Skip to content

Using GitLab CI/CD with a Kubernetes cluster

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

  • Agent connection sharing limit changed from 100 to 500 in GitLab 17.0.

You can use GitLab CI/CD to safely connect, deploy, and update your Kubernetes clusters.

To do so, install an agent in your cluster. When done, you have a Kubernetes context and can run Kubernetes API commands in your GitLab CI/CD pipeline.

To ensure access to your cluster is safe:

  • Each agent has a separate context (kubecontext).
  • Only the project where the agent is configured, and any additional projects you authorize, can access the agent in your cluster.

To use GitLab CI/CD to interact with your cluster, runners must be registered with GitLab. However, these runners do not have to be in the cluster where the agent is.

Use GitLab CI/CD with your cluster

To update a Kubernetes cluster with GitLab CI/CD:

  1. Ensure you have a working Kubernetes cluster and the manifests are in a GitLab project.
  2. In the same GitLab project, register and install the GitLab agent.
  3. Update your .gitlab-ci.yml file to select the agent's Kubernetes context and run the Kubernetes API commands.
  4. Run your pipeline to deploy to or update the cluster.

If you have multiple GitLab projects that contain Kubernetes manifests:

  1. Install the GitLab agent in its own project, or in one of the GitLab projects where you keep Kubernetes manifests.
  2. Authorize the agent to access your GitLab projects.
  3. Optional. For added security, use impersonation.
  4. Update your .gitlab-ci.yml file to select the agent's Kubernetes context and run the Kubernetes API commands.
  5. Run your pipeline to deploy to or update the cluster.

Authorize the agent

If you have multiple GitLab projects, you must authorize the agent to access the project where you keep your Kubernetes manifests. You can authorize the agent to access individual projects, or authorize a group or subgroup, so all projects within have access. For added security, you can also use impersonation.

Authorization configuration can take one or two minutes to propagate.

Authorize the agent to access your projects

  • Changed to remove hierarchy restrictions in GitLab 15.6.
  • Changed to allow authorizing projects in a user namespace in GitLab 15.7.

To authorize the agent to access the GitLab project where you keep Kubernetes manifests:

  1. On the left sidebar, select Search or go to and find the project that contains the agent configuration file (config.yaml).

  2. Edit the config.yaml file. Under the ci_access keyword, add the projects attribute.

  3. For the id, add the path to the project.

    ci_access:
      projects:
        - id: path/to/project
    • Authorized projects must have the same root group or user namespace as the agent's configuration project.
    • You can install additional agents into the same cluster to accommodate additional hierarchies.
    • You can authorize up to 500 projects.

All CI/CD jobs now include a kubeconfig file with contexts for every shared agent connection. The kubeconfig path is available in the environment variable $KUBECONFIG. Choose the context to run kubectl commands from your CI/CD scripts.

Authorize the agent to access projects in your groups

  • Changed to remove hierarchy restrictions in GitLab 15.6.

To authorize the agent to access all of the GitLab projects in a group or subgroup:

  1. On the left sidebar, select Search or go to and find the project that contains the agent configuration file (config.yaml).

  2. Edit the config.yaml file. Under the ci_access keyword, add the groups attribute.

  3. For the id, add the path:

    ci_access:
      groups:
        - id: path/to/group/subgroup
    • Authorized groups must have the same root group as the agent's configuration project.
    • You can install additional agents into the same cluster to accommodate additional hierarchies.
    • All of the subgroups of an authorized group also have access to the same agent (without being specified individually).
    • You can authorize up to 500 groups.

All the projects that belong to the group and its subgroups are now authorized to access the agent. All CI/CD jobs now include a kubeconfig file with contexts for every shared agent connection. The kubeconfig path is available in an environment variable $KUBECONFIG. Choose the context to run kubectl commands from your CI/CD scripts.

Update your .gitlab-ci.yml file to run kubectl commands

In the project where you want to run Kubernetes commands, edit your project's .gitlab-ci.yml file.

In the first command under the script keyword, set your agent's context. Use the format <path/to/agent/project>:<agent-name>. For example:

deploy:
  image:
    name: bitnami/kubectl:latest
    entrypoint: ['']
  script:
    - kubectl config get-contexts
    - kubectl config use-context path/to/agent/project:agent-name
    - kubectl get pods

If you are not sure what your agent's context is, open a terminal and connect to your cluster. Run kubectl config get-contexts.

