[00:18] thumper: yo [00:18] thumper: you were helping a guy who had his juju upgraded from 1.18 to 1.19. Trying to get him to 1.20. I've been trying to replicate things locally [00:19] thumper: I've got 1.19.3 bootstrapped .. but how do I make it think it is 1.18 so the juju upgrade-juju of 1.20 will proceed? [00:26] wwitzel3: i don't think you can - the version number is obtained via version.Current which is compiled into the executable [00:26] oh, wait, the db agent-version needs to be updated i think [00:27] but doing that will trigger downgrade [00:27] the version.Current and agent-version in db need to match, or else juju willtryand make them match [00:28] so you may need to hack 1.20 to run upgrade steps for 1.19 rather than 1.18 [00:31] axw: thumper suggested I ask you to review https://github.com/juju/juju/pull/450 [00:32] axw: It's one half of the issue raised in https://bugs.launchpad.net/juju-core/+bug/1351089 [00:32] <_mup_> Bug #1351089: Isolation failure in sshstorage test [00:32] axw: I have another PR coming soon that addresses this specific isolation error, and a handful of others too, by supressing the load of a users custom BASH_ENV file [00:33] axw: but I thought that this PR was a valid change on it's own merits [00:39] sinzui: i remember why the lock dir is not removed - i read that removing the file could potentially cause issues (i think if other processes held the lock, not sure now). so that's why it is left behind. so i've made it so that bootstrap and destroy remove any lock files at that point [01:15] jcw4: will take a look in a bit, thanks [01:28] wwitzel3: hey [01:28] wwitzel3: I *think* you just need to do the upgrade, then change the agent version in the agent configuration files on the various machines [01:28] wallyworld: hey [01:33] davecheney: care to look at https://github.com/juju/names/pull/22 [01:33] wallyworld: around? [01:38] thumper: /me looks [02:01] thanks axw [02:01] general question... how do I know when CI is ready to start accepting normal merges again? === Beret- is now known as Beret [02:04] jcw4: you'd have to check the bugs I'm afraid [02:04] they're tagged with "regression" [02:05] jcw4: https://bugs.launchpad.net/juju-core/+bugs?field.searchtext=&orderby=-importance&field.status%3Alist=NEW&field.status%3Alist=CONFIRMED&field.status%3Alist=TRIAGED&field.status%3Alist=INPROGRESS&field.status%3Alist=INCOMPLETE_WITH_RESPONSE&field.status%3Alist=INCOMPLETE_WITHOUT_RESPONSE&field.importance%3Alist=CRITICAL&assignee_option=any&field.assignee=&field.bug_reporter=&field.bug_commenter=&field.subscriber=&field.structural_sub [02:07] awesome, thanks axw and davecheney [02:11] thumper: I cannot see an item on the N sprint for addressing the rquirements to support Trusty's 1.18 client [02:12] davecheney: I'll add one [02:12] thumper: thanks [02:12] thumper: for example, https://github.com/juju/juju/pull/449#issuecomment-50825367 [02:18] davecheney: does 1.18 CLI still dial mongo directly? I thought 1.16 was the last to do so [02:20] looking at the 1.18 branch, looks like they all use API [02:20] axw: cool [02:20] i don't know the answer, that is why I asked [02:23] how long are state tests supposed to take? [02:23] * hazmat realizes its been a while [02:23] hazmat: I think it takes 60s-80s on my laptop with SSD [02:24] scratch that, with tmpfs [02:25] hazmat: yup, that is how long it takes [02:25] apiserver also take ~70 s for me [02:26] hmm.. i haven't hit a failure but i am hitting timeouts with tokumx on ssd.. was running with extra verbose those (-gocheck.vv -v) [02:26] axw: are all the build blockers related ? [02:26] 10m and the test gets killed [02:26] hazmat: then the test fuked up [02:27] it happens [02:27] davecheney: nfi [02:27] axw: someone suggested at the team meeting they were all the same underlying cause; mongo startup failures [02:27] hazmat: is tokumx cleanroom, or based on some version of mongo? if the latter, which