[00:29] anyone else seeing new unit test failures on master? [00:29] something changed for me over the past hour or so [00:51] fuck... [00:52] cherylj: I'm going to beg forgiveness [00:52] I'm doing a drive by [00:52] during a bug fix [00:53] because what I found is unneeded code [00:58] wallyworld: ping [00:58] wot [00:58] it wasn't me [00:58] wallyworld: do you recall how to have a JujuConnSuite test use a different user? [00:58] it was someone else [00:58] it was [00:58] I've not even bothered to look who [00:58] hmmm [00:58] I don't care [00:58] in particular [00:58] was just messing with you [00:58] I'm wanting to add a test for bug 1570594 [00:58] Bug #1570594: read access to admin model allows grant

[00:58] i don't recall offhand, but i can look [00:58] there are feature tests that the command is hooked up [00:59] but I want to say "run this command as this user" [00:59] seems like something we would want to do [00:59] and I thought you may have done something like this (or axw) [00:59] yeah. you can log in from the CLI. but for a test, i'd have to look [01:01] wallyworld: in particular, I was thinking around the new local user / cred storage story [01:02] thumper: i see another feature test that invokes "logout" and then "login", which will work,but there's a way to manipulate the suite directly i'm sure [01:02] if I wanted to add somthing that would say "use this user / pwd combo" for the CLI command, how do we do it [01:02] there's a current user in the yaml [01:02] ew [01:02] current-account [01:02] we want a repo suite method that says "set this user as current" [01:03] yes, i'm looking for that [01:03] the other ways invoke setting up yaml that the cli would use [01:04] that's gross [01:04] we can do better [01:04] eg create > 1 account on a controller with different access and set up one as current [01:04] yes [01:04] how about... [01:04] ... [01:04] * thumper thinks [01:04] * thumper looks at the code some more [01:05] thumper: i can't see any feature tests for grant/revoke even. so it may be something we've not added to the suite before [01:05] it appears that the JujuConnSuite uses the default config storage (ie disk) [01:05] thumper: what about OpenAPIAs() [01:06] you get a client connection as the user you want [01:06] featuretests/cmd_juju_model_test.go [01:06] that is where the grant and revoke feature tests are [01:06] it calls the cli [01:06] from the outside [01:06] so I need to set up the "current user" [01:06] thumper: see TestCanPostWithLocalLogin [01:07] wallyworld: wrong level of interaction [01:08] so you want to set up the current user for a CLI call? [01:08] yep [01:08] juju switch-user i think [01:08] ah that's gone [01:08] nah... again wrong level [01:09] I just want to set using the config manager bits [01:09] as in: start of test create read only user, set that user as the current one [01:09] you can set up an in memory config [01:09] then invoke the cli [01:09] one sec [01:10] I'm sure we do this shit elsewhere [01:10] just with all the changes around user cred storage, I'm not sure where to look any more [01:10] it isn't where I last looked [01:11] thumper: see createmodel_test [01:11] in cmd/juju/controller [01:11] the setuptest sets up an in memory config [01:12] I don't want a memory config [01:12] but at least I can see part of what I want there [01:12] you want an on disk one in a fake home? [01:13] where is the disk config? [01:13] that is what the test currently uses [01:13] jujuclient package [01:14] because we don't have enough things with the name juju in it? [01:14] geez [01:14] NewFileClientStore() [01:14] well that's what it is [01:14] functionality for the Juju Client [01:14] we could have called the package George [01:15] heh [01:16] it seeme the JujuConnSute has one already in ControllerStore [01:17] so... what are the minimal steps I need to add a user/pwd pair to the current model for the store? [01:19] thumper: controllerStore.UpdateAccount() [01:19] that will create / update a named account (user) [01:19] for a given controller [01:20] k [01:20] the AccountDetails struct has user/password plus a macaroon which is optional [01:21] well, i think you can leave it off when testing [01:22] thumper: the controllerStore on JujuConnSuite is right now just used to hold controller info when testing CLI command structs directly. if you want to add users etc it may be best to rename it [01:23] thumper: although