16:00:11 <gibi> #startmeeting nova 16:00:12 <openstack> Meeting started Thu Apr 16 16:00:11 2020 UTC and is due to finish in 60 minutes. The chair is gibi. Information about MeetBot at http://wiki.debian.org/MeetBot. 16:00:13 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 16:00:15 <openstack> The meeting name has been set to 'nova' 16:00:32 <gibi> o/ 16:00:39 <artom> o/ 16:00:46 <melwitt> o/ 16:00:56 <alex_xu> o/ 16:00:57 <dansmith> o/ 16:01:36 <bauzas> \o 16:02:02 <gibi> #topic Last meeting 16:02:08 <gibi> #link Minutes from last meeting: http://eavesdrop.openstack.org/meetings/nova/2020/nova.2020-04-09-16.00.log.html 16:02:17 <gibi> Feature Freeze happened last week. We finished 18 blueprints, 1 got FFE, 11 bp got deferred 16:02:34 <gibi> novaclient has been released for Ussuri 16:02:59 <gibi> anything to mention from the last meeting? 16:03:44 <gibi> #topic Release Planning 16:03:53 <gibi> (will go back to bugs in after this) 16:03:59 <gibi> TODOs are tracked in the etherpad #link https://etherpad.opendev.org/p/nova-ussuri-rc-potential 16:04:05 <lyarwood> \o 16:04:08 <gibi> We have one feature open: #link https://review.opendev.org/#/q/topic:bp/policy-defaults-refresh+status:open 16:04:17 <gibi> last week we agreed to finish it until end of this week 16:04:23 <gibi> gmann: where we are with this? 16:04:39 <gmann> sorry forgot about meeting 16:04:42 <gibi> I saw that things are moving but I don't see how much is still needed 16:05:03 <gmann> all the patches need for ussuri are up, few are stuck on gate since yesterday 16:05:10 <gmann> I am trying to get them merged soon 16:05:29 <gmann> doc and release patch is also up which i need to re-work for melwitt comments 16:05:35 <gmann> but that will be last one to merge 16:05:47 <gibi> gmann: cool. so we still has some patches to review but we don't expect new ones 16:06:10 <gmann> yeah most of them are updated for review comment so will be quick to do 16:06:19 <gibi> let's try to approve all the open patches until Friday EOB 16:06:30 <dansmith> I've been trying to grok this whole effort this morning and I'm definitely missing something about what deployers are going to have to do here 16:07:01 <dansmith> hopefully once I figure that out it'll just be some revisions to the spec (or maybe they're already covered in the docs patch, which I haven't read yet) 16:07:40 <gmann> dansmith: i am trying to capture those in this doc with migration plan - https://review.opendev.org/#/c/720129/4/doc/source/configuration/policy-new-defaults.rst 16:07:51 <gibi> as far as I understand the feaure is turned off by default 16:08:02 <dansmith> yeah, I need to read that and see if it clears things up for me 16:08:13 <gmann> yeah it is off by default and we have flag to turn it on without need of overwriting the policy file 16:08:30 <dansmith> gibi: well, I'm more wondering about what we're going to be requiring admins to do at some point before we can complete this migration, not so much the impact of the current ones 16:08:43 <dansmith> gmann: yep, understand that 16:08:55 <melwitt> gmann: I added new comments to the doc patch a little while ago. need to clearly explain whether operators will need to add new keystone roles to use scope types and also explain how end users will need to know how to request scoped tokens from keystone and point to keystone docs for that 16:09:36 <gmann> melwitt: ok, i think pointed to keystone doc but i will check, i agree that is imp 16:09:50 <gibi> dansmith: I see, so what the admins need to do before they can upgrade to V where we remove the old compat code 16:10:13 <melwitt> gmann: yeah, if it is, I missed it. would like to see it very concisely and clearly stated what are the action items for operators 16:10:18 <dansmith> gibi: right, because it sounds like they might have to adjust all their users' roles or something, but again I don't fully grok it 16:10:38 <gmann> melwitt: sure, I will check the comments. 