=== alexlist` is now known as alexlist [02:45] thumper: http://paste.ubuntu.com/12069959/ [02:45] this is from this morning's run [02:45] summary: juju is now as stable on ppc64 as any other platform [02:50] hazaah [03:35] thumper: here's the final allenvwatcher pr: http://reviews.vapour.ws/r/2374/ [04:51] thumper: any chance you can look at the above? ^^ [04:51] * menn0 is keen to be done with this [04:53] * thumper is looking now [04:53] thumper: thanks [05:25] menn0: done [05:25] one (strong) suggestion [05:25] thumper: ok... [05:27] thumper: no problems. I had something like that at one point. [05:29] thumper: to be fair those funcs aren't really for truly public use (hence the comment) [05:40] well... if they have capital letters, they're public [06:55] TheMue, morning [06:56] TheMue, I see you've assigned the constraints card to yourself, but you're yet to land your other branch, so I've reassigned it to myself [07:10] dimitern: morning, the landing card is just waiting, that CI is unblocked [07:10] TheMue, cool [07:11] dimitern: so which card can I start with now? [07:12] TheMue, hmm.. let me have a look [07:13] TheMue, you have a follow-up to do, don't you? :) [07:13] also it seems I've assigned the wrong card to myself - I'm actually working on the other constraints card [07:13] dimitern: hehe [07:15] dimitern: you mean "Test isNetworkingEnvironment() in ..."? we haven't moved it from backlog to iteration backlog, so I thought it's not as important as the spaces cards [07:16] TheMue, yeah, we forgot about this one [07:16] TheMue, the net-cli stuff is the most important once the previous tasks are done :) [07:16] dimitern: ok, so I'll move it [07:16] TheMue, cheers, and feel free to estimate it as well [07:17] let's move it, move it, let's move it, move it *sing* [07:17] dimitern: ok [07:22] dimitern: ping [07:24] voidspace, pong [07:24] dimitern: did you write ListSubnets? [07:25] dimitern: unlike all the other API methods it tages a *names.SpaceTag rather than a names.SpaceTag and I'm wondering why [07:25] I changed it, as it was marginally inconvenient in one place, and a few things broke [07:25] so I probably have to put it back [07:26] voidspace, I did - it's because the space is optional for filtering [07:26] voidspace, it was supposed to make calling the client-side method more convenient :) [07:26] dimitern: and you need to be able to distinguish between not used (nil) and an explicit empty space ("") [07:28] dimitern: so, changing API methods to take tags instead of strings means changing a bunch of CLI calls too [07:28] dimitern: they need to use the tags and do the validation [07:29] so I'm now down to only 23 test failures... [07:29] (the CLI methods were already validating of course - but not always storing a tag. And it's *mostly* tests that need to change.) [07:32] voidspace, :) [07:33] voidspace, sounds good [07:44] TheMue, voidspace, dooferlad, guys, I'm suggesting to cancel the 1:1 meetings today - happy to chat about anything, if you want of course [07:44] but in the face of the feature freeze I'd rather concentrate on that with priority [07:45] dimitern: frome my side nothing special, so sounds good [07:47] TheMue, cheers [07:47] dimitern: sure [07:48] dimitern: thanks [07:55] dimitern: +1 [07:55] thanks guys! [09:01] dimitern: TheMue: dooferlad: ready in a minute or two - coffee!!! [09:32] dimitern: TheMue: dooferlad: not sure if my last message got through, my internet went down [09:32] sorry [09:32] I'm back now but the hangout was empty [09:33] voidspace: you've been frozen and then the connection dropped [09:33] voidspace: so the question is, what has been the last message? ;) [09:33] that was my last message... [09:33] "my internet went down" [09:33] :-) [09:34] voidspace, oh - I think I heard only "own" :) [09:36] hehe [09:36] dimitern: own? [09:40] dimitern: out of "down"? [09:42] voidspace, I think so yeah [09:42] (well it sounded like "on") [09:42] heh [09:42] sorry [09:42] :w [09:42] wrong window... [09:43] voidspace, so it seemed we should chat about subnet add/create vs tagging? [09:43] dimitern: well, I thought we had talked about it a few times already! [09:43] dimitern: that's why I was so surprised [09:44] dimitern: previously you seemed to agree that Create was easier and we didn't need to touch tagging for the MVP [09:44] voidspace, yeah, I