feat: Add standardized instance sizing in AWS (#172)

* Veronica/tshirt deployment (#171) * feat: Add support for t-shirt-sized deployments * Remove duplicates --------- Co-authored-by: Veronica JungYeon Kim <[email protected]> * changes * test * terraform fmt * touchups * a bit o polish --------- Co-authored-by: Veronica JungYeon Kim <[email protected]>
wandb · Feb 8, 2024 · e6b4ab7 · e6b4ab7
1 parent ce44732
commit e6b4ab7
Show file tree

Hide file tree

Showing 9 changed files with 151 additions and 44 deletions.
diff --git a/deployment-size.tf b/deployment-size.tf
@@ -0,0 +1,39 @@
+##################################################
+# standard deployment sizes are specified in 
+# this object
+##################################################
+
+locals {
+  deployment_size = {
+    small = {
+      db            = "db.r6g.large",
+      node_count    = 2,
+      node_instance = "r6i.large"
+      cache         = "cache.m6g.large"
+    },
+    medium = {
+      db            = "db.r6g.xlarge",
+      node_count    = 2,
+      node_instance = "r6i.xlarge"
+      cache         = "cache.m6g.large"
+    },
+    large = {
+      db            = "db.r6g.2xlarge",
+      node_count    = 2,
+      node_instance = "r6i.2xlarge"
+      cache         = "cache.m6g.xlarge"
+    },
+    xlarge = {
+      db            = "db.r6g.4xlarge",
+      node_count    = 3,
+      node_instance = "r6i.4xlarge"
+      cache         = "cache.m6g.xlarge"
+    },
+    xxlarge = {
+      db            = "db.r6g.8xlarge",
+      node_count    = 3,
+      node_instance = "r6i.8xlarge"
+      cache         = "cache.m6g.2xlarge"
+    }
+  }
+}
diff --git a/examples/byo-vpc-eks/main.tf b/examples/byo-vpc-eks/main.tf
@@ -22,8 +22,8 @@ provider "aws" {
 
   default_tags {
     tags = {
-      GithubRepo = "terraform-aws-wandb"
-      GithubOrg  = "wandb"
+      GithubRepo  = "terraform-aws-wandb"
+      GithubOrg   = "wandb"
       Environment = "Production"
     }
   }
@@ -38,16 +38,16 @@ module "wandb_infra" {
   external_dns  = true
 
   create_vpc = false
-  
+
   network_id   = var.vpc_id
   network_cidr = var.vpc_cidr
 
-  network_private_subnets          = var.network_private_subnets
-  network_public_subnets           = var.network_public_subnets
-  network_database_subnets         = var.network_database_subnets 
-  network_private_subnet_cidrs     = var.network_private_subnet_cidrs
-  network_public_subnet_cidrs      = var.network_public_subnet_cidrs
-  network_database_subnet_cidrs    = var.network_database_subnet_cidrs
+  network_private_subnets       = var.network_private_subnets
+  network_public_subnets        = var.network_public_subnets
+  network_database_subnets      = var.network_database_subnets
+  network_private_subnet_cidrs  = var.network_private_subnet_cidrs
+  network_public_subnet_cidrs   = var.network_public_subnet_cidrs
+  network_database_subnet_cidrs = var.network_database_subnet_cidrs
 
   deletion_protection = false
 

diff --git a/examples/byo-vpc-eks/variables.tf b/examples/byo-vpc-eks/variables.tf
@@ -85,45 +85,45 @@ variable "allowed_inbound_ipv6_cidr" {
 }
 
 variable "vpc_id" {
-    type = string
-    description = "VPC network ID"
+  type        = string
+  description = "VPC network ID"
+}
+variable "vpc_cidr" {
+  type        = string
+  description = "VPC network CIDR"
 }
-variable vpc_cidr {
-    type = string
-    description = "VPC network CIDR"
-}   
 
-variable network_private_subnets {
-    type = list[string]
-    description = "Subnet IDs"
+variable "network_private_subnets" {
+  type        = list[string]
+  description = "Subnet IDs"
 }
 
-variable network_public_subnets {
-    type = list[string]
-    description = "Subnet IDs"
+variable "network_public_subnets" {
+  type        = list[string]
+  description = "Subnet IDs"
 }
 
-variable network_database_subnets {
-    type = list[string]
-    description = "Subnet IDs"
+variable "network_database_subnets" {
+  type        = list[string]
+  description = "Subnet IDs"
 }
 
-variable network_private_subnet_cidrs {
-    type = list[string]
-    description = "Subnet CIDRs"
+variable "network_private_subnet_cidrs" {
+  type        = list[string]
+  description = "Subnet CIDRs"
 }
 
-variable network_public_subnet_cidrs {
-    type = list[string]
-    description = "Subnet CIDRs"
+variable "network_public_subnet_cidrs" {
+  type        = list[string]
+  description = "Subnet CIDRs"
 }
 
-variable network_database_subnet_cidrs {
-    type = list[string]
-    description = "Subnet CIDRs"
+variable "network_database_subnet_cidrs" {
+  type        = list[string]
+  description = "Subnet CIDRs"
 }
 
