Contribute to GitLab Runner development

GitLab Runner is a Go binary which can operate in two modes:

  1. GitLab Runner executing jobs locally (“instance” executor).
  2. Runner manager delegating jobs to an autoscaled environment which uses GitLab Runner Helper to pull artifacts.

For developing GitLab Runner in instance executor mode (1) the only setup required is a working Go environment. For developing GitLab Runner in Manager and Helper mode (2) setup also requires a Docker build environment. Additionally running the Manager or Helper in Kubernetes will require a working cluster.

The following instructions setup your Go environment using asdf to manage the Go version. If you already have this or otherwise know what you’re doing, you can skip step 2 (“Install dependencies and Go runtime”).

In order to provide Docker and Kubernetes locally Step 3 has you setting Rancher Desktop. If you don’t need one or both you can skip step 3 (“Install Rancher Desktop”) or just disable k3s (Kubernetes) in Rancher Desktop.

The recommended environment on which to install Go and Rancher Desktop for development is a local laptop or desktop. It is possible to use nested-virtualization to run Rancher Desktop in the cloud (which runs k3s in a VM) but it’s more tricky to setup.

Runner Shorts Video Tutorials

You can also follow along with the Runner Shorts (~20 minute videos) on setting up and making a change:

  1. Please read the recommended environment section above before beginning
  2. Setting up a GitLab Runner development environment
  3. Code walkthrough of GitLab Runner
  4. Making and testing locally a GitLab Runner change

1. Clone GitLab Runner 

git clone https://gitlab.com/gitlab-org/gitlab-runner.git

If you are developing for GitLab Runner in autoscaled mode (Manager and Helper) you might want to check out one or more of Taskscaler, Fleeting and associated plugins. To make local changes from one package visible to the others, use Go workspaces.

git clone https://gitlab.com/gitlab-org/fleeting/taskscaler.git
git clone https://gitlab.com/gitlab-org/fleeting/fleeting.git
git clone https://gitlab.com/gitlab-org/fleeting/fleeting-plugin-aws.git
git clone https://gitlab.com/gitlab-org/fleeting/fleeting-plugin-googlecompute.git
go work init
go work use gitlab-runner
go work use taskscaler
go work use fleeting
go work use fleeting-plugin-aws
go work use fleeting-plugin-googlecompute

2. Install dependencies and Go runtime

The GitLab Runner project uses asdf to manage dependencies. The simplest way to get your development environment setup is to use asdf.

asdf
cd gitlab-runner
asdf plugin add golang
asdf install
Debian/Ubuntu
sudo apt-get install -y mercurial git-core wget make build-essential
wget https://storage.googleapis.com/golang/go1.20.5.linux-amd64.tar.gz
sudo tar -C /usr/local -xzf go*-*.tar.gz
export PATH="$(go env GOBIN):$PATH"
CentOS
sudo yum install mercurial wget make
sudo yum groupinstall 'Development Tools'
wget https://storage.googleapis.com/golang/go1.20.5.linux-amd64.tar.gz
sudo tar -C /usr/local -xzf go*-*.tar.gz
export PATH="$(go env GOBIN):$PATH"
macOS

Using binary package:

wget https://storage.googleapis.com/golang/go1.20.5.darwin-amd64.tar.gz
sudo tar -C /usr/local -xzf go*-*.tar.gz
export PATH="$(go env GOBIN):$PATH"

Using installation package:

wget https://storage.googleapis.com/golang/go1.20.5.darwin-amd64.pkg
open go*-*.pkg
export PATH="$(go env GOBIN):$PATH"
FreeBSD
pkg install go-1.20.5 gmake git mercurial
export PATH="$(go env GOBIN):$PATH"

3. Install Rancher Desktop

The Docker Engine is required to create pre-built image that is embedded into GitLab Runner and loaded when using Docker executor. A local Kubernetes cluster is helpful for developing Kubernetes executor. Rancher Desktop provides both.

To install Rancher Desktop, follow the installation instructions for your OS.

note
Be sure to configure Rancher Desktop to use dockerd (moby) and not containerd.

4. Install GitLab Runner dependencies 

make deps
asdf reshim

For FreeBSD use gmake deps

5. Build GitLab Runner

Compile GitLab Runner using the Go toolchain:

make runner-and-helper-bin-host

make runner-and-helper-bin-host is a superset of make runner-bin-host which in addition takes care of building the Runner Helper Docker archive dependencies.

