Adding mmpose

dev/main.tf

@@ -2,9 +2,9 @@ module "processing" {
     region = "us-west-2"
     num_machines = 0
     opencap_ecr_repository = "660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/opencap-dev"
-    openpose_ecr_repository = "660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/openpose"
+    openpose_ecr_repository = "660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/openpose-dev"
     opencap_api_ecr_repository = "660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/api-dev"
-    mmpose_ecr_repository = "660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/mmpose"
+    mmpose_ecr_repository = "660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap/mmpose-dev"
     opencap_analysis_max_centerofmass_vpos_ecr_repository = "660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap-analysis/max_centerofmass_vpos-dev"
     opencap_gait_analysis_ecr_repository = "660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap-analysis/gait_analysis-dev"
     opencap_treadmill_gait_analysis_ecr_repository = "660440363484.dkr.ecr.us-west-2.amazonaws.com/opencap-analysis/treadmill_gait_analysis-dev"
@@ -21,15 +21,13 @@ module "processing" {
 
     processing_asg_scaling_config = {
         min_size = 0
-        max_size = 2
+        max_size = 0
         desired_size = 0
     }
 
     processing_asg_scaling_target  = 5
     processing_asg_trials_baseline = 3
 
-    processing_asg_use_launch_config = false
-
     # processing_asg_instance_type = "g5.2xlarge"
     # processing_ecs_task_memory = 30146
     processing_asg_instance_type = "g5.xlarge"

modules/processing/autoscaling.tf

@@ -27,15 +27,14 @@ data "aws_ami" "latest_ecs" {
 locals {
     lt_user_data_raw = <<-EOF
     #!/bin/bash
+    # ECS cluster settings
     echo 'ECS_CLUSTER=${aws_ecs_cluster.ecs_cluster.name}' >> /etc/ecs/ecs.config
     echo 'ECS_ENABLE_CONTAINER_METADATA=true' >> /etc/ecs/ecs.config
     echo 'ECS_RESERVED_MEMORY=256' >> /etc/ecs/ecs.config
     EOF
 }
 
 resource "aws_launch_template" "ecs_worker_launch_template" {
-    count = !var.processing_asg_use_launch_config ? 1 : 0
-
     name_prefix               = "${var.app_name}-processing-worker${var.env}"
     image_id                  = data.aws_ami.latest_ecs.image_id
     iam_instance_profile {
@@ -65,12 +64,10 @@ resource "aws_launch_template" "ecs_worker_launch_template" {
 }
 
 resource "aws_autoscaling_group" "worker_lt_asg" {
-    count = !var.processing_asg_use_launch_config ? 1 : 0
-
     name                      = "${var.app_name}-processing-worker-asg${var.env}"
     vpc_zone_identifier       = [values(aws_subnet.pub_subnet)[0].id]
     launch_template {
-        id      = aws_launch_template.ecs_worker_launch_template[0].id
+        id      = aws_launch_template.ecs_worker_launch_template.id
         version = "$Latest"
     }
 
@@ -103,7 +100,7 @@ resource "aws_ecs_capacity_provider" "worker_lt_gpu_provider" {
     name = "${var.app_name}-processing-worker-gpu-capacity${var.env}"
 
     auto_scaling_group_provider {
-        auto_scaling_group_arn         = var.processing_asg_use_launch_config ? aws_autoscaling_group.opencap_processing_asg[0].arn : aws_autoscaling_group.worker_lt_asg[0].arn
+        auto_scaling_group_arn         = aws_autoscaling_group.worker_lt_asg.arn
         managed_termination_protection = "DISABLED"
 
         managed_scaling {
@@ -196,35 +193,3 @@ resource "aws_appautoscaling_policy" "target_tracking" {
     }
   }
 }
-
-### Launch Configuration | Deprecated ###
-resource "aws_launch_configuration" "ecs_launch_config" {
-    count = var.processing_asg_use_launch_config ? 1 : 0
-
-    name                 = "${var.app_name}-processing-workers${var.env}"
-    image_id             = data.aws_ami.latest_ecs.image_id
-    iam_instance_profile = aws_iam_instance_profile.ecs_agent.name
-    security_groups      = [aws_security_group.ecs_sg.id]
-    user_data            = "#!/bin/bash\necho ECS_CLUSTER=${var.app_name}-processing-cluster >> /etc/ecs/ecs.config"
-    instance_type        = var.processing_asg_instance_type
-    key_name             = aws_key_pair.debug.key_name
-    associate_public_ip_address = true
-
-    root_block_device {
-        volume_size = 128
-    }
-}
-
-resource "aws_autoscaling_group" "opencap_processing_asg" {
-    count = var.processing_asg_use_launch_config ? 1 : 0
-
-    name                      = "asg${var.env}"
-    vpc_zone_identifier       = [values(aws_subnet.pub_subnet)[0].id]
-    launch_configuration      = aws_launch_configuration.ecs_launch_config[0].name
-
-    desired_capacity          = var.num_machines
-    min_size                  = 0
-    max_size                  = var.num_machines
-    health_check_grace_period = 300
-    health_check_type         = "EC2"
-}

modules/processing/ecs.tf

@@ -53,12 +53,17 @@ resource "aws_cloudwatch_log_group" "openpose-logs" {
   retention_in_days = 90
 }
 
