You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@calcite.apache.org by "Aitozi (Jira)" <ji...@apache.org> on 2023/06/15 12:44:00 UTC
[jira] [Commented] (CALCITE-5784) Generate the same correlationId for the same query
[ https://issues.apache.org/jira/browse/CALCITE-5784?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17733048#comment-17733048 ]
Aitozi commented on CALCITE-5784:
---------------------------------
IMO, we could deduplicate the creation of the correation id during the sql2rel phase. Same identifier in the same scope can share the same id.
> Generate the same correlationId for the same query
> --------------------------------------------------
>
> Key: CALCITE-5784
> URL: https://issues.apache.org/jira/browse/CALCITE-5784
> Project: Calcite
> Issue Type: Improvement
> Components: core
> Reporter: Aitozi
> Priority: Major
>
> Currently, the CTE query block will be sql2rel multiple times if be referenced multiple times. If it contains a lateral join, it will generate different correlation id. See below:
> {code:java}
> String sql = "WITH a AS (SELECT ename, job, empno, r FROM emp, LATERAL TABLE (ramp(empno)) as T(r))"
> + " SELECT * from a a1, a a2 WHERE a1.r = a2.empno";
> sql(sql).ok();
> {code}
> {code:java}
> LogicalProject(ENAME=[$0], JOB=[$1], EMPNO=[$2], R=[$3], ENAME0=[$4], JOB0=[$5], EMPNO0=[$6], R0=[$7])
> LogicalJoin(condition=[=($3, $6)], joinType=[inner])
> LogicalProject(ENAME=[$1], JOB=[$2], EMPNO=[$0], R=[$9])
> LogicalCorrelate(correlation=[$cor0], joinType=[inner], requiredColumns=[{0}])
> LogicalTableScan(table=[[CATALOG, SALES, EMP]])
> LogicalTableFunctionScan(invocation=[RAMP($cor0.EMPNO)], rowType=[RecordType(INTEGER I)])
> LogicalProject(ENAME=[$1], JOB=[$2], EMPNO=[$0], R=[$9])
> LogicalCorrelate(correlation=[$cor1], joinType=[inner], requiredColumns=[{0}])
> LogicalTableScan(table=[[CATALOG, SALES, EMP]])
> LogicalTableFunctionScan(invocation=[RAMP($cor1.EMPNO)], rowType=[RecordType(INTEGER I)])
> {code}
> It will generate two correlation id. In flink there is a subplan reuse based on the digest, it will break this functionality
--
This message was sent by Atlassian Jira
(v8.20.10#820010)