19:02:04 #startmeeting docteam 19:02:04 Meeting started Wed Dec 3 19:02:04 2014 UTC and is due to finish in 60 minutes. The chair is annegent_. Information about MeetBot at http://wiki.debian.org/MeetBot. 19:02:05 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 19:02:07 The meeting name has been set to 'docteam' 19:02:15 I'll go ahead and use "docteam" here so logs are easy to find 19:02:23 thanks, annegent_ 19:02:29 who all is here? 19:02:32 hi 19:02:34 o/ 19:02:36 \o/ 19:02:51 Enough to start ;) 19:02:54 and o/ 19:02:55 cool 19:03:06 Okay, the agenda is in the mailing list post from today 19:03:38 #link http://lists.openstack.org/pipermail/openstack-docs/2014-December/005568.html 19:03:48 let's start with current status 19:03:50 #topic Description of current state 19:04:14 Lots of fact finding, I actually had to look up some stuff for this :) 19:04:24 So, we currently publish four Install Guides 19:04:34 from one directory in openstack-manuals repo 19:05:03 The files are automatically built and also the translation automation is in place for these. 19:05:56 yep 19:06:27 Last release, this blueprint addressed an audit of the install guides: https://blueprints.launchpad.net/openstack-manuals/+spec/installation-guide-improvements 19:06:47 still not finished, but pretty close. 19:06:53 huge amount of work 19:07:12 just the facts :) 19:07:43 oh also 19:07:50 #link 19:07:52 er 19:07:53 #link https://wiki.openstack.org/wiki/Documentation/InstallationGuideImprovements 19:08:20 The blueprint addressed "Replace openstack-config (crudini) commands with general configuration file editing." 19:09:17 There's a chapter about debconf output for the apt-debian version 19:09:30 #link http://docs.openstack.org/juno/install-guide/install/apt-debian/content/ch_debconf.html 19:09:56 And this one I had to look up, what's documented for DevStack? 19:10:20 what do you mean? 19:10:22 #link http://docs.openstack.org/developer/devstack/ 19:10:31 Only Ubuntu 14.04 (Trusty), Fedora 20 and CentOS/RHEL 6.5 are documented here. OpenStack also runs and is packaged on other flavors of Linux such as OpenSUSE and Debian. 19:10:35 from that page ^^ 19:10:48 I wanted to include that fact finding since Tom brought up DevStack in the thread 19:10:49 ahh 19:11:20 As much as I know, what is written for Devstack in Ubuntu also works for Debian. 19:11:29 good to know 19:11:34 wasn't sure myself! 19:11:56 Then there are install tools documented by different companies in their own domains or doc sites 19:12:00 There used to be a few glitches which were corrected, though it's been a long time I didn't try. 19:12:04 anything else relevant? 19:12:07 Oh I thought of one more thing. 19:12:23 some aspects of openstack are kernel specific (neutron comes to mind) 19:12:34 zigo: The document says that it works for Debian but that it's not documented further... 19:12:37 which might be why they support certain distro releases 19:12:39 I found this in the InstallationGuideImprovments wiki page "Modularize install guides for re-use as basic install guide for training guides.(dguitarbite/Pranav Salunke)" but I don't know how that was resolved exactly? 19:12:52 AJaeger_: Perfect then! 19:13:11 because I think training plays into the goals for the install guide too, so wanted to bring that up while stating current state. 19:13:21 I think current state for training is "working on a blueprint?" 19:13:28 anyone know? 19:13:39 annegent_: they're still playing with the Install Guide, I helped them to build it only for Ubuntu, they did not profile at all 19:13:45 annegent_: the original idea is that the training guide would follow the install guide since it covers basic service installation 19:13:46 AJaeger_: ok 19:13:53 had a guide for "Debian;Ubuntu;Fedora;openSUSE" or something like that 19:13:57 That looked funny ;( 19:14:00 Sam-I-Am: in Paris they wanted their own chapter too 19:14:04 AJaeger_: heh 19:14:15 but I didn't know latest state there 19:14:17 Sam-I-Am: they're moving closer know 19:14:23 s/know/now/ 19:14:31 Okay AJaeger_ was it basically a conditional but new guide? 