Update labels on multiple kube nodepools is unpredictable

MircoT commented 1 year ago

Hi,

I found a problem updating the labels of a Kubernetes pool. It seems that if you have more than one pool, the result state in not predictable and only a partial change is applied.

Terraform Version

Terraform v1.3.6

Affected Resource(s)

project_kube_nodepool

Terraform Configuration Files

terraform {
  required_version = ">= 1.3.6"

  required_providers {
    openstack = {
      source  = "terraform-provider-openstack/openstack"
      version = ">= 1.49.0"
    }

    ovh = {
      source  = "ovh/ovh"
      version = ">= 0.24.0"
    }
  }
}

provider "openstack" {
  auth_url    = "https://auth.cloud.ovh.net/v3/"
  domain_name = "default"
  alias       = "ovh"
}

provider "ovh" {
  alias              = "ovh"
  endpoint           = "ovh-eu"
  application_key    = ""
  application_secret = ""
  consumer_key       = ""
}

resource "ovh_cloud_project_kube" "k8s_cluster" {
  name               = "k8s-cluster"
  region             = "GRA9"
  version            = "1.24"
}

resource "ovh_cloud_project_kube_nodepool" "k8s_pool_a" {
  kube_id       = ovh_cloud_project_kube.k8s_cluster.id
  name          = "k8s-pool-a"
  flavor_name   = "d2-4"
  desired_nodes = 1
  template {
    metadata {
      annotations = {
        "label_a" = "disabled"
        "label_b" = "enabled"
      }
      labels = {
        "label_a" = "disabled"
        "label_b" = "enabled"
      }
    }
  }
}

resource "ovh_cloud_project_kube_nodepool" "k8s_pool_b" {
  kube_id       = ovh_cloud_project_kube.k8s_cluster.id
  name          = "k8s-pool-b"
  flavor_name   = "d2-4"
  desired_nodes = 1
  template {
    metadata {
      annotations = {
        "label_a" = "enabled"
        "label_b" = "enabled"
      }
      labels = {
        "label_a" = "enabled"
        "label_b" = "enabled"
      }
    }
  }
}

Debug Output

There is no useful debug output.

Panic Output

There is no useful panic output.

Expected Behavior

If I change the labels on both pools, the resulting apply of the plan have to be done correctly and the pools have to have the new labels.

Actual Behavior

One of the pool cold not have the new state and, consequently, the labels are not updated.

Steps to Reproduce

terraform apply
Change the labels for both pools
terraform apply

References

I suppose that this issue is relative to the type of metadata section, that is a TypeSet. Also, the type of labels is a map. A similar problem during an update of those types can be found here:

GH-281: that was fixed in PR-286 using a proper hashing function to locate the all changes correctly.

jon4hz commented 1 year ago

Had the same issue a few days ago, but only with one nodepool. I tried to add a label, terraform returned success but in fact the label was never added.