| sidnei | niemeyer, uhm. looks like i digged my own hole there. now i must type really fast. ;) | 00:45 |
|---|---|---|
| sidnei | im sure if i just go back and read all the channel logs i'll learn EVERYTHING. or maybe read the source. :) | 00:47 |
| _mup_ | Bug #829880 was filed: object store doesn't like key with '/' <Ensemble:Triaged by hazmat> <OpenStack Compute (nova):New> < https://launchpad.net/bugs/829880 > | 02:26 |
| hazmat | sidnei, if you script the download of irclogs, i'd be interested in the script ;-) | 02:27 |
| niemeyer | Hey there | 14:00 |
| sidnei | hey niemeyer | 14:32 |
| niemeyer | sidnei: Yo! | 14:32 |
| niemeyer | sidnei: Ready for the talk? :-) | 14:32 |
| sidnei | niemeyer, no ;) | 14:32 |
| niemeyer | sidnei: Hehe :) | 14:33 |
| niemeyer | sidnei: It's no big deal.. just play with it and you'll easily get the idea | 14:33 |
| niemeyer | sidnei: and we'll be here :) | 14:33 |
| sidnei | niemeyer, sure. i'm thinking of starting with a background of 'why ensemble', which should fill at least half the time. then get to the 'how', and hopefully i won't run out of time. :) | 14:34 |
| sidnei | niemeyer, because most people won't have heard of it anyway | 14:34 |
| niemeyer | sidnei: Sounds good.. I have the slides from FISL too, in case you want to get a head start | 14:35 |
| sidnei | niemeyer, definitely helpful | 14:35 |
| niemeyer | sidnei: http://labix.org/presentations/ensemble-fisl12/ | 14:38 |
| niemeyer | sidnei: The demo there is a short (3m) video.. I can send you at some point too if you want | 14:39 |
| sidnei | niemeyer, that's fine thinking of saving the demo for a lightning talk. | 14:40 |
| niemeyer | sidnei: In my experience the demo has been game-changing in presentations | 14:41 |
| niemeyer | sidnei: I could babble for hours and it wouldn't be as effective as showing a couple of in-practice minutes | 14:41 |
| sidnei | niemeyer, right. i'm thinking of redoing the demo so it's more readable on a projector, at fisl i could barely read what was on screen. | 14:43 |
| niemeyer | sidnei: Ohh, sweet! | 14:43 |
| niemeyer | sidnei: We'd love to have that as well! 8) | 14:43 |
| sidnei | i'm sure you do :) | 14:43 |
| sidnei | niemeyer, so changing subject but not quite, how do you see a django formula working? i mean, it's different than most services in the sense that installing a package (python-django) does not bring up a service, but instead you need a django app which is generally a vcs checkout. | 14:46 |
| niemeyer | sidnei: That's not a big deal per se, but let me try to understand what you're really referring to | 14:46 |
| niemeyer | sidnei: Do you mean a generic all-encompassing django formula that can be used for any django app at all, or | 14:47 |
| niemeyer | sidnei: .. bundling of a specific django app from a vcs? | 14:47 |
| sidnei | niemeyer, i think there's room for both. where the former probably means making the vcs location configurable in the formula. | 14:47 |
| niemeyer | sidnei: Agreed, there's room for both, but they're different paths | 14:48 |
| niemeyer | sidnei: I'd actually like to experiment the first option at some point | 14:48 |
| niemeyer | sidnei: The second one is easy | 14:48 |
| niemeyer | sidnei: For the first one, I'd like to try building something a bit smarter, that would enable communicating to the framework that e.g. a given database was made avaialble | 14:49 |
| niemeyer | sidnei: Effectively working more like a PaaS | 14:49 |
| sidnei | right | 14:50 |
| niemeyer | sidnei: Imagine building a thin layer that enabled the Python app to query "Hey, what apps are up?", etc | 14:50 |
| niemeyer | sidnei: We could then offer a few different relations.. e.g. | 14:51 |
| niemeyer | sidnei: mongo-relation-joined, postgres-relation-joined, etc | 14:51 |
| niemeyer | sidnei: These would hook directly into Django and configure the respective database | 14:52 |