19:14:33 annegent_: they could do X and get a version that meets the training guide needs (e.g., stick with neutron, not nova-or-neutron) 19:14:53 annegent_: it was our Install Guide, they just did not profile properly. 19:14:57 Sam-I-Am: yeah it was the mechanics of that we hadn't figured out 19:15:09 and those mechanics may be useful here, not real sure though 19:15:17 AJaeger_: ah ok 19:15:18 AJaeger_: last i spoke with them, they wanted a top-level guide that only includes the files they want 19:15:53 Sam-I-Am: We made some progress in Paris 19:16:05 but let's not discuss that further now 19:16:10 yes 19:16:17 ok, yeah don't want to dig too deep there, just wanted to know current state 19:16:42 which sounds like "training wants a certain install guide but haven't figured out how to build it yet" 19:16:43 yeah? 19:16:51 pretty much 19:16:56 yep 19:17:09 Okay let's move into stated goals then. I did some etherpad digging! 19:17:21 First, all the way back to Grizzly 19:17:23 #link https://etherpad.openstack.org/p/restructure-docs-session 19:17:38 My favorite quote, "What about for Havana, jettisoning the Install guides and focusing on configuration guides. Doc sequestration?" 19:17:44 now we all know that didnt' happen 19:17:54 there were board members who wanted to be sure there's an upstream install guide 19:18:02 so they gave us donated resources 19:18:11 But some other history is in there 19:18:36 themes like "Consistency for IP ranges" and "One happy path" 19:18:57 "One or two compute nodes" vs "all-in-one" was still being discussed 19:19:24 some audience analysis, and also asking "what is the main goal for the install?" 19:19:33 * sgordon sneaks in late 19:19:53 hey sgordon welcome 19:20:02 so, anyone want to throw out "main goal for install" 19:20:05 I can try 19:20:18 "basic working system" 19:20:45 look mum i ran a vm! 19:20:47 "As an admin, I want to try to install OpenStack to learn about how the components work together so that I can automate an install" 19:20:52 and it can download pictures of cats! 19:20:53 services and their interactions are too complex to figure out from a config reference 19:21:09 plus all the underlying dependencies (networking is hard) 19:21:15 "As a beginner, I want to install OpenStack on hardware I have available so that I can learn enough to ask more questions about OpenStack." 19:21:33 nothing works without networking, so we need a place to hand-hold 19:21:42 annegent_: This is *not* a guide about teaching how to do automation, if it was the goal, then we failed it. 19:21:43 sgordon: lol 19:22:16 zigo: but once you have a manual installation, you know what your automation system needs to do 19:22:18 "As an architect, I want to install OpenStack enough times that I understand it well enough to provide an analysis for its applicability to my organization." 19:22:34 Sam-I-Am: Yes, but that's not the goal, IMO. 19:22:52 Just a consequence ... 19:22:55 zigo: but if automation needs to become a goal we'd revise, so this exercise is about us agreeing on the goals 19:23:10 configure prereqs, install and configure openstack services in a specific order, run openstack commands to populate services, etc. 19:23:11 In the early days, say 2 years ago, I always wanted an Install Guide that was one-up from DevStack. 19:23:13 "As an OpenStack user, I want to know how to install OpenStack on multiple machines" 19:23:45 Closer to production than not. However as more complexity has gone into OpenStack, documenting production installs were not really possible. 19:24:06 plus the variations on 'production' 19:24:09 was? anyway, grammar is hard :) 19:24:23 like... do we include HA? how many of each service to we run? etc. 19:24:26 is it one up from devstack just due to the mulit-node-ness? 19:24:35 most peoples eyes would roll back before they even got to the openstack services 19:25:02 you'd always use multi-node in production 19:25:03 If the only goal is multi-node install then yes, HA comes into question. H. 19:25:08 er, I mean Hm. 19:25:14 does it tho 19:25:17 I don't think HA is in the scope of the install-guide, unless we contain it to a specific chapter. 