When a Server Goes Down:
1. Route Withdrawal: The failing server stops advertising its reachability for the Anycast IP address to its neighboring routers, usually triggered by a health check failure.
2. Traffic Reroute: Neighboring routers update their tables, removing the bad route. New traffic destined for that Anycast IP is automatically routed to the next closest server instance that is still advertising the address.
When a Server Comes Up:
1. Route Re-announcement: Once the server is fully up, healthy, and ready to serve traffic, it begins advertising its reachability for the Anycast IP address via BGP again.
2. Traffic Re-convergence: Internet routers gradually update their routing tables to favor the now-closest (re-announced) server. Clients closest to this server will have their requests routed back to it, restoring the optimized low-latency path.