-variable eks_cluster_version {
-    type = string
-    description = "EKS cluster version"
+variable "eks_cluster_version" {
+  type        = string
+  description = "EKS cluster version"
 }
diff --git a/examples/public-dns-external/main.tf b/examples/public-dns-external/main.tf
@@ -41,6 +41,7 @@ module "wandb_infra" {
   bucket_name        = var.bucket_name
   bucket_kms_key_arn = var.bucket_kms_key_arn
   use_internal_queue = true
+  size               = var.size
 }
 
 data "aws_eks_cluster" "app_cluster" {
@@ -100,3 +101,19 @@ output "bucket_name" {
 output "bucket_queue_name" {
   value = module.wandb_infra.bucket_queue_name
 }
+
+output "database_instance_type" {
+  value = module.wandb_infra.database_instance_type
+}
+
+output "eks_node_instance_type" {
+  value = module.wandb_infra.eks_node_instance_type
+}
+
+output "redis_instance_type" {
+  value = module.wandb_infra.redis_instance_type
+}
+
+output "standardized_size" {
+  value = var.size
+}
diff --git a/examples/public-dns-external/variables.tf b/examples/public-dns-external/variables.tf
@@ -13,6 +13,13 @@ variable "zone_id" {
   description = "Id of Route53 zone"
 }
 
+variable "size" {
+  default     = "small"
+  description = "Deployment size"
+  nullable    = true
+  type        = string
+}
+
 variable "subdomain" {
   type        = string
   default     = null

diff --git a/main.tf b/main.tf
@@ -69,7 +69,7 @@ module "database" {
   database_name   = var.database_name
   master_username = var.database_master_username
 
-  instance_class      = var.database_instance_class
+  instance_class      = try(local.deployment_size[var.size].db, var.database_instance_class)
   engine_version      = var.database_engine_version
   snapshot_identifier = var.database_snapshot_identifier
   sort_buffer_size    = var.database_sort_buffer_size
@@ -121,10 +121,11 @@ module "app_eks" {
   namespace   = var.namespace
   kms_key_arn = local.kms_key_arn
 
-  instance_types = var.kubernetes_instance_types
-  map_accounts   = var.kubernetes_map_accounts
-  map_roles      = var.kubernetes_map_roles
-  map_users      = var.kubernetes_map_users
+  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)
+  map_accounts     = var.kubernetes_map_accounts
+  map_roles        = var.kubernetes_map_roles
+  map_users        = var.kubernetes_map_users
 
   bucket_kms_key_arn   = local.use_external_bucket ? var.bucket_kms_key_arn : local.kms_key_arn
   bucket_arn           = data.aws_s3_bucket.file_storage.arn
@@ -195,9 +196,8 @@ module "redis" {
   vpc_id                  = local.network_id
   redis_subnet_group_name = local.network_elasticache_subnet_group_name
   vpc_subnets_cidr_blocks = module.networking.elasticache_subnet_cidrs
-  node_type               = var.elasticache_node_type
-
-  kms_key_arn = local.kms_key_arn
+  node_type               = try(local.deployment_size[var.size].cache, var.elasticache_node_type)
+  kms_key_arn             = local.kms_key_arn
 }
 
 locals {
@@ -285,3 +285,4 @@ module "wandb" {
     }
   }
 }
+
diff --git a/modules/app_eks/variables.tf b/modules/app_eks/variables.tf
@@ -115,3 +115,9 @@ variable "service_port" {
   type    = number
   default = 32543
 }
+
+variable "desired_capacity" {
+  description = "Desired number of worker nodes."
+  type        = number
+  default     = 2
+}
diff --git a/outputs.tf b/outputs.tf
@@ -26,12 +26,27 @@ output "database_password" {
   sensitive = true
   value     = module.database.password
 }
+
+output "database_instance_type" {
+  value = local.deployment_size[var.size].db
+}
+
 output "elasticache_connection_string" {
   value = var.create_elasticache ? module.redis.0.connection_string : null
 }
+
+output "eks_node_count" {
+  value = local.deployment_size[var.size].node_count
+}
+
+output "eks_node_instance_type" {
+  value = local.deployment_size[var.size].node_instance
+}
+
 output "internal_app_port" {
   value = local.internal_app_port
 }
+
 output "kms_key_arn" {
   value       = local.kms_key_arn
   description = "The Amazon Resource Name of the KMS key used to encrypt data at rest."
@@ -51,6 +66,15 @@ output "network_public_subnets" {
   value       = local.network_public_subnets
   description = "The identities of the public subnetworks deployed within the VPC."
 }
+
+output "redis_instance_type" {
+  value = local.deployment_size[var.size].cache
+}
+
+output "standardized_size" {
+  value = var.size
+}
+
 output "url" {
   value       = local.url
   description = "The URL to the W&B application"

diff --git a/variables.tf b/variables.tf
@@ -17,6 +17,13 @@ variable "use_internal_queue" {
   default = false
 }
 
+variable "size" {
+  default     = "small"
+  description = "Deployment size"
+  nullable    = true
+  type        = string
+}
+
 ##########################################
 # Database                               #
 ##########################################
@@ -306,6 +313,12 @@ variable "kubernetes_instance_types" {
   default     = ["m5.large"]
 }
 
+variable "kubernetes_node_count" {
+  description = "Number of nodes"
+  type        = number
+  default     = 2
+}
+
 variable "eks_policy_arns" {
   type        = list(string)
   description = "Additional IAM policy to apply to the EKS cluster"