[linux-lvm] LVM mirror logic bug?

[linux-lvm] LVM mirror logic bug?

[Robert]

Hi,

Today I created a mirror due to some data migrations I am doing, and
the funny thing is that two mirror legs ended up on the same physical
volume:

# lvs -a -o +devices
  LV                VG     Attr   LSize   Origin Snap%  Move Log Copy%
 Convert Devices
  oraarch           sapvg  -wi-ao  20.00G
         /dev/dm-18(0)
  oracle            sapvg  -wi-ao   5.00G
         /dev/dm-18(5120)
  orapta            sapvg  mwi-ao 299.12G
100.00         orapta_mimage_0(0),orapta_mimage_1(0)
  [orapta_mimage_0] sapvg  iwi-ao 299.12G
         /dev/dm-5(11520)
  [orapta_mimage_0] sapvg  iwi-ao 299.12G
         /dev/dm-3(12800)
  [orapta_mimage_0] sapvg  iwi-ao 299.12G
         /dev/dm-3(30002)
  [orapta_mimage_0] sapvg  iwi-ao 299.12G
         /dev/dm-4(0)
  [orapta_mimage_0] sapvg  iwi-ao 299.12G
         /dev/dm-4(9913)
  [orapta_mimage_0] sapvg  iwi-ao 299.12G
         /dev/dm-6(0)
  [orapta_mimage_1] sapvg  iwi-ao 299.12G
         /dev/dm-6(34249)
  [orapta_mimage_1] sapvg  iwi-ao 299.12G
         /dev/dm-16(0)
  [orapta_mimage_1] sapvg  iwi-ao 299.12G
         /dev/dm-17(0)
...
...

Here you can see /dev/dm-6 contains part of both mirror legs.

I have to admit I just used the command:
lvconvert -m1 --corelog sapvg/orapta
not specifying the target pv's as I thought LVM would calculate it for itself.

There should be a rule in place to never use the same PV for the
second leg of a mirror that is used by the first shoudln't there?

I am using a statically compiled version of LVM so I can use it on
different machines that are older than SLES11SP2 as the older distro
won't allow me to mirror multi segmented LV's.

# /tmp/lvm.static version
  LVM version:     2.02.98(2) (2012-10-15)
  Library version: 1.02.77 (2012-10-15)
  Driver version:  4.15.0

Thanks!
Robert

Re: [linux-lvm] LVM mirror logic bug?

[Lars Ellenberg]

On Wed, Feb 06, 2013 at 01:43:05PM +0200, Robert wrote:
> Hi,
> 
> Today I created a mirror due to some data migrations I am doing, and
> the funny thing is that two mirror legs ended up on the same physical
> volume:
> 
> # lvs -a -o +devices
>   LV                VG     Attr   LSize   Origin Snap%  Move Log Copy%
>  Convert Devices
>   oraarch           sapvg  -wi-ao  20.00G
>          /dev/dm-18(0)
>   oracle            sapvg  -wi-ao   5.00G
>          /dev/dm-18(5120)
>   orapta            sapvg  mwi-ao 299.12G
> 100.00         orapta_mimage_0(0),orapta_mimage_1(0)
>   [orapta_mimage_0] sapvg  iwi-ao 299.12G
>          /dev/dm-5(11520)
>   [orapta_mimage_0] sapvg  iwi-ao 299.12G
>          /dev/dm-3(12800)
>   [orapta_mimage_0] sapvg  iwi-ao 299.12G
>          /dev/dm-3(30002)
>   [orapta_mimage_0] sapvg  iwi-ao 299.12G
>          /dev/dm-4(0)
>   [orapta_mimage_0] sapvg  iwi-ao 299.12G
>          /dev/dm-4(9913)
>   [orapta_mimage_0] sapvg  iwi-ao 299.12G
>          /dev/dm-6(0)
>   [orapta_mimage_1] sapvg  iwi-ao 299.12G
>          /dev/dm-6(34249)
>   [orapta_mimage_1] sapvg  iwi-ao 299.12G
>          /dev/dm-16(0)
>   [orapta_mimage_1] sapvg  iwi-ao 299.12G
>          /dev/dm-17(0)
> ...
> ...
> 
> Here you can see /dev/dm-6 contains part of both mirror legs.
> 
> I have to admit I just used the command:
> lvconvert -m1 --corelog sapvg/orapta
> not specifying the target pv's as I thought LVM would calculate it for itself.
> 
> There should be a rule in place to never use the same PV for the
> second leg of a mirror that is used by the first shoudln't there?

I assume you have an --alloc=anywhere setting somewhere.
Try to reproduce with an explicit --alloc=normal.
See also man lvm

	Lars

> I am using a statically compiled version of LVM so I can use it on
> different machines that are older than SLES11SP2 as the older distro
> won't allow me to mirror multi segmented LV's.
> 
> # /tmp/lvm.static version
>   LVM version:     2.02.98(2) (2012-10-15)
>   Library version: 1.02.77 (2012-10-15)
>   Driver version:  4.15.0
> 
> Thanks!
> Robert

-- 
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com

Re: [linux-lvm] LVM mirror logic bug?

[Alasdair G Kergon]

I forget whether this is a case that's meant to work yet or not.
If you can reproduce with the latest upstream code (and 
allocation/maximise_cling not set to 0 of course) post a link to the -vvvv
trace which shows exactly how it's doing the allocation.

Alasdair