14:01:21 <apuimedo> #startmeeting kuryr 14:01:21 <openstack> Meeting started Mon Jun 19 14:01:21 2017 UTC and is due to finish in 60 minutes. The chair is apuimedo. Information about MeetBot at http://wiki.debian.org/MeetBot. 14:01:22 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 14:01:25 <openstack> The meeting name has been set to 'kuryr' 14:01:31 <apuimedo> Who's here for the meeting this week? 14:01:35 <vikasc> o/ 14:01:39 <danil> o/ 14:01:44 <irenab> o/ 14:01:45 <ltomasbo> o/ 14:01:46 <janonymous> o/ 14:01:58 <limao_> o/ 14:02:29 <apuimedo> #topic kuryr-libnetwork 14:02:32 <zengchen> #join #openstack-kuryr 14:02:38 <zengchen> sorry 14:02:39 <garyloug> o/ 14:02:58 <apuimedo> The biggest thing happening in kuryr-libnetwork is hongbin's patch to add support for specifying subnets 14:03:06 <apuimedo> limao_ has been reviewing it 14:03:28 <limao_> apuimedo: yes, discussed with hongbin 14:03:32 <apuimedo> I defer to his good judgement 14:03:39 <apuimedo> limao_: I saw the points you made 14:03:42 <apuimedo> of the three cases 14:03:47 <apuimedo> no subnet, 1:1 and 1:N 14:03:49 <apuimedo> I agree 14:04:24 <apuimedo> does anybody else have some questions, points to make on kuryr-libnetwork? 14:05:51 <apuimedo> alright 14:05:53 <apuimedo> moving on 14:06:02 <apuimedo> #topic kuryr-kubernetes 14:06:10 <danil> I'd like to ask you opinion about my featute in kuryr-k8s project I work on currently. 14:06:18 <danil> This feature is based on feature of Kirill about Sriov support: https://review.openstack.org/#/c/462455/ 14:06:29 <danil> It's about specific ports for SRIOV VFs. I want to specify specific ports (port ids) in pod spec manually while pod creating. Specific ports will be attached to requested VFs firstly. If amount of specific ports is less then amount of requested VFs than others (remaining) ports will be requested from neutron by ordinar way. 14:06:44 <danil> Also this specific ports will not be deleted after deployment deleting 14:07:03 <danil> It's an old version of feature: https://review.openstack.org/#/c/468032/ I work on changes now 14:07:12 <irenab> danil: can it be generic port_id request? 14:07:13 <apuimedo> danil: this sounds quite in line with what kzaitsev1pi was proposing 14:07:34 <irenab> not specific to SR-IOV? 14:07:47 <danil> it's specific to sriov 14:07:55 <apuimedo> irenab: danil: There's two things here 14:08:08 <apuimedo> one of them is that if you annotate the vifs directly in your spec 14:08:13 <apuimedo> then kuryr-controller should inhibit 14:08:17 <apuimedo> and cni should bind 14:08:32 <apuimedo> that works for both sr-iov and for any other kind of neutron port 14:08:37 <apuimedo> s/port/vif/ 14:08:40 <irenab> apuimedo: agree, cni is pecific, but controller part seems generic enough 14:08:46 <apuimedo> the other thing is about requesting 14:09:24 <danil> about requesting. Specific ports are created manually 14:11:19 <irenab> I think requesting specific port ids may be useful if you want to map to some port with qos settings, sgs, etc. 14:12:00 <irenab> danil: apuimedo : added the patch to my review queue 14:12:09 <ltomasbo> in a similar discussion line 14:12:11 <danil> yes, exactly 14:12:14 <danil> thanks 14:12:25 <ltomasbo> could we also add something for the subnet (or network) the pod should be in? 14:12:39 <ltomasbo> so that instead of having to use the hardcoded one at the kuryr.conf 14:12:40 <apuimedo> and that, for now seems good 14:12:55 <danil> currently I worl only on specific ports 14:12:56 <apuimedo> but frankly 14:13:01 <ltomasbo> we could then decide on the subnet the pod should be in? 14:13:08 <apuimedo> I'd rather it'd be a separate driver 14:13:39 <apuimedo> the default subnet driver is meant to be simple, not with overrides 14:14:37 <irenab> apuimedo: correct. the ‘default’ driver was about getting options from config 14:14:48 <ltomasbo> apuimedo, I was thinking about having some annotation with the CIDR, so that we can create the container in the specified subnet (even create it if it does not exists) 14:14:57 <ltomasbo> similarly to how we do in kuryr-libnetowrk 14:15:40 <irenab> ltomasbo: what about having multiple networks in k8s? then it may be the network annotattion 14:15:41 <apuimedo> ltomasbo: that's fine 14:15:48 <apuimedo> but I'd rather that be a different driver 14:15:54 <vikasc> and if annotation not found, pick from config? 