=== jam1 is now known as jam [06:53] jamespage: its working with maas and juju openstack-base bundle. the only thing i had to fix was the network interface name which was ens192 to eth1 in neutron gateway machine [06:54] jamespage: i had to change udev rules and reboot [06:54] jamespage: then it automatically fixed it [06:54] jamespage: i now can go from there [06:56] ybaumy: great! [06:56] ybaumy: yeah we where discussing the configuration for the external networking last week [06:57] ybaumy: at a minimum we might stop attempted auto config of that and make a note in the bundle README that thishas to be set afterwards [06:57] ybaumy: you can config it via the data-port option on the neutron-gateway charm - so br-ex:ens192 for your case [10:02] I guys, I have a quick question. Is there any existing interfaces in neutron-gateway charm to restart neutron-l3-agent and neutron-dhcp-agent openstack services ? [10:03] gabriel: use juju run --application neutron-gateway 'service neutron-l3-agent restart ; service neutron-dhcp-agent restart' [10:06] BlackDex: thanks for your quick reply. I want to do this operation in a charm. For the moment, I have something like: [10:06] for service in ['neutron-dhcp-agent', 'neutron-l3-agent']: [10:06] hookenv.status_set('maintenance', 'Restarting %s' % service) [10:06] host.service_restart(service) [10:06] is it the best way to do so ? [10:07] that is a bit out of my lueage :p. Didn't do that much in charm dev [10:08] BlackDex: ok thanks ! :p However, it really close to your way to restart these services, just from a different point of view [10:10] is it a subbordinate charm? Or an external charm which needs to communicate with neutron-gateway? [10:11] for a subbordinate there are examples i think, but i don't know for external [10:11] yes it is a subordinate charm [10:12] but I did not seen any interfaces to do such [10:13] for example, in neutron-openvswitch charm, you have the neutron-control interface to restart the services [10:15] in fact, i think that it's quite similar :p [10:42] ah :). Well i only did some small changes to existing charms, and that did not include stuff like that [10:45] BlackDex: ok, no problems and thanks again! :) [10:47] :) Good luck! [10:47] thanks! [10:48] maybe there are some people at openstack-charmers which can help you a bit more :) [10:48] i mean #openstack-charms [10:49] oh ok thanks! :) === tinwood is now known as tinwood_swap [12:00] hmz, maas is giving me a headache [12:08] I've spent a couple days working to familarize myself with juju and charms. I'm still failing to understand how best to write a simple layers charm. The tutorials appear to be rather opinionated but quite variable. I'm coming from a docker/docker-compose background. I'd like to add GPU support to kubernetes worker. I've been following along SaMnCo article: [12:08] https://insights.ubuntu.com/2017/02/15/gpus-kubernetes-for-deep-learning%E2%80%8A-%E2%80%8Apart-13/ but not getting to a working nvidia-smi [12:08] any pointers on how best to write a layer that would just be adding a rather picky driver? [12:19] dgonzo: I will let the dev team answer on the later part but can you detail what is not working for Nvidia? What version of my layer do you use? What cloud and what Nvidia card (Latest gives you 375.26 and works for me for Pascal, but will not on k20 from AWS, which require 367 and I didn't model that [12:19] So look at the layer code, try to run it natively on the host, and see if it fails saying that the card is legacy and requires 367 [12:21] If yes, just find out the versions you need. These are set at the very beginning of the layer script [12:21] And easily manageable [12:27] SaMnCo: It doesn't appear that the drivers are installed. Minimum install required to get GPU availability is going to be something like https://github.com/NVIDIA/nvidia-docker/wiki/Deploy-on-Amazon-EC2#gpu-instance-creation [12:29] last run of your nvidia-cuda layers I got a driver error or driver missing -- sorry it was a few iterations ago [12:29] So if you use the g2 instances, you are stuck to 367 and the charm will not work. You can check out the commit history and the version just before the last will work [12:29] ok [12:30] dgonzo: if you use p2, you will be fine [12:30] We are working with nVidia to make that better [12:31] Go back in time to 367.xx essentially [12:33] are you saying there's a specific commit with the 367.xx drivers or I that I should fork and modify to peg at 367.xx? [12:33] ... I am using the g2 series ec2 for cost considerations -- I'm not running these for training but rather for evaluation nodes [12:38] SaMnCo: I see the driver versions in the last commit. Thanks. [12:45] Yes, previous commit is 367 I think [12:45] Check it out [12:45] Or fork