David Wayne Birdsall created TRAFODION-3031:
-----------------------------------------------
Summary: Query with nested subqueries chooses bad plan
Key: TRAFODION-3031
URL: https://issues.apache.org/jira/browse/TRAFODION-3031
Project: Apache Trafodion
Issue Type: Improvement
Components: sql-cmp
Affects Versions: 2.3
Reporter: David Wayne Birdsall
Assignee: David Wayne Birdsall
Attachments: jira.log, jira.sql
The attached files demonstrate the problem. The file jira.sql is a script that
reproduces the problem, while jira.log is a sqlci showing the results.
The query in question does an IN-subquery from T1 to T2, then T2 has an
=-subquery back to T1. T2 contains two indexes, one each on the join columns.
The default plan uses a hybrid hash join of T1 to T2 and is very slow. It does
a full scan of both T1 and T2.
If we set CQD SEMIJOIN_TO_INNERJOIN_TRANSFORMATION 'ON', the plan is a little
bit better. We get a nested join of T1 to T2. But it is inefficient; we still
do a full scan of T2.
If we rename the index T2A to T2Y, and we still have the CQD set, we get a good
nested join plan that uses the index T2Y and reads just one row at each level.
This is very fast.
So, there are two issues here.
# We could do a better job of deciding when to do the semi-join to join
transformation. When the inner table is small, it is profitable to do this.
# The index elimination logic is mistakenly eliminating index T2A so the
Optimizer misses a chance to use it and so does not find the efficient nested
join plan.[^jira.log]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
