topic Re: lvm / mcsg in Operating System - HP-UX

lvm / mcsg

Francois LAURENT — Thu, 09 Oct 2003 10:32:03 GMT

hi all,

My cluster is up and running but the package is halted.
I try to activate a vg of this package on one node, but it failed.

I done :
#vgchange -c n /dev/vg02 --> OK
#vgchange -a y /dev/vg02 --> "vgchange: Activation mode requested for the volume group "/dev/vg02" conflicts with configured mode."

Other test :
#vgchange -c y /dev/vg02 --> OK
#vgchange -a e /dev/vg02 --> "vgchange: Volume group "/dev/vg02" belongs to a different Cluster. Activation failed"

same result on the other node.

Anyone can guide where to check this out?

Thanks for your help

Re: lvm / mcsg

Dietmar Konermann — Thu, 09 Oct 2003 10:52:13 GMT

Try to issue the vgchange -c n from ALL nodes first. Then do the vgchange -c y. Sometime this helps LVM to forget about some buffered information.

Best regards...
Dietmar.

Re: lvm / mcsg

Kevin Wright — Thu, 09 Oct 2003 12:18:02 GMT

Did you change the cluster name? You need to deactivate and make the VG uncluster aware from all nodes. vgchange -c n; vgchange -a n. Then make it cluster aware, -c y, and activate it with -a e on one node.

Re: lvm / mcsg

Kent Ostby — Thu, 09 Oct 2003 12:53:27 GMT

I agree with trying the vgchange -c n from the other nodes in the cluster as well.

If that fails, this document has A fix for it, but you'd probably have to take the clsuter down to execute it: UMCSGKBRC00011516

Re: lvm / mcsg

Dave La Mar — Thu, 09 Oct 2003 14:26:28 GMT

Laurent -
As others have stated, the most common occurrence of this that we have seen is NOT issuing the -c n command from the node that last had it active.

Best of luck.

Regards,

dl

Re: lvm / mcsg

Francois LAURENT — Fri, 10 Oct 2003 02:59:46 GMT

hi all,

thanks for your replys, i tried the vgchange -c n on all nodes and -c y, the vgchange -a e gave the same result ....
same thing with -c n and -a y ...
The cluster name hasn't be changed.
:-(

Re: lvm / mcsg

Dietmar Konermann — Fri, 10 Oct 2003 03:40:48 GMT

Well, that's getting weird.

Here some background information... if you build a new cluster (with no existing configuration present, i.e. no /etc/cmcluster/cmclconfig file) then a fresh ClusterID is generated. This is a unique number, which is used to identify this new cluster in the future.

Now, when you mark a VG cluster aware (vgchange -c y) then this ClusterID is stamped into the LVMREC at the beginning of each disk in the VG.

You can check the ClusterID of a disk like this:

echo "0d8332?U" |/usr/bin/adb /dev/dsk/....

If the result is "0" then the disk is not marked cluster-aware at all.

Please check all your disk devices from all node with this command. Maybe there are still some disks with a foreign ClusterID present, caused this strange behaviour.

Please have a look at the result if this script on each cluster node (prints VGIDs and ClusterIDs) and post the results:
for i in /dev/dsk/*; do
echo$i:
echo "0d8208?U;0d8332?U" |/usr/bin/adb $i
done

Best regards...
Dietmar.

Re: lvm / mcsg

V.Tamilvanan — Fri, 10 Oct 2003 03:41:57 GMT

Hi,

You need to add the VG name as cluster aware VG in Cluster Asci file (/etc/cmcluster/cmcluster.ascii file.)

Re: lvm / mcsg

Francois LAURENT — Fri, 10 Oct 2003 09:57:12 GMT

Hi Dietmar,

effectively, the clusterid is not the same between 2 package on this cluster ... I'll have to rebuild the cluster configuration (cmdeleteconf cmapplyconf) ... in fact, the clusterid comes from a srdf synchronization between 2 emc arrays.

problem is solved.
thanks a lot to all

Best regards