[00:12] wallyworld_: will re-running go get on juju-core ensure the proper dependancies get updated? Or is there an easy way to go get -u all of the entries in dependancies.tsv? [00:13] go get -u on launchpad.net/juju-core should do deps, but [00:14] go has no proper dependency management like python etc [00:14] i think it just pulls trunk [00:14] right, thank you [00:14] the dependencies.tsv file is used with tooling by the CI [00:14] not by go itself [00:14] ahh ok, makes sense, was wondering what that was for [00:15] i lot of us are very disappointed with go's dep management [00:15] the designers think it's ok to build against trunk of all the deps [00:15] which is so crackful i don't know where to start [00:15] heh [00:15] crackful [00:16] so if I want to update ALL THE THINGS can I just go get -u launchpad.net/... ? [00:16] or will that go get everything on launchpad.net ? [00:16] I tried asking in #go-nuts, but they were snarky because they didn't know :P [00:16] not sure actually [00:16] not sure if it just looks on disk [00:16] dyou have a way to update a local tree? [00:17] i normally use bzr to get the required rev [00:17] by hand :-( [00:17] * bodie_ offers wallyworld_ a consolation beer [00:17] cause i don't trust just building against trunk of all the deps [00:17] hm [00:17] * wallyworld_ slurps the virtual beer [00:18] from what i recall, the go guys said that dependency management is hard so let's just ignore it :-( [00:18] looks like go get -u /... works [00:19] great [00:20] wallyworld_: yeah, I've seen projects using two accounts on github, -dev and -stable, which I guess works for ensuring you are working off some random rev of master [00:20] are not [00:21] yeah [01:35] <_thumper_> waigani: next you could do the tests for optional autostart for lxc [01:35] <_thumper_> waigani: although we shouldn't land this until we have "suspend" and "resume" [01:36] <_thumper_> otherwise we have no way to restart a non-autostarting environment [01:41] _thumper_: okay, so where do I get started with the "suspend" and "resume" stuff [01:42] wallyworld_: what was the bug you mentioned I could look at? [01:42] it's on the 1.17.5 milestone page, let me check [01:43] _thumper_: i.e. is there any relevant code or am I introducing this as a new feature, in which case I'll go play with the cli lxc suspend and resume [01:44] waigani: https://launchpad.net/bugs/1291165 [01:45] <_mup_> Bug #1291165: juju bootstrap local cannot bootstrap

[01:45] wallyworld_: cheers [01:45] ah someone has marked it incomplete [01:45] but still [01:45] if there's an empty jenv file, we should be able to nuke it [02:42] davecheney: hey [02:42] davecheney: how's things going? [02:44] wallyworld_: https://codereview.appspot.com/74370044/ [02:44] looking [02:46] axw: that azure branch is a monster [02:46] axw: no changes in gwacl? [02:46] ah, can see it [02:46] yes there is [02:47] axw: but I guess all the gwacl stuff needed is reviewed and landed already [02:48] wallyworld_: I did add some tests around the autostart for the template and the clone [02:49] ok, i missed those then :-) [02:49] https://codereview.appspot.com/74370044/diff2/1:20001/container/lxc/lxc_test.go [02:49] great [02:49] thumper: o/ [02:50] still fighting with proxy prolems [02:50] just adding some more debug to juju [02:50] davecheney: what sort of problems? [02:50] not obeying no_proxy [02:50] i guess [02:50] you have to be very explicit in the no_proxy list [02:51] thumper: sorry, was in the zone [02:51] axw: np [02:51] yes, monstrous in several ways [02:51] davecheney: but no_proxy is at least being propogated? [02:52] axw: I was going to do the initial review for your branch [02:52] and trying not to get put off by the shear size :-) [02:52] thumper: sorry :( [02:53] axw: :-) [02:53] I've tried to make it as minimal as I can [02:53] hey, at least it is just four files :) [02:53] hehe [02:53] reasonably isolated [02:53] I just rewrote those four files :) [02:54] axw: so this branch doesn't put multiple machines into a single cloud