Description
AggregateJoinTransposeRule generates inequivalent nodes if Aggregate relNode contains distinct aggregate function.
T1
| a (String) | b (int) |
|---|---|
| A | 1 |
| A | 2 |
| B | 2 |
| C | 3 |
T2
| c(int) |
|---|
| 1 |
| 1 |
| 2 |
For the following sql,
select count (distinct a) from t1, t2 where t1.b = t2.c
Aggregate would generate following node tree, which is inequivalent with original node tree.
LogicalAggregate(group=[{}], EXPR$0=[$SUM0($4)])
+- LogicalProject(b=[$0], EXPR$0=[$1], c=[$2], $f1=[$3], $f4=[*($1, $3)])
+- LogicalJoin(condition=[=($0, $2)], joinType=[inner])
:- LogicalAggregate(group=[\{1}], EXPR$0=[COUNT(DISTINCT $0)])
: +- LogicalTableScan(table=[[t1]])
+- LogicalAggregate(group=[\{0}], agg#0=[COUNT()])
+- LogicalTableScan(table=[[t2]])
Based on the converted plan, result is 4; while the correct answer is 2.
Attachments
Issue Links
- is duplicated by
-
CALCITE-2790 AggregateJoinTransposeRule incorrectly pushes down distinct count into join
-
- Closed
-