Skip to content

GitLab Kubernetes Agent (PREMIUM ONLY)

WARNING: This feature might not be available to you. Check the version history note above for details.

The GitLab Kubernetes Agent is an active in-cluster component for solving GitLab and Kubernetes integration tasks in a secure and cloud-native way. It enables:

  • Integrating GitLab with a Kubernetes cluster behind a firewall or NAT (network address translation).
  • Pull-based GitOps deployments by leveraging the GitOps Engine.
  • Real-time access to API endpoints within a cluster.

Many more features are planned. Please review our roadmap.

GitLab Agent GitOps workflow

The GitLab Agent uses multiple GitLab projects to provide a flexible workflow that can suit various needs. This diagram shows these repositories and the main actors involved in a deployment:

sequenceDiagram
  participant D as Developer
  participant A as Application code repository
  participant M as Manifest repository
  participant K as Kubernetes agent
  participant C as Agent configuration repository
  K->C: Grab the configuration
  D->>+A: Pushing code changes
  A->>M: Updating manifest
  loop Regularly
    K-->>M: Watching changes
    M-->>K: Pulling and applying changes
  end

There are several components that work in concert for the Agent to accomplish GitOps deployments:

  • A properly-configured Kubernetes cluster.
  • A configuration repository that contains a config.yaml file, which tells the Agent which repositories to synchronize with.
  • A manifest repository that contains a manifest.yaml, which is tracked by the Agent and can be auto-generated. Any changes to manifest.yaml are applied to the cluster.

These repositories might be the same GitLab project or separate projects.

NOTE: GitLab recommends you use the same GitLab project for the agent configuration and manifest repositories. Our backlog contains issues for adding support for private manifest repositories outside of the configuration project and group level agents.

For more details, please refer to our full architecture documentation in the Agent project.

Get started with GitOps and the GitLab Agent

The setup process involves a few steps to enable GitOps deployments:

  1. Install the Agent server.
  2. Define a configuration repository.
  3. Create an Agent record in GitLab.
  4. Generate and copy a Secret token used to connect to the Agent.
  5. Install the Agent into the cluster.
  6. Create a manifest.yaml.

Upgrades and version compatibility

As the GitLab Kubernetes Agent is a new product, we are constantly adding new features to it. As a result, while shipped features are production ready, its internal API is neither stable nor versioned yet. For this reason, GitLab only guarantees compatibility between corresponding major.minor (X.Y) versions of GitLab and its cluster side component, agentk.

Upgrade your agent installations together with GitLab upgrades. To decide which version of agentkto install follow:

  1. Open the GITLAB_KAS_VERSION file from the GitLab Repository, which contains the latest agentk version associated with the master branch.
  2. Change the master branch and select the Git tag associated with your version. For instance, you could change it to GitLab v13.5.3-ee release

The available agentk and kas versions can be found in the container registry.

Install the Kubernetes Agent Server

The GitLab Kubernetes Agent Server (KAS) can be deployed using Omnibus GitLab or the GitLab chart. If you don't already have GitLab installed, please refer to our installation documentation.

NOTE: GitLab plans to include the KAS on GitLab.com.

Install with Omnibus

When using the Omnibus GitLab package:

  1. Edit /etc/gitlab/gitlab.rb:

    gitlab_kas['enable'] = true

  2. Reconfigure GitLab.

To configure any additional options related to GitLab Kubernetes Agent Server, refer to the Enable GitLab KAS section of the gitlab.rb.template.

Install with the Helm chart

When installing or upgrading the GitLab Helm chart, consider the following Helm v3 example. If you're using Helm v2, you must modify this example. See our notes regarding deploy with Helm.

You must set global.kas.enabled=true for the KAS to be properly installed and configured:

helm repo add gitlab https://charts.gitlab.io/
helm repo update
helm upgrade --install gitlab gitlab/gitlab \
  --timeout 600s \
  --set global.hosts.domain=<YOUR_DOMAIN> \
  --set global.hosts.externalIP=<YOUR_IP> \
  --set certmanager-issuer.email=<YOUR_EMAIL> \
  --set global.kas.enabled=true

To specify other options related to the KAS sub-chart, create a gitlab.kas sub-section of your values.yaml file:

gitlab:
  kas:
    # put your KAS custom options here

For details, read Using the GitLab-KAS chart.

