[FLINK-11070] Add stream-stream non-window cross join

• Type: New Feature
• Status: Open
• Priority: Not a Priority
• Resolution: Unresolved
• Component/s: Table SQL / API
• Labels: auto-deprioritized-major auto-deprioritized-minor auto-unassigned
• Assignee: Unassigned
• Reporter: Hequn Cheng
• Created: 04/Dec/18 12:10
• Updated: 26/Dec/21 10:39

Description

Currently, we don't reorder join and rely on the order provided by the user. This is fine for most of the cases, however, it limits the set of supported SQL queries.
Example:

val streamUtil: StreamTableTestUtil = streamTestUtil()
streamUtil.addTable[(Int, String, Long)]("MyTable", 'a, 'b, 'c.rowtime, 'proctime.proctime)
streamUtil.addTable[(Int, String, Long)]("MyTable2", 'a, 'b, 'c.rowtime, 'proctime.proctime)
streamUtil.addTable[(Int, String, Long)]("MyTable3", 'a, 'b, 'c.rowtime, 'proctime.proctime)
val sqlQuery =
      """
        |SELECT t1.a, t3.b
        |FROM MyTable3 t3, MyTable2 t2, MyTable t1
        |WHERE t1.a = t3.a AND t1.a = t2.a
        |""".stripMargin

streamUtil.printSql(sqlQuery)

Given the current rule sets, this query produces a cross join which is not supported and thus leads to:

org.apache.flink.table.api.TableException: Cannot generate a valid execution plan for the given query: 

LogicalProject(a=[$8], b=[$1])
  LogicalFilter(condition=[AND(=($8, $0), =($8, $4))])
    LogicalJoin(condition=[true], joinType=[inner])
      LogicalJoin(condition=[true], joinType=[inner])
        LogicalTableScan(table=[[_DataStreamTable_2]])
        LogicalTableScan(table=[[_DataStreamTable_1]])
      LogicalTableScan(table=[[_DataStreamTable_0]])

This exception indicates that the query uses an unsupported SQL feature.
Please check the documentation for the set of currently supported SQL features.

In order to support more queries, it would be nice to have cross join on streaming. We can start from a simple version, for example, call forceNonParallel() for connectOperator in `DataStreamJoin` when it is a cross join. The performance may be bad. But it works fine if the two tables of cross join are small ones.
We can do some optimizations later, such as broadcasting the smaller side, etc.

Any suggestions are greatly appreciated.

Issue Links

• relates to: FLINK-11020 Reorder joins only to eliminate cross joins