[ 
https://issues.apache.org/jira/browse/CALCITE-4437?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Ruben Q L updated CALCITE-4437:
-------------------------------
    Description: 
The fix applied for CALCITE-4206 was "too drastic" and it resulted in Sort with 
fetch/offset being impossible to decorrelate in all cases.
CALCITE-4333 addressed this issue but only partially (when the Sort with 
fetch/offset is on top on the plan). However, this solution is insufficient, 
because any Sort with fetch/offset that is not inside a Correlate can be 
decorrelated.

Check this test in SqlToRelConverterTest (same test as CALCITE-4333, just with 
an extra LogicalProject on top of the LogicalSort):
{code}
  @Test void testProjectSortLimitWithCorrelateInput() {
    final String sql = ""
        + "SELECT ename||deptno FROM\n"
        + "    (SELECT deptno, ename\n"
        + "    FROM\n"
        + "        (SELECT DISTINCT deptno FROM emp) t1,\n"
        + "          LATERAL (\n"
        + "            SELECT ename, sal\n"
        + "            FROM emp\n"
        + "            WHERE deptno = t1.deptno)\n"
        + "    ORDER BY ename DESC\n"
        + "    LIMIT 3)";
    sql(sql).ok();
  }
{code}

The current plan is:
{noformat}
LogicalProject(EXPR$0=[||($1, CAST($0):VARCHAR NOT NULL)])
  LogicalSort(sort0=[$1], dir0=[DESC], fetch=[3])
    LogicalProject(DEPTNO=[$0], ENAME=[$1])
      LogicalCorrelate(correlation=[$cor0], joinType=[inner], 
requiredColumns=[{0}])
        LogicalAggregate(group=[{0}])
          LogicalProject(DEPTNO=[$7])
            LogicalTableScan(table=[[CATALOG, SALES, EMP]])
        LogicalProject(ENAME=[$1], SAL=[$5])
          LogicalFilter(condition=[=($7, $cor0.DEPTNO)])
            LogicalTableScan(table=[[CATALOG, SALES, EMP]])
{noformat}

It can actually decorrelated as:
{noformat}
LogicalProject(EXPR$0=[||($1, CAST($0):VARCHAR NOT NULL)])
  LogicalSort(sort0=[$1], dir0=[DESC], fetch=[3])
    LogicalProject(DEPTNO=[$0], ENAME=[$1])
      LogicalJoin(condition=[=($0, $3)], joinType=[inner])
        LogicalAggregate(group=[{0}])
          LogicalProject(DEPTNO=[$7])
            LogicalTableScan(table=[[CATALOG, SALES, EMP]])
        LogicalProject(ENAME=[$1], SAL=[$5], DEPTNO=[$7])
          LogicalTableScan(table=[[CATALOG, SALES, EMP]])
{noformat}



  was:
The fix applied for CALCITE-4206 was "too drastic" and it resulted in Sort with 
fetch/offset being impossible to decorrelate in all cases.
CALCITE-4333 addressed this issue but only partially (when the Sort with 
fetch/offset is on top on the plan). However, this solution is insufficient, 
because any Sort with fetch/offset that is not inside a Correlate can be 
decorrelated.

Check this test in SqlToRelConverterTest (same test as CALCITE-4333, just with 
an extra LogicalProject on top of the LogicalSort):
{code}
  @Test void testSortLimitWithCorrelateInput2() {
    final String sql = ""
        + "SELECT ename||deptno FROM\n"
        + "(SELECT deptno, ename\n"
        + "    FROM\n"
        + "        (SELECT DISTINCT deptno FROM emp) t1,\n"
        + "          LATERAL (\n"
        + "            SELECT ename, sal\n"
        + "            FROM emp\n"
        + "            WHERE deptno = t1.deptno)\n"
        + "    ORDER BY ename DESC\n"
        + "    LIMIT 3)";
    sql(sql).ok();
  }
{code}

The current plan is:
{noformat}
LogicalProject(EXPR$0=[||($1, CAST($0):VARCHAR NOT NULL)])
  LogicalSort(sort0=[$1], dir0=[DESC], fetch=[3])
    LogicalProject(DEPTNO=[$0], ENAME=[$1])
      LogicalCorrelate(correlation=[$cor0], joinType=[inner], 
requiredColumns=[{0}])
        LogicalAggregate(group=[{0}])
          LogicalProject(DEPTNO=[$7])
            LogicalTableScan(table=[[CATALOG, SALES, EMP]])
        LogicalProject(ENAME=[$1], SAL=[$5])
          LogicalFilter(condition=[=($7, $cor0.DEPTNO)])
            LogicalTableScan(table=[[CATALOG, SALES, EMP]])
{noformat}

