14:01:19 <dmellado> #startmeeting kuryr 14:01:20 <openstack> Meeting started Mon Mar 19 14:01:19 2018 UTC and is due to finish in 60 minutes. The chair is dmellado. Information about MeetBot at http://wiki.debian.org/MeetBot. 14:01:21 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 14:01:24 <openstack> The meeting name has been set to 'kuryr' 14:01:32 <dmellado> Hi fellow kuryrs, who's here today for the meeting? 14:02:08 <dulek> o/ 14:02:13 <yboaron_> o/ 14:02:17 <ltomasbo> o/ 14:02:55 <irenab> hi 14:03:14 <dmellado> #chair dulek yboaron_ ltomasbo irenab 14:03:15 <openstack> Current chairs: dmellado dulek irenab ltomasbo yboaron_ 14:03:32 <kaliya> o/ 14:03:45 <dmellado> Hi kaliya 14:03:48 <dmellado> so, let's start 14:03:57 <dmellado> #topic kuryr-kubernetes 14:04:18 <dmellado> I wanted to share with you that after a few days of CI crisis we should be fine for the time being 14:04:50 <dmellado> I'll be trying to decouple a few dependencies 14:05:09 <dmellado> so if you've any patch that failed recently, and specially with post_failure, please do recheck it 14:05:14 <dmellado> it might not be your fault 14:05:35 <dmellado> I'll be also adding some multinode gates during the upcoming days, now with proper services split 14:05:42 <dmellado> 2 and 4 nodes, probably 14:05:55 <dmellado> so, besides that, who has something that would love to share? :P 14:07:18 <irenab> dmellado, is there any news regarding sr-iov/dpdk patches? 14:08:01 <dmellado> irenab, not really, maybe kaliya can update us on that 14:08:29 <dmellado> I asked the intel foiks to update the blueprint with their contributions/info 14:08:34 <kaliya> not much. I know danil is working on rebasing the sr-iov part. For the nested part, I saw somewhere some binary file including some Intel patch 14:09:05 <dmellado> https://blueprints.launchpad.net/kuryr-kubernetes/+spec/nested-dpdk-support 14:09:36 <dmellado> kaliya: yeah, they sent us a diff file but I asked them to put a proper review so we could take a look and review 14:09:51 <dmellado> tbh I was hoping that Gary was on the call today to ask him about that 14:09:58 <dmellado> irenab: Will send him an email ad cc you 14:10:11 <irenab> dmellado, thanks 14:10:14 <dmellado> kaliya: thanks on the sriov side, btw 14:10:18 <kaliya> yes, that diff has some syntax issues, thank you dmellado 14:10:36 <yboaron_> Question: Do we plan to merge the multi vif/dpdk/sr-iov patches in current approach, or to wait/sync with what vikasc presented 14:10:44 <dmellado> also tbh I would bet on that diff needing a rebase as well 14:11:13 <dmellado> yboaron_: they have been there for a while, so I'd get there on their current state once they're good to go 14:11:29 <kaliya> what's vikasc "presentation" ? 14:11:35 <dmellado> and add vikasc approach on a follow-up patch 14:11:40 <irenab> dmellado, +1 14:11:42 <dmellado> kaliya: he did send an email to the ML but should be recorded 14:11:54 <kaliya> I missed that 14:12:22 <dmellado> kaliya: I'll ask him for the record link and send an email, no worries 14:12:27 <kaliya> thanks! 14:13:09 <dmellado> yw! 14:13:14 <irenab> kaliya, the slides https://docs.google.com/presentation/d/1_Yu88NA--afhal5uBipQymGqbNKagi-p2kr9acyH8CM/edit#slide=id.g3525172c70_0_0 14:13:59 <kaliya> ahh this one, yes thanks irenab 14:15:13 <dmellado> all right, any another topic or patches that would require attention? 14:15:20 <dmellado> ltomasbo: yboaron_ dulek ? 14:15:34 <dulek> dmellado: Not much from my side regarding kuryr-kubernetes. 14:16:13 <ltomasbo> not much either, perhaps just take a look at the patch: 14:16:25 <ltomasbo> https://review.openstack.org/#/c/548673 14:16:42 <yboaron_> dmellado, neither from my side 14:16:59 <ltomasbo> it will be interesting to agree on going for network ids or subnet ids for the pools 14:17:05 <ltomasbo> with their respective pros and cons 14:17:06 <dmellado> ltomasbo: will do, tbh though this was already merged 14:17:14 <dmellado> will take a look 14:17:30 <ltomasbo> thanks 14:17:40 <dmellado> #topic kuryr-libnetwork 14:17:47 <irenab> ltomasbo, what is the argument? 