[linux-lvm] getting pe_start right

[linux-lvm] getting pe_start right

[Phillip Susi]

While trying to convert my mdadm raid5 array from metadata format 0.9 to 
1.0, I forgot to specify the version and it defaulted to 1.2, which 
damaged the lvm metadata.  After correcting mdadm to 1.0, I decided to 
try to recreate the LVM after managing to locate the volume descriptions 
in the metadata after manually dumping it.

The problem I now have is that pvcreate will not use the same pe_start 
as before.  It was 384 and it is now 3072.  I tried using --metadatasize 
192k --dataalignmentsize 0 and it did not help.  What gives?

Re: [linux-lvm] getting pe_start right

[CHANDRASEKARAN, GIRIDHAR (GIRI) (ATTSI)]

Try "--metadatasize 288k

-----Original Message-----
From: linux-lvm-bounces@redhat.com [mailto:linux-lvm-bounces@redhat.com]
On Behalf Of Phillip Susi
Sent: Friday, May 06, 2011 10:56 AM
To: linux-lvm@redhat.com
Subject: [linux-lvm] getting pe_start right

While trying to convert my mdadm raid5 array from metadata format 0.9 to

1.0, I forgot to specify the version and it defaulted to 1.2, which 
damaged the lvm metadata.  After correcting mdadm to 1.0, I decided to 
try to recreate the LVM after managing to locate the volume descriptions

in the metadata after manually dumping it.

The problem I now have is that pvcreate will not use the same pe_start 
as before.  It was 384 and it is now 3072.  I tried using --metadatasize

192k --dataalignmentsize 0 and it did not help.  What gives?

_______________________________________________
linux-lvm mailing list
linux-lvm@redhat.com
https://www.redhat.com/mailman/listinfo/linux-lvm
read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/

Re: [linux-lvm] getting pe_start right

[Milan Broz]

On 05/06/2011 04:55 PM, Phillip Susi wrote:
> While trying to convert my mdadm raid5 array from metadata format 0.9 to 
> 1.0, I forgot to specify the version and it defaulted to 1.2, which 
> damaged the lvm metadata.  After correcting mdadm to 1.0, I decided to 
> try to recreate the LVM after managing to locate the volume descriptions 
> in the metadata after manually dumping it.
> 
> The problem I now have is that pvcreate will not use the same pe_start 
> as before.  It was 384 and it is now 3072.  I tried using --metadatasize 
> 192k --dataalignmentsize 0 and it did not help.  What gives?

Try to set
default_data_alignment = 0

from lvm.conf:
# Default alignment of the start of a data area in MB.  If set to 0,
# a value of 64KB will be used.  Set to 1 for 1MiB, 2 for 2MiB, etc.

Also, if you use

	pvcreate --uuid <pv uuid from metadata> --restorefile <old VG backup>

it should create PV exacltly as needed for olf VG (with --restorefile it uses
offset from metadata backup, overriding default)

Then vgcfgrestore and you should have exactly the same cfg.

Milan

Re: [linux-lvm] getting pe_start right

[Mike Snitzer]

On Fri, May 06 2011 at 10:55am -0400,
Phillip Susi <psusi@cfl.rr.com> wrote:

> While trying to convert my mdadm raid5 array from metadata format
> 0.9 to 1.0, I forgot to specify the version and it defaulted to 1.2,
> which damaged the lvm metadata.  After correcting mdadm to 1.0, I
> decided to try to recreate the LVM after managing to locate the
> volume descriptions in the metadata after manually dumping it.
> 
> The problem I now have is that pvcreate will not use the same
> pe_start as before.  It was 384 and it is now 3072.  I tried using
> --metadatasize 192k --dataalignmentsize 0 and it did not help.  What
> gives?

Would be quite helpful if you shared kernel and lvm2 version info...

chances are lvm2's data_alignment_detection will shift the data start to
the start of the next full stripe of the raid5.

You should be able to specify the old start with:

--dataalignment 192k --dataalignmentoffset 0

(--dataalignmentoffset 0 likely isn't needed, but it just overrides
lvm2's data_alignment_offset detection if it just so happens that
raid5 has alignment_offset!=0)

Re: [linux-lvm] getting pe_start right

[Phillip Susi]

On 5/6/2011 11:21 AM, Mike Snitzer wrote:
> chances are lvm2's data_alignment_detection will shift the data start to
> the start of the next full stripe of the raid5.

That appears to be what was going on.  I found entries in lvm.conf to 
enable/disable autodetection of the raid stripe width and turned it off.

> You should be able to specify the old start with:
>
> --dataalignment 192k --dataalignmentoffset 0
>
> (--dataalignmentoffset 0 likely isn't needed, but it just overrides
> lvm2's data_alignment_offset detection if it just so happens that
> raid5 has alignment_offset!=0)

The command line switches did not appear to override the autodetection; 
I had to shut it off in the conf file.  Is this a bug?