Environments that use Auto DevOps

If Auto DevOps is enabled, you must define the CI/CD variable KUBE_CONTEXT. Set the value of KUBE_CONTEXT to the context of the agent you want Auto DevOps to use:

deploy:
  variables:
    KUBE_CONTEXT: path/to/agent/project:agent-name

You can assign different agents to separate Auto DevOps jobs. For instance, Auto DevOps can use one agent for staging jobs, and another agent for production jobs. To use multiple agents, define an environment-scoped CI/CD variable for each agent. For example:

  1. Define two variables named KUBE_CONTEXT.
  2. For the first variable:
    1. Set the environment to staging.
    2. Set the value to the context of your staging agent.
  3. For the second variable:
    1. Set the environment to production.
    2. Set the value to the context of your production agent.

Environments with both certificate-based and agent-based connections

When you deploy to an environment that has both a certificate-based cluster (deprecated) and an agent connection:

  • The certificate-based cluster's context is called gitlab-deploy. This context is always selected by default.
  • Agent contexts are included in $KUBECONFIG. You can select them by using kubectl config use-context <path/to/agent/project>:<agent-name>.

To use an agent connection when certificate-based connections are present, you can manually configure a new kubectl configuration context. For example:

deploy:
  variables:
    KUBE_CONTEXT: my-context # The name to use for the new context
    AGENT_ID: 1234 # replace with your agent's numeric ID
    K8S_PROXY_URL: https://<KAS_DOMAIN>/k8s-proxy/ # For agent server (KAS) deployed in Kubernetes cluster (for gitlab.com use kas.gitlab.com); replace with your URL
    # K8S_PROXY_URL: https://<GITLAB_DOMAIN>/-/kubernetes-agent/k8s-proxy/ # For agent server (KAS) in Omnibus
    # ... any other variables you have configured
  before_script:
    - kubectl config set-credentials agent:$AGENT_ID --token="ci:${AGENT_ID}:${CI_JOB_TOKEN}"
    - kubectl config set-cluster gitlab --server="${K8S_PROXY_URL}"
    - kubectl config set-context "$KUBE_CONTEXT" --cluster=gitlab --user="agent:${AGENT_ID}"
    - kubectl config use-context "$KUBE_CONTEXT"
  # ... rest of your job configuration

Environments with KAS that use self-signed certificates

If you use an environment with KAS and a self-signed certificate, you must configure your Kubernetes client to trust the certificate authority (CA) that signed your certificate.

To configure your client, do one of the following:

  • Set a CI/CD variable SSL_CERT_FILE with the KAS certificate in PEM format.
  • Configure the Kubernetes client with --certificate-authority=$KAS_CERTIFICATE, where KAS_CERTIFICATE is a CI/CD variable with the CA certificate of KAS.
  • Place the certificates in an appropriate location in the job container by updating the container image or mounting via the runner.
  • Not recommended. Configure the Kubernetes client with --insecure-skip-tls-verify=true.

Restrict project and group access by using impersonation

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

  • Changed in GitLab 15.5 to add impersonation support for environment tiers.

By default, your CI/CD job inherits all the permissions from the service account used to install the agent in the cluster. To restrict access to your cluster, you can use impersonation.

To specify impersonations, use the access_as attribute in your agent configuration file and use Kubernetes RBAC rules to manage impersonated account permissions.

You can impersonate:

  • The agent itself (default).
  • The CI/CD job that accesses the cluster.
  • A specific user or system account defined within the cluster.

Authorization configuration can take one or two minutes to propagate.

Impersonate the agent

The agent is impersonated by default. You don't need to do anything to impersonate it.

Impersonate the CI/CD job that accesses the cluster

To impersonate the CI/CD job that accesses the cluster, under the access_as key, add the ci_job: {} key-value.

When the agent makes the request to the actual Kubernetes API, it sets the impersonation credentials in the following way:

  • UserName is set to gitlab:ci_job:<job id>. Example: gitlab:ci_job:1074499489.

  • Groups is set to:

    • gitlab:ci_job to identify all requests coming from CI jobs.