version? [02:27] davecheney: the azure one certainly is mongo [02:27] axw, current release is against 2.4 [02:28] axw, but its a mvcc db under the hood.. ie. always journaled [02:28] ok, just asking cos I've been trying to get things working with 2.6 and have had a bunch of issues [02:29] i'm wondering if the mvcc stuff is causing some additional time consumption due to write flushes [02:29] I would hope it wouldn't add *that* much [02:30] how do you control transaction isolation? [02:30] axw, db.beginTransaction(); db.commitTransaction(); db.rollbackTransaction() [02:31] axw, http://docs.tokutek.com/tokumx/tokumx-transactions.html [02:32] axw, actually better http://docs.tokutek.com/tokumx/tokumx-commands.html#tokumx-new-commands-transactions [02:33] cool [02:33] are folks using go 1.3 for dev ? [02:33] I am [02:33] i use trunk [02:33] cool, i am as well, just wanted to double check i had some issues with the rcs and juju [02:34] rsc ? [02:34] rcs ? [02:34] davecheney, release candidates.. http client errors on writes to s3 [02:35] when uploading tools, i would get EOF often [02:35] ok [02:36] anyone know if work has been committed on the api servers not being added to .jenv files until running juju status the first time? [02:36] i looked through the commit logs but didnt see anything [02:36] the state server i should say [02:36] stokachu: it's waiting to land now [02:36] davecheney, ok cool, thanks [02:37] davecheney: they'll be stored by the bootstrap command? [02:37] axw: waigini knows [02:37] but yes [02:37] cool [02:37] axw: I think the failure on precise is because the cloud archive isn't being added [02:38] Setting up rsyslog-gnutls (5.8.6-1ubuntu8.6) ... [02:38] Setting up libsnappy1 (1.0.4-1build1) ... [02:38] Setting up mongodb-clients (1:2.4.6-0ubuntu5~ubuntu12.04.1~juju1) ... [02:38] Setting up mongodb-server (1:2.4.6-0ubuntu5~ubuntu12.04.1~juju1) ... [02:38] Adding system user `mongodb' (UID 107) ... [02:38] ^ this isn't the juju mongo [02:38] it's the system one [02:38] no [02:38] wait [02:38] I don't think we use juju-mongodb on precise [02:38] it _is_ the juju one [02:38] axw: we _have_ to [02:38] precise is ooooooooooooold sauce [02:38] there is logic in cloudinit that does the switch [02:38] something like mongo.*Series [02:39] I don't think it *exists* in precise. I can't remember the logic... will have to dig. but anywya, that hasn't changed AFAIK [02:39] trusty+ uses juju-mongodb, everything else uses mongodb-server. it may be that it's meant to come from cloud-tools tho [02:39] 2.0 is in precise [02:40] anyway [02:40] probably a red herring [02:40] just looking for somehting I can help [02:43] davecheney: wonder if your email is due to https://github.com/juju/txn/pull/6 [02:43] rick_h__: nope [02:43] davecheney: ok, nvm then [02:43] this is something stupid we did to ourselves back in LV [02:44] juju-local depends on juju [02:44] and you can't run the tests without juju-local [02:44] so even in an isolated environment you need to install juju [02:44] then carefully ignore it [02:44] davecheney: ah, gotcha. Ok, just saw testing/dep and juju after seeing this pr today so wondered [02:44] rick_h__: -ENAMEINGOVERLOAD [02:44] lol === menn0-afk is now known as menn0 [02:49] wallyworld: hey, i just pushed up a new round of changes. also exporatory. some tests aren't passing, but i'm not sure if they're transient. don't have time to verify tonight. [03:16] davecheney: added a sprint session to talk 1.18 support, see the bottom of the list [03:17] thumper: thanks [03:17] * davecheney lunch [03:32] thumper: can I get your advice on something [03:32] ? [03:41] wallyworld: thanks for the followup email [03:43] menn0: sure [03:44] thumper: I'm working on adding the ability to roll back the agent version to force a downgrade [03:44] ok [03:44] the upgrader had to be changed because it doesn't normally allow downgrades - that's done [03:44] ok [03:44] but now I'm at the part where I need to