19:25:18 so it gives people an idea of how the services generally fit on different servers 19:25:28 Agreed, this is not a production ready setup - otherwise we would need to cover HA and scalability concerns 19:25:29 because at that point 19:25:37 you end up talking more about pacemaker or whatever 19:25:39 than openstack 19:25:54 "As an OpenStack aficionado, I want to install OpenStack services of interest to me to poke at so I can figure out what all the fuss is about." 19:26:16 I don't know what you call "production", but for many users, it doesn't mean huge deployment with HA and 100s of machines. 19:26:18 Ok, so we're somewhere between "not really production" and "not devstack" -- I think that's fair. 19:26:24 i think one of the main goals is lowering the barrier to entry and giving people a usable system from which they can add more complexity later to support scalability, performance, security, etc. 19:26:40 Many users have only a few compute nodes without HA, and are very happy with it. 19:26:40 We need to have an answer for them too. 19:26:51 I think most companies are doing a better job at lowering the barrier to entry than we ever can do :) 19:26:54 For such a complex setup, I don't think you'd be doing things by hand anyway. 19:26:58 the minimal # of things to do that ends up with a system that doesn't run in screen or disappear when you reboot (devstack) 19:27:03 annegent_: Agreed. 19:27:08 I think a goal is also "I need to install the open source cloud on open source platforms." 19:27:10 (but I'm biased ;) 19:27:21 frankly only jenkins should use devstack 19:27:33 so I actually don't think it's about lowering barrier, I think it's about describing complexity. 19:27:56 which may mean putting up barriers in some cases 19:28:37 If we're about to describe HA setups, would you guys think it's ok to have it only in a single chapter? 19:28:40 Okay, so the goal for the Install Guide is "Readers should have a working OpenStack on minimal hardware using open source tooling." 19:28:49 zigo: we have a whole HA guide for that 19:29:24 annegent_: Ok, so we shouldn't talk about HA at all in the install-guide no? 19:29:27 readers goals: understand complexity, understand OpenStack options, have a working install when they're done? 19:29:33 annegent_: also the case of 'i have an existing system and i want to try adding a new service to it, whats the minimal way to try it out?" 19:29:47 hence the topic-based install guide 19:29:49 zigo: we don't now, but we also could do a better job at pointers to other guides, I think 19:29:54 zigo, HA has it's only guide. I would rather enhance the HA guide so that it builds on top of what the Install Guide does... 19:30:05 I'm all for pointers! :) 19:30:25 zigo: yeah I like pointers, it makes people less angry when the install guide doesn't do what _they_ think it should 19:30:37 I feel pretty comfortable defending and documenting those goals 19:30:51 I'll take an action to write that down in our wiki pages 19:30:59 as long as we're agreed 19:31:22 and we're okay with "minimal hardware" -- hm. 19:31:33 thats the barrier to entry thing 19:31:44 there's even people who dont have three nodes 19:31:48 or three VMs... 19:32:01 I can write use cases up, yep, it's up to us to define minimal. 19:32:11 I'm comfortable with 2 nodes minimal 19:32:21 not one node, I prefer to leave that to devstack 19:32:23 i've thrown around the idea of offering an all-in-one version of the install guide architecture 19:32:27 but can be dissuaded 19:32:43 lots of people understandably get fed up with devstack 19:33:06 The biggest work for an all-in-one guide would be in Neutron, I believe. 19:33:13 Lots of people use devstack for production. Can't be helped apparently. 19:33:21 Otherwise, there's not much change. 19:33:22 its not bad to document, however... the issue comes with troubleshooting. 19:33:23 Oh, and one more bit of history 19:33:25 #link https://etherpad.openstack.org/p/installation-guide-audit 19:33:35 trying to trace networking problems on a single node is a pain 19:33:39 that was from prior to that blueprint 19:33:45 I don't see much about goals in it though. 