Tuesday, 2026-11-17

opendevreviewmelanie witt proposed openstack/cyborg master: Serve agent RPC under the validated resource provider name  https://review.opendev.org/c/openstack/cyborg/+/101596500:50
sean-k-mooneyo/14:08
sean-k-mooneysorry i should have started the irc meeting a few minutes ago 14:09
sean-k-mooneyill start it now in one sec14:09
jgilabernp, was about to ping you14:09
sean-k-mooney#startmeeting cyborg14:10
opendevmeetMeeting started Tue Nov 17 14:10:15 2026 UTC and is due to finish in 60 minutes.  The chair is sean-k-mooney. Information about MeetBot at http://wiki.debian.org/MeetBot.14:10
opendevmeetUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.14:10
opendevmeetThe meeting name has been set to 'cyborg'14:10
sean-k-mooneyping: amoralej bogdando jgilaber rlandy chandankumar skovili melwit14:10
jgilabero/14:10
sean-k-mooneywe will give folks a minute or two to gather then we will start14:10
sean-k-mooney#link https://etherpad.opendev.org/p/openstack-cyborg-irc-meeting#L4914:11
sean-k-mooneythe adgenda is pretty light on topics but we have some revies and bugs to look at14:12
sean-k-mooneyif you have anything to add feel free14:12
sean-k-mooneyfor now lets go directly to reviews and we can loop back ot topics at the end14:12
sean-k-mooney#topic reviews14:13
sean-k-mooney#link https://review.opendev.org/q/topic:%22bug/2157586%2214:13
sean-k-mooneyso these are 2 pathcs form chandan14:13
sean-k-mooneyhttps://review.opendev.org/c/openstack/cyborg/+/997945?usp=search14:14
sean-k-mooney#link https://review.opendev.org/c/openstack/cyborg/+/997945?usp=search14:14
sean-k-mooney#link https://review.opendev.org/c/openstack/cyborg-tempest-plugin/+/992210?usp=search14:14
sean-k-mooneyboth related to https://bugs.launchpad.net/openstack-cyborg/+bug/215758614:14
sean-k-mooneyCyborg Device Profile API rejects OWNER_CYBORG trait14:15
sean-k-mooneylast week we dicussed the validation and whether or not we shoudl rely on os-triats for that14:15
sean-k-mooneywe agreed to remove the validation instead14:15
sean-k-mooneyand rely on placement enforcign the tratis requirements14:15
sean-k-mooneythat is what the first patch does14:16
sean-k-mooney#link https://review.opendev.org/c/openstack/cyborg/+/99794514:16
sean-k-mooneythe second change 14:16
sean-k-mooney#link https://review.opendev.org/c/openstack/cyborg-tempest-plugin/+/99221014:16
sean-k-mooneyadds tempets testing of the placement trait reporting14:17
sean-k-mooneyi think in general these just need code review14:17
sean-k-mooneybut directionally i think they are ok14:17
jgilaberI've reviewed both patches, I think they are ok14:17
sean-k-mooneyack are there any questions or concerns on them?14:18
sean-k-mooneyif not i think we can try to land them this week14:18
jgilaberthere is an open thread on the tempest one about the use of 'placement_max_microversion=latest'14:18
jgilaberit's not wrong, just unnecessary14:18
jgilaberbut yes, I think we can land them soon14:18
sean-k-mooneyya we could use a lower version with the relevent one for the trate info14:18
sean-k-mooney*trait info14:19
sean-k-mooneyi have not reviewd that yet so ill respond in gerrit14:19
sean-k-mooneyok if no other comments lets move on to bugs, we have 4 new ones this week14:20
sean-k-mooney#topic bugs14:20
sean-k-mooney#link https://bugs.launchpad.net/openstack-cyborg/+bug/2174353  obj_make_compatible() raises ObjectActionError for objects with nested object fields (ExtARQ, DriverDevice)14:21
sean-k-mooneymelwitt: i belvie ^ was one you foudn whne auditing our current ovo usage14:21
sean-k-mooneythe tl;dr is that we do not have the code requried to backlevel nested ovos14:22
sean-k-mooneywhich is not surpsing as cybrog was not addign any code to backlevel at all14:23
sean-k-mooneywhich leads into the second bug14:23
sean-k-mooney#link https://bugs.launchpad.net/openstack-cyborg/+bug/217435414:23
sean-k-mooneyARQ object: project_id and attach_handle_uuid were added without version bumps; obj_make_compatible is missing14:23
sean-k-mooneyit also was not alwasy bumping the versions when requried14:23
sean-k-mooneyboth of these bugs violate the intended usage of ovo for  versioning object serisalistion14:24
sean-k-mooneythey would break rooling upgrades in any project that support roolling upgrades14:25
sean-k-mooneyhowever cyborg does not14:25
jgilaberboth seem valid and important14:25
jgilaberis there any dependence between them?14:25
sean-k-mooneyyes i tend to agree, no i do not belive so14:25
sean-k-mooneywe can retoactively add the object version for the latter14:26
