Plan your installation

Version 1.15 

• Version 1.28 (latest)
• Version 1.16
• Version 1.15
• Earlier versions (unsupported)

This page is the first part of a guide that walks you through a small proof-of-concept installation of GDCV for Bare Metal. This document shows you how to set up a minimal hardware environment and plan your IP addresses. The follow-up Create basic clusters shows you how to create an admin cluster and a user cluster. The infrastructure you set up using this guide might not be suitable for your actual production needs and use cases. For more information about production installations, see Choose a deployment model.

Before you begin

1. Read About GKE on Bare Metal.
2. Familiarize yourself with some basic Google Cloud concepts, including projects, IAM permissions, and service accounts.
3. Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.

4. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

   Note: If you don't plan to keep the resources that you create in this procedure, create a project instead of selecting an existing project. After you finish these steps, you can delete the project, removing all resources associated with the project.

   Go to project selector

5. Make sure that billing is enabled for your Google Cloud project.

6. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

   Note: If you don't plan to keep the resources that you create in this procedure, create a project instead of selecting an existing project. After you finish these steps, you can delete the project, removing all resources associated with the project.

   Go to project selector

7. Make sure that billing is enabled for your Google Cloud project.

8. Make a note of the Google Cloud project ID, because it's needed later.

Procedure overview

Minimal infrastructure setup consists of these primary steps:

1. Set up your admin workstation. Set up a Linux admin workstation for on-premises management tasks. This could be an existing or a dedicated machine, which can manage multiple clusters.

2. Set up your cluster node machines. Set up at least three machines for nodes: one admin cluster node, one user cluster control plane node, and one user cluster worker node.

3. Plan your networking. Plan the IP addresses for your node machines, virtual IP addresses (VIPs), and Service and Pod CIDR ranges.

4. Review required Google Cloud resources. To create clusters, your Google Cloud project requires specific Google APIs and service accounts.

1. Set up your admin workstation

The admin workstation hosts tools and configuration files for creating and working with your clusters.

Hardware requirements

The admin workstation requires significant computing power, memory, and storage to run tools and store the resources associated with cluster creation and management.

Ensure your admin workstation meets the following hardware requirements:

• At least 2 CPU cores
• At least 4 GiB of RAM
• At least 128 GiB of storage

Configure the operating system and software

On the admin workstation, you install and configure the following:

• Configure Ubuntu

• Install the gcloud CLI

• Install kubectl

• Install bmctl

Configure the operating system

In order to run bmctl and create a cluster, the admin workstation has the same operating system (OS) requirements as nodes. Each machine must run a supported version of Ubuntu, such as Ubuntu 20.04.

Note: While GKE on Bare Metal is supported on multiple Linux operating systems, this minimal example uses Ubuntu for simplicity. Any steps or behavior that are specific to Ubuntu are highlighted in the documents.

Run the following commands to update firewall settings, install and configure Docker, and ensure each machine uses time synchronization:

1. Disable Uncomplicated Firewall (UFW) and verify its status:

   sudo ufw disable
   sudo ufw status
   

2. Remove any previous Docker version, update your package manager, and install the latest version of Docker:

   sudo apt-get remove docker docker-engine docker.io containerd runc
   sudo apt-get update
   sudo apt-get install \
     apt-transport-https \
     ca-certificates \
     curl \
     gnupg-agent \
     software-properties-common \
     docker.io
   

3. Verify you're now running version 19.03+ of Docker:

   sudo docker version
   

   Both Client and Server versions should be 19.03 or higher as shown in the following sample response:

   Client:
    Version:           20.10.21
    API version:       1.41
    Go version:        go1.18.1
    ...
   
   Server:
    Engine:
     Version:          20.10.21
     API version:      1.41 (minimum version 1.12)
     Go version:       go1.18.1
     ...
   

4. Create the docker group.

   sudo groupadd docker
   

5. Add yourself to the Docker group:

   sudo usermod -aG docker $USER
   

6. Run the following command to activate your group changes:

   newgrp docker
   

7. Run the following command to verify the system clock is synchronized:

   timedatectl
   

   The output of timedatectl should contain the following status:

   System clock synchronized: yes
   

Install the Google Cloud CLI

To install the Google Cloud CLI on Ubuntu, follow the instructions in this installation guide.

