16:31:33 #startmeeting kolla 16:31:34 Meeting started Wed Dec 23 16:31:33 2015 UTC and is due to finish in 60 minutes. The chair is inc0. Information about MeetBot at http://wiki.debian.org/MeetBot. 16:31:35 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 16:31:38 The meeting name has been set to 'kolla' 16:31:43 #topic rollcall 16:31:48 o/ 16:31:50 o/ 16:31:54 hello everyone, show of hands please 16:31:58 o/ 16:31:58 hi once again:) 16:32:27 0/ 16:32:45 #topic agenda 16:32:53 nothing in agenda;) 16:32:57 this will be short one 16:33:17 #topic open discussion 16:33:31 https://review.openstack.org/#/c/260389/ here 16:33:50 yep, the first free topic i wanted to discuss 16:33:55 is the migration to oslo.config 16:34:02 Jeffrey4l did the awesome work on that 16:34:12 the question is whether anyone is ok with that 16:34:28 use oslo.config 16:35:03 I kinda wonder if that would not be overkill 16:35:33 argparse is part of stdlib, I'd like to minimize impact required for running kolla 16:35:50 as the result of talk with nihilifer and mandre today,it is break the back compatibility now. 16:36:09 well, we require a lot of things from pip already 16:36:23 so i don't consider adding oslo.config dep a big issue 16:36:34 oslo.config is a "stdlib" for openstack project. 16:36:41 nihilifer, what besides docker-py and ansible? 16:36:42 but Jeffrey4l reminded one more problematic thing 16:36:45 almost all the project useing it. 16:37:09 inc0: Jinja2, pbr, gitdb, git, six 16:37:14 maybe not a lot 16:37:30 jinja2 comes with ansible 16:38:19 oslo assumes that you have most of global requirements installed, and they will freely add new dependencies 16:38:40 o/ 16:38:44 hey SamYaple 16:38:50 and unless there is real benefit to it, I don't really want to add dependencies to deployment host 16:38:53 o/ 16:39:08 small footprint on hosts is one of our values 16:39:09 inc0 : we do try to hold the line with basic libs like oslo.config and oslo.log :) 16:39:29 dims_, but thing with kolla is - we're kinda different 16:39:37 hi dims_, good to have you here :) 16:39:37 inc0 : that's fair 16:39:46 id be ok with oslo.config since we have the proposal bot in place already 16:39:49 we require like 5 python libs now 16:39:57 not pushing for oslo.config, just making sure we understand concerns 16:40:15 right now it's discussion oslo.log vs python stdlib 16:40:25 oslo.config* 16:40:37 inc0 : yep, no issues using python stdlib, it's all good 16:40:39 unless there is a limitation of the standard lib, i dont see a need to change 16:40:41 oslo.config will introduce two libs more 16:40:47 stevedore and netaddr 16:41:00 I'm not against it, don't get me wrong, just thinking what's real benefit, remember it's really simple python script we're talking about 16:41:03 nihilifer : we are trying to remove netaddr 16:41:13 dims_: oh, ok, nice 16:41:23 ew yea not a fan of netaddr 16:41:27 blocked on some pip work with extra dependencies 16:41:29 would be great to remove that 16:41:30 nihilifer, and stevedore has it's own deps 16:41:39 all I'm asking is...what's the reason for change? 16:41:50 apart from "openstack uses it" 16:41:51 inc0 : +1 good question, 16:42:00 ok, so closing dep things https://github.com/openstack/stevedore/blob/master/requirements.txt 16:42:07 it uses what we already have 16:42:11 going to reasons 16:42:27 the first nice feature, autogenerating sample configs from code 16:42:36 oslo.config handle thing graceful. 16:42:50 so it requires less effort from developer introducing new setting 16:42:56 second nice thing 16:43:05 it already discovers where the config is 16:43:27 whether it's /etc/ or somewhere near project cloned from git 16:43:30 are you suggesting we don't need to have /etc/kolla as a hard-unchanging-path? 16:43:39 yep 16:43:44 i like both of these things 16:43:55 this is worth pursuing for either of those reasons 16:43:56 * dims_ has to run. can take questions later this afternoon on #openstack-oslo :) 16:43:56 and we can get rid of all code in kolla dealing with directory discovery 16:44:04 thanks dims_ ! 