GitLab SaaS runners

If you are using self-managed GitLab or you want to use your own runners on, you can install and configure your own runners.

If you are using GitLab SaaS (, your CI jobs automatically run on shared runners. No configuration is required. Your jobs can run on Linux or Windows.

The number of minutes you can use on these shared runners depends on your quota, which depends on your subscription plan.

Linux shared runners

Linux shared runners on run in autoscale mode and are powered by Google Cloud Platform.

Autoscaling means reduced queue times to spin up CI/CD jobs, and isolated VMs for each job, thus maximizing security. These shared runners are available for users and customers on

GitLab offers Ultimate tier capabilities and included CI/CD minutes per group per month for our Open Source, Education, and Startups programs. For private projects, GitLab offers various plans, starting with a Free tier.

All your CI/CD jobs run on n1-standard-1 instances with 3.75GB of RAM, CoreOS and the latest Docker Engine installed. Instances provide 1 vCPU and 25GB of HDD disk space. The default region of the VMs is US East1. Each instance is used only for one job, this ensures any sensitive data left on the system can't be accessed by other people their CI jobs.

The fleet of runners are dedicated for GitLab projects as well as community forks of them. They use a slightly larger machine type (n1-standard-2) and have a bigger SSD disk size. They don't run untagged jobs and unlike the general fleet of shared runners, the instances are re-used up to 40 times.

Jobs handled by the shared runners on (, time out after 3 hours, regardless of the timeout configured in a project. Check the issues 4010 and 4070 for the reference.

Below are the shared runners settings.

Setting Default
GitLab Runner Runner versions dashboard -
Executor docker+machine -
Default Docker image ruby:2.5 -
privileged (run Docker in Docker) true false

Pre-clone script

Linux shared runners on provide a way to run commands in a CI job before the runner attempts to run git init and git fetch to download a GitLab repository. The pre_clone_script can be used for:

  • Seeding the build directory with repository data
  • Sending a request to a server
  • Downloading assets from a CDN
  • Any other commands that must run before the git init

To use this feature, define a CI/CD variable called CI_PRE_CLONE_SCRIPT that contains a bash script.

This example demonstrates how you might use a pre-clone step to seed the build directory.


The full contents of our config.toml are:

NOTE: Settings that are not public are shown as X.

Google Cloud Platform

concurrent = X
check_interval = 1
metrics_server = "X"
sentry_dsn = "X"

  name = "docker-auto-scale"
  request_concurrency = X
  url = ""
  pre_clone_script = "eval \"$CI_PRE_CLONE_SCRIPT\""
  executor = "docker+machine"
  environment = [
  limit = X
    image = "ruby:2.5"
    privileged = true
    volumes = [
      "/dummy-sys-class-dmi-id:/sys/class/dmi/id:ro" # Make kaniko builds work on GCP.
    IdleCount = 50
    IdleTime = 3600
    MaxBuilds = 1 # For security reasons we delete the VM after job has finished so it's not reused.
    MachineName = "srm-%s"
    MachineDriver = "google"
    MachineOptions = [
      "engine-opt=mtu=1460", # Set MTU for container interface, for more information check
      "engine-opt=ipv6", # This will create IPv6 interfaces in the containers.
      "google-operation-backoff-initial-interval=2" # Custom flag from forked docker-machine, for more information check
      Periods = ["* * * * * sat,sun *"]
      Timezone = "UTC"
      IdleCount = 70
      IdleTime = 3600
      Periods = ["* 30-59 3 * * * *", "* 0-30 4 * * * *"]
      Timezone = "UTC"
      IdleCount = 700
      IdleTime = 3600
    Type = "gcs"
    Shared = true
      CredentialsFile = "/path/to/file"
      BucketName = "bucket-name"

Windows shared runners (beta)

The Windows shared runners are in beta and shouldn't be used for production workloads.

During this beta period, the shared runner pipeline quota applies for groups and projects in the same manner as Linux runners. This may change when the beta period ends, as discussed in this related issue.