set the actual agent version back in state [03:45] * thumper tries not to be distracted by the thought of coffee [03:45] ok [03:45] and state.SetEnvironmentAgentVersion checks to make sure the new version makes sense [03:46] hang on... [03:46] ok [03:46] so one of the checks is that all the machines and agents are starting out at the same version [03:47] and in this case they might not be and that's actually ok [03:47] we want to roll back to last good version anyway [03:48] can I suggest that I make my coffee and then we have a hangout? [03:48] ok sounds good [03:48] coffee wins :) [03:48] what we should identify is all the places in state where we record the agent version [03:48] and also where on disk this is stored [03:48] and at which part of the process the various places change [03:48] so we know what is considered a valid rollback version [03:49] * thumper goes to make coffee [03:50] looking more closely I actually think I don't need to worry. would be good to discuss quickly anyway. [03:55] so, should I not even try to $$merge$$ until the regressions are fixed? [03:56] jcw4: correct [03:56] axw: this is the other half of the test isolation fix: https://github.com/juju/juju/pull/454 [03:56] thumper: thanks [03:57] jcw4: can we just use IsolationSuite instead? [03:58] or OsEnvSuite [03:58] directly [03:58] axw: I actually did that first, but it didn't fix the issue and the test I was testing on needed more refactoring to use it, which I wasn't 100% sure on... I can try again [03:59] axw: I'll do that again and bug you with questions if I have them :) [03:59] no worries [03:59] thanks [03:59] * axw wanders off to get some lunch [03:59] bbs [04:00] menn0: https://plus.google.com/hangouts/_/canonical.com/upgrades?authuser=1 [04:00] "could not start because of an error" [04:00] thumper: ^^ [04:01] * thumper sighs [04:01] menn0: you start one then [04:01] * thumper closes the hangout [04:01] thumper: k [04:01] axw: one point, this one line fix to suppress BASH_ENV in the tests fixes 4 or 5 tests for me. It seems helpful to merge this change *and* refactor the specific test we initially discussed [04:02] thumper: happened again as soon as you answered. killing my chrome now. [04:02] * thumper didn't answer [04:02] thumper: ok. well it keeps happening. even after restarting chrome. [04:03] * menn0 sighs [04:03] firefox? [04:03] menn0: do you have two google identies ? [04:03] i often have to add &authuser=1 to any link I open to make it use the right identify [04:03] (╯°□°)╯︵ ┻━┻ ︵ ╯(°□° ╯) [04:04] davecheney: i have 2 identity but it usually "just works" [04:04] * menn0 tries firefox [04:04] menn0: maybe this is a time when it doesn't [04:05] did the separate logs for each machine go away in juju 1.20.x? [04:05] FU22FA... [04:05] * menn0 goes to retrieve his phone from child [04:06] stokachu: no... [04:06] stokachu: local provider? [04:06] interesting, only machine-0.log is being created [04:06] thumper, yea local provider [04:06] stokachu: multiple local provider environments? [04:06] machine-0.log shows all the logs from each machine correctly [04:06] * thumper recalled a bug [04:06] thumper, nah a single local provider with kvm as the container [04:07] hmm [04:07] ah [04:07] kvm log files aren't mounted like the lxc ones because they can't [04:07] should still be an all-machines.log though [04:07] yea there is, but the nested lxc machines aren't having their logs created either [04:08] i guess is that what you meant by 'multiple' providers [04:08] no... they don't [04:08] no... [04:08] kvm machines used to create machine-x.log files [04:08] the unit logs aren't being created either if that helps [04:09] stokachu: what does juju status say ? [04:09] thumper: https://plus.google.com/hangouts/_/gqu6sqg3xlszyfsl2u4blczvuaa?hl=en-GB [04:09] have any units/services been deployed ? [04:09] http://paste.ubuntu.com/7920401/ [04:09] menn0: I get the party is over :-| [04:09] yea, a bunch [04:10] weird [04:10] thumper: trying again... at least it ain't crashing in FF [04:10] thumper: https://plus.google.com/hangouts/_/gqu6sqg3xlszyfsl2u4blczvuaa?hl=en-GB [04:10] all-machines.log ca-cert.pem machine-0.log rsyslog-cert.pem rsyslog-key.pem [04:10] those are the only files in .juju/local/log [04:10] menn0: july 28 is 2 days ago, not 2 weeks ago :-P [04:10] stokachu: yes, that is right [04:11] menn0: try this one https://plus.google.com/hangouts/_/gvvbftptty735axcajn45l2efia?hl=en-GB [04:11] arg [04:11] now I get error [04:11] yea but there should be more of them :) [04:11] stokachu: no [04:11] stokachu: there shouldn't [04:11] wallyworld: 4-5 ish days, but yes [04:11] stokachu: it doesn't work that way [04:11] wallyworld: I screwed that up [04:11] huh [04:11] then why the did logs used to be there [04:12] for lxc they were [04:12] but kvm, never were [04:12] lxc containers share a log mount point [04:12] jcw4: yeah, seems like it'll be too error prone to get everyone to use IsolationSuite - and it may not be viable in some cases [04:12] * menn0 needs more sleep [04:12] kvm can't [04:12] thumper: try calling me when you're ready [04:12] but i have lxc containers deployed and there aren't logs [04:12] but you lxc containers are inside the kvm containers [04:12] not sure what you are expecting there [04:12] and im pretty sure i had multiple machine logs with kvm [04:13] if you look inside all-machines.log, you should see both machine and unit logs [04:13] if you don't you have a problem [04:13] all i know is juju used to create unit-x.logs and machine-x.logs with this same setup [04:13] also visible with 'juju debug-log' [04:13] and now they dont [04:13] no [04:13] it didn't [04:13] never with kvm [04:13] trust me on this [04:13] I wrote it [04:14] kvm can't [04:14] otherwise it would [04:14] I talked with robbie basak about it [04:14] jcw4: reviewed your original change [04:16] axw: great thanks... will add that comment [04:16] menn0: trying again: https://plus.google.com/hangouts/_/gqu6sqg3xlszyfsl2u4blczvuaa?authuser=1&hl=en [04:16] gah [04:16] * jcw4 is sad; can't seem to find any of the blocking bugs that he can easily fix to clear the CI pipeline [04:16] also issues with couldn't start [04:16] WTF? [04:17] thumper: wondering if it's actually something wrong with hangouts at the moment [04:17] thumper: try calling me one more time [04:17] just did, it starts then crashes [04:18] menn0: ok, so back to previous statement, you have it sorted now? [04:18] maybe [04:19] thumper: it looks like changing the env agent version is allowed as long as all the machines in the env are on either the current or next version [04:19] thumper: which they should be for what I'm looking at [04:19] thumper: so it might all be fine. [04:20] ok, that's good then [04:20] thumper: I'll be doing some extensive manual testing before trying to merge this [04:20] * thumper nods [04:20] thumper: thanks rubber duck / teddy bear :) [04:20] how are you going to make it fail? [04:20] thumper: hack the code I think [04:21] make one of the state agents not upgrade [04:21] menn0: I know [04:21] add an upgrade step [04:21] that checks for some random file on disk [04:21] say /var/lib/juju/kill-upgrade [04:21] and if it is there, the upgrade fails [04:21] that's a nice idea but this is just before the upgrade steps run [04:22] what if the steps fail? [04:22] don't we want to go back? [04:22] this is for the case where one of the state servers fails to come up with the new agent version [04:22] or are we going forwards? [04:22] that depends [04:22] if it's the master, we roll back using the backup [04:22] ok, instead of an upgrade step [04:22] unless there is no backup... then we go in to an error state [04:23] make it something at the start of the machine agent [04:23] if it's the secondary it marks itself as broken but the upgrade continues [04:23] I'm half considering that we should always have something that looks for