| sidnei | niemeyer, so hooking into django.settings, ok. | 14:52 |
| niemeyer | sidnei: Right.. PaaS FTW | 14:53 |
| niemeyer | sidnei: The vcs then is a config option.. | 14:53 |
| niemeyer | sidnei: Which means exactly the same formula can be used to deploy pretty much any django app straight from vcs | 14:54 |
| sidnei | niemeyer, yup, i can see that working. would be simple to export DJANGO_SETTINGS_MODULE to a module generated by things provided by ensemble, which would override the django app defaults. | 14:56 |
| niemeyer | sidnei: Right! | 14:56 |
| sidnei | niemeyer, ok. i think i might give that a try. what about the second option, with bundling the app. | 14:58 |
| niemeyer | sidnei: That'd be awesome.. please send some feedback to the list later if possible. This feels like a big area we should start exploring for several platforms | 14:59 |
| niemeyer | sidnei: The second option is trivial.. you bundle the app itself without a formula and hardcode things up | 14:59 |
| niemeyer | sidnei: Erm | 15:00 |
| niemeyer | sidnei: s/without/within/ | 15:00 |
| sidnei | niemeyer, so the formula would be pushed to ensemble with a tarball included | 15:00 |
| niemeyer | sidnei: Not necessarily.. you can still reference the vcs, for instance | 15:00 |
| niemeyer | sidnei: Except it would be hardcoded inside it | 15:01 |
| niemeyer | sidnei: But the tarball would be doable too | 15:01 |
| sidnei | niemeyer, ok. and how scalable is the deployment of formulas? if i had 1000s of new instances firing up simultaneously, would ensemble fall over from too many concurrent requests? | 15:03 |
| sidnei | niemeyer, i was looking at murder https://github.com/lg/murder and it seems simpler than i thought it was, maybe a murder formula would be interesting too. | 15:03 |
| niemeyer | sidnei: Nope | 15:04 |
| niemeyer | sidnei: I mean, no, it won't fall | 15:04 |
| niemeyer | sidnei: We never tested such a workload, but the communication is relatively simple | 15:05 |
| sidnei | niemeyer, awesome. | 15:07 |
| sidnei | niemeyer, i think a murder formula might be interesting anyway, even if just for the sake of showing how to use murder without capistrano. | 15:09 |
| niemeyer | sidnei: I don't know what's murder.. lookig | 15:11 |
| niemeyer | looking | 15:11 |
| niemeyer | sidnei: Ah, is it the twitter thing? | 15:12 |
| niemeyer | Yeah.. | 15:12 |
| niemeyer | sidnei: Yeah, certainly interesting.. | 15:12 |
| sidnei | niemeyer, the way its deployed seems fairly trivial, it just calls python scripts to fire up a torrent server and multiple peers | 15:16 |
| sidnei | where it == capistrano | 15:16 |
| sidnei | https://github.com/lg/murder/blob/master/lib/murder/murder.rb | 15:17 |
| sidnei | niemeyer, i wonder if it would make more sense as a central service provided by ensemble though | 15:25 |
| niemeyer | sidnei: Probably not | 15:30 |
| niemeyer | sidnei: We we're going to the opposite direction, actually | 15:30 |
| niemeyer | sidnei: Ensemble services themselves will eventually be formulas | 15:30 |
| hazmat | murder is a deploy system, afaicr | 15:39 |
| hazmat | i was thinking about doing a wsgi formula | 15:39 |
| * hazmat tries some jfdi | 15:39 | |
| niemeyer | hazmat: Hey | 15:41 |
| hazmat | niemeyer, hola.. spent the morning trying to track down a touchpad | 15:42 |
| niemeyer | hazmat: I understand it as file-copying | 15:42 |
| hazmat | i saw the annoucement last night before i went to bed | 15:42 |
| hazmat | rushed to stores this morning, just missed them | 15:42 |
| niemeyer | Aw | 15:42 |
| niemeyer | LOL | 15:42 |
| hazmat | 100 for a tablet computer is pretty nice | 15:42 |
| niemeyer | I just googled for "murder" | 15:42 |
| hazmat | doh ;-) | 15:42 |
| hazmat | niemeyer, it uses zk | 15:42 |
| hazmat | its twitter's deploy system afaicr | 15:43 |
| hazmat | in python | 15:43 |
| niemeyer | hazmat: Yeah, I've listened to part of the talk.. it sounded like a generic file-copying thing, but maybe I forget the details | 15:43 |