service yet? [02:55] thumper: no. well, state servers will - but we don't spawn multiple of them yet [02:55] hah [02:55] ok [02:55] just didn't want to get confused more than usual reading the code [02:56] hey thumper [02:56] thumper: let me know if you need any explanations on the azure concepts [02:56] o/ jcastro [02:56] theorize with me. [02:56] axw: sure [02:56] jcastro: oh... kay... [02:56] so you know how you can add servers via the manual provider [02:57] what's to stop me from adding some servers from aws, some from azure, some from hp [02:57] and doing a poor man's cross environment relations? [02:57] jcastro: they won't be able to talk to each other [02:57] jcastro: the units use the internal private address [02:57] jcastro: they won't have the same internal addresses [02:57] which is cloud (or region) private [02:57] what thumper said [02:58] so even if I was to try and trick them with say openvpn or whatever [02:58] bah! so close! [02:58] yes, close [02:58] if you set up all the routing yourself you could do it [02:58] I forsee a time when we use our own SDN for juju, so this would work [02:58] but this isn't in the near future [02:58] perhaps 6-12 months if we're lucky [02:59] * jcastro nods [03:00] thumper: lxc-freeze/lxc-unfreeze == suspend/resume? [03:01] thumper: i think so [03:01] i need to find out which host is getting knocked back by the proxy [03:01] davecheney: cheers [03:01] waigani: no... I don't think so [03:01] waigani: because when a machine stops or is rebooted [03:01] waigani: the machine is in a stopped state, not a frozen state [03:01] so better to be consistent [03:01] perhaps after the initial approach [03:02] we could use freeze/unfreeze in addition to start/stop [03:02] if necessary [03:02] but start/stop are bare minimum necessary [03:03] thumper: oh right, I assumed we already had that and you wanted suspend/resume in addition [03:03] waigani: no... [03:03] it's complicated [03:11] thumper: I think it's best to leave this Azure stuff till after 1.18.0 in case it destabilises - what do you think? [03:12] hmm... [03:12] well, it would only destabalize azure [03:12] and they have kinda asked for it :-) [03:12] I'll punt the question to mramm [03:12] and extras [03:12] ok [03:12] I'll put it on the minutes for tonight's meeting [03:12] and we'll make a decision [03:12] okey dokey [03:31] wallyworld_, thumper: is there a way to look at trunk without switching to it? [03:31] you can browse from launchpad [03:31] from the filesystem? [03:32] from the browser [03:32] it would be handy to have trunk in a sublime window [03:32] or you can branch it locally somewhere [03:32] in that case, just branch locally [03:32] ah okay [03:40] thumper: resp, err := utils.GetNonValidatingHTTPClient().Do(req) [03:40] if err != nil { [03:40] return nil, fmt.Errorf("cannot upload charm: %v", err) [03:40] ^ this http client doesn't observe the proxy [03:42] or doesn't obey the proxy exclusion, or something [03:43] no [03:43] wait [03:43] its somewhere else [03:44] 2014-03-13 03:43:40 ERROR juju.cmd supercommand.go:296 failed to add remote charm "cs:precise/ubuntu-4": cannot upload charm to provider storage: 403 403 Forbidden [03:51] hmm.. [03:52] * thumper is still in axw's first file [03:55] thumper: still chasing [03:55] it's in storage.Puit [03:55] and this must be posting to the little web server the local provider stands up [03:55] i just need to find out which localhost addr it is sitting on [03:58] davecheney: should be in lxcbr0 [04:00] 10.0.3.1 is the host [04:01] davecheney: can you do 10.0.3.0/24 ? [04:04] waigani: what'cha up to? [04:04] waigani: because I have a few refactorings I'd like to see landed ASAP [04:04] :) [04:05] * thumper sighs [04:05] two down, two to go [04:05] thumper: i don't think so [04:05] i migth be able to do 10.0.3 [04:05] i think it's just a substring map [04:05] thumper: i already added 10.0.3.1 to the list [04:05] thumper: girl just got off the ice. I have to head home and do dinner/homework daddy thing. After that? [04:06] or email me