feat(k8s): add DO (Digital Ocean) Kubernetes cluster terraform

Adds Terraform configuration to provision a DO K8s cluster with
auto-scaling node pools for running Codex benchmarks.

Signed-off-by: Chrysostomos Nanakos <chris@include.gr>

k8s/terraform/codex-benchmarks-do/backend.tf

@@ -0,0 +1,19 @@
+terraform {
+  backend "s3" {
+    endpoints = {
+      s3 = "https://<S3_REGION>.digitaloceanspaces.com"
+    }
+    bucket = "codex-infra-terraform"
+    key    = "clusters/codex-benchmarks-do-<DO_REGION>/terraform.tfstate"
+    region = "<S3_REGION>"
+
+    access_key = "<S3_ACCESS_KEY>"
+    secret_key = "<S3_SECRET_KEY>"
+
+    skip_credentials_validation = true
+    skip_requesting_account_id  = true
+    skip_metadata_api_check     = true
+    skip_region_validation      = true
+    skip_s3_checksum            = true
+  }
+}

k8s/terraform/codex-benchmarks-do/main.tf

@@ -0,0 +1,40 @@
+# Kubernetes cluster
+module "doks" {
+  source = "../modules/doks"
+
+  name                            = "codex-benchmarks"
+  region                          = var.region
+  vpc_ip_range                    = "10.1.0.0/20"
+  kubernetes_version              = "1.33.1-do.5"
+  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        = 4
+  kubernetes_node_pool_tags       = ["default", "autoscale"]
+  kubernetes_node_pool_labels = {
+    default-pool  = "true"
+    scaling-type  = "auto"
+    workload-type = "infra"
+  }
+}
+
+# Node pool - Codex
+resource "digitalocean_kubernetes_node_pool" "codex-d-4vcpu-8gb" {
+  cluster_id = module.doks.kubernetes_cluster_id
+  name       = "codex-d-4vcpu-8gb"
+  size       = "c-4"
+  auto_scale = true
+  min_nodes  = 1
+  max_nodes  = 95
+  node_count = 4
+  tags       = ["codex"]
+
+  labels = {
+    default-pool     = "false"
+    scaling-type     = "auto"
+    workload-type    = "benchmarks"
+  }
+}

k8s/terraform/codex-benchmarks-do/providers.tf

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

k8s/terraform/codex-benchmarks-do/terraform.tfvars

@@ -0,0 +1,2 @@
+region            = "<DO_REGION>"
+do_token          = "<DOP_TOKEN>"

k8s/terraform/codex-benchmarks-do/variables.tf

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

k8s/terraform/codex-benchmarks-do/versions.tf

@@ -0,0 +1,10 @@
+# Terraform settings
+terraform {
+  required_version = "~> 1.0"
+  required_providers {
+    digitalocean = {
+      source  = "digitalocean/digitalocean"
+      version = "~> 2.0"
+    }
+  }
+}

k8s/terraform/modules/doks/locals.tf

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

k8s/terraform/modules/doks/main.tf

@@ -0,0 +1,41 @@
+# VPC
+resource "digitalocean_vpc" "this" {
+  name     = local.name
+  region   = var.region
+  ip_range = var.vpc_ip_range
+}
+
+# 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
+  vpc_uuid     = digitalocean_vpc.this.id
+
+  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
+  }
+}

k8s/terraform/modules/doks/outputs.tf

@@ -0,0 +1,5 @@
+# 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."
+}

k8s/terraform/modules/doks/variables.tf

@@ -0,0 +1,103 @@
+# 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."
+}
+
+# VPC
+variable "vpc_ip_range" {
+  type        = string
+  description = " The range of IP addresses for the VPC in CIDR notation."
+}
+
+# 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 = {
+  }
+}

k8s/terraform/modules/doks/versions.tf

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