Tuesday, 2026-04-28

opendevreviewchandan kumar proposed openstack/cyborg master: Add VM setup documentation for Cyborg development  https://review.opendev.org/c/openstack/cyborg/+/98396406:51
opendevreviewchandan kumar proposed openstack/cyborg master: Add VM setup documentation for Cyborg development  https://review.opendev.org/c/openstack/cyborg/+/98396406:52
opendevreviewchandan kumar proposed openstack/cyborg master: Add Cyborg NVMe driver development environment documentation  https://review.opendev.org/c/openstack/cyborg/+/98271106:52
opendevreviewchandan kumar proposed openstack/cyborg master: Add Cyborg NVMe driver development environment documentation  https://review.opendev.org/c/openstack/cyborg/+/98271108:15
opendevreviewchandan kumar proposed openstack/cyborg master: Add user guide for creating instances with accelerators  https://review.opendev.org/c/openstack/cyborg/+/98302708:17
opendevreviewchandan kumar proposed openstack/cyborg master: Add doc around cyborg tempest test and grenade testing  https://review.opendev.org/c/openstack/cyborg/+/98303008:17
opendevreviewchandan kumar proposed openstack/cyborg master: Add doc around cyborg tempest test and grenade testing  https://review.opendev.org/c/openstack/cyborg/+/98303008:21
opendevreviewchandan kumar proposed openstack/cyborg master: Add VM setup documentation for Cyborg development  https://review.opendev.org/c/openstack/cyborg/+/98396410:03
chandankumarCyborg IRC meeting in another 2 hours. Feel free to add your agenda here: https://etherpad.opendev.org/p/openstack-cyborg-irc-meeting#L45 thank you!12:01
opendevreviewchandan kumar proposed openstack/cyborg master: pci: Fix missing OWNER_CYBORG trait in PCI devices  https://review.opendev.org/c/openstack/cyborg/+/98653613:46
chandankumarHello all, who all are around for meeting14:03
chandankumar#startmeeting cyborg14:03
opendevmeetMeeting started Tue Apr 28 14:03:47 2026 UTC and is due to finish in 60 minutes.  The chair is chandankumar. Information about MeetBot at http://wiki.debian.org/MeetBot.14:03
opendevmeetUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.14:03
opendevmeetThe meeting name has been set to 'cyborg'14:03
chandankumaro/14:03
sean-k-mooneyo/14:03
jgilabero/14:04
morenodo/14:04
chandankumar#courtesy ping bogdando[m] rlandy amoralej 14:04
amoralejo/14:04
chandankumarIn the mean time, feel free to add your agenda https://etherpad.opendev.org/p/openstack-cyborg-irc-meeting#L4514:04
chandankumarLet me get started with the first topic14:05
chandankumar#topic 2026.2 hibiscus ptg summary14:05
chandankumarI have added this topic14:05
chandankumarThank you sean-k-mooney for sending the ptg summary to mailing list 14:06
chandankumar#link https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.org/thread/MBE5DX4IQTD4SQOTSWNTDBBHTN5BSCPE/14:06
sean-k-mooneyif there are any corrections please respond or let me know14:06
rlandyo/14:07
jgilaberI looked at the email this morning, everything looked right to me14:07
chandankumarsure, will do that.14:08
chandankumarIf there is no question on above topic, let me move to the next section as there is no other topic14:08
chandankumar#topic reviews14:08
chandankumar#link https://review.opendev.org/q/topic:%22cyborg-dev-doc%2214:09
chandankumarI was working on improving the contributor doc.14:09
chandankumarThank you everyone for all the feedback on the patch series. I have addressed them. Feel free to take a look.14:09
chandankumarMoving to next one14:10
chandankumar#link     https://review.opendev.org/c/openstack/cyborg/+/984508: Add placement client config for devstack multinode14:11
chandankumarthis one fixes the devstack plugin for multinode job and add a new multinode14:11
chandankumardevstack job.14:12
chandankumarPlease add it to your review list.14:12
chandankumarDo we have any more reviews anyone wants to highlight?14:13
chandankumarmoving to the next section14:14
chandankumar#topic bugs14:14
chandankumar#link PCI devices does not have OWNER_CYBORG traits https://bugs.launchpad.net/openstack-cyborg/+bug/215055714:15
chandankumarDuring 2026.2 ptg, under driver development discussion, we talked about traits and pci devices does not OWNER_CYBORG14:15
