Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-04-25
11:00:05 asettle Ah well, now I hang out more often in the nova channel than I like to admit
11:00:42 johnthetubaguy I was talking through an idea on those deployment docs... could we pull out the packages and add in pip install, and let the package folks replace the bits they don't need
11:01:07 johnthetubaguy I am thinking make docs in a way developers think, but re-use them for the packages
11:01:13 asettle johnthetubaguy: don't get me started on that convo dude. I don't disagree with you one bit, but there is more people out there using the package docs than you would like to believe.
11:01:17 johnthetubaguy now I am not sure its feasible...
11:01:26 asettle johnthetubaguy: doesn't work like that. Developers don't think the way that our users and operators do.
11:01:32 johnthetubaguy asettle: right, not saying we don't need the package docs
11:01:40 asettle That is very apparent from the complete disconnection between the communities.
11:01:59 johnthetubaguy asettle: oh, very much so, but I packages are keeping us apart here
11:02:00 cdent I wonder, given the constriction in contributors, if package-related docs isn't the problem of the distros. In upstream we install from git or pip?
11:02:10 johnthetubaguy cdent: thats where I am thinking
11:02:24 sdague johnthetubaguy: there is a big problem with that, in that python doesn't handle a bunch of constructs natively that you need for system packages
11:02:30 cdent it's also aligned with the large swathes of people who puppet or ansible
11:02:32 sdague like CONF_DIR or SHARE_DIR
11:02:32 zioproto johnthetubaguy: thanks !
11:02:39 asettle Hmm. To be honest, I don't know. Might be a able to find out johnthetubaguy and cdent
11:02:42 johnthetubaguy sdague: yeah, I mean theres a lot of hand rolling needed
11:03:03 johnthetubaguy like I say, it totally may not be feasible, but it keeps coming up
11:03:11 johnthetubaguy I think the output has to be package docs, thats what the users needs
11:03:26 johnthetubaguy I am really thinking about how we get something in between thats closer to devs
11:03:26 asettle Well, I mean, it all comes down to this though - even if we decide to go a different route with the docs, we need the peoople to do it, which - we do not have.,
11:03:48 cdent asettle: yeah
11:03:51 sdague johnthetubaguy: I guess I'm coming in late, what's the current issue?
11:03:52 johnthetubaguy asettle: very true, the bootstrap of any of these needs people
11:03:57 asettle I would love to say that this conversation can really go somewhere this release, but I can't guarantee anything. I just lost 4 cores including myself (in theory)
11:04:06 johnthetubaguy sdague: we suck at updating the package docs
11:04:21 asettle johnthetubaguy: Okay can you please write that on a piece of paper and sign that and send it to me? I'dl ike to frame that :p
11:04:57 cdent another option, which is dire, but may represent the reality of lack of people: don't do docs
11:04:58 johnthetubaguy asettle: sure
11:05:08 cdent _really_ make it the distros problem
11:05:30 cdent I know that's unrealistic, but it puts the scale of the problem into the proper perspective
11:05:52 cdent at one end is: no docs and the other is: glorious hand crafted by the community docs for all install scenarios
11:06:20 johnthetubaguy bad docs is worse than no docs, in many ways
11:06:55 johnthetubaguy well, I mean to say, there is great value in knowing the current docs are broken
11:07:29 johnthetubaguy so we though we did enough docs as we went, for peoples needs
11:07:47 johnthetubaguy I guess my thing is how do we make that actually true
11:07:53 johnthetubaguy sdague: is that a better problem statement ^
11:08:47 sdague johnthetubaguy: yeh, probably
11:08:58 sdague sorry, first cup of coffee
11:09:38 johnthetubaguy I am waiting for a coffee delivery, but I think I should be OK, I have one pod left.
11:10:08 sdague in my world of wishes (stuff that would be nice but I don't know if ever would happen)
11:11:14 sdague 1) python install infrastructure understands basic unix semantics which means pip install nova would be totally realistic. Thus the differences between pip and packages is smaller, and things are more likely to work.
11:11:34 sdague which is probably 1 - 2 years upstream python work to get that into the python install infrastructure
11:12:13 sdague 2) the install manuals are marked up in such a way that the install procedure could be executed by a machine, so we don't just need human testers finding issues late in the process
11:12:55 cdent (if the docs can be tested by a machine, then an install can be done by a machine, and thus we don't need the instructions in the docs?)
11:13:47 sdague docs are also there to explain what is going on, they still need some testing
11:13:50 sdague manually
11:14:23 sdague it's just adding unit testing to the docs
11:14:26 cdent On 1, I think I disagree. The issue isn't entirely on the side of python packaging, it is the extent to which we think that packaging should manage the install. We shouldn't, for example, require a conf file, paste file, or policy file. Only if they are different.
11:14:35 mikal BobBall: interestingly, it only happens on my code. There's something subtle happening I'll have to dig into.
