Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2021-01-08
14:24:50 dansmith stephenfin: oof, yeah that probably needs several sets of eyes. I won't get to it today, but maybe next week
14:25:45 stephenfin Agreed. Any time in the near future is good
14:36:14 artom stephenfin, I suppose if every reviewer runs your tool manually it's as good as CI
14:36:39 artom We'd also need to make sure your tool has no bugs, but then that's the same as with a CI job
14:53:12 bauzas jawad_axd: is the PCI device provided in the instance ?
14:53:43 jawad_axd @bauzas Yes
14:54:00 bauzas so, it's either a windows or a nivdia issue, sorry
14:54:06 sean-k-mooney jawad_axd: i assume if you look in the device manager the device is there with the nvida drivae but error 43
14:54:38 sean-k-mooney jawad_axd: it looks like nvidia has updated there driver detection logic recently and broken our ablity to do pci passhtouhg again
14:55:10 sean-k-mooney bauzas: i did pci passhtough of my gtx-780 last week and found our hypervior hiding code nolonger is sufficent
14:55:19 sean-k-mooney error 43 is back
14:55:33 jawad_axd @sean-k-mooney There is no error 43. Device manager shows device working properly.
14:55:34 bauzas what ?
14:56:33 jawad_axd I had error 43 with genforce GTX 780. But with this Tesla gpu, driver installed successfully.
14:56:36 sean-k-mooney jawad_axd: ok then in that case have you disabel the video device so its the only gpu
14:56:55 sean-k-mooney jawad_axd: ya the tesla driver stack allows virutalistation
14:57:07 sean-k-mooney it does not have the same test
14:57:11 sean-k-mooney *check
14:57:31 sean-k-mooney jawad_axd: you could add hw_video_model=none to the image as a test
14:57:55 sean-k-mooney jawad_axd: that would disable the software gpu emulation in qemu so the only gpu would be the nvidia one
14:58:08 jawad_axd @sean-k-mooney Thanks. I try this hw_video_model=none .
14:58:42 sean-k-mooney before you do if you install gpuz in the vm does it detect support for cuda/driectx ectra
14:59:06 jawad_axd @sean-k-mooney gpuz detects gpu.
14:59:20 sean-k-mooney jawad_axd: and it has all the apis available?
14:59:49 sean-k-mooney if so i would guess the other application are just selecting the first gpu which will be the cirros/virtio/qxl one form qemu
15:00:27 sean-k-mooney jawad_axd: tensorflow used to do this for example which is why hw_video_model=none was added
15:00:57 jawad_axd @sean-k-mooney hw_video_model=none would stop using qxl/virtio, right ?
15:01:04 sean-k-mooney yep
15:01:14 sean-k-mooney so the only gpu the guest will see it the pci passthough one
15:07:12 jawad_axd @sean-k-mooney What do you mean by apis in gpuz?
15:08:01 sean-k-mooney gpuz has a bunch of checkboxs on the overview page if i remmeber
15:08:09 sean-k-mooney for cuda directx opengl
15:08:14 sean-k-mooney and some others
15:08:32 sean-k-mooney just wanted to confim the expected api were avaiable
15:08:47 sean-k-mooney and that it was usign the nvida driver not the generic vga windows direver
15:09:33 jawad_axd It has OpenCL, CUDA and OpenGL1.1 checked.
15:09:47 sean-k-mooney yep so the nvidia direr is really working then
15:10:18 sean-k-mooney if they were not check then it might be using the fallback driver instead of the nvidia one
15:10:38 sean-k-mooney so i would guess its just a gpu selection issue cause by both of them being there
15:11:32 jawad_axd Alright. I try hw_video_model=none . Will get back if it works :)
15:25:30 openstackgerrit Balazs Gibizer proposed openstack/nova master: Refactor ResourceRequest constructor https://review.opendev.org/c/openstack/nova/+/769720
15:25:31 openstackgerrit Balazs Gibizer proposed openstack/nova master: Enable mypy on scheduler/utils.py https://review.opendev.org/c/openstack/nova/+/769942
15:27:09 openstackgerrit Balazs Gibizer proposed openstack/nova master: Support interface attach with QoS ports https://review.opendev.org/c/openstack/nova/+/756530
15:41:38 dansmith sean-k-mooney: do you know of any instances where we use the raw keystoneauth stuff to just make a quick call to another service, but with all the request info we got (i.e user token etc)?
15:41:48 dansmith looking for something to copy
15:43:14 sean-k-mooney maybe in the ironic code
15:43:28 dansmith okay I think we used to talk to placement like that didn't we?
15:43:45 sean-k-mooney maybe but i think palcment now uses the sdk
15:43:51 dansmith although not with the user's token I guess
15:43:54 dansmith yeah I know it does now
15:45:25 sean-k-mooney https://github.com/openstack/nova/blob/master/nova/virt/ironic/client_wrapper.py#L66-L73
15:45:33 dansmith ironic lazy-imports ironic client, but it's also not using the user's token right?