19:33:45 also trying to get a mental image of how it all sticks together 19:33:55 this is why i tend to downplay all-in-ones 19:34:01 yeah to me, single node doesn't meet the goal of "understand complexity" 19:34:09 sure, they use less hardware, but they make things overly complex 19:34:15 and you'd never have an AIO in production 19:34:16 Sam-I-Am: I agree. 19:34:20 yeah 19:34:21 so the config doesn't really apply 19:34:30 you can easily grow the two-node/three-node arch 19:34:35 so let's make sure we've got the right audience in mind also 19:34:41 Sam-I-Am, for all in one would we be better off giving people pointers to throwing it up in VMs on a single machine... 19:34:45 much of it is just config duplication with new IPs 19:34:48 "poking at OpenStack" or "OpenStack planner"? 19:34:56 Sam-I-Am, thus allowing us to use same path for rest of instructions 19:35:02 sgordon: we mention using VMs and i think a lot of people choose that route 19:35:09 right 19:35:10 there's some caveats though 19:35:27 we pretty much think the audience doesn't have a lot of spare hardware laying around? 19:35:39 nor are they likely to do a manual install in any permanent system, right? 19:35:42 And the scripts of the training team set up the Install Guide architecture. 19:35:53 So, if you want to have an automatic setup, just uses their scripts! 19:36:00 (all in three VMs) 19:36:14 annegent_: people do manual installs on actual hardware 19:36:16 annegent_, i think our audience depends! 19:36:21 annegent_: It's quite easy to get a single machine running multiple VMs to just try OpenStack. 19:36:25 okay 19:36:38 so our audience, more likely to do VMs or find hardware? 19:36:42 zigo: agreed, I'll do it on my laptop ;) 19:36:48 annegent_: yeah 19:36:49 annegent_: depends on their goals 19:36:58 Sam-I-Am: go on 19:37:14 Sam-I-Am: goes into task analysis, which I also want to hear more about 19:37:18 i really think the answer is both 19:37:20 annegent_: you can't do a lot of realistic testing with VMs due to performance issues 19:37:30 i dont think we could pigeon hole whether they will do VMs or find hardware 19:37:35 task: proof of concept for work vs. task: home lab prior to work (Thinking of the BMW guy) 19:38:05 i've been working with someone the last few weeks who is using the install guide arch as a minimal way to test single-network-node performance 19:38:07 task: install many times to learn enough to automate (that's the task I hear at user groups where I ask about the install guide) 19:38:23 task: install with different configs for compare/contrast? 19:38:35 ^^ wow is that really one of their tasks, ouch 19:38:38 lots of novanet-vs-neutron goes on 19:38:50 I say ouch because that makes my fingers and wrists ache 19:38:51 and soon there will be dvr-vs-l3ha 19:39:13 the networking guide builds minimal dvr and l3ha architectures off the install guide neutron architecture 19:39:13 zigo: I know you had some tasks, one was "task: set up a small private cloud" I believe 19:39:24 Yeah. 19:39:37 zigo: multi node, right? 19:39:53 I had some feedback from users using the install-guide for that. 19:39:58 Yup. 19:40:07 okay 19:40:10 any other tasks? 19:40:34 annegent_: did you cover the "i want to try swift" case? 19:40:38 or $optionalservice 19:40:49 oh yeah that too, we decided 3 nodes was okay for swift, right? 19:40:57 task: set up storage-only cloud 19:40:59 sometimes people build a minimal install, then add $optionalservice to try it out 19:41:03 Single controler that also does networking, multiple compute, and often a small Ceph cluster. 19:41:03 I by the way would think it'd be great to cover Ceph in the install-guide. 19:41:09 swift was a bad choice there. lets say cinder :) 19:41:16 task: add on a service to an existing OpenStack cloud 19:41:40 zigo: I want to document only upstream OpenStack please :) 19:41:43 I would not add ceph setup to the guide 19:41:48 zigo: and not get into governance/politics 19:42:00 where is the fun in that 19:42:00 Ceph is often prefered to swift for small clouds, so that it does both object storage & volume in a distributed way. 19:42:04 sgordon: hee 19:42:22 zigo: sure, but why would I use sparse resources for a non-OpenStack project? 