[00:00] davecheney: is there a particular bug you'd like me to look at next or should I just grab one from the list? [00:11] wallyworld_, thumper: does any of this look familiar and do you have an advice? http://pastebin.ubuntu.com/7224152/ [00:13] sinzui: and juju works fine on hp cloud? [00:16] wallyworld_, yes [00:16] hmmm [00:17] wallyworld_, I just confirmed that both configs use the same keys [00:18] sinzui: in the past, where container permissions have been wrong, the container has been created but subsequent reads failed. here we can't even create the container [00:18] it does seem to imply a canonistack swift issue [00:18] wallyworld_, noted. and in the past creation fails were race conditions. this say auth failure [00:19] sinzui: have you tried the other region? [00:20] wallyworld_, no [00:20] sometimes that can work [00:20] lyc01 vs lyc02 [00:20] wallyworld_, but I just checked the canonistack dashboard for /both/ accounts. The container view shows an error [00:21] hmmm, ok [00:21] and no joy asking in #is? [00:22] wallyworld_, they officially defer to canonical support. I opened a ticket there 10 hours ago and no one will talk to me [00:22] :-( [00:23] I am tempted to send an email notifying canonistack that it will be desupported. Without working accounts, I cannot deliver the next juju to it [00:24] agreed [00:24] we do need an openstack deployment to test against though :-( besides hp cloud [00:25] thumper: this fixes bug 1304132 and also removes the log noise from the critical bug alexis emailed about https://codereview.appspot.com/85770043 [00:25] <_mup_> Bug #1304132: nasty worrying output using local provider

[00:53] arosales: hazmat email sent [00:53] with ppc segfault informatoin [00:54] * arosales looks [00:55] davecheney, k.. trying fresh on new ppc8.. orange box #2 vanquished [00:55] hazmat: that would be a good data point, i only have access to wolfe and winton, which are power7 [00:56] hazmat: if you see panics on your power8 host, you should revert to that kernel I specified [00:56] davecheney, my p7 host has been good wolfe-02.. 3.13-08 [00:57] trying on stilton-5 [00:57] hazmat: yes, that is the working kernel [00:57] it is pre the switch to 64k [00:57] pages [00:57] dfc: so 3.13.0-08.28 is what we need correct? [00:57] hazmat: are you running -08.28? [00:57] the .28 isn't the important bit [00:58] the -08 -18, -23 is [00:58] i've included a link to the old kernel in the archive [00:58] one moment.. switching tracks off maas [01:01] dfc, gotcha, avoid -18 and -23 [01:01] davecheney, so my p7 is -> 3.13.0-8-generic [01:01] davecheney, my p8 is -> 3.13.0-23-generic [01:06] hazmat: right [01:12] davecheney, so.. theory being that's an okay version? .. i'm gonna test and find out either way [01:13] hazmat: i've tested -8, -18 and -23 [01:14] only -8, which was pre the 4k page switch can run juju stabally [01:14] the other kernels radomly kill juju proceses with SEGVs [01:14] solid [01:15] davecheney, cool, thanks for tracking that down.. apparently i lucked into having at least one good demo p machine [01:15] hazmat: yeah me too [01:15] winton-02 is ooooooooooold [01:15] so it was running a very old kernel [01:15] but thumper bodie and timv hit problems [01:17] oh man, 64k pages kill the gccgo runtime? [01:17] somehow that's easy to believe [01:18] mwhudson: yup [01:18] mwhudson: tell me your thoughts [01:18] its signal related [01:19] somehow an invlaid signal is generated, or created, or just pops into existance [01:19] the powerpc/kernel/signal_64.c doesn't know how to handle it, so It calls force_sigsegv [01:19] and the userland thinks it has hit a nil pointer exception and panics [01:19] davecheney: well i think malloc.goc has a #define PAGE_BITS 12 in it [01:19] o [01:19] h [01:19] that sounds pretty messed up [01:19] mwhudson: but why should that matter [01:20] 12 