Refactor out common modules in terraform/gcp_old

terraform/gcp_old/tpu-inference/modules/benchmark/main.tf → terraform/gcp_old/modules/benchmark/main.tf

@@ -5,12 +5,12 @@
 # Runtime: v2-alpha-tpuv6e
 
 data "google_secret_manager_secret_version" "buildkite_agent_token_benchmark_cluster" {
-  secret = "projects/${var.project_id}/secrets/bm-agent-hf-token"
+  secret = "projects/${var.project_id}/secrets/${var.buildkite_agent_token_benchmark_cluster_name}"
   version = "latest"
 }
 
 data "google_secret_manager_secret_version" "huggingface_token" {
-  secret = "projects/${var.project_id}/secrets/tpu_commons_buildkite_hf_token"
+  secret = "projects/${var.project_id}/secrets/${var.huggingface_token_name}"
   version = "latest"
 }
 

terraform/gcp_old/modules/benchmark/variables.tf

@@ -0,0 +1,13 @@
+variable "project_id" {
+  default = "cloud-tpu-inference-test"
+}
+
+variable "buildkite_agent_token_benchmark_cluster_name" {
+  type = string
+  description = "google_secret_manager_secret name for benchmark cluster agent token"
+}
+
+variable "huggingface_token_name" {
+  type = string
+  description = "google_secret_manager_secret name for huggingface token"
+}

terraform/gcp_old/tpu-inference/modules/ci_cpu/main.tf → terraform/gcp_old/modules/ci_cpu/main.tf

@@ -4,7 +4,7 @@
 # Type: e2-standard-2
 
 data "google_secret_manager_secret_version" "buildkite_agent_token_ci_cluster" {
-  secret = "projects/${var.project_id}/secrets/tpu_commons_buildkite_agent_token"
+  secret = "projects/${var.project_id}/secrets/${var.buildkite_agent_token_ci_cluster_name}"
   version = "latest"
 }
 

terraform/gcp_old/modules/ci_cpu/variables.tf

@@ -0,0 +1,12 @@
+variable "project_id" {
+  default = "cloud-tpu-inference-test"
+}
+
+variable "instance_count" {
+  default = 8
+}
+
+variable "buildkite_agent_token_ci_cluster_name" {
+  type = string
+  description = "google_secret_manager_secret name for ci cluster agent token"
+}

terraform/gcp_old/tpu-inference/modules/ci_v5/main.tf → terraform/gcp_old/modules/ci_v5/main.tf

@@ -1,10 +1,10 @@
 data "google_secret_manager_secret_version" "buildkite_agent_token_ci_cluster" {
-  secret = "projects/${var.project_id}/secrets/buildkite_agent_token_ci_cluster"
+  secret = "projects/${var.project_id}/secrets/${var.buildkite_agent_token_ci_cluster_name}"
   version = "latest"
 }
 
 data "google_secret_manager_secret_version" "huggingface_token" {
-  secret = "projects/${var.project_id}/secrets/huggingface_token"
+  secret = "projects/${var.project_id}/secrets/${var.huggingface_token_name}"
   version = "latest"
 }
 

terraform/gcp_old/modules/ci_v5/variables.tf

@@ -0,0 +1,13 @@
+variable "project_id" {
+  default = "vllm-405802"
+}
+
+variable "buildkite_agent_token_ci_cluster_name" {
+  type = string
+  description = "google_secret_manager_secret name for ci cluster agent token"
+}
+
+variable "huggingface_token_name" {
+  type = string
+  description = "google_secret_manager_secret name for huggingface token"
+}

terraform/gcp_old/tpu-inference/modules/ci_v6/main.tf → terraform/gcp_old/modules/ci_v6/main.tf

@@ -5,12 +5,12 @@
 # Runtime: v2-alpha-tpuv6e
 
 data "google_secret_manager_secret_version" "buildkite_agent_token_ci_cluster" {
-  secret = "projects/${var.project_id}/secrets/tpu_commons_buildkite_agent_token"
+  secret = "projects/${var.project_id}/secrets/${var.buildkite_agent_token_ci_cluster_name}"
   version = "latest"
 }
 
 data "google_secret_manager_secret_version" "huggingface_token" {
-  secret  = "projects/${var.project_id}/secrets/tpu_commons_buildkite_hf_token"
+  secret  = "projects/${var.project_id}/secrets/${var.huggingface_token_name}"
   version = "latest"
 }
 
