14:00:36 <apuimedo> #startmeeting kuryr
14:00:37 <openstack> Meeting started Mon May 15 14:00:36 2017 UTC and is due to finish in 60 minutes.  The chair is apuimedo. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:00:38 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
14:00:40 <openstack> The meeting name has been set to 'kuryr'
14:00:53 <apuimedo> Welcome to the post Summit weekly irc meeting
14:00:57 <apuimedo> who's here today?
14:01:01 <vikasc> o/
14:01:03 <garyloug> o/
14:01:08 <irenab> hi
14:01:10 <kzaitsev_ws> \o
14:01:31 <limao> o/
14:01:33 <apuimedo> ltomasbo and dmellado have public holidays today
14:01:41 <ivc_> o/
14:02:06 <apuimedo> alright. Let's get started!
14:02:13 <apuimedo> thanks all for joining
14:02:17 <apuimedo> #topic kuryr-libnetwork
14:02:51 <apuimedo> we have nothing in flight for kuryr-libnetwokr
14:03:06 <kzaitsev_ws> I've updated the uwsgi patch btw
14:03:12 <apuimedo> the libnetwork2 patches got merged
14:03:16 <apuimedo> oh right
14:03:17 <apuimedo> :P
14:03:39 <kzaitsev_ws> also will try to find some time to get the thing running under apache in devstack by default =)
14:03:59 <kzaitsev_ws> and that should conclude the bp and pike goal )
14:04:00 <apuimedo> mod_wsgi is a bit slow :P
14:04:05 <apuimedo> great
14:04:21 <apuimedo> kzaitsev_ws: IIRC there were some issues with the multithread setting in wsgi
14:04:32 <apuimedo> but I don't remember the details :/
14:04:52 <apuimedo> irenab: limao: please, take a look at https://review.openstack.org/#/c/374315/
14:05:00 <irenab> sure
14:05:12 <limao> yes
14:05:16 <apuimedo> #info close to adding wsgi support and thus making one of Pike's community goals
14:05:25 <kzaitsev_ws> we can always make the thing configurable and if we merge it in the middle of the cycle someone might complain in time
14:05:30 <apuimedo> anything else on kuryr-libnetowrk?
14:05:41 <apuimedo> kzaitsev_ws: good point
14:07:06 <limao> I would plan to do some scale test about boot docker with kuryr-libnetwork since I can control some vm resource in my local lab, currently, I would to see what's the performance it would looks like with 10K containers in 100vms
14:07:27 <apuimedo> limao: that would be great
14:07:44 <apuimedo> the best scale testing to date
14:07:46 <apuimedo> :-)
14:07:53 <limao> I'm not sure if anyone has did similar test or has some idea about this
14:07:54 <irenab> limao: do you have some refenrece numbers for native docker network?
14:08:33 <apuimedo> limao: bare metal or on VMs?
14:08:39 <limao> irenab: I think I can do same test in same env with native docker network
14:08:58 <irenab> limao: this will be great, so we can see the comparison
14:09:06 <limao> apuimedo: I think I will test in nested vms
14:09:20 <limao> apuimedo: I do not have 100 bare servers :-)
14:09:23 <apuimedo> cool!
14:10:03 <irenab> you can run in VMs, but not use trunk portd
14:10:05 <apuimedo> irenab: can you make a few #info points of the summary
14:10:09 <apuimedo> #chair irenab
14:10:10 <openstack> Current chairs: apuimedo irenab
14:10:15 <apuimedo> I need to step off for 4 minutes
14:10:21 <limao> apuimedo: irenab: I'd like work out some test case in the next week, and to see how it works.
14:10:29 <irenab> apuimedo: summit?
