[10:47:05] godog: I was told you were involved in the setup of the cassandra's restbaseNNNN-{a,b,c} records, is that true or false information? :) [10:47:42] volans: "yes" *g* [10:47:55] seriously though, yes I was [10:48:20] so, those addresses are set on the hosts as /32 [10:48:29] although they are not VIPs and seems to be regular addresses [10:48:40] was that on purpose, required by cassandra or anything? [10:49:11] in theory would be better to have them with their proper netmask, and as a side effect this will help a lot the import of data into netbox [10:49:27] (we need a way to distinguish VIPs from regular IPs) [10:50:21] I don't remember the exact reason for /32 but it might just work with the subnet's netmask, it was required by cassandra multi instance at the time [10:51:12] yeah I spoke with Eric the other day, but didn't have/recall any specific reason, as long as the IP works it's ok :D [10:51:28] * volans would avoid to ask why additional IPs instead of ports to be gentle :) [10:51:39] s/would/will/ [10:52:23] from a networking POV it will work fine to renumber the /32 into a /somethingelse [10:55:09] dunno if it's puppetized somewhere but happy to help [10:55:32] I didn't looked yet at the puppet size, chasing all those corner cases [15:50:40] godog: thanks a lot for the reviews, do you think we should wait also for Eric's ok on the cassandra side? [15:51:07] I'm adding the additional tests in the meanwhile [15:55:49] the 1.1.1.1 case is trickier as that is a valid ip address... [15:56:21] and to the best of my knowledge ruby's stdlib doesn't have a better support for netmasks without using the netaddr gem [15:56:44] volans: np! IMHO yes better wait at least for an ack on Eric's and/or Hugh's side, unless it is urgent [15:57:40] practically speaking I think/hope the actual deployment would be transparent to cassandra, at least for adding a new address [15:58:07] I hope so too, that's the only thing that worries me, be sure we don't have network blips [15:58:12] thanks for testing it in beta [15:58:21] yeah if it is trickier feel free to skip the test [15:58:40] the test is easy to add, it's triky to make the test pass :D [15:58:48] that was a quick test btw as in didn't test the impact on cassandra, just the OS (FYI) [15:59:08] k [15:59:30] that change doesn't account for aliases to non-primary interfaces [15:59:34] but also, what is a "real address" :) [16:02:06] volans: ^ [16:03:04] paravoid: yes, that's actually a pebcak on my side, I had it stashed locally and forgot to re-add [16:03:10] because by default $interface=$facts['interface_primary'] [16:03:17] in interface::alias but can be overriden [16:03:20] correct [16:03:29] let me fix it [16:03:49] for the real address vs service address, good question [16:04:07] so, we have usage of alias for various things [16:04:19] cassandra-a/b/c instances that are unique to a single host [16:04:27] so a cassandra host has effectively 4 addresses [16:04:53] while for others the usage is to assign a sort of a VIP without being behind LVS [16:05:07] lists.wo, gerrit.w.o are some examples [16:06:03] there is a discussion on what we should actually import and automate through netbox [16:07:05] there is a section in the wikimedia.org zonefile called [16:07:05] ; Service aliases (alphabetical order) [16:07:32] we most likely don't want to automate those as they are manually defined and there is no good way to represent those in netbox [16:07:47] but if they are assigned to a host we will import them [16:08:01] those are what I'm calling "service" IPs [16:08:43] we could discuss the whole genesis of that ofc, I was trying to find a pragmatical solution that doesn't make the current situation worse [16:11:42] paravoid: ^^^ [16:11:59] happy to hear alternative solutions [16:22:18] what does the subnet mask matter? [16:22:21] why* [16:23:01] so, when I first saw the cassandra IPs I thought it was weird/an error that they didn't had the proper netmask [16:23:13] with this we could distinguish between them and the others [16:23:39] that being more like service IPs are created with a /32 or /128 like loopback addresses [16:23:48] I know that this sounds pretty ugly [16:27:26] and we can totally convert those too to the proper netmask after if that helps [16:28:01] those==the service IPs like gerrit, lists, etc.