=== _thumper_ is now known as thumper
thumperaxw_: ping01:07
axw_thumper: gah sorry, overlap with standup01:07
axw_thumper: can we make it in 5-10 please?01:07
axw_wallyworld: last one: https://github.com/juju/juju/pull/810602:55
=== axw_ is now known as axw
thumperwell... a great day of race conditions and deadlocks03:55
babbageclunkthumper: stink04:16
babbageclunkthumper: oh nasty - so the rate-limiting would start causing the agents to timeout and then what - they go into a never-ending loop trying to connect and timing out?04:18
thumperbabbageclunk: the apiserver got stuck ratelimiting every agent04:18
thumperbecause the forwarder tried too often04:19
babbageclunkthumper: approved04:21
thumperlaters peeps05:03
jambabbageclunk: I looked at 810806:23
jamI don't think we want to refuse any downgrade06:23
jambut maybe restrict it within a major.minor series06:24
jamthe most important thing is to get "juju upgrade-juju" to agree with what the actual Upgrade logic will do.06:24
wallyworldaxw: if you get a chance later, could you look at https://github.com/juju/juju/pull/8104. the jujud operator branch depends on this one. i've got other stuff to do so  no huge rush07:12
axwwallyworld: ok. still working on provisioner stuff, will see how I go07:12
wallyworldno worries, it can wait07:12
=== mup_ is now known as mup
jamaxw: reviewed 810912:13
=== petevg_afk is now known as petevg
babbageclunkwallyworld: ping?20:00
wallyworldbabbageclunk: hey20:43
babbageclunkwallyworld: sorry chatting to thumper20:48
wallyworldok, np20:48
babbageclunkwallyworld: hey, did you see the message from jam above?21:18
babbageclunk(off the phone now21:18
babbageclunkthumper points out that the problem with downgrading is that we don't have inverses for the upgrade steps21:19
wallyworldbabbageclunk: i haven't looked at your PR yet, but what he says has sound logic21:20
babbageclunkYeah, and it matched a bit with some of the code in the upgrade-juju command. But I think the check in state.checkUpgradeInfoSanity is there because downgrading would require undoing upgrade steps.21:21
babbageclunkwallyworld: ^21:21
wallyworldi'll need to look at the code21:22
wallyworldbut in general downgrading has been an issue due to not being able to revserse schema changes etc21:23
wallyworldbabbageclunk: it looks like that state code refuses any doengrade, right?21:25
babbageclunkwallyworld: yeah, it's pretty clear, and it's been like that since the beginning.21:31
babbageclunkwallyworld: So that sounds like I probably can't allow downgrades even within the same major.minor.21:37
babbageclunk(by since the beginning, I mean, since upgrading was introduced in 2014.)21:39
wallyworldbabbageclunk: yes, we have never supported downgrading IIANM21:47
wallyworldbabbageclunk: since that's the case, making the CLI match what the server will do is to me the most important bit, and then we can discuss whether to allow downgrades separately21:49
babbageclunkwallyworld: makes sense21:49
=== _thumper_ is now known as thumper
babbageclunkwallyworld: there are a couple of test failures about trying to do downgrades - it's kind of surprising that they were considered to pass when the sanity check would still fail.21:50
wallyworldbabbageclunk: you mean the there are CLI tests? that assert we can do downgrades?21:51
babbageclunkwallyworld: yeah, one in the apiserver tests for model manager, one in featuretests21:52
babbageclunkwallyworld: just looking at them now. Maybe we need to distinguish between allowing downgrades for the controller and hosted models?21:53
wallyworldi'd need to look at those, perhaps we should allow minor21:53
wallyworldcontroller versions needs to be > hosted model version21:53
babbageclunkThat's already checked. But would there be any benefit for downgrading hosted models?21:54
wallyworldthat's up to the user of thse models21:57
babbageclunkwallyworld: it doesn't look like the modelmanager test is actually trying to downgrade - just happens to trip that in setting up the test.21:58
babbageclunkwallyworld: but the feature test is trying to do it.21:58
wallyworldand the feature test passes? how?21:58
wallyworldsince there is that state check21:58
babbageclunkI mean, it did pass until I broke it.22:01
wallyworldbut the feature test is meant to be full stack which mean the state check would have been run22:08
babbageclunkwallyworld: I presume that means that it wasn't checking that there weren't any errors.22:14
wallyworldmaybe, not sure without looking22:15
babbageclunkwallyworld: hmm, the test is about HA - there's a need for the leader to be able to downgrade if one of the followers can't upgrade?22:22
babbageclunkwallyworld: https://github.com/juju/juju/blob/develop/featuretests/upgrade_test.go#L15722:23
wallyworldmakes sense22:24
wallyworldso we must bypass that state check then for that case22:24
wallyworldbut that's internal, not a user driven thing22:24
wallyworldthumper: small one https://github.com/juju/bundlechanges/pull/3322:26
babbageclunkwallyworld: ok, I'll try to chase that code down...22:31
thumperwallyworld: approved22:34
wallyworldthumper: a larger one https://github.com/juju/juju/pull/8111. sadly we still need to use charmstore.v5-unstable, but at least that's only in tests23:48

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