Port terraform cluster creation/destruction from digital ocean to gcp

.github/release/clusters/logos-storage-dist-tests-do-ams3/backend.tf

@@ -1,16 +0,0 @@
-terraform {
-  backend "s3" {
-    endpoints = {
-      s3 = "https://fra1.digitaloceanspaces.com"
-    }
-    bucket = "codex-infra-terraform"
-    key    = "clusters/logos-storage-dist-tests-do-ams3/terraform.tfstate"
-    region = "fra1"
-
-    skip_credentials_validation = true
-    skip_requesting_account_id  = true
-    skip_metadata_api_check     = true
-    skip_region_validation      = true
-    skip_s3_checksum            = true
-  }
-}

.github/release/clusters/logos-storage-dist-tests-do-ams3/main.tf

@@ -1,57 +0,0 @@
-# Kubernetes cluster
-module "doks" {
-  source = "../modules/doks"
-
-  name                            = "logos-storage-dist-tests"
-  region                          = var.region
-  kubernetes_version              = "1.34.5-do.2"
-  kubernetes_ha                   = true
-  kubernetes_auto_upgrade         = false
-  kubernetes_node_pool_name       = "infra-s-4vcpu-16gb-amd"
-  kubernetes_node_pool_size       = "s-4vcpu-16gb-amd"
-  kubernetes_node_pool_auto_scale = true
-  kubernetes_node_pool_min        = 1
-  kubernetes_node_pool_max        = 3
-  kubernetes_node_pool_tags       = ["default", "autoscale"]
-  kubernetes_node_pool_labels = {
-    default-pool  = "true"
-    scaling-type  = "auto"
-    workload-type = "infra"
-  }
-}
-
-# Node pool - Runners CI
-resource "digitalocean_kubernetes_node_pool" "runners-ci" {
-  cluster_id = module.doks.kubernetes_cluster_id
-  name       = "runners-ci-s-2vcpu-8gb-amd"
-  size       = "s-2vcpu-8gb-amd"
-  auto_scale = true
-  min_nodes  = 1
-  max_nodes  = 5
-  tags       = ["runners-ci"]
-
-  labels = {
-    allow-tests-pods = "false"
-    default-pool     = "false"
-    scaling-type     = "auto"
-    workload-type    = "tests-runners-ci"
-  }
-}
-
-# Node pool - Tests Pods
-resource "digitalocean_kubernetes_node_pool" "tests-s-2vcpu-4gb" {
-  cluster_id = module.doks.kubernetes_cluster_id
-  name       = "tests-s-2vcpu-4gb"
-  size       = "s-2vcpu-4gb"
-  auto_scale = true
-  min_nodes  = 1
-  max_nodes  = 10
-  tags       = ["tests-pods"]
-
-  labels = {
-    allow-tests-pods = "true"
-    default-pool     = "false"
-    scaling-type     = "auto"
-    workload-type    = "tests-pods"
-  }
-}

.github/release/clusters/logos-storage-dist-tests-do-ams3/providers.tf

@@ -1,4 +0,0 @@
-# Providers
-provider "digitalocean" {
-  token = var.do_token
-}

.github/release/clusters/logos-storage-dist-tests-do-ams3/variables.tf

@@ -1,10 +0,0 @@
-variable "region" {
-  description = "DigitalOcean region (e.g. ams3)"
-  type        = string
-}
-
-variable "do_token" {
-  description = "DigitalOcean API token"
-  type        = string
-  sensitive   = true
-}

.github/release/clusters/logos-storage-dist-tests-gcp-europe-west4/backend.tf

@@ -0,0 +1,7 @@
+terraform {
+  backend "gcs" {
+    prefix = "clusters/logos-storage-dist-tests-gcp-europe-west4"
+    # bucket is supplied at init time via:
+    #   terraform init -backend-config="bucket=<bucket-name>"
+  }
+}