Define a configuration repository

Next, you need a GitLab repository to contain your Agent configuration. The minimal repository layout looks like this:

.gitlab/agents/<agent-name>/config.yaml

Your config.yaml file can specify multiple manifest projects in the section manifest_projects:

gitops:
  manifest_projects:
  - id: "path-to/your-manifest-project-number1"
  ...

GitLab versions 13.7 and later also supports manifest projects containing multiple directories (or subdirectories) of YAML files. For more information see our documentation on the Kubernetes Agent configuration respository.

Create an Agent record in GitLab

Next, create an GitLab Rails Agent record so the Agent can associate itself with the configuration repository project. Creating this record also creates a Secret needed to configure the Agent in subsequent steps. You can create an Agent record either:

  • Through the Rails console:

    project = ::Project.find_by_full_path("path-to/your-configuration-project")
# agent-name should be the same as specified above in the config.yaml
agent = ::Clusters::Agent.create(name: "<agent-name>", project: project)
token = ::Clusters::AgentToken.create(agent: agent)
token.token # this will print out the token you need to use on the next step

    For full details, read Starting a Rails console session.

  • Through GraphQL: (PREMIUM ONLY)

    mutation createAgent {
  # agent-name should be the same as specified above in the config.yaml
  createClusterAgent(input: { projectPath: "path-to/your-configuration-project", name: "<agent-name>" }) {
    clusterAgent {
      id
      name
    }
    errors
  }
}

mutation createToken {
  clusterAgentTokenCreate(input: { clusterAgentId: <cluster-agent-id-taken-from-the-previous-mutation> }) {
    secret # This is the value you need to use on the next step
    token {
      createdAt
      id
    }
    errors
  }
}

    NOTE: GraphQL only displays the token one time after creating it.

    If you are new to using the GitLab GraphQL API, refer to the Getting started with the GraphQL API page, or the GraphQL Explorer.

Create the Kubernetes secret

After generating the token, you must apply it to the Kubernetes cluster.

  1. If you haven't previously defined or created a namespace, run the following command:

    kubectl create namespace <YOUR-DESIRED-NAMESPACE>

  2. Run the following command to create your Secret:

    kubectl create secret generic -n <YOUR-DESIRED-NAMESPACE> gitlab-agent-token --from-literal=token='YOUR_AGENT_TOKEN'

Install the Agent into the cluster

Next, install the in-cluster component of the Agent. This example file contains the Kubernetes resources required for the Agent to be installed. You can modify this example resources.yml file in the following ways:

  • Replace namespace: gitlab-agent with namespace: <YOUR-DESIRED-NAMESPACE>.
  • You can configure kas-address (Kubernetes Agent Server) in several ways. The agent can use the WebSockets or gRPC protocols to connect to the Agent Server. Select the option appropriate for your cluster configuration and GitLab architecture:
    • The wss scheme (an encrypted WebSockets connection) is specified by default after you install the gitlab-kas sub-chart, or enable gitlab-kas for Omnibus GitLab. When using the sub-chart, you must set wss://kas.host.tld:443 as kas-address, where host.tld is the domain you've setup for your GitLab installation. When using Omnibus GitLab, you must set wss://GitLab.host.tld:443/-/kubernetes-agent as kas-address, where GitLab.host.tld is your GitLab hostname.
    • When using the sub-chart, specify the ws scheme (such as ws://kas.host.tld:80) to use an unencrypted WebSockets connection. When using the Omnibus GitLab, specify the ws scheme (such as ws://GitLab.host.tld:80/-/kubernetes-agent).
    • Specify the grpc scheme if both Agent and Server are installed in one cluster. In this case, you may specify kas-address value as grpc://gitlab-kas.<your-namespace>:5005) to use gRPC directly, where gitlab-kas is the name of the service created by gitlab-kas chart, and your-namespace is the namespace where the chart was installed. Encrypted gRPC is not supported yet. Follow the Support TLS for gRPC communication issue for progress updates.
    • When deploying KAS through the GitLab chart, it's possible to customize the kas-address for wss and ws schemes to whatever you need. Check the chart's KAS Ingress docs to learn more about it.
    • In the near future, Omnibus GitLab intends to provision gitlab-kas under a sub-domain by default, instead of the /-/kubernetes-agent path. Please follow this issue for details.
  • If you defined your own secret name, replace gitlab-agent-token with your secret name in the secretName: section.

