IXGBE driver shows a few seconds delays when notifying link down event of DPDK framework on NSX Edge Bare Metal

Products

• VMware NSX

Issue/Introduction

Symptoms:
Shutting down the interface of its remote peer that is connected to NSX Edge Bare Metal, it takes a few seconds to notify link down event of DPDK framework if you use IXGBE driver for data plane NICs on NSX Edge Bare Metal.

Environment

VMware NSX-T Data Center 2.x
VMware NSX-T Data Center

Cause

IXGBE driver uses a delayed timer on  DPDK framework for NIC state stability in which the link is actually assessed after a few seconds. This is current design of IXGBE driver.

Resolution

This issue is fixed in NSX-T version 2.5.1.

Workaround:
There is no workaround. If you faced unexpected HA failover, recommend to do followings for mitigation.

• Set BFD timeout between NSX Edges to over 5000 msec (ex. 2000 msec * 3 = 6000 msec). Prevent unexpected HA failover before notifying link down event.
• Set LACP timeout to Fast on NSX Edge Bare Metal and its remote peer if you use LACP LAG. Notify link down event on LACP framework faster.

Additional Information

Impact/Risks:
This may cause unexpected HA failover and disconnection to its remote peer because delayed timer is too long.