GitLab Kubernetes Agent

Version history
cautionThis 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 in a cluster.
  • Alert generation based on Container network policy.

Many more features are planned. Please review our roadmap and our development documentation.

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 where the Agent is running.
  • A configuration repository that contains a config.yaml file, which tells the Agent which repositories to synchronize with the cluster.
  • A manifest repository that contains manifest files. Any changes to manifest files are applied to the cluster.

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

noteGitLab 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 manifest files.

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.

noteGitLab 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 repository.

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:

    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
      }
    }
    
    noteGraphQL 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 documentation 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 manifest files

In a previous step, you configured a config.yaml to point to the GitLab projects the Agent should synchronize. Agent monitors each of those projects for changes to the manifest files it contains. You can auto-generate manifest files 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 push a change to a monitored manifest repository, 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 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.

Deploying GitLab Runner with the Agent

You can use the Kubernetes Agent to deploy GitLab Runner in a Kubernetes cluster.

Kubernetes Network Security Alerts

The GitLab Agent also provides an integration with Cilium. This integration provides a simple way to generate network policy-related alerts and to surface those alerts in GitLab.

There are several components that work in concert for the Agent to generate the alerts:

  • A working Kubernetes cluster.
  • Cilium integration through either of these options:
  • One or more network policies through any of these options:
  • Use a configuration repository to inform the Agent through a config.yaml file, which repositories can synchronize with. This repository might be the same, or a separate GitLab project.

The setup process follows the same steps as GitOps, with the following differences:

  • When you define a configuration repository, you must do so with Cilium settings.
  • You do not need to create a manifest.yaml.

Define a configuration repository with Cilium settings

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 must specify the host and port of your Hubble Relay service. If your Cilium integration was performed through GitLab Managed Apps, you can use hubble-relay.gitlab-managed-apps.svc.cluster.local:80:

cilium:
  hubble_relay_address: "<hubble-relay-host>:<hubble-relay-port>"
  ...

Management interfaces

Users with at least the Developer can access the user interface for the GitLab Kubernetes agent at Operations > Kubernetes and selecting the GitLab Agent managed clusters tab. This page lists all registered agents for the current project, and the configuration directory for each agent:

GitLab Kubernetes Agent list UI

Additional management interfaces are planned for the GitLab Kubernetes Agent. Provide more feedback in the related epic.

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 a manifest file is malformed, and Kubernetes can’t create specified objects. Make sure that your manifest files are valid. You may try using them 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.