Perform the following steps on your admin workstation to configure gcloud CLI:

1. Log in to set your gcloud CLI account property:

   gcloud auth login
   

2. Set your gcloud CLI project property:

   gcloud config set project PROJECT_ID
   

   Replace PROJECT_ID with the ID of your Google Cloud project.

3. Verify that your account and project properties are set correctly:

   gcloud config list
   

   The output shows the values of your account and project properties. For example:

   [core]
   account = [e-mail]
   disable_usage_reporting = False
   project = my-project-1234
   Your active configuration is: [default]
   

Install kubectl

To install kubectl:

1. Run the following command on your admin workstation:

   gcloud components install kubectl
   

Install bmctl

bmctl is the proprietary command-line tool for GKE on Bare Metal that you can use for cluster creation and management.

To install bmctl on your admin workstation:

1. Create a baremetal directory and add it to your path. If you are in your home directory, the commands are:

   mkdir baremetal
   export PATH="$HOME/baremetal:$PATH"
   

2. Run the following command to download the latest version of the bmctl binary file and make it executable:

   gsutil cp gs://anthos-baremetal-release/bmctl/1.15.11/linux-amd64/bmctl .
   chmod +x ./bmctl
   

3. Verify that bmctl is installed and executable:

   bmctl version
   

   The response should look something like the following:

   [2023-05-12 17:36:16+0000] bmctl version: 1.14.2-gke.11, git commit: 4ff1347446a93925a079000b50437d4ecebcdf3a, build date: Mon Feb 27 14:07:30 PST 2023
   

Connectivity

The admin workstation needs access to Google Cloud and all your cluster nodes.

Access to Google Cloud

The admin workstation accesses Google Cloud to download and install tools and images, process authorization requests, create service accounts, manage logging and monitoring, and more. You can't create clusters without access to Google Cloud.

Access from the admin workstation

To create and manage clusters from your admin workstation, you need the following access to the node machines:

• Layer 3 connectivity to all cluster node machines.
• Access to the control plane VIP.
• Passwordless root access to all cluster node machines through SSH. SSH access can be either direct or through sudo.

The following section contains instructions for setting up SSH on the admin workstation and the node machines.

2. Set up your cluster node machines

For the minimal installation of a single, non-high availability admin cluster and a single, non-high availability user cluster, you need three machines:

• A machine for an admin cluster with one control plane node.

• Two machines for a user cluster with one control plane node and one worker node.

Hardware requirements

Each node machine must meet the following hardware requirements:

• At least 2 CPU cores
• At least 4 GiB of RAM
• At least 128 GiB of storage

Configure Ubuntu

Configure Ubuntu on each node with the same instructions that were used for the admin workstation.

Set up SSH access to nodes

The admin workstation needs passwordless root access to all cluster node machines through SSH. SSH access can be either direct or through sudo.

These are the high-level steps for setting up SSH for GKE on Bare Metal:

1. Install and configure SSH on all machines

2. Create SSH keys and copy the public key to each node machine

3. Disable password authentication on node machines

4. Verify SSH access between admin workstation and node machines

Install and configure SSH on all machines

GKE on Bare Metal requires passwordless SSH communication between the admin workstation and the cluster nodes. The following steps must be performed on the admin workstation and each node machine.

To configure SSH on machines running Ubuntu:

1. If you do not already have an SSH server running, install one now:

   sudo apt update
   sudo apt install openssh-server
   sudo systemctl status ssh
   

2. Enable root SSH password authentication by uncommenting or adding the PermitRootLogin and PasswordAuthentication lines in the /etc/ssh/sshd_config file and setting the values to yes:

   Note: PasswordAuthentication is enabled temporarily to help with SSH setup. PasswordAuthentication is disabled on the node machines in a subsequent section.

