Merge branch 'main' into bump-tgi

applications/jupyter/main.tf

@@ -70,6 +70,13 @@ module "gcs" {
   bucket_name = var.gcs_bucket
 }
 
+# create namespace
+module "namespace" {
+  source           = "../../modules/kubernetes-namespace"
+  namespace        = var.namespace
+  create_namespace = true
+}
+
 # Creates jupyterhub
 module "jupyterhub" {
   source                            = "../../modules/jupyter"
@@ -94,5 +101,5 @@ module "jupyterhub" {
   url_domain_addr          = var.url_domain_addr
   url_domain_name          = var.url_domain_name
   members_allowlist        = var.members_allowlist
-  depends_on               = [module.gcs]
+  depends_on               = [module.gcs, module.namespace]
 }

applications/rag/frontend/main.tf

@@ -16,20 +16,6 @@ data "google_project" "project" {
 }
 
 
-data "kubernetes_service" "inference_service" {
-  metadata {
-    name      = var.inference_service_name
-    namespace = var.inference_service_namespace
-  }
-}
-
-data "kubernetes_secret" "db_secret" {
-  metadata {
-    name      = var.db_secret_name
-    namespace = var.db_secret_namespace
-  }
-}
-
 locals {
   instance_connection_name = format("%s:%s:%s", var.project_id, var.region, var.cloudsql_instance)
 }
@@ -149,7 +135,7 @@ resource "kubernetes_deployment" "rag_frontend_deployment" {
 
           env {
             name  = "INFERENCE_ENDPOINT"
-            value = data.kubernetes_service.inference_service.status.0.load_balancer.0.ingress.0.ip
+            value = var.inference_service_endpoint
           }
 
           env {
@@ -161,7 +147,7 @@ resource "kubernetes_deployment" "rag_frontend_deployment" {
             name = "DB_USER"
             value_from {
               secret_key_ref {
-                name = data.kubernetes_secret.db_secret.metadata.0.name
+                name = var.db_secret_name
                 key  = "username"
               }
             }
@@ -171,7 +157,7 @@ resource "kubernetes_deployment" "rag_frontend_deployment" {
             name = "DB_PASSWORD"
             value_from {
               secret_key_ref {
-                name = data.kubernetes_secret.db_secret.metadata.0.name
+                name = var.db_secret_name
                 key  = "password"
               }
             }
@@ -181,7 +167,7 @@ resource "kubernetes_deployment" "rag_frontend_deployment" {
             name = "DB_NAME"
             value_from {
               secret_key_ref {
-                name = data.kubernetes_secret.db_secret.metadata.0.name
+                name = var.db_secret_name
                 key  = "database"
               }
             }

applications/rag/frontend/variables.tf

@@ -51,15 +51,9 @@ variable "dataset_embeddings_table_name" {
   description = "Name of the table that stores vector embeddings for input dataset"
 }
 
-variable "inference_service_name" {
-  type        = string
-  description = "Model inference k8s service name"
-}
-
-variable "inference_service_namespace" {
+variable "inference_service_endpoint" {
   type        = string
   description = "Model inference k8s service endpoint"
-  default     = "rag"
 }
 
 variable "create_service_account" {

applications/rag/frontend/versions.tf

@@ -20,5 +20,8 @@ terraform {
     kubernetes = {
       source = "hashicorp/kubernetes"
     }
+    helm = {
+      source = "hashicorp/helm"
+    }
   }
 }

applications/rag/main.tf

@@ -88,6 +88,12 @@ provider "helm" {
   }
 }
 
+module "namespace" {
+  source = "../../modules/kubernetes-namespace"
+  providers              = { helm = helm.rag}
+  create_namespace = true
+  namespace = var.kubernetes_namespace
+}
 
 module "kuberay-operator" {
   source                 = "../../modules/kuberay-operator"
@@ -114,7 +120,7 @@ module "cloudsql" {
   project_id      = var.project_id
   instance_name   = var.cloudsql_instance
   namespace       = var.kubernetes_namespace
-  depends_on      = [module.kuberay-operator]
+  depends_on      = [module.namespace]
 }
 
 module "jupyterhub" {
@@ -142,14 +148,14 @@ module "jupyterhub" {
   url_domain_name          = var.jupyter_url_domain_name
   members_allowlist        = var.jupyter_members_allowlist
 
