An AEM instance is updating and replicating pages at a very high rate.

In the AEM Dispatcher farm configuration, high-frequency replication events can cause "cache stampedes" where the cache is constantly invalidated, forcing all traffic back to the publish instances simultaneously and spiking CPU. The /graceperiod property mitigates this. When set, it defines a window of time (e.g., 2 seconds) during which the Dispatcher will continue to serve "stale" (recently invalidated) cached content to end-users while a single request is sent to the publish tier to fetch the new content. Once fetched, the cache is updated. This effectively throttles invalidation spikes and protects the publish instances from traffic overloads.