2012/06/12/#juju.txt

[00:01] <roy-feldman> Hey Marco - juju bootstrap is working now with my Maas node, without any changes!
[00:01] <roy-feldman> in KVM
[00:02] <roy-feldman> But I had to something a little odd, which may be a MaaS problem
[00:03] <roy-feldman> Basically I had to PXE boot it twice, and I had to select a boot option on the second boot
[00:06] <marcoceppi_> roy-feldman: that's...odd - but I'm glad it works!
[00:07] <roy-feldman> After the initial PXE boot, I got the network error I described
[00:08] <roy-feldman> Then I started guest again, and hit F12 and selected the virtio boot option, not the default boot option which PXE boot
[00:10] <roy-feldman> It seems that MaaS is not completing the PXE boot in one step
[00:10] <roy-feldman> Also, the start button in the MaaS interface is a "noop"
[00:12] <roy-feldman> Now I do have a Juju question
[00:12] <roy-feldman> I can see my MaaS node when I do juju status
[00:14] <roy-feldman> I have done a juju deploy and juju expose of mysql
[00:15] <roy-feldman> How long should it take for local mysql charm to transition from pending to running?
[00:16] <marcoceppi_> roy-feldman: depends on how beefy your machine is. Think of it as having to install the Ubuntu OS, install the juju working parts, then deploy the charm and install the charm.
[00:16] <roy-feldman> I am running a beefy i7 laptop .. I wouldn't think it would take very long
[00:17] <roy-feldman> It doesn't matter if the machine is already running with a OS?
[00:17] <roy-feldman> This was not a cold deploy
[00:17] <marcoceppi_> roy-feldman: not sure, I've typically enlisted machines from the PXE boot screen
[00:18] <roy-feldman> I have done it the other way
[00:18] <roy-feldman> I entered the Mac address to Maas
[00:19] <roy-feldman> Then I booted the machine which registered it with MaaS
[00:19] <roy-feldman> In my case, I had to boot it twice
[00:19] <marcoceppi_> roy-feldman: but did the machine have the Ubuntu OS yet?
[00:19] <roy-feldman> Yes
[00:19] <marcoceppi_> See, I'm not sure what happens here since MaaS builds it's own images to use
[00:20] <marcoceppi_> So I'm not sure if it's going to wipe/re-install or what
[00:20] <roy-feldman> During the second boot it appeared to install the latest server packages
[00:22] <roy-feldman> Perhaps.  In any case I hope to soon have better model of how MaaS + Juju work then I have now. ;-)
[00:23] <roy-feldman> Any suggestions on how I can see what is going on, what logs I can be looking at?
[00:27] <roy-feldman> Looking at the output of Juje debug-log, I see a series messages starting with "ProviderInteractionError: Unexpected Error interacting with provider: 409 CONFLICT"
[00:27] <roy-feldman> I think that would explain why my mysql instance is not coming up
[00:28] <roy-feldman> Should I file a bug report?
[00:29] <marcoceppi_> huh, 409 CONFLICT means MaaS doesn't have any nodes for Juju to use
[00:29] <roy-feldman> Maybe I shouldn't have my only maas node running if I want to deploy a charm
[00:30] <marcoceppi_> roy-feldman: you'll need at least two MaaS nodes to bootstrap and deploy with Juju
[00:30] <marcoceppi_> One for the Bootstrap node and one for the charm you wish to deploy :)
[00:30] <roy-feldman> I do have two nodes
[00:30] <marcoceppi_> Oh
[00:30] <marcoceppi_> What does the MaaS dashboard show?
[00:30] <roy-feldman> If you mean I have a node for the MaaS server and another for deployment
[00:30] <roy-feldman> Hold on
[00:32] <roy-feldman> It show that I have one node which has been allocated
[00:32] <marcoceppi_> And that's it?
[00:33] <roy-feldman> There were 0 nodes when I installed MaaS
[00:33] <marcoceppi_> How many nodes do you have available?
[00:34] <roy-feldman> 1
[00:36] <roy-feldman> Does juju require its own MaaS node?
[00:37] <roy-feldman> I assumed that I could run Juju at the native level to interact with MaaS.  Am I wong?
[00:38] <lifeless> roy-feldman: you can indeed
[00:38] <lifeless> you'll need a MaaS controller, and then a Juju control node, running on a MaaS provisioned node
[00:39] <roy-feldman> I missed that in the configuration steps
[00:39] <roy-feldman> The setup of the Juju control node
[00:39] <roy-feldman> Where is that documented?
[00:40] <lifeless> its what juju bootstrap does
[00:41] <roy-feldman> So I need a free MaaS node when I run Juju bootstrap
[00:41] <roy-feldman> ?
[00:43] <roy-feldman> i.e. So are you saying that every time I run juju bootstrap with a maas envrionment, I need an available Maas Node for the juju controller?
[00:45] <roy-feldman> What is the best way to rollback my juju deployment and go back to adding additional MaaS nodes?
[00:46] <roy-feldman> Should I do a destroy-environment?
[00:48] <lifeless> yes, thats what I'm saying
[00:48] <roy-feldman> thanks
[00:48] <lifeless> uhm, you shouldn't need to destroy it, just add more nodes so it can get them when you go to deploy a charm
[00:49] <roy-feldman> And its Ok to simply cntrl-c out of my original juju deploy and expose?
[00:49] <roy-feldman> No need to do any housecleaning?
[00:50] <roy-feldman> The one that never completed because there wasn
[00:50] <roy-feldman> wasn't an available node
[00:52] <roy-feldman> BTW, shouldn't juju expose give some kind of message if there aren't sufficient nodes to complete the action?
[00:53] <lifeless> please file a bug abou that
[00:53] <roy-feldman> will do
[00:53] <lifeless> I agree it would be good to do so, I don't know why it didn't... may be a MaaS bug, for instance.
[00:54] <roy-feldman> Looking at the trace, it looks like the loop is in juju
[00:54] <roy-feldman> Specifically juju/agents/provision.py
[00:55] <roy-feldman> It just keeps retrying
[00:56] <lifeless> you could leave it where it is and add nodes, if it keeps retrying when a node comes available it will succeed ;)
[00:57] <roy-feldman> Thanks again for all the help, I will try again with more nodes and see what happens and I will file a bug report about juju provision
[01:41]  * negronjl is done for the night
[03:51] <bkerensa> SpamapS, marcoceppi: http://www.omgubuntu.co.uk/2012/06/ubuntu-12-10-development-update-1
[03:51] <bkerensa> you got questions about juju ^ :P
=== almaisan-away is now known as al-maisan
=== tobin_ is now known as tobin
=== tobin is now known as Guest74089
=== Guest74089 is now known as tobin__
=== garyposter is now known as gary_poster
[13:16] <jimbaker> at usenix config mgmt summit today in boston, will be talking later on "service orchestration with juju"
[13:19] <niemeyer> jcastro: ping
[13:19] <niemeyer> jimbaker: Sweet
[13:22] <jimbaker> niemeyer, it's a good lineup of speakers from chef, bcfg2, cfengine, vmware
=== al-maisan is now known as almaisan-away
[14:52] <themiwi> Hi all. Is there an easy way to get essentially the output of `getconf getconf _NPROCESSORS_ONLN` of the remote host in a *-relation-changed hook? Or should I use `n=$(ssh $(relation-get hostname) getconf _NPROCESSORS_ONLN)`?
[14:56] <m_3> themiwi: you could pass that as a relation variable... one side would do `relation-set num-cpus=$(...)` and the other side does `relation-get num-cpus`
[14:58] <m_3> themiwi: that can usually be inferred by instance-type though.. depends on provider.  Also you might take a look at constraints in the juju docs.
[14:59] <m_3> all depends on what you're trying to do... passing relation variables works fine though
[15:34] <negronjl> 'morning all
[15:34] <imbrandon> morn
[15:34] <negronjl> hi imbrandon
[15:34] <m_3> negronjl: yo
[15:35] <negronjl> 'morning m_3
[15:53] <twobottux> aujuju: Is juju specific to ubuntu OS on EC2 <http://askubuntu.com/questions/149952/is-juju-specific-to-ubuntu-os-on-ec2>
[16:33] <SpamapS> I love the askubuntu bot
[16:33] <SpamapS> really.. fantastic idea
[16:52] <m_3> SpamapS: yeah, me too... can we answer it here?
[16:53] <jcastro> imbrandon: you don't need to add "new-charm" anymore, just setting the status Does The Right Thing(tm)
[16:53] <jcastro> also dude, awesome job on the RPMs!
[16:53] <jcastro> has anyone tried them yet?
