ansible_irc_core_meeting
LOGS
19:04:30 <bcoca> #startmeeting ansible irc core meeting
19:04:30 <zodbot> Meeting started Tue Jun  5 19:04:30 2018 UTC.
19:04:30 <zodbot> This meeting is logged and archived in a public location.
19:04:30 <zodbot> The chair is bcoca. Information about MeetBot at http://wiki.debian.org/MeetBot.
19:04:30 <zodbot> Useful Commands: #action #agreed #halp #info #idea #link #topic.
19:04:30 <zodbot> The meeting name has been set to 'ansible_irc_core_meeting'
19:05:16 <bcoca> #topic https://github.com/ansible/ansible/issues/40913
19:05:35 * ryansb waves
19:06:03 * sdoran waves too
19:06:06 <bcoca> #chair ryansb sdoran
19:06:06 <zodbot> Current chairs: bcoca ryansb sdoran
19:06:07 * shertel waves
19:06:11 * jtanner multitasks
19:06:13 <abadger1999> Olá
19:06:16 <bcoca> #chair shertel jtanner abadger1999
19:06:16 <zodbot> Current chairs: abadger1999 bcoca jtanner ryansb sdoran shertel
19:06:21 <schmots_> that change raises an interesteing question.  When does a module change from preview to stable?
19:07:14 <bcoca> schmots_: until now its only been core doing it, its mostly a commitment by maintainer to avoid backwards incompatible changes if possible and used the standard deprecation cycle if it is a must
19:07:51 <bcoca> anyone else have questions or statements to add?
19:08:40 <ryansb> yeah, it's pretty much "the core team/maintainers want to move to using the deprecation cycle and standardize the interface"
19:08:43 <ryansb> no formal process as yet
19:08:45 <jtanner> i think we should let maintainers decide whatever they want for the status field.
19:09:07 <bcoca> jtanner: i was going to add follow up topi, should core even be invovled in this?
19:09:20 <bcoca> for now, lets vote on this case, should we allow him to set to stable?
19:09:27 <jtanner> i think we only need to be involved in a change to the "supported_by" field
19:09:48 <jimi|ansible> o/
19:09:55 <jimi|ansible> jctanner++
19:09:58 <bcoca> #chair jimi|ansible
19:09:58 <zodbot> Current chairs: abadger1999 bcoca jimi|ansible jtanner ryansb sdoran shertel
19:10:15 <bcoca> so im counting both of you as +1 (implied in that its up to him)
19:10:16 <jtanner> imo, if caphrim007 were to a do a PR, the shipit process would have allowed the change
19:10:18 <bcoca> +1 from me
19:10:26 <ryansb> yeah, +1 for maintainers to handle status
19:10:30 <shertel> +1
19:10:49 <bcoca> k, lets keep issues separate, otherwise i have no clue what vote goes where
19:10:52 <jimi|ansible> +1 on the condition that they know it's irrevocable for at least 4 releases
19:11:05 <bcoca> in this case, he knows
19:11:08 <abadger1999> +1 to letting caphrim choose.  But --- fro mthe sound of it I would encourage him not to mark all of bigip* as stableinterface but only the ones that he has to.
19:11:25 <caphrim007> i heard my name
19:11:28 <jtanner> if we don't have automated enforcement, i don't know that we can really dictate anything
19:11:35 <jtanner> shipit rules all
19:11:39 <bcoca> @jtanner you have revert and all
19:11:46 <jtanner> -if- we catch it
19:11:53 <abadger1999> caphrim007: Hey :-) We're talking about the request to mark bigip* as stableinterface :-)
19:12:03 <abadger1999> We seem to be on board with maintaier decides.
19:12:09 <shertel> +1 for maintainers to decide the status field
19:12:11 <abadger1999> I'm just worried about workload and continuity.
19:12:42 <bcoca> #topic hijacking
19:13:25 <caphrim007> can i ask for some clarification on what stableinterface means so i can make a decision on which ones i mark as such?
19:13:46 <abadger1999> workload... stableinterface can require more work to maintain backwards compat and new-fangled goodness at the same time and it sounds like you're all alone from the f5 side.  Continuity... what if you move on?  Then we have modules that your replacement comes into and he may see things differently.
19:14:02 <abadger1999> But those would be for your consideration, I think rather than ours.
