=== slank is now known as slank_away [01:26] Night all [05:20] jam: hi, you around? [05:30] hey wallyworld [05:31] hi, stupid question, i cannot get a go install or build etc to include any of the openstack stuff [05:31] it seems like the openstack package is ignored for me [05:31] wallyworld: what command are you running, and what result are you expecting? You mean being able to run 'juju foo' with openstack as the environ? [05:32] yes [05:32] or you expect something in $GOPATH/bin/... [05:32] so i "go install launchpad.net/juu-core/..." [05:32] and my go bin dire is updatred [05:32] with the latest juju [05:32] but onloy if i change somthing in ec2 package [05:33] if i tweak something in openstack package the build/install process seems to ignore it [05:33] so that implies the openstack stuff is not being included [05:33] wallyworld: right, I'm guessing we are missing the registration of the openstack provider [05:34] and matches what i see cause the openstack init() is not being run [05:34] the registration is done inside init() [05:34] and init() is not being run for me since the go install seems to be ignoring my openstack package [05:34] wallyworld: look at launchpad.net/juju-core/cmd/main.go [05:35] it does "import _ "launchpadb.net/juju-core/environs/ec2" [05:35] we probably just need to add "openstack" there. [05:35] ah ok, that makes sense [05:35] jam, wallyworld: that's right :) [05:36] jam, wallyworld: good mornings [05:36] thanks, i didn't know about that - was driving me crazy [05:36] good early morning to you as well, fwereade [05:36] morninh [05:36] so till now it seems no one has tried to test go juju with openstack apart from running hte tests [05:37] wallyworld, that's weird, I could swear dimitern told me you had it bootstrapping -- maybe that's via the live tests though [05:37] yeah, i suspect so [05:38] and i thought i'd take a moment today to try stuff out so to speak [05:38] wallyworld, +1 [05:38] with no luck till now :-) [05:39] wallyworld: I thought he had, but I thought he had just done a local patch for it. It is how I would have tested it. [05:40] jam: maybe it's time to plug it in then? [05:40] wallyworld, jam, I think so, definitely [05:40] wallyworld, jam, I would be +1 on some very basic tests that just verify that we will accept a trivial config for each of the providers that "ought" to be registered [05:41] fwereade: so with the juju go, running "juju bootstrap" does not create a default environments.yaml like pyjuju, right? [05:41] wallyworld, that's right; I think there's a bug related t it [05:41] ok, np. [05:41] +1 to adding a few simple tests also [05:42] wallyworld, it's https://bugs.launchpad.net/juju-core/+bug/993616 [05:42] wallyworld: so if we are actually at a point where 'juju bootstrap' with a openstack config doesn't barf/cause corruption, I'm happy to expose it. [05:42] <_mup_> Bug #993616: We should ship a boilerplate environments.yaml.

< https://launchpad.net/bugs/993616 > [05:42] wallyworld, it looks like it has some good samples in there [05:42] fwereade: what is the standard for testing actual binaries? [05:42] I don't think I've seen much testing of it in the code. [05:43] jam, we have thus far got away with testing main() in-package [05:44] jam, I would also be +1 on smart improvements to this... there's at least one test which actually builds go code [05:44] jam, I think that's the test for --upload-tools, perhaps [05:44] jam: the main issue i see with exposing openstack by default is perhaps the lack of a suitable public bucket [05:45] wallyworld: we have a shared account now, I think I sent you the access keys [05:45] but if we ship a yaml that has stuff commented out, neither openstack nor ec2 will work out of the box anyway [05:45] we can turn that into the public buckte for now. [05:45] jam: right, but we wouldn't want to ship those keys [05:46] wallyworld: "public bucket" [05:46] we use the shared account to create a readable bucket [05:46] which other people can use [05:46] ok, sorry, brain fade [05:49] wallyworld: so we should be able to do something like set up a test double, override the appropriate env variables, run 'juju bootstrap', and inspect the double afterwards [05:49] if you want a bit of an end-to-end (but with doubles) test. [05:50] that (IMO) makes a pretty decent smoke test [05:50] yeah, agree [05:50] i also want to try it on an openstack instance (ie mine) to see how it goes [05:52] wallyworld: certainly. I just wouldn't