Upgrading GKE on-prem

This page explains how to upgrade GKE on-prem.

To upgrade GKE on-prem, you upgrade your admin workstation. Then, you upgrade your clusters.

Before you begin

• Be sure to follow these instructions from your local workstation or laptop. Don't follow these instructions from your existing admin workstation.
• Check your clusters' current version.
• Review the Release notes and known issues affecting upgrading.
• Review Versions.

Also, read through the following considerations:

About downtime during upgrades

Resource	Description
Admin cluster	When an admin cluster is down, user cluster control planes and workloads on user clusters continue to run, unless they were affected by a failure that caused the downtime
User cluster control plane	Typically, you should expect no noticeable downtime to user cluster control planes. However, long-running connections to the Kubernetes API server might break and would need to be re-established. In those cases, the API caller should retry until it establishes a connection. In the worst case, there can be up to one minute of downtime during an upgrade. Note: If user cluster nodes are unable reach the user control plane during the upgrade, new workloads are not scheduled to the cluster. Existing workloads are unaffected.
User cluster nodes	If an upgrade requires a change to user cluster nodes, GKE on-prem recreates the nodes in a rolling fashion, and reschedules Pods running on these nodes. You can prevent impact to your workloads by configuring appropriate PodDisruptionBudgets and anti-affinity rules.

Sequential upgrading

GKE on-prem supports sequential upgrading, which means that the cluster you want to upgrade must be at the immediate previous patch version.

You can't upgrade your clusters directly to the latest version from a version that is more than one patch version behind. If your cluster is more than one patch version behind, you must sequentially upgrade that cluster through each patch version.

Example

If you want to upgrade to the 1.1.0 version, and your admin workstation and user clusters are running the older 1.0.1 version:

• 1.0.1 (oldest version)
• 1.0.2
• 1.1.0 (latest version)

Then you must sequentially upgrade to the 1.0.2 version and then the 1.1.0 version by performing the following steps:

1. Upgrade your admin workstation from 1.0.1 to 1.0.2.
2. Upgrade your clusters from 1.0.1 to 1.0.2.
3. Upgrade your admin workstation from 1.0.2 to 1.1.
4. Upgrade your clusters from 1.0.2 to 1.1.0.

Back up your GKE on-prem configuration file and kubeconfig files

When you upgrade your admin workstation, Terraform deletes the admin workstation VM and then replaces it with an upgraded admin workstation.

Before you perform the admin workstation upgrade, you must first back up your GKE on-prem configuration file and your clusters' kubeconfig files The gkectl create cluster command creates the kubeconfig files for your admin cluster ([ADMIN_CLUSTER_KUBECONFIG]) and user cluster ([USER_CLUSTER_KUBECONFIG]). See an example in the basic installation. After your admin workstation is upgraded, you copy those same files to the upgraded admin workstation.

Upgrading admin workstation

You use Terraform to upgrade your admin workstation.

Important: You must first backup the admin workstation's files because Terraform deletes the current admin workstation VM and replaces it with a new VM.

The upgraded admin workstation includes the following entities at the same version as the admin workstation's Open Virtualization Appliance (OVA) file:

• gkectl
• full bundle

Downloading the OVA

From Downloads, download the admin workstation OVA file for the version to which you're upgrading.

To download the latest OVA, run the following command:

gcloud storage cp gs://gke-on-prem-release/admin-appliance/1.3.2-gke.1/gke-on-prem-admin-appliance-vsphere-1.3.2-gke.1.{ova,ova.1.sig} ~/

Note: To verify this file, see Verifying the OVA using openssl.

Importing the OVA to vSphere and marking it as a VM template

In the following sections, you:

1. Create some variables declaring elements of your vCenter Server and vSphere environment.
2. Import the admin workstation OVA to vSphere and mark it as a VM template.

