Troubleshooting Management Cluster Issues

This topic includes tips to help you troubleshoot standalone management cluster deployments.

For information about troubleshooting workload clusters, see Troubleshooting Workload Cluster Issues. You can find additional workaround for known issues in this release in the Release Notes or in VMware Knowledge Base articles.

Prerequisite

Some of the procedures below use the kind CLI. To install kind, see Installation in the kind documentation.

Common Tasks

Connect to Cluster Nodes with SSH

You can use SSH to connect to individual nodes of standalone management clusters or workload clusters. To do so, the SSH key pair that you created when you deployed the management cluster must be available on the machine on which you run the SSH command. Consequently, you must run ssh commands on the machine on which you run tanzu commands.

The SSH keys that you register with the management cluster and that are used by any workload clusters that you deploy from the management cluster are associated with the following user accounts:

  • vSphere management cluster and Tanzu Kubernetes nodes running on both Photon OS and Ubuntu: capv
  • AWS bastion nodes: ubuntu
  • AWS management cluster and Tanzu Kubernetes nodes running on Ubuntu: ubuntu
  • AWS management cluster and Tanzu Kubernetes nodes running on Amazon Linux: ec2-user
  • Azure management cluster and Tanzu Kubernetes nodes (always Ubuntu): capi

To connect to a node by using SSH, run one of the following commands from the machine that you use as the bootstrap machine:

  • vSphere nodes: ssh capv@node-address
  • AWS bastion nodes and management cluster and workload nodes on Ubuntu: ssh ubuntu@node-address
  • AWS management cluster and Tanzu Kubernetes nodes running on Amazon Linux: ssh ec2-user@node-address
  • Azure nodes: ssh capi@node-address

Because the SSH key is present on the system on which you are running the ssh command, no password is required.

Delete Users, Contexts, and Clusters with kubectl

To clean up your kubectl state by deleting some or all of its users, contexts, and clusters:

  1. Open your ~/.kube-tkg/config file.

  2. For the user objects that you want to delete, run:

    kubectl config unset users.USERNAME --kubeconfig ~/.kube-tkg/config
    

    Where USERNAME is the name property of each top-level user object, as listed in the config file.

  3. For the context objects that you want to delete, run:

    kubectl config unset contexts.CONTEXT-NAME --kubeconfig ~/.kube-tkg/config
    

    Where CONTEXT-NAME is the name property of each top-level context object, as listed in the config file, typically of the form contexts.mycontext-admin@mycontext.

  4. For the cluster objects that you want to delete, run:

    kubectl config unset clusters.CLUSTER-NAME --kubeconfig ~/.kube-tkg/config
    

    Where CLUSTER-NAME is the name property of each top-level cluster object, as listed in the config file.

  5. If the config files list the current context as a cluster that you deleted, unset the context:

    kubectl config unset current-context --kubeconfig ~/.kube-tkg/config
    
  6. If you deleted management clusters that are tracked by the tanzu CLI, delete them from the tanzu CLI’s state by running tanzu context delete as described in Delete Management Clusters from Your Tanzu CLI Configuration.

Deactivate nfs-utils on Photon OS Nodes

Problem

In Tanzu Kubernetes Grid v1.1.2 and later, nfs-utils is enabled by default. If you do not require nfs-utils, you can remove it from cluster node VMs.

Solution

To deactivate nfs-utils on clusters that you deploy with Tanzu Kubernetes Grid v1.1.2 or later, use SSH to log in to the cluster node VMs and run the following command:

tdnf erase nfs-utils

Target Platform

CAPA resource tagging issue causes reconciliation failure during AWS management cluster deploy and upgrade

Problem

Due to a resource tagging issue in upstream Cluster API Provider AWS (CAPA), offline deployments cannot access the ResourceTagging API, causing reconciliation failures during management cluster creation or upgrade.

Solution

In an offline AWS environment, set EXP_EXTERNAL_RESOURCE_GC=false in your local environment or in the management cluster configuration file before running tanzu mc create or tanzu mc upgrade.

