Skip to main content

Troubleshooting steps for errors during a cluster deployment

The following steps will help you troubleshoot errors in the event issues arise while deploying a cluster.

Scenario - PV/PVC Stuck in Pending Status for EKS Cluster Using AL2023 AMI​

After deploying an Amazon EKS cluster using an Amazon Linux 2023 (AL2023) Amazon Machine Image (AMI), PersistentVolumes (PVs) or PersistentVolumeClaims (PVCs) are stuck in a pending status.

Example
NAMESPACE   NAME                                 STATUS    VOLUME   CAPACITY   ACCESS MODES   STORAGECLASS            VOLUMEATTRIBUTESCLASS   AGE   VOLUMEMODE
wordpress   data-wordpress-wordpress-mariadb-0   Pending                                      spectro-storage-class   <unset>                 16m   Filesystem
wordpress   wordpress-wordpress                  Pending                                      spectro-storage-class   <unset>                 16m   Filesystem

This issue can arise when an add-on pack or Helm chart that requires a PV or PVC is deployed to an existing cluster or while creating a new cluster.

The PV or PVC provisioning fails because IAM Roles for Service Accounts (IRSA) have not been configured for the AWS CSI packs such as Amazon EBS CSI, Amazon EFS CSI, and Amazon Cloud Native. It is also required if using the AWS Application Loadbalancer.

For instances launched on AL2023, IMDSv2 is enforced by default, and IRSA is the recommended approach for providing IAM permissions to Amazon EBS CSI and Amazon EFS CSI.

Debug Steps​

  1. Log in to Palette.

  2. Ensure you are in the correct project scope.

  3. From the left main menu, navigate to the Profiles page. Find and click on your cluster profile.

  4. Create a new version of the cluster profile.

  5. Select the Kubernetes layer of your cluster profile.

  6. Use the YAML editor to configure IRSA roles for the managedControlPlane and managedMachinePool resources.

    Example configuration
    managedControlPlane:
    ---
    irsaRoles:
      - name: "{{.spectro.system.cluster.name}}-irsa-cni"
        policies:
          - arn:aws:iam::aws:policy/AmazonEKS_CNI_Policy
        serviceAccount:
          name: aws-node
          namespace: kube-system
      - name: "{{.spectro.system.cluster.name}}-irsa-csi"
        policies:
          - arn:aws:iam::aws:policy/service-role/AmazonEBSCSIDriverPolicy
    ---
    managedMachinePool:
      roleAdditionalPolicies:
        - "arn:aws:iam::aws:policy/service-role/AmazonEBSCSIDriverPolicy"

  7. Click Confirm Updates after editing.

  8. Select the Storage layer of your cluster profile.

  9. Use the YAML editor to add an IAM role ARN annotation to the AWS CSI Driver so that the IRSA role is correctly referenced. Replace <aws-account-id> with your AWS account ID.

    Example configuration
    charts:
    ...
      aws-ebs-csi-driver:
      ...
        controller:
        ...
          serviceAccount:
            # A service account will be created for you if set to true. Set to false if you want to use your own.
            create: true
            name: ebs-csi-controller-sa
            annotations: {
              "eks.amazonaws.com/role-arn":"arn:aws:iam::<aws-account-id>:role/{{.spectro.system.cluster.name}}-irsa-csi"
            }
            ## Enable if EKS IAM for SA is used
            # eks.amazonaws.com/role-arn: arn:<partition>:iam::<account>:role/ebs-csi-role
            automountServiceAccountToken: true

  10. Update the custom labels for the AWS CSI Driver to retrigger the deployment.

    Example
    charts:
    ...
      aws-ebs-csi-driver:
      ...
        customLabels: {
          restart: "true"
        }

  11. Click Confirm Updates after editing.

  12. Click Save Changes on the cluster profile page.

  13. Update your cluster to use the new cluster profile version that you created with these changes. Refer to Update a Cluster for guidance.

  14. Wait for the nodes to be repaved and the AWS CSI Driver to be redeployed.

  15. Check that the PV or PVC status is Bound by issuing one of the following kubectl commands.

    Example command for PVs
    kubectl get pv --output wide
    Example output for PVs
    NAME               CAPACITY   ACCESS MODES   RECLAIM POLICY   STATUS   CLAIM                                         STORAGECLASS            VOLUMEATTRIBUTESCLASS   AGE   VOLUMEMODE
    pv-xyz...          10Gi       RWO            Delete           Bound    wordpress/data-wordpress-wordpress-mariadb-0  spectro-storage-class   <unset>                 16m   Filesystem
    pv-abc...          8Gi        RWO            Delete           Bound    wordpress/wordpress-wordpress                 spectro-storage-class   <unset>                 16m   Filesystem
    Example command for PVCs
    kubectl get pvc --all-namespaces --output wide
    Example output for PVCs
    NAMESPACE   NAME                                 STATUS    VOLUME      CAPACITY   ACCESS MODES   STORAGECLASS            VOLUMEATTRIBUTESCLASS   AGE   VOLUMEMODE
    wordpress   data-wordpress-wordpress-mariadb-0   Bound     pvc-xyz...  10Gi       RWO            spectro-storage-class   <unset>                 16m   Filesystem
    wordpress   wordpress-wordpress                  Bound     pvc-abc...  8Gi        RWO            spectro-storage-class   <unset>                 16m   Filesystem

Scenario - Instances Continuously Delete Every 30 Minutes​

An instance is launched and terminated every 30 minutes prior to completion of its deployment, and the Events Tab lists errors with the following message:

Failed to update kubeadmControlPlane Connection timeout connecting to Kubernetes Endpoint

This behavior can occur when Kubernetes services for the launched instance fail to start properly. Common reasons for why a service may fail are:

  • The specified image could not be pulled from the image repository.
  • The cloud init process failed.