sean-k-mooneyack review might be a bit slow this week but ill try to pick back up next week14:16
chandankumarfor that one, I have opened the bug14:16
chandankumarhttps://review.opendev.org/c/openstack/cyborg/+/986536 it fixes the same.14:16
sean-k-mooneyok that may be a suitable bug14:16
chandankumarSince it changes the output in openstack accelerator device attribute list14:16
sean-k-mooneythe upgrade impact of that shoudl be effectivly 014:16
chandankumarwill it need a release notes? and test coverage?14:17
sean-k-mooneyi default to yes in both cases14:17
chandankumarok good14:17
sean-k-mooneyit need at least unit test coverage14:17
chandankumarPCI device lacks unit tests.14:18
sean-k-mooney...14:18
jgilabershould we add that trait to all drivers? A quick search shows that other drivers seem to be missing it14:18
chandankumarI need to take a look how much we have coverage14:18
jgilaberhttps://github.com/search?q=repo%3Aopenstack%2Fcyborg%20traits&type=code14:18
chandankumaryes, 14:18
sean-k-mooneyjgilaber: yes so i wonder if we want to track this as more then a bug14:18
chandankumarit is currently available in FGPA, PCI, GPU only14:18
jgilaberyes a specless blueprint might be a good fit I think14:18
sean-k-mooneythis is part of the "how do we model devices in placment" topic14:18
jgilaberparticularly if the testing needs significant work14:19
chandankumarI will take a look at how much test coverage we have there for device spec and whitelist there is no coverage14:19
sean-k-mooneyi added some test when i fixed eliaer bugs in the pci driver by the way14:19
sean-k-mooneyas in unit tests14:20
sean-k-mooneywe will want to have tempest test for this later too14:20
chandankumarhttps://github.com/openstack/cyborg/tree/master/cyborg/tests/unit/accelerator/drivers/pci/pci14:20
chandankumarhttps://github.com/openstack/cyborg/blob/master/cyborg/tests/unit/accelerator/drivers/pci/pci/test_sysinfo.py14:20
sean-k-mooneyhttps://github.com/openstack/cyborg/blob/master/cyborg/tests/unit/accelerator/drivers/pci/pci/test_sysinfo.py14:20
chandankumarit will the test voverage 14:20
sean-k-mooneyya so we have driver specific test although perhpas not where you expect them14:21
sean-k-mooneyso at the very lease we need unit test cover for this in the future functional and tempest test too14:21
chandankumarregarding traits, there is some code duplication about constant names, May be I can clean it in the same patch14:22
sean-k-mooneyin general that better to do in a sperate patch14:22
chandankumarok14:22
sean-k-mooneybut it kind fo depends14:22
sean-k-mooneyon if we woudl abckprot this or not14:22
chandankumarwill do a follow up patch then14:22
sean-k-mooneyin my mind this is maser only14:22
sean-k-mooneyas a general rule you shoudl do code cleanup in a sepreate patch form bug or feature work14:23
chandankumaras you said there is no upgrade impact, i think it is ok to backport14:23
sean-k-mooneywell there is14:23
sean-k-mooneybut its mitigated by how old that specific trait is14:23
chandankumarhttps://github.com/openstack/cyborg/commit/5f896d2931d1a02e4beaf2cb4109a8a3e384b155 unifies pci management14:24
sean-k-mooneyif this trait was new this cycle we coudl nto backprot it to an older relase as an example14:24
sean-k-mooneythat not what you need to check in this case14:25
sean-k-mooneyits https://github.com/openstack/os-traits/blob/master/os_traits/owner/__init__.py14:25
sean-k-mooneythat was added 4 years ago14:25
sean-k-mooneyif it was new in 2026.2 you woudl have to check fi the standard trait exsitns in placment before reporting it14:25
sean-k-mooneyas we have to assuem placment could be at least 2 release older then cyborg14:26
sean-k-mooneywe also need to acont for the fact the cybrog compute agent can be up to 1 slrup older14:26
sean-k-mooneyin thei case we cant assume its newer then 2026.1 and we cant assume it has this backport14:27
chandankumarah in that case, I keep this to master only.14:27
sean-k-mooneyso when we translate teh device spec into a placment request group14:27
sean-k-mooneywe cant unconditonly assume its updated14:28
chandankumarI did not consider these assumption14:28
