| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2025-01-30 | |||
| 10:10:25 | opendevreview | Michael Still proposed openstack/nova master: libvirt: Add extra specs for USB redirection. https://review.opendev.org/c/openstack/nova/+/927354 | |
| 12:07:57 | opendevreview | Stephen Finucane proposed openstack/nova-specs master: Add flavor-search-by-name spec https://review.opendev.org/c/openstack/nova-specs/+/940440 | |
| 17:29:11 | ganso | Hi folks. Given nvidia has dropped support for mdev in their v535+ drivers, what one would have to do to transition from an mdev configuration to a vfio configuration? I've been using mdev and working fine but now with the driver update nothing works anymore. I have installed libvirt 10 but that alone doesn't fix the problem. Does Nova SRIOV support running VGPUs in this mode? What config is necessary to change? I don't see | |
| 17:29:11 | ganso | mention of "vfio" in the nova docs for attaching GPU to guests | |
| 18:10:49 | melwitt | ganso: I don't think yet but there is vfio work in progress toward supporting it https://specs.openstack.org/openstack/nova-specs/specs/2025.1/index.html | |
| 18:15:04 | ganso | melwitt: I saw that spec but I thought it was specific to "variant drivers" (I am not familiar with that). If not, then basically the new nvidia drivers rendered the GPUs completely incompatible with OpenStack Nova until that spec is implemented? | |
| 18:19:26 | melwitt | ganso: I'm not familiar with it either (and not an expert in vgpu) but IIUC that work is needed as a prereq toward supporting sriov vgpu. bauzas would know for sure | |
| 18:57:37 | ganso | melwitt: thanks! let's wait for bauzas to respond | |
| 19:14:17 | mikal | sean-k-mooney: you around today? | |
| 19:29:07 | sean-k-mooney | mikal: sorry sort of. ill be aroudn for a bit longer but them im off until tuesday (pto tomorrow because of public holiday on monday) | |
| 19:30:39 | mikal | sean-k-mooney: ahhh fair enough. I sent an os-traits patch, which I _think_ is correct. I'm arguing with the nova sound patch to integrate it, but I've never done one of these before so I think I'll have missed stuff. I'll move on to the USB one after that. | |
| 19:30:51 | mikal | sean-k-mooney: worst case you'll have some patches to laugh at on Tuesday. | |
| 19:31:47 | sean-k-mooney | i can take a look now | |
| 19:31:58 | sean-k-mooney | im not sure if Uggla has cut the new release yet or not | |
| 19:32:02 | sean-k-mooney | we wanted to do one this week | |
| 19:32:22 | sean-k-mooney | we can always do another one next week if needed | |
| 19:32:54 | sean-k-mooney | so https://review.opendev.org/c/openstack/os-traits/+/940418 | |
| 19:33:01 | mikal | Yep, that's it. | |
| 19:33:36 | mikal | I assume I need that released before CI for the nova patch will pass? (Its failing for other reasons now, but presumably that will eventually block me). | |
| 19:33:59 | sean-k-mooney | it will pass in tempest but not unit tests with depends on | |
| 19:34:09 | sean-k-mooney | you could put the traits stuff at the end | |
| 19:34:43 | mikal | Oh, I've been doing the sound one in the sound patch, not in a separate one. | |
| 19:34:49 | mikal | I could separate them if I need to I suppose. | |
| 19:34:53 | sean-k-mooney | mikal: so the patch looks correct to me ill eave a copule of comments to that effect | |
| 19:35:00 | mikal | Thanks. | |
| 19:35:06 | sean-k-mooney | it can be in one pathc it normlaly is | |
| 19:35:11 | mikal | Ok | |
| 19:35:31 | sean-k-mooney | just the unit test will complain ectra if you try and import the constnat | |
| 19:35:50 | mikal | Even with a depends-on? | |
| 19:36:12 | sean-k-mooney | thre are ways to make ti work in unit test but its complex, ya we would have to properly hook up tox siblings and some other stuff | |
