Home » Server Options » RAC & Failsafe » RAC failover question (10.2.0.4 on Solaris 10)
RAC failover question [message #422347] Tue, 15 September 2009 10:36 Go to next message
Orna
Messages: 62
Registered: November 2008
Member
Hi

I have a 3 nodes 10.2.0.4 RAC cluster ( Solaris 10 )

One of the nodes crushed due to hardware failure and is no longer available.
the other two nodes froze and we had to reboot them as well - but they were rebooted successfully and joined the cluster.

we are using client site load balancing as follows :

TIMSR =
(DESCRIPTION =
(ADDRESS = (PROTOCOL = TCP)(HOST = VIP1)(PORT = 1521))
(ADDRESS = (PROTOCOL = TCP)(HOST = VIP2)(PORT = 1521))
(ADDRESS = (PROTOCOL = TCP)(HOST = VIP3)(PORT = 1521))
(LOAD_BALANCE = yes)
(CONNECT_DATA =
(SERVER = DEDICATED)
(SERVICE_NAME = timsr.ibasis.net)
(FAILOVER_MODE =
(TYPE = SELECT)
(METHOD = BASIC)
(RETRIES = 180)
(DELAY = 5)
)
)
)

The problem is that since one of the machines is dead , the VIP1 virtual IP is down.
As a result the above TNS HANGS ( from time to time- when it happens to choose VIP1 I guess )

The only solution was to manually start the VIP1 on one of the remaining servers in the cluster and edit the listener.ora on that server to listen on VIP1 as well.

Isn't there a better way to handle this ?
can't Oracle failover/load balanced mechanism handle a server crush without me manually intervening ?

Note that I am not intereseted in TAF - existing sessions can fail when a server crush . however , I do want that later sessions will be able to connect to the remaining two nodes wihtout making manual change in listeners, tns or manually start VIPs

thanks
Orna
Re: RAC failover question [message #422433 is a reply to message #422347] Wed, 16 September 2009 07:45 Go to previous messageGo to next message
ebrian
Messages: 2794
Registered: April 2006
Senior Member
Have you configured your listeners for cross-registration ?

You may also want to enable Oracle net tracing on the client side to see if you can glean any additional information.
Re: RAC failover question [message #422438 is a reply to message #422347] Wed, 16 September 2009 07:59 Go to previous messageGo to next message
Mahesh Rajendran
Messages: 10707
Registered: March 2002
Location: oracleDocoVille
Senior Member
Account Moderator
To start with, It could be a known bug.
There are several known issues with VIP failover on Solaris.
Most of these are fixed in 11g and 10gR2 higher patchsets.

It depends on your listener configuration too.
In client side try
failover=on

Above is supposed to be default, but just give a try.
Check the CRS logs for all nodes and post anything relevant.
Re: RAC failover question [message #422716 is a reply to message #422438] Fri, 18 September 2009 08:38 Go to previous messageGo to next message
Orna
Messages: 62
Registered: November 2008
Member
Hi

Can you point me to the Metalink docs or bug numbers please ?

thanks !
Orna
Re: RAC failover question [message #422720 is a reply to message #422438] Fri, 18 September 2009 08:54 Go to previous messageGo to next message
Orna
Messages: 62
Registered: November 2008
Member
I am think that my test case does not qualify as 'failover' - since I manually shutdown the instance and nodeapps - should the VIP failover to the other node even if I do the shutdown myself ?
Re: RAC failover question [message #424378 is a reply to message #422347] Thu, 01 October 2009 22:09 Go to previous message
mkounalis
Messages: 147
Registered: October 2009
Location: Dallas, TX
Senior Member
You need to failover the vip if you do a graceful shutdown of an instance. As previously mentioned, you need to perform a statement similar to this: 'crs_relocate ora.node1.vip -c node2'. Metalink Note# 749160.1. Hope this helps.
Previous Topic: Cluster Installation Error: Not connecting via private node
Next Topic: JDBC Connect Time LB to RAC
Goto Forum:
  


Current Time: Thu Mar 28 16:15:49 CDT 2024