15:45:51 sean-k-mooney ya that im not sure about it look like that is using the config options
15:47:33 sean-k-mooney dansmith: i guess for ironic and placment we just use admin tokens since they are normally admin only apis right
15:47:42 sean-k-mooney so we just use the configs
15:47:43 dansmith right
15:47:58 sean-k-mooney do you need ksa or could you just use the sdk
15:48:26 dansmith well, I'm trying to PoC something where glance calls to itself,
15:48:39 sean-k-mooney too answer you question though i dont know off the top of my head
15:48:46 dansmith so I could just import glanceclient, but I really just need to replay an http call so it seems a little heavy
15:49:05 dansmith ack
15:49:45 sean-k-mooney ah ok
15:53:57 openstackgerrit Elod Illes proposed openstack/nova stable/stein: [stable-only] Fix lower-constraints job https://review.opendev.org/c/openstack/nova/+/769809
15:55:34 stephenfin dansmith: https://review.opendev.org/c/openstack/nova/+/706295/7/nova/network/neutron.py
15:55:55 stephenfin If I understood you correctly, I think that's what we do for port bindings in neutron
15:57:11 dansmith stephenfin: ah, yeah that might be it, thanks
15:57:39 dansmith I figured even though neutron was the right pattern, that everything would be through the client and not helpful,
15:57:48 dansmith but that chunk is like "in case of no client support, break glass" :)
16:05:20 jawad_axd @sean-k-mooney Apparently , hw_video_model doest work with none' anymore? I am getting invalid metadata.. I guess valid options are just graphics adapter model with this property..
16:06:04 openstackgerrit Balazs Gibizer proposed openstack/nova master: Enable mypy on scheduler/report.py https://review.opendev.org/c/openstack/nova/+/765152
16:06:18 openstackgerrit Balazs Gibizer proposed openstack/nova master: Remove unused Instance method https://review.opendev.org/c/openstack/nova/+/757615
16:06:46 gibi stephenfin: I've resolved you comments in the qos interace attach series
16:06:58 stephenfin gibi: ack, I'll take a look straight after this meeting
16:07:05 gibi cool, thanks
16:36:53 gibi stephenfin: went through once again the db migration series, all my questions are answered or fixed so I'm positive all along the series
16:37:06 stephenfin great, thank you
16:37:13 gibi good job!
16:37:38 sean-k-mooney jawad_axd: what release are you on?
16:38:30 sean-k-mooney jawad_axd:https://github.com/openstack/nova/blob/46899968619e4ea0ff2ab380977619bb29578d43/nova/objects/fields.py#L562-L573
16:39:13 sean-k-mooney jawad_axd: i added none in train https://github.com/openstack/nova/commit/35a591d33d8b1a6c30bf40ddc48a07715fd87339
16:39:21 jawad_axd I am using stein
16:39:35 sean-k-mooney ah you one release too old
16:40:40 sean-k-mooney in your case i think your only really choice other then upgrading is to force the applciations to use the gpu via there config or runtime options
16:41:29 sean-k-mooney jawad_axd: the image properties are defiend in oslo versions objects so there is no way to backport support for none with out upgrade issues
16:42:29 sean-k-mooney form a nova poing of view the phsyical gpu passthough is working correctly but the applciaiton is just not selceting the correct gpu of the two avaiable
16:45:07 jawad_axd Alright..
16:46:17 sean-k-mooney sorry cant help more but once nova has succfully booted the vm and made the gpu avaiable to it its jobs is basically done
16:56:55 jawad_axd @sean-k-mooney Atleast I got to know, passthrough is working fine. Now I need to figure out how to use it as default. Thanks
16:57:41 sean-k-mooney depending on your application it will differ
16:58:06 sean-k-mooney games/rendering applcaition like blender usally have an option you can set in there menu or config files
16:58:47 sean-k-mooney with tensorflow for example you can do https://www.tensorflow.org/api_docs/python/tf/config/set_visible_devices
17:16:50 sean-k-mooney huh did people know you can click on the files and expand them in the top review windows
17:16:56 sean-k-mooney https://review.opendev.org/c/openstack/python-openstackclient/+/762396
17:17:28 sean-k-mooney so if you click on any of the boxes beside the file links not on the links it will render the files in a dropdown
17:18:28 sean-k-mooney that is what thos little arrows on earch row on the right are signaling
17:20:42 sean-k-mooney stephenfin: im 50:50 on that reviw by the way. it looks liek ti will work correctly but lee basically made the same point about doing it in parralel in as i did in my first review
17:21:07 sean-k-mooney stephenfin: i think we should do that optimisation but im kind of with gibi that it could be a follow up patch
17:22:55 sean-k-mooney i know you are also looking a provideing a generic way to do the waits in https://review.opendev.org/c/openstack/osc-lib/+/762503
17:23:26 sean-k-mooney so i can see why you would prefer not to do it direcly in https://review.opendev.org/c/openstack/python-openstackclient/+/762396 and wait to reuse that
17:43:42 sean-k-mooney melwitt: just responed to your questions in https://review.opendev.org/c/openstack/nova/+/769614 i proably wont respin that today but let me know if you still have questions

Earlier   Later