[00:03] alexisb: m in 1:1 [00:03] anastasiamac, yep I will be there soon [00:04] alexisb: k. thnx :) [00:33] thumper, you back yet? [00:40] wallyworld or anastasiamac: https://github.com/juju/juju/pull/6431 [00:41] looking [00:41] plz :) [00:42] menn0: lgtm [00:43] anastasiamac: thanks [00:43] menn0: question [00:43] wallyworld: yes? [00:43] in the test, we are passing the same clock value for clock and pingclock [00:43] so how is that doing anything different? [00:44] wallyworld: no we're not... in these tests, PingClock is the test's clock, Clock is WallClock [00:45] wallyworld: sampleConfig() returns a config where Clock is WallClock [00:45] menn0: oh right ok, i didn't realise sampleConfig() returned a clock [00:46] wallyworld: np [00:46] so it's ok to use wall clock [00:46] we werenm'e before [00:46] weren't [00:46] wallyworld: it doesn't matter for these tests [00:46] ok [00:46] wallyworld: the injected clock is used for the code being tested [00:46] wallyworld: the other goroutines can use wallclock and do what they like [00:47] righto, sorry for dumb question [00:47] wallyworld: not at all, glad you're being suspicious [00:50] alexisb: am now [00:51] wallyworld: you happy with these then? [00:51] (that PR) [00:51] menn0: yeah [00:51] i figured you'd hit merge [01:29] wallyworld: FYI capital case should be "Last Connection" not "Last connection" [01:29] IMO [01:30] thumper: it's FirstLetter case [01:30] not CapitalCase [01:30] fugly [01:34] "hi Jay, I missed you today, I love you" [01:34] "mommy I dont love you" [01:34] alexisb: :( [01:34] "you dont, why?" [01:34] "because your working mommy" [01:34] :( [01:34] alexisb: that sucks [01:34] alexisb: last night I got "I hate you daddy, I hate you all and I want to live somewhere else" [01:35] bed time related, not work related though [01:35] ouch :( [01:35] wow [01:35] did you tell her to count to 182? [01:35] bed time is not usually that tough [01:37] heh [01:37] usually, when kids tell their parents "I hate u" they really mean "I hate the power u have over me" ... [01:37] huh, i get "i hate you ian" from everyone all the time [01:38] wallyworld: yup [01:38] u must be very powerful :) [01:39] I don't hate you, wallyworld; I just hate the fact that sensible-browser isn't, and you still use it anyway. :-P [01:40] ah firefox vs chrome, vi vs emacs, star trek vs star wars [01:40] i'd use firefox still if it weren't so slow and bloated [01:41] No, nothing that religious; I was just a bit bemused that they could call something sensible-browser that is so demonstrably not so. [01:42] Bug #1632530 opened: 1.25.6 cannot deploy yakkety lxc units on metal

[02:03] Bug #1632530 changed: 1.25.6 cannot deploy yakkety lxc units on metal

[02:09] Bug #1632530 opened: 1.25.6 cannot deploy yakkety lxc units on metal

[02:12] Bug #1632530 changed: 1.25.6 cannot deploy yakkety lxc units on metal

[02:28] thumper: you might like this one: https://github.com/juju/juju/pull/6432 [02:31] menn0, anastasiamac: I believe you mentioned that Christian was working on the memory leak bug? I saw some fixes from him land so I re-ran the long run test and saw a big difference in the overall memory usage (I have commented on the bug 1625774). [02:31] Bug #1625774: memory leak after repeated model creation/destruction

[02:31] http://people.canonical.com/~leecj2/perfscalemem/ vs http://people.canonical.com/~leecj2/perfscalemem2/ [02:31] veebers: looking [02:32] menn0: +1 [02:32] thumper: thanks [02:42] veebers: \o/ this is awesome!!! [02:43] anastasiamac: it seems that there are a bunch less mongodb actions happening as well [02:45] veebers: yeah both memory and mongodb graphs look neater \o/ [02:46] veebers: m sure alexisb would love this too :D [02:46] yay! [02:47] wallyworld, it works [02:47] thanks [02:47] I will report to aaron in the morning [02:47] awesome [02:47] thanks for testing [02:48] i knew it would work :-D [02:48] wallyworld, well really you tested I pressed buttons [02:48] but you are welcome :) [02:48] with that all I am out [02:48] alexisb: you definitely saw a message saying an update occurred? [02:48] yep [02:48] good :-) [02:49] wallyworld, ding [02:49] ding [02:49] ding [02:49] ;) [02:49] dong [02:51] thumper, wallyworld: huh, this isn't good: http://reports.vapour.ws/releases/4474/job/run-unit-tests-xenial-s390x/attempt/667 [02:52] ../../../golang.org/x/sys/unix/flock.go:18: undefined: Flock_t et. al about 5 undefined in total [02:52] ruh roh [02:53] :( [02:53] filing a bug now [02:54] just in time for GA [02:54] wonderful [02:54] Bug #1632483 changed: juju says error: No selected controller. when there is a typo in the -m parameter it should say "could not find specified model" [02:57] fyi bug 1632541 [02:57] Bug #1632541: Build fails on s390x due to undefined variables

