[00:08] thumper: how long do the cmd/juju tests take for you? on my average laptop, they take about 160s. JujuConnSuite test set up is about 200ms for each [00:08] * thumper runs [00:08] ah... [00:08] my state is somewhat broken right now [00:08] the bot seems way slower [00:09] i'm pretty sure we're using tmpfs [00:09] * thumper scrolls way up [00:09] ~170s [00:09] so that's well within the timeout limit [00:10] we are on fairly high speced machines though [00:10] so yes, the tests are bad, but they should run [00:10] wallyworld_: on my laptop it was 400s [00:10] wow [00:10] I have ssd [00:10] are yoyu using tmpfs? [00:10] yes [00:10] core i5 thinkpad x220 [00:10] there is a 3-4 second delay between each test run [00:10] i'm assuming that is setup/teardown [00:10] i don't see that [00:10] i see 200ms [00:10] don't look at the times [00:11] go test -gocheck.v [00:11] i did [00:11] well, your machine is the odd man out [00:11] it takes 400s on my machine [00:11] and 570+ in CI [00:11] thumper gets the same times as me [00:11] i only have 4 cores [00:11] just like CI [00:12] i have an 8 core i7 [00:13] shouldn't be that much faster [00:13] yes, that is what I said [00:14] i'm not sure CI should be blocked on this - it's not a regression [00:14] i agree [00:14] i'll update the bug [00:14] thanks [00:16] maybe in the interim we can run CI on a larger instance type [00:32] maybe [00:32] the largest that ec2 offer is an 8 core [00:32] and in my testing [00:32] that is still slower than my 2 year old core i5 [00:32] building gccgo as a test [01:17] davecheney: thumper: changing the instance type to "c3.2xlarge" (8 core, 16GB) seems to have helped the cmd/juju tests to pass. More data points needed, but -p 2 was faster than full parallisation for this run. there was a spurious apiserver/client failure which caused -p 2 to run. interestingly, the -p 2 run was faster for many tests this time around [01:18] maybe we try running with -p 2 to start with for a bit [01:18] thumper: are you around? [01:18] yeah... [01:18] thumper: don't sound so happy [01:18] :p [01:18] I'm writing docs/bootstrapping.txt [01:19] thumper: my condoleances [01:19] after having to work it out for the umteenth time... [01:19] I thought I'd write it down [01:19] thumper: how savvy are you on menn0's "upgrade mode" ? [01:19] ish [01:19] whazzup? [01:20] thumper: I am trying to replicate the idea for restore and I would like some overview but apparently I missed menn0 this week [01:21] I worked out a decent part of it [02:04] ok guys braing shutting down, see you all or some of you tomorrow [02:09] cya perrito666 [02:25] I'm getting a "state changing too quickly; try again soon" error [02:26] o_O [02:26] how soon? should I have a cup of tea? [02:27] I'm looping over a collection of state users and adding each one as an environ user via a new AddEnvironmentUser func - maybe I need to add them all as one transaction? [02:28] thumper: ^? [02:42] waigani_: it is lying to you [02:42] that is another error that returns when the assertion fails [02:42] it tries a few times [02:42] and then assumes that the assertion is due to someone else [02:42] in dev, it is most likely you [02:53] waigani_: did that help? [02:53] * thumper goes to turn the coffee machine on [02:53] thumper: sorry I was afk [02:57] wallyworld__: I've got to fix up my bootstrap tools branch, it's incompatible with your changes to use /var/lib/juju in the containers [02:57] gah, not sure how I'm going to fix this... [02:58] axw: that's ok, i mounted that directory to avoid the need for the container to call out to http (if i am thinking of the right thing) [02:58] calling out to http on the host [02:58] wallyworld__: yes that is the one. in my branch, file:// is treated specially. it means read the file contents locally and then add to the cloud-init script [02:59] i did that change because people were seeing errors [02:59] this branch i think obsoletes the need for that [02:59] since the tools are copied into the container via ssh init [02:59] wallyworld__: only for bootstrap ,not for containers... [03:00] the containers will get from state server [03:00] we can handle retries at that point [03:01] it was only a short term quick fix [03:01] wallyworld__: different branch. I guess I can revert it temporarily... I'm checking if we can do better tho. we may be able to load the tools into the userdata for the local provider [03:02] could do. i just wanted a quick way to avoid the observed source of the errors [03:02] i knew