flink中通用的watermark设置代码模板（基于1.14版本）

文章介绍了ApacheFlink中处理乱序事件的watermark策略，包括允许一定延迟的BoundedOutOfOrderness策略、严格有序的MonotonousTimestamps策略以及不使用watermark和自定义watermark的情况。提供了在source和map算子中设置watermark的代码示例。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

flink中通用的watermark设置代码模板，

watermark生成策略通常有四种，长用的为第一、二种

一、WatermarkStrategy.forBoundedOutOfOrderness(Duration.ofMillis(3000)) 允许乱序的watermark // Duration.ofMillis(3000) 设置容忍时间为3000ms
二、WatermarkStrategy.forMonotonousTimestamps()
完全不容忍乱序，紧跟最大事件时间，等于第一个括号内填0，调的就是一的API
三、WatermarkStrategy.noWatermarks() 不使用watermark
四、WatermarkStrategy.forGenerator() 自定义watermark

通常会在source直接设置watermark，后续算子会传递依赖source的watermark

source设置watermark代码模板示例:

package com.syxStudy;

import java.time.Duration;

public class WaterMarkAPI {
    public static void main(String[] args) {
        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
        // 数据源获取socket文本流，假设数据格式为 1684243149,syx,18
        DataStreamSource<String> sourceStream = env.socketTextStream("192.168.xx.xxx", 9999);
        //构造一个watermark生成策略
        WatermarkStrategy<String> stringWatermarkStrategy = WatermarkStrategy.<String>forBoundedOutOfOrderness(Duration.ofMillis(3000)).withTimestampAssigner(new SerializableTimestampAssigner<String>() {
            @Override
            public long extractTimestamp(String s, long l) {
                return Long.parseLong(s.split(",")[0]);
            }
        });
        //将构造的watermark算子分配给source算子。不止可以在source分配watermark，也可以从后续算子开始生成watermark。一般来说是从source生成
        sourceStream.assignTimestampsAndWatermarks(stringWatermarkStrategy);
    }
}

在map算子中生成watermark（不常用）

package com.syxStudy;

import java.time.Duration;

public class WaterMarkAPI {
    public static void main(String[] args) {
        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
        // 数据源获取socket文本流，假设数据格式为 1684243149,syx,18
        DataStreamSource<String> sourceStream = env.socketTextStream("192.168.xx.xxx", 9999);
        
        //在map算子计算逻辑
        SingleOutputStreamOperator<EventBean> map_func = sourceStream.map(new MapFunction<String, EventBean>() {
            @Override
            public EventBean map(String s) throws Exception {
                String[] split_socket_txt = s.split(",");
                return new EventBean(Long.parseLong(split_socket_txt[0]), split_socket_txt[1], Integer.parseInt(split_socket_txt[0]));
            }
        }).returns(EventBean.class);

        //添加到map算子中
        SingleOutputStreamOperator<EventBean> map_add_watermark = map_func.assignTimestampsAndWatermarks(
                WatermarkStrategy.<EventBean>forMonotonousTimestamps()
                        .withTimestampAssigner(new SerializableTimestampAssigner<EventBean>() {
                            @Override
                            public long extractTimestamp(EventBean eventBean, long l) {
                                return eventBean.getTimestamp();
                            }
                        })
        );
    }
}
@Data
@AllArgsConstructor
@NoArgsConstructor
class EventBean{
    private Long timestamp;
    private String name;
    private int age;
}