Skip to main content

Configuring Kubernetes integration

Configuring the Backstage Kubernetes integration involves two steps:

  1. Enabling the backend to collect objects from your Kubernetes cluster(s).
  2. Surfacing your Kubernetes objects in catalog entities

Configuring Kubernetes Clusters

The following is a full example entry in app-config.yaml:

type: 'multiTenant'
- type: 'config'
- url:
name: minikube
authProvider: 'serviceAccount'
skipTLSVerify: false
skipMetricsLookup: true
serviceAccountToken: ${K8S_MINIKUBE_TOKEN}
dashboardUrl: # url copied from running the command: minikube service kubernetes-dashboard -n kubernetes-dashboard
dashboardApp: standard
caFile: '' # local path to CA file
- group: ''
apiVersion: 'v1alpha1'
plural: 'rollouts'
- url:
name: aws-cluster-1
title: 'My AWS Cluster Number One'
authProvider: 'aws'
- type: 'gke'
projectId: 'gke-clusters'
region: 'europe-west1'
skipTLSVerify: true
skipMetricsLookup: true
exposeDashboard: true


This configures how to determine which clusters a component is running in.

Valid values are:

  • multiTenant - This configuration assumes that all components run on all the provided clusters.

  • singleTenant - This configuration assumes that current component run on one cluster in provided clusters.


This is an array used to determine where to retrieve cluster configuration from.

Valid cluster locator methods are:


This cluster locator method will gather Resources of type kubernetes-cluster from the catalog and treat them as clusters for the purposes of the Kubernetes plugin. In order for a resource to be detected by this method, it must also have the following annotations (as seen here in the code):

There are many other annotations that can be applied to a cluster resource to configure the way Backstage communicates, documented here in the API reference. Here is a YAML snippet illustrating an example of a cluster in the catalog:

kind: Resource
name: my-cluster
annotations: '' # base64-encoded CA 'oidc' 'microsoft' 'true'
type: kubernetes-cluster
owner: user:guest

Note that it is insecure to store a Kubernetes service account token in an annotation on a catalog entity (where it could easily be accidentally revealed by the catalog API) -- therefore there is no annotation corresponding to the serviceAccountToken field used by the config cluster locator. Accordingly, the catalog cluster locator does not support the serviceAccount auth strategy.

This method can be quite helpful when used in combination with an ingestion procedure like the GkeEntityProvider (installation documented here) or the AwsEKSClusterProcessor to automatically update the set of clusters tracked by Backstage.


This cluster locator method will read cluster information from your app-config (see below).


Used by the config cluster locator method to construct Kubernetes clients.


The base URL to the Kubernetes control plane. Can be found by using the "Kubernetes master" result from running the kubectl cluster-info command.


A name to represent this cluster, this must be unique within the clusters array. Users will see this value in the Software Catalog Kubernetes plugin.


A human-readable name for the cluster. This value will override the name field for the purposes of display in the catalog.


This determines how the Kubernetes client authenticates with the Kubernetes cluster. Valid values are:

aksThis will use a user's AKS access token from the Microsoft auth provider to access the Kubernetes API on AKS clusters.
awsThis will use AWS credentials to access resources in EKS clusters
azureThis will use Azure Identity to access resources in clusters
googleThis will use a user's Google access token from the Google auth provider to access the Kubernetes API on GKE clusters.
googleServiceAccountThis will use the Google Cloud service account credentials to access resources in clusters
oidcThis will use Oidc Tokens to authenticate to the Kubernetes API. When this is used the oidcTokenProvider field should also be set. Please note the cluster must support OIDC, at the time of writing AKS clusters do not support OIDC.
serviceAccountThis will use a Kubernetes service account to access the Kubernetes API. When this is used the serviceAccountToken field should also be set, or else Backstage should be running in-cluster.

Check the Kubernetes Authentication section for additional explanation.


This determines whether the Kubernetes client verifies the TLS certificate presented by the API server. Defaults to false.


This determines whether the Kubernetes client looks up resource metrics CPU/Memory for pods returned by the API server. Defaults to false.

clusters.\*.serviceAccountToken (optional)

The service account token to be used when using the serviceAccount auth provider. Note that, unless you have an effective credential rotation procedure in place or have a single Kubernetes cluster running both Backstage and all your services, this auth provider is probably not ideal for production.

