[00:08] wallyworld, thumper: sorted out the test failure. it was due to a change in how mgo does retries. I copied the change from master and it now passes. I've tried running big chunks of our state tests and all is looking good. will propose shortly. [00:08] menn0: that's awesome, ty [00:09] restore from ./juju-backup-20140728-1542.tgz completed <-- with failures but :D [00:18] wallyworld: https://github.com/juju/juju/pull/434 for backported mongo change [00:19] \o/ looking [00:23] thumper, wallyworld: this one needs a look before I can merge the big juju 1.20 change: https://github.com/juju/charm/pull/30 [00:23] sure [00:23] menn0: tests pass? [00:23] thumper, wallyworld, sinzui: also, the bot doesn't seem to be noticing this: https://github.com/juju/juju/pull/433 [00:23] what am I doing wrong? [00:23] wallyworld: yep, tests pass [00:24] menn0: did you still say merge? [00:24] menn0, __fixes-1318366__ [00:24] menn0: awesome, juju/charm change merged [00:24] thumper: no. so you need both. [00:24] menn0: AFAIK [00:24] in the same comment or it doesn't matteR? [00:25] thumper: your pr says it has conflicts [00:25] menn0: use figlet(1) [00:25] menn0, oh, but that isn't a merge command $$merge$$ to request merge testing. __fixes-1318366__ is needed to convince CI to test it for the merge bot [00:25] wallyworld: really? [00:25] wtf? [00:25] sinzui: I thought you said 1.20 landings were ungated? [00:25] thumper: this is for master [00:26] menn0, damn, I think github is swallowing __ [00:26] menn0: mine isn [00:26] isnt' [00:26] thumper, they are ungated, or were when I last looked [00:26] sinzui: https://github.com/juju/juju/pull/434#issuecomment-50698948 [00:27] ah, 1350493 is a regressions [00:27] looks like bug 1318366 isn't on the list of allowed fixes for master: "Does not match ['fixes-1350493', 'fixes-1347715', 'fixes-1342725']" [00:27] <_mup_> Bug #1318366: jujud on state server panic misses transaction in queue

[00:27] should I wait? [00:27] or __JFDI__ [00:27] jfdi [00:27] menn0, Yes please, your work predates the discovered of the regressions [00:27] sinzui: ok thanks [00:27] ah fark [00:28] I will remove find something other than __ to avoid the markdown conflict [00:32] CI will accept underscores and non- underscores for fixes-nnn to avoid the issue. [00:32] I will remove the regression tag on 1.20 since there is no test showing the regression at this time [00:33] I mean I removed "ci" from the bug 1350493 since ci didn't see the regressions [00:33] <_mup_> Bug #1350493: 1.20.x local provider not running apt-get update

[00:34] * thumper rolls all the instructions into one [00:35] wallyworld: for some reason my 1.20 base branch didn't have upstream 1.20 [00:36] merge accepted [00:39] thumper, wallyworld: here's the big 1.20 mgo change: https://github.com/juju/juju/pull/435 [00:39] looking [00:39] menn0: oh, only 59 files? [00:40] thumper: yeah sorry. I tried to make it more but couldn't find anything else to search and replace. [00:40] heh [00:40] I say if all the tests pass, shipit [00:42] menn0: ty, +1 [00:43] thumper: I haven't run all the tests but most of what's in the state and worker directories [00:44] merging now [00:48] thumper: good news [00:48] almost all the uses of authorizer.GetAuthEntity [00:49] are in type switches [00:49] so they are the same as a type switch on authorozer.GetAuthTag() [00:50] thumper: and the ones that aren't when they get the entity, they just take it's tag anyway [00:52] spot the bug [00:52] loggedInUser := api.getLoggedInUserTag() [00:52] if loggedInUser != argUser.Tag() { [00:52] result.Results[i].Error = common.ServerError(fmt.Errorf("Can only change the password of the current user (%s)", loggedInUser.Id())) [00:52] continue [00:52] } [00:52] ^ getLoggedInUser returns nil if the logged in user isn't a user [00:52] say a machine or somethig [00:52] so returning the error would panic the server [00:53] heh [00:54] so i guess