16:00:36 #startmeeting Server SIG Weekly Meeting (2016-01-19) 16:00:36 Meeting started Tue Jan 19 16:00:36 2016 UTC. The chair is sgallagh. Information about MeetBot at http://wiki.debian.org/MeetBot. 16:00:36 Useful Commands: #action #agreed #halp #info #idea #link #topic. 16:00:36 The meeting name has been set to 'server_sig_weekly_meeting_(2016-01-19)' 16:00:36 #meetingname ServerSIG 16:00:36 The meeting name has been set to 'serversig' 16:00:36 #chair sgallagh mizmo nirik stefw adamw simo danofsatx mhayden jds2001 16:00:36 #topic roll call 16:00:36 Current chairs: adamw danofsatx jds2001 mhayden mizmo nirik sgallagh simo stefw 16:00:43 .hello mhayden 16:00:44 mhayden: mhayden 'Major Hayden' 16:00:49 .hello dmossor 16:00:50 danofsatx: dmossor 'Dan Mossor' 16:00:56 .hello adamwill 16:00:58 adamw: adamwill 'Adam Williamson' 16:01:22 .hello sgallagh 16:01:23 sgallagh: sgallagh 'Stephen Gallagher' 16:01:28 .hello jstanley 16:01:30 jds2001: jstanley 'Jon Stanley' 16:01:51 .hello simo 16:01:52 simo: simo 'Simo Sorce' 16:02:15 .hello kevin 16:02:16 nirik: kevin 'Kevin Fenzi' 16:02:45 * jds2001 read the (brief) log from last week 16:02:47 .hello jsmith 16:02:48 jsmith: jsmith 'Jared Smith' 16:02:55 * jds2001 is a sad panda :( 16:03:07 This is our best attendance in weeks. I kind of wish we had more topics on the agenda :-/ 16:03:22 jds2001: Well, I'll talk about that a bit once we get started 16:03:22 :) 16:03:23 i can leave, if that would be better 16:03:34 * simo kicks adamw 16:03:36 :P 16:03:48 still here ? >:-) 16:03:57 #topic Agenda 16:04:20 jflory7 wanted to re-raise the topic of the minimal image. 16:04:35 #info Agenda Item: Minimal Image 16:04:58 I had another topic that I was hoping to discuss this week, but unfortunately the stakeholders aren't available until next week 16:05:18 That was the topic of ABI stability checking 16:05:30 So that will happen next week 16:05:43 #info Agenda Item: Server Role Future 16:06:39 .hello andreasn 16:06:40 andreasn: andreasn 'Andreas Nilsson' 16:06:41 sorry, embroiled in a default browser debate in kde meeting 16:06:45 One other topic that I keep getting asked about is whether Server plans to produce an ostree-based image. I suppose we may want to discuss that as well 16:07:07 danofsatx: Just default to Internet Explorer. 16:07:26 heh...that might actually go over well at this point 16:08:28 OK, anyone else want to raise agenda topics? 16:08:42 adamw: Anything from QA? 16:08:54 don't think so, were you expecting something? 16:08:58 dustymabe inquired about a server/cloud FAD 16:09:41 mhayden: I missed that. 16:10:24 adamw: Nothing expected, just QA has been quiet lately and that makes me nervous :-) 16:11:23 OK, let's get started then. 16:11:30 #topic Minimal Image 16:11:34 no news != good news sometimes :D 16:11:36 jflory7: Are you around? 16:11:54 sgallagh: honestly, i'm just kinda digging my way out of my todo pile. 16:12:05 I was wondering what that smell was. 16:12:18 Oh, *to* do. Sorry. 16:12:49 The last time we discussed this, we agreed that the Server product was explicitly not an ultra-minimal image. 16:13:09 But we also agreed that it was installing too much at that time. 16:13:16 sgallagh: tomato, tomahto 16:13:45 I made a number of changes to comps and spin-kickstarts and reduced our DVD ISO image to 1.7GB down from 2.1GB 16:14:06 * jds2001 sees that as good progress. 16:14:15 A significant chunk of what's left is still due to python2 pulled in by IPA/Samba, but that's looking like it's on track to get dropped fairly soon 16:14:39 simo: Care to chime in on that score? 16:15:02 ++sgallagh, nice job 16:15:29 sgallagh++ 16:16:26 sgallagh: for some definition of soon :-) 16:16:32 but yeah progress is being made 16:17:31 simo: Is there an expected ETA? 16:18:25 it's like the CentOS ETAs - it's done when it gets done. 16:19:07 Well, I'm assuming that our downstream probably has certain expectations as well 16:19:50 sgallagh: there are too many actors involved to be able to make accurate ETAs 16:19:56 Fair enough 16:20:03 there are upstreams like python-ldap that are a real problem to work with 16:20:53 Understood. 16:21:05 simo: Fair to say that we should not expect this to happen for F24? 