Change Details

It is best practice to periodically issue repairs on each node of a Cassandra cluster. These repair jobs result in a great deal of processor utilization, and disk and network IO, and so are not recommended to be run concurrently; Ideally, node repairs should happen sequentially. One solution would be to install a job that orchestrates sequential repairs from a single node in the cluster. Such a job could be safely invoked from all hosts at a specified time, but only one (selected from the available nodes using a simple heuristic), would actually issue the repairs.