Update to Kubernetes 1.24.12 + Use Go 1.19

[hrak]

This PR is a continuation of the work in #32 and contains the following changes:

• Use Go 1.19
• Update Kubernetes deps to 1.26.2 1.24.12
• Move the cloudstack-go dep to its new home in the apache org (and update to v2.15.0)
• switch to klog v2

I can provide a test container image if needed.

[hrak]

Please let me know if 1.26 i a bit too ambitious, i will gladly adapt for 1.24 or 1.25.

[joschi36]

If you have the time I would really appreciate if there is a version in between. For instance 1.24.

[hrak]

If you have the time I would really appreciate if there is a version in between. For instance 1.24.

yeah i was afraid of that already, i suddenly realized that compiling this against 1.26 probably means losing compatibility with 1.24 and 1.25 which are still supported. I will get on it!

[hrak]

Updated PR to use Kubernetes v1.24.11 instead

[hrak]

Ship it? Or does anything else need to happen?

[hrak]

Added a switch to klog v2 + updated PR for k8s 1.24.12

[RolphH]

@rohityadavcloud can you help getting this one merged?

[yadvr]

LGTM but I didn’t test it

[yadvr]

@davidjumani could you have a quick look and help merge it.

@RolphH @hrak could you advise what testing you’ve done against any/various versions of CloudStack?

[hrak]

@davidjumani could you have a quick look and help merge it.

@RolphH @hrak could you advise what testing you’ve done against any/various versions of CloudStack?

We are running this without any issues so far on Cloudstack 4.17. We have some more changes in the pipeline once this is merged.

[hrak]

We are running this without any issues so far on Cloudstack 4.17. We have some more changes in the pipeline once this is merged.

Well, maybe i should clarify: The part that works is the loadbalancer/firewall rules part. Most of the logic surrounding instance metadata is broken ever since this was turned into and external cloud provider. GetZone assumes this is running inside kubelet and that hostname=instance hostname, but it isn't, since the provider runs in a pod, causing it to try to look up the instance based on the pod hostname:

I0414 07:08:41.827320       1 cloudstack.go:175] GetZone called from /go/src/github.com/apache/cloudstack-kubernetes-provider/cloudstack_instances.go#198
E0414 07:08:41.874064       1 node_controller.go:220] error syncing 'k8s-mgmt-cp-dev-02': failed to get instance metadata for node k8s-mgmt-cp-dev-02: could not find instance for retrieving the zone: No match found for cloud-controller-manager-789b9fb546-dg46l: &{Count:0 VirtualMachines:[]}, requeuing

Then there are issues with ProviderID, since the provider is currently called external-cloudstack and should probably be called cloudstack, so the logic here matches the kubelet arguments --cloud-provider=external --provider-id=cloudstack:///8411c37d-8d63-4920-8cb2-01fa512c3377 in the same way that this works for f.e. the Openstack CCM. Since most cloud controllers will be external soon and there is no more internal cloudstack CCM the external- bit is kind of redundant anyway.

Passing --cloud-provider=external to the kubelet causes it to taint all nodes with node.cloudprovider.kubernetes.io/uninitialized=true:NoSchedule, which gets removed once the cloud provider is initialized properly. But because of above mentioned issues, the taint never gets removed, causing a non-functional cluster.

I am working on fixes for all these issues but i would really like to get this in first.

[yadvr]

Thanks @RolphH @hrak I'll merge this based on your tests and feedback. Pl do followup with other changes you think we need for the codebase.