ESXi host loses management network connectivity after upgrading from 9.0.2 to 9.1 directly on the host, vmnic uplinks fail to attach to the management VDS with "Not supported"

Products

• VMware vSphere ESXi

Issue/Introduction

After upgrading an ESXi 9.0.2 host to ESXi 9.1 directly on the host (outside of vCenter), the host loses all management network connectivity following the post-upgrade reboot. The host is unreachable by vCenter, SSH, the Host Client web console, and the API, and can only be recovered through out-of-band console access (for example, iLO).

Symptoms:

• Following the post-upgrade reboot to ESXi 9.1, both vmnic0 and vmnic1 fail to register as uplinks to the management vSphere Distributed Switch (VDS), leaving the management port group with zero active physical uplinks.
• The management VMkernel interface (vmk0) remains configured with a valid IP address and the correct VLAN, but has no physical path to the network.
• The vmnics do not recover on subsequent reboots of the host once it is running 9.1.
• Recovery requires per-host manual intervention through the out-of-band console and is not scriptable at scale.
• Any remaining hosts in the same configuration are expected to fail identically on upgrade, blocking further rollout of 9.1.

Log entries:

/var/log/jumpstart-esxcli-stdout.log

records the DVS uplink attach failure for both NICs

Unable to Add Uplink... Not supported: VSI node (187:VSI_NODE_net_pNics_link)


/var/log/vmkernel.log

records the underlying failure for both NICs:

NetTeam: 758: Failed to register uplink vmnic0: Not supported
Net: 4669: Unable to add uplink vmnic0 to portset DvsPortset-0: Not supported

Environment

VMware vSphere ESXi 9.1

Cause

This is caused by a change in NSX IPFIX property-handling behavior in 9.1. As of ESXi 9.1, IPFIX on the Uplink port group is no longer supported.

Because the VDS in this environment has IPFIX monitoring persisted on its Uplink port group, the NSX component in 9.1 hard-fails the uplink attach on every boot. The uplink attach path queries the VSI_NODE_net_pNics_link VSI node, which exposes physical NIC link state, that query returns Not supported, the uplink registration fails, and the physical NICs are never added to DvsPortset-0. As a result, vmk0 retains its IP and VLAN configuration but is left with no physical uplink behind it.

When the upgrade is driven from vCenter, the vCenter upgrade pre-check detects IPFIX enabled on the uplink and blocks the upgrade, prompting for the feature to be disabled first. Upgrading the host directly bypasses that pre-check entirely, so the unsupported configuration is carried into 9.1 and the failure occurs on first boot. With IPFIX enabled on the uplinks, this outcome is expected for any host upgraded directly to 9.1.

Resolution

Upgrading a host directly is not a supported upgrade path when that host is already managed by a vCenter Server.

Supported Upgrade Workflow:

1. Upgrade the vCenter Server first, before any ESXi hosts.
2. Upgrade the ESXi hosts from vCenter, not from the host.
3. Allow the vCenter upgrade pre-check to run. Where IPFIX is enabled on the VDS uplink port group, the pre-check blocks the upgrade and prompts for the configuration to be disabled.
4. Disable IPFIX on the VDS Uplink port group as prompted, then proceed with the host upgrade.

Following this workflow ensures the unsupported IPFIX-on-uplink configuration is removed before the host reaches 9.1, and the uplinks attach normally after the post-upgrade reboot.

Additional Information

Note: Bypassing vCenter and upgrading hosts directly will continue to cause uplink configuration failures after the host reboots, particularly where configurations unsupported in 9.1 such as IPFIX on the uplinks are present.

 

Similar issue: ESXi host not reachable after reboot or upgrade due to VDS(Virtual Distributed Switch) uplinks are lost