Encrypt workload data in-use with Confidential GKE Nodes

This page shows you how to enforce encryption of data in-use in your nodes and workloads by using Confidential Google Kubernetes Engine Nodes. Enforcing encryption can help increase the security of your workloads.

This page is for Security specialists who implement security measures on GKE. To learn more about common roles and example tasks that we reference in Google Cloud content, see Common GKE user roles and tasks.

Before reading this page, ensure that you're familiar with the concept of data-in-use.

What is Confidential GKE Nodes?

You can encrypt your workloads with Confidential GKE Nodes or Confidential mode for Hyperdisk Balanced.

Confidential GKE Nodes

Confidential GKE Nodes uses Compute Engine Confidential VM to protect data in use with hardware-based memory encryption. Confidential GKE Nodes supports the following Confidential Computing technologies:

• AMD Secure Encrypted Virtualization (SEV)
• AMD Secure Encrypted Virtualization-Secure Nested Paging (SEV-SNP)
• Intel Trust Domain Extensions (TDX)

For more information about these technologies and for help choosing the optimal technology for your requirements, see the Confidential VM overview.

Confidential GKE Nodes doesn't change the security measures that GKE applies to cluster control planes. To learn about these measures, see Control plane security. For visibility over who accesses control planes in your Google Cloud projects, use Access Transparency.

You can enable Confidential GKE Nodes for entire clusters, for specific node pools, or for specific workloads. The following table describes these enablement methods:

Confidential mode for Hyperdisk Balanced

You can also enable Confidential mode for Hyperdisk Balanced on your boot disk storage, which encrypts your data on additional hardware-backed enclaves.

You can enable Confidential mode for Hyperdisk Balanced when doing one of the following:

• Create a new cluster
• Create a new node pool

You cannot update an existing cluster or a node pool to change the Confidential mode for Hyperdisk Balanced setting.

The following table shows you the GKE behavior that applies when you enable Confidential mode for Hyperdisk Balanced setting at the cluster level or at the node pool level:

Pricing

The following pricing applies, depending on how you configure Confidential GKE Nodes:

Additionally, Confidential GKE Nodes might generate slightly more log data on startup than nodes that aren't Confidential GKE Nodes. For more information about logs pricing, see Pricing for Google Cloud Observability.

Before you begin

Before you start, make sure that 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.

Requirements

Confidential GKE Nodes has the following requirements:

• Autopilot clusters must run GKE version 1.30.2 or later.

• Standard clusters must run one of the following GKE versions, depending on the Confidential Computing technology that you choose:

  • AMD SEV: any GKE version.
  • AMD SEV-SNP: 1.32.2-gke.1297000 or later.
  • Intel TDX: 1.32.2-gke.1297000 or later.

• To enable Confidential GKE Nodes at the workload level, Autopilot and Standard clusters must run GKE version 1.33.3-gke.1392000 or later.

• Nodes must use a machine type that supports the Confidential Computing technology that you choose. For more information, see Machine types, CPUs, and zones.

• Cluster control planes and nodes must be in a location that supports the Confidential Computing technology that you choose. For more information, see View supported zones.

Configure Confidential GKE Nodes at the workload level

To enable Confidential GKE Nodes in individual workloads, you use a ComputeClass. ComputeClasses are Kubernetes custom resources that let you define the attributes of new nodes that GKE creates and let you control fallback behavior if hardware isn't available. ComputeClasses support all of the available Confidential Computing technologies for GKE, even if a specific technology isn't supported at the cluster or node pool level.

To configure Confidential GKE Nodes at the workload level, follow these steps:

1. Optional: For GKE Standard clusters, enable node-autoprovisioning for the cluster. For Autopilot clusters, skip this step.

   If you choose to use ComputeClasses with only manually created node pools, you must also manually configure node taints and node labels on those node pools. This additional manual configuration might result in you noticing fewer benefits from using a ComputeClass to create Confidential GKE Nodes.