19:14:03 <abadger1999> Sure.
19:14:19 <caphrim007> does it mean i cant add or remove things from the stable ones? or if i do, that i follow the 4 release cycle? or i thought 4 release cycle was for everyone?
19:14:36 <bcoca> adding is ok, removing requires deprecation cycle
19:14:44 <abadger1999> stableinterface means that you try not to break existing playbooks and when you have to, then you have to go through a deprecation cycle to make those types of changes.  In the meantime, you'd have to support both the new and old ways.
19:15:11 <caphrim007> so if, today, i already follow the 4 release cycle, then there's not much change
19:15:25 <abadger1999> (adding can have playbook-breaking effects so you have to think about those just in case they do)
19:15:30 <abadger1999> Correct.
19:15:32 <abadger1999> Although...
19:15:48 <abadger1999> You should be trying harder not to do backwards incompatible things
19:16:01 <abadger1999> "Trying harder" is not quantified though.
19:16:08 <caphrim007> sure. i'm just imagining my company breaking their own apis
19:16:15 <caphrim007> which they've done
19:16:22 <caphrim007> one of them after we shipped a module
19:16:30 <caphrim007> because the api had been stable for 10 years
19:16:37 <caphrim007> and then a new team took over and changed it completely
19:16:48 <abadger1999> Yeah.  If that were the case, the ansible module would need to try to shim in support for the old features using the new api.
19:17:02 <bcoca> well, that is why the module interface is supposed to be an abstraction, when you tightly couple to api you run into those issues
19:17:02 <caphrim007> and/or make different modules?
19:17:06 <abadger1999> That can be a major cause of pain for fast moving apis *cough*docker*cough*
19:17:17 <bcoca> if the module interface is good enough, you can mostly avoid those issues
19:17:36 <bcoca> @abadger1999 docker module changed code a lot, but not the interface as much
19:17:40 <abadger1999> @caphrim007 well, yes and no to that.... the real thing we're trying to achieve is "don't break user's playbooks if they're using a stableinterface module"
19:18:15 <abadger1999> So making a different module and letting the old one stop working is contrary to that goal.
19:19:05 <abadger1999> @bcoca I think you miss the point... the docker python module apis kept changing so the docker ansible module had to grow a lot of shim code to keep the same ansible interface onto all the different docker python module apis.
19:19:09 <caphrim007> i'll have to continue to think about this and discuss it with my org then
19:19:14 <abadger1999> Cool.
19:19:21 <caphrim007> but if ansible is +1 to it, then that is fine
19:19:32 <caphrim007> i'll just go tell my org that at least ansible is not -1
19:19:44 <caphrim007> and then in my org i'll relay to them what you've mentioned here
19:19:51 <ryansb> well, we're not -1 as long as you actually do the supporting piece
19:19:51 <caphrim007> so that they know what they're getting themselves into
19:19:54 <bcoca> @abadger1999 no, that is my point, user was safe from many of the changes, but it was due to maintainers commiting to it and taking the trouble of creating all those shims
19:20:04 <abadger1999> And... you can break it... (/me can remember a few things that broke in core modules) but it should only be broken with great trepedition and a good amount of thought.
19:20:19 <caphrim007> ryansb: i gotcha. i am willing to do the support if my mgmt accepts that overhead
19:20:59 <ryansb> Ok, just important that you include that bit w/ the other bits
19:21:07 <caphrim007> yep
19:21:58 <bcoca> ok, think we can vote on it now:
19:22:00 <bcoca> +1
19:22:43 <bcoca> no one else?
19:23:09 <abadger1999> +1 (voted previously)
19:23:28 <bcoca> well, people have been voting on diff things intermixed, i want clear vote on THIS issue
19:23:43 <abadger1999> #topic stableinterface for community modules
19:23:45 <jtanner> surveymonkey.com
19:24:11 <shertel> +1 (voted previously)
19:24:23 <ryansb> +1 (stuffed ballot box previously)
19:25:26 <bcoca> #topic https://github.com/ansible/ansible/pull/41078
19:26:25 <bcoca> contributor wants to change default behaviour when syslog setting is present, vs current in which we default to journalctl and only use syslog setting if it is not present and we are using syslog directly
19:27:04 <jtanner> fyi hideki works for us in support and is on japanese hours
19:27:42 <bcoca> ah, new the hours, didnt know he works with us, issue still stands, this is a change of behaviour and i'm not against it, but i do think it needs an extra toggle
19:27:48 <bcoca> knew
19:27:52 <abadger1999> hhh
19:28:28 <abadger1999> We use journal.send() either way.
