www.openlinksw.com
docs.openlinksw.com

Book Home

Contents
Preface

Real Application Cluster (RAC) / TAF Support

Real Application Cluster (RAC) / TAF Support
What it is What you have to do to use it Related Links

13.1. Real Application Cluster (RAC) / TAF Support

13.1.1. What it is

Transparent Application Failover (TAF) is a protocol within Oracle whereby, if a connection to a database node fails, it can be re-established against an alternative node.

Figure: 13.1.1.1.

Once a broken connection has failed-over, an application can continue without any special action on its part. However, TAF does not restore all facets of a connection. For instance, TAF does not:

If any of these situations apply to a failed connection, an application may need to take action following failover to return the connection fully to the desired state. In order to do this, the application may request to be notified when failover has occurred. The OpenLink Lite Driver for Oracle 10g and the OpenLink Generic Multi-Tier Client provide this facility through TAF event notifications.

The Oracle 10g Lite Driver and Multi-Tier Agent also allow configuration of the failover retry interval and the maximum number of failover retries, in the event that failover is not successful on the first attempt.

13.1.2. What you have to do to use it

13.1.2.1. Configuration
13.1.2.1.1. Multi-Tier

To use OpenLink's TAF event notification features or configure the failover retry parameters, you must enable OpenLink's TAF support. For the Oracle 10g Multi-Tier Agent, you must set the environment variable OPL_TAF_ENABLE to 'Y' or 'y'. With the agent, this is done by creating an entry in the appropriate 'Environment' section of the OpenLink Request Broker's rulebook, or by navigating through the web-based Administrative Assistant (normally running on <http://servername:8000/>) to configure an agent by form or wizard.
13.1.2.1.2. TAF Retry Configuration

When OpenLink's TAF support is enabled, in the event that a failover attempt fails, the Lite driver or Multi-Tier agent will instruct Oracle to retry. By default, the maximum number of failover attempts is 10 and the interval between retry attempts is 10 seconds. The default values can be overridden.

If using the Oracle 10g Lite Driver for Windows, enter the new values in the 'Max. Retries' and 'Retry Interval' text fields. For the Oracle 10g Lite Driver for Unix or the Multi-Tier Oracle 10g agent, override the defaults by setting the environment variables OPL_TAF_MAX_RETRIES and OPL_TAF_RETRY_INTERVAL. The retry settings operate independently of TAF event call-backs. When OpenLink's TAF support is enabled, they are effective irrespective of whether an application registers an event call-back.

13.1.2.2. Programmatic Considerations

OpenLink ODBC drivers notify ODBC applications that failover has occurred using two mechanisms: event call-backs and SQLSTATEs.
13.1.2.2.1. TAF Event Notification: Call-backs

An application can register a failover call-back routine by setting the proprietary connection attribute SQL_ATTR_EVENT_CALL-BACK (1280) on an open connection. The routine's address is supplied as the value of SQLSetConnectAttr ?'s ValuePtr argument. The same call-back routine can be registered for more than one connection. The call-back routine's signature must take the form:

  void (*call-back) (oplevent_t oplEvent, SQLHANDLE handle, SQLUSMALLINT
  eventInfo)

where:

oplEvent is of type oplevent_t, an enumerated type enumerating the types of events reported to event call-backs. oplevent_t is defined as follows:

 typedef enum {
  OPL_EV_NONE = 0, OPL_EV_FAILOVER_SUCCESS = 16, OPL_EV_FAILOVER_ABORT = 17 }
  oplevent_t;

At the moment, only failover events are supported through the OPL_EV_FAILOVER_xxx event class. Other types of event may be supported in the future using this call-back mechanism. OPL_EV_FAILOVER_SUCCESS indicates that failover was successful, OPL_EV_FAILOVER_ABORT that failover was aborted.

context is the ODBC handle (HDBC) of the connection for which the application wishes to receive event notifications. This handle should be supplied to the ODBC driver when the call-back is registered, using another Oracle-specific connection attribute, SQL_ATTR_EVENT_CONTEXT (1281). If this attribute is not set, the call-back receives SQL_NULL_HANDLE for the handle argument. (If other event types are supported in the future, this attribute may accept other types of ODBC handle, e.g. handles of type SQL_HANDLE_STMT, depending on the scope of the event.)

eventInfo is reserved for future use. All failover events currently return 0.
13.1.2.2.2. TAF Event Notification: SQLSTATEs

As well as an ODBC application being informed of failover through a call-back routine, it also receives notification through SQLSTATEs. After failover completes, the first ODBC call to return, on the affected connection, can return one of two proprietary SQLSTATES, IM500 or IM501:

If the ODBC call returning the failover SQLSTATE fails for some reason, a diagnostic record holding the failover SQLSTATE and message is appended to any diagnostic records already generated by the failing call. In this case, even if failover was successful, a SQLSTATE IM500 may be accompanied by a function return code of SQL_ERROR. For instance, if a transaction was open at the time failover took place, SQLExecute may return SQL_ERROR with two diagnostic records, for example:

Note: Failover notification using TAF-specific SQLSTATEs cannot be used independently of failover call-backs. The use of these SQLSTATEs is only triggered when an application registers a failover call-back.

13.1.3. Related Links