[00:30] davecheney: if you are free, i'd love a review for a critical 1.15 release fix https://codereview.appspot.com/14011043 [00:30] * davecheney looks [00:31] thanks :-) [00:32] wallyworld: LGTM. not a lot to complain about there [00:32] \o/ [00:32] if you want, i *can* complain [00:32] i have had a lot of practice [00:32] i really wish we have CI testing cause these things would have shown up sooner [01:04] does anyone know what 'provider state lookup' is ? [01:43] davecheney: I thought ` quoted strings were raw? [01:43] but it is complaining about \ [01:44] what is the difference between ` and " ? [01:45] ` is raw [01:45] so [01:45] `"tim"` == "\"tim\"" [01:45] what is 'it' ? [01:45] a fuck sticks [01:45] there is a mix of ` and " in the file for error matches [01:55] https://bugs.launchpad.net/charms/+source/haproxy/+bug/1231768 [01:55] <_mup_> Bug #1231768: config-changed: fails when global_maxconns is not set [01:55] what has gone on here ? [02:01] * thumper shrugs [02:01] this is a bug about all bool options being false, perhaps related? [02:10] wallyworld: I'll send an email to the list about the local provider issues [02:10] thumper: ok. did you notice one of your branches had a few small test issues? [02:11] yeah, fixed and resubmitted [02:11] thumper: i'm trying to get legacy tools removal finished. i think we should land that because legacy tools does not pick up checksums and running from trunk now uses simplestreams anyway [02:12] wallyworld: how does this impact the local provider? [02:12] the local provider uses simplestreams now that i have landed a previous fix [02:12] it was failing for you because it was failing bacl to legacy [02:12] hence no checksums [02:13] so removing legacy will catch the error earlier [02:13] s/the/any [02:30] ok [03:09] wallyworld: null-provider is getting "cannot set agent tools for machine 0: empty size or checksum" -- is this related to not implementing SupportsCustomSources? [03:09] axw: yes. it needs to do what i just did for local provider [03:09] I just added the GetToolsSources method to nullEnviron, but I get that still [03:09] wallyworld: thanks, I'll see what else you did [03:09] what did you return [03:10] wallyworld: same as local [03:10] that should have worked then [03:10] it will have the checksum error if it can't find tools via simplestreams [03:10] are tools uploaded? [03:10] is this in a test? [03:11] wallyworld: simplestreams metadata has been uploaded [03:11] wallyworld: in a live test against my VM [03:12] if you run with debug and you see a messaging saying can't find tools, using legacy (or something like that) then it is not finding the tools meatdata [03:13] wallyworld: nothing about legacy, just this: [03:13] 2013-09-27 03:09:57 ERROR juju runner.go:211 worker: exited "upgrader": cannot set agent tools for machine 0: empty size or checksum [03:14] that will happen as a result of not finding tools metadata. or, it could be a Tools{} struct is being constructed somewhere without Size and Checksum being set [03:15] wallyworld: I guess it's not finding it, because what's there does have the size and checksum [03:15] I'll keep digging [03:15] run with debug and paste the output if you like [03:16] are you sure the tols have been put in the right place? [03:18] bbiab [03:30] * thumper upgrades to saucy to look for lxc issues [03:30] I'm having trouble replicating some of the bugs with raring [03:32] thumper: god's speed [03:33] wallyworld: ah, I think I know the problem. It's because of the sftp:// url ;) [03:33] i've finally got this laptop stable after the random reboot o rama of precise and quantal [03:33] there is no way i'm upgrading again [03:33] heh [03:33] axw: ah ok. [03:37] thumper: given this issue that's just popped up, and me remember that Monday is a holiday, one of my tasks probably does want to go red [03:37] remembering* [03:40] davecheney: i forgot to add public bucket to ec2 before https://codereview.appspot.com/14021043 [03:46] 2.4 gig to download to upgrade [03:46] 18 packages are going to be removed. 