谁能解释分组数据集上 运行 combineGroup 和 reduceGroup 转换之间的不同行为？

Any one who can explain the different behavior between run combineGroup and reduceGroup transformation on a grouped dataset?

谁能解释分组数据集上 combineGroup 和 reduceGroup 转换之间的不同行为运行？

ExecutionEnvironment executionEnvironment = ExecutionEnvironment.getExecutionEnvironment();

DataSet<Integer> input = executionEnvironment.fromElements(1,2,2,3,4,5,6,7,8,9);
input.map(new MapFunction<Integer, Tuple1<Integer>>() {
    public Tuple1<Integer> map(Integer value) throws Exception {
        return new Tuple1<Integer>(value
        );
    }
}).groupBy(0).reduceGroup(new GroupReduceFunction<Tuple1<Integer>, Integer>() {
    public void reduce(Iterable<Tuple1<Integer>> values, Collector<Integer> out) throws Exception {
        int sum = 0;

        for (Tuple1<Integer> value : values) {
            sum += value.f0;
        }
        out.collect(sum);
    }
}).print();

分组数据集上的运行s reduceGroup 转换，控制台显示为：

4 6个 4个 1个 7 9 3个 5个 8

input.map(new MapFunction<Integer, Tuple1<Integer>>() {
    public Tuple1<Integer> map(Integer value) throws Exception {
        return new Tuple1<Integer>(value);
    }
}).groupBy(0).combineGroup(new GroupCombineFunction<Tuple1<Integer>, Integer>() {
    public void combine(Iterable<Tuple1<Integer>> values, Collector<Integer> out) throws Exception {
        int sum = 0;

        for (Tuple1<Integer> value : values) {
            sum += value.f0;
        }
        out.collect(sum);
    }
}).print();

分组数据集上的运行s combineGroup，控制台显示为：

4 8个 1个 5个 9 4个 6个 3个 7

很明显，我看不出这个例子有什么不同。

两者计算相同的东西。因此，对于这两种情况，您都会得到相同的结果。差异有些微妙。基本上，"The GroupCombine transformation is the generalized form of the combine step in the Combinable GroupReduceFunction".

阅读文档以获得详细解释，如果还有不明白的地方再问：