put your account details into the test suite :) [05:52] never fear, i was just wanting to run juju locally [05:53] not the tests, the actual tools themselves [05:53] certainly [06:00] bbiab, have to take son to physio [06:08] jam, if you have a moment, would you take a look at https://codereview.appspot.com/7227045/ please? and consider in particular the rambling about InferEndpoints, which I am starting to believe is mildly crackful [06:39] fwereade: looking at the test, it does feel like there should be a call like AddRelation that does the infer first and then adds them. Given you are doing retries in case the service definition changes, it does seem like a combined op. [06:40] jam, cool, thanks [06:41] niemeyer, good morning [06:41] niemeyer, midnight feeding? :) [06:41] niemeyer: is this good morning? or good evening? :) [07:08] TheMue, morning [07:09] dimitern, morning [07:09] TheMue, free for a quick call? [07:10] fwereade: could we do it a bit later? i've to step out to the post office in a few moments, just packing a package i have to sent out today. ;) [07:10] fwereade, TheMue: morning [07:10] TheMue, I expect so -- I'll need to pop out for a bit in an hour or so myself [07:11] TheMue, when will you be back? [07:11] fwereade: i think in 30 minutes, but let's talk now [07:12] TheMue, ok, cool, should be quick [07:49] fwereade, TheMue, jam, dimitern, wallyworld: morning! [07:51] rogpeppe: morning! [08:01] fwereade: Heya [08:01] jam: Yo [08:01] fwereade, jam: Sorry, missed the messages earlier [08:01] fwereade: Yeah, kind of :) [08:02] Making use of the quiet time, though [08:10] niemeyer, enjoy it while you can :) [08:10] gents, popping out for a bit [09:10] rogpeppe, are you free for a call about the API etc? [09:10] fwereade: definitely [09:10] fwereade: one mo, i'll paste you an initial CL [09:11] rogpeppe, cool [09:11] fwereade: https://codereview.appspot.com/6878052 [10:12] jam, mgz, wallyworld: PTAL https://codereview.appspot.com/7228058/ [10:13] * wallyworld looks [10:17] fwereade, rogpeppe: also, this one is closely related: https://codereview.appspot.com/7241045/ to ^^ [10:22] dimitern: done [10:23] wallyworld: thanks! [10:23] np. 'see' you at the standup [10:24] dimitern, cheers, on it shortly [10:44] dimitern: reviewed [10:45] rogpeppe: cheers! [11:02] dimitern, I think I'm in agreement with rogpeppe -- why is it that GetPrivateAddress needs to be exported? [11:02] fwereade: because I need to call it in the tests, which are in a separate module [11:03] dimitern, that sounds like a job for export_test.go [11:04] fwereade: yes, but would it be usable from openstack then? tests are in openstack_test and putting it in export_test will make it available for tests, but not for the openstack module I think [11:04] dimitern, but shouldn't it really be tests against the openstack Instance type anyway? [11:05] fwereade: well, initially it was a method of instance (lower case one - in provider.go), but then creating an empty instance only so I can call it on it seemed wrong [11:05] dimitern, also, shouldn't Instance.DNSName be returning the public address? I suspect I'm hopelessly confused here ;) [11:06] fwereade: no, the same logic is ported from py-juju - private address is used, the public one is not available (at least not always), until you set a floating IP to the instance [11:07] dimitern, ah, ok, sorry, you told me that the other day [11:07] fwereade: yeah, but anyway we don't need the public address there or dns name - just an address to access the instance - private will do [11:08] dimitern, so, wait, don't we need some public address so we can talk to the "bootstrap" node from outside? isn't that done via DNSName()? [11:09] fwereade: I'm not sure, but I think it's not used from outside [11:10] fwereade: maybe we should have a short call to discuss this with whoever knows best about how is this supposed to work [11:10] dimitern, ISTM that we have to talk to the bootstrap node from outside [11:10] rogpeppe, ping [11:10] fwereade: well, how did it work in py-juju then? [11:11] fwereade: pong [11:11] rogpeppe, what's your understanding of DNSName() re public/private address? [11:12] dimitern, I have no idea :) [11:12] fwereade: the only important thing AFAIR is that it should return a publicly usable address [11:12] rogpeppe, that was what I thought :) [11:12] fwereade: because that's what we use to connect to a unit from outside [11:13] fwereade: nothing in juju connects to machines, i