[00:00] sinzui: btw, most likely I will be trying to fix restore failing from good connections issue :) [00:09] wallyworld_: I got the limits lifted on Joyent if you want to give it another try, when you have time. [00:10] arosales: otp, sec [00:10] wallyworld_: ack [00:15] arosales: thanks for that. curtis ran into the same issue originally as well but the CI tests were wteaked to deploy into lxc containers instead of a new instance and it all looks good :-) [00:16] wallyworld_: yup that was on two different accounts too so I am checking with Joyent if we need to include that in our instructions of if the could lift that globally to at least 10 instances or something reasonable like that [00:16] sure [00:17] wallyworld_: would be nice to get confirmation you can add-unit or deploy more than 1 service (without colocating) [00:17] incase I need to ping Joyent again [00:17] ok, i'll try it out as soon as i can today [00:19] wallyworld_: ack and no rush just wanted to let you know before I grabbed some dinner [00:19] wallyworld_: and again thanks for getting that landed wouldn't have happend with out you chipping in [00:19] np, that's what we're here for [00:35] waigani: thanks for lookin into that issue [00:35] is the problem that the code is tryin to consume the body without checking the status code ? [00:36] yep, just writing something up now [00:36] kk [01:20] https://bugs.launchpad.net/bugs/1301663 [01:20] <_mup_> Bug #1301663: cmd/juju: panic during bootstrap [01:27] davechen1y: lboxing.. [01:28] waigani: right o [01:28] on da phone [01:28] ha, can't lbox because of proxies [01:48] anyone tested juju 1.17.7 on trusty with kvm in local provider? [01:48] i can get it to bootstrap but when i deploy none of the vms are being created [01:48] davechen1y: https://codereview.appspot.com/83310044 [01:50] oh interesting looks like im getting a connection refused over the websocket connection [01:50] lemme fix that first [01:51] so it just takes a minute to get the connection and then fails with : http://paste.ubuntu.com/7196827/ [01:53] seems to just repeat that warning about no instances found [01:54] lxc also seems to work just fine [02:13] stokachu: yes, kinda [02:13] stokachu: I tested kvm with saucy when I did it [02:13] but not trusty yet [02:29] thumper: i'm very concernred at the number of unchecked interface conversions in watcher/watcher.go ~400 [02:29] lots of shit to go wrong there [02:30] i have a stack trace from hazmat that is very worrying [02:30] that certainly is a lot of conversions [02:30] i'd like to remove all the v, _ 's [02:31] this code behaves as if the conversion cannot fail [02:31] so we should assert that by not using the two arg version [02:31] for example [02:31] dr, _ := c.Value.(bson.D) [02:32] if this isn't a bson D, then what the fuck is it ? [02:33] and this is just one more reason why I like generics :P [02:33] however... [02:33] I agree, blind type assertions are wrong [02:33] there are a few cases in the codebase where they make sense [02:34] but I don't think this is one [02:35] * davechen1y logs issues [02:36] what the complete fuck [02:36] ... value *errors.errorString = &errors.errorString{s:"cannot assign unit \"s0/0\" to machine 1: machine \"1\" cannot host units"} ("cannot assign unit \"s0/0\" to machine 1: machine \"1\" cannot host units") [02:39] wallyworld_: did your unit address changes get into the 1.18 branch? [02:40] wallyworld_: just about to add a comment to #1259908 [02:40] <_mup_> Bug #1259908: juju status of units report private address (172.x.x.x) instead of public fqdn

