[SPARK-49408] Poor performance in ProjectingInternalRow

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.5.2
• Fix Version/s: 4.0.0, 3.4.4, 3.5.3
• Component/s: SQL
• Labels: pull-request-available
• Assignee: Frank Wong
• Reporter: Frank Wong
• Created: 27/Aug/24 11:27
• Updated: 05/Sep/24 07:00
• Resolved: 05/Sep/24 06:59

Description

In ProjectingInternalRow, the colOrdinals is passed as a List. According to the Scala documentation, the apply method for List has a linear time complexity, and it is used in all methods of ProjectingInternalRow for every row. This can have a significant impact on performance.

The following flame graph was captured in a merge into sql. A considerable amount of time was spent on List.apply. Changing this to IndexedSeq would improve the performance.

 

20240827-172739.html

https://docs.scala-lang.org/overviews/collections-2.13/performance-characteristics.html

Issue Links

• links to: GitHub Pull Request #47890