Github user fhueske commented on a diff in the pull request:
https://github.com/apache/flink/pull/3673#discussion_r115746347
--- Diff:
flink-libraries/flink-table/src/test/scala/org/apache/flink/table/api/scala/batch/sql/DataSetSingleRowJoinTest.scala
---
@@ -187,9 +187,155 @@ class SingleRowJoinTest extends TableTestBase {
),
term("where", "AND(<(a1, b1)", "=(a2, b2))"),
term("join", "a1", "a2", "b1", "b2"),
- term("joinType", "NestedLoopJoin")
+ term("joinType", "NestedLoopInnerJoin")
)
util.verifySql(query, expected)
}
+
+ @Test
+ def testSingleRowJoinLeftOuterJoin(): Unit = {
+ val util = batchTestUtil()
+ util.addTable[(Long, Int)]("A", 'a1, 'a2)
+ util.addTable[(Int, Int)]("B", 'b1, 'b2)
+
+ val queryLeftJoin =
+ "SELECT a2 FROM A " +
+ "LEFT JOIN " +
+ "(SELECT COUNT(*) AS cnt FROM B) " +
+ "AS x " +
+ "ON a1 = cnt"
+
+ val expected =
+ unaryNode(
+ "DataSetCalc",
+ unaryNode(
+ "DataSetSingleRowJoin",
+ batchTableNode(0),
+ term("where", "=(a1, cnt)"),
+ term("join", "a1", "a2", "cnt"),
+ term("joinType", "NestedLoopLeftJoin")
+ ),
+ term("select", "a2")
+ ) + "\n" +
+ unaryNode(
+ "DataSetAggregate",
+ unaryNode(
+ "DataSetUnion",
+ unaryNode(
+ "DataSetValues",
+ unaryNode(
+ "DataSetCalc",
+ batchTableNode(1),
+ term("select", "0 AS $f0")),
+ tuples(List(null)), term("values", "$f0")
+ ),
+ term("union", "$f0")
+ ),
+ term("select", "COUNT(*) AS cnt")
+ )
+
+ util.verifySql(queryLeftJoin, expected)
+ }
+
+ @Test
+ def testSingleRowJoinRightOuterJoin(): Unit = {
+ val util = batchTestUtil()
+ util.addTable[(Long, Int)]("A", 'a1, 'a2)
+ util.addTable[(Int, Int)]("B", 'b1, 'b2)
+
+ val queryRightJoin =
--- End diff --
The generate join is a `RightOuterJoin` but not a `SingleRowJoin`, which
this test should verify.
We had to disable outer joins with predicates that include non-equi
conditions in FLINK-5520 because they were not properly implemented. That
implementation was based on splitting the join predicate into equi-conditions
which were evaluated by the join and non-equi-conditions which were evaluated
in a subsequent filter step. However, this split did not work correctly,
because it would generate too many `null` rows if records passed the equi-join
predicate in the join but not the non-equi predicate in the filter (since each
filter call did only see a single row and would not know if all other rows had
been filtered as well).
In our case the situation is different. We are translating the join into a
`NestedLoopJoin (where one side is at most one record), which can evaluate the
full predicate including the non-equi conditions inside the join and know if we
need to emit a `null` result because there is only a single row that either
matches the predicate or not.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
