GKE on Bare Metal quickstart

Introduction to GKE on Bare Metal

With GKE on Bare Metal, you can define four types of clusters:

  • admin - A cluster used to manage user clusters.
  • user - A cluster used to run workloads.
  • standalone - A single cluster that can administer itself, and that can also run workloads, but can't create or manage other user clusters.
  • hybrid - A single cluster for both admin and workloads, that can also manage user clusters.

In this quickstart, you deploy a two-node hybrid cluster with GKE on Bare Metal. You learn how to create a cluster, and how to monitor the cluster creation process.

This quickstart assumes you have a basic understanding of Kubernetes.

Prepare for GKE on Bare Metal

Before creating a cluster in GKE on Bare Metal, you must do the following:

  1. Create a Google Cloud project.
  2. Configure your admin workstation.

Create a Google Cloud project

For this quickstart, create a new Google Cloud project that organizes all your Google Cloud resources. To create a cluster in GKE on Bare Metal, you need a Google Cloud project in which your account has the Owner role.

See Create and manage projects for details.

Configure a Linux admin workstation

This quickstart uses bmctl and kubectl to create and work with a cluster. This command-line tools are run on a Linux admin workstation. For information about setting up your admin workstation, see Admin workstation prerequisites.

Create your cluster nodes

Create two machines to serve as nodes for your cluster:

  • One machine functions as the control plane node.
  • One machine functions as the worker node.

Go to Cluster node machine prerequisites to learn more about the requirements for the cluster nodes.

Create a cluster

To create a cluster:

  1. Use bmctl to create a config file.
  2. Edit the config file to customize it for your cluster and network.
  3. Use bmctl to create the cluster from the config file.

Create a config file

To create a config file, and enable service accounts and APIs automatically, make sure you are in the baremetal directory, and issue the bmctl command with the following flags:

./bmctl create config -c CLUSTER_NAME \
  --enable-apis --create-service-accounts --project-id=PROJECT_ID

CLUSTER_NAME is the name of your cluster. PROJECT_ID is the project you created in Create a Google Cloud project.

The command above creates a config file under the baremetal directory at the following path: bmctl-workspace/cluster1/cluster1.yaml

Edit the config file

To edit the config file:

