[00:21] davecheney: good to hear [00:25] m_3: i can't remember what I said :) [00:25] it was yesterday [00:27] grr, hp cloud is having a loe down [00:27] it can't talk to anything outside the network [00:27] which makes bootstrapping a bit problematic [10:30] Apr 28 10:29:56 juju-goscale2-machine-0 mongod.37017[9332]: Sun Apr 28 10:29:56 [conn1999] query [10:30] juju.txns.log query: { $query: {}, $orderby: { $natural: -1 } } cursorid:6469952084798152748 ntor [10:30] eturn:10 ntoskip:0 nscanned:11 keyUpdates:0 numYields: 1 locks(micros) r:486329 nreturned:10 resl [10:30] en:948 267ms [10:30] are these log times bad ? [20:53] davecheney, morning [20:54] davecheney, I was wondering if you considered https://codereview.appspot.com/8953044 suitably trivial? it was agreed in the discussions preceding the original CL but we forgot to actually implement it then :/ [20:55] * fwereade goes to do housey things for a bit, will be back soon [21:10] rogpeppe1, heyhey [21:10] yo! [21:10] rogpeppe1, how's it going? [21:11] fwereade: trying to write family emails and you're a welcome distraction :-) [21:11] fwereade: how's you? [21:11] rogpeppe1, I consider https://codereview.appspot.com/8953044 to be plausibly trivial, because this was always the agreed behaviour AIUI: https://codereview.appspot.com/8953044 [21:11] rogpeppe1, I's good :) [21:11] * rogpeppe1 is not looking at anything work related just now [21:11] rogpeppe1, a few cylinders are starting to fire again [21:12] rogpeppe1, no worries :) [21:12] fwereade: that's good. you must've burned out a few [21:12] rogpeppe1, had a nice weekend, lots of laura time [21:12] rogpeppe1, was definitely needed [21:13] fwereade: we had a nice weekend building smart new veg beds from old scaff boards [21:13] rogpeppe1, excellent [21:14] fwereade: not quite done yet, but excellent progress made (and some celebratory g&ts as the sun went down) [21:14] rogpeppe1, I've been reading jonathan strange and mr norrell [21:14] fwereade: good innit? [21:14] rogpeppe1, it's rather good I think [21:14] rogpeppe1, oh yes [21:15] fwereade: i kinda stole it off my sister in law one xmas (she'd just been given it) and read it far too fast before she noticed [21:15] fwereade: i'm on the fourth bear [21:15] fwereade: but not v far in [21:15] rogpeppe1, haha, nice, I have done that on a few occasions [21:17] rogpeppe1, it is not impossible that you will cast it aside with a bewildered WTF, I am not sure how much my enjoyment of that one was based on knowledge of the wider world from his other books [21:17] fwereade: the fire is flaming nicely and the foster cat is happily snoozing on the sofa. [21:17] fwereade: we'll see [21:17] fwereade: i'm reserving judgement for the moment [21:18] fwereade: the context i had from you has been useful [21:18] rogpeppe1, yeah, I don't think that's explained at all [21:18] rogpeppe1, at least not when you kinda need to know it [21:18] fwereade: i think there was one brief aside [21:19] fwereade: interesting to see results of dave's stress tests. [21:19] rogpeppe1, definitely [21:19] fwereade: oops, work related. [21:20] fwereade: must write email :-) [21:20] rogpeppe1, do we ever need that internal api :) [21:20] fwereade: indeed [21:20] rogpeppe1, I have a few threads I need to draw together myself [21:20] fwereade: i was happy that the gui seemed to work fine with all those nodes [21:20] rogpeppe1, that's awesome [21:21] fwereade: that's only one client for the API mind :-) [21:21] * fwereade removes hat [21:21] rogpeppe1, sure [21:21] rogpeppe1, but it's always nice to break after some other component ;p [21:21] :-) [21:22] * rogpeppe1 hasn't quite worked out what to say on the subject of error reporting vs logging yet, but feels there is sensible something lurking around [21:25] rogpeppe1, well, every piece of context you describe *could* be provided by logging [21:25] fwereade: if we were happy with logging 1GB/minute [21:26] rogpeppe1, but it is indeed a tradeoff in many things [21:26] fwereade: yeah [21:27] rogpeppe1, I think that more logging *and* more error context is likely to be generally helpful [21:28] fwereade: agreed [21:28] fwereade: that's the gist of my as yet uncomposed reply [21:28] rogpeppe1, but there have been times when we have erred toward the