and I'll pick it up this evening [04:07] davecheney: http://unix.stackexchange.com/questions/23452/set-a-network-range-in-the-no-proxy-environment-variable [04:08] thumper: intersting [04:41] thumper: was it as good for you as it was for me? [04:41] axw: probably a lot less of a pain for me [04:42] reviews can be quite tiring [06:06] thumper: me is sad [06:06] ERROR juju.cmd supercommand.go:293 empty http-proxy in environment configuration [06:07] save me looking, any idea? [06:07] there is no http-proxy in my env config [06:07] never has been [06:08] ah fark, wrong environment [06:08] that was with local [06:45] hey thumper [07:25] thumper: I accidentally cancelled during a commit. branch is locked. bzr break-lock: The lock for [branch] is in use and cannot be broken. [07:28] thumper: bzr info: Could not acquire lock "...go/src/launchpad.net/juju-core/.bzr/checkout/dirstate": [Errno 11] Resource temporarily unavailable [07:29] waigani_, can you try doing this? bzr break-lock bzr+ssh://example.com/bzr/foo with the bzr+ssh pointing to the remote branch [07:30] ^^ah right during a commit - not a push - I suspect my "advice" is worthless here [07:33] mattyw: yeah during commit $#@$ [07:34] axw? wallyworld_? ^ [07:36] waigani_: I did that once, and thumper had to walk me through fixing it... I can't remember the details sorry [07:37] hmmmm [07:52] thumper: never mind, I fixed it [07:56] wallyworld_: can I bzr switch -b [branch] in a new terminal while lbox proposing in another? [07:56] axw: ^ ? [07:57] waigani_: no [07:57] okay, np [09:21] dimitern: good morning [09:21] jam, morning! [09:21] dimitern: I'd like to have vladk do some work with you to get up to speed on juju-core [09:22] dimitern: do you think you can spend some time pairing with him today ? [09:22] jam, sure, after the weekly meeting perhaps? I need to get this fix done first. [09:23] dimitern: that should be fine, I think he's still got infrastructure stuff to do today [09:38] vladk: did you need something ? It sounded like you were just saying something as I closed the window [09:43] dimitern: o/ [09:43] thumper, hey - seen my mail? [09:44] yes, will get on it [10:02] fwereade: standup ? [10:02] mramm: welcome [10:02] mramm, fwereade, rogpeppe, mgz_, meeting? [10:02] dimitern: ah yes [10:03] voidspace, ^^ [10:14] fwereade: i really want to get this in for 1.18 because API https://codereview.appspot.com/75330043 [10:20] wallyworld_, are you working on bug 1291207? [10:20] <_mup_> Bug #1291207: juju-run symlink is broken after upgrade-juju

[10:20] dimitern: no [10:20] oh, is there a meeting now [10:20] wallyworld_, ok [10:20] dimitern: axw was going to i think [10:21] voidspace: come and join us :-) [10:21] wallyworld_: coming. [10:22] I did not know we had meetings at a different time on Thursdays! [10:22] Thursdays are special [10:22] voidspace, it's just this one [10:22] I wasn't planning to work on that bug... [10:22] I just reported it :) [10:22] dimitern: you mean just today, or just Thursday? [10:23] axw, wallyworld_, ok so if I can get to it I'll fix it today [10:23] voidspace: thursday is everyone, so we go a bit earlier to make it work for NZ [10:23] (currently deep into Azure) [10:23] thanks dimitern [10:23] dimitern: \o/ [10:23] voidspace, every thursday [10:23] dimitern: thanks [10:24] so my code review suggests I should add global state and reduce test coverage! [10:24] amongst some more sensible comments ;-) [10:27] * dimitern raises 2 crossed fingers in disgust... [10:29] hey sinzui, what's your plan for the 1.17.5 release? immediately after fixing the critical issues? [10:31] ah, the suggestion to reduce test coverage *does* suggest removing the code that was being tested too, so not entirely crazy :-) [10:42] axw, re bug 1291292 - the manual bootstrap writes provider-state-url file with a hardcoded file:////provider-state [10:42] <_mup_> Bug #1291292: use a utils/http client for all HTTP(S) calls across the codebase