14:16:05 <apuimedo> honestly 14:16:14 <irenab> vikasc: it depends on the drivers in the deployment 14:16:21 <apuimedo> this is an area where I expect contribution 14:16:30 <apuimedo> people coming up with different and better drivers 14:16:36 <apuimedo> the ones right now are sample level 14:17:01 <irenab> apuimedo: it probably should come with the RFE/bp. Driver is just an implementation detail 14:17:04 <vikasc> like the one, "subnet_of_choice" driver 14:17:35 <vikasc> ofcource a better name :) 14:17:40 <ltomasbo> :D 14:17:42 <vikasc> s/ofcourse 14:17:44 <apuimedo> irenab: bp at most 14:17:55 <apuimedo> I'd require more to change the default 14:18:01 <apuimedo> for adding drivers... Not so much 14:18:54 <ltomasbo> irenab, that is a good point, how to handle both, specifying different subnets, but also enabling multi-networks 14:19:38 <irenab> I wonder if app devs would really wnat to deal with subnets 14:20:14 <ltomasbo> perhaps just network 14:20:33 <apuimedo> irenab: I think this is more of a ops thing 14:20:33 <irenab> especially on pod level. I think it better be defined on network/namespace level 14:20:35 <apuimedo> not app dev 14:20:37 <ltomasbo> I was thinking something like: I want this container in 10.10.0.0/24 14:20:52 <vikasc> i guess providing both, subnet of choice and multiple networks , would add both, lot of flexibility and confusion as well for app devel;oper 14:20:53 <apuimedo> or vnf dev 14:20:58 <irenab> apuimedo: exactly, so pod level is not really good option 14:21:45 <irenab> I think its more : I want this container of extra fast network 14:22:16 <vikasc> wondering what are the use cases which can only be addressed by having subnet_of_choice and not by multiple_networks 14:23:34 <vikasc> network-subnet 1-1 mapping 14:23:57 <apuimedo> frankly, apart of dual stack, I don't really have any idea why would somebody want multiple subnets on a net 14:24:19 <vikasc> +1 14:24:34 <ltomasbo> agree 14:24:39 <irenab> me too 14:25:10 <apuimedo> but as usual, we're open to hear otherwise 14:25:13 <apuimedo> :-) 14:25:33 <vikasc> so instead of focussing on providing a way to request multiple subnets per network, i would vote for focussing on multiple networks 14:25:48 <apuimedo> vikasc: that's right 14:25:48 <irenab> vikasc: +1 14:25:54 <ltomasbo> agree too 14:26:00 <vikasc> wondering how multiple networks effort is converging upstream k8s 14:26:11 <apuimedo> what we really need is somebody to follow the progress of the upsream PoC and make our own based on it 14:26:15 <irenab> vikasc: Are you in details of the k8s multi network implementation status? 14:26:19 <vikasc> if anyone is updated with current state 14:26:46 <vikasc> i could not check in last two weeks irenab 14:26:55 <irenab> apuimedo: I can try to check it for next week 14:27:03 <apuimedo> irenab: that'd be great 14:27:10 <apuimedo> danil: you're welcome to do that as well 14:27:29 <danil> ok, thanks a lot 14:27:47 <irenab> anyway, it seems ortogonal to requesting specific port, that seems to be useful feature 14:27:54 <vikasc> and may be if we see that it seems to be taking still a lot longer 14:28:13 <danil> I will complete and will send to review 14:28:13 <vikasc> we can try implementing multinetwork support using annotations in kuryr 14:28:25 <vikasc> that call we can take 14:29:28 <vikasc> because IIRC , we said last time to wait because it lloke like it was going to come natively in k8s 14:29:45 <apuimedo> vikasc: it's annotations + tpr upstream 14:29:55 <apuimedo> I really would like to know how they are using tpr 14:29:59 <apuimedo> in that workflow 14:30:00 <vikasc> apuimedo, yeah it was around month back 14:30:24 <vikasc> apuimedo, tpr for just holding network object 14:30:51 <vikasc> apuimedo, and propagate info through annotations 14:31:03 <apuimedo> so network reference in the pod 14:31:07 <apuimedo> and netowrk object as tpr 14:31:10 <apuimedo> right 14:31:13 <vikasc> apuimedo, yes 14:31:22 <vikasc> its pretty simple 14:31:43 <apuimedo> I don't see how network object serve us though 14:32:22 <vikasc> its for holding network metadata i guess 14:32:30 <apuimedo> yeah 14:32:35 <apuimedo> but we have it in Neutron 14:32:42 <apuimedo> I remember ivc_ had a proposal for moving most annotations to tprs 14:32:43 <vikasc> for k8s user 14:33:16 <ltomasbo> vikasc, do you have any link with useful information about this? (status, target, design, ...) 