   # Authentication:
   
   #LoginGraceTime 2m
   PermitRootLogin yes
   #StrictModes yes
   #MaxAuthTries 6
   #MaxSessions 10
   ...
   PasswordAuthentication yes
   

3. Set a root password:

   sudo passwd root
   

4. To apply your SSH configuration changes, restart the SSH service:

   sudo systemctl restart ssh.service
   

5. Restart the machine.

6. Verify that SSH is working by establishing an SSH connection from another machine.

Create SSH keys and copy the public key to each node machine

For secure, passwordless connections between the admin workstation and nodes, create an SSH key on your admin workstation and share the public key with the nodes.

1. On your admin workstation, generate a private/public key pair. Don't set a passphrase for the keys:

   ssh-keygen -t rsa
   

2. On your admin workstation, copy the generated public key to each of your node machines:

   ssh-copy-id -i PUBLIC_KEY root@CLUSTER_NODE_IP
   

   Replace the following:

   • PUBLIC_KEY: the path to the file containing the SSH public key. By default, the path is /home/USERNAME/.ssh/id_rsa.pub
   • CLUSTER_NODE_IP: the IP address of the node machine

Disable password authentication on node machines

At this point, you no longer need to have password authentication enabled.

For each node machine:

1. Open /etc/ssh/sshd_config and set PasswordAuthentication to no and save the file.

2. Restart the SSH service.

   sudo systemctl restart ssh.service
   

Verify SSH access between admin workstation and node machines

When SSH is configured properly, you can establish an SSH connection to the node machine from the admin workstation (as root) without a password.

To verify that public key authentication works between your admin workstation and your cluster nodes:

1. On the admin workstation, run the following command for each node machine:

   ssh -o IdentitiesOnly=yes -i PRIVATE_KEY root@CLUSTER_NODE_IP
   

   Replace the following:

   • PRIVATE_KEY: the path to the file containing the SSH private key. By default, the path is /home/USERNAME/.ssh/id_rsa
   • CLUSTER_NODE_IP: the IP address of the node machine

3. Plan your networking

When installing clusters, it's important that you plan your IP addresses, including making sure that you don't create any addressing conflicts. You might need your network administrator to help you find suitable addresses, even for this simple installation. Not counting Pods and Services CIDRs, you need at least 15 unique IP addresses for a minimum admin cluster and user cluster installation.

Plan and specify IP addresses for the following cluster components:

• Cluster nodes: you need an IP address for each node machine
• Virtual IP addresses (VIPs): you need VIPs for access to the Kubernetes API servers, the ingress proxy, and Services of type LoadBalancer
• Pods and Services: you need CIDR address ranges to accommodate every Pod and Service running on your clusters

The rest of this section has illustrative examples of values that work for this installation in a hypothetical network - your values will be different.

For this small installation, put your admin workstation, admin cluster node, and user cluster nodes in the same Layer 2 domain. For example, suppose that all IP addresses in the 172.16.20.0/24 range are routed to a particular Layer 2 domain. Also suppose that your network administrator says you can use 172.16.20.10 - 172.16.20.12 for node machine addresses and 172.16.0.13 - 172.16.20.24 for VIPs.

Important: The IP address examples used in this document meet the requirements for GKE on Bare Metal. However, you must use IP addresses that are available in your network. If you're installing GKE on Bare Metal on a home network, you likely need to adjust addresses for node machines and the cluster VIPs.

The following diagram illustrates a Layer 2 domain that has an admin workstation, an admin cluster, and a user cluster:

IP addresses for an admin cluster and a user cluster (Click to enlarge)

Example cluster node IP addresses

The following table gives an example of how IP addresses could be used for cluster nodes:

Machine	Description	IP address
Admin cluster control plane node	Physical machine that serves as the control plane node for the admin cluster	172.16.20.10
User cluster control plane node	Physical machine that serves as the control plane node for the user cluster	172.16.20.11
User cluster worker node	Physical machine that runs user workloads	172.16.20.12

Example virtual IP addresses (VIPs)

The following table gives an example of how you could specify VIPs for your clusters:

VIP	Description	IP address
Admin cluster control plane VIP address	Admin cluster control plane VIP address (admin cluster Kubernetes API server)	172.16.20.13
User cluster control plane VIP address	User cluster control plane VIP address (user cluster Kubernetes API server)	172.16.20.14
Ingress VIP address	Ingress VIP (included in MetalLB address pool range)	172.16.20.15
Service VIP addresses	Ten addresses for use as external IP addresses for Services of type LoadBalancer. Addresses are allocated as needed on user cluster nodes. This range includes the ingress VIP. This overlap of IP addresses is a requirement for MetalLB, the default bundled load balancer.	172.16.20.15 - 172.16.20.24

