Thursday, 2025-11-27

opendevreviewSeyeong Kim proposed openstack/nova master: libvirt: Support boot_index for multiple block devices  https://review.opendev.org/c/openstack/nova/+/96366502:17
*** mhen_ is now known as mhen02:27
opendevreviewWill Szumski proposed openstack/nova master: Randomise best hosts  https://review.opendev.org/c/openstack/nova/+/96854409:25
opendevreviewJohannes Kulik proposed openstack/nova-specs master: Spec for asynchronous volume-attach API  https://review.opendev.org/c/openstack/nova-specs/+/95890010:38
opendevreviewJohannes Kulik proposed openstack/nova-specs master: Spec for asynchronous volume-attach API  https://review.opendev.org/c/openstack/nova-specs/+/95890011:15
r-taketnHello everyone. I have uploaded the SEV/SEV-ES refactoring codes. Could you please review it and provide some feedback?11:33
r-taketn#link https://blueprints.launchpad.net/nova/+spec/generalize-sev-code11:33
r-taketnIt was discussed at PTG (https://etherpad.opendev.org/p/nova-2026.1-ptg#L687) and decided to make this specless. And, I have run SEV/SEV-ES tests (both manually and using tempest) with uploaded codes. 11:37
opendevreviewDmitriy Rabotyagov proposed openstack/nova-specs master: [spec] Add Cross-AZ scheduling blueprint  https://review.opendev.org/c/openstack/nova-specs/+/90029611:49
masahitohi nova team,  please review the bug fix for 500 Internal Server error in the lock API https://review.opendev.org/c/openstack/nova/+/946223 . Thanks.14:34
noonedeadpunksean-k-mooney: sorry for bothering you again... I'm trying to process reviews on the spec, and got confused myself as well. Given that weights are run per hypervisor, then basically we can't use heapq basically?17:16
noonedeadpunkthere's no reason to build AZ to Host mapping, as we just need to get current hypervisor aggregates, and fetch AZs from it. And then fetch azs foreach request_spec.instance_group.members17:17
noonedeadpunkand then like count how many instances are in the same az, and use that as a negative multiplier?17:18
sean-k-mooneywe can the heap woudl need to be be in the request spec object17:19
sean-k-mooneyalthernitively you can impelent the wigher by overriding the follow fucntion17:19
sean-k-mooneyhttps://github.com/openstack/nova/blob/master/nova/weights.py#L100-L12017:19
sean-k-mooneythat will allow it to process all hosts at once and you jsut return the list of wights17:20
sean-k-mooneynoonedeadpunk so the point i was orgianly trying to make is you rare not ment to fetch data in the weighers17:20
sean-k-mooneyonly use the data form the host state or request spec17:21
sean-k-mooneywe do not have any weigher that operate on aggreats today btu we do have filters that do17:22
sean-k-mooneyhttps://github.com/openstack/nova/blob/master/nova/scheduler/filters/aggregate_instance_extra_specs.py#L5017:22
noonedeadpunkok, I was just looking at existing weighters, but seems none of them is actually using weigh_objects17:23
sean-k-mooneybut even then you lotice that we never do lookups on object or db queries17:23
sean-k-mooneynoonedeadpunk: correct non of the intree ones weigh all the hosts at once17:23
sean-k-mooneyhttps://github.com/openstack/nova/blob/master/nova/scheduler/weights/affinity.py17:23
sean-k-mooneyis the best refence for what you currently want to do17:24
noonedeadpunkyeah, was checking that, and it seems that it should have everything needed17:24
noonedeadpunkexcept not doing it at once17:24
sean-k-mooneyso  i ugess it does indrectly17:25
sean-k-mooneyactully no17:25
sean-k-mooneyit does not have the count of isntace in the curren group in the a given az17:25
sean-k-mooneyif you were to weigh all host you shoudl have the info you need however17:26
noonedeadpunkbut then if weigh_objects is used - does that still run for each hypervisor?17:26
sean-k-mooneysince you can gorup the host_state obejct by AZ then count the number of instnace form the isntnace grou pin each group17:26
sean-k-mooneynoonedeadpunk: no it will run once17:26
noonedeadpunkso.... then also doing a query is not that bad?17:27
sean-k-mooneyyou don tneed to do a query17:27
sean-k-mooneythe info shoudl already be there17:27
sean-k-mooneyif you look at the base impoealtion of weigh objects https://github.com/openstack/nova/blob/master/nova/weights.py#L100-L12017:27
sean-k-mooneythat the forlop that runs the weigher over all host one by one17:28
sean-k-mooneyweighed_obj_list is the list of HostState objects17:28
noonedeadpunkso weighed_obj_list is list of HostState? 17:28
sean-k-mooneyweight_properties is the request spec17:28
noonedeadpunkok17:28
noonedeadpunkyes17:28
sean-k-mooneywe never got a round to fixing the fact they are called diffent thing in diffeent palces17:29
* noonedeadpunk will add description there....17:29
noonedeadpunkyeah, ok, makes sense17:29
noonedeadpunkthanks for clarifying it17:29
noonedeadpunkas I got a little bit confused17:30
sean-k-mooneyya i proably could have explaid this better17:30
sean-k-mooneyit only clicked now that we shoudl ahve all the data we need already 17:30
sean-k-mooneybut that to implement this weigher we shoudl jsut overried weigh_objects17:30
noonedeadpunkyeah17:31
noonedeadpunkI had no idea what these parameters are tbh17:31
noonedeadpunk*arguments17:31
sean-k-mooneyya so in newer filtere/weigher we name them what they are17:31
sean-k-mooneybut in th older once we never fixt them to make sense17:32
noonedeadpunk++17:32
noonedeadpunknow it looks *way* simpler then I originally anticipated17:33
opendevreviewDmitriy Rabotyagov proposed openstack/nova-specs master: [spec] Add Cross-AZ scheduling blueprint  https://review.opendev.org/c/openstack/nova-specs/+/90029617:52
sean-k-mooneynoonedeadpunk: same18:09
noonedeadpunkum?18:19
noonedeadpunksean-k-mooney: sorry, did not get what you meant18:19
sean-k-mooneyits simpler then i was expecting also18:20
noonedeadpunkah18:20
noonedeadpunk:D18:20
noonedeadpunkI just thought it was comment to spec update :D18:23
opendevreviewSean Mooney proposed openstack/nova-specs master: add spec for resouce notifications  https://review.opendev.org/c/openstack/nova-specs/+/96771219:02
noonedeadpunkfwiw, we've found a bug with parallel migrations on ubuntu 24.04 with libvirt 10.0. It's a libvirt bug actually, and I think it's fixed in 10.219:56
noonedeadpunkbut parallel live migrations fail with `libvirt.libvirtError: internal error: unable to execute QEMU command 'blockdev-add': Certificate does not match the hostname`19:56
noonedeadpunkso /o\19:56
opendevreviewDmitriy Rabotyagov proposed openstack/nova-specs master: [spec] Add Cross-AZ scheduling blueprint  https://review.opendev.org/c/openstack/nova-specs/+/90029619:59
noonedeadpunkah. and that is the case only for local storage, so when `--block-migration` is used19:59
noonedeadpunksorry, fixed in 10.3, not 10.220:01
noonedeadpunkhttps://libvirt.org/news.html#v10-3-0-2024-05-0220:01
noonedeadpunkhttps://github.com/libvirt/libvirt/commit/5d48c5d215071526383b8fc50d81ecde62e4111b20:20

Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!