14:18:10 <ltomasbo> irenab, the current patch assumes there is only one subnet per network 14:18:55 <irenab> ltomasbo, so to make it generic case, need to take subnet_id and not network_id into account? 14:18:57 <ltomasbo> irenab, if we use subnet ids instead, we will support multiple subnets per network, but it will have some performance impact, as we need to call neutron to get subnets ids from network ids 14:19:05 <ltomasbo> which is the information at the vif object 14:19:28 <ltomasbo> well, I'm not sure we need multiple subnets per network 14:19:47 <ltomasbo> and celebdor said that their support in neutron is so-so too 14:20:03 <dmellado> ltomasbo: maybe dual stack, but I'll have to think about it 14:20:32 <ltomasbo> that is the only case, and it seems it is not that realistic/useful either 14:20:33 <irenab> we need to see the use case 14:20:47 <ltomasbo> for a generic use case, network_id is enough 14:20:58 <irenab> as long as there is no specific use case, may go with net_id 14:21:04 <ltomasbo> there may be some corner cases were subnet id may be needed I guess 14:21:15 <dmellado> yeah, I don't really think that it might be worth the performance hit 14:21:15 <ltomasbo> *where 14:21:30 <dmellado> ltomasbo: you're more than welcome to add tests regarding the corner cases 14:21:30 <irenab> and the framwork should be flexible enough to add multi subnets in the future is needed 14:21:45 <celebdor> ltomasbo: can you remind me why there is a perf impact on using subnet id? 14:21:59 <ltomasbo> yes, it is switching one key by the other andthe way to obtain it 14:22:13 <ltomasbo> celebdor, vif obj has already the network id 14:22:13 <celebdor> where do we obtain the netid from? 14:22:20 <celebdor> that's what I thought 14:22:29 <celebdor> but I thought that vif also has subnets 14:22:34 <ltomasbo> but if we want to use the subnet_id we need to retrieve it from the network id + subnet range 14:22:51 <celebdor> in fact I thought we even get the IP address from subnets 14:23:03 <celebdor> (of the vif object) 14:23:06 <ltomasbo> celebdor, we do get the subnet cidr 14:23:13 <ltomasbo> but not the uuid 14:23:45 <yboaron_> ltomasbo, I Assume that caching will help with performance, how many network to we intend to support ? 14:23:58 <yboaron_> networks 14:24:00 <ltomasbo> I thought about that too 14:24:04 <ltomasbo> but it may have some problems 14:24:10 <irenab> the performance impact can be itigatred if kuryr will fetch subnets on start-up 14:24:12 <celebdor> ltomasbo: ah... Another instance of crappy (for our purposes) os-vif information structure 14:24:14 <ltomasbo> we need to base it on the network ui and the subnet range 14:24:33 <ltomasbo> and then you can actually remove the subnet and create another one, with same range (and different uuid) 14:25:36 <ltomasbo> so, caching could be tricky, and assuming the subnets are already there could not cover all the use cases, for instance creating new subnets for each namespace 14:26:00 <dmellado> yeah, if we cache and create new subnets then we're screwed again 14:26:20 <yboaron_> ltomasbo, yes, that's a good point 14:26:34 <ltomasbo> another option could be to use network_id + subnet cidr 14:26:41 <ltomasbo> that should be unique 14:26:52 <ltomasbo> but perhaps a bit ugly... 14:26:55 <irenab> do we capture per namespace subnet discussion somewhere? 14:27:19 <irenab> not pool specific parts only? 14:27:20 <ltomasbo> not yet, I'm working on a PoC, and will write a devref soon 14:28:05 <celebdor> dulek: I sort of feel that with KuryrPort CRDs we would not have that much of a problem 14:28:08 <irenab> maybe worth to discuss the the full picture 14:28:28 <dmellado> irenab: +1 14:28:29 <yboaron_> +1 14:28:39 <dulek> celebdor: Well, K8s API is much more reliable than Neutorn. 