[02:40] axw: the ones aligned with jyent provider, yes [02:40] right, they had to - thanks [02:40] so the uniter doesn't set unit address on start up anymore [02:40] yup [02:47] davechen1y: have you played much with web sockets? [02:47] davechen1y: I'm trying to work out how to write a test that needs to read from a web socket [02:48] thumper: i haven't [02:48] not at all [02:48] sorry [02:48] np [02:48] I'll keep poking with a stick from a distance [02:49] thumper, do you need to talk websocket to it? [02:50] thumper, its start of as http 1.1.. then does an immediate upgrade. [02:50] off [02:50] yes, I think I do, as this is how debug-log is streamed back to the client from the api [02:51] yup, that is about all I know [02:51] thumper, just using an existing websocket client and setup the server on a high socket for test... its separate from the api server afaicr. [02:51] no... [02:52] it is part of the apiserver [02:52] but I have it working so far [02:52] at least tests for the error conditions are good [02:52] perhaps I just make a client like the client would :-) [02:52] cool [02:52] * thumper goes to read more code [03:14] davechen1y: the proxy nuke lines were not meant to be there sorry, taken out now: https://codereview.appspot.com/83310044 [03:15] waigani: thanks [03:15] waigani: let me test it [03:15] does anyone else have time for a review ? [03:24] davechen1y: I can probably spare some time, what needs reviewing? [03:24] https://codereview.appspot.com/83310044 [03:24] looks pretty staight forward [03:28] waigani: LGTM, passes for me [03:28] thank you [03:28] sweeeet :) [03:28] waigani: please fix Body.Close before landing [03:28] waigani: I added a comment [03:28] axw: yep, will do [03:29] ta [03:29] axw: thanks, good catch [03:44] axw: moved Body.Close to above if, logged the body (debug level), checked log and test on ppc. [03:44] shall I push to lp, or do you want another lbox? [03:44] axw: ^ [03:44] waigani: lbox propose [03:45] okay [03:45] waigani: thanks [03:45] just give lbox half an hour ... [03:48] :) [03:49] axw: davechen1y: https://codereview.appspot.com/83310044 [03:51] shipit [03:51] :D [03:51] hodl up [03:51] one more thing [03:53] yeah, good point about Errorf [03:53] waigani: ^^ dontshipit [03:53] haha [03:54] I had my finger on the button [03:55] davechen1y: oh nice. That looks a lot better! [03:55] http://paste.ubuntu.com/7197095/ [03:55] umm, this is bad [03:56] I'll just push that last one to lp right davechen1y? [03:57] waigani: lbox propose [03:58] oh shit, I missed the log as error, sorry hang on [04:00] * thumper goes to make dinner [04:00] back later tonight for meeting [04:01] hoping to catch rogpeppe before meeting to talk about testing websockets [04:01] my girl just got off the ice so I'm going to have to disappear [04:01] back later tonight [04:02] davechen1y: https://codereview.appspot.com/83310044 [04:03] ok [04:04] davechen1y: just taking off skates etc so I can ship it if I get an lgtm from you [04:04] waigani: i'll review in a sec [04:04] i'll mark the review as approved and make sure the bot lands it [04:05] davechen1y: great. Thank you. [04:12] oh crap, the joyent test take > 10 minutes [04:12] and get killed by the watchdog [04:14] davechen1y: what's the issue with stripping the jujud binary? i didn't add that bit in, just curious [04:14] striping go binaries breaks them [04:15] we don't strip them in the release build [04:15] but I found a place in environs/tools where building local tools it still strips [04:15] yeah. i guess it still works somewhat cause i think that's what upload tools uses [04:15] and people have been running ok with that [04:15] wallyworld_: only on amd64 [04:16] ah [04:16] it really really doens't work for arm [04:16] rightio [04:16] i've assigned it to me to fix [04:16] i've already got a fix [04:16] will propose in a sec [04:16] just running tests [04:16] mark the bug as in progress then!! [04:16] ffs [04:16] i only just logged it [04:16] hold your horsees [04:17] i was trying to be hepful :-P [04:17] i know [04:17] * davechen1y pats wallyworld_ in a fatherly way [04:17] i saw a few bugs, all raised by you and thought you may need some extra hands on deck :-) [04:18] good thinking [04:18] wallyworld_: if you have time [04:18] i'm really getting nowhere with the simplestreams problems [04:18] juju doesn't even build on 386 [04:18] sorry, i mean the tests don't pass on 386 [04:18] which one? [04:18] so there is a lot of fixture data [04:19] wallyworld_: http://paste.ubuntu.com/7197153/ [04:19] i'm sorry i can't get better reports [04:19] it takes fucking forever to run these tests [04:19] these tests are running inside a 386 vm? [04:20] wallyworld_: http://cloud-images.ubuntu.com/releases/14.04/beta-2/ [04:20] cloud images, just push the button [04:21] oooh. nice [04:21] yeah [04:21] really nice [04:21] need a ubuntu image [04:21] push the button [04:21] ok, i have a wip branch i need to get done but i can look at the tests after that [04:22] ok [04:22] thanks === vladk|offline is now known as vladk [05:38] axw: got a sec ? [05:39] davechen1y: yup? [05:55] afk for a bit [06:06] gaah it just took me *far* too long that filepath.Walk does not follow symlinks [06:06] s/long/long to realise/ [06:08] fwereade: heh :( been there [06:55] thumper: i'm here if you want me [06:56] fwereade: did you want it to? === vladk is now known as vladk|offline [07:17] rogpeppe, I wouldn't usually, but I was passing it a symlink unawares and totally baffled as to why it wasn't walking any further :) [07:17] fwereade: ah... [07:18] davecheney, ping [07:18] fwereade: early morning amusing reading, BTW, if you hadn't seen it: https://www.usenix.org/system/files/1403_02-08_mickens.pdf [07:19] rogpeppe, I like that guy's stuff, don't think I've seen that one, thanks :) [07:19] fwereade: me too, and me too [07:21] fwereade: do you want to do any final review before I land this? (grouping is not enabled until azure-mode lands) -- https://codereview.appspot.com/73910043 [07:22] axw, I'll do a quick pass now, if you haven't heard from me in 20 mins don't let me block you [07:22] fwereade: okay, thanks [07:22] "So, yes, it would be great if fixing your browser involved actions that were not semantically equivalent to voodoo. But, on the bright side, things could always be worse. For example, it would definitely be horrible if your browser’s scripting lan- guage combined the prototype-based inheritance of Self, a quasi-functional aspect borrowed from LISP, a structured syntax adapted from C, and an aggressively asynchronous I/O model that [07:22] requires elaborate callback chains that span multiple generations of hard-working Americans. OH NO I’VE JUST DESCRIBED JAVASCRIPT. What an unpleasant turn of events! People were begging for a combination of Self, LISP, and C in the same way that the denizens of Middle Earth were begging Saruman to breed Orcs and men to make Uruk-hai." [07:24] fwereade: main change after merging is that it now uses DistributionGroup and takes the cloud-service name from the first instance.Id belonging to Azure, whereas the old code would use labels based on principal units [07:26] rogpeppe: he's hilarious. I wish more people wrote like him; I might learn more :) [07:26] axw: +1 [07:31] axw, LGTM [07:32] fwereade: thanks === vladk|offline is now known as vladk [07:59] fwereade: do you want https://codereview.appspot.com/81340043/ in before I enable azure? bear in mind that azure-mode will not allow add-machine or --to [08:00] *enable azure-mode [08:00] fwereade: btw, availability-sets-enabled is a bit wordy, do you think we should actually call it azure-mode? [08:01] axw, sorry, I'm hitting dangling pointers in my brain [08:01] no worries :) [08:01] it's a bit twisty [08:03] axw, so, that CL appears to lack tests a bit [08:03] fwereade: yeah, it was just a WIP to get your thoughts [08:04] I meant, do you want me to flesh it out and continue on with it before calling azure done. This one really is about ec2 and others where you would be allowed to add-machine and --to [08:04] axw, so AFAICT they won't interact -- unless azure-mode allows for assignment to clean/empty machines, but IIRC it doesn't [08:04] right [08:05] fwereade: you made a comment on the DistributionGroup CL: "LGTM. Taking DG into account when choosing pre-existing instances is not [08:05] in scope for this CL, but please make sure it's on your list of [08:05] necessary-features-before-done-done." [08:05] axw, ok, I think we should get that in, because we want to enable service distribution in the other providers: once we have the interface in place, it becomes a "simple" matter of adding bugs for individual providers [08:06] fwereade: agreed, but can I go ahead and land the remaining azure stuff first? [08:06] axw, certainly [08:06] cool [08:06] axw, sorry slow to catch up :) [08:06] no problemo [08:31] morning all [08:32] frankban, can I get your +1 on https://code.launchpad.net/~wallyworld/juju-core/watcher-address-issues/+merge/213968 