16:10:38 <dansmith> gibi: and since more people are skipping releases these days, 16:10:58 <dansmith> we need to make sure we're not expecting them to have to do somethign where they have to turn the cloud off, migrate a bunch of users, and then turn it back on 16:11:17 <gibi> dansmith: worst case we cannot remove the compat code in V 16:11:21 <dansmith> or worse, some problem where they do that and it affects a user in another service because they have to for nova, or something 16:11:38 <dansmith> gibi: well, worst case we do all this and then never remove it, 16:11:40 <gmann> we are keeping compat code till W. for 2 cycle as this is big change for migration 16:11:47 <dansmith> and then there's two ways, two sets of docs, and confusion :) 16:11:56 <gibi> dansmith: like notifications :) 16:12:02 <dansmith> gibi: yep 16:12:38 <gibi> it is 95% merged so merging the last 5% does not change much for us now 16:12:52 <gibi> but I agree that we need a comprehensive doc 16:13:48 <gibi> if it turns out that it is hard to migrate to the new policies then we have to figure out a way to help 16:14:15 <gibi> but I think not merging the last 5% does not help much with that 16:14:33 <dansmith> I'm not arguing against merging it, 16:14:37 <gmann> i think at some point keystone is also going to remove the support of old role in their policy. 16:15:00 <dansmith> I'm just saying I'm *personally* missing context here, and I hope I'm in the minority, and we need to make sure we've got our attack plan well thought-out 16:15:35 <dansmith> personally missing the big picture and the end game, I mean 16:16:09 <gibi> OK lets work on those attack plans together if it turns out something is missing 16:16:12 <gibi> can we move on? 16:16:25 <dansmith> surely 16:16:37 <gibi> cool 16:16:39 <gibi> Next week (Apr 23) we have to cut RC1 16:16:49 <gibi> please tag any RC critical bug with ussuri-rc-potential tag in launchpad 16:16:56 <gibi> bauzas volunteered to write the release notes prelude. It is due before RC1 16:17:04 <gibi> I will propose an patch with an RPC alias for Ussuri 16:17:10 <gibi> Do we want to make a major compute RPC version bump? 16:17:12 <bauzas> thanks for the reminder 16:17:44 <bauzas> gibi: we generally made such bumps, nope ? 16:18:02 <bauzas> #link https://wiki.openstack.org/wiki/Nova/ReleaseChecklist 16:18:02 <gibi> bauzas: it was a loong time when we bumped to 5.x in compute RPC 16:18:15 <bauzas> I know 16:18:17 <melwitt> there's a patch proposed already for rpc alias fyi https://review.opendev.org/719315 16:18:23 <bauzas> I only remember the old ages 16:18:41 <gibi> https://github.com/openstack/nova/blob/master/nova/compute/rpcapi.py#L392 16:18:42 <sean-k-mooney> gibi: as an fyi ill try and rebase https://review.opendev.org/#/c/706276/ today and after rc1 ill create a similar patch to updated the implemented specs for ussuri 16:19:06 <sean-k-mooney> we forgot to do it last cycle. 16:19:11 <gibi> melwitt: thanks. I did not notice that. 16:19:13 <dansmith> gibi: an rpc bump at this point would likely be difficult 16:19:28 <dansmith> we need to plan that a little further in advance, IMHO 16:19:55 <gibi> dansmith: OK then this answers my question. 16:20:29 <gibi> sean-k-mooney: thanks 16:21:05 <gibi> #info we won't bump compute RPC to 6.0 in Ussuri 16:21:22 <gibi> dansmith: for the future, how much time would need to do such bump? 16:21:56 <artom> Also, and it sucks to leak "downstream" stuff like this, but don't we have to think about what releases operators/vendors are skipping, and make sure to *not* major-bump RPC in those? 16:22:23 <dansmith> artom: no 16:22:30 <sean-k-mooney> we can pin the rpc version so no 16:22:31 <dansmith> gibi: depends on the delta really 16:22:46 <dansmith> artom: people skipping versions can't use rpc compatibility anyway 16:22:52 <artom> dansmith, so in theory we could bump to 6.0 in U, and not break OSP 16 -> 17 FFWD upgrade? 16:23:06 <artom> Oh right, the control plane is down anyways 16:23:10 <dansmith> artom: rpc versions have no impact on an FFU 16:23:21 <dansmith> and even still OSP 16 is T, so no either :) 16:23:23 <artom> dansmith, yep, thanks for setting me straight 16:24:25 <gibi> dansmith: OK, I will add some note to the V release schedule to plan a such bump earlier, maybe around M3 16:24:33 <sean-k-mooney> too gibi question i assume we can move to 6.0 in train but likely would not want to do a major version bump after m3 16:24:43 <sean-k-mooney> * in Victoria 16:25:40 <gibi> sean-k-mooney: so it is need to be planned even eariler? 