[00:09] wallyworld: ok, will fixup the er's .. I can probably do it as two branches, I will look in to that. As for the results, I already define a params.go to hold the result, so I can just store the value that I am marshalling in there instead of manually marshalling it and the RPC layer will do that for me? [00:10] wwitzel3: i believe so yes [00:11] the gui makes other api calls which just return the structs [00:12] wwitzel3: don't waste time on 2 branches if it's too difficult to split out. bigger branches can be harder to land though since one bit might be ok but other bits may need fixes and the ok bit can't land separately [00:14] wwitzel3: also, don't forget to add SupportedArchitectures to the results [00:14] wallyworld: I actually removed it, since it is send along with each instanceType [00:14] sent [00:14] we also need to handle that some providers can not supply InstanceTypes but do have SupportedAchitectures etc [00:15] oh, let me look [00:16] wwitzel3: i think we need supported arches as an explicit field since otherwise the caller will have to iterate over the instance types and pull out the arches, also not all providers will suport returning instance types but will be able to supply supported arches [00:16] wallyworld: which ones? I didn't see that, it is easy to re-add them as their own top level key in the map [00:17] manual provider for example [00:17] maas also [00:18] the supported arches value will be used to guide the user in creating vlid constraints with arch=blah [00:18] ahh ok [00:18] should I still include it with the instanceType data? [00:18] no, it is a separate concept [00:18] each provider has a SupportedArchitecures() api [00:19] wallyworld: right, I just notice that InstanceType had an Arches field and it was the same type as the return value of SupportedArchitectures() [00:20] wwitzel3: yes, that's right. for ec2 (only i think), instances can run on i386 or amd64. i think for all other providers instance types run on one arch [00:20] so we record for individual instance types what arch(es) that instance type can run on [00:21] wallyworld: ok, so that needs to be specific to the instance, not the provider level SupportedArchitectures() [00:21] no, it's at the provider level [00:22] the supported rches for the provider comes from reading the simplestreams image metadata [00:22] and adding together all the arches listed [00:23] so that the system can disallow arch=foo constraints for arches for which there is no available image [00:23] wallyworld: right and that is different than the supported arches for an instance [00:23] wwitzel3: yes, so when an image is chosen, it needs to be matched up with an instance type and the arches also need to match there too [00:24] wallyworld: right, so really what I am doing now in the API InstanceType.Arches = provider.SupportedArchiectures() is wrong, I need to move supported up to its own field and only populate Arches from the specific provider instance/package/flavor response [00:24] wallyworld: found one leak [00:24] wwitzel3: so when the user says mem=64G arch=i386, the matching finds an i386 image from simplestreams and then looks at all the instance types which can provid 64G which can run on i386 [00:25] wwitzel3: you don't really need all the instance type data - just the valid names [00:25] wwitzel3: the gui only cares about knowing what instance type names are valid [00:25] wallyworld: oh, ok, so I'll just drop it from the results then and move supportedArchiectures up to its own field [00:26] i guess we could return all the instance type data [00:26] not just the names [00:26] but names is all that's needed for now [00:26] and regardless, yes, we do need supported aches as top level field [00:26] wallyworld: sounds good [00:27] wwitzel3: since we have api versioning, let's start with just names, and add stuff later if needed [00:27] wallyworld: also if you have any