GitLab Runner

Tier: Free, Premium, Ultimate Offering: GitLab.com, Self-managed

GitLab Runner is an application that works with GitLab CI/CD to run jobs in a pipeline.

Use GitLab-hosted runners

If you use GitLab.com or GitLab Dedicated, you can run your CI/CD jobs on GitLab-hosted runners. These runners are managed by GitLab and fully integrated with GitLab.com. By default these runners are enabled for all projects. You can disable the runners if you have the Owner role for the project.

Use self-managed runners

Tier: Free, Premium, Ultimate Offering: Self-managed

Alternatively, you can install GitLab Runner and register your own runners on GitLab.com or on your own instance. To use self-managed runners, you install GitLab Runner on infrastructure that you own or manage.

Scale a fleet of runners

When your organization scales to having a fleet of runners, you should plan for how you will monitor and adjust performance for these runners.

GitLab Runner versions

For compatibility reasons, the GitLab Runner major.minor version should stay in sync with the GitLab major and minor version. Older runners may still work with newer GitLab versions, and vice versa. However, features may not be available or work properly if a version difference exists.

Backward compatibility is guaranteed between minor version updates. However, sometimes minor version updates of GitLab can introduce new features that require GitLab Runner to be on the same minor version.

note
GitLab Runner 15.0 introduced a change to the registration API request format. It prevents the GitLab Runner from communicating with GitLab versions lower than 14.8. You must use a Runner version that is appropriate for the GitLab version, or upgrade the GitLab application.

If you host your own runners but host your repositories on GitLab.com, keep GitLab Runner updated to the latest version, as GitLab.com is updated continuously.

Runner registration

After you install the application, you register individual runners. Runners are the agents that run the CI/CD jobs that come from GitLab.

When you register a runner, you are setting up communication between your GitLab instance and the machine where GitLab Runner is installed.

Runners usually process jobs on the same machine where you installed GitLab Runner. However, you can also have a runner process jobs in a container, in a Kubernetes cluster, or in auto-scaled instances in the cloud.

Executors

When you register a runner, you must choose an executor.

An executor determines the environment each job runs in.

For example:

  • If you want your CI/CD job to run PowerShell commands, you might install GitLab Runner on a Windows server and then register a runner that uses the shell executor.
  • If you want your CI/CD job to run commands in a custom Docker container, you might install GitLab Runner on a Linux server and register a runner that uses the Docker executor.

These are only a few of the possible configurations. You can install GitLab Runner on a virtual machine and have it use another virtual machine as an executor.

When you install GitLab Runner in a Docker container and choose the Docker executor to run your jobs, it’s sometimes referred to as a “Docker-in-Docker” configuration.

Who has access to runners in the GitLab UI

Before you register a runner, you should determine if everyone in GitLab should have access to it, or if you want to limit it to a specific GitLab group or project.

There are three types of runners, based on who you want to have access:

The scope of a runner is defined during the registration. This is how the runner knows which projects it’s available for.

Tags

When you register a runner, you can add tags to it.

When a CI/CD job runs, it knows which runner to use by looking at the assigned tags. Tags are the only way to filter the list of available runners for a job.

For example, if a runner has the ruby tag, you would add this code to your project’s .gitlab-ci.yml file: 

job:
  tags:
    - ruby

When the job runs, it uses the runner with the ruby tag.

Configuring runners

You can configure the runner by editing the config.toml file. This is a file that is installed during the runner installation process.

In this file you can edit settings for a specific runner, or for all runners.

You can specify settings like logging and cache. You can set concurrency, memory, CPU limits, and more.

Monitoring runners

You can use Prometheus to monitor your runners. You can view things like the number of currently-running jobs and how much CPU your runners are using.

Use a runner to run jobs

After a runner is configured and available for your project, your CI/CD jobs can use the runner.

Features

GitLab Runner has the following features.

  • Run multiple jobs concurrently.
  • Use multiple tokens with multiple servers (even per-project).
  • Limit the number of concurrent jobs per-token.
  • Jobs can be run:
    • Locally.
    • Using Docker containers.
    • Using Docker containers and executing job over SSH.
    • Using Docker containers with autoscaling on different clouds and virtualization hypervisors.
    • Connecting to a remote SSH server.
  • Is written in Go and distributed as single binary without any other requirements.
  • Supports Bash, PowerShell Core, and Windows PowerShell.
  • Works on GNU/Linux, macOS, and Windows (pretty much anywhere you can run Docker).
  • Allows customization of the job running environment.
  • Automatic configuration reload without restart.
  • Easy to use setup with support for Docker, Docker-SSH, Parallels, or SSH running environments.
  • Enables caching of Docker containers.
  • Easy installation as a service for GNU/Linux, macOS, and Windows.
  • Embedded Prometheus metrics HTTP server.
  • Referee workers to monitor and pass Prometheus metrics and other job-specific data to GitLab.

Runner execution flow

This diagram shows how runners are registered and how jobs are requested and handled. It also shows which actions use registration, authentication, and job tokens.

sequenceDiagram participant GitLab participant GitLabRunner participant Executor opt registration GitLabRunner ->>+ GitLab: POST /api/v4/runners with registration_token GitLab -->>- GitLabRunner: Registered with runner_token end loop job requesting and handling GitLabRunner ->>+ GitLab: POST /api/v4/jobs/request with runner_token GitLab -->>+ GitLabRunner: job payload with job_token GitLabRunner ->>+ Executor: Job payload Executor ->>+ GitLab: clone sources with job_token Executor ->>+ GitLab: download artifacts with job_token Executor -->>- GitLabRunner: return job output and status GitLabRunner -->>- GitLab: updating job output and status with job_token end

Glossary

This glossary provides definitions for terms related to GitLab Runner.

  • GitLab Runner: The application that you install that executes GitLab CI jobs on a target computing platform.
  • runner configuration: A single [[runner]] entry in the config.toml that displays as a runner in the UI.
  • runner manager: The process that reads the config.toml and runs all the runner configurations concurrently.
  • runner: The process that executes the job on a selected machine. Depending on the type of executor, this machine could be local to the runner manager (shell or docker executor) or a remote machine created by an autoscaler (docker-autoscaler or kubernetes).
  • machine: A virtual machine (VM) or pod that the runner operates in. GitLab Runner automatically generates a unique, persistent machine ID so that when multiple machines are given the same runner configuration, jobs can be routed separately but the runner configurations are grouped in the UI.

See also the official GitLab Word List and the GitLab Architecture entry for GitLab Runner.

Troubleshooting

Learn how to troubleshoot common issues.

Contributing

Contributions are welcome. See CONTRIBUTING.md and the development documentation for details.

If you’re a reviewer of GitLab Runner project, take a moment to read the Reviewing GitLab Runner document.

You can also review the release process for the GitLab Runner project.

Changelog

See the CHANGELOG to view recent changes.

License

This code is distributed under the MIT license. View the LICENSE file.