    • The list of IDs of groups the project is in.

    • The project ID.

    • The slug and tier of the environment this job belongs to.

      Example: for a CI job in group1/group1-1/project1 where:

      • Group group1 has ID 23.
      • Group group1/group1-1 has ID 25.
      • Project group1/group1-1/project1 has ID 150.
      • Job running in the prod environment, which has the production environment tier.

    Group list would be [gitlab:ci_job, gitlab:group:23, gitlab:group_env_tier:23:production, gitlab:group:25, gitlab:group_env_tier:25:production, gitlab:project:150, gitlab:project_env:150:prod, gitlab:project_env_tier:150:production].

  • Extra carries extra information about the request. The following properties are set on the impersonated identity:

Property Description
agent.gitlab.com/id Contains the agent ID.
agent.gitlab.com/config_project_id Contains the agent's configuration project ID.
agent.gitlab.com/project_id Contains the CI project ID.
agent.gitlab.com/ci_pipeline_id Contains the CI pipeline ID.
agent.gitlab.com/ci_job_id Contains the CI job ID.
agent.gitlab.com/username Contains the username of the user the CI job is running as.
agent.gitlab.com/environment_slug Contains the slug of the environment. Only set if running in an environment.
agent.gitlab.com/environment_tier Contains the tier of the environment. Only set if running in an environment.

Example config.yaml to restrict access by the CI/CD job's identity:

ci_access:
  projects:
    - id: path/to/project
      access_as:
        ci_job: {}

Example RBAC to restrict CI/CD jobs

The following RoleBinding resource restricts all CI/CD jobs to view rights only.

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: ci-job-view
roleRef:
  name: view
  kind: ClusterRole
  apiGroup: rbac.authorization.k8s.io
subjects:
  - name: gitlab:ci_job
    kind: Group

Impersonate a static identity

For a given connection, you can use a static identity for the impersonation.

Under the access_as key, add the impersonate key to make the request using the provided identity.

The identity can be specified with the following keys:

  • username (required)
  • uid
  • groups
  • extra

See the official Kubernetes documentation for details.

Restrict project and group access to specific environments

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

By default, if your agent is available to a project, all of the project's CI/CD jobs can use that agent.

To restrict access to the agent to only jobs with specific environments, add environments to ci_access.projects or ci_access.groups. For example:

ci_access:
  projects:
    - id: path/to/project-1
    - id: path/to/project-2
      environments:
        - staging
        - review/*
  groups:
    - id: path/to/group-1
      environments:
        - production

In this example:

  • All CI/CD jobs under project-1 can access the agent.
  • CI/CD jobs under project-2 with staging or review/* environments can access the agent.
    • * is a wildcard, so review/* matches all environments under review.
  • CI/CD jobs for projects under group-1 with production environments can access the agent.

Related topics

Troubleshooting

Grant write permissions to ~/.kube/cache

Tools like kubectl, Helm, kpt, and kustomize cache information about the cluster in ~/.kube/cache. If this directory is not writable, the tool fetches information on each invocation, making interactions slower and creating unnecessary load on the cluster. For the best experience, in the image you use in your .gitlab-ci.yml file, ensure this directory is writable.

Enable TLS

If you are on a self-managed GitLab instance, ensure your instance is configured with Transport Layer Security (TLS).

If you attempt to use kubectl without TLS, you might get an error like:

$ kubectl get pods
error: You must be logged in to the server (the server has asked for the client to provide credentials)

Unable to connect to the server: certificate signed by unknown authority

If you use an environment with KAS and a self-signed certificate, your kubectl call might return this error:

kubectl get pods
Unable to connect to the server: x509: certificate signed by unknown authority

The error occurs because the job does not trust the certificate authority (CA) that signed the KAS certificate.

To resolve the issue, configure kubectl to trust the CA.

Validation errors

If you use kubectl versions v1.27.0 or v.1.27.1, you might get the following error:

error: error validating "file.yml": error validating data: the server responded with the status code 426 but did not return more information; if you choose to ignore these errors, turn validation off with --validate=false

This issue is caused by a bug with kubectl and other tools that use the shared Kubernetes libraries.

To resolve the issue, use another version of kubectl.