A BGP session reset can be very disruptive – each router participating in the failed session must delete the routes it received from its peer, recalculate new best paths, update forwarding tables (depending on the types of routes), and send route withdrawals and advertisements to other peers. It makes sense then that session resets should be avoided as much as possible and when a session reset cannot be avoided the disruption to the network should be minimized.
To support these objectives, the BGP implementation in SR OS supports two key features:
BGP high availability (HA)
BGP graceful restart (GR)
BGP HA refers to the capability of a router with redundant CPMs to keep established BGP sessions up whenever a planned or unplanned CPM switchover occurs. A planned CPM switchover can occur during In-Service Software Upgrade (ISSU). An unplanned CPM switchover can occur if there is an unexpected failure of the primary CPM.
BGP HA is always enabled on routers with redundant CPMs; it cannot be disabled. BGP HA keeps the standby CPM in-sync with the primary CPM, with respect to BGP and associated TCP state, so that the standby CPM is ready to take over for the primary CPM at any time. The primary CPM is responsible for building and sending the BGP messages to peers but the standby CPM reliably receives a copy of all outgoing UPDATE messages so that it has a synchronized view of the RIB-OUT.