Spark延长SparkContext初始化时间

有些应用中可能希望先在driver上运行一段java单机程序，然后再初始化SparkContext用集群模式操作java程序返回值。从而避免过早建立SparkContext对象分配集群资源，使资源长时间空闲。

这里涉及到两个yarn参数：

  <property> 
    <name>yarn.am.liveness-monitor.expiry-interval-ms</name>  
    <value>6000000</value> 
  </property>
   <property> 
    <name>yarn.resourcemanager.am.max-retries</name>  
    <value>10</value> 
  </property>

Yarn会周期性遍历所有的ApplicationMaster，如果一个ApplicationMaster在一定时间（可通过参数yarn.am.liveness-monitor.expiry-interval-ms配置，默认为10min）内未汇报心跳信息，则认为它死掉了，它上面所有正在运行的Container将被置为运行失败（RM不会重新执行这些Container，它只会通过心跳机制告诉对应的AM，由AM决定是否重新执行，如果需要，则AM重新向RM申请资源），AM本身会被重新分配到另外一个节点上（管理员可通过参数yarn.resourcemanager.am.max-retries指定每个ApplicationMaster的尝试次数，默认是1次）执行。

还需要两个spark参数：

<property> 
    <name>spark.yarn.am.waitTime</name>  
    <value>6000000</value> 
  </property>
   <property> 
    <name>spark.yarn.applicationMaster.waitTries</name>  
    <value>200</value> 
  </property>

集群管理

Spark On YARN

属性名称	默认值	含义
spark.yarn.scheduler.heartbeat.interval-ms	5000	Spark AppMaster发送心跳信息给YARN RM的时间间隔
spark.yarn.am.waitTime	100000	启动时等待时间
spark.yarn.applicationMaster.waitTries	10	RM等待Spark AppMaster启动重试次数，也就是SparkContext初始化次数。超过这个数值，启动失败

下面是一个测试用例，现在driver打印30分钟的信息，然后再初始化SparkContext

import iie.udps.common.hcatalog.SerHCatInputFormat;
import iie.udps.common.hcatalog.SerHCatOutputFormat;
import java.io.IOException;
import java.util.HashMap;
import java.util.List;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.io.NullWritable;
import org.apache.hadoop.io.WritableComparable;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hive.hcatalog.data.DefaultHCatRecord;
import org.apache.hive.hcatalog.data.HCatRecord;
import org.apache.hive.hcatalog.data.schema.HCatSchema;
import org.apache.hive.hcatalog.mapreduce.OutputJobInfo;
import org.apache.spark.SerializableWritable;
import org.apache.spark.SparkConf;
import org.apache.spark.api.java.JavaPairRDD;
import org.apache.spark.api.java.JavaRDD;
import org.apache.spark.api.java.JavaSparkContext;
import org.apache.spark.api.java.function.Function;
import org.apache.spark.api.java.function.PairFunction;
import org.apache.spark.api.java.function.Function2;
import scala.Tuple2;

/**
 * 实现功能：首先在driver上单机打印30分钟数据，然后初始化SparkContext开启集群模式，用spark+hcatlog 读hive表数据，实现GroupByAge功能，
 * 输出结果到hive表中，同时打印xml信息到hdfs文件。
 * spark-submit --class iie.udps.example.spark.SparkTest --master yarn-cluster 
 * --num-executors 2 --executor-memory 1g --executor-cores 1 --driver-memory 1g 
 * --conf spark.yarn.applicationMaster.waitTries=200,--conf spark.yarn.am.waitTime=1800000 --jars /home/xdf/udps-sdk-0.3.jar,/home/xdf/udps-sdk-0.3.jar
 *  /home/xdf/sparktest.jar -c /user/hdfs/TestStdin2.xml
 */
public class SparkTest {

	@SuppressWarnings("rawtypes")
	public static void main(String[] args) throws Exception {
		if (args.length < 2) {
			System.err.println("Usage: <-c> <stdin.xml>");
			System.exit(1);
		}
		
		String stdinXml = args[1];
		OperatorParamXml operXML = new OperatorParamXml();
		List<java.util.Map> stdinList = operXML.parseStdinXml(stdinXml);// 参数列表