IP addresses for Pods and Services

In addition to the IP addresses you have specified for cluster nodes and VIPs, you need to specify addresses for Pods and Services. You specify a CIDR range to be used for Pod IP addresses and another CIDR range to be used for the ClusterIP addresses of Kubernetes Services. Use IP addresses in the private address space, as described in RFC 1918. These addresses are specified as part of cluster configuration, as illustrated in the next part of this guide.

As part of your IP planning, decide what CIDR ranges you want to use for Pods and Services. Unless you have a reason to do otherwise, use the following suggested ranges:

Purpose	Prepopulated CIDR range
Admin cluster Pods	192.168.0.0/16
Admin cluster Services	10.96.0.0/20
User cluster Pods	192.168.0.0/16
User cluster Services	10.96.0.0/20

The suggested ranges illustrate these points:

• The Pod CIDR range can be the same for multiple clusters in the default, island mode network model.

• The Service CIDR range can be the same for multiple clusters.

• Typically you need more Pods than Services in a cluster, so you probably want a Pod CIDR range that is larger than the Service CIDR range. For example, the suggested Pod range for a user cluster has 2^(32-16) = 2¹⁶ addresses, but the suggested Service range for a user cluster has only 2^(32-20) = 2¹² addresses.

Avoid overlap

To avoid overlapping with IP addresses that are reachable on your network, you might need to use CIDR ranges that are different from the preceding suggestions. The Service and Pod ranges must not overlap with any address outside the cluster that you want to reach from inside the cluster.

For example, suppose that your Service range is 10.96.232.0/24, and your Pod range is 192.168.0.0/16. Traffic sent from a Pod to an address in either of those ranges is treated as in-cluster and can't reach any destination outside the cluster.

In particular, the Service and Pod ranges must not overlap with:

• IP addresses of nodes in any cluster

• IP addresses used by load balancer machines

• VIPs used by control-plane nodes and load balancers

• IP address of DNS servers or NTP servers

4. Review required Google Cloud resources

Before you can create clusters, GKE on Bare Metal requires a specific set of Google APIs to be enabled in your associated Google Cloud project. To use the Google APIs, GKE on Bare Metal requires service accounts configured with specific IAM roles in your associated Google Cloud project.

The process for creating clusters in the next part of this guide, Create basic clusters, enables APIs and creates service accounts for you automatically.

Here are the Google APIs that are automatically enabled:

• anthos.googleapis.com
• anthosaudit.googleapis.com
• anthosgke.googleapis.com
• cloudresourcemanager.googleapis.com
• connectgateway.googleapis.com
• container.googleapis.com
• gkeconnect.googleapis.com
• gkehub.googleapis.com
• gkeonprem.googleapis.com
• iam.googleapis.com
• logging.googleapis.com
• monitoring.googleapis.com
• opsconfigmonitoring.googleapis.com
• serviceusage.googleapis.com
• stackdriver.googleapis.com
• storage.googleapis.com

The following table describes the service accounts that are created automatically:

Service account	Purpose	Roles
anthos-baremetal-gcr	GDCV for Bare Metal uses this service account to download container images from Container Registry.	None
anthos-baremetal-connect	Connect Agent uses this service account to maintain a connection between your cluster and Google Cloud. This enables access to the cluster and to workload management features, including the Google Cloud console and the Connect gateway to interact with your cluster.	roles/gkehub.connect
anthos-baremetal-register	Connect Agent uses this service account to register your clusters with a fleet.	roles/gkehub.admin
anthos-baremetal-cloud-ops	Stackdriver Agent uses this service account to export logs and metrics from clusters to Cloud Logging and Cloud Monitoring.	roles/logging.logWriter roles/monitoring.metricWriter roles/stackdriver.resourceMetadata.writer roles/opsconfigmonitoring.resourceMetadata.writer roles/monitoring.dashboardEditor

What's next

Create basic clusters