19:28:51 <abadger1999> The only difference is whether facility gets used or not.
19:29:04 <bcoca> no, we use syslog module in other case
19:29:16 <bcoca> syslog.syslog(syslog.LOG_INFO, msg)
19:29:31 <bcoca> vs journal.send(u"%s %s" % (module, journal_msg), **dict(journal_args))
19:30:10 <abadger1999> No...
19:30:13 <abadger1999> +                        journal.send(MESSAGE=u"%s %s" % (module, journal_msg),
19:30:15 <abadger1999> +                                     SYSLOG_FACILITY=facility,
19:30:16 <abadger1999> +                                     **dict(journal_args))
19:30:18 <abadger1999> +                    else:
19:30:19 <abadger1999> +                        journal.send(MESSAGE=u"%s %s" % (module, journal_msg),
19:30:21 <abadger1999> +                                     **dict(journal_args))
19:30:24 <abadger1999> Both code paths end in journal.send()
19:30:44 <bcoca> im looking at code now
19:30:48 <bcoca> except IOError:
19:30:49 <bcoca> # fall back to syslog since logging to journal failed
19:30:51 <bcoca> self._log_to_syslog(syslog_msg)
19:30:52 <bcoca> else:
19:30:54 <bcoca> self._log_to_syslog(syslog_msg)
19:31:04 <bcoca> the else is on no journalctl, the except is if journalctl.send fails
19:31:16 <bcoca> and log_to_syslog, has syslog module
19:31:24 <bcoca> ^ current devel
19:31:36 <bcoca> its been like that a long time
19:32:42 <bcoca> the change here makes a) using journalctl depend on having syslog available, b) introduce the facility to journalctl which was only used for syslog before
19:32:47 <abadger1999> I Okay, obviously one of us is reading this code wrong....
19:32:54 <abadger1999> I don't see any else on journalctl.
19:33:18 <abadger1999> journalctl isn't present in the devel module_utils/basic.py code
19:33:30 <bcoca> if has_journal:
19:33:34 <bcoca> ^ else: syslog
19:33:51 <bcoca> from systemd import journal
19:33:52 <bcoca> has_journal = True
19:33:53 <abadger1999> And that's still present with hideki's change.
19:34:00 <bcoca> not saying its not
19:34:27 <bcoca> please read what i posted above a) journalctl now depends on HAS_SYSLOG (didnt before his change) b) the syslog facility now affects journalctl
19:34:42 <abadger1999> a) untrue.
19:34:48 <abadger1999> b) true.
19:34:49 <bcoca> if HAS_SYSLOG
19:34:53 <abadger1999> b looks like a bugfix.
19:34:56 <abadger1999> So what?
19:35:04 <bcoca> ^ that is his first change
19:35:09 <abadger1999> So what?
19:35:12 <bcoca> changes journalctl to depend on HAS_SYSLOG
19:35:18 <abadger1999> No it doesn't
19:35:21 <bcoca> how not?
19:35:30 <abadger1999> https://github.com/saito-hideki/ansible/blob/a3216fbdad9bef6f2fd60675b7cc2ed3e99619f2/lib/ansible/module_utils/basic.py#L2198
19:35:46 <bcoca> ah, missed last else
19:36:22 <bcoca> still, changes behaviour of logging, which can change the file where logs go to
19:36:25 <abadger1999> That was what I posted above.
19:36:33 <abadger1999> yes, but that's the point I think.
19:36:36 <bcoca> thought you refered to the syslog else
19:36:42 <jtanner> is this codepath tested anywhere?
19:36:45 <abadger1999> It looks like a bugfix.
19:37:01 <abadger1999> We provide confuration of the logging facility to use.
19:37:08 <bcoca> he is changing a setting that previuoslly affected only syslog to also affect journalctl
19:37:21 <abadger1999> We're currently not horing that configuration if logging goes through journald.
19:37:34 <abadger1999> Thi sPR fixes that bug.