a file in order to die [04:23] think the chaos monkey stuff [04:24] thumper: +1 [04:24] /var/lib/juju/chaos [04:24] :) [04:24] so we can have CI tests that controllably cause things to fail [04:24] * thumper goes to add another agenda item to next week's meeting [04:25] thumper: that's not a silly idea [04:26] sinzui: you there next week? [04:26] thumper: i'll see if I can leave something permanent and extensible in place for testing this stuff [04:26] menn0: I suggest $datadir/chaos/machine-agent [04:26] thumper: I like it [04:27] so we have 6 blocking regression bugs, but only two are assigned and in progress... since no-one can land anything til these are resolved, shouldn't everyone be working on these? [04:27] s/everyone/someone/ [04:27] jcw4: and yes [04:27] :D [04:27] * thumper is prepping for a trip away leaving tomorrow [04:27] * jcw4 bravely looks at installing precise on a virtualbox instance [04:40] jcw4: you don't need to install precise [04:40] set your default series to precise [04:40] and bootstrap ec2 [04:42] davecheney: I haven't done that yet... do I need to provide ec2 tokens in my env. or something? [04:42] i.e. do I use my own ec2 credentials [04:42] jcw4: yes [04:42] just set the usual ec2 env vars and it works (majic)! [04:42] davecheney: okay... I want to try that [04:44] jcw4: you can use any provider [04:44] well, not aszure, that is properly borken [04:44] hehe [04:44] but openstack if you use rackspace or hp cloud [04:45] I'm most familiar with ec2, but I've got a couple rackspace accounts too [04:45] wallyworld: I've got some changes to hopefully make bootstrap/mongo more robust, but unit tests on 1.20 are not happy [04:45] wallyworld: I can't run 1.20 tests without something mongo related failing [04:45] with or without my changes [04:45] axw: did you want a quick chat [04:45] sure [04:46] ehrm, what's up with hangouts [04:46] i'm in 1:1 [04:47] wallyworld: it's dying on me [04:47] axw: yeah, you joined and then keft, we can try standup one [04:47] wallyworld: same thing [04:47] thumper: did you get hangouts to work? I'm getting the same thing [04:47] jcw4: protip: do a bootstrap without default-series set [04:47] so it _should_ choose trusty (dunno) [04:48] and maybe will work before you start trying to debug why it doesn't [04:48] axw: hmmm, restart chromium? [04:48] davecheney: I see, so do a pass through on trusty using ec2, and then change my series to precise? [04:48] wallyworld: didn't help :/ [04:48] wtf [04:49] I'll try firefox [04:49] may have to install the plugin [04:49] davecheney: and to make sure I'm clear... I initiate the tests on my local trusty box, and if my ~/.juju/... config is pointed at ec2 it should "just work" ? [04:50] juju switch "the name of your ec2 environment" [04:50] juju bootstrap --upload-tools [04:50] jcw4: i am concerned that bootstraping an environment is not second nature to you [04:50] given the number of bugs we have that are not caught in CI [04:50] * jcw4 blushes [04:50] you shouldn't just assume "works on my machine" == "ship it" [04:51] davecheney: almost all of my work has been in areas that don't seem as susceptible to environmental issues [04:51] davecheney: and my testing has been 100% via unit tests. [04:52] davecheney: my first real dogfood testing of actions on the command line was yesterday [04:52] * jcw4 slinks off and hides [04:52] oh wtf [04:52] jcw4: better late than never [04:52] firefox doesn't like it either [04:52] hmmm [04:53] axw: how about you try and start one? [04:53] yeah trying now [04:54] wallyworld: nope. [04:54] jeez [04:54] axw: wallyworld menn0 why not use skype ? [04:55] yuk [04:55] * wallyworld doesn't have skype installed [04:55] * axw has never used skype before [04:56] axw: so do the trunk tests pass? is it just 1.20? [04:56] wallyworld: I'll try now [04:56] axw, wallyworld: if you guys are having trouble too then hangouts is broken right now [04:57] menn0: I am, dunno about wallyworld [04:57] thumper and I were unable to get a call up despite many attempts in chrome and firefox [04:57] yeah, I guess it's busted [04:57] seems to work for me [04:57] it was crashing for us while the call rang [05:02] wallyworld: I did get an error on master just now [05:02] Panic: not authorized for query on presence.presence.beings (PC=0x40EF7D) [05:02] ffs [05:03] the precise failures on CI were uniter related [05:04] davecheney: so after juju bootstrap --upload-tools... I do "go test ./..." locally and it'll magically connect to ec2? Or do I need to first ssh into ec2 and then somehow initiate the tests? [05:05] jcw4: nope [05:05] jcw4: deploy some charms and see if your environment works [05:05] davecheney: oh... I see. What if I want to reproduce the test errors that are blocking CI? [05:06] the error is "can't bootstrap a precise image" [05:06] this isn't tested in the unit tests [05:06] it's done in some intergration test build that curtis runs [05:06] which is roughtly the steps you just did [05:07] there might be some juiggery pokery we need to do to make sure that boostraps a 32bit environment (getting pretty rare in ec2 these days) [05:07] davecheney: oh, ok. [05:07] * jcw4 scrambles to keep up [05:07] :) [05:08] I have a friend who always says "Don't judge me Weldon" [05:08] I know how he feels :) [05:09] wallyworld: I think a lot of the mongo errors we get in tests are to do with how we reuse Mongo databases [05:10] axw: you mean in the test setup? [05:10] wallyworld: MgoTestPackage creates a single Mongo database and then MgoSuite reuses it [05:11] restarting it only on authorisation errors [05:11] sometimes those errors don't get caught in the right place tho [05:11] yeah, that has bothered me [05:11] * thumper eagerly awaits daylight savings time so the team lead calls are at a sane time [05:12] * thumper is brain dead [05:12] thumper is soft [05:18] axw: i've tried removing the lock file and it ran into various issues, so easiest for now just to leave it but use a different name to previously [05:19] so diff is 1 line [05:19] wallyworld: cool, I think that'll be fine [05:19] yep,for now [05:19] gotta get this release out [05:21] wallyworld: the change I'm trying to do is to add the admin user to mongo before initialising state [05:22] that way we can always open state authenticated, and use session.Copy [05:22] axw: is that the cause of some issues? [05:22] wallyworld: I couldn't reproduce the i/o timeout, but it's the only thing I can think of [05:23] wallyworld: sorry, this is for azure [05:23] hmm, i would hope that this sort of bug would at least be consistent across providers [05:23] bootstrap fails because the mgo socket gets a timeout [05:23] well it's timing related [05:24] but adminuser first,then a consistent way to open state is good [05:24] then that leaves precise tests :-( [05:25] well I need to get tests running for 1.20 on my trusty machine too... [05:25] sigh, at least it fails for both series then [05:28] axw: i'll gotta go and do some packing for the flight tomorrow, how far away is the io timeout fix? [05:28] wallyworld: it's done, but I can't verify that the tests are unaffected [05:28] ah right [05:28] davecheney: it's aliiive :) [05:28] so, however long it takes to get the unit tests to run, which could be quite a while at this rate [05:29] does it bootstrap ok? [05:29] could you see the azure issue? [05:29] wallyworld: it bootstraps fine. I couldn't reproduce the error in the first place. [05:29] sigh [05:30] axw: i'm wondering if it's worth pushing up the fix - do the relevant state tests pass? [05:30] I'll try them in isolation [05:30] at least CI can have a go then [05:31] jcw4: ok, so the next thing is to ensure you're bootstrapping a 386 environment [05:31] so [05:31] juju destroy-environment -y $(juju switch) [05:32] then juju bootstrap --constraints="arch=386" --upload-tools [05:32] this may work [05:32] or it might fail horribly [05:32]