15:59:50 <sdake> #startmeeting kolla 15:59:51 <openstack> Meeting started Wed Jun 22 15:59:50 2016 UTC and is due to finish in 60 minutes. The chair is sdake. Information about MeetBot at http://wiki.debian.org/MeetBot. 15:59:52 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 15:59:54 <openstack> The meeting name has been set to 'kolla' 15:59:56 <sdake> #topic rollcall 16:00:02 <rhallisey> hello 16:00:04 <wirehead_> o/ 16:00:06 <sdake> \o/ containers FTW 16:00:14 <wirehead_> I can hardly contain my excitement. 16:00:20 <sdake> me either wirehead_ 16:00:38 <vhosakot> o/ 16:00:39 <inc0> o/ 16:00:40 <coolsvap> o/ 16:00:51 <britthouser> 0/ 16:01:04 <sdake> #topic announcements 16:01:14 <sdake> midcycle july 12,13th in Ansible HQ 16:01:21 <sdake> #link https://wiki.openstack.org/wiki/Meetings/Kolla 16:01:29 <sdake> I am going to put up an eventbrite page for folks to register 16:01:36 <sdake> please register if you intend to come 16:01:38 <sdake> if you intend to come i'd dbook travel now 16:01:55 <sdake> within 14 days most companies wont approve travel 16:02:00 <wirehead_> yah 16:02:19 <vhosakot> sorry, I'll have to miss the mid-cycle meet :( traveling on work... I'll attend remote thru webex and participate in etherpad 16:02:26 <sdake> #topic refactor build.py 16:02:44 <inc0> ok 16:02:47 <inc0> I guess that's me 16:02:54 <inc0> so guys, build.py is a mess 16:03:11 <sdake> before that, the link for the midcycle is here: 16:03:15 <sdake> #link https://wiki.openstack.org/wiki/Sprints/KollaNewtonSprint 16:03:18 <sdake> sorry about that 16:03:20 <inc0> we all know it, we kinda want to ingnore it but that won't make problem go away 16:03:20 <akwasnie> o/ 16:04:04 <inc0> I was meant to say "show of hands who will help with that" but since akwasnie already volounteered... 16:04:05 <inc0> :P 16:04:14 <sdake> ;-) 16:04:22 <sdake> ya agree build.py is a charlie foxtrot 16:04:45 <sdake> i think we dont know what to do with it inc0 16:04:57 <akwasnie> ;) 16:05:04 <sdake> perhps the way to get started is to brainstorm a list of things we want to fix 16:05:06 <inc0> well there are few things 16:05:23 <inc0> one would be separation of jinja2 generation and dockerfile builds 16:05:39 <inc0> decouple whole giant script into importable submodules 16:06:16 <sdake> i'd like the build.py to be usable as a python library as well ;) 16:06:16 <inc0> I'm still kinda reluctant to copy whole thing to tmp but since we don't have better idea...we might think about it 16:06:25 <inc0> yeah, that's part of it 16:06:29 <inc0> technically it already is 16:06:36 <sdake> right 16:06:49 <inc0> I already moved it out of command 16:06:51 <sdake> well first step is to get a blueprint up 16:07:00 <sdake> second step is to write down list of work items 16:07:02 <inc0> however we still need to go forwards 16:07:22 <inc0> bottom line, is there anyone feeling like leading this effort? 16:07:37 <wirehead_> Well, the “I want to build a pile of templated containers from the CLI” is an interesting and non-Kolla specific issue. 16:08:57 <coolsvap> inc0, if you can have the bp and list of tasks i can help with some tasks 16:09:07 <sdake> one opton people have kicked out is to use ansible to bulid the containers 16:09:21 <vhosakot> inc0: I can help too.. is there a link for the bp? 16:09:26 <sdake> i dont know if that owuld be flexible enough or us 16:09:34 <coolsvap> thats one option i think we should be considering 16:09:37 <coolsvap> atleast PoC 16:09:43 <wirehead_> I think it might be less imposing to first split the pieces up, then find people who can lead each piece. 