利用Spark监听listener来监控任务完成进度
2019-10-24 本文已影响0人
alexlee666
一、背景
当时在做数据湖的项目,需要使用Spark SQL做数据ETL,即并发地将全表数据从RDBMS经过数据转换等导入到HDFS中。由于Web UI上需要显示ETL的进度,因此需要能够指导当前导了多少个row。但是由于是多个executor并发地读取数据,而如何获取每个executor导了多少个row就是一个问题了,Spark SQL本身并没有提供这样的API。本文将介绍如何使用Spark监听listener来预估任务完成的进度。
二、实现方法
- 首先,自定义一个监听类,并继承SparkListener并override方法;
- 实例化该监听类得到监听器对象,sparkcontex添加该监听器对象即可。
三、业务代码示例
import org.apache.spark.scheduler._
import org.slf4j.LoggerFactory
/*
* This class is used to listen the progress of submitted spark job
* The number of completed tasks will be counted
* In this way, the rough progress of submitted spark job can be estimated
* */
class MySparkListener(instanceName:String,schemaName:String,tableName:String,ceilNum:Long,rowCount:Long,parallelismNum:Int,partitionNum:Int) extends SparkListener{
val logger = LoggerFactory.getLogger(classOf[MySparkListener])
var taskCount: Int = 0
override def onApplicationStart(applicationStart: SparkListenerApplicationStart): Unit = {
super.onApplicationStart(applicationStart)
logger.info("\n\n\n>>>>>> Spark application started")
}
override def onApplicationEnd(applicationEnd: SparkListenerApplicationEnd): Unit = {
super.onApplicationEnd(applicationEnd)
logger.info("\n\n\n>>>>>> Spark application ended")
}
override def onJobEnd(jobEnd: SparkListenerJobEnd): Unit = {
super.onJobEnd(jobEnd)
}
override def onStageCompleted(stageCompleted: SparkListenerStageCompleted): Unit = {
super.onStageCompleted(stageCompleted)
}
override def onTaskEnd(taskEnd: SparkListenerTaskEnd): Unit = {
super.onTaskEnd(taskEnd)
taskCount = taskCount + 1
if(taskCount <= parallelismNum + 1){
val sparkJobProgress = Math.floor(rowCount*(0.1+taskCount*0.76/(parallelismNum+1))).toLong
if(sparkJobProgress <= rowCount){
// 处理逻辑,更新进度......
}
}
}
}
object Main {
def main(args: Array[String]): Unit = {
val sparkSession = SparkSession.builder().master("yarn").appName("Datalake")getOrCreate()
val sc = sparkSession.sparkContext
logger.info(">>>>>> start spark listener")
val sparkListener = new MySparkListener(instanceName,schemaName,tableName,ceilNum,rowCount,parallelismNum,partitionNum)
sc.addSparkListener(sparkListener)
如有错误,敬请指正!