10:01:16 #startmeeting containers 10:01:17 Meeting started Tue Feb 20 10:01:16 2018 UTC and is due to finish in 60 minutes. The chair is strigazi. Information about MeetBot at http://wiki.debian.org/MeetBot. 10:01:18 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 10:01:20 The meeting name has been set to 'containers' 10:01:36 #topic Roll Call 10:01:43 o/ 10:02:06 hi 10:02:16 hi slunkad flwang1 10:02:33 strigazi: hello 10:03:16 meeting agenda #link https://wiki.openstack.org/wiki/Meetings/Containers#Agenda_for_2018-02-20_1000_UTC 10:03:33 #topic Announcements 10:03:37 None 10:03:43 #topic Blueprints/Bugs/Ideas 10:04:14 Calico network driver for kubernetes https://review.openstack.org/#/c/540352/ flwang1, do you want to log the status? 10:05:20 I'll do it then :) 10:05:23 yep, with strigazi's help, we have tested most of the functions 10:05:27 and it just works 10:05:41 problem found: 10:05:58 calico needed extra mounts for the kubelet container 10:06:23 and coredns needed the pods_network_cidr to be passed in its configuration 10:07:15 kubelet now has an interface to add new mounts 10:07:21 next 10:07:26 Cluster Federation remaining patch for queens https://review.openstack.org/#/c/546093/ 10:07:38 strigazi: yep, i'm going to add another document/note to highlight the calico driver need v1.9.3 kube_tag 10:08:22 yes we need that 10:08:34 1.9.3 or later 10:08:55 yep, would you mind me adding it with a following patch? 10:09:38 it is better to be in this one. since we cherry-pick we double the patches 10:09:47 ok, no problem 10:10:03 I will post a new patchset then 10:10:12 thanks 10:12:31 Cluster Federation remaining patch for queens https://review.openstack.org/#/c/546093/ the feature is implemented server side. We can iterate in this cycle both upstream and downstream and complete the feature in rocky. kubernetes clusters created by magnum are ready to be federated if a user wants to do it manually. 10:13:37 next, 10:13:47 ingress controller https://review.openstack.org/#/c/528756/ 10:14:55 with this patch, traefik is deployed with magnum. Users can then create ingresses specific to their deployement. 10:15:17 strigazi: that's a good one, but i haven't got time to test it 10:15:29 strigazi: do we want to get it in Queens? 10:15:53 flwang1 it works better if the cloud that the cluster runs in has dns for the cluster ndoes 10:16:15 flwang1: yes 10:16:56 next, 10:17:01 k8s: allow passing extra options to kube daemons https://review.openstack.org/#/c/481062/ 10:17:30 We came with this issue for users that need high customization for kubernetes. 10:18:13 The most prominent use case we had was for cluster with over 300 nodes, (we tried up to 800) 10:18:48 there are parameters to tune like --qps which affects the performance of the cluster. 10:18:49 strigazi: wow 10:20:21 sounds interesting, will take a look 10:20:21 nice! 10:20:47 I'm looking for the param documentation, just a moment 10:21:25 eg: --event-qps int32 If > 0, limit event creations per second to this value. If 0, unlimited. (default 5) 10:21:42 this is for kubelet 10:22:48 large clusters are subject to tuning, otherwise their performance is degraded 10:23:11 next, 10:23:16 strigazi: it's a great feature 10:23:44 disable_drivers option https://review.openstack.org/#/c/541663/ flwang1 it's ready this one right? 10:24:07 strigazi: yep, it's ready 10:24:10 I'll test after the meeting 10:24:15 strigazi: great 10:24:16 thanks 10:24:34 I have three more, all for queens :) 10:24:42 Run etcd and flanneld in a system container https://review.openstack.org/524116 10:25:25 can it be tested with fedora 26? 