16:09:59 <inc0> coolsvap vhosakot I'd really like one of you to take lead then;) 16:10:06 <sdake> ya inc0 - first and second steps above, then we can distribute 16:10:23 <inc0> I can contribute ideas, but I would lie to you if I'd say I did enough research whether or not they are any good;) 16:10:30 <Mech422__> I need to add nova-docker to my nova containers - so I could work on something like that? maybe something with the new 'customization' stuff? 16:10:36 <sdake> inc0 i think the problem is coolsvap and vhosakot don't know what work items to write in the blueprint 16:10:55 <inc0> ok, so 3 of us, let's work on it together than ok? 16:11:20 <inc0> I just don't want to dominate this discussion, I just speculate what could be done 16:11:21 <vhosakot> inc0: does refactoring build.py include chaning the way it logs output as well? 16:11:29 <vhosakot> changing* 16:11:37 <sdake> a proper refacor changes no functionality 16:11:38 <vhosakot> sdake: yep, I'll need help writing work items 16:11:39 <inc0> potentially, I know that harlowja already started this work 16:11:41 <sdake> its a restructure of the code 16:11:42 <coolsvap> inc0, wfm 16:12:06 <inc0> sdake, yeah, but fixing what's broken/stupid/could be better could also be good 16:12:29 <sdake> i think separating the jinja2 generaton from the build into modules is a good starting point 16:12:32 <inc0> vhosakot, so how about we'll do good old etherpad brainstorm later? 16:12:41 <sdake> a third module would be building the dependency graph 16:12:57 <inc0> maybe include multiple base images while we're at it 16:12:58 <wirehead_> Easier to first refactor with the understanding of what needs to change, then the change, IMHO. 16:13:03 <inc0> since it seems to be recurring topic 16:13:21 <sdake> i recommend not changing functionality during a refactor 16:13:24 <inc0> wirehead_, yeah, totally agree, let's brainstorm ideas first 16:13:28 <sdake> it makes it harder to keep correct 16:13:31 <vhosakot> yep, we'll add an etherpad to brainstorm the work items first... I have some ideas about avoiding bad cross-builds (building Ubuntu images on CentOS and vewversa) 16:13:35 <vhosakot> vice-versa* 16:13:41 <inc0> sdake, but we can pave work so this feature will be easier to implement 16:14:01 <wirehead_> inc0: we’ve got our own parallel non-openstack-oriented implementation of some of build.py 16:14:06 <sdake> i think hands will be ffull turning this into 3 submodules 16:14:08 <Jeffrey4l> o/ 16:14:16 <sdake> hey Jeffrey4l 16:14:35 <Jeffrey4l> sdake, sorry, i am late. 16:14:39 <wirehead_> inc0: So at the very least, I can see if dcwangmit01_ and I can offer some perspective. 16:14:39 <sdake> ok gotta move on :) 16:14:44 <sdake> #topic do we need tqdm 16:14:44 <inc0> wirehead_, I'd be interested to hear about reasons you made it and improvements 16:15:00 <coolsvap> harlowja, around? 16:15:05 <inc0> tqdm? 16:15:07 <inc0> what's that? 16:15:13 <sdake> no idea what it is or who added to agenda 16:15:19 <wirehead_> https://github.com/noamraph/tqdm 16:15:22 <mandre> inc0: that's to show a progress bar 16:15:23 <vhosakot> I think we should solve bugs like https://bugs.launchpad.net/kolla/+bug/1546652 when re-factoring build.py 16:15:23 <openstack> Launchpad bug 1546652 in kolla "Add "--force-cross-build" argument to kolla-build" [Low,Triaged] - Assigned to Vikram Hosakote (vhosakot) 16:15:23 <coolsvap> its a library included by harlowja in one of the logging reviews 16:15:40 <rhallisey> covered it last week 16:15:44 <coolsvap> yes 16:15:47 <rhallisey> but we could review 16:15:48 <rhallisey> :) 16:15:57 <coolsvap> the review on requirements queue is abandoned 16:16:09 <coolsvap> so unless we want it it will not be restored again 16:16:36 <sdake> ok, so this has been covered rhallisey ? 