TestAddUserAndRegister uses the account bits [01:24] in cmd_juju_register_test [01:24] I think I have what I need... [01:25] thumper: ok, np. if you get stuck, the above test shows how to setup NewAPIConnectionParams to create an api connection [01:25] * thumper nods [01:25] using the account details you added to the store [01:33] and so it was foretold [01:33] that tests would fail [01:33] on this, the 21st day of April [01:33] when thine distro-info revealed [01:33] that the lts [01:33] is [01:33] xenial [01:33] http://juju-ci.vapour.ws:8080/job/github-merge-juju/7506/console [01:34] and no merges shall merge [01:35] Bug #1572798 opened: list-clouds or show-cloud does not show default region [01:37] * thumper headdesks [01:38] mgz, sinzui ping? [01:38] hi cherylj [01:38] hey hey sinzui [01:38] happy release day [01:38] nothing can merge now [01:38] :D [01:39] what can we do to get things to merge so we can fix the failing tests incrementally? [01:40] cherylj: 1. I open the lst cherry soda. [01:40] can we hack the distro-info on whatever instance we provision? [01:40] cherylj: 2. Just after the test script injects angsty Antelope, We change Xenial to supported or devel [01:41] sinzui: 3 - we assign these bugs POST HASTE! [01:42] alright wallyworld and thumper, time to pony up some bodies to fix unit tests :) [01:42] wallyworld: how do I set the newly added user as the default user? [01:42] cherylj: I thnk merges/*ix unit tests are fixable in 30 minutes. I hope the window's unit tests don't need sucha a hack. I cannot think how they would think xenial is lts [01:42] cherylj: redir is already doing a whole bunch :-) [01:42] cherylj: and perrito666 fixes the windows ones [01:43] sinzui: that would probably be the fallback default series, which has not been updated to be xenial yet [01:43] cherylj: , No, first open soda and order child to bring ice cream, then fix test [01:43] check for shellfish [01:43] thumper: SetCurrentAccount() [01:43] since your kids seem bent on killing you [01:43] wallyworld: yeah... that didn't work [01:43] wallyworld: there are 7 bugs for different packages with failing tests [01:43] and that's just in juju/juju [01:43] I didn't test the other repos [01:45] wallyworld: perhaps a quick hangout for a screenshare and you can tell me what I'm doing wrong [01:45] cherylj: the guys have been told to pick stuff from the board and/or ask you, i'll ask them to prioritise unit test fixes [01:45] thumper: ok [01:45] cherylj: it seems like the no matching tools one should be a quick fix [01:46] TBH I'm surprised there weren't xenial tools added to the tests before [01:46] thumper: some might be solved by changing testing/environ.go:const FakeDefaultSeries = "trusty" to "xenial" [01:46] * thumper nods [01:46] but not all [01:46] wallyworld: I'm in the hangout [01:47] cherylj: can we tag the cards with "unit test" so they are easily identified? [01:47] there's a lot of cards on that board [01:47] wallyworld: sure. These are the only critical cards in tech-debt [01:47] Bug #1572798 changed: list-clouds or show-cloud does not show default region [01:48] cherylj: they are the ones i've already asked redir to look at [01:48] the LTS ones [01:48] cherylj: I think changing the born date to 2016-05-21 will work. DO you agree [01:48] * cherylj checks locally [01:49] sinzui: yes that works [01:49] :) [01:49] if by "born date" you mean "release" [01:50] same thing [01:59] Bug #1572798 opened: list-clouds or show-cloud does not show default region [02:00] cherylj: I will send an update to all the slaves. It will take about 20 minutes. We might se everything start to work then [02:00] tyvm, sinzui [02:01] * redir is eod [02:01] * thumper read that initially as redir is dead [02:01] hehe [02:02] redir: i was just letting cherylj know we did assign resources to fixing the LTS test issues and weren't ignoring her :-) [02:03] redir: didn't expect you to do any more work past EOD [02:03] cherylj: wallyworld I'll have a qq about those tomorrow. I modified stuff to work with xenial and got all but 2 passing in cmd/jj/service [02:03] wallyworld: I'm not [02:03] :) [02:03] redir: I'll be here, just ping me [02:03] cool [02:04] redir: we should make sure it works with trusty too (as we're updating the merge bot to forget about xenial for now) [02:04] and hopefully we'll save ourselves pain for the next lts [02:04] cherylj: it can't really work with both at once [02:04] we need to choose [02:05] cherylj: I modified environs/config/config.go to say xenial [02:05] wallyworld: I think it would be a case of checking not against a hard coded series, but rather against the default series [02:05] Bug #1572798 changed: list-clouds or show-cloud does not show default region [02:05] and updated tests. THey worked (execpt 2) then reverted environs... and they still passed. [02:05] cherylj: that's where is disagree - tests should use hard coded values [02:05] cherylj: so that when stuff changes, we are forced to fix the tests [02:06] and we do this every two years [02:06] ok [02:06] that's just IMHO [02:06] others may disagree [02:06] but I think the multi series test charm code is looking for precise or trusty. [02:07] redir: that's because the test charm data contains precise and trustyu [02:07] that data doesn't come from lts [02:07] I think it should use the value of the system the tests are running on, wallyworld, cherylj $.02:) [02:07] from from the metadata.yaml for the test charm [02:07] wallyworld: I figured that but hadn't found where yet. [02:08] wallyworld: of course:| [02:08] redir: testcharms/charm-repo [02:08] wallyworld: thanks [02:09] redir: that charm metadata doesn't necessarily need to include xenial - it could, but not strictly necessary to get the current tests passing as that's not a root cause issue IIANM [02:11] wallyworld: this is so terrible [02:11] wallyworld: dumped the yaml, and there is "kontroller" and "local.kontroller" [02:13] wallyworld: if you have a minute i wouldn't mind you vetting what I am doing, before I do it all over the place... [02:13] redir: sure, jump in to https://plus.google.com/hangouts/_/canonical.com/juju-core-team [02:18] axw: it is your fault [02:19] jujuclient/file.go line 541 [02:19] apparently we shouldn't have multiple users using one controller [02:20] thumper: you can have multiple users, just not logged in at once [02:20] thumper: from the same client [02:20] cherylj: The update is in place and I have requeued alexis's and mgz's merges [02:20] thanks, sinzui! [02:20] axw: the code says no [02:20] see comment on line 533 [02:20] thumper: this is client-side. you can logout and login [02:21] thumper: the reason for the change was to avoid accidentally leaving your client logged into a controller [02:22] all our juju conn suite tests use the admin model... [02:23] thumper: can you just remove the admin user and then add your new user? [02:23] how? [02:23] how do I programatically logout? [02:23] axw: you have accidentally voluneered to prove we can delay xenial's release by one month in the tests. I will replay your lp1571832-tools-layered merge if needed [02:23] thumper: RemoveAccount(controller, "admin@local") [02:24] sinzui: wat? [02:24] how do you guys stop goimports deleting "gopkg.in/mgo.v2" all the time? [02:24] axw: nothing is merging because unit tests are failing now that xenial is the lts [02:25] axw: so we're hacking the merge bot to think that it's not release day for xenial [02:25] axw: the juju test suite fails if xenial is released, and it is in some timezones. I have a hack in place to prevent the tests from seeing xenial's release. Your branch got in before mgz's [02:26] cherylj sinzui: I'm confused, did I do something wrong? cancel my job if you need to [02:26] axw: no :) [02:26] okey dokey [02:26] axw: you're just our first victim to see if sinzui's hack is working [02:26] axw, you did nothing wrong, you just got into the queue first [02:26] ah, heh :) [02:26] * axw squeaks like a guinea pig [02:27] sinzui: so are the i386 tests gone now? [02:27] cherylj: yes they are :) [02:27] yay! [02:27] because I'm fixing the arm64 tests at the expense of i386 [02:27] :) [02:27] mwhudson: I've been meaning to hack goimports to be aware of context. there's a TODO in the code to prefer import paths that are used in other files in the same package IIRC [02:28] mwhudson: it's a PITA [02:28] axw: grumble [02:36] mwhudson: is there a different mgo package that it's choosing? What I do is just remove the incorrect package from my gopath [02:37] natefinch: i don't think so [02:38] mwhudson: when is it deleting the import? I sometimes get the problem where I manually add the import before I add the code, and then save, and goimports runs and removes the unused import. :/ [02:38] natefinch: on any save [02:39] natefinch: i think it's because the package name is different