actions-runner-controller/test/platforms/aws-eks

Context

Terraform templates to quickly create an EKS cluster with a managed node group. This is not a reference setup! It's a vanilla setup to be used when attempting to replicate issues and/or to test new features.

⚠️ Do not use this setup in production.

Pre-requisites

• Terraform v1.3+ installed locally.
• an AWS account
• the AWS CLI v2.7.0/v1.24.0 or newer, installed and configured
• AWS IAM Authenticator
• kubectl v1.24.0 or newer

Download & Authenticate

brew install awscli aws-iam-authenticator terraform

Configure & authenticate AWS CLI. This will vary based on your AWS account and IAM setup

Setup

# Export AWS region & profile env variables
export AWS_REGION="eu-west-2"           # Replace with your region
export AWS_PROFILE="actions-compute"    # Replace with your profile

# You're free to use terraform cloud but you need to update main.tf first
terraform init

# Run terraform plan
terraform plan

# Verify the plan output from the previous step
# Run terraform apply
terraform apply

# Retrieve access credentials for the cluster and configure kubectl
aws eks --region "${AWS_REGION}" update-kubeconfig \
    --name "$(terraform output -raw cluster_name)" \
    --profile "${AWS_PROFILE}"

# If you get this error: 'NoneType' object is not iterable
# Remove the ~/.kube/config file and try again
# https://github.com/aws/aws-cli/issues/4843

# Verify your installation
kubectl cluster-info

Setup ARC by following this quick-start guide.

Troubleshooting

dial tcp: lookup api.github.com: i/o timeout

If you see this error in the controller pod logs:

ERROR   AutoscalingRunnerSet    Failed to initialize Actions service client for creating a new runner scale set {"autoscalingrunnerset": "arc-runners/arc-runner-set", "error": "failed to get runner registration token: Post \"https://api.github.com/app/installations/33454774/access_tokens\": POST https://api.github.com/app/installations/33454774/access_tokens giving up after 5 attempt(s): Post \"https://api.github.com/app/installations/33454774/access_tokens\": dial tcp: lookup api.github.com: i/o timeout"}

This is because the controller pod is not able to resolve the api.github.com domain name. This is a good guide for troubleshooting DNS failures in EKS. For a fresh setup this is most likely a security group configuration problem.

The controller could have allocated to a node that cannot reach coredns. You need to allow the DNS (TCP / UDP) traffic to flow between the worker nodes' security groups.