[00:32] hi night shift === arosales_ is now known as arosales [00:51] sweet my house is finally on street view,I am in the 21st century [00:56] welcome to the internet perrito666 [01:01] anyone know if its possible to deploy things to the juju bootstrap nodes, if you're using HA? I'm trying to go do lxc:0, lxc:1 type of thing [01:02] in the past we've done a juju deploy of cs:ubuntu to the units based on constraints [01:11] ah, I might have to use native juju deploy to get the cs:ubuntu deployed, then use juju-deployer [01:26] bradm: yes you can [01:26] bradm: however the boostrap node isn't as special in an HA world [01:26] bradm: it just becomes one of the state servers [01:26] don't rely on hard coded numbers [01:26] you can use placement directives for stateserver nodes just like normal [01:27] thumper: I'm trying to deploy openstack HA things to those 3 nodes [01:27] bradm: we don't have any magic short cut names [01:27] bradm: but if this is something that you envision doing a lot [01:27] thumper: I'm pretty sure this is a juju deployer bug [01:27] it might make sense to add some sort of alias [01:27] probably deployer [01:28] thumper: I just did a juju deploy cs:trusty/ubuntu --to 0, and then juju add-unit ubuntu --to 1 [01:28] thumper: the use case here is bootstack, we want to have a HA juju bootstrap node, and then use those same 3 nodes as the HA for the rest of openstack [01:28] thumper: by deploying ubuntu to the 3 HA bootstrap nodes, we can then deploy the rest via lxc [01:29] jam: any idea what's up with the old landing bot? https://code.launchpad.net/~mark-sheahan-ms/gwacl/cert-args/+merge/235889/comments/577883 [01:29] thumper: aha, bug 1324129 seems relevant to my interests, just added a comment to it. [01:29] Bug #1324129: unit placement can only go to machine 0 [01:30] there's a real mix of inequality between juju and juju deployer about what you can do placement to [01:30] bradm: why are you bothering to deploy ubuntu? it is the do nothing charm [01:31] thumper: hrm, I probably could just use lxc:1, couldn't I, now I'm outside of juju-deployer [01:31] or whatever the format is [01:32] thumper: our standard way of deploying it has been to deploy ubuntu using constraints and give it a name, and then use that to deploy the units to [01:32] thumper: oh, because I need to be able to refer to those hosts by service name, thats why [01:32] whaaa??? [01:32] thumper: juju-deployer will only let you refer to machine id 0, and thats it. [01:33] bradm: yes [01:33] bradm: yes to the "lxc:1" [01:33] thumper: right, which won't work inside juju-deployer yaml [01:35] I can't see any other way around it until juju-deployer loses the idea that machine id 0 is special [01:40] ideally both juju and juju deployer would agree on what you can deploy to [01:40] ideally [01:40] juju deployer will let you use service names, but only machine id 0 [01:41] and juju will let you use whatever machine id, but not service names [01:41] so we're having to switch and change a bit [01:45] thumper: well, there's now bugs on both programs to make it equal.. :) [01:45] bradm: how does the deployer let you specify a service name as a target? [01:46] and what does it mean when there are 0 units [01:46] or more than 1 unit? [01:46] thumper: you can do things like lxc:ceph=2 [01:46] thumper: and that'll do the 2nd ceph unit [01:47] bradm: file a bug, and I'll make someone do it :) [01:48] thumper: already done days ago [01:48] thumper: bug 1372759 [01:48] Bug #1372759: Extend juju deploy --to format to allow specifying service name