6. Run GitLab Runner 

./out/binaries/gitlab-runner run

You can use the any of the usual command-line arguments (including --debug):

./out/binaries/gitlab-runner --debug run

Building the Docker images

If you want to build the Docker images, run make runner-and-helper-docker-host, which will:

  1. Build gitlab-runner-helper and create a helper Docker image from it.
  2. Compile GitLab Runner for linux/amd64.
  3. Build a DEB package for Runner. The official GitLab Runner images are based on Alpine and Ubuntu, and the Ubuntu image build uses the DEB package.
  4. Build the Alpine and Ubuntu versions of the gitlab/gitlab-runner image.

New auto-scaling (Taskscaler) in GitLab Runner (since 15.6.0)

The Next Runner Auto-scaling Architecture adds a new mechanism for autoscaling which will work with all environments. It will replace all current autoscaling mechanisms (e.g. Docker Machine). This new mechanism is in a pre-alpha state and actively being developed. There are two new libraries being used in GitLab Runner:

  1. Taskscaler
  2. Fleeting

You don’t need to check out these libraries to use GitLab Runner at HEAD, but some development in the autoscaling space may take place there. In addition Taskscaler and Fleeting, there are a number of Fleeting Plugins which adapt GitLab Runner to a specific cloud providers (e.g. Google Computer or AWS EC2). The written instructions above (“Clone GitLab Runner”) show how to check out the code and the videos (“Runner Shorts”) show how to use it. These instructions show how to use GitLab Runner with a plugin.

Each plugin will come with instructions on how to build the binary and configure the underlying instance group. This work is being done in this issue. The canonical build and configuration instructions will live with each plugin, but in the meantime, here are some general instructions.

Build the plugin

To run GitLab Runner with a plugin, generate an executable binary and place it on your system’s PATH.

To generate the binary, ensure $GOPATH/bin is on your PATH, then use go install.

Each plugin contains a path to ./cmd/<plugin-name>. For example, from the fleeting-plugin-aws directory:

cd cmd/fleeting-plugin-aws/
go install

If you manage go versions with asdf, run this command after the binary generates:

asdf reshim

Use the plugin

GitLab Runner is started in the usual way but specifies an instance executor. It also specifies under plugin_config and connector_config an Instance Group, its location, and some details about how to connect to the underlying instances. GitLab Runner should find the Instance Group and create an initial number of idle VMs. When a job is picked up the configured instance runner, it will consume a running VM and replace it via AWS service calls in the fleeting-plugin-aws plugin.

[[runners]]
  name = "local-taskrunner"
  url = "https://gitlab.com/"
  token = "REDACTED"
  executor = "instance"
  shell = "bash"
  [runners.autoscaler]
    max_use_count = 1
    max_instances = 20
    plugin = "fleeting-plugin-aws"                                 # Fleeting plugin name as built above [1].
    [runners.autoscaler.plugin_config]
      credentials_file = "/Users/josephburnett/.aws/credentials".  # Credentials which can scale an Autoscaling Group (ASG) [2].
      name = "jburnett-taskrunner-asg"                             # ASG name.
      project = "jburnett-ad8e5d54"                                # ASG project.
      region = "us-east-2"                                         # ASG region.
    [runners.autoscaler.connector_config]
      username = "ubuntu"                                          # ASG instance template username for login.
    [[runners.autoscaler.policy]]
      idle_count = 5
      idle_time = 0
      scale_factor = 0.0
      scale_factor_limit = 0

If you terminate GitLab Runner with SIGTERM you may see some of these processes hanging around. Instead terminate with SIGQUIT.

Note that ASGs should have autoscaling disabled. GitLab Runner takes care of autoscaling via the Taskscaler library.

7. Run test suite locally

GitLab Runner test suite consists of “core” tests and tests for executors. Tests for executors require certain binaries to be installed on your local machine. Some of these binaries cannot be installed on all operating systems. If a binary is not installed tests requiring this binary will be skipped.

