OpenTracing, Worker Healthcheck, Filter EIPs by a service name, fix D…

…atadog, Balancer-less EC apps - Added OpenTracing configuration for Datadog - Added healthcheck parameter for worker apps - Filter EIPs by service name during instance launch - Fix datadog agent hostname to work in host network mode - EC2 port mappings exposure and Balancer-less instances
hazelops · May 25, 2022 · ce9d119 · ce9d119
1 parent 99be7b5
commit ce9d119
Show file tree

Hide file tree

Showing 10 changed files with 123 additions and 46 deletions.
diff --git a/autoscaling.tf b/autoscaling.tf
@@ -2,10 +2,12 @@ resource "aws_eip" "autoscaling" {
   # If ec2_eip_count is set, use that number for number of EIPs, otherwise use var.max_size + 1 (but that might not be the best during downscaling and deletion of EIPs
   count             = var.ec2_eip_enabled ? (var.ec2_eip_count > 0 ? var.ec2_eip_count : var.max_size + 1) : 0
   public_ipv4_pool  = "amazon"
+  vpc = true
 
   tags = {
     Name = "${local.name}-${count.index + 1}"
     env  = var.env
+    service = local.name
   }
 }
 

diff --git a/data.tf b/data.tf
@@ -7,6 +7,7 @@ data "template_file" "asg_ecs_ec2_user_data" {
 
   vars = {
     ecs_cluster_name  = local.ecs_cluster_name
+    service           = local.name
     env               = var.env
     ec2_service_group = var.ec2_service_group
     ec2_eip_enabled   = tostring(var.ec2_eip_enabled)

diff --git a/ecs-modules/ecs-service/main.tf b/ecs-modules/ecs-service/main.tf
@@ -4,11 +4,12 @@ module "task" {
   env  = var.env
   name = var.name
 
-  ecs_task_family_name = var.ecs_service_name != "" ? var.ecs_service_name : ""
-  ecs_launch_type      = var.ecs_launch_type
-  ecs_network_mode     = var.ecs_network_mode
-  ecs_volumes_from     = var.ecs_volumes_from
-  ecs_exec_enabled     = var.ecs_exec_enabled
+  ecs_task_family_name          = var.ecs_service_name != "" ? var.ecs_service_name : ""
+  ecs_launch_type               = var.ecs_launch_type
+  ecs_network_mode              = var.ecs_network_mode
+  ecs_volumes_from              = var.ecs_volumes_from
+  ecs_exec_enabled              = var.ecs_exec_enabled
+  ecs_task_health_check_command = var.ecs_task_health_check_command
 
   docker_image_name           = var.docker_image_name
   docker_image_tag            = var.docker_image_tag
@@ -81,7 +82,7 @@ resource "aws_ecs_service" "this" {
 
 
   dynamic "service_registries" {
-    for_each = (var.ecs_launch_type == "FARGATE" && var.ecs_service_discovery_enabled) ? [
+    for_each = (var.ecs_launch_type == "FARGATE" && var.ecs_service_discovery_enabled && var.app_type == "web") ? [
       1
     ] : []
     content {
@@ -102,14 +103,14 @@ resource "aws_ecs_service" "this" {
   }
 
   dynamic "load_balancer" {
-    for_each = [
+    for_each = var.app_type == "web" ? [
     for p in var.port_mappings : {
       container_name   = p.container_name
       target_group_arn = p.target_group_arn
       container_port   = p.container_port
 
     }
-    ]
+    ]: []
 
     content {
       target_group_arn = load_balancer.value.target_group_arn
@@ -176,13 +177,14 @@ resource "aws_ecs_service" "this_deployed" {
   }
 
   dynamic "load_balancer" {
-    for_each = [
+    for_each = var.app_type == "web" ? [
     for p in var.port_mappings : {
       container_name   = p.container_name
       container_port   = p.container_port
       target_group_arn = p.target_group_arn
     }
-    ]
+    ] : []
+
 
     content {
       target_group_arn = load_balancer.value.target_group_arn

diff --git a/ecs-modules/ecs-service/variables.tf b/ecs-modules/ecs-service/variables.tf
@@ -14,6 +14,17 @@ variable "name" {
   description = "ECS app name"
 }
 