16:26:02 <dansmith> gibi: I'll have to look at the current delta, but.. is there anything begging for a bump? 16:26:08 <dansmith> I mean, the only reason to do it is really if we desperately need to drop something, 16:26:17 <dansmith> but if not it's just perceptual purity, which isn't a good reason 16:26:29 <bauzas> fwiw https://github.com/openstack/nova/blob/master/nova/compute/rpcapi.py#L372-L379 16:26:31 <bauzas> dansmith: ^ 16:26:54 <gibi> dansmith: I have to look at the delta too but I vaguely remember discussing the 6.0 bump for reasons 16:27:08 <dansmith> bauzas: right, none of those are things we need to drop 16:27:26 <bauzas> dansmith: I only see additive changes between 5.1 and now 16:27:28 <dansmith> bauzas: that's my point.. a bunch of additions will be identical in 6.0 16:27:32 <dansmith> right 16:27:43 <bauzas> actually, 5.1 could be debatable 16:27:53 <melwitt> yeah so would more be looking at how many things have notes like "remove this in the next major bump" 16:27:57 <bauzas> dansmith: I agree 16:28:09 <bauzas> except 5.1 maybe 16:28:09 <dansmith> melwitt: not really, those are often no-ops anyway 16:28:19 <dansmith> melwitt: more like thing like 5.1, but even that is just uuber minor, IMHO 16:28:25 <bauzas> right 16:28:35 <melwitt> I guess I thought eventually would be looking to clean them up 16:28:36 <dansmith> I mean things like "if <5.5, then do an extra DB lookup to reconstruct some information we're missing" 16:28:40 <bauzas> now we objectified all rpc apis, changes are mostly additive 16:28:56 <melwitt> oh 16:28:58 <dansmith> melwitt: right, that's what I'm saying.. just cleanups aren't worth it.. things that actually have an impact are why we need to 16:29:17 <melwitt> ok 16:29:24 <gibi> OK. I made a note to myself to double check it but let's move on 16:29:26 <bauzas> moving on ? :) 16:29:28 <bauzas> heh 16:29:43 <gibi> anything else for the release planning before we jump back to bugs? 16:29:44 <dansmith> moving....on. 16:30:07 <gibi> #topic Bugs (stuck/critical) 16:30:13 <gibi> No critical 16:30:24 <gibi> #link 71 new untriaged bugs (-40 since the last meeting): https://bugs.launchpad.net/nova/+bugs?search=Search&field.status=New 16:30:36 <gibi> good progress to eating our backlog 16:30:47 <gibi> but we keep continue as RC1 is close 16:30:53 <gibi> RC critical bugs are tagged with ussuri-rc-potential #link https://bugs.launchpad.net/nova/+bugs?field.tag=ussuri-rc-potential 16:31:17 <gibi> there is now 1 bug on that list 16:31:35 <gibi> it has a patch up and I gave review feedback 16:31:50 <gibi> I'm not 100% sure it is that critical though 16:31:57 <bauzas> not RC holding 16:32:25 <bauzas> i mean it's not a U regression 16:32:37 <bauzas> but I thought it was worth looking at it before we branch 16:33:07 <gibi> sure 16:33:27 <gibi> somebody added an bug to the agenad 16:33:36 <gibi> how to solve the bug ? #link https://bugs.launchpad.net/nova/+bug/1841932 Could we add compat for os:hide_hypervisor_id and reference that in docs and code?so that we can use AggregateInstanceExtraSpecsFilter and hide_hypervisor_id at the same time. 16:33:36 <openstack> Launchpad bug 1841932 in OpenStack Compute (nova) "hide_hypervisor_id extra_specs in nova flavor cannot pass AggregateInstanceExtraSpecsFilter" [Undecided,Expired] 16:33:57 <bauzas> leave it for open discussion ? 16:34:22 <gibi> OK, I hope the person added it to the agenda will show up until then 16:34:36 <gibi> any other bug related business? 16:34:41 <sean-k-mooney> so the problem with that extra spec is we did not namespace it 16:34:57 <sean-k-mooney> so it breaks the AggregateInstanceExtraSpecsFilter 16:35:17 <gibi> OK so it is not a regression at least 16:35:23 <sean-k-mooney> no 16:35:25 <gibi> good 16:35:27 <gibi> moving on 16:35:32 <tframbo> yeah ,i added the bug 16:35:48 <gibi> tframbo: let's discuss it a bit at the end of the meeitng 16:35:51 <gibi> #topic Stable Branches 16:35:53 <bauzas> the whole hide_hypervisor_id thing is... hacky :p 16:35:56 <tframbo> ok 16:36:00 <gibi> lyarwood any news? 16:36:16 <gibi> we discussed a need for a train release last week 16:36:29 <bauzas> tl;dr: shhhhht, QEMU, don't tell the nvidia driver I don't have the right to use you 16:36:55 <melwitt> stable train and stein were recently broken bc placement started requiring python 3.6 and our func