3V0-24.25 Advanced VMware Cloud Foundation 9.0 vSphere Kubernetes Service Questions and Answers

The errors described—specifically the memcache.go failure, the inability to fetch resource lists for Antrea, and the YAML context error—are classic symptoms of aConfiguration Context mismatch. In VCF 9.0, there are two distinct layers of API interaction: theSupervisor Cluster API(used for management tasks like creating clusters) and theGuest Cluster API(used for deploying workloads within the VKS).

When an administrator upgrades a Supervisor, the API endpoint or the available API groups may change. If the administrator attempts to run kubectl commands against a VKS cluster while their kubeconfig context is still pointing to the Supervisor (or vice versa), the client will encounter " mapping values " errors and " unable to handle request " errors because it is sending requests to an endpoint that does not recognize those specific resource definitions (like Antrea stats in the wrong context). To resolve this, the administrator must ensure they have switched to the correct context using kubectl config use-context < cluster-name > after the Supervisor update to ensure the local client is communicating with the correct API server and version of the Kubernetes binaries.

When you build Kubernetes-aware policy constructs, “groups” are commonly used to collect objects so you can apply consistent controls (security posture, traffic rules, observability scope, etc.) to a set of endpoints. In VCF 9.0 documentation, the Kubernetes member types that can be used for group-based collection includeKubernetes NodeandKubernetes Serviceas supported member object categories. Nodes represent the worker compute endpoints that run workloads, while Services represent stable networking front-ends for sets of pods (and are often the anchoring object for policy and routing decisions at the Kubernetes layer). Using Node-based grouping helps apply policies to the infrastructure execution points where workloads run, and Service-based grouping helps apply policies consistently to application entry points and east-west communication targets, regardless of pod churn. This combination is especially useful in Kubernetes-centric operational models because it aligns policy scope with both (1) where workloads execute (nodes) and (2) how workloads are exposed and discovered (services).

VCF 9.0 states that you can provision Kubernetes clusters using both GUI and CLI approaches, and it explicitly calls out the CLI clients: “the VCF CLI and kubectl provide command-line interfaces for provisioning Kubernetes clusters.” That directly maps tokubectl (A)andVCF CLI (E)as supported clients for provisioning and lifecycle operations. Separately, VCF 9.0 explains that vSphere administrators can “manage and monitor vSphere Pods, VMs, and VKS clusters by using the vSphere Client,” which corresponds tovSphere UI (C)in the question. In addition, the vSphere Client is used to access Supervisor-facing self-service interfaces (for example, the Local Consumption Interface through the vSphere Client), reinforcing vSphere UI as an operational entry point for managing Supervisor-backed services and workloads.

By contrast,Cluster APIis a controller framework (not an operator “client” for admins in this context), andesxtop/esxcliare ESXi host tools that do not represent the documented, supported interfaces for provisioning and managing VKS clusters at the Kubernetes service layer.

Storage

Resources

When a stateful pod (like Harbor’s database) is stuck Pending/ContainerCreating or repeatedly failing to attach/mount storage, one of the first checks is whether the PVC’sAccess Modesmatch what the workload and backing storage support (for example, ReadWriteOnce vs ReadWriteMany). In Kubernetes, the PVC and PV explicitly showAccess Modes(for example Access Modes: RWO) when you inspect them.

In the vSphere Namespace UI, theStoragetab is where namespace storage is managed and reviewed (including storage policy assignments that back storage classes/PVCs), making it the natural place to validate the storage configuration being consumed by the Harbor database PVC. VMware’s VCF documentation directs administrators to navigate to a namespace and use theStoragetab for storage-related settings.

TheResourcestab is the Kubernetes-object oriented view for the namespace, where you can locate the PVC object associated with the failing pod and inspect its properties/YAML (including the requested storage class and access mode intent). Namespace permissions explicitly include the ability to view Kubernetes resources within the namespace, which aligns with checking PVC details there.

cert-manager addresses the operational risk described (manual creation/renewal causing outages) by making certificate lifecycle management anative, declarative Kubernetes workflow. Instead of treating TLS certificates as manually managed files, cert-manager extends the Kubernetes API with custom resources such asCertificate,Issuer, andClusterIssuer, so certificates and their issuing policies become first-class objects that can be version-controlled and automatically reconciled. This directly satisfies the requirement to usetrusted certificates issued through the corporate CA, because an Issuer/ClusterIssuer can represent that corporate CA integration and define how certificate requests are fulfilled. Once configured, cert-manager continuously monitors certificate validity andautomatically renews and rotatescertificates before expiration, then updates the referenced Kubernetes Secrets so Ingress endpoints remain protected without human intervention. In a vSphere Supervisor / VKS environment, VMware also uses cert-manager on the Supervisor for automated certificate rotation in platform integrations (for example, rotating certificates used by monitoring components), reinforcing the model of automated rotation rather than manual certificate handling.

