Apache Beam: What is the difference between DoFn and SimpleFunction?

Question

While reading about processing streaming elements in apache beam using Java, I came across DoFn<InputT, OutputT> and then across SimpleFunction<InputT, OutputT>.

Both of these look similar to me and I find it difficult to understand the difference.

Can someone explain the difference in layman terms?

Answer 1

Conceptually you can think of SimpleFunction is a simple case of DoFn:

• SimpleFunction<InputT, OutputT>:

  • simple input to output mapping function;
  • single input produces single output;
  • statically typed, you have to @Override the apply() method;
  • doesn't depend on computation context;
  • can't use Beam state APIs;
  • example use case: MapElements.via(simpleFunction) to convert/modify elements one by one, producing one output for each element;

• DoFn<InputT, OutputT>:

  • executed with ParDo;
  • exposed to the context (timestamp, window pane, etc);
  • can consume side inputs;
  • can produce multiple outputs or no outputs at all;
  • can produce side outputs;
  • can use Beam's persistent state APIs;
  • dynamically typed;
  • example use case: read objects from a stream, filter, accumulate them, perform aggregations, convert them, and dispatch to different outputs;

You can find more specific examples and use cases for ParDos in the dev guide.

This part mentions the MapElements, which is the use case for SimpleFunctions