Earlier  
Posted Nick Remark
#openstack-nova - 2021-02-17
18:13:35 bauzas /o\
18:13:52 bauzas stephenfin: we can't call neutron_client.list_segments() as you proposed
18:14:05 bauzas because the oldest neutron client we support doesn't have it
18:14:11 stephenfin just bump neutronclient?
18:14:31 sean-k-mooney well that is why having lower constratins is still valuable
18:14:32 bauzas stephenfin: with this change ? woah
18:14:41 bauzas sean-k-mooney: yeah, I kiss this job
18:14:47 sean-k-mooney even if we only list our direct dependcies
18:14:49 bauzas because it showed a potential issue
18:14:55 sean-k-mooney yep
18:15:08 stephenfin bauzas: Then rebase your work on top of https://review.opendev.org/c/openstack/nova/+/706295 :)
18:15:17 bauzas sean-k-mooney: stephenfin: you guys okay with bumping the neutron client for such purpose ?
18:15:22 stephenfin that'll do the bump for you
18:15:27 stephenfin yup, no issues from me
18:15:30 sean-k-mooney bauzas: the alterneive might be to use the sdk but if neutron client did not have it id ont think the sdk will
18:15:51 stephenfin nah, let's not complicate things further. neutronclient works fine
18:16:06 sean-k-mooney am how big a bump
18:16:09 sean-k-mooney i think yes
18:16:16 sean-k-mooney espcally if our current min is quite old
18:16:18 artom bauzas, sorry, 1:1. Oui, c'est un point.
18:16:25 stephenfin bauzas: If you've a choice, can you choose 7.1.0 to avoid conflicts with https://review.opendev.org/c/openstack/nova/+/706295 please
18:16:44 bauzas sean-k-mooney: stephenfin: yeah, will then bump neutronclient to 7.1.0 to avoid the conflict
18:17:00 bauzas hope that gibi will be happy too
18:17:01 stephenfin ta
18:17:12 sean-k-mooney oph i shoudl review https://review.opendev.org/c/openstack/nova/+/706295
18:17:21 sean-k-mooney i did not know you were doing that
18:17:39 sean-k-mooney why are you doing that by the way
18:17:44 sean-k-mooney is it needed for something
18:17:47 sean-k-mooney or just because
18:18:35 stephenfin it's less confusing IMO. I don't think we should reinvent the wheel
18:18:49 sean-k-mooney going form ksa back to neutron client is a regression though
18:18:50 stephenfin if there's a client that does things for us, let's delegate to them
18:18:57 sean-k-mooney we should be movign to the sdk
18:19:10 sean-k-mooney using ksa was a step on that move
18:19:20 stephenfin Maybe a sideways step more so than a regression, I think
18:19:28 stephenfin Nah, it was because neutronclient didn't support the thing
18:19:38 sean-k-mooney well it ws converted from neutorn client to ksa at somepoint i think
18:19:43 stephenfin Nope
18:19:52 stephenfin I added the feature to neutronclient. It didn't exist before :)
18:20:04 sean-k-mooney ok well using neutorn clinet more feels liek a regression since eventually we want to remove it
18:20:05 stephenfin Also, moving to sdk is a _lot_ of work. We saw that with the ironic series previously
18:20:29 sean-k-mooney we still use the ironic clinet in places
18:20:37 stephenfin My point exactly :)
18:20:40 sean-k-mooney that was never finsihed
18:20:49 stephenfin Wouldn't the ksa calls be replaced by sdk also?
