When BGP Graceful Restart Wasnt Graceful @TeamNANOG
When BGP Graceful Restart Wasnt Graceful  @TeamNANOG
Uploaded October 2026 | Updated November 2026, 3 hours ago
BGP Graceful Restart should reduce traffic impact during short control-plane interruptions. During our firewall HA upgrade, we expected it to hide the failover. Instead, after GR was enabled, the failover caused almost two minutes of traffic loss.

This talk walks through the troubleshooting: Established BGP sessions with zero prefixes, delayed End-of-RIB, different IPv4 and IPv6 behavior, packet captures, vendor logs and a small BIRD lab test.

The takeaway is simple: GR is only graceful when the implementation matches real HA behavior. Otherwise, it can make an outage longer instead of shorter.

Marian Rychtecky:
Marian Rychtecky is CTO of NIX.CZ, where he is responsible for the technical development and operation of a multi-site Internet Exchange infrastructure across Central Europe. His work focuses on routing, BGP, EVPN/VXLAN, network automation, resilience, and practical troubleshooting of large-scale networks.
When BGP Graceful Restart Wasnt GracefulKeynote: Global Infrastructure and Layer 3 Interconnectionn96 Newcomers OrientationEpistemology of Incidents & Problem SolvingWhen Photonics Becomes a Network Architecture Constraint in AI Data CentersImproving Routing Security for Global Research and EducationHPN-SSH: High Performance Data Transfers with SSHIP Geofeeds: Trust, Accuracy, and AbuseSecuring the Internal Control Plane with Standards & OSSWhen the World Cup Moved to YouTube: 416 Gbps and the Limits of Local Delivery in BrazilNetworking at the Speed of Light - A Peek at Networking Inside a High Frequency Trading FirmARIN Update: What’s Changing and What’s Next for Network Operators
NANOG |

When BGP Graceful Restart Wasn't Graceful

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER