What is the key characteristic of "lazy evaluation" in Spark?

The key characteristic of "lazy evaluation" in Spark is that transformations are executed only when an action is called. This means that when you define transformations (such as map or filter) on a DataFrame or RDD, Spark does not immediately execute those operations. Instead, it builds up a logical plan of the operations to be performed. The actual execution occurs only when an action (like count, collect, or save) requires the result of the transformations.

This behavior allows Spark to optimize the execution plan by analyzing all the transformations collectively before performing any computations. By delaying execution, Spark can eliminate unnecessary calculations, optimize the execution strategy, and reduce the amount of data shuffled between nodes, leading to improved performance and resource efficiency.