Creating variables for govc

Before you import the admin workstation OVA to vSphere, you need to provide govc some variables declaring elements of your vCenter Server and vSphere environment:

Note: You can find many of these values by logging in to the vCenter Server.

export GOVC_URL=https://[VCENTER_SERVER_ADDRESS]/sdk
export GOVC_USERNAME=[VCENTER_SERVER_USERNAME]
export GOVC_PASSWORD=[VCENTER_SERVER_PASSWORD]
export GOVC_DATASTORE=[VSPHERE_DATASTORE]
export GOVC_DATACENTER=[VSPHERE_DATACENTER]
export GOVC_INSECURE=true

You can choose to use vSphere's default resource pool or create your own:

# If you want to use a resource pool you've configured yourself, export this variable:
export GOVC_RESOURCE_POOL=[VSPHERE_CLUSTER]/Resources/[VSPHERE_RESOURCE_POOL]

# If you want to use vSphere's default resource pool, export this variable instead:
export GOVC_RESOURCE_POOL=[VSPHERE_CLUSTER]/Resources

where:

• [VCENTER_SERVER_ADDRESS] is your vCenter Server's IP address or hostname.
• [VCENTER_SERVER_USERNAME] is the username of an account that holds the Administrator role or equivalent privileges in vCenter Server.
• [VCENTER_SERVER_PASSWORD] is the vCenter Server account's password.
• [VSPHERE_DATASTORE] is the name of the datastore you've configured in your vSphere environment.
• [VSPHERE_DATACENTER] is the name of the datacenter you've configured in your vSphere environment.
• [VSPHERE_CLUSTER] is the name of the cluster you've configured in your vSphere environment.
For using a non-default resource pool,
• [VSPHERE_RESOURCE_POOL] is the name of the resource pool you've configured to your vSphere environment.

Importing the OVA to vSphere: Standard switch

If you are using a vSphere Standard Switch, import the OVA to vSphere using this command:

govc import.ova -options - ~/gke-on-prem-admin-appliance-vsphere-1.3.2-gke.1.ova <<EOF
{
  "DiskProvisioning": "thin",
  "MarkAsTemplate": true
}
EOF

Importing the OVA to vSphere: Distributed switch

If you are using a vSphere Distributed Switch, import the OVA to vSphere using this command, where [YOUR_DISTRIBUTED_PORT_GROUP_NAME] is the name of your distributed port group:

govc import.ova -options - ~/gke-on-prem-admin-appliance-vsphere-1.3.2-gke.1.ova <<EOF
{
  "DiskProvisioning": "thin",
  "MarkAsTemplate": true,
  "NetworkMapping": [
      {
          "Name": "VM Network",
          "Network": "[YOUR_DISTRIBUTED_PORT_GROUP_NAME]"
      }
  ]
}
EOF

Setting the Terraform template variable for the new admin workstation VM

In your admin workstation's TFVARS file, set vm_template to the version to which you're upgrading. The value of vm_template looks like this, where [VERSION] is the OVA's version:

gke-on-prem-admin-appliance-vsphere-[VERSION]

Using Terraform to upgrade your admin workstation

To upgrade your admin workstation, run the following command. This command deletes the current admin workstation VM and replaces it with an upgraded VM:

Note: Be sure that you import the OVA to vSphere before running this command.

terraform init && terraform apply -auto-approve -input=false

Connecting to your admin workstation

SSH in to your admin workstation by running the following command:

ssh -i ~/.ssh/vsphere_workstation ubuntu@[IP_ADDRESS]

Copy your backed up configuration and kubeconfig files

Earlier, you backed up your GKE on-prem configuration file and your clusters' kubeconfig files. Now, you should copy those files back to your upgraded admin workstation.

Upgrading clusters

After upgrading your admin workstation and connecting to it, perform the following steps:

Verify that enough IP addresses are available

Before you upgrade, be sure that you have enough IP addresses available for your clusters.