14:10:38 <apuimedo> yeah
14:10:46 <irenab> limao: great
14:11:19 <irenab> #info limao is about to add test cases for libnetwork and execute performance/scale tests
14:11:38 <irenab> quick summary of Os Boston summit
14:12:06 <irenab> there was quite interest in the kuryr project. Onboarding session was quite full
14:12:59 <irenab> Fuxi seems to gain interest from the cinder folks
14:14:05 <apuimedo> irenab: yes. it's a pity that hongbin is not around
14:14:15 <irenab> We presented the kuryr project update with plans for Pike/Queens and some rough agenda for Rocky
14:14:20 <apuimedo> anyway, the gist of it is that there may be effort to join the golang cinder driver to fuxi
14:14:24 <apuimedo> and have fuxi be golang
14:14:24 <irenab> let me find a link to share
14:14:28 <apuimedo> we'll see about that though
14:14:49 <kzaitsev_ws> wow +)
14:14:56 <kzaitsev_ws> about the golang I mean
14:15:21 <hongbin> o/
14:15:31 <apuimedo> hongbin: nice to see you :-)
14:15:38 <irenab> #link https://www.openstack.org/videos/boston-2017/project-update-kuryr
14:16:13 <apuimedo> kzaitsev_ws: there's also something intriguing that dims reported about some golang cni from they hyper.sh folks being submitted to openstack
14:16:28 <apuimedo> we'll have to be alert to that and check how it all fits
14:17:12 <apuimedo> okay
14:17:18 <apuimedo> moving to other kuryr-kubernetes things
14:17:32 <apuimedo> I started working on the loadbalancer service type
14:17:48 <apuimedo> it will come with small additions to the service handler
14:17:54 <apuimedo> and an external ip driver
14:17:58 <apuimedo> based on floating ip
14:18:10 <apuimedo> so if somebody wants to add support for k8s externalip field
14:18:15 <apuimedo> the same driver could be used
14:18:22 <irenab> apuimedo: will it cover the clusterIP type getting external IP use case?
14:18:37 <apuimedo> irenab: nope. The driver would serve, but I'm not implementing that
14:18:44 <apuimedo> it would be probably 5 lines more only
14:18:46 <apuimedo> :P
14:18:49 <irenab> ok
14:18:57 <irenab> do you have bp?
14:19:03 <apuimedo> not yet
14:19:20 <irenab> apuimedo: you should serve an example :-)
14:19:26 <apuimedo> but I have it for the token support
14:19:34 <apuimedo> irenab: I'll have the spec before I send the patch
14:19:38 <apuimedo> I promise this time!
14:19:40 <apuimedo> xD
14:19:46 <kzaitsev_ws> I'm having an internal meeting about my sriov experiments tomorrow. would probably share the end result and would start drafting the bp/spec
14:19:56 <apuimedo> kzaitsev_ws: cool!
14:20:03 <apuimedo> I haven't had the time to look at it
14:20:15 <apuimedo> and unfortunately I think none of my servers has an sriov card atm
14:20:17 <kzaitsev_ws> I don't expect all of you guys to like my design outright but =)
14:20:33 <apuimedo> kzaitsev_ws: if need be, we'll have a design call ;-)
14:20:43 <kzaitsev_ws> there should be smth to start with )
14:20:47 <apuimedo> cool
14:21:00 <irenab> kzaitsev_ws: no worries, we will have the review process
14:21:13 <apuimedo> of course!
14:21:52 <apuimedo> irenab: ivc_; kzaitsev_ws: vikasc: we should give another round to garyloug's macvlan patch
14:22:10 <irenab> apuimedo: I already did this morning
14:22:14 <apuimedo> If you have something to discuss about it now, we can use 5 minutes of this meeting
14:22:18 <vikasc> will do!
14:22:26 <apuimedo> thanks irenab !
14:22:56 <kzaitsev_ws> For now — I've implemented a separate handler and a separate cni binary, that reuses a lot of code from kuryr-k8s and inherit's it's controll flow. In case the idea doesn't fit general direction it should be separatable into some sort of contrib folder and be pluggable.
14:22:59 <garyloug> thanks guys :-)
14:23:55 <apuimedo> kzaitsev_ws: That's a very failsafe approach
14:24:11 <apuimedo> we should also give another round to ltomasbo's pool patches
14:24:29 <ivc_> kzaitsev_ws apuimedo it should become much more pluggable once we get to cni daemon
14:24:36 <apuimedo> port pooling is very important for the reference implementation in pike
14:24:41 <irenab> apuimedo: I also reviewed them today, basically the main concern is lack of unit tests
14:24:41 <kzaitsev_ws> I need start reviewing the hard patches too )
14:25:37 <apuimedo> I also wanted to propose work on being installable with kubeadm apply -f kuryr.yaml
14:25:52 <vikasc> apuimedo, +1
14:26:14 <apuimedo> If somebody wants to take this. Ping me and I can give some pointers
14:26:15 <irenab> apuimedo: alternative to helm?