16:16:48 <rhallisey> ya I mentioned it last week 16:16:53 <wirehead_> Yeah, I am all for friendly logging, but I dont think we had any actionable discussion items. :/ 16:16:55 <sdake> #topic customization of dockerfiles 16:17:15 <inc0> sooo 16:17:27 <inc0> we already have mechanism partially reviewed/merged 16:17:46 <sdake> and by partial he means it does about 25% of what it needs to do :) 16:17:47 <inc0> sdake, you mentioned that you would like to take lead on this? 16:18:06 <sdake> yes of course - as soon as i am unburried from my current work 16:18:12 <sdake> then distribute the work to the community 16:18:24 <sdake> we should be able to knock it out in a week or so once the foundation is done 16:19:16 <mandre> what are the outstanding patches that need to merge? 16:19:25 <sdake> i'm hopeful the core review team can spare 30min-2 hours to port our container sets (40 of em) to this new model 16:19:42 <sdake> before we do that, we need the model to be correct 16:19:44 <sdake> i'll be working on that first 16:19:52 <sdake> mandre they are unwritten as of yet 16:19:52 <Mech422__> inc0: I need to learn the customization stuff - might as well have me do somethign useful while I learn it 16:19:55 <britthouser> I'm game for helping the port sdake. 16:20:12 <sdake> britthouser cool - this work will be easy - look at one docker file make it look like another 16:20:18 <vhosakot> sdake: me too.. will help ports 16:20:19 <inc0> Mech422__, ofc.. 16:20:29 <inc0> lots of work to do;) 16:20:39 <sdake> but there is 130 containers in our tree 16:20:42 <sdake> so it takes awhile to do 16:20:58 <sdake> but the real problem is that work can't begin until i finish the prototype inc0 started 16:21:18 <sdake> so thts where that is 16:21:26 <sdake> any Q? 16:22:00 <sdake> #tiopic non-ini config 16:22:01 <inc0> #link https://review.openstack.org/#/q/topic:bp/third-party-plugin-support 16:22:09 <inc0> ok, that's important 16:22:13 <mandre> sdake: thanks inc0 16:22:25 <inc0> we have merge config for ini 16:22:35 <inc0> but we have no way of allowing customizations of non-ini config 16:22:40 <sdake> #topic non-ini config 16:22:52 <inc0> also as it turns out, even our merge config isn't perfect 16:23:06 <inc0> so I thought of changing way we deploy configs slightly 16:23:19 <vhosakot> merge_config does not work for non-ini config files.. we need to just over-write instead of merging 16:23:21 <inc0> and figure out how to decouple generation of configs from deployment of them 16:23:40 <inc0> like, prep all the configs per hostname 16:23:57 <inc0> and then op will be able to change them to their liking 16:24:10 <inc0> and only then deploy 16:24:17 <inc0> what do you think of such solution? 16:24:19 <sdake> that adds another action 16:24:23 <mandre> inc0: seems like something that will be useful for kolla-kubernetes 16:24:29 <Mech422__> inc0: I like the simplicity 16:24:40 <Mech422__> inc0: just have a tree of configs, edit it, deploy it 16:24:56 <inc0> well, it will be a lot of files tho 16:25:00 <inc0> as we need this per-node 16:25:11 <inc0> anyway, UX will be important in this one 16:25:22 <wirehead_> Note that kolla-kubernetes is anti-hostname. 16:25:26 <inc0> and as sdake said, we might change deploy to actually 2 separate steps 16:25:29 <Mech422__> inc0: yeah - but its conceptually very easy to explain to people "go to configs/node1/etc/globals.yml and edit foo' 16:25:38 <wirehead_> Most of my patches are turning off per-hostname config generation for the kube case. 16:25:54 <inc0> yeah, this will be ansible specific to a point 16:26:14 <inc0> so, do you guys like this approach? 