Merge pull request #9 from So1S/SO1S-393-infra-gpu-노드-terraform-프로비저닝

SO1S-393 GPU 노드 Terraform 프로비저닝

Author: NEM-NE

live/bootstrap.sh

@@ -31,6 +31,20 @@ elif [ $SO1S_ENV_NUMBER -eq 1 ]; then
   SO1S_ENV_PATH="./prod"
 fi
 
+SO1S_REGEX="^[1-2]$"
+while [[ ! $SO1S_USE_GPU =~ $SO1S_REGEX ]]
+do
+  echo -e "GPU를 사용할 것인지 번호를 입력해주세요. \n-> (1) 미사용 (2) 사용 "
+  read SO1S_USE_GPU
+done
+
+# 환경에 따른 글로벌 이름 설정 -> Prod은 고정 값
+if [ $SO1S_USE_GPU -eq 2 ]; then
+  INFERENCE_INSTANCE='["g4dn.xlarge"]'
+elif [ $SO1S_USE_GPU -eq 1 ]; then
+  INFERENCE_INSTANCE='["t3a.large"]'
+fi
+
 cd $SO1S_ENV_PATH
 
 # Check Terraform Version
@@ -51,7 +65,7 @@ fi
 echo -e "\n"
 echo "Start Resource Provisioning"
 echo "-> terraform apply -var=global_name=$SO1S_GLOBAL_NAME"
-terraform apply -var="global_name=$SO1S_GLOBAL_NAME"
+terraform apply -var="global_name=$SO1S_GLOBAL_NAME" -var="inference_node_instance_types=$INFERENCE_INSTANCE"
 
 # Using for ALB, External DNS Chart
 RESULT=`terraform output`

live/dev/main.tf

@@ -44,7 +44,7 @@ module "eks" {
 
     disk_size = 30
   }
-  inference_node_instance_types = ["t3a.medium"]
+  inference_node_instance_types = var.inference_node_instance_types
 
   api_node_spot = true
   api_node_size_spec = {

live/dev/variables.tf

@@ -1,4 +1,10 @@
 variable "global_name" {
   description = "This name will use as prefix for AWS resource so, Please write unique and distinct word e.g. sungbin, backend-test..."
   type = string
+}
+
+variable "inference_node_instance_types" {
+  description = "This type will use inference node that use eks"
+  type        = list(string)
+  default     = ["t3a.large"]
 }

live/prod/main.tf

@@ -44,7 +44,7 @@ module "eks" {
 
     disk_size = 30
   }
-  inference_node_instance_types = ["t3a.large"]
+  inference_node_instance_types = var.inference_node_instance_types
 
   api_node_spot = false
   api_node_size_spec = {

live/prod/variables.tf

@@ -3,20 +3,9 @@ variable "global_name" {
   type        = string
   default     = "prod"
 }
-variable "public_instance_types" {
-  description = "This type will use public node that use eks"
-  type        = list(string)
-  default     = ["t3a.medium"]
-}
 
-variable "inference_instance_types" {
+variable "inference_node_instance_types" {
   description = "This type will use inference node that use eks"
   type        = list(string)
   default     = ["t3a.large"]
 }
-
-variable "api_instance_types" {
-  description = "This type will use api node that use eks"
-  type        = list(string)
-  default     = ["t3a.large"]
-}

modules/cluster/variables.tf

@@ -55,7 +55,7 @@ variable "inference_node_spot" {
 variable "inference_node_instance_types" {
   description = "This type will use inference node that use eks"
   type        = list(string)
-  default     = ["t3a.medium"]
+  default     = ["t3a.large"]
 }
 
 variable "api_node_size_spec" {