18:00:45 <richm> #startmeeting CentOS OpsTools SIG
18:00:45 <centbot> Meeting started Wed Jun 28 18:00:45 2017 UTC.  The chair is richm. Information about MeetBot at http://wiki.debian.org/MeetBot.
18:00:45 <centbot> Useful Commands: #action #agreed #help #info #idea #link #topic.
18:00:51 <jdetiber> tdawson: I'll probably be around
18:00:53 <richm> #chair mrunge
18:00:53 <centbot> Current chairs: mrunge richm
18:02:06 <richm> so jbadiap looked at adding the fluent-plugin-kubernetes_metadata_filter plugin + build and runtime deps
18:02:16 <mrunge> o/
18:02:18 <mrunge> hello
18:02:19 <richm> It's is about 230+ rubygem packages
18:02:40 <richm> that would need to be converted to rpms
18:03:20 <mrunge> that sounds like a HUGE task
18:03:30 <mrunge> is there any deadline for this?
18:03:33 <richm> we desperately need some sort of automation around this
18:03:47 <richm> I don't know - is there an rdo/openstack requirement?  Kolla perhaps?
18:04:25 <mrunge> rdo doesn't run on kubernetes
18:04:36 <mrunge> iirc, kolla is building docker images
18:04:48 <mrunge> but I may be wrong
18:05:02 <mrunge> it's not that a big step from docker to kubernetes
18:05:27 <mrunge> aha, there is https://github.com/openstack/kolla-kubernetes
18:05:36 <richm> What do other groups do?  Do they sign up to support 300+ packages in order to get 1 program?  If so, do they have an army of maintainers, or some really advanced automation?
18:05:57 <mrunge> you mean, like for elasticsearch in fedora?
18:06:00 <richm> With respect to these ruby packages - How much overlap with other groups is there?  For example, satellite/foreman may already have a lot of these packages, same with puppet, other ruby packages in centos/rdo?
18:06:11 <richm> mrunge: or opendaylight?
18:06:24 <mrunge> isn't opendaylight written in java?
18:06:51 <mrunge> it would be good to check how many of those deps are already in fedora, for example
18:07:04 <richm> https://paste.fedoraproject.org/paste/I5YlWWyWld81iaaGQxM-uQ
18:07:06 <mrunge> next issue is the old ruby stack in centos
18:07:21 <richm> yes, opendaylight is java, so I imagine it has the same problems as elasticsearch
18:07:34 <mrunge> ok, now I get you
18:07:44 <richm> we need some sort of workflow like ^^^ in the paste
18:08:08 <mrunge> yes! and maybe no, as java packagers tend to bundle everything
18:08:26 <richm> I'm just strictly speaking about ruby packages for fluentd
18:08:33 <richm> for now, anyway
18:08:39 <mrunge> yup
18:08:55 <mrunge> is there something like gem2rpm?
18:09:02 <richm> yes, there is gem2rpm
18:09:05 <richm> but it has some issues
18:09:24 <richm> for example - it does not add a %check section
18:09:39 <mrunge> it's in the paste, sorry for the question then
18:09:43 <richm> because there is no standard way to test ruby packages - a lot of packages use `rake test` but there are many that do not
18:10:21 <mrunge> fedora-review seems to consider running rake test as a build issue
18:10:44 <richm> and then there are the circular dependences e.g. rspec-xxx depends on rspec-yyy, but rspec-yyy depends on rspec-xxx
18:10:52 <mrunge> right
18:11:13 <richm> at this rate, we'll have a full fluentd stack by 2018
18:11:25 <mrunge> I looked into packaging grafana, ~ 200 golang and javascript deps
18:11:37 <richm> at which point 0.12 will be deprecated, and we'll have to start all over again with 0.14
18:12:43 <richm> If we really want a fully rpm-ized fluentd stack, including all of the build deps, we need to start building out a CI infrastructure that can do https://paste.fedoraproject.org/paste/I5YlWWyWld81iaaGQxM-uQ and probaby more
18:13:06 <richm> there is no way we are ever going to be able to support this with human power only
18:13:16 <mrunge> richm, so this is for fluentd talking to kubernetes?
18:13:32 <mrunge> I mean, we are already using fluentd in rdo and in rhos
18:13:54 <mrunge> I have the feeling, I'm missing something
18:13:56 <richm> mrunge: just the fluent-plugin-kubernetes pulls in 200+ packages - many of those overlap with other fluentd + dependencies
18:14:12 <richm> mrunge: right - we deliberately turned off the %check in many packages
18:14:16 <richm> or crippled the %check
18:14:27 <mrunge> and why do we need that plugin now?