[16:53] <imbrandon> umm not sure
[16:53] <imbrandon> i just put the word out a few hours ago
[16:53] <imbrandon> so no feedback yet
[16:54] <jcastro> you blog it or need me to?
[16:54] <imbrandon> i'm about to blog about that and my new "download for ubnutu" button soon, so shuld get some
[16:54] <imbrandon> soonish
[16:54] <jcastro> k, poke me and I'll syndicate it on cloud.u.c
[16:54] <imbrandon> your more than welcome to , more ppl read yours i think
[16:54] <imbrandon> ll
[16:54] <imbrandon> kk
[16:55] <imbrandon> i'm pretty sure my whole juju category is syndicated already
[16:55] <imbrandon> on cloud.u.c
[16:55] <imbrandon> i'll make sure tho
[16:55] <jcastro> no I need to post it, it doesn't automatically publish
[16:55] <imbrandon> ahh, kk
[16:55] <SpamapS> m_3: no, how would it give you credit? ;)
[16:56] <imbrandon> SpamapS: loging to the bot via api :)
=== koolhead17|zzZZ is now known as koolhead17
[16:59] <SpamapS> imbrandon: good luck with that
[16:59] <SpamapS> sounds like a yak to be shaved later
[17:00] <m_3> hmmm... travelng to nepal lately?
[17:00] <imbrandon> SpamapS: heh
[17:05] <adam_g> hazmat: ping
[17:06] <hazmat> adam_g, pong
[17:08] <adam_g> hazmat: looking at snapshot.py of charm runner for the first time... is clean_juju_state() something that can be easily adpated to work with a non-local environment?
[17:08] <hazmat> adam_g, about to get into a meeting..
[17:08] <hazmat> adam_g, the state cleaning yes, the storage cleaning no
[17:09] <hazmat> adam_g, we don't have a provider storage method for killing files
[17:09] <adam_g> hazmat: if i wanted to just script around it via ssh/paramiko, i'd just be deleting the related files from the web storage right?
[17:10] <hazmat> adam_g, what provider?
[17:10] <adam_g> MAAS
[17:10] <adam_g> actually, i wouldn't even need to do that. ive got local access to the MAAS server
[17:10] <hazmat> adam_g, if its maas, i'd check if they have an api for deleting files and just sniff ~/.enviroments.yaml by hand for the creds to delete the files
[17:10] <adam_g> hazmat: ill start there, thanks
[17:47] <m_3> negronjl: yo... gotta sec?
[17:47] <negronjl> m_3: sure
[17:47] <m_3> g+?
[17:48] <negronjl> m_3: sure .. give me a sec.  I'll invite you when I'm there
[17:48] <m_3> ok
[17:49] <negronjl> m_3: started invite sent
[18:20] <themiwi> m_3: Sorry, got interrupted and then had to dash away to catch the train. Yes, passing this information as a relation variable to override the default would be great. However, I'd like to also provide a sensible default choice which is not just a pessimistic 1.
[18:53] <m_3> whoohoo... jim's talk/demo went well
[19:01] <SpamapS> sweet
[19:01] <jcastro> nice
[19:03] <SpamapS> themiwi: interesting problem you're trying to solve. In what case is it important to know the *remote* CPU count for service configuration?
[19:03] <SpamapS> themiwi: either way, you don't have to provide a default. Just keep running your changed hook until the other side has *set* that value.
[19:04] <SpamapS> themiwi: the changed hooks can ping-pong back and forth a few times.
[19:04] <themiwi> SpamapS: I'm trying to cook up a charm for the Sun Grid Engine (SGE), where the master/head node needs to know the number of slots it can allocate on the compute node.
[19:05] <themiwi> SpamapS: yes, that's what I'm going to do now. I set it in the *-relation-joined hook of the compute charm and the keep querying it in the *-relation-changed of the head charm
[19:21] <SpamapS> themiwi: perfect :)
[19:22] <SpamapS> themiwi: wi would have thought grid engine would have its own RPC to talk to its nodes and figure things like CPU's out
[19:22] <SpamapS> anyway.. time to find nourishment
[19:23] <themiwi> SpamapS: I suppose it could, but the thing with cluster administration is, that administrators want to be able to adjust every tiny detail and would probably hate such automatisms...