[01:48] ta [02:01] well, way to paint myself into a corner - I can't use juju deployer to deploy to units 0, 1 and 2, but I can't use juju to deploy via service name [02:09] bradm: fixing that is part of stuff we'll be looking at in the coming weeks. We've got a call next week with the deployer devs to make sure we move the parts in sync [02:09] rick_h_: awesome, that'll be good. [02:10] I still need to figure out a way to deploy with HA bootstrap nodes, and with HA openstack, both on the same 3 nodes. [02:10] bradm: hmm, we're looking at specifying new machines in the work. This is slightly different. We'll keep this in mind as we complete out the task list. [02:11] rick_h_: basically, we're trying to use the 3 HA bootstrap nodes as a target for lxc containers for HA openstack, if that makes ense. [02:11] er, sense. [02:11] bradm: so I'd be curious how an HA juju env looks in the GUI machine [02:11] machine view that we put out today [02:12] bradm: and then you can use the webUI to create containers on each of the state servers and place services on it [02:12] rick_h_: easy enough to deploy it, is it in the normal charmstore? [02:12] bradm: yes, it was published today [02:13] rick_h_: ok, I'll have a look at that shortly, just have to answer the door [02:13] bradm: rgr [02:14] bradm: shared a video of the machine view stuff on the orange box doing containers/etc. So it's missing the HA juju part which I've not tried out. It might give some idea if it'd be useful or not if I'm following you. [02:14] bradm: what provider are you using? MAAS? [02:48] thumper: ping [02:49] menn0: hey [02:49] thumper: just back from kids swimming. [02:49] thumper: I've just figured out why those units aren't coming up after upgrade [02:49] and? [02:50] thumper: the unit agents aren't starting up after shutting down to restart into the new tools version [02:50] thumper: once I manually start the agent, all is fine [02:50] thumper: I have another unit that's still in that state so I'm going to poke around [02:50] thumper: but do you have any ideas? [02:51] only that if the process exits with a particular exit code, upstart won't try to restart it [02:51] but I don't recall what it is, axw might [02:52] I wasn't aware of special exit codes [02:52] *I'm not aware [02:52] hmm... pretty sure there is one [02:54] normal exit 0 [02:54] in the upstart config file [02:56] menn0: so if the user agent exited with a value of 0, it wouldn't have been restarted by upstart [02:56] thumper: yep I got that [02:57] thumper: I'm not seeing anything in the unit agent's logs that indicates why this might have happened. [02:57] thumper: I'm going to try with master just to make sure this isn't a more general problem. [02:57] * menn0 is glad he scripted the envrionment setup [02:57] ok [03:00] rick_h_: yes, we're using maas [03:00] rick_h_: this is on softlayer hosts [03:08] davecheney: thanks for all those reviews [03:08] davecheney: would you mind taking one more look at http://reviews.vapour.ws/r/88/? [03:12] ericsnow: looking [03:22] davecheney: thanks [03:22] thumper: it's definitely my changes causing the unit agents to not start [03:22] stragne [03:24] review done [03:27] menn0: did you want to talk through it? [03:28] thumper: probably a good idea [03:28] thumper: standup hangout? [03:28] menn0: I'm in the standup hangout :) [03:31] HOLY SHIT [03:32] did everyone know that in RB you can attac a comment to MORE THAN ONE LINE ? [03:33] oh? [03:34] davecheney: that's one thing I like about reviewboard [03:34] click and drag, baby! [03:34] davecheney: I did not but I just tried it and you totally can :) [03:35] ericsnow: it occured to me in the review comment mode, why it would only show one single line with no context [03:35] so, that's the reason [03:35] now if the bastard would only show comments and diffs on the same page [03:36] davecheney: the comments are on the diff: the little numbered boxes on the left (click to expand and reply from the diff page) [03:36] ericsnow: yes, people keep saying that [03:36] so, click on the box [03:36] davecheney: :) [03:36] and you want to reply to their reply [03:36] try it [03:36] you'll see why it doesn't work [03:37] ericsnow: is there an optoin to have RB auto expand those little boxes [03:37] ? [03:37] davecheney: you mean that it redirects you away from the diff? [03:37] davecheney: not natively [03:38] ericsnow: bingo [03:38] davecheney: apparently reviewboard has quite a few extensions out there so I expect there's something [03:43] rick_h_: https://chinstrap.canonical.com/~bradm/ha-juju.png is what HA juju looks like in the gui, in the machine view [03:54] restarting into utopic. bbs.. or possibly not [03:59] meint gott, it just worked [04:36] thumper: ping [04:36] menn0: ya? [04:36] thumper: so scratch what I thought it might