AWS CDK GitLab Runner autoscaling on EC2 instances using docker+machine executor.
npm install @pepperize/cdk-autoscaling-gitlab-runner
This project provides a CDK construct to execute jobs on auto-scaled EC2 instances using the Docker Machine executor.
> Running out of Runner minutes,
> using Docker-in-Docker (dind),
> speed up jobs with shared S3 Cache,
> cross compiling/building environment multiarch,
> cost effective autoscaling on EC2,
> deploy directly from AWS accounts (without AWS Access Key),
> running on Spot instances,
> having a bigger build log size
``shell`
npm install @pepperize/cdk-autoscaling-gitlab-runner
or
`shell`
yarn add @pepperize/cdk-autoscaling-gitlab-runner
`shell`
pip install pepperize.cdk-autoscaling-gitlab-runner
`
dotnet add package Pepperize.CDK.AutoscalingGitlabRunner
`xml`
1. Create a new AWS CDK App in TypeScript with projen
`shell`
mkdir gitlab-runner
cd gitlab-runner
git init
npx projen new awscdk-app-ts
2. Configure your project in .projenrc.js
- Add deps: ["@pepperize/cdk-autoscaling-gitlab-runner"],
3. Update project files and install dependencies
`shell`
npx projen
4. Register a new runner
- For a shared runner, go to the GitLab Admin Area and click Overview > Runners
- For a group runner, go to Settings > CI/CD and expand the Runners section
- For a project runner, go to Settings > CI/CD and expand the Runners section
_Optionally enable: Run untagged jobs [x]
Indicates whether this runner can pick jobs without tags_
See also _Registration token vs. Authentication token_
5. Retrieve a new runner authentication token
`shell`
curl --request POST "https://gitlab.com/api/v4/runners" --form "token=
6. Store runner authentication token in SSM ParameterStore
`shell`
aws ssm put-parameter --name "/gitlab-runner/token" --value "
7. Add to your main.ts
`typescript
import { Vpc } from "@aws-cdk/aws-ec2";
import { App, Stack } from "@aws-cdk/core";
import { GitlabRunnerAutoscaling } from "@pepperize/cdk-autoscaling-gitlab-runner";
const app = new App();
const stack = new Stack(app, "GitLabRunnerStack");
const vpc = Vpc.fromLookup(app, "ExistingVpc", {
vpcId: "
});
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(stack, "GitlabRunner", {
network: {
vpc: vpc,
},
runners: [
{
token: token,
configuration: {
// optionally configure your runner
},
},
],
});
`
8. Create service linked role
_(If requesting spot instances, default: true)_
`sh`
aws iam create-service-linked-role --aws-service-name spot.amazonaws.com
9. Configure the AWS CLI
- AWSume
- Configuring the AWS CLI
- AWS Single Sign-On
10. Deploy the GitLab Runner
`shell`
npm run deploy
By default, an AWS S3 Bucket is created as GitLab Runner's distributed cache.
It's encrypted and public access is blocked.
A custom S3 Bucket can be configured:
`typescript
const cache = new Bucket(this, "Cache", {
// Your custom bucket
});
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
token: token,
},
],
cache: { bucket: cache },
});
`
See example,
GitlabRunnerAutoscalingCacheProps
By default, the amazonec2 driver will create an EC2 key pair for each runner. To use custom ssh credentials provide a SecretsManager Secret with the private and public key file:
1. Create a key pair, download the private key file and remember the created key pair name
2. Generate the public key file
`
ssh-keygen -f
3. Create an AWS SecretsManager Secret from the key pair
shell`
aws secretsmanager create-secret --name
4. Configure the job runner
`typescript
const keyPair = Secret.fromSecretNameV2(stack, "Secret", "CustomEC2KeyPair");
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
keyPair: keyPair,
configuration: {
machine: {
machineOptions: {
keypairName: "
},
},
},
},
],
cache: { bucket: cache },
});
`
By default, docker machine is configured to run privileged with CAP_SYS_ADMIN to support Docker-in-Docker using the OverlayFS driver
and cross compiling/building with multiarch.
See runners.docker section
in Advanced configuration
`typescript
import { GitlabRunnerAutoscaling } from "@pepperize/cdk-autoscaling-gitlab-runner";
import { StringParameter } from "aws-cdk-lib/aws-ssm";
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
token: token,
configuration: {
environment: [], // Reset the OverlayFS driver for every project
docker: {
capAdd: [], // Remove the CAP_SYS_ADMIN
privileged: false, // Run unprivileged
},
machine: {
idleCount: 2, // Number of idle machine
idleTime: 3000, // Waiting time in idle state
maxBuilds: 1, // Max builds before instance is removed
},
},
},
],
});
`
See example,
DockerConfiguration
By default, t3.nano is used for the manager/coordinator and t3.micro instances will be spawned.
For bigger projects, for example with webpack, this won't be enough memory.
`typescript
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
manager: {
instanceType: InstanceType.of(InstanceClass.T3, InstanceSize.SMALL),
},
runners: [
{
instanceType: InstanceType.of(InstanceClass.T3, InstanceSize.LARGE),
token: token,
configuration: {
// optionally configure your runner
},
},
],
});
`
> You may have to disable or configure Spot instances
See example,
GitlabRunnerAutoscalingManagerProps,
GitlabRunnerAutoscalingJobRunnerProps
By default, the latest Amazon 2 Linux will be used for the manager/coordinator.
