Flink-intervalJoin源码和并行度问题

1.源码

底层用的是connect

把两个流的数据先保存到状态中

 先判断有没有迟到，迟到就放到侧输出流

再根据范围找数据

然后根据上界删除数据

 

 

package com.atguigu.gmall.realtime.test;

import org.apache.flink.api.common.eventtime.SerializableTimestampAssigner;
import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.co.ProcessJoinFunction;
import org.apache.flink.streaming.api.windowing.time.Time;
import org.apache.flink.util.Collector;

/**
 * @author 
 * @date 2023/7/8
 * 该案例演示了Flink的intervalJoin
 */
public class Test03_IntervalJoin {
    public static void main(String[] args) throws Exception {
        //TODO 1.基本环境准备
        //1.1 指定流处理环境
        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
        //1.2 设置并行度
        env.setParallelism(2);

        //TODO 2.检查点相关的设置(略)
        //TODO 3.从指定的网络端口获取员工数据，指定Watermark以及提取事件时间字段
        SingleOutputStreamOperator empDS = env
            .socketTextStream("hadoop102", 8888)
            .map(
                listStr -> {
                    String[] fieldArr = listStr.split(",");
                    return new Emp(Integer.valueOf(fieldArr[0]), fieldArr[1],
                        Integer.valueOf(fieldArr[2]), Long.valueOf(fieldArr[3]));
                }
            ).assignTimestampsAndWatermarks(
                // WatermarkStrategy.forBoundedOutOfOrderness(Duration.ofSeconds(3))
                WatermarkStrategy
                    .forMonotonousTimestamps()
                    .withTimestampAssigner(
                        new SerializableTimestampAssigner() {
                            @Override
                            public long extractTimestamp(Emp emp, long recordTimestamp) {
                                return emp.getTs();
                            }
                        }
                    )
            );
        empDS.print("emp:");
        //TODO 4.从指定的网络端口获取部门数据，指定Watermark以及提取事件时间字段
        SingleOutputStreamOperator deptDS = env
            .socketTextStream("hadoop102", 8889)
            .map(
                lineStr -> {
                    String[] fieldArr = lineStr.split(",");
                    return new Dept(Integer.valueOf(fieldArr[0]), fieldArr[1], Long.valueOf(fieldArr[2]));
                }
            ).assignTimestampsAndWatermarks(
                WatermarkStrategy
                    .forMonotonousTimestamps()
                    .withTimestampAssigner(
                        new SerializableTimestampAssigner() {
                            @Override
                            public long extractTimestamp(Dept dept, long recordTimestamp) {
                                return dept.getTs();
                            }
                        }
                    )
            );

        deptDS.print("dept:");

        //TODO 5.使用intervalJoin关联员工和部门
        empDS
            .keyBy(Emp::getDeptno)
            .intervalJoin(deptDS.keyBy(Dept::getDeptno))
            .between(Time.milliseconds(-5),Time.milliseconds(5))
            .process(
                new ProcessJoinFunction>() {
                    @Override
                    public void processElement(Emp emp, Dept dept, Context ctx, Collector> out) throws Exception {
                        out.collect(Tuple2.of(emp,dept));
                    }
                }
            ).print(">>>");
        env.execute();
    }
}

2.并行度问题

多个上游，一个下游，下游取最小的

一个上游，多个下游，广播

多个上游，多个下游，先广播，再取最小的

注意：水位线会减1ms

有可能出现上游出现了某个水位线，但是不触发，是因为别的并行度没有数据，水位线是long的最小值

尽量在source处设置水位线

只有kafka source是多个并行度，其他source都是一个并行度