[
https://issues.apache.org/jira/browse/FLINK-26362?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
luoyuxia updated FLINK-26362:
-----------------------------
Description:
With hive dialect, can be reproduced using following code:
{code:java}
tableEnv.executeSql("CREATE TABLE t1 (c1 INT, c2 CHAR(100))");
tableEnv.executeSql("CREATE TABLE t2 (c1 INT)");
List<Row> results = CollectionUtil.iteratorToList(tableEnv.executeSql("SELECT
c1 FROM t1 WHERE c1 IN (SELECT c1 FROM t2)").
collect());
{code}
Then it will throw IndexOutOfBoundsException: 0, the reason is from the
following comments:
?? If it's a subquery and the project is identity, we skip creating this
project. This is to handle an issue with calcite SubQueryRemoveRule. The rule
checks col uniqueness by calling RelMetadataQuery::areColumnsUnique with an
empty col set, which always returns null for a project and thus introduces
unnecessary agg node.
??
So there could be no project node and only tablescan node in subquery, but when
we try to do type conversion for the subquery, with the following code, it'll
throw exception when there's no project node.
{code:java}
if (queryRelNode instanceof Project) {
return replaceProjectForTypeConversion(
rexBuilder,
(Project) queryRelNode,
targetCalcTypes,
targetHiveTypes,
funcConverter);
} else {
RelNode newInput =
addTypeConversions(
rexBuilder,
queryRelNode.getInput(0),
targetCalcTypes,
targetHiveTypes,
funcConverter);
queryRelNode.replaceInput(0, newInput);
return queryRelNode;
}
{code}
was:
With hive dialect, can be reproduced using following code:
{code:java}
tableEnv.executeSql("CREATE TABLE t1 (c1 INT, c2 CHAR(100))");
tableEnv.executeSql("CREATE TABLE t2 (c1 INT)");
List<Row> results = CollectionUtil.iteratorToList(tableEnv.executeSql("SELECT
c1 FROM t1 WHERE c1 IN (SELECT c1 FROM t2)").
collect());
{code}
Then it will throw IndexOutOfBoundsException: 0, the reason is
" If it's a subquery and the project is identity, we skip creating this
project. This is to handle an issue with calcite SubQueryRemoveRule. The rule
checks col uniqueness by calling RelMetadataQuery::areColumnsUnique with an
empty col set, which always returns null for a project and thus introduces
unnecessary agg node.
"
So there could be no project node and only tablescan node in subquery, but when
we try to do type conversion for the subquery, with the following code, it'll
throw exception when there's no project node.
{code:java}
if (queryRelNode instanceof Project) {
return replaceProjectForTypeConversion(
rexBuilder,
(Project) queryRelNode,
targetCalcTypes,
targetHiveTypes,
funcConverter);
} else {
RelNode newInput =
addTypeConversions(
rexBuilder,
queryRelNode.getInput(0),
targetCalcTypes,
targetHiveTypes,
funcConverter);
queryRelNode.replaceInput(0, newInput);
return queryRelNode;
}
{code}
> "IndexOutOfBoundsException" when subquery select all field from using hive
> dialect
> ----------------------------------------------------------------------------------
>
> Key: FLINK-26362
> URL: https://issues.apache.org/jira/browse/FLINK-26362
> Project: Flink
> Issue Type: Sub-task
> Reporter: luoyuxia
> Priority: Major
> Labels: pull-request-available
>
> With hive dialect, can be reproduced using following code:
> {code:java}
> tableEnv.executeSql("CREATE TABLE t1 (c1 INT, c2 CHAR(100))");
> tableEnv.executeSql("CREATE TABLE t2 (c1 INT)");
> List<Row> results = CollectionUtil.iteratorToList(tableEnv.executeSql("SELECT
> c1 FROM t1 WHERE c1 IN (SELECT c1 FROM t2)").
> collect());
> {code}
> Then it will throw IndexOutOfBoundsException: 0, the reason is from the
> following comments:
> ?? If it's a subquery and the project is identity, we skip creating this
> project. This is to handle an issue with calcite SubQueryRemoveRule. The rule
> checks col uniqueness by calling RelMetadataQuery::areColumnsUnique with an
> empty col set, which always returns null for a project and thus introduces
> unnecessary agg node.
> ??
> So there could be no project node and only tablescan node in subquery, but
> when we try to do type conversion for the subquery, with the following code,
> it'll throw exception when there's no project node.
> {code:java}
> if (queryRelNode instanceof Project) {
> return replaceProjectForTypeConversion(
> rexBuilder,
> (Project) queryRelNode,
> targetCalcTypes,
> targetHiveTypes,
> funcConverter);
> } else {
> RelNode newInput =
> addTypeConversions(
> rexBuilder,
> queryRelNode.getInput(0),
> targetCalcTypes,
> targetHiveTypes,
> funcConverter);
> queryRelNode.replaceInput(0, newInput);
> return queryRelNode;
> }
> {code}
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
