15:02:01 <ihrachys> #startmeeting neutron_upgrades 15:02:02 <openstack> Meeting started Mon Jan 30 15:02:01 2017 UTC and is due to finish in 60 minutes. The chair is ihrachys. Information about MeetBot at http://wiki.debian.org/MeetBot. 15:02:03 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 15:02:05 <openstack> The meeting name has been set to 'neutron_upgrades' 15:02:15 <korzen> hello 15:02:16 <sindhu> hi 15:02:16 <ihrachys> hello everyone 15:02:50 <ihrachys> #link https://wiki.openstack.org/wiki/Meetings/Neutron-Upgrades-Subteam Agenda 15:03:03 <ihrachys> #topic Announcements 15:03:23 <ihrachys> Ocata-3 released the previous week, so we are in freeze mode 15:03:33 <ihrachys> meaning, we don't land non-critical bits 15:04:02 <ihrachys> from those that we care about, I think only https://review.openstack.org/#/c/407868/ has a chance for exception 15:04:11 <ihrachys> others will need to wait for Pike to open 15:04:19 <ihrachys> which should happen when rc1 is cut off master 15:04:22 <ihrachys> which is this week 15:05:01 <ihrachys> any other patches that we think are critical to merge before rc1 cut off? 15:05:33 * electrocucaracha is thinking 15:05:35 <korzen> no, I guess that port binding adoption patch is the one 15:06:03 <korzen> I'm working on it, but need more time to adopt to extended PortBinding Db model 15:06:26 <korzen> I can say what is wrong on next section 15:06:30 <ihrachys> yeah I see all test fail. is it some new development in code base that triggered that? 15:06:39 <ihrachys> ok let's do in next section 15:06:49 <ihrachys> #topic Partial Multinode Grenade 15:07:01 <ihrachys> korzen: did you mean this section, or another one? 15:07:16 <korzen> another 15:07:17 <korzen> ;) 15:08:07 <ihrachys> ok, re partial 15:08:22 <ihrachys> I was going to report bugs for linuxbridge job. I haven't :-o 15:08:27 <ihrachys> I will follow up after the meeting 15:08:50 <ihrachys> as for another matter of business, multiple api servers, have you korzen set the env you planned to? 15:09:07 <korzen> sadly no 15:09:15 <korzen> I'm targeting to test in this week 15:09:26 <ihrachys> ok we both slag, good :) 15:09:48 <ihrachys> slag, is it a word? nevermind 15:10:09 <ihrachys> #topic Object implementation 15:10:19 <ihrachys> korzen: your stage 15:10:46 <korzen> ok, so the port binding table was extended with status field and 'host' was added to primary keys 15:11:12 <korzen> and the port binding adoption patch was not adjusted to have host as primary key 15:11:30 <korzen> one issue is to add host in get_object 15:11:56 <korzen> the other issue is that current flow of creating port binding is udpating the host in later steps 15:12:05 <korzen> updating* 15:12:30 <korzen> now having the host as primary key, OVO is forbidding to update it 15:13:18 <korzen> so I'm working on delete and create instead of update the port binding 15:13:30 <ihrachys> oh. we can either delete/create, or we may need to relax rules for primary keys not being updatable 15:14:03 <ihrachys> what was the reason why we enforced that in the first place? any good explanation? 15:14:06 <korzen> it is the requirement not to update the primary keys because of objects DB API 15:14:29 <korzen> in object DB API we are doing fetch and then save the changes 15:14:42 <korzen> if you change the primary key, the fetch will fail 15:15:25 <korzen> and you do not have a way of look back the original value of PK 15:15:30 <korzen> in object state 15:15:52 <ihrachys> oh right. 15:16:12 <ihrachys> so we would need to pass two sets of filters - one for original keys, another for new 15:16:26 <ihrachys> yeah better keep it simple while it's possible and do delete/create 15:16:49 <korzen> yes 15:17:16 <korzen> so I'm looking into delete&create scenario, the work is in progress 15:17:29 <ihrachys> that being said, an updated object == fetched object, so we have the info on original fields; but probably not worth of exploring. 