-  depends_on = [module.kuberay-operator, module.gcs]
+  depends_on = [module.namespace, module.gcs]
 }
 
 module "kuberay-logging" {
   source     = "../../modules/kuberay-logging"
   providers  = { kubernetes = kubernetes.rag }
   namespace  = var.kubernetes_namespace
-  depends_on = [module.kuberay-operator]
+  depends_on = [module.namespace]
 }
 
 module "kuberay-cluster" {
@@ -164,7 +170,7 @@ module "kuberay-cluster" {
   autopilot_cluster      = local.enable_autopilot
   google_service_account = var.ray_service_account
   grafana_host           = module.kuberay-monitoring.grafana_uri
-  depends_on             = [module.kuberay-operator, module.kuberay-monitoring]
+  depends_on             = [module.kuberay-operator]
 }
 
 module "kuberay-monitoring" {
@@ -175,15 +181,14 @@ module "kuberay-monitoring" {
   create_namespace                = true
   enable_grafana_on_ray_dashboard = var.enable_grafana_on_ray_dashboard
   k8s_service_account             = var.ray_service_account
-  depends_on                      = [module.kuberay-operator]
 }
 
 module "inference-server" {
   source            = "../../tutorials/hf-tgi"
   providers         = { kubernetes = kubernetes.rag }
   namespace         = var.kubernetes_namespace
   autopilot_cluster = local.enable_autopilot
-  depends_on        = [module.kuberay-operator]
+  depends_on        = [module.namespace]
 }
 
 module "frontend" {
@@ -193,8 +198,7 @@ module "frontend" {
   create_service_account        = var.create_rag_service_account
   google_service_account        = var.rag_service_account
   namespace                     = var.kubernetes_namespace
-  inference_service_name        = module.inference-server.inference_service_name
-  inference_service_namespace   = module.inference-server.inference_service_namespace
+  inference_service_endpoint    = module.inference-server.inference_service_endpoint
   db_secret_name                = module.cloudsql.db_secret_name
   db_secret_namespace           = module.cloudsql.db_secret_namespace
   dataset_embeddings_table_name = var.dataset_embeddings_table_name
@@ -214,5 +218,5 @@ module "frontend" {
   url_domain_addr          = var.frontend_url_domain_addr
   url_domain_name          = var.frontend_url_domain_name
   members_allowlist        = var.frontend_members_allowlist
-  depends_on = [ module.gcs ]
+  depends_on = [ module.namespace ]
 }

applications/rag/variables.tf

@@ -347,7 +347,7 @@ variable "gpu_pools" {
     {
       name               = "gpu-pool-l4"
       machine_type       = "g2-standard-24"
-      node_locations     = "us-central1-a,us-central1-b"
+      node_locations     = "us-central1-a"
       autoscaling        = true
       min_count          = 1
       max_count          = 3

applications/rag/workloads.tfvars

@@ -24,6 +24,7 @@ kubernetes_namespace = "rag"
 create_gcs_bucket    = true
 gcs_bucket           = "rag-data-xyzu" # Choose a globally unique bucket name.
 
+cloudsql_instance = "pgvector-instance"
 ## Service accounts
 # Creates a google service account & k8s service account & configures workload identity with appropriate permissions.
 # Set to false & update the variable `ray_service_account` to use an existing IAM service account.

