/srv/irclogs.ubuntu.com/2018/08/28/#juju.txt

anastasiamaca very simple review PTAL https://github.com/juju/juju/pull/9118 - exposes lisitng of credential models from state00:47
anastasiamacwallyworld: babbageclunk: veebers: thumper ^^01:15
* thumper looks01:15
babbageclunkanastasiamac: Already looking, should have said, sorry!01:15
babbageclunkanastasiamac: approved01:19
thumperI had a suggestion01:20
wallyworldanastasiamac: i had a suggestion also01:21
anastasiamacbrilliant, thnx \o/01:28
anastasiamacwallyworld: i have been thinking uuid vs tag too.. the trick is that i know where it'll be used and they will all go back to tag....01:30
anastasiamaci'll change anyway to b inline with convention01:30
anastasiamacthumper: my only bleh with map[string]string keyed on uuid is that it's not blatantly obvious what is actually in the map... with map[names.ModelTag]string, a casual observer can deduce it from type ;D01:34
anastasiamacbut since it's my bleh, i'll change to uuid01:34
thumperthanks01:34
thumperbabbageclunk: I'm going to make a coffee and then work through some intermittent failures with watchers01:41
thumperif you want to chat while I do that, I'd be up for it01:41
thumperkinda like pair programming01:41
babbageclunkthumper: ok - are those the ones you think the raftleases might be causing?01:42
thumperI think vscode has some code sharing features too for this01:42
babbageclunkOr just a second brain making light work01:42
thumperthe later01:42
thumperthere is a lease issue01:42
babbageclunkSure!01:42
thumperin the state package01:42
thumperso I have two types of problems to fix01:42
* thumper goes to make coffee first01:42
thumperbabbageclunk: heading to 1:1 HO01:53
thumperbabbageclunk: ugh, installing a kernel before restarting02:01
* thumper waits02:01
babbageclunkheh02:01
veebersHmm, at some point I've accidently pasted the APGL license header into my terminal or so my history shows me :-P02:33
veebersgotta love pasting the wrong clipboard into bash and hoping none of the text is actually destructive commands02:34
wallyworldbabbageclunk: i'm seeing several of these in the logs02:58
wallyworldERROR juju.worker.dependency "lease-clock-updater" manifold worker returned unexpected error: updating global clock: lease operation timed out02:58
babbageclunkwallyworld: in startup?03:01
babbageclunkOr after that?03:01
wallyworldafter, when system has been running for a while03:01
babbageclunkhmm03:01
wallyworldcould just be slow connectiivity or something, it's on ec2 and i also see occasional mongo timeouts03:02
wallyworldprobs nothing to worry about immediately03:03
babbageclunkMight be that I've got the timeout threshold too low.03:03
babbageclunkAre you running with multiple controllers?03:03
wallyworldbabbageclunk: no, just the one03:21
babbageclunkwallyworld: weird - I could see the timeout being maybe too low for forwarding from a follower to the leader.03:23
wallyworldmight just be crappy environment03:24
wallyworldmog sometimes times out too according to the logs03:24
wallyworld*mgo03:24
babbageclunkwallyworld: I think jam knows about the mongo errors - I'm not sure they're timeouts exactly.03:25
wallyworldthe logs claim they are03:26
wallyworldio.timeout03:26
wallyworldcan't recall the exact message now03:26
babbageclunkI think it's trying to connect to old addresses.03:26
wallyworldwhen i see it again i'll try and add some debugging03:28
babbageclunkyeah, sounds good03:28
veeberswallyworld: you have a couple of moments? Slighlty stuck with the status bits, I can the cloud container status is being updated; but status shows "Creating mysql container" which is a unit status (set by the charm), it seems where I think the status message for 'juju status' is not right?03:34
wallyworldjuju sets "creating mysql container" initially03:35
wallyworldIIANM03:35
veebersaye, it's setting it as unit status, and my cloudcontainer status bits are being ignored. I can't see where that "creating mysql . . " is intially getting set03:36
wallyworldit's set when the unit is first created i seem to recall03:37
wallyworldand then the charm can overrite it as it comes up03:37