think [11:13] rogpeppe, define "nothing in juju" please :) [11:14] fwereade: bah. ykwim :-) [11:14] dimitern: when you say "would it be usable from openstack", what openstack package are you referring to there? [11:14] rogpeppe: environs/openstack [11:15] dimitern: any function defined in openstack is callable from openstack [11:15] rogpeppe: the problem is the tests are in openstack_test [11:15] dimitern: then you can use export_test [11:16] rogpeppe: and still use it in both openstack and openstack_test ? [11:16] dimitern: or define some tests in openstack instead, if that's more appropriate [11:16] dimitern: sure [11:16] dimitern, hence export_test.go -- that will be in package openstack, but thanks to the _test.go name will only be built with the tests [11:16] dimitern: in export_test: [11:16] dimitern, you won't be able to use export_test.go stuff in openstack [11:16] func GetPrivateAddress(addresses map[string][]nova.IPAddress) (string, error) {return getPrivateAddress(addresses) } [11:16] dimitern, so you have an exported name that calls the unexported function [11:17] dimitern, what rogpeppe said :) [11:17] rogpeppe: ok, I'll that then [11:17] fwereade: still not sure about the DNSName semantics though [11:17] dimitern: generally we try not to put significant code in export_test - it's used to make things available for the external tests only. [11:18] dimitern, it kinda looks like you're always returning a private address there [11:18] dimitern, I think I'd be ok with a private address if and only if no public one is available [11:18] dimitern: private addresses are pretty much useless for the purposes juju uses DNSName for [11:18] dimitern, but DNSName() should usually return a public one [11:18] fwereade: I still think we better discuss this - with mgz as well, he has better view on the py openstack provider [11:19] fwereade: i tihnk i disagree - it's misleading if juju ssh tries to connect to an address that doesn't work [11:19] I need to wrap my mind around it a bit [11:19] dimitern, ISTM that the python uses a private one only if it can't get a public one [11:19] rogpeppe, ^ [11:19] fwereade: where do you see this? [11:19] fwereade: what's the use of doing that? [11:20] rogpeppe, dimitern, mgz: my unlettered reading of this is that some openstacks have nothing public, but that the client is connecting from the same network, so it's "ok" [11:20] fwereade: hmm, if that's really true, then it might be ok [11:20] dimitern, providers/openstack/machine.py:62 [11:21] fwereade: yeah, I see this - so it's fine to do it like this? [11:22] dimitern, I think that falling back to a private address when no public one exists is *probably* ok, assuming we get mgz to confirm we're not smoking crack [11:23] mgz: ping [11:23] dimitern, but the current approach of always returning private is not ok, I think [11:23] fwereade: well, if we need definitely a public IP always it changes things quite a bit [11:24] fwereade: we i'll then need to provision a floating IP to each machine we're starting in the environment [11:24] dimitern, I think I'm failing to communicate [11:24] dimitern, if a public address is available, we should definitely use that [11:24] fwereade: maybe we should have an environ parameter, say private-addresses, saying whether we want to use private or public addresses [11:24] dimitern, if not, ISTM that falling back to a private address is roughly sane [11:25] fwereade: it's never available for canonistack and I doubt it will be anything but rare in other cases [11:25] fwereade: when it's false, we could try to allocate floating ip addresses, or whatever's necessary to make dns names public [11:25] dimitern, assuming my reasoning holds -- I was not privy to the thinking behind this behaviour in python, and I would like some input from someone who's done this before ;) [11:26] rogpeppe: dns name allocation is also supported somewhat, but it's a different API call and likely to be similar to how floating IPs are assigned [11:26] rogpeppe, yeah, why does it have to be DNSName()? is it not SomeStringThatWillGetYouToTheInstance()? [11:27] fwereade: well, it's SomeStringThatWillGetYouTotheInstanceFromSomewhere() :-) [11:27] rogpeppe, ha, yeah [11:27] fwereade: and it's the "somewhere" that we can't easily second guess [11:27] rogpeppe: I was thinking about that :) [11:27] dimitern, what time's your standup again? [11:28] fwereade: now [11:29] dimitern, please ask mgz then, and we shall reconvene