suggestions of where I should look to create a proper testing env to exercise this code that would be helpful. [00:27] maybe not... [00:27] geez [00:27] wallyworld: well I already have all the other stuff in there, but I can remove it if you prefer [00:27] wwitzel3: i'm wary of yagni [00:28] and the extra testing, maintenance etc for stuff thats not used [00:28] wwitzel3: which bit specifically do you need help testing? the client calls? [00:28] thumper: be with you in a sec [00:29] wallyworld: yes, starting from state/api .. the env I am creating is throwing nil pointer exceptions because it doesn't have the InstanceTypes method or AvailabilityZones methods .. but I've added them to the dummy provider .. [00:30] wwitzel3: i normally add an assertion that a struct implements an interface so i get a compile time error if i haven't got the right methods eg var _ environs.EnvironProvider = (*environProvider)(nil) [00:30] wallyworld: https://github.com/wwitzel3/juju/commit/e1598a5300cd20c4c23c02e09952d1b6136fdad3 here is my broken test if that helps [00:31] the above says that the environProvider struct implements the EnvironProvider interface [00:31] and it won't compile if it doesn't [00:31] we have a bunch of those in the various providers [00:31] wallyworld: ok [00:31] since there's several interfaces that can be implemented [00:32] wwitzel3: dummy provider has these already [00:32] var _ imagemetadata.SupportsCustomSources = (*environ)(nil) [00:32] var _ tools.SupportsCustomSources = (*environ)(nil) [00:32] var _ environs.Environ = (*environ)(nil) [00:33] i can't see trivially what's wrong with your implementation [00:33] wallyworld: ok, thanks, I'll keep poking at it [00:34] wwitzel3: thank you. let me know how you go [00:45] wallyworld: your tip help, added common.EnvironCapabilities like you mentioned and it revealed that the dummy provider wasn't properly implementing the interface [00:45] cool :-) [00:45] wallyworld: I guess now I actually have to write some tests that assert something useful [00:45] bugger :P [00:45] lol yeah [00:46] wallyworld: chat when you have a minute [00:46] thumper: sure, 2 mins [00:51] https://github.com/juju/juju/pull/195 [00:51] i am blocked on this review [00:52] wallyworld: https://github.com/juju/juju/pull/201 [00:52] thumper: https://plus.google.com/hangouts/_/canonical.com/tanzanite-daily [01:05] wallyworld: gonna spend some time today looking into test failures [01:06] clearly something is very broken [01:06] axw: hey, yes please, i was going to ask you to do that. thumper already has one fix landing https://github.com/juju/juju/pull/201 [01:06] cool [01:06] axw: I'm just testing that precise upgrades fix now [01:07] menn0: cool, thanks [01:18] grr, version package [01:28] axw: also, don't forget to backport stuff to the 1.20 branch eg i think the ssh timeout fix was proposed for master only [01:30] wallyworld: should i move version.SeriesVersion so it is only visible when compiled on linux ? [01:30] wallyworld: I've been waiting for them to merge into master first. I guess I'd better do it now before I forget [01:30] axw: np :-) [01:31] davecheney: it's needed on other clients too [01:31] to set up ubuntu workloads using custom image metadata [01:31] wallyworld: that is a problem [01:31] how can I run it on osx ? [01:31] fuckit [01:31] i'll just remove that comment [01:31] davecheney: see my comment? [01:31] yup [01:31] it can run [01:31] i'll just remove my comment on that function and back away [01:32] ;) [01:32] why not just make the trivial change i suggested? [01:32] then it can run on other clients [01:32] looked harder than just backing out a comment I wrote that was a mistake [01:32] my comment was in error [01:32] i have deleted it [01:32] sure, but it seems like there's a legitimate problem there that is trivially fixable [01:32] so lets just fix it [01:33] wallyworld: i'll fix it in a followu [01:33] ok [01:33] the goal is to fix 1.20 [01:33] i didn't think this was broken in 1.20 [01:33] wallyworld: i am only going on the reports I have been given [01:33] the 1.20 branch was forked of master before this change? [01:34] i cannot validate the problem or the fix myself [01:34] i could be wrong [01:34] let me check [01:38] axw: did this make it in to 1.20 as well? https://github.com/juju/juju/pull/188 [01:39] axw: my manual upgrade test worked fine btw [01:39] menn0: sweet. not yet, I've just created a PR and sent it to the bot now [01:40] axw: great, just making sure [01:40] menn0: thanks for testing [01:43] davecheney: 1.20 was forked for trunk right before that os_version change that broke stuff [01:43] so any fixes just need to go to trunk [01:44] axw: once stuff lands in 1.20, mark the bugs as fix committed as well [01:45] i think there are 2 branches for bug 1334273 [01:45] wallyworld: will do [01:45] <_mup_> Bug #1334273: Upgrades of precise localhost fail

[01:45] wallyworld: yep, there's one more coming for 1.20 [01:45] axw: I've tested the 1.18.4 to 1.19.5 upgrade on precise 3 times now and all looks well [01:45] \o/ [01:45] menn0: awesome [01:58] wallyworld: i don't understand [01:59] how can i help? [01:59] wallyworld: dunno, does 1.20 cli work on windows ? [01:59] if so, job done [02:00] it should i think since the landings to trunk which broke things happened after 1.20 was forked [02:00] but [02:00] even on 1.20, if SUpportedSeries is called on bootstrap, it will fail [02:01] wallyworld: the bug i have is 1.20 cli doesn't work on windows [02:01] i cannot validate this muself [02:01] only go on the bug report I have [02:01] and at the moment, things are getting less clear by the second [02:01] hmmm. i don't know about that then. perhpas that bug was raised before the decision was made to branch 1.20 off an earlier rev from master [02:02] we'd need to check with curtis [02:03] sinzui: ping ? [02:05] davecheney: what bug number? [02:07] lp # 1334493 [02:07] <_mup_> Bug #1334493: Cannot compile/exec win client

[02:08] davecheney: that bug is not marked as affecting 1.20 [02:08] it was taken off 1.20 on the 28th [02:08] wallyworld: mate all I can tell you is what is in the bug [02:08] i'm guessing that's when the 1.20 branch was cut off an earlier rev before that bug was introduced [02:08] somethigns broken [02:09] i fixed something [02:09] maybe it helped [02:09] the bug was introduced in https://github.com/juju/juju/pull/95/files [02:09] davecheney: sure, but look at the affected series - 1.20 was removed [02:09] which was 17 days ago [02:10] * thumper has father in law drop in for a quick visit [02:11] thumper: is that a euphamism for something ? [02:12] boom tish [02:13] davecheney: 1.20 was forked before that rev from what i can tell, which is why the bug was changed to no longer affecting it [02:14] 1.20 was forked on the 15th of June ? [02:14] really ? [02:14] no [02:14] last common rev was merging pull reuqest 160 [02:14] i think your date is a bit out [02:15] pr 160 was 5 days ago [02:15] wallyworld: oh [02:15] you are right [02:15] i am sorry [02:15] np [02:15] i was lookig at the first comment on that PR [02:15] not when it finally merged [02:15] so [02:15] ok [02:15] so 1.20 isn't broken and trunk isn't blocked then [02:16] so, i *think* that bug was originally raised as affecting 1.20, and then retracted [02:16] yes, 1.20 is good, trunk not blocked [02:16] phew [02:16] * wallyworld is not 100% sure the issue is fixed on trunk, but seems to be [02:17] with the pr to be merged [02:17] ok, here's the plan [02:17] i'll merge this one [02:18] then i have a followup waiting on it that adds more tests [02:18] ok [02:18] we can keep the fun going there [02:18] deal ? [02:18] deal, thanks :-) [02:18] \o/ [02:18] sorry for confusion [02:18] nope [02:18] it was my fault [02:19] np [02:46] wallyworld: I guess my branch didn't fix the panic [02:46] wallyworld: sorry... [02:46] thumper: hey, thanks for trying. you did a lot of good fixes regardless [03:04] oh fark... [03:04] * thumper had an annoying thought [03:11] menn0: are you around to chat? [03:11] thumper: yep [03:13] thumper: hang out? [03:14] menn0: yeah [03:15] thumper: https://plus.google.com/hangouts/_/gy36zsk3j2lx4ffgfvotsvzxmya [03:25] thumper: did you foresee a charm's resources directory on disk being a flat list of files? [03:26] i'd think we'd want a dir heirarchy [03:26] wallyworld: yes... [03:26] hmmm [03:26] it was strongly suggested that the dir be flat by the time a charm hook is using it [03:27] by whom? our fearless leader? [03:28] thumper: also, there seem to be devel and stable streams. but what about charm revisions? would we have a stream for each revision, or namespace the resources under each stream by revision? [03:35] axw: do my eyes deceive me, your branch landed [03:35] wow, first attempt [04:03] wallyworld: charm revisions are independent of stream revisions [04:03] wallyworld: the full charm version would then be "charm revision - stream - stream revision" tuple [04:03] all mashed together into a string [04:04] that was the idea anyway [04:47] umm [04:47] https://github.com/juju/juju/pull/195 [04:47] the bot ate my merge requst [04:48] looks like our landing instance might have gone down [04:48] it was processing a pr and got interrupted [04:50] weird, mine too [04:50] hmmm, and now the lander says no pull requests to merge [04:50] wallyworld: it thinks it's still processing them [04:51] gotta add a "Build failed: whatever" message [04:51] i'll poke around and see if i can find out where it keeps its in progress queue [04:52] wallyworld: it's dumb, it just checks for "Status: merge request accepted." and then a "Build failed:" [04:52] oh wait, i see what you're saying [04:52] sigh [04:53] I think what we should do is add a link to the Jenkins job, so the the lander can periodically check the status of the jobs it thinks are running [04:53] yep, let's talk to martin about it [04:53] atm we just get a link to the job type, but not the actual run [04:54] until it fails anyway [04:54] so, should i $$merge$$ again ? [04:54] davechen1y: already done [04:54] thanks mate [05:52] can someone kill this job [05:52] http://juju-ci.vapour.ws:8080/job/github-merge-juju/310/console [05:52] it's not going to pass [05:53] gooone [05:54] today is giving me the shits [05:55] yeah :-( [06:07] win12 === vladk|offline is now known as vladk [06:27] wallyworld: https://github.com/juju/juju/pull/205 [06:28] looking [06:29] # github.com/juju/juju/state [06:29] state/action.go:86: too many arguments in call to names.NewActionTag [06:29] state/state.go:1389: tag.UnitTag undefined (type names.ActionTag has no field or method UnitTag) [06:29] state/state.go:1390: tag.Sequence undefined (type names.ActionTag has no field or method Sequence) [06:29] is anyone seeing this on master ? [06:29] yes, i've run godeps [06:29] davechen1y: I was before I ran godeps [06:29] * axw tries again [06:29] lucky(~/src/github.com/juju/juju) % godeps -u dependencies.tsv [06:29] "/home/dfc/src/github.com/juju/names" now at b2e06a0ab1c09f138853d1ef6b11f94ca9f7b675 [06:30] commit 780947ad0e66382af782c65eec6b86796409f0c7 [06:30] Author: Roger Peppe [06:30] Date: Thu Jun 26 15:46:34 2014 +0100 [06:30] use earlier names dependency [06:30] umm [06:30] why ? [06:30] no problems here [06:30] axw: what sha do you have for juju/names ? [06:30] b2e06a0ab1c09f138853d1ef6b11f94ca9f7b675 [06:31] ta [06:31] oh fuck [06:31] godeps is broken [06:31] lucky(~/src/github.com/juju/juju) % godeps -u dependencies.tsv [06:31] "/home/dfc/src/github.com/juju/names" now at b2e06a0ab1c09f138853d1ef6b11f94ca9f7b675 [06:32] lucky(~/src/github.com/juju/names) % git log | head -n2 [06:32] commit d6e9f06b936da18e4feeef3e788bf0dde0cc2d99 [06:32] Merge: 01e6ac7 f52c443 [06:32] greap, godeps doens't