  1. Open the bmctl-workspace/cluster1/cluster1.yaml config file in an editor.
  2. Edit the file with your specific node and network requirements. Use the sample config file below for reference. This quickstart doesn't use or include information on OpenID Connect (OIDC).
# gcrKeyPath:  < to GCR service account key>
gcrKeyPath: baremetal/gcr.json
# sshPrivateKeyPath:  < to SSH private key, used for node access>
sshPrivateKeyPath: .ssh/id_rsa
# gkeConnectAgentServiceAccountKeyPath:  < to Connect agent service account key>
gkeConnectAgentServiceAccountKeyPath: baremetal/connect-agent.json
# gkeConnectRegisterServiceAccountKeyPath:  < to Hub registration service account key>
gkeConnectRegisterServiceAccountKeyPath: baremetal/connect-register.json
# cloudOperationsServiceAccountKeyPath:  < to Cloud Operations service account key>
cloudOperationsServiceAccountKeyPath: baremetal/cloud-ops.json
apiVersion: v1
kind: Namespace
  name: cluster-cluster1
# Cluster configuration. Note that some of these fields are immutable once the cluster is created.
# For more info, see https://cloud.google.com/anthos/clusters/docs/bare-metal/1.13/reference/cluster-config-ref#cluster_configuration_fields
apiVersion: baremetal.cluster.gke.io/v1
kind: Cluster
  name: cluster1
  namespace: cluster-cluster1
  # Cluster type. This can be:
  #   1) admin:  to create an admin cluster. This can later be used to create user clusters.
  #   2) user:   to create a user cluster. Requires an existing admin cluster.
  #   3) hybrid: to create a hybrid cluster that runs admin cluster components and user workloads.
  #   4) standalone: to create a cluster that manages itself, runs user workloads, but does not manage other clusters.
  type: hybrid
  # Anthos cluster version.
  anthosBareMetalVersion: 1.13.10
  # GKE connect configuration
    projectID: PROJECT_ID
  # Control plane configuration
      # Control plane node pools. Typically, this is either a single machine
      # or 3 machines if using a high availability deployment.
      - address:  CONTROL_PLANE_NODE_IP
  # Cluster networking configuration
    # Pods specify the IP ranges from which pod networks are allocated.
    # Services specify the network ranges from which service virtual IPs are allocated.
    # This can be any RFC 1918 range that does not conflict with any other IP range
    # in the cluster and node pool resources.
  # Load balancer configuration
    # Load balancer mode can be either 'bundled' or 'manual'.
    # In 'bundled' mode a load balancer will be installed on load balancer nodes during cluster creation.
    # In 'manual' mode the cluster relies on a manually-configured external load balancer.
    mode: bundled
    # Load balancer port configuration
      # Specifies the port the load balancer serves the Kubernetes control plane on.
      # In 'manual' mode the external load balancer must be listening on this port.
      controlPlaneLBPort: 443
    # There are two load balancer virtual IP (VIP) addresses: one for the control plane
    # and one for the L7 Ingress service. The VIPs must be in the same subnet as the load balancer nodes.
    # These IP addresses do not correspond to physical network interfaces.
      # ControlPlaneVIP specifies the VIP to connect to the Kubernetes API server.
      # This address must not be in the address pools below.
      controlPlaneVIP: CONTROL_PLANE_VIP
      # IngressVIP specifies the VIP shared by all services for ingress traffic.
      # Allowed only in non-admin clusters.
      # This address must be in the address pools below.
      ingressVIP: INGRESS_VIP
    # AddressPools is a list of non-overlapping IP ranges for the data plane load balancer.
    # All addresses must be in the same subnet as the load balancer nodes.
    # Address pool configuration is only valid for 'bundled' LB mode in non-admin clusters.
    # addressPools:
    # - name: pool1
    #   addresses:
    #   # Each address must be either in the CIDR form (
    #   # or range form (
    # A load balancer nodepool can be configured to specify nodes used for load balancing.
    # These nodes are part of the kubernetes cluster and run regular workloads as well as load balancers.
    # If the node pool config is absent then the control plane nodes are used.
    # Node pool configuration is only valid for 'bundled' LB mode.
    # nodePoolSpec:
    #   nodes:
    #   - address: LOAD_BALANCER_NODE_IP;
  # Proxy configuration
  # proxy:
  #   url: http://[username:password@]domain
  #   # A list of IPs, hostnames or domains that should not be proxied.
  #   noProxy:
  #   -
  #   - localhost
  # Logging and Monitoring
    # Cloud project for logs and metrics.
    projectID: PROJECT_ID
    # Cloud location for logs and metrics.
    location: us-central1
    # Whether collection of application logs/metrics should be enabled (in addition to
    # collection of system logs/metrics which correspond to system components such as
    # Kubernetes control plane or cluster management agents).
    # enableApplication: false
  # Storage configuration
    # lvpNodeMounts specifies the config for local PersistentVolumes backed by mounted disks.
    # These disks need to be formatted and mounted by the user, which can be done before or after
    # cluster creation.
      # path specifies the host machine path where mounted disks will be discovered and a local PV
      # will be created for each mount.
      path: /mnt/localpv-disk
      # storageClassName specifies the StorageClass that PVs will be created with. The StorageClass
      # is created during cluster creation.
      storageClassName: local-disks
    # lvpShare specifies the config for local PersistentVolumes backed by subdirectories in a shared filesystem.
    # These subdirectories are automatically created during cluster creation.
      # path specifies the host machine path where subdirectories will be created on each host. A local PV
      # will be created for each subdirectory.
      path: /mnt/localpv-share
      # storageClassName specifies the StorageClass that PVs will be created with. The StorageClass
      # is created during cluster creation.
      storageClassName: local-shared
      # numPVUnderSharedPath specifies the number of subdirectories to create under path.
      numPVUnderSharedPath: 5
  # NodeConfig specifies the configuration that applies to all nodes in the cluster.
    # podDensity specifies the pod density configuration.
      # maxPodsPerNode specifies the maximum number of pods allowed on a single node.
      maxPodsPerNode: 250

# Node pools for worker nodes
apiVersion: baremetal.cluster.gke.io/v1
kind: NodePool
  name: node-pool-1
  namespace: cluster-cluster1
  clusterName: cluster1
  - address: WORKER_NODE_1_IP
  - address: WORKER_NODE_2_IP

Run preflight checks and create the cluster

The bmctl command runs preflight checks on your cluster config file before it creates a cluster. If the checks are successful, bmctl creates the cluster.

To run preflight checks and create the cluster:

  1. Ensure that you are in the baremetal directory.
  2. Use the following command to create the cluster:
  3. ./bmctl create cluster -c CLUSTER_NAME
    For example:
    ./bmctl create cluster -c cluster1

    The bmctl command monitors the preflight checks and cluster creation, displays output to the screen, and writes verbose information to the bmctl logs.

