Flink 会话窗口&全局窗口的实践

416 阅读1分钟

会话窗口

会话窗口,没有固定的窗口开始时间和结束时间,只有一个固定的会话间隔,如果超过此gap没有数据的话,会触发此窗口的输出

image.png

全局窗口

全局窗口是程序开始一直都存在的窗口,必须定义窗口的触发器,不然程序不知道什么时间输出。这里有值得注意的是,窗口没有数据了,程序也会继续输出。

image.png

示例代码

StreamExecutionEnvironment executionEnvironment =
        StreamExecutionEnvironment.getExecutionEnvironment();
executionEnvironment.setParallelism(1);
DataStreamSource<String> dataStreamSource = executionEnvironment.socketTextStream(
        "localhost", 9097);
SingleOutputStreamOperator<City> map = dataStreamSource.map(x -> {
    String[] split = x.split(",");
    City city = new City();
    city.setName(split[0]);
    city.setNum(Long.valueOf(split[1]));
    return city;
});
map.keyBy(x -> x.getName()).window(ProcessingTimeSessionWindows.withGap(Time.seconds(10))).reduce(new ReduceFunction<City>() {
    @Override
    public City reduce(City value1, City value2) throws Exception {
        value1.setNum(value1.getNum() + value2.getNum());
        return value1;
    }
}).print();


map.keyBy(x -> x.getName()).window(GlobalWindows.create()).trigger(ContinuousProcessingTimeTrigger.of(Time.seconds(10))).reduce(new ReduceFunction<City>() {
    @Override
    public City reduce(City value1, City value2) throws Exception {
        value1.setNum(value1.getNum() + value2.getNum());
        return value1;
    }
}).print();

try {
    executionEnvironment.execute("test session/global windows");
} catch (Exception e) {
    e.printStackTrace();
}