2. Save the following ComputeClass manifest as a YAML file. For more information about the supported fields and values for ComputeClasses, see the ComputeClass CustomResourceDefinition.

   apiVersion: cloud.google.com/v1
   kind: ComputeClass
   metadata:
     name: COMPUTE_CLASS
   spec:
     nodePoolConfig:
       confidentialNodeType: CONFIDENTIAL_COMPUTE_TECHNOLOGY
     nodePoolAutoCreation:
       enabled: true
     activeMigration:
       optimizeRulePriority: true
     priorityDefaults:
       location:
         zones: ['ZONE1','ZONE2']
     priorities:
     - machineFamily: MACHINE_SERIES
     - machineType: MACHINE_TYPE
     whenUnsatisfiable: DoNotScaleUp
   

   Replace the following:

   • COMPUTE_CLASS: a name for your new compute class.

   • CONFIDENTIAL_COMPUTE_TECHNOLOGY: the Confidential Computing technology to use. The following values are supported:

     • SEV for AMD SEV
     • SEV_SNP for AMD SEV-SNP
     • TDX for Intel TDX

   • ZONE1,ZONE2,...: the zones to create nodes in, such as ['us-central1-a','us-central1-b']. Specify zones that support the Confidential Computing technology that you specified. For more information, see View supported zones.

   • MACHINE_SERIES: the machine series for the nodes, such as n2d. Specify a machine series that supports the Confidential Computing technology that you specified. For more information, see Machine types, CPUs, and zones.

   • MACHINE_TYPE the machine type for the nodes, such as n2d-standard-4. Specify a machine type that supports the Confidential Computing technology that you specified. For more information, see Machine types, CPUs, and zones.

   This manifest includes the spec.nodePoolAutoCreation field with a value of enabled: true, which lets GKE create new node pools by using node auto-provisioning.

3. Create the ComputeClass in your cluster:

   kubectl apply -f PATH_TO_COMPUTE_CLASS_MANIFEST
   

   Replace PATH_TO_COMPUTE_CLASS_MANIFEST with the path to the ComputeClass manifest file.

4. Save the following Deployment manifest as a YAML file:

   apiVersion: apps/v1
   kind: Deployment
   metadata:
     name: helloweb
     labels:
       app: hello
   spec:
     selector:
       matchLabels:
         app: hello
     template:
       metadata:
         labels:
           app: hello
       spec:
         nodeSelector:
           # Replace with the name of a compute class
           cloud.google.com/compute-class: COMPUTE_CLASS 
         containers:
         - name: hello-app
           image: us-docker.pkg.dev/google-samples/containers/gke/hello-app:1.0
           ports:
           - containerPort: 8080
           resources:
             requests:
               cpu: "250m"
               memory: "4Gi"

5. Create the Deployment:

   kubectl apply -f PATH_TO_DEPLOYMENT_MANIFEST
   

   Replace PATH_TO_DEPLOYMENT_MANIFEST with the path to the Deployment manifest file.

New nodes that GKE creates for this workload use the Confidential Computing technology that you specified. These nodes run in one of the zones that you selected, on a machine type that you defined in the ComputeClass. GKE adds node taints and node labels so that only Pods that select the same ComputeClass can run on the nodes.

Configure Confidential GKE Nodes at the cluster level

You can enable Confidential GKE Nodes for an entire Autopilot or Standard cluster, which means that every node in the clusters uses Confidential GKE Nodes. As a result, all of the workloads that you deploy in that cluster run on Confidential GKE Nodes. You can't override the cluster setting in individual workloads or node pools.

Create a new Autopilot cluster

Autopilot clusters support only AMD SEV at the cluster level. When you enable Confidential GKE Nodes, the default machine series in the Autopilot cluster changes to N2D. All of your workloads run on these confidential nodes without needing to change your workload manifests.

• Run the following command:

  gcloud container clusters create-auto CLUSTER_NAME \
      --location=CONTROL_PLANE_LOCATION \
      --enable-confidential-nodes
  

  Replace the following:

  • CLUSTER_NAME: the name of the Autopilot cluster.
  • CONTROL_PLANE_LOCATION: the Compute Engine region for the cluster control plane, such as us-central1.