11:15:16 cdent And files that must exist can either be used as pkg_data, or be created on the fly
11:15:45 cdent or to put another way: the issues may be with the existing complexity, not the limitations of python packaging
11:15:52 asettle Jesus this is what happens when I don't look for 5 minutes
11:16:21 cdent we could look at the difficulties with the tooling as a smell in _our_ stuff, not the other stuff
11:16:26 asettle Uhhh *raises hand* I would say that no docs is probably not beneficial
11:16:49 cdent asettle: that's why it is over on the bad end of the scale
11:17:09 sdague cdent: there is plenty of existing complexity to prune down (paste is a good example), but I very much disagree that python having no first order constructs for CONF_DIR and SHARE_DIR is not a problem.
11:17:33 cdent what's wrong with pkg data for those things?
11:18:20 sdague write me the setup.py that configures apache
11:18:28 sdague for placement
11:18:31 cdent that's what I'm saying we don't want that
11:18:57 sdague that's what I'm saying we absolutely want if we expect the pip install method to be a thing
11:19:03 cdent why should pip install nova configure apache?
11:19:15 cdent ansible, puppet, my fingers should do that
11:19:25 sdague then it's not an installer
11:20:00 sdague if "pip install nova" doesn't get to me to "service nova start" and things work, then there is a bunch of missing space there
11:20:24 cdent yes, and it is in that missing space where most people want their own control
11:20:45 cdent if we try to, ourselves, fill in that space we are giving ourselves a bunch of work that we will most likely get wrong, and will also not reach everyone
11:20:46 sdague cdent: not for the most part people using our install guide
11:20:51 cdent better to make that space blank
11:21:19 cdent the current install guide has a _ton_ of by hand work
11:21:22 sdague cdent: just to be clear, you are mostly just arguing for the status quo
11:21:27 sdague which, is fine
11:21:28 cdent which at the moment is at risk of constantly wrong
11:22:34 alex_xu johnthetubaguy: sdague I guess this is just waiting for +2 https://review.openstack.org/#/c/445864/ :)
11:22:45 zioproto johnthetubaguy: running mitaka I found out that I have rows with the same instance_uuid in the table instance_extra
11:22:46 cdent I would much rather the docs start from a common denominator (which would be be 'pip install <something' and then proceed (in the hard-way trope that's common these days) to educate people in the system they are messing with
11:22:53 cdent for the people who want automation they can go use it
11:22:55 zioproto should I understand why ?
11:23:03 zioproto it sounds like a possible bug ?
11:25:01 johnthetubaguy zioproto: good question, I am not sure, sounds bad
11:25:43 johnthetubaguy sdague: I think I was with cdent (in the crazy idea in my head), its about documenting using pip to do the install, not about making pip do all the install
11:26:05 johnthetubaguy sdague: but it does seem odd writing it down when people want a script to do it for them
11:27:02 cdent johnthetubaguy, sdague I don't think people who are trying things _do_ want a script to do it for them. They want to try it on for size by really understanding the pieces and parts. Then once they decide it is worth it _then_ they want automation
11:28:22 sdague cdent: that's probably an area where we don't agree. I've never seen that perspective in anyone I've interacted with.
11:29:05 cdent the install guide shouldn't really be an install guide, it's a "what's this openstack thing guide". Anybody who is _really_ using it will have transcended that stage.
11:30:50 cdent such a document would start with something like "if you don't care about these details, see <these OSA docs>"
11:31:18 cdent (of course in the absence of the people, all of this is moot)
11:35:26 cdent I guess underlying all this for me is that the point of upstream install docs in something as huge and infrastructurey as openstack is not to create users, it is to create contributors.
11:35:34 sdague maybe, I think that's just punting the problem. And I don't feel like those level of installers actually scale down well to small clusters.
11:38:21 sdague so, I'm assuming the starting point of this conversation is issues with placement docs?
11:39:11 cdent but of course, but we resolved that relatively quickly and moved on to general notions of how do we improve the communication about creation of docs, and then into more esoteric notions of "what are docs, anyway?"
11:39:47 cdent the original problem was that there's a mismatch in the docs between the configured endpoint for placement in apache in the packages and the service catalog endpoint
11:41:36 sdague and that it took a while to notice?
11:42:39 cdent sdague: I think more that the various threads of noticing took a while to cohere into shared awareness?
11:43:05 sdague ok, I'm not sure I understand what that sentence means
11:43:58 cdent lots of people noticed, at different times, and talked about with different people, at different times, but because the right people weren't there at the time of talking it didn't result in plans for change until yesterday afternoon
11:44:11 openstackgerrit Feodor Tersin proposed openstack/nova master: Implement ScaleIO image backend https://review.openstack.org/407440
11:44:18 cdent (that's captures the subthread in this discussion of "IRC may be hurting us, sometimes")
11:45:36 cdent jaypipes: I hope you had a good sleep and coffee already

Earlier   Later