Configuring ServiceGuard and shared volume group

[rayvd]

I'm working on setting up a two node MC/ServiceGuard cluster using two A500 systems with HP-UX 11.00, a single 36 GB shared disk, and MC/ServiceGuard A.11.14. I did a pvcreate on the shared disk and then created a new vg, vg01, using vgcreate. But when I run cmclcheckconf, I see a bunch of SCSI errors in my syslog, and cmclcheckconf gives an error, as follows
Error: Unable to determine a unique identifier for physical volume /dev/dsk/c4t8d0 on node castor. Use pvcreate to give the disk an identifier.

This seems odd, I've already created a vg using this disk, and was able to create a logical volume. I also did a vgexport, then copied over the lvm_map file to the other node and used vgimport.

On both nodes, the 36 GB disk has the same hardware path, 0/4/0/0.8.0, and is connected to a SCSI C895 Ultra2 Wide LVD controller, which is located at 0/4/0/0 on both systems, with the Initiator at 0/4/0/0.7. Is this right or is ther something I'm doing wrong?

[A. Clay Stephenson · Accepted solution]

Something is very wrong. You have both controllers set to SCSI ID 7? That can't work. You must reset the SCSI ID on one of the nodes to 6 (or at least not 7). The other thing I don't like about your configuration is that this disk is not mirrored. Normally, in MC/SG all disks are mirrored or are arrays with multiple SCSI paths so that long before MC/SG has kicked-in normal Mirror/UX has already fixed that part of the problem.

[Jeff Schussele]

Hi Dion,

Make the VG cluster aware

vgchange -c y /dev/vg01

Then try to activate the VG exclusive on both systems

vgchange -a e /dev/vg01 (host #1)
vgchange -a n /dev/vg01

vgchange -a e /dev/vg01 (host #2)
vgchange -a n /dev/vg01

If that's OK then try to check cluster conf

HTH,
Jeff

[rayvd]

Clay,
I think you're right, I remember the SG cluster we set up in class had the controllers on different IDs. My question then becomes how can I control what Scsi ID the controller uses. These systems are on loan from HP for us to use in testing our software line for compatibility with ServiceGuard. So I agree the configuration isn't the best, but it won't be used for a HA production application, just to get our apps to failover correctly on a SG cluster.

[A. Clay Stephenson]

I'm not familiar with a C895 (and that doesn't look like an HP part no.) but in general you have to enter the boot menu and issue a SCSI INIT command to change the setting. I suggest that you look up your controller in docs.hp.com; it's probably here: http://docs.hp.com/hpux/netcom/index.html#SCSI%20Host%20Bus%20Adapters

You should set one of the controllers to SCSI ID 6 and the other should be left at the default value (7). I would check both. Good SA's assume nothing. Termination should also be enabled on both controllers and no other terminators should be on the bus. You may have other MC/SG-LVM problems but until you get this fixed the rest is moot.

[rayvd]

Clay,
You are correct, I was able to interrupt the boot on one of the systems, then do the following:
1. Service Menu
2. scsi init

When I booted the system, the controller had the correct id of 6.

[rayvd]

Thanks to both of you, I have my SG cluster up and running now.