[Gllug] DD Ext3 Move

Nix nix at esperi.org.uk
Wed Mar 1 20:40:09 UTC 2006


On Tue, 28 Feb 2006, Christian Smith wrote:
> On Sat, 25 Feb 2006, Nix wrote:
>>Look out, though; pvmoving swap partitions seems to be decidedly
>>unreliable in 2.6.15, even if you've swapoffed: I've had cause to try it
>>three or four times and ended up with my swap device turned into a
>>SUSPENDED device-mapper target, the pvmove stalled, and the system
>>unusable (all fork()s hit D state).
> 
> 
> Of swapoffed, why bother moving if you're having problems? Just blow it
> away. You only have to re-mkswap(*) the new volume.

I'd have *done* that if I'd known about the problem in advance, but I
just did a

pvmove /dev/sdb

and watched it freeze shortly thereafter...

>>Upon reboot pvmove noticed that it had been interrupted and restarted
>>happily. LVM's approach to robustness is truly excellent. :)
> 
> I've lost a VG, but that was me not removing a physical volume correctly.

I made that mistake as well, and recovered thanks to a lot of gdb'ing, a
hex editor, and vgcfgrestore. (The idiot in question had no backups, and
no it wasn't me. He has backups *now*.)

> LVM2, and the use of device-mapper, is really nice. Shame old RH used to
> enable LVM before MD, thus you couldn't have LVM on top of MD! I think
> that's probably fixed now?

Err, you can enable LVM more than once: indeed if you're starting LVM
from an initrd or initramfs, you probably need to do just that to get
the device nodes created on both root filesystems.

-- 
`... follow the bouncing internment camps.' --- Peter da Silva
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list