Yet another Cluster lock missing - Full info provided

[donna hofmeister]

ami correct in thinking that 'vgchange -a' does not work on your system?  what about the other side of the cluster -- does vgchange know about '-a'?

[rventura]

Well, i believe this command will deactivate and activate the vg, therefore, i cant execute this command since the ERP data resides in that VG and people is using the ERP... Unless this is a non dirructive operation..?

Thanks

[donna hofmeister]

a 'man vgchange' should tell you if '-a' is a valid option

[rventura]

Yeah, the man page show the '-a' option.

I also read about the '-c' option, maybe i should had used that one instead of '-a'...?

[donna hofmeister]

to remove the volume group, you'll need to do N lvremoves (one for each lvol), N vgreduces (one for each disk minus one) and one vgremove (to remove the last physical volume).

when you go to make the replacement vg, you'll want to start with a "pvcreate -f" to write fresh/untainted lvm headers onto each disk.

------------

that's the lovely theoretical part!

since these disks are visible to both nodes of your cluster -- if you touch these disks, the application WILL BE impacted.  so you'll need to arrange for application downtime!!

double and triple check that you've got backups (plural!!) of whatever is currently on this volume before you start anything!!

use this command to record the PVRA information of the disk that will be the cluster lock disk:

# echo "0x2008?4X" | adb /dev/dsk/cXtYdZ

following the vgimport of the map file on the 2nd node, issue this command (on the 2nd node) to make sure cXtYdZ is the same on both systems. (the values returned must match!).

cmapplyconf will distribute the SG information but also modify the LVM headers (the PVRA).

as a reminder -- mirroring does NOT replicate lvm headers.  when you have to replace a disk -- especially one that is part of a SG cluster you need to do a vgcfgrestore as well.

[rventura]

Wont the PVCREATE destroy all my data in the drives..?

[donna hofmeister]

it sure will....hence the recommendation to do multiple backups and plan for an application outtage.

[Stephen Doud]

You stated "We are running a single node cluster for reason beyond my control" hence the cluster lock functionality is not being used.

However, if you cannot afford the down-time but need to reinstall the cluster lock structure on disk, open a support case with the HP Support Center and request the cminitlock tool.  It has the ability to restore the cluster lock structure on the disk designated in the cluster binary file.