DHCP

During an upgrade, GKE on-prem creates one temporary node in the admin cluster and one temporary node in each associated user cluster. Make sure that your DHCP server can provide enough IP addresses for these temporary nodes. For more information, see IP addresses needed for admin and user clusters.

Static IPs

During an upgrade, GKE on-prem creates one temporary node in the admin cluster and one temporary node in each associated user cluster. For your admin cluster and each of your user clusters, verify that you have reserved enough IP addresses. For each cluster, you need to have reserved at least one more IP address than the number of cluster nodes. For more information, see Configuring static IP addresses.

Determine the number of nodes in your admin cluster:

kubectl --kubeconfig [ADMIN_CLUSTER_KUBECONFIG] get nodes

where [ADMIN_CLUSTER_KUBECONFIG] is the path of your admin cluster's kubeconfig file.

Next, view the addresses reserved for your admin cluster:

kubectl get cluster --kubeconfig [ADMIN_CLUSTER_KUBECONFIG] -o yaml

In the output, in the reservedAddresses field, you can see the number of IP addresses that are reserved for the admin cluster nodes. For example, the following output shows that there are five IP addresses reserved for the admin cluster nodes::

...
reservedAddresses:
- gateway: 21.0.135.254
  hostname: admin-node-1
  ip: 21.0.133.41
  netmask: 21
- gateway: 21.0.135.254
  hostname: admin-node-2
  ip: 21.0.133.50
  netmask: 21
- gateway: 21.0.135.254
  hostname: admin-node-3
  ip: 21.0.133.56
  netmask: 21
- gateway: 21.0.135.254
  hostname: admin-node-4
  ip: 21.0.133.47
  netmask: 21
- gateway: 21.0.135.254
  hostname: admin-node-5
  ip: 21.0.133.44
  netmask: 21

The number of reserved IP addresses should be at least one more than the number of nodes in the admin cluster. If this is not the case, you can reserve an additional address by editing the Cluster object.

Open the Cluster object for editing:

kubectl edit cluster --kubeconfig [ADMIN_CLUSTER_KUBECONFIG]

Under reservedAddresses, add an additional block that has gateway, hostname, ip, and netmask.

Go through the same procedure for each of your user clusters.

To determine the number of nodes in a user cluster:

kubectl --kubeconfig [USER_CLUSTER_KUBECONFIG] get nodes

where [USER_CLUSTER_KUBECONFIG] is the path of your user cluster's kubeconfig file.

To view the addresses reserved for a user cluster:

kubectl get cluster --kubeconfig [ADMIN_CLUSTER_KUBECONFIG] \
-n [USER_CLUSTER_NAME] [USER_CLUSTER_NAME] -o yaml

where:

• [ADMIN_CLUSTER_KUBECONFIG] is the path of your admin cluster's kubeconfig file.

• [USER_CLUSTER_NAME] is the name of the user cluster.

To edit the Cluster object of a user cluster:

kubectl edit cluster --kubeconfig [ADMIN_CLUSTER_KUBECONFIG] \
-n [USER_CLUSTER_NAME] [USER_CLUSTER_NAME]

Modifying the configuration file

On your admin workstation VM, edit the configuration file that you used to create your admin and user clusters. Set the value of bundlepath, where [VERSION] is the GKE on-prem version to which you're upgrading your clusters:

bundlepath: /var/lib/gke/bundles/gke-onprem-vsphere-[VERSION].tgz

About automatically-enabled features

A new GKE on-prem version might include new features or support for specific VMware vSphere features. Sometimes, upgrading to a GKE on-prem version automatically enables such features. You learn about new features in GKE on-prem's Release notes. New features are sometimes surfaced in the GKE on-prem configuration file.