18:20:57 sean-k-mooney yes
18:21:13 stephenfin Right, so neutronclient gets us simpler code for now, and when we switch we're replacing one thing rather than two
18:21:22 sean-k-mooney but move backward
18:21:41 sean-k-mooney it would be better to go striat form ksa to sdk
18:21:43 stephenfin I don't see why though. neutronclient is actively supported
18:22:12 stephenfin I mean, I realise that one client to rule them all is better in the long run
18:22:26 sean-k-mooney well this isnt an osc thng
18:22:33 stephenfin one library so
18:22:34 openstackgerrit Artom Lifshitz proposed openstack/nova master: WIP: Regression test https://review.opendev.org/c/openstack/nova/+/776250
18:22:37 sean-k-mooney its related
18:22:50 sean-k-mooney ya its a libvary that is not ment to be used by any project outside of openstack
18:22:55 stephenfin I don't get why neutronclient is worse than a homemade ksa invocation
18:23:03 sean-k-mooney but enven then we wanted ot encurage the project to stop using them
18:23:17 stephenfin eh, yes and no
18:23:30 stephenfin we want to drop use of the project specific CLIs
18:23:50 sean-k-mooney i just dont see why we would bother changin it when it snot broken if we are not going to change it to sdk since that what we want to use long term
18:24:07 sean-k-mooney stephenfin: we want to drop novaclint as a lib too eventually
18:24:16 sean-k-mooney not just as a cli
18:24:28 stephenfin because the SDK change could be 5 years off, and that ksa change and the related code it needs is there now
18:24:45 sean-k-mooney it does not have to be monoltic
18:24:50 sean-k-mooney we have sdk usage already
18:25:23 stephenfin not for neutron though
18:25:33 sean-k-mooney sure but lets add it incrementally
18:26:23 stephenfin eh, idk. While that changeover happens everything is worse and more confusing that it was previously
18:26:35 stephenfin including things like the NeutronFixture
18:26:55 sean-k-mooney well this more or less hould be nitten form mos of nova
18:26:56 stephenfin this makes my life easier, with far less effort than switching to sdk does
18:27:03 openstackgerrit Artom Lifshitz proposed openstack/nova master: libvirt: start tracking NUMACell.socket for hosts https://review.opendev.org/c/openstack/nova/+/766816
18:27:07 stephenfin huh?
18:27:11 openstackgerrit Artom Lifshitz proposed openstack/nova master: extra specs/image props: add `socket` PCI NUMA affinity https://review.opendev.org/c/openstack/nova/+/772748
18:27:11 sean-k-mooney the nova/network/neutorn module will hide that form the rest of nova
18:27:17 openstackgerrit Artom Lifshitz proposed openstack/nova master: Add `socket` PCI NUMA affinity policy request prefilter https://review.opendev.org/c/openstack/nova/+/772749
18:27:22 openstackgerrit Artom Lifshitz proposed openstack/nova master: Track host NUMA topology in PCI manager https://review.opendev.org/c/openstack/nova/+/774149
18:27:27 openstackgerrit Artom Lifshitz proposed openstack/nova master: WIP: pci: implement the `socket` NUMA affinity policy https://review.opendev.org/c/openstack/nova/+/772779
18:28:08 stephenfin sean-k-mooney: If you do look at the patch, ignore the fact that it's using neutronclient for a bit and tell me it doesn't make e.g. the NeutronFixture easier to grok
18:28:28 stephenfin that's where I'm coming from
18:28:58 stephenfin I'll do the sdk work eventually, once the OSC gaps are closed, but I think this is a win for now
18:29:32 sean-k-mooney well it not going to change the public api of that module so im not sure it will
18:29:42 sean-k-mooney im looking at it now
18:30:32 sean-k-mooney but the frecatorign you are ding im not sure require the code changes
18:31:58 sean-k-mooney none of those private api shoudl be needed in the generic neutron fixture https://review.opendev.org/c/openstack/nova/+/706295/10/nova/tests/fixtures.py#b1769
18:32:11 sean-k-mooney they should not be called outside the neutron moduule
18:32:22 sean-k-mooney only the public funtion will be
18:33:05 stephenfin They're called inside the neutron module
18:33:21 sean-k-mooney yep which can stub them in its own unit tests
18:33:35 stephenfin NeutronFixture is weird. It mocks APIs from both neutronclient and nova.network.neutron
18:33:48 sean-k-mooney ya
18:34:07 sean-k-mooney i then to use the libvirt version of it too at least when wriging func test
18:34:18 sean-k-mooney there is value in this yes
18:34:18 stephenfin no, it we don't stub them out then we need to provide an entire mock of e.g. 'bind_ports_to_host'
18:34:48 sean-k-mooney well we proably shoudl do htat since we need it for live migration right
18:35:01 stephenfin well we do
18:35:09 stephenfin by mocking out the private APIs it calls
18:35:39 stephenfin that API is the glue. We don't want to mock that out
18:35:50 stephenfin but we do want to mock out the neutron API calls

Earlier   Later