@@ -21,17 +21,17 @@ locals {
 
 resource "google_compute_disk" "disk_east5_b" {
   provider = google-beta.us-east5-b
-  count = 24
+  count = "${var.ci_v6_instance_count}"
 
   name  = "tpu-disk-east5-b-${count.index}"
-  size  = 2048
+  size  = "${var.ci_v6_disk_size}"
   type  = "hyperdisk-balanced"
   zone  = "us-east5-b"
 }
 
 resource "google_tpu_v2_vm" "tpu_v6_ci" {
   provider = google-beta.us-east5-b
-  count = 24
+  count = "${var.ci_v6_instance_count}"
   name = "vllm-tpu-v6-ci-${count.index}"
   zone = "us-east5-b" 
 

terraform/gcp_old/modules/ci_v6/variables.tf

@@ -0,0 +1,23 @@
+variable "project_id" {
+  default = "cloud-tpu-inference-test"
+}
+
+variable "buildkite_agent_token_ci_cluster_name" {
+  type = string
+  description = "google_secret_manager_secret name for ci cluster agent token"
+}
+
+variable "huggingface_token_name" {
+  type = string
+  description = "google_secret_manager_secret name for huggingface token"
+}
+
+variable "ci_v6_instance_count" {
+  type = number
+  description = "number of instances to spawn for ci_v6"
+}
+
+variable "ci_v6_disk_size" {
+  type = number
+  description = "disk size for ci_v6"
+}

terraform/gcp_old/tpu-inference/main.tf

@@ -1,35 +1,38 @@
 # module "benchmark" {
-#   source = "./modules/benchmark"
+#   source = "../modules/benchmark"
 #   providers = {
 #     google-beta.us-east1-d = google-beta.us-east1-d
 #   }
-
-#   buildkite_agent_token_benchmark_cluster = var.buildkite_agent_token_benchmark_cluster
-#   huggingface_token     = var.huggingface_token
+#   project_id            = var.project_id
 # }
 
 module "ci_v6" {
-  source = "./modules/ci_v6"
+  source = "../modules/ci_v6"
   providers = {
     google-beta.us-east5-b = google-beta.us-east5-b
   }
   project_id            = var.project_id
+  buildkite_agent_token_ci_cluster_name = var.buildkite_agent_token_ci_cluster_name
+  huggingface_token_name = var.huggingface_token_name
+  ci_v6_instance_count = var.ci_v6_instance_count
+  ci_v6_disk_size = var.ci_v6_disk_size
 }
 
 module "ci_cpu" {
-  source    = "./modules/ci_cpu"
+  source    = "../modules/ci_cpu"
   providers = {
     google-beta.us-east5-b = google-beta.us-east5-b
   }
-  project_id                       = var.project_id
+  project_id = var.project_id
+  buildkite_agent_token_ci_cluster_name = var.buildkite_agent_token_ci_cluster_name
+
+
 }
 
 # module "ci_v5" {
-#   source = "./modules/ci_v5"
+#   source = "../modules/ci_v5"
 #   providers = {
 #     google-beta.us-south1-a = google-beta.us-south1-a
 #   }
-
-#   buildkite_agent_token_ci_cluster = var.buildkite_agent_token_ci_cluster
-#   huggingface_token     = var.huggingface_token
+#   project_id = var.project_id
 # }

terraform/gcp_old/tpu-inference/variables.tf

@@ -1,3 +1,23 @@
 variable "project_id" {
   default = "cloud-tpu-inference-test"
 }
+
+variable "buildkite_agent_token_benchmark_cluster_name" {
+  default = "bm-agent-hf-token"
+}
+
+variable "buildkite_agent_token_ci_cluster_name" {
+  default = "tpu_commons_buildkite_agent_token"
+}
+
+variable "huggingface_token_name" {
+  default = "tpu_commons_buildkite_hf_token"
+}
+
+variable "ci_v6_instance_count" {
+  default = 24
+}
+
+variable "ci_v6_disk_size" {
+  default = 2048
+}

terraform/gcp_old/vllm/main.tf

@@ -1,24 +1,32 @@
 module "benchmark" {
-  source = "./modules/benchmark"
+  source = "../modules/benchmark"
   providers = {
     google-beta.us-east1-d = google-beta.us-east1-d
   }
 
   project_id = var.project_id
+  buildkite_agent_token_benchmark_cluster_name = var.buildkite_agent_token_benchmark_cluster_name
+  huggingface_token_name = var.huggingface_token_name
 }
 module "ci_v6" {
-  source = "./modules/ci_v6"
+  source = "../modules/ci_v6"
   providers = {
     google-beta.us-east5-b = google-beta.us-east5-b
   }  
   project_id = var.project_id
+  buildkite_agent_token_ci_cluster_name = var.buildkite_agent_token_ci_cluster_name
+  huggingface_token_name = var.huggingface_token_name
+  ci_v6_instance_count = var.ci_v6_instance_count
+  ci_v6_disk_size = var.ci_v6_disk_size
 }
 
 module "ci_v5" {
-  source = "./modules/ci_v5"
+  source = "../modules/ci_v5"
   providers = {
     google-beta.us-south1-a = google-beta.us-south1-a
   }
 
   project_id = var.project_id
+  buildkite_agent_token_ci_cluster_name = var.buildkite_agent_token_ci_cluster_name
+  huggingface_token_name = var.huggingface_token_name
 }