from the last component of the import path [02:39] but it's also possible i'm holding it wrong [02:39] mwhudson: hmm.. weird. I've done a lot of work using gopkg.in import paths, and it seems to work ok [02:39] hmm [02:41] mwhudson: just double checked and it adds the right import for mgo.v2 if it's missing, and won't remove it if it's there (and used). [02:41] natefinch: huh, ok, good to know [02:41] now why is it not working for me ? :( [02:42] is it an old version of goimports? maybe there was a bug fix at some point [02:42] i just rebuilt it [02:42] although let's rebuild again just to be sure [02:43] ok works now, my rebuild must not have actually rebuilt it ?! [02:43] anyway [02:43] sorry for the noise [02:43] glad it's working :) [02:43] no wait, no it's not :( [02:43] aww crap [02:45] natefinch: aah works outside of emacs [02:47] * mwhudson updates things [02:50] bah doesn't help [02:54] fuck fuck fuckity fuck [02:54] * thumper bitches some more [02:58] can I get a review? http://reviews.vapour.ws/r/4669/ [03:00] cherylj: can you slap a comment about the i386 about the fact that we don't support it anymore? I wouldn't know that by looking at that code. [03:00] above the i386 that is [03:01] cherylj: otherwise, ship it :) [03:06] sorry axw my fix worked, but your merge had one failure in cloudinitSuite.TestWindowsCloudInit [03:06] sinzui: yeah, I forgot to update a test, just rescheduled it [03:07] sinzui: sorry for holding up your fix [03:07] hmm... [03:08] cherylj, natefinch: powerpc seems a pretty safe "juju will never ever support" arch [03:10] oh fark... [03:10] * thumper headdesks again [03:12] * thumper headdesks again [03:12] * thumper headdesks again [03:16] mwhudson: it's really just a string. If we want, we could use arch=abcdef ... but I'm guessing that might fail elsewhere.... i386 is something we can parse as a valid arch that we don't support. [03:20] natefinch: yeah, that's the thing. It has to be a valid arch to juju (as defined in juju/utils/arch), but not one we support [03:21] and it was decreed yesterday that we are not building for i386 anymore [03:21] yay [03:21] cherylj: it's honestly kind of funny that we even have arches defined that we don't support. You'd think we could just remove those and then... problem solved. [03:30] weird, somehow half the tests in this suite aren't getting run [03:37] * thumper headdesks [03:37] * thumper headdesks [03:37] * thumper headdesks [03:37] * thumper headdesks [03:37] * thumper headdesks [03:43] thumper: having fun? [03:44] mwhudson: no, why? [03:44] thumper: just a suspicion i had [03:44] mwhudson: you must be very astute [03:45] lol [03:45] thumper: well i'm running juju tests so i have lots of time to be observant [03:46] oh um, this doesn't seem like good news [03:46] with juju-mongodb (i.e. 2.4): [03:47] PASS: allwatcher_internal_test.go:421: allWatcherStateSuite.TestChangeAnnotations 0.309s [03:47] with 3.2: [03:47] PASS: allwatcher_internal_test.go:421: allWatcherStateSuite.TestChangeAnnotations 24.110s [03:47] FARK!!! [03:47] wow [03:47] shit [03:47] * thumper passes that to wallyworld [03:47] told you [03:48] yep, horatio has highlighted that [03:48] ah ok cool [03:48] it sucks, we need to figure out why exactly [03:49] wallyworld: http://reviews.vapour.ws/r/4670/ [03:49] ok [03:50] wallyworld: it has a drive by cleanup I couldn't let lie [03:51] wallyworld: the important bit was that sharing models currently only done by controller admins [03:51] which is crazy [03:51] but that's what we have [03:52] but read only access to the controller model made you a controller admin [03:52] which I've fixed [03:52] in this branch [03:52] good :-) [03:55] thumper: so what happened to all that angst that the grant/revoke methods were on the wrong facade? [03:55] that is a different issue [03:55] that's still fucked [03:55] but we are no more fucked than before [03:55] but more unfucking needed [03:55] * thumper is emailing... [03:55] so the root issue here was a Find() was missing a condition [03:57] wallyworld: for admins, yes [03:57] we didn't have readonly before [04:08] oh FFS === hazmat_ is now known as hazmat [06:53] wallyworld: https://github.com/juju/juju/pull/5244 -- fixes critical issue [06:53] sure [06:54] axw: oh, i thought we had that filtering where needed [06:54] wallyworld: so did I ... [06:54] is this a regression? [06:54] wallyworld: apparently only for LXC [06:54] I don't