[linux-lvm] safe to ru-run vgscan after mounting LV root?

[linux-lvm] safe to ru-run vgscan after mounting LV root?

[Mark van Walraven]

Hi All,

First, let me say how impressed I am with LVM.  I've just set up a 2.4.2
system with all filesystems in logical volumes, on an MD mirror.  I am
an experienced (i.e., "jaded") sysadmin, so please take my enthusiasm
as a truly singular occurrence.  Kudos to you.

Now for my question: The system above runs vgscan (and vgchange) from an
initrd before mounting the root filesystem.  After that, the /etc/lvmtab*
files and directories need to be refreshed from the initrd.  Instead of
keeping the initrd hanging around until then, could vgscan simply be
run again to recreate /etc/lvmtab*, even though PVs and LVs are active?

Thanks,

Mark.

Re: [linux-lvm] safe to ru-run vgscan after mounting LV root?

[Patrick Caulfield]

On Wed, Mar 28, 2001 at 02:14:21AM +1200, Mark van Walraven wrote:
> Hi All,
> 
> First, let me say how impressed I am with LVM.  I've just set up a 2.4.2
> system with all filesystems in logical volumes, on an MD mirror.  I am
> an experienced (i.e., "jaded") sysadmin, so please take my enthusiasm
> as a truly singular occurrence.  Kudos to you.
> 
> Now for my question: The system above runs vgscan (and vgchange) from an
> initrd before mounting the root filesystem.  After that, the /etc/lvmtab*
> files and directories need to be refreshed from the initrd.  Instead of
> keeping the initrd hanging around until then, could vgscan simply be
> run again to recreate /etc/lvmtab*, even though PVs and LVs are active?

You can run vgscan as many times as you like. The main reason you wouldn't is
the time it can take. But even that is small compared to the initialisation time
of most SCSI cards so at bootup it shouldn't be a problem.

patrick

Re: [linux-lvm] safe to ru-run vgscan after mounting LV root?

[Jay Sissom]

Just a warning.  When I run vgscan on my system, it deletes my LV.  I
still don't know why.  There seems to be a terrible bug in vgscan.

Be careful if you run it.  You might have to do a vgcfgrestore to get your
LV's back.

I am using the beta6 code, so I don't have old code.

Jay

On Tue, 27 Mar 2001, Patrick Caulfield wrote:

> On Wed, Mar 28, 2001 at 02:14:21AM +1200, Mark van Walraven wrote:
> > Hi All,
> >
> > First, let me say how impressed I am with LVM.  I've just set up a 2.4.2
> > system with all filesystems in logical volumes, on an MD mirror.  I am
> > an experienced (i.e., "jaded") sysadmin, so please take my enthusiasm
> > as a truly singular occurrence.  Kudos to you.
> >
> > Now for my question: The system above runs vgscan (and vgchange) from an
> > initrd before mounting the root filesystem.  After that, the /etc/lvmtab*
> > files and directories need to be refreshed from the initrd.  Instead of
> > keeping the initrd hanging around until then, could vgscan simply be
> > run again to recreate /etc/lvmtab*, even though PVs and LVs are active?
>
> You can run vgscan as many times as you like. The main reason you wouldn't is
> the time it can take. But even that is small compared to the initialisation time
> of most SCSI cards so at bootup it shouldn't be a problem.
>
> patrick
>
> _______________________________________________
> linux-lvm mailing list
> linux-lvm@sistina.com
> http://lists.sistina.com/mailman/listinfo/linux-lvm
>

Re: [linux-lvm] safe to ru-run vgscan after mounting LV root?

[Mark van Walraven]

On Tue, 27 Mar 2001, Patrick Caulfield wrote:
> You can run vgscan as many times as you like. The main reason you wouldn't is
> the time it can take. But even that is small compared to the initialisation time

Thanks for the feedback.  That's what I figured, but ...

On Tue, Mar 27, 2001 at 11:01:50AM -0500, Jay Sissom wrote:
> Just a warning.  When I run vgscan on my system, it deletes my LV.  I
> still don't know why.  There seems to be a terrible bug in vgscan.

... this worries me.

Could anyone else comment on this, please?  Is Jay's problem bizzare
and irreproducable, or is there possibly a problem with replacing the
lvmtab files from beneath the noses of the other tools?

Thanks,

Mark.

Re: [linux-lvm] safe to ru-run vgscan after mounting LV root?

[Ed Tomlinson]

On Wednesday 28 March 2001 23:38, you wrote:
> On Tue, 27 Mar 2001, Patrick Caulfield wrote:
> > You can run vgscan as many times as you like. The main reason you
> > wouldn't is the time it can take. But even that is small compared to the
> > initialisation time
>
> Thanks for the feedback.  That's what I figured, but ...
>
> On Tue, Mar 27, 2001 at 11:01:50AM -0500, Jay Sissom wrote:
> > Just a warning.  When I run vgscan on my system, it deletes my LV.  I
> > still don't know why.  There seems to be a terrible bug in vgscan.

