Skip to content

This is a custom GitHub action to provision and manage self-hosted runners using AWS EC2 On-Demand and/or Spot instances.

License

Notifications You must be signed in to change notification settings

NextChapterSoftware/ec2-action-builder

Use this GitHub action with your project
Add this Action to an existing workflow or create a new one
View on Marketplace

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

31 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

EC2 Github Action Builder

TL;DR

Jump to examples

Overview

This is a custom GitHub action to provision and manage self-hosted runners using AWS EC2 On-Demand and/or Spot instances.

It offers multiple spot instance provisioning modes:

  • None: (default) Strictly On-Demand instances only
  • SpotOnly: Strictly Spot instances only
  • BestEffort: Use a Spot instance of same class and size when price is <= On-Demand
    • (Automatic fallback to On-Demand)
  • MaxPerformance: Use the largest spot instance in the same class for <= the On-Demand price
    • (Automatic fallback to On-Demand)

Supported operating system AMIs:

  • Amazon Linux
  • Ubuntu
  • Debian

Why?

Cost Savings

Operating system vCPUs Per-minute rate (USD)

OS     vCPU   GH Price/Minute      EC2 Price/Minute
Linux	2      $0.008             $0.001284 (c5a.large)
Linux	4      $0.016             $0.00257  (c5a.xlarge)
Linux	8      $0.032             $0.00514  (c5a.2xlarge)
Linux	16     $0.064             $0.0114   (c5.4xlarge)
Linux	32     $0.128             $0.02054  (c5a.8xlarge)
Linux	64     $0.256             $0.041067 (c5a.16xlarge)

Sources:

Customizable Machine Image

  • Set custom EC2 root volume size
  • Set custom storage class for root volume
  • Users can provide their own custom AMI image pre-loaded with all the necessary tooling of their choice saving time and cost.

Enhance Security

  • EC2 instances run within your infrastructure
  • Easier to harden runner instances using custom AMIs, Security Groups etc
  • Easier monitoring and vulnerability scanning using existing tools (e.g CloudWatch, GuardDuty, AWS Inspector etc)
  • Secure networking setup by eliminating any need to expose ports to external service or using Bastion hosts!
  • Lower data transfer costs (e.g ECR images, S3 objects etc)

Setup

1. Create GitHub Personal Access Token

  1. Create a fine-grained personal access token
  2. Edit the token permissions and select Only select repositories for Repository access
  3. Select any repositories you wish to use with this action
  4. Grant Read and Write access for Administration access level under Repository permissions
  5. Add the token to GitHub Action secrets and note the secret name

2. Setup GitHub Secrets for IAM credentials

2a. Use IAM keys

  1. Add your IAM Access Key ID and Secret Access Key to GitHub Secrets and note the secret names!
  2. Modify ${{ secrets.DEPLOY_AWS_ACCESS_KEY_ID }} and ${{ secrets.DEPLOY_AWS_SECRET_ACCESS_KEY }} in examples below to match the names of your GH secrets

2b. Use OIDC

  1. Configure your EC2 backend to allow a federated connection from github
  2. use configure-aws-credentials or similar to authenticate against your backend using OIDC. See the OIDC example for guidance on how this is done. The documentation in the configure-aws-credendials is very detailed.

Note: For information about required IAM permissions check IAM role policy here

3. Collect EC2 information:

  • AWS Region (e.g us-west-2)
  • EC2 AMI ID for your desired instance type in the region (Ubuntu AMI Locator)
    • Important Note: Only Ubuntu, Amazon Linux and Debian AMIs have been tested
    • To find AMIs for other operating systems follow instructions here
  • EC2 Subnet ID for any subnet with internet access (Can be private with NAT)
  • EC2 Security Group ID for any security group which allows all outbound traffic (Default configuration for an empty Security Group)

Note: The security group does not require any in-bound rules. You can add in-bound rules based on your needs (e.g open SSH port 22)

