[04:36] <jtv> bigjools: I get the impression our support for DNS management is brittle in the face of multiple networks...  we always map a node's hostname to its first-registered NIC, instead of picking a NIC that's on the DNS-managed network.
[04:37] <jtv> The right thing to do AIUI is to map the hostname to each IP address that the node has on a network for which we manage DNS.
[04:39] <jtv> This means AFAICS that the fix will change DNS mappings for setups where the nodes are on multiple networks, when the first interface that gets registered is not on the network for which we manage DNS.
[05:05] <bigjools> "for which", I like it :)
[05:06] <bigjools> jtv: I think we need to set the "primary" network to which the first IP is assigned in the DNS
[05:06] <bigjools> jtv: however not fixing this now would not be a regression
[05:07] <bigjools> anyway can you get "make run" actually serving anything on trusty?
[05:08] <jtv> I'll try it.
[05:09] <jtv> My gramatically awkwardly laboured point is that this behaviour _does_ change if we're going to manage DNS for multiple networks.
[05:09] <bigjools> jtv: what I meant was that we don't have to manage multiple DNS networks
[05:09] <bigjools> stick to the single for now
[05:10] <bigjools> and we can solve this problem later
[05:10] <jtv> !
[05:10] <bigjools> why the !
[05:10] <jtv> That's what I documented in my phased approach, and you said absolutely we have to do DNS on multiple interfaces.
[05:10] <bigjools> we do - but I am saying it can wait until the end
[05:11] <jtv> The end of what?
[05:11] <bigjools> the cycle/feature
[05:11] <bigjools> anyway there's two things getting conflated here
[05:12] <bigjools> management of cluster interfaces does not have anything to do with which NIC on a node has its IP used as its dns lookup
[05:12] <bigjools> since the cluster may not be connected to all the same networks as the node
[05:12] <bigjools> they only need to share one
[05:13] <bigjools> jtv: want to do a hangout to talk more?
[05:13] <jtv> Yes please.  Meanwhile, no dice on Trusty — ISE.  :(
[05:13] <jtv> Let me get my equipment ready for a hangout.
[05:13] <bigjools> fnar
[05:13] <jtv> This will include a heavy object to hurl at the crying baby.
[05:13] <bigjools> I didn't want that kind of hangout
[05:13] <jtv> Oh zark off.
[05:15] <bigjools> call me when yer ready
[05:16] <jtv> Well, _that_ boot didn't work, so let me try another one.
[05:16] <jtv> Logging in with your password sort of requires a response to keyboard input.
[05:18] <jtv> Ah, I can switch to a text console and back, and then the login screen accepts input.
[05:19] <bigjools> !
[05:19] <jtv> It's an alpha.
[05:19] <jtv> By the way, the error I get from maas running from branch in Trusty is an import error involving django's debug toolbar.  Let's try uninstalling that.
[05:19] <jtv> Good thing we have oops reports!
[05:19] <bigjools> all I get is an oops page
[05:20] <jtv> Me too, but it does generate an oops report with the traceback.
[05:20] <lifeless> oops
[05:20] <bigjools> jtv: I don't get any traceback or nuffink
[05:20] <bigjools> hey lifeless
[05:20] <bigjools> you lurker you
[05:21] <lifeless> noone expects the NZ inquisition!
[05:23] <jtv> Hey there lifeless
[06:02] <bigjools> jtv: so I think our django debug toolbar is an old egg, I'm trying to get a new one
[06:04] <jtv> Trying that here too.
[06:05] <bigjools> jtv: setting it to 1.0.1
[06:05] <bigjools> jtv: btw djorm-ext-pgarray is at 0.9 now
[06:05] <jtv> Oh, then we may no longer need our workaround.
[06:06] <bigjools> have a looksee and let me know
[06:06] <jtv> Prioritising the nodegroupinterface work though...
