upgrading_to_v13.0.md

Upgrading to v13.0

The v13.0 release of kubernetes-engine is a backwards incompatible release.

kubernetes provider removed from the module

• kubernetes provider has been removed across all modules/submodules and need to be specified in the calling module.

To leverage Terraform v0.13 features such as custom variable validation and using count, for_each or depends_on in modules, it is required that a module does not contain any nested provider configuration and receives all of its provider configurations from the calling module. This release adapts to this requirement.

+  data "google_client_config" "default" {}

+  provider "kubernetes" {
+    load_config_file       = false
+    host                   = "https://${module.gke.endpoint}"
+    token                  = data.google_client_config.default.access_token
+    cluster_ca_certificate = base64decode(module.gke.ca_certificate)
+  }

 module "gke" {
   source                  = "terraform-google-modules/kubernetes-engine/google"
-  version                 = "~> 12.0"
+  version                 = "~> 13.0"
}

ACM submodule local_file removed

ACM submodule no longer creates a local file called operator_cr.yaml. The yaml contents are rendered dynamically and passed via STDIN which fixes errors due to operator_cr.yaml file not being present between ephemeral pipeline runs.

This is destructive and will result in deletion and recreation of the ACM operator.

Wait for cluster script removed

wait-for-cluster.sh used to ensure that the cluster is in a ready state has been removed due to improvements in the provider. As part of the upgrade, Terraform might indicate that the wait_for_cluster null_resource must be destroyed. This is no-op and can be safely applied:

  # module.example.module.gke.module.gcloud_wait_for_cluster.null_resource.module_depends_on[0] will be destroyed
  - resource "null_resource" "module_depends_on" {
      - id       = "8092231570921454387" -> null
      - triggers = {
          - "value" = "2"
        } -> null
    }

  # module.example.module.gke.module.gcloud_wait_for_cluster.null_resource.run_command[0] will be destroyed
  - resource "null_resource" "run_command" {
      - id       = "5198854261759708357" -> null
      - triggers = {
          - "arguments"             = "de7f9e59f7436cdec4c8fc34fb8e9833"
          - "create_cmd_body"       = "PROJECT_ID CLUSTER_NAME CLUSTER_LOCATION "
          - "create_cmd_entrypoint" = "../../../scripts/wait-for-cluster.sh"
          - "gcloud_bin_abs_path"   = "/google-cloud-sdk/bin"
          - "md5"                   = "4149df8255e42ea6781ec241e82e9e69"
        } -> null
    }

  # module.example.module.gke.module.gcloud_wait_for_cluster.null_resource.run_destroy_command[0] will be destroyed
  - resource "null_resource" "run_destroy_command" {
      - id       = "6787683359757867074" -> null
      - triggers = {
          - "destroy_cmd_body"       = "PROJECT_ID CLUSTER_NAME CLUSTER_LOCATION "
          - "destroy_cmd_entrypoint" = "../../../scripts/wait-for-cluster.sh"
          - "gcloud_bin_abs_path"    = "/google-cloud-sdk/bin"
        } -> null
    }

Plan: 0 to add, 0 to change, 3 to destroy.

Node Pool Random ID Keepers Modified

Note: This change only applies to the update variant submodules.

As reported in issue #842, the v13.0 release has added the node pool taints to the keepers for the random_id resource used in the naming of node pools.

This addition forces a replacement of the random_id resource, and therefore the node pools themselves.

To avoid this, it is possible to edit the remote state of the random_id resource to add the "missing" keeper attribute.

1. Perform a terraform plan as normal, identifying the random_id resources changing and the new "taints" attribute
2. Pull the remote state locally: terraform state pull > default.tfstate
3. Back up the original remote state: cp default.tfstate original.tfstate
4. Edit the random_id resources to add in the new "taints" attributes from the terraform plan step
5. Bump the serial number at the top
6. Push the modified state to the remote terraform state push default.tfstate
7. Confirm the random_id resource no longer changes (or the corresponding nodepool) in a terraform plan