Hi, tl;dr it's not possible to "reverse-engineer" tasks to functions.
In essence, Spark SQL is an abstraction layer over RDD API that's made up of partitions and tasks. Tasks are Scala functions (possibly with some Python for PySpark). A simple-looking high-level operator like DataFrame.join can end up with multiple RDDs, each with a set of partitions (and hence tasks). What the tasks do is an implementation detail that you'd have to know about by reading the source code of Spark SQL that produces the "bytecode". Just looking at the DAG or the tasks screenshots won't give you that level of detail. You'd have to intercept execution events and correlate them. Not an easy task yet doable. HTH. Pozdrawiam, Jacek Laskowski ---- "The Internals Of" Online Books <https://books.japila.pl/> Follow me on https://twitter.com/jaceklaskowski <https://twitter.com/jaceklaskowski> On Tue, Apr 11, 2023 at 6:53 PM Trường Trần Phan An <truong...@vlute.edu.vn> wrote: > Hi all, > > I am conducting a study comparing the execution time of Bloom Filter Join > operation on two environments: Apache Spark Cluster and Apache Spark. I > have compared the overall time of the two environments, but I want to > compare specific "tasks on each stage" to see which computation has the > most significant difference. > > I have taken a screenshot of the DAG of Stage 0 and the list of tasks > executed in Stage 0. > - DAG.png > - Task.png > > *I have questions:* > 1. Can we determine which tasks are responsible for executing each step > scheduled on the DAG during the processing? > 2. Is it possible to know the function of each task (e.g., what is task ID > 0 responsible for? What is task ID 1 responsible for? ... )? > > Best regards, > Truong > > --------------------------------------------------------------------- > To unsubscribe e-mail: user-unsubscr...@spark.apache.org