Examples

Standard

  • Modify ec2_spot_instance_strategy for other deployment strategies. List of all values can be found here
  • Modify github_token value to match the name for your Personal Access Token secret name
jobs:
    start-runner:
        timeout-minutes: 5              # normally it only takes 1-2 minutes
        name: Start self-hosted EC2 runner   
        runs-on: ubuntu-latest
        permissions:
          actions: write        
        steps:      
          - name: Start EC2 runner
            id: start-ec2-runner
            uses: NextChapterSoftware/ec2-action-builder@v1.10
            with:
              github_token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
              aws_access_key_id: ${{ secrets.DEPLOY_AWS_ACCESS_KEY_ID }}
              aws_secret_access_key: ${{ secrets.DEPLOY_AWS_SECRET_ACCESS_KEY }}
              aws_region: "us-west-2"
              ec2_instance_type: c5.4xlarge
              ec2_ami_id: ami-008fe2fc65df48dac
              ec2_iam_instance_profile: AWSInstanceProfile
              ec2_subnet_id: "SUBNET_ID_REDACTED"
              ec2_security_group_id: "SECURITY_GROUP_ID_REDACTED"
              ec2_instance_ttl: 40                # Optional (default is 60 minutes)
              ec2_spot_instance_strategy: None    # Other options are: SpotOnly, BestEffort, MaxPerformance 

    # Job that runs on the self-hosted runner 
    run-build:
        timeout-minutes: 1
        needs:
          - start-runner
        runs-on: ${{ github.run_id }}          
        steps:              
          - run: env

Advanced

  • IAM policy and role setup instructions can be found here
  • Modify ec2_spot_instance_strategy for other deployment strategies. List of all values can be found here

IMPORTANT NOTE

An error occured: Runner version vX.YZ is deprecated and cannot receive messages.

Error message above is usually caused by --disableupdate custom configuration argument used with a deprecated Runner version. Make sure to use a runner that has not been deprecated or omit github_action_runner_version to use the latest available version.

jobs:
    start-runner:
        timeout-minutes: 5                  # normally it only takes 1-2 minutes
        name: Start self-hosted EC2 runner   
        runs-on: ubuntu-latest
        permissions:
          actions: write        
        steps:      
          - name: Start EC2 runner
            id: start-ec2-runner
            uses: NextChapterSoftware/ec2-action-builder@v1.10
            with:
              aws_access_key_id: ${{ secrets.DEPLOY_AWS_ACCESS_KEY_ID }}
              aws_secret_access_key: ${{ secrets.DEPLOY_AWS_SECRET_ACCESS_KEY }}
              aws_iam_role_arn: "arn:aws:iam::REDACTED:role/REDACTED"
              aws_region: "us-west-2"
              github_token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
              github_action_runner_version: v2.300.2                  # Optional (default is latest release)
              github_action_runner_extra_cli_args: "--disableupdate"  # Extra cli args for runner startup command
              github_job_start_ttl_seconds: 60                        # Optional - (defaults to 0 disabling this feature)
              github_api_retry_delay: 10                              # Optional - Delay when polling for runner registration (default is 10 seconds)
              ec2_instance_type: c5.4xlarge
              ec2_ami_id: ami-008fe2fc65df48dac
              ec2_root_disk_size_gb: "100"                 # Optional - (defaults to AMI settings)
              ec2_root_disk_ebs_class: "gp2"               # Optional - Only used with custom volume root size (defaults to gp2)
              ec2_subnet_id: "SUBNET_ID_REDACTED"
              ec2_security_group_id: "SECURITY_GROUP_ID_REDACTED"
              ec2_instance_ttl: 40                          # Optional - (default is 60 minutes)
              ec2_spot_instance_strategy: MaxPerformance    # Other options are: None, BestEffort, MaxPerformance 
              ec2_instance_tags: >                          # Required for IAM role resource permission scoping
                [
                  {"Key": "Owner", "Value": "deploybot"}
                ]

