监控 AWS EMR 作业运行进度的最佳实践是什么？答案

【问题标题】：What is the best practice to monitor AWS EMR job running progress?监控 AWS EMR 作业运行进度的最佳实践是什么？
【发布时间】：2015-02-06 11:25:54
【问题描述】：

我有以下代码来运行 EMR 作业，并且它运行成功。而且我还想监控运行状态。我使用DescribeJobFlows API，但它说根据http://docs.aws.amazon.com/AWSJavaSDK/latest/javadoc/com/amazonaws/services/elasticmapreduce/AmazonElasticMapReduceClient.html，该API已被弃用。

有人可以帮忙了解监控 EMR 运行进度的最佳做法吗？

public class EmrJobRunner {
  public static void main(String[] args) {
    // args is [input_file_path, output_directory], make sure output_directory does not exist
    String inputFilePath = "s3://mybucket/emr/input";
    String outputDirectory = "s3://mybucket/emr/output/" + System.currentTimeMillis();
    String jarName = "WordCount.jar";
    String jarPath = "s3://mybucket/emr/" + jarName;
    String logPath = "s3://mybucket/emr/logs";

    String TERMINATE_JOB_FLOW = "TERMINATE_JOB_FLOW";
    String CONTINUE = "CONTINUE";

    AWSCredentials credentials = new BasicAWSCredentials("pub_key", "sec_key");
    StepFactory stepFactory = new StepFactory();

    AmazonElasticMapReduce emr = new AmazonElasticMapReduceClient(credentials);
    emr.setRegion(Region.getRegion(Regions.AP_SOUTHEAST_1));

    StepConfig enableDebugging = new StepConfig()
      .withName("Enable debugging")
      .withActionOnFailure(TERMINATE_JOB_FLOW)
      .withHadoopJarStep(stepFactory.newEnableDebuggingStep());

    StepConfig installHive = new StepConfig()
      .withName("Install Hive")
      .withActionOnFailure(TERMINATE_JOB_FLOW)
      .withHadoopJarStep(stepFactory.newInstallHiveStep());

    StepConfig runScript = new StepConfig()
      .withName("Run Script")
      .withActionOnFailure(CONTINUE)
      .withHadoopJarStep(stepFactory.newRunHiveScriptStep("s3://dummy/dummy.hive"));

    List<String> jarArgs = Arrays.asList(inputFilePath, outputDirectory);
    HadoopJarStepConfig jarCfg= new HadoopJarStepConfig()
      .withJar(jarPath)
      .withArgs(jarArgs);
    StepConfig runJar = new StepConfig()
      .withName(jarName)
      .withActionOnFailure(TERMINATE_JOB_FLOW)
      .withHadoopJarStep(jarCfg);

    JobFlowInstancesConfig instanceCfg = new JobFlowInstancesConfig()
      .withKeepJobFlowAliveWhenNoSteps(false)
      .withTerminationProtected(true)
      .withInstanceCount(3)
      .withMasterInstanceType(InstanceType.C1Medium.toString())
      .withSlaveInstanceType(InstanceType.C1Medium.toString())
      .withHadoopVersion("2.4.0");

    List<StepConfig> steps = Arrays.asList(enableDebugging, installHive, runScript, runJar);

    RunJobFlowRequest request = new RunJobFlowRequest()
      .withName("My EMR Job Flow")
      .withAmiVersion("3.3.2")
      .withInstances(instanceCfg)
      .withLogUri(logPath);
      .withSteps(steps);

    RunJobFlowResult result = emr.runJobFlow(request);
    // saying DescribeJobFlows is deprecated
    // DescribeJobFlowsResult jobFlowDescResult = emr.DescribeJobFlows(DescribeJobFlowsRequest describeJobFlowsRequest);
  }

}

【问题讨论】：

标签： java amazon-web-services emr elastic-map-reduce amazon-emr

【解决方案1】：

由于 DescribeJobFlows 已弃用，因此监控集群状态是监控作业运行进度的另一种方法。

    RunJobFlowResult runJobResult = emr.runJobFlow(runJobFlowRequest);
    System.out.printf("Run JobFlowId is: %s\n", runJobResult.getJobFlowId());

    while(true) {
      DescribeClusterRequest desc = new DescribeClusterRequest()
        .withClusterId(runJobResult.getJobFlowId());
      DescribeClusterResult clusterResult = emr.describeCluster(desc);
      Cluster cluster = clusterResult.getCluster();
      String status = cluster.getStatus().getState();
      System.out.printf("Status: %s\n", status);
      if(status.equals(ClusterState.TERMINATED.toString()) || status.equals(ClusterState.TERMINATED_WITH_ERRORS.toString())) {
        break;
      }
      try {
        TimeUnit.SECONDS.sleep(30);
      } catch (InterruptedException e) {
        e.printStackTrace();
      }
      // maybe other handle
    }

【讨论】：