| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-02-06 | |||
| 17:35:02 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] New objects to transmit NUMA config from dest to source https://review.openstack.org/634827 | |
| 17:35:03 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [Very WIP] Update NUMA XML during live migration https://review.openstack.org/635229 | |
| 17:36:01 | artom | cfriesen, ^^ I know you wanted to play with early code. So it's still very early, but once that top [Very WIP] patch is just WIP it'll be ready for testing - by which I mean, I'll have enough code written to start actually running it myself in my dev env | |
| 17:36:24 | artom | So it'll most likely still be broken, actually :) | |
| 17:36:55 | artom | My plan is to write some automated tests with a tempest plugin we have, and iterate running those/fixing bugs until the feature works. Then unit tests will get fixed/written. | |
| 17:38:36 | cfriesen | artom: cool. I'm just finishing up some other downstream stuff, but thanks for the heads-up. | |
| 17:47:31 | mriedem | speaking of gibi's bw provider series, i'm +2 on the bottom 2 patches https://review.openstack.org/#/c/567268/ | |
| 17:47:36 | mriedem | to help keep that moving | |
| 18:10:07 | artom | cfriesen, yeah, no huge rush, just an FYI | |
| 18:16:05 | openstackgerrit | Matt Riedemann proposed openstack/nova master: WIP: Fix legacy-grenade-dsvm-neutron-multinode-live-migration https://review.openstack.org/634962 | |
| 18:31:39 | openstackgerrit | Stephen Finucane proposed openstack/nova master: docs: Update the PCI passthrough guides https://review.openstack.org/635243 | |
| 18:31:57 | stephenfin | mriedem: You might have some interest in that little doc patch ^ | |
| 18:32:24 | stephenfin | Inspired by stephenfin's struggles getting a PCI environment set up locally again | |
| 18:40:15 | cdent | mriedem, melwitt, jaypipes: Assuming we all still have jobs and openstack is still a thing come the opening of train, I'd like to work on making shared disk a thing. Do we have anything other than effort that is blocking that? | |
| 19:08:19 | mordred | cdent: I think that also assumes that the rule of law still holds somewhat and the seas haven't risen and swallowed us right? | |
| 19:08:38 | mriedem | cdent: unknown unknowns | |
| 19:08:52 | mriedem | i have a devstack-plugin-ceph patch that sets up a shared DISK_GB provider | |
| 19:08:59 | mriedem | but that job needs to be made multi-node to be interesting | |
| 19:09:13 | mriedem | https://review.openstack.org/#/c/586363/ | |
| 19:09:22 | mriedem | plus that existing bug which enumerates some known issues and the ptg etherpad | |
| 19:09:41 | mriedem | the todo from the ptg was write up a spec with the known issues and how we deal with those so we have a clear picture | |
| 19:09:55 | mriedem | then the rest i guess we find out from integration testing | |
| 19:11:22 | mriedem | cdent: assuming the vmware case is something like multiple vcenters sharing the same datastore? | |
| 19:11:41 | mriedem | or maybe that's already supported albeit poorly in nova | |
| 19:12:01 | cdent | mriedem: that's one. the other is N>1 datastores in the same vcenter | |
| 19:12:06 | cdent | shared can work there too | |
| 19:12:18 | cdent | https://github.com/cdent/placement-disk-models | |
| 19:12:36 | cdent | or a mix of both | |
| 19:13:07 | cdent | the main problem facing vcenters right now is the N>1 situation where it appears like there is a larger free slice than there actually is | |
| 19:13:22 | cdent | right now max_unit is manipulated to cover that, but it is tetchy | |
| 19:15:18 | cdent | mordred: you make a good point. I should have forewarned that I'm unable to enumerate all risks, and past performance is not an indicator | |
| 19:15:57 | mordred | cdent: fair | |
| 19:26:14 | openstackgerrit | sean mooney proposed openstack/nova master: update flavor admin docs https://review.openstack.org/635198 | |
