[00:08] top - 11:08:40 up 16 days, 1:34, 9 users, load average: 5.09, 2.39, 1.02 [00:27] davecheney: is it fixed already? [00:44] i have a partial fix that fixes the link error during compilation [00:45] i'm going (20 mins so far ...) [00:45] doing a [00:45] full test run to see where we're at [00:45] the fix isn't suitable for upstream, but it gives me a starting point [00:45] thumper: http://paste.ubuntu.com/7154127/ [00:45] * thumper nods [00:46] that isn't looking to bad [00:46] the fix is basically to reverse the order of the .a files presented to the linker [00:46] but a more sophisticated fix will probably be needed [00:46] looks like an ordering issue for those failures [00:47] yeah, i suspect that once the tests build [00:47] the fixes to the failed test cases will be fairly straight forward [00:54] wallyworld: https://codereview.appspot.com/22320043/ [00:54] wallyworld: resurrecting a 4 month old branch [00:54] ok [00:54] because I want it for my plugin [00:54] davecheney: agreed [01:09] thumper: email sent [01:09] davecheney: oh... kay... [01:09] as i said in there, i'm working on a less horrible fix [01:10] thumper: you'll see, don't panic [01:13] * thumper isn't panicing [01:31] thumper: standup [01:31] thumper: wallyworld axw alexisb waigani are al waiting for you to grace us with your presence [02:03] thumper: did you say you don't know what the minimum version of lxc we require is [02:03] ? [02:03] axw: correct... [02:03] ok [02:03] you should ask hallyn [02:04] mk [02:04] really, it is the one that makes lxc-start move the container into the STARTING state [02:21] wallyworld: so the user has to resolve provisioning errors, the provisioner won't just retry? [02:21] like resolving charm errors? [02:21] axw: for now yes. the provisioner can't retry without extra logic, since if it is a rate limit issue, it would just compound the problem [02:22] we need to add extra logic to be smart about how and when auto retry happens [02:22] yeah okay. it'd need exponential backoff and so on [02:22] thanks [02:22] np [02:22] so first cut is just to provide the machinery and a manual trigger [02:33] wallyworld: does it not make sense to have the machine watcher get notified when the machine's status changes? [02:34] wallyworld: which would make the MachinesWithTransientErrors API redundant I think [02:35] axw: sorta. but i wanted to keep the implementations separate because 1. we are close to release and i didn't want to dick with that logic too much, 2. we may well want to have separate queues / processing requirements for handling machines with errors [02:36] ok [02:36] it's all a a bit rushed sadly [02:37] well it's not being used, so if it turns out it can be done without the API then we can just not use it. but at least it'll be there if we do need it [02:37] perhaps we can revert to using the normal machine watcher, but i didn't want to risk breaking existing stuff what was expecting certain semantics for that watcher [02:37] yep fair call [02:37] axw: it's used in the branch i'm about to propose [02:37] wallyworld: ok [02:38] by the provisioner task [02:38] it's internal anyway [02:38] but we can easily change that in future if we want to do it differently as it's an internal api [02:38] yrp [02:38] the joys of trying to get stuff into a release [02:39] maybe i'm being overly cautious, but i do like having this stuff go in as an orthogonal implementation to what's there [02:39] to reduce risk [02:44] thumper: https://codereview.appspot.com/80300043/ [02:44] fix proposed for the gccgo link order bug [02:44] now to make some spaghetti and see if I can roll a custom version of gccgo-go with that fix [02:45] davecheney: ack [02:47] davecheney: seems reasonable [02:47] davecheney: except for this 'a.p.' malarky [02:47] :-) [03:01] thumper: good question about the race condition - I honestly don't know, but I think you're right and it could race. I'll use locals for those vars updated in the goroutine [03:15] axw: ok [03:16] wallyworld: got a minute for a hangout? [03:16] sure [03:18] thumper: you setting one up or me? [03:19] I'll do it [03:19] typical of me [03:19] ask a question... [03:19] then "oh, look, shiney" [03:19] https://plus.google.com/hangouts/_/72cpjm79q3a5ke7h17c6in5qas?hl=en [04:03] axw: what are you looking at now? [04:03] axw: because two things leap to mind... [04:03] thumper: was going to do some more azure stuff, but I can shelve that again [04:04] axw: the more urgent one is "warn about empty .jenv file" as opposed to being too dumb [04:04] axw: second one is 'juju unset-env' [04:04] thumper: warn about empty file and then act as if it didn't exist? [04:05] umm... I'm not sure [04:05] I'm tempted to just tell them it is there, and stop [04:05] the whole reason it is created empty is to grab the namespace [04:05] it is only weird edge cases where it would be empty and there [04:06] yeah true. I guess we'll need a utility to help people clean stuff up when we actually do have the config hosted [04:06] * thumper nods [04:06] I'll just have it exit for now with a message pointing at the location [04:07] I think that is reasonable [04:08] it should (fingers crossed) no longer happen [04:16] thumper: was just looking at IRC logs and looks like mgz is still working on it. I'll do unset-env for now, if you have no objections [04:16] ok [04:20] axw: I've emailed fwereade and jam about it [04:20] thumper: okey dokey [04:31] derp [04:31] thumper: I mucked up the Infof in cmd/juju/common.go [04:31] it says bootstrap fails even when it doesn't [04:31] hmm... [04:31] * axw fixes [04:32] oops [04:34] davecheney, ping? [04:41] wallyworld: quick review please? https://codereview.appspot.com/80330043 [04:41] sure [04:41] wallyworld: thanks for updating the client thing - will take another look in a bit [04:41] np, i was glad to get a 2nd opinion [04:46] mattyw: pong [04:57] wallyworld: reviewed again - sorry, missed something before [04:57] sure, looking [06:38] wallyworld: I can't think of a way to solve the attribute error thing without a tonne of extra work, so if you don't mind I'll land this and we can iterate later [06:38] at least then the API will be in === vladk|offline is now known as vladk === vladk is now known as vladk|offline [06:58] jam: the go.crypto/ssh ProxyCommand doesn't shell out to "ssh", it shells out to "juju ssh" [06:58] so effectively calling back into the same go.crypto/ssh code [07:17] axw: ok [07:49] mornin' all [09:03] morning rogpeppe [09:06] jam: morning [09:07] jam: are you around for a quick chat? [09:31] morning all [09:31] voidspace: hiya === vladk|offline is now known as vladk [10:03] * fwereade is just about to do some reviews, but would like it if someone would take a quick look at https://codereview.appspot.com/80540043/ [10:05] good morning eeryone [10:09] fwereade: would you be available for a quick chat? [10:12] fwereade: looks good altough you might want to have opinion from a more expert dev [10:22] fwereade: what problems did the tests find? [10:22] rogpeppe, sure [10:23] rogpeppe, just a couple of things like Checking an error and not returning [10:23] rogpeppe, so a subsequent look at a FileInfo would panic [10:23] rogpeppe, I had it in one place and not another [10:23] rogpeppe, dirs weren't checking perms or something? [10:24] https://plus.google.com/hangouts/_/canonical.com/juju-core?authuser=1 [10:33] if someone has a moment, I would appreciate this to be looked at https://codereview.appspot.com/80560043/patch/1/10002 [10:36] perrito666, looking [10:38] perrito666, reviewed [10:38] dimitern: thank yu [10:46] perrito666, i forgot to mention that this won't work on windows, but that's ok I guess [10:47] perrito666, or perhaps it will.. not sure (testing needed) [10:47] dimitern: utils ssh lacks an implementation for windows? (I guess that would depend on cygwin or smth on that area) [10:48] perrito666, no, but on windows because we're not using openssh's scp no extra arguments can be passed to the command (not sure if that's relevant) [10:48] call is a bit bad for me this morning... [11:03] dimitern: fixed [11:05] perrito666, cheers [11:05] fwereade, rogpeppe, mgz, https://codereview.appspot.com/76910044/ is ready to land - want a last look? [11:08] dimitern: sure [11:17] anyone else wants to take a look at? https://codereview.appspot.com/80560043/patch/1/10002 it is fairly short [11:21] dimitern: extra tx if you paste the links here [11:22] :) different computer [11:28] perrito666, sure: https://wiki.edubuntu.org/SecurityTeam/TestingMAAS http://marcoceppi.com/2012/05/juju-maas-virtualbox/ [11:29] dimitern: sory, I use another machine to be able to hold hangouts that long [11:29] perrito666, I should start following your lead in that soon :) [11:30] I use a mac, which has hardware accel for that, yet it gets kind of hot, I guess I should get an android tablet or