[19:24] <SpamapS> hah
[19:24] <SpamapS> tweakers
[19:26] <themiwi> SpamapS: ;-) yep. they take pride in squeezing every single flop from their clusters.
[19:30]  * m_3 is a card carrying tweaker
[20:22] <themiwi> Another question: Whenever a relation is added, the variable nodecpus is set to the number of processors/cores of the slave. so far so good. however, the master needs to maintain a *sum* of all node cpus. Where would I store this persistent information?
[20:26] <m_3> themiwi: juju doesn't choose this for you... I'd stick it in a map or db on the filesystem (we often use /var/lib/juju/ to house such things).  some folks like facter.. there're lots of options.  key is that this map will update over time as nodes die and/or join
[20:27] <m_3> themiwi: hooks xxx-relation-changed, xxx-relation-departed, and xxx-relation-broken will have to recalculate that number over the lifetime of the cluster
[20:27] <themiwi> m_3: was just wondering whether i could do something like "unit-set cpucount $ncpus"...
[20:28] <themiwi> m_3: but maintaining a simple text file containing that figure is also no problem.
[20:28] <m_3> themiwi: relation-list is useful here
[20:28] <m_3> themiwi: oh, so config allows you to set stuff from the command line too...
[20:28] <m_3> there's a config-changed hook
[20:28] <themiwi> m_3: sure. writing these charms is quite intricate and the documentation rather scarce, it seems :-)
[20:28] <m_3> and then `juju set var=value` commands
[20:29] <m_3> yeah... some are really easy... some really hard... depends on the service
[20:30] <themiwi> especially keeping things straight with all that ping-ponging between xxx-relation-changed hooks is a bit daunting
[20:30] <m_3> yes
=== marcoceppi_ is now known as marcoceppi
[20:31] <m_3> it's super-powerful though
[20:32] <m_3> it's auto-negotiation for config params that you usually have to set manually or in config scripts
[20:34] <themiwi> sure, but you also have to be super-careful
[20:57] <imbrandon> its a one time investment though :)
[20:57] <imbrandon> or should be ... heh
[20:57] <m_3> maybe once per long-term release :)
[20:58] <imbrandon> right
[20:58] <imbrandon> :)
[21:01] <jcastro> imbrandon: heya
[21:01] <jcastro> hazmat never got your ubuntufication of the sphinx template
[21:01] <jcastro> can you hook him up so our docs look awesome?
[21:02] <jcastro> hazmat: this is handy btw, http://jujucharms.com/tools/store-missing
[21:02] <jcastro> nice one on that one
[21:06] <imbrandon> jcastro: sure
[21:06] <imbrandon> jcastro: yea i hadent sent it to anyone really yet
[21:07] <imbrandon> like the download button ? heh u should pimp it a lil for me :)
[21:07] <jcastro> no, you said you styled the stuff for juju.ubuntu.com/docs
[21:07] <jcastro> the sphinx stuff
[21:08] <imbrandon> doing it now, is there a reason not to just do a merge req , i mean i dont mind sending them directly to hazmat , but just curious
[21:09] <imbrandon> and yes thats what i said "sure" to, and then added the bit about the button
[21:09] <imbrandon> :)
[21:09] <imbrandon> but yea, let me dig that branch up its burried in my ~/Projects folder and then i'll pass it arround , give me ~30ish min
[21:09] <jcastro> well remember I was like "show daniel too a bunch of other ubuntu stuff uses sphinx"
[21:09] <imbrandon> yup yup
[21:09] <jcastro> no rush or whatever
[21:10] <hazmat> imbrandon, just do a merge request
[21:10] <jcastro> I just happened to be talking about it with kapil
[21:10] <hazmat> imbrandon, my only concern is if it requires updates to the installation
[21:10] <imbrandon> hazmat: cool cool , yea i dont mind, was just like ummm ? heh
[21:10] <imbrandon> nah
[21:10] <imbrandon> its just some new files in _templates
[21:11] <hazmat> imbrandon, the build should cron update and run automatically, so as long as there aren't any new sphinx extensions it should just work as a mp proposal
[21:11] <imbrandon> thats about it, dont even think the biuld rules need updates
[21:11] <imbrandon> if i rember right
[21:11] <imbrandon> yup yup
[21:12] <imbrandon> btw it does use the html build target right ?