[03:01] thumper, wallyworld we need to get on veebers bug asap ^^^ [03:03] * wallyworld is finishing addmodel upgrade steps, after that i can look [03:03] thanks wallyworld [03:04] wallyworld, thumper: should just be a godeps update to newer version of golang/x/sys [03:05] ok, ta [03:07] * natefinch goes back to twiddling with simplestreams [03:12] Bug #1629817 changed: [arm64] bad cpu-cores detection

[03:15] * redir eod [03:16] anastasiamac, thumper or wallyworld : https://github.com/juju/juju/pull/6433 [03:17] oooh, a test fix [03:18] nice [03:19] alright all, I am eod [03:20] see y'all tomorrow === thumper is now known as thumper-afk [03:46] menn0: not sur eif you have a moment for a small review. similar code to existing upgrade steps https://github.com/juju/juju/pull/6435 [03:58] or anastasiamac ? ^^^^ [04:13] wallyworld: give me a sec and i'll do it [04:13] ty [04:19] menn0: and also a one liner please https://github.com/juju/juju/pull/6436 [04:19] nfi how such an old version of that dependency go committed :-( [04:22] wallyworld: it probably hasn't changed since 2015 [04:22] wallyworld: ship it [04:22] ty [04:22] menn0: i think it was a recent dep addition [04:22] wallyworld: ok well that's dumb then [04:22] whoever did it just used an outdated version perhaps [04:24] menn0: i checked the commit history of that golang project and s390 was a recent addition, so i *think* it will fix it :-) [04:24] wallyworld: seems plausible [04:24] certainly worth updating [04:24] if we had wanted to use that package before then we would have been screwed [04:29] wallyworld: other one reviewed. a few suggestions but looks fine. [04:30] menn0: ty, appreciate it, will fix the issues [04:31] menn0: i see you suggestions - that will teach me just to copy and tweak existing code without thinking too hard [04:31] wallyworld: haha [04:31] too much to do, sigh [04:31] I think we've all done it [04:31] menn0: also did you se email about moving tech board meeting forward 2 hours? [04:32] wallyworld: yeah, I was just about to write about that [04:32] wallyworld: works for me but I don't know about the jam [04:32] it *should* do, it's his 10am [04:33] wallyworld: jam has been idle for over 41 hrs... I wonder if he's on leave [04:33] might just be me and you :-) [04:33] which kind of defeats the point [04:33] * menn0 checks calendar [04:33] yeah [04:33] although we do need to discuss at least the add-model change [04:34] wallyworld: nothing in the calendar [04:34] i did check yesterday i now recall [04:34] wallyworld: there's certainly important things to discuss [04:35] we can make a start even if he is not there. he may come on later for his SOD [04:38] wallyworld: ok so meet 2 hours earlier then? [04:38] yep, let's [04:38] wallyworld: ok [04:39] wallyworld: I'll reply to that email in case John notices that. [04:39] sgtm, ta [05:29] menn0: replied [05:29] will remove critical logging :) [06:00] menn0: I just don't chat as much, just Lurk. :) [06:00] menn0: I'm there now, though it seems wallyworld is away again? [06:03] jam: he emailed to say he had to pick up his kid. he'll be back at 20 past. [06:03] jam: delay until then? [06:03] menn0: either way. I don't know what your day is like. Middle of my day so anytime is fine. [06:03] works better for me to delay as kids are going mental right now [06:04] can help my wife [06:04] for a bit [06:04] jam: see you in 15mins [06:22] menn0: her enow === frankban|afk is now known as frankban === thumper-afk is now known as thumper [07:45] oh... you had the tech board meeting well before what the calendar said [07:45] oh well [08:06] frobware: ping [08:12] voidspace: pong [08:12] voidspace: this is happening because it is latest+greatest lxd (2.4.x) [08:14] frobware: is that right? ah, damn - it means it might not fix the vsphere issue at the same time [08:14] frobware: does the proposed fix work for lxd < 2.4 ? [08:15] voidspace: trying now [08:15] heh, I could try it too [08:15] voidspace: well... just started looking at this so my hypothesis may turn out to be wrong [08:16] o/ [08:16] are we expecting babbageclunk? [08:17] thumper: as far as I know, yes [08:19] mgz: ping [08:20] frobware: works fine for me so far with lxd 2.0 (xenial default) [08:22] frobware: my containers (controller plus two units) have started fine with IPv4 addresses and are active [08:22] frobware: so the config change hasn't broken the "default case" [08:22] voidspace: which bug did https://github.com/juju/juju/pull/6430 fix? [08:22] frobware: I'm not sure there's a specific bug to track this [08:23] voidspace: my mistake: which bug did https://github.com/juju/juju/pull/6338 fix? [08:24] frobware: ah, looking [08:24] frobware: https://bugs.launchpad.net/juju-ci-tools/+bug/1624495 [08:24] Bug #1624495: operations fails on rackspace because of ipv6 address in dns-name

