Simple auto-scaling for Amazon Lightsail Containers written in Javascript

About

Amazon Lightsail Containers is one of the easiest way to run container based app on AWS. It automatically sets up a load balanced TLS endpoint, custom DNS, free private image registry and more. But I think it lacks one important feature which is auto scaling. That's why I build this tool.

lightsail-containers-autoscaler support both dynamic scaling and scheduled scaling. The aim of this tool is to be as simple as possible hence it is only consist of single Javascript file src/index.js. You just need to schedule this tool to run in regular interval and it will fetch metrics from the Lightsail Container and do the auto-scaling.

I recommend running this tool using Amazon EventBridge and AWS Lambda so you do not have to maintain any servers.

WARNING: This project is in early development and not many tests has been conducted, use it at your own risk.

Requirements

This tool has been tested using following:

• Node.js v16.x
• Serverless Framework 3.21 (Used to deploy to AWS)

How to Install

Clone this repo or download the archive and extract to some directory.

git clone [email protected]:rioastamal/lightsail-containers-autoscaler.git

Install all required dependencies.

cd lightsail-containers-autoscaler

npm install

If you want to deploy to your AWS account, then you need to install Serverless Framework.

npm install -g serverless

How to Deploy

Several resources will be created during deployment such as:

• Amazon DynamoDB
• Amazon EventBridge
• AWS Lambda

This deployment uses Serverless Framework so make sure to edit serverless.yml to suit your needs. Settings that you will likely to edit are functions section.

functions:
  autoscaling:
    handler: src/index.handler
    events:
      - schedule:
          rate: rate(10 minutes)
          enabled: true
          input: 
            name: lightsail-containers-autoscaler
            rules:
            - enabled: true
              dry_run: false
              scaling_type: dynamic
              nodes: 3
              power: nano
              metric: cpu
              average: 20
              average_operator: gte
              average_duration_minutes: 10
              wait_after_last_deployment_minutes: 10
            - enabled: true
              dry_run: false
              scaling_type: dynamic
              nodes: 1
              power: nano
              metric: cpu
              average: 5
              average_operator: lte
              average_duration_minutes: 15
              wait_after_last_deployment_minutes: 30
            - enabled: true
              dry_run: false
              scaling_type: scheduled
              nodes: 2
              power: micro
              run_at: "* * 21 * * *"
              wait_after_last_deployment_minutes: 30

Change schedule.rate to run the function as your required interval. See AWS Schedule Syntax for more details. The next thing you may want to change is input. It takes the same format as input.sample.json but in YAML format.

Make sure you already configure your AWS credentials via ~/.aws or via AWS_* environment variables. To deploy just run:

export AWS_DEFAULT_REGION=ap-southeast-1
export APP_CONTAINER_SVC_NAME=demo-auto

serverless deploy

It will run the Lambda function every 10 minutes and do the auto scaling based on rules matched.

How to Run (local)

To run locally, configure your AWS credentials via ~/.aws or via AWS_* environment variables. Then make sure you have permissions to write to DynamoDB table and Lightsail Container service. Take a look at IAM permissions defined in serverless.yml file.

First, define required environment variables. In this case I am using ap-southeast-1 and demo-auto as example values.

export NODE_ENV=development
export APP_REGION=ap-southeast-1
export APP_TABLE_NAME=ls-containers-autoscaling-$NODE_ENV
export APP_CONTAINER_SVC_NAME=demo-auto
export APP_NO_LAMBDA=true

You can use sample configuration input.sample.json as an input.

cat input.sample.json | node src/index.js

It will scale in/out the container service based on the rule matched.

If you want to simulate particular date then pass APP_CURRENT_DATE environment when running this tool.

cat input.sample.json | APP_CURRENT_DATE=2020-08-08T06:30:00+07:00 node src/index.js

Configuration

Explanation of each attribute on configuration that used in as an input for the function.

Supported attributes for each element on rules.

Sample 1

1. Do scale out to 4 nodes and power to micro when average CPU metric is above 70 percent for the last 15 minutes. Only do scaling if last deployment is more than 10 minutes ago.
2. Do Scale in 1 node and power to nano when average CPU metric is below 5 percent for the last 45 minutes. Only do scaling if last deployment is more than 30 minutes ago.

{
  "name": "lightsail-containers-autoscaler",
  "rules": [
    {
      "enabled": true,
      "dry_run": false,
      "scaling_type": "dynamic",
      "nodes": 4,
      "power": "micro",
      "metric": "cpu",
      "average": 70.0,
      "average_operator": "gte",
      "average_duration_minutes": 15,
      "wait_after_last_deployment_minutes": 10
    },
    {
      "enabled": true,
      "dry_run": false,
      "scaling_type": "dynamic",
      "nodes": 1,
      "power": "nano",
      "metric": "cpu",
      "average": 5.0,
      "average_operator": "lte",
      "average_duration_minutes": 15,
      "wait_after_last_deployment_minutes": 30
    }
  ]
}

Sample 2

1. Do scale out to 6 nodes and power to micro every Monday to Friday at 8am to 4pm. Only do scaling if last deployment is more than 10 minutes ago.
2. Do scale in to 1 node and power to nano every Monday to Friday at 5pm to 7am. Only do scaling if last deployment is more than 30 minutes ago.
3. Do scale in to 1 node and power to nano every weekend. Only do scaling if last deployment is more than 30 minutes ago.

{
  "name": "lightsail-containers-autoscaler",
  "rules": [
    {
      "enabled": true,
      "dry_run": false,
      "scaling_type": "scheduled",
      "nodes": 6,
      "power": "micro",
      "run_at": "* 0-59 8-16 * * 1-5",
      "wait_after_last_deployment_minutes": 10
    },
    {
      "enabled": true,
      "dry_run": false,
      "scaling_type": "scheduled",
      "nodes": 1,
      "power": "nano",
      "run_at": "0-59 0-7,17-23 * * 1-5",
      "wait_after_last_deployment_minutes": 30
    },
    {
      "enabled": true,
      "dry_run": false,
      "scaling_type": "scheduled",
      "nodes": 1,
      "power": "nano",
      "run_at": "* * * * 0,6",
      "wait_after_last_deployment_minutes": 30
    },
  ]
}

To do

• Add unit tests
• Improve documentation

Contributing

Fork this repo and send me a Pull Request (PR).

License

This project is licensed under MIT License. See LICENSE file.