To apply this file, run the following command:

kubectl apply -n <YOUR-DESIRED-NAMESPACE> -f ./resources.yml

To review your configuration, run the following command:

$ kubectl get pods -n <YOUR-DESIRED-NAMESPACE>

NAMESPACE     NAME                               READY   STATUS    RESTARTS   AGE
gitlab-agent  gitlab-agent-77689f7dcb-5skqk      1/1     Running   0          51s

Example resources.yml file

apiVersion: v1
kind: ServiceAccount
metadata:
  name: gitlab-agent
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: gitlab-agent
spec:
  replicas: 1
  selector:
    matchLabels:
      app: gitlab-agent
  template:
    metadata:
      labels:
        app: gitlab-agent
    spec:
      serviceAccountName: gitlab-agent
      containers:
      - name: agent
        image: "registry.gitlab.com/gitlab-org/cluster-integration/gitlab-agent/agentk:latest"
        args:
        - --token-file=/config/token
        - --kas-address
        - wss://kas.host.tld:443 # change this line for the one below if using Omnibus GitLab
        # - wss://gitlab.host.tld:443/-/kubernetes-agent
        volumeMounts:
        - name: token-volume
          mountPath: /config
      volumes:
      - name: token-volume
        secret:
          secretName: gitlab-agent-token
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxSurge: 0
      maxUnavailable: 1
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: gitlab-agent-write
rules:
- resources:
  - '*'
  apiGroups:
  - '*'
  verbs:
  - create
  - update
  - delete
  - patch
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: gitlab-agent-write-binding
roleRef:
  name: gitlab-agent-write
  kind: ClusterRole
  apiGroup: rbac.authorization.k8s.io
subjects:
- name: gitlab-agent
  kind: ServiceAccount
  namespace: gitlab-agent
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: gitlab-agent-read
rules:
- resources:
  - '*'
  apiGroups:
  - '*'
  verbs:
  - get
  - list
  - watch
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: gitlab-agent-read-binding
roleRef:
  name: gitlab-agent-read
  kind: ClusterRole
  apiGroup: rbac.authorization.k8s.io
subjects:
- name: gitlab-agent
  kind: ServiceAccount
  namespace: gitlab-agent

Create a manifest.yaml

In a previous step, you configured a config.yaml to point to the GitLab projects the Agent should synchronize. In each of those projects, you must create a manifest.yaml file for the Agent to monitor. You can auto-generate this manifest.yaml with a templating engine or other means.

The agent is authorized to download manifests for the configuration project, and public projects. Support for other private projects is planned in the issue Agent authorization for private manifest projects.

Each time you commit and push a change to this file, the Agent logs the change:

2020-09-15_14:09:04.87946 gitlab-k8s-agent      : time="2020-09-15T10:09:04-04:00" level=info msg="Config: new commit" agent_id=1 commit_id=e6a3651f1faa2e928fe6120e254c122451be4eea

Example manifest.yaml file

This file creates an NGINX deployment.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-deployment
  namespace: gitlab-agent  # Can be any namespace managed by you that the agent has access to.
spec:
  selector:
    matchLabels:
      app: nginx
  replicas: 2
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - name: nginx
        image: nginx:1.14.2
        ports:
        - containerPort: 80

Example projects

The following example projects can help you get started with the Kubernetes Agent.

Simple NGINX deployment

This basic GitOps example deploys NGINX:

Deploying GitLab Runner with the Agent

These instructions assume that the Agent is already set up as described in the Get started with GitOps:

  1. Check the possible Runner chart YAML values on the Runner chart documentation, and create a runner-chart-values.yaml file with the configuration that fits your needs, such as:

    ## The GitLab Server URL (with protocol) that want to register the runner against
## ref: https://docs.gitlab.com/runner/commands/README.html#gitlab-runner-register
##
gitlabUrl: https://gitlab.my.domain.com/

## The Registration Token for adding new Runners to the GitLab Server. This must
## be retrieved from your GitLab Instance.
## ref: https://docs.gitlab.com/ce/ci/runners/README.html
##
runnerRegistrationToken: "XXXXXXYYYYYYZZZZZZ"

## For RBAC support:
rbac:
  create: true

## Run all containers with the privileged flag enabled
## This will allow the docker:dind image to run if you need to run Docker
## commands. Please read the docs before turning this on:
## ref: https://docs.gitlab.com/runner/executors/kubernetes.html#using-dockerdind
runners:
  privileged: true

  2. Create a single manifest file to install the Runner chart with your cluster agent:

    helm template --namespace gitlab gitlab-runner -f runner-chart-values.yaml gitlab/gitlab-runner > manifest.yaml

  3. Push your manifest.yaml to your manifest repository.

Troubleshooting

If you face any issues while using GitLab Kubernetes Agent, you can read the service logs with the following commands:

  • KAS pod logs - Tail these logs with the kubectl logs -f -l=app=kas -n <YOUR-GITLAB-NAMESPACE> command. In Omnibus GitLab, the logs reside in /var/log/gitlab/gitlab-kas/.
  • Agent pod logs - Tail these logs with the kubectl logs -f -l=app=gitlab-agent -n <YOUR-DESIRED-NAMESPACE> command.

