ESXi host disconnects from vCenter due to VMkernel Multihoming heartbeat failures

Products

• VMware vSphere ESXi

Issue/Introduction

This article addresses a scenario where ESXi hosts repeatedly disconnect from vCenter Server approximately one minute after reconnection. This is typically caused by an invalid multihoming configuration where multiple VMkernel adapters share the same subnet on the default TCP/IP stack.

vpxa log : "Agent can't send heartbeats: Host is down"

Interface Configuration: Multiple vmknics on the same subnet and using the same stack, command "esxcfg-vmknic -l"

Routing Table Conflict: The management interface (vmk0) is missing from the active routing table, or the default route is bound to the wrong interface, command "esxcfg-route -l"

 

 

Environment

VMware vSphere ESXi

Cause

The issue is caused by an invalid multihoming configuration. When vmk0 (Management) and vmk1 (vMotion) share the same subnet on the defaultTcpipStack, a routing conflict occurs. The ESXi kernel may suppress the management interface or bind the default gateway to the incorrect interface, preventing the vpxa agent from sending heartbeats.

Resolution

To resolve the routing conflict and restore stable management connectivity, perform one of the following:

1. Isolate vMotion Traffic: Assign the vMotion VMkernel adapter (vmk1) to a different, non-overlapping IP subnet and VLAN.
2. Use Dedicated TCP/IP Stack: Move the vMotion interface (vmk1) to a dedicated vMotion TCP/IP stack. This isolates the vMotion routing table from the defaultTcpipStack used for management traffic.
3. Verify Routing: Ensure that the default gateway is correctly associated with the management interface (vmk0) by running esxcfg-route -l.

Additional Information

https://knowledge.broadcom.com/external/article/318546/multihoming-on-esxi.html