[01:11] http://paste.ubuntu.com/24709556/ [01:11] hehehe [01:11] anyone who can see whats going on there gets a gold star === thumper is now known as thumper-afk === thumper-afk is now known as thumper === frankban is now known as frankban|afk [07:15] Good morning Juju world! [11:05] hello there [11:05] i'm getting this message [11:05] ERROR some agents have not upgraded to the current model version 2.1.3.1: machine-1 [11:06] i have a ha juju running [11:06] and apperently one agent wasn't up/failed [11:06] how can i fix this? [11:23] is it possible to run the etcd charm without persistent storage? [11:39] nvm, just skip the --to flag =) [15:11] hey lazyPower, arch question for ya... do you know why k8s went with etcd vs something like zookeeper? [15:11] kwmonroe: that question has come up quite a bit on their lists, but with consul instead of zookeeper. I think it was because coreos was in there early and did that heavy lifting but its hard to say [15:12] ah, roger dodger. just curious. not saying zookeeper is some kind of panacea by any stretch ;) [15:12] kwmonroe: yeah, there was talk at one time of doing a libkv that would abstract the providers [15:14] something with a small footprint like https://jujucharms.com/hadoop-hbase/ would have fit the kv bill nicely. [15:14] because why have a microservice when you can just have 10 machines [15:14] :) no comment [15:14] :) [15:18] kwmonroe: probably go vs java [15:54] o/ juju world! [15:58] what's up Budgie^Smore [15:59] not much, still evangelising juju in interviews :) [15:59] :) [16:00] had someone say the master / slave model is bad though :-/ [16:02] has anyone figured out what juju's performance is in terms of cpu & mem per host managed? [16:03] yea, 2.2 does a lot of work on that as it's a limiting factor in JAAS being effective [16:04] so there's regular testing of scalability and chattiness and such [16:15] yeah it is a limiting factor in any master / slave environment... is the testing documented anywhere? [16:22] Budgie^Smore: hmm, there's notes in the bugs/etc where things are identified but yea, nothing really to point to other than the various bugs as things are identified. [16:24] rick_h it would definitely be useful to have since there is a fear that master / slave environments are not scalable, which we both know is bull and makes it impossible to auto scale in and out [16:24] Budgie^Smore: fair enough. I might take some time to show off the 2.2 improvements in a visible way or something. [16:24] Budgie^Smore: thanks for the nudge [16:25] rick_h no worries, just giving back the feedback that I am getting when talking about Juju :) [16:25] good stuff [16:27] rick_h at the very least it would be good to have documentation on scaling controllers beyond a minimal ha setup [16:27] Budgie^Smore: yes, that's firmly in my current working list [16:27] Budgie^Smore: I'm doing the troubleshooting, but the goal post-troubleshooting is to turn into the "best practices" for juju taking expected scale into account [16:27] Budgie^Smore: so look for "operating Juju" documentation [16:28] in the near future [16:28] rick_h is there anything else I haven't though about that is in your work list ? ;-) [16:28] Budgie^Smore: let's see, I've got some metrics, blogs posts, and 2.2 testing on my todo list :P [16:32] Budgie^Smore: adding cpu cores scales how many units a controller can operate pretty much linearly. last updates to 2.2, a 8-core machine was able to run 1400 machines with 2300 units, using 10GB of RAM and 70% of CPU [16:32] (as an example) [16:34] urulama oh awesome, thanks :) that funnily enough is in line with the example of about 1000 slaves / master that I was talking about in an interview last week [16:35] Budgie^Smore: there are some changes before RC1 is out, that will help lower CPU usage from what you've seen with 2.1.x even more. 2.2 will be awesome :D [16:35] it is worth knowing that I am usually dealing with networks of systems in the 10k+ range [16:35] we'll get there :) [16:36] urulama I don't doubt it, but I do think 10 masters for 10k system is not a bad ratio to start with [16:37] hm. [16:37] but that depends if the controllers can share the world load or if HA is more a active / passive affair [16:37] work* [16:38] HA doesn't do full load balancing atm [16:38] ah ... on the roadmap? [16:39] yep [16:39] :P [16:39] and not "some distant future" either :D [16:39] oh if you are talking being able to hand hyperscale it better not be :P [16:39] but until then, current HA is not so much about scaling as it is about redundancy [16:40] handle* [16:48] sometimes I wish I was a stronger dev than I am, just this discussion has given me ideas :) [21:52] hi there: how can set an EIP on AWS for the controller at bootstrap phase? juju bootstrap set a dinamic public ip and if i reboot machin i can' login again [22:36] i understand that becuase there is HA configuration it is not necessary to have a EIP : is it correct?