feat: Implement autoscaling in Kubernetes for Terraform EKS

deployment-size.tf

@@ -6,34 +6,44 @@
 locals {
   deployment_size = {
     small = {
-      db            = "db.r6g.large",
-      node_count    = 2,
-      node_instance = "r6i.xlarge"
-      cache         = "cache.m6g.large"
+      db             = "db.r6g.large",
+      node_count     = 2,
+      node_instance  = "r6i.xlarge",
+      cache          = "cache.m6g.large",
+      min_node_count = 1,
+      max_node_count = 3
     },
     medium = {
-      db            = "db.r6g.xlarge",
-      node_count    = 2,
-      node_instance = "r6i.xlarge"
-      cache         = "cache.m6g.large"
+      db             = "db.r6g.xlarge",
+      node_count     = 2,
+      node_instance  = "r6i.xlarge",
+      cache          = "cache.m6g.large",
+      min_node_count = 2,
+      max_node_count = 5
     },
     large = {
-      db            = "db.r6g.2xlarge",
-      node_count    = 2,
-      node_instance = "r6i.2xlarge"
-      cache         = "cache.m6g.xlarge"
+      db             = "db.r6g.2xlarge",
+      node_count     = 2,
+      node_instance  = "r6i.2xlarge",
+      cache          = "cache.m6g.xlarge",
+      min_node_count = 2,
+      max_node_count = 6
     },
     xlarge = {
-      db            = "db.r6g.4xlarge",
-      node_count    = 3,
-      node_instance = "r6i.2xlarge"
-      cache         = "cache.m6g.xlarge"
+      db             = "db.r6g.4xlarge",
+      node_count     = 3,
+      node_instance  = "r6i.2xlarge",
+      cache          = "cache.m6g.xlarge",
+      min_node_count = 2,
+      max_node_count = 8
     },
     xxlarge = {
-      db            = "db.r6g.8xlarge",
-      node_count    = 3,
-      node_instance = "r6i.4xlarge"
-      cache         = "cache.m6g.2xlarge"
+      db             = "db.r6g.8xlarge",
+      node_count     = 3,
+      node_instance  = "r6i.4xlarge",
+      cache          = "cache.m6g.2xlarge",
+      min_node_count = 3,
+      max_node_count = 10
     }
   }
 }

main.tf

@@ -136,6 +136,8 @@ module "app_eks" {
 
   instance_types   = try([local.deployment_size[var.size].node_instance], var.kubernetes_instance_types)
   desired_capacity = try(local.deployment_size[var.size].node_count, var.kubernetes_node_count)
+  min_capacity     = try(local.deployment_size[var.size].min_node_count, var.min_node_count)
+  max_capacity     = try(local.deployment_size[var.size].max_node_count, var.max_node_count)
   map_accounts     = var.kubernetes_map_accounts
   map_roles        = var.kubernetes_map_roles
   map_users        = var.kubernetes_map_users
@@ -383,7 +385,6 @@ module "wandb" {
           }
         }
       }
-
       mysql = { install = false }
       redis = { install = false }
 

modules/app_eks/add-ons.tf

@@ -32,43 +32,43 @@ resource "aws_iam_role" "oidc" {
 ### add-ons for eks version 1.28
 
 resource "aws_eks_addon" "aws_efs_csi_driver" {
-   depends_on = [
-     aws_eks_addon.vpc_cni
-   ]
-   cluster_name               = var.namespace
-   addon_name                 = "aws-efs-csi-driver"
-   addon_version              = "v2.0.4-eksbuild.1"
-   resolve_conflicts          = "OVERWRITE"
+  depends_on = [
+    aws_eks_addon.vpc_cni
+  ]
+  cluster_name      = var.namespace
+  addon_name        = "aws-efs-csi-driver"
+  addon_version     = "v2.0.4-eksbuild.1"
+  resolve_conflicts = "OVERWRITE"
 }
 
 resource "aws_eks_addon" "aws_ebs_csi_driver" {
   depends_on = [
     aws_eks_addon.vpc_cni
   ]
-  cluster_name                = var.namespace
-  addon_name                  = "aws-ebs-csi-driver"
-  addon_version               = "v1.31.0-eksbuild.1"
-  resolve_conflicts           = "OVERWRITE"
+  cluster_name      = var.namespace
+  addon_name        = "aws-ebs-csi-driver"
+  addon_version     = "v1.31.0-eksbuild.1"
+  resolve_conflicts = "OVERWRITE"
 }
 
 resource "aws_eks_addon" "coredns" {
   depends_on = [
     aws_eks_addon.vpc_cni
   ]
-  cluster_name                = var.namespace
-  addon_name                  = "coredns"
-  addon_version               = "v1.10.1-eksbuild.11"
-  resolve_conflicts           = "OVERWRITE"
+  cluster_name      = var.namespace
+  addon_name        = "coredns"
+  addon_version     = "v1.10.1-eksbuild.11"
+  resolve_conflicts = "OVERWRITE"
 }
 
 resource "aws_eks_addon" "kube_proxy" {
   depends_on = [
     aws_eks_addon.vpc_cni
   ]
-  cluster_name                = var.namespace
-  addon_name                  = "kube-proxy"
-  addon_version               = "v1.28.8-eksbuild.5"
-  resolve_conflicts           = "OVERWRITE"
+  cluster_name      = var.namespace
+  addon_name        = "kube-proxy"
+  addon_version     = "v1.28.8-eksbuild.5"
+  resolve_conflicts = "OVERWRITE"
 }
 