14:28:49 <dulek> celebdor: But we still get synchronization issues, right? 14:28:57 <ltomasbo> irenab, yes, but adding the network id to the pools is related but not only targeting the subnet per namespace 14:29:15 <dmellado> ltomasbo: let's add in any case some docs to your patch 14:29:20 <dmellado> I'll fetch you tomorrow at the office 14:29:23 <ltomasbo> as if you have pools enabled, and change the subnet_id at kuryr.conf and restart the controller, it will lead you to error 14:29:31 <ltomasbo> not to error, to the pods in a different subnet 14:29:43 <celebdor> dulek: like? 14:29:56 <irenab> ltomasbo, agree, but some scenarious you consider seems to be driven by multi-subnet network 14:30:46 <ltomasbo> irenab, could be, but the initial though was on making kuryr namespace aware, more than the multi-subnet driver 14:30:49 <celebdor> With KuryrPort you'd just get the subnet_id from the driver, create the KuryrPort objects in the K8s API 14:31:05 <celebdor> that are assigned to specific Host 14:31:08 <dulek> celebdor: Like when creating and changing subnets? Don't know yet TBH. 14:31:34 <dulek> celebdor: Mhm. 14:31:37 <irenab> celebdor, how driver will fetch the subnet_id? 14:31:54 <celebdor> irenab: that's driver specific 14:32:02 <celebdor> one can be configuration 14:32:05 <ltomasbo> celebdor, for the subnet per namespace object we could also add subnet id to the CRDs, and use them for the pools 14:32:06 <irenab> call neutron? 14:32:25 <celebdor> ltomasbo: yes, I thought about KuryrSubnet 14:32:26 <ltomasbo> but that will solve only that case 14:32:34 <celebdor> and have the namespace annotated to point to it 14:32:44 <ltomasbo> celebdor, I already have a kuryrNet in my Pod :) 14:32:47 <ltomasbo> PoC 14:32:54 <celebdor> irenab: "call neutron" That usually only makes things worse :P 14:33:04 <celebdor> ltomasbo: that's good 14:34:24 <celebdor> so... I feel like this needs a videochat discussion :P 14:34:30 <ltomasbo> xD 14:34:38 <yboaron_> Yeppppppp 14:35:20 <ltomasbo> I agree, but why if we start with 1 subnet per network, and if we find out a use case for more than a subnet per network, then move to subnet_ids 14:35:44 <ltomasbo> it will be much simple and if there is no such use case, it will not pay off 14:35:57 <irenab> +1 14:36:09 <ltomasbo> (and perhaps the best solution is to include the subnet id at the os-vif object actually) 14:36:13 <celebdor> ltomasbo: okay, okay 14:36:39 <ltomasbo> that will bring consistency across drivers/pools 14:36:50 <celebdor> ltomasbo: you have my +2 14:37:01 <celebdor> but if we get to the end of the cycle in this state I'll get grumpy 14:37:09 <dmellado> celebdor: like always 14:37:11 <celebdor> I really want to move state to CRDs 14:37:12 <dmellado> xD 14:37:14 <yboaron_> ltomasbo, you have my +1 for ... video chat :-) 14:37:15 <celebdor> dmellado: but a bit more 14:37:27 <ltomasbo> xD 14:37:41 <dmellado> in any case yah, let's do a bluejeans call on the topic 14:38:03 <celebdor> ltomasbo: since you prolly have more meetings, send the invite :P 14:38:25 <ltomasbo> celebdor, what do you want to move to CRDs? for all the pools? or the cni pool ones? 14:38:58 <celebdor> ltomasbo: I want the Pools to be de facto backed by KuryrPort objects in the API 14:39:09 <ltomasbo> or you mean to move pools to CRD regardless of the driver 14:39:12 <celebdor> so that reloading is not a Neutron search operation, but a GET on K8s 14:39:42 <celebdor> like what kuryr CNI side pools PoC by dulek is doing 14:39:56 <ltomasbo> well, reloading is cheap enough now 14:40:09 <celebdor> ltomasbo: I don't mean it just because of this 14:40:22 <celebdor> I'm just trying to consolidate state on the K8s side 14:40:35 <celebdor> having much in both sides makes me uneasy 14:40:55 <ltomasbo> ok, anyway, I'll send an invite for a call 14:41:30 <dulek> Are w talking kuryr-libnetwork now? 14:41:36 <ltomasbo> go! 14:41:52 <dulek> There's this fix I've wrote for Neutron extensions getting deprecated: https://review.openstack.org/#/c/553756 14:41:58 <dulek> s/deprecated/removed 14:42:10 <dulek> Neutron patch is depending on this one, so it'll be great to get it reviewed. 