Flink-1.12 - 之如何构建一个简单的TopN应用

Flink-1.12 - 之如何构建一个简单的TopN应用

本文主要介绍通过Flink-1.12如何构建一个简单的TopN应用，这里介绍

• DataStream API构建
• Flink SQL构建

1 maven依赖如下

    <!--当前版本的控制~~-->
    <properties>
        <maven.compiler.source>8</maven.compiler.source>
        <maven.compiler.target>8</maven.compiler.target>
        <flink.system.version>1.12.2</flink.system.version>
        <scala.version>2.12</scala.version>
    </properties>

    <dependencies>

        <!-- https://mvnrepository.com/artifact/org.apache.flink/flink-java -->
        <dependency>
            <groupId>org.apache.flink</groupId>
            <artifactId>flink-java</artifactId>
            <version>${flink.system.version}</version>
        </dependency>

        <!-- https://mvnrepository.com/artifact/org.apache.flink/flink-table-planner-blink -->
        <dependency>
            <groupId>org.apache.flink</groupId>
            <artifactId>flink-table-planner-blink_${scala.version}</artifactId>
            <version>${flink.system.version}</version>
        </dependency>

        <!-- https://mvnrepository.com/artifact/org.apache.flink/flink-streaming-java -->
        <dependency>
            <groupId>org.apache.flink</groupId>
            <artifactId>flink-streaming-java_${scala.version}</artifactId>
            <version>${flink.system.version}</version>
        </dependency>

        <!-- https://mvnrepository.com/artifact/org.apache.flink/flink-clients -->
        <dependency>
            <groupId>org.apache.flink</groupId>
            <artifactId>flink-clients_${scala.version}</artifactId>
            <version>${flink.system.version}</version>
        </dependency>


        <!-- https://mvnrepository.com/artifact/org.apache.flink/flink-streaming-scala -->
        <dependency>
            <groupId>org.apache.flink</groupId>
            <artifactId>flink-streaming-scala_${scala.version}</artifactId>
            <version>${flink.system.version}</version>
        </dependency>

2 使用DataStream API构建

package com.shufang.stream;

import com.shufang.bean.Orders;
import com.shufang.bean.WindowOrderCount;
import com.shufang.func.MyOrderSourceFunction;
import com.shufang.util.MyUtil;
import org.apache.commons.compress.utils.Lists;
import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.api.common.functions.AggregateFunction;
import org.apache.flink.api.common.state.MapState;
import org.apache.flink.api.common.state.MapStateDescriptor;
import org.apache.flink.configuration.Configuration;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.streaming.api.functions.KeyedProcessFunction;
import org.apache.flink.streaming.api.functions.windowing.WindowFunction;
import org.apache.flink.streaming.api.windowing.assigners.SlidingEventTimeWindows;
import org.apache.flink.streaming.api.windowing.time.Time;
import org.apache.flink.streaming.api.windowing.windows.TimeWindow;
import org.apache.flink.util.Collector;
import java.text.SimpleDateFormat;
import java.time.Duration;
import java.util.ArrayList;
import java.util.Comparator;
import java.util.Date;
import java.util.Map;


public class WindowAggrFunction_TopN_Optimize {
    public static void main(String[] args) throws Exception {

        //1 获取执行环境，1.12.0之后默认时间语义是EventTime，但是可以在EventTime mode下明确指定使用processingTime
        StreamExecutionEnvironment env = MyUtil.getStreamEnv();
        env.setParallelism(10);

        // TODO 正则匹配，不以.css|.js|.png|.ico结束，通常可以用来过滤
        String regexpPattern = "^((?!\\.(css|js|png|ico)$).)*&";

        //2 从数据源获取数据，
        SingleOutputStreamOperator<Orders> orderDtlStream = env.addSource(new MyOrderSourceFunction()).assignTimestampsAndWatermarks(
                WatermarkStrategy
                        .<Orders>forBoundedOutOfOrderness(Duration.ofSeconds(20))
                        .withTimestampAssigner((order, timestamp) -> order.getTimestamp())

        );

        orderDtlStream.print("detail");


        //3 主要是统计最近10s钟内不同货币的交易次数，每5s钟更新一次结果输出,找出热门的交易货币，以及排名
        SingleOutputStreamOperator<WindowOrderCount> aggregateStream = orderDtlStream
                .keyBy(order -> order.getCurrency())
                .window(SlidingEventTimeWindows.of(Time.seconds(10), Time.seconds(5)))
                .allowedLateness(Time.minutes(1))  // 当到了窗口的endTime,窗口会输出一个计算结果，但是窗口不会关闭，迟到的数据在一分钟进来都会参与计算并更新结果状态
                .aggregate(new MyOrderAggr(), new MyAllWindowFunction());

        orderDtlStream.print("agg");