232 new packages are going to be installed. 1943 packages are going to be upgraded. [03:46] * thumper goes to make a coffee [03:47] wallyworld: LGTM, trivial [03:47] fire at will [03:47] thanks :-) [03:47] o.O [03:47] 2.4 G to download ? [03:47] i don't think precise ocupied 2.4 Gb on disk [04:27] davecheney: raring [04:30] wow [04:30] DOWNLOAD ALL THE THINGS! [06:13] can anyone help me with this issue [06:13] p1/0:config-changed % config-get --format json global_maxconn [06:13] "" [06:13] ^ inside the unit the default value is not being reported [06:13] this value defaults to 4096 in the config.yaml [06:14] but is being returned as "" in the hook context [06:37] mornin' all [06:37] mornning rogpeppe [06:37] any idea on the questio above ? [06:37] it's blocking me atm [06:37] davecheney: just looking [06:38] i'm redoin the environment [06:38] davecheney: i wonder if this is related to the recent problems with boolean in a hook context [06:38] i REALLY hope this isn't a transient failure [06:39] rogpeppe: we copy the config settings into the state don't we ? [06:39] well, obviously [06:39] that is how juju get works [06:39] and get reports the value correctly, and understands its got a default, and is an int [06:40] davecheney: this stuff has changed not too long ago and i'm not familiar with the code any more [06:45] oh poo [06:45] change is bad [06:46] davecheney: how can i get sudo to use my normal environment vars? [06:47] davecheney: scratch that, found it [06:48] -E from mempory [06:48] probably not right [06:49] davecheney: yeah, but it doesn't preserve $PATH. crappy frickin' thing. a recipe for unexpected behaviour [06:52] davecheney: so is this *really* the hoop one needs to jump through to bootstrap the local provider? sudo -E sh -c "export PATH=$PATH; juju bootstrap -e local --debug [06:52] axw: ^ [06:52] rogpeppe: juju will already be in the path [06:52] unless you're a developer [06:52] which sucks for you [06:53] davecheney: true [06:53] ahh [06:53] now I understand your comment on unexpected behavior [06:53] and why none of you guys test the juju ppa's :) [06:53] davecheney: and non devs won't need GOROOT set either, i suppose [06:53] rogpeppe: ? [06:53] rogpeppe: sudo `which juju` ... [06:53] axw: you've been using the local provider quite a bit [06:53] rogpeppe: NOBODY needs $GOROOT set [06:54] but lets not get distracted [06:54] axw: that doesn't work for me [06:54] davecheney, https://bugs.launchpad.net/juju-core/+bug/1231457 [06:54] <_mup_> Bug #1231457: uniter API discards non-string config settings [06:54] davecheney: ahem [06:54] rogpeppe: what's the problem? [06:54] davecheney: yeah, it's just path [06:54] I mean, what's the outcome [06:54] axw: i bootstrapped it ok - i just found it... unintuitive [06:55] fwereade: shit, i'm using yesterdays rev [06:55] well, that is a day I won't get back [06:55] axw: if you've got a version of juju that isn't at /usr/bin/juju, it fails [06:56] rogpeppe: yeah, it's not great. IIANM, future versions of Ubuntu/lxc/upstart will allow us to have local without root [06:56] axw: that would definitely be an improvement! [06:56] davecheney, chit, bad luck:( [06:57] fwereade: checking now [06:57] thanks for the answer [06:57] i take it that the cli doesn't use the api in the same way the unit agent does [06:57] davecheney, the CLI still hardly uses the api [06:57] davecheney, but even when it does there will be approximately no overlap with the agent API [06:58] that would explain the disparity [06:59] davecheney, yeah -- and particularly in terms of config settings the results are very different *anyway* [07:00] davecheney, because ServiceGet or whatever it is returns a great ugly pile of human-targeted bs like "description" and whether or not a value is the default [07:01] fwereade: oh fuck, don't mention default [07:01] i'll cry [07:01] davecheney, when really it should be a charm url + basic config settings, and the client can dress