[00:20] thumper: looking now [00:33] thumper: ship it with some small fixes [00:45] Bug #1538241 changed: 2.0-beta1 stabilization

[00:54] Bug #1538241 opened: 2.0-beta1 stabilization

[01:01] Bug #1538241 changed: 2.0-beta1 stabilization

[01:17] menn0: ta [01:25] rick_h_: I'll test azure and see what's up [01:25] axw: ok, there's some general "stuff left behind" i get every once in a while with kill-controller. I'm not sure if this is part of that or just how much is meant to get cleaned up. [01:25] axw: ty for following up [01:26] rick_h_: you should expect there to be nothing at all left behind [01:26] rick_h_: everything is scoped by the resource group, each model gets a resource group [01:27] axw: going to see if we can get a GSoC student to help automate that process of pulling the info for the credentials. Wow that's painful. [01:27] rick_h_: the azure creds? [01:27] axw: ok, yea that's what I assumed, but I need to look more at what you create with the azure cli process vs resource group vs model [01:27] axw: working through the whole nodejs cli to create the AD entries and such [01:28] rick_h_: sounds good. yes, very painful. amazingly, it's actually much better than when I started working on the provider :) [01:28] ugh [01:28] * rick_h_ can't imagine [01:28] the fact that I have to install nodejs makes me cry much less the multi-step "run this command with this UUID, and this one with this UUID...and..." [01:28] rick_h_: I had envisaged a CLI tool to automate all the nodejs scripty bits [01:29] * axw nods vigorously [01:29] axw: right, it must be something that we can even build into juju/Go based on using the same api in the same way [01:29] start out as a plugin and see how far we can get [01:29] sounds good [01:30] rick_h_: FWIW, MS have acknowledged that it sucks. I'm not sure what they're doing about it though. [01:30] axw: yea, well I got pulled into a "juju sucks on AWS" meeting with our folks that work with them [01:30] and I promised it'd get all better with the new ARM...and now I'm not so sure heh [01:31] the creds bit is worse, I think the rest is much better [01:31] definitely [01:31] hopefully just a scripting headache to hide from users [01:47] hmm... [01:47] we have this thing called annotations [01:47] and it can be set on "anything" [01:47] but in practice, it seems like it is only ever stored on services [01:47] anyone know of other places it is used? [01:48] thumper: not on anything... [01:48] thum [01:49] thumper: on "gloab entities"... essentially, anything that has a global key and a tag [01:50] thumper: machines, units, services, charms, models [01:50] well... sure [01:50] but in practice, I think it is just services [01:50] thumper: juju itself may not use all of these but clients might.. i believe GUI is a prime example [01:50] yeah [01:50] * thumper goes for the JFDI everything card [01:51] thumper: it would b lovely to know how customers are using Juju (as in feature-focused customer feedback) [01:51] rick_h_: I can't reproduce leftovers after kill-controller [01:51] * anastasiamac dreaming [01:52] rick_h_: sure they weren't there before you bootstrapped? can you delete manually and see if it happens again? [01:53] axw: will see. I gave my creds to mark to try out so have to check if he's using it first [01:53] axw: did you deploy anything into it? [01:54] rick_h_: no rush I guess. I think CI will notice if junk gets left behind [01:54] rick_h_: nope. did you? [01:54] axw: yes, jenkins/haproxy, related them, exposed haproxy to get the public IP addr [01:55] axw: hmm, maybe I was impatient, I don't see the resource group in the portal at all now [01:55] rick_h_: deleting resource groups is super duper slow [01:55] axw: I might have caught it with no machines/etc but with the IP/availability set left or something [01:55] axw: hmm, ok. I'll try again and see and report back [01:55] axw: ty for attempting to reproduce [01:55] thanks, no worries [02:22] thumper: here's the State reference counter: http://reviews.vapour.ws/r/3921/ [02:22] axw: ping [02:22] menn0: pong [02:22] axw: I noticed something strange in master [02:22] * menn0 digs out some links [02:24] axw: https://github.com/juju/juju/blob/master/cmd/jujud/agent/machine.go#L1062-L1074 [02:24] axw: in master the stateworker has some logic to call unregisterSimplestreamsDataSource when the *state.State is closed (when the worker terminates) [02:25] axw: but the related code to NewStorage and registerSimplestreamsDataSource is no longer present [02:25] axw: openStateForUpgrade still has it though [02:25] axw: and it's in the 1.25 branch [02:26] axw: do you think this is an unintended deletion? or are the calls to NewStorage and registerSimplestreamsDataSource