test stuff in tox.ini was pulling placement from master, 16:37:11 <melwitt> train is fixed and stein fix is en route to the gate 16:37:18 <gmann> #link https://review.opendev.org/#/c/719121/ 16:37:23 <gibi> melwitt: thanks 16:37:35 <gmann> hope it get node in gate 16:37:58 <melwitt> yeah there might be some gate issue, I've been discussing in -infra. might be a nova bug even. yay! (not) 16:38:18 <gibi> it seems lyarwood is not here I will ping him about the stable release tomorrow 16:38:20 <lyarwood> yeah sorry ^ I've also hit a few other issues slowing things down 16:38:59 <lyarwood> we should still be able to release before RC (iirc that was the plan) 16:39:21 <gibi> lyarwood: cool thanks 16:39:37 <gibi> #topic Sub/related team Highlights 16:39:50 <gibi> API (gmann) 16:39:56 <gibi> anything besides the policy work? 16:40:17 <gmann> no, all what we discussed earlier . 16:40:26 <gibi> cool 16:40:27 <gibi> moving on 16:40:34 <gibi> #topic Stuck Reviews 16:40:39 <gibi> (bauzas) https://review.opendev.org/#/c/589085/8 (Merging virt API changes after FeatureFreeze) 16:41:00 <bauzas> nothing crucial here but I think we need a consensus on things we can accept 16:41:24 <bauzas> tl;dr: I'm touching the virt API signature of revert_resize and resize') 16:41:25 <gibi> we need to choose between allowing the virt driver interface to change post FF and allow the same interfacechange on stable/ussuri branch later 16:41:38 <bauzas> both are internal APIs but are in use by out-of-tree drivers 16:41:57 <bauzas> so for convenience (since virt APIs aren't public), we notify them 16:42:06 <bauzas> there are two options 16:42:20 <bauzas> 1/ we consider it's a breaking change hence not acceptable post FF 16:42:37 <bauzas> and in this case, we shouldn't allow this to be backportable 16:43:02 <bauzas> 2/ we assume this isn't a breaking change and in this case, we also consider it as acceptable post-FF 16:43:18 <gibi> just looking at the interface change patch it feels lower risk to do it now than to do it on the stable branch 16:43:26 <bauzas> and the corollar being, if this punts to V, this can be backportable to U 16:43:35 <bauzas> so, 1/ or 2/ ? 16:43:46 <gibi> 2/ 16:44:07 <bauzas> I thought we had disagreements on such things 16:44:17 <bauzas> I won't vote on it, since I'm involved in the change 16:44:20 <melwitt> and there is no desire to backport this earlier than ussuri right? I see that the bug was reported a long time ago 16:44:30 <bauzas> (but you get my opinion by seeing the change) 16:44:45 <bauzas> melwitt: I honestly have no energy for it 16:45:03 <bauzas> or, in other words, this wouldn't be backported upstream 16:45:15 <stephenfin> yeah, maybe downstream. Can't see us doing that upstream, tbh 16:45:23 <melwitt> ok. well, the change looks simple at least. so seems low risk 16:45:37 <bauzas> melwitt: but your question remains, if we go with 2/, in theory we could accept virt API changes to be backportable down to T or whatever 16:46:22 <bauzas> or we just say "no policy, it's case-by-case" and we move on 16:46:56 <gibi> if I were a out of tree virt driver maintainer I would expect changes on that interface to happen more at major version than at minor version 16:47:18 <stephenfin> that's where I'm at too ^ 16:47:20 <melwitt> yeah, agree with that 16:47:37 <gibi> (for the record I'm not maintainting out of tree drivers) 16:47:56 <gibi> so we agree to change the interface then do it where it is expected to change 16:48:07 <stephenfin> but, with that said, if this was a more critical issue with a move operation that affected more people, I imagine out-of-tree drivers would have to deal with it 16:48:12 <sean-k-mooney> stephenfin: i dont really want use to do it downstream either 16:48:43 <bauzas> stephenfin: out-of-tree driver maintainers would have to maintain compatibility with the backport, whatever the use is 16:48:58 <bauzas> it's not about the use of the parameter 16:49:17 <bauzas> it's just that if we backport it to Train, Stein or Rocky, then we break them 16:49:27 <bauzas> (for resize and revert resize here) 16:49:28 <sean-k-mooney> bauzas: unless you catch the type error and call the function again without the parmater which we have dont in the past 16:49:30 <gibi> dansmith: do you have a hard NO