16:22:59 The other side of this is of course that we want to keep reducing the minimum install size as best as *reasonable* 16:23:22 For example, I'm still trying to track down exceptionally-large dependency chains 16:23:55 Any time we can break out a Requires (particularly in languages like Python), it usually gets rid of a lot of packages. 16:24:28 If anyone out there would like to help write some more modeling tools to figure out how best to discover the worst offenders, please see me after class :) 16:25:06 #info Work continues on reducing unnecessary dependencies in Fedora Server 16:25:14 * danofsatx would love to if he only knew how 16:25:20 * jds2001 too 16:25:25 danofsatx: How's your python? 16:25:29 not knowing how to code sux 16:25:47 jds2001: same question 16:25:49 danofsatx: python is easy to pick up :) 16:25:53 sgallagh: reasonable 16:25:56 i've python'd from time to time :) 16:26:16 So I've got a few scripts I used to generate the initial data dumps I sent to the list a few months ago 16:26:34 They're reasonably-approachable examples of how to use python-rpm 16:26:36 my pyhton skilz are about as hackish as adamw's were when he started on his pet QA project 16:27:05 danofsatx: eh, spend a day or two at the university of stackoverflow, you'll be fine. 16:27:12 in other words: I know how to effectivly use stackoverflow 16:27:18 It might be a good time to actually start a python-rpm-analysis project 16:28:20 OK, let's talk more about that in the coming days. 16:28:31 #topic Server Role Future 16:29:33 So, as mentioned last week, Red Hat is at least temporarily deprioritizing their investment into rolekit. 16:29:40 * nirik notes https://pagure.io/insim might be of use in the last topic 16:29:51 it is? 16:29:56 * danofsatx digs up logs 16:30:12 danofsatx: thats pretty much what the logs say, save your time :) 16:30:21 We've been given permission to finish the major refactor that are underway. 16:30:35 oh. I wasn't here last week, missed that. ;( 16:30:35 The good news is that this refactor should result in *extremely* easy role-creation 16:30:51 any reason? Or just a command from on high? 16:31:20 sgallagh: i guess the quesiton is what is the new direction for the requirement? 16:31:25 danofsatx: Mostly a matter of prioritization; the three people who were working on it are needed more urgently on other things for a while 16:31:37 roger that. 16:32:38 Aside, back on the minimal image topic: a fresh install of Rawhide's nightly with the Fedora Server env group takes 1.2GB of space. 16:33:02 I'm not giving up on rolekit; I still think it's a project with a future. 16:33:19 I'm going to be expending at least some of my own personal time maintaining it and developing new roles. 16:33:59 But if the Server SIG wants this to remain a key piece of our value-add, I'm going to need some volunteer help 16:34:39 * danofsatx curses his Appalachain blue-collar upbringing 16:34:44 Once this refactor is done (and we hope to have some example documentation to show in two or three weeks), creation of new roles should be quite simple. 16:34:52 I'm happy to help (time permitting) 16:36:05 nirik: Thanks, that would be good. We've got a backlog of potential role ideas on the github issues page. 16:36:16 I'll let people know as soon as the refactor lands 16:36:23 yeah. Sounds good. 16:36:42 do we need to time those/make changes around them? or they can just land anytime they are done? 16:37:02 nirik: Sorry, antecedent for those/they? 16:37:48 roles. Say we make a new role... should that be a change for the next fedora release? or just go in as it's done... 16:38:52 nirik: I'd say judge each one as they come. 16:39:08 ok. Or they could go in as done, but only advertise them next cycle. 16:39:10 We may want to break things into a supported and experimental subpackage, for example 16:39:15 (so we have prevew time) 16:39:18 yeah 16:40:14 #info Volunteer help requested in creation of new roles once the big refactor lands 16:41:27 It seems like there's at least enough *interest* that we may not need to seriously reprioritize Server efforts just yet. 16:41:40 Let's see where we stand around F24 Alpha 16:41:48 * nirik nods 16:41:59 OK, one last topic on the agenda. 