J7m4/infra 240 aws redis in cluster

examples/public-dns-external/main.tf

@@ -11,6 +11,18 @@ provider "aws" {
   }
 }
 
+
+locals {
+  env_vars = merge(
+    {
+      "TAG_CUSTOMER_NS"    = var.namespace
+      "TAG_CLOUD"          = "AWS"
+    },
+    var.other_wandb_env
+  )
+}
+
+
 module "wandb_infra" {
   source = "../../"
 
@@ -38,6 +50,8 @@ module "wandb_infra" {
 
   license = var.wandb_license
 
+  other_wandb_env = local.env_vars
+
   bucket_name        = var.bucket_name
   bucket_path        = var.bucket_path
   bucket_kms_key_arn = var.bucket_kms_key_arn
@@ -50,6 +64,14 @@ module "wandb_infra" {
   system_reserved_pid                 = var.system_reserved_pid
 
   aws_loadbalancer_controller_tags = var.aws_loadbalancer_controller_tags
+
+  create_elasticache        = var.create_elasticache
+  create_redis_in_cluster   = var.create_redis_in_cluster
+  use_redis_in_cluster      = var.use_redis_in_cluster
+  redis_master_name         = var.redis_master_name
+  redis_service_name_prefix = var.redis_service_name_prefix
+
+  kubernetes_alb_internet_facing = var.kubernetes_alb_internet_facing
 }
 
 data "aws_eks_cluster" "app_cluster" {

examples/public-dns-external/variables.tf

@@ -131,3 +131,39 @@ variable "aws_loadbalancer_controller_tags" {
   type        = map(string)
   default     = {}
 }
+
+variable "use_redis_in_cluster" {
+  type        = bool
+  default     = false
+  description = "whether to actually use redis in cluster"
+}
+
+variable "create_redis_in_cluster" {
+  type        = bool
+  default     = false
+  description = "whether to create an in-cluster redis"
+}
+
+variable "create_elasticache" {
+  type        = bool
+  default     = true
+  description = "whether to create an elasticache redis"
+}
+
+variable "redis_service_name_prefix" {
+  type        = string
+  default     = null
+  description = "Prefix of the redis service name"
+}
+
+variable "redis_master_name" {
+  type        = string
+  default     = null
+  description = "Name of the redis master that sentinel uses to monitor"
+}
+
+variable "kubernetes_alb_internet_facing" {
+  type        = bool
+  default     = true
+  description = "Whether to enable private only traffic"
+}

main.tf

@@ -146,6 +146,11 @@ module "app_eks" {
   map_roles    = var.kubernetes_map_roles
   map_users    = var.kubernetes_map_users
 
+  depends_on = [
+    module.app_lb,
+    module.networking,
+  ]
+
   bucket_kms_key_arns = compact([
     local.default_kms_key,
     var.bucket_kms_key_arn != "" && var.bucket_kms_key_arn != null ? var.bucket_kms_key_arn : null
@@ -250,9 +255,10 @@ locals {
   weave_trace_sa_name = "wandb-weave-trace"
 }
 
-module "wandb" {
-  source  = "wandb/wandb/helm"
-  version = "2.0.0"
+module "wandb_with_cleanup_delay" {
+  # source  = "wandb/wandb/helm"
+  # version = "2.0.0"
+  source  = "../terraform-helm-wandb"
 
   depends_on = [
     module.database,
@@ -264,6 +270,10 @@ module "wandb" {
   controller_image_tag   = var.controller_image_tag
   enable_helm_release    = var.enable_helm_release
 
+  redis_master_name          = var.redis_master_name
+  redis_service_name_prefix  = var.redis_service_name_prefix
+  create_redis_in_cluster    = var.create_redis_in_cluster
+
   spec = {
     values = {
       global = {
@@ -289,8 +299,8 @@ module "wandb" {
         }
 
         redis = {
-          host = module.redis[0].host
-          port = "${module.redis[0].port}?tls=true&ttlInSeconds=604800"
+          host = var.use_redis_in_cluster ? "${var.redis_service_name_prefix}-redis" : module.redis[0].host
+          port = var.use_redis_in_cluster ? "26379?master=${var.redis_master_name}&ttlInSeconds=604800" : "${module.redis[0].port}?tls=true&ttlInSeconds=604800"
         }
       }
 
@@ -358,6 +368,12 @@ module "wandb" {
       mysql = { install = false }
       redis = { install = false }
 
+      redisInCluster = var.use_redis_in_cluster ? {
+        enabled = true
+      } : {
+        enabled = false
+      }
+
       weave = {
         persistence = {
           provider = "efs"

modules/app_eks/lb_controller/controller.tf

@@ -6,11 +6,13 @@ locals {
 }
 
 resource "helm_release" "aws_load_balancer_controller" {
-  name       = "aws-load-balancer-controller"
-  repository = "https://aws.github.io/eks-charts"
-  chart      = "aws-load-balancer-controller"
-  namespace  = "kube-system"
-  version    = "1.7.2"
+  name          = "aws-load-balancer-controller"
+  repository    = "https://aws.github.io/eks-charts"
+  chart         = "aws-load-balancer-controller"
+  namespace     = "kube-system"
+  version       = "1.7.2"
+  wait          = true
+  wait_for_jobs = true
 
   set {
     name  = "clusterName"

modules/app_eks/main.tf

@@ -60,6 +60,9 @@ module "eks" {
     metadata_http_put_response_hop_limit = 2
     metadata_http_tokens                 = "required",
     version                              = var.cluster_version,
+    tags                                 = {
+      "customer-ns" = var.namespace
+    }
   }
 
   node_groups = {

modules/database/main.tf

@@ -133,12 +133,16 @@ module "aurora" {
   performance_insights_retention_period = 7
   preferred_backup_window               = var.preferred_backup_window
   preferred_maintenance_window          = var.preferred_maintenance_window
-  security_group_tags                   = { "Namespace" : "${var.namespace}" }
+  security_group_tags                   = { "Namespace" : var.namespace }
   skip_final_snapshot                   = true
   snapshot_identifier                   = var.snapshot_identifier
   storage_encrypted                     = true
   subnets                               = var.subnets
   vpc_id                                = var.vpc_id
 
+  tags = {
+    Namespace = var.namespace
+  }
+
 
 }

modules/redis/main.tf

@@ -29,6 +29,10 @@ resource "aws_elasticache_replication_group" "default" {
   kms_key_id                 = var.kms_key_arn
   at_rest_encryption_enabled = true
   transit_encryption_enabled = true
+
+  tags = {
+    Namespace = var.namespace
+  }
 }
 
 resource "aws_security_group" "redis" {

variables.tf

@@ -480,6 +480,30 @@ variable "elasticache_node_type" {
   default     = null
 }
 
+variable "create_redis_in_cluster" {
+  type        = bool
+  default     = false
+  description = "Create redis in cluster"
+}
+
+variable "use_redis_in_cluster" {
+  type        = bool
+  default     = false
+  description = "whether to actually use the redis in cluster"
+}
+
+variable "redis_service_name_prefix" {
+  type        = string
+  default     = null
+  description = "Used to infer the service name as `redis-service-name-prefix`-redis.`namespace`"
+}
+
+variable "redis_master_name" {
+  type        = string
+  default     = null
+  description = "Name used by sentinel to identify the set of redis nodes"
+}
+
 ##########################################
 # Weights & Biases                       #
 ##########################################