similar [11:40] marcoceppi, nice article btw :) ^^ [11:41] dimitern: thanks, it's too bad virtualbox and maas don't play together nicely [11:41] marcoceppi, yeah, I tried myself twice before deploying maas + vbox (unsuccessful) [11:42] dimitern: it's much better to use maas + qemu/kvm [11:43] marcoceppi, I tried that yesterday and almost got it working - pxe boot working (with both cluster controller vm and a blank node using single NIC bridged to virbr0), but after commissioning the outbound networking on the new node is not working [11:43] brb [11:44] dimitern: how is your maas setup? does it own dchp, etc? [11:45] marcoceppi, yeah, it has 2 nic (lo and eth0 - the bridged one; also managed in maas for dhcp + dns) [11:45] marcoceppi, had to set the default domain config from local to "", because it seems avahi boot is not working [11:45] dimitern: is ip_forwarding setup properly with the masquard rules, etc? [11:46] dimitern: it took me a long time to get maas-master networking configured correctly [11:46] marcoceppi, and I also set the iptables rules according to the TestingMAAS article [11:46] * marcoceppi reads article [11:46] dimitern: there are some scripts that might help you setup the networking, actually [11:46] marcoceppi, ip_fwd on cluster vm or on the host (the latter is on) [11:46] latter [11:47] marcoceppi, so pxe boot works, but as the new node starts apt-getting stuff it fails to resolve archive.ubuntu.com urls [11:47] the fact that this is saying cobbler convinces me it's out of date. I don't think maas uses cobbler anymore [11:48] yeah, it doesn't [11:48] but I used the alternative (installing maas-dhcp and configuring it through the web ui) [11:48] dimitern: well, there should be a squid-deb-proxy running on maas master, you should be able to set the apt-mirror url in maas to piont to the maas master's internal network ip [11:48] but that won't solve needed to resolve addresses outside of the machine [11:49] marcoceppi, so just install squid-deb-proxy on cluster controller, add a new archive from the web ui? [11:49] dimitern: also, is it not connecting to the outside or is it a dns lookup issue? [11:50] marcoceppi, not sure if it's just dns resolving issue or outbound networking in general, need to look into it [11:50] dimitern: if you can get to a node in maas, you should be able to ping a network address, and if you can, you'll just need to set up the proper DNS forwarding route for the maas dns server [11:50] marcoceppi, the frustrating thing is that the guide wasn't specific on how to setup the new nodes' networking when creating the vm [11:51] dimitern: just DHCP should be sufficent [11:51] marcoceppi, but following yours, i decided all vms have nic bridged to the virbr0 [11:51] dimitern: I had to set the external DNS server for my nodes to resolve archive.ubuntu [11:51] marcoceppi, so the node appears and stays in commissioning (in some cases even ready), but the fqdn is something like ";; cannot resolve xyz ;;" [11:52] wwitzel3, with libvirt? [11:52] well as an alternative, you can use virtulbox in nat mode and then do a reverse ssh tunnel from within the maas machine [11:52] the master that is [11:53] dimitern: yes, in MAAS/settings [11:53] dimitern: there is an Upstream DNS [11:53] wwitzel3, ah, ok, I'll try that [11:53] dimitern: also, I set the default domain to some fake internal domain, like my.maas and added the maas controller to my resolv.conf as the nameserver for that domain [11:54] wwitzel3, cheers [11:56] rogpeppe: want to rejoin the hangout? [11:56] rogpeppe: it's empty [11:56] well, except for me... [11:57] voidspace: will do [11:59] dimitern: could you ack the chanes, just to be sure? https://codereview.appspot.com/80560043/patch/20001/30002 [11:59] rogpeppe: https://codereview.appspot.com/72500043/ [11:59] * perrito666 has the button hanging over the submit [12:00] natefinch: would you be able to join us in the hangout? [12:00] natefinch: (https://plus.google.com/hangouts/_/canonical.com/juju-core?v=1395698400&authuser=1) [12:04] morning all [12:09] perrito666, looking [12:10] perrito666, reviewed, just one minor thing [12:10] wwitzel3: was it you that asked me how to make gofmt go trough vim? [12:10] fwereade, rogpeppe, mgz, poke about https://codereview.appspot.com/76910044/ [12:11] perrito666: well I can make it automatically run, I just can't get it to highlight the lines with errors [12:11] wwitzel3: ah I got it