| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-cyborg - 2026-06-23 | |||
| 14:26:49 | chandankumar | nope, I hope to finish the cleanup rewrite with clear_mode and clear_method by today | |
| 14:27:21 | sean-k-mooney | ack | |
| 14:27:44 | sean-k-mooney | i guess if there are no questions we can move on | |
| 14:27:55 | sean-k-mooney | #topic generic mdev driver | |
| 14:27:59 | sean-k-mooney | #link https://review.opendev.org/c/openstack/cyborg-specs/+/982276 | |
| 14:28:13 | sean-k-mooney | jgilaber: anythign you want to rasie on your spec? | |
| 14:28:24 | sean-k-mooney | im hoping to review that tomorow if its ready for review? | |
| 14:28:43 | opendevreview | chandan kumar proposed openstack/cyborg-specs master: Add generic NVMe driver spec with secure cleanup https://review.opendev.org/c/openstack/cyborg-specs/+/985349 | |
| 14:28:54 | jgilaber | nothing to raise, I think I addressed all comments | |
| 14:29:07 | chandankumar | Ok I have updated it here, But a one more pass on my side before asking for review | |
| 14:29:21 | jgilaber | the only point left I think is how to handle the multiple mdev types with placement | |
| 14:29:34 | sean-k-mooney | ack anythin you want to call out on the nova side? | |
| 14:30:16 | jgilaber | on the nova side the mtty series is close to merge I think | |
| 14:30:20 | jgilaber | it's working well in CI | |
| 14:30:29 | jgilaber | and I pushed a patch for the OWNER_NOVA trait | |
| 14:30:32 | jgilaber | #link https://review.opendev.org/c/openstack/nova/+/994299 | |
| 14:30:54 | jgilaber | on the last one I'm still working on the tests, but hopefully will be ready for review by the end of the week | |
| 14:31:06 | sean-k-mooney | our initall plan was to reused that for the cyborg ci. we may want to revaluate if we will instead port mdev supprot into the pci sim | |
| 14:31:17 | sean-k-mooney | but i think we choudl continue with the inial plan for now | |
| 14:31:49 | sean-k-mooney | ack ill add the owner trait patch to my review list | |
| 14:32:42 | jgilaber | yes, we can evaluate once I'm done with the nova part if we want to use pci sim, but for now using the nova devstack plugin should work and be easy to use | |
| 14:32:52 | sean-k-mooney | we may need to think about the upgrade impact of https://review.opendev.org/c/openstack/nova/+/994299/2/nova/compute/pci_placement_translator.py | |
| 14:33:09 | sean-k-mooney | that likely need to be gated behind a min compute service version check | |
| 14:33:24 | sean-k-mooney | well no | |
| 14:33:28 | sean-k-mooney | that where your addign the trait | |
| 14:33:32 | sean-k-mooney | that proably fine | |
| 14:33:35 | jgilaber | the pre filter is behind a service check | |
| 14:33:51 | jgilaber | adding the trait should be fine I think since it would not prevent any allocation | |
| 14:33:56 | sean-k-mooney | ah yes https://review.opendev.org/c/openstack/nova/+/994299/2/nova/scheduler/request_filter.py | |
| 14:34:36 | sean-k-mooney | ok ill do a proper review later | |
| 14:34:57 | jgilaber | ack, thanks | |
| 14:34:58 | sean-k-mooney | for the prefilter i think we want to add @functools.cache to the min version check | |
| 14:35:31 | sean-k-mooney | an all cell scatter gather on ever schdule is expensive | |
| 14:35:40 | sean-k-mooney | but we can do it once per scdhuler restart | |
| 14:36:01 | sean-k-mooney | we woudl just need to document that in the upgrade release note | |
| 14:36:13 | sean-k-mooney | i.e. that a schduler restart is needed after all compute are upgraded | |
| 14:36:19 | sean-k-mooney | but that kind of needed anyway for other easons | |
| 14:37:00 | sean-k-mooney | im not conviced the prefilter is correct by the way | |
| 14:37:02 | jgilaber | ack, I think I'll need to update the release note for that | |
| 14:37:10 | sean-k-mooney | we can go into this in more detail in teh review | |
| 14:37:16 | jgilaber | sure | |
| 14:37:25 | sean-k-mooney | effectivly you will ned to include the triat request in the request group | |
| 14:37:30 | sean-k-mooney | not just root reqiures | |
| 14:37:47 | sean-k-mooney | but ya lets move to bugs unless there are other questions? | |
| 14:38:11 | sean-k-mooney | melwitt: ^ might be good to get your input on too | |
| 14:38:59 | sean-k-mooney | #topic bugs | |
| 14:39:11 | sean-k-mooney | first one | |
| 14:39:17 | sean-k-mooney | #link https://bugs.launchpad.net/openstack-cyborg/+bug/2157586 | |
| 14:39:22 | sean-k-mooney | Cyborg Device Profile API rejects OWNER_CYBORG trait | |
| 14:39:46 | sean-k-mooney | this came up while reviewing chandands spec i belvie | |