18:14:44 <richm> mrunge: I don't know - I was hoping jbadiap would be here
18:15:01 <mrunge> ok, I see
18:15:06 <richm> if we only look at runtime dependencies, we have < 50 packages
18:15:21 <mrunge> that would be my starter
18:15:28 <richm> but once you start enabling %check for unit tests, that number explodes by a factor of 4 or 5
18:16:02 <richm> and it isn't just the kubernetes plugin, although that may be particularly egregious
18:16:20 <mrunge> with ci in place, we might be able to skip checks without having a to bad feeling
18:16:28 <richm> right
18:16:45 <mrunge> I totally agree, checks would be great
18:17:16 <richm> do we have CI?
18:17:38 <mrunge> there is a little bit
18:18:01 <mrunge> https://github.com/mrunge/opstools-gate
18:18:13 <mrunge> but that is not integrated into centos ci yet
18:18:22 <mrunge> not sure if it's feasible at all
18:18:33 <mrunge> at least, there are ideas for test cases
18:18:58 <mrunge> like: make sure an rpm is installable, it runs and returns that output, which is expected
18:19:27 <mrunge> the playbooks currently don't run on multiple hosts
18:20:05 <mrunge> as you can see here: https://github.com/mrunge/opstools-gate/blob/master/roles/log_target/tasks/main.yml#L34
18:20:19 <mrunge> fluentd is just tested to the bare minimum
18:20:27 <mrunge> but we can enhance that
18:21:39 <richm> we don't need multiple hosts, at least for now
18:22:00 <richm> we can run multiple fluentd on the same machine, somehow
18:22:10 <mrunge> right
18:22:20 <richm> e.g. for secure_forward testing
18:22:55 <mrunge> with that openstack stuff, we can easily spawn multiple instances. I didn't implement that, but that won't be much work
18:23:18 <mrunge> the question is, if we want to go that route at all (using openstack etc)
18:23:35 <mrunge> or if we would like to try to hook that up into centos ci
18:24:14 <richm> It doesn't matter to me which ci we use
18:24:58 <richm> the logging team is in full on bug fix mode for the next 2 weeks - openshift 3.6 code freeze is July 7
18:25:15 <richm> after that, we start a major testing/CI initiative for both upstream and downstream
18:25:22 <richm> this will be part of that
18:25:45 <mrunge> ok, we are concentrating on the next openstack release, we are past milestone 2 of 3
18:26:09 <richm> mrunge: what's next on your roadmap?
18:26:38 <mrunge> currently we are implementing fully containerized openstack services
18:27:02 <richm> ok - I've been seeing a lot of traffic on that topic
18:27:16 <mrunge> and then my focus will shift a bit from opstools actually
18:27:40 <mrunge> but that being said, I'd love to see more automation for opstools in place
18:31:04 <mrunge> richm, in terms of automation, next step would be to run some tests in rpms in -candidate or -testing and tag those rpms into -stable
18:31:25 <richm> mrunge: ok
18:31:26 <mrunge> one can already trigger builds by submitting changes to specs
18:31:32 <richm> ah, cool
18:31:52 <richm> mrunge: that's in https://github.com/mrunge/opstools-gate ?
18:31:53 <mrunge> an example is here: https://review.rdoproject.org/r/#/c/7085/5
18:32:08 <mrunge> opstools-gate is some kind of prototype
18:32:17 <mrunge> but I'm open to other suggestions
18:33:03 <mrunge> the linked change fails due to missing dependencies on arm or powerpc
18:33:20 <mrunge> I forgot where. there is an open ticket on bugs.centos.org for that
18:33:50 <richm> cool.io is a C wrapper package - must be compiled
18:34:06 <richm> i.e. not a noarch package
18:34:55 <mrunge> yes, it depends on some noarch packages, which seem to be missing on the builder
18:35:17 <mrunge> right, rubygem(rspec) is missing
18:36:18 <mrunge> in this case https://cbs.centos.org/koji/taskinfo?taskID=188085 rubygem-diff-lcs, rubygem-rspec-core and rspec-mocks were missing
18:36:37 <mrunge> and rubygem-expectations
18:36:59 <richm> ah, I see - the ppc build failed first and cancelled the other two arches
18:37:11 <mrunge> right
18:37:19 <richm> ok
18:37:28 <richm> I'll take a look at that
18:37:33 <richm> I think sandro is out
18:37:54 <mrunge> funny, the packages seem to be there: http://cbs.centos.org/repos/opstools7-fluentd-012-testing/ppc64le/os/Packages/
18:38:26 <richm> yeah - odd
18:40:33 <mrunge> not sure who would be the best person to look at that
18:42:40 <richm> mrunge: I'll investigate
18:43:04 <mrunge> thank you, richm
18:43:15 <mrunge> that must be something like a typo or so
18:43:18 <richm> #action richm investigate cool.io build failures
18:44:04 <richm> any other topics for discussion today?
18:44:19 <mrunge> not from my side
18:44:25 <richm> anyone else?
18:45:21 <richm> I think we can end then
18:45:33 <mrunge> yes.
18:45:37 <richm> #endmeeting