Migrating Azure In-tree to Out-of-tree

Kubernetes is moving away from maintaining cloud providers in-tree.

Starting with Kubernetes 1.29, in-tree cloud providers have been disabled. You must disable DisableCloudProviders and DisableKubeletCloudCredentialProvider to use the in-tree Azure cloud provider or migrate from in-tree cloud provider to out-of-tree provider. You can disable the required feature gates by setting feature-gates=DisableCloudProviders=false as an additional argument for the cluster’s Kubelet, Controller Manager, and API Server in the advanced cluster configuration. Additionally, set DisableKubeletCloudCredentialProvider=false in the Kubelet’s arguments to enable in-tree functionality for authenticating to Azure container registries for image pull credentials. See upstream docs for more details.

In Kubernetes v1.30 and later, the in-tree cloud providers have been removed. Rancher allows you to upgrade to Kubernetes v1.30 when you migrate from an in-tree to out-of-tree provider.

To migrate from the in-tree cloud provider to the out-of-tree Azure cloud provider, you must stop the existing cluster’s kube controller manager and install the Azure cloud controller manager.

If it’s acceptable to have some downtime during migration, follow the instructions to set up an external cloud provider. These instructions outline how to configure the out-of-tree cloud provider for a newly provisioned cluster. During set up, there will be some downtime, as there is a time gap between when the old cloud provider stops running and when the new cloud provider starts to run.

If your setup can’t tolerate any control plane downtime, you must enable leader migration. This facilitates a smooth transition from the controllers in the kube controller manager to their counterparts in the cloud controller manager.

Important:

The Kubernetes cloud controller migration documentation states that it’s possible to migrate with the same Kubernetes version, but assumes that the migration is part of a Kubernetes upgrade. Refer to the Kubernetes documentation on migrating to use the cloud controller manager to see if you need to customize your setup before migrating. Confirm your migration configuration values. If your cloud provider provides an implementation of the Node IPAM controller, you also need to migrate the IPAM controller.

Starting with Kubernetes v1.26, in-tree persistent volume types kubernetes.io/azure-disk and kubernetes.io/azure-file are deprecated and no longer supported. There are no plans to remove these drivers following their deprecation, however you should migrate to the corresponding CSI drivers, disk.csi.azure.com and file.csi.azure.com. To review the migration options for your storage classes and upgrade your cluster to use Azure Disks and Azure Files CSI drivers, see Migrate from in-tree to CSI drivers.

• RKE2
• RKE

1. Update the cluster config to enable leader migration:

spec:
  rkeConfig:
    machineSelectorConfig:
      - config:
          kube-controller-manager-arg:
            - enable-leader-migration
        machineLabelSelector:
          matchExpressions:
            - key: rke.cattle.io/control-plane-role
              operator: In
              values:
                - 'true'

Note that the cloud provider is still azure at this step:

spec:
  rkeConfig:
    machineGlobalConfig:
      cloud-provider-name: azure

1. Cordon control plane nodes so that Azure cloud controller pods run on nodes only after upgrading to the external cloud provider:

kubectl cordon -l "node-role.kubernetes.io/control-plane=true"

1. To deploy the Azure cloud controller manager, use any of the available options:

• UI: Follow steps 1-10 of Helm chart installation from UI to install the cloud controller manager chart.
• CLI: Follow steps 1-4 of Helm chart installation from CLI.
• Update the cluster’s additional manifest: Follow steps 2-3 to install the cloud controller manager chart.

Confirm that the chart is installed but that the new pods aren’t running yet due to cordoned controlplane nodes.

1. To enable leader migration, add --enable-leader-migration to the container arguments of cloud-controller-manager:

kubectl -n kube-system patch deployment cloud-controller-manager \
--type=json \
-p='[{"op": "add", "path": "/spec/template/spec/containers/0/args/-", "value": "--enable-leader-migration"}]'

1. Update the provisioning cluster to change the cloud provider and remove leader migration args from the kube controller manager. If upgrading the Kubernetes version, set the Kubernetes version as well in the spec.kubernetesVersion section of the cluster YAML file.

spec:
  rkeConfig:
    machineGlobalConfig:
      cloud-provider-name: external

Remove enable-leader-migration from the kube controller manager:

spec:
  rkeConfig:
    machineSelectorConfig:
      - config:
          kube-controller-manager-arg:
            - enable-leader-migration
        machineLabelSelector:
          matchExpressions:
            - key: rke.cattle.io/control-plane-role
              operator: In
              values:
                - 'true'

1. Uncordon control plane nodes so that Azure cloud controller pods now run on nodes:

kubectl uncordon -l "node-role.kubernetes.io/control-plane=true"

1. Update the cluster. The cloud-controller-manager pods should now be running.

kubectl rollout status deployment -n kube-system cloud-controller-manager
kubectl rollout status daemonset -n kube-system cloud-node-manager

1. The cloud provider is responsible for setting the ProviderID of the node. Check if all nodes are initialized with the ProviderID:

kubectl describe nodes | grep "ProviderID"

1. (Optional) You can also disable leader migration after the upgrade, as leader migration is not required with only one cloud-controller-manager. Update the cloud-controller-manager deployment to remove leader migration from the container arguments:

- --enable-leader-migration=true

1. Update the cluster config to enable leader migration in cluster.yml:

services:
  kube-controller:
    extra_args:
      enable-leader-migration: "true"

Note that the cloud provider is still azure at this step:

cloud_provider:
  name: azure

1. Cordon the control plane nodes, so that Azure cloud controller pods run on nodes only after upgrading to the external cloud provider:

kubectl cordon -l "node-role.kubernetes.io/controlplane=true"

1. To install the Azure cloud controller manager, follow the same steps as when installing Azure cloud provider on a new cluster:

• UI: Follow steps 1-10 of Helm chart installation from UI to install the cloud controller manager chart.
• CLI: Follow steps 1-4 of Helm chart installation from CLI to install the cloud controller manager chart.

1. Confirm that the chart is installed but that the new pods aren’t running yet due to cordoned controlplane nodes. After updating the cluster in the next step, RKE will upgrade and uncordon each node, and schedule cloud-controller-manager pods.

2. To enable leader migration, add --enable-leader-migration to the container arguments of cloud-controller-manager:

kubectl -n kube-system patch deployment cloud-controller-manager \
--type=json \
-p='[{"op": "add", "path": "/spec/template/spec/containers/0/args/-", "value": "--enable-leader-migration"}]'

1. Update cluster.yml to change the cloud provider to external and remove the leader migration arguments from the kube-controller.

rancher_kubernetes_engine_config:
  cloud_provider:
    name: external

Remove enable-leader-migration if you don’t want it enabled in your cluster:

services:
  kube-controller:
    extra_args:
      enable-leader-migration: "true"

1. If you’re upgrading the cluster’s Kubernetes version, set the Kubernetes version as well.

2. Update the cluster. The cloud-controller-manager pods should now be running.

kubectl rollout status deployment -n kube-system cloud-controller-manager
kubectl rollout status daemonset -n kube-system cloud-node-manager

1. The cloud provider is responsible for setting the ProviderID of the node. Verify that all nodes are initialized with the ProviderID:

kubectl describe nodes | grep "ProviderID"

1. (Optional) You can also disable leader migration after the upgrade, as leader migration is not required with only one cloud-controller-manager. Update the cloud-controller-manager deployment to remove leader migration from the container arguments:

- --enable-leader-migration=true