[00:00:13] Everytime I watch a talk on testing I get motivated to do unit and integration testing, then after about a week I once again realise that writing tests takes 5 times longer than writing code D: [00:00:23] yeah [00:00:41] I had that experience with the bgp support I added to nova [00:00:53] they don't accept code without tests, though [00:02:28] I'm hoping travisci released private repo support so I can run tests on pull requests.... doing the whole code review thing is sorta pointless when you have 3 developers that are spread all over the world and you need to deploy every few hours. [00:03:36] ugh [00:03:48] someone −1'd me saying the code probably belongs in quantum [00:03:59] but quantum doesn't have floating IP support, so how the hell am I going to add it there [00:04:25] lol [00:05:05] I assume your patch was to route public ips directly to the interfaces rather than snating everything off the lan? [00:05:23] no [00:05:48] it was to use BGP for floating IP announcements rather than ARP [00:05:54] so that you don't need subnets [00:06:02] and can use arbitrary IPs [00:06:10] Oooh [00:06:25] That makes sense, I forgot openstack expects blocks at a time [00:06:34] actually, it doesn't [00:06:48] you can add individual IPs to the floating list (that's how I've been doing it) [00:07:07] we have a single network node, and so we have static routes set to go to that one node [00:08:36] Hmm [00:09:24] if we use BGP, we can remove the static routes and use multi-node networking [00:09:29] I'm trying to picture how the nodes are actually laied out logically for labs specific use case and just confused myself lol. [00:09:45] then each network node will announce their floating IPs [00:10:41] So all the nodes have public ips and you want to put a network node on every compute node then have each node advertise to the edge routers and have the floating ips routed to the node ip then nat it to the host? [00:11:02] NAT to the instance, but otherwise yes [00:11:17] errr yeah I meant instance [00:11:49] That makes sense, I was thinking that the nodes just arped for the ips then NAT'd them but that would have all sorts of weird failover issues with timeouts on a node failure I guess [00:11:59] it's an elegant solution, and best part is that it should make the switchover on migrations faster, too [00:12:15] it currently does artp [00:12:17] *arp [00:12:30] in our set up, the arps don't matter, though [00:12:40] since we have static routes configured [00:13:12] my laptop's keyboard keys are starting to pop off :( [00:15:06] It really sucks when that happens :( [00:15:46] yep. maybe I need one of those new macbook pros with the pretty screeen [00:15:59] can't really replace the keyboard [00:16:24] I quite like my not shiny mbp, shiny mbp's are too expensive for a general use laptop [02:44:51] 07/07/2012 - 02:44:51 - User laner may have been modified in LDAP or locally, updating key in project(s): deployment-prep [04:52:10] 07/07/2012 - 04:52:10 - Creating a home directory for ning at /export/keys/ning [04:53:10] 07/07/2012 - 04:53:10 - Updating keys for ning at /export/keys/ning [06:24:10] 07/07/2012 - 06:24:10 - Updating keys for ning at /export/keys/ning [08:06:26] !info [08:06:26] https://www.mediawiki.org/wiki/WMF_Projects/Wikimedia_Labs [08:06:35] oh yeah [08:06:37] wrong channel [08:06:39] * Beetstra hides [21:42:10] 07/07/2012 - 21:42:10 - Creating a home directory for kghbln at /export/keys/kghbln [21:43:10] 07/07/2012 - 21:43:10 - Updating keys for kghbln at /export/keys/kghbln