terraform-provider-rancher2: Cluster Config loading error: Cannot read properties of undefined (reading 'length') when machineSelectorConfig is set to 'null'

Internal reference: SURE-6770 Reported in 2.7.5

Issue description: A user encountered an issue loading the Config tab for specific clusters from the Cluster Management page.

When going to the Cluster Management page, selecting a cluster and clicking the Config button, the page displays a “Loading …” text but ends up with the following message: Cannot read properties of undefined (reading 'length')

What the UI dev console outputs as errors are the following:

- console.js:31 Cannot read properties of null (reading 'cloud-provider-name')
- console.js:31 Cannot read properties of undefined (reading 'length')

The UI loads fine for 1 out of 3 clusters, and the user can’t find any configuration for the working cluster that mentions “cloud-provider-name”.

When comparing the cluster YAMLs again, they found the following difference: for the two clusters that have an issue with the config UI, they see the following property in the YAML:

• For the one cluster that works, this property is not present at all:

machineSelectorConfig:
  - config: null

This config block can have “cloud-provider-name”, so we advised the customer to try removing it or setting it to an empty string or an actual string value E.g.: machineSelectorConfig:

    machineSelectorConfig:
    - config:
        cloud-provider-name: azure
        protect-kernel-defaults: false
    registries: {} 

Removing this property or setting the value of config to {} instead of null fixed the UI issue in this case.

The user does not know what has caused this property to be updated (with the null value),  all their clusters are created in the same way using Terraform, and they have been able to read and modify the config from the management console for all clusters, but recently it has stopped working for 2 out of 3 clusters in one of their setups.

Business impact: The Cluster config page failed to load with any useful message, making it difficult to say what went wrong.

Workaround: Remove the "machineSelectorConfig: - config: null" code block completely, and the config starts loading.

Actual behavior: The Cluster config page failed to load or provide a useful message as to why it failed to load.

Expected behavior: The page should load okay, or the message should highlight that null is not an accepted config value.

Additional notes:

• config “null” should be an accepted setting for machineSelectorConfig as per this https://github.com/rancher/dashboard/pull/8242
• -  “cloud-provider-name” might not be okay with defaulting to null,  https://github.com/rancher/kontainer-driver-metadata/blob/dev-v2.7/channels-rke2.yaml#L101-L110

• Update from the user:

I was experimenting with changing RKE config parameters via Terraform and noticed that the rancher2 provider wants to change the machineSelectorConfig parameter when changing another parameter, such as etcd snapshot retention. I attempted this with the newest version of the rancher2 provider: https://registry.terraform.io/providers/rancher/rancher2/3.1.1

As a workaround, I tried adding this particular parameter to the ignore_changes lifecycle section like this:

resource "rancher2_cluster_v2" "cluster" {
  name = var.cluster_name
  kubernetes_version = var.kubernetes_version
  lifecycle {
    ignore_changes = [
      rke_config[0].machine_selector_config,
   ]
  }

When I do this, I no longer see that Terraform wants to change this parameter in the Terraform plan output. However, after running an apply operation, the machineSelectorConfig is still changed to null, which then causes the UI issue again when attempting to view the config. It appears as if any kind of change to the RKE config on an RKE2 cluster resource via the rancher2 Terraform provider will result in this bug.