sean-k-mooneyit wont really have an impact14:26
sean-k-mooneyas the fiels were added so long ago that that14:26
sean-k-mooneybranches that did nto have them are long eol14:27
sean-k-mooneythe first bug however will need to be fixed before i can add the object verion tests as aprt of the ovo indirection sepc14:27
sean-k-mooneyim inclidne to set https://bugs.launchpad.net/openstack-cyborg/+bug/2174354 as high14:28
sean-k-mooneyand https://bugs.launchpad.net/openstack-cyborg/+bug/2174353 as medium14:28
sean-k-mooneysorry 14:28
sean-k-mooneyother wasy around 14:28
sean-k-mooneyhttps://bugs.launchpad.net/openstack-cyborg/+bug/2174353 the issue with nested object fields 14:28
sean-k-mooneyshould be high14:28
jgilaberok, sounds good14:28
sean-k-mooneyhttps://bugs.launchpad.net/openstack-cyborg/+bug/2174354 the fact that an object version was not done is medium14:29
sean-k-mooneybecause everyoen who woudl have been broken by that wer boken in 201914:29
jgilaberyes, that makes them less critical14:31
sean-k-mooneyok i added a comment ot each set triaged and added tags14:32
sean-k-mooneyhttps://bugs.launchpad.net/openstack-cyborg/+bug/2174355 is more of the same14:32
sean-k-mooney#link https://bugs.launchpad.net/openstack-cyborg/+bug/217435514:32
sean-k-mooneyExtARQ obj_make_compatible does not handle version 1.1 changes14:32
jgilaberyeah same, medium seems fine for this14:33
sean-k-mooneyso before last cycle when i asked folks to start addign obj_make_compatiable14:33
melwitto/14:33
sean-k-mooneycybrog did not have any defintions of that if i recall14:33
jgilaberthis does have a dependency on the first one14:33
sean-k-mooneyi know some of the change perhaps in chandands serise backfile some of these methods14:33
sean-k-mooneyi guess so yes14:34
sean-k-mooneyat some point in the next 12-18 months im expecting use to repalce all our ovos with v2 objects14:35
sean-k-mooneymainly because i think we need to concodliate so we nolonge have top levlel and driver copies of the objects14:35
sean-k-mooneythta does not mean we cant add the backlevleing supprot14:36
sean-k-mooneyhowever there are no supproted deployment senarios wehre that code would be requried14:36
sean-k-mooneyi.e. all stable release supprot the latest version of the ovos so we should never have to backlevel14:37
sean-k-mooneyfor the saem reason as the other bug i think this should be traiges as medium14:37
sean-k-mooneywhat do folks think?14:37
jgilaber+1, medium seems right14:38
jgilaberthey are not critical, but would be good to fix them14:39
sean-k-mooneyok left a comment14:41
sean-k-mooneythat brings us to the final bug reported by melwitt 14:41
sean-k-mooney#link https://bugs.launchpad.net/openstack-cyborg/+bug/217445914:41
sean-k-mooneyAPI -> agent RPC (fpga_program) targets the agent by device hostname, which does not match the agent's RPC server name 14:42
sean-k-mooneyso ^ is a disconnecte between confg.HOST adn hypervior_hostname14:42
sean-k-mooneyi think14:42
sean-k-mooneymelwitt: do you want to cover this topic14:43
melwittthe fpga programming test is apparently the only one that makes a RPC call to the agent, somehow14:43
sean-k-mooneyi have some comemtn on older reviwe wher ei point out that https://github.com/openstack/cyborg/blob/master/cyborg/accelerator/drivers/gpu/nvidia/sysinfo.py#L31814:44
sean-k-mooneyis kind fo problematic14:44
melwittit's fine if you do :) but yeah it's that the RPC queue name is based off the short name (CONF.host) but then the RPC call is targeting a name based off the FQDN14:44
melwittI proposed a patch that may or may not be the right fix, but it's just something to look at14:45
sean-k-mooneyack are you happy to assing https://bugs.launchpad.net/openstack-cyborg/+bug/2174459 to yourself14:46
sean-k-mooneyi thinkt his is valid and medium14:47
sean-k-mooneyah i was wonderign if it was related to  https://github.com/openstack/cyborg/commit/0e9a632f3165af8d108301d7838ba2792de64222 or not14:48
melwittsure. I forgot the automation no longer does assignment14:48
sean-k-mooneyreally we need to have host and hostname on the devices at least internally14:48
melwittyes I think it is related to that, that's what created the deviation14:48
sean-k-mooneythe same way an instnace has host and hypervioer_hostname14:48
sean-k-mooneyhum i see your fixing this by changing the rpc endpoint14:49
sean-k-mooneyim not sure that is really correct14:50
sean-k-mooneythe reaosn i did the split is its not genrelaly correct to assume the rpc endpoint name and placment resouce provier for a comute are the same14:50