+variable "app_type" {
+  type        = string
+  description = "ECS application type. Valid values: web (with load balancer), worker (scheduled task without ALB)."
+  default     = "web"
+
+  validation {
+    condition     = var.app_type == "web" || var.app_type == "worker"
+    error_message = "The app_type value must be a valid type: web or worker."
+  }
+}
+
 variable "namespace" {
   type        = string
   description = "Namespace name within the infrastructure"
@@ -284,6 +295,11 @@ variable "aws_service_discovery_private_dns_namespace" {
   default     = ""
 }
 
+variable "ecs_task_health_check_command" {
+  type        = string
+  description = "Command to check for the health of the container"
+}
+
 variable "ecs_network_mode" {
   type        = string
   description = "Corresponds to networkMode in an ECS task definition. Supported values are none, bridge, host, or awsvpc"

diff --git a/ecs-modules/ecs-task/variables.tf b/ecs-modules/ecs-task/variables.tf
@@ -21,6 +21,16 @@ locals {
       cpu               = var.ecs_launch_type == "FARGATE" ? var.cpu : null
       memoryReservation = var.memory_reservation
       essential         = true
+      healthCheck       = length(var.ecs_task_health_check_command) > 0 ? {
+        retries    = 3
+        timeout    = 5
+        interval   = 30
+        startPerid = null
+        command    = [
+          "CMD-SHELL",
+          var.ecs_task_health_check_command
+        ]
+      } : null
 
       linuxParameters = {
         sharedMemorySize = (var.shared_memory_size > 0 && var.ecs_launch_type != "FARGATE") ? var.shared_memory_size : null
@@ -74,7 +84,7 @@ locals {
           "Name"           = "datadog"
           "Host"           = "http-intake.logs.datadoghq.com"
           "apiKey"         = data.aws_ssm_parameter.dd_api_key[0].value
-          "dd_service"     = "${var.name}"
+          "dd_service"     = var.name
           "dd_source"      = "ecs"
           "dd_tags"        = "fluentbit:true,env:${var.env},service:${var.env}-${var.name}"
           "dd_message_key" = "log"
@@ -338,6 +348,11 @@ variable "ecs_volumes_from" {
   default     = []
 }
 
+variable "ecs_task_health_check_command" {
+  type        = string
+  description = "Command to check for the health of the container"
+}
+
 variable "resource_requirements" {
   type        = list(any)
   description = "The ResourceRequirement property specifies the type and amount of a resource to assign to a container. The only supported resource is a GPU"

diff --git a/ecs_ec2_user_data.sh.tpl b/ecs_ec2_user_data.sh.tpl
@@ -70,7 +70,7 @@ aws_get_instance_environment() {
 }
 
 aws_get_unassigned_eips() {
-	local describe_addreses_response=$(aws ec2 describe-addresses --region $instance_region --filters "Name=tag:env,Values=$instance_environment" --query "Addresses[?AssociationId==null].AllocationId" --output text)
+	local describe_addreses_response=$(aws ec2 describe-addresses --region $instance_region --filters "Name=tag:env,Values=$instance_environment" "Name=tag:service,Values=${service}" --query "Addresses[?AssociationId==null].AllocationId" --output text)
 	eips=($${describe_addreses_response///})
 	if [ -n "$describe_addreses_response" ]; then return 0; else return 1; fi
 }

diff --git a/main.tf b/main.tf
@@ -98,21 +98,23 @@ module "service" {
   env              = var.env
   name             = var.name
   namespace        = var.namespace
+  app_type         = var.app_type
   ecs_cluster_name = local.ecs_cluster_name
   ecs_service_name = local.ecs_service_name
 
