Necessity of using failover sinks

I have a tough time seeing why one would need a failover sink where both
write to the same HDFS cluster. Can one describe to me why this is a good
idea?