Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2026-09-14
16:27:48 gibi_ So I will dump into the PTG one my ideas. Lets try to focus on understanding each other before we start debating. Shall we?
16:28:09 gmaan yeah, thanks gibi_
16:28:20 Uggla sounds good
16:28:31 gmaan i mean sounds good to me but other can tell their preferece too
16:28:50 gmaan but i agree that email are hard to read and get consensus
16:28:53 bauzas I'd be prefering to help nova reviewers instead of other service reviewers but let's discuss this in the etherpad
16:29:27 gibi_ OK sounds like we have the next step. Thanks
16:29:30 gibi_ Uggla: back to you
16:29:53 Uggla thx gibi_
16:30:36 Uggla Anything else you'd like to discuss ?
16:30:55 Leo[m] Hello
16:31:21 Leo[m] Just wanted to circle back on my spec about using hostname for ceph mon references: https://review.opendev.org/c/openstack/nova-specs/+/1000098
16:31:32 sean-k-mooney o/
16:31:46 Leo[m] It's been up for a while and I was hoping to get someone to look at it
16:32:01 Leo[m] Please let me know if there is anything else I can do on my end
16:32:01 sean-k-mooney Leo[m]: i assume you mean dns name so FQDN
16:32:03 sean-k-mooney not hostname
16:32:51 Leo[m] Yes, the intention is to use a generic CNAME record so the ceph mons can change without needing to change the references in Nova
16:33:01 sean-k-mooney Uggla: for my topic i wanted to ask can we reappove the cybrog manage mode specless bluepint for this cycle
16:33:34 sean-k-mooney it missed last cycel because of the cyborg side not nova so im hopign stephen and Melanie will be fine with revieing it this cycle
16:34:20 sean-k-mooney Leo[m]: if the entire tech stack supprot that i think that woudl be infinetly better then the situation we have today
16:34:57 Uggla sean-k-mooney ok but will it need changes on the nova side ?
16:35:09 tkajinam Leo[m], maybe the right path is to propose that topic to the upcoming ptg and discuss details there, while that would attract attentions from cores.
16:35:18 sean-k-mooney yes but i have written those its a very small patch
16:35:26 Uggla tkajinam ++
16:35:30 sean-k-mooney Uggla: like -20+300 lines
16:35:39 tkajinam maybe it's a good timing to create an etherpad to gather topics ?
16:35:53 sean-k-mooney Uggla: if it didnt need nova change it woudl not have a nova blueprint :)
16:35:54 Uggla tkajinam done
16:36:01 tkajinam Uggla++
16:36:37 sean-k-mooney https://review.opendev.org/c/openstack/nova/+/994579
16:36:44 sean-k-mooney that is the entirity of the patch
16:36:54 Leo[m] What would it take to propose the change at the upcoming PTG?
16:37:12 sean-k-mooney Leo[m]: you just need to add it to the adgenda
16:37:19 sean-k-mooney and Uggla will help you find a spot
16:37:27 sean-k-mooney *time
16:38:09 Leo[m] Ok. I don't want to take up too much time for such a small change
16:38:33 sean-k-mooney its an imporant operational pain point
16:38:46 sean-k-mooney so i think its size is not the thing to focus on :)
16:39:11 sean-k-mooney Leo[m]: i have not looked at the spec yet
16:39:22 sean-k-mooney but the main aspect i would ask is how do we upgrade to this
16:39:30 sean-k-mooney per host config options are not a good path
16:39:36 sean-k-mooney at least not long term
16:39:42 sean-k-mooney as they cause issues with live migration
16:39:54 sean-k-mooney but ya we can work out those details later
16:39:57 Uggla Leo[m] usually we discuss topic at PTG and then the changes are reviewed.
16:40:33 Leo[m] What do you mean by "per host config options"?
16:40:48 Leo[m] Sounds good. Does that mean I should prepare the PR now?
16:41:21 sean-k-mooney ideally we woudl not bave a boolen config flage to opt into using the hostname
16:41:26 sean-k-mooney ```As currently proposed, one config option will be added
16:41:28 sean-k-mooney (a boolean in ``[libvirt]``, default ``False``) that allows operators
16:41:30 sean-k-mooney to opt in to using hostnames for the mon addresses```
16:41:51 Uggla only spec is required, but if you have a "draft" patch that something helpful.
