Serviceguard
cancel
Showing results for 
Search instead for 
Did you mean: 

ProLiant DL380 ServiceGuard & bonding problem

 
Timo J
Frequent Advisor

ProLiant DL380 ServiceGuard & bonding problem

RHEL3.0, kernel 2.4.21-9, bonding driver 1.0.4f-1 (bonding driver from HP)

I have 2-node cluster, four nics on each host. I've created two bonds on both hosts (two nics on one bond, though all nics at same subnet). Bonding is working ok, so when I'm unplugging lan cables, the other nic gets active and so on.

But when I'm trying to query the ServiceGuard cluster with command 'cmquerycl -v -L /my/lock/disk -n host1 -n host2 -C $SGCONF/foo.ascii' , i'm getting the following error:

"Error: Unable to determine network configuration: failed to receive net probe reply from node host1.
Failed to probe network.
Warning: Network interface bond0 on node host2 couldn't talk to itself.
Warning: Network interface bond1 on node host2 couldn't talk to itself."

Exactly same error if I execute that command on host or host2.
N/A
2 REPLIES
Steven E. Protter
Exalted Contributor

Re: ProLiant DL380 ServiceGuard & bonding problem

You need to check if ServiceGuard can handle bonding.

I know it can handle fast failover but it might not be compatible with a bonding configuration.

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
Serviceguard for Linux
Honored Contributor

Re: ProLiant DL380 ServiceGuard & bonding problem

SG uses bonding.

The most common problem is using the wrong driver. If you have TG3 - try the BCM5700 driver. Can't remember the appropriate driver for Intel but both are documented in the release notes.

If you need to add the BCM5700 driver and download it - read the install instructions VERY carefully. It is to easy to do it wrong.

Another possible problem (or one to watch out for) is cmclnodelist (and rhosts) not being correctly set up. See section Editing Security Files (chapter 5) in the Users manual.

Happy clustering