previously [04:37] thumper: that unit died when the state server restarted [04:37] thumper: (into the new tools version) [04:37] thumper: lots of workers died because they lost their API connection (expected) [04:37] thumper: but for some reason the uniter decided it should die too [04:37] thumper: logs here: http://paste.ubuntu.com/8430245/ [04:38] thumper: that's the tail of the unit's log [04:38] thumper: the key part is: INFO juju.worker.uniter uniter.go:144 unit "rsyslog-forwarder-ha/1" shutting down: tomb: dying [04:38] and probably DEBUG juju.worker.uniter modes.go:398 ModeAbide exiting [04:38] that shouldn't have happened [04:39] the broken pipes, dying workers and watcher errors are expected when the state server goes away [04:40] the other units had units restart when the state server died but they came back after a few retries, once the state server was available again [04:40] but for some reason this unit (and others in previous upgrade attempts) decided to die [04:40] I'm guessing it's a timing thing somehow [04:41] hmm... [04:41] ModeAbide? [04:41] I learned about that in my first week thanks to talk by Will [04:41] :) [04:42] it's a function in the uniter [04:42] but also the steady-state mode of the unit agent [04:44] I would have thought that any exit of the agent due to an error should be non-zero, and hence restarted [04:44] what you seem to be saying is that there is some situations where that doesn't seem to be the case [04:45] yes [04:45] hmm... [04:45] that seems to be the case [04:45] look at agentDone function [04:45] and beyond that, the unit agent shouldn't have even been exiting at that point [04:45] jujud/agent.go:315 [04:46] worker.ErrTerminateAgent [04:46] I'm wondering if that is the error being returned somehow [04:46] what it looks like from looking at how ModeAbide is exited is that the unit's state went to Dying [04:46] which might explain why the agent exited with 0 [04:46] it was as if the unit had been told to terminate [04:46] huh? [04:47] the uniter thought it was dying? [04:47] yep [04:47] that would explain why it didn't restart [04:47] exactly [04:47] but seems weird that it might think it is dying because it can't talk to the api server [04:47] I'm just checking what the state server sent back over the API about this unit [04:52] thumper: nope. the "life" calls for that unit always respond with "alive" [04:53] probably worthwhile running this test a few times on master if it is timing based [04:53] as pretty sure it isn't your change that did this [04:53] yes it is important, but perhaps a decent bug report would be best right now [04:54] thumper: you think it's not my change? I'm not sure yet. [04:54] thumper: but good to check on master a few times to be sure [04:54] fairly sure... [04:54] not 100% [04:54] but well up there [04:55] ok. recreating the env and switching to master now. === urulama-afk is now known as urulama === alexlist` is now known as alexlist [08:44] any reviewers around? http://reviews.vapour.ws/r/109/ [08:44] TheMue, cmars, as OCR can you have a look please? ^^ [08:44] dimitern: yep, will do [08:44] morning btw [08:44] fwereade, you might be interested, if you have time ^^ [08:44] TheMue, morning :) [09:30] fwereade, TheMue, cmars, another trivial cleanup follow-up - http://reviews.vapour.ws/r/110/ [09:32] dimitern: I'm almost throught with 109, will take a look then [09:33] TheMue, cheers! [09:46] Is there a document somewhere with details on the upcoming Brussels sprint? === fabrice is now known as fabrice|lunchpar [09:50] dimitern: so, you've got a review [09:50] Spads: afaik only the typical wiki page [09:51] TheMue: where is it? [09:51] https://wiki.canonical.com/CDO/Sprints/JujuOct14 <-- oh, this? [09:51] TheMue, tyvm [09:56] Spads: yep [09:56] TheMue: many thanks [09:56] Spads: yw [10:05] TheMue, cmars, fwereade, and another small follow-up for the uniter - http://reviews.vapour.ws/r/111/ [10:08] dimitern: hey, I wonna be productive today too, but you don't let me :D [10:09] TheMue, sorry to be a pest, but I really want to finish with the port ranges stuff today if I can, so we can start again on container addressability next week :) [10:11] dimitern: hehe, yeah, and it's my job as ocr today [10:11] TheMue, +1 [10:23] dimitern: #111 is marked as WIP, so waiting with review? Oh, and btw, #110 is done. [10:25] TheMue, yes, all of the follow-up are WIP, because they can't land before their parent is merged into juju:master [10:26] dimitern: ok [10:45] dimitern: having a quick talk? [10:46] TheMue, yep, sorry, omw [11:52] TheMue, cmars, fwereade, last one for today :) I promise - http://reviews.vapour.ws/r/112/ [11:53] dimitern: 112 is the german number for emergency calls, so I'm not sure :D [11:57] TheMue, :D it's the same here [11:58] TheMue: its the argentinian number for helpdesk, that never works btw [12:03] perrito666: *lol* [12:06] axw: standup? [12:07] katco: I'm there [12:08] axw: huh... so am i? [12:08] I'll rejoin [12:20] http://www.reddit.com/r/juju/comments/2his7a/juju_gui_now_has_an_awesome_machine_view/ for your upvotes please [12:24] rick_h_: sexy [12:24] rick_h_: looking really sharp +1 [12:24] * katco watching your video right now [12:24] shame about the lack of hardware details in the old maas provider [12:24] yea, glad to see the bug get fixed [12:24] you actually got me pausing my music to look at the vid [12:25] so hopefully it'll just get better in time as juju support hardware info more [12:25] rick_h_: they all do now [12:25] so yep :) [12:25] (all do in master) [12:25] .. and 1.20.8 I think? [12:25] rick_h_: you need music on those vids :p and perhaps that sound from a boats horn that is used in all the movie trailers nowadays [12:26] +1 [12:27] lol [12:27] something like [12:27] THIS SUMMER JUJU WILL HAVE LESS BUGS [12:28] SUNDAY SUNDAY SUNDAY [12:28] *explosion* [12:28] rofl [12:28] then DEPLOYS BUNDLES HA NEW VIEW [12:28] i'm just wondering when they'll start combinging the buzzfeed headlines with trailers [12:29] "YOU (DUNNNNN) WON'T (DUNNN) BELIEVE (pkaw!) HOW THESE TWO PEOPLE FALL IN LOVE" [12:29] rick_h_, upvoted! [12:30] dimitern: ty! [12:30] re [12:30] k [12:30] sdadg [12:30] sorry kb hit the ground [12:31] rick_h_: upvoted [12:31] if you tweet something Ill also retweet you [12:32] you want like an officially tweet https://twitter.com/jujuui/status/515467739951923200 [12:32] or a personal tweet https://twitter.com/mitechie/status/515462052106629121 [12:33] or another point of view tweet https://twitter.com/jaycee/status/515468467567603712 [12:33] :) [12:33] rk [12:33] j [12:33] mm the fall clearly was not good on this kb [12:33] now sometimes shift enters [12:33] rick_h_: retweets are for free so I can go with all three of them [12:37] thanks for all the points/sharing, go juju! :) [12:38] bradm, deployer will support arbitrary machines.. it was a safety belt for portability.. but it reality is its very useful [12:38] to go to arbitrary machines === jheroux_away is now known as jheroux [13:02] ahh the old "gmail is unavailable, check our status page" ..... Status page says everything is fine [13:03] reloading fixes it, though, so, I guess I won't ragequit gmail today [13:03] well, they dont say that the status page is working [13:03] lol [13:03] where's the status page for the status page? [13:04] natefinch: "check the status page" is an euphemism for "stop hitting our mail servers while we try to get them working again you refresh maniac" [13:07] I'm very sad that juju.io is not something we own [13:08] juju.wtf is available, though. [13:09] juju.bike [13:10] there's juju.works which is kinda cute [13:14] +1 === fabrice|lunchpar is now known as fabrice [13:21] and most likely juju.ninja should be available too [13:21] nope, taken [13:21] maaan, I finally got to know how this is done http://cdn.diply.com/img/6d874ee0-a5db-43f2-ad34-b358b618bd5f.jpg [13:21] I can die happily [13:22] neat! [13:22] I was a whole day trying to get http://ide.ninja/ :p and finally got it, they are oddly unexpensive [13:23] gojuju.io is available, which is a pretty good compromise I think. Not ideal, but since we already failed to get any juju domain names when deciding on the name of the product (seems like bad planning.... everyone knows the first thing you do when starting a project is buy the domain name!) [13:26] natefinch: lol [13:26] when I started my