against merging the interface change now? (I'm asking you as you expressed a preference to not do it in U) 16:49:49 <sean-k-mooney> that would support out of tree driver but i dont think we shoudl have to do that 16:49:58 <bauzas> we can honestly defer it to V if we are able to merge it sooner than later 16:50:06 <bauzas> sean-k-mooney: I'm opposed to it 16:50:22 <bauzas> sean-k-mooney: there is one thing to notify and have a gentleman agrement 16:50:39 <bauzas> there is another thing to implement some kind of public and maintained API for them 16:50:48 <stephenfin> yeah, that would imply actual support 16:51:15 <bauzas> okay, so, I don't hear any disagreement on merging it post-FF if we can 16:51:21 <bauzas> and if we can't, no backport to U 16:51:27 <bauzas> I'm cool with both things 16:51:36 <sean-k-mooney> i know im jsut saying what we have done in the past. im pretty sure we did this recently for one of the migration specs. my preference is to defer to V 16:51:54 <sean-k-mooney> with no backport 16:51:57 <stephenfin> Cool. I'll take another look at that in the morning and send it through so 16:52:07 <bauzas> let's just leave a chance of this thing to be reviewed during this RC period and if we are not able to merge it, merge it in V 16:52:22 <bauzas> stephenfin: the last patch needs a revision, hold your pen 16:52:37 <gibi> #info give a chance to https://review.opendev.org/#/c/589085/ to merge before RC1 16:52:43 <bauzas> I think we're mostly on agreement, let's move on 16:52:59 <bauzas> gibi: also, the consensus is "don't backport any virt API changes" 16:53:09 <bauzas> which is a solid argument 16:53:33 <gibi> let's discuss that part of the agreemet at another time. 16:53:38 <bauzas> all good, thanks folks 16:53:42 <gibi> moveing on 16:53:43 <gibi> #topic Open discussion 16:53:52 <gibi> (gibi): Virtual PTG planning. #link http://lists.openstack.org/pipermail/openstack-discuss/2020-April/014148.html 16:54:01 <gibi> I need to book slots in the common schedule so we need to know how much slots we need, and what time slots we want to use. 16:54:13 <gibi> doodle link is in the mail 16:54:47 <gibi> and if you have any feeling about the amount of time we need the please ping me on #openstack-nova later 16:55:23 <gibi> tframbo: your turn 16:55:31 <gibi> #link https://bugs.launchpad.net/nova/+bug/1841932 16:55:31 <openstack> Launchpad bug 1841932 in OpenStack Compute (nova) "hide_hypervisor_id extra_specs in nova flavor cannot pass AggregateInstanceExtraSpecsFilter" [Undecided,Expired] 16:55:35 <bauzas> I'm sad the Foundation only provided a week slot 16:55:40 <bauzas> but meh 16:55:47 <tframbo> Maybe we could add compat for "os:hide_hypervisor_id” like other extra specs and reference that in docs and code to solve it? 16:56:20 <sean-k-mooney> tframbo: we coudl specical case for that yes for backport only 16:56:39 <bauzas> tframbo: can't you just use aggregates with random keys ? 16:56:41 <sean-k-mooney> but i think we shoudl be removing support for unnamespaced extra specs form the filter 16:57:23 <bauzas> I'm personnally opposed to expose any single libvirt knob as an image metadata property 16:57:47 <sean-k-mooney> bauzas: the image metadata is how we are ment to expose them 16:57:55 <sean-k-mooney> its why this feature originally only supported that 16:58:03 <sean-k-mooney> and not extra specs 16:58:13 <bauzas> sean-k-mooney: no 16:58:13 <sean-k-mooney> the extra sepcs supprot was added after 16:58:25 <bauzas> you can achieve the same thing without automatically exposing your capabilities 16:58:41 <bauzas> we have traits, we have filters that do this for you 16:58:55 <sean-k-mooney> this change how the vm is virutalsed 16:58:59 <bauzas> I don't see the need to add yet another knob 16:59:02 <sean-k-mooney> so trait wont help 16:59:11 <gibi> we are running out of time 16:59:13 <bauzas> that's a scheduling problem 16:59:18 <bauzas> not a boot issue 16:59:33 <sean-k-mooney> lets talk about this on #openstack-nova 16:59:35 <gibi> tframbo: I see that bauzas and sean-k-mooney has oppinion about the bugfix so let's talk to them on #openstack-nova 16:59:36 <bauzas> yeah, agreed, let's punt this to -nova 16:59:49 <gibi> thanks everyone 16:59:56 <tframbo> ok 17:00:00 <sean-k-mooney> o/ 17:00:01 <gibi> #endmeeting