[08:00:42] greetings [09:58:12] morning [16:39:58] andrewbogott: Do you know if there is any equivalent of the https://docs.openstack.org/magnum/latest/user/#supported-labels documentation that covers the magnum-cluster-api driver relevant settings? This is what I have used in the past to configure new magnum templates. [16:42:12] I have been having a pretty hard time finding any docs on the magnum-cluster-api provider in the "normal" places. I did eventually stumble on https://vexxhost.github.io/magnum-cluster-api/user/configs/ [16:48:32] * dhinus off [17:02:50] * andrewbogott is in a jet-lagged haze [17:03:20] * bd808 forgot SREs are in not-normal places this week [17:03:55] bd808, the most interesting label is kube_tag which should needs to correspond to the version of kubernetes in the worker image [17:04:11] of which only one is really provided at the moment [17:04:23] Are there other things you want to set? [17:05:26] * andrewbogott digs up yubikey to check [17:05:55] andrewbogott: *nod* so I should stick with kube_tag=v1.36.5 because v1.36.6 is out but not in the allowed base image. [17:07:35] The heat centric docs documented various combinations of components that had been tested together. Does that all happen in the base image now rather than being a thing that can be tuned in the tags? [17:07:57] correct. And if you did change the kube_tag it would have to go along with a change to image_id [17:09:11] If most everything is baked into the base image now that makes things simpler from the config POV, although maybe also more toil that we have to push down to your team for updates. [17:09:15] I thought there was a half-baked support matrix someplace... but for the most part IME the magnum-capi driver isn't super configurable and you generally get all the features turned on like it or not. [17:10:00] capi is pretty configurable but the marshalling would have to be in the magnum driver [17:10:30] I can paste a known good template if you want, although maybe you're already looking at that [17:11:24] cinder_csi_enabled and cloud_provider_enabled are I think the things I'm still expecting to set true in my config. I haven't actually spun up a cluster yet for testing. [17:11:32] * andrewbogott discovers the phrase 'Ubuntu Bubernetes' in the docs he wrote [17:13:23] You should get cinder_csi by default with the new driver [17:13:48] you're using cloud_provider_enabled just to enable the cinder bits? [17:16:36] btw I'm sorry that I broke all of your work and/or invalidated the docs you wrote for the previous driver. In theory the new driver is more reliable, but at this point it's what we get, like it or not. [17:21:42] The 'kubernetes-full-v1.36.5' template will attach an 80GB cinder volume to each worker node, which I think is similar to what you want. [17:29:56] In the zuul cluster I was/Dan is using cloud_provider_enabled in part to get access to the keystone auth management stuff so we can delegate admins for the cluster. https://gitlab.wikimedia.org/repos/releng/zuul/tofu-provisioning#kubeconfig-for-accessing-managed-cluster [17:33:23] A `docker_volume_type=high-iops` label was needed in the CAPI stuff. [17:36:08] ok. [17:36:56] I think the best bet is to try it and at least see if you can get a cluster up. And then if bits don't work we can go code-diving (and potentially I can submit patches to forward on the labels if they don't get handled already) [17:37:43] I'm pretty sure there's no comprehensive list of "here are the magnum things that actually work with this driver" because i've asked upstream for it before. But the right way to get one is probably to submit a patch that includes a wrong one :D [17:38:54] (right now I need to go walk around so I don't fall asleep again but I'll be back in a few hours) [17:39:24] Also, dduvall has been down this road already so he may have some answers. [17:39:43] ty for working on this! [17:47:21] * dduvall reads [17:52:35] bd808: https://gitlab.wikimedia.org/repos/releng/zuul/tofu-provisioning/-/commit/0592d881b6410aef240ae77dd61bc8702bdda36d is the closest thing i can find that summarizes what i changed to get a CAPI based cluster up (without trying to thaw my brain's cold storage) [17:54:28] dduvall: thanks. I had skimmed that but I will look more closely. [17:55:39] let me know if you want to pair on it [17:56:01] I was maybe confused about if the current config in that repo is for the right backend. T437498 reads like that cluster is not right yet either. [17:56:02] T437498: Rebuild the cloud-runner-staging magnum cluster in the zuul3 project with the new magnum-cluster-api driver - https://phabricator.wikimedia.org/T437498 [20:25:32] (much later) it's two different clusters. 'zuul-k8s-v134' is running with the new driver, 'cloud-runner-staging' is not.