Failed Validation, Credentials Error on AWS

Problem

Running tanzu management-cluster create or tanzu mc create fails with an error similar to the following:

Validating the pre-requisites...
Looking for AWS credentials in the default credentials provider chain

Error: : Tkg configuration validation failed: failed to get AWS client: NoCredentialProviders: no valid providers in chain
caused by: EnvAccessKeyNotFound: AWS_ACCESS_KEY_ID or AWS_ACCESS_KEY not found in environment
SharedCredsLoad: failed to load shared credentials file
caused by: FailedRead: unable to open file
caused by: open /root/.aws/credentials: no such file or directory
EC2RoleRequestError: no EC2 instance role found
caused by: EC2MetadataError: failed to make EC2Metadata request

Solution

Tanzu Kubernetes Grid uses the default AWS credentials provider chain. Before creating a management or a workload cluster on AWS, you must configure your AWS account credentials as described in Configure AWS Credentials.

Failed Validation, Legal Terms Error on Azure

Problem

Before creating a standalone management or workload cluster on Azure, you must accept the legal terms that cover the VM image used by cluster nodes. Running tanzu mc create or tanzu cluster create without having accepted the license fails with an error like:

User failed validation to purchase resources. Error message: 'You have not accepted the legal terms on this subscription: '*********' for this plan. Before the subscription can be used, you need to accept the legal terms of the image.

Solution

If this happens, accept the legal terms and try again:

Management Cluster

Clean Up After an Unsuccessful Management Cluster Deployment

Problem

An unsuccessful attempt to deploy a standalone management cluster leaves orphaned objects in your cloud infrastructure and on your bootstrap machine.

Solution

  1. Monitor your tanzu mc create command output either in the terminal or Tanzu Kubernetes Grid installer interface. If the command fails, it prints a help message that includes the following: “Failure while deploying management cluster… To clean up the resources created by the management cluster: tkg delete mc….”
  2. Run tanzu mc delete YOUR-CLUSTER-NAME. This command removes the objects that it created in your infrastructure and locally.

You can also use the alternative methods described below:

  • Bootstrap machine cleanup:

    • To remove a kind cluster, use the kind CLI. For example:

      kind get clusters
      kind delete cluster --name tkg-kind-example1234567abcdef
      
    • To remove Docker objects, use the docker CLI. For example, docker rm, docker rmi, and docker system prune -a --volumes.

      Caution

      If you are running Docker processes that are not related to Tanzu Kubernetes Grid on your system, remove unneeded Docker objects individually.

  • Target platform cleanup:

    • vSphere: Locate, power off, and delete the VMs and other resources that were created by Tanzu Kubernetes Grid.
    • AWS: Log in to your Amazon EC2 dashboard and delete the resources manually or use an automated solution.
    • Azure: In Resource Groups, open your AZURE_RESOURCE_GROUP. Use checkboxes to select and Delete the resources that were created by Tanzu Kubernetes Grid, which contain a timestamp in their names.

Cannot Delete Management Cluster on AWS

Problem

After running tanzu mc delete on AWS, tanzu mc get and other Tanzu CLI commands no longer list the deleted management cluster, but:

  • The cluster is not deleted from AWS infrastructure, and its nodes still appear in the Amazon EC2 dashboard
  • Management cluster logs show error cluster infrastructure is still being provisioned: VpcReconciliationFailed.

Solution

This behavior occurs when TKG uses expired or otherwise invalid AWS account credentials. To prevent or recover from this situation:

  • Prevent: Update AWS account credentials as described in Configure AWS Account Credentials using either AWS Credential Profiles or local, static environment variables.

    • You cannot use AWS instance profile credentials to delete a management cluster.
  • Recover using the EC2 Dashboard: Delete the management cluster nodes from the EC2 dashboard manually

  • Recover using the CLI:

    1. In the kind cluster that remains on the bootstrap machine due to failed management cluster deletion, correct the AWS credential secret:

      kubectl get secret capa-manager-bootstrap-credentials -n capa-system -ojsonpath="{.data.credentials}"| base64 -d
      
    2. Edit the secret to include the AWS credentials:

      [default]
      aws_access_key_id = <key id>
      aws_secret_access_key = <access_key>
      region = <region>
      
    3. Run tanzu mc delete again.

Kind Cluster Remains after Deleting Management Cluster

Problem

Running tanzu mc delete removes the management cluster, but fails to delete the local kind cluster from the bootstrap machine.

Solution

  1. List all running kind clusters and remove the one that looks like tkg-kind-unique_ID:

    kind delete cluster --name tkg-kind-unique_ID
    
  2. List all running clusters and identify the kind cluster.

    docker ps -a
    
  3. Copy the container ID of the kind cluster and remove it.

    docker kill container-ID
    

Machines Stuck After Management Cluster Fails to Deploy

Problem

Your stanalone management cluster fails to deploy because machines are stuck, waiting for remediation.