company I decided the name based on the one with the lowest google hits and domains taken [13:26] * natefinch has several projects that never got past the "buy the domain name" stage, for example... [13:26] so while we did that we had a functioning company for like 4 months called thing2009 [13:26] haha [13:26] in spanish, which is coso 2009 [13:33] natefinch: my calendar says 1:1 [13:36] TheMue: could you get me a review on http://reviews.vapour.ws/r/107/? It's pretty small. [13:39] ericsnow: will do it next, right now I'm doing the 113 [13:39] TheMue: cool, thanks! [13:45] TheMue: I could also really use http://reviews.vapour.ws/r/88/, but 107 is a bigger priority for me :) [13:51] ericsnow: hmm, just wanted to make you happy with a done #107 and you've got already the next one for me. ;) [13:51] TheMue: I'm a slave driver ;) [13:52] ericsnow: yeah, I see it. this morning already dimitern has been a slive driver. [13:53] TheMue: ah, to be OCR and free! [13:56] for backups upload and download we could follow the precedent of tools and charms (add hooks directly to the API server to handle HTTP requests directly)... [13:56] ...or teach the RPC server to handle files too. [13:56] any quick thoughts? [13:57] (I'll probably end up taking this to the ML) [14:09] wwitzel3: standup? [14:22] anyone know off the top of their heads if add-unit -n 3 --to=1,2,3 is possible? [14:25] rick_h_: I don't think so, but not 100% sure [14:26] natefinch: thanks, will see if I can try it out [14:30] marcoceppi: the summary formatter just landed :) [14:30] * marcoceppi revs up the git machine [14:33] katco: I had a huge 20 machine deployment yesterday, but completely forgot about testing as it was late at night, I'll be spinning up another one soon though [14:34] marcoceppi: no worries. it's been pretty well banged on, i'm just excited to see what large deployments look like :) [14:52] oh, just got notified my new provider will install the new connection (100 mbps) on tuesday. let's press thumbs that e'thing works [14:54] 100MPS is pretty sweet [14:55] dimitern, TheMue: could you guys talk to johnmc on #juju and try to give him a hand? I spent a few hours working with him yesterday. He has a machine that used to be able to deploy lxc containers and now can't [14:58] natefinch: sadly have to step out after my actions meeting in 2 min [15:01] * perrito666 has 12 m and is happy enough [15:35] ericsnow: is there anything backups related I can assist with? [15:36] wwitzel3: you could take a look at upload/download or at adding more functional tests [15:37] wwitzel3: preferably the functional tests :) [15:37] wwitzel3: thanks for the offer, BTW [15:37] ericsnow: sure, any specific tests you're looking to have done? [15:39] wwitzel3: I guess one each on the backups interface (state/backups), on the API (api/backups), and the command (cmd/juju/backups) [15:40] ericsnow: so by more, you meant all? :P [15:41] ericsnow: sounds good, I'll ping you went I'm getting ready to start on them, I'm wrapping up the runcmd stuff now. [15:41] wwitzel3: well, we have functional tests on CI that we will switch to the new backups once we ditch the old, but yeah :) [15:41] wwitzel3: cool [15:42] ericsnow: I learned that case Foo, Bar: expression .. is much more useful than case Foo:\ncase Bar: expression [15:43] ericsnow: coincidentally learning that also fixed the last known bug I was stuggling with in apiserver/runcmd [15:44] which then made me feel dumb [15:44] yay! [15:44] wwitzel3: oh, you mean how each case has an implicit break and you must explicitly ask it to chain them? === fabrice_ is now known as fabrice|out [15:45] ericsnow: yep [15:45] wwitzel3: I hate stuff like that [17:44] anybody using 14.10? [18:22] ericsnow: you around? [18:22] natefinch: yeah [18:23] ericsnow: finally got to look at the backups list stuff [18:23] natefinch: cool [18:24] ericsnow: how is this safe? https://github.com/ericsnowcurrently/juju/blob/backups-list/state/backups/backups.go#L99 [18:25] ericsnow: by definition, you're letting people create a backup value with anything that implements filestorage.Filestorage [18:25] natefinch: we control what goes in so we have a guarantee as to what