wallyworldi don't quite follow how the cloud container bits are being ignore03:37
wallyworldFullStatus() would need to read them03:38
wallyworldso it's in our control03:38
veeberswallyworld: juju defaults it to 'waiting for container' and the charm sets for mysql: status_set('maintenance', 'Creating mysql container')03:38
wallyworldah that sounds right03:38
veebersoh, by ignores it I mean; I'm not changing the right part to thread it through03:38
wallyworldi think FullStatus is the apiserver side method whuich reads stuff out of hte db and populates the status params struct03:39
wallyworldhave you changed that bit?03:40
veeberswallyworld: that might be it (only hit (u *Unit) Status() at the moment)03:42
veebersI'll check that out03:42
wallyworldveebers: right, that won't do anything03:43
wallyworldjuju status calls client facade FullStatus()03:43
wallyworldthere's a bunch of logic in there03:43
wallyworldit might call unit.Status() though, not sure03:43
veebersack, that looks more promising, thanks!03:43
wallyworldbut that's the place to look03:43
thumperwallyworld: you busy?03:54
thumperhttps://github.com/juju/juju/pull/9116/files03:55
wallyworldthumper: about to have a m,eeting with carmine03:55
wallyworldcan look after03:55
thumperthanks04:23
anastasiamacwallyworld: babbageclunk: m finding myself in this nice, cozy spot where I wan to return []params.ErrorResults from apiserver facade... how r u feeling about that?05:26
anastasiamac(or a struct that contains[]params.ErrorResults)05:26
wallyworldhmm, i guess it depends on the call05:28
babbageclunkanastasiamac: so, effectively a list of lists of errors?05:29
anastasiamacyes, altho i could break it down i guess05:30
anastasiamacthe reason - we have a bulk call to update credential05:31
anastasiamacand each credential can have a number of models that could error out05:31
anastasiamacso my choices are - list of lists of errors05:31
wallyworldtypically the bulk call would be a slice of things and a corresponding slice of errors. well that's been what we've dome till now05:31
anastasiamacof a struct where i can identify each credential with a map[model]error and an error05:32
anastasiamacyes, so slice of errorresults05:32
anastasiamacwas tryingt o not change the actual method much05:32
anastasiamacbut i'll have to to return different result05:32
wallyworldit could work05:33
anastasiamacso, right now this bulk call does return params.ErrorResults but this, of course, does not cater for the cred models check where each can error out too... hence []params.ErrorResults...05:33
anastasiamacit culd, for sure :) i'll sleep on it and experiment a bit more... just wanted to know if ppl will immediately dislike or b k with this...05:34
wallyworldit's different but worth considering i think05:35
anastasiamacion one hand, returning []params.ErrorResults will be neater.. but i feel like there will be a lot of complex understanding that is kind of implied and only comes naturally if u read the code... that is the only consideration that is holding me back :)05:36
anastasiamacthe other option, 3rd one that is, is map[cred]params.ErrorResults :D this way we r going to be very explicit...05:37
anastasiamacbut kind of unprecedented, so i think i talked myself into breaking a separate struct per credential where each item contains []params.Error or similar....05:39
anastasiamacthnx for ur help and listening \o/05:39
wallyworldkelvinliu_: this PR makes operators use statefulsets / storage and also i think fixes that termination bounce bug at startup https://github.com/juju/juju/pull/912006:24
* jhebden is back from [afk] - 426510h:37m:11s away06:37
kelvinliu_wallyworld, looking. and https://github.com/juju/juju/pull/9119 this small fix, would u take a look when u got time, thanks06:39
wallyworldsure06:40
wallyworldkelvinliu_: lgtm, ty06:42
kelvinliu_thanks06:43
kelvinliu_wallyworld, the pr looks great.07:04
wallyworldtyvm07:04
wallyworldi'm happy with it07:04
wallyworldseems to solve a few issues07:04
kelvinliu_is the terminate bug the `terminating-> creating issue happens it gets deployed in the first time?07:18
