Graceful Restart (GR) is also referred to as Nonstop Forwarding (NSF) and describes a router’s ability to maintaining forwarding state through a protocol-level restart or GRES event, leveraging the fact that modern routers use a separated control and dataplane, which in turn allows decoupling such that a restart of one no longer forces the restart of the other.
A protocol restart can occur due to intentional or unintentional reasons. For example, an operator choosing to restart the routing process, rebooting the router, or upgrading its software are examples of the former, whereas a routing process crash or hardware-induced RE switchover fall into the latter category.
GR is not a panacea of guaranteed success, and as mentioned previously the trend is to move to Nonstop Routing as support for the protocol you need in your network becomes available. One upside to GR is that it can be used on routers with a single RE; both GRES and NSR require redundant REs to work.
The primary drawback to GR is the need for protocol extensions and helper support in neighboring routers, and a stable network topology. If any neighbors do not support GR, or if other changes occur in the network, GR ends up aborting and loss results in the dataplane. Even when all goes to plan and the GR event succeeds, which means there is zero loss in the data plane, there is still control plane flap and a subsequent need for protocol reconvergence between the restarting router and its helpers, ...