Service guard Local LAN failover question

[Dominic Saul]

I have a two node cluster just being built using HPUX-11.23 and Service guard version 11.18. During testing the cluster, I found if primary LAN fails it switches to standby LAN. Now if standby LAN is happily running the app and during that time primary card recovers it fails back the application to Primary LAN. So that means there will be twice small delay in response. So is there way, i could change behaviour that when LAN3 fails it fails to lan11 but when lan3 recovers it doesnt' fail back the IP back to it even lan11 is happily running service

ds84501a:/root# cmviewcl -v

CLUSTER STATUS
tsafeclus up

NODE STATUS STATE
ds84501a up running

Cluster_Lock_LVM:
VOLUME_GROUP PHYSICAL_VOLUME STATUS
/dev/vg_lock /dev/dsk/c2t0d2 up

Network_Parameters:
INTERFACE STATUS PATH NAME
PRIMARY up 0/2/1/0/4/1 lan1
PRIMARY up 0/2/1/0/6/1 lan3
PRIMARY up 0/4/2/0 lan6
PRIMARY up 0/6/1/0/4/1 lan9
STANDBY up 0/6/1/0/6/1 lan11

PACKAGE STATUS STATE AUTO_RUN NODE
tdsdbpkg up running enabled ds84501a

[Ganesan R]

Hi,

If you set FAILBACK_POLICY value to MANUAL in package configuration file, the package will not failback automatically.

FAILBACK_POLICY MANUAL

[Dominic Saul]

Hi,

Sorry, I meant about local lan ports failures on same node and it is is not package failover but IP address failover on local lans on same node

[Sajjad Sahir]

Dear Team

if ip for local node is failed u have to fix it there is no swithover then

if ip failed in packeges then only it will be swith over because we are assigning two ip there.

sajjad

[Dominic Saul]

Guys,

Let me explain again, 2 node cluster A and B. On node A, lan3 is primary and lan11 standby. Now lan3 fails it switches to lan11 happily. After a while lan3 recovers ie comes up, NOW LAN11 fails back to lan3. So my question is how to stop this.

[Duncan Edmonstone]

...and the answer is, you can't. The IP will always fail back to the primary LAN card as soon as it becomes available.

Do you see this as an issue? Do you expect to suffer multiple transitory loss of connecitivity issues on your primary LAN?

I've never seen this be a problem in practice. The small delay in response is to my mind almost invisible for most applications

HTH

Duncan

[Ganesan R]

Hi,

I go with Duncan answer. There is no parameter in cluster/pkg config files to control that behaviour.

Look at this also

http://forums11.itrc.hp.com/service/forums/questionanswer.do?admit=109447626+1238510870797+28353475&threadId=743357

[Dominic Saul]

Thanks Guys.Ok,I thought there would be some way to stop this because if app is happily running on lan11 and if primary(lan3 in this case) recovers it should not take action and only when if lan11 has issues it should take it over.

[John Bigg]

You already have the answer, but it appears there is a lack of confidence in it.

I can confirm there is no way to prevent the automatic failback of IP addresses from a standby lan to a primary when the primary recovers.

The only possibility is to use 11.18 and the new cmmodnet -d option which allows a lan to be disabled. This way you could disable the primary before you fix it, and then use -e to re-enable at a more convenient time.

The downside of this is that of course, unless you have multiple standbys you are exposed to further failures while the recovered primary is disabled.

[John Bigg]

You will get this same answer from me (indirectly) via your official support call too :-)

[Dominic Saul]

Guys, I think we got the answer here that it would fail back