14:05:11 <apuimedo> #startmeeting kuryr
14:05:12 <openstack> Meeting started Mon May 22 14:05:11 2017 UTC and is due to finish in 60 minutes.  The chair is apuimedo. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:05:13 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
14:05:15 <openstack> The meeting name has been set to 'kuryr'
14:05:27 <apuimedo> Welcome everybody to this slightly delayed kuryr meeting
14:05:31 <apuimedo> who's here today?
14:05:33 <mchiappero> o/
14:05:36 <janonymous> o/
14:05:37 <vikasc> o/
14:05:37 <dmellado> o/
14:05:37 <ltomasbo> o/
14:05:39 <irenab> hi
14:05:41 <ivc_> o/
14:05:43 <kzaitsev_ws> o/
14:05:55 <irenab> apuimedo: will have to leave in 15 mins, sorry
14:06:20 <garyloug> o/
14:06:48 <apuimedo> #topic kuryr-libnetwork
14:06:59 <danil> o/
14:07:41 <apuimedo> #info There's now a docker plugin for devstack. We should move kuryr-libnetwork to use it in devstack so we do not conflict when being used with other services that depend on docker like zun
14:08:00 <dmellado> +1
14:08:20 <dmellado> are there any another recent changes in devstack that could raise
14:08:23 <dmellado> etcd3?
14:08:35 <apuimedo> dmellado: that is not merged yet
14:08:49 <dmellado> but we should be aware of it
14:09:17 <kzaitsev_ws> #link https://review.openstack.org/#/c/445432/
14:09:34 <apuimedo> sure
14:09:34 <kzaitsev_ws> for the logs )
14:09:38 <apuimedo> thanks kzaitsev_ws
14:09:44 <apuimedo> anything else on kuryr-libnetwork front?
14:09:49 <dmellado> thanks kzaitsev_ws ;)
14:11:20 <apuimedo> alright, moving on then
14:11:34 <apuimedo> #topic kuryr-kubernetes
14:11:44 <apuimedo> (we'll do fuxi once hongbin logs in)
14:12:09 <apuimedo> #info devstack is now based on k8s 1.6.2
14:12:27 <vikasc> apuimedo, why not 1.6.3? :)
14:12:43 <apuimedo> vikasc: I heard it had some regressions. Once we test with 1.6.3, we can move to that
14:12:57 <vikasc> cool
14:13:08 <vikasc> thanks apuimedo
14:13:12 <kzaitsev_ws> #link https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG.md#v163
14:13:15 <dmellado> vikasc: you can test that by putting KURYR_HYPERKUBE_VERSION="v1.6.3" on your local.conf ;)
14:13:31 <apuimedo> #info We should be moving to kubeadm instead of hyperkube due to the latter being on life support
14:13:47 <apuimedo> #link https://blueprints.launchpad.net/kuryr-kubernetes/+spec/kubeadm-based-devstack
14:13:49 <vikasc> apuimedo, +2
14:13:53 <kzaitsev_ws> ok. I've drafted my SRIOV spec and published both spec and code
14:13:55 <vikasc> oops +1
14:14:00 <vikasc> :)
14:14:05 <kzaitsev_ws> #link https://review.openstack.org/#/q/topic:bp/kuryr-kubernetes-sriov-support
14:14:28 <kzaitsev_ws> would be happy to get some reviews from you folks =)
14:14:29 <dmellado> +2A
14:14:31 <dmellado> xD
14:14:34 <apuimedo> I started work on that. The biggest issue right now is that the openstack Docker plugin installs the latest docker and kubeadm is not advised with anything older than 1.12 IIRC
14:14:42 <vikasc> kubeadm seems to be having some issues lately.
14:15:07 <vikasc> like changing network_args
14:15:15 <apuimedo> In order to make openstack docker devstack plugin nicely configurable, I added created an issue in k8s
14:15:19 <apuimedo> #link https://github.com/kubernetes/kubernetes.github.io/issues/3866
14:15:30 <apuimedo> vikasc: what kind of change?
14:15:55 <apuimedo> #info sriov spec submitted for discussion
14:16:24 <kzaitsev_ws> the 2 most controversial points in the spec are a separate handler for SRIOV requests (this one I feel is a good idea) and separate cni (this one I'm not sure =)).