These are the binaries that you can install:

  1. VirtualBox and Vagrant; the Vagrant Parallels plugin is also required
  2. kubectl with minikube
  3. Parallels Pro or Business edition
  4. PowerShell

After installing the binaries run:

make development_setup

To execute the tests run:

make test

Kubernetes Integration tests

To run correctly, some Kubernetes integration tests require specific configuration or runtime arguments of the Kubernetes cluster they run against. These tests will be skipped if the cluster configuration is incorrect. Below is a sample configuration for Kubernetes clusters that would commonly be used on a developer workstation:

  • minikube
minikube delete
minikube config set container-runtime containerd
minikube config set feature-gates "ProcMountType=true"
minikube start
  • k3s
k3s server --tls-san=k3s --kube-apiserver-arg=feature-gates=ProcMountType=true

8. Run tests with helper image version of choice

If you are developing functionality inside a helper, you’ll most likely want to run tests with the version of the Docker image that contains the newest changes.

If you run tests without passing -ldflags, the default version in version.go is development. This means that the runner defaults to pulling a helper image with the latest tag.

Make targets

make targets inject -ldflags automatically. You can run all tests by using:

make simple-test

make targets also inject -ldflags for parallel_test_execute, which is most commonly used by the CI/CD jobs.

Custom go test arguments

In case you want a more customized go test command, you can use print_ldflags as make target:

go test -ldflags "$(make print_ldflags)" -run TestDockerCommandBuildCancel -v ./executors/docker/...

In GoLand

Currently, GoLand doesn’t support dynamic Go tool arguments, so you’ll need to run make print_ldflags first and then paste it in the configuration.

note
To use the debugger, make sure to remove the last two flags (-s -w).

Helper image

Build the newest version of the helper image with:

make helper-dockerarchive-host

Then you’ll have the image ready for use:

REPOSITORY                                                    TAG                      IMAGE ID            CREATED             SIZE
gitlab/gitlab-runner-helper                                   x86_64-a6bc0800          f10d9b5bbb41        32 seconds ago      57.2MB

Helper image with Kubernetes

If you are running a local Kubernetes cluster make sure to reuse the cluster’s Docker daemon to build images. For example, with minikube:

eval $(minikube docker-env)

9. Install optional tools

  • Install golangci-lint, used for the make lint target.
  • Install markdown-lint and vale, used for the make lint-docs target.

Installation instructions will pop up when running a Makefile target if a tool is missing.

10. Contribute

You can start hacking gitlab-runner code. If you need an IDE to edit and debug code, there are a few free suggestions you can use:

Managing build dependencies

GitLab Runner uses Go Modules to manage its dependencies.

Don’t add dependency from upstream default branch when version tags are available.

Tests

The Runner codebase makes a distinction between unit and integration tests in the following way:

  • Unit test files have a suffix of _test.go and contain the following build directive in the header:

      // go:build !integration

  • Integration test files have a suffix of _integration_test.go and contain the following build directive in the header:

      // go:build integration

    They can be run by adding -tags=integration to the go test command.

To test the state of the build directives in test files, make check_test_directives can be used.

Developing for Windows on a non-windows environment

We provide a Vagrantfile to help you run a Windows Server 2019 or Windows 10 instance, since we are using multiple machines inside of Vagrant.

The following are required:

  • Vagrant installed.
  • Virtualbox installed.
  • Around 30GB of free hard disk space on your computer.

Which virtual machine to use depends on your use case:

  • The Windows Server machine has Docker pre-installed and should always be used when you are developing on GitLab Runner for Windows.
  • The Windows 10 machine is there for you to have a windows environment with a GUI which sometimes can help you debugging some Windows features. Note that you cannot have Docker running inside of Windows 10 because nested virtualization is not supported.

Running vagrant up windows_10 will start the Windows 10 machine for you. To:

  • SSH inside of the Windows 10 machine, run vagrant ssh windows_10.
  • Access the GUI for the Windows 10, you can connect via RDP by running vagrant rdp windows_10, which will connect to the machine using a locally installed RDP program.

For both machines, the GitLab Runner source code is synced bi-directionally so that you can edit from your machine with your favorite editor. The source code can be found under the $GOROOT environment variable. We have a RUNNER_SRC environment variable which you can use to find out the full path so when using PowerShell, you can use cd $Env:RUNNER_SRC.

Other resources

  1. Reviewing GitLab Runner merge requests
  2. Add support for new Windows Version
  3. Runner Group - Team Resources