Flink的KeyedProcessFunction基于Event Time和Process Time的定时器用法实例分析

FLink处理函数简介

在Flink底层，我们可以不定义任何具体的算子（比如 map，filter，或者 window），而只是提炼出一个统一的【处理】（process）操作——它是所有转换算子的一个概括性的表达，可以自定义处理逻辑，所以这一层接口就被叫作【处理函数】（process function）。在处理函数中，我们直面的就是数据流中最基本的元素：数据事件（event）、状态（state）以及时间（time）。这就相当于对流有了完全的控制权。处理函数比较抽象，没有具体的操作，所以对于一些常见的简单应用（比如求和、开窗口）会显得有些麻烦；不过正是因为它不限定具体做什么，所以理论上我们可以做任何事情，实现所有需求。

Flink几种处理函数简介

1. ProcessFunction是用于处理数据流的通用函数。它是一个抽象类，定义了处理数据流的常用方法，如processElement，onTimer等。您可以扩展ProcessFunction类并重写这些方法，以便在Flink程序中执行复杂的数据流处理逻辑。
2. KeyedProcessFunction是ProcessFunction的特殊类型，用于处理带有键的数据流。它定义了额外的方法，如getKey，context.timerService()等，用于访问数据流中每个元素的键以及在处理函数中安排定时器。
3. ProcessWindowFunction和ProcessAllWindowFunction是用于处理时间窗口的特殊函数。它们提供了一个process方法，用于在每个窗口中对数据进行处理。ProcessWindowFunction接受带有键的数据流，并且每个窗口都对应于一个键，而ProcessAllWindowFunction接受不带键的数据流，并且每个窗口都包含整个数据流。

这里重点介绍KeyedProcessFunction，KeyedProcessFunction是用来处理KeyedStream的。每有一个数据进入算子，则会触发一次processElement()的处理。它还提供了定时器的功能，在在预警、监控等场景特定场景下，非常适合。
KeyedProcessFunction定时器包分为两种：基于事件时间、基于处理时间。下面以统计计数的方式展示这两种定时器的用法，并附上详细的分析思路。以下用例基于Flink1.14

实例分析

KeyedProcessFunction基于事件时间的定时器

代码：

import org.apache.flink.api.common.eventtime.SerializableTimestampAssigner;
import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.common.state.ValueState;
import org.apache.flink.api.common.state.ValueStateDescriptor;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.configuration.Configuration;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.KeyedProcessFunction;
import org.apache.flink.util.Collector;
import org.apache.flink.util.OutputTag;

import java.text.SimpleDateFormat;
import java.time.Duration;
import java.util.Date;

/**
 * @description:
 *
 * @author pony
 * @date 2024/1/17 20:55
 * @version 1.0
 * nc -l 9999
 */
public class KeyedProcessFunctionOnTimerEventTime {
    public static void main(String[] args) throws Exception {

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        WatermarkStrategy<String> watermarkStrategy = WatermarkStrategy
                .<String>forBoundedOutOfOrderness(Duration.ofSeconds(60))
                .withTimestampAssigner(new SerializableTimestampAssigner<String>() {
                    @Override
                    public long extractTimestamp(String element, long recordTimestamp) {
                        return Long.valueOf(element.split(",")[1]);
                    }
                })
                .withIdleness(Duration.ofSeconds(1));

        DataStream<Tuple2<String, Long>> stream0 = env.socketTextStream("x.x.x.x", 9999)
                .assignTimestampsAndWatermarks(watermarkStrategy) //必须在数据源上指定watermark
                .map(new MapFunction<String, Tuple2<String, Long>>() {
                    @Override
                    public Tuple2<String, Long> map(String value) throws Exception {
                        return new Tuple2<String, Long>(value.split(",")[0], Long.valueOf(value.split(",")[1]));
                    }
                });

        // apply the process function onto a keyed stream
        DataStream<Tuple2<String, Long>> result = stream0
                .keyBy(value -> value.f0)
                .process(new CountEventTimeWithTimeoutFunction());

        result.print();

        env.execute("KeyedProcessFunction wordCount");
    }

    /**
     * The implementation of the ProcessFunction that maintains the count and timeouts
     */
    static class CountEventTimeWithTimeoutFunction
            extends KeyedProcessFunction<String, Tuple2<String, Long>, Tuple2<String, Long>> {

        private ValueState<Long> state;
        private static final Integer DELAY = 1000; //1s

