Authorize actions in clusters using role-based access control

This page shows you how to authorize actions on resources in your Google Kubernetes Engine (GKE) clusters using the built-in role-based access control (RBAC) mechanism in Kubernetes.

RBAC is a core security feature in Kubernetes that lets you create fine-grained permissions to manage what actions users and workloads can perform on resources in your clusters. As a platform administrator, you create RBAC roles and bind those roles to subjects, which are authenticated users such as service accounts or Groups. Kubernetes RBAC is enabled by default.

Before you begin

Before you start, make sure you have performed the following tasks:

  • Enable the Google Kubernetes Engine API.
  • Enable Google Kubernetes Engine API
  • If you want to use the Google Cloud CLI for this task, install and then initialize the gcloud CLI. If you previously installed the gcloud CLI, get the latest version by running gcloud components update.

Interaction with Identity and Access Management

You can use both Identity and Access Management (IAM) and Kubernetes RBAC to control access to your GKE cluster:

  • IAM is not specific to Kubernetes; it provides identity management for multiple Google Cloud products, and operates primarily at the level of the Google Cloud project.

  • Kubernetes RBAC is a core component of Kubernetes and lets you create and grant roles (sets of permissions) for any object or type of object within the cluster.

  • To authorize an action, GKE checks for an RBAC policy first. If there isn't an RBAC policy, GKE checks for IAM permissions.

In GKE, IAM and Kubernetes RBAC are integrated to authorize users to perform actions if they have sufficient permissions according to either tool. This is an important part of bootstrapping a GKE cluster, since by default Google Cloud users do not have any Kubernetes RBAC RoleBindings.

To authorize users using Google Cloud accounts, the client must be correctly configured to authenticate using those accounts first. For example, if you are using kubectl, you must configure the kubectl command to authenticate to Google Cloud before running any commands that require authorization.

In almost all cases, Kubernetes RBAC can be used instead of IAM. GKE users require at minimum, the container.clusters.get IAM permission in the project that contains the cluster. This permission is included in the container.clusterViewer role, and in other more highly privileged roles. The container.clusters.get permission is required for users to authenticate to the clusters in the project, but does not authorize them to perform any actions inside those clusters. Authorization may then be provided by either IAM or Kubernetes RBAC.

Define and assign permissions

You can define RBAC rules in ClusterRole and Role objects, and then assign those rules with ClusterRoleBinding and RoleBinding objects as follows:

  • ClusterRole: a cluster-level grouping of resources and allowed operations that you can assign to a user or a group using a RoleBinding or a ClusterRoleBinding.
  • Role: a namespaced grouping of resources and allowed operations that you can assign to a user or a group of users using a RoleBinding.
  • ClusterRoleBinding: assign a ClusterRole to a user or a group for all namespaces in the cluster.
  • RoleBinding: assign a Role or a ClusterRole to a user or a group within a specific namespace.

When you use a RoleBinding to assign a ClusterRole to a user or group, those users and groups can only access resources in the namespace you specify in the RoleBinding. If you want the users or groups to access resource across all namespaces, use a ClusterRoleBinding instead.

Define permissions using Roles or ClusterRoles

You define permissions within a Role or ClusterRole object. A Role defines access to resources within a single Namespace, while a ClusterRole defines access to resources in the entire cluster.

Roles and ClusterRoles have the same syntax. Each has a rules section, where you define the resources the rule applies to and allowed operations for the Role. For example, the following Role grants read access (get, watch, and list) to all pods in the accounting Namespace:

kind: Role
  namespace: accounting
  name: pod-reader
- apiGroups: [""] # "" indicates the core API group
  resources: ["pods"]
  verbs: ["get", "watch", "list"]

Refer to the Role and ClusterRole API documentation for a full list of allowed fields.

Role vs. ClusterRole

Because permissions granted by a ClusterRole apply across the entire cluster, you can use ClusterRoles to control access to different kinds of resources than you can with Roles. These include:

  • Cluster-scoped resources such as nodes
  • Non-resource REST Endpoints such as /healthz
  • Namespaced resources across all Namespaces (for example, all Pods across the entire cluster, regardless of Namespace)

Assign Roles using RoleBindings or ClusterRoleBindings

After creating a Role or ClusterRole, you assign it to a user or group of users by creating a RoleBinding or ClusterRoleBinding. Users and groups are called subjects, and can be any of the following:

Subject type Value for kind Value for name
Google Cloud user account User Google Cloud registered email address
Kubernetes service account ServiceAccount The name of a Kubernetes ServiceAccount object in the cluster
IAM service account User Automatically generated IAM service account email address
Google Group address on a verified domain Group Email address of a Google Workspace Group that is a member of the gke-security-groups group. For instructions to set up Google Groups for RBAC, refer to Configure Google Groups for RBAC.