sean-k-mooneyso this will need more work to make sure we dont break upgrades14:28
sean-k-mooneythe reporting of the triat is fine14:28
sean-k-mooneythe requesting of it need to check the compute-agent verion and gate it on the min version14:28
sean-k-mooneyor a config option14:29
sean-k-mooneygiven we are not reprotign it today we must not be requesting it either14:29
sean-k-mooneylets loop bakc to that in more detail on teh review or outside the meeting14:29
chandankumarsure14:29
chandankumarthank you sean-k-mooney!14:30
chandankumarSince there is no more bugs, moving to open discussion14:30
chandankumar#topic Open Discussion14:30
chandankumarFeel free to bring topic here14:31
sean-k-mooneyi guess now is as good a time as any to say14:32
sean-k-mooneyi will be on pto friday and monday and im thinking of also taking next tuesday14:32
sean-k-mooneyso i may not be be here next week for the meeting14:33
chandankumarsure, enjoy your time off!14:33
chandankumarMoving to last one14:33
chandankumar#topic volunteer to chair for next meeting14:34
chandankumarAny volunteer up for chairing the next meeting?14:34
jgilaberI can do it14:34
chandankumarthank you jgilaber !14:34
chandankumarNow closing the meeting early, thank you all for attending!14:35
chandankumar#endmeeting cyborg14:35
opendevmeetMeeting ended Tue Apr 28 14:35:18 2026 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)14:35
opendevmeetMinutes:        https://meetings.opendev.org/meetings/cyborg/2026/cyborg.2026-04-28-14.03.html14:35
opendevmeetMinutes (text): https://meetings.opendev.org/meetings/cyborg/2026/cyborg.2026-04-28-14.03.txt14:35
opendevmeetLog:            https://meetings.opendev.org/meetings/cyborg/2026/cyborg.2026-04-28-14.03.log.html14:35
sean-k-mooneychandankumar: jgilaber i dont know if we want to continue to discuss the upgrade condieration or not right now or if ye want to reflect on my comments14:36
sean-k-mooneyi think before we proceed with reproting the owner triat (that shoudl be safe) we need to do a deeper review of how that currently interact with t hegeenrat request groups in the arq14:37
sean-k-mooneyand look at the exact mechanis when we have up to n-2 compute-agent runnign with a fully upgraded api/conductor14:38
chandankumar+1 to do a deep review on interaction with placement and cyborg on owner traits14:38
chandankumarWe might discover more bugs in that.14:39
sean-k-mooneyopenstack --os-cloud devstack-admin resource provider trait list 45593a00-c178-3159-9687-6d7efa55f64d14:39
sean-k-mooney+----------------------------+14:39
sean-k-mooney| name                       |14:39
sean-k-mooney+----------------------------+14:39
sean-k-mooney| CUSTOM_PCI_INTEL           |14:39
chandankumarand then revisit this again in another meeting.14:39
sean-k-mooney| CUSTOM_PCI_PRODUCT_ID_154C |14:39
sean-k-mooney+----------------------------+14:39
sean-k-mooneyi can confirm that the trait is indeed missing14:39
sean-k-mooneyso its not stricly a bug14:40
sean-k-mooneythe suprpot for this was part of a new spec14:40
sean-k-mooneythat was not fully impmented14:40
sean-k-mooneyi.e. owner traits were only intoduced a long time after cybrog was creted 14:40
sean-k-mooneyas part fo a feature that spaned nova and cybrog that was never impmented on teh nova side14:41
sean-k-mooneyand only partly doen in cyborg14:41
sean-k-mooneyso im wondering if this shoudl really be a bug at all given it was orgially a feature qruieding a spec14:41
chandankumarhttps://github.com/search?q=repo%3Aopenstack%2Fnova-specs%20OWNER_CYBORG&type=code there are two specs14:41
sean-k-mooneyand given the nvova part was never done (its part of the nova spec for vgpus)14:41
chandankumarthis one https://github.com/openstack/nova-specs/blob/e53cb185c680e479a44a94ef0cfb7e1f52c762d6/specs/wallaby/approved/cyborg-vgpu-support.rst#L125 ?14:42
sean-k-mooneyyep both are appvoed not implemtned14:42
sean-k-mooneyso the nvida driver was partly updated in cybrog because of this14:42
sean-k-mooneybut it was never completed and the canges for the other cybrog driver were not done14:42
sean-k-mooneyhttps://github.com/openstack/cyborg-specs/blob/master/specs/wallaby/approved/vgpu-driver-proposal.rst14:43
