[linux-lvm] vgscan goes into an infinite loop (migration issues)

[linux-lvm] vgscan goes into an infinite loop (migration issues)

[Navindra Umanee]

Hi,

I have an LVM1 system working under 2.4.  I've installed 2.6.0 with
devmapper and the LVM2 tools.  Now I'm trying to get 2.6.0 to mount my
LVM1 system or at least to convert LVM1 to LVM2.

I have a single IDE harddrive with one VG named /dev/bytepool
containing four LVM partitions.  I have a few LVs under
/dev/bytepool/.

vgconvert runs for a long long time, uses more than 3G of memory
(including swap) and then fails with an out of memory error.  I can
increase swap if it will help, but I don't think it will. 

vgscan does the same thing with global/format set to LVM1.  So I put
vgscan in verbose and debug mode:

locking/file_locking.c:158   Locking /var/lock/lvm/V_bytepool RB
toollib.c:305   Finding volume group "bytepool"
device/dev-io.c:325   Opened /dev/hda6
format1/disk-rep.c:353   Found /dev/hda6 in VG bytepool
device/dev-io.c:325   Opened /dev/hda7
format1/disk-rep.c:353   Found /dev/hda7 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda7
device/dev-io.c:325   Opened /dev/hda8
format1/disk-rep.c:353   Found /dev/hda8 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda8
device/dev-io.c:325   Opened /dev/hda9
format1/disk-rep.c:353   Found /dev/hda9 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda9
format1/disk-rep.c:353   Found /dev/hda6 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda6
format1/disk-rep.c:353   Found /dev/hda7 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda7
format1/disk-rep.c:353   Found /dev/hda8 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda8
format1/disk-rep.c:353   Found /dev/hda9 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda9
format1/disk-rep.c:353   Found /dev/hda6 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda6
format1/disk-rep.c:353   Found /dev/hda7 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda7
format1/disk-rep.c:353   Found /dev/hda8 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda8
format1/disk-rep.c:353   Found /dev/hda9 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda9
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda6
format1/disk-rep.c:353   Found /dev/hda7 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda7
format1/disk-rep.c:353   Found /dev/hda8 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda8
format1/disk-rep.c:353   Found /dev/hda9 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda9
format1/disk-rep.c:353   Found /dev/hda6 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda6
format1/disk-rep.c:353   Found /dev/hda7 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda7
format1/disk-rep.c:353   Found /dev/hda8 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda8
format1/disk-rep.c:353   Found /dev/hda9 in VG bytepool
format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda9
[...]

Basically it's an infinite loop.  Any idea what is going on?  I would
greatly appreciate any suggestions.

Btw, it was quite an unpleasant shock to find that Linux 2.6.0 no
longer supported my LVM filesystems -- the kernel documentation
completely fails to mention anything at all about LVM changes.  Can
someone take care of this oversight?  

I *am* really glad my partitions and data have not been toasted so
far...  :-)

Thanks,
Navin.

Re: [linux-lvm] vgscan goes into an infinite loop (migration issues)

[Alasdair G Kergon]

On Tue, Jan 06, 2004 at 04:57:19PM -0500, Navindra Umanee wrote:
> format1/disk-rep.c:392   Ignoring duplicate PV  on /dev/hda7

Whatever goes wrong, it shouldn't loop!

And there's a clue in that error message: the PV's UUID is meant 
to appear between 'PV' and 'on' - but it's empty.

Please send me a copy of your metadata so I can try to reproduce
the problem.

Alasdair
-- 
agk@uk.sistina.com

Re: [linux-lvm] vgscan goes into an infinite loop (migration issues)

[Navindra Umanee]

Alasdair G Kergon <agk@uk.sistina.com> wrote:
> And there's a clue in that error message: the PV's UUID is meant 
> to appear between 'PV' and 'on' - but it's empty.

Hmmm.  Does this help:

[navindra@vinata navindra]# lvm1-pvdisplay /dev/hda6
--- Physical volume ---
PV Name               /dev/ide/host0/bus0/target0/lun0/part6
VG Name               bytepool
PV Size               10 GB [20964762 secs] / NOT usable 32.19 MB [LVM: 129 KB]
PV#                   1
PV Status             available
Allocatable           yes (but full)
Cur LV                5
PE Size (KByte)       32768
Total PE              318
Free PE               0
Allocated PE          318
PV UUID               none

As you see, it claims there is no UUID.

> Please send me a copy of your metadata so I can try to reproduce
> the problem.

Thanks a lot for your followup!  I've sent the information you requested.

Cheers,
Navin.

Re: [linux-lvm] vgscan goes into an infinite loop (migration issues)

[Alasdair G Kergon]

On Tue, Jan 06, 2004 at 06:13:11PM -0500, Navindra Umanee wrote:
> PV UUID               none

LVM2 uses the UUID to distinguish between PVs - so it sees all the
devices as duplicates of each other.

I'll try to change the code so it generates a UUID if it's missing.
 
Alasdair
-- 
agk@uk.sistina.com