Doubled PV

Hello everybody

I just read that it's not recommended to run LVM on top of Softraid devices. Why is that? Looks like a perfect combination to me.

However. Recently two (?!?) disks of a RAID5 device got bad sectors. The RAID array stopped immediately. Recreating the RAID with 'mkraid -R', I could save most of its data. I did move it away with 'pvmove'.

Then - only 10 PE remaining on the bad PV - I decided to use 'pvmove -i' to remove those faulty PE's. That worked. But now my VG is in a silly
state: because 'pvmove' touched the bad sectors, the RAID shut down, i.e. the whole PV (/dev/md/20) was no longer writeable nor readable. So pvmove couldn't write to /dev/md/20, not even LVM metadata in an intact area.

Now /dev/md/20 is still registered as part of myVG. Aktivating it again with 'mkraid -R /dev/md20' I can pvdisplay on it and all metadata seems okay, but like before the 'pvmove -i'. Those 10 LE's are still on there.

'vgscan -v' outputs the following, somewhat silly message:

vgscan -- removing "/etc/lvmtab" and "/etc/lvmtab.d"
vgscan -- creating empty "/etc/lvmtab" and "/etc/lvmtab.d" vgscan -- reading all physical volumes (this may take a while...) vgscan -- scanning for all active volume group(s) first vgscan -- reading data of volume group "myVG" from physical volume(s) vgscan -- only found 6561 of 6551 LEs for LV /dev/myVG/data (2) vgscan -- ERROR "vg_read_with_pv_and_lv(): allocated LE of LV" can't get data of volume group "myVG" from physical volume(s)
vgscan -- "/etc/lvmtab" and "/etc/lvmtab.d" successfully created vgscan -- WARNING: This program does not do a VGDA backup of your volume group

That's exactly 10 LEs too much!

I don't mind too much about the data in those 10 LEs. But I would like to be able to activate myVG again!

How can I tell it, to forget about /dev/md/20? Do I have to use LVM2 tools? I've read through the mailing list, but didn't find the proper tool / command.

Thanks for your help

Markus

Markus,

what _exactly_ happened during the pvmove ?
Would help to analyze any pvmove bug.

Using LVM2 tools is a good idea.

You'll have text metadata that way, would be able to hand-edit it removing the 10 LEs and vgcfgrestore the changed metadata back onto the PVs.

Regards,
Heinz    -- The LVM Guy --

*** Software bugs are stupid.
    Nevertheless it needs not so stupid people to solve them ***

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Heinz Mauelshagen                                 Sistina Software Inc.
Senior Consultant/Developer                       Am Sonnenhang 11
                                                  56242 Marienrachdorf
                                                  Germany
[email protected]                           +49 2626 141200
                                                       FAX 924446
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Hard to reconstruct in detail, but as soon as the raid device has 'enough' failed devices, it doesn't respond to any request. It simply isn't readable anymore.

Don't think that's a bug in pvmove. Neigther in linux-softraid - just a very bad accident. (If a bug at all, I would say it's an ATA bug...)

I could restore the single PV to a state, where it had no LEs on it. (using vgcfgrestore from lvm 1). That way vgscan now finds just enough LEs and I can activate the VG again.

Lucky me... ;-)

Thanks anyway.

Markus