to jut fix the file :p [12:11] which is faster [12:11] perrito666: well mine does that too as long as there aren't any synxtax errors [12:12] perrito666: but if there are syntax errors, it just outputs "go fmt returned an error" and doesn't tell me what line, so I have to drop to the command line and run go fmt to see where the error actually is. [12:13] wwitzel3: i think I can add that to fmt.vim :) ill try on the weekend [12:14] dimitern: going over it now [12:20] dimitern: lgtm [12:20] mgz, ta! [12:21] dimitern: fixed, merging then :) [12:21] dimitern: thank you very much [12:21] perrito666, go for it [12:26] mm, apparently has a conflict with my other branch, how should I fix that? [12:29] pull trunk, merge trunk in to the branch you're trying to land, [12:30] ack [12:30] perrito666, switch to trunk, pull it in, then switch to your other branch and do "bzr merge :parent" [12:30] resolve the conflicts, commit, push, then tell the bot to try again [12:30] :) [12:30] rogpeppe, do you have a minute? [12:31] rogpeppe, I'm wondering why this fails http://play.golang.org/p/VPC3P2nGYQ ? [12:31] dimitern: gimme 5 mins [12:31] rogpeppe, sure [12:32] perrito666, ah, one thing you might need - after merging, resolving, committing and pushing, you'll probably need to add a comment on the MP to self-approve the changes [12:38] after merging and committing I du another propose right? [12:41] just the landing step [12:41] no really need to lbox [12:41] flip the approved on the mp is all you really need [12:41] ok, I just bzr push then go to the mp in launchpad and hit approved [12:41] yup [12:42] ok, makes sense [12:46] jam: fwereade: i added support for multiple machines also [12:57] http://en.wikipedia.org/wiki/Toast_sandwich [13:03] rogpeppe, I'm really banging my head here on this one [13:03] dimitern: ok, looking [13:04] dimitern: what does jc.DeepEquals print? [13:04] rogpeppe, mismatch at ["services"]["network-service"]: validity mismatch; obtained map[interface {}]interface {}{"charm":"cs:quantal/dummy-1", "exposed":false, "networks":map[interface {}]interface {}{"disabled":[]interface {}{"net3", "net4"}, "enabled":[]interface {}{"net1", "net2"}}}; expected [13:04] rogpeppe, expected is actually not nil, but IsValid == false [13:05] rogpeppe, here's the full output with some added logging: http://paste.ubuntu.com/7156594/ [13:05] dimitern: the second one has "networks-service" not "network-service" [13:05] dimitern: http://play.golang.org/p/BWtpWAaiEf [13:06] rogpeppe, ah! [13:06] rogpeppe, tyvm [13:06] dimitern: np. i've been that same situation *so* many times! [13:06] dimitern: it was worse without jc.DeepEquals... [13:07] rogpeppe, indeed :) at least it tell you at which point it fails [13:07] darn plurals :) [13:17] fwereade, mgz, updated https://codereview.appspot.com/76400049/ - now we show networks for services, not machines (ignore the branch name, don't know how to rename it) [13:17] fwereade, I'd especially like your comments due to your not lgtm [13:24] wwitzel3, rogpeppe: back in action, going to do some cleanup of the proposal branch [13:25] voidspace: http://paste.ubuntu.com/7156661/ [13:26] dimitern: are you free for a quick chat? [13:26] natefinch: ok, cool. i'm afraid i haven't got any further with my review [13:26] rogpeppe, sure [13:26] rogpeppe: no problem [13:26] rogpeppe, mumble? [13:27] dimitern: https://plus.google.com/hangouts/_/canonical.com/juju-core?v=1395698400&authuser=1 ? [13:27] dimitern: (i've forgotten how to use mumble...) [13:27] rogpeppe, ok [13:28] sinzui: hey, with the last landing you should be able to run the test for https://bugs.launchpad.net/juju-core/+bug/1291022 [13:28] <_mup_> Bug #1291022: Cannot restore a state-server on ec2 and openstack

[13:29] perrito666, fab, the test is being run with every revision. I will look for a pass with r2488 [13:33] sinzui: plz keep me posted, I ran it successfully by integrating my branches, but ymmv [13:33] HA! I love this [13:33] https://github.com/bradfitz/iter [13:39] bodie_: I just saw that the other day, pretty great [13:41] natefinch: can I call? [13:41] wwitzel3: sure [14:17] perrito666: I ended up using the syntastic bundle for vim and it does proper line highlighting or errors [14:17] of [14:17] :) that is very good news [14:18] perrito666: yeah, it hilights the offending line and provideds the message [14:20] looking for a quick leg up understanding /state/api/apiclient.go