[
https://issues.apache.org/jira/browse/FLINK-24716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17437098#comment-17437098
]
xuyang commented on FLINK-24716:
--------------------------------
The plan seems correct, because the condition "d > '2012'" has been applied
into the source and it seems not necessary to still keep this information in
push down filter of the source or in the calc. The semantic of this plan is
right.
> Non-equality predicates on partition columns lead to incorrect plans
> --------------------------------------------------------------------
>
> Key: FLINK-24716
> URL: https://issues.apache.org/jira/browse/FLINK-24716
> Project: Flink
> Issue Type: Sub-task
> Components: Table SQL / Planner
> Reporter: Timo Walther
> Priority: Major
>
> Queries such as
> {code}
> SELECT d FROM T1 WHERE c = 100 AND d > '2012'
> {code}
> where {{d}} is a partition column leads to incorrect plans:
> {code}
> == Abstract Syntax Tree ==
> LogicalProject(d=[$2])
> +- LogicalFilter(condition=[AND(=($0, 100), >($2, _UTF-16LE'2012'))])
> +- LogicalTableScan(table=[[default_catalog, default_database, T1]])
> == Optimized Physical Plan ==
> Calc(select=[d], where=[=(c, 100)])
> +- TableSourceScan(table=[[default_catalog, default_database, T1, filter=[],
> partitions=[], project=[c, d]]], fields=[c, d])
> == Optimized Execution Plan ==
> Calc(select=[d], where=[(c = 100)])
> +- TableSourceScan(table=[[default_catalog, default_database, T1, filter=[],
> partitions=[], project=[c, d]]], fields=[c, d])
> {code}
> It seems in many cases (with SupportsFilterPushDown and without) the {{<}}
> predicate is swallowed and not part of the final execution plan anymore.
> Reproducible code can be found
> [here|https://github.com/twalthr/flink/blob/e5a2cc9bcc9b38cf2b94c9ea7c7296ce94434343/flink-table/flink-table-planner/src/test/java/org/apache/flink/table/planner/plan/rules/logical/TestClass.java]
> with new testing infrastructure.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