VCF 9.0 describes a vSphere Namespace as the control point where administrators defineresource boundariesfor workloads, explicitly stating that vSphere administrators can create namespaces and “configure them with specified amount ofmemory, CPU, and storage,” and that you can “set limits forCPU, memory, storage” for a namespace. This directly supports stepAas a day-2 control to keep multi-tenant clusters governed and prevent resource contention across different teams and workload types.

For monitoring and optimization, VCF 9.0 explains that day-2 operations include visibility into utilization and operational metrics for VKS clusters, noting that application teams can use day-2 actions and gain insights intoCPU and memory utilizationand advanced metrics (including contention and availability) for VKS clusters. In addition, VCF 9.0 monitoring guidance for VKS clusters states thatTelegraf and Prometheusmust be installed and configured on each VKS cluster before metrics and object details are sent for monitoring, and that VCF Operations supports metrics collection for Kubernetes objects (namespaces, nodes, pods, containers) via Prometheus. Since the Prometheus stack commonly includes Grafana dashboards for visualization, deployingPrometheus + Grafanamatches the required monitoring/optimization outcome inC.

VCF 9.0 distinguishes betweenbacking up the Supervisor control planeandbacking up workloadsthat run on the Supervisor, includingvSphere Pods. In the “Considerations for Backing Up and Restoring Workload Management” table, the scenario “Backup and restore vSphere Pods” explicitly lists the required tool as“Velero Plugin for vSphere”, with the guidance to “Install and configure the plug-in on the Supervisor.”

The same document is explicit thatstandalone Velero with Restic is not valid for vSphere Pods, stating: “You cannot use Velero standalone with Restic to backup and restore vSphere Pods. You must use the Velero Plugin for vSphere installed on the Supervisor.”

vCenter file-based backup is documented for restoring theSupervisor control plane state, not for backing up and restoring vSphere Pod workloads themselves. Therefore, to meet the requirement to protect third-party applications running asvSphere Pods, the architect should propose theVelero Plugin for vSphere.

Configuration Sequence (in order):

Create the vSphere Namespace

Assign the zones

Select workload networking

Assign the zonal storage policy

Define resource quotas / limits

Grant RBAC / permissions

A zonal vSphere Namespace is created as a standard namespace first, then “zonalized” by associating it with one or morevSphere Zonesso workloads can be scheduled according to zone placement rules. You start bycreating the namespacebecause it is the tenancy and governance container where networking, storage access, quotas, and permissions are applied. Next, youassign the zones, since zone association is what makes the namespace “zonal” and determines where Kubernetes workloads (and their node pools) are allowed to land.

With zones set, you configureworkload networking, because namespaces must have the correct network attachment and IP behavior for the workloads that will be placed across the selected zones. Then youassign the zonal storage policy, ensuring that persistent volumes can be provisioned using storage that is valid/available for the zone placement model you selected. After networking and storage access are defined, you setresource quotas/limits(CPU, memory, storage) so multi-tenant consumption stays within governance boundaries. Finally, yougrant RBAC/permissionsso the right DevOps/users can consume the namespace and provision clusters/workloads under the enforced controls.

In VCF 9.0 Workload Management, workload backup for VKS clusters (and vSphere Pods) is performed using Velero, specifically theVelero Plugin for vSpherefor Supervisor/VKS scenarios. The documentation describes that you must firstprovide an S3-compatible object storeas part of installing and configuring the Velero Plugin for vSphere, because backup data (Kubernetes metadata and volume snapshot data movement) relies on object storage rather than “namespace storage.” The backup workflow further indicates that when you create a backup, the system uploadsKubernetes metadata to the object store, and persistent volume snapshot handling is coordinated through the plugin components (Snapshot/Upload custom resources).

Therefore, the correct configuration pattern is to point Velero’s backup target (BackupStorageLocation) at anS3-compatible object storeendpoint so metadata and snapshot payloads have a durable destination. This aligns with the documented prerequisite that object storage is required to enable backup/restore operations for persistent workloads, and it is the mechanism Velero uses to store backup artifacts for later restore operations in VKS environments.

In VCF 9.0 VKS clusters, network policy is a core Kubernetes networking control implemented by the cluster’s CNI (Antrea or Calico). The VCF documentation’s “VKS Cluster Networking” table describesNetwork policyas the feature that “controls what traffic is allowed to and from selected pods and network endpoints,” and identifies Antrea or Calico as the providers for this capability. That definition precisely matches optionB: it governs pod-to-pod and pod-to-external endpoint communication rules. This is different from ingress routing (which the same table describes separately as “Cluster ingress … routing for inbound pod traffic”), so option C is not correct for “network policy.” It is also different from NodePort behavior (external access via a port on each worker node through the Kubernetes network proxy), which is explicitly listed as “Service type: NodePort.” Finally, creating/operating clusters natively in Supervisor is a broader lifecycle function (Cluster API/VKS API), not the definition of network policy. Therefore,NetworkPolicyis the Kubernetes-layer mechanism to define and enforce allowed traffic flows.

