13:00:46 #startmeeting rolekit (2015-10-27) 13:00:46 Meeting started Tue Oct 27 13:00:46 2015 UTC. The chair is sgallagh. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:00:46 Useful Commands: #action #agreed #halp #info #idea #link #topic. 13:00:46 #meetingname rolekitweekly 13:00:46 #chair sgallagh twoerner nilsph 13:00:46 #topic init process 13:00:46 The meeting name has been set to 'rolekitweekly' 13:00:46 Current chairs: nilsph sgallagh twoerner 13:00:52 .hello nphilipp 13:00:53 nilsph: nphilipp 'Nils Philippsen' 13:00:54 .hello twoerner 13:00:55 .hello sgallagh 13:00:56 twoerner: twoerner 'Thomas Woerner' 13:00:59 sgallagh: sgallagh 'Stephen Gallagher' 13:01:18 How is everyone doing? 13:02:13 laboring on the last remnants of a cold 13:02:18 Ouch, sorry to hear that. 13:02:30 I am busy.. working on some firewalld extesion 13:02:41 but fine 13:03:05 I meant to ask: will either of you be at the systemd conference next week? 13:03:17 nope 13:04:13 I will not be there... 13:05:00 OK, just curious. 13:05:06 #topic Status Update 13:05:41 I don't have much to report here; I've spent the last two weeks alternately herding Fedora cats and working on a massive glibc patch. So not much time for rolekit. 13:05:50 sgallagh: will you be there? 13:05:55 I will 13:06:37 nice 13:06:43 Hopefully I'll be back on rolekit later this week; I'm going to work on the pendingrole cancellation stuff first and then move on to the progress updater patches. 13:08:10 twoerner: Have you had any time for rolekit lately? 13:09:01 I think I will have some time for rolekit again next week 13:09:54 nilsph: ? 13:10:00 but did not had lots of time in the last two weeks for rolekit 13:10:10 sgallagh: no, sorry 13:10:18 (was distracted, so doubly sorry ;) 13:10:29 re: systemd conf I mean 13:10:58 I've thought about out-of-tree role plugins 13:11:33 #topic Out-of-tree Role Plugins 13:11:37 Go on :) 13:12:07 heh 13:12:54 I think one of the bigger challenges with this is that we need to store these plugins somewhere (probably just beside the instance settings), and how to address them in the UI 13:13:21 nilsph: Sorry, I'm not quite following. 13:13:36 That was the very condensed versions. The abstract. :) 13:13:39 -s 13:13:43 :) 13:13:52 Sorry, I'll let you talk before interrupting. 13:14:02 Now you did it again. :P 13:15:27 OK. 1) Store the role plugin somewhere. Probably needs to go right beside the instance which is created from it, because we need to be able to redeploy/decommission it later and we can't guarantee that one out-of-tree "Froboz" role is like the other. 13:15:55 Does that make sense? 13:15:59 nilsph: No, actually. 13:16:03 Good. 13:16:05 :) 13:16:32 I'm sure there's a longer version of it :) 13:16:41 A role plugin should never be removed without also removing any instances of it. 13:17:04 We should probably provide guidelines to packagers about this. 13:17:46 I'm pretty sure this is an acceptable doc issue rather than a technical one. 13:18:08 We can minimize it in packaging scripts, but realistically it falls under "don't do something obviously broken", to me :) 13:18:14 sgallagh: so do you think along the lines of 1) an admin creates a new role by importing its plugin and 2) he creates an instance from it as separate activities? 13:18:27 I do 13:18:32 Aaaah. 13:19:07 Realistically, we can't deploy at all without the plugin already being present. 13:19:36 Well, I didn't understand it like that unfortunately. We should add a little prose to the issue then, so I won't chase down other blind alleys :). 13:19:37 Though I suppose we could talk about having auto-retrievable plugins from some sort of a repository. 13:20:00 But not in the first pass. 13:20:21 sgallagh: My initial understanding was s.th. like "rolectl deploy --out-of-tree froboz-role.py ..." 13:20:38 Turns out, it was wrong. Good we talked about it :). 13:21:04 nilsph: That could still be useful, but only as a helper mechanism. 13:21:13 Remember, we want rolekit to be driveable from its API. 13:21:38 rolectl should always be a thin layer around the API except where impossible (like --deferred) 13:22:35 That said, there's value to having rolectl be able to register and install a new role plugin. 13:22:53 Yes. We should somehow distinguish "user-supplied" roles from "officially shipped" ones, so if a user adds a role "foobar" and we do add another "foobar" in the next version things don't disintegrate. 13:22:54 so something like `rolectl register frobozz-role` 13:23:06 Yes. 13:23:08 Where frobozz-role is either frobozz-role.py or frobozz-role/ 13:23:35 And then it gets copied somewhere the admin is unlikely to mess with it (except through the API and/or rolectl) 13:23:42 nilsph: Exactly. 13:24:06 I'd still call that "phase 2" though. 13:24:41 The first piece of this effort is figuring out what parts of the code need to be documented and made more public so that building out-of-tree roles is possible (and testable) 13:24:48 okay 13:24:59 I think I understand better now. 13:25:25 In phase 1, I think it's okay if we tell people "`make install` must put rolename.py " 13:26:40 Aaahhh. 13:27:05 err, role.py, of course 13:27:53 Perhaps we should reserve a namespace for user-supplied roles? S.th. like "We guarantee that rolekit will never ship roles starting with "user-"." 13:28:00 The biggest piece of this, I think, is going to be refactoring and documenting a "-devel" module that provides all of the useful helper routines. 13:28:42 nilsph: Probably a good idea, but maybe "custom-" would be better. 13:28:55 yes 13:29:04 "custom-" it is 13:29:59 nilsph: Let's not *enforce* this, however. It may be that we will want to have some official roles live out of tree. 13:30:14 So that they would be maintained by the other project rather than us. 13:30:49 (In Fedora, we would just subpackage those and Requires: or Recommends: them if we want them available by default) 13:33:12 sgallagh: I wouldn't suggest "anybody else than us must use the so-not-official-it-hurts prefix". But perhaps 3rd parties could kind of "register" their role names with us so we don't end up with conflicts? I mean there's no way we can force them to play nice, but we don't need to make promises not to step on their toes either. 13:33:57 nilsph: Probably reasonable. 13:34:16 And as we go along, if we do end up with a registry/repository of plugins, that will sort of work itself out 13:35:07 sgallagh: not sure if we shall follow the lead of others that use reversed domain names for self-organizing namespaces. Makes stuff a little clunky for end users, but it's pretty safe. 13:35:31 I'd really prefer not to do that. 13:35:50 But we'll see if it becomes necessary 13:36:21 I feel like that only becomes an issue once we have a lot of participants, at which point it will be a good problem to have 13:37:37 yeah 13:40:02 OK, anything else on this topic? 13:40:54 no, I think I have a better understanding now 13:42:42 #topic Open Floor 13:42:50 Anything for Open Floor this week? 13:43:47 not from me 13:43:55 twoerner: ^^ ? 13:44:25 nothing from me 13:44:48 OK, then thanks for coming. 13:44:57 #endmeeting