KAS logs - GitOps: failed to get project info

{"level":"warn","time":"2020-10-30T08:37:26.123Z","msg":"GitOps: failed to get project info","agent_id":4,"project_id":"root/kas-manifest001","error":"error kind: 0; status: 404"}

This error is shown if the specified manifest project root/kas-manifest001 doesn't exist, or if a project is private. To fix it, make sure the project exists and its visibility is set to public.

KAS logs - Configuration file not found

time="2020-10-29T04:44:14Z" level=warning msg="Config: failed to fetch" agent_id=2 error="configuration file not found: \".gitlab/agents/test-agent/config.yaml\

This error is shown if the path to the configuration project was specified incorrectly, or if the path to config.yaml inside the project is not valid.

Agent logs - Transport: Error while dialing failed to WebSocket dial

{"level":"warn","time":"2020-11-04T10:14:39.368Z","msg":"GetConfiguration failed","error":"rpc error: code = Unavailable desc = connection error: desc = \"transport: Error while dialing failed to WebSocket dial: failed to send handshake request: Get \\\"https://gitlab-kas:443/-/kubernetes-agent\\\": dial tcp: lookup gitlab-kas on 10.60.0.10:53: no such host\""}

This error is shown if there are some connectivity issues between the address specified as kas-address, and your Agent pod. To fix it, make sure that you specified the kas-address correctly.

Agent logs - ValidationError(Deployment.metadata

{"level":"info","time":"2020-10-30T08:56:54.329Z","msg":"Synced","project_id":"root/kas-manifest001","resource_key":"apps/Deployment/kas-test001/nginx-deployment","sync_result":"error validating data: [ValidationError(Deployment.metadata): unknown field \"replicas\" in io.k8s.apimachinery.pkg.apis.meta.v1.ObjectMeta, ValidationError(Deployment.metadata): unknown field \"selector\" in io.k8s.apimachinery.pkg.apis.meta.v1.ObjectMeta, ValidationError(Deployment.metadata): unknown field \"template\" in io.k8s.apimachinery.pkg.apis.meta.v1.ObjectMeta]"}

This error is shown if your manifest.yaml file is malformed, and Kubernetes can't create specified objects. Make sure that your manifest.yaml file is valid. You may try using it to create objects in Kubernetes directly for more troubleshooting.

Agent logs - Error while dialing failed to WebSocket dial: failed to send handshake request

{"level":"warn","time":"2020-10-30T09:50:51.173Z","msg":"GetConfiguration failed","error":"rpc error: code = Unavailable desc = connection error: desc = \"transport: Error while dialing failed to WebSocket dial: failed to send handshake request: Get \\\"https://GitLabhost.tld:443/-/kubernetes-agent\\\": net/http: HTTP/1.x transport connection broken: malformed HTTP response \\\"\\\\x00\\\\x00\\\\x06\\\\x04\\\\x00\\\\x00\\\\x00\\\\x00\\\\x00\\\\x00\\\\x05\\\\x00\\\\x00@\\\\x00\\\"\""}

This error is shown if you configured wss as kas-address on the agent side, but KAS on the server side is not available via wss. To fix it, make sure the same schemes are configured on both sides.

It's not possible to set the grpc scheme due to the issue It is not possible to configure KAS to work with grpc without directly editing GitLab KAS deployment. To use grpc while the issue is in progress, directly edit the deployment with the kubectl edit deployment gitlab-kas command, and change --listen-websocket=true to --listen-websocket=false. After running that command, you should be able to use grpc://gitlab-kas.<YOUR-NAMESPACE>:5005.

Agent logs - Decompressor is not installed for grpc-encoding

{"level":"warn","time":"2020-11-05T05:25:46.916Z","msg":"GetConfiguration.Recv failed","error":"rpc error: code = Unimplemented desc = grpc: Decompressor is not installed for grpc-encoding \"gzip\""}

This error is shown if the version of the agent is newer that the version of KAS. To fix it, make sure that both agentk and KAS use the same versions.