Transparent Application Failover (OPS)

From: Dirk Kiehne <dirk_at_liberate.com>
Date: Sat, 27 Jan 2001 00:02:53 GMT
Message-ID: <3A723011.7B33E682@liberate.com>

A TAF question for OPS experts,

I'm trying to get our JDBC thick client to talk to our OPS and when I shutdown the public ethernet with ifconfig, the client should failover to the other instance.

This is happening, although when the failover occurs, the select that was in process fails. This is even though, I've specified the following in the tnsnames.ora:

OPS1 =
  (DESCRIPTION =

    (ADDRESS = (PROTOCOL = TCP)(HOST = ops1)(PORT = 1521))
    (ADDRESS = (PROTOCOL = TCP)(HOST = ops2)(PORT = 1521))
    (LOAD_BALANCE = yes)
    (FAILOVER = yes)
    (CONNECT_DATA =
      (SERVICE_NAME = ops1.mib.liberate.com)
      (FAILOVER_MODE =
        (TYPE = SELECT)
        (METHOD = BASIC)
        (RETRIES=3)
        (DELAY=2)

    )
  )

The "TYPE" field set to "SELECT" should maintain the select and after failover continue with the call of the client through the other instance.

I've made the same modifications to the client side where our servers reside with Oracle client, as well as the two server instance tnsnames.ora.

I've also verified that the servers are configured with:

	SVRMGR> select sid, failover_type, failover_method,


	> failed_over from v$session;

But the select never continues...

My question: Do I make these changes to all 3 tnsnames.ora?? (The docs are not to clear on this)

Another stupid question: After failover, I'm left with two connections, the original (now dead connection) and the failed over one. After I disconnect the client, the second one goes away, but the dead one is still in limbo. To avoid confusion, I'm shutting down the instance on both nodes, but am forced to do shutdown abort (on the dead connection node) to clean up connections. Is there a clean way, within svrmgrl to kill the connection or should I just do a kill <pid> ??

Thanks in advance,
-Dirk Received on Fri Jan 26 2001 - 18:02:53 CST