16:26:31 <sdake> i like the merge approach because it has less steps 16:26:36 <sdake> as in one :) 16:26:43 <inc0> sdake, but it's not possible for non-ini 16:26:47 <sdake> i think this is a good midcycle topic 16:26:51 <sdake> right for non-ini ust copy over 16:26:55 <rhallisey> sounds good to me 16:27:03 <inc0> and it won't be possible for non-ini as we don't really want to write parser for every config format there is 16:27:13 <sdake> inc0 i think we all agree on that point 16:27:15 <inc0> ok, we cna talk on midcycle about it 16:27:28 <sdake> #topic gate instability 16:27:59 <sdake> who added to agenda? 16:28:01 <inc0> we still have non-voting gates 16:28:09 <inc0> hold on, looking for link 16:28:37 <vhosakot_> sorry, got disconnected... 16:29:12 <inc0> #link http://lists.openstack.org/pipermail/openstack-dev/2016-June/097401.html 16:29:23 <inc0> we still have non-voting, unstable gates 16:29:26 <britthouser> we just gave you all the AIs vhosakot 16:29:29 <Mech422__> inc0: /me whispers augeas :-) 16:29:52 <sdake> the unstable gates need to be fixed before there is to be eany voting 16:29:56 <inc0> FAILED on gate doesn't mean we didn't do good patch 16:30:04 <vhosakot_> wow ;) 16:30:24 <sdake> at one point the gates were pertty stable 16:30:29 <sdake> and i was almost going to make them voting 16:30:36 <sdake> or atlleast one of them 16:30:37 <coolsvap> inc0, yes but its very unpredictable 16:30:37 <inc0> soo, we *really* need to get to the bottom of problem and fix it 16:30:41 <sdake> now the gates fail all the damn time 16:31:06 <coolsvap> and many time just a couple of rechecks solve the issue 16:31:12 <inc0> if it's infra problem, we need to work with infra to find solution 16:31:14 <sdake> recheck is the wrong answer 16:31:21 <coolsvap> yes 16:31:23 <inc0> yeah, we shouldnt ever recheck 16:31:33 <inc0> recheck means gates are busted 16:31:36 <sdake> 5% recheck ok 16:31:36 <sdake> 50% recheck not ok 16:31:50 <coolsvap> inc0, when we look at logs we know its not related to fix 16:32:06 <sdake> coolsvap yes but that doesn't tell you patch is correct 16:32:08 <inc0> coolsvap, then let's start gathering issues it WAS related to 16:32:12 <sdake> no develoepr trusts the gates 16:32:14 <inc0> and see if we can fix those 16:32:28 <sdake> inc0 good idea 16:32:29 <sdake> moment 16:32:37 <inc0> at some point we want to say "gates failed so I failed" 16:32:49 <coolsvap> inc0, agreed 16:32:57 <sdake> #link https://etherpad.openstack.org/p/kolla-gate-failures 16:33:03 <inc0> after that we make them voting, and after that we start going towards "gates are good so patch is good" 16:33:06 <sdake> lets start first by classifying the fialures 16:33:24 <sdake> inc0 we can't make them voting until we are using mirrors in infra 16:33:29 <inc0> sdake, I don't thinkw e should do this right now, as it will take lots of time to figure out real issue 16:33:43 <sdake> lets classify the issues 16:34:00 <sdake> we dont have to do it during meeting 16:34:06 <sdake> but it needs to be done over time 16:34:29 <inc0> anyway, show of hands, who volounteer for "gate task force" 16:34:31 <inc0> o/ 16:35:14 <inc0> ehh...always alone ;_( 16:35:21 <sdake> before we commit to fixing stuff 16:35:27 <sdake> lets figure out what needs fixing 16:35:46 <inc0> I mean people who will commit time to look at number of logs and figure out together common problems 16:36:07 <sdake> oh, i expect the core team during their reviews to help identify the common problems 16:36:08 <sdake> not debug them 16:36:10 <sdake> but identify them 16:36:20 <sdake> for example, I see sometimes the gate tries to pull from docker.io 16:36:24 <Jeffrey4l> i am always debugging the gate. it is pain without the gate. 16:36:26 <sdake> it should never do that except the centos image 16:37:08 <inc0> and why centos is a special beast? 