might have misinformed you unintentionally [09:44] dimitern: if we *have* to do Add, then so-be-it, but I think it's *more work* [09:44] voidspace, we'll implement create, but not include in the MVP [09:45] sure, I'm talking about what is the least work we can do which gets something useful [09:45] voidspace, also add doesn't need to respect pre-existing tags on a subnet in ec2 [09:45] for the MVP [09:45] so what does it do? [09:45] voidspace, it adds it to state, validating it exists on the provider [09:45] just assumes the subnet exists and adds it to state [09:46] so it does go out to the provider [09:46] (the current implementation doesn't...) [09:46] voidspace, yes, as we have already a Subnets call we can call - it shouldn't be too bad to add this check [09:46] ok [09:46] so if we ignore tagging then Add is easier :-) [09:48] voidspace, indeed - we'll add tagging (discovery and updates) later if we manage - it's strictly an improvement in UX and observability, not so much in functionality [09:52] dimitern: fair enough I guess [09:52] dimitern: like I said, based on our previous conversations I was just surprised [10:12] dimitern: updated PR http://reviews.vapour.ws/r/2360/ [10:25] voidspace, cheers, looking [10:31] dooferlad, others: http://reviews.vapour.ws/r/2376/ spaces constraints [10:31] dimitern: already reading :-) [10:34] dooferlad, ta! [10:35] voidspace, didn't we talk about ListSubnets returning something like []params.SubnetInfo? [10:36] voidspace, I mean the client-side api method [10:37] dimitern: I don't recall a conversation about that, let me look and see what it does return [10:37] it's been a while since I wrote that code [10:38] dimitern: hmm... it returns a list of params.Subnet [10:39] dimitern: there is no params.SubnetInfo is there? [10:39] voidspace, sorry, I meant params.Subnet [10:39] dimitern: well, it has a result type that holds a list of params.Subnet [10:39] voidspace, well, looking at your branch it returns ListSubnetsResults and an error [10:39] dimitern: do you mean removing the ListSubnetsResults part from the returned value [10:40] voidspace, that's pretty unusual for client-side apis [10:40] voidspace, just return the Results part of it and the error [10:40] dimitern: ah [10:40] dimitern: ListSpaces does the same [10:41] dimitern: ah, and the CLI expects just a list of subnets [10:41] dimitern: so that needs fixing [10:41] probably the same for ListSpaces too [10:41] dimitern: thanks [10:43] dimitern: just return response.Results or can we get rid of the response type completely and just return a slice of Subnet ? [10:44] (I mean can the apiserver return []params.Subnet or do we need a Result type to hold it) [10:46] I'm just changing the api List* methods [10:46] voidspace, nope, the apiserver-side still needs to use the result type like this [10:46] yeah [10:46] voidspace, you have a review [10:46] dimitern: thanks [10:47] dimitern: why do we need to verify that the command members are set correctly on error? [10:47] dimitern: given that we report the error to the user and exist [10:48] voidspace, I'd appreciate a second look on my review as well - esp. if any of the TODOs reminds you of something I might've missed [10:48] dimitern: ok [10:49] voidspace, it's not a big deal to leave this in place [10:49] voidspace, is it? [10:51] dimitern: well, as we're now using SubnetTag I can't construct a SubnetTag for an invalid CIDR [10:51] dimitern: so in the case of error we can't check that [10:52] dimitern: and if there's no *need* for the others then it's better not to test them [10:52] voidspace, ah [10:52] we should only test behaviour we actually need [10:52] voidspace, sure, then it's fine [10:52] thanks [10:52] voidspace, unlike APIs, those CLI methods are not usable from outside the package [10:53] yep [10:53] that was my thinking [10:53] dimitern: what's the value in changing providerId to network.Id? [10:53] dimitern: there's no validation, so it doesn't really offer any benefit like SubnetTag and SpaceTag do [10:54] it can take arbitrary strings [10:54] I believe [10:54] and it probably means finding and fixing another crap-load of tests :-p [10:54] voidspace, fwereade reminded me on more than one occasion that we should use typed strings for foreign ids, like the provider ones [10:54] ok [10:54] fair enough [10:54] I'll