19:42:26 pretty sure the audience doesn't need more complexity 19:42:28 Looking at the November OpenStack users survey, the majority of installations uses ceph - so integration of an existing ceph cluster is something we probably should document somewhere 19:42:28 my take is that if someone wants to write a module for X 19:42:32 would we reject it 19:42:44 annegent_: Because it's what people use in real deployments. 19:42:50 ceph feels like an HA thing? 19:42:55 not really 19:42:59 I worry about maintenance ongoing 19:43:00 any more than swift is 19:43:04 This is something that belongs in one of the other guides 19:43:09 ceph implements the swift api I think 19:43:22 yeah I want more narrow definition for install guide 19:43:25 well, for another meeting :) 19:43:29 heck we barely get to newly incubated now 19:43:36 newly integrated I mean 19:43:43 ok is that all the tasks? 19:43:49 We document the OpenStack setup in the Install Guide. And any other hardware or software enhancements will be documented elsewhere 19:44:03 annegent_: It does, and soon we'll be able to use only swift-proxy with everything else on Ceph. 19:44:13 annegent_: "where do i go from here" probably covers HA, ceph, automation, etc. 19:44:21 pointers, yep. 19:44:25 (currently, there's only a ceph back-end for the object filesystem storage, so it's not great...) 19:44:25 security... 19:44:56 scope: chosen audiences and tasks 19:45:01 we can include things like that at the end of the install guide "hey your VM works... now you can install $optionalservices and if you're ready to join the big leagues, check these docs" 19:45:04 scope: only integrated OpenStack projects 19:45:11 annegent_: i hear there is an openstack-ansible thing :) 19:45:14 scope: open source tooling 19:45:24 scope: enable translation 19:45:35 I think all those scopes are agreed to? 19:45:35 such a generic chapter "Next steps" would be a welcome addition. But let's make it short and abstract 19:45:49 oh one more scope: 19:45:51 AJaeger_: we have the structure already. 19:46:11 AJaeger_: it was one of the icehouse neutron reorg topics 19:46:14 who are the maintainers as we go towards more and more added services? the project teams or distro teams or upstream team? 19:46:52 and then really we can dive into solutions, I think we've laid out a lot of the considerations to keep in mind 19:46:57 annegent_: i dont mind adding content to the install guide, but i would like a source of "here are the minimal requirements, here is a minimal tested working config, maybe some diagrams" 19:47:07 (hopefully 15 mins is enough to work on solutions) 19:47:30 those of us that maintain the install guide generally know how services need to fit in 19:48:16 so upstream docs is the maintainer? Or is it a chapter maintainer per service? 19:48:25 Sam-I-Am: Next steps currently points to the next chapter. I was thinking of a total new chapter - but then let's call it differently. But that's another discussion, not for now 19:48:33 i think it depends 19:48:34 annegent_: we'd lose consistency with chapter maintainers 19:48:42 this has happened before 19:48:43 trove is an example where it seems there is a dedicated resource 19:48:53 but i think most people contribute across the guide 19:48:57 not to a single section 19:49:08 i dont necessarily need to write the content, but i definitely want to be involved before it gets merged 19:49:10 sgordon: yeah there's definitely not a "glance maintainer" for example 19:49:19 there's a "nova neutron glance keystone" set 19:49:24 and a "swift" set 19:49:36 and a "added on services" set that may be more easily distributed 19:49:37 we might see more of the trove-style approach as more *aaS are added 19:49:40 if we had project docs liasons, they would get poked at prior to release saying "did anything change that affects the install guide? here's what we do now" 19:49:42 Do we really have to establish this kind of rules? Sometimes, chaos works out well ... :P 19:49:42 but not sure it applies now 19:50:08 zigo: heh. yeah I hear that a lot, and will have to get more towards chaos this release, completely understand that 19:50:12 zigo: any way to reduce the chaos that is openstack is a good idea. 19:50:15 but for now I'm holding the line for install / config 19:50:24 since those are release deliverables 19:50:40 people have one goal: launch a VM 19:51:07 When I play out the distributed teams route, I nearly always figure we'd have to give up conditional per-distro output. 