14:26:30 <apuimedo> irenab: it's the main k8s installer
14:26:46 <apuimedo> we'll also be collaborating with openstack-helm people for kuryr-k8s chart
14:26:49 <vikasc> apuimedo, i can take this.
14:27:03 <apuimedo> vikasc: very well
14:27:22 <apuimedo> vikasc: do you have time to start this before Friday?
14:27:43 <vikasc> apuimedo, yes, i can start anytime
14:27:54 <apuimedo> right
14:28:06 <vikasc> apuimedo, i got some free cycles
14:28:07 <apuimedo> vikasc: so ping me tomorrow and I'll give you some details
14:28:13 <vikasc> apuimedo, sure.
14:28:16 <kzaitsev_ws> ivc_: any sneak peak on what a cni daemon would be? (like is there some document or a bp?)
14:28:18 <apuimedo> maybe I'll have the bp ready
14:28:29 <apuimedo> kzaitsev_ws: using oslo services
14:28:42 <apuimedo> it would bind to a socket file
14:28:50 <ivc_> kzaitsev_ws we discussed it couple of times. prolly there's a videoconf link in ml
14:28:55 <apuimedo> and it would run the pod event handlers
14:29:13 <apuimedo> then the kuryr cni exec just connects to the socket and writes the requests
14:29:23 <irenab> apuimedo: we need bp with list of points
14:29:32 <apuimedo> points?
14:29:47 <irenab> basic direction in points
14:29:53 <apuimedo> oh! Right!
14:29:59 <apuimedo> ok!
14:30:10 <irenab> ivc_: did you start working on it?
14:30:10 <apuimedo> I'll have BPs for all these stuff this week
14:30:34 <kzaitsev_ws> bp all the things!!! =)
14:30:38 <apuimedo> indeed
14:30:56 <ivc_> irenab nope
14:31:29 <apuimedo> ivc_: I do not remember what was the reason exactly, but I had the urge of proposing the cni split had the daemon only watching and returning data and then the executable performing the NET_ADMIN actions
14:32:03 <apuimedo> maybe it was to have the daemon container not need to be privileged
14:32:36 <ivc_> apuimedo i'd prefer daemon to do the job. don't forget about pyroute2 overhead
14:32:40 <apuimedo> but since we do the baremetal binding in two steps...
14:32:45 <apuimedo> ivc_: oh, right!
14:33:04 <apuimedo> anything else on kuryr-k8s?
14:33:36 <apuimedo> ivc_: oh, I found a bug deploying kube-dns related to services and endpoints
14:33:43 <kzaitsev_ws> apuimedo: btw here is the project you were referring to https://review.openstack.org/#/c/462453/
14:33:49 <apuimedo> I'll try to dig it out by adding it to devstack optionally
14:34:01 <kzaitsev_ws> at least thats what dims menthined in #sig-openstack
14:34:07 <ivc_> apuimedo isnt kube-dns udp-based?
14:34:12 <kzaitsev_ws> #link https://review.openstack.org/#/c/462453/
14:34:23 <dims> yep, that's the one
14:34:27 <apuimedo> right
14:34:46 <apuimedo> and the idea they have is to start cni/controller from scratch in golang inside openstack
14:34:48 <dims> they seem to be coming in with empty repo, so we should engage quickly so they use what we have here
14:34:49 <kzaitsev_ws> and here's their github https://github.com/hyperhq/kubestack
14:35:21 <dims> kzaitsev_ws : they won't be importing that directly to openstack git repo
14:35:28 <kzaitsev_ws> ah, ok )
14:35:55 <apuimedo> dims: where's the answer to all those questions you sent them?
14:36:48 <dims> apuimedo : i haven't seen an answer yet, i met with "resouer" who was at the boston summit, was promised some write up to send around.
14:37:01 <apuimedo> dims: very well
14:37:07 <apuimedo> at some point maybe there should be a meeting
14:38:05 <dims> apuimedo : ++ i requested them to find/engage with projects we have
14:38:42 <apuimedo> good
14:39:02 <apuimedo> pengfei ni knows kuryr
14:39:18 <dims> resouer is Harry Zhang - https://github.com/resouer
14:39:49 <apuimedo> cool
14:40:23 <dims> that's all i had :)
14:40:30 <apuimedo> thanks a lot dims
14:40:40 <apuimedo> anything else on kuryr-k8s?