Disabling new features via the configuration file

If you need to disable a new feature that is automatically enabled in a new GKE on-prem version and driven by the configuration file, perform the following steps before you upgrade your cluster:

1. From your upgraded admin workstation, create a new configuration file with a different name from your current configuration file:

   gkectl create-config --config [CONFIG_NAME]

2. Open the new configuration file and the feature's field. Close the file.

3. Open your current configuration file and add the new feature's field in the appropriate specification.

4. Provide the field a false or equivalent value.

5. Save the configuration file. Proceed with upgrading your clusters.

You should always review the Release notes before you upgrade your clusters. You cannot declaratively change an existing cluster's configuration after you upgrade it.

Running gkectl prepare

The gkectl prepare command performs the following tasks:

• If necessary, copy a new node OS image to your vSphere environment, and mark the OS image as a template.

• Push updated Docker images, specified in the new bundle, to your private Docker registry, if you have configured one.

To perform the preceding tasks, run the following command:

gkectl prepare --config [CONFIG_FILE] [FLAGS]

where:

• [CONFIG_FILE] is the GKE on-prem configuration file you're using to perform the upgrade.

• [FLAGS] is an optional set of flags. For example, you could include the --skip-validation-infra flag to skip checking of your vSphere infrastructure.

Upgrading your admin cluster

Run the following command:

gkectl upgrade admin \
--kubeconfig [ADMIN_CLUSTER_KUBECONFIG] \
--config [CONFIG_FILE] \
[FLAGS]

where:

• [ADMIN_CLUSTER_KUBECONFIG] is the admin cluster's kubeconfig file.

• [CONFIG_FILE] is the GKE on-prem configuration file you're using to perform the upgrade.

• [FLAGS] is an optional set of flags. For example, you could include the --skip-validation-infra flag to skip checking of your vSphere infrastructure.

Upgrading your user cluster

To upgrade a user cluster, your admin cluster must be upgraded to the desired version or higher prior to upgrading your user cluster.

gkectl

From your admin workstation, run the following command:

gkectl upgrade cluster \
--kubeconfig [ADMIN_CLUSTER_KUBECONFIG] \
--config [CONFIG_FILE] \
--cluster-name [CLUSTER_NAME] \
[FLAGS]

where:

• [ADMIN_CLUSTER_KUBECONFIG] is the admin cluster's kubeconfig file.

• [CLUSTER_NAME] is the name of the user cluster you're upgrading.

• [CONFIG_FILE] is the GKE on-prem configuration file you're using to perform the upgrade.

• [FLAGS] is an optional set of flags. For example, you could include the --skip-validation-infra flag to skip checking of your vSphere infrastructure.

Console

You can choose to register your user clusters with Google Cloud console during installation or after you've created them. You can view and log in to your registered GKE on-prem clusters and your Google Kubernetes Engine clusters from Google Cloud console's GKE menu.

When an upgrade becomes available for GKE on-prem user clusters, a notification appears in Google Cloud console. Clicking this notification displays a list of available versions and a gkectl command you can run to upgrade the cluster:

1. Visit the GKE menu in Google Cloud console.

   Visit the GKE menu

2. Under the Notifications column for the user cluster, click Upgrade available, if available.

3. Copy the gkectl upgrade cluster command.

4. From your admin workstation, run the gkectl upgrade cluster command, where [ADMIN_CLUSTER_KUBECONFIG] is the admin cluster's kubeconfig file, [CLUSTER_NAME] is the name of the user cluster you're upgrading, and [CONFIG_FILE] is the GKE on-prem configuration file you're using to perform the upgrade.

Resuming an upgrade

If a user cluster upgrade is interrupted after the admin cluster is successfully upgraded, you can resume the user cluster upgrade by running the same upgrade command with the --skip-validation-all flag:

gkectl upgrade cluster \
--kubeconfig [ADMIN_CLUSTER_KUBECONFIG] \
--config [CONFIG_FILE] \
--cluster-name [CLUSTER_NAME] \
--skip-validation-all

About resuming an admin cluster upgrade

You shouldn't interrupt an admin cluster upgrade. Currently, admin cluster upgrades aren't always resumable. If an admin cluster upgrade is interrupted for any reason, you should contact support for assistance.

