| opendevreview | sean mooney proposed openstack/cyborg master: Add release liaison guide for Cyborg https://review.opendev.org/c/openstack/cyborg/+/977177 | 01:24 |
|---|---|---|
| opendevreview | chandan kumar proposed openstack/cyborg master: [WIP] nvmevirt plugin testing https://review.opendev.org/c/openstack/cyborg/+/977083 | 10:52 |
| opendevreview | chandan kumar proposed openstack/cyborg master: [WIP] nvmevirt plugin testing https://review.opendev.org/c/openstack/cyborg/+/977083 | 12:06 |
| chandankumar | sean-k-mooney, Hello, is it possible to disable teim-ci on work in progress gerrit patch? | 12:07 |
| sean-k-mooney | yes i just enabled it last week | 12:08 |
| sean-k-mooney | but i can revert that in the automatic pipeline | 12:09 |
| chandankumar | sean-k-mooney, wonderful, thank you! | 12:09 |
| sean-k-mooney | done https://github.com/SeanMooney/ci-sean-mooney/commit/7b0dda13220463fb4ae954cfb12c444ac3175028 | 12:15 |
| sean-k-mooney | you cna still triger it manualy with a comment but it wont auto run on workflow -1 | 12:16 |
| sean-k-mooney | this friday is a rechacrge day but im planing to spend a day working on teim-ci imporment next week | 12:17 |
| sean-k-mooney | there are a few things i want to to do improve the quailty of the output more. it has been useful but i canmake it better. if there is anything you want me to consider let me know | 12:17 |
| sean-k-mooney | one of the things im oging to fix is the treshold for it to report comments | 12:18 |
| sean-k-mooney | its ment to be 80% confidance for warning and above and 90 for anything less like suggestion but i think its currenly ~60% | 12:19 |
| sean-k-mooney | i have a refactor of how it scors and verifies issues with a secound pass planned too | 12:19 |
| chandankumar | Overall feedback provided by teim-ci on the code review is helpful. | 12:55 |
| chandankumar | Just one question what is confidence scoring ? and How you are calculating? | 12:56 |
| chandankumar | is it linked with the severity? | 12:56 |
| sean-k-mooney | its calulated by the llm. its ment to be a self correction mechanium where it evaulates if the feedback si actully corect. i have some guidance on it today in the subagent but i know i can impvoe it. | 12:57 |
| sean-k-mooney | https://github.com/SeanMooney/openstack-ai-style-guide/blob/master/agents/code-review-agent.md#field-requirements-by-severity and https://github.com/SeanMooney/openstack-ai-style-guide/blob/master/agents/code-review-agent.md#severity-guidelines | 12:58 |
| sean-k-mooney | im going to incoperate part of how antopic does this in there code review plugin. and also have that checked in a second pass | 12:59 |
| sean-k-mooney | https://github.com/anthropics/claude-code/blob/main/plugins/code-review/commands/code-review.md?plain=1#L24-L53 they do a mulit pass review in there plugin and i think haveing a second agent verify the finding fo the first helps reduice fals posiitves so i want to see if i can incopert some of that into my tools as well | 13:04 |
| chandankumar | ah, thanks now it is clear. | 13:05 |
| chandankumar | Thank you for creating it. I am already using quick-rules.md from openstack-ai-style-guide in cursor | 13:05 |
| sean-k-mooney | stephen shared his agent mds yesterday https://github.com/stephenfin/openstack-agentsmd/blob/main/AGENTS.md | 13:06 |
| sean-k-mooney | im plannign to give it a try as well and incoperate some of the guidatnce into it too. | 13:07 |
| sean-k-mooney | but to do all of this properly i woudl need to spend more time on it then i currently am | 13:07 |
| sean-k-mooney | implanning to spend 1 day amonth or so incoperating impovments over time | 13:07 |
| chandankumar | I will also try stephen one. | 13:08 |
| sean-k-mooney | for our day of learnign last fristay i created https://github.com/SeanMooney/grt which is a rust port of git-review wiht the abile to also retirve the comments and a skill for ai to be able to use it https://github.com/SeanMooney/grt/blob/master/skills/grt/SKILL.md | 13:09 |