sean-k-mooneywas the cyborg spec for this14:44
sean-k-mooneythat was finally implemtned in 2023.1 https://github.com/openstack/cyborg-specs/blob/master/specs/2023.1/implemented/vgpu-driver-proposal.rst14:45
sean-k-mooneywithout the nova half and without updatign the other drivers to also use the cybrog owner trait14:45
chandankumarwhen you say nova half, how much work is still left to implement to make it work properly?14:47
chandankumarhttps://review.opendev.org/q/topic:%22bp/cyborg-vgpu-support%2214:47
chandankumarhttps://review.opendev.org/c/openstack/nova/+/812247: vGPU support by Cyborg is an open patch 14:47
sean-k-mooneynone of the nova change merged 14:47
sean-k-mooneyso on the nova side there are 2 related parts14:48
sean-k-mooneyfirst nova need to report all its resouce provider with OWNER_COMPUTE or OWNER_NOVA i dont recall14:48
sean-k-mooneyit thne need to have a prefilter to include that for its device requests that are not cyborg related14:49
chandankumarSince it is not properly implemented.14:51
chandankumarHow do you suggest here next course of action? 14:52
chandankumarI am going to do a deep review of how placement nova and cyborg works on traits14:52
chandankumaror we want to table this discussion for some other day?14:52
jgilaberonce we implement the owner trait in nova, after upgrading will the trait be added to an already existing resource class or does the operator need to do some action?14:53
sean-k-mooneynova shoudl do it14:55
sean-k-mooneyfor its resouce classes14:55
sean-k-mooneybut the requesting of it need to be gated by a min compute service version check14:55
chandankumarcan you share one example of min service version check?14:56
chandankumarfrom other project14:56
sean-k-mooneysure but i need to check if cybrog has the same concpet today14:58
sean-k-mooneyso in the service object14:58
sean-k-mooneyhttps://github.com/openstack/nova/blob/master/nova/objects/service.py#L4014:58
sean-k-mooneywe have an integer constant that we increment when we add specific feautres14:59
sean-k-mooneyeacg compute serivce looks at there local constant and report that value to the db14:59
sean-k-mooneynova maintian a history of what those are https://github.com/openstack/nova/blob/master/nova/objects/service.py#L6514:59
sean-k-mooneywe use them both for feature discoverya and detecting nodes that are tooo old15:00
sean-k-mooneyin the compute api among other places we can then check the compute service version and block actions15:01
sean-k-mooneyhttps://github.com/openstack/nova/blob/master/nova/compute/api.py#L373-L39415:01
sean-k-mooneyso operation that dont work with cybrog are decorated with block_accelerators15:01
sean-k-mooneywhich acppate a verion15:01
sean-k-mooneyif the min version fo any compute agent cant supprot it15:02
sean-k-mooneyhttps://github.com/openstack/nova/blob/master/nova/objects/service.py#L621-L68615:02
sean-k-mooneywe block teh request15:02
sean-k-mooneywe also have other hleper https://github.com/openstack/nova/blob/master/nova/objects/service.py#L495-L51815:03
sean-k-mooneybut the simple version of it is nova will refuse to starte some serices if a compute agent is older then it can supprot15:04
sean-k-mooneynova/cyborg or any service with a distibuted compont officlaly only supprot the current release the prior release and the prior slup release15:05
sean-k-mooneyso at most 2 relases back15:06
chandankumarI was aware about block_accelerators usage regarding controlling instance actions15:09
chandankumarah, thank you for explaining, Now i understand service version history15:11
chandankumarI just looked into the cyborg codebase, there is no such mapping15:11
chandankumarWe need to implement similar minimum version check for cyborg that we can use it during upgrade.15:14
sean-k-mooneyits actully more invaidve then that because we do not currently track compute-agent isntace in the db at all15:21
sean-k-mooneyso for cybrog we need to isntead have a lot of infra before we coudl do this automaticly so we woudl have to gate this behind a config option i think15:22

Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!