infrastructure/main.tf

@@ -57,6 +57,7 @@ module "public-gke-standard-cluster" {
   cluster_name                         = var.cluster_name
   cluster_labels                       = var.cluster_labels
   kubernetes_version                   = var.kubernetes_version
+  release_channel                      = var.release_channel
   cluster_region                       = var.cluster_region
   cluster_zones                        = var.cluster_zones
   ip_range_pods                        = var.ip_range_pods
@@ -93,6 +94,7 @@ module "public-gke-autopilot-cluster" {
   cluster_name               = var.cluster_name
   cluster_labels             = var.cluster_labels
   kubernetes_version         = var.kubernetes_version
+  release_channel            = var.release_channel
   cluster_region             = var.cluster_region
   cluster_zones              = var.cluster_zones
   ip_range_pods              = var.ip_range_pods
@@ -117,6 +119,7 @@ module "private-gke-standard-cluster" {
   cluster_name                         = var.cluster_name
   cluster_labels                       = var.cluster_labels
   kubernetes_version                   = var.kubernetes_version
+  release_channel                      = var.release_channel
   cluster_region                       = var.cluster_region
   cluster_zones                        = var.cluster_zones
   ip_range_pods                        = var.ip_range_pods
@@ -154,6 +157,7 @@ module "private-gke-autopilot-cluster" {
   cluster_name               = var.cluster_name
   cluster_labels             = var.cluster_labels
   kubernetes_version         = var.kubernetes_version
+  release_channel            = var.release_channel
   cluster_region             = var.cluster_region
   cluster_zones              = var.cluster_zones
   ip_range_pods              = var.ip_range_pods

infrastructure/variables.tf

@@ -98,6 +98,11 @@ variable "kubernetes_version" {
   default = "latest"
 }
 
+variable "release_channel" {
+  type    = string
+  default = "RAPID"
+}
+
 variable "cluster_region" {
   type = string
 }

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

@@ -33,6 +33,7 @@
 | <a name="input_network_name"></a> [network\_name](#input\_network\_name) | # network variables | `string` | n/a | yes |
 | <a name="input_project_id"></a> [project\_id](#input\_project\_id) | GCP project id | `string` | n/a | yes |
 | <a name="input_region"></a> [region](#input\_region) | GCP project region or zone | `string` | `"us-central1"` | no |
+| <a name="input_release_channel"></a> [release\_channel](#input\_release\_channel) | n/a | `string` | n/a | yes |
 | <a name="input_subnetwork_name"></a> [subnetwork\_name](#input\_subnetwork\_name) | n/a | `string` | n/a | yes |
 
 ## Outputs

modules/gke-autopilot-private-cluster/main.tf

@@ -25,6 +25,7 @@ module "gke" {
   name                    = var.cluster_name
   cluster_resource_labels = var.cluster_labels
   kubernetes_version      = var.kubernetes_version
+  release_channel         = var.release_channel
   region                  = var.cluster_region
   zones                   = var.cluster_zones
   network                 = var.network_name

modules/gke-autopilot-private-cluster/variables.tf

@@ -46,6 +46,10 @@ variable "kubernetes_version" {
   type = string
 }
 
+variable "release_channel" {
+  type = string
+}
+
 variable "cluster_regional" {
   type = bool
 }

modules/gke-autopilot-public-cluster/README.md

@@ -25,10 +25,11 @@ No resources.
 | <a name="input_network_name"></a> [network\_name](#input\_network\_name) | # network variables | `string` | n/a | yes |
 | <a name="input_project_id"></a> [project\_id](#input\_project\_id) | GCP project id | `string` | n/a | yes |
 | <a name="input_region"></a> [region](#input\_region) | GCP project region or zone | `string` | `"us-central1"` | no |
+| <a name="input_release_channel"></a> [release\_channel](#input\_release\_channel) | n/a | `string` | n/a | yes |
 | <a name="input_subnetwork_name"></a> [subnetwork\_name](#input\_subnetwork\_name) | n/a | `string` | n/a | yes |
 
 ## Outputs
 
 | Name | Description |
 |------|-------------|
-| <a name="output_cluster"></a> [cluster](#output\_cluster) | n/a |
+| <a name="output_cluster"></a> [cluster](#output\_cluster) | n/a |

modules/gke-autopilot-public-cluster/main.tf

@@ -21,12 +21,12 @@ module "gke" {
   cluster_resource_labels    = var.cluster_labels
   region                     = var.cluster_region
   kubernetes_version         = var.kubernetes_version
+  release_channel            = var.release_channel
   zones                      = var.cluster_zones
   network                    = var.network_name
   subnetwork                 = var.subnetwork_name
   ip_range_pods              = var.ip_range_pods
   ip_range_services          = var.ip_range_services
   master_authorized_networks = var.master_authorized_networks
   deletion_protection        = var.deletion_protection
-
 }

