Upgrading Google Distributed Cloud

This page explains how to upgrade Google Distributed Cloud. If your user cluster is managed by the Anthos On-Prem API, and you want to use the Google Cloud console or the Google Cloud CLI to upgrade the cluster, see Upgrade a user cluster using Anthos On-Prem API clients.

Overview of the upgrade process

You can upgrade directly to any version that is in the same minor release or the next minor release. For example, you can upgrade from 1.14.0 to 1.14.1, or from 1.13.1 to 1.14.0.

If you are upgrading to a version that is not part of the next minor release, you must upgrade through one version of each minor release between your current version and your desired version. For example, if you want to upgrade from version 1.12.2 to version 1.14.0, you can't upgrade directly. You must first upgrade from 1.12.2 to 1.13.x, and then upgrade to 1.14.0.

This topic discusses how to upgrade from version 1.13.x to version 1.14.y.

Review the cluster upgrade best practices before you start your upgrade process.

Here is the general workflow for upgrading.

  1. Upgrade your admin workstation to the target version of your upgrade.

  2. From your admin workstation, upgrade your user clusters.

  3. After all of the user clusters have been upgraded, you can upgrade your admin cluster from the admin workstation. This step is optional unless you require the features available in the upgrade.

Asynchronous user cluster upgrade

For a user cluster upgrade, there are two variations of the gkectl upgrade cluster command:

  • Asynchronous (recommended)
  • Synchronous

With the asynchronous variation, the command starts the upgrade and then completes. You don't need to watch the output of the command for the entire duration of the upgrade. Instead, you can periodically check on the upgrade progress by running gkectl list clusters and gkectl describe clusters.

To use the asynchronous variation, include the --async flag in the command. For details, see Upgrade a user cluster.

Certificate rotation during upgrade

During an upgrade, leaf certificates are rotated, but CA certificates are not rotated. You must manually rotate your CA certificates at least once every five years. For more information, see Rotate user cluster certificate authorities and Rotate admin cluster CA certificates.

Prepare for upgrade

Before you start an upgrade, take a snapshot of your cluster. The snapshot will help with troubleshooting if there is an issue during the upgrade.

Before you created your admin workstation, you filled in an admin workstation configuration file that was generated by gkeadm create config. The default name for this file is admin-ws-config.yaml.

In addition, your workstation has an information file. The default name of this file is the same as the name of your admin workstation.

Locate your admin workstation configuration file and your information file. You need them to do the upgrade steps. If these files are in your current directory and they have their default names, then you won't need to specify them when you run the upgrade commands. If these files are in another directory, or if you have changed the filenames, then you specify them by using the --config and --info-file flags.

If your output information file is missing, you can re-create it. See Re-create an information file if missing.

Upgrade your admin workstation

  1. Download gkeadm:

    gkeadm upgrade gkeadm --target-version TARGET_VERSION
    

    Replace TARGET_VERSION with the target version of your upgrade.

  2. Upgrade your admin workstation:

    gkeadm upgrade admin-workstation --config AW_CONFIG_FILE --info-file INFO_FILE
    

    Replace the following:

    • AW_CONFIG_FILE: the path of your admin workstation configuration file. You can omit this flag if the file is in your current directory and has the name admin-ws-config.yaml.

    • INFO_FILE: the path of your information file. You can omit this flag if the file is in your current directory. The default name of this file is the same as the name of your admin workstation.

The preceding command performs the following tasks:

  • Backs up all files in the home directory of your current admin workstation. These include:

    • Your admin cluster configuration file. The default name is admin-cluster.yaml.
    • Your user cluster configuration file. The default name is user-cluster.yaml.
    • The kubeconfig files for your admin cluster and your user clusters.
    • The root certificate for your vCenter server. Note that this file must have owner read and owner write permission.
    • The JSON key file for your component access service account. Note that this file must have owner read and owner write permission.
    • The JSON key files for your connect-register and logging-monitoring service accounts.
  • Creates a new admin workstation, and copies all the backed-up files to the new admin workstation.

  • Deletes the old admin workstation.

Verify that enough IP addresses are available

Before you upgrade your clusters, be sure that you have allocated enough IP addresses. You can allocate additional IP addresses as needed. See Manage node IP addresses to determine how many IP addresses you need.