| 19:28:21 | mriedem | need another stable core for https://review.openstack.org/#/c/633807/ | |
| 19:39:22 | openstackgerrit | Matt Riedemann proposed openstack/nova master: update flavor admin docs https://review.openstack.org/635198 | |
| 19:43:03 | openstackgerrit | Andrey Volkov proposed openstack/nova master: Check hosts have no instances for AZ rename https://review.openstack.org/509206 | |
| 19:45:42 | sean-k-mooney | mriedem: thanks for adressing the remaining nits. i dont mind respinning for things like that but docs are not really my strong suit. | |
| 19:46:14 | openstackgerrit | sean mooney proposed openstack/nova master: cleanup *.pyc files in docs tox envs https://review.openstack.org/635210 | |
| 19:48:01 | openstackgerrit | Matt Riedemann proposed openstack/nova master: api-ref: fix link to flavor extra specs docs https://review.openstack.org/635252 | |
| 19:49:44 | openstackgerrit | Matt Riedemann proposed openstack/nova master: api-ref: fix link to flavor extra specs docs https://review.openstack.org/635252 | |
| 19:59:41 | openstackgerrit | sean mooney proposed openstack/nova stable/ocata: PCI: do not force remove allocated devices https://review.openstack.org/635075 | |
| 20:14:44 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] New objects to transmit NUMA config from dest to source https://review.openstack.org/634827 | |
| 20:14:45 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] Update NUMA XML during live migration https://review.openstack.org/635229 | |
| 20:14:45 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] Send dest NUMA info inside LiveMigrateData https://review.openstack.org/634828 | |
| 20:18:50 | artom | OK, that's "done". Now to automate some tests. | |
| 20:19:13 | sean-k-mooney | artom: done as in works? | |
| 20:19:40 | artom | sean-k-mooney, no idea :) But I'm at the point there the XML on the dest should be updated. So I can start writing tests that assert stuff about it | |
| 20:20:11 | sean-k-mooney | im planning to pull it down in tommorow anyway but i can try it out later if you think it will work | |
| 20:20:14 | artom | It's an MVP - claims aren't cleaned up where they should be, no rollback handling, etc etc | |
| 20:20:33 | artom | sean-k-mooney, unlikely, I haven't run it myself yet | |
| 20:20:50 | artom | (Unless you count the gate) | |
| 20:21:36 | sean-k-mooney | well it wont hurt to try. ill spin it up on my ci cloud in the backgound and take a look | |
| 20:21:48 | artom | I fully expect it to blow up immediately :D | |
| 20:22:15 | sean-k-mooney | well thats what devstack vms are for | |
| 20:29:12 | openstackgerrit | Matt Riedemann proposed openstack/nova master: doc: mention description field in user flavors docs https://review.openstack.org/635263 | |
| 20:41:53 | cfriesen | stephenfin: you around? | |
| 20:43:36 | sean-k-mooney | stephenfin: he normally finishes about 2 hours before now | |
| 20:43:39 | cfriesen | sean-k-mooney: mriedem: you guys might know this too...we expect both KVM-based and tcg-based qemu to report a hypervisor type of "QEMU", right? | |
| 20:43:50 | sean-k-mooney | yes | |
| 20:45:00 | sean-k-mooney | kvm is just a software/kernel accleration for qemu so it does not change the hypervisor type if used | |
| 20:45:24 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Fix legacy-grenade-dsvm-neutron-multinode-live-migration https://review.openstack.org/634962 | |
| 20:45:26 | cfriesen | I think there's a bug in stephenfin's "fail live migration if instance has numa topologies" commit then. https://github.com/openstack/nova/blob/master/nova/conductor/tasks/live_migrate.py#L166 | |
| 20:45:33 | sean-k-mooney | similarly if you were to use the tcg multi thread backend "which we dont support yet" it would still be qemu | |
| 20:46:57 | sean-k-mooney | hum maybe lets see if that is the same hypervior type that is in the hyperviors api | |