work [06:32] great, that's fucking great [06:32] now the new version of godeps [06:32] lies [06:33] rather than telling you it can't find that rev [06:33] hmm. I'm sure it failed for me last time I didn't specify -f [06:33] ಠ╭╮ಠ [06:33] axw: i recently upraded to rog's new version [06:33] imma gonna downgrade again [06:34] davechen1y: yeah I'm on the new version too [06:34] it didn't work for me [06:34] unless there' a new new version [06:34] it kept saying "i've updated" [06:34] but it didn't [06:34] see above === vladk is now known as vladk|offline [06:46] shit . === vladk|offline is now known as vladk [07:02] axw: https://github.com/juju/juju/pull/206 [07:06] https://github.com/juju/loggo/pull/2 [07:12] there are SO MANY races in the apiserver package [07:15] morning all [07:15] davechen1y: sorry went to pick up my daughter. looking now [07:15] davechen1y: does the detector pick them up? didn't find any last time I ran it [07:15] mattyw: morning [07:16] axw, thanks for taking a look at my branch - sorry about putting the wrong bug link [07:16] A few minutes before I wrote that I commented on the wrong bug in lp as well [07:16] mattyw: that's ok, I figured it out from the commit description :) [07:17] axw: it does, if you're persistant enough [07:17] http://paste.ubuntu.com/7730028/ [07:17] davechen1y: ah, thanks [07:17] PR's for both in play [07:17] sweet [07:31] morning all [07:33] morning [07:34] axw: shit [07:34] that race is worse than I thought [07:35] davechen1y: which one? [07:36] dimitern: ping [07:37] vladk, pong [07:37] axw: the one on the loggo/TestLogWriter [07:38] axw: i'm fixing it now [07:38] davechen1y: ah, I see. [07:38] hrm, thought I fixed that.. maybe it was a similar case [07:38] dimitern: I wrote a test for WatchInterfaces API client, It gives a 5 sec delay between state modification and notify signal [07:39] dimitern: test is working fast on API server, but with delays on API client [07:39] axw: i cna't see a way we can assert that the TestLogWriter is single threaded [07:40] davechen1y: perhaps just call loggo.RemoveWriter before ranging over tw.Log [07:40] axw: fuck it, i've gone and made tw.Log a function that returns a copy of writer.Log [07:40] okey dokey [07:40] vladk, probably the statetesting code used for the watcher need tweaking a bit - can i have a look? [07:40] if it's worth doing; it's worth over doing [07:40] less band-aid-ish [07:40] sgtm [07:43] axw: PTAL [07:44] this will need some work to integrate into juju and other callers [07:44] vladk, look for BackingState and StartSync it once you're sure there's a notification waiting [07:45] vladk, (even if that's not *directly* applicable, the point is that the *State that's driving the api server needs to be synced -- and that's not necessarily the *State you're usually manipulating in the tests) [07:47] fwereade, statetesting does that internally for the Assert methods [07:47] ..IIRC [07:47] dimitern, indeed, but it's not necessarily using the right *State [07:47] dimitern, and if you're using the wrong one you'll see those 5s delays [07:47] fwereade, right [07:48] ttps://bugs.launchpad.net/juju-core/+bug/1336180 [07:48] <_mup_> Bug #1336180: state/apiserver: yet more data races [07:48] fwereade, vladk, it uses c.State.StartSync() [07:49] dimitern, vladk: so it *should* just be a matter of starting it with s.BackingState, if that's accessible; and figuring out how to set one up if it's not [07:49] dimitern, fwereade: https://github.com/juju/juju/pull/207 [07:50] vladk, line 278 in networker_test.go - use s.BackingState instead of s.State for NewNotifyWatcherC [07:51] (and in other similar cases) === rogpeppe2 is now known as rogpeppe [08:06] morning [08:06] *yawn* [08:17] dimitern: thanks, that works, could you review my PR [08:24] vladk, will do in a bit === rogpeppe1 is now known as rogpeppe [09:05] dimitern: so a network interface always has a subnet (in the