there isn't any coverage of this path ... [00:55] I guess not [00:55] thumper: you're a winner [01:10] thumper: your change merged \0/, now just waiting for menn0 's [01:10] phew [01:10] hurry up bot [01:18] is there any way to make the test runner print the name of test before it starts running it? [01:18] only with -gocheck.vv I think [01:18] which gives you all the logs too [01:18] i'm using that, sadly that's not doing the trick working w/ mgo though [01:19] menn0: FWIW I got this failure on master: FAIL: upgrade_test.go:274: UpgradeSuite.TestLoginsDuringUpgrade [01:19] A recent commit to master broke deploy in 4 clouds. Something about /var/lib/juju/db not existing [01:19] * sinzui gathers logs [01:19] axw, maybe it is.. its just lost in the undifferentiated output [01:20] thumper: https://github.com/juju/juju/pull/436 [01:20] ^ with fix and test [01:20] * thumper wonders if he broke trunk again [01:22] thumper: can I land that fix ? [01:23] menn0: you branch fialed to land [01:23] menn0: good news for you - gomaasapi needs updating too! [01:23] yeah [01:23] davecheney: not yet I don't think [01:23] just about to say that [01:23] thumper: ok [01:23] thumper: it wasn't really a question, more 'can I use the magic string to land it' [01:23] davecheney: I'd rather you didn't just now [01:23] ok [01:26] thumper: https://github.com/juju/juju/pull/437 [01:34] wallyworld: https://github.com/juju/juju/pull/431 [01:34] fixes the manual provider regression [01:34] looking [01:34] for real this time [01:34] verified on the CI machine [01:34] thumper, https://bugs.launchpad.net/juju-core/+bug/1350633 [01:34] wallyworld: also https://github.com/juju/juju/pull/430 unbreaks something I changed in the previous change [01:48] wallyworld: thanks [01:48] np, thanks for fixing [01:48] just thinking, maybe it'd be better to return an error if the output is something else [01:49] i.e. not "no-agent-dir" or whatever [01:49] i wondered that [01:49] then we'd see it in the output [01:49] yep [01:49] so long as it works in all circumstances [01:50] it's only going to consider stdout still [01:50] so it should be fine [01:52] sinzui: ok, on it [01:54] axw: tx for the review on restore [01:54] perrito666: nps [01:54] I was really needing a second pair of eyes on that thing [01:56] sinzui: fuck fuck fuckity fuck, found it (I think) [01:56] yep [01:56] * thumper gets a patch ready [01:56] * thumper thinks of how to test it [01:57] * thumper wonders why the existing test didn't fail [01:58] davecheney, wallyworld: back from lunch. I'll deal with the failed merge after the team meeting. [01:59] np [01:59] thumper: do you have more detail on the UpgradeSuite.TestLoginsDuringUpgrade failure you saw? [01:59] menn0: only the scrollback [01:59] I'll look in a miute [02:02] thumper, great the --debug I ran didn't give me any more information than the console output. I did find another test instance that has be deleting for more than a day in HP. good I get to write a support ticket [02:02] sinzui: I know what the problem is, just not sure why the tests didn't catch it [02:17] natefinch: you are not around for the meeting right? [02:27] davecheney: its odd, you are the one I hear the best next to jeese [02:28] and I think we are as far apart (network wise) as topologically possible [02:28] perrito666: hangouts appear to be point to point based, so some sound ok, others are complete disasters [02:29] well, mystery :) Ill try to have more meetings with you :p I hear you better than almost everyone else [02:29] ok ppl almost midnight here, see you all tomorrow, cheers [02:30] * perrito666 turns into a pumpkin [02:32] perrito666: it's because I have a loud mouth [02:33] davecheney, wallyworld: I don't understand why this happened: https://github.com/juju/juju/pull/435#issuecomment-50702449 [02:33] menn0: toptip: don't trust godeps [02:33] i think it's fucked [02:33] it tells you it's updated to the correct rev [02:33] but