Known issues

The following known issues affect upgrading clusters.

Version 1.1.0-gke.6, 1.2.0-gke.6: stackdriver.proxyconfigsecretname field removed

The stackdriver.proxyconfigsecretname field was removed in version 1.1.0-gke.6. GKE on-prem's preflight checks will return an error if the field is present in your configuration file.

To work around this, before you upgrade to 1.2.0-gke.6, delete the proxyconfigsecretname field from your configuration file.

Stackdriver references old version

Before version 1.2.0-gke.6, a known issue prevents Stackdriver from updating its configuration after cluster upgrades. Stackdriver still references an old version, which prevents Stackdriver from receiving the latest features of its telemetry pipeline. This issue can make it difficult for Google Support to troubleshoot clusters.

After you upgrade clusters to 1.2.0-gke.6, run the following command against admin and user clusters:

kubectl --kubeconfig=[KUBECONFIG] \
-n kube-system --type=json patch stackdrivers stackdriver \
-p '[{"op":"remove","path":"/spec/version"}]'

where [KUBECONFIG] is the path to the cluster's kubeconfig file.

Disruption for workloads with PodDisruptionBudgets

Currently, upgrading clusters can cause disruption or downtime for workloads that use PodDisruptionBudgets (PDBs).

Version 1.2.0-gke.6: Prometheus and Grafana disabled after upgrading

In user clusters, Prometheus and Grafana get automatically disabled during upgrade. However, the configuration and metrics data are not lost. In admin clusters, Prometheus and Grafana stay enabled.

For instructions, refer to the GKE on-prem release notes.

Version 1.1.2-gke.0: Deleted user cluster nodes aren't removed from vSAN datastore

For instructions, refer to the GKE on-prem release notes.

Version 1.1.1-gke.2: Data disk in vSAN datastore folder can be deleted

If you're using a vSAN datastore, you need to create a folder in which to save the VMDK. A known issue requires that you provide the folder's universally unique identifier (UUID) path, rather than its file path, to vcenter.datadisk. This mismatch can cause upgrades to fail.

For instructions, refer to the GKE on-prem release notes.

Upgrading to version 1.1.0-gke.6 from version 1.0.2-gke.3: OIDC issue

Version 1.0.11, 1.0.1-gke.5, and 1.0.2-gke.3 clusters that have OpenID Connect (OIDC) configured cannot be upgraded to version 1.1.0-gke.6. This issue is fixed in version 1.1.1-gke.2.

If you configured a version 1.0.11, 1.0.1-gke.5, or 1.0.2-gke.3 cluster with OIDC during installation, you are not able to upgrade it. Instead, you should create new clusters.

Upgrading to version 1.0.2-gke.3 from version 1.0.11

Version 1.0.2-gke.3 introduces the following OIDC fields (usercluster.oidc). These fields enable logging in to a cluster from Google Cloud console:

• usercluster.oidc.kubectlredirecturl
• usercluster.oidc.clientsecret
• usercluster.oidc.usehttpproxy

If you want to use OIDC, the clientsecret field is required even if you don't want to log in to a cluster from Google Cloud console. To use OIDC, you might need to provide a placeholder value for clientsecret:

oidc:
  clientsecret: "secret"

Nodes fail to complete their upgrade process

If you have PodDisruptionBudget objects configured that are unable to allow any additional disruptions, node upgrades might fail to upgrade to the control plane version after repeated attempts. To prevent this failure, we recommend that you scale up the Deployment or HorizontalPodAutoscaler to allow the node to drain while still respecting the PodDisruptionBudget configuration.