16:37:20 <coolsvap> inc0, the centos base image ;) 16:37:23 <sdake> ubuntu same thing 16:37:26 <coolsvap> or rather any base image 16:37:36 <sdake> ya centos:7 16:37:39 <sdake> or ubuntu:16.04 16:37:46 <Jeffrey4l> shouldn't we never pull the images from docker.io? 16:37:49 <sdake> that is the only thing that hsould be coming from docker.io 16:37:56 <Jeffrey4l> i mean the kolla images. 16:38:04 <britthouser> There is a setting in docker config to block docker.io 16:38:06 <sdake> Jeffrey4l right - but i see gat efialures where the gate is tring to pull kolla images from docker.io 16:38:14 <Jeffrey4l> yes. 16:38:20 <sdake> britthouser we need docker.io for the base image 16:38:36 <sdake> anyway that is just one i saw of many 16:38:40 <sdake> rabbitmq implodes 16:38:42 <sdake> often 16:38:45 <inc0> guys, back to the topic at hand, let's not look for solution here, just figure out process of finding problems 16:38:47 <Jeffrey4l> i think there may be something wrong in kolla_docker script. 16:38:49 <britthouser> oh....so it can't find the image locally, so it starts searching. 16:38:57 <sdake> the pocess of finding the problems is as follows: 16:39:12 <sdake> 1. when reviewing, look at the failed gate - and provide a link to the fialure in the etherpad 16:39:14 <inc0> so we can use etherpad as scratchpad and add bugs 16:39:31 <sdake> in 1-2 weeks we can come back and have lots of data to work with 16:39:41 <sdake> and actually solve it a person per gate problem 16:39:52 <sdake> any disagreement? 16:40:19 <inc0> let's use specific tag for gate-related bugs 16:40:20 <Jeffrey4l> agree. 16:40:27 <coolsvap> nope looks like a good plan 16:40:30 <inc0> and submit bugs for every issue we see 16:40:36 <sdake> lets not even file bugs yet 16:40:42 <sdake> we have over 60 bugs in new state 16:40:53 <sdake> filing bugs ins't helpful because they aren't being managed effectively by the core team 16:41:00 <sdake> and I really can't do it alone 16:41:02 <inc0> that's different issue 16:41:06 <sdake> project too big 16:41:12 <sdake> i realize its a different issue 16:41:21 <inc0> let's manage them well;) 16:41:24 <sdake> however filing bugs compounds it 16:41:49 <inc0> then let's not avoid filing, let's manage these bugs well so it wont compound this issue 16:41:58 <sdake> its hard to tell the noise from th stuff we need to look at immediately 16:42:00 <inc0> and in the process learn to manage rest of bugs well too;) 16:42:16 <sdake> i intend to teach people at midcycle how to triage properly 16:42:36 <sdake> anyway moving on :) 16:42:40 <coolsvap> i agree with sdake every morning i look at bugs i find more bugs to update things than look at them for resolution 16:42:43 <britthouser> seems like summarizing gate issues into buckets, and file a bug for each bucket would keep the signal-to-noise ratio down? 16:43:12 <sdake> britthouser right thatis the purpose of the etherpad - t o determien a list o critical bugs we can track as gate problems 16:43:17 <inc0> britthouser, yeah, we just need good process to determine buckets;) 16:43:33 * britthouser likes to state the obvious I guess 16:43:41 <sdake> softball ftw ;) 16:43:49 <sdake> #topic kolla-kubernetess 16:43:55 <inc0> so bottom line, look at suspiciously failed gate logs and see what's wrong 16:44:10 <sdake> i'd recommend not diagnosing 16:44:15 <sdake> unless your keen to do so 16:44:24 <sdake> diagnosing a gate bug takes 2-3 weeks 16:44:38 <sdake> lets figure out which ones we have 16:44:44 <rhallisey> I've been looking at ansible for deployment of kolla-kube. The ansible kubernetes module needs a few patches which I'm almost done with 16:45:36 <sdake> rhallisey please link the specification 16:45:38 <rhallisey> doing ansible for deployment would take some of the load off of kube to handle deployemnt 16:45:46 <rhallisey> didn't make a spec 16:45:48 <rhallisey> it's just a poc 16:45:57 <sdake> nah i mean the k8s + olla spec 16:46:01 <rhallisey> k 16:46:10 <rhallisey> https://review.openstack.org/#/c/304182/ 16:46:14 <sdake> ok cats 16:46:19 <sdake> need +2s on the spec please 16:46:20 <inc0> rhallisey, so what do you think? does it seem like a good idea at all? 16:46:30 <rhallisey> 9 +2s is enought to +a 16:46:37 <sdake> oh are there 9 16:46:41 <sdake> wow you guys move fast :) 16:46:56 <rhallisey> inc0, well, it beats python + etcd + bash scripts 16:47:06 <sdake> ok w+1 :) 16:47:06 <inc0> rhallisey did good job about this one;) I +2 without reading;) (j/k) 16:47:08 <rhallisey> the downside would be that kube doesn't not have as much control 16:47:32 <inc0> does it need this kind of control tho? 16:47:33 <rhallisey> inc0, :) 16:48:04 <rhallisey> inc0, need, not really. But it would make use of the native upgrade workflow 16:48:25 <rhallisey> but even with the native upgrade workflow, it wouldn't be possible to upgrade openstack without life cycle hooks 16:48:26 <sdake> i don't think the native upgrade workflow actually works in our use case 16:48:28 <inc0> well I'm afraid that you won't be able to do all the upgrades with k8s native stuff 16:48:31 <sdake> but that is down the road imo 16:48:33 <rhallisey> which dont' exists atm 16:48:55 <inc0> anyway, I'm happy to help with figuring upgrades out 16:48:59 <rhallisey> inc0, we coudl still do the upgrades 16:49:05 <rhallisey> it would just be ansible driven 16:49:07 <inc0> one way or another 16:49:11 <wirehead_> Yeah, I’m not entirely sure where the k8s community is going to head as far as upgrades go. 16:49:28 <wirehead_> So far, the feeling I’ve been getting is that the community doesn’t see it as part of core. 16:49:54 <rhallisey> wirehead_, I expect them to build in more advanced life cycle hooks so that you can actaully do some stuff in the middle of scaling down/up the containers 16:49:55 <wirehead_> So we might be stuck with our upgrades regardless. 16:50:00 <wirehead_> Yeah. 16:50:08 <rhallisey> but we could be waiting a long time 16:50:11 <rhallisey> ~9 months 16:50:12 <wirehead_> Yeah. 16:50:16 <inc0> we can gradually remove ansible from it 16:50:23 <rhallisey> inc0, that's also true 16:50:25 <pbourke> that just seems bizzare 16:50:33 <rhallisey> pbourke, what's that 16:50:34 <pbourke> kubernetes literally has no concept of upgrades? 16:50:36 <sdake> pbourke shaking my head too 16:50:46 <rhallisey> pbourke, it does, but not to the level we need it 16:50:47 <wirehead_> Also, there’s weirdness at the moment with the kolla-kubernetes CLI and how some parts of it are very Ansible and other parts of it aren’t. 16:50:48 <inc0> pbourke, upgrades -> scale down and scale up 16:50:49 <inc0> that's it 16:50:50 <sdake> pbourke it does but they lack flexilbity 16:50:55 <pbourke> inc0: that's not really upgrades though 16:50:58 <rhallisey> pbourke, it's exactly like inc0 said it 16:51:04 <inc0> true...but that's what it is 16:51:23 <wirehead_> I think a lot of the early adopters have made their systems less reliant upon harsh database upgrades. 