Github Runner AutoScaling

This is an archive of our previous autoscaling infrastructure. This repository will not receive further updates.

Architecture Diagram

Project Overview

This repository handles auto-scaling of our VM runners for dgraph github projects. We want this mostly to reduce the maintenance pain points & cost. The project has 2 pieces primarily:

• Infrastructure
• Source Code

Infrastructure

We use Packer to create an AWS machine image (AMI) for our EC2 self-hosted runners. This image contains all the dependencies init-dependencies.sh that we need to run a Github actions job. It will also register itself as a runner, which is done by init-runner.sh which runs when the instance comes up. In summary, every instance that is created from this AMI is ready to run a Github actions job without any further configurations. The configurations for packer build step is maintained in the respective *.pkr.hcl files for different architectures.

How to build the AMI?

1. Install Packer on your machine
2. Authenticate to AWS
3. Build the AMI

   packer build . 
   

4. Visit the AWS AMI page to verify that Packer successfully built your AMI.

Once the AMI is built - it's important to associate them with the Launch Template for use.

Source Code

The Source Code is our Orchestrator. This has 2 modes of operations manual & auto. The code is written in python and it communicates with AWS (EC2, SSM) & GitHub.

The idea is as follows:

• during scale-ups - the program will obtain a token from GitHub and save it on SSM using the instance-id as the unique deliminator. When the instance bootsup, it reads this token & self-registers with GitHub repository.
• during scale-downs - the program will deregister the runner, and delete the instance on AWS.

Regarding the modes:

• manual, we can manually add or remove instances
• auto, the program will run indefinitely and will manage auto-scaling for multiple repositories based on what is defined inside of config.yaml

Operations

Setup

• Install Python3+
• Install Requirements (you can do it with PyEnv or Venv)

  cd src
  pip install -r requirements.txt
  

• Edit the src/config.template.yaml by replacing the templated values to create src/config.yaml

Manual

• scale-up

  export GH_OWNER="XXXXX"
  export GH_REPOSITORY="XXXXX"
  export GH_TOKEN="XXXXX"
  export AWS_ACCESS_KEY_ID="XXXXX"
  export AWS_SECRET_ACCESS_KEY="XXXXX"
  cd src
  python scale.py up -a arm64 # for arm64
  python scale.py up -a amd64 # for amd64
  

• scale-down

  export GH_OWNER="XXXXX"
  export GH_REPOSITORY="XXXXX"
  export GH_TOKEN="XXXXX"
  export AWS_ACCESS_KEY_ID="XXXXX"
  export AWS_SECRET_ACCESS_KEY="XXXXX"
  cd src
  python scale.py down -r i-XXXXX
  

Automatic

• automatic

  # make sure to add your repository to `config.yaml`
  export GH_TOKEN="XXXXX"
  export AWS_ACCESS_KEY_ID="XXXXX"
  export AWS_SECRET_ACCESS_KEY="XXXXX"
  cd src
  python scale.py auto