16:44:09 tha'ts the main reasom i introduced it in kolla-mesos 16:44:13 thanks dims_ 16:44:24 we didn't want to write a code checking paths 16:44:34 ok, you got me convinced 16:44:51 ive recently made alot of effort to remove the path things in ansible with the "simmplfy config" patch 16:45:08 so with oslo.config we can most likely remove the hard dep on /etc/kolla 16:45:12 for kolla-ansible 16:45:14 yeah, I don't like us having hardcoded config path as well, we should gradually get rid of it 16:45:24 for example, we could try to get rid of using something like that https://github.com/openstack/kolla/blob/master/kolla/cmd/build.py#L242 16:45:27 same for build.py 16:45:40 nihilifer: i dont even have to look it up, i know that section 16:45:45 i was thinking about it as well 16:45:58 ok, let's do this, thanks for clarification guys 16:46:13 +2 for proceeding from me 16:46:13 ok 16:46:22 there is a second thing in oslo.config 16:46:25 we must discuss 16:46:39 when you register cli argument "foo" 16:46:45 and it's a boolean setting 16:46:53 oslo.config automatically defines --nofoo 16:46:57 aha! first make them agree then push forward! well played sir 16:47:00 as a logical negation 16:47:06 nice 16:47:15 it will change the --no-cache arg then 16:47:31 Jeffrey4l's concern was 16:47:36 whether changing --no-cache 16:47:39 to --cache 16:47:46 and automatically generated --nocache 16:47:51 is fine for you 16:47:59 as long as the default is unchanged, i think its fine 16:48:01 it will break a backwards compatibility 16:48:03 --nocache != --no-cache 16:48:17 but imo it's not huge change and benefits are better 16:48:17 nihilifer: not necessarily, we can put in --no-cache and deprecate it 16:48:20 yep. it break the compatibility. 16:48:35 SamYaple: --no-cache will be negated as --nono-cache 16:48:38 if we add --no-cache option to oslo.config 16:48:38 that's the thing 16:48:41 nihilifer: i know 16:48:49 so we deprecate it 16:49:01 it will create --nono-cache right? but old --no-cache works 16:49:05 a little migration cruft we cary for a cycle 16:49:33 inc0: yes, it will create --nono-cache and there is no way to disable such a behavior in oslo.config 16:49:48 So, we will have two option? --no-cache and --cache ? with the automatically inversed `--nono-cache` and `--nocach` ? 16:49:51 nihilifer, so if we do that and deprecate --no-cache 16:50:04 Jeffrey4l: yea 16:50:05 it will be our migration cruft 16:50:16 we deprecate --no-cache and remove it in N 16:50:20 ok 16:50:21 I agreed. 16:50:30 good to heat that you're fine with that 16:50:39 oslo.config also support the deprected thing. 16:50:47 yup :) 16:51:01 with the removal of /etc/kolla we can go full non-root 16:51:10 another thing is that `--template` options. it is a time to remove? 16:51:15 yep 16:51:26 Jeffrey4l: i mean we can 16:51:30 yeah, we already removed non-templated dockers 16:51:30 but it hasn't been a cycle 16:51:55 so we should do it later? right? 16:51:55 it was something that was only temp in the liberty branch anyway 16:52:12 i dont think it really matters Jeffrey4l, we can leave it to be safe but if its a problem remove it 16:52:46 OK. got it. 16:52:48 I'm ok with just deleting it 16:52:54 me too 16:53:08 it doesn't work anyway 16:53:09 its been a dead option for longer than it was a working one 16:53:14 yup 16:54:00 and it was broken in 1.0 so unless you have like mid-liberty stuff running (and I'm talking about 1 month time window) you're ok 16:54:08 and if you do, well, sorry to hear that 16:54:29 i think we can move on, whats on the docket next? 16:54:31 there is another back compatility. the [kolla-build] section should be renamed to [DEFAULT]. 16:54:35 it that ok for you? 16:54:42 oh hmm 16:54:49 so, config changes are different 16:54:50 can we somehow make that compatible? 16:55:05 i know oslo supports options from different sections 16:55:10 I'm -1 unless it's backward compatible 16:55:16 hmmm... we may create section [kolla-build] using oslo.config 16:55:21 agreed, we do have to be backwards compatible 16:55:23 and just leave [DEFAULT] empty 16:55:31 that's what we're doing in kolla-mesos 16:55:35 nihilifer: thats fine 16:55:47 we may can use deprecate session feature. 16:55:58 we could also do that, then move the options and deprecate the old spots 16:55:58 Jeffrey4l: no need for that 16:56:10 we can just create cfg.OptGroup('kolla-build') 16:56:11 s/session/section/ 16:56:14 and put all stuff there 16:56:25 there is really no wrong answer here as long as its backwards compatible 16:56:29 nihilifer, the cli will add a `--sectionname-opts` 16:56:47 Jeffrey4l: ah, you're right, sorry... 