the repo and edit the reactive/cuda.sh [12:46] To repoint to these === scuttle|afk is now known as scuttlemonkey [15:54] hi all [15:55] in https://jujucharms.com/docs/2.0/models-config i read i can use noproxy for local addresses, but it seems not working [15:55] juju bootstrap --config noproxy=10.0.0.0/8 maas juju-controller --constraints tags=juju WARNING unknown config field "noproxy" WARNING unknown config field "noproxy" [15:56] and the no_proxy defined in /etc/environment seems ignored too [16:04] Hetfield: did you mean to type noproxy there? the config opt is no-proxy, and btw, --config will affect the bootstrap node, you can specify it again if you want your models to use it with "--model-default no-proxy=foo". [16:04] Hetfield: see more discussion about that here: https://github.com/juju/docs/issues/1676 [16:10] SaMnCo: I built the resource with updated driver. I'm getting an error "hook failed: "install" [16:11] I'm still entering commands line-by-line as the template file in the repo doesn't yet work [16:12] kwmonroe: yes, i need both, for bootstrap and later for day by day [16:29] trying to install Ubuntu OpenStack with Autopilot and receive the following JuJu Bootstrap Failure during installation "problem with juju bootstrap". [16:30] dgonzo: hmmm OK. I'll spin up a g2 when I can to reproduce and fix. These Nvidia drivers drive me crazy [16:30] At MWC right now so not before tomorrow [16:30] from the log file: wait 15s\nAttempt 5 to download tools from https://streams.canonical.com/juju/tools/agent/1.25.6/juju-1.25.6-trusty-amd64.tgz...\ncurl: (56) SSL read: error:1408F119:SSL routines:SSL3_GET_RECORD:decryption failed or bad record mac, errno 0\ntools from https://streams.canonical.com/juju/tools/agent/1.25.6/juju-1.25.6-trusty-amd64.tgz downloaded: HTTP 200; time 20.655s; size 770048 bytes; speed 37281.000 bytes/s Tools [16:31] any help would be appreciated ;/ === vagarwal_ is now known as vagarwal [16:39] xpmaven: oo ssl decryption error? is there a proxy or anything between you and the url its attempting to fetch? [16:40] no proxy [16:41] well, unless one considers the MaaS controller a proxy... [16:47] SaMnCo: no problem. Thanks for forging the way on this. Being able to manage a "hybrid" cluster with gpu resources is exactly what i've been trying to do "by hand" even with the frustrations you've got a solution that shows real promise [17:04] xpmaven: nothing that would cause an ssl error that i'm aware of [17:04] xpmaven: the only thing i coudl think of was mitm proxying your ssl query and failing validation. but that doesn't seem to be the case... i'm uncertain. I would certainly file a bug however so we can get the right people looking at it [17:09] thanks [17:10] submitted: https://github.com/Ubuntu-Solutions-Engineering/openstack-installer/issues/1038 [17:13] xpmaven: if you curl the directory (without the file) does it display an ssl error? [17:15] no it does not [18:03] anyone familiar with configuring the haproxy charm to do https redirects? [18:04] bdx: I am. [18:07] jrwren: so here, https://gist.github.com/jamesbeedy/fa588b242ccffe5ce52c1e41c895e274#file-haproxy-bundle-yaml-L15 [18:09] I'm trying to add the redirect scheme to the service options, but am getting denied due to what looks like a unicode error bc of my "!" http://paste.ubuntu.com/24079429/ [18:09] bdx: yeah, quote the whole thing. [18:10] service_options: ['redirect scheme https code 301 if !{ ssl_fc }'] [18:11] jrwren: I'll give this a whirl. thanks [18:16] bdx: that charm is deceptively awesome. I thought did didnt' support a bunch of stuff, but it turns out it supports almost everything. Its great. [18:18] o/ [18:22] wish the maas team were as responsive as you guys! [18:23] jrwren: I've had similar mileage ... the more I dig in, the better it gets :-) [18:23] is zsh completion for juju 2.0 available somewhere? [18:24] gah [18:24] jrwren: I spoke to soon [18:25] http://paste.ubuntu.com/24079515/ [18:25] bdx: that didn't work? It must be some yaml thing. In my case it works fine quoted, but when I don't use json array syntax (the []) but instead use yaml list syntax [18:26] bdx: here is exactly what I use: http://paste.ubuntu.com/24079518/ [18:27] jrwren: ahh nice, I'll give that a try [18:41] jrwren: sadly, same result using your config [18:41] jrwren: I'm wondering, are you using a trusty harproxy? [18:43] bdx: xenial [19:18] jacekn: ping, trying to use your grafana charm but getting a failure to get wheeze from packagecloud? [19:31] jrwren: got it working, thanks [19:31] bdx: great! [19:38] really wish I could figure out what maas-enlist isn't working :-/ [20:20] hey guys I need to customize the charmstore openstack bundle, can someone point me in the right direction to start doing some charm coding etc