Upgrade a user cluster

There are two types of user cluster upgrade you can do at the command line:

  • Asynchronous
  • Synchronous

Asynchronous upgrade

Proceed with the following steps on your admin workstation:

  1. Run gkectl prepare to import OS images to vSphere:

    gkectl prepare \
      --bundle-path /var/lib/gke/bundles/gke-onprem-vsphere-TARGET_VERSION.tgz \
      --kubeconfig ADMIN_CLUSTER_KUBECONFIG
    
  2. If your cluster has a Windows node pool, run gkectl prepare windows, and update the osImage field for the node pool. For detailed instructions, see Upgrade user cluster with Windows node pools.

  3. Run the pre-upgrade tool to check the cluster health and configuration.

  4. In the user cluster configuration file, set gkeOnPremVersion to the target version of your upgrade.

  5. On your admin workstation, start an asynchronous upgrade:

    gkectl upgrade cluster \
      --kubeconfig ADMIN_CLUSTER_KUBECONFIG \
      --config USER_CLUSTER_CONFIG \
      --async
    

    The preceding command completes, and you can continue to use your user cluster while the upgrade is in progress.

  6. To see the status of the upgrade:

    gkectl list clusters --kubeconfig ADMIN_CLUSTER_KUBECONFIG
    

    The output shows a value for the cluster STATE. If the cluster is still upgrading, the value of STATE is UPGRADING. For example:

    NAMESPACE             NAME    READY   STATE       AGE   VERSION
    my-uc-gkeonprem-mgmt  my-uc   False   UPGRADING   9h    1.14.0-gke.1
    

    The possible values for STATE are PROVISIONING, UPGRADING, DELETING, UPDATING, RUNNING, RECONCILING, ERROR, and UNKNOWN.

  7. To get more details about the upgrade progress and cluster events:

    gkectl describe clusters --kubeconfig ADMIN_CLUSTER_KUBECONFIG \
      --cluster USER_CLUSTER_NAME -v 5
    

    The output shows the OnPremUserCluster custom resource for the specified user cluster, which includes cluster status, conditions, and events.

    We record events for the start and end of each critical upgrade phase, including:

    • ControlPlaneUpgrade
    • MasterNodeUpgrade
    • AddonsUpgrade
    • NodePoolsUpgrade

    Example output:

    Events:
    Type    Reason                      Age    From                            Message
    ----     ------                     ----   ----                            -------
    Normal  NodePoolsUpgradeStarted     22m    onprem-user-cluster-controller  Creating or updating node pools: pool-2: Creating or updating node pool
    Normal  AddonsUpgradeStarted        22m    onprem-user-cluster-controller  Creating or updating addon workloads
    Normal  ControlPlaneUpgradeStarted  25m    onprem-user-cluster-controller  Creating or updating cluster control plane workloads: deploying user-kube-apiserver-base, ...: 14/15 pods are ready
    Normal  ControlPlaneUpgradeFinished 23m    onprem-user-cluster-controller  Control plane is running
    
  8. When the upgrade is complete, gkectl list clusters shows a STATUS of RUNNING:

    NAMESPACE             NAME    READY   STATE     AGE     VERSION
    my-uc-gkeonprem-mgmt  my-uc   True    RUNNING   9h      1.14.0-gke.1
    

    Also, when the upgrade is complete, gkectl describe clusters shows a LastGKEOnPremVersion field under Status. For example:

    Status:
    Cluster State:  RUNNING
    LastGKEOnOremVersion:  1.14.0-gke.1
    

Troubleshoot asynchronous upgrade

For an asynchronous upgrade, the timeout duration is based on the number of nodes in the cluster. If the upgrade takes longer than the timeout duration, the cluster state is changed from UPGRADING to ERROR, with an event saying that the upgrade operation timed out. Note that the ERROR state here means the upgrade is taking longer than expected, but has not been terminated. The controller continues the reconciliation and keeps retrying the operation.

Usually a timeout is the result of a deadlock caused by a PodDisruptionBudget (PDB). In that case, Pods cannot be evicted from old nodes, and the old nodes cannot be drained. If the Pod eviction takes longer than 10 minutes, we write an event to the OnPremUserCluster object. You can capture the event by running gkectl describe clusters. Then you can adjust the PDB to allow the node to drain. After that, the upgrade can proceed and eventually complete.