Assuming you have already created a service account named SERVICE_ACCOUNT_NAME in namespace NAMESPACE and it has adequate permissions, here are some sample procedures to procure a long-lived service account token for use with this provider:

  • On versions of Kubernetes prior to 1.24, you could get an (automatically-generated) token for a service account with:

    kubectl -n <NAMESPACE> get secret $(kubectl -n <NAMESPACE> get sa <SERVICE_ACCOUNT_NAME> -o=json \
    | jq -r '.secrets[0].name') -o=json \
    | jq -r '.data["token"]' \
    | base64 --decode
  • For Kubernetes 1.24+, as described in this guide, you can obtain a long-lived token by creating a secret:

    kubectl apply -f - <<EOF
    apiVersion: v1
    kind: Secret
    name: <SECRET_NAME>
    namespace: <NAMESPACE>
    annotations: <SERVICE_ACCOUNT_NAME>

    waiting for the token controller to populate a token, and retrieving it with:

    kubectl -n <NAMESPACE> get secret <SECRET_NAME> -o go-template='{{.data.token | base64decode}}'

If a cluster has authProvider: serviceAccount and the serviceAccountToken field is omitted, Backstage will ignore the configured URL and certificate data, instead attempting to access the Kubernetes API via an in-cluster client as in this example.

clusters.\*.oidcTokenProvider (optional)

This field is to be used when using the oidc auth provider. It will use the id tokens from a configured backstage auth provider to authenticate to the cluster. The selected oidcTokenProvider needs to be properly configured under auth for this to work.

- type: 'config'
- name: test-cluster
url: http://localhost:8080
authProvider: oidc
oidcTokenProvider: okta # This value needs to match a config under auth.providers

The following values are supported out-of-the-box by the frontend: gitlab (the application whose clientId is used by the auth provider should be granted the openid scope), google, microsoft, okta, onelogin.

Take note that oidcTokenProvider is just the issuer for the token, you can use any of these with an OIDC enabled cluster, like using microsoft as the issuer for a EKS cluster.

clusters.\*.dashboardUrl (optional)

Specifies the link to the Kubernetes dashboard managing this cluster.

Note that you should specify the app used for the dashboard using the dashboardApp property, in order to properly format links to kubernetes resources, otherwise it will assume that you're running the standard one.

Note also that this attribute is optional for some kinds of dashboards, such as GKE, which requires additional parameters specified in the dashboardParameters option.

clusters.\*.dashboardApp (optional)

Specifies the app that provides the Kubernetes dashboard.

This will be used for formatting links to kubernetes objects inside the dashboard.

The supported dashboards are: standard, rancher, openshift, gke, aks, eks. However, not all of them are implemented yet, so please contribute!

Note that it will default to the regular dashboard provided by the Kubernetes project (standard), that can run in any Kubernetes cluster.

Note that for the gke app, you must provide additional information in the dashboardParameters option.

Note that you can add your own formatter by registering it to the clusterLinksFormatters dictionary, in the app project.


import { clusterLinksFormatters } from '@backstage/plugin-kubernetes';
clusterLinksFormatters.myDashboard = (options) => ...;

See also for real examples.

clusters.\*.dashboardParameters (optional)

Specifies additional information for the selected dashboardApp formatter.

Note that, even though dashboardParameters is optional, it might be mandatory for some dashboards, such as GKE.

required parameters for GKE
projectIdthe ID of the GCP project containing your Kubernetes clusters
regionthe region of GCP containing your Kubernetes clusters
clusterNamethe name of your kubernetes cluster, within your projectId GCP project

Note that the GKE cluster locator can automatically provide the values for the dashboardApp and dashboardParameters options if you set the exposeDashboard property to true.


type: 'multiTenant'
- type: 'config'
- url:
name: my-cluster
dashboardApp: gke
projectId: my-project
region: us-east1
clusterName: my-cluster
clusters.\*.caData (optional)

Base64-encoded certificate authority bundle in PEM format. The Kubernetes client will verify that the TLS certificate presented by the API server is signed by this CA.

This value could be obtained via inspecting the kubeconfig file (usually at ~/.kube/config) under clusters[*].cluster.certificate-authority-data. For GKE, execute the following command to obtain the value

gcloud container clusters describe <YOUR_CLUSTER_NAME> \

See also for complete docs about GKE without gcloud.

clusters.\*.caFile (optional)

Filesystem path (on the host where the Backstage process is running) to a certificate authority bundle in PEM format. The Kubernetes client will verify that the TLS certificate presented by the API server is signed by this CA. Note that only clusters defined in the app-config via the config cluster locator method can be configured in this way.

clusters.\*.customResources (optional)

Configures which custom resources to look for when returning an entity's Kubernetes resources belonging to the cluster. Same specification as customResources


This cluster locator is designed to work with Kubernetes clusters running in GKE. It will configure the Kubernetes backend plugin to make requests to clusters running within a Google Cloud project.

This cluster locator method will use the google authentication mechanism.

The Google Cloud service account to use can be configured through the GOOGLE_APPLICATION_CREDENTIALS environment variable. Consult the Google Cloud docs for more information.

For example:

- type: 'gke'
projectId: 'gke-clusters'
region: 'europe-west1' # optional
authProvider: 'google' # optional
skipTLSVerify: false # optional
skipMetricsLookup: false # optional
exposeDashboard: false # optional
matchingResourceLabels: # optional
- key: 'environment'
value: 'production'

Will configure the Kubernetes plugin to connect to all GKE clusters in the project gke-clusters in the region europe-west1.

