| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-09-18 | |||
| 08:24:00 | lyarwood | yes correct | |
| 08:24:39 | lyarwood | the train UCA has the new MIN_{LIBVIRT,QEMU}_VERSIONs we are looking to set in the current bump on master | |
| 08:24:59 | lyarwood | as well | |
| 08:26:24 | gibi | this means that we now change what libvirt version we use for testing the upgrade from ussuri | |
| 08:36:22 | lyarwood | gibi: we would also change the version of libvirt installed when testing stable/ussuri as a whole | |
| 08:36:39 | lyarwood | gibi: but tbh I don't think anyone is actually using the default versions installed in bionic | |
| 08:38:59 | bauzas | sorry I was focusing on some downstream bugs | |
| 08:39:03 | bauzas | a tl:dr: please ? | |
| 08:40:47 | lyarwood | bauzas: sean-k-mooney suggested that we use the Train UCA in Victoria so we can land the libvirt version bump *without* hitting the QEMU issue found in focal and the Ussuri UCA. | |
| 08:41:17 | bauzas | lyarwood: for all jobs ? | |
| 08:41:17 | lyarwood | bauzas: This mostly works but we need to also use the Train UCA on stable/ussuri to allow grenade jobs on master to pass | |
| 08:41:26 | lyarwood | bauzas: yes for all jobs | |
| 08:41:35 | bauzas | lyarwood: so it would be a Focal + train UCA for all jobs, right? | |
| 08:41:50 | lyarwood | bauzas: no just bionic | |
| 08:41:58 | lyarwood | bauzas: this doesn't change the focal situation | |
| 08:42:10 | lyarwood | bauzas: it just means we can land the libvirt version bump ahead of the focal change | |
| 08:42:33 | lyarwood | bauzas: so we don't end up landing the version bump after rc | |
| 08:42:35 | bauzas | lyarwood: I raised a concern yesterday about us bumping minimums so close to RC1 | |
| 08:42:50 | bauzas | but I heard people wanting us to bump | |
| 08:43:01 | lyarwood | bauzas: we don't have a choice if the TC are pushing us to move to focal | |
| 08:43:20 | bauzas | well, TC aren't operators | |
| 08:43:25 | bauzas | some of them are | |
| 08:43:48 | bauzas | but I don't want us to hold all our gate so close to RC1 because we found some bugs unexpectedly | |
| 08:44:10 | bauzas | telling "we had to do it because $TC" is a terrible explanation | |
| 08:44:13 | lyarwood | I also hate making changes like this so close to RC tbh | |
| 08:44:25 | lyarwood | but if we don't support bionic | |
| 08:44:33 | bauzas | okay, let's be pragmatic and discuss about the change surface to be as minimal as we can | |
| 08:44:48 | gibi | do we need to change the min libvirt version to be able to switch to Focal? | |
| 08:44:58 | bauzas | there are two different concerns here | |
| 08:45:04 | bauzas | 1/ people want to bump versions | |
| 08:45:10 | bauzas | 2/ TC wants us to move to Focal | |
| 08:45:19 | bauzas | those two concerns aren't interrelated | |
| 08:45:44 | bauzas | (AFAICT) | |
| 08:46:00 | lyarwood | right, we did however say that we would bump the min versions in V | |
| 08:46:08 | lyarwood | and that we would also move the next versions in V | |
| 08:46:46 | lyarwood | gibi: correct we don't need to change the min to move to focal | |
| 08:46:49 | bauzas | bumping minimums is somehow understandable | |
| 08:47:03 | bauzas | more than us moving to Focal, tbh | |
| 08:47:11 | openstackgerrit | Liang Fang proposed openstack/nova master: [WIP] rbd patch for volume local cache https://review.opendev.org/726762 | |
| 08:47:28 | stephenfin | I personally class the bumping of minimums more important and less risky than the focal move | |
| 08:47:28 | bauzas | anyway, I want us to stop focusing on distros, and rather be discussing on QEMU versions | |
| 08:47:32 | lyarwood | tbh I'd personally drop the entire move to focal at this point | |
| 08:47:34 | stephenfin | oh, jinx :) | |
| 08:47:44 | lyarwood | \o/ | |
| 08:48:03 | gibi | droping Focal needs to be agreed with the TC | |
| 08:48:12 | bauzas | stephenfin: yeah I'm kinda leading to this conclusion | |
| 08:48:24 | lyarwood | I'm amazed that we continued with it after M2 tbh | |
| 08:48:31 | lyarwood | move early or don't move at all | |