struct State [14:20] if anyone has a sec, pm? [14:21] thanks in advance :) [14:30] sigh, *of course* mongodb charm doesn't implement any nagios relations [14:31] alexisb: ping [14:31] oops, that was meant for #juju === hatch__ is now known as hatch [14:35] bodie_, api.State is the client-facing api facade, i.e. it emulates state.State, but through the api (it contains the underlying api connection as well) [14:36] yeah, I saw that -- wasn't aware that it is the client frontend for State.state though. (useful!) [14:37] I was going on the assumption that the API is an http endpoint but I'm thinking that is wrong. correct? [14:42] is JUJU_HOME ~user or ~user/.juju ? [14:42] it's the latter [14:46] bodie_, strictly speaking it's a front-end of state.State as seen through the API only [14:47] by API you mean the actual programmatic API, right? Not the HTTP API [14:48] I was under the impression the gui communicates with the program via HTTP API but now I'm recalling hearing that it uses websockets [14:48] bodie_, the API endpoint itself is either https-based or websocket-based (the former is only used for uploading charms and fetching stuff from charms, the latter exposes the rest of the api) [14:48] I see [14:48] so if I want to add verbs to the API, I don't need to define them in the HTTPS frontend [14:49] bodie_, if by verbs you mean api methods that can be used by clients (cli, gui, deployer, etc.), then you need to add them to api.Client as methods which call their apiserver.Client counterparts [14:50] fwereade, rogpeppe, mgz, juju deploy --networks/--no-networks support, PTAL https://codereview.appspot.com/80600044/ [14:51] OK, I think that makes sense :) we had talked about doing it via hooks but I'm not positive that's the right way [14:51] or at least, it gives me something to go on [14:51] thanks for the help [14:51] bodie_, what do you have in mind? === vladk is now known as vladk|offline [14:52] bodie_, things hooks can do are very limited compared to the api [14:52] service actions such as mysql snapshot [14:53] there are a few of us who got roped in from the outside, trying to put it together over the next few weeks [14:53] bodie_, what will that do? please, expand a bit. [14:53] I think it's either an experiment to see how frustrating it is for newbies to dive into the deep end unguided, or we're on reality tv [14:53] but here I am haha [14:54] :) [14:54] basically, a few Action words added to the command interface [14:54] juju do, juju queue, juju cancel, and a couple of others [14:54] the flow is like [14:55] the api code can be daunting at first, but it's really straightforward once you grok the way things pass through client->api->websocket->apiserver->state [14:55] yeah, I'm kind of briding the first bit of that right now [14:55] s/briding/bridging/ [14:55] ah, so juju can be added either in cmd/juju or as plugins named "juju-" [14:56] the flow is like -- juju do [14:56] sounds like you could use a plugin that calls juju run internally [14:56] yeah, I think I get how to add a command now, I just want to make sure I understand what needs to happen once it gets triggered [14:56] hm [14:57] well juju run gets called by the command line already right? regardless of argument [14:57] so we would just add a file to /cmd/juju/ [14:57] juju run allows you to run scripts remotely on a unit or all machines inside the same context (env vars etc.) that a hook runs [14:57] okay [14:57] yeah, that's what I've been digging through over the last two days [14:58] either in cmd/juju/mycmd.go (for example) or as an executable juju-mycmd in $PATH - both will give you the ability to do "juju mycmd" [14:59] right... and then I'll have to --upload-tools to test it (right?) [14:59] no [14:59] ok [14:59] so let's say I want to have arguments to that command trigger stuff on the remote [14:59] e.g. mysql dump [15:00] the "run" on the node gets triggered by uh (checks notes0 [15:00] ) [15:00] --upload-tools is needed in general to have the latest jujud binaries in the cloud; for client-side stuff (on your machine) you just need to rebuild cmd/juju or whatever other plugin/script you're using [15:00] apiClient in /cmd/juju/run.go Run function [15:00] triggers execution on the remote via rpc [15:01] or -- triggers the remote function rather [15:01] which is in /cmd/juju [15:01] the simplest way to do that is to first check if you can simulate what you need via juju run (or