[10:43] axw, and it does that because I guess p-s-url being an url, it's read with bootstrap.LoadStateFromURL [10:45] dimitern: it should be http:// though. I will try to remember to take a look later [10:45] machine-0 starts an httpstorage listener [10:46] axw, it starts an sshstorage in either bootstrap or provisioning mode (can't remember) [10:46] only command line should be using sshstorage - could be an issue around there [10:47] axw, seemed so to me as well [11:01] fwereade, i've assigned to myself all bugs i'm working on today - basically i need to get upgrades from 1.16 going smoothly [11:01] dimitern, ok, excellent; you and mgz will need to do vlans as a matter of some urgency though [11:06] jam: fwereade: i've tested this live upgrading from all of 1.16, 1.17, 1.18. I would love to get it in to 1.18 so we can delete lots of code post 1.18 https://codereview.appspot.com/75330043 [11:06] mgz_: I'm just butchering our mp in response to the review from rogpeppe [11:06] rogpeppe: did you see my comment on the br0 bug? [11:06] dimitern: rogpeppe: vladk: Dimiter has mentioned that he's a bit swamped with some critical bug fixes. Roger, is it possible for you to spend some time with vladk looking over your shoulder about the work your on? [11:07] jam: sure [11:07] wwitzel3: no [11:07] mgz_: although rogpeppe and natefinch suggest adding global state to the tests [11:07] voidspace: ?? [11:07] voidspace: what global state? [11:07] rogpeppe: using a global variable instead of a hard coded exception [11:07] rogpeppe: I solved it a different way instead :-) [11:07] voidspace: that's not global state actually [11:07] rogpeppe: which you *also* hinted at in a different part of the review [11:07] voidspace: but i also suggested another way [11:08] voidspace: yeah [11:08] rogpeppe: short version, I did the same set of tests we did yesterday using trunk and it all worked. [11:08] wwitzel3: bugger [11:08] oh for a real maas [11:09] rogpeppe: using aggregator.instanceInfo instead of the reqc channel directly improves the tests a lot [11:09] rogpeppe: so thanks for that [11:09] voidspace: cool [11:09] rogpeppe: just going through making all those changes [11:09] rogpeppe: and I have a question about the removal of the loop exit - but it's in my reply to the review so you can read it and answer when I'm done [11:10] voidspace: the only test coverage i suggested removing was inappropriately messing with private state - if we're going to do that, then there are *many* more tests we can do :-) [11:10] night all [11:10] thumper: g'night [11:10] rogpeppe: we have code to handle the channel dying [11:10] voidspace: i suggested removing that [11:10] rogpeppe: right [11:10] voidspace: because we never actually close the channel [11:11] rogpeppe: however, my follow up question is [11:11] rogpeppe: inside newAggregator we have the inner func which wraps the loop call in tomb.Kill with a defered tomb.Done [11:11] rogpeppe: which we didn't see an easy way to test [11:11] rogpeppe: as loop exit is now not *possble* shouldn't that code be removed too? [11:12] voidspace: the loop does exit when the tomb is killed [11:12] right, there's still the tomb exit [11:12] I forgot about that return [11:12] ok [11:12] voidspace: when that happens, tomb.Done will be called [11:12] voidspace: and if that's not called, then Wait won't return [11:12] voidspace: so we are testing it AFAICS [11:12] rogpeppe: I thought we were removing the only return from the loop, but we aren't [11:13] so fine, that's my answer - thanks [11:13] voidspace: cool [11:13] Hi, I just got stung by Bug#1243827 , is there a work around for it or a fix in the pipeline ? [11:13] <_mup_> Bug #1243827: juju is stripping underscore from options

