[FLINK-26505] Fix IndexOutOfBoundsException for non equality condition exists in left semi join in Hive dialect

• Type: Sub-task
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.16.0
• Component/s: Connectors / Hive
• Labels: pull-request-available
• Assignee: luoyuxia
• Reporter: luoyuxia
• Created: 07/Mar/22 03:49
• Updated: 16/Aug/22 13:07
• Resolved: 16/Aug/22 13:07

Description

It'll throw exception "java.lang.IndexOutOfBoundsException: Index: 0, Size: 0" with the following sql which can be executed in Hive:

select count(1)
from
  (select key
  from t1
  where key = 0) t1
left semi join
  (select key
  from t2
  where key = 0) t2
on 1 = 1;

From the source code, it will call `RelOptUtil.splitJoinCondition` to split join condition and initialize the `leftJoinKeys` and `rightJoinKeys` passed in the method:

  public static RexNode splitJoinCondition(
      List<RelDataTypeField> sysFieldList,
      RelNode leftRel,
      RelNode rightRel,
      RexNode condition,
      List<RexNode> leftJoinKeys,
      List<RexNode> rightJoinKeys,
      List<Integer> filterNulls,
      List<SqlOperator> rangeOp)

But when meet the case that there's no equality join keys found, the behavior will be wired that the `leftJoinKeys` will contain two RexNodes and the `rightJoinKeys` contains nothing. Then it'll cause the exception for we expect the size of `leftJoinKeys` is equal to `rightJoinKeys`.
It seems a issue of Calcite CALCITE-5032.

To fix it, just following what hive has done, we can rewrite the method `RelOptUtil.splitJoinCondition` to fix it. And what's more,
it makes it's possible to support non-equi LEFT SEMI JOIN supported in Hive HIVE-17766.

Issue Links

• links to: GitHub Pull Request #18994