-  ecs_platform_version  = var.ecs_launch_type == "FARGATE" ? var.ecs_platform_version : null
-  ecs_launch_type       = var.ecs_launch_type
-  ec2_service_group     = var.ec2_service_group
-  docker_container_port = var.docker_container_port
-  ecs_network_mode      = var.ecs_network_mode
-  ecs_volumes_from      = var.ecs_volumes_from
-  cpu                   = var.cpu
-  memory                = var.memory
-  memory_reservation    = var.memory_reservation
-  volumes               = local.volumes
-  assign_public_ip      = var.assign_public_ip
-  security_groups       = var.security_groups
+  ecs_platform_version          = var.ecs_launch_type == "FARGATE" ? var.ecs_platform_version : null
+  ecs_launch_type               = var.ecs_launch_type
+  ecs_task_health_check_command = var.ecs_task_health_check_command
+  ec2_service_group             = var.ec2_service_group
+  docker_container_port         = var.docker_container_port
+  ecs_network_mode              = var.ecs_network_mode
+  ecs_volumes_from              = var.ecs_volumes_from
+  cpu                           = var.cpu
+  memory                        = var.memory
+  memory_reservation            = var.memory_reservation
+  volumes                       = local.volumes
+  assign_public_ip              = var.assign_public_ip
+  security_groups               = var.security_groups
 
   web_proxy_enabled = var.web_proxy_enabled
   ecs_exec_enabled  = var.ecs_exec_enabled
@@ -189,35 +191,42 @@ module "service" {
   # TODO: instead of hardcoding the index, better use dynamic lookup by a canonical name
   target_group_arn = var.app_type == "web" && length(module.alb[*].target_group_arns) >= 1 ? module.alb[0].target_group_arns[0] : null
 
-  port_mappings = var.app_type == "web" ? [
+  port_mappings = jsondecode(var.app_type == "web" ? jsonencode([
     {
       container_name   = var.web_proxy_enabled ? "nginx" : var.name
       container_port   = var.web_proxy_enabled ? var.web_proxy_docker_container_port : var.docker_container_port
       host_port        = var.ecs_network_mode == "awsvpc" ? (var.web_proxy_enabled ? var.web_proxy_docker_container_port : var.docker_container_port) : var.docker_host_port
       target_group_arn = length(module.alb[*].target_group_arns) >= 1 ? module.alb[0].target_group_arns[0] : ""
     }
-  ] : []
+  ]) : jsonencode(var.port_mappings))
 
   environment = merge(var.environment, local.datadog_env_vars, local.ecs_exec_env_vars, {
     APP_NAME      = var.name
     ENV           = var.env
     PROXY_ENABLED = var.web_proxy_enabled ? "true" : "false"
-  }, var.ecs_launch_type == "EC2" ? {
-    DD_AGENT_HOST = "datadog-agent"
-  } : {}
+  }
   )
 }
 
-resource "aws_route53_record" "this" {
+resource "aws_route53_record" "alb" {
   count   = var.app_type == "web" ? length(local.domain_names) : 0
   zone_id = var.zone_id
   name    = local.domain_names[count.index]
   type    = "A"
 
-  alias {
-    name                   = module.alb[0].this_lb_dns_name
-    zone_id                = module.alb[0].this_lb_zone_id
-    evaluate_target_health = true
-  }
+  alias  {
+      name                   = module.alb[0].this_lb_dns_name
+      zone_id                = module.alb[0].this_lb_zone_id
+      evaluate_target_health = true
+    }
+}
+
+resource "aws_route53_record" "ec2" {
+  count   = (var.ecs_launch_type == "EC2" && var.ec2_eip_enabled && var.ec2_eip_dns_enabled) ? length(local.domain_names) : 0
+  zone_id = var.zone_id
+  name    = local.domain_names[count.index]
+  type    = "A"
+
+  records = var.ec2_eip_enabled ? aws_eip.autoscaling.*.public_ip : []
 }
 
diff --git a/monitoring.tf b/monitoring.tf
@@ -1,7 +1,7 @@
 # Datadog Logging/Monitoring Module (can be enabled/disabled via datadog_enabled)
 module "datadog" {
   source  = "hazelops/ecs-datadog-agent/aws"
-  version = "~> 3.0"
+  version = "~> 3.2"
 
   app_name             = var.name
   env                  = var.env
@@ -20,7 +20,7 @@ module "route_53_health_check" {
 
