| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-05-29 | |||
| 15:48:41 | aspiers | kashyap: it's just SVG | |
| 15:48:58 | aspiers | stephenfin: search and replace? ;-p | |
| 15:49:09 | kashyap | aspiers: Ah, okay. | |
| 15:49:11 | stephenfin | aspiers: Deleting cells v1 | |
| 15:49:21 | cdent | that'll show you! | |
| 15:49:22 | aspiers | ROFL | |
| 15:49:47 | stephenfin | :D | |
| 15:50:07 | aspiers | WTF, I'm #4 by LoC | |
| 15:50:15 | aspiers | How is that possible | |
| 15:50:30 | stephenfin | specs? | |
| 15:50:36 | cdent | ugh, this is not great: https://www.stackalytics.com/?metric=loc&module=placement-group | |
| 15:51:48 | edleafe | cdent: they have been saying that their numbers have been out of whack for a while | |
| 15:52:04 | cdent | edleafe: they reported recently that they fixed it. that looks right | |
| 15:52:23 | edleafe | I thought they were still working on it | |
| 15:52:39 | cdent | there was a post couple days ago | |
| 15:52:50 | cdent | but who knows, I don't think we can ever really fully rely on it | |
| 15:52:51 | openstackgerrit | Stephen Finucane proposed openstack/nova stable/stein: docs: Don't version links to reno docs https://review.opendev.org/662025 | |
| 15:52:53 | melwitt | mriedem, johnthetubaguy: thanks for the reviews on counting quotas <3 I'm going to add a follow up patch for the nits/notes and then also propose the spec amendment for the deviations from the spec | |
| 15:53:27 | edleafe | You mean Eric has only contribute 25 lines of code?? | |
| 15:53:33 | aspiers | edleafe: they were missing a whole bunch of commits for nova, but I think that got fixed in the last day or 2 | |
| 15:54:39 | cdent | edleafe: of merged stuff in train, yeah, I'd say that's right | |
| 15:54:40 | aspiers | edleafe: it says 809 LoC for Eric | |
| 15:55:05 | aspiers | I'm looking at https://www.stackalytics.com/?metric=loc&release=train&module=nova-group | |
| 15:55:07 | edleafe | aspiers: for Train? | |
| 15:55:16 | edleafe | aspiers: ah, we're looking at placement | |
| 15:55:20 | aspiers | ah sorry | |
| 15:56:17 | mriedem | melwitt: just posted comments on https://review.opendev.org/#/c/638073/ | |
| 15:56:31 | mriedem | melwitt: i'm going to post a DNM patch to test the pg flow | |
| 15:57:16 | melwitt | whoa, that is a lot of comments | |
| 15:59:58 | mriedem | only 12, that's nothing | |
| 16:01:36 | openstackgerrit | Stephen Finucane proposed openstack/nova stable/queens: Fail to live migration if instance has a NUMA topology https://review.opendev.org/629597 | |
| 16:01:37 | openstackgerrit | Stephen Finucane proposed openstack/nova stable/queens: fix up numa-topology live migration hypervisor check https://review.opendev.org/648939 | |
| 16:09:57 | openstackgerrit | Matt Riedemann proposed openstack/nova master: DNM: Test count_usage_from_placement=True with postgresql https://review.opendev.org/662028 | |
| 16:11:43 | phillw | Hi good people, a quick question if I may. Regarding using GPUs in compute nodes, I managed to find an article from CERN on the subject but nothing else (http://openstack-in-production.blogspot.com/2018/05/introducing-gpus-to-cern-cloud.html) Does anyone know of anything else current;y on-going / in the pipe line for this? | |
| 16:12:52 | mriedem | phillw: vGPUs? because that's documented here https://docs.openstack.org/nova/latest/admin/virtual-gpu.html | |
| 16:13:06 | phillw | mriedem: thanks :) | |
| 16:13:13 | mriedem | physical GPUs can be accessed using pci-passthrough since...juno? | |
| 16:13:26 | mriedem | vGPU support in the libvirt and xenapi drivers is since queens | |
| 16:13:47 | mriedem | https://docs.openstack.org/nova/latest/admin/pci-passthrough.html for the latter | |
| 16:13:56 | mriedem | *er pcpu pci passthrough | |
| 16:14:01 | mriedem | *pgpu | |
| 16:14:14 | mriedem | that doc mostly talks about sriov ports though | |
| 16:14:58 | sean-k-mooney | you sould also be about to use amds sriov based vgpu "MxGpu" or something like that via pci passthrough from juno or whenever we added that | |
| 16:15:44 | sean-k-mooney | i.e. if your gpu support virtualisation via sriov then that will jsut work like any other sriov vf with pci passhtough | |
| 16:15:57 | sean-k-mooney | the same we we supprot qat virtualistion via sriov | |
| 16:17:13 | phillw | I'll go and have a read up. A colleague of mine wants to use the gpu power for AI learning and I'm at the start of re-learning openstack after a break since 2014 :) | |
| 16:18:09 | mriedem | scanning the cern article it looks pretty comprehensive | |
| 16:18:14 | mriedem | and says what we just told you anyway :) | |
| 16:18:31 | mriedem | i'm not aware of who is using vgpus yet | |
| 16:18:32 | sean-k-mooney | the current state is we have dedicated support for the nvida data center gpus( which need a license) and untested support for sriov stuff which shoudl jsut work via pcie passthough | |
