I'm looking to see if I can rescue a guest VM from a LVM Storage Pool.

I have a server with 2 disks. The CentOS 6.4 HOST OS was on /dev/sda and I had /dev/sdb dedicated to a LVM based storage pool where I had only one CentOS guest VM.

The HOST OS disk /dev/sda is broken. Doesn't appear to be recoverable and no backups (I know, that will be my next project after I get this all back up and running). All rescue tools I've tried result in many IO errors on sda.

The 2'nd disk sdb is intact but I don't know how to retrieve anything from it.

When I set this up, I followed the RedHat Virtualization Admin manual "12.1.4. LVM-based storage pools". So sdb had one physical volume and a volume group.


I inserted the good sdb disk into another/different healthy CentOS 6.5 server (where it shows up as /dev/sdc). lsblk shows this:

sdc 8:32 0 223.6G 0 disk
\_sdc1 8:33 0 223G 0 part

pvs and vgs don't show anything for the disk I'm trying to recover (sdc on the healthy server).

Is there a way to rescue the healthy disk to retrieve files from it or even boot the guest VM?

pvs just reports volumes the system already knows about. You need to run pvscan to search for volumes not already known to the system.

Should we infer from this you have a similar setup as the other machine (sda/sdb) ?. Same named vgs and lvs maybe ?.

If so, you can't see the added disk entities until you rename them. Used to be a right PITA, but now I think vgrename, lvrename might help - haven't had a need to try them.

You need may to activate lvm on the disk you're trying to retrieve data from. If so, the command is:

vgchange -a y

There will be a problem if the testing system has any logical volumes with the same name as the recover disk had, if so you could try using a system booted off a live linux CD (fedora live cd works for me).

The old machine had sda+sdb, where sda is the HOST OS and broken/not-bootable hard drive. sdb is the drive I want to rescue the virtual machine from. In the new machine there is sda+sdb+sdc, where sda is a USB key with ESX on it (not using it, should have removed it), sdb is fresh install of CentOS and sdc is the drive I'm trying to rescue (pulled sdb from the other machine).

The VGs are named differently so shouldn't have collision on names. The VG has the machine name in it and the machines have different names.

pvscan & vgscan didn't show me any other physical volumes or volume groups from sdc.

Am I to presume that when you create PV's/VG's/LV's, that info lives on the originating hard drive the original OS was booted from? This would prevent the hard drives from being portable to other machines. I don't know too much about how PV/VG/LV's work. This would also mean, if you loose the boot OS, you loose ALL the disks in the machine. I can't believe linux would be that crippling - I must be doing something wrong here.

The LVM metadata is part of the LVM volume. The files under /etc/lvm are just a textual presentation and backup of that metadata.

Let's see what the disk is labeled as:

I used a program called testdisk to search for partitions and it found a lvm partition and updated the partition table. After powering down the machine and booting up, pvscan and vgscan found a VG+LV and I'm able to mount it and read all the data.

I doubt I'll be able to boot it though as the /boot partition isn't there. I wish I knew how this disappeared to begin with, but I'm glad I was able to rescue the filesystem to make a backup of it.

fdisk -l /dev/sdc output:

from "parted -l"

Sounds like you got the important stuff back. Please consider marking this thread as "Solved" (in the Thread Tools near the top of the page).