Kubernetes Service

When adding a Kubernetes cluster to anexisting vSphere Namespace, the workflow is initiated from within that namespace’s context in the vSphere Client. The administrator first opens the target namespace (as shown in the left navigation where the namespace is selected), then uses theKubernetes Servicearea to create and manage Kubernetes clusters associated with that namespace. In the namespaceSummaryview, theKubernetes Servicetile/card provides the entry point for cluster lifecycle actions, typically via an action link such asManage(and from thereCreate Cluster). This is because the Kubernetes cluster is anamespace-scoped resourcein the Supervisor environment: it inherits namespace-level policies and configuration such as permissions/RBAC, resource limits, storage policies, and networking selections that have already been defined for that namespace. Navigating to other areas likeStorageorNetworkis used to validate or adjust prerequisites (for example, storage policy availability or network settings), but the actual “create cluster” operation is launched fromKubernetes Servicewithin the namespace to ensure the cluster is created under the correct tenant boundary and governance model.

In Workload Management, updating the Supervisor and updating VKS clusters are related but distinct lifecycle operations. The Supervisor runs its own Kubernetes distribution, while VKS clusters consume vSphere Kubernetes releases (VKrs). These are “delivered differently,” with Supervisor Kubernetes releases and VKrs each having their own release cadence and compatibility constraints. As a result, successfully updating the Supervisor control plane does not automatically change the Kubernetes version running inside an existing VKS workload cluster; the VKS cluster must be updated to a compatible VKr separately. This mismatch is exactly why an application can still fail after a Supervisor update: the DevOps engineer is still deploying onto a cluster that hasn’t been updated to the Kubernetes version required by the application (or by the API versions/features it depends on). Additionally, Workload Management enforces sequential minor-version updates and compatibility checks between Supervisor and VKrs, so the correct remediation is to update the VKS cluster to an appropriate VKr that satisfies both application needs and Supervisor compatibility.

VCF 9.0 design guidance for theSupervisor NSX Load Balancer modelstates that theNSX load balancers run on NSX Edges, and sets explicit sizing requirements at theEdge nodelevel. In the “NSX Load Balancer Design Requirements,” VCF 9.0 requires that theNSX Edge cluster must be deployedbecause “NSX Load Balancers run on NSX Edges.” It further requires that“NSX Edge nodes must be deployed with a minimum of Large form factor”because NSX load balancers havefixed resource allocations on NSX Edge nodes, and the Large form factor is needed to accommodate basic system and workload needs.

This directly ties “deployment size” of the load balancer service capacity to theEdge node form factor(Small/Medium/Large, etc.) rather than the number of pods or the number of clusters. The document also notes that if additional load balancers are required, NSX Edges can bescaled up or out, again reinforcing that sizing is anEdge nodesizing decision.

Configuration Sequence (in order):

Deactivate the service.

Confirm uninstallation.

Delete the service.

Confirm deletion.

Uninstalling a Supervisor Service cleanly follows a “stop/remove/cleanup” pattern to avoid orphaned components and to ensure the Supervisor isn’t still reconciling the service while you remove it. First, youDeactivate the serviceso the Supervisor stops managing and running the Harbor service components. This prevents new service resources from being created while you are attempting to remove them and allows the platform to transition the service into a non-operational state safely.

Next, youConfirm uninstallationto initiate the removal workflow. This step acknowledges that the service’s deployed components (such as pods, controllers, and any associated objects) will be removed from the Supervisor-managed environment. After the uninstall workflow is initiated and the service is no longer active, you proceed toDelete the serviceto remove the Harbor service registration/entry from the Supervisor Services list, ensuring it is no longer available to namespaces or consumers. Finally, youConfirm deletionas the last safeguard, since deletion is typically destructive and removes the service definition from the environment’s available services catalog.

VCF 9.0 explains that VKS releases are distributed throughcontent libraries, and that when you configure asubscribed content libraryyou can choose how content is downloaded and synchronized. Critically for constrained storage sites, the documentation states you can selecton-demand (“when needed”)download so thatonly the metadatais synchronized, and theactual image content is not downloadeduntil you explicitly need it—“as a result storage space is saved.”

This directly addresses the problem of a remote site with minimal datastore capacity: choosingWhen neededprevents the subscribed library from immediately caching all published VKS images locally. The vSphere content library guidance reinforces this behavior: with “download contents only when needed,” the system synchronizesmetadata only, which “saves storage space,” and you download a specific item’s content only when you synchronize that item for use.

Therefore, selectingDownload content → When neededis the correct action to avoid consuming all available datastore space.