19:38:28 <bcoca> i would argue that the config was for a specific system and that this change will suprise users by having the logs disappear from where they expect them
19:42:20 <abadger1999> It won't surprise new users.  it could surprise existing users who have syslog_facility set and have been using it with systemd.  It won't surprise existing users who don't have syslog facility set.  It won't surprise existing users who are either switching over to systemd or are enabling syslog_facility for the first time.
19:42:46 <abadger1999> this is one of those shims that we were talking about earlier.
19:43:02 <bcoca> well, not yet
19:43:24 <abadger1999> and the shim is supposed to abstract which backend syslog is in place so that the user gets the same behaviour either way.
19:44:28 <bcoca> well they normally behave differently, though systemd has been comming closer to syslog
19:44:38 <bcoca> and in this case the paramter woudl mirror across them
19:45:00 <bcoca> i still think its a behaviour chagne and would requrie another toggle, but im not going to strongly opose
19:45:02 <bcoca> -0
19:46:03 <bcoca> ^ votes?
19:46:10 <abadger1999> bugfix +1 to PR... should be added to the porting guide and changelog so that people know to remove the configuration if they don't want to log to a different location.
19:46:34 <bcoca> abadger1999: since config is global, they cannot set per host/group
19:46:53 <bcoca> so if they have mixed systemd/syslog .. they are going to have logs moved one way or the other
19:47:15 <sdoran> Been trying to keep up and think this through.
19:47:33 <bcoca> if it were host/group configurable, i would have less of an issue
19:47:37 <sdoran> I think it feels like a bug: the `syslog_facility` should be unified regardless of the logging mechanism.
19:47:53 <sdoran> So that it's different between journal and syslog seems wrong.
19:48:14 * sdoran awaits thrashing :)
19:48:42 <bcoca> abadger1999:  in the end you added teh feature as it is, so i'll defer
19:48:47 <bcoca> sdoran: so vote?
19:49:07 <abadger1999> bcoca: which is as it should be.
19:49:15 <bcoca> sdoran: iirc systemd didnt' have facilities initially, so the diff is less of an issue today
19:49:29 <abadger1999> You don't want to have half of your machines logging to one location and half to another.
19:49:45 <bcoca> abadger1999: not saying I want that, its what people currently have
19:49:52 <sdoran> I do see how if an existing user has gotten accustomed to/figured out that journalctl messages go one place and syslog another, that could be jarring.
19:50:11 <bcoca> most people dont care about logs, but those that do .. care a lot
19:50:27 * bcoca points at jtanner
19:50:31 <sdoran> I vote +1 to merge, document it as bugfix/porting guide item.
19:50:54 <bcoca> ok, i'll merge, you guys follow up with him to add docs/warnings/whistles?
19:50:57 <jtanner> clint eastwood quote about logs and dead hands and cold weather, or something like that
19:51:29 <bcoca> so going to add 'adhoc topic'
19:51:39 <bcoca> #topic should core still controll status in metadata?
19:52:12 <abadger1999> How is this different from what we already voted on?
19:52:35 <bcoca> @abadger1999 we voted on particular case, now we are voting if we should have those cases at all in future
19:52:57 <abadger1999> Hmm...
19:53:16 <bcoca> why i insiste on not cross voting on topic, makes things unlcear
19:53:19 <abadger1999> If we can dfer judgement, I'm willing to.
19:53:30 <bcoca> dfer to what?
19:53:32 <abadger1999> Proposal: Keep it ad hoc for now.
19:53:46 <abadger1999> people continue to bring such changes to us.
19:53:50 <bcoca> i'll make it an option
19:54:01 <bcoca> but that implies we are still involved
19:54:02 <abadger1999> If we get an upswell in requests revisit.
19:54:19 <abadger1999> *requests, revisit.
19:54:22 <bcoca> a) not involved anymore: update docs that say we are
19:54:23 <jtanner> i have a feeling that we need to address some things internally in regards to metadata, and then come back to community when we figure out what is for what
19:54:39 <bcoca> b) stay involved but tentatively revisit on volume: no changes
19:54:56 <bcoca> c) stay involved and enforce :need to detect/prevent if community changes status
19:55:23 <bcoca> d) defer vote till we get our stories straight?
