Windows Failover Cluster does not work as expected when using NSX segments

Products

• VMware NSX

Issue/Introduction

• VMs configured as members of a Windows Failover Cluster do not work as expected when failing over when connected to an NSX segment.
• When failing over the VIP moves to the standby member however traffic does not resolve to the new owner of the VIP for a period of time (up to 30 mins).
• If the VIP is failed back to the original member, traffic resolves instantly.
• Segments IP discovery is configured as per Configure and apply NSX-T Segment IP discovery profile when using high availability (HA) for Virtual Machines.
• GARP not sent from the VM (at OS or switchport level).

Environment

VMware NSX

VMware NSX-T Data Center

Cause

The failover is intended to announce the new owner of the VIP via a GARP. In some circumstances this is not broadcast by the VM at an OS level. This is prior to NSX or any part of the VMware infra being involved and is an OS level issue. 

As the GARP is not announced the network does not learn of the new location of the IP and does not amend the forwarding tables. This continues until the ARP expires in which case the new owner will work as expected until another failover occurs.

Resolution

This is not a VMware issue and is related to an OS issue. The software vendor should be contacted for assistance. 

Additional Information

Required configuration for HA-VIP usage on NSX segments is documented by KB - Configure and apply NSX-T Segment IP discovery profile when using high availability (HA) for Virtual Machines.