Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2026-05-11
17:29:18 womax yes that's it
17:29:23 dansmith gotcha
17:37:03 womax and linked to that, I tried to experiement a bit with the download handler code path, to see what the implmemention could look like. And I was under the impression that it cannot works if glance show_multiple_locations config is false (which is the default (https://docs.openstack.org/glance/latest/configuration/glance_api.html#DEFAULT.show_multiple_locations), as nova never seems to call the dedicated route to get locations
17:37:04 womax (/v2/images/{image_id}/locations), (https://opendev.org/openstack/nova/src/branch/master/nova/image/glance.py#L253). I don't know if it normal/bug, or if I missed something else, but I could never make it even try to see if download_handlers existed without a bit of modification in the nova code
#openstack-nova - 2026-05-12
07:42:34 gibi Hi folks it seems nova-alt-configuration job is started to fail since yesterday
07:42:38 gibi https://zuul.opendev.org/t/openstack/builds?job_name=nova-alt-configurations&branch=master&skip=0
07:43:10 gibi May 11 20:41:52.111833 npfa8fa206b4014 nova-compute[32688]: ERROR nova.compute.manager [instance: 0de2d300-db70-420c-a502-c6015ad70e94] libvirt.libvirtError: unsupported configuration: spice graphics are not supported with this QEMU
07:43:14 gibi May 11 20:41:52.111833 npfa8fa206b4014 nova-compute[32688]: ERROR nova.compute.manager [instance: 0de2d300-db70-420c-a502-c6015ad70e94] libvirt.libvirtError: unsupported configuration: spice graphics are not supported with this QEMU
07:43:19 gibi https://zuul.opendev.org/t/openstack/build/7f08461295e24d3b8f1673f29dd9f711/log/compute1/logs/screen-n-cpu.txt?severity=4
07:45:43 gibi this can be relevant https://review.opendev.org/c/openstack/devstack/+/968354
07:49:57 frickler gibi: iiuc you need to install an extra pkg for that, see also https://review.opendev.org/c/openstack/kolla/+/972441/4/docker/nova/nova-libvirt/Dockerfile.j2
07:52:26 gibi frickler: thanks for the pointer
07:56:03 gibi I will push a fix in devstack then
07:57:42 frickler looks like the failure was already present in the testing patch, but wasn't noticed https://review.opendev.org/c/openstack/nova/+/968892
08:06:15 gibi indeed
08:06:39 gibi it is partially me who approved the devstack patch yesterday who missed that failure.
08:07:09 gibi I pushed https://review.opendev.org/c/openstack/devstack/+/988214 to fix it
08:09:06 opendevreview Balazs Gibizer proposed openstack/nova master: DNM: test nova-alt-configuration https://review.opendev.org/c/openstack/nova/+/988215
08:09:15 gibi testing it in ^^
08:15:43 mikal gibi: yeah, the Debian Trixie qemu requires an additional package for SPICE. Same deal with Ubuntu FWIW.
08:34:24 gibi mikal: TIL :)
08:34:32 gibi elodilles: for your eyes https://review.opendev.org/q/I469688be203c319dd048d6a99057b8d98369de0e
08:42:20 elodilles gibi: ACK, i'll keep an eye on them
09:16:50 sean-k-mooney gibi: am i belive speice is an optional package on debian 13 and 12 + backport likely has the same pattern
09:17:08 sean-k-mooney ah yep
09:17:25 sean-k-mooney gibi: mikal so i have been meaning to move that job to debian 13 for a while now
09:21:22 mikal sean-k-mooney: yeah fair enough. I am working on a job for Kolla-Ansible at the moment, but that's likely heavier weight than what you'd want in Nova.
09:22:05 sean-k-mooney https://review.opendev.org/c/openstack/nova/+/966479
09:22:44 sean-k-mooney was the change i worked on 6 months ago i just didnt get around to the followup for addign the spice module expelcitly
09:22:52 sean-k-mooney at least not that isee upstream
09:23:45 mikal I'd like to do stronger assertions that SPICE is actually working, not just getting the server "banner" back, but that's a thing I've wanted for a while and am only just getting to now.
