| opendevreview | Merged openstack/nova master: Use SDK for deprecated baremetal proxy API https://review.opendev.org/c/openstack/nova/+/867796 | 00:07 |
|---|---|---|
| opendevreview | Takashi Kajinami proposed openstack/nova master: Remove deprecated [api] use_forwarded_for https://review.opendev.org/c/openstack/nova/+/903339 | 01:01 |
| opendevreview | Nobuhiro MIKI proposed openstack/nova master: scheduler: Request COMPUTE_ADDRESS_SPACE_* traits https://review.opendev.org/c/openstack/nova/+/895761 | 02:31 |
| opendevreview | OpenStack Proposal Bot proposed openstack/nova master: Imported Translations from Zanata https://review.opendev.org/c/openstack/nova/+/903427 | 03:31 |
| opendevreview | Takashi Kajinami proposed openstack/nova master: Remove deprecated [api] use_forwarded_for https://review.opendev.org/c/openstack/nova/+/903339 | 05:41 |
| gibi | JayF: Done | 08:11 |
| bauzas | gibi: sean-k-mooney: it would be nice if you could review my bugfix series https://review.opendev.org/q/topic:%22bug/2041519%22 | 08:22 |
| bauzas | it would also be nice if cores could look at my mdev live-migration spec https://review.opendev.org/c/openstack/nova-specs/+/900636 | 10:10 |
| bauzas | wow | 10:17 |
| noonedeadpunk | hey there! I have quite "trivial" question about limit_tenants_to_placement_aggregate / placement_aggregate_required_for_tenants / AggregateMultiTenancyIsolation things just to confirm I understand that correctly. | 10:37 |
| noonedeadpunk | So basically "user story" is to make set of computes be available only for specific tenant, meaning that it should be able to spawn instances _only_ on that aggregate, while all other tenants should be able to spawn anywhere _except_ this aggregate | 10:38 |
| bauzas | noonedeadpunk: shit, I need to leave in a sec | 10:38 |
| noonedeadpunk | Am I right that I should then use limit_tenants_to_placement_aggregate: true, placement_aggregate_required_for_tenants: false and ignore AggregateMultiTenancyIsolation | 10:39 |
| bauzas | there are multiple ways to support tenancy limitations | 10:39 |
| noonedeadpunk | Or, use AggregateMultiTenancyIsolation but then add another property and set of flavors for this tenant? | 10:39 |
| bauzas | either by the filter or by our pre-filters | 10:39 |
| bauzas | noonedeadpunk: there is another change from sahid related to this https://review.opendev.org/c/openstack/nova/+/896512 | 10:41 |
| * bauzas disappears for a couple of mins | 10:41 | |
| noonedeadpunk | I need just 1 tenant to be isolated, so it shouldn;t be an issue... | 10:42 |
| noonedeadpunk | sure... | 10:42 |
| noonedeadpunk | will still write a question for the future :) | 10:42 |
| noonedeadpunk | So reading description for placement_aggregate_required_for_tenants it made me think, that using placement will limit tenant to the aggregate, so won't allow to escape these computes, but will allow others to use them | 10:43 |
| noonedeadpunk | Or maybe I need both limit_tenants_to_placement_aggregate and AggregateMultiTenancyIsolation? Then on pre-filter THE tenant will get only suggestion for the aggregate, and then filter will cut down all rest tenants to be able to spawn there? | 10:44 |
| opendevreview | Stephen Finucane proposed openstack/nova master: Bump hacking version https://review.opendev.org/c/openstack/nova/+/903529 | 10:55 |
| opendevreview | Stephen Finucane proposed openstack/nova master: Resolve mypy error https://review.opendev.org/c/openstack/nova/+/903530 | 10:55 |
| gibi | bauzas: I will try... | 10:56 |
| opendevreview | Stephen Finucane proposed openstack/nova master: tox: Use pre-commit for pep8 target https://review.opendev.org/c/openstack/nova/+/896227 | 10:57 |
| opendevreview | Stephen Finucane proposed openstack/nova master: pre-commit: Add mypy https://review.opendev.org/c/openstack/nova/+/896228 | 10:57 |
| bauzas | gibi: no worries, as you can | 11:12 |
| opendevreview | Merged openstack/nova master: tests: Use clearer syntax for empty generator https://review.opendev.org/c/openstack/nova/+/903076 | 11:18 |
| dvo-plv | sean-k-mooney, thanks, it works with venv, but I can not figure out how it works previously, couple days ago | 12:40 |