Example event:

Warning  PodEvictionTooLong  96s (x2 over 4m7s)  onprem-user-cluster-controller
Waiting too long(>10m0.00000003s) for (kube-system/coredns-856d6dbfdf-dl6nz) eviction.

In addition, when an upgrade is blocked or fails, you can run gkectl diagnose to check for common cluster issues. Based on the result, you can decide whether to perform a manual fix or contact the Anthos support team for further assistance.

Synchronous upgrade

The gkectl upgrade command runs preflight checks. If the preflight checks fail, the command is blocked. You must fix the failures, or use the --skip-preflight-check-blocking flag. You should only skip the preflight checks if you are confident there are no critical failures.

Proceed with these steps on your admin workstation:

  1. Run gkectl prepare to import OS images to vSphere:

    gkectl prepare \
      --bundle-path /var/lib/gke/bundles/gke-onprem-vsphere-TARGET_VERSION.tgz \
      --kubeconfig ADMIN_CLUSTER_KUBECONFIG
    
  2. If your cluster has a Windows node pool, run gkectl prepare windows, and update the osImage field for the node pool. For detailed instructions, see Upgrade user cluster with Windows node pools.

  3. Run the pre-upgrade tool to check the cluster health and configuration.

  4. In the user cluster configuration file, set gkeOnPremVersion to the target version of your upgrade.

  5. Upgrade the cluster:

    gkectl upgrade cluster \
      --kubeconfig ADMIN_CLUSTER_KUBECONFIG \
      --config USER_CLUSTER_CONFIG_FILE
    
  6. If you are upgrading to version 1.14.0 or higher, a new kubeconfig file is generated for the user cluster that overwrites any existing file. To view cluster details in the file, run the following command:

    kubectl config view --kubeconfig USER_CLUSTER_KUBECONFIG

Resume an upgrade

If a user cluster upgrade is interrupted, 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 USER_CLUSTER_CONFIG_FILE \
    --skip-validation-all

Upgrade the admin cluster

Before you begin:

  • Determine if your certificates are up to date, and renew them if necessary.

  • If you are upgrading to version 1.13 or higher, you must first register the admin cluster by filling out the gkeConnect section in the admin cluster configuration file. Run the update cluster command with the configuration file changes.

Do the steps in this section on your new admin workstation. Make sure your gkectl and clusters are the appropriate version for an upgrade, and that you have downloaded the appropriate bundle.

  1. Make sure the bundlepath field in the admin cluster configuration file matches the path of the bundle to which you want to upgrade.

    If you make any other changes to the fields in the admin cluster configuration file, these changes are ignored during the upgrade. To make those changes take effect, you must first upgrade the cluster, and then run an update cluster command with the configuration file changes to make other changes to the cluster.

  2. Run the pre-upgrade tool to check the cluster health and configuration.

  3. Run the following command:

    gkectl upgrade admin \
        --kubeconfig ADMIN_CLUSTER_KUBECONFIG \
        --config ADMIN_CLUSTER_CONFIG_FILE \
        FLAGS
    

    Replace the following:

    • ADMIN_CLUSTER_KUBECONFIG: the admin cluster's kubeconfig file.

    • ADMIN_CLUSTER_CONFIG_FILE: the Google Distributed Cloud admin cluster configuration file on your new admin workstation.

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

If you are upgrading to version 1.14.0 or higher, a new kubeconfig file is generated for the admin cluster that overwrites any existing file. To view cluster details in the file, run the following command:

  kubectl config view --kubeconfig ADMIN_CLUSTER_KUBECONFIG 

If you downloaded a full bundle, and you have successfully run the gkectl prepare and gkectl upgrade admin commands, you should now delete the full bundle to save disk space on the admin workstation. Use this command:

rm /var/lib/gke/bundles/gke-onprem-vsphere-${TARGET_VERSION}-full.tgz

Resuming an admin cluster upgrade

If an admin cluster upgrade is interrupted or fails, the upgrade can be resumed if the admin cluster checkpoint contains the state required to restore the state prior to the interruption.