| 20:47:57 | sean-k-mooney | so that is refering to https://github.com/openstack/nova/blob/master/nova/objects/compute_node.py#L69 | |
| 20:48:29 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Set bash xtrace and errexit globally in nova/tests/live_migration/hooks/* https://review.openstack.org/635291 | |
| 20:51:01 | sean-k-mooney | so the fact its not a HVTypeField is not promissing but it proably should be https://github.com/openstack/nova/blob/master/nova/objects/fields.py#L379 | |
| 20:51:58 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Replace ansible --sudo with --become in live_migration/hooks scripts https://review.openstack.org/635308 | |
| 20:53:40 | cfriesen | sean-k-mooney: agreed, that's what I was thinking. QEMU rather than KVM | |
| 20:53:40 | sean-k-mooney | cfriesen: so ya when virt tipe is set to kvm it is still qemu i just check on a local devstack. | |
| 20:54:52 | cfriesen | the tricky bit is that I think qemu with TCG doesn't support numa stuff | |
| 20:55:07 | sean-k-mooney | it does | |
| 20:55:18 | sean-k-mooney | it just does not support per cpu pinning | |
| 20:55:33 | cfriesen | ah, okay. | |
| 20:55:33 | sean-k-mooney | you can create numa nodes and use hugepages | |
| 20:55:38 | cfriesen | should be all right then | |
| 20:55:53 | sean-k-mooney | but nova's implematnion does not work due to the per cpu pinning it adds on top | |
| 20:56:34 | sean-k-mooney | the mttcg backend (multi thread TCG) remvoes the cpu pinning limitation too | |
| 20:56:49 | sean-k-mooney | libvirt will have support for that soon | |
| 21:00:23 | cfriesen | I'll propose a fix, see what stephen has to say tomorrow. | |
| 21:00:56 | sean-k-mooney | cfriesen: ya looking at the compute node object in the db it does not have the virt type | |
| 21:01:04 | sean-k-mooney | so i guess just update it to qemu | |
| 21:01:55 | sean-k-mooney | at the moment numa stuff wont work with qemu via nova so its fine that this applies to both | |
| 21:02:34 | sean-k-mooney | if the instance has a numa toplogy and we are migrating it it means it was spawned on a kvm host | |
| 21:07:17 | sean-k-mooney | artom: fyi this is the most imporant comment out of the one i just left on the review https://review.openstack.org/#/c/634606/9/nova/compute/manager.py@6011 | |
| 21:07:37 | artom | sean-k-mooney, that explains so much! | |
| 21:08:06 | artom | I hadn't rebased locally, line numbers in CI output didn't match, and _get_resource_tracker() was clearly present | |
| 21:08:07 | sean-k-mooney | it broke the sriov migrtaion series too | |
| 21:08:08 | artom | Thanks! | |
| 21:08:50 | sean-k-mooney | i fixed it in that yesterday so when i saw the fuction and the filing ci it clicked :) | |
| 21:08:54 | openstackgerrit | Merged openstack/nova stable/queens: Fix port dns_name reset https://review.openstack.org/633807 | |
| 21:09:04 | openstackgerrit | Merged openstack/nova master: update flavor admin docs https://review.openstack.org/635198 | |
| 21:09:22 | artom | Well, CI was going to fail regardless, but I was hoping it'd fail less | |
| 21:15:04 | openstackgerrit | Matt Riedemann proposed openstack/nova master: api-ref: warn about changing/unsetting AZ name with instances https://review.openstack.org/635315 | |
| 21:18:39 | openstackgerrit | Matt Riedemann proposed openstack/nova master: api-ref: mention policy defaults for aggregates https://review.openstack.org/635317 | |
| 21:48:40 | mriedem | cfriesen: jackding: https://review.openstack.org/#/c/603844/ | |
| 21:48:53 | mriedem | i like this approach better, but the patch needs work | |
| 21:49:02 | mriedem | and i think we probably don't want to mess with instances whose task_state is not None | |
| 21:50:11 | mriedem | also funny that i was looking at a very similar thing in something i was writing yesterday for cross-cell resize https://review.openstack.org/#/c/635080/1/nova/compute/manager.py@5029 | |
| 21:55:03 | jackding | mriedem: thanks. Yeh rollback can be messy. | |