feat(TPG>=5.40.0)!: Add support for RayOperator Addon (#2032)

Co-authored-by: Gen Lu <[email protected]>
Co-authored-by: Andrew Peabody <[email protected]>

Author: genlu2011

README.md

@@ -228,6 +228,7 @@ Then perform the following commands on the root folder:
 | notification\_config\_topic | The desired Pub/Sub topic to which notifications will be sent by GKE. Format is projects/{project}/topics/{topic}. | `string` | `""` | no |
 | notification\_filter\_event\_type | Choose what type of notifications you want to receive. If no filters are applied, you'll receive all notification types. Can be used to filter what notifications are sent. Accepted values are UPGRADE\_AVAILABLE\_EVENT, UPGRADE\_EVENT, and SECURITY\_BULLETIN\_EVENT. | `list(string)` | `[]` | no |
 | project\_id | The project ID to host the cluster in (required) | `string` | n/a | yes |
+| ray\_operator\_config | The Ray Operator Addon configuration for this cluster. | <pre>object({<br>    enabled            = bool<br>    logging_enabled    = optional(bool, false)<br>    monitoring_enabled = optional(bool, false)<br>  })</pre> | <pre>{<br>  "enabled": false,<br>  "logging_enabled": false,<br>  "monitoring_enabled": false<br>}</pre> | no |
 | region | The region to host the cluster in (optional if zonal cluster / required if regional) | `string` | `null` | no |
 | regional | Whether is a regional cluster (zonal cluster if set false. WARNING: changing this after cluster creation is destructive!) | `bool` | `true` | no |
 | registry\_project\_ids | Projects holding Google Container Registries. If empty, we use the cluster project. If a service account is created and the `grant_registry_access` variable is set to `true`, the `storage.objectViewer` and `artifactregsitry.reader` roles are assigned on these projects. | `list(string)` | `[]` | no |

autogen/main/cluster.tf.tmpl

@@ -361,6 +361,23 @@ resource "google_container_cluster" "primary" {
         enabled = stateful_ha_config.value.enabled
       }
     }
+
+    dynamic "ray_operator_config" {
+      for_each = local.ray_operator_config
+    
+      content {
+        
+        enabled = ray_operator_config.value.enabled
+
+        ray_cluster_logging_config {
+          enabled = ray_operator_config.value.logging_enabled
+        }
+        ray_cluster_monitoring_config {
+          enabled = ray_operator_config.value.monitoring_enabled
+        }
+      }
+    }
+
     {% if beta_cluster and autopilot_cluster != true %}
     istio_config {
       disabled = !var.istio

autogen/main/main.tf.tmpl

@@ -111,6 +111,7 @@ locals {
   gke_backup_agent_config      = var.gke_backup_agent_config ? [{ enabled = true }] : [{ enabled = false }]
   gcs_fuse_csi_driver_config   = var.gcs_fuse_csi_driver ? [{ enabled = true }] : []
   stateful_ha_config           = var.stateful_ha ? [{ enabled = true }] : []
+  ray_operator_config          =  length(var.ray_operator_config) > 0 && lookup(var.ray_operator_config, "enabled", false)? [var.ray_operator_config]: []
 {% if beta_cluster and autopilot_cluster != true %}
   cluster_cloudrun_config_load_balancer_config = (var.cloudrun && var.cloudrun_load_balancer_type != "") ? {
     load_balancer_type = var.cloudrun_load_balancer_type

autogen/main/variables.tf.tmpl

@@ -814,6 +814,20 @@ variable "stateful_ha" {
   default     = false
 }
 
+variable "ray_operator_config" {
+  type        = object({
+    enabled = bool
+    logging_enabled = optional(bool, false)
+    monitoring_enabled = optional(bool, false) 
+  })
+  description = "The Ray Operator Addon configuration for this cluster."
+  default     = {
+    enabled = false
+    logging_enabled = false
+    monitoring_enabled = false 
+  }
+}
+
 variable "timeouts" {
   type        = map(string)
   description = "Timeout for cluster operations."

