Will have the same parallelism with last processor And each task in current processor will co-locate with task of last processor
Only make sense when the message has implemented the hashCode() Otherwise, it will use Object.hashCode(), which will not return same hash code after serialization and deserialization.
The idea of ShuffleGroupingPartitioner is derived from Storm.
The idea of ShuffleGroupingPartitioner is derived from Storm. Messages are randomly distributed across the downstream's tasks in a way such that each task is guaranteed to get an equal number of messages.
Round Robin partition the data.