2012/07/20/#maas.txt

=== matsubara is now known as matsubara-afk
[13:28] <realnorth_> anyone here have a link or know what environment variables I can feed juju?
[13:28] <realnorth_> also is it possible to choose which server to add a charm to?
=== realnorth_ is now known as realnorth
[13:31] <melmoth> realnorth, i have been told that jitsu has a -deploy-to option
[13:31] <melmoth> but i never try it yet
[13:33] <smoser> i'm guessing just about everyone is sleeping, lifeless and bigjools at least. allenap maybe around?
[13:33] <smoser> i hadnt' followed all the dns thread
[13:33] <allenap> smoser: Hi there, what's up?
[13:33] <smoser> but i thought that lifeless suggested doing ec2 like names, which i think is generally useful.
[13:34] <smoser> but then others didn't like that because of "no human friendly hostname"
[13:34] <smoser> wouldn't cnames fix that?
[13:34] <smoser> and having dhcp hand out the hostname as the human-friendly-cname?
[13:34] <allenap> smoser: Yes. One of the questions I had was: what do names with the IP address in add beyond just using IP addresses?
[13:35] <allenap> rvba: ^ maybe you can help too?
[13:35] <smoser> not much i guess.
[13:35] <smoser> but if you add in the cname, then you have both, dont you?
[13:35] <allenap> Yeah.
[13:36] <allenap> For smaller clouds, having a nice name is more useful. For massive clouds it's a burden.
[13:37] <smoser> random bit of info . i think that there was some dns service that just converted "v.x.y.z.somedomain.com' to v.x.y.z Ip address for you
[13:37] <smoser> i tihnk ii think you'd just allow it as configurable.
[13:37] <smoser> an dif someone sets the "hostname " for something then they get it.
[13:38] <smoser> who cares.
[13:38] <rvba> smoser: so a node would have an ec2 like name and a "human" CNAME then?
[13:38] <smoser> yeah.
[13:38] <smoser> well, that was my thought.
[13:38] <smoser> reverse looking up its IP would get the ec2-like name
[13:38] <smoser> but who cares.
[13:38] <rvba> right
[13:39] <rvba> The main problem with writing DNS config files is with reverse config files (because of the classful/classless pb) so yeah, that sounds like something we should definitely look into.
[13:39] <smoser> if you allow the api to set the hostname of a node, then someone can easily take a DB that has "MAC:preferred-hostname" and populate maas from that.
[13:40] <rvba> allenap: actually, we sort of mentioned that on the call this morning: having "two" hostnames: the ec2 like one and the manual one.
[13:40] <rvba> But we thought about that on the DB level.  As smoser said, CNAME is exactly that at the DNS level.
[13:41] <allenap> Yeah. We also mooted allowing nodes to specify their preferred hostname in the DHCP request and making that a CNAME of the "ugly" name.
[13:42] <allenap> But that doesn't cope with collisions well.
[13:42] <smoser> well, its easily fixable anyway.
[13:42] <smoser> the only place that that has value is when maas is not running the dhcp/dns server
[13:43] <smoser> and it assumes that something gave this system a relevant IP address based on its mac address.
[13:43] <smoser> er.. relavant hostname based on mac
[13:43] <smoser> and whatever that something is can then be just dumped/queried. and a tool then update the node once its in maas.
[13:44] <rvba> allenap: collisions are not really a problem if we can write the reverse zone files using only ec2 hostnames.
[13:44] <smoser> collisions are an issue if maas isn't serving the dhcp
[13:44] <rvba> allenap: for each zone, we compute the superset classful network and we write the full reverse zone file for that with ec2-like names.
[13:44] <allenap> rvba: I mean collisions of the "nice" names if done by dhcp; two nodes could ask for the same thing.
[13:45] <rvba> Right, that's another problem.  I was thinking about the clasless network collision.
[13:45] <rvba> Maybe the term "collision" is not really proper for that.
[13:45] <allenap> I think my brain has had enough of that stuff for this week ;)
[13:59] <rvba> smoser: that's a great idea I think.  Maybe there is something I'm missing but it definitely looks like the missing piece of the puzzle.
[14:23] <roaksoax> allenap: howdy :) so I was looking at your recen changes and was wondeirng, wher's that maas-provision binary?