14:16:25 <vikasc> apuimedo, not sure if 1.6.3 has this problem or not, but on 1.6.2 NETWORK_ARGS needs to be removed from kubelet config to get node READY
14:16:34 <apuimedo> kzaitsev_ws: I think that the CNI part should just be another handler of the same CNI (just like in the controller it is second handler for vif)
14:16:40 <vikasc> otherwise kubelet keeeps waiting for cni
14:17:02 <apuimedo> vikasc: I don't recall having to do that on my home setup
14:17:05 <apuimedo> but I'll check
14:17:12 <kzaitsev_ws> apuimedo: since I tend to agree, will update spec and code accordingly.
14:17:14 <vikasc> apuimedo, let me search link
14:17:17 <vikasc> wait
14:17:20 <apuimedo> thanks kzaitsev_ws
14:17:22 <dmellado> vikasc: didn't happen to me using 1.6.2
14:18:10 <apuimedo> #info macvlan already gained +1 from two cores irenab and apuimedo
14:18:14 <kzaitsev_ws> vikasc: deployed a devstack last friday with 1.6.2 and it worked out of the box. although I might have not tested well enough..
14:18:18 <apuimedo> #link https://review.openstack.org/#/c/440669/
14:18:33 <apuimedo> kzaitsev_ws: vikasc means with kubeadm, not hyperkube
14:18:40 <kzaitsev_ws> oh ok
14:18:48 <vikasc> apuimedo, https://github.com/kubernetes/kubernetes/issues/43815
14:19:02 <apuimedo> mchiappero: both irenab and I would like to run it first, so it would be good if the patch included a readme section on setting it up
14:19:23 <apuimedo> because we would surely figure it out. Others may not :P
14:19:28 <dmellado> oh, got it
14:19:35 <mchiappero> apuimedo: sure, I think garyloug will complete the patch, maybe the documentation will be a follow up, would that be ok?
14:19:35 <dmellado> I was thinking about hyperkube as well
14:20:05 <apuimedo> mchiappero: Just a bit of README in this patch. Proper documentation in a follow-up
14:20:08 <dmellado> mchiappero: I'd like to run it as well, how about putting a depends-on patch with some quick docs?
14:20:18 <mchiappero> ok, got it!
14:20:25 <dmellado> heh, apuimedo was faster xD
14:20:32 <apuimedo> dmellado: and more lenient
14:21:19 <ltomasbo> I would like to receive some reviews on the ports-pool patches
14:21:25 <apuimedo> #info we're working on getting the pluggable (driver/handler) config out of the config module
14:21:27 <apuimedo> #link https://review.openstack.org/#/c/450113/
14:21:38 <apuimedo> I think we should get that one merged
14:21:42 <ltomasbo> I already got a few +1s from irenab, and I'll be adding unittest this week
14:21:50 <apuimedo> and then work in a nicer way
14:21:56 <apuimedo> ltomasbo proposed the nova approach
14:22:15 <apuimedo> I'm not over the moon with it because they just put everything in a config package
14:22:29 <apuimedo> and I'd rather have the config live with the pluggable code
14:22:32 <apuimedo> but we'll see
14:22:54 <apuimedo> ltomasbo: put a link to the patch series ;-)
14:23:02 <apuimedo> (ports pool)
14:23:08 <ltomasbo> https://blueprints.launchpad.net/kuryr-kubernetes/+spec/ports-pool
14:23:15 <ltomasbo> this is the blueprint ^
14:23:21 <kzaitsev_ws> There's one more thing we're working on here.
14:23:23 <ltomasbo> covering all the patches (plus devref)
14:23:49 <kzaitsev_ws> danil: is working on making it possible to request specific IP/Mac for Pods in kuryr-k8s, much like from neutron
14:24:05 <apuimedo> #link https://blueprints.launchpad.net/kuryr-kubernetes/+spec/ports-pool
14:24:22 <kzaitsev_ws> this is mostly required by our team, that would use SRIOV
14:24:24 <apuimedo> kzaitsev_ws: what kind of annotation does he have in mind?
14:25:02 <kzaitsev_ws> and frankly I'm not 100% sure it's a good idea after all (since it would mean you will not be able to scale deployments easily)
14:25:41 <apuimedo> kzaitsev_ws: well, this should probably apply to plain pod objects, which are not scalable anyway, are they?
