Calcite has a Spool operator, maybe you can check that. Anjali Shrishrimal <[email protected]>于2020年2月20日 周四下午3:20写道：
> Hi everybody, > > I would like to have your suggestions on CALCITE-3806. > > Asking it here as suggested by Julian. > > > > > > If RelNode tree contains a subtree whose result can be obtained by some > other part of the same tree, > > can we optimize it ? and how to express it in plan ? > > > > For example, > > Let's say input structure looks like this : > > > > LogicalUnion(all=[true]) > > LogicalProject(EMPNO=[$0]) > > LogicalFilter(condition=[>=($0, 7369)]) > > LogicalTableScan(table=[[scott, EMP]]) > > LogicalProject(EMPNO=[$0]) > > LogicalFilter(condition=[>=($0, 7369)]) > > LogicalTableScan(table=[[scott, EMP]]) > > > > > > In this case, > > > > LogicalProject(EMPNO=[$0]) > > LogicalFilter(condition=[>=($0, 7369)]) > > LogicalTableScan(table=[[scott, EMP]]) > > > > is repeated. It is going to fetch same data twice. > > Can we save one fetch? Can we somehow tell 2nd input of union to make use > of union's 1st input. Is there any way to express that in plan? > > > > Also, > If the structure was like this : > > > > LogicalUnion(all=[true]) > > LogicalProject(EMPNO=[$0]) > > LogicalFilter(condition=[>=($0, 7369)]) > > LogicalTableScan(table=[[scott, EMP]]) > > LogicalProject(EMPNO=[$0]) > > LogicalFilter(condition=[>=($0, 8000)]) > > LogicalTableScan(table=[[scott, EMP]]) > > > > Second part of union can perform filtering on fetched data of 1st part. > (As second's output is subset of first's output) > > > > Does calcite provide such kind of optimizations ? > > If not, what are the challenges to do so? > > > > > > > > Would love to hear your thoughts. > > > > > Thank you, > Anjali Shrishrimal >