massively redundant in our error messages, and I think that's also a problem [21:29] fwereade: if we err in that direction and prune as needed, i think that's a better approach than the opposite [21:29] rogpeppe1, yeah [21:29] rogpeppe1, I find myself pondering state though [21:30] rogpeppe1, there are a lot of places we pass up unadorned errors from mgo [21:30] fwereade: indeed [21:30] fwereade: let's not [21:31] rogpeppe1, hmm -- this to me is actually a case where I think logging wins out [21:32] rogpeppe1, "mgo said no" is generally independent of context -- the connection fell over, basically, and there's nowt to be done [21:32] fwereade: logging errors or logging actions? [21:32] rogpeppe1, logging actions [21:32] rogpeppe1, and leaving errors we can't account for to fall out, and the detailed context be pieced together as needed [21:32] fwereade: we generate enormous log files as is - i'm concerned about the volume of logging data we generate. [21:32] fwereade: or *will* generate. [21:33] rogpeppe1, I think that actually implementing fricking log levels will address that [21:33] fwereade: the problem is that anyone in any significantly sized environment will dial back the log levels [21:34] fwereade: and then we'll be left with io.EOF [21:34] rogpeppe1, we don't have to send everything everywhere [21:35] fwereade: that's a good point actually. [21:36] fwereade: but still, those API servers are gonna be helluva busy [21:36] fwereade: when everything's going through the api [21:36] rogpeppe1, consider me undecided on how best to address logging [21:37] fwereade: yeah me too. i agree with you that a combination of better logging *and* better error reporting will be useful [21:37] rogpeppe1, the distribution problem smells a little of being amenable to charming, if we can figure out the right interface to expose [21:38] fwereade: perhaps. some volume estimations may well be useful. [21:39] rogpeppe1, yeah, just an idle thought [21:40] rogpeppe1, have been having slightly less idle thoughts about an environs/resource package with Image and Spec structs and Map and Kind interfaces [21:40] rogpeppe1, but I have been doing that when I should have been reviewing ian's stuff [21:41] * fwereade gets he to a codereview [21:41] fwereade: interesting. i think there's something nice struggling to get out there but i haven't spent any time on it since maas appeared. [21:42] rogpeppe1, it has kinda clicked -- not in the "it's perfect" sense but the "it's definitely *closer* to perfect" sense [21:43] fwereade: i want something that's not just a "this appeared vaguely similar in these providers" thing; something that's obviously genuinely helpful in current and future contexts. [21:43] rogpeppe1, eg it will be as stupid as the individual implementations of Map are wrt caching [21:43] rogpeppe1, well, it's a step that way -- I don't think Image should *really* be a struct, but I don't want to speculate too far [21:44] fwereade: i really need to get back to what i'm supposed to be doing :-) look forward to continuing discussions around this topic [21:44] rogpeppe1, likewise [21:44] rogpeppe1, hey sorry 1 thing? [21:44] fwereade: 1 :-) [21:45] fwereade: go on [21:45] rogpeppe1, davecheney mentioned you were going to fix the unassigned unit thing seen with juju-gui to machine 0; please chat to me about it at your convenience, ideally prior to implementation, I have thoughts [21:45] rogpeppe1, but not now [21:46] fwereade: bug #? [21:46] rogpeppe1, not sure [21:47] rogpeppe1, if it's not currently on your mind you can leave it on mine, but I'll cc you in in case it rings a bell, I have a half-reply open [21:49] fwereade: doesn't ring a bell currently, unless it's the failed deploy thing, in which case i just reported the bug, and had no immediate intention of fixing it [21:50] rogpeppe1, ah great, I'll find it and assign it to myself then, I have a CL up that addresses the best currently-addressable subset of the problem, I think [21:58] fwereade: #1173089 ? [21:58] <_mup_> Bug #1173089: deploy can fail partially [21:59] rogpeppe1, yeah, that's the one [21:59] rogpeppe1, got distracted by another bug ;p [22:00] rogpeppe1, fwiw I don't think we should rollback -- I think create/assign should be a single txn [22:29] fwereade: yeah, probs