15:17:48 <ihrachys> korzen: thanks, I hope we can get it in this week to avoid dangerous backports 15:18:14 <korzen> I hope it also 15:18:47 <ihrachys> looking at the merged list from previous week 15:18:55 <ihrachys> https://review.openstack.org/#/q/status:merged+project:openstack/neutron+branch:master+topic:bp/adopt-oslo-versioned-objects-for-db 15:19:12 <ihrachys> seems like we landed ipam only 15:19:32 <ihrachys> as for open patches... 15:19:33 <ihrachys> https://review.openstack.org/#/q/status:open+project:openstack/neutron+branch:master+topic:bp/adopt-oslo-versioned-objects-for-db 15:20:06 <ihrachys> some were ready to go, like Tag or ProvisioningBlock, but will need to wait master to reopen 15:20:16 <korzen> I was also looking into subnet ovo adoption patch 15:20:16 <electrocucaracha> ihrachys: what about Tag? it was so close 15:20:39 <ihrachys> electrocucaracha: we bumped it off gate after Ocata-3 was released. 15:20:49 <ihrachys> we will land in due course start of Pike 15:21:04 <ihrachys> NetworkSegment turned out to have an issue with passing gate: https://review.openstack.org/385178 15:21:10 <ihrachys> electrocucaracha: any news on that gate failure? 15:21:24 <ihrachys> the one where neutron apparently hanged on segment allocation 15:21:41 <electrocucaracha> I tried using dicts instead of objects but it's still failing 15:22:08 <electrocucaracha> I noticed that most of the suscriptors use them as dicts 15:22:32 <electrocucaracha> I'll need extra eyes on that 15:22:53 <ihrachys> ok. have you tried to reproduce it locally? 15:23:55 <electrocucaracha> I think that I couldn't 15:24:04 <electrocucaracha> but I can retry today 15:24:31 <ihrachys> ok. it's one of those cases that is very expensive to debug in gate. 15:25:13 <electrocucaracha> agree 15:26:32 <korzen> the problem with subnet/network/port OVOs would be the extensions, I hoped that enginefacade work will help with detached db_obj, but it is not so simple, I need to add the db_obj to session in make_subnet_dict and detach it again after extension is done adding its db stuff 15:28:18 <ihrachys> korzen: as long as they pre-fetch their relationships on fetch, isn't it enough for them to get access to them? 15:28:33 <ihrachys> even while the db_obj may be detached? 15:29:04 <korzen> ihrachys, yes, all extensions need to be pre-fetched, because db_obj won't be able to do any feching 15:29:46 <korzen> I guess with current implementation of extensions, we need to register it somehow 15:29:59 <korzen> and do the prefetching 15:30:14 <ihrachys> isn't it done with lazy= directive in db model? 15:30:29 <korzen> yes, lazy should work 15:31:35 <korzen> so I guess every extension should define its db relationship with lazy joined 15:32:06 <ihrachys> maybe not joined, see https://review.openstack.org/#/c/408143/ 15:32:31 <ihrachys> but the premise is the same - their models may require some tweaking to pre-fetch data 15:32:52 <ihrachys> which should also help keeping the number of queries to fetch a resource at constant. 15:33:44 <korzen> yes 15:34:05 <korzen> btw, electrocucaracha are you going to PTG? 15:34:16 <electrocucaracha> Nope :( 15:34:24 <korzen> :( 15:34:26 <electrocucaracha> actually I was checking the neutron agenda 15:34:34 <electrocucaracha> https://etherpad.openstack.org/p/neutron-ptg-pike 15:34:55 <electrocucaracha> it seems like neutron upgrades will be covered by your topics korzen right? 15:35:20 <korzen> yes, I wanted to setup the goals for Pike 15:35:38 <korzen> if not at wide audience, at least at upgrades subteam forum 15:36:17 <electrocucaracha> in terms of scope, how can we make to complete it in pike 15:36:19 <electrocucaracha> ? 15:37:11 <korzen> by our hard work ;) 15:37:24 <electrocucaracha> :) 15:37:31 <korzen> what do you mean exactly? 