bite the bullet [10:55] voidspace, yeah, no validation for now, having a type makes adding this easier [10:55] if/when needed [10:55] voidspace, thanks [10:55] so for our imaginery future use case then [10:55] ;-) [10:56] well :) [10:56] dimitern: network.Id is used for too many *different* things for us to ever be able to add meaningful validation [10:56] we'd have to create a *specific* type [10:56] good design and architecture includes some need for precognition and such [10:57] voidspace, it's used for provider-specific ids [10:57] dimitern: many *different* provider-specific ids with different specifications [10:58] voidspace, at least having them marked as such will simplify changing them to separate types if needed IMO [10:59] voidspace, I hope you don't mind too much for a string cast here and there :) [11:01] dimitern: I'm not totally convinced about the *semantic* value of using types (that offer no tangible other benefit) to represent "things" that actually *are* strings, and having to case to and from everywhere they're used [11:01] dimitern: I'll do it, I'm just not convinced it adds value [11:01] voidspace, I'll wait for your and mine branches to land and then create the net-cli-mvp branch [11:02] voidspace, right [11:02] dimitern: cool [11:03] dimitern: I'll take lunch shortly [11:03] I'll try and get the branch landable first [11:04] voidspace, ta! [11:25] dimitern: right, made those changes [11:25] dimitern: switching to network.Id touched 6 files, so you might want to take another brief look [11:25] :-D [11:25] maybe "want" is the wrong word [11:26] voidspace, sure, looking [11:27] dimitern: https://github.com/juju/juju/compare/net-cli...dooferlad:net-cli-add-deployment-support-for-spaces?expand=1 seems to be all https://canonical.leankit.com/Boards/View/101652562/116603811 needs. [11:27] just let me know about printing/logging the deprecation warning and I will tidy up that comment and get it in for review [11:27] dooferlad, ok, I'll have a look shortly [11:27] dimitern: thanks. [11:28] voidspace, get some lunch first [11:28] voidspace, you're eating tests otherwise :) [11:29] :-) [11:31] dimitern: well, I want to meditate and it will be a little bit of time before I can persuade my wife to clear out of the room I need :-D [11:31] voidspace, I see ok :) [11:34] voidspace, dooferlad, only now I realize ListSpaces API doesn't conform to the usual (results, error) behavior - it returns results and an error, but only the first error when fetching a space's subnets [11:35] dimitern: you mean it should collect all errors [11:35] it needs to use a Results and Error fields of the results for such errors [11:35] dimitern: as the only possible error is connection with state failing (so all should fail or none) [11:35] dimitern: I don't think it's a big issue [11:35] voidspace, yeah, only return an error if it can't return anything [11:36] I mean, we should fix it - but add a card for after mvp maybe [11:36] voidspace, well, just because we're catching this now means it lacks bulk call coverage [11:36] "bulk call coverage"? [11:36] it's a bulk response not a bulk call... [11:37] it's a single call with a bulk result [11:37] voidspace, yeah - a call to it with e.g. one valid, one invalid, and another valid - checking 2 results and 1 error is returned [11:37] voidspace, the apiserver method I mean [11:37] right [11:38] voidspace, it's trivial to fix and we should do it for the mvp by all means, as changing the api later is worse [11:39] dimitern: ok [11:39] voidspace, I'd add a card for it, but please add a TODO about it in your PR - apiserver/spaces [11:40] dimitern: TODO added [11:41] well, pushing now [11:42] voidspace, thanks! [11:42] done [11:44] right, room free [11:44] off to meditate [11:46] dooferlad, hey [11:47] dooferlad, the help docs need to mention the spaces constraints but not --networks btw [11:49] dooferlad, but there are also - tests we do ignore --networks and dropping those that don't (or at least adding TODOs/skipping them) [11:50] dooferlad, hmm sorry, I might have described it badly in the card [11:51] dooferlad, the idea is not only to give a warning for it, but also not to pass it through the layers as well, so it can't interfere with the other work around spaces for deployments [11:51] dooferlad, does that make sense? === ahasenac` is now known as