melwittyeah, I wasn't sure how to approach this14:50
sean-k-mooneyya its tricky14:52
sean-k-mooneyim tryign to think what the upgrade impactg ot both aprohces are14:52
sean-k-mooneyif we proceded with yoru change conf.host nolonger has any value14:52
sean-k-mooneyit exist to defein the rpc endpoint that identifies the agent14:52
sean-k-mooneyand we use the hostname as the identifyer by defualt14:53
sean-k-mooneysince it should be unique, it also has to be a value that is agreed between nova and neutron and cybrog14:53
sean-k-mooneyseperatly the compute node root resouce provdier name is not CONG.host14:54
sean-k-mooneytis the comptue nodes hypervioer_hostname14:54
melwittyeah, it does feel weird to make it act different than the usual way presumably all other services work. so the other way would be to make the caller somehow use the right queue name rather than the resource provider name?14:54
sean-k-mooneyya so i think we need ot add a second filed on the device objects14:55
sean-k-mooneyto hold a refence to the service (or the queue topic) i.e CONF.host14:55
sean-k-mooneyim just not sure how to do that in a backprotable way14:55
sean-k-mooneythe orgianl problem i was tryign to solve is nova and cybrog did not agree on the name of the resouce provider14:56
sean-k-mooneycyborg expected the short hostname, libvirt provided the fqdn when there was a relvent domain14:56
melwittyeah not sure how adding a second field on device could be backportable14:58
sean-k-mooneythe only way to do it would be to stashit in an exitng colum14:58
melwitt(I have to join another call in a minute, sorry)14:58
sean-k-mooneyno worries we could revert my eiarlry patch adn reintuduce it in master with a new filed14:59
sean-k-mooneyim nust concerned that woudl also break peole14:59
sean-k-mooneyit was possible to change CONF.hsot to "make it work" before14:59
sean-k-mooneyand that also a workaround here14:59
sean-k-mooneylets reviesit this next week14:59
sean-k-mooneyor on irc14:59
sean-k-mooneyare we agreed its valid and high or medium15:00
jgilabersounds good, I need to spend some time looking at this15:00
sean-k-mooneyill set it to high15:00
sean-k-mooneythat more or less it for this week15:00
jgilaberyes, sounds right15:00
sean-k-mooneyanyone open to takign the meting next week?15:01
jgilaberI can do it15:01
sean-k-mooneythank jgilaber 15:01
sean-k-mooneyin that case thanks for coming and lets call it there15:01
sean-k-mooney#endmeeting15:01
opendevmeetMeeting ended Tue Nov 17 15:01:55 2026 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)15:01
opendevmeetMinutes:        https://meetings.opendev.org/meetings/cyborg/2026/cyborg.2026-11-17-14.10.html15:01
opendevmeetMinutes (text): https://meetings.opendev.org/meetings/cyborg/2026/cyborg.2026-11-17-14.10.txt15:01
opendevmeetLog:            https://meetings.opendev.org/meetings/cyborg/2026/cyborg.2026-11-17-14.10.log.html15:01
*** user is now known as Guest118218:00
opendevreviewmelanie witt proposed openstack/cyborg master: Carry the agent RPC identity on reported devices  https://review.opendev.org/c/openstack/cyborg/+/101596518:47
melwittsean-k-mooney: changed ^ to add a service_host field to devices. note I will be out next week but we can discuss before/after that18:51
sean-k-mooneymelwitt: ack cool19:06
sean-k-mooneyi think that is certenly a better long term fix for master19:07
sean-k-mooneythe real question is for stable are we ok with jsut documentign that CONF.host shoudl equal CONF.agent.resource_provider_name19:08
sean-k-mooneythat is the workaorund for now just make the two the same. today they only differ if socket.getHostname() and socket.getFQDN() retun differnet values19:09
melwittthat's what I'm thinking19:12
opendevreviewmelanie witt proposed openstack/cyborg master: Carry the agent RPC identity on reported devices  https://review.opendev.org/c/openstack/cyborg/+/101596519:13
sean-k-mooneymelwitt: are you on pto form tomorrow?19:21
melwittsean-k-mooney: no, not until next week19:23
sean-k-mooneyoh ok im going to finish up shortly be we can caht about the the patch more tomroow so19:23
melwittbasically I am out next week only19:24
melwittnot sure how much of a rush it is, just didn't want you to be disappointed next week if I'm not around19:25
melwittand I figured better to put upload something sooner than later so we don't forget about it19:25
sean-k-mooneyi think now that we know its an issue as long as we reolve it this cycle its fine19:29
sean-k-mooneyso important but not urgent19:30
* melwitt nods19:31

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