Open Stack

On Wed, Jan 29, 2014 at 12:39:23PM -0500, Russell Bryant wrote:
> On 01/29/2014 12:27 PM, Daniel P. Berrange wrote:
> > On Wed, Jan 29, 2014 at 11:47:07AM -0500, Russell Bryant wrote:
> >> Greetings,
> >>
> >> A while back I mentioned that we would revisit the potential deprecation
> >> of nova-network in Icehouse after the icehouse-2 milestone.  The time
> >> has come.  :-)
> >>
> >> First, let me recap my high level view of the blockers to deprecating
> >> nova-network in favor of Neutron:
> >>
> >>   - Feature parity
> >>     - The biggest gap here has been nova-network's multi-host mode.
> >>       Neutron needs some sort of HA for l3 agents, as well as the
> >>       ability to run in a mode that enables a single tenant's traffic
> >>       to be actively handled by multiple nodes.
> >>
> >>   - Testing / Quality parity
> >>     - Neutron needs to reach testing and quality parity in CI.  This
> >>       includes running the full tempest suite, for example.  For all
> >>       tests run against nova with nova-network that are applicable, they
> >>       need to be run against Neutron, as well.  All of these jobs should
> >>       have comparable or better reliability than the ones with
> >>       nova-network.
> >>
> >>   - Production-ready open source components
> >>     - nova-network provides basic, but usable in production networking
> >>       based purely on open source components.  Neutron must have
> >>       production-ready options based purely on open source components,
> >>       as well, that provides comparable or better performance and
> >>       reliability.
> > 
> > What, no mention of providing an automated upgrade path ? Given how
> > we go to great lengths to enable continuous deployment with automated
> > upgrade paths, I'd really expect to see something to deal with migrating
> > people from nova-network to neutron with existing tenants unaffected.
> 
> That's a good point.  This is actually a very sticky situation.  We have
> a upgrade path already, which is why I didn't mention it.  It's not
> really great though, so it's worth further discussion.  The path is roughly:
> 
> 1) Deploy a parallel nova install that uses Neutron, but shares all
> other services with the existing Nova that uses nova-network.
> (Keystone, glance, cinder, etc)
> 
> 2) Spawn new instances in the new Nova.
> 
> 3) For any instances that you want to migrate over to Neutron, snapshot
> them to glance, and then re-spawn them in the new Nova.
> 
> This is the only plan that I've heard that we *know* should work for all
> deployment variations.  I've seen very little effort go into
> investigating or documenting any more advanced upgrade paths.
> 
> The other upgrade piece is some sort of data migration.  There are some
> bits of data, such as security group definitions, that we should be able
> to automatically export from nova and import into neutron.  I don't
> think anyone has worked on that, either.

I was thinking of an upgrade path more akin to what users got when we
removed the nova volume driver, in favour of cinder.

  https://wiki.openstack.org/wiki/MigrateToCinder

ie no guest visible downtime / interuption of service, nor running of
multiple Nova instances in parallel.

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 :|