[08:25] frobware: and it seemed to fix that bug [08:25] frobware: that pull request should ensure we get IPv4 *if* we have an IPv4 address that matches the requested scope (public) [08:26] frobware: it won't help if we have a public IPv6 address and only a cloud local IPv4 [08:26] frobware: which I assume is what's happening [08:26] voidspace: so in this case the defaults [auto] for IPv6 is enabled when running lxd init on lxd 2.4 [08:26] frobware: we can tweak that code to prefer a cloud local (fallbackScope) IPv4 over a public IPv6 if that's what we want [08:26] frobware: and that's new in 2.4? [08:27] voidspace: looking at the steps in: https://bugs.launchpad.net/juju/+bug/1632440, I would say yes. [08:27] Bug #1632440: rc3, lxd 2.4.1, 16.04/16.10 use ipv6 if configured via lxd init

[08:28] voidspace: the question I have (and I think rick_h_ too) is why didn't our 'prefer ipv4 over ipv6' choose the former? [08:28] frobware: see my comment above [08:29] frobware: that pull request should ensure we get IPv4 *if* we have an IPv4 address that matches the requested scope (public) [08:29] frobware: it won't help if we have a public IPv6 address and only a cloud local IPv4 [08:29] frobware: we can tweak that code to prefer a cloud local (fallbackScope) IPv4 over a public IPv6 if that's what we want [08:29] I've added this as a comment to the bug [08:29] voidspace: if we did as you suggest wouldn't that a) be better and b) catch other cases like this? [08:30] frobware: it would be better if we were happy to return a cloud local address when a public one is requested [08:30] hmm [08:30] frobware: I can prepare a PR with that change [08:30] frobware: can you repro the bug? [08:30] voidspace: was about to try [08:30] frobware: you could test my change to see if it fixes it [08:31] voidspace: ok, let's do that. if you prepare a patch, I'll look into getting a machine with lxd 2.4 on it [08:31] frobware: cool [08:34] voidspace: I cannot repro with 2.0. I changed my local LXD setup to have IPv6 so not sure why we run into this with lxd 2.4 [08:35] frobware: *sigh* [08:35] voidspace: I really want to understand the 2.0 case with IPv6 enabled first. [08:40] voidspace: bbiab (10 mins) [08:58] voidspace: oh... so the reason I cannot make it fail with lxd 2.0 on my machine is I disabled IPv6 for some other bug... :) [08:58] frobware: yay. The future is here :-) [09:00] voidspace: bug #1614953 [09:00] Bug #1614953: hw csum failure when IPv6 interfaces configured in netdev_rx_csum_fault+0x38/0x40

Yakkety):In Progress by jsalisbury> [09:00] dooferlad: ^^ :) [09:01] dooferlad: most of the time I got the csum failure and occasionally my machine panic'd. so I disabled the whole shebang. I did leave myself a note in /etc/sysctl.conf. [09:02] * frobware tries to recall what he was doing... [09:07] mgz: ping? [09:07] * dooferlad tries to think through the haze of fluffy cold brain. Stupid viruses - we have a release! [09:08] an ultra-small PR to fix a bug (8 character change). could someone review please? https://github.com/juju/juju/pull/6438 [09:09] rogpeppe: looking [09:09] voidspace: thanks [09:10] rogpeppe: hah, LGTM [09:10] voidspace: ta! [09:11] I was too slow to review, that was a great one! :-D [09:11] oh dammit, it causes tests to fail! [09:12] macgreagoir: great but flawed! [09:13] Will I look more sensibly slow when it breaks the tests? [09:14] macgreagoir: showoff [09:15] :-D [09:16] macgreagoir: this fixes the fix: https://github.com/juju/juju/pull/6438/files [09:18] *coughs* LGTM [09:24] rogpeppe: too long now, I'm withdrawing my approval. [09:24] I've got a dependency update PR! It's short! https://github.com/juju/juju/pull/6439 [09:28] * babbageclunk doesn't really withdraw his approval it was just a joke. [09:30] babbageclunk: lgtm, fwiw [09:31] macgreagoir: Thanks! You're a graduated reviewer in my heart! [09:31] I believe babbageclunk is already drinking in an NZ timezone. [09:32] * rogpeppe never "graduated" as a juju reviewer [09:37] voidspace: can repro with LXD 2.0 and -rc3 now: http://pastebin.ubuntu.com/23312073/ [09:37] voidspace: so baby steps. trying tych0's patch [09:41] macgreagoir: your patch is tremendously helpful. http://pastebin.ubuntu.com/23312086/ [09:41] macgreagoir: we *must* land this for GA [09:42] voidspace, dooferlad, frobware can I get another +1? Or rogpeppe, even though you're not graduated? :) [09:42] frobware: You can only blame yourself! :-D [09:42] voidspace: ^^ that's LXD 2.0 with tych0's patch [09:42] * macgreagoir is working on it [09:45] voidspace: I don't see tych0's patch having a positive impact on LXD 2.0... double-checking... [09:52] voidspace: and that's because it only looks for the config enabled in the new style networking. [09:57] babbageclunk: I previously ran into build issues when bumping the lxd dependencies - mgz has the detail. [09:58] frobware: Hmm. Good point - I'll confirm with him. [09:58] frobware: Maybe this will be better since there aren't API changes? [09:59] babbageclunk: added to the PR comment. but I think the issues are when they build the .debs. [10:00] frobware: does my PR help? [10:04] voidspace: let me load that on top. [10:05] voidspace: looking at your PR - this is quite a change "cloud local is picked over hostname" [10:12] I can't raise mgz - frobware do you know whether he's around today? Or is he just head down on release stuff? [10:12] babbageclunk: given the release you might find he's starting his day later for overlap. [10:12] Ah right. [10:13] frobware: yes it is quite a change [10:13] frobware: does it work? [10:13] voidspace: for a sample size of 1. :) [10:14] voidspace: just looking at adding a model, adding 10 containers, verifying no IPv6, destroy model. Lather. Rinse. Repeat. [10:14] frobware: cool [10:15] frobware: it does mean that where we would have got a public hostname we can now get a cloud local IP address [10:15] frobware: I could fix that with a more complex change [10:15] but maybe we would usually prefer the IP address? [10:15] voidspace: unless I screwed up... http://pastebin.ubuntu.com/23312299/ [10:16] frobware: so I want to know (from the logs - probably need INFO or TRACE level) what IP addresses those machines have [10:16] voidspace: looking [10:16] frobware: I'm pretty confident the address picking code works, so the problem is more likely to be that they *only* have IPv6 addresses [10:16] voidspace: your branch has only 1 commit, correct? [10:16] correct [10:17] the production change is trivial [10:17] voidspace: I see IPv4 and 6. http://pastebin.ubuntu.com/23312308/ [10:22] If anybody has worked on status output, I'm looking for where indentation is specified, please. [10:25] frobware: right, but what does the machine agent see [10:26] frobware: are they still showing ipv6? [10:26] frobware: in juju? [10:26] voidspace: http://pastebin.ubuntu.com/23312331/ [10:26] macgreagoir: I've worked on that a little bit. Do you mean in yaml serialisation? [10:27] voidspace: just adding the logs [10:27] babbageclunk: I think so. I need add some indentation to a new list in status and