search

vexxhost / magnum-cluster-api

Cluster API driver for OpenStack Magnum
Apache License 2.0
41 stars 17 forks source link

Support Kubernetes 1.29.x and 1.30.x #334

Open 0x00ace opened 3 months ago

0x00ace commented 3 months ago

Successfully booted Kubernetes 1.27.11 and upgraded to new Kubernetes versions 1.27.11 -> 1.28.7 -> 1.29.2 using commands:

 openstack coe cluster upgrade cluster_1 k8s-v1.28.7
 openstack coe cluster upgrade cluster_1 k8s-v1.29.2

Also, tested new images uploading to the custom container registry without any issues

0x00ace commented 3 months ago

@mnaser I have all sets of the images added. Also, I have updated k8s versions.

mnaser commented 2 months ago

@okozachenko1203 can you take into consideration your patch there?

KeplerSysAdmin commented 2 months ago

Kepler Team here, We would like to have this k8s version supported ASAP. Thank you

KeplerSysAdmin commented 1 month ago

v1.27 end of life is 2024-06-28 When can we merge this PR?

mnaser commented 4 weeks ago

recheck

0x00ace commented 3 weeks ago

@mnaser I have successfully created a 1.28.x cluster and have upgraded it to 1.29.x But to create a 1.29.x one needs to add a label keystone_auth_enabled=false

okozachenko1203 commented 3 weeks ago

recheck

0x00ace commented 2 weeks ago

For kubernetes >= 1.29.0 there is an issue with k8s-keystone-auth

Because of that issue, I need to pass a label keystone_auth_enabled=false

0x00ace commented 2 weeks ago

@mnaser Should I also include Kubernetes 1.30.x support in this PR? The issue with k8s-keystone-auth I have mentioned above is also valid for 1.30.x. Everything else looks to be working.

cgroschupp commented 1 week ago

It might be a good idea to split this PR to get a newer k8s version than 1.27 as soon as possible, since the end of life is in 6 days. And then work on the keystone-auth problems in another PR and release it later.

mnaser commented 1 week ago

I am struggling with finding a good way to manage the life cycle of these images other than "write a manual script" at this point. It seems like it's hard to find a way to get CI to react for changes, and the image builds are flakey, and the Sonobuoy tests are flakey too, and on top of it they eat a lot of resources...

0x00ace commented 1 week ago

@mnaser I will rework this PR to support only k8s 1.29 and 1.30 k8s 1.28 support in this PR

mnaser commented 1 week ago

@0x00ace thanks, actually I figured out a nice way to build images that is far more reliable.

https://github.com/vexxhost/capo-image-elements

I am almost done with the CI bit and then we can ignore the image build here and ship the images from there to consume here.

mnaser commented 1 week ago

Alright, 1.28 is in right now, I'm also trying to get image builds for EL9 working for 1.28, and then we can tackle this issue.

0x00ace commented 1 week ago

@mnaser Updated PR for support k8s 1.29.x and 1.30.x I'm using Rocky 9 with cilium and keystone_auth_enabled=false. Cluster creation, resizing, and adding new nodegroups look to be working without issues. Also, I have no issues with a cluster upgrade from 1.28 -> 1.29 -> 1.30.

mnaser commented 1 week ago

https://github.com/vexxhost/capo-image-elements/issues/4

This is what I'm struggling for EL9