Debug Steps​

  1. Initiate an SSH session with the Kubernetes instance using the SSH key provided during provisioning, and log in as user spectro. If you are initiating an SSH session into an installer instance, log in as user ubuntu.

    ssh --identity_file <_pathToYourSSHkey_> spectro@X.X.X.X

  2. Elevate the user access.

    sudo -i

  3. Verify the Kubelet service is operational.

    systemctl status kubelet.service

  4. If the Kubelet service does not work as expected, do the following. If the service operates correctly, you can skip this step.

    1. Navigate to the /var/log/ folder.

      cd /var/log/

    2. Scan the cloud-init-output file for any errors. Take note of any errors and address them.

      cat cloud-init-output.log

  5. If the kubelet service works as expected, do the following.

    • Export the kubeconfig file.

      export KUBECONFIG=/etc/kubernetes/admin.conf

    • Connect with the cluster's Kubernetes API.

      kubectl get pods --all-namespaces

    • When the connection is established, verify the pods are in a Running state. Take note of any pods that are not in Running state.

      kubectl get pods -o wide

    • If all the pods are operating correctly, verify their connection with the Palette API.

      • For clusters using Gateway, verify the connection between the Installer and Gateway instance:

        curl -k https://<KUBE_API_SERVER_IP>:6443

      • For Public Clouds that do not use Gateway, verify the connection between the public Internet and the Kubernetes endpoint:

        curl -k https://<KUBE_API_SERVER_IP>:6443
        info

        You can obtain the URL for the Kubernetes API using this command: kubectl cluster-info.

  6. Check stdout for errors. You can also open a support ticket. Visit our support page.

Scenario - Deployment Violates Pod Security​

Cluster deployment fails with the following message.

Error creating: pods <name of pod> is forbidden: violates PodSecurity "baseline:v<k8s version>": non-default capabilities …

This can happen when the cluster profile uses Kubernetes 1.25 or later and also includes packs that create pods that require elevated privileges.

Debug Steps​

To address this issue, you can change the Pod Security Standards of the namespace where the pod is being created.

  1. Log in to Palette.

  2. Navigate to the left Main Menu and click on Profiles.

  3. Select the profile you are using to deploy the cluster. Palette displays the profile stack and details.

  4. Click on the pack layer in the profile stack that contains the pack configuration.

  5. In the pack's YAML file, add a subfield in the pack section called namespaceLabels if it does not already exist.

  6. In the namespaceLabels section, add a line with the name of your namespace as the key and add pod-security.kubernetes.io/enforce=privileged,pod-security.kubernetes.io/enforce-version=v<k8s_version> as its value. Replace <k8s_version> with the version of Kubernetes on your cluster and only include the major and minor version following the lowercase letter v. For example, v1.25 and v1.28.

  7. If a key matching your namespace already exists, add the labels to the value corresponding to that key.

warning

We recommend only applying the labels to namespaces where pods fail to be created. If your pack creates multiple namespaces, and you are unsure which ones contain pods that need the elevated privileges, you can access the cluster with the kubectl CLI and use the kubectl get pods command. This command lists pods and their namespaces so you can identify the pods that are failing at creation.

For guidance in using the CLI, review Access Cluster with CLI. To learn more about kubectl pod commands, refer to the Kubernetes documentation.

Examples​

The following example shows a pack that creates a namespace called "monitoring". In this example, the monitoring namespace does not have any pre-existing labels. You need to add the namespaceLabels line as well as the corresponding key-value pair under it to apply the labels to the monitoring namespace.

pack:
  namespace: "monitoring"

  namespaceLabels:
    "monitoring": "pod-security.kubernetes.io/enforce=privileged,pod-security.kubernetes.io/enforce-version=v1.28"

This second example is similar to the first one. However, in this example, the monitoring key already exists under namespaceLabels, with its original value being "org=spectro,team=dev". Therefore, you add the labels to the existing value:

pack:
  namespace: "monitoring"

  namespaceLabels:
    "monitoring": "org=spectro,team=dev,pod-security.kubernetes.io/enforce=privileged,pod-security.kubernetes.io/enforce-version=v1.28"

Scenario - Nutanix CAPI Deployment Updates​

In the event that the internal Nutanix cluster Cluster API (CAPI) configurations are updated, there is a possibility that the cluster's Kubernetes deployments may encounter issues, resulting in an unhealthy cluster. This can occur when the CAPI changes may be incompatible with the newer version of Palette. The following steps will help you troubleshoot and resolve this issue.

Debug Steps​

  1. Open a terminal session and ensure you have the kubectl CLI installed. If you do not have the CLI installed, you can download it from the Kubernetes website.

  2. Set up your terminal session to use the kubeconfig file for your Nutanix cluster. You can find the kubeconfig for your cluster in the Palette UI by visiting the Nutanix cluster's details page. Check out the Access Cluster with CLI guide for guidance on how to set up your terminal session to use the kubeconfig file.

  3. To restore the cluster to a healthy state, you need to delete the following deployments so that Palette can re-create them with the updated machine template. Issue the following commands to delete the following three deployments.

     kubectl delete deployment capi-controller-manager --namespace capi-system
     kubectl delete deployment capi-kubeadm-bootstrap-controller-manager --namespace capi-kubeadm-bootstrap-system
     kubectl delete deployment capi-kubeadm-control-plane-controller-manager --namespace capi-kubeadm-control-plane-system

  4. Palette will automatically re-create the deleted deployments with the updated machine template. You can monitor the progress of the re-creation by checking the status of the deployments using the following command.

    kubectl get deployments --all-namespaces