.github/release/clusters/logos-storage-dist-tests-gcp-europe-west4/main.tf

@@ -0,0 +1,70 @@
+# Kubernetes cluster
+module "gke" {
+  source = "../modules/gke"
+
+  name                       = "logos-storage-dist-tests"
+  project                    = var.project
+  region                     = var.region
+  kubernetes_release_channel = "STABLE"
+  node_pool_name             = "infra-e2-standard-4"
+  node_pool_machine_type     = "e2-standard-4"
+  node_pool_min              = 1
+  node_pool_max              = 3
+  node_pool_labels = {
+    default-pool  = "true"
+    scaling-type  = "auto"
+    workload-type = "infra"
+  }
+}
+
+# Node pool - Runners CI
+resource "google_container_node_pool" "runners-ci" {
+  name     = "runners-ci-e2-standard-2"
+  cluster  = module.gke.kubernetes_cluster_id
+  location = var.region
+  project  = var.project
+
+  autoscaling {
+    min_node_count = 1
+    max_node_count = 5
+  }
+
+  node_config {
+    machine_type = "e2-standard-2"
+    labels = {
+      allow-tests-pods = "false"
+      default-pool     = "false"
+      scaling-type     = "auto"
+      workload-type    = "tests-runners-ci"
+    }
+    oauth_scopes = [
+      "https://www.googleapis.com/auth/cloud-platform",
+    ]
+  }
+}
+
+# Node pool - Tests Pods
+resource "google_container_node_pool" "tests-pods" {
+  name     = "tests-e2-medium"
+  cluster  = module.gke.kubernetes_cluster_id
+  location = var.region
+  project  = var.project
+
+  autoscaling {
+    min_node_count = 1
+    max_node_count = 10
+  }
+
+  node_config {
+    machine_type = "e2-medium"
+    labels = {
+      allow-tests-pods = "true"
+      default-pool     = "false"
+      scaling-type     = "auto"
+      workload-type    = "tests-pods"
+    }
+    oauth_scopes = [
+      "https://www.googleapis.com/auth/cloud-platform",
+    ]
+  }
+}

.github/release/clusters/logos-storage-dist-tests-gcp-europe-west4/providers.tf

@@ -0,0 +1,5 @@
+# Providers
+provider "google" {
+  project = var.project
+  region  = var.region
+}

.github/release/clusters/logos-storage-dist-tests-gcp-europe-west4/variables.tf

@@ -0,0 +1,9 @@
+variable "project" {
+  description = "GCP project ID"
+  type        = string
+}
+
+variable "region" {
+  description = "GCP region (e.g. europe-west4)"
+  type        = string
+}

.github/release/clusters/logos-storage-dist-tests-gcp-europe-west4/versions.tf

@@ -2,9 +2,9 @@
 terraform {
   required_version = "~> 1.0"
   required_providers {
-    digitalocean = {
-      source  = "digitalocean/digitalocean"
-      version = "~> 2.0"
+    google = {
+      source  = "hashicorp/google"
+      version = "~> 6.0"
     }
   }
 }

.github/release/clusters/modules/doks/locals.tf

@@ -1,4 +0,0 @@
-locals {
-  name           = "${var.name}-do-${var.region}"
-  node_pool_name = "pool-${var.kubernetes_node_pool_size}"
-}

.github/release/clusters/modules/doks/main.tf

@@ -1,33 +0,0 @@
-# Kubernetes cluster
-resource "digitalocean_kubernetes_cluster" "this" {
-  name         = local.name
-  region       = var.region
-  version      = var.kubernetes_version
-  ha           = var.kubernetes_ha
-  auto_upgrade = var.kubernetes_auto_upgrade
-
-  node_pool {
-    name       = var.kubernetes_node_pool_name
-    size       = var.kubernetes_node_pool_size
-    node_count = var.kubernetes_node_pool_count
-    auto_scale = var.kubernetes_node_pool_auto_scale
-    min_nodes  = var.kubernetes_node_pool_min
-    max_nodes  = var.kubernetes_node_pool_max
-    tags       = var.kubernetes_node_pool_tags
-    labels     = var.kubernetes_node_pool_labels
-
-    dynamic "taint" {
-      for_each = length(var.kubernetes_node_pool_taint) == 0 ? {} : { taint = true }
-      content {
-        key    = lookup(var.kubernetes_node_pool_taint, "key")
-        value  = lookup(var.kubernetes_node_pool_taint, "value")
-        effect = lookup(var.kubernetes_node_pool_taint, "effect")
-      }
-    }
-  }
-
-  maintenance_policy {
-    day        = var.kubernetes_maintenance_day
-    start_time = var.kubernetes_maintenance_start_time
-  }
-}

.github/release/clusters/modules/doks/outputs.tf

@@ -1,5 +0,0 @@
-# Kubernetes cluster
-output "kubernetes_cluster_id" {
-  value       = digitalocean_kubernetes_cluster.this.id
-  description = "A unique ID that can be used to identify and reference a Kubernetes cluster."
-}