modules/gke-autopilot-public-cluster/variables.tf

@@ -50,6 +50,10 @@ variable "kubernetes_version" {
   type = string
 }
 
+variable "release_channel" {
+  type = string
+}
+
 variable "cluster_region" {
   type = string
 }
@@ -74,4 +78,4 @@ variable "master_authorized_networks" {
 variable "deletion_protection" {
   type    = bool
   default = false
-}
+}

modules/gke-standard-private-cluster/README.md

@@ -43,11 +43,12 @@
 | <a name="input_network_name"></a> [network\_name](#input\_network\_name) | # network variables | `string` | n/a | yes |
 | <a name="input_project_id"></a> [project\_id](#input\_project\_id) | GCP project id | `string` | n/a | yes |
 | <a name="input_region"></a> [region](#input\_region) | GCP project region or zone | `string` | `"us-central1"` | no |
+| <a name="input_release_channel"></a> [release\_channel](#input\_release\_channel) | n/a | `string` | n/a | yes |
 | <a name="input_subnetwork_name"></a> [subnetwork\_name](#input\_subnetwork\_name) | n/a | `string` | n/a | yes |
 | <a name="input_tpu_pools"></a> [tpu\_pools](#input\_tpu\_pools) | n/a | `list(map(any))` | n/a | yes |
 
 ## Outputs
 
 | Name | Description |
 |------|-------------|
-| <a name="output_cluster"></a> [cluster](#output\_cluster) | n/a |
+| <a name="output_cluster"></a> [cluster](#output\_cluster) | n/a |

modules/gke-standard-private-cluster/main.tf

@@ -29,6 +29,7 @@ module "gke" {
   name                                 = var.cluster_name
   cluster_resource_labels              = var.cluster_labels
   kubernetes_version                   = var.kubernetes_version
+  release_channel                      = var.release_channel
   region                               = var.cluster_region
   zones                                = var.cluster_zones
   network                              = var.network_name

modules/gke-standard-private-cluster/variables.tf

@@ -46,6 +46,10 @@ variable "kubernetes_version" {
   type = string
 }
 
+variable "release_channel" {
+  type = string
+}
+
 variable "cluster_regional" {
   type = bool
 }

modules/gke-standard-public-cluster/README.md

@@ -35,11 +35,12 @@ No resources.
 | <a name="input_network_name"></a> [network\_name](#input\_network\_name) | # network variables | `string` | n/a | yes |
 | <a name="input_project_id"></a> [project\_id](#input\_project\_id) | GCP project id | `string` | n/a | yes |
 | <a name="input_region"></a> [region](#input\_region) | GCP project region or zone | `string` | `"us-central1"` | no |
+| <a name="input_release_channel"></a> [release\_channel](#input\_release\_channel) | n/a | `string` | n/a | yes |
 | <a name="input_subnetwork_name"></a> [subnetwork\_name](#input\_subnetwork\_name) | n/a | `string` | n/a | yes |
 | <a name="input_tpu_pools"></a> [tpu\_pools](#input\_tpu\_pools) | n/a | `list(map(any))` | n/a | yes |
 
 ## Outputs
 
 | Name | Description |
 |------|-------------|
-| <a name="output_cluster"></a> [cluster](#output\_cluster) | n/a |
+| <a name="output_cluster"></a> [cluster](#output\_cluster) | n/a |

modules/gke-standard-public-cluster/main.tf

@@ -25,6 +25,7 @@ module "gke" {
   cluster_resource_labels              = var.cluster_labels
   region                               = var.cluster_region
   kubernetes_version                   = var.kubernetes_version
+  release_channel                      = var.release_channel
   zones                                = var.cluster_zones
   network                              = var.network_name
   subnetwork                           = var.subnetwork_name

modules/gke-standard-public-cluster/variables.tf

@@ -50,6 +50,10 @@ variable "kubernetes_version" {
   type = string
 }
 
+variable "release_channel" {
+  type = string
+}
+
 variable "cluster_region" {
   type = string
 }
@@ -117,4 +121,4 @@ variable "gpu_pools" {
 
 variable "tpu_pools" {
   type = list(map(any))
-}
+}