| sean-k-mooney | so that allows me to ask claude to summerise teh open comment on a gerrit review and plan how to adress tehm iteritivly | 13:09 |
| sean-k-mooney | it seams to work but i need to refine it. a bit more before its ready for a lot of use | 13:11 |
| sean-k-mooney | i will proably demo this to some folks once i have got more expirnce with using it myself | 13:12 |
| chandankumar | One more question How are you managing different Agents/Claude.md file for different projects since we cannot commit to the project repo? | 13:13 |
| sean-k-mooney | copy paste. i also simlink CLAUDE.md to point to AGENTS.md what stephen does is they reun ther llm form one directoy higher and put that AGENTS.md in the directory isntead adn it auto applies to lal the child repos | 13:52 |
| sean-k-mooney | if you dont wnat to do that claude and cursor bot supprot creating a CLAUDE/AGENTS.md in a subfolder of your user home direcotry | 13:53 |
| sean-k-mooney | that will be used for all folders | 13:53 |
| sean-k-mooney | i just leave them untracked in git | 13:54 |
| sean-k-mooney | i am debating if i want ot eventully commit someint into cyborg/watcher but for now i dont plan to do that | 13:54 |
| opendevreview | Takashi Kajinami proposed openstack/cyborg master: Remove unused sphinxcontrib-seqdiag https://review.opendev.org/c/openstack/cyborg/+/977223 | 14:15 |
| sean-k-mooney | rlandy: i created https://launchpad.net/~cyborg-bug-team so you or anyone else interested in triage can join | 15:20 |
| opendevreview | Merged openstack/cyborg-tempest-plugin master: Remove unused babel.cfg https://review.opendev.org/c/openstack/cyborg-tempest-plugin/+/976042 | 15:44 |
| opendevreview | Takashi Kajinami proposed openstack/cyborg master: Bump hacking https://review.opendev.org/c/openstack/cyborg/+/977237 | 16:19 |
| opendevreview | Takashi Kajinami proposed openstack/cyborg master: Remove unnecessary options from tox https://review.opendev.org/c/openstack/cyborg/+/977238 | 16:19 |
| rlandy | thank you sean-k-mooney | 16:27 |
| opendevreview | Takashi Kajinami proposed openstack/cyborg master: Bump hacking https://review.opendev.org/c/openstack/cyborg/+/977237 | 17:14 |
| opendevreview | Takashi Kajinami proposed openstack/cyborg master: Remove unnecessary options from tox https://review.opendev.org/c/openstack/cyborg/+/977238 | 17:14 |
| fungi | sean-k-mooney: (or anyone else involved) i'm working with the openinfra marketing and pr teams in preparation for 2026.1/gazpacho release press, and curious if there's any plans or motivations any of the maintainers can share publicly for the cyborg reinvigoration | 19:36 |
| fungi | all the media hype around ai/llm training and inference workloads makes coprocessor lifecycle management more interesting to talk about, but if it's something else that's cool to know too | 19:37 |
| fungi | bonus points if there's something that can make it into https://releases.openstack.org/gazpacho/highlights.html for the cycle | 19:38 |
| sean-k-mooney | that why my manamgent are intrested in it :) | 19:38 |
| sean-k-mooney | fungi: i have adding somthing to the cycle highlights on my todo list but not sure what if anything makes sense this cycle | 19:39 |
| fungi | i'll be on a conference call with the marketing/pr folks in 3 weeks, so there's time to come up with ideas | 19:39 |
| sean-k-mooney | one of hte non llm relatein gfature we woudl liek to adress is secure mulit tenatnt nvme managment | 19:39 |
| sean-k-mooney | nova does not offically suypport pci passthhough fo statefule devices like nvme becuase we cant clena them and provide mulit tenancy | 19:40 |
| sean-k-mooney | that is a gape that we want to adress with cyborg. there is a lot of folks intersted in better gpu managmenet as well | 19:40 |