other commands) [15:01] implementing RunCommand [15:01] * bodie_ reads [15:01] if it is sufficient, you can put them in a script and use that [15:02] juju run is quite versatile - check its help [15:02] ok [15:05] bodie_, juju ssh and juju scp are also useful to run stuff remotely or copy files across local/remote machines (but these do not give you hook contexts like run) [15:06] OK, so maybe what we're actually doing is trying to make a Run alias that knows how to do certain things based on what's in the respective charm [15:06] rogpeppe, got a moment? [15:06] mattyw: sure [15:06] mattyw: hangout? [15:06] rogpeppe, we can do - irc is probably fine for this [15:07] unless you want to hear my voice ;) [15:07] mattyw: that's cool too, though i ache for the melodious timbre of your voice, obviously [15:08] rogpeppe, you're only human [15:08] how poetic [15:08] rogpeppe, this review from fwereade https://codereview.appspot.com/75600044/diff/20001/state/user.go#newcode32 [15:08] admin user being allowed an empty password [15:09] bodie_, i wouldn't call it "alias" (in CLI terms only actual commands can have aliases), but I got you - it's just a script that calls one or more times juju run and can take arguments [15:09] I'm sure there was a reason - but can't remember what/ if there was - do you have any wisdom to impart? [15:10] mattyw: i'm not really sure why we consider "admin" special in any way actually [15:10] mattyw: apart from for mongo access, which isn't relevant to this [15:11] mattyw: i'd prefer to avoid any name special-casing [15:11] rogpeppe, I'm starting to wonder if I snuck that code in so I didn't have to change loads of tests [15:11] mattyw: ha ha [15:12] rogpeppe, I'll try removing it and see what happens [15:12] rogpeppe, if you don't think it should be there either then I'll just have to bullet the bite [15:12] mattyw: looks like there are in fact a fair number of tests that call AddUser with an empty password [15:12] I looking for reviews on https://codereview.appspot.com/76400049/ and https://codereview.appspot.com/80600044/ please [15:12] mattyw: why don't we just allow the empty password? [15:13] rogpeppe, I've changed a number of tests in that cl aready - you mean for all users? [15:13] mattyw: yeah [15:14] rogpeppe, I suppose so that makes sense [15:14] mattyw: given that a user can change a password to "password" or "a", i don't really see what preventing a blank password gives us [15:14] rogpeppe, very true [15:15] rogpeppe, fwereade I'm going to change the mp to allow the empty password for all users for now then if that's ok? [15:19] rogpeppe, mattyw: that sounds fair to me [15:20] rogpeppe, mattyw: state is probably not the place for that sort of policy enforcement anyway [15:20] fwereade: +1 [15:22] fwereade, rogpeppe where would be? (out of interest) [15:22] mattyw: if we wanted to, probably in the API [15:22] mattyw, rogpeppe: API layer IMO [15:23] nice when we concur :) [15:23] fwereade, got some time to review my CLs? [15:41] dimitern, first one LGTM [15:41] fwereade, \o/ [15:46] is there a way I can get a list of the charms loaded into the juju env? (uploaded and not necessarily deployed) [15:46] maybe poking a db or something? [15:47] rogpeppe: why is go test -c so dang slow? [15:47] hatch, you could check the storage [15:47] natefinch: it's quicker if you do go test -i first [15:48] dimitern cool thanks [15:48] dimitern, couple of thoughts on the second, let me know when you've read it [15:49] fwereade, cheers [15:50] rogpeppe: that doesn't seem to make a noticeable difference. It's just, I can compile in .2 seconds, but test -c takes 1.2-1.5 seconds for the same package [15:51] rogpeppe: the reason I care is that I want to be able to find compile errors in tests when I make changes to code, and not have to actually run the tests to check. I can recursively call go test -c in all subdirectories, but it takes forever [15:51] fwereade, right now we can't verify if the provider supports networks, but we can do it later if we have a way to introspect the provider capabilities reliably === vladk|offline is now known as vladk [17:55] alexisb: here [17:55] err [17:55] sorry [17:55] allenap: here [17:55] mgz: Can I borrow your brain? roaksoax (or one of his team) is seeing http://paste.ubuntu.com/7153235/ with Juju 1.5 but not 1.4. I can see what’s going on in MAAS, but I can’t figure out what has happened to get to this point. [17:56]