16:57:21 you can move section options though, so we can deprecate all the options out of [kolla-build] 16:57:39 ok, so i'm +1 for deprecating 16:57:43 http://paste.openstack.org/show/482506/ here is the reaon. 16:57:48 s/reaon/reason/ 16:58:25 Jeffrey4l: yep, that's right 16:58:38 yea and we can deprecate that stuff out in N 16:58:49 we just have a waiting game on our hands 16:58:58 ok, so anyone is against making a deprecated [kolla-build] section and putting the stuff in [DEFAULT] too? 16:59:21 *deprecating* means it is works in both [DEFAULT] ans [kolla-build]. And should be remove in the future. 16:59:25 right? 16:59:28 yes 16:59:29 yes 16:59:31 OK. 16:59:38 I think this is fine. 16:59:47 agreed 17:00:20 cool. anything else to discuuss about oslo.config? 17:00:35 i have no. 17:00:41 Jeffrey4l: are you taking on the work? 17:01:13 Jeffrey4l, what do u mean? 17:01:26 Jeffrey4l: whether you'll continue work on it :) 17:01:27 will you do this we just talked about?;) 17:01:38 I think he is asking himself 17:01:42 nihilifer, yes. 17:01:52 ok, thanks! 17:01:57 cool 17:01:57 awesome then I have no futher questions your honor 17:02:05 I will push a commit soon. 17:02:15 so, may I take over? 17:02:21 next thing i'd like to discuss 17:02:32 ok go on nihilifer 17:02:54 in kolla-mesos we're currently using docker images from here https://hub.docker.com/r/mesoscloud/ 17:02:57 to set up mesos 17:03:10 these images are outdated in terms of i.e. version of marathon 17:03:26 i took a look on images from https://hub.docker.com/r/mesosphere/ 17:03:39 but one of them has no open repo available... 17:04:00 and mesos from mesosphere images doesn't work with docker 17:04:03 so... 17:04:13 i'd like to create own dockerfiles 17:04:20 I'm ok with that 17:04:23 +2 17:04:26 and i'm thinking about putting them into main kolla repo 17:04:29 you can deploy them with kolla-ansible ;) 17:04:39 and to build them with kolla-build 17:04:53 nihilifer: i am under the personal opinion that all docker files should go in kolla main repo 17:05:11 SamYaple: i have same opinion 17:05:13 I'm ok with having our own mesos dockerfiles 17:05:14 all deploy tools (kolla-ansible kolla-mesos kolla-kubernetes) should not have dockerfiles 17:05:21 we can control version pins and everything 17:05:43 cool 17:05:59 yeah we might want to deploy mesos with ansible, this wouldn't be that bad really 17:06:17 inc0: yep, we can try it 17:06:41 having easy way to deploy everything from scratch would be good for kolla-mesos 17:07:03 that may help with multinode installation of mesos 17:07:17 I thing currently biggest barrier of it is that you can't do something you can with kolla-ansible - deploy everything with one command 17:07:19 nihilifer: one question here 17:07:40 will these mesos images still be able to follwo the 1-pid per container, or would they require supervisord? 17:07:49 1-pid per container 17:07:54 ok cool 17:08:16 one image for mesos, one for marathon and one for zookeeper I guess 17:08:39 elemoine: there will be separate images for mesos-master and mesos-slave 17:08:50 yeah, right 17:09:06 the zookeeper container may be useful for kolla-ansible as well 17:09:10 but yeah, zookeeper will go to the sepatate image 17:09:17 and chronos as well 17:10:02 ok, so if we're fine with that 17:10:14 alright weve got 20 minutes left and I have a subject to bring up 17:10:32 * britthouser hopes its mid-cycle dates 17:10:43 * nihilifer hopes too 17:10:52 I lost my hopes 17:10:53 sdake was supposed to handle that a month ago. 17:11:02 where do you guys want to vote on dates? 17:11:07 ill set something up today 17:11:15 whenisgood? 17:11:23 is that a website? 