.github/release/clusters/modules/doks/variables.tf

@@ -1,97 +0,0 @@
-# Main
-variable "name" {
-  type        = string
-  description = "A name for the created resources."
-}
-
-variable "region" {
-  type        = string
-  description = "The DigitalOcean region slug for the resources location."
-}
-
-# Kubernetes Control Plane
-variable "kubernetes_version" {
-  type        = string
-  description = "The slug identifier for the version of Kubernetes used for the cluster."
-}
-
-variable "kubernetes_ha" {
-  type        = bool
-  description = "Enable/disable the high availability control plane for a cluster."
-}
-
-variable "kubernetes_auto_upgrade" {
-  type        = bool
-  description = "A boolean value indicating whether the cluster will be automatically upgraded to new patch releases during its maintenance window."
-}
-
-variable "kubernetes_maintenance_day" {
-  type        = string
-  description = "The day of the maintenance window policy."
-  default     = "sunday"
-}
-
-variable "kubernetes_maintenance_start_time" {
-  type        = string
-  description = "The start time in UTC of the maintenance window policy in 24-hour clock format / HH:MM notation (e.g., 15:00)."
-  default     = "04:00"
-}
-
-# Kubernetes default Node Pool
-variable "kubernetes_node_pool_name" {
-  type        = string
-  description = "A name for the node pool."
-  default     = null
-}
-
-variable "kubernetes_node_pool_size" {
-  type        = string
-  description = "The slug identifier for the type of Droplet to be used as workers in the node pool."
-  default     = null
-}
-
-variable "kubernetes_node_pool_count" {
-  type        = number
-  default     = null
-  description = "The number of Droplet instances in the node pool."
-}
-
-variable "kubernetes_node_pool_auto_scale" {
-  type        = bool
-  description = "Enable auto-scaling of the number of nodes in the node pool within the given min/max range."
-  default     = null
-}
-
-variable "kubernetes_node_pool_min" {
-  type        = number
-  description = "If auto-scaling is enabled, this represents the minimum number of nodes that the node pool can be scaled down to."
-  default     = null
-}
-
-variable "kubernetes_node_pool_max" {
-  type        = number
-  description = "If auto-scaling is enabled, this represents the maximum number of nodes that the node pool can be scaled up to."
-  default     = null
-}
-
-variable "kubernetes_node_pool_tags" {
-  type        = list(any)
-  description = "A list of tag names applied to the node pool."
-  default     = ["default", "autoscale"]
-}
-
-variable "kubernetes_node_pool_labels" {
-  type        = map(string)
-  description = "A map of key/value pairs to apply to nodes in the pool."
-  default = {
-    default-pool = "true"
-    scaling-type = "auto"
-  }
-}
-
-variable "kubernetes_node_pool_taint" {
-  type        = map(string)
-  description = "A block representing a taint applied to all nodes in the pool."
-  default = {
-  }
-}

.github/release/clusters/modules/doks/versions.tf

@@ -1,9 +0,0 @@
-# Terraform settings
-terraform {
-  required_providers {
-    digitalocean = {
-      source  = "digitalocean/digitalocean"
-      version = "~> 2.0"
-    }
-  }
-}

.github/release/clusters/modules/gke/locals.tf

@@ -0,0 +1,3 @@
+locals {
+  name = "${var.name}-gcp-${var.region}"
+}

.github/release/clusters/modules/gke/main.tf

@@ -0,0 +1,47 @@
+# Kubernetes cluster
+resource "google_container_cluster" "this" {
+  name     = local.name
+  location = var.region
+  project  = var.project
+
+  # Create an empty cluster — all node pools are managed as separate resources
+  remove_default_node_pool = true
+  initial_node_count       = 1
+
+  deletion_protection = false
+
+  release_channel {
+    channel = var.kubernetes_release_channel
+  }
+
+  # Enable Workload Identity
+  workload_identity_config {
+    workload_pool = "${var.project}.svc.id.goog"
+  }
+
+  # Send pod stdout/stderr to Cloud Logging automatically
+  logging_service    = "logging.googleapis.com/kubernetes"
+  monitoring_service = "monitoring.googleapis.com/kubernetes"
+}
+
+# Default (infra) node pool
+resource "google_container_node_pool" "default" {
+  name     = var.node_pool_name
+  cluster  = google_container_cluster.this.id
+  location = var.region
+  project  = var.project
+
+  autoscaling {
+    min_node_count = var.node_pool_min
+    max_node_count = var.node_pool_max
+  }
+
+  node_config {
+    machine_type = var.node_pool_machine_type
+    labels       = var.node_pool_labels
+
+    oauth_scopes = [
+      "https://www.googleapis.com/auth/cloud-platform",
+    ]
+  }
+}

.github/release/clusters/modules/gke/outputs.tf

@@ -0,0 +1,10 @@
+# Kubernetes cluster
+output "kubernetes_cluster_id" {
+  value       = google_container_cluster.this.id
+  description = "The fully-qualified ID of the GKE cluster."
+}
+
+output "kubernetes_cluster_name" {
+  value       = google_container_cluster.this.name
+  description = "The name of the GKE cluster."
+}