16:42:10 #topic Fedora Server OSTree Image 16:42:43 intersting idea, but im slightly confuzzled 16:42:50 I've been asked several times by various sources if it might make sense to do a Fedora Server deliverable using ostree for updates rather than (or alternative to) dnf updates 16:42:56 * jds2001 thought the server product was more aimed towards pets :) 16:43:18 how would that look? 16:43:21 jds2001: It is, but that doesn't mean we couldn't essentially just replace the update mechanism 16:43:53 roles become containers? 16:43:56 sgallagh: what filesystem does ostree support ? 16:44:09 simo: Last I checked, it was FS-independent 16:44:22 how does it swap them ? 16:44:29 jds2001: So there is some work in rpm-ostree to allow an ostree base with RPM addons 16:44:30 * nirik doesn't think that would be all that useful really, given that you couldn't install more packages on it... 16:44:42 I mean it seem like you need a quite different OS structure for that to work reasonably 16:44:43 if that work was done it would be more interesting. ;) 16:44:48 sgallagh: how do you update those addons? dnf? 16:44:57 or is that some sort of layered ostree thingy? 16:45:03 nirik: The other option of course is producing a Server Role like "OSTree Factory" 16:45:06 it's pretty neat that you can roll back changes from an update 16:45:18 Where end-users could produce their own atomically updated installs 16:45:38 that would be much more interesting yeah... 16:45:49 sgallagh: neat but sounds out of server purview 16:45:57 then you install, setup the ostree factory for your needs/packages and generate new ones as you like. 16:46:03 sgallagh: that is an awesome idea 16:46:09 (but sadly that still requires rebooting for each update right?) 16:46:28 nirik: I do not see how you avoid it in the os-tree world 16:46:49 nirik: thats the entire point 16:46:50 nirik: I think one could kexec(), but that's virtually identical 16:46:52 yeah, me either 16:48:17 So, such a Factory would require a pretty considerable investment in effort. I suspect it would basically amount to repackaging a large portion of the rel-eng infrastructure. 16:48:39 so, uh, good luck with that. 16:48:40 * nirik could see using that for koji builders. Then you could bisect when something broke on them and find the trouble package 16:48:50 sgallagh: i've built ostree images before, it's not easy but not impossible either 16:48:52 yeah, it seems like it would be a lot of work indeed. 16:49:24 jds2001: Right, we would really need the right tools 16:49:54 Much of this work is probably already done by the Project Atomic folks 16:50:02 They just only expect to produce a single tree 16:50:08 (And revisions of that tree) 16:51:08 sgallagh: I could see a ostree storage server 16:51:14 or ostree dns server 16:51:34 things that would be consumed as an applaince 16:51:41 appliance 16:51:50 dgilmore: I guess I'm thinking about ways that a user would produce their own appliance images with these tools 16:52:02 Or possibly a CSB workstation image, etc. 16:52:35 sgallagh: using ostree to make 1000 webservers exactly the same should be doable 16:52:50 sgallagh: or bank teller workstation etc 16:53:07 what it needs is documentation on how you do it 16:53:16 Yes, absolutely 16:53:57 dgilmore: I was thinking something on the order of: write a spin-kickstart, pump it through a service, get an ostree as output 16:54:25 sgallagh: ostree, and installer 16:54:31 so you can pxe install it 16:54:35 ack 16:54:39 or dvd install it 16:54:59 or make cloud images easily 16:55:21 jds2001, mhayden, danofsatx: I'm curious what your thoughts are on this, as our resident Real World users. 16:56:16 personally. I don't like it. but that's only because I'm unfamiliar with it and I don't like change that I don't at least partially understand 16:56:21 (And whether this is something that seems like it would be useful or if you'd be more likely to use Atomic) 16:56:37 so, don't take my opinion - it doesn't count ;) 16:57:21 sgallagh: i think being able to do the entire thing on-premises is essential 16:57:25 with varying repos, etc. 16:59:09 OK, I'll talk with some of the Atomic people and get an idea of what kinds of tooling they have around and whether it could be adapted for such a purpose. (Or if we'd be looking at a whole new approach) 16:59:33 Alright, I have to get to another meeting, so is everyone alright if we skip Open Floor today? 17:00:11 silence == consent 17:00:14 #endmeeting