[HIVE-17716] Not pushing postaggregations into Druid due to CAST on constant

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: Druid integration
• Target Version/s: 3.0.0
• Assignee: Unassigned
• Reporter: Jesús Camacho Rodríguez
• Created: 05/Oct/17 22:32
• Updated: 27/Feb/24 22:13

Description

After Calcite is upgraded to 1.14 and the rule to push post-aggregations to Druid is enabled, the following query fails to create a postaggregation:

EXPLAIN
SELECT language, sum(added) + 100 AS a
FROM druid_table_1
GROUP BY language
ORDER BY a DESC;

Problem seems to be that CAST is getting on the way for the rule to be applied. In particular, this is the final Calcite plan:

 HiveSortLimit(sort0=[$1], dir0=[DESC-nulls-last])
  HiveProject(language=[$0], a=[+($1, CAST(100):DOUBLE)])
    DruidQuery(table=[[default.druid_table_1]], intervals=[[1900-01-01T00:00:00.000/3000-01-01T00:00:00.000]], groups=[{6}], aggs=[[sum($10)]])

There are two different parts to explore to seek a solution: 1) why CAST(100):DOUBLE) is not folded to 100.0d, and 2) whether the rule to push post-aggregations to Druid could handle the CAST in some particular cases.

Issue Links

• is broken by: HIVE-17717 Enable rule to push post-aggregations into Druid
• Is contained by: CALCITE-2170 Use Druid Expressions capabilities to improve the amount of work that can be pushed to Druid

Sub-Tasks

• handle UDF to double/int over aggregate: Resolved