09:24:08 sean-k-mooney mikal: i mena we could do that
09:24:28 sean-k-mooney it really just depends on what that looks liek and how much time it addes to the job
09:25:05 mikal Well, I have most of a headless SPICE client working now. I want to have a bit more a test workload which enumerates a bit of the protocol as well though. I don't think it would make the job much slower, apart from needing to download (or cache) the client and target instance image.
09:25:14 mikal But forward looking statements etc etc.
09:26:12 sean-k-mooney well if you end up packagign this as a tempest/devstack plugin that we coudl teadilly enable we coudl consider addign it
09:27:04 sean-k-mooney s/teadilly/readally or eaisilly/?
09:27:11 sean-k-mooney oen of those words
09:29:07 mikal Just as soon as I learn what a tempest plugin (or Claude tells me), I will consider that thing.
09:29:28 sean-k-mooney its just a way of packaging tempst test in a seperate repo
09:29:55 sean-k-mooney as an example https://opendev.org/openstack/whitebox-tempest-plugin/
09:30:10 mikal Huh, so potentially sharable between Nova and Kolla-Ansible? That would be nice.
09:30:24 sean-k-mooney yes
09:30:39 mikal Ok, I shall read your newsletter...
09:30:41 sean-k-mooney and if you do that you could provide a devstack plugin to just set up the deps/generate the config
09:30:56 sean-k-mooney https://docs.openstack.org/tempest/latest/plugins/plugin.html is proably more useful
09:31:54 sean-k-mooney of course tempest (which is just stdlib unittest as the framework) may not be what you want fo rthis but i think it could be a decent fit
09:41:23 mikal Ok, I will consider this all tomorrow but it sounds promising to me.
09:41:25 mikal Thank you.
09:42:23 sean-k-mooney if gibi has time to update teh devstack patch to make it work on debian 13 we will likely proceed with it today, if they dont we may still proceed with it today to unblock the gate and follwo up next week
09:47:56 mikal Oh definitely adding the package now is the way to go. My thing will take a while to navigate I think. I'd offer to send a change for adding the patch, but I am also about to be pulled away for family duties.
09:48:41 sean-k-mooney im 99% sure i have a patch also locally but the host i run my devstack vms lost a hardrive at teh weekend
09:48:54 sean-k-mooney and while it can run with one failure i have it off until i have time to replace it
09:49:17 sean-k-mooney so my debian 13 spice enve where i was testing that fix is ofline :)
09:49:45 sean-k-mooney gibi's patch shoudl fix the current case and it would be a minor tweak to also supprot debain 13
10:50:04 frickler gibi: it there anything that can be done to make progress with https://bugs.launchpad.net/os-vif/+bug/2109396 ? there was a new pyroute2 release recently, but from the changelog it didn't look to me like it would help with this issue. the pyroute2 cap is the last remaining item we have until we can go back to having "normal" u-c updates again
10:58:11 opendevreview Balazs Gibizer proposed openstack/nova master: move nova-alt-config to debian-13 https://review.opendev.org/c/openstack/nova/+/966479
10:59:33 gibi I respin the devstack fix to include debian 13 and updated ^^ to point to that devstack change
11:00:08 gibi frickler: somebody should either fix pyroute2 with eventlet or we should drop eventlet from our codebase
11:00:15 gibi I have no other ideas
11:00:30 sean-k-mooney if only one of those was almost complete
11:01:11 sean-k-mooney frickler: i assume its not posisbel to have diffent version depending on if your using eventlet or not like we can based on python verion?
11:02:17 sean-k-mooney nova i now runign nova-compute without eventlet by default so could perhaps downgrade the pyroute2 used in eventlet mode and uncap it in threaded mode?