It can actually decorrelated as:
{noformat}
LogicalProject(EXPR$0=[||($1, CAST($0):VARCHAR NOT NULL)])
  LogicalSort(sort0=[$1], dir0=[DESC], fetch=[3])
    LogicalProject(DEPTNO=[$0], ENAME=[$1])
      LogicalJoin(condition=[=($0, $3)], joinType=[inner])
        LogicalAggregate(group=[{0}])
          LogicalProject(DEPTNO=[$7])
            LogicalTableScan(table=[[CATALOG, SALES, EMP]])
        LogicalProject(ENAME=[$1], SAL=[$5], DEPTNO=[$7])
          LogicalTableScan(table=[[CATALOG, SALES, EMP]])
{noformat}




> The Sort rel should be decorrelated even though it has fetch or limit when it 
> is not inside a Correlate
> -------------------------------------------------------------------------------------------------------
>
>                 Key: CALCITE-4437
>                 URL: https://issues.apache.org/jira/browse/CALCITE-4437
>             Project: Calcite
>          Issue Type: Bug
>          Components: core
>    Affects Versions: 1.26.0
>            Reporter: Ruben Q L
>            Priority: Major
>             Fix For: 1.27.0
>
>
> The fix applied for CALCITE-4206 was "too drastic" and it resulted in Sort 
> with fetch/offset being impossible to decorrelate in all cases.
> CALCITE-4333 addressed this issue but only partially (when the Sort with 
> fetch/offset is on top on the plan). However, this solution is insufficient, 
> because any Sort with fetch/offset that is not inside a Correlate can be 
> decorrelated.
> Check this test in SqlToRelConverterTest (same test as CALCITE-4333, just 
> with an extra LogicalProject on top of the LogicalSort):
> {code}
>   @Test void testProjectSortLimitWithCorrelateInput() {
>     final String sql = ""
>         + "SELECT ename||deptno FROM\n"
>         + "    (SELECT deptno, ename\n"
>         + "    FROM\n"
>         + "        (SELECT DISTINCT deptno FROM emp) t1,\n"
>         + "          LATERAL (\n"
>         + "            SELECT ename, sal\n"
>         + "            FROM emp\n"
>         + "            WHERE deptno = t1.deptno)\n"
>         + "    ORDER BY ename DESC\n"
>         + "    LIMIT 3)";
>     sql(sql).ok();
>   }
> {code}
> The current plan is:
> {noformat}
> LogicalProject(EXPR$0=[||($1, CAST($0):VARCHAR NOT NULL)])
>   LogicalSort(sort0=[$1], dir0=[DESC], fetch=[3])
>     LogicalProject(DEPTNO=[$0], ENAME=[$1])
>       LogicalCorrelate(correlation=[$cor0], joinType=[inner], 
> requiredColumns=[{0}])
>         LogicalAggregate(group=[{0}])
>           LogicalProject(DEPTNO=[$7])
>             LogicalTableScan(table=[[CATALOG, SALES, EMP]])
>         LogicalProject(ENAME=[$1], SAL=[$5])
>           LogicalFilter(condition=[=($7, $cor0.DEPTNO)])
>             LogicalTableScan(table=[[CATALOG, SALES, EMP]])
> {noformat}
> It can actually decorrelated as:
> {noformat}
> LogicalProject(EXPR$0=[||($1, CAST($0):VARCHAR NOT NULL)])
>   LogicalSort(sort0=[$1], dir0=[DESC], fetch=[3])
>     LogicalProject(DEPTNO=[$0], ENAME=[$1])
>       LogicalJoin(condition=[=($0, $3)], joinType=[inner])
>         LogicalAggregate(group=[{0}])
>           LogicalProject(DEPTNO=[$7])
>             LogicalTableScan(table=[[CATALOG, SALES, EMP]])
>         LogicalProject(ENAME=[$1], SAL=[$5], DEPTNO=[$7])
>           LogicalTableScan(table=[[CATALOG, SALES, EMP]])
> {noformat}



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to