no longer required any more? [02:27] menn0: not actually sure. anastasiamac, ideas on above? ^^ I think this is related to the structured metadata changes [02:27] * axw pokes around code [02:27] axw: m not sure why it was removed? an overzealous merge maybe? [02:28] axw: wow, 6min at "All hosted models reclaimed, cleaning up controller machines [02:28] and counting [02:28] rick_h_: heh, yeah :/ [02:28] rick_h_: that's the cost of being 100% sure we're not leaking resources on azure, sadly [02:28] * menn0 digs into the history a bit [02:28] axw: but yes, all things look gone this time [02:29] axw: so I'll chalk it up to impatience last time ty [02:29] rick_h_: cool, thanks for confirming [02:29] menn0: eyes of the eagle \o/ [02:29] *an [02:30] anastasiamac: well I'm completely changing how the state worker starts so I need to understand how this works :) [02:37] menn0: I *think* we don't need it anymore, because we asynchronously populate the state db with metadata, and then return those results to StartInstanceParams. anastasiamac, you did that work... does that sound right? [02:37] anastasiamac: that wasn't done for 1.25 was it? which would explain why it's different in the 1.25 branch? [02:38] axw: m not remembering it being asynch.... but i *hope* that it was in 1.25 rathern than not... [02:38] anastasiamac, axw : except what's in master looks wrong... like someone half completed a change or messed up a merge [02:38] anastasiamac: asynchronous, as in we have an image metadata worker. [02:39] 1.25 looks sensible [02:39] menn0: right, the unregister certainly shouldn't be there if we don't need to register it anymore. still looking.. [02:42] menn0: removed by wallyworld in https://github.com/juju/juju/commit/66a6121c6598b2de7d8ac3acec7afbeccdcd958a [02:43] wallyworld: do you recall why this change was made? https://github.com/juju/juju/commit/66a6121c6598b2de7d8ac3acec7afbeccdcd958a#diff-f5ec9ed405cc8f3a833355afdc629bd3L1110 [02:43] axw: otp, give me a few minutes [02:56] axw, anastasiamac: thanks for the thinking and digging so far. [03:00] menn0: axw:wallyworld so the review for it http://reviews.vapour.ws/r/3459/ [03:00] has comment [03:01] still in meeting sorry [03:01] to my question of "why is this removed/no longer needed" [03:01] the response is [03:01] We only use simplestreams datasources are first searching in state database. Why add state database as a simplestreams datasource? What ends up happening is we search twice in the same location. And the logs show the search the second time errors anyway, so it's doublely evil. [03:02] .... as a consequence, the registration was removed (registerSimplestreamsDataSource) [03:02] so we just need to remove the unregister call [03:02] looks like \o/ [03:02] anastasiamac: ok, so that means that the openStateForUpgrade should be similarly updated too [03:03] I might JFDI that into master now [03:03] actually no, i'll do it in my feature branch [03:03] it's relevant to the changes there and is almost done [03:04] menn0: i think so :D... [03:04] axw: the call to unregsiter is not doing anything, i think... removing it would be clean \o/ [03:18] axw: menn0: i think that change was because we were using the stream setting for tools to filter image metadata [03:39] wallyworld, anastasiamac: http://reviews.vapour.ws/r/3922/ [03:39] looking [03:40] menn0: awesome, ty === ses is now known as Guest3272 === gberginc_ is now known as gberginc === blahdeblah_ is now known as blahdeblah [08:50] wallyworld: a handful of juju/api.go-related changes, when you have a moment: https://github.com/juju/juju/pull/4490 [08:50] * axw bbl [10:08] halp [10:31] frobware: you want to do 1:1? [10:32] voidspace: yep [10:32] frobware: I'd like coffee first ~5mins? [10:32] voidspace: ok [10:41] frobware: omw [10:42] frobware: uhm... I joined the wrong room. Really on my way now... === kadams54_ is now known as kadams54 [12:53] morning all [13:29] dooferlad, dimitern, voidspace: http://reviews.vapour.ws/r/3914/ the very small master->maas-spaces2 merge we talked about this morning [13:31] frobware, why not bring in the latest master instead? [13:34] dimitern: can do. I thought we said this morning we would just do this and then the next... [13:34] dimitern: happy to drop and go for the bigger change. [13:34] dimitern: btw, do you have a link to your PR from this morning? [13:36] dimitern: if it doesn't turn up in RB my experience is that RB does not like much markdown formatting. Remove all that and it will probably turn up. This is why I have a bunch of discarded reviews - It took me a while to realise that it was the commit message formatting that was "objectionable". [13:37] frobware, I think it's better to keep tracking the latest