When partitioning is present, each partition receives only a part of the input stream; if it is computing any aggregates, such as means, sums, or counts for example, they need to be combined downstream of all the partitions to yield correct results. That function is performed by a unifier. For example, if the tuples are numbers and each partition is computing the per-window sum and emitting it at the end of each streaming window, the downstream unifier would need to add all the per-partition sums and emit that value at the end of each window. Thus, the results coming out of the unifier would be indistinguishable from the situation where partitioning was not present.
By default, if no unifier is defined, the platform supplies a pass-through ...