Solution

For a management cluster that you deployed with the dev plan, which has only one control plane node, you must redeploy the management cluster. For management clusters with more than one control plane node, you can identify and delete the stuck machines.

  1. Retrieve the status of the management cluster. For example:

    kubectl -n tkg-system get cluster my-mgmt-cluster -o yaml
    
  2. Find the names of the stuck machines from the output of the previous step. A stuck machine is marked as WaitingForRemediation. For example, the name of the stuck machine is my-mgmt-cluster-zpc7t in the following output:

    status:
      conditions:
      - lastTransitionTime: "2021-08-25T15:44:23Z"
        message: KCP can't remediate if current replicas are less or equal then 1
        reason: WaitingForRemediation @ Machine/my-mgmt-cluster-zpc7t
        severity: Warning
        status: "False"
        type: Ready
    
  3. Increase the machine health check timeout values for the control plane nodes to greater than the default, 5m. For example:

    tanzu cluster machinehealthcheck control-plane set my-cluster --mhc-name my-control-plane-mhc --unhealthy-conditions "Ready:False:10m,Ready:Unknown:10m"
    

    For more information about updating a MachineHealthCheck object, see Create or Update a MachineHealthCheck Object in Configure Machine Health Checks for Workload Clusters.

  4. Set kubectl to the context of your management cluster. For example:

    kubectl config use-context mgmt-cluster-admin@mgmt-cluster
    
  5. Delete the stuck machines.

    kubectl delete machine MACHINE-NAME
    

    Where MACHINE-NAME is the name of the machine you located in an earlier step.

  6. Wait for the KubeadmControlPlane controller to redeploy the machine.

Restore the ~/.config/tanzu Directory

Problem

The ~/.config/tanzu directory on the bootstrap machine has been accidentally deleted or corrupted. The Tanzu CLI creates and uses this directory, and cannot function without it.

Solution

To restore the contents of the ~/.config/tanzu directory:

  1. To identify existing Tanzu Kubernetes Grid management clusters, run:

    kubectl --kubeconfig ~/.kube-tkg/config config get-contexts
    

    The command output lists names and contexts of all management clusters created or added by the Tanzu CLI.

  2. For each management cluster listed in the output, restore it to the ~/.config/tanzu directory and CLI by running:

    tanzu context create --management-cluster --kubeconfig ~/.kube-tkg/config --context MGMT-CLUSTER-CONTEXT --name MGMT-CLUSTER
    

Running tanzu management-cluster create on macOS Results in kubectl Version Error

Problem

If you run the tanzu management-cluster create or tanzu mc create command on macOS with the latest stable version of Docker Desktop, it fails with the error message like:

Error: : kubectl prerequisites validation failed: kubectl client version v1.15.5 is less than minimum supported kubectl client version 1.27.5

This happens because Docker Desktop symlinks an older version of kubectl into the path.

Solution

Place a newer supported version of kubectl in the path before Docker’s version.

Recover Management Cluster Credentials

If you have lost the credentials for a standalone management cluster, for example, by inadvertently deleting the .kube-tkg/config file on the system on which you run tanzu commands, you can recover the credentials from the management cluster control plane node.

  1. Run tanzu mc create to recreate the .kube-tkg/config file.
  2. Obtain the public IP address of the management cluster control plane node, from vSphere, AWS, or Azure.
  3. Use SSH to log in to the management cluster control plane node.

    See Connect to Cluster Nodes with SSH above for the credentials to use for each target platform.

  4. Access the admin.conf file for the management cluster.

    sudo vi /etc/kubernetes/admin.conf
    

    The admin.conf file contains the cluster name, the cluster user name, the cluster context, and the client certificate data.

  5. Copy the cluster name, the cluster user name, the cluster context, and the client certificate data into the .kube-tkg/config file on the system on which you run tanzu commands.

Pinniped

Adding external identity management to an existing deployment may require setting dummy VSPHERE_CONTROL_PLANE_ENDPOINT value

Problem

Integrating an external identity provider with an existing TKG deployment may require setting a dummy VSPHERE_CONTROL_PLANE_ENDPOINT value in the management cluster configuration file used to create the add-on secret, as described in Generate the Pinniped Add-on Secret for the Management Cluster

Deactivating Pinniped requires manual Secret delete on legacy clusters

Problem

When you deactivate external identity management on a management cluster, the unused Pinniped Secret object remains present on legacy workload clusters.

If a user then tries to access the cluster using an old kubeconfig, a login popup will appear and fail.

