Deep Dive of the Day Commissioning hosts and network pools in VCF 9.1: the groundwork for every new cluster
Every host you add to VCF, whether for a new workload domain or to grow a cluster, goes through commissioning first. Per the VCF 9.1 docs, commissioning adds ESX hosts to the global inventory, and they then sit in the Unassigned Hosts list until a domain or cluster claims them. In 9.1 you start it from the vSphere Client (Global Inventory Lists, Hosts, Unassigned Hosts, Commission Hosts) on the management domain vCenter, or from a vCenter linked through VCF Single Sign-On; the SDDC Manager UI is still there as an alternative. Christopher Lewis notes that it isn't in VCF Operations, which catches people who expect everything there now.
The checklist is where most failures start. Hosts need forward and reverse DNS with the host name matching the FQDN, a self-signed certificate regenerated for that FQDN, a management VMkernel interface on a standard switch with a gateway in the management subnet, hardware on the compatibility guide, and every old disk partition removed. If your environment uses CA-signed host certificates, new hosts must use them too. Commissioning briefly turns on SSH to read the host key and turns it off again, and it creates test VMkernel adapters to check vMotion, so MAC address filtering on the physical network can make that check fail.
You choose a storage type per host, and that choice sticks. The bulk JSON import accepts VSAN, VSAN_ESA, VSAN_REMOTE (vSAN compute), VSAN_MAX (vSAN storage), NFS, VMFS_FC and VVOL, and the docs say a commissioned host can only be used with clusters of the same storage type: vSAN ESA hosts for ESA clusters, OSA hosts for OSA clusters, and so on. Getting it right at commissioning saves you a decommission later. Validation runs against every host, the wizard asks you to confirm each fingerprint, and by default failed hosts are skipped so the rest carry on.
Every host also needs a network pool, which must exist first and match the storage type. A pool holds the VLAN, MTU, subnet and IP ranges for the vMotion and storage VMkernel ports, so you only give an address for vmk0 and VCF hands out the rest. vSAN needs vMotion and vSAN networks, vSAN storage clusters add a vSAN Storage Client network, and NFS needs vMotion and NFS. The management domain gets a default pool at deployment, which can't be deleted, and you need a new pool for hosts on a different Layer 2 network.
The decision you can't undo is the subnet. The docs say a pool's subnet cannot be changed after it's deployed; you can add IP ranges to grow into it, and the sizing guidance recommends starting with a larger subnet than you need today. A pool can only be deleted once no host with an address from it belongs to a workload domain, and you can't mix IPv4 and IPv6 for the same traffic type within a cluster. Size vMotion and vSAN subnets for the hosts you expect in three years, not the four you're commissioning this week.
In a nested lab the same rules apply, with a couple of extra hurdles. Nested hosts usually fail the 10GbE NIC speed and vSAN ESA disk compatibility checks, and Christopher Lewis describes the usual lab-only workaround: adding enable.speed.of.physical.nics.validation=false and vsan.esa.sddc.managed.disk.claim=true to the operationsmanager application.properties on SDDC Manager and restarting the service. That's for labs only. Broadcom's own route for proof-of-concept hardware is KB 408300, which the docs cite for using non-certified ESA disks from VCF 9.0.1.
Based on Commission ESX Hosts (VCF 9.1) Official Network Pool Management (VCF 9.1) Official VCF 9 Day 2: Commissioning new ESX Hosts (thecloudxpert.net)