16:51:35 <wirehead_> So they just plain don’t need an OpenStack-styled upgrade. 16:51:51 <inc0> yeah, container apps are different 16:52:00 <rhallisey> I highly doubt anyone in the kube community has upgraded an 'application' as complex as openstack that needs to be cluster aware 16:52:06 <inc0> we have to deal with not exactly containerish apps tho 16:52:08 <rhallisey> the killer here is cluster aware 16:52:19 <inc0> yeah, kinda 16:52:29 <sdake> topic change 16:52:31 <rhallisey> we either build that awareness into ansible or into the containers 16:52:36 <inc0> so bottom line, I'd rather see proper upgrades than true to k8s 100% 16:52:39 <sdake> networking security in kubernetes 16:52:51 <sdake> i see this 0.0.0.0 bind all going on 16:53:01 <sdake> i think that may be secure in a kubernetes world 16:53:20 <sdake> but it sure does make it easy for data to leak to the wrong location 16:53:21 <wirehead_> It’s more-or-less the only way to fly in the kube world. 16:53:26 <inc0> well, it will be 0.0.0.0 without net host sdake 16:53:30 <inc0> so it is secure really 16:53:36 <rhallisey> ya there's no net=host 16:53:42 <inc0> 0.0.0.0 is problem with net=host 16:53:55 <sdake> i get it 16:54:08 <sdake> but the management network should be different from the api network for example 16:54:16 <sdake> does k8s offer that kind of functionality? 16:54:21 <inc0> mgmt==api 16:54:27 <rhallisey> we can't use net=host for multinode because say if two of the same pods end up on the same node. Bad news 16:54:30 <sdake> right - thats a security problem 16:54:33 <inc0> data plane is different, but that won't be affected by k8s 16:55:04 <wirehead_> So, right now, none of the services are exposed outside of the k8s cluster. 16:55:26 <wirehead_> We need to create Ingress resources to expose the API services. 16:55:37 <inc0> sdake, I think we need more knowledge to determine security of k8s tho 16:55:38 <sdake> would someone write up state of the union on security in k8s with the work going on 16:55:41 <sdake> i'd like to understand it 16:55:50 <sdake> because 1 of 2 things that consistently kills projects is poor security 16:56:05 <rhallisey> wirehead_, doc? :) 16:56:25 <wirehead_> Sure. 16:56:30 <sdake> cool 16:56:33 <rhallisey> I can start a doc and we can tackle it together if you like 16:56:47 <sdake> specifically I'd like to understand how if i have access to the api network i can't get at the management network :) 16:57:05 <inc0> sdake, api network == public endpoints? 16:57:21 <wirehead_> So, we’re going to need to write that up for the docs regardless because you need to be careful about a few configuration details in the everything-in-one-kube-cluster model. 16:57:38 <sdake> inc0 yup 16:57:51 <Jeffrey4l> if there is net=host in k8s, what the openvswitch_db/libvirt will be like? 16:58:10 <inc0> Jeffrey4l, I don't think we can use k8s for neutron services 16:58:12 <rhallisey> they will be net=host I believe 16:58:12 <inc0> or nova-compute 16:58:18 <Jeffrey4l> ok. 16:58:19 <inc0> neutron agents * 16:58:25 <rhallisey> all compute service will be anchored too 16:58:32 <inc0> not in it's usual way anyway 16:58:39 <wirehead_> net=host services can access kube services. 16:58:48 <Jeffrey4l> got it . 16:58:51 <wirehead_> It’s just you can’t expose an API that uses net=host as a kube service without work. 16:59:14 <sdake> ok meeting time is up 16:59:21 <inc0> thanks folks 16:59:23 <rhallisey> thx 16:59:26 <sdake> hvee to overflow - apologies for no open discussion 16:59:28 <wirehead_> Oh, and I’ve got it to a demoable state. 16:59:28 <sdake> #endmeeting