Recover Volume group info after boot disk crash.

[Jack Daniels_1]

We have a 2 node HP/UX 10.20 SG cluster. The boot disk failed on one of the node.So All packages were failed over to the other node.

After the disk was replaced, i couldn't bring up the system on the mirrored disk. After talking with HP, it was suggested to recover from make_recovery tape.Fortunately, i had done a make_recovery tape before shutting down. So, i was able to boot of the tape and recover everything in vg00.

However the /etc/lvmconf doesn't have the details about the other 15 shared VGs which are part of the cluster.

What is the safest/sure method to recover the VG information on this box before failing over the packages to the original setup.

The /etc/cmcluster directory is intact and the cmclconfig binary is the same as on the working node.

I am hoping that i wouldn't have to reconfigure the packages with the cmapplyconf.

Here is what i have planned.

1.On the working node,
#vgexport -m mapfile -v -p $VG_NAME
(preview mode)

2.Copy the map files to the failed node.

3.Shutdown the packages on the working node.

4.On the failed node,
Delete the lv device files from /dev/$VG_NAME
leaving just the group file.

5.On the failed node,
#vgimport -m mapfile /dev/$VG_NAME.
This will update /etc/lvmtab and create the device files for the LVs in /dev/$VG_NAME.

6.Join the failed node back to the cluster.

7.Startup the packages on the failed node and test.

Any suggestions/help would be greatly appreciated.

TIA

[Jeff Machols]

You are 100% right, doing the export and import is the best way to recover the LVM information. However I think you will have to run cmapplyconf -f to get the MC information on the failed node

[Sridhar Bhaskarla · Accepted solution]

Hi Jack,

You don't need to reconfigure the packages. You just need to import the volume groups and that should be enough apart from making sure the network configuration is restored properly.

**YOU DON'T NEED TO HALT THE CLUSTER/PACKAGE**

1. On the working node do

vgexport -p -s -m /tmp/vg##.map vg##

If you don't like -s, you can use -f

vgexport -p -m /tmp/vg##.map -f /tmp/vg##.disks vg##

Copy vg##.map and/or vg##.disks to the other system.

2. vgexport all the vg##s except the local volume groups like vg00 etc.,

3. Create volume group directories and group device files

mkdir /dev/vg##
mknod /dev/vg##/group c 64 0x0?0000

Make sure the group numbers match on both the systems. (not mandatory but a good practice)

4. Import the volume groups

vgimport -s -v -m /tmp/vg##.map vg##

If you don't like to use -s, you can use -f option to import. Make sure you edit vg##.disks and modify the C# numbers to match the hardware on the failed node.

5. Once all the VGs are imported, you can safely bring up the node as /etc/cmcluster has already been restored through mak_recovery.

cmrunnode (on the failed node)

6. Do a cmviewcl -v and make sure everything looks ok.

7. Some people edit /etc/lvmrc and disable the volume groups from being checked during the startup. Not mandatory as vgchange -c y takes care of these volume groups.

-Sri

[harry d brown jr]

JD,

When you say you did a "make_recovery" before you shutdown, did you include the -A option??


If not, you will need to restore the other parts of vg00

live free or die
harry

[Jack Daniels_1]

thanks for the feedback.

The make_recovery tape was created with -A option. So my entire vg00 got restored.

Sridhar,

Can i import VG information on the failed node when the volume groups are still active on the other node. Don't i need to vgchange to deactivate them on the other node before doing the import on the failed node.

TIA

[Jeff Machols]

In order to export a volume group it must be deactivated.

[G. Vrijhoeven]

Hi,

No.
You can import volume groups that are active on an other node. You can not activate them.

Gideon