[21:12] <imbrandon> as far as whats published
[21:12] <imbrandon> not like the all-in-one html or something weird
[21:17] <jcastro> SpamapS: ok so like brandon is winning with his packages. :)
[21:22] <SpamapS> jcastro: I can't do anything to thwart the Debian NEW queue
[21:25] <imbrandon> hahah i'm sure SpamapS will mucho win in the end, i actually had alot of help unknowingly from hazmat doing the footwork over at pypi.python.com a while back
[21:25] <imbrandon> so i cheated :)
[21:29] <imbrandon> SpamapS: btw your interview over at omg is on fire, there are over 200 ppl current viewing it and a ton of positive twitter comments ( despite some of the "ohh its not desktop" idiots )
[21:29] <imbrandon> been like that for like 3 or 4 hours now
[21:31] <SpamapS> imbrandon: heh cool. :) #2 this week
[21:45] <koolhead17> SpamapS, :)
[21:49] <imbrandon> SpamapS: ugh , can we ( read: you ) rename jitsu binary to juju-jitsu ? hehe ( i know i know probably not ) ... there is a conflict with another deployment tool named jitsu and both dont like to live in bin/jitsu at the some time heh
[21:49] <imbrandon> no biggie though, lets just hope that I'm not most ppl and have both installed , or want to at least
[21:50] <imbrandon> ( the other jitsu is the Joyent's tool to deploy and manage http://no.de and http://jit.su accounts
[21:50] <imbrandon> for nodejs )
[21:54] <SpamapS> imbrandon: oh there's another jitsu?
[21:54] <imbrandon> yea
[21:54] <imbrandon> i found out the hard way
[21:54] <SpamapS> why does node.js take all the best names
[21:54] <imbrandon> heh
[21:54] <imbrandon> i went to install juju-jitsu and it was like ummm there is somethign there already
[21:55] <imbrandon> i looked and rembered
[21:55] <imbrandon> heh
[21:55] <imbrandon> i was like damn!
[21:55] <SpamapS> node also conflicts withs ome broke ass old AX25 packet thing caled node
[21:55] <imbrandon> but yea its probably a corner case
[21:55] <imbrandon> heh
[21:55] <SpamapS> not really a corner case
[21:56] <SpamapS> I can see them conflicting
[21:56] <SpamapS> there ought to be a single registry for bin names
[21:56] <imbrandon> really its used as a wrapper anyhow, so if it was just named juju-jitsu for the bin
[21:56] <imbrandon> it would probably be fine
[21:56] <imbrandon> mostly* a wrapper
[21:57] <SpamapS> I want it called jitsu
[21:57] <SpamapS> typing juju-jitsu sucks
[21:57] <imbrandon> cuz once youve done juju-jitsu wrap-juju , it wont matter tho :)
[21:57] <imbrandon> and only rename the bin not everything :)
[21:57] <imbrandon> i dont really care, but i did run into it today heh
[21:57] <SpamapS> I'd be willing to relent and do jjitsu
[21:58] <imbrandon> heh
[21:58] <imbrandon> jj :) that would make a good alias, /me adds to .bash_profile
[21:59] <imbrandon> i wanna add real alias support too btw, like hub has for git , you just "alias git=hub" in bash_profile and go on about your business
[21:59] <imbrandon> its sweet
[22:00] <imbrandon> thats one of those $sometimes things
[22:01] <imbrandon> actually its more like .... eval "$(hub alias -s)"
[22:01] <imbrandon> but still, same thing, just works if you use zsh or any shell
[22:48] <negronjl> SpamapS, jcastro, m_3:  Do we know which day is the Charm School planned for during Velocity ?
[22:49] <SpamapS> negronjl: did we decide to do one w/o the conference blessing?
[22:49] <negronjl> SpamapS: not sure yet, hence the question :)
[22:50] <SpamapS> negronjl: IIRC, it was rejected by them
[23:07] <negronjl> SpamapS: ah.  thx
[23:31] <SpamapS> http://packages.qa.debian.org/j/juju.html
[23:48] <m_3> negronjl: nope... dunno
[23:51] <negronjl> m_3: thx
[23:52] <imbrandon> SpamapS: saweet
[23:52] <imbrandon> :P
[23:53] <imbrandon> SpamapS: here, this one is all for you :) http://www.brandonholtsclaw.com/blog/2012/juju-everywhere#comment-555572949
[23:54] <imbrandon> Mr Spaleta
[23:55] <SpamapS> Our biggest fan