Workaround

Manually delete the legacy cluster’s Pinniped Secret as described in Deactivate Identity Management.

Post-Deploy Pinniped Job Fails

Problem

Upgrading to Tanzu Kubernetes Grid v2.4 returns an error similar to the following:

 Operation cannot be fulfilled on certificates.cert-manager.io "pinniped-cert": the object has been modified; please apply your changes to the latest version and try again

Solution

This error may occur if the Pinniped post-deploy job conflicts with a component’s upgrade process. Follow these steps to delete and redeploy the job.

  1. Delete the Pinniped post-deploy job.

    kubectl delete jobs.batch -n pinniped-supervisor pinniped-post-deploy-job
    
  2. Wait about 5 minutes for kapp-controller to redeploy the post-deploy job.

  3. Check the status of the Pinniped app.

    kubectl get app -n tkg-system pinniped
    NAME       DESCRIPTION           SINCE-DEPLOY   AGE
    pinniped   Reconcile succeeded   5s             49m
    

    If the DESCRIPTION shows Reconciling, wait a few minutes, then check again. Once it shows Reconcile succeeded continue to the next step.

  4. Check the status of the Pinniped post-deploy job.

    kubectl get jobs -n pinniped-supervisor
    NAME                             COMPLETIONS   DURATION   AGE
    pinniped-post-deploy-job-ver-1   1/1           9s         62s
    

Pinniped Authentication Error on Workload Cluster After Management Cluster Upgrade

Problem

You recently upgraded your management cluster. When attempting to authenticate to a workload cluster associated with this management cluster, you receive an error message similar to the following:

Error: could not complete Pinniped login: could not perform OIDC discovery for "https://IP:PORT": Get "https://IP:PORT/.well-known/openid-configuration": x509: certificate signed by unknown authority

Solution

This happens because the copy of the Pinniped supervisor CA bundle that the workload cluster is using is out of date. To update the supervisor CA bundle, follow the steps below:

  1. Set the kubectl context to the management cluster.

  2. Obtain the base64-encoded CA bundle and the issuer endpoint from the pinniped-info ConfigMap:

    kubectl get configmap pinniped-info -n kube-public -o jsonpath={.data.issuer_ca_bundle_data} > /tmp/ca-bundle && kubectl get configmap pinniped-info -n kube-public -o jsonpath={.data.issuer} > /tmp/supervisor-endpoint
    
  3. Obtain the values.yaml section from the Pinniped add-on secret for the workload cluster:

    kubectl get secret WORKLOAD-CLUSTER-NAME-pinniped-addon -n WORKLOAD-CLUSTER-NAMESPACE -o jsonpath="{.data.values\.yaml}" | base64 -d > values.yaml
    

    This secret is located on the management cluster.

  4. In the values.yaml file created above, update the supervisor_ca_bundle_data key to match the CA bundle from the pinniped-info ConfigMap. Additionally, ensure that the supervisor_svc_endpoint matches the issuer endpoint.

  5. Apply your update by base64 encoding the edited values.yaml file and replacing it in the workload cluster secret. This command differs depending on the OS of your environment. For example:

    Linux:

    kubectl patch secret/WORKLOAD-CLUSTER-NAME-pinniped-addon -n WORKLOAD-CLUSTER-NAMESPACE -p "{\"data\":{\"values.yaml\":\"$(base64 -w 0 < values.yaml)\"}}" --type=merge
    

    macOS:

    kubectl patch secret/WORKLOAD-CLUSTER-NAME-pinniped-addon -n WORKLOAD-CLUSTER-NAMESPACE -p "{\"data\":{\"values.yaml\":\"$(base64 < values.yaml)\"}}" --type=merge
    
  6. On the workload cluster, confirm that the Pinniped app successfully reconciled the changes:

    kubectl get app pinniped -n tkg-system
    
  7. Authenticate to the cluster. For example:

    kubectl get pods -A --kubeconfig my-cluster-credentials
    

Pods

Pods Are Stuck in Pending on Cluster Due to vCenter Connectivity

Problem

When you run kubectl get pods -A on the created cluster, some pods remain in pending state.

You run kubectl describe pod -n pod-namespace pod-name on an affected pod and see the following event:

n node(s) had taint {node.cloudprovider.kubernetes.io/uninitialized: true}, that the pod didn't tolerate

Solution

Ensure there is connectivity and firewall rules in place to ensure communication between the cluster and vCenter. For firewall ports and protocols requirements, see the vSphere listings in VMware Ports and Protocols.

