← Back to team overview

openstack team mailing list archive

Re: Should we discourage KVM block-based live migration?

 

On Wed, Apr 24, 2013 at 05:23:11PM -0400, Lorin Hochstein wrote:
> On Wed, Apr 24, 2013 at 11:59 AM, Daniel P. Berrange <dan@xxxxxxxxxxxx>wrote:
> 
> > On Wed, Apr 24, 2013 at 11:48:35AM -0400, Lorin Hochstein wrote:
> > > In the docs, we describe how to configure KVM block-based live migration,
> > > and it has the advantage of avoiding the need for shared storage of
> > > instances.
> > >
> > > However, there's this email from Daniel Berrangé from back in Aug 2012:
> > > http://osdir.com/ml/openstack-cloud-computing/2012-08/msg00293.html
> > >
> > > "Block migration is a part of the KVM that none of the upstream
> > developers
> > > really like, is not entirely reliable, and most distros typically do not
> > > want to support it due to its poor design (eg not supported in RHEL).
> > >
> > > It is quite likely that it will be removed in favour of an alternative
> > > implementation. What that alternative impl will be, and when I will
> > > arrive, I can't say right now."
> > >
> > > Based on this info, the OpenStack Ops guide currently recommends against
> > > using block-based live migration, but the Compute Admin guide has no
> > > warnings about this.
> > >
> > > I wanted to sanity-check against the mailing list to verify that this was
> > > still the case. What's the state of block-based live migration with KVM?
> > > Should we say be dissuading people from using it, or is it reasonable for
> > > people to use it?
> >
> > What I wrote above about the existing impl is still accurate. The new
> > block migration code is now merged into libvirt and makes use of an
> > NBD server built-in to the QMEU process todo block migration. API
> > wise it should actually work in the same way as the existing deprecated
> > block migration code.  So if you have new enough libvirt and new enough
> > KVM, it probably ought to 'just work' with openstack without needing
> > any code changes in nova. I have not actually tested this myself
> > though.
> >
> > So we can probably update the docs - but we'd want to checkout just
> > what precise versions of libvirt + qemu are needed, and have someone
> > check that it does in fact work.
> >
> >
> Thanks, Daniel. I can update the docs accordingly. How can I find out what
> are the minimum versions of libvirt and qemu are needed?
> 
> Also, I noticed you said "qemu" and not "kvm", and I see that
> http://wiki.qemu.org/KVM says that qemu-kvm fork for x86 is "deprecated,
> use upstream QEMU now".  Is it the case now that when using KVM as the
> hypervisor for a host, an admin will just install a "qemu" package instead
> of a "qemu-kvm" package to get the userspace stuff?

It depends on the distro to be honest. eg on Fedora you'd use qemu-kvm
which is a virtual package which will pull in qemu-system-$ARCH for your
particular host.

Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|


References