[06:08] <bigjools> debug_toolbar/base.html missing now
[06:08] <bigjools> fair enough
[06:09] <bigjools> hurray fixed
[06:10] <jtv> \o/
[06:10] <jtv> Failing for me:
[06:10] <jtv> Error: Picked: sqlparse = 0.1.10
[06:11] <bigjools> landing a branch
[06:11] <jtv> While installing repl.
[06:11] <bigjools> one mo
[06:11] <jtv> OK
[06:13] <bigjools> give the lander 10 minutes
[06:13]  * bigjools makes coffee
[06:55] <jtv> bigjools: works for me too now!
[06:55] <bigjools> hurray!
[08:22] <roaksoax> bigjools: ping
[08:23] <jtv> Hi roaksoax — try again in an hour, or after our standup.
[08:23] <roaksoax> jtv: hehe, i'll be asleep in an hour (and probably should be sleeping now)
[08:23] <jtv> !
[08:23] <jtv> I thought you were just up early!
[08:23] <roaksoax> nah
[08:23] <roaksoax> i'
[08:23] <roaksoax> :)
[08:24] <roaksoax> jtv: anyway going to bed but will leave bigjools with my opne question
[08:24] <jtv> OK nn!
[08:24] <roaksoax> bigjools: since otherwise we wont be able to talk until sunday night (my time), your monday morning, I was wondering what's the status of adding upstream support to MAAS for HA? The charm is coming along pretty good and will soon start looking at that HA stuff for maas-region!
[08:25]  * roaksoax out
[08:26] <jtv> rvba: I put a few cards on the board for immediate next steps towards multiple-managed-interfaces.  They have detailed descriptions.
[08:26] <rvba> jtv: okay, I'll have a look.
[08:27] <jtv> Thanks.
[08:35] <rvba> jtv: AFAIK, we can have the same DHCP server listen to multiple interfaces and have one config per interface.  That's all good.  But can we do something similar with DNS?
[08:36] <jtv> rvba: depends on how similar you mean...  that's the region controller, of course, so it's going to listen on all of the controller's interfaces.
[08:37] <jtv> But any given node is only going to be on one managed network, apparently.
[08:37] <jtv> And that's why there is no confusion between the networks.
[08:37] <rvba> Do you mean that there is one (and only one) network marked as "the management network"?
[08:42] <jtv> From the cluster controller's perspective, no.  But from the node's perspective, yes.
[08:51] <rvba> Care to explain?
[08:56] <rvba> jtv: also, shouldn't we try splitting the card named "Handle multiple cluster net interfaces" into smaller tasks?
[09:00] <jtv> rvba: why do you think there are cards for jobs that are part of that project?
[09:01] <jtv> About the networks: a cluster will manage multiple networks, and a node will be in only one of them.
[09:01] <rvba> jtv: then I think there are things that are missing :).  Like update the generated DHCP config to account for the multiple interfaces.
[09:02] <rvba> Also, the parsing of the lease file will have to be changed (I think).
[09:03] <jtv> You're right — we'll have to update the DHCP config for multiple networks.
[09:04]  * rvba adds a card
[09:07] <jtv> That's currently all built around the assumption of a single network.  :(
[09:08] <rvba> Right, that's why I'm saying it will probably have to be changed.
[09:10] <jtv> I can't think of anything that would need changing in the parser.
[09:11] <rvba> Well, I'm just saying that's worth a look :).
[09:19] <rvba> jtv: btw, did you fix the problem in maas-test where the machine was not cleaned up?
[09:21] <jtv> rvba: no, and I can't test it now.  :(
[09:21] <rvba> jtv: you mean you can't reproduce the problem?
[09:22] <jtv> That's right.
[09:22] <jtv> And as I said I just don't see how the "finally" clause could fail to run...
[09:22]  * rvba tries
[09:53] <bigjools> roaksoax: HA is not going to happen I'm afraid.
[09:53] <bigjools> unless we find time out of nowhere
[11:48]  * gmb lunches
[13:37] <gmb> allenap: Are you able to help me unfuck my dev box? I accidentally rebooted halfway through apt-get remove maas last night and now maas-region-controller can't be fully installed or removed.
[13:50] <allenap> gmb: Sure :)