| 16:18:36 | mriedem | bauzas might know | |
| 16:18:38 | phillw | it missed out about deployment :) | |
| 16:18:49 | mriedem | phillw: yeah but it linked to those same docs | |
| 16:18:56 | mriedem | that's your homework | |
| 16:20:11 | sean-k-mooney | phillw: so you your colleage wants to use a consomer gpu they are out of luck unless they do a complte passthough which technicall violates nvidas eula | |
| 16:20:31 | phillw | mriedem: it's Tims home work, he wants to add gpu's :) As both our instances are private we'll be using packstack. | |
| 16:20:53 | sean-k-mooney | phillw: is packstack still a thing? | |
| 16:20:55 | phillw | sean-k-mooney: so not possible with sriov without breaching EULA? | |
| 16:21:05 | sean-k-mooney | on the nvida side | |
| 16:21:11 | phillw | sean-k-mooney: yeah, it's still hanging on in there :P | |
| 16:21:19 | sean-k-mooney | technically if you dont use the binary driver then your fine | |
| 16:21:31 | sean-k-mooney | but if you use the opensocrce driver you dont have gpu compute | |
| 16:21:38 | sean-k-mooney | so its kind of a catch 22 | |
| 16:21:49 | phillw | ah, understood. | |
| 16:22:20 | sean-k-mooney | nvida really want people bying the datacenter gpus insteadk of ther gaming ones for viutalisation | |
| 16:23:08 | mriedem | there is a workaround i believe | |
| 16:23:08 | mriedem | https://review.opendev.org/#/c/579897/ | |
| 16:23:13 | sean-k-mooney | yes there is | |
| 16:23:22 | phillw | yeah, i can understand that. I recall playstation upgrade being done which would prevent USAF using a cluster of them as cloud :) | |
| 16:23:25 | sean-k-mooney | you set the hypervior to hidden | |
| 16:23:29 | mriedem | ^ is from CERN btw | |
| 16:23:56 | sean-k-mooney | yes that is for hyperv i think | |
| 16:23:57 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Stop handling 'InstanceUnknownCell' exception https://review.opendev.org/651307 | |
| 16:23:58 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Remove unnecessary wrapper https://review.opendev.org/651308 | |
| 16:23:58 | openstackgerrit | Stephen Finucane proposed openstack/nova master: db: Remove cell APIs https://review.opendev.org/651309 | |
| 16:23:59 | openstackgerrit | Stephen Finucane proposed openstack/nova master: conf: Remove cells v1 options, group https://review.opendev.org/651310 | |
| 16:23:59 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Remove 'ComputeManager._reschedule' https://review.opendev.org/660774 | |
| 16:24:00 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Remove cells v1 parameter from 'ComputeTaskAPI.resize_instance' https://review.opendev.org/651311 | |
| 16:24:00 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Stop passing 'kwargs' to 'rebuild_instance' https://review.opendev.org/651312 | |
| 16:24:01 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Stop passing 'delete_type' to 'terminate_instance' https://review.opendev.org/651313 | |
| 16:24:01 | openstackgerrit | Stephen Finucane proposed openstack/nova master: filters: Stop handling cells v1 https://review.opendev.org/651314 | |
| 16:24:02 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Remove nova.compute.*API() shims https://review.opendev.org/660527 | |
| 16:24:02 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Ensure controllers all call super https://review.opendev.org/660950 | |
| 16:24:04 | sean-k-mooney | for libvirt it already works | |
| 16:24:41 | phillw | that's cool, I'm on nodding terms with libvirt as I use it for my VMs :) | |
| 16:25:13 | mriedem | sean-k-mooney: that patch is for making nvidia work with windows guests in libvirt | |
| 16:25:19 | mriedem | the existing support is for libvirt guests only | |
| 16:25:33 | sean-k-mooney | * linux guests | |
| 16:26:05 | mriedem | i never really understood the opposition to that patch since it's just doing the same thing for windows guests on kvm that we allow for linux guests | |
| 16:26:12 | mriedem | i.e. we've already opened pandora's box there | |
| 16:26:23 | sean-k-mooney | we nack it however as we did not want to add a feature to specically workaound a driver check that is there to prevent you breaching nvidas eula | |
| 16:26:38 | mriedem | *we already do* | |
| 16:26:54 | mriedem | https://blueprints.launchpad.net/nova/+spec/hide-hypervisor-id-flavor-extra-spec | |
| 16:26:54 | sean-k-mooney | the hypervior hidden flag was not added for that use however | |
| 16:27:00 | sean-k-mooney | it was already there | |
| 16:27:06 | mriedem | yes it was | |
| 16:27:14 | mriedem | https://blueprints.launchpad.net/nova/+spec/add-kvm-hidden-feature | |
| 16:27:40 | mriedem | https://review.opendev.org/#/c/459753/18//COMMIT_MSG@14 | |
| 16:28:03 | mriedem | so jay was +W on that change, but not this other one to allow the same thing for windows guests | |
| 16:28:06 | mriedem | which i don't understand | |