The JMX Mbeans used to check the number of active and pending repair sessions has changed in Cassandra 2.2, making Reaper unable to check if a repair is already running before running a new segment.
This release fixes the issue by using the right MBean depending on the version of Cassandra.
It also adds a check to prevent running new segments on nodes that are performing validation compaction, and raise the timeout for each segments when running incremental repair (there's one "fake" segment per node in incremental repair, which may take longer than 30mn due to anticompaction).