know [06:54] right ok, lxc in 1.25/2.0 worked [06:54] but not lxd [06:54] wallyworld: not sure if we were ever doing the right thing for KVM, seems unlikely that we'd have changed it [06:55] kvm never got much love [06:55] on non amd64 [07:04] axw: i just asked for an extra test [07:15] wallyworld: ok, thanks === urulama|____ is now known as urulama [08:15] mgz: ping [08:24] frobware: ping [08:41] morning [08:42] TheMue: o/ [08:52] voidspace: twittering about management experiences ;) [08:54] TheMue: :-) [09:01] dimitern, frobware, voidspace: standup? [09:01] dooferlad: omw [09:31] dimitern: so on a maas controller with extra interfaces juju bootstrap fails due to a schema error in gomaasapi [09:32] dimitern: the way networks are returned fails thumper's json validation code :-) [09:32] dimitern: so a fix in gomaasapi needed [09:32] I'm looking :-) [09:32] good to find these things out I guess [09:32] voidspace: oh, what's the error? [09:33] dimitern: ERROR failed to bootstrap model: cannot start bootstrap instance: cannot run instances: cannot run instance: space 0: subnet 3: subnet 2.0 schema check failed: dns_servers: expected list, got nothing [09:34] dimitern: it looks like it can't get all the subnet info - which is possibly a problem on the configuration [09:34] dimitern: but the controller is running fine, so gomaasapi shouldn't refuse to talk to it [09:35] voidspace: ah! it already is so much better at figuring out how exactly maas api changed [09:35] voidspace: yeah, the dns_servers should be optional [09:37] dimitern: yeah, but the old approach was tolerant against schema changes like that ;-) [09:37] oh the good old days [09:38] :D [09:41] dimitern, voidspace, dooferlad: morning - running late today :/ [09:42] frobware: morning o/ [09:42] fwereade: are you around? [09:42] dimitern, o/ [09:42] dimitern: have you commissioned a node in MAAS with xenial recently? [09:43] frobware: not since BOW IIRC [09:44] fwereade: do you have 10m or so for a quick HO? [09:44] dimitern: I have no working nodes this morning... :( [09:45] fwereade: context: I'm almost done with refactoring state ports to associate them with subnets, rather than networks [09:45] frobware: I'll try recommissioning one of the nucs now, as I've upgraded to 2.0 beta3 this morning [09:46] dimitern, sure, joining juju-sapphire [09:46] fwereade: omw === sinzui_ is now known as sinzui [10:20] frobware: thanks for the review. I don't think there's much point in tacking on the type to the name, since you'll never see the name out of the context of the type anyway [10:20] frobware: unless space awareness would change that ... [10:20] frobware: would private networks like these be exposed to the user? [10:21] axw: my comment was more from a debugging standpoint should you want to identify these "things" separately. [10:22] frobware: ok then, I guess I can buy that. I'll change it tomorrow [10:22] thanks [10:25] voidspace: after all the twitter harrassment, you got the day wrong again ;) [10:26] voidspace: anything I can do to help with the bootstrap issue? [10:26] babbageclunk: I just haven't corrected it yet :-p [10:26] babbageclunk: I'm on it, thanks [10:27] voidspace: ok, I'll keep going through the tests [10:27] babbageclunk: hah [10:27] babbageclunk: you could test my branch deploying a container in the standard case [10:27] babbageclunk: I can't test due to fixing gomaasapi [10:27] ok [10:27] voidspace: what's the standard case/ [10:28] ? [10:28] babbageclunk: add a machine (used to be juju add-machine) [10:28] babbageclunk: then juju deploy wordpress --to lxc:1 [10:28] or whatever the number of the new machine is [10:28] might be 0 if you've not switched to admin, not sure [10:28] voidspace: ok cool - trying that now [10:28] babbageclunk: standard case is without extra subnets / nics which is what I was attemptin [10:28] *attemptin [10:30] what the hell [10:30] voidspace: nice correction [10:30] *attempting [10:30] a failed attemp [10:37] frobware: that nuc I tried to recommission failed\ [10:38] frobware: telltale sign from the rackd.log: http://paste.ubuntu.com/15962619/ [10:39] dimitern: I'm totally broken too... [10:41] frobware: it's even reported: https://bugs.launchpad.net/maas/+bug/1389811 [10:41] Bug #1389811: tgtadm: out of memory crash [10:41] Man that is really annoying - sometimes my machine just freezes for about 2 minutes. [10:42] I think it's journald freezing. [10:43]