that up however he wants [07:12] fwereade: do you have anything to say about https://codereview.appspot.com/13969043/ before i approve it? [07:22] rogpeppe, I still don't love it but I'm not going to fight it,because an extension *will* be useful if those files are floating around freely, and it'll be better to have all such files with a consistent convention [07:22] fwereade: well, i'm very happy to go with any alternative suggestion [07:26] rogpeppe, if you're just thinking of the actual choice of extension I'm fine with .jenv... can't see anything elseobviously using it [07:26] fwereade: yeah, that's what i was thinking of [07:26] fwereade: ok, cool [07:26] fwereade: you'd prefer no extension at all? [07:27] rogpeppe, only slightly, and I don't have strong arguments to justify it [07:28] rogpeppe, whereas your arguments is favour of an extension doseem pretty strong [07:28] fwereade: ok, cool [07:28] fwereade: i'll approve then [08:23] fairly simple review, anyone? https://codereview.appspot.com/14028043 [08:24] dimitern, fwereade, axw, TheMue, wallyworld: ^ [08:26] rogpeppe: looking [08:26] TheMue: thanks [08:34] g [08:35] axw: thanks for the review [08:35] nps [08:44] fwereade: i have to do a few things now, but can we talk about the release etc later [08:46] rogpeppe: reviewed [08:49] wallyworld, sure [08:49] wallyworld, pre-standup? [08:52] fwereade: maybe, depends when i'm back. i have a branch which removes all legacy tools. there are about 6 upgrader tests commented out because of a slight change in tools finding logic (it stops looking earlier than legacy) and the tools in private hide the ones in public [08:52] i could out that up for a pre-review if you want [08:52] put [08:52] i had to rename all our fake charms from "series" to "quantal" [08:52] since simplestreams does not allow fake series [08:52] and no test machine will be running quantal [08:53] so there's a lot of one line changes all over the place [08:54] fwereade: so i might propose it with the upgrader tests mentioned above commented out and come back later to talk about it [08:58] fwereade: https://codereview.appspot.com/14031043/ i'll come back later, i may miss the standup but do want to discuss the release [09:03] fwereade, rogpeppe, dimitern: one one too complex review, https://codereview.appspot.com/14030043/ [09:05] TheMue: why do we need to keep the old Status method around? [09:06] TheMue: can't you just return some more data from Status (like you added the data param to SetStatus) ? [09:06] rogpeppe: only want to do this in two CLs [09:07] rogpeppe: just keep this one small [09:07] TheMue, looking [09:08] TheMue: i don't think a few extra underscores add much to a review burden, but fair enough i guess [09:11] TheMue, reviewed [09:12] dimitern: thx, just seen [09:13] dimitern: maybe I should really change Status immediatelly ;) [09:27] if anyone has cycles, could I please get a review on this? https://codereview.appspot.com/14011046/ [09:27] null provider is broken without it. [09:27] fwereade: I've POCd some changes to wire up prechecker; let me know if it's too evil ;) https://codereview.appspot.com/14032043 [09:28] fwereade: only thing I'm not really sure about is the modification to cmd/jujud. it's a new dependency, but it seems like the setup of State does belong here [09:30] wallyworld: any particular reason you changed "series" to "quantal" in your MP? [09:36] axw, cheers, I'll take a look [09:44] axw: i think your changes there are a good reason for us to avoid optional environs interfaces. [09:44] axw: but it may well be too late for that. [10:24] mgz: ping [10:25] axw: i'm wondering about bootstrap storage [10:26] axw: would it be possible for a provider to know if it has already been bootstrapped and therefore whether it's appropriate to use bootstrap storage or not? [10:27] axw: then we could lose all externally visible BootstrapStorage stuff [10:34] rogpeppe: hey [10:34] mgz: could we have a chat about the addressing stuff [10:34] ? [10:35] yup, before