ahasenack [12:11] does anyone else get build errors for gwacl on master after updating and running godeps ? [12:12] axw, ping [12:40] dimitern: yea, that isn't what I got from the card. I read it as we would still accept --networks, but we should mark it as deprecated to discourage its use before removing it in a later version. [12:42] dooferlad, we would've done that if currently it was at least remotely useful [12:43] dooferlad, and it will soon go away to be replaced by deploy time bindings of the right kind, with spaces (once we figure out how'd they look like) [13:02] anyone up for a trivial review? http://reviews.vapour.ws/r/2377/ [13:05] TheMue, voidspace, dooferlad ^^ [13:07] dimitern: *click* === lp|conference is now known as lazyPower [13:24] dimitern: you said that space contraints are handled elsewhere in the card I am looking at. Where is that? [13:25] dimitern: also, with network constraints, do you want me to just remove them or just not wire them up in the CLI? [13:25] or just leave as-is I guess. [13:26] dooferlad, they need to go, but that's a separate card, which will require a lot of prerequisites (mostly provider and provisioner related) landed first [13:27] dooferlad, not sure what do you mean by "handled" - my PR implements parsing and tests for them, and since they're constraints that makes them usable everywhere [13:27] dimitern: doh! Haven't merged that in :-) [13:28] whether we do anything with them is another thing [13:34] dooferlad, did that answer your questions btw? :) [13:35] dimitern: I am leaving the constraints code as it is, so it will accept networks. I have removed the code path that --networks used for juju deploy. [13:35] dimitern: I have had a look at add-machine and add-unit and both seem to have no explicit --network support, so they are just constraints code again [13:35] dimitern: if that is right, yes. [13:37] dooferlad, indeed - I just double checked [13:37] dooferlad, are there a lot of tests affected by --networks being ignored ? [13:38] dimitern: I just removed the two that started to fail (correctly) once support for the --network flag was removed. [13:38] dimitern: I didn't look at any already skipped ones. [13:40] dooferlad, by "removed" do you mean it returns an error like "flag provided but not defined: --networks" or it accepts it, prints a warning and continues? [13:40] prints a warning and continues as if it isn't there [13:40] dooferlad, great! [13:40] dimitern: http://reviews.vapour.ws/r/2382/ [13:41] dimitern: I haven't added any support, so much as removed other support. [13:41] dooferlad, that's a lot better - less fallout than my original attempt which did the former [13:41] so the title is rubbish [13:49] dooferlad, it's fine :) [13:49] dooferlad, almost done with the review btw [14:05] dooferlad, was the removal of that status test case necessary? [14:09] dimitern: yes, it failed. [14:14] dooferlad, that's because we're not stopping tests from calling deploy APIs with networks [14:15] dooferlad, rather than removing it, can you comment it out for now with a TODO about fixing it once the API and state layers also ignore (requested)networks (which --networks sets effectively) ? [14:15] dooferlad, I'll add a comment to the review [14:21] dooferlad, reviewed [14:43] dimitern: ping [14:47] voidspace, pong [14:47] dimitern: so, I add a *Error field to params.Space so we can return partial results [14:48] dimitern: the top level params.ListSpaceResult has an Error field (in case fetching the spaces fails - the new error is for an error fetching subnets for an individual space) [14:48] dimitern: and ListSpaces also returns an error [14:49] dimitern: so if we get an error fetching the subnets for a single space obviously we set the Error field on Space [14:49] voidspace, you did all this in your subnets api branch? [14:49] dimitern: do we set an error on ListSpaceResult and do we return an error too? [14:49] voidspace, or in a follow-up? [14:49] dimitern: no, new branch - looking at it now [14:49] voidspace, ah, great [14:49] dimitern: not sure if I can get it done before I go as I go to the gym in 40minutes [14:49] dimitern: that branch landed on net-cli by the way [14:49] voidspace, did the other one land? [14:50] yep [14:50] voidspace, awesome! [14:50] voidspace, so I added a card for the ListSpace/Subnet errors [14:50] I