[14:25] <rvba> smoser: I tried to summarize why I think your idea is great in the email I just sent.  We would have the full forward/reverse DNS config written using ec2-like names.  Then on top of that we could use nsupdate to add CNAME record to accommodate custom hostnames.
[14:25] <rvba> records*
[14:25] <smoser> rvba, thanks.
[14:25] <smoser> we expose "friendly-hostname" setting in maas via api
[14:25] <smoser> right?
[14:26] <rvba> Yes we do.
[14:26] <smoser> so my 'mac-addr:friendly-hostname' population script is easy.
[14:27] <smoser> (ie, it takes a file full of lines with 'mac-addr friendly-hostname' and just iterates over it calling maas and setting the hostname for each node based on mac
[14:27] <smoser> )
[14:27] <allenap> roaksoax: It's a front-end for several commands. It will probably evolve into the front-end for all of MAAS (so that we no longer expose twistd or django-admin).
[14:27] <rvba> smoser: yes, that would work.
[14:28] <roaksoax> allenap: right, but where is it? do we need a wrapper for it (like we did with maas binary)? (see debian/extras/maas in packaging branch)
[14:28] <rvba> smoser: hum, the key we use in the api is the system_id.
[14:28] <roaksoax> rvba: that is indeed a good idea as I was mentioning to smoser
[14:28] <roaksoax> but that means that servers wont have <rack>.<city>.<blahblah>.domain.com
[14:28] <roaksoax> which might come handy
[14:29] <roaksoax> when you do inventory
[14:29] <roaksoax> that's why I think we should consult this with sysadmins
[14:29] <rvba> Well, we might provide a way to set the custom hostname based on tags.
[14:29] <rvba> So that would be possible.
[14:30] <smoser> rvba, somehow i can surely get a node by its mac-id, no?
[14:30] <rvba> Let me check.
[14:30] <smoser> err.. get a system_id by its mac_id.
[14:30] <roaksoax> rvba: right, there's various possiblities to approach the issues yes
[14:30] <roaksoax> we need to consider invetory from now on
[14:30] <roaksoax> and we need to consider Administrators, and Datacenters and the standards they use
[14:30] <roaksoax> and provide a sokution based on that
[14:31] <roaksoax> we can't simply impose them something we believe is better when they already follow standards
[14:31] <rvba> smoser: no, system_id is the key really.  But that's something we could add very easily.
[14:33] <smoser> rvba, well i think its essential, really.
[14:33] <rvba> smoser: I completely agree.
[14:33] <rvba> smoser: please file a bug.
[14:34] <rvba> smoser: that's something really easy to add.
[14:34] <rvba> The mac is the information that the outside world knows so it makes sense.
[14:35] <roaksoax> allenap: btw is that part of the maas cli?
[14:37] <allenap> roaksoax: It is, and it will need a wrapper. It's very simple though: python -m provisioningserver
[14:39] <roaksoax> allenap: right, and what is the maas-cli stuff
[14:39] <roaksoax> allenap: err apiclient
[14:39] <allenap> roaksoax: Don't know :-/
[14:43] <roaksoax> ok :)
[16:17] <roaksoax> rvba: ping
[16:17] <roaksoax> rvba: do you how how could you effectively run yui tests?
[16:20] <rvba> roaksoax: the yui tests are run as part of the test suite but you can run only the yui tests.
[16:20] <roaksoax> rvba: right, I'm looking towards running them in the actual yui3 source package from the archives
[16:20] <roaksoax> rvba: so I was wondering how is it that you run them?
[16:21] <roaksoax> and get the ooutput of them as?
[16:21] <rvba> In MAAS, we use SST to do that (see src/maasserver/tests/test_js.py).
[16:30] <smoser> rvba, bug 1027154
[16:30] <ubot5> Launchpad bug 1027154 in maas (Ubuntu) "need way to get system based on mac address" [Undecided,New] https://launchpad.net/bugs/1027154
[16:31] <rvba> smoser: ta
[16:32] <roaksoax> rvba: cool thanks!
[20:12] <smoser> anyone able to +1 this:
[20:12] <smoser> https://code.launchpad.net/~smoser/maas/vdenv-fix-mawk/+merge/116067
[20:12] <smoser> roaksoax, ^
[20:12] <roaksoax> smoser: give me a sec :)
[20:16] <roaksoax> smoser: done!