Ops netstack default route on dual DPU ESXi host missing after reboot

Products

• VMware vSphere ESXi

Issue/Introduction

VMware ESXi with Dual DPUs.

• The default route (gateway) for the ops netstack vmkernel interface on a dual DPU ESXi host fails to persist after a host reboot.
• After the host restarts, services relying on this route (such as vmkping to VCF Operations for Networks collectors using the ops netstack) will fail until the route is manually re-added via CLI.
  
  
• Observed in hostd.log (directory path:  /var/run/log/)
  
  ####-##-##T##:##:##.060Z In(166) Hostd[525754]: [Originator@6876 sub=Libs opID=1cc21aca sid=5252ba8c user=#########] VmKernelNicInfo::AddVmKernelNic: Added vmkNic, netstack:'ops', interface:'vmk5'
  
  
• Observed in vmkernel.log (directory path:  /var/run/log/) 
  
  ####-##-##T##:##:##.211Z In(182) vmkernel: cpu5:524849)Tcpip: 1653: Cleaning 4 Leaked routes on 'vmk5'
  ####-##-##T##:##:##.211Z In(182) vmkernel: cpu5:524849)Tcpip: 1661: Freeing interface 'vmk5'

Environment

VMware ESXi 8.0 Update 3i and previous 8.0 releases

Cause

When a vmkernel adapter is added to an ops or mirror netstack on a dual DPU host, it is deployed on the DPU side and is not visible under the host's "TCP/IP configuration" tab in the vCenter Server UI.

Due to this design, default routes overridden during vmknic creation do not properly persist through the standard host configuration upon reboot.

Resolution

This is a known issue impacting VMware vSphere ESXi. 

If you encounter this issue, please open a Broadcom support case using the instructions at KB 142884 - Creating and managing Broadcom cases