please? [08:32] frankban, (or not ofc ;)) [08:33] fwereade: sure, looking and qaing it [08:38] fwereade: does the comment at line 8-9 reflect reality? [08:39] frankban, wait, yeah, I think that's crack [08:39] wallyworld_, ^ -- what about instancepoller? [08:40] wallyworld_, it doesn't workin the local provider but it should do everywhere else [08:42] rebooting [09:05] fwereade: commented on the MP [09:09] dimitern: I have a problem with MAAS on VirtualBox [09:09] LXC from MAAS node can not get IP address from DHCP on MAAS cluster controller, because DHCP Request is sent w/o 'broadcast' flag and unicast DHCP Offer is not delivered inside of VirtualBox. [09:09] May be the problem related to VirtualBox, but I suggest to run DHCPD on MAAS with 'always-broadcast' option. [09:09] Does it make sense? [09:10] vladk, I have very little knowledge of setting up dhcp internals and details - it seems you might be on a right track there [09:11] vladk: #maas would be the place to bring it up, I think [09:13] dimitern: LXC container inside MAAS choose their MACs automatically and may even have different MACs on each lxc-start. [09:13] I offer to add some option to 'juju add-machine' to specify either MAC of LXC container or CIDR+GW. [09:14] vladk, dimitern: I really don't want MAC addresses leaking into the UI, but I'm just fine with uspicking one and sticking to it [09:14] s/uspicking/us picking/ [09:18] dimitern, but in this case it will be impossible to have a predictable external IP address bound to LXC [09:24] fwereade, rogpeppe quick question for one of you (sorry if I'm repeating - I got disconnected soon after asking the first time so I don't think I got through)... [09:24] rogpeppe: got time for a hangout? I need help [09:25] I've started a branch for getting the current user set as the service owner when juju deploy is called. The wip mp is here: https://codereview.appspot.com/83060049/. I've added a test: https://codereview.appspot.com/83060049/patch/1/10004 but what I should do in the test is actually change the current user (as defined in the .jenv file) but I couldn't work out how to do it [09:25] thumper: sure [09:25] rogpeppe: thanks https://plus.google.com/hangouts/_/7acpj31hohv230rlfvf7d3jqec?hl=en [09:25] mattyw: are you around for a while? [09:25] rogpeppe, I am yes - no hurry [09:26] mattyw: cool, will answer after talking with thumper [09:27] mattyw, I don't think that owner should be in the args over the API -- it's implicit in the connection [09:28] fwereade, I guessed that would be the case but I couldn't see it [09:28] mattyw, I think you just need to get the current tag off the auth object and pass that through [09:28] infact - idiot - of course it is [09:28] will do [09:28] mattyw, it may not currently be exposed but it's definitely there [09:28] mattyw, cheers [09:28] sorry - should have seen that [09:28] and how should I switch the current env in the tests? [09:29] mattyw, that should then leave cmd/juju and state/api untouched, I think [09:29] mattyw, so long as the apiserver tests check that it works with a different connected user, I don't think yu need to touch those other packages at all [09:30] ok [09:34] fwereade: we can haz HA azure now - all merged [09:35] axw, w00t! [09:35] dimitern: is any way in MAAS api to understand what MAC address is of PRIMARY interface. you may suggest the first one, but if I remove and recreate the first MAC, it become the last one [09:35] * axw tests that it works once more to be sure [09:38] axw, much appreciated :) [09:39] fwereade: can I have an ear for a sec? [09:39] * fwereade listens [09:39] fwereade: arm vs armhf [09:39] the tool building process is generating simplestreams with armhf and tarballs with armhf in them [09:39] but AFAICT the binaries inside them have "arm" [09:39] jam, yeah, I pinged davecheney, I don't understand his objection to calling it armhf inside juju [09:40] fwereade: all *I* care about is that they are consistent, and I feel it is easier to fix juju-core at this stage than fixing stuff outside of juju-core [09:40] if only because we have a patch :) [09:40] jam, and (from my perspective at least) it's