tried looking for other examples in the apiserver but failed to find any and thought it would be quicker to ask [14:50] dimitern: it should just be ListSpace I believe [14:50] dimitern: ListSubnets doesn't make extra calls per subnet, so either it all works or it all fails [14:50] voidspace, so usually bulk calls that return many results for many entities [14:50] (I believe) [14:51] dimitern: with ListSpaces we can fetch the spaces and then fail to fetch the subnets for some of the spaces [14:51] so partial results make sense in theory [14:51] voidspace, ..only return errors if it's not possible to do its job [14:51] I say in theory, because in practise the only error is that the connection to state goes [14:51] so again, in reality it will be all or nothing [14:51] but it's better to stick to the pattern [14:52] dimitern: right, so we don't return a top level error [14:52] dimitern: what about setting an error on ListSpacesResult? [14:52] voidspace, let me have a look at the code now for a moment [14:52] dimitern: or should we rely on the client checking the Error field of each Space [14:55] voidspace, for ListSpaces you're correct - the backing error should be returned as top-level, others can only be per-space (fetching subnets failed) [14:56] wallyworld, http://reviews.vapour.ws/r/2387/ please [14:57] dimitern: I'm sorry I don't understand [14:57] dimitern: what is "the backing error"? [14:57] dimitern: if we get an error fetching the subnets for a space obviously we set the Error on the params.Space [14:57] voidspace, coming from the state (backing) [14:57] dimitern: do we set an error anywhere else? [14:58] dimitern: so we also set an error on ListSpaceResults? or are you saying we also return the error at the very top level? [14:59] voidspace, sorry, let me explain better [14:59] voidspace, https://github.com/juju/juju/pull/2980/files#diff-ec0483c931b7d3d2d378df66d465deb1R100 < this is a top-level error (with empty results) [15:00] dimitern: yes [15:00] voidspace, https://github.com/juju/juju/pull/2980/files#diff-ec0483c931b7d3d2d378df66d465deb1L109 < this is a per result error [15:00] yes [15:00] dimitern: that's what we do currently [15:00] voidspace, in the latter case, the results != empty, err = nil [15:00] right [15:01] and we set the error on the individual params.Space we create [15:01] (we should I mean) [15:01] I'm asking do we *also* do anything else with it? [15:01] voidspace, there's no individual params.Space [15:01] voidspace, ListSpaces takes no args at apiserver side [15:02] the object there "result" is a params.Space [15:02] so instead of "return results, err" it would be [15:03] result.Error = common.ServerError(err); result.Results = append(result.Results, result);continure [15:03] *continue [15:03] voidspace, ah, got you [15:03] dimitern: that's the change you have said we ought to make [15:03] voidspace, yes [15:03] I'm asking if we should *also* set results.Error [15:03] voidspace, and I've just realized ListSubnets apiserver side is not yet done, so it's fine [15:03] and we've agreed that the returned err should still be nil [15:04] ListSubnets won't have the same problem [15:04] I'm asking if we should *also* set results.Error [15:04] and we've agreed that the returned err should still be nil [15:04] voidspace, yes, if you can fetch spaces ok, but not all subnets -> return results(with Error set as needed), nil [15:05] voidspace, if you can't fetch spaces - as it is now [15:05] so I should set result.Error for each Space where fetching subnets fail, and set a results.Error to indicate that there are only partial results [15:06] voidspace, sorry, I'm confused - what do you mean by result.Error for each space? [15:06] dimitern: every space we fetch is represented by a params.Space [15:06] voidspace, yes, yes [15:06] dimitern: that will grow an Error field [15:06] voidspace, sorry [15:06] dimitern: that is settled [15:06] voidspace, I need to get in a call [15:06] voidspace, it is, yes [15:06] dimitern: I am asking, do we *also* set an Error on ListSpaceResults [15:07] voidspace, where is that ListSpaceResults defined? [15:08] sorry params.ListSpacesResults [15:08] at the moment it doesn't have an Error field because the method would either return all results or an error [15:08] so it didn't need one there were no partial results [15:08] we're now