spark提交任务的三种的方法

Posted 2021-01-21 itboys

在学习Spark过程中，资料中介绍的提交Spark Job的方式主要有三种：

第一种:

   通过命令行的方式提交Job,使用spark 自带的spark-submit工具提交，官网和大多数参考资料都是已这种方式提交的，提交命令示例如下：
./spark-submit --class com.learn.spark.SimpleApp --master yarn --deploy-mode client --driver-memory 2g --executor-memory 2g --executor-cores 3 ../spark-demo.jar
参数含义就不解释了，请参考官网资料。
 第二种:

   提交方式是已JAVA API编程的方式提交，这种方式不需要使用命令行，直接可以在IDEA中点击Run 运行包含Job的Main类就行，Spark 提供了以SparkLanuncher 作为唯一入口的API来实现。这种方式很方便（试想如果某个任务需要重复执行，但是又不会写linux 脚本怎么搞？我想到的是以JAV API的方式提交Job, 还可以和Spring整合，让应用在tomcat中运行），官网的示例：http://spark.apache.org/docs/latest/api/java/index.html?org/apache/spark/launcher/package-summary.html

根据官网的示例，通过JAVA API编程的方式提交有两种方式：

        第一种是调用SparkLanuncher实例的startApplication方法，但是这种方式在所有配置都正确的情况下使用运行都会失败的，原因是startApplication方法会调用LauncherServer启动一个进程与集群交互，这个操作貌似是异步的，所以可能结果是main主线程结束了这个进程都没有起起来，导致运行失败。解决办法是调用new SparkLanuncher().startApplication后需要让主线程休眠一定的时间后者是使用下面的例子：

package com.learn.spark; 

import org.apache.spark.launcher.SparkAppHandle; 
import org.apache.spark.launcher.SparkLauncher; 

import java.io.IOException; 
import java.util.HashMap; 
import java.util.concurrent.CountDownLatch; 

public class LanuncherAppV { 
    public static void main(String[] args) throws IOException, InterruptedException { 

        HashMap env = new HashMap(); 
        //这两个属性必须设置 
        env.put("HADOOP_CONF_DIR", "/usr/local/hadoop/etc/overriterHaoopConf"); 
        env.put("JAVA_HOME", "/usr/local/java/jdk1.8.0_151"); 
        //可以不设置 
        //env.put("YARN_CONF_DIR",""); 
        CountDownLatch countDownLatch = new CountDownLatch(1); 
        //这里调用setJavaHome()方法后，JAVA_HOME is not set 错误依然存在 
        SparkAppHandle handle = new SparkLauncher(env) 
        .setSparkHome("/usr/local/spark") 
        .setAppResource("/usr/local/spark/spark-demo.jar") 
        .setMainClass("com.learn.spark.SimpleApp") 
        .setMaster("yarn") 
        .setDeployMode("cluster") 
        .setConf("spark.app.id", "11222") 
        .setConf("spark.driver.memory", "2g") 
        .setConf("spark.akka.frameSize", "200") 
        .setConf("spark.executor.memory", "1g") 
        .setConf("spark.executor.instances", "32") 
        .setConf("spark.executor.cores", "3") 
        .setConf("spark.default.parallelism", "10") 
        .setConf("spark.driver.allowMultipleContexts", "true") 
        .setVerbose(true).startApplication(new SparkAppHandle.Listener() { 
        //这里监听任务状态，当任务结束时（不管是什么原因结束）,isFinal（）方法会返回true,否则返回false 
         @Override 
        public void stateChanged(SparkAppHandle sparkAppHandle) { 
            if (sparkAppHandle.getState().isFinal()) { 
                countDownLatch.countDown(); 
            } 
            System.out.println("state:" + sparkAppHandle.getState().toString()); 
        } 

        @Override 
        public void infoChanged(SparkAppHandle sparkAppHandle) { 
            System.out.println("Info:" + sparkAppHandle.getState().toString()); 
        } 
    }); 
    System.out.println("The task is executing, please wait ...."); 
    //线程等待任务结束 
    countDownLatch.await(); 
    System.out.println("The task is finished!"); 


    } 
}

 注意：如果部署模式是cluster,但是代码中有标准输出的话将看不到，需要把结果写到HDFS中，如果是client模式则可以看到输出。

第二种方式是：通过SparkLanuncher.lanunch()方法获取一个进程，然后调用进程的process.waitFor()方法等待线程返回结果，但是使用这种方式需要自己管理运行过程中的输出信息，比较麻烦，好处是一切都在掌握之中，即获取的输出信息和通过命令提交的方式一样，很详细，实现如下：

package com.learn.spark; 

import org.apache.spark.launcher.SparkAppHandle; 
import org.apache.spark.launcher.SparkLauncher; 

import java.io.IOException; 
import java.util.HashMap; 

public class LauncherApp { 

public static void main(String[] args) throws IOException, InterruptedException { 

