Spark architecture

From your world. Redshift leader/compute nodes vs. Spark driver/executors

Practice quiz with instant feedback. For a scored run that counts toward your local progress, use uv run de quiz SPK-01.

1The driver process of an application crashes while a long job runs on the executors. What happens to the application? AThe executors elect a new driver and continue. BThe whole Spark application fails; the executors are torn down. CThe cluster manager finishes the current stage, then stops. DNothing; the driver is only needed at startup.

4Which statement about Spark Connect is correct? AIt requires every client to bundle the full Spark JVM and jars. BThe client sends an unresolved logical plan to a remote server over gRPC; the server plans and executes it. CIt replaces the Catalyst optimizer with a client-side planner. DIt only works for streaming workloads.

6You have 40 GB of Parquet in S3. A transformation that you can write entirely in SQL already runs well in your Redshift warehouse. What is the best reason to NOT use Spark here? ASpark cannot read Parquet. BAt this size and with pure SQL, a warehouse (or DuckDB/Polars on one box) is simpler and typically faster; Spark's shuffle and JVM overhead is a cost you don't need. CSpark cannot express SQL. DSpark requires the data to be under 1 GB.