| opendevreview | Stephen Finucane proposed openstack/nova master: docs: Further clarifications to the SG doc https://review.opendev.org/c/openstack/nova/+/903675 | 12:57 |
| stephenfin | sean-k-mooney: per our discussion ^ | 12:57 |
| bauzas | gibi: want to ship your recent backport into Antelope? I was asked to +1 https://review.opendev.org/c/openstack/releases/+/899603 | 13:22 |
| bauzas | I think I'll release first the antelope reelase and then we will create another stable release for https://review.opendev.org/c/openstack/nova/+/903553 | 13:24 |
| gibi | bauzas: left feedback on the mdev live migration series | 13:31 |
| bauzas | gibi: <3 | 13:31 |
| bauzas | gibi: replied https://review.opendev.org/c/openstack/nova-specs/+/900636 | 13:42 |
| bauzas | you asked a very good question about the conductor option | 13:42 |
| bauzas | I already thought about all the possibilities but I don't know which one is the better | 13:42 |
| bauzas | and I also tried to explain you why and how we need a libvirtdriver dict | 13:43 |
| bauzas | another possibility instead of an internal dict would be to persist *something* *somewhere* about instances/mdevs | 13:44 |
| bauzas | either in DB or in some file | 13:44 |
| bauzas | I know sean-k-mooney has concerns about the fact we check the usage by looking at the guest XMLs | 13:44 |
| bauzas | (I mean, if we're about persisting anything, I would way prefer to do it in some file like /var/lib/instances rather than with our Nova DB, for multiple reasons, one being I'm afraid of all the possible wrong values we would write in the DB that would mean to be fixed by hand) | 13:46 |
| sean-k-mooney | bauzas: i have a personal goal of not requiring persitent xmls at all in the future | 14:08 |
| sean-k-mooney | i want nova to be able to recreate a vm evne ifn the /etc/libvirt/qemu directiory is deleted | 14:08 |
| sean-k-mooney | i would much prefer to use the db but /var/lib/instances is much better then what we have today | 14:09 |
| sean-k-mooney | bauzas: if we coudl converge the pci tracker and medev code paths into a common device tracker that would be awsome in my view but i dont think that is goign to happen near term | 14:10 |
| bauzas | sean-k-mooney: sorry in a meeting, let's discussing in 10-ish | 14:18 |
| bauzas | discuss* | 14:18 |
| *** haleyb|out is now known as haleyb | 14:24 | |
| bauzas | sean-k-mooney: back, readinh | 14:25 |
| bauzas | sean-k-mooney: sure, tbc, lemme explain why we're discussing it now : gibi is wondering why we need an internal dict to "persist" which mdevs are taken for live-migration (before the target guest runs with a domain) | 14:26 |
| bauzas | so I was just offering to persist some 'mdevs' file in the related /var/lib/instances/<inst_uuid> directory that would tell 'yo, those are the mdevs I use' | 14:27 |
| bauzas | but there be dragon | 14:28 |
| bauzas | (I'm pretty sure that a domain is created on target's /var/lib/instances as soon as we start the live-migration) | 14:28 |
| opendevreview | Merged openstack/nova master: Remove Ironic client wrapper https://review.opendev.org/c/openstack/nova/+/867797 | 14:28 |
| opendevreview | Merged openstack/nova master: Remove unnecessary aliasing https://review.opendev.org/c/openstack/nova/+/867798 | 14:28 |
| opendevreview | Merged openstack/nova master: Use real SDK objects in tests https://review.opendev.org/c/openstack/nova/+/867799 | 14:29 |
| bauzas | wow the gate sounds definitely better :) | 14:29 |
| bauzas | (tbh, what I said about writing a file in /var/lib/instances per instance dir could be a problem with shared storage, or not, depending on whether we are able to say "look, this mdev doesn't exist, I'm gonna recreate it" | 14:31 |
| sean-k-mooney | we can always nest it in a $hostname folder if needed | 14:38 |
| bauzas | sean-k-mooney: say I have a magical wand, what would be your preference ? using an internal dict like I wrote in the spec or persist some file in /var/lib/instances/<inst> ? | 14:53 |
| bauzas | the latter would require me to update the spec but I can live with that | 14:53 |
| bauzas | (this wouldn't impact the current behaviour which would be to lookup the domain definitions, this is just additive either way) | 14:54 |