        //3.1 要求出每个时间窗口的TopN，我们需要按照窗口分组，按照counts进行排序
        //windowEnd,key,count
        SingleOutputStreamOperator<String> top5Stream = aggregateStream.keyBy(wc -> wc.getWindowEnd())
                .process(new MyHotTopN(5));

        //4 进行输出
        top5Stream.print();


        env.execute("should specify a name");
    }


    /**
     * 定义一个processFunction，每来一次数据就存储State中，最终等到ontimer()的时候触发排序计算操作
     */
    static class MyHotTopN extends KeyedProcessFunction<Long, WindowOrderCount, String> {
        // 定义一个控制TopN 的N的属性
        private Integer topSize;
        private SimpleDateFormat sdf = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss");

        // 定义一个MapState，用来保存每个窗口中的所有的<currency,counts>，最终使用onTimer()触发输出topN
        MapState<String, Long> mapState;

        public MyHotTopN(Integer topSize) {
            this.topSize = topSize;
        }

        // 初始化mapState状态
        @Override
        public void open(Configuration parameters) throws Exception {
            mapState = getRuntimeContext().getMapState(new MapStateDescriptor<String, Long>("mapState", String.class, Long.class));
        }

        // 每来一条数据我们处理一次
        @Override
        public void processElement(WindowOrderCount value, Context ctx, Collector<String> out) throws Exception {
            //1 将信息放入到mapState中
            mapState.put(value.getCurrency(), value.getCounts());
            //2 注册定时器1,等到每个窗口的endTime + 1，触发窗口的输出操作
            ctx.timerService().registerEventTimeTimer(value.getWindowEnd() + 1);
            //3 注册一个定时器，在窗口关闭之后清空该窗口的mapState
            ctx.timerService().registerEventTimeTimer(value.getWindowEnd() + 60 * 1000);

        }

        // 定时器内管理的生命周期的操作
        @Override
        public void onTimer(long timestamp, OnTimerContext ctx, Collector<String> out) throws Exception {
            //1 在process的时候为每个窗口注册了2个定时器，此时先判断是清空状态的定时器，还是输出窗口TopN的定时器
            if (timestamp == ctx.getCurrentKey() + 60 * 1000) {
                // 如果走进来，此时应该触发的定时器是清空的定时器，那么清空窗口的状态，并退出
                mapState.clear();
                return;
            }

            //2 如果走到这里，说明是输出结果的定时器,那么就进行topN的排序并输出结果
            String windowEndString = sdf.format(new Date(timestamp - 1));

            //3 拿到map中的所有的数据，进行排序
            ArrayList<Map.Entry<String, Long>> topNs = Lists.newArrayList(mapState.iterator());
            topNs.sort(new Comparator<Map.Entry<String, Long>>() {
                @Override
                public int compare(Map.Entry<String, Long> o1, Map.Entry<String, Long> o2) {
                    if (o1.getValue() < o2.getValue())
                        return 1;
                    else if (o1.getValue() > o2.getValue())
                        return -1;
                    else
                        return 0;
                }
            });

            //4 最终按照@topsize取topN，为了方便打印好看，以String类型遍历输出
            StringBuilder sb = new StringBuilder("=====================================");
            sb.append("窗口结束时间为：").append(windowEndString).append("\n");
            for (int i = 0; i < Math.min(topNs.size(), topSize); i++) {
                sb.append("当前的货币为：").append(topNs.get(i).getKey()).append(" || ");
                sb.append("当前的货币的在该时间段的交易次数为：").append(topNs.get(i).getValue()).append(" || ");
                sb.append("当前的交以次数排名为：").append(i + 1).append("\n");
            }
            sb.append("=====================================");


            out.collect(sb.toString());


        }
    }


    /**
     * 实现一个增量聚合的窗口函数 agg function ，该类型的窗口函数可以改变输出的类型
     * reduce不能改变输出的类型，输入输出的类型必须保持一致
     * Type parameters:
     * <IN> – 输出的event类型
     * <ACC> – 累加器的类型，每来一条数据更新一次累加器的状态 ，The type of the accumulator (intermediate aggregate state).
     * <OUT> – 最终聚合的结果类型 ，The type of the aggregated result
     */
    static class MyOrderAggr implements AggregateFunction<Orders, Long, Long> {

        // 初始化累加器
        @Override
        public Long createAccumulator() {
            return 0L;
        }

        // 累加器的计算逻辑，来一个event => + 1
        @Override
        public Long add(Orders orders, Long acc) {
            return acc + 1;
        }

        // 获取累加器的值
        @Override
        public Long getResult(Long acc) {
            return acc;
        }

        // 不同的累加器的merge操作
        @Override
        public Long merge(Long aLong, Long acc1) {
            return aLong + acc1;
        }
    }


    /**
     * 定义一个全窗口函数，用来接收agg function的输出的 value类型，
     * Type parameters:
     * <IN> – 从AggFunction的输出类型作为输入类型 The type of the input value.
     * <OUT> – 最终的输出类型，可以随意定义 The type of the output value.
     * <KEY> – keyedStream的key的类型 ，The type of the key.
     * <W> – 这个应用所在的窗口的类型 ，The type of Window that this window function can be applied on.
     */
    static class MyAllWindowFunction implements WindowFunction<Long, WindowOrderCount, String, TimeWindow> {

