• Prerequisites
  • Cluster
  • Ingress controller
  • TLS certificates
  • Metrics
  • Configure Domain Name Services
• Installing the GitLab Operator
• Installing GitLab
• Recommended next steps
  • OpenShift
• Uninstall the GitLab Operator
  • Uninstall an instance of GitLab
  • Uninstall the GitLab Operator
• Troubleshoot the GitLab Operator

Installation

This document describes how to deploy the GitLab Operator by using manifests in your Kubernetes or OpenShift cluster.

If using OpenShift, installation is typically handled by the Operator Lifecycle Manager (OLM). Installation using OLM is considered experimental. GitLab does not support any issues related to instances deployed using OLM. For more information on potential issues with OLM, see issue 241.

Prerequisites

1. Create or use an existing Kubernetes or OpenShift cluster
2. Install pre-requisite services and software
   • Ingress controller
   • cert-manager
   • Metrics server
3. Configure Domain Name Services

Cluster

The GitLab Operator aims to support new minor Kubernetes and OpenShift versions three months after their initial releases. We welcome any compatibility issues with releases newer than those listed above in our issue tracker.

Some GitLab features might not work on deprecated versions and versions older than the versions listed above.

For some components, like the agent for Kubernetes and GitLab Charts, GitLab might support different cluster versions.

Starting with 16.7, the Operator is built for x86-64 and arm64. The arm64 images are not tested in CI and are not recommended for production use.

If you are on a multi-arch cluster you may want to add a node selector for the kubernetes.io/arch label to the Operator Deployment.

Patch the Deployment to be scheduled on x86-64/amd64 nodes only:

kubectl patch deployments gitlab-controller-manager \
  -p '{"spec": {"template": {"spec": {"nodeSelector": {"kubernetes.io/arch": "amd64"}}}}}'

If you are using the Operator Helm chart you can add the node selector to your values.yaml instead:

nodeSelector:
  kubernetes.io/arch: amd64

This ensures that the Operator runs on amd64 nodes, using the platform that we currently test.

See epic 10928 for more information on the arm64 support for CNG images.

Ingress controller

An Ingress controller is required to provide external access to the application and secure communication between components.

The GitLab Operator deploys our forked NGINX chart from the GitLab Helm Chart by default.

If you prefer to use an external Ingress controller, use NGINX Ingress by the Kubernetes community to deploy an Ingress Controller. Follow the relevant instructions in the link based on your platform and preferred tooling. Take note of the Ingress class value for later (it typically defaults to nginx). When configuring the GitLab CR, be sure to set nginx-ingress.enabled=false to disable the NGINX objects from the GitLab Helm Chart.

TLS certificates

To create a certificate for the operator’s Kubernetes webhook, cert-manager is used. You should use cert-manager for the GitLab certificates as well.

Because the operator needs a certificate for the Kubernetes webhook, you can’t use the cert-manager bundled with the GitLab Chart. Instead, install cert-manager before you install the operator.

Follow the installation documentation to install a supported cert-manager release for your platform and tooling.

Metrics

Configure Domain Name Services

You need an internet-accessible domain to which you can add a DNS record.

See our networking and DNS documentation for more details on connecting your domain to the GitLab components. You use the configuration mentioned in this section when defining your GitLab custom resource (CR).

Ingress in OpenShift requires extra consideration. See our notes on OpenShift Ingress for more information.

Installing the GitLab Operator

Start by selecting an installation method.

kubectl create namespace gitlab-system

kubectl apply -f gitlab-operator-<platform>.yaml

helm repo add gitlab https://charts.gitlab.io
helm repo update

helm install gitlab-operator gitlab/gitlab-operator \
  --create-namespace \
  --namespace gitlab-system

Confirm the installation by checking the status of the Operator Deployment:

kubectl -n gitlab-system get deployment gitlab-controller-manager

Installing GitLab

1. Create a GitLab custom resource (CR).

   Create a new file named something like mygitlab.yaml.

   Here is an example of the content to put in this file:

   apiVersion: apps.gitlab.com/v1beta1
   kind: GitLab
   metadata:
     name: gitlab
   spec:
     chart:
       version: "X.Y.Z" # https://gitlab.com/gitlab-org/cloud-native/gitlab-operator/-/blob/<OPERATOR_VERSION>/CHART_VERSIONS
       values:
         global:
           hosts:
             domain: example.com # use a real domain here
           ingress:
             configureCertmanager: true
         certmanager-issuer:
           email: youremail@example.com # use your real email address here
   

   For more details on configuration options to use under spec.chart.values, see the GitLab Helm Chart documentation.

2. Deploy a GitLab instance using your new GitLab CR.

   kubectl -n gitlab-system apply -f mygitlab.yaml
   

   This command sends your GitLab CR up to the cluster for the GitLab Operator to reconcile. You can watch the progress by tailing the logs from the controller pod:

   kubectl -n gitlab-system logs deployment/gitlab-controller-manager -c manager -f
   

   You can also list GitLab resources and check their status:

   $ kubectl -n gitlab-system get gitlab
   NAME     STATUS   VERSION
   gitlab   Ready    5.2.4
   

When the CR is reconciled (the status of the GitLab resource is Running), you can access GitLab in your browser at https://gitlab.example.com.

To log in you need to retrieve the initial root password for your deployment. See the Helm Chart documentation for further instructions.

Recommended next steps

After completing your installation, consider taking the recommended next steps, including authentication options and sign-up restrictions.

OpenShift

If you run OpenShift, change the approval strategy for the GitLab Operator from automatic (the default) to manual. This prevents OpenShift from installing new operator versions until approval is given.

You can also set a custom startingCSV to pin the Operator’s version or to upgrade to a non-latest version.

• The approval strategy can be changed from the OpenShift web console or by editing the Subscription.
• Set .spec.approved to true of the InstallPlan to approve an manual upgrade.
• Each GitLab Operator supports a defined subset of GitLab chart versions: upgrades to the GitLab Operator must also involve updating the chart version in the GitLab custom resource.
• If the Operator and specified chart version are incompatible, configuration changes to the chart can fail with errors about the chart version.

Uninstall the GitLab Operator

Follow the steps below to remove the GitLab Operator and its associated resources.

Items to note prior to uninstalling the operator:

• The operator does not delete the Persistent Volume Claims or Secrets when a GitLab instance is deleted.
• When deleting the Operator, the namespace where it is installed (gitlab-system by default) is not deleted automatically. This ensures that persistent volumes are not lost unintentionally.

Uninstall an instance of GitLab

kubectl -n gitlab-system delete -f mygitlab.yaml

This removes the GitLab instance, and all associated objects except for Persistent Volume Claims as noted above).

Uninstall the GitLab Operator

GL_OPERATOR_VERSION=<your_installed_version> # https://gitlab.com/gitlab-org/cloud-native/gitlab-operator/-/releases
PLATFORM=kubernetes # or "openshift"
kubectl delete -f https://gitlab.com/api/v4/projects/18899486/packages/generic/gitlab-operator/${GL_OPERATOR_VERSION}/gitlab-operator-${PLATFORM}-${GL_OPERATOR_VERSION}.yaml

This deletes the Operator’s resources, including the running Deployment of the Operator. This does not delete objects associated with a GitLab instance.

Troubleshoot the GitLab Operator

Troubleshooting the Operator can be found in troubleshooting.md.