11:02:33 sean-k-mooney but that likely to requrie some toolign changes in devstack to do properly
11:03:09 gibi hm if we allow downgrading to our small amount of eventlet based jobs then the rest can probably run with the latest pyroute2
11:03:19 gibi s/to/in/
11:03:27 frickler how long will the eventlet mode be needed? 2027.1 for migration purposes?
11:03:44 gibi yes 2027.1 will still have eventlet as a last slurp
11:04:23 sean-k-mooney and we will be deleting it in about 12 months time basiclly we will do the deletion early in 2027.2
11:04:44 frickler we could override u-c within devstack, but that won't help deployers
11:05:10 sean-k-mooney 2026.2 is the first release where we are defaultign to threaded mode for nbova-compute
11:05:34 frickler so let me update https://review.opendev.org/c/openstack/requirements/+/973210 and then we can see where we actually stand currently
11:05:41 sean-k-mooney so this release would be compatibaly with newer pyroute2
11:07:00 sean-k-mooney frickler: we could also just say "if you want to use py3.14 you shoudl use threaded mode adn only keep the cap for 3.13 and below
11:07:35 sean-k-mooney but cool lets see what the updated run shows
11:14:00 opendevreview Dr. Jens Harbott proposed openstack/nova master: DNM: test pyroute2 bump https://review.opendev.org/c/openstack/nova/+/988242
11:20:11 gibi the nova-alt-configuration is the job that tests the eventlet mode. So we need a good result from that job in ^^ But that job is broken and being fixed in devstack :)
11:28:24 frickler ah, I wasn't aware that that is the same job. so waiting for it to be fixed first
11:30:21 gibi yeah I relized as we are chatting here :)
13:08:47 sean-k-mooney gibi: so it failed but on an undrealted test the spice tests passed but some fo the live migraiton test failed in test cleanup
13:13:20 gibi I will recheck the nova patch but I think the devstack one is good to go
13:13:32 sean-k-mooney yep i +2'd the devstack one
13:14:12 sean-k-mooney did you have a test patch on deb-12 on nova? i just check the patch that moved nova-alt-config to debin 13
13:15:21 gibi sean-k-mooney: this https://review.opendev.org/c/openstack/nova/+/988215 shows that the devstack patch unblock the nova-alt-config job
13:15:31 gibi with debian 12
13:15:54 sean-k-mooney great
13:16:18 sean-k-mooney then ya we shoudl proceed with the devstack patch is frickler and other are happy with it
13:16:52 sean-k-mooney as i said the test failure looked unrelated on debian 13
13:19:56 sean-k-mooney gibi: while i have been in meeting i have been having ai agents test my kernel module and it apears to be workign enough for the device to show up in lspci
13:20:25 sean-k-mooney im going to play around with it a bit more and see if i can get it to a porint where the VF are usabel by a nested qemu quickly
13:21:14 gibi sean-k-mooney: thanks
13:22:58 elodilles sean-k-mooney: sorry for pinging, can you please revisit my placement gatefix patch / check my comment on it? o:) https://review.opendev.org/c/openstack/placement/+/980761/2#message-6b13614a07d353e22186487083fe57b46b3aa02f
13:25:32 sean-k-mooney elodilles if the grenade job is working im kind of ok with it existing
13:25:43 sean-k-mooney but i dont really like having a know broken job
13:25:46 sean-k-mooney that is non voting
13:26:29 sean-k-mooney i would prefer if the skip level job was removed until it was fixed but if other agree then we can keep it
13:27:17 elodilles sean-k-mooney: yes, i usually remove as well broken / unfixable CI jobs, but with this i feel some hope o:) (also, it's included via a job template)
13:30:11 sean-k-mooney ill merge it for now and we can discuss it more once the cid is unblocked
13:31:52 elodilles sean-k-mooney: sure, thanks \o/
13:32:01 opendevreview ribaudr proposed openstack/nova master: Add reproducer for cold migration stuck after compute restart https://review.opendev.org/c/openstack/nova/+/988257

Earlier   Later