17:11:28 http://whenisgood.net/ 17:12:07 ok ill look at that, sometime in febuary is all i know, i dont care when 17:12:17 that wasnt the subject though 17:12:23 =) 17:12:25 just put whole month there 17:12:44 inc0: if you know how to use that site, do you want to do that? 17:12:54 the last I haerd from sdake was last week in Jan? 17:13:07 britthouser: since im hosting that date is news to me 17:13:19 britthouser: i hope no, it may collide with fosdem :( 17:13:21 lol....I'll go with what you say. =) 17:13:42 ok something will go out today with dates, ill see to it personally 17:14:05 thx. =) 17:14:05 Now what i wanted to talk about was named volumes with docker 17:14:26 if you use docker 1.9.x and do `docker volume ls` youll see all the volumes on your system 17:14:31 you can create named voluems as well 17:14:39 so we can actually get rid of these data containers we have 17:15:11 this doesn't require anything on the container side of things, so it shouldnt affect mesos 17:15:33 but it does mean we have to run our own docker module 17:15:42 Is the named volume functionally equivalent to data container? 17:15:57 http://whenisgood.net/qrb4db9 17:16:02 britthouser: the data contaienr previously was the only way to keep track of a volume 17:16:13 this way you can name a volume and reference it by name 17:17:09 thanks inc0 looks good 17:17:18 I'll pop a ML 17:17:27 inc0: do you have the address? 17:17:44 and for the record, I'm +2 on our own docker module 17:17:52 SamYaple, just the poll 17:18:00 if you want to pop a mail with full detail 17:18:05 totally +2 for named volumes and docker module 17:18:06 that will be better 17:18:20 there is one big problem with our own docker module tho 17:18:28 ok well the reason i bring it up is because we can't be 1:1 with the docker module upstream which people wanted 17:18:35 i never did, but it was brought up 17:18:36 Sounds like a better solution. I was just thinking about backup/upgrade/etc. Does named volume change any of that? 17:18:46 britthouser: it helps it alot, yes 17:19:34 so problem is 17:19:43 according to GPLv3 17:20:02 our own docker module cannot be based in any way on upstream docker module 17:20:07 it is not 17:20:12 i have written it from scratch 17:20:23 i havent even used the same arg names 17:20:24 to a point that technically if someone saw original one 17:20:55 he/she can't write new one as "inspiration" might also be treated as licence breakage 17:21:22 well we can just throw out all of ansible then. the only documentation for alot of the module code is lookin at other modules 17:21:23 and if so, kolla will have to be turn to gplv3, which means removal from openstack namespace 17:21:54 im not worried about this inc0. it is an original module 17:21:58 SamYaple, I'm just saying how that looks like in legal eyes 17:22:27 then basically all of OSA would be gone. and Kolla to (we have code like this) 17:22:37 and any other ansible project honeslty 17:22:41 all I'm saying, when we do write our own docker module, we need to be careful about that 17:22:54 even more reason to not try to 1:1 parity the options 17:23:04 yeah 17:23:09 some names overlap like 'name' or 'image' but past that its my own creation 17:23:53 I am not worried about a licensing problem, but ill make sure to not look at the docker module while writing it ;) 17:24:04 yeah;) 17:24:23 all is clear Sam? 17:24:28 can we move on? 17:24:32 ok well the docker module is almost done, so you should see a patchset in the next few days 17:24:38 yup im good 17:24:39 cool, thanks 17:24:41 cool 17:24:42 so 17:24:56 https://review.openstack.org/#/c/257568/ -> this is first ready play 17:24:59 for upgrades 17:25:01 this is keystone 17:25:18 it's important as it will provide standard and template for other services 17:25:36 so once again please we need as much eyes on this one as possible 17:26:15 if we merge this one, I'll pop a ML thread "upgrade open season is open" 17:26:35 to encourage everyone to take on one of projects and write upgrade playbook on it 17:27:12 sounds good to you guys> 17:27:35 Yup. 17:27:48 yes 17:27:48 ok, 3 minutes left, anyone else? 17:27:55 Austin Summit: looks like hotels are ready for booking 17:28:20 that is all 17:28:27 quick notice 17:28:35 that i left a comment on k8s blueprint 17:28:38 ah, also speaker proposal is open 17:28:45 i found a way to run bootstrap tasks in k8s 17:28:51 cool! 17:28:54 so no need of chronos or involving ansible on that 17:29:04 thanks 17:29:12 if Ryan will be ready, i'll be +2 for k8s 17:29:27 if you want to speak about kolla, let's get more presence on this summit 17:29:44 I think Mesos would make a good topic 17:29:53 yeah, I'll try upgrades 17:30:01 that too! 17:30:12 ok guys, we ran out of time 17:30:15 thanks everyone 17:30:18 cya! 17:30:18 thx 17:30:19 #endmeeting kolla