master - i'll deal with merging stuff once my PRs are approved [13:37] frobware, here it is: https://github.com/juju/juju/pull/4481 [13:38] dimitern: looking at that I'm sure the bullet lists are "objectionable" for RB [13:42] frobware, I added those later; I suspect the issue is with PRs that have only first line but no description, so the RB bot fails to add the link to the description [13:42] dimitern: I had problems where I had 4 spaces instead of 2 at the beginning of a line [13:43] weird.. [13:43] * dimitern will be back in about 1h [14:21] Bug #1548333 opened: 2.0 bootstrap error messages and help are not helpful [14:24] Bug #1548333 changed: 2.0 bootstrap error messages and help are not helpful [14:27] Bug #1548333 opened: 2.0 bootstrap error messages and help are not helpful [14:30] Bug #1548333 changed: 2.0 bootstrap error messages and help are not helpful [14:42] Bug #1548333 opened: 2.0 bootstrap error messages and help are not helpful [15:35] morning ericsnow [15:35] natefinch: hey [15:36] ericsnow: almost got the uniter stuff in over the weekend, but adding a resource to the wordpress testcharm caused a lot of failures in tests that use that charm but didn't register resources :/ [15:36] natefinch: nice (and not so nice) [15:37] ericsnow: yeah, and trying to register resources caused import cycles.... so now I'm just changing all of the uniter tests to hardcode starsay rather than wordpress (after making starsay a duplicate of wordpress, with some resources added). [15:38] natefinch: yeah, I ran into those same import cycles (remember the last time I was complaining about internal tests) [15:38] ericsnow: heh [15:38] ericsnow: yeah [15:39] ericsnow: (lack of) import cycles *are* a benefit of external tests :) [15:39] natefinch: our future selves always thank us :) [15:40] natefinch: anyway, I too ended up going a different route to avoid the circular imports, rather than fixing them (a too-large project) [15:42] ericsnow: yeah, kinda sucks. [15:46] natefinch: regardless, let me know if there's anything I can do to help [15:48] ericsnow: I'm making a copy of the wordpress charm to use only in the uniter tests where we need wordpress+resources, and hopefully that'll isolate it so it doesn't break everything else that wanted the old wordpress charm. I think that's the last thing to do for this. [15:48] natefinch: sounds good [15:50] ericsnow: now to wait 5 minutes for the uniter tests to run [16:22] natefinch: see my reply comment about adding the IncCharmModifiedVersionOps in Staged.Activate() [16:23] ericsnow: ok, thanks for pinging me about it. [16:23] natefinch: yep [16:24] ericsnow: ahh, thanks for the note on what you were thinking. I was overcomplicating it in my head. [16:25] natefinch: yeah, I could tell what you are thinking [16:25] natefinch: dealing with that *in* the transaction is indeed a pain, though not quite a nightmare :) [16:28] ericsnow: also, I just stumbled on a better way to test resources, I think: https://github.com/juju/juju/blob/master/worker/uniter/uniter_test.go#L1976 [16:29] ericsnow: we can append resources to the existing metadata.yaml for just the tests we care about [16:29] natefinch: smart! [16:29] ericsnow: figured it out because this test was failing with invalid yaml [16:30] ericsnow: sort of surprised this test works at all... it seems to assume there's already a newline at the end of the yaml [16:48] dimitern: heh, so I have to reimplement environ.Spaces() ... [16:49] dimitern: the assumption that SpaceProviderId is a name is built into the way we fetch subnets and spaces [16:49] dimitern: and so switching requires changing the way we fetch subnets, which means rewriting Spaces because it builds the spaces from the subnets... [16:49] dimitern: ah well [16:54] voidspace, in a call, sorry - will get back to you shortly [16:54] dimitern: no need, was just commenting [16:55] voidspace, ah, ok then [17:06] ericsnow: keeping error values and call names in sync in the stub tests is killing me [17:07] natefinch: sorry [17:08] ericsnow: we might want to build in a test that makes sure the number of errors prepopulated is the same as the number of calls... I think I see a place where I had more errors than calls [17:09] natefinch: maybe so [17:27] dooferlad, can I trouble you with a couple of small reviews? just deps updates to fix LP:#1534643 [17:27] Bug #1534643: cookies file locked for too long

