorp.com)
(failover_mode=
(backup=opshp2)
(type=session)
(method=preconnect)
(retries=0)
(delay=10))))
WebSphere works with OPS/RAC/CTF failover configurations. WebSphere also works with some of OPS/RAC/TAF failover configurations, for example TYPE of SESSION, and METHOD of both BASIC and PRECONNECT.
WebSphere is a trademark of the IBM Corporation in the United States, other countries, or both.
IBM is a trademark of the IBM Corporation in the United States, other countries, or both.