.github/release/clusters/modules/gke/variables.tf

@@ -0,0 +1,52 @@
+# Main
+variable "name" {
+  type        = string
+  description = "A name for the created resources."
+}
+
+variable "project" {
+  type        = string
+  description = "The GCP project ID."
+}
+
+variable "region" {
+  type        = string
+  description = "The GCP region for the cluster (regional cluster spans 3 zones)."
+}
+
+# Kubernetes Control Plane
+variable "kubernetes_release_channel" {
+  type        = string
+  description = "The GKE release channel: RAPID, REGULAR, or STABLE."
+  default     = "STABLE"
+}
+
+# Kubernetes default Node Pool
+variable "node_pool_name" {
+  type        = string
+  description = "A name for the default node pool."
+}
+
+variable "node_pool_machine_type" {
+  type        = string
+  description = "The GCE machine type for nodes in the default pool."
+}
+
+variable "node_pool_min" {
+  type        = number
+  description = "Minimum number of nodes per zone in the default pool (autoscaling)."
+}
+
+variable "node_pool_max" {
+  type        = number
+  description = "Maximum number of nodes per zone in the default pool (autoscaling)."
+}
+
+variable "node_pool_labels" {
+  type        = map(string)
+  description = "A map of key/value pairs to apply as Kubernetes labels to nodes in the default pool."
+  default = {
+    default-pool = "true"
+    scaling-type = "auto"
+  }
+}

.github/release/clusters/modules/gke/versions.tf

@@ -0,0 +1,9 @@
+# Terraform settings
+terraform {
+  required_providers {
+    google = {
+      source  = "hashicorp/google"
+      version = "~> 6.0"
+    }
+  }
+}

.github/workflows/release.yml

@@ -206,27 +206,29 @@ jobs:
     if: github.ref_type == 'tag' || github.event_name == 'workflow_dispatch'
     needs: build-docker-dist-tests
     env:
-      TF_VAR_region: ams3
-      TF_VAR_do_token: ${{ secrets.RELEASE_TESTS_DO_TOKEN }}
-      AWS_ACCESS_KEY_ID: ${{ secrets.RELEASE_TESTS_SPACES_ACCESS_KEY }}
-      AWS_SECRET_ACCESS_KEY: ${{ secrets.RELEASE_TESTS_SPACES_SECRET_KEY }}
+      TF_VAR_project: ${{ secrets.RELEASE_TESTS_GCP_PROJECT }}
+      TF_VAR_region: europe-west4
       STORAGEDOCKERIMAGE: ${{ needs.build-docker-dist-tests.outputs.logos_storage_image }}
       TEST_TYPE: release-tests
       BRANCH: ${{ inputs.branch || 'master' }}
       SOURCE: https://github.com/logos-storage/logos-storage-nim-cs-dist-tests
-      TF_DIR: .github/release/clusters/logos-storage-dist-tests-do-ams3
+      TF_DIR: .github/release/clusters/logos-storage-dist-tests-gcp-europe-west4
     steps:
       - name: Checkout
         uses: actions/checkout@v4
 
+      - name: Authenticate to GCP
+        uses: google-github-actions/auth@v2
+        with:
+          workload_identity_provider: ${{ secrets.RELEASE_TESTS_GCP_WORKLOAD_IDENTITY_PROVIDER }}
+          service_account: ${{ secrets.RELEASE_TESTS_GCP_SERVICE_ACCOUNT }}
+
+      - name: Setup gcloud
+        uses: google-github-actions/setup-gcloud@v2
+
       - name: Setup Terraform
         uses: hashicorp/setup-terraform@v3
 
-      - name: Setup doctl
-        uses: digitalocean/action-doctl@v2
-        with:
-          token: ${{ secrets.RELEASE_TESTS_DO_TOKEN }}
-
       - name: Setup kubectl
         uses: azure/setup-kubectl@v4
         with:
@@ -234,14 +236,18 @@ jobs:
 
       - name: Terraform init
         working-directory: ${{ env.TF_DIR }}
-        run: terraform init
+        run: terraform init -backend-config="bucket=${{ secrets.RELEASE_TESTS_TF_STATE_BUCKET }}"
 
       - name: Terraform apply
+        id: tf-apply
         working-directory: ${{ env.TF_DIR }}
         run: terraform apply -auto-approve
 
       - name: Get kubeconfig
-        run: doctl kubernetes cluster kubeconfig save logos-storage-dist-tests-do-ams3
+        run: |
+          gcloud container clusters get-credentials logos-storage-dist-tests-gcp-europe-west4 \
+            --region europe-west4 \
+            --project ${{ secrets.RELEASE_TESTS_GCP_PROJECT }}
 
       - name: Create in-cluster app kubeconfig secret
         run: |
@@ -291,7 +297,7 @@ jobs:
           [[ "$job_status" == "SuccessCriteriaMet" ]] || exit 1
 
       - name: Terraform destroy
-        if: always()
+        if: always() && steps.tf-apply.conclusion != 'skipped'
         working-directory: ${{ env.TF_DIR }}
         run: terraform destroy -auto-approve