Available as of v2.5+

Kubernetes is moving away from maintaining cloud providers in-tree. vSphere has an out-of-tree cloud provider that can be used by installing the vSphere cloud provider and cloud storage plugins.

This page covers how to migrate from the in-tree vSphere cloud provider to out-of-tree, and manage the existing VMs post migration.

It follows the steps provided in the official vSphere migration documentation and provides the steps to be performed in Rancher.

Cloud-config Format Limitation

Existing volumes that were provisioned using the following cloud-config format will NOT get migrated due to an existing bug in vsphere CSI.

If the cloud-config has this format for datastore and resource pool path, vsphere CSI driver cannot recognize it:

  1. default-datastore: </datacenter>/datastore/<default-datastore-name>
  2. resourcepool-path: "</datacenter>/host/<cluster-name>/Resources/<resource-pool-name>"

Volumes provisioned with the in-tree provider using the following format will get migrated correctly:

  1. default-datastore: <default-datastore-name>
  2. resourcepool-path: "<cluster-name>/Resources/<resource-pool-name>"

Upstream bug: https://github.com/kubernetes-sigs/vsphere-csi-driver/issues/628

Rancher issue tracking this bug: https://github.com/rancher/rancher/issues/31105

Prerequisites

  • vSphere CSI Migration requires vSphere 7.0u1. In order to be able to manage existing in-tree vSphere volumes, upgrade vSphere to 7.0u1.
  • The Kubernetes version must be 1.19 or higher.

Migration

1. Install the CPI plugin

Before installing CPI, we need to taint all nodes with node.cloudprovider.kubernetes.io/uninitialized=true:NoSchedule.

This can be done by running the following commands:

  1. curl -O https://raw.githubusercontent.com/rancher/helm3-charts/56b622f519728378abeddfe95074f1b87ab73b1e/charts/vsphere-cpi/taints.sh

Or:

  1. wget https://raw.githubusercontent.com/rancher/helm3-charts/56b622f519728378abeddfe95074f1b87ab73b1e/charts/vsphere-cpi/taints.sh
  2. chmod +x taints.sh
  3. ./taints.sh <path to kubeconfig if running the command outside the cluster>

Once all nodes are tainted by the running the script, launch the Helm vSphere CPI chart.

  1. From the Cluster Explorer view, go to the top left dropdown menu and click Apps & Marketplace.
  2. Select the vSphere CPI chart.
  3. Fill out the required vCenter details and click Launch.

vSphere CPI initializes all nodes with ProviderID, which is needed by the vSphere CSI driver.

Check if all nodes are initialized with the ProviderID with the following command:

  1. kubectl describe nodes | grep "ProviderID"

2. Install the CSI driver

  1. From the Cluster Explorer view, go to the top left dropdown menu and click Apps & Marketplace.
  2. Select the vSphere CSI chart.
  3. Fill out the required vCenter details and click Launch.
  4. Set Enable CSI Migration to true.
  5. This chart creates a StorageClass with the csi.vsphere.vmware.com as the provisioner. You can provide the URL of the datastore to be used for CSI volume provisioning while creating this StorageClass. The datastore URL can be found in the vSphere client by selecting the datastore and going to the Summary tab. Fill out the details for the StorageClass and click Launch.

3. Edit the cluster to enable CSI migration feature flags

  1. While editing the cluster, if the Kubernetes version is less than 1.19, select Kubernetes version 1.19 or higher from the Kubernetes Version dropdown.
  2. For enabling feature flags, click on “Edit as YAML”, and add the following under kube-controller and kubelet:

    1. extra_args:
    2. feature-gates: "CSIMigration=true,CSIMigrationvSphere=true"

4. Drain worker nodes

Worker nodes must be drained during the upgrade before changing the kubelet and kube-controller-manager args.

  1. Click Edit as Form and then click on “Advanced Options.”
  2. Set the field Maximum Worker Nodes Unavailable to count of 1.
  3. To drain the nodes during upgrade, select Drain Nodes > Yes.
  4. Set Force and Delete Local Data to true.
  5. Click Save to upgrade the cluster.