19:51:13 and we get them there in a step-by-step process so it isn't all magic 19:51:29 and I'm not ready to give that up yet 19:51:48 cept for those who just want to store something 19:52:02 and ping the vm 19:52:21 annegent_: given persistent packaging issues, i'd like to install just openstack from source, but rely on the distro for dependencies (read: rpc) 19:52:37 then we can pretty much be distro-agnostic as long as you meet the dependency requirements 19:52:42 Sam-I-Am: for ease of documenting as a goal? 19:53:08 dependency-management are the value add from a distro right? 19:53:18 the conditionals and testing all these distros is a lot of overhead... and we also don't see packages until after release. 19:53:28 annegent_: and also integration into the distribution, for example start/stop scripts 19:53:52 AJaeger_: thats one caveat, but we have things to write those. 19:53:58 real basic question, do distros need their OpenStack Install docs on docs.openstack.org? 19:54:23 It's also distros versus products. 19:54:28 annegent_: It's not at least for Ubuntu, as I'm doing all of their work in this regard! :) 19:54:50 (I mean, they did zero dependency packaging for Juno...) 19:54:50 For our OpenStack product - SUSE Cloud - we have the documentation including Crowbar for setup at the SUSE pages. I think Red Hat does it similar... 19:54:55 not their value add? Heh. We document for them too. All take no give? 19:55:09 annegent_, well i think we put them there for manual install instructions 19:55:20 if it wasnt using the packaged install there we would focus on RDO site 19:55:27 (same as we do for automation w/ packstack) 19:55:54 some other distros have their own install guides, but often wrapped around products (read: automation)... sometimes pay products. 19:55:54 I do believe it's important for me to have docs in the official OpenStack site. 19:56:05 so Sam-I-Am if there was this non-packaged way to install, what becomes the doc dependency for maintenance? Ansible scripts? 19:56:20 i would hope not 19:56:24 i also hear that the distro-specific docs dont always go to the level of detail we do with explaining stuff. explaining stuff is a BIG part of the install guide. 19:56:31 zigo: okay, seems that way for debian and ubuntu (I dont' know of Canonical docs site ?) 19:56:32 There's no "products" or "vendor" in my case, so we do rely on community, and nobody will do a specific Debian install-guide. 19:57:05 sgordon: you would hope not, referring to? 19:57:16 Sam-I-Am: Yeah, that's the point. The install-guide really goes into details which is good. 19:57:17 annegent_, your comment wrt ansible 19:57:24 annegent_: yeah, ansible scripts. 19:57:30 it would seem weird to go to all this effort to keep specific automation tools out of the guide 19:57:32 then annoint one 19:57:37 annegent_: we have a good source of those, though. 19:57:42 sgordon: ok, yeah want to uncover what we would be dependent upon working (and what we would have to troubleshoot) 19:57:48 annegent_: Do we have to finish soon or can we stay longer in the room? 19:57:56 AJaeger_: ah, time got away 19:57:58 Looking 19:58:13 yeah Octavia's in here next 19:58:21 Ok, let's make sure we know what we can do next. 19:58:22 can we beat them up? 19:58:24 I want to document our goals 19:58:28 Sam-I-Am: T_T 19:58:30 state the audience 19:58:31 ... 19:58:33 lol 19:58:33 state the tasks 19:58:43 on the mailing list and wiki 19:58:52 I'd also like to hear from training team 19:59:04 we didnt get to solutions, so maybe another meeting? 19:59:18 and I'll add it to the doc team agenda for next week, which is 1400 UTC (right?) Wed. 19:59:29 yep 19:59:31 "it" being "solution discussion for Install Guide" 19:59:39 does that work? 19:59:49 sure. might consume the entire hour :P 20:00:07 there's only four to choose from :) 20:00:17 Okay, handing it over, thanks for making it work all. 20:00:19 #endmeeting