[linux-lvm] LVM commands won't run w/failed PV [solved?]

[Ross Boylan]

On Sun, 2011-07-03 at 23:36 -0700, Ross Boylan wrote:
> I have a VG made from several PV's, one of which (mostly) failed.
> Anticipating this, I had moved all key LVs to be backed by other disks.
> The system runs off those LVs.
> 
> When I try to run LVM commands, I get lots of errors, e.g.,
> # lvs daisy
>   /dev/dm-9: read failed after 0 of 4096 at 0: Input/output error
>   /dev/dm-10: read failed after 0 of 4096 at 0: Input/output error
>   /dev/dm-11: read failed after 0 of 4096 at 0: Input/output error
>   /dev/dm-12: read failed after 0 of 4096 at 0: Input/output error
> [etc]
>   /dev/sdb: read failed after 0 of 4096 at 0: Input/output error
>   /dev/sdb1: read failed after 0 of 2048 at 0: Input/output error
>   /dev/sdb2: read failed after 0 of 2048 at 0: Input/output error
>   /dev/sdb5: read failed after 0 of 4096 at 0: Input/output error
>   Couldn't find device with uuid 'qqWQc6-Ucv9-8htm-TnOz-n1Va-9L0g-H3WA6o'.
>   Couldn't find all physical volumes for volume group daisy.
>   Volume group "daisy" not found
> 
> The last message in particular is weird, since many LVs from daisy are
> successfully mounted.
> 
> How can I get LVM to overlook the problem, or access the setup enough to
> remove the bad PV from the VG?  I believe sdb5 is the only PV from sdb
> that is in the VG.
vgreduce --removemissing daisy
seems to have worked.  Of course the LVs on the bad PV are gone.  The
man page says that even if an LV is only partly on the bad PV it is
still elimated entirely, and recommends --partial (see below) if you
want to attempt recovery first.

vgreduce daisy /dev/sdb
would have been disastrous, I think.  The failed disk was at sdb, but
when I pulled it the good disk (holding all the remaining PVs for daisy)
was at sdb.

I'd be curious if there's any other way.
vgchange with --partial did activate what it could, but it said it put
everything in read-only mode.
> 
> It may be relevant that the bad drive sort of comes up: during system
> startup it was detected and some of its LVs were mounted.  This led to
> an apparently successful replay of the log on one LV, and the start of
> an fsck on another before errors apparently caused it to be dropped.
I hoped that if I removed the bad disk entirely it would help.  Instead,
my initrd was completely unable to bring up daisy, including its root
partition.  So the system would not run until the vgreduce
--removemissing.  I did my repairs from the initrd.

Ross
> 
> # vgs --version
>   LVM version:     2.02.39 (2008-06-27)
>   Library version: 1.02.27 (2008-06-25)
>   Driver version:  4.13.0
> The failed disk is SATA.  Running Debian Lenny with Linux 2.6.26-2-686
> kernel.
> 
> Thanks.
> Ross Boylan
> 
> P.S. Yes, I'm thinking about RAID.  No, there is no RAID in the current
> system at any level (lvm, dm, hardware).
>