Tanzu CLI

Tanzu CLI Cannot Reach Management Cluster

Problem

Running tanzu CLI commands returns an error similar to the following:

Failed to invoke API on cluster : the server has asked for the client to provide credentials, retrying

Solution

See Update Management Cluster Certificate in Your Tanzu CLI Configuration and Cannot access the clusters using tkg/tanzu cli commands in Tanzu Kubernetes Grid.

Windows CMD: Extraneous characters in CLI output column headings

Problem

In the Windows command prompt (CMD), Tanzu CLI command output that is formatted in columns includes extraneous characters in column headings. The issue does not occur in Windows Terminal or PowerShell.

Solution

On Windows bootstrap machines, run the Tanzu CLI from Windows Terminal.

CLI temporarily misreports status of recently deleted nodes when MHC are deactivated

Problem

When machine health checks (MHCs) are deactivated, then Tanzu CLI commands such as tanzu cluster status may not report up-to-date node state while infrastructure is being recreated.

Solution

None.

Tanzu Kubernetes Grid Installer Interface

Tanzu Kubernetes Grid UI Does Not Display Correctly on Windows

Problem

When you run the tanzu management-cluster create --ui or tanzu mc create --ui command on a Windows system, the UI opens in your default browser, but the graphics and styling are not applied. This happens because a Windows registry is set to application/x-css.

Solution

  1. In Windows search, enter regedit to open the Registry Editor utility.
  2. Expand HKEY_CLASSES_ROOT and select .js.
  3. Right-click Content Type and select Modify.
  4. Set the Value to application/javascript and click OK.
  5. Run the tanzu mc create --ui command again to relaunch the UI.

NSX Advanced Load Balancer

With NSX ALB, you cannot create clusters with identical names

Problem

If you are using NSX Advanced Load Balancer for workloads (AVI_ENABLE) or the control plane (AVI_CONTROL_PLANE_HA_PROVIDER) the Avi Controller may fail to distinguish between identically-named clusters.

Solution:

Set a unique CLUSTER_NAME value for each cluster. Do not create multiple management clusters with the same CLUSTER_NAME value, even from different bootstrap machines.

Requests to NSX Advanced Load Balancer VIP fail with the message no route to host

Problem

If the total number of LoadBalancer type Service is large, and if all of the Service Engines are deployed in the same L2 network, requests to the NSX Advanced Load Balancer VIP can fail with the message no route to host.

This occurs because the default ARP rate limit on Service Engines is 100.

Solution

Set the ARP rate limit to a larger number. This parameter is not tunable in NSX Advanced Load Balancer Essentials, but it is tunable in NSX Advanced Load Balancer Enterprise Edition.

Ignorable AKODeploymentConfig error during management cluster creation

Problem

Running tanzu management-cluster create to create a management cluster with NSX ALB outputs the error no matches for kind AKODeploymentConfig in version networking.tkg.tanzu.vmware.com/v1alpha1.

Solution

The error can be ignored. For more information, see this KB article.

Multus CNI fails on medium and smaller pods with NSX Advanced Load Balancer

Problem

On vSphere, workload clusters with medium or smaller worker nodes running the Multus CNI package with NSX ALB can fail with Insufficient CPU or other errors.

Solution

To use Multus CNI with NSX ALB, deploy workload clusters with worker nodes of size large or extra-large.

Upgrade

Upgrading clusters on Azure fails

Problem

On Azure, upgrading management clusters and workload clusters fails with errors such as context deadline exceeded or unable to upgrade management cluster: error waiting for kubernetes version update for kubeadm control plane. This happens because operations on Azure sometimes take longer than on other platforms.

Solution

Run the tanzu management-cluster upgrade or tanzu cluster upgrade again, specifying a longer timeout in the --timeout flag. The default timeout is 30m0s.

Upgrade fails for standalone management clusters originally created in TKG v1.3 or earlier

Problem

In TKG v2.4, the components that turn a generic cluster into a TKG standalone management cluster are packaged in a Carvel package tkg-pkg. Standalone management clusters that were originally created in TKG v1.3 or earlier lack a configuration secret that the upgrade process requires in order to install tkg-pkg, causing upgrade to fail.

Solution

Perform the additional steps listed in Upgrade Standalone Management Clusters for standalone management clusters created in TKG v1.3 or earlier.

check-circle-line exclamation-circle-line close-line
Scroll to top icon