I read somewhere that for better performance
For maximum performance, the number of reducers should be slightly less than
the number of reduce slots in the cluster. This allows the reducers to
one wave and fully utilizes the cluster during the reduce phase.
I don't quite understand this, Can you please help me understand?