  The cluster must run version 1.30.2 or later. For more information, see Create an Autopilot cluster with a specific release channel and version.

Create a new Standard cluster

You can specify a Confidential Computing technology for your nodes when you create a Standard cluster. Specifying the technology when you create a cluster has the following effects:

• You can create only node pools or workloads that use Confidential GKE Nodes in that cluster.
• You can't update the cluster to disable Confidential GKE Nodes.
• You can't override the cluster-level Confidential Computing technology in individual node pools or workloads.

Configuring a Confidential Computing setting at the cluster level is permanent. As a result, consider the following use cases before you create your cluster:

• To use node auto-provisioning in your cluster, you must configure specific options in your cluster creation command. For more information, see the Use Confidential GKE Nodes with node auto-provisioning section.
• To use different Confidential Computing technologies to encrypt specific node pools in the cluster, skip this section and configure Confidential GKE Nodes at the node pool level.

To create a Standard mode cluster that uses Confidential GKE Nodes, select one of the following options:

gcloud container clusters create CLUSTER_NAME \
    --location=LOCATION \
    --machine-type=MACHINE_TYPE \
    --node-locations=ZONE1,ZONE2,... \
    --confidential-node-type=CONFIDENTIAL_COMPUTE_TECHNOLOGY

For more information about creating clusters, see Creating a regional cluster.

For any node pool created with the Confidential mode for Hyperdisk Balanced setting, only the nodes in the node pool are restricted to the setup configuration. For any new node pools created in the cluster, you must set up confidential mode at creation.

Configure node auto-provisioning

You can configure node auto-provisioning to use Confidential GKE Nodes in auto-provisioned node pools. Node auto-provisioning supports the following Confidential Computing technologies:

• AMD SEV
• AMD SEV-SNP

To use Confidential GKE Nodes with node auto-provisioning, specify the --enable-confidential-nodes gcloud CLI flag when you create a cluster. The following additional considerations apply:

• When you create the cluster, choose a Confidential Computing technology that node auto-provisioning supports. This choice is irreversible at the cluster level.
• When you enable node auto-provisioning on an existing cluster, the cluster must already use a Confidential GKE Nodes technology that node auto-provisioning supports.
• When you deploy workloads, verify that the workloads select the Confidential Computing technology of the cluster and a compatible machine series for that technology.

Configure Confidential GKE Nodes at the node pool level

You can enable Confidential GKE Nodes on specific node pools if Confidential GKE Nodes is disabled at the cluster level.

Confidential mode for Hyperdisk Balanced setting must be specified during node pool creation request.

Create a new node pool

To create a new node pool with Confidential GKE Nodes enabled, select one of the following options:

gcloud container node-pools create NODE_POOL_NAME \
    --location=LOCATION \
    --cluster=CLUSTER_NAME \
    --machine-type=MACHINE_TYPE \
    --node-locations=ZONE1,ZONE2,... \
    --confidential-node-type=CONFIDENTIAL_COMPUTE_TECHNOLOGY

Update an existing node pool

This change requires recreating the nodes, which can cause disruption to your running workloads. For details about this specific change, find the corresponding row in the manual changes that recreate the nodes using a node upgrade strategy without respecting maintenance policies table. To learn more about node updates, see Planning for node update disruptions.

You can update existing node pools to use Confidential GKE Nodes or to switch the Confidential Computing technology that the nodes use. The node pool must meet all of the following requirements:

• The nodes use a machine type that supports the chosen Confidential Computing technology.
• The nodes run in zones that support the chosen Confidential Computing technology.

Because the machine type and location availability might vary between Confidential Computing technologies, you might need to update these attributes for your node pool before you enable a specific Confidential Computing technology.