+resource "aws_cloudwatch_log_group" "mmpose-logs" {
+  name              = "/ecs/${var.app_name}-mmpose${var.env}"
+  retention_in_days = 90
+}
+
 resource "aws_ecs_task_definition" "task_definition" {
   family                = "worker${var.env}"
   container_definitions = data.template_file.task_definition_template.rendered
   execution_role_arn    = aws_iam_role.ecs_tasks_execution_role.arn
   task_role_arn         = aws_iam_role.processing_worker_role.arn
-
+  
   memory                = var.processing_ecs_task_memory
   volume {
     name = "data${var.env}"

modules/processing/task_definition.json.tpl

@@ -20,10 +20,6 @@
         {
           "name": "DOCKERCOMPOSE",
 	        "value": "1"
-        },
-        {
-          "name": "ECS_ENABLE_CONTAINER_METADATA",
-          "value": "true"
         }
       ],
       "secrets": [
@@ -36,7 +32,12 @@
           "valueFrom": "${API_URL}"
         }
       ],
-      "resourceRequirements": null,
+      "resourceRequirements" : [
+          {
+             "type" : "GPU",
+             "value" : "1"
+          }
+      ],
       "ulimits": null,
       "dnsServers": [],
       "mountPoints": [
@@ -89,12 +90,11 @@
       "command": [],
       "linuxParameters": null,
       "cpu": 0,
-      "environment": [],
-      "resourceRequirements" : [
-          {
-             "type" : "GPU", 
-             "value" : "1"
-          }
+      "environment": [
+        {
+          "name": "NVIDIA_VISIBLE_DEVICES",
+          "value": "0"
+        }
       ],
       "ulimits": null,
       "dnsServers": [],
@@ -130,5 +130,64 @@
       "systemControls": [],
       "privileged": null,
       "name": "openpose"
+    },
+    {
+      "dnsSearchDomains": [],
+      "environmentFiles": null,
+      "logConfiguration": {
+        "logDriver": "awslogs",
+        "secretOptions": null,
+        "options": {
+          "awslogs-group": "/ecs/${APP_NAME}-mmpose${ENV}",
+          "awslogs-region": "${REGION}",
+          "awslogs-stream-prefix": "ecs"
+        }
+      },
+      "entryPoint": [],
+      "portMappings": [
+      ],
+      "command": [],
+      "linuxParameters": null,
+      "cpu": 0,
+      "environment": [
+        {
+          "name": "NVIDIA_VISIBLE_DEVICES",
+          "value": "0"
+        }
+      ],
+      "ulimits": null,
+      "dnsServers": [],
+      "mountPoints": [
+        {
+          "readOnly": false,
+          "containerPath": "/mmpose/data",
+          "sourceVolume": "data${ENV}"
+        }
+      ],
+      "workingDirectory": null,
+      "secrets": null,
+      "dockerSecurityOptions": [],
+      "memory": null,
+      "memoryReservation": null,
+      "volumesFrom": [],
+      "stopTimeout": null,
+      "image": "${MMPOSE}",
+      "startTimeout": null,
+      "firelensConfiguration": null,
+      "dependsOn": null,
+      "disableNetworking": null,
+      "interactive": null,
+      "healthCheck": null,
+      "essential": true,
+      "links": [],
+      "hostname": null,
+      "extraHosts": null,
+      "pseudoTerminal": null,
+      "user": null,
+      "readonlyRootFilesystem": null,
+      "dockerLabels": {},
+      "systemControls": [],
+      "privileged": null,
+      "name": "mmpose"
     }
   ]

modules/processing/template_file.tf

@@ -5,6 +5,7 @@ data "template_file" "task_definition_template" {
         ENV = "${var.env}"
         APP_NAME = "${var.app_name}"
         OPENPOSE = "${var.openpose_ecr_repository}"
+        MMPOSE = "${var.mmpose_ecr_repository}"
         OPENCAP = "${var.opencap_ecr_repository}"
         API_TOKEN = var.env == "-dev" ? "arn:aws:secretsmanager:us-west-2:660440363484:secret:OpenCapProcessingCredentials-oXYoTR:api_token_dev::" : "arn:aws:secretsmanager:us-west-2:660440363484:secret:OpenCapProcessingCredentials-oXYoTR:api_token::"
         API_URL = var.env == "-dev" ? "arn:aws:secretsmanager:us-west-2:660440363484:secret:OpenCapProcessingCredentials-oXYoTR:api_url_dev::" : "arn:aws:secretsmanager:us-west-2:660440363484:secret:OpenCapProcessingCredentials-oXYoTR:api_url::"

modules/processing/variables.tf

@@ -40,15 +40,6 @@ variable "num_machines" {
   default     = 1
 }
 
-variable "processing_asg_use_launch_config" {
-  default = true
-  description =<<-EOF
-  true: Use launch config for processing autoscaling group, false: use launch template
-  This is for providing a smooth transition from launch config to launch template as LCs are deprecated
-  the goal is to delete this variable along with LC code when all services are using launch templates
-  EOF
-}
-
 variable "processing_asg_scaling_config" {
   default = {
     min_size = 0