What release of the utilities/patch?  I had this problem with beta2, which
is what ships with the kernel.  Upgrading to beta6 fixed it _here_.

> ... this worries me.

Me too...  Here hoping that beta6 is the answer.

> Could anyone else comment on this, please?  Is Jay's problem bizzare
> and irreproducable, or is there possibly a problem with replacing the
> lvmtab files from beneath the noses of the other tools?

Ed Tomlinson

Re: [linux-lvm] safe to ru-run vgscan after mounting LV root?

[Jay Sissom]

Nope, I am running kernel 2.4.2 with the beta6 patch and the beta6
utilities.

This started after I resized a LV.  The answer from this list was, "don't
run vgscan".  Now people on the list say you can run it as many times as
you like.  If I run it once, all my LV's are gone.  I'd like to figure out
what the problem is, but don't know where to start looking.

On Thu, 29 Mar 2001, Ed Tomlinson wrote:

> On Wednesday 28 March 2001 23:38, you wrote:
> > On Tue, 27 Mar 2001, Patrick Caulfield wrote:
> > > You can run vgscan as many times as you like. The main reason you
> > > wouldn't is the time it can take. But even that is small compared to the
> > > initialisation time
> >
> > Thanks for the feedback.  That's what I figured, but ...
> >
> > On Tue, Mar 27, 2001 at 11:01:50AM -0500, Jay Sissom wrote:
> > > Just a warning.  When I run vgscan on my system, it deletes my LV.  I
> > > still don't know why.  There seems to be a terrible bug in vgscan.
>
> What release of the utilities/patch?  I had this problem with beta2, which
> is what ships with the kernel.  Upgrading to beta6 fixed it _here_.
>
> > ... this worries me.
>
> Me too...  Here hoping that beta6 is the answer.
>
> > Could anyone else comment on this, please?  Is Jay's problem bizzare
> > and irreproducable, or is there possibly a problem with replacing the
> > lvmtab files from beneath the noses of the other tools?
>
> Ed Tomlinson
> _______________________________________________
> linux-lvm mailing list
> linux-lvm@sistina.com
> http://lists.sistina.com/mailman/listinfo/linux-lvm
>

Re: [linux-lvm] safe to ru-run vgscan after mounting LV root?

[Andreas Dilger]

Jay Sissom writes:
> Nope, I am running kernel 2.4.2 with the beta6 patch and the beta6
> utilities.

OK, that's a bad sign.

> This started after I resized a LV.  The answer from this list was, "don't
> run vgscan".  Now people on the list say you can run it as many times as
> you like.  If I run it once, all my LV's are gone.  I'd like to figure out
> what the problem is, but don't know where to start looking.

You _should_ be able to re-run vgscan, but in most cases (as you have
seen) it is not _necessary_ to do so.

In any case, how to track down the problem - run "vgscan -v -d" and
start going through the output with a fine-tooth comb.  Basically,
at some point you will start getting errors, and that is what is bad
(it is OK to get error codes in a few places, like when checking if
a partition/disk has an LVM header on it).

For a VG to be operative, you need all of the PVs in it to be OK.
This seems to be a common source of problems.  I have a modified
pv_read_all_pv() and pv_read_all_pv_of_vg() which is a lot better
in terms of reporting why a PV was discarded from the PV list. This
may make finding the source of the problem easier (even obvious ;-).

You can find it in an email to the lvm-devel mailing list dated
March 15 "Re: [lvm-devel] Extent list sorting considered buggy".

Cheers, Andreas
-- 
Andreas Dilger  \ "If a man ate a pound of pasta and a pound of antipasto,
                 \  would they cancel out, leaving him still hungry?"
http://www-mddsp.enel.ucalgary.ca/People/adilger/               -- Dogbert

Re: [linux-lvm] safe to ru-run vgscan after mounting LV root?

[Patrick Caulfield]

On Thu, Mar 29, 2001 at 04:38:16PM +1200, Mark van Walraven wrote:
> On Tue, 27 Mar 2001, Patrick Caulfield wrote:
> > You can run vgscan as many times as you like. The main reason you wouldn't is
> > the time it can take. But even that is small compared to the initialisation time
> 
> Thanks for the feedback.  That's what I figured, but ...
> 
> On Tue, Mar 27, 2001 at 11:01:50AM -0500, Jay Sissom wrote:
> > Just a warning.  When I run vgscan on my system, it deletes my LV.  I
> > still don't know why.  There seems to be a terrible bug in vgscan.
> 
> ... this worries me.
> 
> Could anyone else comment on this, please?  Is Jay's problem bizzare
> and irreproducable, or is there possibly a problem with replacing the
> lvmtab files from beneath the noses of the other tools?

I've never had vgscan do that to me, though I'm not denying that it happens.
There certainly have been some odd vgscan reports on the list over the last few
months. Most of them seem to be from systems that have been upgraded from 0.8 I
don't know if it is true in this case.

There is no fundamental problem with replacing the lvmtab files under the noses
of other tools because the LVM locking takes care of that. If stuff is getting
corrupted then it's a bug of course.

patrick