14:42:26 <celebdor> dulek: alrighty 14:42:28 <dulek> I've already tested it here: https://review.openstack.org/#/c/553763, looks like things are working. 14:42:34 <dmellado> dulek: saw it, will review 14:43:28 <dulek> Thanks! 14:43:31 <irenab> dulek, will +2 once zuul is happy 14:43:37 <dmellado> yep 14:44:08 <dmellado> # open discussion 14:44:14 <dmellado> #topic open discussion 14:44:19 <celebdor> dulek: looks good to me 14:44:55 <celebdor> irenab: dmellado: are you both going to the summit? 14:45:13 <celebdor> Cause I got some requests to confirm my attendance due to the accepted talks 14:45:21 <celebdor> and as you know I won't be able to make it this time 14:45:29 <irenab> celebdor, I am not sure yet, will update 14:45:35 <dmellado> celebdor: I am, I'll need still ack fron irena, yep 14:46:19 <yboaron_> the Israeli summit ? :=) 14:46:27 <dmellado> yboaron_: the Vancouver one 14:46:33 <dmellado> :P 14:46:40 <celebdor> yboaron_: that one I'm counting with you and irenab for 14:46:41 <yboaron_> dmellado, kidding 14:46:52 <celebdor> I'd love to go though 14:47:06 <celebdor> and the weather is nicer than in Vancouver 14:47:22 <dmellado> celebdor: in any case throw the session to me and I'll see how can I manage 14:47:27 <irenab> celebdor, not sure about the weather 14:47:47 <yboaron_> I'm sure about the weather! 14:47:52 <dmellado> it can't be worse than Dublin 14:47:54 <irenab> its about +33 here today 14:48:15 <ltomasbo> dmellado, are you also taking the other one (SF one?) 14:48:18 <dmellado> +33? how much humidity? 14:48:27 <celebdor> dmellado: I already did 14:48:30 <dmellado> ltomasbo: yeah, those are lightning talks so I could handle 14:48:39 <dmellado> but I'll see if I can get cospeakers 14:48:42 <ltomasbo> ok 14:48:50 <celebdor> dmellado: the humidity depends on how much you sweat 14:48:51 <celebdor> xD 14:48:55 <irenab> ltomasbo, are you planning to attend the summit? 14:49:00 <dmellado> ltomasbo: you're always welcome to come along, though 14:49:15 <dmellado> irenab: we split the SF project and the Summit and ltomasbo will travel closer xD 14:49:23 <ltomasbo> irenab, I don't think so, going all the way to Vancouver for just 10 minutes talk... 14:49:41 <dmellado> ltomasbo: you can also come along for some another session xD 14:49:45 <ltomasbo> dmellado, closer but more boring... 14:49:49 <irenab> I think it has potential to become 40 mins 14:49:54 <dmellado> so far I've got 2 lightning talks 14:49:57 <dmellado> + one full talk 14:50:00 <dmellado> + project updates 14:50:04 <dmellado> + project oboarding 14:50:07 <dmellado> man, so much for me xD 14:50:10 <ltomasbo> xD 14:50:46 <irenab> dmellado, sounds like you will miss the Dublin summit ... 14:50:49 <dmellado> and the worst thing is that my girlfriend expects me to go to the wedding from her friend the next day after I'm back xD 14:51:15 <kaliya> you can bring a nice present:) 14:51:18 <dmellado> irenab: you go and grab someone from toga if you can't make it in the end xD 14:51:30 <irenab> :-) 14:52:00 <dmellado> anyways, it's being a looong meeting day and I still didn't get to have lunch 14:52:05 <dmellado> so thanks for attending 14:52:12 <dmellado> and bon apetit! xD 14:52:13 <kaliya> bye 14:52:14 <dmellado> xD 14:52:18 <celebdor> thank you all! 14:52:19 <irenab> bye 14:52:20 <yboaron_> bbye 14:52:27 <dmellado> #endmeeting kuryr