The manager/coordinator instance's cloud init scripts requires yum is installed, any RHEL flavor should work.
The requested runner instances by default using Ubuntu 20.04, any OS implemented by the Docker Machine provisioner should work.
`typescript
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
manager: {
machineImage: MachineImage.genericLinux(managerAmiMap),
},
runners: [
{
machineImage: MachineImage.genericLinux(runnerAmiMap),
token: token,
configuration: {
// optionally configure your runner
},
},
],
});
`
See example,
GitlabRunnerAutoscalingManagerProps,
GitlabRunnerAutoscalingJobRunnerProps
Each runner defines one [[runners]] section in the configuration file.
Use Specific runners when you want to use runners for specific projects.
`typescript
const privilegedRole = new Role(this, "PrivilegedRunnersRole", {
// role 1
});
const restrictedRole = new Role(this, "RestrictedRunnersRole", {
// role 2
});
const token1 = StringParameter.fromStringParameterAttributes(stack, "Token1", {
parameterName: "/gitlab-runner/token1",
});
const token2 = StringParameter.fromStringParameterAttributes(stack, "Token2", {
parameterName: "/gitlab-runner/token2",
});
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
token: token1,
configuration: {
name: "privileged-runner",
},
role: privilegedRole,
},
{
token: token2,
configuration: {
name: "restricted-runner",
docker: {
privileged: false, // Run unprivileged
},
},
role: restrictedRole,
},
],
});
`
See example,
GitlabRunnerAutoscalingProps
By default, EC2 Spot Instances are requested.
`typescript
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
token: token,
configuration: {
machine: {
machineOptions: {
requestSpotInstance: false,
spotPrice: 0.5,
},
},
},
},
],
});
`
See example,
EC2 spot price,
MachineConfiguration,
MachineOptions,
Advanced configuration - runners.machine.autoscaling
To build binaries of different architectures can also use Multiarch
`typescript
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
token: token,
configuration: {
docker: {
privileged: true,
},
},
},
],
});
`
Configure your .gitlab-ci.yml file
`yaml`
build:
image: multiarch/debian-debootstrap:armhf-buster
services:
- docker:stable-dind
- name: multiarch/qemu-user-static:register
command:
- "--reset"
script:
- make build
See multiarch/qemu-user-static
To run your jobs on AWS Graviton you have to provide an AMI for arm64 architecture.
`typescript
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
token: token,
configuration: {
instanceType: InstanceType.of(InstanceClass.M6G, InstanceSize.LARGE),
machineImage: MachineImage.genericLinux({
[this.region]: new LookupMachineImage({
name: "ubuntu/images/hvm-ssd/ubuntu-focal-20.04--server-",
owners: ["099720109477"],
filters: {
architecture: [InstanceArchitecture.ARM_64],
"image-type": ["machine"],
state: ["available"],
"root-device-type": ["ebs"],
"virtualization-type": ["hvm"],
},
}).getImage(this).imageId,
}),
},
},
],
});
`
See Ubuntu Amazon EC2 AMI Locator
To deploy from within your GitLab Runner Instances, you may pass a Role with the IAM Policies attached.
`typescript
const role = new Role(this, "RunnersRole", {
assumedBy: new ServicePrincipal("ec2.amazonaws.com", {}),
inlinePolicies: {},
});
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
role: role,
token: token,
configuration: {
// optionally configure your runner
},
},
],
});
`
See example,
GitlabRunnerAutoscalingProps
If no existing Vpc is passed, a cheap VPC with a NatInstance (t3.nano) and a single AZ will be created.
`typescript
const natInstanceProvider = aws_ec2.NatProvider.instance({
instanceType: aws_ec2.InstanceType.of(InstanceClass.T3, InstanceSize.NANO), // using a cheaper gateway (not scalable)
});
const vpc = new Vpc(this, "Vpc", {
// Your custom vpc, i.e.:
natGatewayProvider: natInstanceProvider,
maxAzs: 1,
});
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
token: token,
configuration: {
// optionally configure your runner
},
},
],
network: { vpc: vpc },
});
`
See example,
GitlabRunnerAutoscalingProps
Deploys the Autoscaling GitLab Runner on AWS EC2 with the default settings mentioned above.
Happy with the presets?
`typescript
const token = StringParameter.fromStringParameterAttributes(stack, "Token", {
parameterName: "/gitlab-runner/token",
});
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
token: token,
configuration: {
// optionally configure your runner
},
},
],
});
`
See example,
GitlabRunnerAutoscalingProps
By default, the GitLab amzonec2 driver will be configured to install the
amazon-ecr-credential-helper
on the runner's instances.
To configure, override the default job runners environment:
`typescript`
new GitlabRunnerAutoscaling(this, "Runner", {
runners: [
{
// ...
environment: [
"DOCKER_DRIVER=overlay2",
"DOCKER_TLS_CERTDIR=/certs",
'DOCKER_AUTH_CONFIG={"credHelpers": { "public.ecr.aws": "ecr-login", "
],
},
],
});
This project uses projen to maintain project configuration through code. Thus, the synthesized files with projen should never be manually edited (in fact, projen enforces that).
To modify the project setup, you should interact with rich strongly-typed
class AwsCdkTypeScriptApp and
execute npx projen to update project configuration files.
> In simple words, developers can only modify .projenrc.js file for configuration/maintenance and files under /src` directory for development.
See also Create and Publish CDK Constructs Using projen and jsii.