it would be throwaway [03:44] wallyworld__: can you PTAL: https://github.com/juju/juju/pull/600/commits [03:44] sure [03:44] from the 5th commit on [03:45] wallyworld__: it looks like lxc doesn't have the same limit on userdata size, so if we need to we have the option of serialising the tools in there for all lxc containers [03:46] good to know [03:46] i think that would be useful [03:46] avoid networking calls back to the state server to get the tools [03:47] axw: supportedArchitecturesCount is just for testing? [03:48] ah nevermind [03:48] i thought it was in production code [03:53] * thumper needs to work out how to poke inside of mongo [03:54] also noticed some wonderful weirdness in our code [03:54] the password hash of the admin-secret is used as the actual password for the admin user to mongo (which is "machince-0" btw) [03:55] * thumper goes to make that coffee [03:56] bugger, timer on the machine would have turned it off again by now [04:12] ugh... [04:12] spent the day teasing apart the layers of juju to work out where to put my change [04:13] still not entirely sure... but closer [04:15] where does juju log its debugging output when attempting to bootstrap into openstack? [04:15] i ran with --debug but its just sitting at apt-get update http://paste.ubuntu.com/8146668/ [04:20] grr... [04:20] stokachu: not sure sorry [04:21] no worries [04:21] * thumper can't push because master is dirty [04:21] ericsnow: you need to fix your pre-push hooks [04:36] thumper: hang on [04:36] martin fixed that, twice [04:37] how did a change get past the bot [04:37] wallyworld__: chat? [04:37] davecheney: no idea [04:37] thumper: hiya, ok [04:37] onyx standup? [04:37] https://plus.google.com/hangouts/_/g6ga27vzkwgy3dz4s7xly5sivia?authuser=1&hl=en [04:38] ok [04:47] wallyworld__: https://github.com/juju/juju/pull/604 [05:02] davecheney: https://github.com/juju/juju/pull/605 [05:04] * davecheney looks [05:04] thumper: not logm [05:04] do not use NewXXTag in productoin code [05:04] unless you are 100% sure that the tag is valid [05:05] davecheney: the line above it validates [05:05] davecheney: how else should we do it? [05:05] names.ParseEnvironTag("environment-"+string) [05:05] if you're sure the tag is valid then LGTM [05:06] but this is a warning [05:06] hmm... [05:06] honestly those NewXXTag functoins shouldn't be in the package [05:06] they are a footgun [05:06] davecheney: but they have to be created somewhere, right? [05:06] in most cases they come as strings on the wire [05:06] kind-id [05:06] so we use parse [05:06] but something puts them on the wire [05:07] yup tag.String() [05:07] but something creates the tag [05:07] you have to have trust somwhere [05:07] no argument there [05:07] but you should look suspciously at every case [05:08] the same method is used in state.Initialize [05:08] we have a uuid [05:08] then create an environment tag from it [05:08] sure [05:09] but you are arguing that using a dangrous weapon is ok because others have used it heaps of times in the past with nothing going wrong [05:09] past performance is no guarentee of future profit [05:09] no, I am saying that this is one of the places where you use the dangerous weapon carefully [05:09] sure [05:09] there are always places where we need to create tags with known data [05:10] yes [05:10] but I don't see any validation there [05:10] you just take what comes out of hte jenv fil [05:10] no, this isn't the jenv [05:10] +ssInfo, err := st.StateServerInfo() [05:10] right, st here is *state.State [05:10] ssIfnfo should have an envTag field or method [05:11] no, StateServerInfo is not a *state.State [05:11] it's some turd that got passed back from the api [05:11] no, st is [05:11] ssInfo [05:11] StateServerInfo is a method on *State [05:11] +st.environTag = names.NewEnvironTag(ssInfo.EnvUUID) [05:11] you have a LGTM with resevatoins [05:12] there is no value in griding on this point [05:12] right, here is some ickyness... [05:12] which we can fix [05:12] StateServerInfo is a POD structure [05:12] Plain Old Data [05:12] all public [05:12] yeah, this is the same POS that infests the state and the mongo packages [05:12] and binds them tightly to the _client_ api [05:13] * thumper nods [05:13] we should separate the serialization structure from the info structure [05:13] it's fine for it to be public fields [05:13] it's returned by value [05:13] we can't change the copy that state has [05:13] but then adding a method that creates an environ tag from the uuid in the struct is