 resource "aws_eks_addon" "vpc_cni" {

modules/app_eks/main.tf

@@ -55,10 +55,10 @@ module "eks" {
       iam_role_arn                         = aws_iam_role.node.arn,
       instance_types                       = var.instance_types,
       kubelet_extra_args                   = local.system_reserved != "" ? "--system-reserved=${local.system_reserved}" : "",
-      max_capacity                         = 5,
+      max_capacity                         = var.max_capacity,
       metadata_http_put_response_hop_limit = 2
       metadata_http_tokens                 = "required",
-      min_capacity                         = var.desired_capacity,
+      min_capacity                         = var.min_capacity,
       version                              = var.cluster_version,
     }
   }

modules/app_eks/variables.tf

@@ -122,6 +122,18 @@ variable "desired_capacity" {
   default     = 2
 }
 
+variable "min_capacity" {
+  description = "Minimum number of worker nodes."
+  type        = number
+  default     = 1
+}
+
+variable "max_capacity" {
+  description = "Maximum number of worker nodes."
+  type        = number
+  default     = 6
+}
+
 variable "system_reserved_cpu_millicores" {
   description = "(Optional) The amount of 'system-reserved' CPU millicores to pass to the kubelet. For example: 100.  A value of -1 disables the flag."
   type        = number

modules/app_lb/outputs.tf

@@ -15,9 +15,9 @@ output "tg_app_arn" {
 }
 
 output "alb_name" {
-value =  aws_lb.alb.arn
+  value = aws_lb.alb.arn
 }
 
 output "nlb_security_group" {
-  value = var.enable_private_only_traffic? aws_security_group.inbound_private[0].id : null
+  value = var.enable_private_only_traffic ? aws_security_group.inbound_private[0].id : null
 }

modules/endpoint/main.tf

@@ -1,9 +1,9 @@
 resource "aws_vpc_endpoint" "default" {
-  vpc_id              = var.network_id
-  service_name        = var.service_name
-  vpc_endpoint_type   = "Gateway" 
-  auto_accept = true
-  route_table_ids  = var.private_route_table_id
+  vpc_id            = var.network_id
+  service_name      = var.service_name
+  vpc_endpoint_type = "Gateway"
+  auto_accept       = true
+  route_table_ids   = var.private_route_table_id
 
   policy = <<POLICY
 {

modules/endpoint/variables.tf

@@ -1,14 +1,14 @@
 variable "network_id" {
-   type = string
-   description = "ID of the network (VPC) where infrastructure resources will be deployed."
- }
+  type        = string
+  description = "ID of the network (VPC) where infrastructure resources will be deployed."
+}
 
 variable "private_route_table_id" {
-  type = list(string)
+  type        = list(string)
   description = "Private route table ID within the specified network (VPC) where resources will be deployed"
 }
 
 variable "service_name" {
-  type = string
+  type        = string
   description = "Name of the service or vpc endpoint"
 }

modules/iam_role/variables.tf

@@ -4,7 +4,7 @@ variable "namespace" {
 }
 
 variable "aws_iam_openid_connect_provider_url" {
-  type        = string
+  type = string
 }
 
 variable "yace_sa_name" {

modules/private_link/main.tf

@@ -1,6 +1,6 @@
 locals {
   max_lb_name_length = 32 - length("-nlb")
-  lb_name_truncated  = var.enable_private_only_traffic ?  "${substr(var.namespace, 0, local.max_lb_name_length)}-private-link-nlb" : "${substr(var.namespace, 0, local.max_lb_name_length)}-nlb"
+  lb_name_truncated  = var.enable_private_only_traffic ? "${substr(var.namespace, 0, local.max_lb_name_length)}-private-link-nlb" : "${substr(var.namespace, 0, local.max_lb_name_length)}-nlb"
 }
 
 resource "aws_lb" "nlb" {
@@ -9,10 +9,10 @@ resource "aws_lb" "nlb" {
   load_balancer_type         = "network"
   subnets                    = var.network_private_subnets
   enable_deletion_protection = var.deletion_protection
-  security_groups = var.enable_private_only_traffic ? [var.nlb_security_group] : []
-lifecycle {
-  create_before_destroy = true
-}
+  security_groups            = var.enable_private_only_traffic ? [var.nlb_security_group] : []
+  lifecycle {
+    create_before_destroy = true
+  }
 }
 
 resource "aws_lb_target_group" "nlb" {

modules/private_link/variables.tf

@@ -29,8 +29,8 @@ variable "vpc_id" {
 }
 
 variable "enable_private_only_traffic" {
-  type        = bool
+  type = bool
 }
 variable "nlb_security_group" {
-  type        = string
+  type = string
 }

variables.tf

@@ -370,6 +370,18 @@ variable "kubernetes_node_count" {
   default     = 2
 }
 
+variable "min_node_count" {
+  description = "Number of nodes"
+  type        = number
+  default     = 1
+}
+
+variable "max_node_count" {
+  description = "Number of nodes"
+  type        = number
+  default     = 6
+}
+
 variable "eks_policy_arns" {
   type        = list(string)
   description = "Additional IAM policy to apply to the EKS cluster"