Open Stack

On 08/05/2014 12:18 PM, Jay Pipes wrote:
> Hello stackers, TC, Neutron contributors,
> 
> At the Nova mid-cycle meetup last week in Oregon, during the discussion
> about the future of nova-network, the topic of nova-network -> Neutron
> migration came up.
> 
> For some reason, I had been clueless about the details of one of the
> items in the gap analysis the TC had requested [1]. Namely, the 5th
> item, about nova-network -> Neutron migration, which is detailed in the
> following specification:
> 
> https://review.openstack.org/#/c/101921/12/specs/juno/neutron-migration.rst
> 
> The above specification outlines a plan to allow migration of *running*
> instances from an OpenStack deployment using nova-network (both with and
> without multi-host mode) to an OpenStack deployment using Neutron, with
> little to no downtime using live migration techniques and an array of
> post-vm-migrate strategies to wire up the new VIFs to the Neutron ports.
> 
> I personally believe that this requirement to support a live migration
> with no downtime of running instances between a nova-network and a
> Neutron deployment *is neither realistic, nor worth the extensive time
> and technical debt needed to make this happen*.
> 
> I suggest that it would be better to instead provide good instructions
> for doing cold migration (snapshot VMs in old nova-network deployment,
> store in Swift or something, then launch VM from a snapshot in new
> Neutron deployment) -- which should cover the majority of deployments --
> and then write some instructions for what to look out for when doing a
> custom migration for environments that simply cannot afford any downtime
> and *really* want to migrate to Neutron. For these deployments, it's
> almost guaranteed that they will need to mangle their existing databases
> and do manual data migration anyway -- like RAX did when moving from
> nova-network to Neutron. The variables are too many to list here, and
> the number of deployments actually *needing* this work seems to me to be
> very limited. Someone suggested Metacloud *might* be the only deployment
> that might meet the needs for a live nova-network -> Neutron migration.
> Metacloud folks, please do respond here!
> 
> In short, I don't think the live migration requirement for nova-network
> to Neutron is either realistic or valuable, and suggest relaxing it to
> be good instructions for cold migration of instances from an older
> deployment to a newer deployment. There are other more valuable things
> that Neutron contributors could focus on, IMO -- such as the DVR
> functionality that brings parity to Neutron with nova-network's
> multi-host mode.
> 
> Thoughts?
> 
> -jay
> 
> [1]
> https://wiki.openstack.org/wiki/Governance/TechnicalCommittee/Neutron_Gap_Coverage

Yes, I agree with what you're suggesting here.  This was the approach I
was advocating for a cycle or two ago.  In a design summit session,
there were folks that seemed to really want to go off and investigate
live migration options.  Given what has (or hasn't) been done so far, I
maintain the same opinion as you've presented here, which is that it's
really not a worthwhile investment overall.  We should just provide some
good documentation on how a cold migration would work.

-- 
Russell Bryant