Windows shared runners on autoscale by launching virtual machines on the Google Cloud Platform. This solution uses an autoscaling driver developed by GitLab for the custom executor. Windows shared runners execute your CI/CD jobs on n1-standard-2 instances with 2 vCPUs and 7.5 GB RAM. You can find a full list of available Windows packages in the package documentation.

We want to keep iterating to get Windows shared runners in a stable state and generally available. You can follow our work towards this goal in the related epic.


The full contents of our config.toml are:

NOTE: Settings that aren't public are shown as X.

concurrent = X
check_interval = 3

  name = "windows-runner"
  url = ""
  token = "TOKEN"
  executor = "custom"
  builds_dir = "C:\\GitLab-Runner\\builds"
  cache_dir = "C:\\GitLab-Runner\\cache"
  shell  = "powershell"
    config_exec = "C:\\GitLab-Runner\\autoscaler\\autoscaler.exe"
    config_args = ["--config", "C:\\GitLab-Runner\\autoscaler\\config.toml", "custom", "config"]
    prepare_exec = "C:\\GitLab-Runner\\autoscaler\\autoscaler.exe"
    prepare_args = ["--config", "C:\\GitLab-Runner\\autoscaler\\config.toml", "custom", "prepare"]
    run_exec = "C:\\GitLab-Runner\\autoscaler\\autoscaler.exe"
    run_args = ["--config", "C:\\GitLab-Runner\\autoscaler\\config.toml", "custom", "run"]
    cleanup_exec = "C:\\GitLab-Runner\\autoscaler\\autoscaler.exe"
    cleanup_args = ["--config", "C:\\GitLab-Runner\\autoscaler\\config.toml", "custom", "cleanup"]

The full contents of our autoscaler/config.toml are:

Provider = "gcp"
Executor = "winrm"
OS = "windows"
LogLevel = "info"
LogFormat = "text"
LogFile = "C:\\GitLab-Runner\\autoscaler\\autoscaler.log"
VMTag = "windows"

  ServiceAccountFile = "PATH"
  Project = "some-project-df9323"
  Zone = "us-east1-c"
  MachineType = "n1-standard-2"
  Image = "IMAGE"
  DiskSize = 50
  DiskType = "pd-standard"
  Subnetwork = "default"
  Network = "default"
  Tags = ["TAGS"]
  Username = "gitlab_runner"

  MaximumTimeout = 3600
  ExecutionMaxRetries = 0

  Enabled = true
  Directory = "C:\\GitLab-Runner\\autoscaler\\machines"


Below is a simple .gitlab-ci.yml file to show how to start using the Windows shared runners:

    - shared-windows
    - windows
    - windows-1809

  - build
  - test

 - Set-Variable -Name "time" -Value (date -Format "%H:%m")
 - echo ${time}
 - echo "started by ${GITLAB_USER_NAME}"

    - .shared_windows_runners
  stage: build
    - echo "running scripts in the build job"

    - .shared_windows_runners
  stage: test
    - echo "running scripts in the test job"

Limitations and known issues

  • All the limitations mentioned in our beta definition.
  • The average provisioning time for a new Windows VM is 5 minutes. This means that you may notice slower build start times on the Windows shared runner fleet during the beta. In a future release we intend to update the autoscaler to enable the pre-provisioning of virtual machines. This is intended to significantly reduce the time it takes to provision a VM on the Windows fleet. You can follow along in the related issue.
  • The Windows shared runner fleet may be unavailable occasionally for maintenance or updates.
  • The Windows shared runner virtual machine instances do not use the GitLab Docker executor. This means that you can't specify image or services in your pipeline configuration.
  • For the beta release, we have included a set of software packages in the base VM image. If your CI job requires additional software that's not included in this list, then you must add installation commands to before_script or script to install the required software. Note that each job runs on a new VM instance, so the installation of additional software packages needs to be repeated for each job in your pipeline.
  • The job may stay in a pending state for longer than the Linux shared runners.
  • There is the possibility that we introduce breaking changes which will require updates to pipelines that are using the Windows shared runner fleet.