        @Override
        public void open(Configuration parameters) throws Exception {
            state = getRuntimeContext().getState(new ValueStateDescriptor<>("myState", Long.class));
        }

        @Override
        public void processElement(
                Tuple2<String, Long> value,
                Context ctx,
                Collector<Tuple2<String, Long>> out) throws Exception {

            Long current = state.value();
            if (current == null) {
                current = 0L;
            }
            current++;
            state.update(current);
            //获取当前数据流的水位线
            long currentWatermark = ctx.timerService().currentWatermark();

//            long timer = ctx.timestamp() + DELAY;//设置定时器的时间为当前event time+DELAY
            long timer = currentWatermark + DELAY;//设置定时器的时间为当前水位线+DELAY
            //注册事件时间定时器，与watermark绑定，必须满足条件: watermark >= timer 来触发特定event的定时器
            ctx.timerService().registerEventTimeTimer(timer);

            //删除事件时间定时器
            if (currentWatermark < 0) {
                ctx.timerService().deleteEventTimeTimer(timer);
            }

            System.out.println("last Watermark: " + currentWatermark + ", format: " + time(currentWatermark));

            // 打印信息，用于核对数据
            System.out.println(String.format("processElement: %s, %d, ctx.timestamp() : %d (%s), timer : %d (%s)\n",
                    ctx.getCurrentKey(),
                    current,
                    ctx.timestamp(),
                    time(ctx.timestamp()),
                    timer,
                    time(timer)));

        }

        @Override
        public void onTimer(
                long timestamp, //定时器触发时间，等于以上的timer
                OnTimerContext ctx,
                Collector<Tuple2<String, Long>> out) throws Exception {
            // 取得当前单词
            String currentKey = ctx.getCurrentKey();
            // get the state for the key that scheduled the timer
            Long result = state.value();

            // 打印数据，用于核对是否符合预期
            System.out.println(String.format("onTimer: %s, %d, ctx.timestamp() : %d (%s), timestamp : %d (%s)\n",
                    currentKey,
                    result,
                    ctx.timestamp(),
                    time(ctx.timestamp()),
                    timestamp,
                    time(timestamp)));
            System.out.println("current Watermark: " + ctx.timerService().currentWatermark() + ", format: " + time(ctx.timerService().currentWatermark()));
            
            out.collect(new Tuple2<String, Long>(currentKey, result));

        }

        @Override
        public void close() throws Exception {
            super.close();
            state.clear();
        }
    }

    public static String time(long timeStamp) {
        return new SimpleDateFormat("yyyy-MM-dd HH:mm:ss.SSS").format(new Date(timeStamp));
    }
}

测试数据：

nc -l 9999
a1,1704038400000
a1,1704038401000
a1,1704038403000

运行结果：

KeyedProcessFunction基于处理时间的定时器

代码：

import org.apache.flink.api.common.eventtime.SerializableTimestampAssigner;
import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.api.common.functions.MapFunction;
import org.apache.flink.api.common.state.ValueState;
import org.apache.flink.api.common.state.ValueStateDescriptor;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.configuration.Configuration;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.KeyedProcessFunction;
import org.apache.flink.util.Collector;
import org.apache.flink.util.OutputTag;

import java.text.SimpleDateFormat;
import java.time.Duration;
import java.util.Date;

/**
 * @description:
 *
 * @author pony
 * @date 2024/1/17 20:55
 * @version 1.0
 * nc -l 9999
 */
public class KeyedProcessFunctionOnTimerProcessTime {
    public static void main(String[] args) throws Exception {

        StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();

        WatermarkStrategy<String> watermarkStrategy = WatermarkStrategy
                .<String>forBoundedOutOfOrderness(Duration.ofSeconds(60))
                .withTimestampAssigner(new SerializableTimestampAssigner<String>() {
                    @Override
                    public long extractTimestamp(String element, long recordTimestamp) {
//                        return System.currentTimeMillis();
                        return Long.valueOf(element.split(",")[1]);
                    }
                })
                .withIdleness(Duration.ofSeconds(1));

