Add default node pool to GKE cluster

tools/kubernetes/terraform/examples/v5e/main.tf

@@ -2,6 +2,7 @@ variable "project_id" {}
 variable "resource_name_prefix" {}
 variable "region" {}
 variable "tpu_node_pools" {}
+variable "cpu_node_pool" {}
 variable "maintenance_interval" {}
 
 
@@ -11,5 +12,6 @@ module "tpu-gke" {
   resource_name_prefix = var.resource_name_prefix
   region               = var.region
   tpu_node_pools       = var.tpu_node_pools
+  cpu_node_pool        = var.cpu_node_pool
   maintenance_interval = var.maintenance_interval
 }

tools/kubernetes/terraform/examples/v5e/terraform.tfvars

@@ -98,4 +98,11 @@ tpu_node_pools = [{
   topology     = "16x16"
   policy       = "sb-compact-4d"
 }]
+cpu_node_pool = {
+  zone = ["us-east5-a", "us-east5-b", "us-east5-c"]
+  machine_type = "n2-standard-64",
+  initial_node_count_per_zone = 1,
+  min_node_count_per_zone = 1,
+  max_node_count_per_zone = 10
+}
 maintenance_interval = "PERIODIC"

tools/kubernetes/terraform/module/main.tf

@@ -117,3 +117,30 @@ resource "google_container_node_pool" "multihost_tpu" {
     policy_name  = var.tpu_node_pools[count.index].policy
   }
 }
+
+resource "google_container_node_pool" "cpu_node_pool" {
+  provider           = google-beta
+  project            = var.project_id
+  name               = "cpu-node-pool"
+  location           = var.region
+  node_locations     = var.cpu_node_pool.zone
+  cluster            = google_container_cluster.tpu_cluster.name
+  initial_node_count = var.cpu_node_pool.initial_node_count_per_zone
+  autoscaling {
+    min_node_count = var.cpu_node_pool.min_node_count_per_zone
+    max_node_count = var.cpu_node_pool.max_node_count_per_zone
+  }
+  node_config {
+    oauth_scopes = [
+      "https://www.googleapis.com/auth/cloud-platform"
+    ]
+    machine_type = var.cpu_node_pool.machine_type
+
+    metadata = {
+      disable-legacy-endpoints = "true"
+    }
+    gcfs_config {
+      enabled = true
+    }
+  }
+}

tools/kubernetes/terraform/module/terraform.tfvars

@@ -18,4 +18,11 @@ tpu_node_pools = [{
   machine_type = "ct4p-hightpu-4t"
   topology     = "2x2x2"
 }]
+cpu_node_pool = {
+  zone = ["us-central2-a", "us-central2-b", "us-central2-c"]
+  machine_type = "n2-standard-64",
+  initial_node_count_per_zone = 1,
+  min_node_count_per_zone = 1,
+  max_node_count_per_zone = 10
+}
 maintenance_interval = "AS_NEEDED"

tools/kubernetes/terraform/module/variables.tf

@@ -38,6 +38,23 @@ variable "tpu_node_pools" {
   }))
 }
 
+variable "cpu_node_pool" {
+  description = "cpu nodepool config"
+  type = object({
+    zone         = list(string),
+    machine_type = string,
+    initial_node_count_per_zone = number,
+    min_node_count_per_zone = number,
+    max_node_count_per_zone = number
+  })
+  validation {
+    condition = (
+      (var.cpu_node_pool.min_node_count_per_zone >=0 && var.cpu_node_pool.min_node_count_per_zone <= var.cpu_node_pool.max_node_count_per_zone)
+    )
+    error_message = "cpu_node_pool.min_node_count_per_zone must be >= 0 and <= cpu_node_pool.max_node_count_per_zone."
+  }
+}
+
 variable "maintenance_interval" {
   default = "AS_NEEDED"
   description = "maintenance interval for TPU machines."