Flink优化总结

Flink优化

tuples 是否比pojo性能好呢? 是的.

类继承自 Tuple

public class ProvinceEvent extends Tuple3<Long,String,String> {
//    private Long timestamps;
//    private String phonenum;
//    private String province;

    public ProvinceEvent( Long timestamps, String phonenum, String province) {
        super(timestamps,phonenum,province);
        this.f0 = timestamps;
        this.f1 = phonenum;
        this.f2 = province;
    }

    public Long getTimestamps(){
        return this.f0;
    }

    public void setTimestamps(Long timestamps) {
        this.f0 = timestamps;
    }

    public String getPhonenum() {
        return this.f1;
    }

    public void setPhonenum(String phonenum) {
        this.f1 = phonenum;
    }

    public String getProvince() {
        return this.f2;
    }

    public void setProvince(String province) {
        this.f2 = province;
    }
}

2.复用Flink对象

错误示例:

stream
    .apply(new WindowFunction<WikipediaEditEvent, Tuple2<String, Long>, String, TimeWindow>() {
        @Override
        public void apply(String userName, TimeWindow timeWindow, Iterable<WikipediaEditEvent> iterable, Collector<Tuple2<String, Long>> collector) throws Exception {
            long changesCount = ...
            // A new Tuple instance is created on every execution
            collector.collect(new Tuple2<>(userName, changesCount));
        }
    }

可以看出,apply函数每执行一次,都会新建一个Tuple2类的实例,因此增加了对垃圾收集器的压力。解决这个问题的一种方法是反复使用相同的实例:

stream
    .apply(new WindowFunction<WikipediaEditEvent, Tuple2<String, Long>, String, TimeWindow>() {
        // Create an instance that we will reuse on every call
        private Tuple2<String, Long> result = new Tuple<>();

        @Override
        public void apply(String userName, TimeWindow timeWindow, Iterable<WikipediaEditEvent> iterable, Collector<Tuple2<String, Long>> collector) throws Exception {
            long changesCount = ...

            // Set fields on an existing object instead of creating a new one
            result.f0 = userName;
            // Auto-boxing!! A new Long value may be created
            result.f1 = changesCount;

            // Reuse the same Tuple2 object
            collector.collect(result);
        }
    }

这种做法更好一点。虽然每次调用时都新建一个Tuple2的实例,但是其实还间接创建了Long类的实例。为了解决这个问题,Flink有许多所谓的value class:IntValueLongValueStringValueFloatValue等。下面介绍一下如何使用它们:

最优方案:

stream
    .apply(new WindowFunction<WikipediaEditEvent, Tuple2<String, Long>, String, TimeWindow>() {
        // 创建一个可变的计算实例
        private LongValue count = new IntValue();
        // 分配可变的元组
        private Tuple2<String, LongValue> result = new Tuple<>("", count);

        @Override
        // 请注意,现在我们有不同的返回类型
        public void apply(String userName, TimeWindow timeWindow, Iterable<WikipediaEditEvent> iterable, Collector<Tuple2<String, LongValue>> collector) throws Exception {
            long changesCount = ...

            // 设置现有对象上的字段,而不是创建一个新对象
            result.f0 = userName;
            // 设置现有对象上的字段,而不是创建一个新对象
            count.setValue(changesCount);

            // 重用相同的元组,同一个LongValue实例
            // 每次发送出去的对象要一样
            collector.collect(result);
        }
    }

3. 使用注解功能

4.Select Join Type

5 给算子添加uid

6window内数据倾斜

原理:

1.首先将key打散,我们加入将key转化为 key-随机数 ,保证数据散列

2.对打散后的数据进行聚合统计,这时我们会得到数据比如 : (key1-12,1),(key1-13,19),(key1-1,20),(key2-123,11),(key2-123,10)

3.将散列key还原成我们之前传入的key,这时我们的到数据是聚合统计后的结果,不是最初的原数据

4.二次keyby进行结果统计,输出到addSink

 
 
 
import org.apache.flink.api.common.functions.AggregateFunction
import org.apache.flink.api.common.state.{ValueState, ValueStateDescriptor}
import org.apache.flink.api.java.tuple.Tuple
import org.apache.flink.api.scala.typeutils.Types
import org.apache.flink.streaming.api.functions.KeyedProcessFunction
import org.apache.flink.streaming.api.functions.windowing.WindowFunction
import org.apache.flink.streaming.api.scala._
import org.apache.flink.streaming.api.windowing.time.Time
import org.apache.flink.streaming.api.windowing.windows.TimeWindow
import org.apache.flink.util.Collector
 