shortly? [11:32] fwereade: sure, once he comes round [11:32] dimitern, cool [11:33] dimitern: hey [11:34] * mgz reads discussion [11:34] mramm: we're mumbling if you want to unmute [11:35] cool be on in 1 min [11:36] mramm: you were echoing [11:40] so, for now returning whatever you can (by the logic from Python) for DNSName will get things working [11:49] fwereade: ping [11:49] rogpeppe, pong [11:49] fwereade: i'm looking at auth stuff [11:49] fwereade: and wondering about the "admin" of SetAdminPassword [11:49] fwereade: i'm wondering if it might be more appropriate if the "admin" was actually "client" [11:50] fwereade: because Admin sounds like it has superuser properties, but i don't think that's what we want to give it [11:50] rogpeppe, -1, I think we will before too long have to worry about other users with fewer privileges [11:51] fwereade: indeed, and i think that "client" can more easily mutate into other usernames [11:51] rogpeppe, who should have those privileges if not whoever set up the environment? [11:51] fwereade: i'm not sure that anyone has all the privileges [11:51] rogpeppe, this may be something we need a call to cover properly [11:51] fwereade: for instance, i don't think we want to enable a dodgy admin to act as a uniter [11:52] rogpeppe, ok -- but I think whoever set up the environment should have all applicable privileges on that environment -- and "admin" seems to fit that better than just "client" which is IMO way too generi [11:53] rogpeppe, at least we're not calling them "root" ;0 [11:53] rogpeppe, btw, we don't have any ssh proxy support in state.Open, do we..? [11:53] fwereade: i'm also looking at it from the perspective of the actual mechanics of logging in. here's the code i currently have: http://paste.ubuntu.com/1585690/ [11:54] fwereade: no. do we want it? [11:54] rogpeppe, given what mgz said above, yes, probably, if we want canonistack to work [11:54] fwereade: where would the proxy live? [11:55] rogpeppe, you ssh into canonistack instances via chinstrap AIUI [11:55] fwereade: the question wrt the above code is: what underlying type does State.Entity return when EntityName=="admin" ? [11:55] * rogpeppe hasn't heard of chinstrap [11:56] rogpeppe, IMO we need a state.User [11:56] fwereade: i think setting passwords directly on the applicable entities is fine. but we may need state.User too [11:57] rogpeppe, I'm 90% sure we need a state.User, but I need to talk to you about it [11:57] rogpeppe, I imagine it would have worked transparently when we were using SSH directly, because everybody would have had their proxies set up [11:58] fwereade: what is chinstrap? [11:58] fwereade: ah, it's a machine! [11:58] fwereade: we don't need ssh proxying, just a straightforward tcp tunnel [11:58] s/tunnel/forwarder/ [11:59] fwereade: but maybe sshd can provide that. [12:00] rogpeppe, just realised I need to eat before team meeting [12:00] fwereade: ok, enjoy [12:00] bbiab :) === TheRealMue is now known as TheMue [12:31] jam: i'm trying to lbox submit those branches and the very first one is failing with a sad gofmt, complaining about a file which doesn't even exist. have you seen that before? [12:49] wallyworld_: that sounds a bit weird. if gofmt is talking about a file, i'm pretty sure it must be there. [12:49] wallyworld_: have you tried go fmt ./... from the project root? [12:49] yeah, tell me about it. but 'ls' says otherwise [12:49] yep, no issues [12:50] wallyworld_: try running the same command as lbox does: find * -name '*.go' | xargs gofmt -l [12:50] wallyworld_: and see if it prints anything [12:50] yeah, did that too [12:50] let me see what it said [12:51] no output, so happy [12:51] yet lbox submit complains [12:51] i did wipe my local go bin dir and did a go install launchpad/net/lbox earlier [12:52] wallyworld_: try putting a debug statement in .lbox.check to see if it's actually running; and include the value of $BADFMT too. [12:52] s/a debug/an echo/ [12:53] yep, will do [12:54] ah, complains branch is not clean, i'll have to comment out that bit [12:54] wallyworld_: just commit it... [12:54] wallyworld_: (then revert it later) [12:55] ok [12:57] hmmm. my echo wasn't printed [12:57] now i'm confused [12:57] wallyworld_: sanity check failed :-) [12:58] yeah, but that's not something i expected to fail, it's just worked previously [12:58] and yet something is running go fmt [12:58] wallyworld_: that's what sanity checks are all about :-) [12:59] wallyworld_: more: something