or after standup? [10:35] mgz: how about now, same hangout? [10:35] I'm there [10:48] aand I'm back through no logical means that I can discern [11:02] rogpeppe: I couldn't think of a good way to do that before, but I'd certainly prefer that if possible [11:02] going out now - have a nice weekend [11:02] axw, have a good one as well [11:04] gaah [11:05] rogpeppe, dimitern, fwereade: to be sure please once again https://codereview.appspot.com/14030043/ [11:11] TheMue, reviewed [11:15] fwereade: thanks [11:47] rogpeppe, I'm trying to add a test to prove state.Open fails for unit agents [11:48] rogpeppe, the problem is it always fails, even when I call SetMongoPassword on the unit in the test [11:48] rogpeppe, mongo passwords are not hashed, right? Whatever I pass in SetMongoPassword should be it, right? [11:50] rogpeppe, I want to make the test fail reliably, so when I remove SetMongoPassword on the unit it will succeed [11:52] dimitern: how does it fail? [11:52] dimitern: perhaps you could paste the test code? [11:52] rogpeppe, I doesn't fail, that's the problem [11:52] rogpeppe, here it is http://paste.ubuntu.com/6162450/ [11:52] rogpeppe, and I have unit.SetMongoPassword(initialUnitPassword) in primeAgent [11:53] dimitern: you said "the problem is it always fails" ... then "it doesn't fail"... which one? [11:53] dimitern: so that test fails, because it succeeds in connecting to the state? [11:53] rogpeppe, the test case, as written should pass when I *didn't* set a mongo password [11:54] rogpeppe, but it should fail if I did, and it passes in both cases [11:54] it fails by not failing [11:54] not failing is passing [11:54] oh no, that's wrong [11:54] fail test is failing [11:54] c'mon :) [11:55] maybe I'm not explaining it well enough [11:55] if at first you succeed, try again until you fail [11:55] dimitern: so you're saying that that test successfully connects to the state, even though the mongo password isn't set (supposedly) ? [11:55] if only.. [11:55] dimitern: perhaps you should paste the output from the test too... [11:55] rogpeppe, no, I'm saying the test (as written) always passes (i.e. fails to connect to state), whether I set a mongo password or not in primeAgent [11:56] this is like watching Who's On First :) [11:56] something whizzes over roger's head [11:56] rogpeppe, does that make more sense? [11:56] dimitern: that would imply that a unit agent could never connect to the state... [11:57] rogpeppe, the agent is not even running yet [11:57] rogpeppe, that's step 2 [11:57] rogpeppe, at step 1 here, I want to make sure I'm setting a mongo password and with it I can connect to state as that unit [11:57] rogpeppe, and when I remove the password setting it'll fail to connect [11:58] rogpeppe, istm state connections are not based on tag+password, just a password? [11:58] dimitern: that does seem a bit odd [11:58] rogpeppe, how can we tell who's connecting? [11:58] dimitern: no, they're based on tag and password [11:59] rogpeppe, ah, so maybe I'm missing Tag = unit.Tag() in state.Info [11:59] dimitern: maybe so [11:59] dimitern: you'll be setting the admin password if tag is blank [11:59] rogpeppe, I see [12:00] rogpeppe, now it works! :) [12:00] great! [12:00] rogpeppe, what's an admin password and what it is used for? [12:01] dimitern: that's what the juju clients use [12:01] dimitern: it's guarded by admin-secret [12:01] dimitern: we should be able to delete that account entirely in time, probably [12:02] rogpeppe, I see, ok [12:02] rogpeppe, thanks [12:03] dimitern, rogpeppe: any chance we could drop the ""->"user-admin" business? it doesn't seem like a win to me [12:04] fwereade: i'd love to, but given that it's going away and has compatibility issues, i think it's probably not worth changing at this point. [12:04] s/has comp/changing it has comp/ [12:04] rogpeppe, how's it going away? that would entirely satisfy me :) [12:05] fwereade: because nobody needs to connect to mongo with the admin password, in the future [12:06] rogpeppe, oh! it's mongo-only is it? sorry, I see [12:06] fwereade: yea [12:06] h [12:26] rogpeppe, heh, we *really* want UUID in environ, don't we [12:26] fwereade: context? [12:27] rogpeppe, advising people not to use environment name as a unique identifier when writing environs [12:27] rogpeppe, I think they might reasonably ask wtf else they're expected to use [12:27] fwereade: ha, yes [12:27] but the name is so handy... [12:28] mgz, for users, yes [12:28] (the main advantage being anywhere the value resurfaces, it's easy to see at a glance what juju thing it's related to) [12:28] mgz, internally, for distinguishing between environments? it's hopeless [12:29] mgz, using it as such leads to great pain when developers use shared provider credentials [12:29] right, I do have sympath with that [12:29] +y [12:30] I just suffer from dealing with uuids in nova daily [12:30] :) [12:30] is this one a volume or a machine... did I repaste the right thing to the commandline [12:30] ha, yeah [12:50] fwereade: https://codereview.appspot.com/14030043/ looks better now [13:21] rogpeppe, fwereade: so right now, because of the firewaller we need to enable state workers for JobManageEnviron as well [13:22] dimitern, yep, sounds right [13:22] fwereade, ok [13:26] hey, cool, that *was* the red arrows who flew across the bay with fancy coloured smoke, must be practising for the airshow tomorrow [13:27] fwereade: send pics *lol* [13:44] are compilation complaints in trunk goyaml and juju-core expected, or is this indicative of something wrong on my local saucy system? http://pastebin.ubuntu.com/6162806/ [13:52] gary_poster, that looks more like goamz to me, but I wasn't aware of changes there [13:53] gary_poster, and also goose [13:54] fwereade, gary_poster: i had the goose one too. clearing all 3rd party pkg and src and re-get them helped [13:55] fwereade, TheMue thanks, was just thinking of doing the same thing [13:56] fwereade: btw, could you find a moment to take a look into the reproposed CL [13:56] ? [13:56] TheMue, I'm churning through a big one from ian before he goes to sleep [13:56] fwereade: ok, makes sense [13:59] rogpeppe, fwereade: why do we have only PasswordHash() method on agent.Config, but not the actual password? [14:00] dimitern: because we never store the actual plaintext [14:00] dimitern: um [14:00] rogpeppe, well I need it for testing now :) [14:00] dimitern: hmm, actually i'm talking rubbish [14:00] rogpeppe, how can I access the password that gets changed after opening state the first time? [14:00] dimitern: one mo, i'll have a look [14:01] dimitern: it seems to me that there's no reason [14:01] dimitern: it could just return the password [14:02] rogpeppe, you mean rename agent.Config.PasswordHash() to Password() and return the plain test? [14:02] text [14:02] dimitern: i think so, yeah [14:03] TheMue, fwereade killing all third party and rebuilding did the trick, thanks [14:03] rogpeppe, it's actually used only in 2 places in tests only [14:03] fwereade: i just noticed that the hp cloud image metadata on cloud-images has a trailing / for the auth url whereas our boilerplate does not. so bootstrapping hp cloud fails. i could add that change into the tools branch, or else a new branch will be required [14:05] wallyworld, well I'd really prefer a separate branch [14:05] ok [14:05] gary_poster: yw [14:12] rogpeppe, ISTM that the only things that use Instances are the firewaller and the ssh command (which should be getting machine addresses) [14:12] mgz, speaking of which, is there anything I can do to support you in the addresses stuff? [14:13] fwereade: rog and I are making pretty good progress here I think [14:13] fwereade: me and mgz are rattling it off as we speak... [14:13] mgz, I am primarily talking emotional support to be fair, but let me know if there's something practical I can do [14:13] mgz, rogpeppe: <3 [14:13] fwereade: the forthcoming worker/addresspublisher will use Instances [14:13] rogpeppe, can you take a