9
Configuring High Availability Features for Real Application Clusters

This chapter describes how to configure the high availability features of Real Application Clusters. The topics in this chapter are:

• Transparent Application Failover

• Primary and Secondary Instances

• Configuring Clients for Real Application Clusters

Transparent Application Failover

This section discusses the Real Application Clusters-specific aspects of transparent application failover (TAF). This section covers the following topics:

• FAILOVER_MODE Parameters

• TAF Implementation

• Transparent Application Failover Verification

To enable TAF, manually configure a net service name in the CONNECT_DATA portion of the connect descriptor. In the net service name you can include the FAILOVER_MODE and INSTANCE_ROLE parameters.

TAF instructs Oracle Net to move a connection that has failed to a different instance. This enables the user to continue working by using the new connection as if the original connection had never failed.

FAILOVER_MODE Parameters

To configure TAF, include FAILOVER_MODE parameter settings in the CONNECT_DATA portion of a connect descriptor. There are several sub-parameters you can use to specify FAILOVER_MODE. For example, you can specify the type of failover and speed at which Oracle should process it, as well as the retry behavior Oracle should use.

TAF Implementation

Depending on the FAILOVER_MODE parameters, you can implement TAF using any of the following methods:

• Implementing TAF with Connect-Time Failover and Client Load Balancing

• Retrying Connections

• Pre-Establishing Connections

Implementing TAF with Connect-Time Failover and Client Load Balancing

You can implement TAF with connect-time failover and client load balancing for multiple addresses. In the following example, Oracle Net connects randomly to one of the listener addresses on node db1 or db2. If the instance fails after the connection, OracleNet fails over to the other node's instance, preserving any SELECT statements in progress.

db.us.acme.com= 
 (description=  
  (load_balance=on)    /* only connect time load balancing and connection load 
balancing */
  (failover=on)            /* only connect time failover */
  (address=  
       (protocol=tcp)  
       (host=db1)  
       (port=1521)) 
 (address=  
       (protocol=tcp)  
       (host=db2)  
       (port=1521)) 
  (connect_data= 
     (service_name=db.us.acme.com) 
     (failover_mode=
       (type=select) 
       (method=basic))))

Retrying Connections

TAF also provides the ability to automatically retry connecting with the RETRIES and DELAY parameters if the first connection attempt fails. In the following example, Oracle Net attempts to connect to the listener on db1. If the initial connection fails, it fails over to addresses in the description lists.

db.us.acme.com=  
 (description= 
  (address=  
       (protocol=tcp)  
       (host=db1)  
       (port=1521)) 
  (connect_data= 
     (service_name=db.us.acme.com) 
     (failover_mode=
       (type=select) 
       (method=basic)
       (retries=20)
       (delay=15))))

Pre-Establishing Connections

A backup connection can be pre-established. The initial and backup connections must be explicitly specified. In the following example, Oracle Net connects to the listener on db1. If db1 fails after the connection, Oracle Net fails over to db2, preserving any SELECT statements in progress. If the pre-connect to the failed instance does not succeed at connect time, then fail back to this instance is no longer possible.

db.acme.com= 
 (description=  
  (address=  
       (protocol=tcp)  
       (host=db1)  
       (port=1521)) 
  (connect_data= 
     (service_name=db.us.acme.com) 
     (instance_name=db1) 
     (failover_mode=
       (backup=db2.acme.com) 
       (type=select) 
       (method=preconnect))))
db2.acme.com= 
 (description=  
  (address=  
       (protocol=tcp)  
       (host=db2)  
       (port=1521)) 
  (connect_data= 
     (service_name=db.us.acme.com) 
     (instance_name=db2)
     (failover_mode=
     (backup=db1.acme.com)
     (type=select)
     (method=preconnect))
     ))

Transparent Application Failover Verification

Use the V$SESSION view to obtain information about the connected clients and their TAF status. For example, query the FAILOVER_TYPE, FAILOVER_METHOD, and FAILED_OVER columns to verify that you have correctly configured TAF as in the following SQL statement:

SELECT MACHINE, FAILOVER_TYPE, FAILOVER_METHOD, FAILED_OVER, COUNT(*)
FROM V$SESSION
GROUP BY MACHINE, FAILOVER_TYPE, FAILOVER_METHOD, FAILED_OVER;

The output before failover resembles the following:

MACHINE              FAILOVER_TYPE FAILOVER_M FAI   COUNT(*)
-------------------- ------------- ---------- --- ----------
db1                  NONE          NONE       NO          11
db2                  SELECT        PRECONNECT NO           1

The output after failover is:

MACHINE              FAILOVER_TYPE FAILOVER_M FAI   COUNT(*)
-------------------- ------------- ---------- --- ----------
db2                  NONE          NONE       NO          10
db2                  SELECT        PRECONNECT YES          1

Primary and Secondary Instances

The primary and secondary instances feature specifies that the primary instance accepts primary connections and the secondary instance only accepts secondary connections. You can only implement this feature for two-instance Real Application Clusters environments. This section describes the primary and secondary instance feature in more detail and contains the following topics:

• Overview of Primary and Secondary Instances

• Initialization File Configuration

• Client Configuration

• Listener Configuration

• Connecting to Secondary Instances

• Warming the Library Cache on the Secondary Instance

Overview of Primary and Secondary Instances

An instance is the primary instance when ACTIVE_INSTANCE_COUNT=1 is set in the initialization file and it has been started first. The primary instance registers its status and database service information with its local listener through dynamic service registration.

If you configure shared server with the LISTENER attribute, the primary instance can also register with the secondary instance's listener. The LISTENER parameter can specify a listener name alias for the listener to which the dispatcher(s) register information. This is resolved to a list of listener addresses through a naming method, such as a tnsnames.ora file.

