[linux-lvm] Persistence of disk-based LVM data after rebuild

justin clancyj at tiscali.co.uk
Thu Feb 23 16:54:21 UTC 2006


Hi guys,

I came across a weird problem with LVM after a server rebuild.  I
thought you might like to know about it.

Details:

  LVM version:     2.01.14 (2005-08-04)
  Library version: 1.01.04 (2005-08-02)
  Driver version:  4.4.0

  SUSE 10 with 2.6.13-15-default kernel.

I had two PVs (/dev/hdc2 and /dev/hdd2) in a VG called vg01 on the
original system.  After the rebuild I assigned the first partition of
each disk to swap (as before) and the other two partitions to a raid
mirror and an LV.

Everything worked well until I tried to issue commands like vgdisplay,
when I got messages like the following:

  LV work: inconsistent LE count 16 != 32

  Internal error: LV segments corrupted in work.

  Volume group "vg01" doesn't exist


...which, of course, it didn't anymore.  I did an strace on vgdisplay to
see where the data came from and saw that it was the first partition of
each disk.  I took the swap offline and issued the command:

dd if=/dev/zero of=/dev/hdc1 count=10

...on each disk to zap out any metadata on those partitions.  Now it's fine.

Any comments?

  Regards,

    Justin.




More information about the linux-lvm mailing list