Warning: Don't repair the admin master with gkectl repair admin-master after a failed upgrade attempt. This will cause the admin cluster to get into a bad state.

Follow these steps:

  1. Check if the admin control plane is healthy before you begin the initial upgrade attempt. See Diagnosing cluster issues. As discussed in that topic, run the gkectl diagnose cluster command for the admin cluster.

  2. If the admin control plane is unhealthy prior to the initial upgrade attempt, repair the admin control plane with the gkectl repair admin-master command.

  3. When you rerun the upgrade command after an upgrade has been interrupted or has failed, use the same bundle and target version as you did in the previous upgrade attempt.

When you rerun the upgrade command, the resumed upgrade recreates admin cluster state from the checkpoint and reruns the entire upgrade. Starting from 1.12.0, if the admin control plane is unhealthy, the upgrade process will directly upgrade to the target version without trying to restore the admin cluster at the source version before proceeding to upgrade.

The upgrade will resume from the point where it failed or exited if the admin cluster checkpoint is available. If the checkpoint is unavailable, the upgrade will fall back to relying on the admin control plane, and therefore the admin control plane must be healthy in order to proceed with the upgrade. After a successful upgrade, the checkpoint is regenerated.

If gkectl exits unexpectedly during an admin cluster upgrade, the kind cluster is not cleaned up. Before you rerun the upgrade command to resume the upgrade, delete the kind cluster:

docker stop gkectl-control-plane && docker rm gkectl-control-plane

After deleting the kind cluster, rerun the upgrade command again.

Roll back an admin workstation after an upgrade

You can roll back the admin workstation to the version used before the upgrade.

During the upgrade, gkeadm records the version before it was upgraded in the output information file. During the rollback, gkeadm uses the version listed to download the older file.

To roll back your admin workstation to the previous version:

gkeadm rollback admin-workstation --config=AW_CONFIG_FILE

You can omit --config=AW_CONFIG_FILE if your admin workstation configuration file is the default admin-ws-config.yaml. Otherwise, replace AW_CONFIG_FILE with the path to the admin workstation configuration file.

The rollback command performs these steps:

  1. Downloads the rollback version of gkeadm.
  2. Backs up the home directory of the current admin workstation.
  3. Creates a new admin workstation using the rollback version of gkeadm.
  4. Deletes the original admin workstation.

Install bundle with a different version for upgrade

If you upgrade your workstation, a bundle with a corresponding version is installed there for upgrading your clusters. If you want a different version, follow these steps to install a bundle for TARGET_VERSION, which is the version to which you want to upgrade.

  1. To check the current gkectl and cluster versions, run this command. Use the flag --details/-d for more detailed information.

    gkectl version --kubeconfig ADMIN_CLUSTER_KUBECONFIG --details
    

    The output provides information about your cluster versions.

  2. Based on the output you get, look for the following issues, and fix them as needed.

    • If the current admin cluster version is more than one minor version lower than the TARGET_VERSION, upgrade all your clusters to be one minor version lower than the TARGET_VERSION.

    • If the gkectl version is lower than 1.11, and you want to upgrade to 1.12.x, you will have to perform multiple upgrades. Upgrade one minor version at a time, until you get to 1.11.x, and then proceed with the instructions in this topic.

    • If the gkectl version is lower than the TARGET_VERSION, upgrade the admin workstation to the TARGET_VERSION.

  3. When you have determined that your gkectl and cluster versions are appropriate for an upgrade, download the bundle.

    Check whether the bundle tarball already exists on the admin workstation.

    stat /var/lib/gke/bundles/gke-onprem-vsphere-TARGET_VERSION.tgz

    If the bundle is not on the admin workstation, download it.

    gcloud storage cp gs://gke-on-prem-release/gke-onprem-bundle/TARGET_VERSION/gke-onprem-vsphere-TARGET_VERSION.tgz /var/lib/gke/bundles/
    

  4. Install the bundle.

    gkectl prepare --bundle-path /var/lib/gke/bundles/gke-onprem-vsphere-TARGET_VERSION.tgz --kubeconfig ADMIN_CLUSTER_KUBECONFIG
    

    Replace ADMIN_CLUSTER_KUBECONFIG with the path of your kubeconfig file. You can omit this flag if the file is in your current directory and has the name kubeconfig.

