Flink SQL:在纯SQL语法中用时间戳连接表

Question

我在使用Flink的SQL语法连接多个表时遇到了一些问题，其中至少有一个表具有时间属性列。

我有一个表 Table1，当 rowtime 用作 flink rowtime 时，它使用架构（id、value1、rowtime）。

我想将此表与使用模式（id， value 2）的表Table2连接。连接必须在匹配id时完成。

最后，我想通过使用滚动时间窗口对这个连接的结果进行分组。

仅仅使用SQL语法有可能做到这一点吗？

下面是我想做的一个例子:

SELECT 
    Table1.id as id, 
    TUMBLE_END(rowtime, INTERVAL '10' SECOND),
    MAX(value1) as value1,
    MAX(value2) as value2       
FROM Table1 JOIN TABLE2 ON Table1.id = Table2.id
GROUP BY Table1.id, TUMBLE(rowtime, INTERVAL '10' SECOND)

但它给了我以下错误：

2019-11-12 16:37:57.191 [main] ERROR - Cannot generate a valid execution plan for the given query: 

FlinkLogicalCalc(expr#0..6=[{inputs}], id=[$t0], EXPR$1=[$t4], value1=[$t1], value2=[$t2])
  FlinkLogicalWindowAggregate(group=[{0}], value1=[MAX($2)], value2=[MAX($3)])
    FlinkLogicalCalc(expr#0..2=[{inputs}], expr#3=[0], proj#0..1=[{exprs}], value1=[$t3], value2=[$t3])
      FlinkLogicalJoin(condition=[=($0, $2)], joinType=[inner])
        FlinkLogicalTableSourceScan(table=[[Table1]], fields=[id, value1, rowtime], source=[KafkaTableSource(id, value1, rowtime)])
        FlinkLogicalTableSourceScan(table=[[Table2]], fields=[id, value2], source=[Table2_Type(id, value2)])

Rowtime attributes must not be in the input rows of a regular join. As a workaround you can cast the time attributes of input tables to TIMESTAMP before.
Please check the documentation for the set of currently supported SQL features.
org.apache.flink.table.api.TableException: Cannot generate a valid execution plan for the given query: 

FlinkLogicalCalc(expr#0..6=[{inputs}], id=[$t0], EXPR$1=[$t4], value1=[$t1], value2=[$t2])
  FlinkLogicalWindowAggregate(group=[{0}], value1=[MAX($2)], value2=[MAX($3)])
    FlinkLogicalCalc(expr#0..2=[{inputs}], expr#3=[0], proj#0..1=[{exprs}], value1=[$t3], value2=[$t3])
      FlinkLogicalJoin(condition=[=($0, $2)], joinType=[inner])
        FlinkLogicalTableSourceScan(table=[[kafkaDataStream]], fields=[id, value1, rowtime], source=[KafkaTableSource(id, value1, rowtime)])
        FlinkLogicalTableSourceScan(table=[[SensorConfigurationUpdateHTTP]], fields=[id, value2], source=[Table2_Type(id, value2)])

Rowtime attributes must not be in the input rows of a regular join. As a workaround you can cast the time attributes of input tables to TIMESTAMP before.
Please check the documentation for the set of currently supported SQL features.
    at org.apache.flink.table.api.TableEnvironment.runVolcanoPlanner(TableEnvironment.scala:387)
    at org.apache.flink.table.api.TableEnvironment.optimizePhysicalPlan(TableEnvironment.scala:302)
    at org.apache.flink.table.api.StreamTableEnvironment.optimize(StreamTableEnvironment.scala:816)
    at org.apache.flink.table.api.StreamTableEnvironment.writeToSink(StreamTableEnvironment.scala:379)
    at org.apache.flink.table.api.TableEnvironment.insertInto(TableEnvironment.scala:879)
    at org.apache.flink.table.api.Table.insertInto(table.scala:1126)
    ...

我还尝试将我的rowtime转换为TIMESTAMP类型（根据错误消息的建议），但我无法再处理时间窗口。它会导致以下错误：

2019-11-12 16:44:52.473 [main] ERROR - Window can only be defined over a time attribute column.
org.apache.flink.table.api.ValidationException: Window can only be defined over a time attribute column.
    at org.apache.flink.table.plan.rules.datastream.DataStreamLogicalWindowAggregateRule.getOperandAsTimeIndicator$1(DataStreamLogicalWindowAggregateRule.scala:84)
    at org.apache.flink.table.plan.rules.datastream.DataStreamLogicalWindowAggregateRule.translateWindowExpression(DataStreamLogicalWindowAggregateRule.scala:89)
    at org.apache.flink.table.plan.rules.common.LogicalWindowAggregateRule.onMatch(LogicalWindowAggregateRule.scala:65)
    at org.apache.calcite.plan.AbstractRelOptPlanner.fireRule(AbstractRelOptPlanner.java:315)
    at org.apache.calcite.plan.hep.HepPlanner.applyRule(HepPlanner.java:556)
    at org.apache.calcite.plan.hep.HepPlanner.applyRules(HepPlanner.java:415)
    at org.apache.calcite.plan.hep.HepPlanner.executeInstruction(HepPlanner.java:252)
    at org.apache.calcite.plan.hep.HepInstruction$RuleInstance.execute(HepInstruction.java:127)
    at org.apache.calcite.plan.hep.HepPlanner.executeProgram(HepPlanner.java:211)
    at org.apache.calcite.plan.hep.HepPlanner.findBestExp(HepPlanner.java:198)
    at org.apache.flink.table.api.TableEnvironment.runHepPlanner(TableEnvironment.scala:360)
    at org.apache.flink.table.api.TableEnvironment.runHepPlannerSequentially(TableEnvironment.scala:326)
    at org.apache.flink.table.api.TableEnvironment.optimizeNormalizeLogicalPlan(TableEnvironment.scala:282)
    at org.apache.flink.table.api.StreamTableEnvironment.optimize(StreamTableEnvironment.scala:813)
    at org.apache.flink.table.api.StreamTableEnvironment.writeToSink(StreamTableEnvironment.scala:379)
    at org.apache.flink.table.api.TableEnvironment.insertInto(TableEnvironment.scala:879)
    at org.apache.flink.table.api.Table.insertInto(table.scala:1126)

Answer 1

联接结果不能包含时间属性，因为联接不能保证时间戳的顺序得到保留。Flink假设两个表都是动态的，可以在任何时间点更改。表表2中的一条新记录可能与表格1的第一条记录连接，产生时间戳为“随机”顺序的结果。

可以通过向联接添加时态约束来更改此设置。您可以使用时间窗口联接定义查询，或者将 Table2 建模为时态表，然后将 Table1 与它联接。