autogen/main/versions.tf.tmpl

@@ -24,11 +24,11 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = ">= 5.33.0, < 6"
+      version = ">= 5.40.0, < 6"
     }
     google-beta = {
       source  = "hashicorp/google-beta"
-      version = ">= 5.33.0, < 6"
+      version = ">= 5.40.0, < 6"
     }
     kubernetes = {
       source  = "hashicorp/kubernetes"
@@ -46,7 +46,7 @@ terraform {
   required_providers {
     google = {
       source  = "hashicorp/google"
-      version = ">= 5.25.0, < 6"
+      version = ">= 5.40.0, < 6"
     }
     kubernetes = {
       source  = "hashicorp/kubernetes"

cluster.tf

@@ -280,6 +280,23 @@ resource "google_container_cluster" "primary" {
         enabled = stateful_ha_config.value.enabled
       }
     }
+
+    dynamic "ray_operator_config" {
+      for_each = local.ray_operator_config
+
+      content {
+
+        enabled = ray_operator_config.value.enabled
+
+        ray_cluster_logging_config {
+          enabled = ray_operator_config.value.logging_enabled
+        }
+        ray_cluster_monitoring_config {
+          enabled = ray_operator_config.value.monitoring_enabled
+        }
+      }
+    }
+
   }
 
   datapath_provider = var.datapath_provider

examples/simple_autopilot_public/main.tf

@@ -44,12 +44,17 @@ module "gke" {
   subnetwork                      = local.subnet_names[index(module.gcp-network.subnets_names, local.subnet_name)]
   ip_range_pods                   = local.pods_range_name
   ip_range_services               = local.svc_range_name
-  release_channel                 = "REGULAR"
+  release_channel                 = "RAPID"
   enable_vertical_pod_autoscaling = true
   network_tags                    = [local.cluster_type]
   deletion_protection             = false
   enable_l4_ilb_subsetting        = true
   gcs_fuse_csi_driver             = true
   stateful_ha                     = false
   gke_backup_agent_config         = false
+  ray_operator_config = {
+    enabled            = true
+    logging_enabled    = true
+    monitoring_enabled = true
+  }
 }

main.tf

@@ -95,6 +95,7 @@ locals {
   gke_backup_agent_config    = var.gke_backup_agent_config ? [{ enabled = true }] : [{ enabled = false }]
   gcs_fuse_csi_driver_config = var.gcs_fuse_csi_driver ? [{ enabled = true }] : []
   stateful_ha_config         = var.stateful_ha ? [{ enabled = true }] : []
+  ray_operator_config        = length(var.ray_operator_config) > 0 && lookup(var.ray_operator_config, "enabled", false) ? [var.ray_operator_config] : []
 
   cluster_authenticator_security_group = var.authenticator_security_group == null ? [] : [{
     security_group = var.authenticator_security_group

modules/beta-autopilot-private-cluster/README.md

@@ -135,6 +135,7 @@ Then perform the following commands on the root folder:
 | notification\_filter\_event\_type | Choose what type of notifications you want to receive. If no filters are applied, you'll receive all notification types. Can be used to filter what notifications are sent. Accepted values are UPGRADE\_AVAILABLE\_EVENT, UPGRADE\_EVENT, and SECURITY\_BULLETIN\_EVENT. | `list(string)` | `[]` | no |
 | private\_endpoint\_subnetwork | The subnetwork to use for the hosted master network. | `string` | `null` | no |
 | project\_id | The project ID to host the cluster in (required) | `string` | n/a | yes |
+| ray\_operator\_config | The Ray Operator Addon configuration for this cluster. | <pre>object({<br>    enabled            = bool<br>    logging_enabled    = optional(bool, false)<br>    monitoring_enabled = optional(bool, false)<br>  })</pre> | <pre>{<br>  "enabled": false,<br>  "logging_enabled": false,<br>  "monitoring_enabled": false<br>}</pre> | no |
 | region | The region to host the cluster in (optional if zonal cluster / required if regional) | `string` | `null` | no |
 | regional | Whether is a regional cluster (zonal cluster if set false. WARNING: changing this after cluster creation is destructive!) | `bool` | `true` | no |
 | registry\_project\_ids | Projects holding Google Container Registries. If empty, we use the cluster project. If a service account is created and the `grant_registry_access` variable is set to `true`, the `storage.objectViewer` and `artifactregsitry.reader` roles are assigned on these projects. | `list(string)` | `[]` | no |

modules/beta-autopilot-private-cluster/cluster.tf

@@ -173,6 +173,23 @@ resource "google_container_cluster" "primary" {
         enabled = stateful_ha_config.value.enabled
       }
     }
+
+    dynamic "ray_operator_config" {
+      for_each = local.ray_operator_config
+
+      content {
+
+        enabled = ray_operator_config.value.enabled
+
+        ray_cluster_logging_config {
+          enabled = ray_operator_config.value.logging_enabled
+        }
+        ray_cluster_monitoring_config {
+          enabled = ray_operator_config.value.monitoring_enabled
+        }
+      }
+    }
+
   }
 
   allow_net_admin = var.allow_net_admin