[ARROW-12725] [C++][Compute] GroupBy: improve performance by encoding keys in row format only when they are inserted into hash table

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 4.0.0
• Fix Version/s: 6.0.0
• Component/s: C++
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/28467
• Assignee: Michal Nowakiewicz
• Reporter: Michal Nowakiewicz
• Created: 10/May/21 18:03
• Updated: 11/Jan/23 08:28
• Resolved: 30/Aug/21 17:42

Description

Previous implementation of hash group by converts input ExecBatches to row-oriented format,
then hashes and compares rows as if they were a single column. 
It is more efficient (especially for small number of key columns) to avoid relatively costly 
encoding and instead compute hashes of individual columns in column-oriented format mixing them together, and similarly comparing column-oriented data to row-oriented data in the hash table without converting. 
Encoding only happens for a subset of input rows that are inserted into the hash table - they introduce new groups. 
Keys in hash table remain stored as row-oriented.

Issue Links

• is a child of: ARROW-12633 [C++] Query engine umbrella issue
• links to: GitHub Pull Request #10290