a hell of a lot easier to fix juju than to fix the tool-building [09:40] jam, yeah [09:40] jam great minds ;) [09:41] davechen1y, davecheney: ping again? [09:45] uh oh, spaghetti oh. I broke something [09:48] axw, change your name to max power and try again [09:48] lol [09:50] fwereade: ack [09:52] vladk, maas doesn't enforce ordering of macs - what you enter is what you can use to link to networks [09:54] fwereade: I'm thinking we need james' patch to land on the 1.18 branch as well, right? [09:54] frankban: I believe LXC has an empty scope because we can't tell whether it is public or private. [09:55] It is just the address we got from eth0 [09:55] on Clouds that is actually a private address [09:55] for LXC it is probably the "most public address we can get" [09:55] though still since it is on LXCBR0 it is probably a private address. [09:56] there's no real public at all for the local provider, and lxc on other clouds will give you a machine-local address from that [09:56] jam: yeah, I am just concerned about client logic. From the client perspective, an internal eth0 address of an LXC in a local env could be also be considered "public", meaning reachable. [09:56] mgz: well for MaaS it is a public-ish address, but for where we end up with EC2 it would be private. [09:57] right, maas is special [09:57] frankban: so I'm pretty sure it means "we need to do more coding to parameterize this on where the LXC is running" [09:57] which is... Is it worth it for now? [09:57] jam: I just want to make sure the logic I described in the bug is what must be implemented by clients [09:57] davechen1y, what's the objection to calling it armhf inside juju? it feels like the cleanest way to get consistency [09:58] fwereade: sorry, was at soccer, missed your ping [09:58] wallyworld_, I forget what it was about now ;p [09:58] goo :-) [09:58] good [09:58] wallyworld_, oh, yeah, instancepoller [09:59] sorry brb [09:59] wallyworld_: so there is a bug that frankban noticed, and the question about what to do with "unknown" networks. [10:00] (appending to a list that we preseed with length) [10:00] trivially fixed with "addresses = ... 0, len(merged))" [10:00] fwereade: quick one when you have a moment https://codereview.appspot.com/83940043 [10:00] fwereade: ok, i'll look. btu all i'm doing is exposing the exisitng address getters on unit [10:00] mgz: I think frankban's logic for "use the first public address, otherwise fall back to the first unknown address" is the logic we're going to have to go with for now. [10:00] it will still be wrong in the ec2 case [10:00] so if they're wrong then it's also wrong elsewhere [10:01] but hopefully by then we can do the "oh private address X maps back to public address Y" and shove that in the information for the container [10:01] which means the logic will still be correct. [10:01] wallyworld_: the other code didn't do "make" [10:01] fwereade, jam, rogpeppe, meeting? [10:01] fwereade: integration testing fail - actually works with this change [10:01] jam: why is it wrong in the ec2 case? [10:01] it just started with the nil slice and appended to it. [10:02] frankban: because if you have an LXC on EC2 it can *only* ever see its machine private addres. [10:02] you have to go back to the provider [10:02] and lookup "what is the actual public address assigned to the outer machine that matches the private address inside the LXC" [10:02] jam: i needed to return an empty slice rather than a nil slice cause that's what the mega watchr wants [10:02] wallyworld_: that is fine, the problem is that you do "make([]foo, len(merged)" and then append [10:02] just change that to [10:03] make([]foo, 0, len(merged)) [10:03] I have to switch machines, [10:03] oh ok [10:04] jam: so, is it ok to use the logic I described in the bug (public + fallback)? Do you think that work for all providers? [10:06] frankban: that works as well as we can do right now [10:06] when we fix EC2 to know about how to do that mapping, we can put a public address for those LXC containers [10:07] and have it continue to work [10:08] jam1: ok, so, when that branch lands, we can go ahead with quickstart and the GUI, using that logic to retrieve the