  5. List available cluster versions, and make sure the target version is included in the available user cluster versions.

    gkectl version --kubeconfig ADMIN_CLUSTER_KUBECONFIG --details

You can now create a user cluster at the target version, or upgrade a user cluster to the target version.

Troubleshooting the upgrade process

If you experience an issue when following the recommended upgrade process, follow these recommendations to resolve them. These suggestions assume that you have begun with a version 1.11.x setup, and are proceeding through the recommended upgrade process.

See also: Troubleshooting cluster creation and upgrade

Troubleshooting a user cluster upgrade issue

Suppose you find an issue with the upgrade version when upgrading a user cluster. You determine from Google Support that the issue will be fixed in an upcoming patch release. You can proceed as follows:

  1. Continue using the current version for production.
  2. Test the patch release in a non-production cluster when it is released.
  3. Upgrade all production user clusters to the patch release version when you are confident.
  4. Upgrade the admin cluster to the patch release version.

Troubleshooting an admin cluster upgrade issue

If you encounter an issue when upgrading the admin cluster, you must contact Google Support to resolve the issue with the admin cluster.

In the meantime, with the new upgrade flow, you can still benefit from new user cluster features without being blocked by the admin cluster upgrade, which allows you to reduce the upgrade frequency of the admin cluster if you want. Your upgrade process can proceed as follows:

  1. Upgrade production user clusters to 1.12.x.
  2. Keep the admin cluster at its earlier version and continue receiving security patches.
  3. Test admin cluster upgrade from 1.11.x to 1.12.x in a test environment, and report issues if there are any;
  4. If your issue is solved by a 1.12.x patch release, you can then choose to upgrade the production admin cluster to this patch release if desired.

Known issues for recent versions

The following known issues might affect upgrades if you are upgrading from version 1.7 or later.

See also: Known issues

Upgrading the admin workstation might fail if the data disk is nearly full

If you upgrade the admin workstation with the gkectl upgrade admin-workstation command, the upgrade might fail if the data disk is nearly full, because the system attempts to back up the current admin workstation locally while upgrading to a new admin workstation. If you cannot clear sufficient space on the data disk, use the gkectl upgrade admin-workstation command with the additional flag --backup-to-local=false to prevent making a local backup of the current admin workstation.

Disruption for workloads with PodDisruptionBudgets

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

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, Google Distributed Cloud 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 username provided in your credentials configuration file has the Host.Inventory.EditCluster permission.

  • There are at least three physical hosts available.

If your vSphere environment does not meet the preceding conditions, you can still upgrade, but for upgrading a user cluster from 1.3.x to 1.4.x, you need to disable anti-affinity groups. For more information, see this known issue in the Google Distributed Cloud release notes.

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.

User cluster nodes

If an upgrade requires a change to user cluster nodes, Google Distributed Cloud 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.

Re-create an information file if missing

If the output information file for your admin workstation is missing, you must re-create this file so you can then proceed with the upgrade. This file was created when you initially created your workstation, and if you have since done an upgrade, it was updated with new information.

The output information file has this format:

Admin workstation version: GKEADM_VERSION
Created using gkeadm version: GKEADM_VERSION
VM name: ADMIN_WS_NAME
IP: ADMIN_WS_IP
SSH key used: FULL_PATH_TO_ADMIN_WS_SSH_KEY
To access your admin workstation:
ssh -i FULL-PATH-TO-ADMIN-WS-SSH-KEY ubuntu@ADMIN-WS-IP

Here is a sample output information file:

Admin workstation version: v1.10.3-gke.49
Created using gkeadm version: v1.10.3-gke.49
VM name: admin-ws-janedoe
IP: 172.16.91.21
SSH key used: /usr/local/google/home/janedoe/.ssh/gke-admin-workstation
Upgraded from (rollback version): v1.10.0-gke.194
To access your admin workstation:
ssh -i /usr/local/google/home/janedoe/.ssh/gke-admin-workstation ubuntu@172.16.91.21

Create the file in an editor, substituting the appropriate parameters. Save the file with a filename that is the same as the VM name in the directory from which gkeadm is run. For example, if the VM name is admin-ws-janedoe, save the file as admin-ws-janedoe.