meaninless [05:13] as it gives you a false sense of security [05:14] when there is none [05:14] obiously we'd remove the envUUID field [05:14] nah... I just put it there [05:14] well shit [05:14] this is all about our shitty data structures === Ursinha is now known as Ursinha-afk [05:21] wallyworld__: https://github.com/axw/juju/commit/b56a48d3bd760f9ab58ccada562dd663b1786a0d#commitcomment-7526685 [05:21] rightio [05:21] davecheney: let me ponder this env tag for a bit [05:21] can you please take a look at that and see if I'm making sense [05:21] before I merge it [05:21] I'd like to ensure that we do it right [05:26] wallyworld__: thanks. going to do some more testing before I land, and double check coverage [05:27] ok, it's got potential to break things does this branch [05:28] jam1: i got the bot "fixed" by throwing a larger instance at it - our tests are still horrible [05:28] wallyworld__: well it's pretty invasive so yeah... I have done some targeted testing with non-amd64 arch, will test the whole lot before attempting merge tho [05:29] ty [05:42] morning all [05:56] wallyworld__: :( [06:07] hazmat: in case this got lost in the noise of github activity: https://github.com/juju/juju/pull/596 === JoshStrobl[ZZZ] is now known as JoshStrobl [08:10] morning [08:31] morning TheMue [08:37] jam: seen your mail, could you tell me a bit more about it? [08:59] morning jam, TheMue [08:59] jam, the meeting will start any minute now :) [08:59] dimitern: heya [09:02] jam, ping [10:18] dimitern: pong [10:18] sorry about the delay [10:18] dimitern: dang it [10:18] sorrI missed it [10:18] I have to take the dog out now, will be back in 20 min or so [10:21] jam, no worries, i'll bring you up-to-speed at the standup [10:44] dimitern: speaking of which :) [10:45] TheMue: ^^ [10:46] brt [10:54] dimitern, ping - when you have a moment [10:57] mattyw, a tentative pong (doing standup now)? :) === Ursinha-afk is now known as Ursinha [11:49] do actions use a hook context? (ie. long running) [11:53] bodie_, fwereade ^ [11:55] hazmat: hi. There does not seam to be any difference between Hooks and Actions aside from location (when it comes to running). https://github.com/juju/juju/blob/master/worker/uniter/context.go#L330 [11:55] gsamfira, thanks [11:58] glad I could help :) === Ursinha is now known as Ursinha-afk === Ursinha-afk is now known as Ursinha [14:26] eh, maybe I’m blind, but do we have a github.com/juju/juju/upstart? [14:27] cmd/jujud/machine.go imports it, but I cannot find it (neither can my compiler) [14:28] dimitern: you’re around for a little crosscheck? [14:31] mattyw: ping [14:32] TheMue, hey there [14:33] mattyw: could you take a look please? it seems the repo has a problem [14:33] mgz, I am on the hangout whenever you are ready [14:34] TheMue, I certainly can but I might not be the best person for the job [14:34] alexisb: thanks for the poke [14:36] mattyw: maybe I already found the checkin [14:37] so currently our master won’t build, a package is missing [14:37] TheMue, now I'm here [14:38] TheMue, this one yeah? https://github.com/juju/juju/commit/1f7148c5e2ae9f68eb9f8b0c94f6c00b82ee4a18 [14:38] dimitern: thx, but found it already. [14:38] mattyw: yeah, exactly [14:38] ok [14:39] dimitern: in jujud machine.go imports a non-existing package :( [14:39] TheMue, dimitern but it looks like the package isn't used either [14:39] TheMue, what? [14:39] mattyw: yep [14:39] mattyw: I’m only wondering how it passed the bot [14:40] TheMue, me too - who's the best person to ask about the bot? [14:43] TheMue, the tests run http://juju-ci.vapour.ws:8080/job/github-merge-juju/431/console [14:43] TheMue, but that error is in main - do tests on main get run? [14:43] mattyw, TheMue, is this about upstart? [14:43] mattyw: this seems to be the problem [14:44] dimitern: yep [14:44] TheMue, mattyw, it seems juju/upstart moved into juju/service/upstart [14:45] and perhaps wallyworld had goimports installed and juju/upstart code was in GOPATH before juju/service/upstart, and probably the same happened on the bot [14:46] dimitern: hmm, could be the reason [14:46] it happened in https://github.com/juju/juju/commit/190f98fcab118b5dce269e8c0021a563455fee39#diff-88ad1ca7d18fe89a76f6348caf6ddd42 [14:47] dimitern, makes sense [14:48] dimitern, TheMue anyidea how we can stop this from happening next time? [14:48] mattyw, TheMue, it just happened so that code importing the old path was merged last [14:48] https://github.com/juju/juju/commit/880aaa83f1a474ef7856f1237c3781ab6a51dbfe [14:48] also upgrades isn’t used in machine.go [14:49] mattyw, I'm not sure if that's the case, but if it is, then we should look into the bot and see how it does fetch dependencies, etc. [14:49] dimitern, where is the code for the bot? [14:50] it might be that ian added the import line manually, rather than using goimports [14:50] mattyw, mgz would know that I guess [14:51] which bot bit? [14:52] mattyw: you want to look at the make-release-tarball script in lp:juju-release-tools [14:52] mgz, ok thanks TheMue ^^ [14:53] yep [14:53] what was the symptom exactly? I'm a little confused from the log [14:53] we had broken import that got past the landing? [14:53] or didn't get past the landing but did get past the build? [14:54] mgz: cmd/jujud/machine.go imports packages it doesn’t use and that doesn’t exist [14:54] I see, on trunk currently. [14:55] we should file a critical ci blocker bug for that [14:55] so nothing lands until it gets fixed [14:55] and blame is on the last rev of master? or an earlier one? [14:57] mgz, last rev [14:57] tasdomas, dimitern in other news this is ready for more reviews when you have a moment https://github.com/juju/juju/pull/562 [14:57] mattyw, will have a look shortly [14:57] oh, I see [14:57] go fmt passes... [14:58] and the go build line has gone [14:59] dimitern: my suggestion, I land a backout of the last rev [14:59] add `go build ./...` back to the tarball script [14:59] mgz, there seems to be another issue [14:59] reland the earlier rev and see that it borks? [14:59] mgz, ../../state/backups/metadata/metadata.go:10:2: cannot find package "github.com/juju/utils/filestorage" in any of: [14:59] /usr/lib/go/src/pkg/github.com/juju/utils/filestorage (from $GOROOT) [15:00] and it's not in dependencies.tsv as well [15:00] also the same rev? [15:00] if so, covered by the backout [15:00] mgz, let me check [15:00] seems not.. [15:01] probably eric's change? [15:01] mgz, yes, on trunk [15:01] mgz, but it's not the same rev I think [15:01] yeah, looks like that's ericsnowcurrently backups-storage [15:02] mgz, yep https://github.com/juju/juju/commit/f4da7f542947abb798da7da730a5482a029eee44 [15:02] so, two borked landings from the build line going... now, why was that removed... [15:02] mgz, so we're not even trying if it builds ? lol.. [15:04] well, not at the tarball stage [15:04] mgz, iirc there was a unit test for deps.tsv.. [15:04] we do before running the tests, and that's working for some reason [15:04] mgz, it takes like 10 secs - we should do it before running tests, not as late as tarball packaging time i think [15:04] tar [15:04] ah, I see [15:04] sorry, [15:05] tarball build comes before tests [15:05] mgz, hmm.. I wonder why that is [15:05] we get deps and make tarball on the main juju machine [15:05] then send the tarball to a new instance to run the tests [15:05] so, the bot *should* still be failing before we run tests, but from the logs it's not for some reason [15:07] can see the line `go build github.com/juju/juju/...` in the landing console log, and it's not got the error [15:08] a little worried it's not actually testing the right juju at present [15:11] hm, [15:11] I'm tempted to blame a godeps change [15:11] nothing on the ci side has changed [15:14] heh [15:14] okay, got it [15:15] mgz, yep? what is it? [15:16] ah? [15:16] for some reason, fixDetachedHead from cmd/go/vcs.go is now getting called, when it wasn't before [15:17] and that does checkout master... overwriting the merge [15:17] yay! :D [15:18] not sure *what* has triggered this, but can fix at least [15:18] lots of fun [15:18] godeps perhaps [15:18] lets hope it was recent [15:18] strange kind of error [15:18] it depends on how it gets missing revisions from git - if it does not use fetch but pull it can happen [15:18] because we've only been testing the current head, rather than the pending merge, for the last few landings at least [15:20] I’ve seen it first when testing the dummy provider. here I got it in github.com/juju/juju/environs/jujutest/livetests.go:124: build command „go“ failed … [15:20] nope, scratch that - godeps uses git fetch, at list in lp:godeps trunk [15:20] for now, I want to just back those two changes out [15:20] and eat lunch... === kwmonroe_ is now known as kwmonroe [16:02] jcw4, bodie_ we are on the hangout when you guys are ready [16:02] woo hoo [16:08] TheMue: #jujuskunkworks [17:00] hello everyone [17:00] hey! [17:07] anyone: pr #609 [17:09] gd [17:09] urk [17:09] gsamfira, perrito666: ^ [17:13] mgz: on which