        @Override
        public void apply(String key, TimeWindow window, Iterable<Long> input, Collector<WindowOrderCount> out) throws Exception {
            Long count = input.iterator().next(); //从累加器获取的统计累加值
            long windowEnd = window.getEnd(); //窗口的标识：这里是窗口的endTime

            //最终返回我们需要的类型WindowOrderCount(windowEnd,currency,counts)
            out.collect(new WindowOrderCount(windowEnd, key, count));
        }
    }
}

3 通过Flink SQL构建

package com.shufang.stream;

import com.shufang.bean.Orders;
import com.shufang.func.MyOrderSourceFunction;
import com.shufang.util.MyUtil;
import org.apache.flink.api.common.eventtime.WatermarkStrategy;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.streaming.api.datastream.DataStream;
import org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator;
import org.apache.flink.streaming.api.environment.StreamExecutionEnvironment;
import org.apache.flink.table.api.Slide;
import org.apache.flink.table.api.Table;
import org.apache.flink.table.api.bridge.java.StreamTableEnvironment;
import org.apache.flink.types.Row;

import java.time.Duration;

import static org.apache.flink.table.api.Expressions.$;
import static org.apache.flink.table.api.Expressions.lit;

public class TableSQLAPi_TopN {
    public static void main(String[] args) throws Exception {
        //1 获取执行环境，1.12.0之后默认时间语义是EventTime，但是可以在EventTime mode下明确指定使用processingTime
        StreamExecutionEnvironment env = MyUtil.getStreamEnv();
        StreamTableEnvironment tableEnv = MyUtil.getBlinkStreamTableEnv();
        env.setParallelism(1);

        /* 这是order中的字段,这是一个pojo类
         * public Long timestamp;
         * public Long amount;
         * public String currency;
         */
        //2 从数据源获取数据:Stream
        SingleOutputStreamOperator<Orders> orderDtlStream = env.addSource(new MyOrderSourceFunction()).assignTimestampsAndWatermarks(
                WatermarkStrategy
                        .<Orders>forBoundedOutOfOrderness(Duration.ofSeconds(20))
                        .withTimestampAssigner((order, timestamp) -> order.getTimestamp())

        );

        //3 由于没有外部的数据源，我们假装从stream中获取数据，这个Expression虽然好看，但是难用啊
        Table orders = tableEnv.fromDataStream(
                orderDtlStream,
                $("currency"), $("amount"), $("timestamp").rowtime().as("ts")
        );

        orders.printSchema();


        //4 获取到我们想要的统计表
        Table windowOrderCounts = orders.window(
                Slide.over(lit(10).seconds()).every(lit(5).seconds()).on($("ts")).as("sw")
        ).groupBy($("sw"), $("currency"))
                .select(
                    $("currency"),
                    $("sw").end().as("windowEnd"),
                    $("amount").count().as("counts")
                );

        //5 根据counts排序,TableAPi不支持rank、row_number以及dense_rank 排序，所以还需要使用SQL来处理
        //createTemporaryView("windowOrderCounts",windowOrderCounts)并不是StreamTableEnv的方法
        //所以我们需要调用：tableEnv.createTemporaryView("windowOrderCounts",stream)来注册表
        // TODO can't use this !!! tableEnv.createTemporaryView("windowOrderCounts",windowOrderCounts);

        //6 将windowOrderCounts转成流，并进行表的注册：windowOrderCounts
        DataStream<Row> stream = tableEnv.toAppendStream(windowOrderCounts, Row.class);
        tableEnv.createTemporaryView("windowOrderCounts", stream);


        /**
         * root
         *  |-- currency: STRING
         *  |-- windowEnd: TIMESTAMP(3)
         *  |-- counts: BIGINT
         */
        //7 使用Over窗口实现排序取TopN
        String sql = "" +
                "SELECT  \n" +
                "   windowEnd, \n" +
                "   currency, \n" +
                "   counts,  \n" +
                "   rn  \n" +
                "FROM ( \n" +
                "   SELECT  \n" +
                "\t  *, \n" +
                "\t  ROW_NUMBER() OVER w AS rn   \n" +
                "   FROM windowOrderCounts  \n" +
                "   WINDOW w AS (PARTITION BY windowEnd ORDER BY counts DESC)  \n" +
                ") tmp_table  \n" +
                "WHERE rn <= 5";

        Table topN = tableEnv.sqlQuery(sql);

        DataStream<Tuple2<Boolean, Row>> topNStream = tableEnv.toRetractStream(topN, Row.class);

        topNStream.print("final top5");

        env.execute("sql top5");

    }
}

SQL代码比DataStream代码要整洁很多，内部帮我们构建了状态。