Implement Performance Tuning Strategies & Optimize Cluster Utilization, Including Partitioning, Repartitioning, Coalescing, Identifying Data Skew, and Reducing Shuffling ASSOCIATE-DEVELOPER-APACHE-SPARK Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   What is the main advantage of using coalesce() over repartition() when reducing the number of partitions?

   • Acoalesce() sorts the data during the shuffle
   • Bcoalesce() always results in a more even data distribution
   • Ccoalesce() can increase the number of partitions without shuffle
   • Dcoalesce() avoids a full shuffle by combining partitions on the same executor
   Reveal Solution Discussion Select an answer first
2. 22expert · hard

   A Spark job performs a join between two large DataFrames. The Spark UI shows that one task takes 30 minutes while all other tasks complete in under 5 minutes. The join key has a few highly skewed values. The engineer tries salting the join key, but the job still shows a bottleneck. What is the most likely reason the salting approach is not fully effective?

   • AThe broadcast join was used, which bypasses the shuffle and ignores salting.
   • BThe number of shuffle partitions is too low, causing multiple salted keys to land in the same partition.
   • CThe salting was applied after the join, not before.
   • DThe salting was applied only to the fact table, not the dimension table.
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   Which strategy is commonly used to mitigate data skew in a join operation?

   • AApplying a salting technique to the join key to distribute the skewed data
   • BFiltering out the skewed keys from the dataset
   • CUsing a broadcast join to send the larger table to all executors
   • DIncreasing the number of partitions using repartition() without changing the join keys
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   What is the primary behavior of the repartition() method in Spark?

   • AIt reduces the number of partitions without a full shuffle
   • BIt performs a full shuffle to redistribute data evenly across the specified number of partitions
   • CIt only increases the number of partitions, never decreases
   • DIt sorts the data within each partition to improve query performance
   Reveal Solution Discussion Select an answer first
6. 25application · easy

   A Spark job is running slowly, and the engineer suspects data skew. Which evidence in the Spark UI would most strongly confirm data skew?

   • AThe job has a low number of tasks, but each task processes a large amount of data.
   • BAll tasks have similar durations and process similar amounts of data.
   • CA few tasks have much longer durations and process significantly more data than the majority.
   • DThe job has a high number of tasks, but each task processes a small amount of data.
   Reveal Solution Discussion Select an answer first