14:33:42 <vikasc> i will seach and send 14:33:49 <vikasc> s/search 14:33:55 <vikasc> and share ok kuryr channel 14:34:00 <vikasc> s/on 14:34:13 <apuimedo> thanks vikasc 14:34:15 <vikasc> ltomasbo, it is on google groups , signetwork 14:34:28 <vikasc> i will find out 14:34:39 <ltomasbo> vikasc, thanks! 14:34:44 <vikasc> yw! 14:35:34 <irenab> apuimedo: regarding https://review.openstack.org/#/c/472763/ 14:36:09 <irenab> still does not work for me completely, but all the trunk setting is very useful 14:36:15 <apuimedo> #info Installation instructions were merged https://docs.openstack.org/developer/kuryr-kubernetes/installation/index.html 14:36:53 <apuimedo> irenab: I'll try to get some time this week to go step by step with you so I can spot where I put the bug 14:37:11 <irenab> I also posted some question there 14:37:19 <apuimedo> ah, cool 14:37:21 <apuimedo> I'll check 14:37:51 <irenab> regarding the installation instructions 14:38:21 <irenab> where the special configuration of the service account/role/service-router is supposed to be covered? 14:38:23 <apuimedo> #info janonymous fixed vagrant devstack 14:39:19 <apuimedo> #info Also work is progressing into making kuryr-kubernetes deployable in containers https://review.openstack.org/466675 14:39:38 <apuimedo> #info We're also getting the tempest plugin soon 14:39:48 <apuimedo> It should be merged this week 14:40:32 <janonymous> Please have a look at https://review.openstack.org/#/c/454555/ patch also in free time :) 14:41:02 <apuimedo> janonymous: yes. We should get that by pike-3 14:41:17 <apuimedo> I would like to run it a few days before I'm comfortable with it 14:41:26 <janonymous> apuimedo: cool, thanks! 14:41:36 <janonymous> apuimedo: agree 14:41:50 <apuimedo> Hopefully I'll get machines to test the SR-IOV kzaitsev is working on 14:42:06 <apuimedo> Anything else on kuryr-k8s 14:43:09 <apuimedo> alright. Moving on 14:43:20 <apuimedo> #topic general 14:43:27 <ltomasbo> apuimedo, irenab: https://review.openstack.org/#/c/474576/ 14:43:29 <apuimedo> Any other topic to close today's meeting? 14:43:34 <ltomasbo> take a look when you have time ^^ 14:43:41 <irenab> apuimedo: fuxi? 14:43:43 <apuimedo> Next week we'll maybe skip due to business travel 14:43:53 <janonymous> Any updates on ptg? 14:43:53 <apuimedo> irenab: darn. I checked if hongbin was present 14:43:56 <apuimedo> and didn't see him 14:43:57 <hongbin> hi 14:43:59 <apuimedo> Le'ts to fuxi 14:44:02 <apuimedo> #topic fuxi 14:44:25 <hongbin> last week, i believe zengchen is working on several patches on fuxi-k8s 14:44:33 <hongbin> zengchen: want to give a brief update? 14:44:41 <zengchen> ok 14:45:01 <zengchen> I am working on flex-volume driver. 14:45:18 <zengchen> i believe i will finish cinder-driver this week 14:45:25 <apuimedo> zengchen: Hi! Didn't notice you were here too :-) 14:45:54 <zengchen> i am seeing you are taling about kuryr 14:46:06 <zengchen> s/taling/talking 14:46:09 <apuimedo> :-) 14:46:20 <apuimedo> any updates on fuxi-golang? 14:46:36 <hongbin> apuimedo: i believe the first task is for you to setup the core team: https://review.openstack.org/#/admin/groups/1784,members 14:46:48 <zengchen> i am afraid i don't hava enough time. 14:47:03 <zengchen> maybe next week i will start on it 14:47:04 <apuimedo> hongbin: good point 14:47:17 <apuimedo> #action apuimedo to set up fuxi-golang core team 14:47:22 <apuimedo> #link https://review.openstack.org/#/admin/groups/1784,members 14:47:33 <hongbin> zengchen: thanks zengchen , apuimedo i have nothing else from my side 14:47:39 <apuimedo> very well 14:47:51 <apuimedo> zengchen: anything else from your side? 14:48:15 <zengchen> maybe we should invite more guys from cinder/manila to contribute on fuxi-go 14:49:29 <apuimedo> zengchen: agreed. Let's try to reach out 14:49:35 <apuimedo> #topic general 14:49:42 <apuimedo> Anything else? 14:49:53 <apuimedo> #action apuimedo to look at https://review.openstack.org/#/c/474576/ 14:50:08 <janonymous> Any updates on ptg? 14:50:28 <apuimedo> VTG 14:50:50 <janonymous> :D 14:50:57 <apuimedo> that's the only update 14:51:00 <apuimedo> I don't have a date 14:51:15 <apuimedo> but I'll send some email about it in a week or two 14:52:44 <apuimedo> Thanks all for joining 14:52:47 <apuimedo> #endmeeting