RAC VIP

[bvinod]

I have configured RAC10g with SGeRAC on HPUX 11i. The configuration is a two node cluster. The RAC vip is assigned to eth0:1 on both the nodes. When i shutdown node 1 completely the VIP service fails over to the node2. But now both the vip's are competeing for eth0:1 resulting in the ip address of eth0:1 switching rapidly. node 2 eth0:1 now takes vip1 for 30 secs and vip2 for 30 secs where as the vip1 when shifted to node 2 should take eth0:2. Does anybody have an idea of why this can happen. Pls share ur experiance on this if any.

[Steven E. Protter]

Shalom,

Both nodes can have eth0:1.

What won't work is the same IP address on both nodes at the same time.

At this point you have not made clear that this is the case.

Your theory that when on the second node eth0:2 should be used is simply not correct.

Also note that HP-UX uses lan0 and lan0:1 not eth0 as device names.

I'd say its time to look at the logs /var/adm/syslog/syslog.log and the oracle logs and do a reasonable problem determination.

SEP

[Matti_Kurkela]

Each time ServiceGuard starts up a package that requires a VIP, it finds the lowest unused alias number on the respective interface.

So the alias numbers can and will change.

Look into the package control scripts, especially the service commands and customer-defined functions. These should make no assumptions about the interface name.

Which version of HP-UX 11i are you running? Which version of ServiceGuard?
The "eth" in interface names sounds like Linux: in HP-UX it's usually "lan".

MK

[whiz_kid]

i dont think u need to configure any VIP from service guard if its a RAC installation. Oracle does take care of VIP, is asked at the time of Oracle RAC installation.

[Andrew C Fieldsend]

If memory serves - it's a year or so since I last installed a RAC on ServiceGuard - you install a separate SG package on each cluster node, both of which run in parallel, and Oracle handles the disk sharing itself with the help of the SGeRAC.

There is no VIP; each package has its own IP address, and the Oracle client software handles which instance it should connect to, as well as failover between the instances (which are the same database, of course).

[bvinod]

Its actually lan0 not eth0. What I am taking is a post vip failover scenario. After a vip failover to the node2(node1 is shutdown), the lan0:1 will aquire vip1(vip of node1) for few minutes and then vip2(vip of node2) for some few minutes. This alternate aquiring of IP by lan0:1 keeps continuing. Also although my server interface cards are Gigabit cards, the swithh i am using is a 100mbps switch. Is that going to matter and can be a cause to my problem.

kindly help me with ur suggestions

[Andrew C Fieldsend]

Sorry, I might not have been clear in my earlier post. There shouldn't be a VIP failover between nodes; each node has its own private IP for its own Oracle package.

Also, yes it will make a difference if you connect your gigabit cards through a 100 Mb switch - they will autonegotiate to the lower speed, and you might have performance problems with RAC cache fusion (depending on how heavily used your RAC is).

[Orhan Biyiklioglu]

Hi,

This is actually a RAC issue, not related to ServiceGuard. It is a bug of Oracle EE 10.2.0.1.0 which is fixed in 10.2.0.2.0.

You can move to 10.2.0.2.0 or just edit the file $CRS_HOME/bin/racgvip and change the line

_LIN=`listif | $GREP -w "$_LOCAL_IF:" | $SED -e's/^.*://' | $SORT -n`

to

_LIN=`listif | $GREP "$_LOCAL_IF:" | $SED -e's/^.*://' | $SORT -n`

Needless to say, you should make a backup first and try this at your own risk.

PS: Consult Oracle Metalink Doc ID 370286.1 for more info.

[bvinod]

Thanks for the INFO. Do i have to do any re-installation/configuration after doing the specified change

[Andrew C Fieldsend]

Apologies, I should have checked the changes betwen 9i and 10g. What I described was Oracle 9i RAC, which didn't have a VIP. 10g does, to improve client response time in the event of node failure.

My comment about the switch speed still stands though.