BGP Graceful Restart settings when ECMP routing is enabled

Products

• VMware NSX

Issue/Introduction

When Graceful Restart (GR) is configured, shut down or reboot of a BGP peer ESG/DLR failover to another existing path to reach a specific prefix takes more time than expected.

• For a shutdown of the peer, it takes 120 seconds as it is the default GR timer (currently can not be modified)
• For a reboot, it takes 120 seconds or less if the time to reboot and re-establish the BGP session is shorter.

It is a global setting that applies to all the BGP sessions configured as it is intended to keep learned prefixes from another peer in the routing table while the peer is rebooting.

Environment

VMware NSX Datacenter for vSphere

Resolution

It is not recommended to enable GR when ECMP is enabled. This applies for both DLR and ESGs.

Steps to disable the Graceful Restart:

• Log on to the vSphere web-Client. Navigate to Networking & security > NSX Edges > Select the DLR(s)/ESG(s) > Manage > routing > BGP.
• In the BGP Configuration section, Click on edit on the top right and de-select the Graceful Restart and click ok.
  
  Note: This will restart all the BGP sessions configured as GR is negotiated when the session is established. It is recommended to perform this change within the Maintenance window.