comes out [18:26] This says you will accept anything that implements the interface: https://github.com/ericsnowcurrently/juju/blob/backups-list/state/backups/backups.go#L51 [18:27] ericsnow: either we accept anythign with the interface, in which case, the downcast is bad, or we don't, and we should only accept the concrete type we really expect to get passed to NewBackups [18:27] natefinch: hmm [18:28] natefinch: you make a good point [18:28] ericsnow: it's perfectly ok to take a concrete type for NewBackups.... it's not really a big deal [18:28] natefinch: an error would be better there than a panic [18:28] natefinch: it makes testing a pain [18:29] well, I can write a test that will make List() panic right now ;) [18:29] natefinch: right [18:29] natefinch: sorry, the seconds statement was not in relation to the first :) [18:30] natefinch: I agree that the code should be doing an "ok" check on the type assertion [18:31] natefinch: and passing a concrete type to NewBackups makes testing a pain [18:32] ericsnow: yeah, but there's a difference between making testing difficult and making your interface lie. [18:34] natefinch: at present the two are mutually exclusive so I picked the latter :) [18:35] I think the problem is that people are trying too hard to be generic with interfaces, when they should be generic with plain old structures (or POS as they say in the business ;) [18:36] natefinch: you're missing a closing paren there, friend. [18:37] (friday humor) ;) [18:37] I usually count the paren in the smiley face as the close paren (since I so often end asides with smilies :) [18:38] blasphemy! [18:38] parens must bring balance to the universe [18:38] i often wonder if only programmers notice those kinds of things [18:39] They said that about Anakin Skywalker, forgetting that, at that point, all the Jedi were currently good guys. [18:39] or maybe grammar nazis [18:39] katco: to have balance you dont need another parens just to begin with (; [18:39] * katco is a trekkie [18:39] perrito666: lol [18:40] also ;)) looks like a guy with double chin [18:41] no, it's a big grin! [18:41] yeah, the problem is that my smilies are generally supposed to be inside the context of the parens, and doing (foo :) ) just looks weird [18:41] I just have the compiler automatically insert the extra parens where needed [18:41] natefinch: when inside parens I just :p just in case [18:42] syntactically, commas and em-dashes are acceptable for asides. they have the same meaning as parens, it's just a matter of how much of a pause you want to cause the reader [18:43] ericsnow: I am still thinking about your code, btw [18:43] natefinch: k [18:43] natefinch: thanks! [18:43] katco: I am pretty sure grammatical rules dont include smileys [18:43] at least in spanish [18:43] katco: I like parens because it's immediately clear that where there's an open paren, there's going to be a close paren - whereas with a dash, it might just be a single dash. [18:43] perrito666: true, true [18:44] natefinch: but as i lex your sentence, i parse :) as a single token, so i'm still waiting for the closing paren! [18:47] was that too nerdy even for this crowd? :p [18:47] katco: nah, just trying to get some work done too. I Was going to make some joke about keeping parsers flexible enough for real-world inputs [18:48] hrm. aren't parsers generally very pedantic? [18:49] browsers aren't *shrug* [18:50] many think to their detriment... quirks mode promotes horrible practice [18:50] oh I know [18:52] katco: your tokenizer could be a bit more fault proof [18:53] an actuall lexer for english would have parsed that correctly and dropped the ; since there is no construction that includes ;) and ending in a ; is most likely an error [18:53] * perrito666 worked in NLP before juju [18:53] oooh fun stuff! [18:54] one of the 1st go programs i wrote was a NLP using markovian chains [18:54] katco: I used heavy stuff made by linguists, awful c++ [18:54] perrito666: i will have to buy you a beer sometime [18:54] and slow as a snail in reverse [18:59] ericsnow: who calls newapi? [18:59] perrito666: look in the init func [19:00] oh I understand [19:03] fatal: remote eric already exists [19:06]