[13:54:04] hello folks [13:54:23] got bitten by https://phabricator.wikimedia.org/T335285 while trying to switch to ipv6 dual stack [13:55:01] it looks like all clusters have /64s at the moment, at least from what I see in kubernetes.yaml [13:56:41] shall we open a task to update all those configs? Not necessarily to update to dual stack, but at least to have /116 everywhere in case it will be needed [13:57:08] the next question would be - for ml-staging, this is the current subnet https://netbox.wikimedia.org/ipam/prefixes/541/ [13:57:37] what is the best procedure? Delete and re-create a new /116 under 2620:0:860:300::/56 ? [14:03:03] created https://phabricator.wikimedia.org/T353705 [14:05:43] I see from https://netbox.wikimedia.org/ipam/prefixes/387/prefixes/ that Janis used "NEW" for the /116, so I believe that I should just drop and re-create [14:05:52] akosiaris: o/ (if you have time) --^ [14:14:53] I would assume that drop and recreate is fine [14:14:59] super [14:15:53] o/ [14:16:05] yeah, I think you 'll need to drop and recreate [14:17:56] super proceeding [14:23:11] and now I don't recall if netbox can suggest a /116, if not listed among child prefixes [14:38:11] akosiaris: in terms of allocations, should we follow any practice? Like, I dropped 2620:0:860:303::/64 but I can pick up 2620:0:860:303::/116 and be done. Not sure if k8s will support more etc.., but probably not worth to come up with some scheme for our clusters? [14:39:03] I quickly checked with Riccardo (For netbox tips), we could allocate a /64 for each cluster and pick up subnets from there [14:40:32] elukey: it's probably not worth it to fret too much. We were targetting /64s to begin with and suddenly found ourselves having to target something 52bits smaller. [14:41:13] akosiaris: ack, last favor - do you mind to check if https://netbox.wikimedia.org/ipam/prefixes/886/ make sense? [14:41:22] My only suggestion would be, in order to avoid too much visual clutter in netbox to utilize the /11X that are close to the ones already allocated [14:41:32] and even then... it's not really a strong incentive [14:41:54] elukey: yeah, sure [14:43:24] okok I see what you mean with visual clutter [14:43:28] lemme try to pick another one [14:44:36] really, not a very strong incentive. [14:46:06] okok :) [14:51:56] all right https://gerrit.wikimedia.org/r/c/operations/puppet/+/984209 [15:55:07] rolled out ipv6 in ml-staging, all worked :) [16:01:24] aaaand https://github.com/istio/istio/issues/46625 [16:01:46] I am very sad