VMware Cloud Foundation 9.0 introduces a major architectural redesign that replaces the traditional SDDC Manager–centric domain management model with aunified Fleet Management architectureimplemented throughVCF Operations Fleet Manager. In this model, each Workload Domain operates withits own vCenter, but Enhanced Linked Mode (ELM) isremovedto improve isolation, reduce blast radius, and support multi-site scalability. As a result, administrators logged into the vSphere Client of the Management Domaincan no longer manage or expand clusters in other Workload Domains, which explains why the vSphere UI blocks the attempted expansion of WLD-02.

Fleet Manager becomes the new authoritative control plane for lifecycle, topology, host commissioning, and workload domain expansion. Only Fleet Manager maintains the fullglobal viewnecessary to orchestrate cluster addition operations across distributed vCenters and domains. Because WLD-02 is running VCF 9.0 and is fully fleet-aware, its expansion must occur throughVCF Operations Fleet Manager, not through the vSphere Client or legacy SDDC Manager workflows.

Options involving WLD-03 are invalid since that domain is running VCF 5.2.1, is isolated, and cannot participate in fleet-aware operations. SDDC Manager (A) is no longer the correct interface for VCF 9.0 domain expansion operations.

The vSphere Client “no healthy upstream” error is a classic indicator that one or morevCenter backend services are not running or responding, preventing the reverse proxy layer (envoy / nginx) from routing requests to the appropriate upstream services.

Two services in particular are known root causes:

A. vpxd service not running

vpxd is the core vCenter Server service responsible for inventory, host management, and client interaction. If vpxd is stopped, crashed, or restarting, the vSphere Client cannot communicate with backend APIs, resulting in the “no healthy upstream” condition.

B. SSO (vmware-stsd / identity service) not running

Authentication in vCenter depends on the SSO/Identity service. If SSO is unavailable, login sessions cannot be validated, and vCenter marks the upstream service as unhealthy.

Other options donotmatch the behavior:

C (Port 443 closed)would produce a connection failure, not the upstream error.

D (logging in with root)is fully supported and does not trigger this message.

E (vmware-rbd-watchdog)relates to backup/restore health, not core authentication/management planes.

In VMware Cloud Foundation, deletion of a workload domain requires that VCF Operations can correctly discover and process the NSX components attached to that domain. The workload domain delete workflow explicitly includes removal of the NSX Manager and NSX Edge components associated with the domain, unless those NSX components are shared.

In earlier and current VCF guidance, VMware state that NSX Edge clusters for a workload domain must be removed using the documented/VCF-aware method (for example, using the NSX Edge removal process referenced in KB 78635, not by deleting objects directly in NSX Manager). If an administrator deletes theNSX Edge cluster directly in NSX Manager, the VCF inventory and orchestration logic still “believes” the Edge cluster exists. When the workload domain delete workflow reaches the stage“Gather input for deletion of NSX component”, it queries NSX / internal state for Edge cluster data. Because the underlying object has been manually removed, the returned structure is null, which results in an internal“Cannot read the array length because "<locall9>" is null”style error.

Using theNSX Edge Cluster Deployment Removal Toolas per documentation keeps VCF and NSX in sync and is thesupportedpath, so option A is not the likely cause. Network pools and shared NSX Manager configurations do not match the specific NSX-component array/null condition described.