it lies [02:33] huh? [02:33] gomaas was never updated [02:33] hmm ok [02:33] was it? [02:34] gomaas also needs the import path changes [02:34] afaik, godeps works ok now that it has the -f option [02:35] wallyworld: your experience differs from mine [02:35] :- [02:35] ( [02:35] i have found it to be unreliable since growing the ability to update if a rev is missing [02:35] ok, shame [02:35] i wish Go as a platform supported something better [02:36] not like this problem is new or anything [02:36] davecheney, wallyworld: well I look at my current gomaasapi copy and it indeed has references to labix.org/v2/mgo/ [02:36] so I have no idea how builds and tests are working [02:36] menn0: then the hash your using isn't the same one as dependencies.tsv [02:37] menn0: i think you still have labix source locally [02:37] but the bot doesn't [02:37] aarg!!! [02:37] found out why the tests passed [02:37] and...... [02:37] wallyworld: menn0 yeah, that could be it [02:37] drumroll [02:37] yep, that's exactly it [02:37] * davecheney holds breath [02:38] * menn0 deletes the old repo [02:38] * davecheney is breathless with ................ anticipation [02:38] * wallyworld taps fingers [02:38] * wallyworld holds breath [02:38] OUT WITH IT MAN! [02:39] wallyworld: mocking [02:39] ah [02:39] ffs [02:39] \o/ [02:39] s.PatchValue(mongo.AvailSpace, <-- said a constant [02:39] new function makes sure the directory exists [02:39] wallyworld: let me tell you about apiserver/testing/fakeauthorizer some day [02:40] davecheney: noooooooo! [02:40] wallyworld: thumper will tell you all about it when he tucks you in next week [02:40] ooooh [02:40] can't wait [02:41] * wallyworld likes thumper's good night kisses [02:41] menn0: did you update the gomaasapi rev in dependencies.tsv in 1.20? [02:41] I bumped it in master only [02:42] axw: that'll be it [02:44] axw: yep I'm just pushing that now [02:45] wallyworld: https://github.com/juju/juju/pull/438 [02:45] axw: it took me a while to figure out why everything worked on my machine without that change. turns out I still had the old labix mgo hanging around [02:45] looking [02:46] * thumper fixes 1.20 [02:46] thumper: goes to show one has to be careful when writing mocks - just the correct bits need to be mocked out and nothing else [02:47] * thumper nods [02:48] damn git and it's auto commit shit [02:48] "damn git" without the qualifier is what you meant i think [02:51] wallyworld: and the backport https://github.com/juju/juju/pull/439 [02:52] hmm, where has i seen that diff before? [02:53] sinzui: I do like the __JFDI__ flag, and the bold actually fits [02:54] * thumper has quick dogwalk before school ends [03:09] review for https://github.com/juju/juju/pull/440 please? should fix the windows bootstrap regression [03:09] wallyworld: I can see that the test run for the 1.20 changes already has at least one failure [03:10] :( [03:10] looking in to it now [03:10] ok [03:17] wallyworld: I don't follow your comment about precise [03:17] I'm just using "trusty" because the series is used to infer the OS [03:17] it could be precise, quantal, trusty, ... [03:18] axw: i thought that people running precise would get confused if they saw a /var/lib/juju/data/trusty dir [03:18] "wtf, i'm on precise, why is juju saying trusty" [03:18] axw: is there no other reasonable series we can use at that callsite? [03:18] it won't be in there [03:18] not one that is passed in? [03:18] wallyworld: it'll just be /var/lib/juju [03:19] wallyworld, axw: does this ring any bells? http://paste.ubuntu.com/7910716/ [03:19] axw: oh, ignore me [03:19] it's failing on 1.20 for me with the newer mgo [03:19] thumper: I could just change it to "/var/lib/juju" and do away with the call altogether [03:19] axw: i misread it, sorry [03:19] in CI too [03:19] menn0: looking [03:19] axw: that helper function is what i was referring to [03:19] doh [03:19] axw: the test itself is the same in master [03:20] menn0: hmm no sorry [03:20] axw: thanks. I'll keep digging. [03:21] menn0: I think I have seen that intermittently once or twice [03:21] thumper: it could be that the timeout is a little tight [03:22] and perhaps the fix to this panic changes mgo's timing. the test is about watchers. [03:23] menn0: that error doesn't ring a bell. the only wather timing issues i know of are in state/watcher [03:24] axw: my vote is just to use /var/lib/juju with a TODO [03:25] righto, will change [03:25] now I can't get that test to fail on my machine. I'm going to increase the timeout a bit. [03:29] menn0: don't you just love tests with timeouts? sigh [03:30] wallyworld: well the timeout is only in case the code under test doesn't work ... but unfortunately the code needs most of that timeout to do it's thing [03:30] menn0: "we" should be using channels and signals for tests that need coordination, not timeouts [03:31] but most of our tests are bad in that respect [03:32] hmm.... MachineSuite.TearDownTest... Panic: not authorized for query on presence.presence.beings (PC=0x414686) [03:32] wallyworld: it IS using channels. the timeout is used for the maximum time to wait for something to come back from the channel. the code is dependent on the watcher frequency which is why it has to wait. [03:32] so the test isn't THAT bad [03:32] menn0: ok, sorry :-) i spoke without looking at the code [03:32] it should probably patch the watcher frequency so it doesn't have to wait so long [03:32] * wallyworld is scared by bad tests [03:32] but I don't know if that would have other bad effects [03:33] also... Panic: local error: bad record MAC (PC=0x414676) [03:33] menn0: there's a way to manually trigger the watcher [03:33] thumper: that's a mongo 2.4 bug [03:33] wallyworld: ok [03:33] hopefully fixed in 2.6 [03:33] hence the drive to port juju to mongo 2.6 [03:34] wallyworld: this test appears to be fairly high level so I don't know if that applies [03:34] I'm sure there's a way to make it faster [03:34] but I'm not going to try and solve that now [03:34] there's a TODO already [03:34] menn0: yep. for now i'd to the minimum [03:35] actually I think I see the problem [03:35] there could be a race [03:35] thumper: that not authorised panic is a worry - it can happen with the new session copy stuff if there's a race [03:35] since a wrong session without credentials is used to connect [03:35] does it only happen every so often? [03:35] wallyworld: cmd/jujud tests [03:36] * thumper is running the tests again [03:36] from what i've seen, given al the improvements to tests over the past several weeks, the cmd/jujud tests are the main ones still failing [03:42] wallyworld: succeeded that time [03:42] ah, intermittent tests [03:44] anyone know, in the actions spec [03:44] is it action.yaml or actions.yaml? [03:44] yes [03:44] :-P [03:44] * marcoceppi waddles away [03:44] sorry :-) [03:44] i don't know which one [03:44] haha, it's fine [03:45] I'm trying to dig up the spec [03:45] marcoceppi: in the code, it seems to be actions.yaml [03:45] in juju-core code that is [03:45] wallyworld: interesting that it's pluarl but metadata and config aren't [03:45] yeah [03:46] i have had no involvement so can't comment [03:46] * marcoceppi wonders if he should call it benchmark or benchmarks.yaml [03:46] marcoceppi: well, it is environments.yaml for example [03:46] wallyworld: ah, true, good point [03:46] I guess pluarl is depending on how well the word sounds :) [03:47] i'd go with plural [03:47] I'll go with plural form [03:47] yeah [04:11] thumper: sorry, I thought you were working on the master branch - I shouldn't have changed the status of that bug :) === menn0_ is now known as menn0 [04:24] uh,oh [04:24] lucky(~/src/github.com/juju/juju) % git pull upstream master [04:24] fatal: unable to access 'https://github.com/juju/juju.git/': Could not resolve host: github.com [04:49] wallyworld, sinzui: the mgo fixes for 1.20 are