In mapreduce framework each map function generates key values. The partition function accepts these key values and in return provides the index for a reduce. Generally the key is hashed and a modulo is done to the number of reducers.