| 19:36:31 | sean-k-mooney | bacislly the unit test alwasy pull the libs form pypi | |
| 19:36:37 | mikal | Yeah, that doesn't sound like a good use of time. | |
| 19:36:38 | sean-k-mooney | we woudl need to make the job install them form souce | |
| 19:36:54 | sean-k-mooney | devstack has explcit supprot for this | |
| 19:37:20 | mikal | I am a big concerned about the feature freeze deadline, but not concerned enough to do something like that quite yet. | |
| 19:39:31 | sean-k-mooney | well even if we made it pass we still cant merge it for two reaons one we need to have a release or if anyone tries to use it for real it would break | |
| 19:39:46 | sean-k-mooney | second the depend on will prevent it form merging until the other patch is merged | |
| 19:40:00 | sean-k-mooney | so mechanically it would not speed anything up really | |
| 19:40:05 | sean-k-mooney | maybe slighlty | |
| 19:40:20 | sean-k-mooney | but only by the tiem ti takes to cut a release | |
| 19:40:43 | mikal | Yeah cool. I'll keep working on traits for sound and usb then and hope a release happens in the mean time. | |
| 19:41:15 | mikal | If you've got time for a quick scan of my traits attempt in https://review.opendev.org/c/openstack/nova/+/926126/12 that would be cool, but if you don't that's also fine. | |
| 19:42:40 | sean-k-mooney | i have +2'd the triats patch melwitt if your about care to take a look https://review.opendev.org/c/openstack/os-traits/+/940418 | |
| 19:43:42 | sean-k-mooney | mikal: ah https://review.opendev.org/c/openstack/releases/+/940363 merged 6 hours ago | |
| 19:44:10 | mikal | How often do they do those releases? On demand? | |
| 19:44:21 | sean-k-mooney | bauzas: ^ we shoudl plan an os-traits 3.4.0 once we merge mikal's patch | |
| 19:44:34 | sean-k-mooney | mikal: yes on demand | |
| 19:44:40 | melwitt | sean-k-mooney: not sure I can do a useful review. what are the sources for the list enums? i.e. where do they come from | |
| 19:45:01 | sean-k-mooney | libvirt models. i was debating if we shoudl add the links | |
| 19:45:18 | sean-k-mooney | i decied nto to bring that up but since you asked | |
| 19:45:20 | mikal | Links to the domain XML docs? | |
| 19:45:33 | sean-k-mooney | https://libvirt.org/formatdomain.html#sound-devices | |
| 19:45:53 | sean-k-mooney | mikal: ya we sometime link to the libvirt docs or other refences | |
| 19:46:06 | mikal | I can do that if you want me to. | |
| 19:46:19 | sean-k-mooney | so for the nic models https://github.com/openstack/os-traits/blob/master/os_traits/compute/net.py#L25 | |
| 19:46:27 | sean-k-mooney | we lined to the def in nova | |
| 19:47:31 | sean-k-mooney | in other case we link to specs https://github.com/openstack/os-traits/blob/master/os_traits/compute/security.py#L32 | |
| 19:48:30 | sean-k-mooney | we dont require it but it can be useful, these are just comment for future us | |
| 19:48:44 | mikal | Please hold. | |
| 19:49:09 | sean-k-mooney | so like the link to wiki pages for all the cpu feature flags in https://github.com/openstack/os-traits/blob/master/os_traits/hw/cpu/x86/__init__.py | |
| 19:49:59 | opendevreview | Michael Still proposed openstack/os-traits master: Add traits for sound and USB devices on instances. https://review.opendev.org/c/openstack/os-traits/+/940418 | |
| 19:51:27 | melwitt | sean-k-mooney: yeah I mean, otherwise I'm just like reviewing "hm yes, these look like all caps names in a list" 🙂 | |
| 19:51:40 | melwitt | ✅ | |
| 19:51:50 | mikal | Well in all fairness, it is a list of all caps names. | |
| 19:52:40 | sean-k-mooney | melwitt: no worries its a good callout | |