    HashMap env = new HashMap(); 
    //这两个属性必须设置 
    env.put("HADOOP_CONF_DIR","/usr/local/hadoop/etc/overriterHaoopConf"); 
    env.put("JAVA_HOME","/usr/local/java/jdk1.8.0_151"); 
    //env.put("YARN_CONF_DIR",""); 

    SparkLauncher handle = new SparkLauncher(env) 
        .setSparkHome("/usr/local/spark") 
        .setAppResource("/usr/local/spark/spark-demo.jar") 
        .setMainClass("com.learn.spark.SimpleApp") 
        .setMaster("yarn") 
        .setDeployMode("cluster") 
        .setConf("spark.app.id", "11222") 
        .setConf("spark.driver.memory", "2g") 
        .setConf("spark.akka.frameSize", "200") 
        .setConf("spark.executor.memory", "1g") 
        .setConf("spark.executor.instances", "32") 
        .setConf("spark.executor.cores", "3") 
        .setConf("spark.default.parallelism", "10") 
        .setConf("spark.driver.allowMultipleContexts","true") 
        .setVerbose(true); 


    Process process =handle.launch(); 
    InputStreamReaderRunnable inputStreamReaderRunnable = new InputStreamReaderRunnable(process.getInputStream(), "input"); 
    Thread inputThread = new Thread(inputStreamReaderRunnable, "LogStreamReader input"); 
    inputThread.start(); 

    InputStreamReaderRunnable errorStreamReaderRunnable = new InputStreamReaderRunnable(process.getErrorStream(), "error"); 
    Thread errorThread = new Thread(errorStreamReaderRunnable, "LogStreamReader error"); 
    errorThread.start(); 

    System.out.println("Waiting for finish..."); 
    int exitCode = process.waitFor(); 
    System.out.println("Finished! Exit code:" + exitCode); 

    } 
}

使用的自定义InputStreamReaderRunnable类实现如下：

package com.learn.spark; 

import java.io.BufferedReader; 
import java.io.IOException; 
import java.io.InputStream; 
import java.io.InputStreamReader; 

public class InputStreamReaderRunnable implements Runnable { 

　　private BufferedReader reader; 

　　private String name; 

　　public InputStreamReaderRunnable(InputStream is, String name) { 
　　　　this.reader = new BufferedReader(new InputStreamReader(is)); 
　　　　this.name = name; 
　　} 

　　public void run() {
 
　　　　System.out.println("InputStream " + name + ":"); 
　　　　try { 
　　　　　　　　String line = reader.readLine(); 
　　　　　　　　while (line != null) { 
　　　　　　　　　　　System.out.println(line); 
　　　　　　　　　　　line = reader.readLine(); 
　　　　　　　　} 
　　　　　　　　reader.close(); 
　　　 　　} catch (IOException e) { 
　　　　　 　　e.printStackTrace(); 
　　　　　　} 
　 　} 
}

第三种方式是通过yarn的rest api的方式提交(不太常用但在这里也介绍一下)：

Post请求示例： * http://<rm http address:port>/ws/v1/cluster/apps

请求所带的参数列表：

Item	Data Type	Description
application-id	string	The application id
application-name	string	The application name
queue	string	The name of the queue to which the application should be submitted
priority	int	The priority of the application
am-container-spec	object	The application master container launch context, described below
unmanaged-AM	boolean	Is the application using an unmanaged application master
max-app-attempts	int	The max number of attempts for this application
resource	object	The resources the application master requires, described below
application-type	string	The application type(MapReduce, Pig, Hive, etc)
keep-containers-across-application-attempts	boolean	Should YARN keep the containers used by this application instead of destroying them
application-tags	object	List of application tags, please see the request examples on how to speciy the tags
log-aggregation-context	object	Represents all of the information needed by the NodeManager to handle the logs for this application
attempt-failures-validity-interval	long	The failure number will no take attempt failures which happen out of the validityInterval into failure count
reservation-id	string	Represent the unique id of the corresponding reserved resource allocation in the scheduler
am-black-listing-requests	object	Contains blacklisting information such as “enable/disable AM blacklisting” and “disable failure threshold”