reachable address [10:08] jam1: sounds ok? [10:08] frankban: sgtm [10:16] hazmat, do you have the paste that davecheney linked with the casts in the watcher code? [10:16] fwereade, i don't.. i have the tracebacks.. [10:17] hazmat, sorry, it was that to which I referred [10:18] fwereade, here's the watcher one.. http://pastebin.ubuntu.com/7198067/ [10:18] i'll forward the rest via emial [10:33] good thing I am not joining near a major release.. ah, wait :| [10:33] :p [11:06] thumper: yea saucy works with kvm [11:14] fwereade: I looked at https://codereview.appspot.com/81110043/ (manifestDeployer) [11:15] it seems ok, though the SetAbortWait is confusing for me. [11:16] jam1, I tried to make it clear, but clearly failed [11:16] fwereade: well, I sort of understand what it is doing, but it took a bit because the function has side effects *and* returns an object. [11:16] It also wasn't clear why we needed it. [11:17] fwereade: is that just part of the interface? [11:17] it seems like without the wait form, the other function can't return at all. [11:17] ah, it just always ignored abort before [11:18] jam1, it's really just a way of checking that the abort chan is passed through to the bundle reader [11:18] jam1, better ways of doing it greatly appreciated :) [11:18] fwereade: sure, I actually see the side effect, but it isn't very straight forward [11:18] so *if* you called SetWaitAbort [11:18] then we will wait on <-abort [11:18] jam1, because it certainly felt unhelpfully convoluted to me [11:18] otherwise we just always fall through with the jam, yeah, exactly [11:19] doing the "create one and close it" looks strange. [11:19] I'll propose something [11:19] jam1, cool, thanks [11:26] rogpeppe: you there? the hangout is awfully quiet :-) [11:26] fwereade: suggestion sent [11:26] you certainly don't have to pick it up, and it is more verbose [11:26] but I feel it is a bit easier to read [11:26] voidspace: I'm sorry, you're not allowed to go on vacation, ever. [11:26] jam1: ah, ok - that must be a new policy... [11:27] voidspace: certainly not to something so seedy as a place to Con people with Pie [11:27] jam1: what about if I promise to proselatyse Juju? [11:27] (forgive the spelling) [11:27] If I'm going to be conned, I hope I get pie [11:27] voidspace: approved! [11:27] jam1: thanks :-) [11:27] * natefinch brings fork to pycon, is disappointed. [11:28] natefinch: silly man :-) [11:28] natefinch: should have brought a spork [11:28] fwereade: is there actually anything blocking https://code.launchpad.net/~fwereade/juju-core/filetesting-package/+merge/212811 other than rogpeppe didn't actually give it an LGTM ? [11:29] it looks like you did all of his suggestions. [11:29] so yesterday, whilst attempting to print, I discovered that Ubuntu has a keyboard shortcut for "change my resolution, futz with my screen layout and disable the mouse" [11:29] *really* useful [11:29] ah, i had a couple of draft comments [11:29] all you have to do is press instead of [11:31] there will be enough juju people at pycon that we could do an openspace [11:31] though the last thing I need is one more thing to coordinate [11:31] voidspace: mess-with-displays enough that Pidgin can't find its window anymore :) [11:32] maybe tvansteenburgh will do it :) [11:32] voidspace: apparently it is "detect displays" which changes it from separate to mirrored [11:32] so if you have different sized screens (as I do) it gets messed up [11:32] nice [11:32] it also messes up the "my second screen is on the left" [11:32] settings [11:32] http://askubuntu.com/questions/20113/how-to-stop-mod4-p-from-switching-the-display and http://askubuntu.com/questions/68463/how-to-disable-global-super-p-shortcut [11:32] voidspace: you're not the only one that dislikes it [11:32] yeah, apparently it's a *feature* [11:33] I'm actually impressed with trusty multi monitor support in general though [11:33] I'm guessing this is "I connected to a Projector, help!" [11:33] wow, yeah, that's terrible [11:33] "just works" (mostly) [11:33] it would be fine if it just toggled from mirror display to [11:33] so having a command sequence to switch to Mirrored displays isn't that uncommon [11:33]