[et-mgmt-tools] koan --virt: lvm support
mlkb at x-tend.be
Wed Sep 5 09:18:49 UTC 2007
On Wed, 2007-09-05 at 04:28 +0100, Daniel P. Berrange wrote:
> On Tue, Sep 04, 2007 at 01:55:53PM -0400, Michael DeHaan wrote:
> > Johan Huysmans wrote:
> > >well... it is possible ;)
> > >because we are allready doing this for over a year.
> > >
> > The word on the street (from Dan Berrange) is that this feature is not
> > guaranteed to work in future
> > versions of Xen.
> To be a little more verbose....
> Virtual machines are assigned virtual *disks*. ie hdN, sdN (for full virt) or
> xvdN (for paravirt). Names like sda1 are partitions, and not disks. It makes
> no conceptual sense to have a partition without an enclosing disk. If you want
> to have a 1-1 mapping between the backing files & disks inside the guest, then
> simply don't partition the disk inside the guest - format /dev/sda directly.
> Xen paravirt only happens to allow you to map a file straight through as a
> partition, and while it may currently work, I can make absolutely no guarentees
> it will work in the future as the Xen paravirt kernel evolves - particularly
> as Xen increasingly merges into LKML trees. NB with the LKML merge the ability
> to hijack hdN and sdN in paravirt guests has been removed - paravirt can only
> use xvdN devices now. I've not checked if hijacking partitions as described
> below works, but I would not be surprised if it doesn't. Finally its also not
> portable to Xen fullyvirt, or KVM, or QEMU, etc. Basically unsupportable in
> the medium-to-long term.
I think it is extremely important to continue to be able to use
filesystems on LVM volumes in dom0 as filesystems in domU's. Using
them as full disks with a partition on them would be fine.
Kris Buytaert <mlkb at x-tend.be>
More information about the et-mgmt-tools