| bauzas | if we decide later to use that file too for mdevs we spawn, then we would eventually stop to lookup the domain XMLs | 14:55 |
| bauzas | this is Christmas time, make a wish. | 14:55 |
| bauzas | gibi: sean-k-mooney: ^ | 14:55 |
| bauzas | (note : I don't know *yet* how to implement a persistent file but I can arrange that by remembering the service uuid file thing | 14:57 |
| gibi | > we use the internal dict for persisting which mdevs are already used for live-migration for making sure that other instances being created in the target compute wouldn't take them while we live-migrate. | 15:09 |
| gibi | OK so this is basically a lock between the instances | 15:09 |
| gibi | so that again resource tracker territory. But I got it that it is too big of a move | 15:09 |
| gibi | so I drop my question about persistence | 15:10 |
| gibi | an in memory dict is OK to me | 15:10 |
| bauzas | fair enough | 15:11 |
| JayF | Thanks gibi \o/ was quite a gift to wake up to the rest of that chain being not only approved but through the gate! | 15:11 |
| bauzas | gibi: so we only have one open question about the conductor conf option | 15:12 |
| bauzas | gibi: see my comment about it and all the alternatives | 15:12 |
| gibi | JayF: it is almost Christmas anyhow :) | 15:13 |
| gibi | bauzas: I'm on it | 15:13 |
| bauzas | thanks | 15:13 |
| JayF | Let me rebase my ironic<>nova driver bugfix and I'll add something else to my list ;) | 15:13 |
| gibi | :) | 15:13 |
| gibi | > We could avoid the pre-flight check that adds a conductor option and just | 15:25 |
| gibi | assume to test the libvirt versions and the types during the pre-livemigration | 15:25 |
| gibi | conductor checks that claim the resources. | 15:25 |
| gibi | bauzas: does it mean we can do all our checks in pre-livemigration without the extra conductor option? | 15:26 |
| gibi | if so then I vote for not having a conductor option and the the pre-livemigration do the checking | 15:26 |
| gibi | pre-livemigration is there for many other checks like cpu compatibility | 15:27 |
| gibi | bauzas: ^^ | 15:28 |
| *** jamesdenton_ is now known as jamesdenton | 15:37 | |
| opendevreview | ribaudr proposed openstack/nova master: Reproducers for bug 1869804 https://review.opendev.org/c/openstack/nova/+/877772 | 15:39 |
| opendevreview | ribaudr proposed openstack/nova master: Fix [backport]: Live migrating to a host with cpu_shared_set configured will now update the VM's configuration accordingly. https://review.opendev.org/c/openstack/nova/+/877773 | 15:39 |
| opendevreview | ribaudr proposed openstack/nova master: Fix: Live migrating to a host with cpu_shared_set configured will now update the VM's configuration accordingly. https://review.opendev.org/c/openstack/nova/+/903706 | 15:39 |
| bauzas | gibi: cool with me so | 15:43 |
| bauzas | I'll doublecheck with the code if that would work, since we'll go to the target first | 15:43 |
| gibi | bauzas: ack | 15:58 |
| opendevreview | Sahid Orentino Ferdjaoui proposed openstack/nova master: scheduler: AggregateMultitenancyIsolation to support unlimited tenant https://review.opendev.org/c/openstack/nova/+/896512 | 15:58 |
| opendevreview | Merged openstack/nova stable/2023.1: [pwmgmt]ignore missin governor when cpu_state used https://review.opendev.org/c/openstack/nova/+/903553 | 16:49 |
| opendevreview | Merged openstack/nova master: Remove outdated comment about allocation ratios https://review.opendev.org/c/openstack/nova/+/900242 | 16:50 |
| opendevreview | Pierre Riteau proposed openstack/nova stable/2023.2: Remove outdated comment about allocation ratios https://review.opendev.org/c/openstack/nova/+/903684 | 16:55 |
| bauzas | gibi: doh, we missed the target for releasing your bugfix shortly by 1 hour | 17:29 |
| gibi | bauzas: now worries | 17:36 |
| gibi | bauzas: that is not a blocker until we start testing with cpu_dedicated_set | 17:36 |
| opendevreview | Jay Faulkner proposed openstack/nova master: [ironic] Partition & use cache for list_instance* https://review.opendev.org/c/openstack/nova/+/900831 | 20:56 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!