To update an existing node pool to use Confidential GKE Nodes, select one of the following options:

Place workloads on only confidential node pools

If you enable Confidential GKE Nodes at the cluster level, all of your workloads run on confidential nodes. You don't need to make changes to your manifests. However, if you enable Confidential GKE Nodes only for specific Standard mode node pools at the node pool level, you should declaratively express that your workloads must run only on node pools with Confidential GKE Nodes.

• To require that a workload runs on a specific Confidential Computing technology, use a node selector with the cloud.google.com/gke-confidential-nodes-instance-type label, like in the following example:

  apiVersion: v1
  kind: Pod
  spec:
  # For readability, lines are omitted from this example manifest
    nodeSelector:
      cloud.google.com/gke-confidential-nodes-instance-type: "CONFIDENTIAL_COMPUTE_SELECTOR"
  

  Replace CONFIDENTIAL_COMPUTE_SELECTOR with the name of the technology that the node pool uses. This field supports the following values, which you must specify in uppercase:

  • SEV: AMD SEV
  • SEV_SNP: AMD SEV-SNP
  • TDX: Intel TDX

• To let a workload run on any confidential nodes, regardless of the Confidential Computing technology, use a node affinity rule, like in the following example:

  apiVersion: v1
  kind: Pod
  spec:
    containers:
    - name: confidential-app
      image: us-docker.pkg.dev/google-samples/containers/gke/hello-app:1.0
    affinity:
      nodeAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          nodeSelectorTerms:
          - matchExpressions:
            - key: cloud.google.com/gke-confidential-nodes-instance-type
              operator: Exists
  

• To let a workload run on nodes that use only a subset of the available Confidential Computing technologies, use a node affinity rule that's similar to the following example:

  apiVersion: v1
  kind: Pod
  spec:
    containers:
    - name: confidential-app
      image: us-docker.pkg.dev/google-samples/containers/gke/hello-app:1.0
    affinity:
      nodeAffinity:
        requiredDuringSchedulingIgnoredDuringExecution:
          nodeSelectorTerms:
          - matchExpressions:
            - key: cloud.google.com/gke-confidential-nodes-instance-type
              operator: In
              values:
              - SEV
              - SEV_SNP
              - TDX
  

  In the values field, specify only the Confidential Computing technologies that you want to run the workload on.

Verify that Confidential GKE Nodes is enabled

You can check whether your clusters or nodes use Confidential GKE Nodes by inspecting the clusters, node pools, nodes, or Compute Engine instances.

Verify for clusters

Select one of the following options:

gcloud container clusters describe CLUSTER_NAME \
    --location=CONTROL_PLANE_LOCATION \
    --format='yaml(confidentialNodes.confidentialInstanceType)'

confidentialNodes:
  confidentialInstanceType: CONFIDENTIAL_COMPUTE_SELECTOR

Verify for node pools

Select one of the following options:

gcloud container node-pools describe NODE_POOL_NAME \
    --location=CONTROL_PLANE_LOCATION \
    --format='yaml(confidentialNodes.confidentialInstanceType,enableConfidentialStorage)'

confidentialNodes:
  confidentialInstanceType: CONFIDENTIAL_COMPUTE_SELECTOR

confidentialNodes:
  confidentialInstanceType: CONFIDENTIAL_COMPUTE_SELECTOR
enableConfidentialStorage: true

Verify for a specific node

1. Find the node name:

   kubectl get nodes
   

2. Describe the node:

   kubectl describe NODE_NAME
   

   Replace NODE_NAME with the name of a node to inspect.

   The output is similar to the following:

   Name:               gke-cluster-1-default-pool-affsf335r-asdf
   Roles:              <none>
   Labels:             cloud.google.com/gke-boot-disk=pd-balanced
                       cloud.google.com/gke-container-runtime=containerd
                       cloud.google.com/gke-confidential-nodes-instance-type=CONFIDENTIAL_COMPUTE_SELECTOR
                       cloud.google.com/gke-nodepool=default-pool
                       cloud.google.com/gke-os-distribution=cos
                       cloud.google.com/machine-family=e2
   # lines omitted for clarity
   

   In this output, the cloud.google.com/gke-confidential-nodes-instance-type node label indicates that the node is a confidential node.

