[IMPALA-5095] Use parquet::Statistics for simple min/max aggregates

• Type: Sub-task
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: Impala 2.8.0
• Component/s: Backend
• Labels: parquet performance ramp-up
• Epic Color: ghx-label-3
• Assignee: Unassigned
• Reporter: Alexander Behm
• Created: 17/Mar/17 20:19
• Updated: 07/Feb/23 11:10

Description

select min(int_col), max(bigint_col) from parquet_table;
select min(int_col), max(bigint_col) from parquet_table group by partition_col;
select min(int_col), max(int_col) from parquet_table; <--- case a little trickier because int_col refd twice

The slot values for int_col and bigint_col can be directly filled in from the parquet::Statistics, assuming stats are available for both columns. No columns need to be scanned/materialized.

This JIRA focuses on implementing this optimization in the simple case where all scanned columns feed into min/max aggregates and where all columns have parquet::Statistics. Those conditions can be relaxed, but should be addressed separately.

This optimization opportunity must be detected by the planner and is not applicable when there are scan predicates.