| sean-k-mooney | for ai/llms | 19:40 |
| sean-k-mooney | but also for vdi | 19:40 |
| fungi | oh neat, so basically dedicated nvme storage pools scoped to tenants/projects? | 19:40 |
| sean-k-mooney | not nessically dedicated althoguh we could supprot that | 19:41 |
| fungi | gpu management for graphics acceleration in vdi use cases, i guess you mean? | 19:41 |
| sean-k-mooney | more just making sure we can do the end to end cleaning of the device when its released initally | 19:41 |
| sean-k-mooney | fungi: yes so nova can do that already but we do plan to invest in improving gpu manage ment in general | 19:42 |
| sean-k-mooney | vdi and ai are both usecases drivng that interest | 19:42 |
| fungi | is the nvme wiping traditional overwrites, or the modern always-encrypted approach where you just discard the encryption key? | 19:42 |
| sean-k-mooney | vdi for vmware usescases mostly | 19:43 |
| sean-k-mooney | fungi: we have not implmented it yet it willl be propsoed as a spec for 2026.2 but it will do it the modeern way | 19:43 |
| fungi | good call | 19:43 |
| sean-k-mooney | dan wrote a script to do it last cycle as a poc | 19:43 |
| sean-k-mooney | in nova-contrib | 19:44 |
| fungi | especially with nvme wear leveling, you have no idea from the system layer whether the blocks you're overwriting are the ones containing the original data | 19:44 |
| sean-k-mooney | we plan ot port/adapt that to be a nvme driver in cyborg | 19:44 |
| fungi | when you say "vmware use cases" you mean nova with esx(i) hypervisors, or vmware replacement on kvm? | 19:44 |
| sean-k-mooney | we have some customer movign form vmware deployment to openstack on libvirt/kvm | 19:45 |
| sean-k-mooney | and they have legacy vdi workloads | 19:45 |
| sean-k-mooney | now they can use nova's vgpu support already | 19:45 |
| sean-k-mooney | or gpu passthough | 19:45 |
| sean-k-mooney | but we want to ensure that we achive partiy in the cyborbg project for that and potically look at dynmic gpu slicing down the road | 19:46 |
| sean-k-mooney | nova would only ever manage static pools fo resocues | 19:46 |
| sean-k-mooney | cyborg has the potical to go behond that | 19:46 |
| sean-k-mooney | https://opendev.org/openstack/nova-contrib/src/branch/master/clean-on-delete.py is the poc i was refeint to | 19:46 |
| sean-k-mooney | fungi: honestly howver i dont nessarly have a hyped salses plan for my cyborg effort at least not yet | 19:48 |
| sean-k-mooney | i think this will evolve alot of the next 12 months | 19:48 |
| fungi | don't worry, i don't expect you to ;) | 19:49 |
| sean-k-mooney | we are looking at nvme managment and pci grouping and complting time slice gpus as our first 3 itmes manily because we can test them and under stand the poblem space | 19:49 |
| fungi | basically, the folks drafting the 2026.1/gazpacho press release are interested in including a few sentences about cyborg getting revitalized and some reasons | 19:49 |
| fungi | so this is all great info | 19:50 |
| sean-k-mooney | i put some of this in my ptl candacy https://opendev.org/openstack/election/src/branch/master/candidates/2026.2/Cyborg/smooney@redhat.com if that hleps | 19:51 |
| fungi | i know from their perspective if some company wanted to go on record saying they had increased customer interest in x/y/z that's great, but it sounds like there's plenty to talk about even without | 19:51 |
| sean-k-mooney | well i wont name names but we did have intrest form one protecail customer in cybrog but they decied nto to continue with that | 19:52 |
| fungi | i'm sure there's plenty of that too. not everyone's use cases align with a particular solution | 19:53 |
| opendevreview | sean mooney proposed openstack/cyborg master: Eventlet removal https://review.opendev.org/c/openstack/cyborg/+/970152 | 21:04 |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!