juju model) [09:05] dimitern: what's the default subnet? [09:08] dimitern: and why do we reference count networks? [09:09] voidspace, sorry, in a call, will get back to you a bit later [09:09] dimitern: no problem [09:32] * fwereade bbiab [09:33] is juju-bot on holiday today? [09:34] voidspace, back [09:34] voidspace, so there are two default networks - juju-private and juju-public (if available) [09:35] voidspace, they are discovered and created at bootstrap time, if the provider supports that [09:36] voidspace, we use refcounts for networks, because they can be referenced by services and/or machines (i.e. they are considered in-use, when the refcount>0) [09:37] voidspace, removing networks in use is not allowed; refcounts are used as a simple sanity check on the same doc as the network, as opposed to checking multiple documents in other collections [09:41] axw, are you still awake/ working? [09:43] mattyw: I am here [09:43] the bot is not on holiday, it's just a bit sick [09:44] mattyw: which PR are you trying to land? [09:44] axw, https://github.com/juju/juju/pull/108 [09:44] axw, but I was going to ask you about this one: https://github.com/juju/juju/pull/198/ [09:45] vladk, reviewed [09:45] mattyw: nfi why it didn't pick that up... [09:45] mattyw: ask away [09:46] axw, just wanted to make sure I understand your feedback properly [09:46] axw, do you mean if you specifiy add-machine -n you shouldn't be able to specify lxc:2 as well? [09:47] mattyw: right [09:47] mattyw: just like with "juju deploy", which doesn't allow you to do --to and -n together [09:49] axw, deploy uses the UnitCommandBase stuff for all of that, I did wonder if I should be using something generic to provide the -n flag, but I figured it probably wouldn't be worth it, I guess I should just implement the same logic in AddMachineCommand.Init? [09:50] mattyw: yeah I think so [09:50] axw, also I wasn't sure if I should just be sending a list of items to the existing AddMachine api call - or making a new api that would take a NumMachine int, do you have an opinion either way? [09:52] one small command change for review: https://github.com/juju/juju/pull/208 [09:52] mattyw: I think a list makes sense [09:52] dimitern: thanks [09:52] dimitern: but, I was asking specifically about subnets [09:53] dimitern: every network must have a subnet [09:53] voidspace, yes [09:53] dimitern: so what is the "default subnet"? [09:53] voidspace, a subnet of the default network [09:53] dimitern: specifically [09:53] dimitern: what will the netmask be [09:53] dimitern: how do we determine it [09:54] voidspace, the provider will implement ListNetworks() and/or ListSubnets(), depending on what's supported, and tell us [09:54] dimitern: heh, so for local provider then [09:54] voidspace, in network.BasicInfo struct used as result will have an IsDefault field [09:55] dimitern: what does the implementation of ListSubnets do? or what if the provider doesn't support subnets [09:55] dimitern: what will we use as the default [09:55] voidspace, local and manual are special (yet undefined fully wrt networking) [09:55] dimitern: I just wonder if the concept of "you have to have a subnet" really reflects reality [09:56] dimitern: and for refcounts, it seems to be the case that we want refcounts in order to destroy networks [09:56] voidspace, there are SupportsNetworks() and there will be SupportsSubnets() as well, which each provider implements to tell juju what it can handle, and we call ListSubnets/Networks or both accordingly [09:56] dimitern: i.e. know when they are not references [09:56] dimitern: (sure, but as a network *must* have a subnet I wonder what we do if the provider doesn't support subnets - what is "the default") [09:56] voidspace, read the doc :) [09:56] dimitern: why do we need to destroy networks? as they're abstract concepts, what's the cost in just leaving them around [09:57] dimitern: I am! [09:57]