15:38:17 <electrocucaracha> well, OSIC is so interested in enable Neutron upgrades in Pike, so if it's needed more resources or something else 15:38:31 <electrocucaracha> maybe we can help somehow 15:39:14 <korzen> we will groom the efforts and tasks to do, and we will start doing it :) 15:39:14 <electrocucaracha> the good news is that we have a lot of progress in many ongoing patches the bad news is that they're on progress 15:39:22 <ihrachys> neutron upgrades is a vague term. which specific use cases do you consider? if that's mixed api versions coverage in gate, that's I believe something that we will need people looking at and sync with other projects. 15:39:44 <ihrachys> in the meantime, korzen was planning to test it manually for Ocata 15:40:44 <korzen> I think that online data migration is the critical piece we are missing to support zero-downtime upgrades 15:40:59 <electrocucaracha> yeah, maybe it should be better to define the scope and be realistic on the things that we can accomplish 15:41:12 <electrocucaracha> this seems more like cathedral model instead of bazaar 15:42:35 <electrocucaracha> this is something that we can't discuss in 18 minutes, but at least I'd like to consider it in the agenda of PTG 15:42:37 <ihrachys> korzen: that would assume that you have a valid case to cover with data migration rules. so far we did not have one. maybe live migration will give us one. I am all for working on framework for that, but do we grasp what will need to be done right now so that we can start? 15:42:41 <korzen> ihrachys, do you know the exact schedule for the PTG? or are there any free spot to create upgrade subteam meeting before we go to the neutron team>? 15:43:33 <ihrachys> korzen: afaiu ptg is not a single stream event, it's more like multiple groups running with their ideas in smaller groups. so we should be able to sync on spot. 15:44:17 <electrocucaracha> korzen ihrachys : dasanind will be there to also representing OSIC, just try to involve her in the discussions 15:44:46 <korzen> ok, I will work on the agenda 15:44:50 <ihrachys> gotcha 15:45:19 <dasanind> electrocucaracha: thank you 15:45:51 <ihrachys> #action korzen to work on PTG upgrades agenda 15:46:02 <dasanind> korzen: and ihrachys : all the meetings in PTG are in one room? 15:46:27 <ihrachys> korzen: I believe it's like a room with multiple tables. honestly it's first PTG so we may need to check. 15:46:55 <ihrachys> korzen: anyway, my question on the strategy for data migration from above is still up. what do you think on how to make progress without a use case? come up with a fake one? 15:48:03 <korzen> ihrachys, I'm planning to take a look at nova's and cinder's way of completing this action. I have a feeling that the framework should be done before (or at least in parallel) to actual first usage of it 15:48:49 <ihrachys> I agree with the goal 15:48:49 <korzen> if some would need to do the data migration, he/she would be blocked because of non existing framework 15:49:22 <electrocucaracha> +1 15:49:39 <korzen> I will update the team upfront after initial research 15:49:51 <korzen> if the online data migration is critical 15:50:49 <ihrachys> ack 15:50:51 <ihrachys> thanks 15:51:01 <korzen> I'm not so keen on API versioning, but I guess that there will be a separate discussion on the API future 15:51:24 <korzen> and the upgrades use-case can be a addition to such discussion 15:52:03 <reedip_> ihrachys : do we have an etherpad for the ptg ? 15:52:15 <ihrachys> reedip_: https://etherpad.openstack.org/p/neutron-ptg-pike 15:52:40 <ihrachys> korzen: yes api seems to be a hot potato right now 15:53:50 <ihrachys> ok, let's move one 15:54:16 <ihrachys> I don't see any motion on patches with UpgradeImpact tag, so skipping the section 15:54:23 <ihrachys> #topic Open discussio 15:54:25 <ihrachys> #topic Open discussion 15:54:44 <ihrachys> the list on the agenda page is empty as usual 15:54:59 <ihrachys> unless we have something on spot to raise... 15:55:27 <electrocucaracha> hehehe, well we started the open discussion before 15:55:34 <electrocucaracha> talking about PTG 15:55:37 <ihrachys> kinda yes 15:55:41 <korzen> sorry about that ;) 15:56:18 <ihrachys> ok if there is nothing to raise at this point, I call it a day. thanks folks. 15:56:19 <korzen> no other opens from me 15:56:25 <ihrachys> #endmeeting