wallyworldkelvinliu_: yeah, that one07:38
kelvinliu_wallyworld, nice!07:38
rmcdHey all, having an issue building a charm... It's saying there's no fetcher for a relation I've written. Recently switched to a new laptop so gone from having the interface locally to grabbing it from Gitlab. Any idea how I can fix this?07:53
stickupkidmanadart: did we port the lxd stuff to 2.4?08:59
stickupkidmanadart: just checking we've not regressed the 2.4 branch of lxd 2.0.x08:59
stickupkidmanadart: we're good, sorry for the noise09:00
stickupkidmanadart: do you have 5 minutes for a quick hangout?10:18
stickupkidanyone know where the provision of machines logs out to?14:00
stickupkidcan I use "juju debug-log"?14:00
pmatulisstickupkid, i would try to use that command and apply it to the controller model (-m)14:20
hmlstickupkid: sometimes its easier to grep the /var/log/juju/machine-0.log on the controller14:22
hmlstickupkid: if you up the root logging level to trace, you can see the api calls too, which may help14:22
hmland their payload14:22
stickupkidhml: pmatulis: thanks14:23
stickupkidmanadart: LXD 2.0 doesn't have a server name :|16:25
stickupkidhml: https://github.com/juju/juju/pull/9121 can you give a quick look at this, I need to fix the deploy still, but it should make sense at least16:56
hmlstickupkid: i’ll look shortly16:56
stickupkidhml: take your time, EOD for me :D16:57
hmlstickupkid:  have a good evening16:57
=== gurmble is now known as grumble
rick_h_kwmonroe: bdx cory_fu zeestrat looking to do the Juju Show tomorrow. You all around to join? I've started a discourse post for show planning/notes afterwards https://discourse.jujucharms.com/t/juju-show-38-wed-aug-29th/202 if you have anything20:30
zeestratrick_h_: I'll try to be there20:37
rick_h_zeestrat: cool20:38
veebersMorning all o/20:43
magicaltroutrick_h_: what time is it these days?20:50
rick_h_magicaltrout: it'll be normal 2pm EST20:51
rick_h_magicaltrout: though maybe I should re-evaluate the timing if folks want me to move it up in the day more20:51
rick_h_veebers: woot woot20:51
magicaltroutmy biggest issue is that its kids bedtime =/ hour earlier or later would be better for myself, but don't switch stuff on my account.20:52
magicaltroutI'll swing by tomorrow, I want to discuss big data/big data ldn and some other stuff20:52
rick_h_magicaltrout: sweet20:52
rick_h_hmmm, I could do an hour earlier. anyone else have feedback?20:52
veebers\o/20:55
* jhebden is back from [afk] - 426524h:56m:12s away20:56
magicaltrout...20:58
magicaltroutthats a long time20:58
rick_h_lol21:00
jhebdenepoch!21:09
thumperhttps://github.com/juju/juju/pull/912322:33
veeberswth, I had this working last night just before eod, and now it's not :-\ /me digs in22:36
babbageclunkveebers: time of day related? ;)22:43
veebersheh, unlikely :-)22:49
thumperbabbageclunk, wallyworld: https://github.com/juju/juju/pull/9123 has the hubwatcher fix22:52
wallyworldthumper: is there a test? i think i saw one in the other PR?22:57
magicaltroutyou shoudn't discount time of day related bugs.. in one platform i develop we have a unit test that passes in any other timezone than GMT23:00
magicaltroutwhen you run it in the UK during BST is passes23:00
magicaltroutthen daylight savings end and it returns to failure....23:00
veebersmagicaltrout: hah, actually I've had issues with something like that before. Tests always worked in NZ-TZ but my English coworker had occasional failures :-)23:02
thumperwallyworld: no, there isn't a test, it is very racy23:04
wallyworldok23:09
wallyworldlgtm23:09
thumperanastasiamac: I'm pretty sure the default value is an interafce{} type, so a missing value will be nil.23:34
thumperso any type could have no default23:35
anastasiamacthumper: k23:35
thumperand all missing values should just be the empty string23:35
thumperI think23:35
anastasiamaci agree23:35
veeberswallyworld: this is really odd, the status stuff is messed up if I use the model name that I've been using over and over (but destroying controllers between each run). If I use a new model name everything is fine23:53

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!