| 08:48:36 | stephenfin | apologies if this was discussed yesterday, but couldn't we have a single non-voting tempest job on Focal? | |
| 08:48:50 | bauzas | gibi: I don't understand why the OpenStack community makes an hard sell on a OS thing | |
| 08:48:59 | stephenfin | to satisfy that request | |
| 08:49:11 | bauzas | I thought we only considered dependencies and not OSes | |
| 08:49:13 | lyarwood | yeah but I get the feeling that gmann's changes are to the base jobs | |
| 08:49:15 | bauzas | to be frank | |
| 08:49:23 | lyarwood | we could overwrite the node labels in our tree however | |
| 08:49:25 | gibi | bauzas: me neither but I think to make the communication clear we have to be on the same side with the TC | |
| 08:49:37 | bauzas | okay, let's be pragmatic | |
| 08:50:08 | bauzas | the problem is with a specific QEMU version that races on volume attach/detach, right? | |
| 08:50:13 | stephenfin | ah, you mean the parent 'openstack-*' and 'legacy-*' jobs? | |
| 08:50:25 | bauzas | and past this version, QEMU isn't fixed yet | |
| 08:50:30 | bauzas | am i correct? | |
| 08:50:54 | lyarwood | stephenfin: yeah | |
| 08:51:04 | bauzas | so, we can legitamately bump the minimums without being impacted by the QEMU issue | |
| 08:51:11 | bauzas | legitimatly | |
| 08:51:14 | lyarwood | bauzas: that's the issue with focal and the ussuri UCA yes | |
| 08:51:41 | lyarwood | bauzas: but we can only bump the mins *if* we use the train UCA, ussuri UCA or move to focal | |
| 08:51:42 | bauzas | lyarwood: because the ussuri UCA ships a qemu version that's impacted, right? | |
| 08:51:55 | lyarwood | bauzas: correct it's the same versions as shipped in focal | |
| 08:52:08 | lyarwood | bauzas: with the train UCA being a little older and not impacted by the bug | |
| 08:52:11 | bauzas | ok, so the problem is with UCA, not the distro | |
| 08:52:26 | lyarwood | bauzas: well both, it's in the core distro from focal | |
| 08:52:32 | bauzas | ah I see | |
| 08:52:39 | bauzas | gotcha | |
| 08:52:41 | bauzas | so | |
| 08:52:49 | bauzas | I'm all documenting this | |
| 08:53:07 | gibi | buming minimum means we change the libvirt version on stable/ussuri too | |
| 08:53:13 | gibi | is it OK? | |
| 08:53:18 | bauzas | gibi: why ? | |
| 08:53:22 | gibi | did we do that before in other stable branches? | |
| 08:53:30 | lyarwood | gibi: yeah we've done that in the past | |
| 08:53:42 | stephenfin | I guess we'll change the version tested if we backport, yes. We won't change the versions allowed though | |
| 08:53:43 | lyarwood | bauzas: grenade | |
| 08:53:45 | gibi | lyarwood: so it is sort of a normal thing at a min bump? | |
| 08:53:48 | brinzhang_ | gibi: IMO, in V release is enough | |
| 08:54:03 | lyarwood | gibi: iirc the last time we did this was due to QEMU/libvirt bugs | |
| 08:54:16 | bauzas | lyarwood: what? | |
| 08:54:24 | lyarwood | gibi: so we enabled the UCA to workaround these and had to do it on an older stable release for grenade as well | |
| 08:54:31 | bauzas | lyarwood: I thought grenade was upgrading on the target | |
| 08:54:42 | brinzhang_ | Cyborg done this today https://review.opendev.org/#/q/project:openstack/cyborg+branch:master+topic:migrate-to-focal | |
| 08:54:43 | lyarwood | bauzas: it does but it doesn't rerun devstack | |
| 08:54:54 | brinzhang_ | bump to the min version | |
| 08:55:06 | lyarwood | bauzas: so we'd need to use the train UCA on stable/ussuri to ensure n-cpu from master would work | |
| 08:55:23 | lyarwood | stephenfin: correct, it's just the tested version in stable/ussuri being changed | |
| 08:55:42 | lyarwood | and again I don't think anyone is using the min versions of libvirt and QEMU at the moment | |
| 08:56:06 | lyarwood | if you're using Ubuntu you're using the UCAs | |
| 08:56:20 | lyarwood | if you're using el distro you;re already way past these versions on stable/ussuri | |
| 08:56:56 | bauzas | ok, either way | |
| 08:57:04 | bauzas | if we merge the bumps | |
| 08:57:28 | bauzas | are we sure that the new minimums are satisfied by the train UCA ? | |
| 08:57:40 | bauzas | because | |