When a node reports another as down, the node that is still up will store structures, known as hints, for the down node. Hints are essentially writes meant for one node that are temporarily stored on another. When the down node returns to a healthy status, the hints are then streamed to that node in an attempt to re-sync its data.
As storage is finite, hints will be stored for up to three hours, by default. This means that a down node must be brought back within that time frame, or data loss will occur.
With Apache Cassandra versions older than 3.0, hints were stored in the hints table of the system keyspace. The problem with this approach is that once the stored hints were replayed, they were then deleted. This frequently ...