Note that the GKE cluster locator can automatically provide the values for the dashboardApp and dashboardParameters options if you enable the exposeDashboard option.


The Google Cloud project to look for Kubernetes clusters in.

region (optional)

The Google Cloud region to look for Kubernetes clusters in. Defaults to all regions.

authProvider (optional)

Set the authentication method for discovering clusters and gathering information about resources.

Defaults to google which leverages the logged in user's Google OAuth credentials.

Set to googleServiceAccount to leverage Application Default Credentials ( To use a service account JSON key (not recommended), set the GOOGLE_APPLICATION_CREDENTIALS environment variable on the Backstage backend to the path of the service account key file.

skipTLSVerify (optional)

This determines whether the Kubernetes client verifies the TLS certificate presented by the API server. Defaults to false.

skipMetricsLookup (optional)

This determines whether the Kubernetes client looks up resource metrics CPU/Memory for pods returned by the API server. Defaults to false.

exposeDashboard (optional)

This determines whether the dashboardApp and dashboardParameters should be automatically configured in order to expose the GKE dashboard from the Kubernetes plugin.

Defaults to false.

matchingResourceLabels (optional)

Array of key value labels used to filter out clusters which don't have the matching resource labels.


This cluster locator method will assume a locally running kubectl proxy process using the default port (8001).

NOTE: This cluster locator method is for local development only and should not be used in production.

Custom KubernetesClustersSupplier

If the configuration-based cluster locators do not work for your use-case, it is also possible to implement a custom KubernetesClustersSupplier.

customResources (optional)

Configures which custom resources to look for by default when returning an entity's Kubernetes resources.


  • The optional kubernetes.customResources property is overrode by customResources at the clusters level.

Defaults to empty array. Example:

- group: ''
apiVersion: 'v1alpha1'
plural: 'rollouts'


The custom resource's group.


The custom resource's apiVersion.


The plural representing the custom resource.

apiVersionOverrides (optional)

Overrides for the API versions used to make requests for the corresponding objects. If using a legacy Kubernetes version, you may use this config to override the default API versions to ones that are supported by your cluster.


cronjobs: 'v1beta1'

For more information on which API versions are supported by your cluster, please view the Kubernetes API docs for your Kubernetes version (e.g. API Groups for v1.22 )

objectTypes (optional)

Overrides for the Kubernetes object types fetched from the cluster. The default object types are:

  • pods
  • services
  • configmaps
  • limitranges
  • resourcequotas
  • deployments
  • replicasets
  • horizontalpodautoscalers
  • jobs
  • cronjobs
  • ingresses
  • statefulsets
  • daemonsets

You may use this config to override the default object types if you only want a subset of the default ones. However, it's currently not supported to fetch object types other than the ones specified in the default types.


- configmaps
- deployments
- limitranges
- pods
- services
- statefulsets

Role Based Access Control

The current RBAC permissions required are read-only cluster wide, the below Kubernetes manifest describes which objects are required and will ensure the plugin functions correctly:

kind: ClusterRole
name: backstage-read-only
- apiGroups:
- '*'
- pods
- configmaps
- services
- deployments
- replicasets
- horizontalpodautoscalers
- ingresses
- statefulsets
- limitranges
- resourcequotas
- daemonsets
- get
- list
- watch
- apiGroups:
- batch
- jobs
- cronjobs
- get
- list
- watch
- apiGroups:
- pods
- get
- list

Surfacing your Kubernetes components as part of an entity

There are two ways to surface your Kubernetes components as part of an entity. The label selector takes precedence over the annotation/service id.

Common label

Adding the entity annotation

In order for Backstage to detect that an entity has Kubernetes components, the following annotation should be added to the entity's catalog-info.yaml:

'': dice-roller

Adding the namespace annotation

Entities can have the annotation, this will cause the entity's Kubernetes resources to by looked up via that namespace.

'': dice-space

Labeling Kubernetes components

In order for Kubernetes components to show up in the software catalog as a part of an entity, Kubernetes components themselves can have the following label:


Label selector query annotation

You can write your own custom label selector query that Backstage will use to lookup the objects (similar to kubectl --selector="your query here"). Review the labels and selectors Kubernetes documentation for more info.

'': 'app=my-app,component=front-end'

Cluster Selection annotation

This is applicable only for singleTenant serviceLocatorMethod.

You can now select single kubernetes cluster that the entity is part-of from all your defined kubernetes clusters. To apply this use the following annotation.

SingleTenant Cluster:

'': dice-cluster

In the example above, we configured the "" annotation on the entity catalog-info.yaml file to specify that the current component is running in a single cluster called "dice-cluster", so this cluster must have been specified in the app-config.yaml, under the Kubernetes clusters configuration (for more details, see Configuring Kubernetes clusters).

If you do not specify the annotation, by default Backstage fetches from all defined Kubernetes clusters.