To see all PodDisruptionBudget objects that do not allow any disruptions:

kubectl get poddisruptionbudget --all-namespaces -o jsonpath='{range .items[?(@.status.disruptionsAllowed==0)]}{.metadata.name}/{.metadata.namespace}{"\n"}{end}'

Appendix

About VMware DRS rules enabled in version 1.1.0-gke.6

As of version 1.1.0-gke.6, GKE on-prem automatically creates VMware Distributed Resource Scheduler (DRS) anti-affinity rules for your user cluster's nodes, causing them to be spread across at least three physical hosts in your datacenter. As of version 1.1.0-gke.6, this feature is automatically enabled for new clusters and existing clusters.

Before you upgrade, be sure that your vSphere environment meets the following conditions:

• VMware DRS is enabled. VMware DRS requires vSphere Enterprise Plus license edition. To learn how to enable DRS, see Enabling VMware DRS in a cluster
• The vSphere user account provided in the vcenter field has the Host.Inventory.EditCluster permission.
• There are at least three physical hosts available.

Disabling VMware DRS before upgrading to 1.1.0-gke.6

If you do not want to enable this feature for your existing user clusters—for example, if you don't have enough hosts to accommodate the feature—perform the following steps before you upgrade your user clusters:

1. Open your existing GKE on-prem configuration file.
2. Under the usercluster specification, add the antiaffinitygroups field:

   usercluster:
         ...
         antiaffinitygroups:
           enabled: false
   

3. Save the file.
4. Use the configuration file to upgrade. Your clusters are upgraded, but the feature is not enabled.

Alternate upgrade scenario

If security patches like Common Vulnerabilities and Exposures (CVEs), don't exist for your version of the admin workstation, you can use the alternative method to upgrade GKE on-prem. The alternative method upgrades only gkectl and your user clusters. You don't upgrade your admin workstation in this scenario. Therefore, you must only use this alternate method after all security updates are applied to your existing admin workstation.

Important: There is a known issue with this alternate upgrade method when you upgrade your user clusters from version 1.2.2 to 1.3.0 using the Bundle download. During the upgrade process a timeout might occur that causes your user cluster upgrade to fail. To avoid this issue, you must perform the full upgrade process that includes upgrading your admin workstation using the OVA file.

1. Update your Google Cloud CLI components: gcloud components update.
2. Download and install the latest gkectl tool.
3. Download the bundle.
4. Upgrade your user clusters.

Troubleshooting

For more information, refer to Troubleshooting.

New nodes created but not healthy

Symptoms

New nodes don't register themselves to the user cluster control plane when using manual load balancing mode.

Possible causes

In-node Ingress validation might be enabled that blocks the boot up process of the nodes.

Resolution

To disable the validation, run:

kubectl patch machinedeployment [MACHINE_DEPLOYMENT_NAME] -p '{"spec":{"template":{"spec":{"providerSpec":{"value":{"machineVariables":{"net_validation_ports": null}}}}}}}' --type=merge

Diagnosing cluster issues using gkectl

Use gkectl diagnosecommands to identify cluster issues and share cluster information with Google. See Diagnosing cluster issues.

Running gkectl commands verbosely

-v5

Logging gkectl errors to stderr

--alsologtostderr

Locating gkectl logs in the admin workstation

Even if you don't pass in its debugging flags, you can view gkectl logs in the following admin workstation directory:

/home/ubuntu/.config/gke-on-prem/logs

Locating Cluster API logs in the admin cluster

If a VM fails to start after the admin control plane has started, you can try debugging this by inspecting the Cluster API controllers' logs in the admin cluster:

1. Find the name of the Cluster API controllers Pod in the kube-system namespace, where [ADMIN_CLUSTER_KUBECONFIG] is the path to the admin cluster's kubeconfig file:

   kubectl --kubeconfig [ADMIN_CLUSTER_KUBECONFIG] -n kube-system get pods | grep clusterapi-controllers

2. Open the Pod's logs, where [POD_NAME] is the name of the Pod. Optionally, use grep or a similar tool to search for errors:

   kubectl --kubeconfig [ADMIN_CLUSTER_KUBECONFIG] -n kube-system logs [POD_NAME] vsphere-controller-manager