09:00:33 <jakeyip> #startmeeting magnum
09:00:33 <opendevmeet> Meeting started Wed Oct  4 09:00:33 2023 UTC and is due to finish in 60 minutes.  The chair is jakeyip. Information about MeetBot at http://wiki.debian.org/MeetBot.
09:00:33 <opendevmeet> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
09:00:33 <opendevmeet> The meeting name has been set to 'magnum'
09:00:55 <jakeyip> Agenda:
09:00:56 <jakeyip> #link https://etherpad.opendev.org/p/magnum-weekly-meetingy
09:01:03 <jakeyip> #topic Roll Call
09:01:13 <dalees> o/
09:01:17 <jakeyip> o/
09:01:59 <mnasiadka> o/
09:02:36 <jakeyip> let's wait 5 mins
09:08:28 <dalees> hi mkjpryor , currently rollcall for meeting if you're here to join/listen in.
09:08:38 <dalees> (we're pausing 5min for others to join)
09:09:07 <mkjpryor> :)
09:09:47 <jakeyip> hi mkjpryor
09:10:23 <jakeyip> alright let's have a quick CAPI update
09:11:19 <mkjpryor> from our side, we are now working mostly in our out-of-tree driver
09:11:28 <jakeyip> #topic CAPI
09:11:41 <mkjpryor> I think JohnG is attempting to keep the upstream patches in sync
09:12:00 <dalees> mkjpryor: thanks from us, we're using and testing with it. It builds and upgrades clusters :)
09:12:25 <mkjpryor> That's good! It didn't for a while during the refactor :)
09:12:41 <jakeyip> mkjpryor: did I break it? :P
09:12:53 <dalees> yup, i noticed. John fixed things before I got the PRs created :)
09:13:04 <mkjpryor> I have also been making changes to the underlying Helm charts to improve the CI
09:13:18 <jakeyip> nice
09:13:20 <mkjpryor> So we now test upgrading between Kubernetes versions and chart versions in the CI
09:14:03 <mkjpryor> As of yesterday, we have added an explicit approval step that means we are now more comfortable running the CI on external PRs as well
09:14:12 <dalees> mkjpryor: we'd be keen to keep the flatcar support PR moving, it's our primary target for deployment. I'll add a patch for the driver that will go alongside (and be updated as the chart changes to a better model from your suggestions)
09:14:46 <mkjpryor> dalees: There are some changes about to merge that change the way we construct the kubeadmConfigSpec
09:15:43 <mkjpryor> Basically, the spec can now be built in layers that are merged together instead of having to pull out bits and render them later
09:16:06 <mkjpryor> It should make the Flatcar and Keystone webhook patches much simpler
09:16:55 <dalees> great, we'll keep an eye out and look to rebase the PRs onto that.
09:17:56 <mkjpryor> :)
09:18:21 <mkjpryor> We also have another external contributor adding Manila support
09:18:39 <mkjpryor> And we support 1.28 :)
09:19:48 <mkjpryor> I have also been working on a patch for CAPO to allow us to have additional data volumes for machines. The primary use case for this is to put etcd on a different disk, for sites that only have a small pool of SSD.
09:20:19 <mkjpryor> There will be a follow-up patch to allow making use of the ephemeral disk for a flavor, so etcd can be put on that while the root volume comes from Cinder.
09:21:59 <dalees> cool - I have a patch that will go up soon to set the root volume type and size within Magnum (just sending through values that Magnum config already has settings for)
09:22:46 <mkjpryor> Sounds good
09:23:04 <jakeyip> is that from the current labels like docker_volume_size, ?
09:23:54 <dalees> no, config file - CONF.cinder.default_boot_volume_size and CONF.cinder.default_boot_volume_type
09:24:27 <dalees> there is more detail there my patch misses, i suspect. Especially with multiple volumes
09:25:53 <jakeyip> ah ok.
09:26:32 <opendevreview> Michal Nasiadka proposed openstack/magnum master: devstack: Install sonobuoy and kubectl  https://review.opendev.org/c/openstack/magnum/+/893823
09:28:02 <jakeyip> keep chatting... :D
09:29:38 <opendevreview> Michal Nasiadka proposed openstack/magnum master: Add kubernetes fedora coreos v1 jobs  https://review.opendev.org/c/openstack/magnum/+/896244
09:30:43 <dalees> mkjpryor: got a question for you about the cluster-api-addon-provider - we're keen to get it supporting `oci://` hosted charts, which might mean a library change to something that supports helm3. any thoughts before we look closer?