| niemeyer | hazmat: I imagined that it might be put in a relation, etc | 15:44 |
| hazmat | niemeyer, basically bittorrent + zk | 15:44 |
| niemeyer | hazmat: Ensemble won't get in the way of the copying/deploying bit | 15:44 |
| hazmat | niemeyer, yeah.. its not really germane to a service its pretty custom imo | 15:44 |
| niemeyer | hazmat: E.g.: """ | 15:45 |
| niemeyer | Murder (which by the way is the name for a flock of crows) is a combination of scripts written in Python and Ruby to easily deploy large binaries throughout your company’s datacenter(s). | 15:45 |
| niemeyer | """ | 15:45 |
| hazmat | a formula interestested in doing app deploy, just does a set on the version in the config file | 15:45 |
| hazmat | for upgrade | 15:45 |
| niemeyer | Yeah | 15:45 |
| niemeyer | hazmat: Still, how would one put the first version in place | 15:46 |
| hazmat | niemeyer, deploy --with-config | 15:46 |
| hazmat | niemeyer, just finished the config.yaml ;-) | 15:46 |
| niemeyer | hazmat: I imagine, e.g., a few formulas like murder-server, murder-service, etc | 15:47 |
| hazmat | niemeyer, i just don't think its that useful | 15:47 |
| niemeyer | Agreed, it's less exciting | 15:47 |
| smoser | hazmat, cloud-init runs fine to my knowledge on canonistack | 16:43 |
| hazmat | smoser, yeah.. it does.. i was having some issues logging in with keys that where put in place my cloud-init.. but cloud-init works fine | 16:43 |
| hazmat | smoser, there's a few minor things i noticed.. the metadata server/hostname set seems to hang | 16:44 |
| smoser | "metadata server/hostname set seems to hang" | 16:44 |
| smoser | ? | 16:44 |
| hazmat | smoser, the login issue disappeared once i started using a keypair when launching instances, haven't bothered to track it down.. it was probably a red -herring | 16:44 |
| smoser | hostname is https://bugs.launchpad.net/nova/+bug/820962 | 16:45 |
| _mup_ | Bug #820962: Generating hostname from display name incorrect <OpenStack Compute (nova):Triaged> < https://launchpad.net/bugs/820962 > | 16:45 |
| hazmat | smoser, ah.. yeah that's it | 16:46 |
| hazmat | smoser, do you know if current openstack has a metadata server of some type? | 16:46 |
| smoser | yes, it does. | 16:46 |
| smoser | its not perfect (bug 823520, bug 827569) but it is present and generally working | 16:47 |
| _mup_ | Bug #823520: EC2 instance-type metadata returns SQLAlchemy object string <OpenStack Compute (nova):Fix Committed by dan-prince> < https://launchpad.net/bugs/823520 > | 16:47 |
| _mup_ | Bug #827569: ec2metadata service does not include 2011-01-01 <OpenStack Compute (nova):Confirmed> < https://launchpad.net/bugs/827569 > | 16:47 |
| hazmat | thanks, i'll look into that then, there's a little bit of metadata that we'd like for ensemble from the instances | 16:47 |
| smoser | you're probably running natty ? | 16:48 |
| hazmat | smoser, yeah.. mostly natty for ec2.. oneiric for lxc | 16:48 |
| smoser | so user data issue , natty? | 16:49 |
| hazmat | smoser, honestly atm it all appears to just magically work | 16:49 |
| hazmat | i'm able to deploy services on canonistack | 16:50 |
| hazmat | i'm just not sure how in all cases ;-) | 16:50 |
| hazmat | off to enjoy a beautiful day.. have a good weekend folks | 16:50 |
| smoser | verified that onieric image on canonicalstack works with no --key (launched and accessed via ssh_import_id in cloud-config) | 16:51 |
| smoser | and verified natty does too | 16:53 |
| smoser | but... there is a bug in the loader kernels it would appear that sometimes they dont' boot all the way | 16:53 |
| smoser | it is the loader kernel that is at fault likely | 16:53 |
| smoser | a reboot will usually fix it | 16:53 |
| sidnei | niemeyer, yes, it's pretty generic file copying, peer-to-peer due to bittorrent. | 17:04 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!