Repair is a Cassandra operation which ensures that data consistency is eventually attained across the ring. It is I/O intensive but required to ensure well-being of the cluster.
The graphs display 2 metrics related to repairs: Active repairs are the number of repair related messages (validation/streaming) that are currently being executed in a thread in Cassandra, while pending repairs is the number of repair messages queued up waiting for a thread.
Repairs are a scheduled operation. If you believe repairs may be causing issues with your cluster then please contact email@example.com and we will investigate.