Moving existing VG to multipath devices

I have a RHEL 4.4 system that has been set-up incorrectly.

The disk storage is provided by an IBM SAN using IBM SDD multipath software. The system can see 4 paths to each LUN, i.e. /dev/sda, /dev/sdb, /dev/sdc and /dev/sdd are 4 paths to a single LUN. Each LUN also has a multipath device associated with it, i.e. /dev/vpatha.

Whoever built the volume group for data has not used the multipath device /dev/vpatha as a physical volume, but instead used /dev/sda. This does not give any resilience if that path goes off-line. The UUID of the 4 paths is recognized as being the same.

Is there a way to change the volume group to use the multipath devices instead of the single path devices without having to re-build?

The volume group is not a boot device, it is where the application / database data is stored.

Thanks,

I think your problem is mainly one of LVM filtering ("filter =" in /etc/lvm/lvm.conf).

LVM is probably seeing the /dev/sda device before it sees the /dev/vpatha device so that is what it is using.

(BTW, the devices you see in the files in /etc/lvm/archive and /etc/lvm/backup are just what they say. They're hints. LVM works off the metadata written to each block device it controls. Since /dev/sda, /dev/sdb, /dev/sdc, /dev/sdd and /dev/vpatha are really all the same device, the metadata is the same.)

You should be filtering out at least the underlying "/dev/sd" devices. There are good comments in /etc/lvm/lvm.conf.

My guess is you have the default filter.

    # By default we accept every block device:
    filter = [ "a/.*/" ]

If that's the case, you could probably get away with

filter =[ "r|/dev/sda|", "r|/dev/sdb|", "r|/dev/sdc|", "r|/dev/sdd|" ]

When you do a 'pvscan -vvv' you need to see somewhere in the output this

      /dev/sda: Skipping (regex)
      /dev/sdb: Skipping (regex)
      /dev/sdc: Skipping (regex)
      /dev/sdd: Skipping (regex)

Hope this helps.

Thanks for the reply - I've checked my filter setting in lvm.conf and it does appear to be wrong.

I have;
filter = [ "a/.*/" ]
filter = [ "r/sd.*/", "a/.*/" ]

My devices are sda, sdb etc. not sd.a sd.b so the second line is not working because it is wrong.

I'll get this updated and run a pvscan. I presume for a pvscan I'b be better with filesystems unmounted and applications stopped? Is pvscan destructive?

No, 'pvscan' is not destructive. It simply reads the metadata off of all of the eligible devices. I'm not altogether sure what LVM would do in this case. But if you can stop your applications, and umount the filesystems as you mentioned, it would be safer.

You could also do a 'vgchange -an <vgname>' before you do your 'pvscan', with a 'vgscan' and a 'vgchange -ay <vgname>' afterwards just to make sure you pick up the new PVs used by that volume group. ('pvscan -v3' will tell you what it's "thinking".)

In terms of your lvm.conf, you should only have one uncommented filter statement. Right now, the last one encountered is the one that is used, so you're safe; but I'd comment out that first one just to be consistent with the comments in lvm.conf:

# Don't have more than one filter line active at once: only one gets used. # Run vgscan after you change this parameter to ensure that # the cache file gets regenerated (see below). # If it doesn't do what you expect, check the output of 'vgscan -vvvv'.

Good luck with the change.

https://access.redhat.com/knowledge/solutions/47894