adding partial results [15:09] so I wonder if ListSpacesResults (apiserver/params/network.go I believe) [15:09] I wonder if it now also needs an Error field [15:09] used to indicate partial results [15:09] voidspace, I believe we need to make api.ListSpaces() ([]params.Space, error) [15:10] dimitern: earlier today you said that apiserver couldn't return []params.Space and ListSpacesResults was still needed :-) [15:11] voidspace, then also change ListSpacesResults' Results field to be of type ListSpacesResult (or whatever), which has a Space and Error *Error fields [15:12] instead of adding an Error to Space... [15:12] add a new intermediate type [15:12] voidspace, it's just confusing because of the api/apiserver transition [15:12] I think you're still slightly misundestanding my question/suggestion :-( [15:12] I think you have too much in your head [15:12] we don't need a new intermediate result type, we can just add an Error field to Space [15:13] voidspace, ok, I see how params.Space can be used as what I'm describing as ListSpacesResult (singular) [15:13] I am asking if the *plural*, the container results type, ListSpacesResults also needs an Error field [15:13] because if we get partial results (an error occurred somewhere) don't we want a top level Error field to indicate that the results are partial? [15:14] maybe not, and just having the error printed inline with the space description is fine [15:15] voidspace, it's not common to have ErrPartialResults from API server methods [15:15] voidspace, and in this case I don't think it's even useful [15:15] ok [15:15] so we won't [15:15] thanks [15:15] voidspace, thanks for bearing with me :) [15:15] :-) [15:16] I go soon [15:16] so I'll be leaving you to it [15:17] ericsnow: if we always flush immediately after set, we should be able to remove the mergeProcMaps [15:17] voidspace, sure [15:18] natefinch: right [15:21] dimitern: cmd/juju/spaces/list.go - ListSpaces command fetches the subnets using api.AllSubnets() [15:21] dimitern: it doesn't use the subnets on the Space!! [15:21] I guess space.Subnets() is newer than the ListSpaces command... [15:22] I don't think I can fix that in the time I have [15:22] it will make the code simpler though [15:22] I'll add it to the card [15:23] voidspace, ack, otp - thanks [15:26] dimitern: done, card updated - including the code to set the error correctly [15:26] dimitern: have a good week [15:26] TheMue: dooferlad: EOW [15:26] I begin my journey home [15:26] good luck with the MVP [15:26] voidspace: have a safe trip [15:26] thanks [15:28] ericsnow: updated the code according to your suggestions, would like a re-review so we can get it done [15:28] dimitern: maybe I'm missing something and you can point me to the right way. I've got my test, I've got my mock, but the way to inject is seems difficult [15:28] natefinch: will do [15:47] ericsnow: realized I didn't publish my responses to your earlier comments. Just did so. [15:47] natefinch: thanks [15:47] (only one real comment) [16:01] TheMue, can I look at the code you're having issues with? [16:03] dimitern: do you have an opinion about what card I should do next? [16:05] dooferlad, I don't see a card for apiserver subnets - create and lest [16:05] list [16:05] dimitern: it's not a concrete issue, more a missing idea. [16:06] dimitern: you meen the apiserver side of what voidspace has done? [16:06] dimitern: in the test in http://paste.ubuntu.com/12109009/ I try to inject the mock [16:06] dooferlad, yeah [16:07] dooferlad, we seem to have missed this - can you add a card and start on Add first perhaps? [16:07] TheMue, I'll have a look shortly [16:07] dimitern: isn't it in the archive? https://canonical.leankit.com/Boards/View/101652562/115620070 [16:07] dimitern: also https://canonical.leankit.com/Boards/View/101652562/115620186 [16:08] dooferlad, both of these are for spaces [16:08] dimitern: doh! [16:08] dimitern: yes, you are right [16:09] dooferlad, I know it's confusing - it gets me every time :) [16:12] dooferlad, and I'm wrong - apiserver has AddSubnets, but no list yet [16:12] dooferlad, we don't need CreateSubnets for the MVP, so we can just do Add and List [16:13] dooferlad, there's a card I added re ListSpacesResults and ListSubnetsResults - if you prefer [16:13] dooferlad, also voidspace has a branch with some progress on it [16:15]