    # Job that runs on the self-hosted runner 
    run-build:
        timeout-minutes: 1
        needs:
          - start-runner
        runs-on: ${{ github.run_id }}          
        steps:              
          - run: env

Use OIDC

  • IAM policy and role setup instructions can be found here
  • Modify ec2_spot_instance_strategy for other deployment strategies. List of all values can be found here
jobs:
  start-runner:
    timeout-minutes: 5              # normally it only takes 1-2 minutes
    name: Start self-hosted EC2 runner   
    runs-on: ubuntu-latest
    permissions:
      actions: write        
      contents: read
      id-token: write
    steps:      
      - name: Configure AWS credentials
        id: creds                                  # name of step, to allow access to outputs 
        uses: aws-actions/configure-aws-credentials@v4
        with:
          aws-region: "AWS_REGION"
          role-to-assume: "arn:aws:iam::REDACTED:role/REDACTED"
          output-credentials: true                 # output the credentials
      - name: Start EC2 runner
        id: start-ec2-runner
        uses: NextChapterSoftware/ec2-action-builder@v1.10
        with:
          github_token: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
          aws_access_key_id: ${{ steps.creds.outputs.aws-access-key-id }}         # generated by configure-aws-credentials
          aws_secret_access_key: ${{ steps.creds.outputs.aws-secret-access-key }} # generated by configure-aws-credentials
          aws_session_token: ${{ steps.creds.outputs.aws-session-token }}         # generated by configure-aws-credentials
          aws_region: "AWS_REGION"
          ec2_subnet_id: "SUBNET_ID_REDACTED"
          ec2_security_group_id: "SECURITY_GROUP_ID_REDACTED"
          ec2_instance_type: t4g.large
          ec2_ami_id: ami-0c29a2c5cf69b5a9c
          ec2_instance_ttl: 40                      # Optional (default is 60 minutes)
          ec2_spot_instance_strategy: BestEffort    # Other options are: None, SpotOnly, BestEffort, MaxPerformance 
          ec2_instance_tags: >                      # Required for IAM role resource permission scoping
            [
              {"Key": "Owner", "Value": "deploybot"}
            ]

    # Job that runs on the self-hosted runner 
    run-build:
        timeout-minutes: 1
        needs:
          - start-runner
        runs-on: ${{ github.run_id }}          
        steps:              
          - run: env

How it all works under the hood

General instance launch flow

  • Your GitHub personal token is used to obtain a Runner Registration token
  • If no explicit runner version has been provided, it will retrieve the latest version number
  • It then uses all the provided info to compile an EC2 user-data script which does the following:
    • Set a max TTL on the EC2 instance on startup
    • Create a shutdown script which is executed when jobs end
    • Downloads GitHub Action Runner bundle
    • Unpack Action Runner bundle
    • Configure Runner agent as an ephemeral agent
  • EC2 instance is launched with the user-data script from previous step
  • Once EC2 boot has completed, user-data script is executed
  • Runner binary registers itself with GitHub API using the current job ID
  • Once the Runner is registered, control is transferred to the next job (this is your build job)
  • Upon a job completion (failure/success), Shutdown script is triggered to kill the instance with a 1 minute delay

Spot instance provisioning

  • Script looks up On-Demand price for the supplied instance type
  • It will then look up EC2 Spot instance prices using AWS API
  • Depending on the mode
    • SpotOnly: It will try to launch a spot instance with On-Demand price as the max price cut-off
    • BestEffort: It will try to launch a spot instance but falls back to On-Demand if prices are too high!
    • MaxPerformance: It will try to get the largest spot instance in class for the On-Demand price of the supplied instance type. It falls back to On-Demand if prices are too high!

Other EC2 Considerations

  • Each instance is named as "{repo}-{jobID}"
  • Default EC2 TTL is 60 minutes
  • Other EC2 tags are github_job_id and github_ref
  • Spot instances might be taken away by AWS without any prior notice