When a BGP peering session is reset, the system performs an exhaustive cleanup of the dynamic state that was learned as part of that session. This state includes (although is not limited to) Adj-RIB-In routes, that is, routes that were learned from the peer. Until cleanup is completed, no new peering session is permitted to establish. In a highly scaled environment, when many peers and many routes are involved, cleanup can take a long time, with the time to clean up the Adj-RIB-In be the most dominant time contributor. This delays restoration of service.
This feature introduces improvements in (that is, reduction of) the delay before peers can reestablish, with focus on allowing peer reestablishment before the completion of Adj-RIB-In cleanup, if the peer is coming back up with the unchanged capabilities. Cleanup is completed in a background job in parallel with peer establishment. This feature also introduces a more aggressive peer establishment scheme to benefit any case of peer flaps, but especially for large-scale BGP deployments.