show-machine. [10:27] babbageclunk: https://github.com/juju/juju/pull/6424 [10:28] voidspace: http://pastebin.ubuntu.com/23312333/ [10:30] voidspace: the machine-0.log http://pastebin.ubuntu.com/23312338/ [10:31] voidspace: actually that log ^ is of little use [10:31] frobware: I have a failing test case where it is picking IPv6 over IPv4 - debugging [10:32] babbageclunk: Don't let me eat your time, though thanks. I'm just looking for a shortcut under time pressure, but I'm sure I can find it. [10:35] frobware: oh for goodness sake [10:35] frobware: network.NewAddress("10.229.221.170") creates an address with type "hostname" [10:35] voidspace: controller log: http://178.62.20.154/~aim/controller-machine-0.log [10:35] frobware: so then juju treats it with equal priority to ipv6 [10:35] voidspace: ugh? [10:36] frobware: http://pastebin.ubuntu.com/23312355/ [10:36] frobware: I believe this is the cause of the problem [10:36] voidspace: we lack distinction between address and hostname, IMO. [10:37] voidspace: addresses are bytes. should never be strings until you want to print them. [10:37] frobware: well, evidently... ;-) [10:37] voidspace: wow [10:38] what's the status on landing stuff in juju master now? is it supposed to be only critical stuff, or has the release been forked now? [10:39] rogpeppe: my understanding is only critical [10:39] frobware: thanks [10:40] frobware: gah, no - my mistake [10:40] frobware: I gave it the address 10.299.221.170 [10:40] frobware: which is obviously invalid (299) [10:41] frobware: given the choice between 10.199.121.170 and the ipv6 address the address selection does work - even on current master without my change [10:41] frobware: so SelectPublicAddress is not being given that IPv4 address [10:41] coffee [10:42] macgreagoir: I think frobware's comment about the indentation is wrong (no offence frobware!). That's just how a list of items is represented in YAML. You can see the same format in the Projects section on yaml.org. [10:42] voidspace: FWIW I think a string is a perfectly reasonable representation of an address :) [10:42] babbageclunk: fair enough. I just tried various /other/ indenters and they ... indented lists. [10:43] rogpeppe: semantically broken [10:43] voidspace: but i don't think that Type should be part of Address - it should be a method [10:43] frobware: how so? [10:43] rogpeppe: agreed [10:43] rogpeppe: it shouldn't be possible for it to be wrong [10:43] voidspace: it's always possible for it to be wrong, even if it's just a set of bytes [10:44] I meant Type [10:44] voidspace: ah yes, indeed [10:44] rogpeppe: because you only need it as a string for display. the type system should aid us to disntinguish between hostnames and addresses. if we need a hostname (string) then convert. but the I believe the base type of an IP address is []byte. [10:45] really coffee [10:45] babbageclunk: Cheers, I'm not seeing any standard flags to indent either. frobware, maybe this could be a nicety for later if we decide it's better? [10:45] frobware: it's really useful to be able to connect to an address whether it's a DNS name or IP address [10:45] frobware: we do it on the command line all the time [10:46] frobware: given that there's no syntactic overlap between the two, having a single string representation seems good to me [10:46] rogpeppe: sure. it's more about the internal type. [10:46] frobware: why does it matter? [10:47] rogpeppe: ^^ network.NewAddress("10.229.221.170") creates an address with type "hostname" [10:47] frobware: in my view, very little code should need to care about addresses as other than an opaque string [10:47] frobware: it doesn't [10:47] frobware: (i checked) [10:48] rogpeppe: we've also given this problem to the charmers - when we say "address" they say I want to both "address" and "name" [10:48] frobware: ? [10:48] rogpeppe: network-get address [10:49] rogpeppe: it's not clear what that returns. they faff around trying IPaddress? symbolic-hostname? should I always convert? [10:49] rogpeppe: if we had: network-get hostname and network-get address then it would be clearer [10:49] frobware: why should it matter? [10:49] rogpeppe: apparently it matters to the. hadoop for one. [10:49] frobware: I think it's just a neither-a-bug-nor-a-feature of the particular yaml serialiser we use. [10:50] frobware: if you need an ip address and your software doesn't know how to resolve it, then you can resolve it easily [10:51] frobware: FWIW network-get seems to document that it returns an IP address [10:51] rogpeppe: except for the times it returns a symbolic name [10:51]