=== natefinch is now known as natefinch-lunch [17:50] anyone up for some short & sweet reviews? http://reviews.vapour.ws/r/3931/ and http://reviews.vapour.ws/r/3932/ [18:29] I think the unit tests just popped up a browser window :/ [18:29] Juju Charms API -> log in with Ubuntu One === natefinch-lunch is now known as natefinch [18:32] dang [18:32] that is new? [18:34] I dunno.. was just running a full juju-core test [18:36] cmars, if you havent gotten your reviews, tim should be on soon [18:48] * perrito666 cuts storage bucket out of providers [18:48] I have been removing stuff for a month from juju :p [18:57] perrito666: you are doing god's work [18:57] lol [18:57] I wouldn't put ian so high :p [19:03] heh [19:16] natefinch: there is a mail talking about the browser issue [19:17] perrito666: yeah, cmars let me know it's fixed in latest master [19:47] neat, a new sporadic failure: /home/ubuntu/juju-core_2.0-beta1/src/github.com/juju/testing/mgo.go:489: &errors.errorString{s:"no reachable servers"} ("no reachable servers") [19:48] natefinch, we dont need more of those [19:48] alexisb: indeed [19:48] and with that I go to lunch :) === alexisb is now known as alexisb-afk [20:20] is it common that i see this with a longer-running beta-2 model-controller? consistently when i creat-models and destroy them after about the third one i start seeing this plague my units http://paste.ubuntu.com/15173762/ [20:20] if i upgrade-charm it typically goes away and triggers the install [20:22] if user==lazypower && rand.Under(0.33) { return failwhale() } [20:22] natefinch - git-blame schenanigans.go [20:23] lazyPower: it's weird that it's a sha hash mismatch... I wonder if it's a network problem screwing up the download [20:32] I'm hitting them all today: github.com/juju/juju/state/cloudimagemetadata/image.go:86: cannot save cloud image metadata [20:46] rick_h_, ericsnow: are charmers allowed to update a charm and add/remove resources from the metadata.yaml? I presume yes. I don't think that's something we've thought about [20:47] natefinch: adding new ones should work fine as-is, removing old ones should be covered once we handle removal [20:50] ericsnow: your charmstore poller looks like it's just marking what resources are out of date, but not storing the data on what's in the charmstore anywhere... don't we need both? [20:50] natefinch: yeah, I'm working on that right now [20:51] ericsnow: ok, no problem... wanted to make sure I wasn't misunderstanding either your code or the spec :) [20:52] ericsnow: heh, we're going to have like 4 or 5 representations of a resource in the DB, depending on its state... staged, pending, active, waiting to be updated from the store... yeesh. [20:53] yep [20:53] we should scrap it all and turn it into a state machine ;) [20:54] * natefinch waits for eric to say he already did that ;) [20:55] OMG... third sporadic test failure... good old "bad record MAC" [21:14] thumper: with the new manifold stuff, do you know how to conditionally start a worker? previously, you'd just use an if statement in jujud/machine.go === alexisb-afk is now known as alexisb [21:18] wallyworld, heyhey [21:18] * thumper lets fwereade answer [21:18] fwereade: hey, maybe you can answer :-) [21:18] wallyworld, an if statement in th start func should do the trick -- depending on how changeable the value you're depending on is [21:19] fwereade: what do i return from the start func? can't be nil [21:19] is there a specific error? [21:19] wallyworld, what people always used to do was return FinishedWorker IIRC [21:19] wallyworld, that squicks me out [21:19] wallyworld, dependency.ErrMissing would probably do the trick though [21:20] wallyworld, that means "I'm not going to do anything with my current set of dependencies" [21:20] sounds good, i'll try that [21:20] wallyworld, but will try again if any dependency does come up or go doown [21:20] yeah, that was sort of my issue [21:20] i didn't want it to retry [21:48] cherylj: where can i see the last bless of master? [21:49] katco: http://reports.vapour.ws/releases#master [21:50] cherylj: duh ty [21:50] np :) [22:01] natefinch: yes, they can add/remove. I'm not sure what removing will do. Since you're always upgrading ahead and can downgrade, it'd have to keep it around for a while just like the charms themselves? [22:02] rick_h_: yeah, I don't think removing will be a problem if we keep them around (which we do). [22:03] natefinch: yea, I think we'll need some admin way to garden those out at some point, as the infinite charms has shown to be a problem in large production deploys [22:04] alexisb: you have frozen for me [22:04] thumper, lost you on the hangout === blahdeblah_ is now known as blahdeblah [22:47] natefinch - looks like a repeat offender. i just hit again on my second model :( [22:48] is there anything i can tail/append here to help other than shasum mismatch? [23:36] can I get a quick / easy review? http://reviews.vapour.ws/r/3936/ [23:36] cherylj: looking :D [23:37] cherylj: LGTM :D [23:55] thanks, anastasiamac!