Stamatis Zampetakis created HIVE-28931:
------------------------------------------
Summary: Convert IN to SEARCH in RexNodeConverter
Key: HIVE-28931
URL: https://issues.apache.org/jira/browse/HIVE-28931
Project: Hive
Issue Type: Improvement
Components: CBO
Reporter: Stamatis Zampetakis
Assignee: Stamatis Zampetakis
RexNodeConverter has some logic
([transformInToOrOperands|https://github.com/apache/hive/blob/4554364b94e60f10d3e4e723c8016ed3148b9198/ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/translator/RexNodeConverter.java#L520],
[rewriteInClauseChildren|https://github.com/apache/hive/blob/4554364b94e60f10d3e4e723c8016ed3148b9198/ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/translator/RexNodeConverter.java#L570])
for transforming IN to OR when certain conditions hold. The initial logic was
introduced in HIVE-19097 to facilitate the simplification/unification of
expressions (by RexSimplify) since the IN operator is mostly a blackbox for CBO
so it cannot be easily combined with other equivalent expressions.
In CALCITE-4173, the SEARCH operator was introduced to represent IN, BETWEEN,
and various other expressions. The SEARCH operator is used extensively for
simplifications and there is also some logic in RexSimplify that tries to
transform OR to SEARCH when that's possible.
After upgrading to Calcite version 1.33.0 (HIVE-27102), Hive also takes
advantage of the SEARCH operator. Doing an early conversion from IN to SEARCH
(instead of OR) in RexNodeConverter seems beneficial for enabling further
simplifications and it could possibly reduce compilation latency since we would
avoid the intermediate conversion to OR when that is not necessary. In a
nutshell instead of doing IN -> OR -> SEARCH we could do IN -> SEARCH and we
could possibly exploit the RexBuilder.makeIn method.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