object ProcessFunctionScalaV2 {
 
 
  def main(args: Array[String]): Unit = {
    val env: StreamExecutionEnvironment = StreamExecutionEnvironment.getExecutionEnvironment
    env.enableCheckpointing(2000)
    val stream: DataStream[String] = env.socketTextStream("localhost", 9999)
    val typeAndData: DataStream[(String, Long)] = stream.map(x => (x.split(",")(0), x.split(",")(1).toLong))
    val dataStream: DataStream[(String, Long)] = typeAndData
      .map(x => (x._1 + "-" + scala.util.Random.nextInt(100), x._2))
    val keyByAgg: DataStream[DataJast] = dataStream.keyBy(_._1)
      .timeWindow(Time.seconds(10))
      .aggregate(new CountAggregate())
    keyByAgg.print("第一次keyby输出")
    val result: DataStream[DataJast] = keyByAgg.map(data => {
      val newKey: String = data.key.substring(0, data.key.indexOf("-"))
      println(newKey)
      DataJast(newKey, data.count)
    }).keyBy(_.key)
      .process(new MyProcessFunction())
    result.print("第二次keyby输出")
 
 
    env.execute()
  }
 
  case class DataJast(key :String,count:Long)
 
  //计算keyby后,每个Window中的数据总和
  class CountAggregate extends AggregateFunction[(String, Long),DataJast, DataJast] {
 
    override def createAccumulator(): DataJast = {
      println("初始化")
      DataJast(null,0)
    }
 
    override def add(value: (String, Long), accumulator: DataJast): DataJast = {
      if(accumulator.key==null){
        printf("第一次加载,key:%s,value:%d\n",value._1,value._2)
        DataJast(value._1,value._2)
      }else{
        printf("数据累加,key:%s,value:%d\n",value._1,accumulator.count+value._2)
        DataJast(value._1,accumulator.count + value._2)
      }
    }
 
    override def getResult(accumulator: DataJast): DataJast = {
      println("返回结果:"+accumulator)
      accumulator
    }
 
    override def merge(a: DataJast, b: DataJast): DataJast = {
      DataJast(a.key,a.count+b.count)
    }
  }
 
 
  /**
   * 实现:
   *    根据key分类,统计每个key进来的数据量,定期统计数量
   */
  class MyProcessFunction extends  KeyedProcessFunction[String,DataJast,DataJast]{
 
    val delayTime : Long = 1000L * 30
 
    lazy val valueState:ValueState[Long] = getRuntimeContext.getState[Long](new ValueStateDescriptor[Long]("ccount",classOf[Long]))
 
    override def processElement(value: DataJast, ctx: KeyedProcessFunction[String, DataJast, DataJast]#Context, out: Collector[DataJast]): Unit = {
      if(valueState.value()==0){
        valueState.update(value.count)
        printf("运行task:%s,第一次初始化数量:%s\n",getRuntimeContext.getIndexOfThisSubtask,value.count)
        val currentTime: Long = ctx.timerService().currentProcessingTime()
        //注册定时器
        ctx.timerService().registerProcessingTimeTimer(currentTime + delayTime)
      }else{
        valueState.update(valueState.value()+value.count)
        printf("运行task:%s,更新统计结果:%s\n" ,getRuntimeContext.getIndexOfThisSubtask,valueState.value())
      }
    }
 
    override def onTimer(timestamp: Long, ctx: KeyedProcessFunction[String, DataJast, DataJast]#OnTimerContext, out: Collector[DataJast]): Unit = {
      //定时器执行,可加入业务操作
      printf("运行task:%s,触发定时器,30秒内数据一共,key:%s,value:%s\n",getRuntimeContext.getIndexOfThisSubtask,ctx.getCurrentKey,valueState.value())
 
      //定时统计完成,初始化统计数据
      valueState.update(0)
      //注册定时器
      val currentTime: Long = ctx.timerService().currentProcessingTime()
      ctx.timerService().registerProcessingTimeTimer(currentTime + delayTime)
    }
  }
 
 
 
}

7 算子优化

1.ReduceFunction,AggregateFunction

在每一个窗口中,增量的计算每一个到达的元素。

就是来一个数据,计算一个数据。

2. WindowFunction,ProcessWindowFunction

先把窗口的数据都缓存下来,然后再一起计算,数据在集合里面。(不够高效)

可以拿到窗口的开始,结束时间。

8 批处理的优化

  • 语义注解

  • 使用StringValue

读取文本文件内容,将文件内容转换成DataSet[StringValue]类型数据集。StringValue是一种可变的String类型,通过StringValue存储文本数据可以有效降低String对象创建数量,从而降低系统性能上的开销。

  • 开启对象重用
env.getConfig().enableObjectReuse();

9去重的优化

使用bitmap

posted @ 2020-11-25 11:32  宁君  阅读(473)  评论(0编辑  收藏  举报