Deploy containers (GKE, GKE clusters)

This page explains how to deploy a container image to a Google Kubernetes Engine (GKE) cluster or an GKE cluster where Binary Authorization is enabled. The kubectl commands you use to deploy the image are the same as the ones you use to deploy images to clusters that do not use Binary Authorization.

Before you begin

Make sure you have the Binary Authorization API enabled in your project and a GKE cluster with Binary Authorization enabled. See setting up on Google Kubernetes Engine or setting up on GKE clusters.

Install kubectl for interacting with GKE.

Configure kubectl

You must update the local kubeconfig file for your kubectl installation. This provides the credentials and endpoint information required to access the cluster in GKE or GKE clusters.

To configure kubectl, run the following gcloud command:

GKE 

gcloud container clusters get-credentials \
    --zone ZONE \
    CLUSTER_NAME

Replace the following:

  • ZONE: the name of the GKE zone where the cluster is running, for example, us-central1-a
  • CLUSTER_NAME: the name of the cluster

GKE clusters 

gcloud container fleet memberships get-credentials \
    --location LOCATION \
    MEMBERSHIP_NAME

Replace the following:

  • LOCATION: the location of the fleet membership of the GKE cluster, for example, global
  • MEMBERSHIP_NAME: the name of the fleet membership of the GKE cluster

Deploy the container image

Deploy your container image as follows:

  1. Configure environment variables:

    POD_NAME=POD_NAME
IMAGE_PATH=IMAGE_PATH
IMAGE_DIGEST=IMAGE_DIGEST

    Replace the following:

    • POD_NAME: the name you want to use for the GKE workload
    • IMAGE_PATH: path of the image in Artifact Registry, Container Registry, or another registry.

    • IMAGE_DIGEST: the digest of the image manifest. Examples are as follows:

      • Artifact Registry:
        • Path: us-docker.pkg.dev/google-samples/containers/gke/hello-app
        • Digest: sha256:37e5287945774f27b418ce567cd77f4bbc9ef44a1bcd1a2312369f31f9cce567
      • Container Registry:
        • Path: gcr.io/google-samples/hello-app
        • Digest: sha256:c62ead5b8c15c231f9e786250b07909daf6c266d0fcddd93fea882eb722c3be4

      To learn how to get the digest of an image in Artifact Registry, see Managing images; for an image in Container Registry, see Listing the versions of an image.

  2. Deploy your image using the kubectl run command.

    You must deploy the image using the digest rather than a tag like 1.0 or latest, as Binary Authorization uses the digest to look up attestations.

    To deploy the image, run the following kubectl command:

    kubectl run ${POD_NAME} \
    --image ${IMAGE_PATH}@${IMAGE_DIGEST}

    Now, verify that the deployment was blocked by Binary Authorization:

    kubectl get pods

    You see your Pod listed.

Fail open

If GKE is unable to reach the Binary Authorization server for any reason, or if the server returns an error, GKE cannot determine whether Binary Authorization would allow or deny the image. In this case, GKE fails open: it defaults to allowing the image to be deployed, but creates a log entry in Cloud Audit Logs to record why the image was allowed.

GKE enforcement fails open because of a tradeoff between reliability and security. GKE sends a request to the Binary Authorization whenever a Pod is created or updated. This includes scenarios where Pods are automatically created or updated by higher level Kubernetes workload controllers, like ReplicaSets and StatefulSets. If GKE failed closed instead of open, any Binary Authorization outage would stop these Pods from running. Moreover, when Pods are denied, failover can lead to cascading failures as redirected traffic overloads Pods that are still running. Any Binary Authorization outage could trigger a complete outage for your cluster, even without deploying any new images.

Deploy images that violate the policy

Binary Authorization supports a feature known as breakglass that allows an image to be deployed, even if it violates the policy.

For more information, see Using breakglass

Clean up

To clean up, delete the Pod by executing the following command:

  kubectl delete pod ${POD_NAME}

What's next