is saying "gofmt is sad", right? [12:59] yes [12:59] wallyworld_: which means *a* lbox.check script is running, from somewhere... [12:59] for a non existant file [12:59] yeah, i'll have to try and track it down [12:59] wallyworld_: how about running lbox --versose --debug ? [12:59] s/versose/verbose/ [12:59] poke for our hangout? [13:00] good idea, didn't realise those optins were there [13:00] on my way [13:00] * fwereade joins [13:01] hangout link? [13:01] mgz: poke for the hangout [13:02] https://plus.google.com/hangouts/_/33acfc2c8af8792568274fa110371db956f9fde7 [13:02] w7z: https://plus.google.com/hangouts/_/33acfc2c8af8792568274fa110371db956f9fde7 [13:05] niemeyer: meeting? [13:11] rogpeppe: Sorry, yep [13:11] rogpeppe: figured it out - jam's fault :-P [13:12] sad go fmt file in trunk which was not there when i started my branch [13:16] wallyworld_: I'm willing to take the blame, I accidentally committed something directly that I was trying to submit via botd [13:16] sorry about that. [13:16] np [13:28] jam: i've submitted my stuff, needing to merge trunk (and hence your changes) first. there were some conflicts which i think i've done correctly. but it would be cool if you could check [13:56] fwereade: i'm just considering adding a user api to the state. we'd use it for admin only to start with, but potentially expandable in several directions. what do think of this? http://paste.ubuntu.com/1585936/ [14:06] rogpeppe, that looks roughly sane, I'll want to have a call again in a bit but my brain is currently a touch overloaded [14:06] fwereade: np! [14:38] rogpeppe, wallyworld: follow-up https://codereview.appspot.com/7228058 [14:39] rogpeppe: what was your trick for doing an initial codereview thing? I can just propose current trunk against an empty (or r1) branch for the current state to come up in diff, for instance === slank_away is now known as slank [15:08] * niemeyer => lunch [15:10] rogpeppe, mgz: follow-up on https://codereview.appspot.com/7241045/ [15:16] dimitern: looking [15:41] ...I still don't like this struct-list-test+loop thing [15:42] mgz: well, I like it more each time :) so compact and expressive [15:43] it's longer than the original! [15:43] mgz: it's not yes, I changed it as we discussed - use public if available, otherwise use the private - as a compromise [15:44] mgz: aah longer [15:44] each test in python is 5-7 lines, with pretty formatting [15:44] and that's the *complete* test [15:44] not just a test definition that then needs overly complicated testing code that lives somewhere off the page [15:44] mgz: anyway it does what's supposed and it's easy to extend, etc. [15:45] each of these definitions is 8 lines, and that's without any test code, or the struct defintion [15:45] it's harder to do particular customisations for special cases though [15:45] you'd end up having to add a new field to your struct as a marker [15:46] mgz: it's definitely shorter than individual test case methods + comments for each [15:46] add that to the each struct constructor (or used the named param idiom), and then special case that in the testing codde [15:46] yeah, because there are no comments [15:46] so, you've no idea what the expected result is... [15:47] anyway, looks fine apart from me disliking the style (which is idiomatic here, so correct for you to do...) [15:48] mgz: ok :) is it LGTM then? [15:48] sec, hitting m shortly [15:50] ok gents, I've been on since 6am and I'm a bit fried now; I'll probably swing by later, but I'm off to decompress for a bit [15:53] dimitern: okay, get landing :) [15:53] have a nice rest fwereade :) [15:53] mgz: cheers [16:17] rogpeppe: sorry I missed your last comments, will do a follow-up CL [16:17] dimitern: np [16:18] dimitern: just makes the docs look a bit more professional, i think [16:18] rogpeppe: yeah, I agree - wasn't sure if using func-style comments for struct fields is ok [16:19] dimitern: it's not always done, but i think it looks better when it is. particularly when some are longer. [16:20] rogpeppe: yes, if it's a whole line, rather than a short one at the end of the line [16:21] dimitern: yeah. [16:30] rogpeppe: and here it is: https://codereview.appspot.com/7235058 [16:32] dimitern: LGTM with one typo [16:32] rogpeppe: cheers [17:11] Yeah, build of Juju on OS X works. Will test it later. [17:30] a CL to review, if anyone fancies, nothing too onerous: https://codereview.appspot.com/7226056/ [18:04] that's me for the day [18:04] g'night all!