        DataStream<Tuple2<String, Long>> stream0 = env.socketTextStream("x.x.x.x", 9999)
                .assignTimestampsAndWatermarks(watermarkStrategy) //必须在数据源上指定watermark
                .map(new MapFunction<String, Tuple2<String, Long>>() {
                    @Override
                    public Tuple2<String, Long> map(String value) throws Exception {
                        return new Tuple2<String, Long>(value.split(",")[0], Long.valueOf(value.split(",")[1]));
                    }
                });

        // apply the process function onto a keyed stream
        DataStream<Tuple2<String, Long>> result = stream0
                .keyBy(value -> value.f0)
                .process(new CountProcessTimeWithTimeoutFunction());

        result.print();

        env.execute("KeyedProcessFunction wordCount");
    }

    static class CountProcessTimeWithTimeoutFunction
            extends KeyedProcessFunction<String, Tuple2<String, Long>, Tuple2<String, Long>> {

        private ValueState<Long> state;
        private static final Integer DELAY = 60 * 1000; //1s

        @Override
        public void open(Configuration parameters) throws Exception {
            state = getRuntimeContext().getState(new ValueStateDescriptor<>("myState", Long.class));
        }

        @Override
        public void processElement(
                Tuple2<String, Long> value,
                Context ctx,
                Collector<Tuple2<String, Long>> out) throws Exception {

            Long current = state.value();
            if (current == null) {
                current = 0L;
            }
            current++;
            state.update(current);

            long timer = ctx.timestamp() + DELAY;//设置定时器的时间为当前event time+DELAY
            //注册处理时间定时器, 与watermark无关，定时器触发条件：当前系统时间>timer
            ctx.timerService().registerProcessingTimeTimer(timer);
            //删除处理时间定时器
//            ctx.timerService().deleteProcessingTimeTimer(timer);

            System.out.println("processElement currentProcessingTime: " + ctx.timerService().currentProcessingTime() + ", format: " + time(ctx.timerService().currentProcessingTime()));
            // 打印所有信息，用于核对数据
            System.out.println(String.format("processElement: %s, %d, ctx.timestamp() : %d (%s), timer : %d (%s)\n",
                    ctx.getCurrentKey(),
                    current,
                    ctx.timestamp(),
                    time(ctx.timestamp()),
                    timer,
                    time(timer)));
        }

        @Override
        public void onTimer(
                long timestamp,
                OnTimerContext ctx,
                Collector<Tuple2<String, Long>> out) throws Exception {
            // 取得当前单词
            String currentKey = ctx.getCurrentKey();
            // get the state for the key that scheduled the timer
            Long result = state.value();

            System.out.println("onTimer currentProcessingTime: " + ctx.timerService().currentProcessingTime() + ", format: " + time(ctx.timerService().currentProcessingTime()));
            // 打印数据，用于核对是否符合预期
            System.out.println(String.format("onTimer: %s, %d, ctx.timestamp() : %d (%s), timestamp : %d (%s)\n",
                    currentKey,
                    result,
                    ctx.timestamp(),
                    time(ctx.timestamp()),
                    timestamp,
                    time(timestamp)));

            //另外还支持侧流
            OutputTag<Tuple2<String, Long>> outputTag = new OutputTag<Tuple2<String, Long>>("single"){};
            if (result < 2) {
                ctx.output(outputTag, new Tuple2<>(currentKey, result));
            } else {
                out.collect(new Tuple2<String, Long>(currentKey, result));
            }

        }

        @Override
        public void close() throws Exception {
            super.close();
            state.clear();
        }
    }

    public static String time(long timeStamp) {
        return new SimpleDateFormat("yyyy-MM-dd HH:mm:ss.SSS").format(new Date(timeStamp));
    }
}

测试数据：

nc -l 9999
a,1705568024000    
a,1705568024000

运行结果：

总结

在真实业务场景中【 KeyedProcessFunction基于处理时间的定时器】用的比较多，比较符合业务场景，即根据事件的时间来指定处理时间去定时触发定时器。因此在此场景中，可以不指定watermarkStrategy，可以获取传输参数的时间时间来定时触发定时器。

参考：
Process Function
Generating Watermarks