The following RoleBinding grants the pod-reader Role to a user, a Kubernetes service account, an IAM service account, and a Google Group:

kind: RoleBinding
  name: pod-reader-binding
  namespace: accounting
# Google Cloud user account
- kind: User
# Kubernetes service account
- kind: ServiceAccount
  name: johndoe
# IAM service account
- kind: User
# Google Group
- kind: Group
  kind: Role
  name: pod-reader

Verify API access using kubectl

kubectl provides the auth can-i subcommand for quickly querying the API authorization layer. As a platform administrator, you might need to impersonate users to determine what actions they can perform. You can use the auth can-i and pass an additional --as flag.

When you run the command kubectl auth can-i without the --as flag, Identity and Access Management (IAM) performs the authorization. Whereas, when you append the --as flag, Kubernetes RBAC performs the authorization. Therefore, you will need to create the necessary Role and RoleBinding objects for RBAC.

For more information, see Verifying API Access.

API Usage and Examples

For complete information on using the Kubernetes API to create the necessary Role, ClusterRole, RoleBinding, and ClusterRoleBinding objects for RBAC, see Using Role-Based Access Control Authorization in the Kubernetes documentation.

Troubleshooting and debugging

To debug issues with RBAC, use the Admin activity audit log, which is enabled on all clusters by default. If access to a resource or operation is denied due to lack of sufficient permissions, the API server logs an RBAC DENY error, along with additional information such as the user's implicit and explicit group membership. If you are using Google Groups for RBAC, google groups appears in the log message.


The following sections describe interactions that might not seem obvious when working with Kubernetes RBAC and IAM.

Default discovery roles

Clusters are created with a set of default ClusterRoles and ClusterRoleBindings. Requests made with valid credentials are placed in the system:authenticated group, whereas all other requests fall into system:unauthenticated.

The system:basic-user ClusterRole lets users make SelfSubjectAccessReviews to test their permissions in the cluster. The system:discovery role lets users read discovery APIs, which can reveal information about CustomResourceDefinitions added to the cluster.

Anonymous users (system:unauthenticated) receive the system:public-info-viewer ClusterRole instead, which grants read-only access to /healthz and /version APIs.

To see the API endpoints allowed by the system:discovery ClusterRole, run the following command:

kubectl get clusterroles system:discovery -o yaml

Forbidden error for service accounts on Google Cloud VM instances

The following error can occur when the VM instance does not have the userinfo-email scope:

Error from server (Forbidden): error when creating ... "role-name" is forbidden: attempt to grant extra privileges:...

For example, suppose the VM has cloud-platform scope but does not have userinfo-email scope. When the VM gets an access token, Google Cloud associates that token with the cloud-platform scope. When the Kubernetes API server asks Google Cloud for the identity associated with the access token, it receives the service account's unique ID, not the service account's email.

To authenticate successfully, either create a new VM with the userinfo-email scope or create a new role binding that uses the unique ID.

To create a new VM instance with the userinfo-email scope, run the following command:

gcloud compute instances create INSTANCE_NAME \
    --service-account SERVICE_ACCOUNT_EMAIL \
    --scopes userinfo-email

To create a new role binding that uses the service account's unique ID for an existing VM, perform the following steps:

  1. Identify the service account's unique ID:

    gcloud iam service-accounts describe SERVICE_ACCOUNT_EMAIL

    For example, the following output displays the uniqueId for the service account:

    displayName: Some Domain IAM service account
    etag: BwWWja0YfJA
    name: projects/project-name/serviceAccounts/
    oauth2ClientId: '123456789012345678901'
    projectId: project-name
    uniqueId: '123456789012345678901'
  2. Create a role binding using the uniqueId of the service account:

    kubectl create clusterrolebinding CLUSTERROLEBINDING_NAME \
        --clusterrole cluster-admin \
        --user UNIQUE_ID

Permission to create or update roles and role bindings

In Kubernetes, you can only create or update a role or a role binding with specific permissions if you meet the following conditions:

  • Create or update a role: You must already have the same permissions that you want to grant to the role. Alternatively, you must have authorization to perform the escalate verb on the role.
  • Create or update a role binding: You must already have the same permissions that are granted in the role being bound, with the same scope as the role binding. Alternatively, you must have authorization to perform the bind verb on the referenced role.

If the permissions that you're granting in the role were originally granted to you using an IAM allow policy instead of RBAC, your role or role binding request might fail. For example, consider the following role creation request, from a user who has been granted the IAM permissions container.pods.* and container.roles.create:

kubectl create role allowed-to-view-pods --resource pods --verb list,get

If the user was only given the permissions using IAM, the following error could occur:

Error from server (Forbidden): "allowed-to-view-pods" is forbidden:
user "" (groups=["system:authenticated"]) is attempting to grant RBAC permissions not currently held:
{APIGroups:[""], Resources:["pods"], Verbs:["list" "get"]}

To mitigate this limitation, grant the caller the permissions in the role using RBAC instead of IAM.

You can alternatively use either RBAC or IAM to grant the caller the escalate verb, the bind verb, or both. However, GKE does not recommend this approach, because the caller can then grant any permission to any role.

What's next