is < 16 [01:20] davecheney: dunno [01:20] but is a multiple [01:20] all that happens is if you call mmap(0, 4096) you get a 64k allocation [01:20] mwhudson: i'm logging this all in a bug now [01:20] then i have a juju test to fix [01:21] then i'll try to create a smaller reproduction case [01:21] there is additional debugging in that file [01:21] but it appears to be turned off in this build [01:21] maybe spinning a new kernel with it enabled is the next step [01:22] wallyworld_: back from the gym now [01:22] davecheney: an invalid signal number is generated? [01:22] wow [01:22] what is the userspace doing when this signal arrives? [01:22] chlling [01:22] thumper: ok. i have 2 fixes for that critical bug https://codereview.appspot.com/85770043 and https://codereview.appspot.com/85750045 [01:23] so it's an async signal? [01:23] not sure if more work is needed [01:23] [18519.444748] jujud[19277]: bad frame in setup_rt_frame: [01:23] 0000000000000000 nip 0000000000000000 lr 0000000000000000 [01:23] [18519.673632] init: juju-agent-ubuntu-local main process (19220) [01:23] killed by SEGV signal [01:23] [18519.673651] init: juju-agent-ubuntu-local main process ended, respawning [01:23] wallyworld_: so what what going wrong? [01:23] thumper: i'll get those landed and will have to either test or ask axw if there's anything else obvious that needs looking at [01:24] thumper: 2 things 1. instance poller noise due to it not ignoring unprovisioned machines [01:24] +1 for that [01:24] 2. bad schema def for storage-port config attr on manual provider causing provisioner startup to fail [01:24] due to json serialisation issue [01:24] float64 vs int and all that [01:25] so those 2 fixes i did just by looking at logs [01:25] i had a look at the code to see if i could relate the fixes to the actual observed issue, but didn't get far enough [01:26] so i figured we could fire up some arm instances and test and/or ask axw for input when he comes online [01:26] I am online [01:26] what input do you need? [01:26] \o/ [01:26] I have LGTM'd your two fixes [01:26] axw: bug 1302205 [01:26] <_mup_> Bug #1302205: manual provisioned systems stuck in pending on arm64

[01:26] ok :-) [01:26] i am not sure if my fixes are sufficient [01:27] they are needed, but is there more to be done [01:27] davecheney, confirmed btw re 23.. panic while doing nothing detected in the log [01:27] hrm [01:27] have you seen similar issues when developing the manual provider? [01:27] wallyworld_: nope [01:28] or maybe we just need to test with the fixes [01:28] could yet be an arm issue i guess [01:28] davecheney: can you run my test program from https://sourceware.org/bugzilla/show_bug.cgi?id=16629 ? [01:28] looking at the logs now... [01:28] hazmat: i've even seen /usr/bin/go panic while running tests [01:28] https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1304754 [01:28] <_mup_> Bug #1304754: gccgo compiled binaries are killed by SEGV on 64k ppc64el kernels [01:28] certainly that storage-port issue is pretty fatal [01:28] arosales: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1304754 [01:28] <_mup_> Bug #1304754: gccgo compiled binaries are killed by SEGV on 64k ppc64el kernels [01:28] wallyworld_: ah, the "close of closed channel" is something rogpeppe brought up last night [01:29] there's a bug in cmd/jujud [01:29] not sure if he fixed it yet... [01:29] (although i can't see why 64k pages would matter here) [01:29] axw: is that in the machine 0 log attached to the bug? [01:29] wallyworld_: yeah [01:29] let me look [01:29] wallyworld_: https://codereview.appspot.com/85450044/ [01:29] wallyworld_: I broke the machine agent when I allowed upgrade steps to get a state connection [01:30] davecheney: also, it would sure be nice to follow the execution of handle_rt_signal64 with gdb [01:30] axw: does that mp fix the close channel