   enabled                        = var.route53_health_check_enabled
   env                            = var.env
-  fqdn                           = var.app_type == "web" ? aws_route53_record.this[0].name : null
+  fqdn                           = var.app_type == "web" ? aws_route53_record.alb[0].name : null
   domain_name                    = var.root_domain_name
   name                           = var.name
   subscription_endpoint_protocol = var.sns_service_subscription_endpoint_protocol

diff --git a/outputs.tf b/outputs.tf
@@ -29,3 +29,11 @@ output "efs" {
 output "eips" {
   value = var.ec2_eip_enabled ? aws_eip.autoscaling.*.public_ip : []
 }
+
+output "public_ip" {
+  value = (var.ec2_eip_enabled && length(aws_eip.autoscaling)>0) ? aws_eip.autoscaling.0.public_ip : ""
+}
+
+output "ec2_dns_name" {
+  value = var.ec2_eip_dns_enabled ? aws_route53_record.ec2.0.fqdn : ""
+}
diff --git a/variables.tf b/variables.tf
@@ -10,15 +10,21 @@ locals {
   # Datadog Environment Variables: https://docs.datadoghq.com/agent/guide/environment-variables/
   #                                https://docs.datadoghq.com/agent/docker/apm/?tab=linux#docker-apm-agent-environment-variables
   datadog_env_vars = var.datadog_enabled ? {
-    DD_PROFILING_ENABLED       = "true"
-    DD_TRACE_ENABLED           = "true"
-    DD_RUNTIME_METRICS_ENABLED = "true"
-    DD_APM_ENABLED             = "true"
-    DD_SERVICE                 = var.name
-    DD_SERVICE_NAME            = var.name
-    DD_ENV                     = var.env
+    DD_PROFILING_ENABLED        = "true"
+    DD_TRACE_ENABLED            = "true"
+    DD_RUNTIME_METRICS_ENABLED  = "true"
+    DD_APM_ENABLED              = "true"
+    DD_SERVICE                  = var.name
+    DD_SERVICE_NAME             = var.name
+    DD_ENV                      = var.env
+    DD_AGENT_HOST               = local.datadog_agent_host
+    OTEL_EXPORTER_OTLP_ENDPOINT = "http://${local.datadog_agent_host}:4318"
+    OTEL_TRACES_EXPORTER        = "otlp"
+    OTEL_RESOURCE_ATTRIBUTES    = "service.name=${var.name}"
   } : {}
 
+  datadog_agent_host = (var.ecs_network_mode != "host" && var.ecs_network_mode != "awsvpc") ? "datadog-agent" : "localhost"
+
   ecs_exec_env_vars = var.ecs_exec_custom_prompt_enabled ? {
     PS1 = var.ecs_exec_prompt_string
   } : {}
@@ -115,7 +121,7 @@ variable "name" {
 
 variable "app_type" {
   type        = string
-  description = "ECS application type. Valid values: web (with load balancer), worker (scheduled task without ALB)."
+  description = "ECS application type. Valid values: web (with ALB), worker (without ALB)."
   default     = "web"
 
   validation {
@@ -146,7 +152,7 @@ variable "ec2_service_group" {
 variable "instance_type" {
   type        = string
   description = "EC2 instance type for ECS"
-  default     = "t3.micro"
+  default     = "t3.small"
 }
 
 variable "environment" {
@@ -298,6 +304,12 @@ variable "docker_host_port" {
   default     = 0
 }
 
+variable "port_mappings" {
+  description = "List of ports to open from a service"
+  type = list(any)
+  default = []
+}
+
 variable "docker_container_entrypoint" {
   type        = list(string)
   description = "Docker container entrypoint"
@@ -382,6 +394,12 @@ variable "ec2_eip_count" {
   default     = 0
 }
 
+variable "ec2_eip_dns_enabled" {
+  type        = bool
+  description = "Whether to manage DNS records to be attached to the EIP"
+  default     = false
+}
+
 
 variable "ecs_cluster_name" {
   type        = string
@@ -395,6 +413,12 @@ variable "autoscaling_health_check_type" {
   default     = "EC2"
 }
 
+variable "ecs_task_health_check_command" {
+  type        = string
+  description = "Command to check for the health of the container"
+  default     = ""
+}
+
 variable "alb_health_check_path" {
   type        = string
   description = "ALB health check path"