Re: [libvirt] [PATCH] Drop virStorageBackendLogicalMatchPoolSource

On 06/17/2016 11:16 AM, Ján Tomko wrote: > On Thu, Jun 16, 2016 at 07:44:05AM -0400, John Ferlan wrote: >> >> >> On 06/16/2016 06:03 AM, Ján Tomko wrote: >>> On Wed, Jun 15, 2016 at 06:45:59PM -0400, John Ferlan wrote: >>>> >>>> >>>> On 06/15/2016 01:19 PM, Ján Tomko wrote: >>>>> Regression introduced by commit 71b803a for [1] that prevents >>>>> starting up >>>>> a logical pool created with <source><device path=''></source> >>>>> after it has been moved to a different physical volume. >>>> >>>> Is there a bug for this? XML examples? >>>> >>>> Is an empty source device path string a valid value? Reading >>>> http://libvirt.org/formatstorage.html doesn't give me that impression. >>> >>> I meant any pool with a <device> specified, not an empty path. >> >> But the patch is specifically targeted at removing the _logical pool >> check. So the question is, is it "legal" for a logical pool to "start" >> with an empty string as a source device path? Secondarily I'm curious >> what's the use case of that? > > I do not know what happens with an empty string. > <source><device path=''></source> was not an exact snippet, > just a hint that there is a device path specified. > > If you create a pool like this: > > <pool type='logical'> > <name>vgname</name> > <source> > <device path='/dev/sda4'/> > <name>vgname</name> As I see things, this more or less lets libvirt "know" that "/dev/sda4" and "vgname" are associated. > </source> > <target> > <path>/dev/vgname</path> > </target> > </pool> > > Then pvmove /dev/sda4 to /dev/sda5, libvirt will refuse that pool > even though <name>vgname</name> is enough to uniquely identify a storage > pool. > As an admin you take this step to pvmove your data from /dev/sda4 to /dev/sda5. But then you expect libvirt to know that? Do you expect libvirt to automagically update the XML to /dev/sda5 during 'build' or 'start'? >> >>> >>> The whole point of LVM is abstraction from the lower layers so we >>> shouldn't ever be checking this. >>> >> >> So it's OK to activate/start a pool where the source device path is >> incorrect? > > For LVM, the important path is in <source><name>. > But isn't there a 1-to-1 relationship between the VG and the PV? A PV cannot be in more than one VG, so if you move your PV, then you have to update your XML to let libvirt know.

John