Read repair is a feature that allows for inconsistent data to be fixed at query time. read_repair_chance (and dclocal_read_repair_chance) is configured for each table, defaulting to 0.1 (10% chance). When it occurs (only with read consistency levels higher than ONE), Apache Cassandra will verify the consistency of the replica with all other nodes not involved in the read request. If the requested data is found to be inconsistent, it is fixed immediately.
Do note that read repairs do not occur when read consistency is set to ONE or LOCAL_ONE. Also, reads at a consistency level of all will force a read repair to happen 100% of the time. However, these read repairs do incur latency, which is why they are typically set to low percentages. ...
