Good morning everyone, I'm seeing some unexpected behavior that I'm hoping the dev team can verify: 1. cluster_v2 machine pools has a

labels

field that should be for node labels but for some reason the provider logic transforms a map of

labels

into

machineDeploymentLabels

. I think think this might be the function that's responsible: https://github.com/rancher/terraform-provider-rancher2/blob/master/rancher2/structure_cluster_v2_rke_config_machine_pool.go#L66 2. I'm trying to pass in a map of

kubelet-arg

values into cluster_v2 machine_selector_configs but it's also undocumented in the provider docs. I'm forced to pass in an HCL map to the

config

key but the

kubelet-arg

subkey should be an array while the provider expects only expects a string. I've tried using properly formatted HCL and even hacking it with multiline strings or formatted single-line yaml arrays but no luck so far. 3. The two last issues make me wonder why I'm passing in rke_config values as HCL maps in the first place when Rancher/RKE2 expects them as YAML? I think we can maintain closer parity between the Rancher/RKE2 configs and the Terraform plans if we just pass in those configs as YAML while also making the codebase more maintainable. Thoughts?

(1) cluster_v2 machine pools has a

labels

field that should be for node labels but for some reason the provider logic transforms a map of

labels

into

machineDeploymentLabels

I noticed (1) as well. There is an issue for it on github as well: • https://github.com/rancher/terraform-provider-rancher2/issues/949 And I created a PR to "fix" it. But wasn't quite sure what the intended behaviour should be.

@polite-honey-55746 from RGS relayed this solution from the provider dev team:

(2) I'm trying to pass in a map of

kubelet-arg

values into cluster_v2 machine_selector_configs but it's also undocumented in the provider docs.

Instead of passing the values in to

machine_selector_config

, we instead passed it in to

machine_global_config

so the relevant block on the terraform plan looks like this when I try to set a static CPU policy on the cluster

machine_global_config = <<-EOF
kubelet-arg:
- cpu-manager-policy=static
- kube-reserved=cpu=500m,memory=1Gi,ephemeral-storage=1Gi
- system-reserved=cpu=300m,memory=500Mi,ephemeral-storage=1Gi
- eviction-hard=memory.available<500Mi,nodefs.available<10%
EOF

Verified kubelet started with the correct args and I was able to create

guaranteed

workloads with this method.