Detail
When a node with a strong consistency namespace cold starts, the records on that node have no replication status. This is because the replication status is held in memory for performance reasons and so is lost when the node cold starts.A process called 'appeals' allows the node to contact other nodes in the cluster and, on a per-partition basis, compare the records it has to those records with a known replication status. This is important as when an unreplicated record is read it triggers a write to re-replicate the record, so a large number of unreplicated records can have a negative impact on read latency.
What threads are used to do this? How many of those threads are there? Is the number configurable?