repo? [17:13] juju/juju [17:16] * perrito666 receives confirmation from msdn of subscription... I wonder when did I subscribe [17:16] it was at least one month ago [17:21] so what's the deal with upstart and how far back do we have to roll back to get it to build? [17:22] well, 2 options. there is one commit that was ported forward, and if we remove that one, it will build [17:23] or, we can do a PR, that removes the extra imports and calls agentConfig.Tag().String() instead of agentConfig.Tag() in a couple of places [17:23] and it will also build [17:23] but I have not investigated the issues related to the second PR that is being reverted by https://github.com/juju/juju/pull/609 [17:23] mgz might have more info on that [17:24] mgz: btw, can https://bugs.launchpad.net/juju-core/+bug/1361721 be reproduced in something other than utopic? [17:24] Bug #1361721: MachineSuite.TestDyingMachine failing frequently [17:24] perrito666: that's the only job it's on I think, but it's been a dodgy testfor a while [17:26] bodie_: I'm not sure, which upstart what? [17:26] the failing build on the latest master [17:26] bodie_ : the upstart package was moved to the service package quite a while ago. [17:27] bodie_: mgz has a revert in the pipeline now [17:27] bodie_: my pr should fix the failing build [17:27] ah, great [17:28] if there is no other issue with the commits that are being reverted, the change to get it to build without reverting is about 4 lines. I am running the tests now. Should I let them finish and see if that fixes it? [17:31] gsamfira: I want to just revert, because the tests were never run on those changes [17:31] then fix the landing before putting in new code [17:33] fair enough. As you wish. I am running the tests on that code now, with the fix. If you prefer to revert, its fine with me :). I was just offering an alternative that would be shorter === jheroux_away is now known as jheroux === urulama is now known as urulama-afk [19:26] perrito666: how's your morning go? [19:26] ericsnow: wonderful [19:27] perrito666: glad to hear it [19:27] ericsnow: btw, one of your PRs has just been reverted, please contact mgz for more info [19:27] perrito666: yeah, I saw :( [19:29] mgz: how exactly was my patch failing? [19:30] mgz: I'm guessing it's related to updating dependencies.tsv [19:31] 16:02 < dimitern> mgz, ../../state/backups/metadata/metadata.go:10:2: cannot find package "github.com/juju/utils/filestorage" in any of: [19:31] 16:02 < dimitern> I/usr/lib/go/src/pkg/github.com/juju/utils/filestorage (from $GOROOT) [19:31] 16:02 < dimitern> and it's not in dependencies.tsv as well [19:32] mgz: weird [19:33] worth trying the merge again and building locally, see if it's actually okay [19:33] mgz: github.com/juju/utils/filestorage has existed for some time and it's in the revision listed in depenedencies.tsv [19:34] mgz: at least as long as that revision didn't get reverted too [19:36] all: I just sent an email to juju-dev in the thread "getting rid of all-machines.log" you opinion will be greatly appreciated === arosales_ is now known as arosales [20:18] ok good news is: I don't need utopic to fix 1.20 tests [20:19] ace [20:21] sweet, 8G of ram really did the trick for test running [20:28] why cant I see builds before #545 for http://juju-ci.vapour.ws:8080/job/run-unit-tests-utopic-amd64/ ? [20:41] abentley: mgz jog anyone can tell since when has 1.20 -> http://juju-ci.vapour.ws:8080/job/run-unit-tests-utopic-amd64/ been broken? I know it failed for the last revision, but do we know if this is indeed something new [20:41] ? [20:41] sorry, shift too close to enter [20:42] perrito666: The last revision that passed was eba6e37f [20:43] abentley: thanks a lot [20:43] perrito666: r6dc9a588 was tested and failed, but I need to check to see whether it was the same failure mode. [20:44] perrito666: Silly me, that's the candidate. [20:44] weird, gitk says r6dc9a588 is not part of 1.20 [20:46] perrito666: I'm sorry, the last to pass was eba6e37f. The way jenkins displays this is confusing. [20:47] abentley: I know, dont worry, confuses me each time [20:48] mm, changes from eba6e37f to head of 1.20 contain a patch mgz just reverted from master [20:51] ping ericsnow, perrito666 [20:51] wwitzel3: hey [20:52] ericsnow, perrito666: got time for a quick meeting / standup? [20:52] wwitzel3: sure [20:53] ok, going to moonstone [20:59] wwitzel3: sorry, thought I had joined! [21:17] sorry was in another window, you guys still there? [21:17] perrito666: nope [21:18]