| 14:39:56 | sean-k-mooney | chandankumar: do you want to expand on it | |
| 14:40:23 | chandankumar | it came from here https://review.opendev.org/c/openstack/cyborg-tempest-plugin/+/992210/4//COMMIT_MSG | |
| 14:40:33 | sean-k-mooney | oh yes your right | |
| 14:40:46 | chandankumar | I was adding scenario test for pci driver to create device profile there I found that | |
| 14:40:57 | sean-k-mooney | so the tl;dr is | |
| 14:41:02 | sean-k-mooney | cybrog device profiles | |
| 14:41:10 | sean-k-mooney | cannot currently request any standard traits | |
| 14:41:26 | sean-k-mooney | such as OWNER_CYBORG | |
| 14:41:50 | sean-k-mooney | this is a limitation that we should adress | |
| 14:41:57 | chandankumar | https://github.com/openstack/cyborg/blob/master/cyborg/api/controllers/v2/device_profiles.py#L211 | |
| 14:42:40 | sean-k-mooney | https://docs.openstack.org/api-ref/accelerator/#create-device-profile | |
| 14:42:47 | sean-k-mooney | if we look at the api ref | |
| 14:43:01 | sean-k-mooney | this is the descripton of the group field | |
| 14:43:05 | sean-k-mooney | """This is a list of dictionaries to describe returned accelerator resource by users, where users use keys to describe the resource_classes or traits and values to indicate its quantity or property. This is intentionally similar to extra_specs in nova flavor, and uses the same keywords for resources and traits. The key-value pair can either be a resource/trait or a Cyborg | |
| 14:43:07 | sean-k-mooney | property. Cyborg property is of the form “accel:<key>”: “<value>”. The valid key-value pairs can be found below.""" | |
| 14:43:41 | sean-k-mooney | the grousp is a dict of string key to string value | |
| 14:43:53 | sean-k-mooney | so from an api poirnt of view this is unversioned | |
| 14:44:03 | sean-k-mooney | and the limitation to CUSTOM_ triats is undocuemnted | |
| 14:44:41 | sean-k-mooney | there is some docuemition fo the valid accel: keys https://opendev.org/openstack/cyborg-specs/src/branch/master/specs/train/implemented/device-profiles.rst#valid-accel-keys | |
| 14:44:46 | sean-k-mooney | in the orginal spec | |
| 14:45:03 | sean-k-mooney | whic we really shoudl move into the api ref | |
| 14:45:13 | sean-k-mooney | but i think we can treat this as a valid bug | |
| 14:45:19 | sean-k-mooney | do others agree? | |
| 14:46:04 | sean-k-mooney | note requesting a standard trait via the device profile i think is ok. allowing you to add standard trait via the atirbute api is not | |
| 14:46:24 | sean-k-mooney | or at least not genericlly ok so this si really about the request | |
| 14:46:42 | sean-k-mooney | auto requestign OWNER_CYBORG while releated is also sperate | |
| 14:46:49 | chandankumar | yes, you mean to say device profile should accept any triat | |
| 14:46:50 | sean-k-mooney | as that has an upgrade impact | |
| 14:46:59 | sean-k-mooney | yes | |
| 14:47:18 | sean-k-mooney | and if the triat does not eixsti in the delooymen then you will just get a schduler fialure | |
| 14:47:45 | sean-k-mooney | but we shoudl jsut check it conforms to placment trait nameign shape | |
| 14:48:11 | chandankumar | good idea, first check about trait existance and then accept | |
| 14:48:35 | sean-k-mooney | i.e [A-Z_]* | |
| 14:48:48 | sean-k-mooney | well i dont think we need to check with placmeent if the trait exists although we could | |
| 14:49:11 | sean-k-mooney | but we should validate that it conforms to placment requirement for the name of a valid triat | |
| 14:49:25 | chandankumar | ok | |
| 14:49:37 | chandankumar | I will propose a patch | |
| 14:49:45 | sean-k-mooney | we coudl decied if we ant to check with the placment api in teh implemation review | |
| 14:49:55 | sean-k-mooney | woudl folks agree this shoudl be medium? | |
| 14:50:01 | chandankumar | yes | |
| 14:50:37 | jgilaber | yes, since this can be worked around | |
| 14:50:40 | sean-k-mooney | ok done feel free to assign this to yourself if you start working on it | |
| 14:51:10 | sean-k-mooney | next bug | |
| 14:51:31 | sean-k-mooney | #topic Set literals used instead of dicts in placement_client.py LOG calls | |
| 14:51:41 | sean-k-mooney | #link https://bugs.launchpad.net/openstack-cyborg/+bug/2157892 | |
| 14:51:57 | sean-k-mooney | melwitt: you found this one | |
| 14:52:29 | melwitt | oh yes that was opened by me yesterday. just something I noticed while going over code in general | |
| 14:53:28 | sean-k-mooney | https://github.com/openstack/cyborg/blob/master/cyborg/common/placement_client.py#L313 | |
| 14:53:33 | sean-k-mooney | that looks pretty valid to me | |
| 14:53:55 | jgilaber | yes, this looks valid and a good find | |