You can find the bmctl, preflight checks, and node installation logs in the following directory: baremetal/bmctl-workspace/CLUSTER_NAME/log

The bmctl preflight checks the proposed cluster installation for the following conditions:

  • The Linux distribution and version are supported.
  • SELinux is not in "enforcing" mode.
  • On Ubuntu, Uncomplicated Firewall (UFW) is not active.
  • Google Container Registry is reachable.
  • The VIPs are available.
  • The cluster machines have connectivity to each other.
  • Load balancer machines are on the same Layer 2 subnet.

Cluster creation can take several minutes to finish.

Get information about your cluster

After you successfully create a cluster, use the kubectl command to show information about the new cluster. During cluster creation, the bmctl command writes a kubeconfig file for the cluster that you can query with kubectl. The kubeconfig file is written to bmctl-workspace/CLUSTER_NAME/CLUSTER_NAME-kubeconfig.

For example:

kubectl --kubeconfig bmctl-workspace/cluster1/cluster1-kubeconfig get nodes

This command returns:

node-01   Ready    master   16h   v1.17.8-gke.16
node-02   Ready    <none>   16h   v1.17.8-gke.16

If your cluster creation fails preflight checks, then check the preflight check logs for errors, and correct them in the cluster config file. The preflight check logs are located in the /log directory at


The preflight check logs for each machine in the cluster are in the CLUSTER_NAME directory, and are organized by IP address. For example:

└── preflight-20201007-034844
    └── node-network

Ignore pre-flight check errors

If your cluster creation fails after pre-flight checks, you can try to re-install the cluster using the --force flag in the bmctl command.

The --force flag installs over an existing cluster, but ignores the results from any preflight check failure due to already allocated server ports.

  1. Ensure that you are in the baremetal directory.
  2. Use the following command with the --force flag to re-create the cluster:
  3. ./bmctl create cluster -c CLUSTER_NAME --force
    For example:
    ./bmctl create cluster -c cluster1 --force

Create a Deployment and a Service

Here's a manifest for a Deployment:

apiVersion: apps/v1
kind: Deployment
  name: my-deployment
      app: metrics
      department: sales
  replicas: 3
        app: metrics
        department: sales
      - name: hello
        image: "gcr.io/google-samples/hello-app:2.0"

Save the manifest as my-deployment.yaml.

Create the Deployment with the following command:

kubectl --kubeconfig bmctl-workspace/cluster1/cluster1-kubeconfig create -f my-deployment.yaml

View the Deployment:

kubectl --kubeconfig bmctl-workspace/cluster1/cluster1-kubeconfig get deployments

The output shows that the Deployment has three available and ready Pods:

my-deployment      3/3     3            3           16s

The following manifest defines a Service of type LoadBalancer:

apiVersion: v1
kind: Service
  name: my-service
    app: metrics
    department: sales
  type: LoadBalancer
  - port: 80
    targetPort: 8080

Save the manifest as my-service.yaml.

Create the Service with the following command:

kubectl --kubeconfig bmctl-workspace/cluster1/cluster1-kubeconfig create -f my-service.yaml

View the Service:

kubectl --kubeconfig bmctl-workspace/cluster1/cluster1-kubeconfig get service my-service


my-service   LoadBalancer   80:30060/TCP

GKE on Bare Metal gives the service an external IP address. Use the external IP address to call the service:


The output is a hello world message:

Hello, world!
Version: 2.0.0
Hostname: my-deployment-75d45b64f9-6clxj

Create a high availability control plane

The quickstart created a simple two-node hybrid cluster. If you want to create a high availability control plane, create a cluster that has three control plane nodes.

For example, edit the config file to add two additional nodes to the control plane:

    clusterName: cluster1
    # Control Plane node pools. Typically, this is either a single machine
    # or 3 machines if using a high availability deployment.
    - address: <Machine 1 IP>
    - address: <Machine 2 IP>
    - address: <Machine 3 IP>

Run the load balancer in its own node pool

The quickstart created a simple two-node hybrid cluster. Thus, the load balancer runs on the same node that runs the control plane.

If you want the load balancer to run in its own node pool, edit the nodePoolSpec values of the loadBalancer section of your config file:

      clusterName: "cluster1"
      - address: <LB Machine 1 IP>
      - address: <LB Machine 2 IP>