10:25:28 if f27 both components are removed and we need to run them in syscontainers like kube. 10:27:27 strigazi: btw the fedora atomic image link is broken here https://docs.openstack.org/magnum/pike/user/#kubernetes-on-fedora-atomic 10:28:01 flwang1 it could, but it's not quite the same. we are bitten by docker-storage-setup 10:28:19 flwang1: I need to update it properly to work with both 10:28:23 strigazi: ok, then it would be tricky 10:28:35 we will default to overlay2 10:28:52 you mean network driver? 10:29:06 and devicemapper will be adviced to be used only with external volumes 10:29:13 flwang1 not storage 10:29:16 flwang1 no, storage 10:29:20 ok, i see 10:29:45 make sense for me 10:29:59 slunkad: that would be the correct https://dl.fedoraproject.org/pub/alt/atomic/stable/ it changed :( 10:30:28 ah i see 10:31:19 next, 10:31:26 strigazi: btw, as for image, should we update the image in devstack? 10:31:30 from 26 to 27? 10:31:41 yes, when it works :) 10:32:07 and given the k8s 1.7.x will EOL at March, should we update the default version of k8s? 10:32:24 yes, we will default to 1.9.3 10:32:36 in Queens? 10:32:40 yes 10:32:41 is there a patch? 10:33:07 no 10:33:16 mind me proposing? 10:33:21 I don't 10:33:30 cool 10:33:32 go for it :) 10:33:38 will do after this meeting 10:34:17 flwang1: 1.7 is EOL in March? 10:34:27 I though it is one year 10:34:34 strigazi: i will find the link for you 10:37:33 ok, next is the dashboard. 10:37:50 I missed it in the agenda. 10:38:27 I'll update the patch to work with kubernetes 1.9 and RBAC. https://review.openstack.org/#/c/507465/ 10:39:10 strigazi: awesome 10:39:17 And last one, is upgrades. The feature I want to go in as beta. 10:39:59 We need to have a working prototype deployed and finalize the feature in Rocky 10:41:04 I have tested going from 1.7 to 1.8 with success but this path is not enough I would say. 10:41:30 flwang1: how does that sound? 10:41:54 sounds good for me 10:41:57 i'm keen to test it 10:42:26 strigazi: IIRC, you haven't posted your patch on gerrit, yes? 10:42:58 I did, it's incomplete though. https://review.openstack.org/#/q/status:open+project:openstack/magnum+branch:master+topic:bp/cluster-upgrades 10:44:02 sorry, it's on my list actually 10:44:13 after completed the calico work, i will revisit the patch 10:44:25 flwang1 sure, 10:44:31 A note here 10:45:00 Magnum is designed to manage a lot of clusters 10:45:57 do we really need the note ? ;) 10:46:26 it should definitely support A LOT OF clusters 10:46:31 So we make some decisions keeping in mind that the clusters are built to serve almost all use cases 10:46:44 strigazi: +1 10:47:41 applications should be closely attached to nodes 10:48:19 this concept needs some time to grasp that you need to stop caring about the node that your app is running on :) 10:48:37 from my experience is difficult to pass it to new users 10:48:51 that was my note :) 10:50:17 I think we can move to 10:50:17 #topic Open Discussion 10:50:32 slunkad: are you coming in the end at the PTG? 10:50:44 strigazi: yes i will be there 10:50:53 slunkad: great! 10:51:06 strigazi: what was the link to the planning etherpad? 10:51:23 https://wiki.openstack.org/wiki/PTG/Rocky/Etherpads 10:51:50 thanks! 10:52:09 strigazi: thanks for the link though i won't be there 10:52:19 flwang1: which time will you be able to join remotely? 10:53:12 depends on the timezone, i nneed to check 10:53:26 it uk/ireland 10:53:30 it;s uk/ireland 10:53:36 it's uk/ireland :) 10:54:05 i don't mind working at night for one or two days 10:55:36 It will be mostly one I guess, we will see, depends on the attendance 10:56:43 ok 10:56:50 let's wrap this meeting 10:57:01 3' left 10:57:36 Next meeting at the 6th of March 10:57:47 Next week is the PTG 10:57:48 got 10:58:08 #endmeeting