19:55:26 <bcoca> ^ jtanner
19:55:29 <sdoran> We don't currently have anything in place to prevent metadata merges.  We would need something like that to really enforce it.
19:55:38 <bcoca> sdoran: that ic c)
19:55:44 <jtanner> +1 for "d"
19:56:12 <bcoca> i was b/c ... but considering all new info i got today +1 for a)
19:56:32 <bcoca> @jimi|ansible, @privateip_ @shertel ... allofrest?
19:56:53 <bcoca> abadger1999: assuming you are for b)?
19:57:26 <abadger1999> +1 b or d
19:57:37 <bcoca> i'll put you down for 0.5
19:57:53 <bcoca> or should we do rank choice voting?
19:57:53 <abadger1999> i'm a full +1 to either one
19:57:57 <sdoran> I don't want to dilute the meaning of the terms since they could be valuable in the future Galaxy.
19:58:05 <sdoran> +1 for D
19:58:06 <abadger1999> bcoca: approval voting
19:58:11 <sdoran> More discussion needed.
19:58:12 <bcoca> abadger1999: same thing
19:58:17 <jimi|ansible> i think b) should be "not involved, but reserve the right to revisit on volume and/or complaints of non-stability"
19:58:39 <bcoca> jimi|ansible: i think a-d all reserve that right
19:58:49 <jimi|ansible> well we don't say that for any of them :)
19:58:53 <jimi|ansible> in that case a) +1
19:59:00 <bcoca> we currently have docs that say 'we do get involved'
19:59:01 <abadger1999> bcoca: rank choice would be four options, assign each a separate weight for 0 to 3.  approval voting is four options, assign each either 0 or 1.
19:59:19 * abadger1999 had to write the Fedora elections app so got exposed to a bunch of voting systems.
19:59:55 <bcoca> i thought ranked was particular case of approval
20:00:04 * bcoca sits corrected, but prefers ranked
20:00:06 <abadger1999> And after I did, some group that wanted to convert the US to range voting kept trying to get us to release the voter's voting records....
20:00:14 <abadger1999> ranked choice is a subset of range voting.
20:00:50 <bcoca> nex topic: which voting method we use
20:00:54 <abadger1999> ;-)
20:01:17 <bcoca> ok, so a+2, b+0.5, c+1 d+1.5
20:01:26 <abadger1999> I think it's 2 for a 3 for d
20:01:28 <bcoca> abadger1999: im counting each person as single whole vote
20:01:48 <abadger1999> *sigh*
20:01:54 <bcoca> did i miss someone?
20:01:57 <abadger1999> well, then I'll only vote for d
20:01:58 <abadger1999> yes.
20:02:02 <shertel> +1 for d too
20:02:04 <abadger1999> you missed either sdoran or jtanner
20:02:15 <bcoca> ok a+2 b+0,c+1 d+3
20:02:23 <abadger1999> (no2 _4 for d)
20:02:26 <abadger1999> *now
20:02:29 <abadger1999> +4
20:02:29 <jtanner> sorry, multitasking
20:02:31 <jtanner> so many fires
20:02:40 <bcoca> yes d+4 .. in the lead
20:03:04 <bcoca> well, d basically means b for now, but to revise soon?
20:03:18 <abadger1999> yep.
20:03:31 <bcoca> ok, closing vote in 10s
20:03:47 * bcoca really needs voting app in zoidbot
20:03:56 <bcoca> #topic open floor
20:04:14 <jimi|ansible> a+3, b+1, c+0, d+0
20:04:26 <bcoca> jimi|ansible: an ACCURATE voting app
20:04:33 <jimi|ansible> :)
20:05:04 <schmots_> how do we mark an update for a 2.6.1 release and not a 2.7 release?
20:05:33 <bcoca> you shouldn't
20:05:40 <bcoca> everything should go thorugh devel 1st
20:05:54 <schmots_> oh, even easier.
20:05:58 <bcoca> then you cherry pick to a backport PR that points at stable-2.6
20:06:05 <bcoca> AFTER the PR for devel was merged
20:06:23 <schmots_> Thanks as always bcoca
20:06:36 <bcoca> np, i just take your first born or beer as payment
20:06:52 <bcoca> no new topics, so
20:06:58 <bcoca> #endmeeting