issue? [01:31] mwhudson: way above my pay grade [01:31] i'm not even qualified for pointer arythmetic [01:31] wallyworld_: yeah [01:31] davecheney: looks like the latest in the archies is -23 [01:31] yup [01:31] axw: great. so i'll land my branches and we can re-test i guess [01:32] wallyworld_: sgtm [01:32] wallyworld_: it would be nice to silence "cannot get instance info for instance "manual:10.0.128.7": no instances found" too, but it's not critical [01:33] axw: i haven't look into that on yet - what's the cause? [01:33] i wonder if there is an arm64 kernel with 64k pages i can try with [01:33] wallyworld_: manually provisioned machines are not managed by the provider - they just should not be polled [01:33] mwhudson: thta would be a good test [01:34] i tried to test using gccgo/amd64 [01:34] axw: in that case i'll add some code to my first branch [01:34] but lxc was all fucked on amd64 yesterday [01:34] do both fixes in one go [01:34] wallyworld_: there's a state.Machine.IsManual method that'll help there [01:34] i don't know anything about legacy architectures like amd64 [01:38] davecheney: do you have link hand to the matching initrd to the -28 .deb you pointed at? [01:41] arosales, so i removed the other kernels .. sudo upgrade-grub.. currently doing shutdown -r now .. to see if it worked ;-) [01:41] arosales_, removed via pkgs that is === arosales_ is now known as arosales [01:42] where's this -28 kernel at, I don't see it in proposed? [01:42] jcastro, its on the machines that barf.. ls /boot [01:42] jcastro: o/ [01:42] jcastro: I have a version of debug-log on my machine that works with the local provider [01:42] arosales, don't do what i just suggested it.. it doesn't like that ;-) [01:42] thumper, hey! we made a plugin, heh [01:43] yeah, but it doesn't do filtering [01:43] * thumper guesses [01:43] oooh [01:44] or replay or exclude/include by unit/machine or channel [01:44] jcastro, we have parity.. [01:45] * hazmat sheds a tear [01:45] for debug-log [01:45] heh [01:45] thumper, also, one thing we should talk about [01:45] arosales: not -28 [01:45] is the debug-hooks <-> retry --resolved thing makes me cry [01:45] hazmat, ack :-) [01:45] you want uname -a [01:45] Linux winton-02 3.13.0-8-generic #28-Ubuntu SMP Mon Feb 17 08:22:39 UTC 2014 ppc64le ppc64le ppc64le GNU/Linux [01:45] oh _8_ [01:45] wget https://launchpad.net/ubuntu/+source/linux/3.13.0-8.28/+build/5602341/+files/linux-image-3.13.0-8-generic_3.13.0-8.28_ppc64el.deb [01:45] not 28 [01:46] davecheney, so the _only_ fix is to revert to -08? [01:46] jcastro: i'll forward you the email [01:46] davecheney, do I need an initrd for that? [01:46] arosales, atm yes [01:46] arosales: the only workaround I have at this time [01:46] jcastro: no [01:46] davecheney, thanks! [01:46] jcastro: I don't understand [01:46] so I do debug-hooks [01:47] and in order to be able to fire off a hook to debug it [01:47] also, know this [01:47] I can only make one person happy at a time [01:47] I need to open a new terminal and do resolved --retry [01:47] it isn't your turn [01:47] it is hazmat's [01:47] <3 [01:47] local log will keep me happy. :D [01:48] yeah, I'm open to looking to fix it... [01:48] davecheney, I am confused in your email you state, "workarounds: you should install this kernel [01:48] wget https://launchpad.net/ubuntu/+source/linux/3.13.0-8.28/+build/5602341/+files/linux-image-3.13.0-8-generic_3.13.0-8.28_ppc64el.deb" [01:49] davecheney, ah I should have said -8.28 not -28 [01:49] davecheney, gotcha [01:49] with is a revert [01:49] thumper, and all of cts :-) [01:49] sorry long day [01:49] * arosales better just grab some dinner [01:50] arosales: yup, we're also lucky that -28.8 isn't a think [01:51] both those numbers appear to be increasing [01:51] s/think/thing [02:08] c.Assert(err, gc.IsNil) [02:08] ... value *mgo.QueryError = &mgo.QueryError{Code:16149, Message:"exception: cannot run map reduce without the js engine", Assertion:false} ("exception: cannot run map reduce without the js engine") [02:08] store tests are failing again [02:09] i thought that the store tests wouldn't run unless we passed a flag ? [02:09] cmars: didn't you fix this ? [02:09] davecheney, thought so, yes. is this trunk or 1.18? [02:09] cmars: trunk [02:09] hmm [02:10] davecheney, which test is it? is there a file & line #? [02:10] cmars: please hold [02:11] go test launchpad.net/juju-core/store 2>&1 | pastebinit [02:11] Failed to contact the server: [Errno socket error] [Errno socket error] timed out [02:11] oh for fucks sake [02:11] does nothing work today ? [02:13] thumper: what is the env var to lower logging ? [02:13] JUJU_LOG= ? [02:14] here's mine: JUJU_LOGGING_CONFIG==INFO; juju.container=TRACE; juju.provisioner=TRACE [02:14] that is ready by bootstrap [02:14] ta [02:14] hnn, that isn't it [02:14] rog had a different one [02:14] a flag to testing [02:15] -juju.log WARNING [02:17] cmars: http://paste.ubuntu.com/7224466/ [02:17] ok, thanks. looking [02:19] ta [02:22] davecheney, i thought it had landed, but it hasn't [02:22] https://code.launchpad.net/~cmars/juju-core/cs-mongo-tests/+merge/213563 [02:23] i think we can land it, if CI will support running the store tests with full mongodb tests [02:25] davecheney, what do you think? will you take that as an action item? [02:25] cmars: no [02:25] ok :) [02:25] i cannot take that as an action item [02:26] i'll follow up w/curtis tmw then [02:26] cool [02:26] cheers [03:19] arosales: do you have any doc or otherwise that tells me what i need to do get access to some arm vms to test a fix for that bug [03:27] wallyworld_: i can help w/ that [03:27] \o/ [03:27] wallyworld_: do you have an account on batuan? [03:27] that's the gateway into our network - if you don't, you can ask for one in #is [03:27] yes, since i have logged onto power vms previously [03:27] sweet [03:28] i think you know it's bug 1302205 [03:28] <_mup_> Bug #1302205: manual provisioned systems stuck in pending on arm64

[03:28] right [03:28] i don't think it's specifcally an arm issue [03:28] but want to test on arm anyway [03:28] yep - just a sec... wonder if someone just took our host down [03:28] there have been 3 branches which landed today or last night which should hopefully fix it [03:34] wallyworld_: yeah, looks like its in use debugging an unrelated issue. i'll send you an e-mail with access info and ping you (or have someone ping you) when its ready [03:34] great thanks :-) [03:36] davecheney: how can I make sure that the bufio.Scanner doesn't consume too much? [03:36] davecheney: I have an io.ReadCloser [03:36] davecheney: and I want to read up to the first new line, and no more [03:36] the Scanner when I call scan reads 4k [03:36] which consumes way more than I want [03:45] wallyworld_: e-mail sent - system isn't quite ready yet, stay tuned.. [03:45] ok [03:51] dannf, thanks re wallyworld arm access question :-) [03:53] np, very happy to have help w/ it :) two days and i've just managed to learn how to kinda read go :/ [04:06] wallyworld_: ok, have at it [04:06] dannf: awesome, firing up ssh now [04:07] dannf: "ssh 10.229.41.200" should work right? [04:07] it should, but its not working for me either - lemme ask [04:40] is that because the ssh config can't work out where to proxy through? [04:42] thumper: fyi I filed a merge request for the changes you suggested yesterday about isolating the git tests [04:42] jcw4: awesome, I saw that in my inbox [04:42] jcw4: I'll take a look once I submit this change :-) === vladk|offline is now known as vladk [04:43] thumper: thanks; no rush... just excited about contributing ;) [04:46] wallyworld_: fyi instance type constraint branch has conflicts [04:46] yes it does, fixed loclly [04:46] still wip [04:47] jcw4: did you push your changes? [04:47] yes; to a new branch [04:47] the last one was too messy [04:47] :-) [04:48] jcw4: ok, there is a resubmit option on the RHS of the merge proposal page, that includes a "start over" [04:48] which would have marked the old as superseded [04:48] but that's OK, I'll just reject the old one. [04:48] I see. Thanks [04:50] jcw4: what happens when you change the LC_ALL to "C" ? [04:50] I was planning on doing that after running all the tests without it. [04:50] after they all passed I was too excited and forgot [04:50] testing now [04:52] thumper: worker/uniter/charm/... tests passed [04:52] I'll push that change too? [04:52] move that patch env into the base git test suite [04:52] with the other env patches [04:53] jcw4: then you can delete the SetUpTest for GitDirSuite [04:53] cool, right [04:53] as it won't be doing anything [04:53] then yes, push that [05:03] thumper: the LoggingSuite TearDownTest(c) needs to be called in the GitSuite TearDownTest? I'd add that back in if necessary [05:03] jcw4: if the only line of the tear down is to upcall the tear down, then you can just delete it [05:03] thumper: okay; that's all there is. tx [05:15] * thumper EODs [05:20] wallyworld_: I've responded to your comments, but I'm now looking at HA [05:20] np [05:21] just wanted to get some thoughts down [05:21] i'm stuck on otherthings also [05:33] hi all, I am using 1.16.6 stable juju-core to bootstrap an Openstack Havana cloud, but fails with can't find index.json [05:34] it seems that juju is trying to find the meta file in the path /streams/ but swift has tools/streams/ [05:41] morning all [05:57] fwereade, can you take a look at this please ? https://codereview.appspot.com/85220044/ [06:51] dimitern: howdy! How's the vlan work coming along? [06:51] hey bigjools [06:52] bigjools, i'm in the final steps - cloudinit scripts that bring up network interfaces [06:53] bigjools, vladk is working on a few extensions to gomaasapi to allow us to unit test the new api calls [06:53] bigjools, capabilities; lshw dump of a node; networks?op=list_connected_macs [06:54] dimitern: nice, all going to make it for the release? And any issues with maas I need to know about? [06:54] bigjools, but all these were live tested on my local maas using daily builds ppa [06:55] bigjools, we're aiming for feature completeness by friday, but should be ready before that [06:55] excellent [06:56] bigjools, bug 1303617 hit me after a recent upgrade and i can no longer use the fast installer (fails at boot and doesn't recover), which is slow and tedious [06:56] <_mup_> Bug #1303617: pc-grub install path broken in curtin [06:56] dimitern: weird, I did a fast install today and it was fine [06:56] hmm Fix Released - i'll try it now [06:57] bigjools, we have a few wishlist items for the maas api [06:58] bigjools, like the ability to see networks + connected macs in one place (either in GET node/system_id or in GET networks/(all)) [06:59] dimitern: please file bugs [06:59] bigjools, will do [06:59] I will triage them as wishlist and we'll put them on the stack [06:59] bigjools, otherwise now we need to do several api calls at startinstance time to get all we need [06:59] ok we can optimise that [07:10] cmars: I'm not sure why your test failed, but it would seem that we could tell the landing bot to always run the mongojs tests [07:10] cmars: though because of that, I'd actually rather have the CI tests disable it, rather than disabled by default. [07:11] experience has shown that ENV vars play nicer with go test than flags, because flags are only valid per package, and "go test ./..." tries to pass all flags to all packages. [07:11] bigjools, filed bug 1304857 [07:11] <_mup_> Bug #1304857: API should report networks and connected macs in the response of a single node [07:12] jam, fancy a review ? https://codereview.appspot.com/85220044/ [07:15] bigjools, another question re gomaasapi - how do you feel about adding high level wrappers around common APIs? like having AcquireNode method that the provider calls, rather than constructing a URL internally? Other similar examples are ListNetworks, ListNodes, etc.? === vladk is now known as vladk|offline [07:20] dimitern: I can't remember much about gomaasapi [07:22] bigjools, :) well, that was just a thought [07:22] dimitern: one of the other guys will have an opinion I'm sure [07:23] bigjools, we'll ask them for reviews when changes are proposed === BjornT_ is now known as BjornT [07:30] dimitern, I'm getting progressively more nervous about NetworkName vs making it clear that it's a provider-specific id like instance.Id [07:31] fwereade, can't we say yes, it's provider specific, but it's also used by juju to identify the network internally? [07:31] dimitern, it will be, indeed [07:31] dimitern, but we're going to want network names as well [07:31] fwereade, ok, how are we going to make it clearer? [07:32] dimitern, when openstack gives us network abcdef638746328756865198, and we call that NetworkName, what field will we use for the "private" name users will want to use [07:32] dimitern, or "my_network" or whatever [07:33] fwereade, openstack has labels for networks just the same [07:33] dimitern, and so does every provider ever? [07:33] dimitern, that's quite the prediction ;p [07:33] fwereade, i can't say that :P [07:34] fwereade, so tell me how to alleviate your nervousness about it? :) [07:34] dimitern, call it NetworkId :) [07:35] dimitern, you know -- we have machine ids, and instance ids, and they are not the same [07:35] dimitern, (and machine ids are machine names really, but hysterical raisins) [07:36] fwereade, so, basically change it everywhere from NetworkName to NetworkId ? [07:36] fwereade, I need a follow-up for that [07:36] dimitern, I'm more concerned about the API [07:37] fwereade, you're thinking about network tags? [07:37] dimitern, and that the terminology that's hard to change be consistent with what we expect to do [07:37] dimitern, well, that was my first thought [07:37] dimitern, but then I realised that converting these names into tags would be completely wrong [07:37] mornin' all [07:37] morning rogpeppe [07:37] dimitern, because they're provider vocabulary, not juju vocabulary [07:37] dimitern: hiya [07:38] fwereade, ok, so then what? [07:38] fwereade, i'm trying to follow but can't see what's needed [07:38] axw: ping [07:38] dimitern, although -- wait, don't you use tags in the client api? I think we should... [07:38] rogpeppe: pong [07:39] fwereade, we use tags everywhere in the api [07:39] fwereade, but not for networks [07:39] axw: about removing JobManageEnviron: [07:39] * fwereade grumbles [07:39] axw: the reason we don't want to remove JobManageEnviron from a voting state server is that when a machine hasn't got JobManageEnviron we allow it to be removed [07:40] dimitern, we don't identify machines in the client API by provider-specific instance id, and we shouldn't identify networks that way either [07:40] axw: and if that happens we could break the invariant that we only ever have an odd number of voting state servers [07:40] axw: or rather, and odd number of state servers that *want* to vote [07:40] fwereade, i agree, but the only way we can deal with networks so far is if we get them from the provider [07:40] fwereade, at provisioning time [07:40] s/and odd/an odd/ [07:41] dimitern, we *can* impose a requirement that network names match provider ids exactly, this is MVP after all [07:41] fwereade, i guess you're suggesting to require the user to add any networks to juju before being able to deploy with them [07:42] fwereade, and i can see how this is the way we wanna go eventually, but not for nwo [07:42]