Set organization policy constraints

You can define an organization policy constraint to ensure that all VM resources created across your organization are Confidential VM instances. For GKE, you can customize the Restrict Non-Confidential Computing constraint to require that all new clusters are created with one of the available Confidential Computing technologies enabled. Add the container.googleapis.com API Service name to the deny list when enforcing organization policy constraints, like in the following example:

gcloud resource-manager org-policies deny \
    constraints/compute.restrictNonConfidentialComputing compute.googleapis.com container.googleapis.com \
    --project=PROJECT_ID

Replace PROJECT_ID with your project ID.

Create a PersistentVolume for Confidential mode for Hyperdisk Balanced

For guidance on allowable values for throughput or IOPS, see Plan the performance level for your Hyperdisk volume.

The following examples show how you can create a Confidential mode for Hyperdisk Balanced StorageClass for each Hyperdisk type:

To find the name of the StorageClasses available in your cluster, run the following command:

kubectl get sc

Limitations

Confidential GKE Nodes has the following limitations:

• All of the limitations of Compute Engine Confidential VM instances also apply to Confidential GKE Nodes.
• Confidential GKE Nodes isn't compatible with sole tenant nodes.
• Confidential GKE Nodes only supports using ephemeral storage on local SSDs, but doesn't support using local SSDs in general.
• Confidential GKE Nodes supports only Container-Optimized OS node images. Ubuntu and Windows nodes aren't supported.
• Autopilot clusters support only AMD SEV at the cluster level. To use a different Confidential Computing technology, configure Confidential GKE Nodes at the workload level.
• Node auto-provisioning supports only AMD SEV and AMD SEV-SNP at the node pool level. To use a different Confidential Computing technology, configure Confidential GKE Nodes at the workload level.

Confidential mode for Hyperdisk Balanced has the following limitations:

• All of the limitations of using customer-managed encryption keys (CMEK) to encrypt disks apply to Confidential mode for Hyperdisk Balanced.
• Confidential mode for Hyperdisk Balanced is supported only on Confidential GKE Nodes that use AMD SEV as the Confidential Computing technology.

Live migration limitations

Compute Engine Confidential VM that use the N2D machine type and use AMD SEV as the Confidential Computing technology support live migration, which minimizes the potential workload disruption from a host maintenance event. Live migration occurs in the following GKE versions:

• 1.27.10-gke.1218000 and later
• 1.28.6-gke.1393000 and later
• 1.29.1-gke.1621000 and later

If your node pools were already running a supported version when live migration was added, manually upgrade the node pools to the same or a different supported version. Upgrading the nodes triggers node recreation, and the new nodes have live migration enabled.

For details about which Compute Engine machine types support live migration, see Supported configurations.

If a host maintenance event occurs on a node that doesn't support live migration, the node enters a NotReady state. Running Pods will experience disruptions until the node becomes ready again. If the maintenance takes more than five minutes, GKE might try to recreate the Pods on other nodes.

Disable Confidential GKE Nodes

This change requires recreating the nodes, which can cause disruption to your running workloads. For details about this specific change, find the corresponding row in the manual changes that recreate the nodes using a node upgrade strategy without respecting maintenance policies table. To learn more about node updates, see Planning for node update disruptions.

You can only disable Confidential GKE Nodes in Standard mode node pools. If the node pool is in a cluster that uses Confidential GKE Nodes at the cluster level, you can't disable the feature at the node pool level. To disable Confidential GKE Nodes, select one of the following options:

gcloud container node-pools update NODE_POOL_NAME \
    --cluster=CLUSTER_NAME \
    --location=CLUSTER_LOCATION \
    --no-enable-confidential-nodes

What's next

• Learn more about Confidential VM
• Learn more about Google Cloud encryption at rest
• Learn more about Google Cloud encryption in transit
• Learn more about customer-managed encryption keys (CMEK)
• Learn how to remotely attest that workloads are running on Confidential VM
• Learn how to run GPUs on Confidential GKE Nodes