import org.apache.flink.streaming.api.functions.source.SourceFunction
import org.apache.flink.streaming.api.scala.StreamExecutionEnvironment
import org.apache.flink.api.scala._

case class Type1(){}
case class Type2(){}

object MultipleOutputJob {
  def main(args: Array[String]) {
    val env = StreamExecutionEnvironment.getExecutionEnvironment

    // Stream of Type1
    val stream1 = env.addSource((sc: SourceFunction.SourceContext[Type1]) => {
      while(true){
        Thread.sleep(1000)
        sc.collect(Type1())
      }
    })

    // Mapping from Type1 to Type2
    val stream2 = stream1.map(t1 => Type2())

    // Collect both the original and the derived data
    stream1.print
    stream2.print

    env.execute()
  }
}

您需要先创建一个 Package 器类型，然后拆分并选择流。对于 Package 器，只有一个成员不为null；

class TypeWrapper {
    // keeping this short for brevity
    public TypeA firstType;
    public TypeB secondType;
}

拆分并选择：

DataStream<TypeWrapper> stream1 = ...

DataStream<TypeA> streamA = stream1.filter(new FilterFunction<TypeWrapper>() {
    @Override
    public boolean filter(TypeWrapper value) throws Exception {
        return value.firstType != null;
    }
})
.map(new MapFunction<TypeWrapper, TypeA>() {
    @Override
    public TypeA map(TypeWrapper value) throws Exception {
        return value.firstType;
    }
});

DataStream<TypeB> streamB = stream1.filter(new FilterFunction<TypeWrapper>() {
    @Override
    public boolean filter(TypeWrapper value) throws Exception {
        return value.secondType != null;
    }
})
.map(new MapFunction<TypeWrapper, TypeB>() {
    @Override
    public TypeB map(TypeWrapper value) throws Exception {
        return value.secondType;
    }
});

因为 filter() 以及 map() 将被锁在 stream1 两者都在同一个线程上执行，而且操作成本很低。