split和side-output分流源码分析
在日常开发中，我们常常需要对数据流进行拆分处理，flink提供了split/side output方式进行分流（filter分流方式使用场景有限，不作讨论）。首先，同一个流上不能同时使用split和side-output进行分流。否则，会抛出如下异常：
throw new UnsupportedOperationException("getSideOutput() and split() may not be called on the same DataStream. " +
"As a work-around, please add a no-op map function before the split() call.");
split分流源码分析
@Override
public SplitStream<T> split(OutputSelector<T> outputSelector) {
if (requestedSideOutputs.isEmpty()) {
wasSplitApplied = true;
return super.split(outputSelector);
} else {
throw new UnsupportedOperationException("getSideOutput() and split() may not be called on the same DataStream. " +
"As a work-around, please add a no-op map function before the split() call.");
}
}
当在SingleOutputStreamOperator上调用split方法后，会首先对requestedSideOutputs进行检查，requestedSideOutputs是一个Map<OutputTag<?>, TypeInformation>集合，如果此流没有进行过side-output分流，则将wasSplitApplied = true;然后调用SingleOutputStreamOperator父类DataStream上的split方法，如下：
public SplitStream<T> split(OutputSelector<T> outputSelector) {
return new SplitStream<>(this, clean(outputSelector));
}
返回一个SplitStream，SplitStream继承了DataStream,SplitStream通过selectOutput方法将SplitStream再次转化成DataStream流。
- 注意：split不能连续分流，经过split分流后的DataStream不能再继续进行split分流。否则会抛如下异常：
Exception in thread "main" java.lang.IllegalStateException: Consecutive multiple splits are not supported. Splits are deprecated. Please use side-outputs.
side-outputs分流源码分析（推荐使用）
如果想对流进行多次拆分，则必须使用side-output的方式进行分流，SingleOutputStreamOperator类提供了getSideOutput方法对流进行拆分，源码如下：
public <X> DataStream<X> getSideOutput(OutputTag<X> sideOutputTag) {
if (wasSplitApplied) {
throw new UnsupportedOperationException("getSideOutput() and split() may not be called on the same DataStream. " +
"As a work-around, please add a no-op map function before the split() call.");
}
sideOutputTag = clean(requireNonNull(sideOutputTag));
// make a defensive copy
sideOutputTag = new OutputTag<X>(sideOutputTag.getId(), sideOutputTag.getTypeInfo());
TypeInformation<?> type = requestedSideOutputs.get(sideOutputTag);
if (type != null && !type.equals(sideOutputTag.getTypeInfo())) {
throw new UnsupportedOperationException("A side output with a matching id was " +
"already requested with a different type. This is not allowed, side output " +
"ids need to be unique.");
}
requestedSideOutputs.put(sideOutputTag, sideOutputTag.getTypeInfo());
SideOutputTransformation<X> sideOutputTransformation = new SideOutputTransformation<>(this.getTransformation(), sideOutputTag);
return new DataStream<>(this.getExecutionEnvironment(), sideOutputTransformation);
}
该方法需要传入一个OutputTag，然后检查是否已经执行过split方法对流进行拆分，同时检查sideOutputTag的全局唯一性，sideOutputTag的id必须唯一，然后返回一个新的DataStream。