in! *\O/* [04:50] davecheney: I've had that a bit today as well (but my internet link has been a bit flaky too) [04:50] menn0: whoo hoo [04:50] ty [04:51] * menn0 updates the LP bug [04:55] thumper, wallyworld I would like permission to land this, https://github.com/juju/juju/pull/442 [04:55] I'm tired to skipping over this test failure [04:56] davecheney: go for it [04:57] wallyworld: thanks, i helps as I'm working in that area daily [04:58] no problem at all, nice change [05:05] Oh, this change was actuall Go 1.4 [05:05] nm [05:05] what is important is there are some improvements in the race detector coming soon to 1.4 dev [05:05] and I want to be able to run juju under this improved race detector [05:05] given the plethora of problems we have at ht emoment [05:10] ericsnow: ping? [05:11] * ericsnow hides [05:11] menn0: hi :) [05:12] ericsnow: do you want to do a quick hangout re the backup work? it might be more efficient. [05:13] menn0: sure, I really should have gone to bed hours ago, but at this point I'm not going to sweat it :) [05:13] ericsnow: I don't have much time before I need to go so I promise it'll be quick :) [05:14] ericsnow: https://plus.google.com/hangouts/_/gthfw6mxzrcsonn5bzqoqdb5iya?authuser=1&hl=en-GB [05:14] ericsnow: https://plus.google.com/hangouts/_/gthfw6mxzrcsonn5bzqoqdb5iya?authuser=1&hl=en-GB [05:31] thumper, wallyworld large email about FakeAuthorizer sent to the list, enjoy! [05:35] jcw4|away: this test has been broken for weeks [05:35] http://paste.ubuntu.com/7911495/ [05:35] is it on your schedule to fix ? [05:38] jcw4|away: the test failure stems from the caller expecting actions to be returned in a set order === Guest46318 is now known as wallyworld [05:42] *END* *OF* *DAY* [06:18] davecheney: jeez, it's not good when a mock breaks the contract implemented by the thing it is mocking [06:19] wallyworld: indeed [06:19] wallyworld: ideally, if I do this properly, we won't need a mock at all [06:20] \o/ [06:20] there will just be apiserver.TagAuthenticator [06:32] wallyworld: https://github.com/juju/juju/pull/443 [06:32] looking [06:34] wallyworld: not ugrent [06:34] urgent [06:35] davecheney: already +1 :-) [06:35] this is part of unfucking FakeAuthoriser [06:35] wallyworld: it'll wait til the trunk is unblocked [06:35] ok [06:35] i'm doing a poor man's git pipeline [06:35] i miss lp for that [06:57] Why can't I use juju run against a subordinate? [07:07] is there an easy way to tell the current blocked status of commits to juju-core? i *think* that it's currently blocked by #1345638 (only) but it would be nice to have a definitive way to tell other than trying a $$merge$$ [07:07] <_mup_> Bug #1345638: ec2 changes? rising failure rate in ec2 health checks

[07:07] marcoceppi: i don't see why you shouldn't be able to [07:08] marcoceppi: what error do you see? [07:08] rogpeppe: currently getting this [07:08] juju run --unit subordinate/0 "cabs status" [07:08] ERROR subodinate/0 is not a principal unit [07:08] rogpeppe: should I file a bug? [07:10] marcoceppi: yes - it does seem to be a deliberate decision, but i can't immediately think of a reason why you shouldn't be able to do it. [07:10] rogpeppe: cool, this will really help unblock me with some stuff around benchmarking [07:11] marcoceppi: note: i wasn't involved in doing juju-run at all - there may well be a good reason why it doesn't allow it [07:11] rogpeppe: anyone you know off the top of your head i could ping for more clarification? [07:12] marcoceppi: fwereade might know. or axw. otherwise i think thumper is the man. [07:13] hmm, I don't know why [07:13] marcoceppi, rogpeppe: can't think why offhand. are there maybe issues around ssh to subordinate units that juju run falls foul of? [07:14] fwereade: i don't see why it should [07:14] fwereade: I can juju ssh to subordinate/0 without issue [07:14] * marcoceppi is scouring the mailing list atm [07:14] marcoceppi, it's evidently not that