This enables the primary instance to accept connections from its local listener, as well as from the secondary instance's listener. A secondary instance registers with its local listener as a secondary instance, and the secondary instance has its ACTIVE_INSTANCE_COUNT set to 1 in the initsid.ora file.

If the primary instance fails, the secondary instance assumes the primary role and registers with its listeners. When the failed instance re-starts, it starts up as the secondary instance.

Clients connected to the failed primary instance are failed over to the secondary instance if you have configured TAF. Clients connecting to the Real Application Clusters database after the primary instance fails are automatically routed to the secondary instance.

Initialization File Configuration

To enable primary and secondary instance configuration, configure the initialization file by setting the ACTIVE_INSTANCE_COUNT parameter to 1 on both instances.

Client Configuration

Oracle Corporation recommends configuring a connect descriptor on clients that use an address list containing the listener addresses for the primary instance and the secondary instance. Set the LOAD_BALANCE parameter to OFF because all client connections can only go to the primary instance. FAILOVER is set to ON by default for a list of addresses, so it does not need to be explicitly specified. An example of the client configuration follows:

db.us.acme.com= 
 (description= 
  (load_balance=off) /* connection load balancing */ 
  (address=(protocol=tcp)(host=db1)(port=1521))
  (address=(protocol=tcp)(host=db2)(port=1521)) 
  (connect_data=
     (service_name=db.us.oracle.com)))

Oracle does not recommend setting LOAD_BALANCE=ON. If you do, half of the connections attempt to connect to the listener on the secondary instance that fails to provide connections. The client then tries the listener on the primary instance, which succeeds.

Listener Configuration

Remove the static service information sid_LIST_listener_name entry from the listener.ora file. This way, the listener only uses information obtained from dynamic service registration. For example, the sid_list_listener entry has been removed from the listener.ora file as shown in Table 9-1:

Table 9-1 Listener.ora Example without sid_list_listener Entry

listener= 
 (description=  
  (address=
   (protocol=tcp)
   (host=db1)
   (port=1521)))
sid_list_listener= 
 (sid_desc=
   (oracle_home=/orahome81)
   (sid_name=db1))

listener= 
 (description=  
  (address=
   (protocol=tcp)
   (host=db1)
   (port=1521)))

Connecting to Secondary Instances

In some situations, you may wish to connect to the secondary instance even when the primary instance is active. For example, you may want to perform a batch operation on the database. To do this, use the INSTANCE_ROLE parameter in the connect data portion of the connect descriptor to configure explicit secondary instance connections.

The optional INSTANCE_ROLE parameter in the CONNECT_DATA section of a connect descriptor enables you to specify connections to primary or secondary instances. This parameter is useful for explicitly connecting to primary or secondary instances and for using Transparent Application Failover (TAF) to pre-connect to secondary instances.

Warming the Library Cache on the Secondary Instance

Maintaining information about frequently executed SQL and PL/SQL statements in the library cache improves the performance of the Oracle database server. In Real Application Clusters primary and secondary instance configurations, the library cache associated with the primary instance contains up-to-date information. During failover, the benefit of that information is lost unless the library cache on the secondary instance was previously populated.

Use the DBMS_LIBCACHE package to transfer information in the library cache of the primary instance to the library cache of the secondary instance. This process is called warming the library cache. It improves performance immediately after failover because the new primary library cache does not need to be populated with parsed SQL statements and compiled PL/SQL units.

Configuring Clients for Real Application Clusters

This section describes client configuration issues not covered by the database creation process. You should configure the client with a net service name for the database. This entry should have an address list of all the listeners in the database. Additionally, set the connect-time failover and client load balancing options.

If the first listener fails, connect time failover instructs the client to failover to the next listener in the address list. Client load balancing instructs the client to randomly select a listener address. This randomization distributes the load to avoid overburdening a single listener.

There are two cases in which a client attempts to connect to another address:

• If the listener is down

• If the listener is up, but it has not received registration for the given SERVICE_NAME, so the instance is down

  Note : The PMON process waits 60 seconds to reconnect to a failed listener. You can use the ALTER SYSTEM REGISTER SQL statement to force a reconnection attempt to occur immediately. You can only execute this statement, however, when the listener is running.

The second case implies that the client only attempts to connect to the next listener if the first listener fails. The client also tries the next listener if the first listener is up, but does not know about the SERVICE_NAME given in CONNECT DATA.

Together, connect-time failover and client load balancing instruct the client to randomly choose an address. If the chosen address fails, the connection request fails over to the next address. If an instance fails, the client can connect using another instance.

Implementation of Client Configurations

To control how the client executes these connection attempts, configure multiple listening addresses and use FAILOVER=ON and LOAD_BALANCE=ON for the address list. For example:

db.us.acme.com= 
 (description= 
  (load_balance=on)
  (failover=on)
  (address=(protocol=tcp)(host=db1)(port=1521))
  (address=(protocol=tcp)(host=db2)(port=1521)) 
  (connect_data=
     (service_name=db.us.acme.com)))

Testing the Oracle Net Configuration

To ensure the files are configured correctly:

1. On any node or client machine, connect to an instance:

   CONNECT SYSTEM/password@net_service_name
   
   

   Oracle displays a "Connected" message.

   If there is a connection error, troubleshoot your installation. Typically, this is a result of a problem with the IP address, host name, service name, or instance name.

2. On one node, increase Miller's salary by $1000 and commit the change:

   UPDATE EMP
   set sal = sal + 1000
   where ename = 'miller';
   commit;
   
   

3. On the other nodes, select the EMP table again:

   SELECT * FROM EMP;
   
   

   Miller's salary should now be $2,300, indicating that all the instances can access the records in the database.