In a dataflow, how many transformation steps will fold?

• A. All
• B. None
• C. Some
• D. Most

Answer: (C)

Folding (operator fusion) is an optimization where adjacent transformations are merged into one execution unit to reduce overhead and data movement. A step can be folded when the fused result stays the same, meaning the transformations can be combined without changing correctness. Stateless operations that process the same input stream and don’t introduce new shuffle boundaries are good candidates for fusion. But steps that require grouping, windowing, or repartitioning, or those with external side effects or strict timing/ordering requirements, cannot be safely fused. Because of these limitations, only some transformation steps will fold.