ESP - ECS Stats Plotter

Description

ESP Retrives AWS ECS Container Stats from Metadata Endpoint and Plots Stats to Cloudwatch Detailed Metrics. ESP enables faster autoscaling of AWS ECS Tasks by indexing AWS Cloudwatch high-resolution Custom Metrics.

Requirement

Add cloudwatch:PutMetricData Permission to Task Role.

add cloudwatch:PutMetricData to task role.

  statement {
    effect = "Allow"
    actions = [
      "cloudwatch:PutMetricData"
    ]
    resources = ["*"]
  }

Usage

Build and Push to Repository

Build docker image and push to ECR repository.

docker build --platform linux/amd64 -t esp .
docker push *******.dkr.ecr.us-east-1.amazonaws.com/esp:latest

Add ESP as a sidecar container

Add ESP as a ECS sidecar container in the container definition below.

[
  {
    "name": "web",
    ...<snip>...
  {
    "name": "esp",
    "image": "********.dkr.ecr.us-east-1.amazonaws.com/esp:latest",
    "essential": true,
    "environment": [
      {
        "name": "CONTAINER_NAME",
        "value": "web"
      },
      {
        "name": "REGION",
        "value": "us-east-1"
      },
      {
        "name": "NAMESPACE",
        "value": "FOO"
      }
    ], 
    "dependsOn": [
      {
        "containerName": "web",
        "condition": "START"
      }
    ]
  }
]

Environment Name	Description
CONTAINER_NAME	Container name for which you want to measure impossibility
REGION	AWS Region Name
NAMESPACE	Namespace name for Cloudwatch high resolution custom metrics

Set Autoscalling

Configure Cloudwatch Metric Alarm for Autoscalling. Below is an example of Terraform Code.

resource "aws_cloudwatch_metric_alarm" "foo" {
  alarm_name          = "foo-CPU-Utilization-High-30"
  comparison_operator = "GreaterThanOrEqualToThreshold"
  evaluation_periods  = "1"
  metric_name         = "CPUUtilization"
  namespace           = "FOO"
  period              = "60"
  statistic           = "Average"
  threshold           = "15"
  dimensions = {
    ClusterName = aws_ecs_cluster.foo.name
    ServiceName = aws_ecs_service.foo.name
  }
  alarm_actions = [aws_appautoscaling_policy.scale_out.arn]
}

resource "aws_cloudwatch_metric_alarm" "foo" {
  alarm_name          = "foo-CPU-Utilization-Low-5"
  comparison_operator = "LessThanThreshold"
  evaluation_periods  = "1"
  metric_name         = "CPUUtilization"
  namespace           = "FOO"
  period              = "60"
  statistic           = "Average"
  threshold           = "5"
  dimensions = {
    ClusterName = aws_ecs_cluster.foo.name
    ServiceName = aws_ecs_service.foo.name
  }
  alarm_actions = [aws_appautoscaling_policy.scale_in.arn]
}

Author

jedipunkz

License

The source code is licensed MIT. The website content is licensed CC BY 4.0,see LICENSE.