| 19:53:16 | sean-k-mooney | future it will help future us anyway | |
| 20:04:15 | mikal | So yeah, I think I've done that thing. | |
| 20:23:24 | melwitt | sean-k-mooney: commented on the review with a couple of questions with the main one being about if should have MODEL_NONE in this case | |
| 20:28:52 | sean-k-mooney | am, do we need ot report the ablity to not do somehitng | |
| 20:29:41 | sean-k-mooney | COMPUTE_USB_MODEL_NONE woudl mean the ablity to do emulation without any usb contoler | |
| 20:30:18 | sean-k-mooney | it is part of the grapics triats because when usign vgpus we wanted a way to say dont add an emulated grapics card | |
| 20:30:46 | melwitt | I dunno but we have it as a trait for hw_video_model | |
| 20:30:51 | sean-k-mooney | i guess we coudl add it for completeness | |
| 20:31:19 | melwitt | I'm ok with not having it (obviously I don't know this stuff) but wanted to bring it up in case it needs to be a thing | |
| 20:31:29 | sean-k-mooney | or well for video models we have a usecase to reqeust None because you get one by default if the host is set up for vnc or spice | |
| 20:32:07 | sean-k-mooney | so we added hw_graphic_model=none to opt out of that behavior | |
| 20:32:12 | melwitt | ah ok. and usb is not something that is enabled by default | |
| 20:32:19 | sean-k-mooney | it actully is | |
| 20:32:24 | sean-k-mooney | so im thinking your right | |
| 20:32:39 | sean-k-mooney | melwitt: you can configure the default pointer model to ps2 or usb | |
| 20:32:48 | melwitt | not sure if there is a use case for wanting to explicitly disable it | |
| 20:33:03 | sean-k-mooney | the use case to still use ps2 is lower overhead for realtime guest | |
| 20:33:20 | sean-k-mooney | which is whey they prefer serial console or no console over spice/vnc | |
| 20:33:30 | sean-k-mooney | less overhead to the vm due to interupts | |
| 20:33:31 | melwitt | oh... now that you say that it sounds vaguely familiar | |
| 20:34:20 | sean-k-mooney | https://docs.openstack.org/nova/latest/configuration/config.html#DEFAULT.pointer_model | |
| 20:34:49 | sean-k-mooney | the thing is i dont know if none is a valid libvirt value or if you just dont create the device xml | |
| 20:35:05 | melwitt | ah, good ol usbtablet | |
| 20:35:47 | sean-k-mooney | so for sound modle https://libvirt.org/formatdomain.html#sound-devices it looks like you jsut dont create it but i also know that you can get a sound card added becasule fo q35 i think | |
| 20:36:30 | melwitt | oh, I got the idea from the libvirt doc. it says "Additionally, since 0.10.0, if the USB bus needs to be explicitly disabled for the guest, model='none' may be used." https://libvirt.org/formatdomain.html#controllers | |
| 20:36:32 | sean-k-mooney | for usb contooler i think you just dont add one | |
| 20:36:40 | melwitt | so that's why I asked | |
| 20:36:44 | sean-k-mooney | oh i missed htat | |
| 20:37:20 | sean-k-mooney | so we could supprot it but that would need more work to chnage the pointer model if the config would use usb | |
| 20:37:55 | sean-k-mooney | melwitt: your raisign very reasonable points to which i dont have a simple answer | |
| 20:37:59 | melwitt | hm ok | |
| 20:38:53 | sean-k-mooney | my my concern is delaying the spice direct feature again for this. im wondering if we could add none later if we needed it? | |
| 20:39:08 | sean-k-mooney | basiclly not to expand the scope more | |
| 20:39:36 | melwitt | yeah.. I'm assuming we can punt that to later. this was just in case you already knew it's a thing we need we could add it | |
| 20:40:07 | sean-k-mooney | we didn tactully dicsus traits in the spec https://specs.openstack.org/openstack/nova-specs/specs/2025.1/approved/libvirt-spice-direct-consoles.html | |