[openstack-dev] [nova] [neutron] the nova network facade that isn't
mriedem at linux.vnet.ibm.com
Mon Apr 18 20:48:38 UTC 2016
On 4/18/2016 3:33 PM, Sean Dague wrote:
> When doing bug triage this morning a few bugs popped up:
> - https://bugs.launchpad.net/nova/+bug/1456899 - nova absolute-limits
> Security groups count incorrect when using Neutron
> - https://bugs.launchpad.net/nova/+bug/1376316 - nova absolute-limits
> floating ip count is incorrect in a neutron based deployment
> - https://bugs.launchpad.net/nova/+bug/1456897 - nova absolute-limits
> Floating ip
> The crux of this is the Nova limits API basically returns junk about
> resources it doesn't own. It's been this way forever.
> Last year there was a spec to add proxying to Neutron to the Nova API -
> https://review.openstack.org/#/c/206735/ - which died on the vine.
> I think we've moved to a point in time where we need to stop thinking
> about nova-net / neutron parity in our API. Neutron is the predominate
> stack out there. Where things don't work correctly with Neutron from our
> proxy API we should start saying "yes, that's not supported, please go
> talk to Neutron" one way or another.
> I feel like in this case it would be dropping the keys which we know are
> lies (terrible lies). If using OpenStack client, it can smooth over
> this. In general, people should assume they should be talking to neutron
> when getting this kind of data. I feel like in other cases where we
> don't return good neutron data today, we should accept that as status
> quo, and not fix it.
> I'd like to propose an alternative spec which is this kind of approach,
> to by policy not enhance any of the proxies and instead focus on ways in
> which we can aggressively deprecate them. But figured it was worth
> discussion first. Flame away!
I guess at a high level my thinking was always, if nova-network isn't
deprecated, and these APIs are broken when using Neutron, it's (mostly)
trivial to add a proxy to fill those gaps (like my spec for
os-virtual-interfaces). So then when people move from deprecated
nova-network to neutron, all of their tooling doesn't start breaking.
In thinking about it another way, if we just say nova-network is
deprecated again and therefore we have no incentive to make these APIs
work in the Neutron case, and want to force people off them, then I can
see that point.
It was different back in Havana when I was originally looking at this
because Neutron adoption was very different. With the recent survey,
however, it looks like nova-network is 7% of deployments now, and that's
including non-production. So I concede that it's making less sense to
put effort into making the APIs work with a proxy.
More information about the OpenStack-dev