modules-gcp-tf-gke (WiP)

GKE cluster with GPU Nodes

resource "google_container_cluster" "container_cluster" {
  provider                 = google-beta
  name                     = var.cluster_name
  location                 = var.region
  network                  = var.vpc_network_name
  subnetwork               = var.vpc_network_subnet_name
  remove_default_node_pool = true
  initial_node_count       = var.initial_node_count

  node_locations = [
    "europe-west4-a"
//    "europe-west4-b",
//    "europe-west4-c"
  ]
}

resource "google_container_node_pool" "primary_preemptible_nodes" {
  name       = "my-node-pool"
  location   = var.region
  cluster    = google_container_cluster.container_cluster.name
  node_count = 1

  node_config {
    spot         = true
    image_type   = "ubuntu_containerd"
    machine_type = var.machine_type
    disk_type    = var.disk_type
    disk_size_gb = 32
    service_account = var.node_service_account
    oauth_scopes    = [
      "https://www.googleapis.com/auth/cloud-platform"
    ]
    guest_accelerator {
      type  = "nvidia-tesla-t4"
      count = 1
      }
    }

  autoscaling {
    total_min_node_count = 1
    total_max_node_count = 1
    location_policy = var.autoscaling_strategy
  }
}

output "endpoint" {
  value = google_container_cluster.container_cluster.endpoint
}

Connecting to your cluster

• get your kubeconfig from gcloud

  gcloud container clusters get-credentials <cluster-name> --zone=<your zone>

Requirements

Name	Version
helm	2.9.0
kubernetes	2.20.0

Providers

Name	Version
google	n/a
google-beta	n/a
helm	2.9.0

Modules

No modules.

Resources

Name	Type
google-beta_google_container_cluster.container_cluster	resource
google_container_node_pool.primary_preemptible_nodes	resource
helm_release.nginx_ingress	resource
google_client_config.client_config	data source
google_client_config.provider	data source
google_compute_network.network	data source
google_compute_subnetwork.network-with-private-secondary-ip-ranges	data source

Inputs

Name	Description	Type	Default	Required
autoscaling_enabled	set autoscaling true or false	bool	false	no
autoscaling_max_nodes	max number of nodes allowed	number	1	no
autoscaling_min_nodes	min number of nodes allocation	number	1	no
autoscaling_strategy	GKE autoscaling strategy	string	n/a	yes
cluster_name	Name of the GKE cluster we will create	string	n/a	yes
disk_size	Default size of the node Disk	string	n/a	yes
disk_type	'pd-standard', 'pd-balanced' or 'pd-ssd'	string	n/a	yes
guest_accelerator	GPU or TPU to attach to the virtual-machine.	string	n/a	yes
guest_accelerator_count	Number of accelerators to attach to each machine	number	n/a	yes
initial_node_count	Number of nodes the GKE cluster starts with	number	1	no
machine_type	The virtual amchine type to use for the node pool	string	n/a	yes
main_availability_zone	the gcp region we do all of this in	string	n/a	yes
node_service_account	The SA we will use to control nodes on the GKE cluster	string	n/a	yes
project_id	GCP project id (gcloud projects list)	string	n/a	yes
region	the gcp region we do all of this in	string	n/a	yes
state_bucket_name	the name of the bucker we are going to store our state in	string	n/a	yes
state_path	directory where we store state	string	n/a	yes
use_default_node_pool	True=use the deafult GKE node pool, Fale=use seprately managed pool	bool	n/a	yes
vpc_network_name	name of the VPC network our machines will live in	string	n/a	yes
vpc_network_subnet_name	name of the VPC subnetwork our machines will live in	string	n/a	yes

Outputs

Name	Description
endpoint	n/a