Restarting management services on ESXi prepared for NSX disconnects VM data path

Products

• VMware NSX
• VMware vSphere ESXi

Issue/Introduction

Put the ESXi host in maintenance mode before restarting the management services.
Restarting management services on a ESXi (either through the services.sh script or through the DCUI) prepared for NSX and connected to NSX Logical switches in unicast mode, may disconnect the VM's data path.

Environment

VMware NSX for vSphere 6.4.x

Cause

When you restart the management services, netcpa is the first process to be stopped and last process to start. NSX Controller has a 5 minute grace time before it flushes the VTEP table for a host that has lost netcpa connectivity. After the 5 minute grace time, BUM traffic over unicast logical switches gets impacted due to loss of netcpa connectivity.

Generally, restarting the management services has no impact on running Virtual Machines if all the services are restarted with in the 5 minutes time. But, if services like USB Arbitrator is involved, which may take several minutes to start, causes the netcpa service to be off for several minutes.

Resolution

This is a known issue affecting NSX for vSphere 6.4.x.

 

To avoid this issue, put the ESXi host in maintenance mode before proceeding with the restart.