14:42:01 <apuimedo> alright
14:42:07 <apuimedo> #topic fuxi
14:42:11 <apuimedo> #chair hongbin
14:42:12 <openstack> Current chairs: apuimedo hongbin irenab
14:42:16 <hongbin> hi
14:42:30 <hongbin> since last week is boston summit, i don't have too much to update
14:42:59 <hongbin> there are a few patches that need reviews as ususal :)
14:43:01 <hongbin> https://review.openstack.org/#/c/461956/
14:43:09 <hongbin> https://review.openstack.org/#/c/461212/
14:43:18 <hongbin> https://review.openstack.org/#/c/454474/
14:43:35 <apuimedo> Looking forward to the plugin api support :P
14:43:47 <hongbin> apuimedo: plugin api?
14:44:23 <apuimedo> docker plugin install kuryr/fuxi
14:44:24 <apuimedo> :-)
14:44:30 <hongbin> apuimedo: i see
14:44:44 <hongbin> apuimedo: we have a bug to track that AFAIK
14:44:56 <kzaitsev_ws> I need to add fuxi to my gerrit review dashboard...
14:45:00 <hongbin> apuimedo: yes, it will be available soon, it si on my priority
14:45:15 <apuimedo> great hongbin!
14:45:27 <apuimedo> anything else hongbin ?
14:45:33 <hongbin> apuimedo: that is all from me
14:45:38 <apuimedo> thanks hongbin
14:45:47 <apuimedo> it was great to meet you again in the summit
14:45:50 <apuimedo> #topic general
14:46:22 <apuimedo> I posted the demos and slides we did in the summit
14:46:30 <apuimedo> and there's also the recordings on the youtube channel
14:46:38 <irenab> apuimedo: link?
14:46:43 <apuimedo> irenab: right
14:47:05 <apuimedo> #link https://www.youtube.com/watch?v=3DSskYvOZvY
14:47:11 <apuimedo> #link https://www.youtube.com/watch?v=gFqfNZ3iBBg
14:47:21 <apuimedo> #link https://www.youtube.com/watch?v=pOOAM06B39s&t=58s
14:47:26 <apuimedo> oops
14:47:30 <apuimedo> #link https://www.youtube.com/watch?v=pOOAM06B39s
14:47:44 <apuimedo> #link https://www.youtube.com/watch?v=6veNkVIAQfs
14:48:06 <apuimedo> #link https://www.youtube.com/watch?v=dZZD0rJjgAo
14:48:55 <kzaitsev_ws> and could you pls share the slides somewhere too? =)
14:49:15 <apuimedo> #link https://www.slideshare.net/celebdor/kuryr-kubernetes-the-seamless-path-to-adding-pods-to-your-datacenter-networking
14:49:23 <apuimedo> #link https://www.slideshare.net/celebdor/openstack-boston-summit-kuryr-project-updates
14:49:29 <apuimedo> #link https://www.slideshare.net/celebdor/kuryr-fuxi-openstack-networking-and-storage-for-docker-swarm-containers
14:49:39 <apuimedo> kzaitsev_ws: there you go ^^
14:49:40 <kzaitsev_ws> awesome, thanks =)
14:49:41 <apuimedo> :-)
14:50:12 <apuimedo> there's also the onboarding slides, but I think we'll keep those on demand
14:50:25 <apuimedo> or when new people show up in the channel :P
14:50:33 <apuimedo> that's all from me today
14:50:36 <apuimedo> anything else anybody?
14:51:32 <irenab> nope
14:51:58 <apuimedo> oh yea. In case somebody needs to do kuryr with pod/container in VM, you can take a look at https://github.com/danielmellado/kuryr_heat
14:52:04 <apuimedo> #link https://github.com/danielmellado/kuryr_heat
14:52:30 <apuimedo> probably that can be a base for magnum support
14:52:39 <apuimedo> oh, only for trunk mode pod-in-vm
14:52:50 <apuimedo> macvlan and ipvlan obviously don't need the trunking
14:54:28 <apuimedo> ok. Thank you all for joining!
14:54:36 <apuimedo> #endmeeting