In an NSX-T environment GARP packets are not received by all VMs on multi-TEP ESXi hosts

Products

• VMware NSX

Issue/Introduction

Symptoms:

•  NSX-T environment prior to NSX 2.4.0  
•  ESXi host is configured for load balancing and has multiple TEP interfaces e.g. vmk10, vmk11, etc.
•  GARP packets reach some VMs on an ESXi host but not others

   Note: The impact of this issue is seen when an application VM is configured for HA.
             On failover, a VIP becomes active on the standby VM and sends a GARP.
             If a VM does not receive this GARP it can continue to send traffic to the VM which is no longer active.

Environment

VMware NSX-T Data Center

Cause

When a VM on an Overlay logical switch sends a GARP, it is sent to the TEP List of the Routing Domain.
The Routing Domain TEP List contains one TEP from each ESXi host and this is the TEP mapped to the DLR port on the host.
This ensures that the Distributed T1 and T0 Routers can be updated of the new MAC/IP relationship as well as VMs on that logical switch.
Due to a software issue when the GARP packet was received on an ESXi host it was only sent to VMs mapped to that TEP.
VMs mapped to other TEPs on the host did not receive the GARP packet.
NSX-T 2.4.0 and above ensures all VMs on a logical switch regardless of TEP mapping receive the GARP packet.

Resolution

This issue is resolved in NSX-T Datacenter 2.4.0 and above.

Workaround:
If an upgrade is not possible the ESXi host should be reconfigured to a single TEP configuration.