Cisco NX-OS supports nonstop forwarding and graceful restart for BGP.
You can use nonstop forwarding (NSF) for BGP to forward data packets along known routes in the Forward Information Base (FIB)
while the BGP routing protocol information is being restored following a failover. With NSF, BGP peers do not experience routing
flaps. During a failover, the data traffic is forwarded through intelligent modules while the standby supervisor becomes active.
If a Cisco NX-OS router experiences a cold reboot, the network does not forward traffic to the router and removes the router
from the network topology. In this scenario, BGP experiences a nongraceful restart and removes all routes. When Cisco NX-OS
applies the startup configuration, BGP reestablishes peering sessions and relearns the routes.
A Cisco NX-OS router that has dual supervisors can experience a stateful supervisor switchover. During the switchover, BGP
uses nonstop forwarding to forward traffic based on the information in the FIB, and the system is not removed from the network
topology. A router whose neighbor is restarting is referred to as a "helper." After the switchover, a graceful restart operation
begins. When it is in progress, both routers reestablish their neighbor relationship and exchange their BGP routes. The helper
continues to forward prefixes pointing to the restarting peer, and the restarting router continues to forward traffic to peers
even though those neighbor relationships are restarting. When the restarting router has all route updates from all BGP peers
that are graceful restart capable, the graceful restart is complete, and BGP informs the neighbors that it is operational
again.
When a router detects that a graceful restart operation is in progress, both routers exchange their topology tables. When
the router has route updates from all BGP peers, it removes all the stale routes and runs the best-path algorithm on the updated
routes.
After the switchover, Cisco NX-OS applies the running configuration, and BGP informs the neighbors that it is operational
again.
For single-hop iBGP peers with update-source configured under neighbor configuration mode, the peer supports fast external
fall-over.
With the additional BGP paths feature, if the number of paths advertised for a given prefix is the same before and after restart,
the choice of path ID guarantees the final state and removal of stale paths. If fewer paths are advertised for a given prefix
after a restart, stale paths can occur on the graceful restart helper peer.