14:26:22 <apuimedo> kzaitsev_ws: The clear workaround is to create the port manually and place the vif annotation already on the pod definition you send to 8s
14:26:24 <apuimedo> *k8s
14:26:42 <kzaitsev_ws> well currently we're using an annotation to Pods: a list of objects ip/macs for a specific physnet
14:26:51 <apuimedo> but having an annotation for the port uuid could be good
14:27:15 <apuimedo> could it be done based on ports?
14:27:52 <kzaitsev_ws> hm. you mean ask for a specific port instead of id.
14:28:07 <kzaitsev_ws> ugh
14:28:16 <kzaitsev_ws> ask for port-id instead of mac/ip address
14:28:36 <apuimedo> yes
14:28:39 <ivc_> apuimedo kzaitsev_ws +1 on port id
14:28:40 <apuimedo> it may be easier
14:28:49 <kzaitsev_ws> there's not much difference though (except that port-ids are harder to write)
14:29:00 <ivc_> that also allows more cases than just ip/mac to be handled that way
14:29:20 <kzaitsev_ws> ivc_: thats a valid point
14:29:23 <apuimedo> kzaitsev_ws: ips can be ambiguousu
14:29:30 <apuimedo> *ambiguous
14:30:35 <kzaitsev_ws> and asking for a port would also be in-line with how you boot a nova sriov instance. (at least how I booted it)
14:30:48 <apuimedo> :-)
14:31:09 <apuimedo> I would also like to ask for reviews/comments to https://review.openstack.org/#/c/466291/
14:31:16 <apuimedo> I see irenab posted some already :-)
14:31:24 <kzaitsev_ws> yeah, that sounds like a good idea to me. danil?
14:31:39 <apuimedo> It is something that makes my life easier. So I was wondering if people would find it useful too
14:31:51 <apuimedo> otherwise I'll spin it off in another repo
14:31:52 <danil> yes, me too
14:32:00 <apuimedo> great danil
14:32:04 <apuimedo> thanks for the work on this!
14:32:20 <Irenab_> I will check
14:33:05 <kzaitsev_ws> the work is currently based on sriov patches, but we'll probably update it to work with general vifs
14:33:54 <apuimedo> #info vikasc has been working on making kuryr-k8s deployable by kubernetes
14:33:56 <apuimedo> #link https://review.openstack.org/#/c/466675/
14:34:56 <apuimedo> This patch ^^ makes kuryr-cni a binary  without deps so that it can easily be copied from a daemonset to the CNI bin dir
14:35:27 <apuimedo> Anything else on kuryr-k8s?
14:37:31 <apuimedo> hongbin: I'm trying to make the devstack docker plugin source configurable
14:37:42 <apuimedo> so that we can change which repo gets it from
14:37:44 <hongbin> o/
14:37:54 <apuimedo> right now it only installs latest
14:38:07 <apuimedo> and if not pinning, at least repo configuration is necessary
14:38:19 <hongbin> apuimedo: you want to propose a patch to the plugin?
14:39:15 <hongbin> apuimedo: i like the idea :)
14:39:17 <apuimedo> yes. I'll try
14:39:25 <hongbin> ack
14:39:34 <apuimedo> #topic fuxi
14:39:38 <apuimedo> #chair hongbin
14:39:39 <openstack> Current chairs: apuimedo hongbin
14:39:49 <hongbin> hi all
14:40:12 * hongbin is finding what happened last week
14:40:40 <hongbin> in last week, we merged a similar fix for the breakage cuased by systemd: https://review.openstack.org/#/c/461956/
14:40:41 <apuimedo> :-)
14:40:47 <apuimedo> good
14:41:15 <hongbin> in addition, there is a patch in review queue for switchin g to devstack docker plugin: https://review.openstack.org/#/c/461212/
14:41:22 <hongbin> that is all from my side
14:42:04 <hongbin> apuimedo: ^^
14:42:50 <apuimedo> good
14:42:56 <apuimedo> thanks hongbin
14:43:02 <apuimedo> #topic general
14:43:14 <apuimedo> Anybody has any other topic?
14:43:25 <apuimedo> s/topic/item/ ?
14:44:56 <kzaitsev_ws> not me =) looks like I've mentioned everything I had =)
14:45:09 <Irenab_> Me neither
14:45:54 <apuimedo> very well. Thank you all for joining!
14:46:02 <apuimedo> #endmeeting