When customers are adding and removing replicas from a topology, the Replication Update Vector is increasing in size, and so is the ChangeLogDB.
Customers would like to have a way to indicate that a replica (and the associated replicaID) has been fully removed and should never reappear.
When doing this, we expect the replicaID to be removed from the Replication Update Vector (RUV), and from the replication Changelog, but we expect replication to continue to work without restoring these IDs.
Removing the replicaID from the RUV and Changelog will prevent both of these to grow in size in an uncontrolled way over time.
In order to keep replication working (because data state and change to replicate are computed based on RUV), Replication should maintain a list of removed Replicas, that should be ignored when a server presents a RUV that contains it. And there should also be a way to clean that list of removed replicas over time (either once all servers have indicated that they've removed a replicaID, or after a delay).