Load shedding drops some proportion of load by dropping traffic as the server approaches overload conditions. The goal is to keep the server from running out of RAM, failing health checks, serving with extremely high latency, or any of the other symptoms associated with overload, while still doing as much useful work as it can.