16:42:54 sean-k-mooney nova get the ceph mon info from the ceph.conf
16:43:23 Leo[m] Ok, great. Thank you for your help. Is there a link to information about the PTG and its agenda?
16:43:25 sean-k-mooney so ideally the mon adresses there woudl jsut eb updated to the fqdn and nova would "just work"
16:44:04 sean-k-mooney note that nova access ceph in 2 ways
16:44:05 Uggla Leo[m] https://etherpad.opendev.org/p/nova-2027.1-ptg
16:44:22 sean-k-mooney we use the ceph cli and the pythyon bi9nding so we need a coherent story for both
16:44:41 tkajinam put your topic at the bottom
16:46:07 sean-k-mooney so in the interest of time, are we ok with updateign https://blueprints.launchpad.net/nova/+spec/support-vfio-variant-driver-managed-mode-via-cyborg to 2027.1 and seting it back to approved?
16:46:46 sean-k-mooney Uggla: you marked it as implemnted on last week but i reset it because its still in progress hence its current state
16:46:57 Uggla sean-k-mooney ok
16:47:11 gibi_ sean-k-mooney: any findings from the original plans during the implementation?
16:47:21 Uggla so I might have write something wrong in HL/prelude.
16:47:42 sean-k-mooney gibi_: just that the cybrog side needed me to implment ovo indirection api..
16:48:06 gibi_ sean-k-mooney: ack that probably not affecting nova :)
16:48:06 sean-k-mooney gibi_: i realised late that we didnt have that in cybrog so i paused the work to focus on other features
16:48:33 sean-k-mooney so my plan is to complete the cybrog work includign ci then update the nova patch once im happy with all of that
16:49:14 sean-k-mooney im not expecting any scope change on the nova side for this specificly
16:49:37 sean-k-mooney i hae some ptg topic like using the sdk to talk to cybrog proerply
16:49:43 sean-k-mooney but that independint of this feature
16:49:45 gibi_ sean-k-mooney: sounds good to me
16:50:06 gibi_ Uggla: if this is in the prelude and highlight the we need to get remove it :)
16:50:37 sean-k-mooney i dont htink this is is there
16:50:48 sean-k-mooney i think we called out the mdev support
16:50:51 sean-k-mooney that did merge
16:50:58 sean-k-mooney so the nvidia drvier in thery shoudl now work
16:51:26 Uggla yep but I extended with variant because I thought it was merged.
16:51:56 sean-k-mooney oh ok
16:52:01 sean-k-mooney almost but not quite
16:53:01 sean-k-mooney ah i see https://docs.openstack.org/releasenotes/nova/2026.2.html
16:53:10 sean-k-mooney ya i can push a fix for that if you want
16:53:48 sean-k-mooney well this is the current text
16:54:01 sean-k-mooney ```Nova’s libvirt driver now supports Cyborg-managed mediated devices (such as vGPUs) via a new MDEV accelerator request binding type. Operators can choose to manage vGPU/mdev lifecycle through Cyborg as an alternative to Nova’s native support. Both management paths coexist safely thanks to a new OWNER_NOVA trait that prevents scheduling collisions between Nova-managed and
16:54:03 sean-k-mooney Cyborg-managed devices.```
16:54:16 sean-k-mooney that does not mention manage mode
16:54:19 sean-k-mooney so i think we are good
16:55:09 Uggla yep
16:55:37 gibi_ cool then
16:55:49 sean-k-mooney ok so jsut to be clear gibi_ Uggla are you ok with markign this as approved again for this cycle
16:56:00 Uggla yep
16:56:03 sean-k-mooney if so we can move on/finsh up for this week
16:56:08 gibi_ 6yepp
16:56:27 sean-k-mooney cool thanks
16:56:30 gibi_ one queston before we stop
16:57:04 gibi_ are we then OK no to have an RC2 for the os-vif change being bumped into nova's min requirements?
16:57:37 sean-k-mooney we could its not stricly needed but i dont object
16:57:57 gibi_ https://review.opendev.org/c/openstack/os-vif/+/1005496
16:58:08 sean-k-mooney that a diffent question

Earlier   Later