Spark中的多线程并发处理

Spark中的多任务处理

Spark的一个非常常见的用例是并行运行许多作业。 构建作业DAG后，Spark将这些任务分配到多个Executor上并行处理。
但这并不能帮助我们在同一个Spark应用程序中同时运行两个完全独立的作业，例如同时从多个数据源读取数据并将它们写到对应的存储，或同时处理多个文件等。

每个spark应用程序都需要一个SparkSession(Context)来配置和执行操作。 SparkSession对象是线程安全的，可以根据需要传递给你的Spark应用程序。

顺序执行的例子

import org.apache.spark.sql.SparkSession

object FancyApp {
  def def appMain(args: Array[String]) = {
    // configure spark
    val spark = SparkSession
        .builder
        .appName("parjobs")
        .getOrCreate()

    val df = spark.sparkContext.parallelize( to ).toDF
    doFancyDistinct(df, "hdfs:///dis.parquet")
    doFancySum(df, "hdfs:///sum.parquet")
  }

  def doFancyDistinct(df: DataFrame, outPath: String) = df.distinct.write.parquet(outPath)

  def doFancySum(df: DataFrame, outPath: String) = df.agg(sum("value")).write.parquet(outPath)

}

优化后的例子

import org.apache.spark.sql.SparkSession
import import java.util.concurrent.Executors
import scala.concurrent._
import scala.concurrent.duration._

object FancyApp {
  def def appMain(args: Array[String]) = {
    // configure spark
    val spark = SparkSession
        .builder
        .appName("parjobs")
        .getOrCreate()

    // Set number of threads via a configuration property
    val pool = Executors.newFixedThreadPool()
    // create the implicit ExecutionContext based on our thread pool
    implicit val xc = ExecutionContext.fromExecutorService(pool)
    val df = spark.sparkContext.parallelize( to ).toDF
    val taskA = doFancyDistinct(df, "hdfs:///dis.parquet")
    val taskB = doFancySum(df, "hdfs:///sum.parquet")
    // Now wait for the tasks to finish before exiting the app
    Await.result(Future.sequence(Seq(taskA,taskB)), Duration(, MINUTES))
  }

  def doFancyDistinct(df: DataFrame, outPath: String)(implicit xc: ExecutionContext) = Future {
    df.distinct.write.parquet(outPath)
  }

  def doFancySum(df: DataFrame, outPath: String)(implicit xc: ExecutionContext) = Future {
    df.agg(sum("value")).write.parquet(outPath)
  }
}

java 实现例子

    val executors = Executors.newFixedThreadPool(threadPoolNum)
    val completionService = new ExecutorCompletionService[String](executors)
    for ((branch_id, dataList) <- summary) {
      logInfo(s"************** applicationId is ${applicationId} about Multi-threading starting: file is ${branch_id}")
      completionService.submit(new Callable[String] {
        override def call(): String = {
          new VerificationTest(spark, branch_id, dataList, separator).runJob()
          branch_id
        }
      })
    }