look at this please http://paste.ubuntu.com/6162906/, more specifically line 9 [14:13] rogpeppe, how will this ever work? [14:14] rogpeppe, is Instances likely to move onto InstanceBroker then? [14:14] rogpeppe, that would be convenient :) [14:14] fwereade: not as far as i was aware [14:15] rogpeppe, and what's the original intent? [14:15] fwereade: at the moment we're thinking of the address publisher as something that runs independently of the provisioner. [14:15] rogpeppe, ah ok, I guess it's not part of the provisioner [14:15] rogpeppe, jinx :) [14:15] fwereade: hold on, i'll give you a sneak preview if you like [14:15] fwereade: then we can get some feedback [14:16] rogpeppe, cool [14:16] fwereade: [14:16] rogpeppe, I should do TheMue's first though [14:16] package addresspublisher [14:16] import ( [14:16] "fmt" [14:16] stdtesting "testing" [14:16] "time" [14:16] gc "launchpad.net/gocheck" [14:16] "launchpad.net/juju-core/instance" [14:16] "launchpad.net/juju-core/state" [14:16] coretesting "launchpad.net/juju-core/testing" [14:16] jc "launchpad.net/juju-core/testing/checkers" [14:16] "launchpad.net/juju-core/testing/testbase" [14:16] ) [14:16] func TestPackage(t *stdtesting.T) { [14:16] gc.TestingT(t) [14:16] } [14:16] var _ = gc.Suite(&publisherSuite{}) [14:16] type publisherSuite struct { [14:16] oh nooo :) [14:16] testbase.LoggingSuite [14:16] } [14:16] * fwereade goes off for a 5 minute break? ;p [14:16] var testAddrs = []instance.Address{instance.NewAddress("127.0.0.1")} [14:16] func (*publisherSuite) TestSetsAddressInitially(c *gc.C) { [14:16] ctxt := &testMachineContext{ [14:16] getAddresses: addressesGetter(c, "i1234", testAddrs, nil), [14:16] dyingc: make(chan struct{}), [14:16] lol [14:16] } [14:16] m := &testMachine{ [14:16] instanceId: "i1234", [14:16] refresh: func() error { return nil }, [14:17] life: state.Alive, [14:17] } [14:17] died := make(chan machine) [14:17] publisherc := make(chan machineAddress, 100) [14:17] paste error :) [14:17] // Change the poll intervals to be short, so that we know [14:17] * dimitern ducks for cover [14:17] // that we've polled (probably) at least a few times while we're [14:17] // waiting for anything to be sent on publisherc. [14:17] abort abort! [14:17] defer testbase.PatchValue(&shortPoll, coretesting.ShortWait/10).Restore() [14:17] defer testbase.PatchValue(&longPoll, coretesting.ShortWait/10).Restore() [14:17] who can kick? :) [14:17] go runMachine(ctxt, m, nil, died, publisherc) [14:17] select { [14:17] case addr := <-publisherc: [14:17] c.Assert(addr.addresses, gc.DeepEquals, testAddrs) [14:17] c.Assert(addr.machine, gc.Equals, m) [14:17] case <-time.After(coretesting.LongWait): [14:17] c.Fatalf("publisher never published the expected address") [14:17] } [14:17] select { [14:17] ah, sorry about that [14:17] haha [14:17] fwereade: http://paste.ubuntu.com/6162921/ [14:18] at least i know now that reconnecting aborts [14:18] wow, weird, goyaml recovers from panics inside a type's GetYAML and SetYAML and returns them as errors... didn't see that one coming. [14:19] I guess they don't return an error value, so that's sort of all you can do. [14:19] natefinch: they should definitely return errors. it's a design flaw that they don't [14:20] dimitern: what's the issue you have with that paste? [14:20] dimitern: it looks roughly plausible to me [14:20] dimitern: we connect using the password, falling back to the old password if the password isn't set or we can't connect with it [14:21] mgz: i'm gonna grab some lunch too [14:25] rogpeppe: I'm like || this close to rewriting the whole stupid thing. I already looked around for alternatives and didn't see anything. The other yaml parsers just build up a document tree in memory... which is like, sorta useless by itself [14:28] rogpeppe, we construct and empty state.Info, filling in some stuff, not setting the password, and the immediately after that we check if the password is set [14:29] sinzui: I've filed RT #1224057 asking for a VPN. [14:29]