[11:13] I'm very happy to remove tests if we also remove the code being tested [11:13] voidspace: it occurred to me that there are actually no tests that the requests are issued at the expected rate [11:14] gnuoy: put the data into quotation marks [11:14] rogpeppe: that gatherTime is honoured? [11:14] voidspace: yeah [11:14] rogpeppe: well, that's a ratelimiter detail - so that should be tested inside ratelimiter [11:14] adeuring, I'll give that a try, thanks [11:14] rogpeppe: we test that they're batched by the ratelimiter [11:14] rogpeppe: we don't test that our gatherTime is *passed* to the ratelimiter, but if they're batched (which we do test) I think it's fair to just assume that [11:15] voidspace: kinda. we test some batching, but we don't really check that we're using the ratelimiter appropriately. [11:15] gnuoy: ...or use double underscores. But that will require a change in the config once the bug is fixed. (and I am quite close to having a fix) [11:15] rogpeppe: so I could test that the time taken for the batched request to return is greater than gatherTime [11:15] voidspace: i believe it works, but i woke up this morning thinking that it didn't, and had to write a little piece of code to convince me itdid [11:15] rogpeppe: a simple assertion that timeTakenToBatch > gatherTIme then? [11:16] rogpeppe: I don't want to test the semantics of the ratelimiter inside the aggregator tests - the ratelimiter should be tested inside ratelimiter [11:16] voidspace: i'd actually prefer something that calls instanceInfo several times (say 100 times at 1ms intervals, with gatherTime say to 10ms) and check that the total number of requests issued is no more than 10 [11:17] rogpeppe: that does sound like testing the ratelimiter works [11:18] voidspace: i do hear you there, but it's actually just testing one incarnation of the ratelimiter, and that we are using it appropriately. [11:18] voidspace: it would be quite easy to do it wrongly [11:18] voidspace: that said, i believe the code works, so i'm happy to go forward with the existing tests if you're happy with them [11:18] rogpeppe: I could add a (threadsafe *sigh*) counter to testInstanceGetter [11:18] rogpeppe: so it wouldn't be hard to write [11:19] rogpeppe: just fire off a bunch of requests, ensure none of them error and that testInstanceGetter.counter <= 10 [11:19] voidspace: yeah [11:20] voidspace: sync/atomic can be useful for thread-safe counters, if you haven't already added a mutex [11:20] rogpeppe: the existing batch test also ensures that the right response goes back to the right request - so it still has value "as is" [11:20] voidspace: definitely [11:20] rogpeppe: I don't think we need a mutex in the place you suggested one [11:20] voidspace: ok [11:20] rogpeppe: as we're doing a blocking read so we *can* actually know there is no concurrency issue in that specific place [11:20] rogpeppe: the batched requests that set instanceGetter.ids can only happen after the first request completes [11:21] voidspace: that does assume a particular implementation, but i'm happy with that [11:21] voidspace: we don't do black-box testing [11:21] rogpeppe: well, it assumes that we can only get a reply after the request has been sent [11:22] rogpeppe: so we're assuming a particular implementation of the *universe* [11:22] ;-) [11:22] as we wait for the reply before making the next request [11:22] I don't *think* that makes assumptions about the implementation [11:22] voidspace: it assumes that the loop doesn't call Instances again after sending the reply [11:23] rogpeppe: we know there are no other requests coming in [11:23] voidspace: but that would indeed be a hard mistake to make [11:23] voidspace: i was not seriously suggesting adding a mutex, in all honesty [11:23] ok [11:24] voidspace: just pointing out the potential issue - it's a common pattern to see, and raises warning signs in my head [11:24] I've worked with threads before [11:24] so I'm no stranger to concurrency pain [11:24] voidspace: :-) [11:25] voidspace: given the above, you probably won't need a thread-safe counter either [11:25] voidspace: as you know that instances is always called from a single thread [11:25] no, we need to batch 100 calls concurrently as we can't use instanceInfo synchronously if we want batching [11:25] and a plain increment won't be thread safe [11:26] so I think we will [11:26] although you maybe right - let me think about it [11:26] voidspace: we can make 100 calls concurrently, but Instances will only be called from a single thread [11:26] ah yes [11:27] we make the requests from 100 different goroutines - but the instance is accessed just from the loop goroutine [11:27] voidspace: yeah [11:27] cool [11:27] voidspace: that said, we really are assuming an implementation here - the previous implementation of instanceInfo *would* have called Instances concurrently [11:28] voidspace: so probably best to use atomic.AddInt32 or whatever , just to be on the safe side [11:28] voidspace: then you can easily verify that the tests fail against the old version of the code if you wish [11:28] fair enough [11:29] a big part of my work with paymentservice was implementing correct locking to solve our concurrent database access problems - that was *worse* though because the concurrent access was from different processes, and actions that acquired the lock could trigger other actions in other processes that needed the lock too [11:29] so we had to add locking without deadlocking [11:29] we ended up with a state machine to handle the actions (state transitions) and also do the locking in basically *one place* [11:29] but before then we had to add careful locking everywhere we found a race condition [11:30] coffee [11:30] voidspace: yeah, mutexes work ok in simple situations, but for more complex situations, a single thread makes things easier [11:31] and William and I have some fun horror stories about threading from Resolver Systems - we changed the main calculation engine to work in a background thread (to not block the UI and allow multiple concurrent calclulations) [11:31] voidspace: database locking is awkward because you don't really want a monitor process either [11:31] but syncing the results back to the UI had to happen on the UI thread [11:31] voidspace: ha ha [11:31] we spent *days* finding and fixing race conditions [11:31] including multiple iterations of "I can *prove* that what just happened can't happen" [11:32] voidspace: a nice idiom in Go can be to send a function on a channel, which then gets executed by the monitor thread [11:32] rogpeppe: yes, being able to send functions on channels is very nice [11:33] similar to InvokeOnThread (or whatever it's called) in C# - which kinda-sorta has function pointers in the form of delegates [11:33] actually, it's functional programming support improved massively after I stopped using it [11:34] voidspace: yes, except that InvokeOnThread sounds like you can tell some code to execute on any arbitrary thread [11:34] only if there's an event loop running [11:34] so for the main GUI event loop you can pass a delegate and say "execute this delegate on the GUI thread and return the results to me" [11:34] voidspace: right [11:35] which yes, isn't the same as sending a function across a channel - just reminded me of it I guess [11:35] voidspace: if you have generalised "event loops" it makes sense [11:35] yep, you could run multiple event loops in different threads [11:36] anyway, coffee time for me [11:36] voidspace: enjoy [11:36] :-) [11:39] vladk: hi [11:41] vladk: when you feel that you've caught up with what you want to do, perhaps we could hang out together, as john suggested [11:42] rogpeppe: may we hangout in an hour? I'm a little busy right now [11:43] vladk: sounds good [11:43] vladk: ping me when you're ready [11:43] rogpeppe: ok [11:47] rogpeppe: https://codereview.appspot.com/72270044/ [11:47] rogpeppe: I made those changes we discussed yesterday [11:47] wwitzel3: looking [12:04] rogpeppe: also I was able to replicate the error from your ticket (1271144) [12:04] rogpeppe: but am not able to replicate it on your branch :) [12:04] rogpeppe: it is intermitten on trunk, but I can make it happen [12:04] gnuoy: doies the string where the "_" is removed start with a "-" of "+" (just to confirm that I'm on the right track) [12:04] wwitzel3: great! [12:04] wwitzel3: that's a relief [12:05] http://paste.ubuntu.com/7084302/ [12:05] adeuring, yes, it starts with a "-" [12:05]