then :) [07:15] I'll open a bug and see if thumper or axw chime in [07:15] marcoceppi: how about proposing a trivial fix for the issue (just remove the unit.IsPrincipal check) and try to get a review off thumper? [07:15] marcoceppi: or that [07:16] rogpeppe: ah, was going to check next how hard it woudl be to patch [07:16] if it's just a line or two I'll do that [07:16] marcoceppi: it's just one if condition [07:16] rogpeppe: cool, I'll go that route then [07:16] marcoceppi: github.com/juju/juju/state/apiserver/client/run.go:68 [07:16] and duke it out in a pull request [07:16] rogpeppe: ta! [07:17] marcoceppi, yeah, should be a trivial fix -- IIRC AssignedMachineId does work for subordinates, so just drop the IsPrincipal() check and see what happens [07:18] marcoceppi, get thumper's opinion on the PR, though, he might remember some important reason I've forgotten [07:18] * fwereade observes that rogpeppe said all that already [07:18] * fwereade slinks off [07:18] fwereade: ack, I'll take a stab at it later tonight and just work around it for now. I'd want to compile/test as well to make sure it addresses my issue and actually works [07:19] :-) [07:19] thanks for the help fwereade and rogpeppe o/ === psivaa is now known as psivaa-brb [08:03] can someone review please: https://github.com/juju/juju/pull/445 <- fixes a critical bootstrap bug on master, 1.20 will come shortly [08:04] axw: *click* [08:07] axw: change looks fine to me (I'll let TheMue do formal review) [08:08] thanks [08:08] axw: the PR says ""With recent changes, we were leaving the installation of mongodb until too late" [08:08] axw: *which* changes made this difference? [08:09] voidspace: thumper fixed a bug where we were reinstalling/restarting the upstart service unnecessarily [08:09] but left the installation bit where it was [08:09] axw: +1 [08:09] ok, cool [08:09] TheMue: thanks [08:09] just wanted to understand the need [08:10] axw: and btw thx for the initial comment describing WHY and not WHAT has changed. [08:10] axw: just reviewd another PR this morning where the motivation for the change was missing [08:11] no worries (though evidently could've been better). I like to know why things are done too ;) [08:12] axw: yeah, helps when doing the review [08:12] * TheMue has to admit that he has to improve himself here [08:13] heh, all round I think :-) [08:16] btw, this morning I’ve seen a pattern to avoid a large number of arguments. here the args are in a struct and this is passed as variadic [08:16] inside the function it is checked if 0 or 1 of these structs is passed [08:16] and then the values are checked or a default is taken [08:17] somehow this looks a bit strange [08:17] what I like more is a pattern by Rob which can be found here http://commandcenter.blogspot.nl/2014/01/self-referential-functions-and-design.html [08:18] it’s a bit more code, but when used it reads really good [08:18] fwiw ;) [08:19] that would be quite a lot of code, seems like it'd be overkill except in public APIs [08:19] it is neat [08:20] can't say I've seen any optional/variadic struct signatures - that sounds a bit unusual for the codebase [08:20] axw: I found it in state, one already exists and one has copied this in the PR I’ve reviewd [08:21] axw: one moment, will look for the file === psivaa-brb is now known as psivaa [08:23] axw: oh, just have seen that it’s in a testing package, testing/factory/factory.go, line 88 [08:24] axw: so for testing code it may be ok [08:25] I see [08:26] I'd be tempted to make a second function, MakeUserDefaults [08:26] which calls the other with default values [08:30] Is there someone I need to ping or a review queue for pull requests agianst core? [08:30] or you have to pass a struct but it can be empty (which leads to defaults) [08:32] marcoceppi: the OCR (see https://docs.google.com/a/canonical.com/spreadsheets/d/1iQLLOWrjzxddm5VhYWYi0-2k3xI6wTMlpkvnVNJCYGY/edit#gid=0) can be pinged [08:32]