09:32:55 <jakeyip> So we met with mnaser. We talked about supporting VEXXHOST as an out of tree driver and how CI will work. He made good points about how Kubernetes development is now in Github and it's easier to attract devs there.
09:34:15 <jakeyip> I think the plan may be to shrink Magnum into an API shell with the ability to support multiple drivers. So we do away with all the issues that has been crippling Magnum.
09:36:03 <jakeyip> whether we will have an in tree reference CAPI driver or not is still unclear. first we have to find an elegant solution to support multiple drivers.
09:39:45 <jakeyip> mnasiadka think this will require a spec and I agree.
09:41:03 <jakeyip> anything else for CAPI?
09:42:04 <dalees> for the capi helm driver, all the manifests live in the chart repo, and the magnum template can specify the version. It may be that with this, an old Magnum can support a very new k8s with no release or code changes.
09:42:32 <dalees> ^ referring to the driver model (whether it's in tree or out, i suppose)
09:43:12 <dalees> nothing else on CAPI from me.
09:43:42 <opendevreview> Michal Nasiadka proposed openstack/magnum-tempest-plugin master: Add job for cluster tests  https://review.opendev.org/c/openstack/magnum-tempest-plugin/+/893131
09:44:02 <jakeyip> nice, thanks
09:44:33 <jakeyip> #topic CI
09:44:44 <mnasiadka> Yay
09:44:59 <jakeyip> as we can all see mnasiadka is hard at work doing CI, and it's coming together. very exciting
09:45:02 <mnasiadka> I've been working on a patch
09:45:04 <mnasiadka> #link https://review.opendev.org/c/openstack/magnum-tempest-plugin/+/893131
09:45:23 <mnasiadka> there's a depends-on that downloads kubectl and sonobuoy
09:45:30 <mnasiadka> #link https://review.opendev.org/c/openstack/magnum/+/893823
09:45:53 <mnasiadka> and then finally there's a patch to magnum repo to use those jobs with different kube_tags
09:45:56 <mnasiadka> #link https://review.opendev.org/c/openstack/magnum/+/896244
09:46:28 <mnasiadka> For now those jobs don't use a load balancer - but will follow up after it gets merged with using OVN LBs to save on memory in the runner nodes (amphora VMs are quite big)
09:46:40 <mnasiadka> And then add sonobuoy, and add multinode
09:47:00 <mnasiadka> And then we'll get into hell of maintaining those jobs and catching out things that fail :)
09:47:32 <dalees> nice work mnasiadka
09:47:48 <jakeyip> great work :)
09:48:21 <mnasiadka> I was thinking for starters we could have two jobs - one for 1.26 and one for 1.27 (since that's in our support matrix for B)
09:48:33 <mnasiadka> and each of those versions with a flannel and calico variant
09:49:11 <mnasiadka> any comments? ideas?
09:50:49 <jakeyip> sounds good. I'm anxious about carbon footprint of our tests; they take so long
09:52:28 <mnasiadka> You haven't seen other projects testing :)
09:52:59 <jakeyip> if only devstack can be containerised...
09:53:39 <mkjpryor> I would like a devstack that runs in a standard GitHub runner :)
09:54:03 <jakeyip> I'd be happy to stick with calico only and expand if there is a need. compared to what we had which is nothing, it's already great
09:54:03 <mkjpryor> But there would be no space left for test workloads unfortunately
09:54:35 <mnasiadka> mkjpryor: vexxhost is using paid github actions nodes for that...
09:54:44 <jakeyip> hahaha
09:54:57 <mnasiadka> so just do a Calico version of the CI jobs? fine by me
09:55:01 <mnasiadka> I never liked flannel :)
09:55:19 <dalees> +1 Calico
09:55:34 <jakeyip> we use fannel but +1 calico :P
09:55:59 * dalees eyes up Cilium... one day.
09:56:09 <jakeyip> yeah I am interested too
09:56:11 <jakeyip> CAPI can do it?
09:56:22 <dalees> CAPI can do anything
09:57:19 <dalees> yeah, it's just a CNI so it *can*; not sure if it's in the helm charts yet though. Vexxhost driver is calico only for now.
09:57:19 <jakeyip> goody
09:57:35 <jakeyip> ok. think we are at the end. anyone has anything else?
09:57:40 <jakeyip> #topic Open Discussion
09:58:20 <jakeyip> I think matt let so nothing else, let's end the meeting before the hour :)
09:58:44 <dalees> all good, thanks
09:58:48 <jakeyip> #endmeeting