		// 获得输入参数
		String inputDBName = stdinList.get(0).get("inputDBName").toString();
		String inputTabName = stdinList.get(0).get("inputTabName").toString();
		String outputDBName = stdinList.get(0).get("outputDBName").toString();
		String outputTabName = stdinList.get(0).get("outputTabName").toString();
		String tempHdfsBasePath = stdinList.get(0).get("tempHdfsBasePath")
				.toString();
		String jobinstanceid = stdinList.get(0).get("jobinstanceid").toString();
		
		System.out.println(inputDBName+": "+ inputTabName +": "+outputDBName+": "+ outputTabName
				+": "+ tempHdfsBasePath+": "+ jobinstanceid);

		long begin = System.currentTimeMillis(); 
		int count = 600;// 写文件行数
		for (int i = 0; i < count; i++) {
			System.out.println("aaaaaaaaaaaaaaa"+i);
			Thread.sleep(3000);
		}
		long end = System.currentTimeMillis();   
        System.out.println("FileOutputStream执行耗时:" + (end - begin) + "ms");   
		
		
		if (inputDBName == "" || inputTabName == "" || jobinstanceid == ""
				|| outputDBName == "" || outputTabName == ""
				|| tempHdfsBasePath == "" || jobinstanceid == "") {

			// 设置异常输出参数
			java.util.Map<String, String> stderrMap = new HashMap<String, String>();
			String errorMessage = "Some operating parameters is empty!!!";
			String errotCode = "80001";
			stderrMap.put("errorMessage", errorMessage);
			stderrMap.put("errotCode", errotCode);
			stderrMap.put("jobinstanceid", jobinstanceid);
			String fileName = "";
			if (tempHdfsBasePath.endsWith("/")) {
				fileName = tempHdfsBasePath + "stderr.xml";
			} else {
				fileName = tempHdfsBasePath + "/stderr.xml";
			}
			
			// 生成异常输出文件
			operXML.genStderrXml(fileName, stderrMap);
		} else {			
			// 根据输入表结构，创建与输入表同样结构的输出表
			HCatSchema schema = operXML
					.getHCatSchema(inputDBName, inputTabName);

			// Spark程序第一件事情就是创建一个JavaSparkContext告诉Spark怎么连接集群
			SparkConf sparkConf = new SparkConf().setAppName("SparkExample");
			
			JavaSparkContext jsc = new JavaSparkContext(sparkConf);
			
			// 读取并处理hive表中的数据，生成RDD数据并处理后返回
			JavaRDD<SerializableWritable<HCatRecord>> LastRDD = getProcessedData(
					jsc, inputDBName, inputTabName, schema);
			
			// 将处理后的数据存到hive输出表中
			storeToTable(LastRDD, outputDBName, outputTabName);

			jsc.stop();

			// 设置正常输出参数
			java.util.Map<String, String> stdoutMap = new HashMap<String, String>();
			stdoutMap.put("outputDBName", outputDBName);
			stdoutMap.put("outputTabName", outputTabName);
			stdoutMap.put("jobinstanceid", jobinstanceid);
			String fileName = "";
			if (tempHdfsBasePath.endsWith("/")) {
				fileName = tempHdfsBasePath + "stdout.xml";
			} else {
				fileName = tempHdfsBasePath + "/stdout.xml";
			}
			
			// 生成正常输出文件
			operXML.genStdoutXml(fileName, stdoutMap);
		}
		System.out.println(inputDBName+": "+ inputTabName +": "+outputDBName+": "+ outputTabName
				+": "+ tempHdfsBasePath+": "+ jobinstanceid);
		System.exit(0);
	}

	/**
	 * 
	 * @param jsc
	 * @param dbName
	 * @param inputTable
	 * @param fieldPosition
	 * @return
	 * @throws IOException
	 */
	@SuppressWarnings("rawtypes")
	public static JavaRDD<SerializableWritable<HCatRecord>> getProcessedData(
			JavaSparkContext jsc, String dbName, String inputTable,
			final HCatSchema schema) throws IOException {
		// 获取hive表数据
		Configuration inputConf = new Configuration();
		Job job = Job.getInstance(inputConf);
		SerHCatInputFormat.setInput(job.getConfiguration(), dbName, inputTable);
		JavaPairRDD<WritableComparable, SerializableWritable> rdd = jsc
				.newAPIHadoopRDD(job.getConfiguration(),
						SerHCatInputFormat.class, WritableComparable.class,
						SerializableWritable.class);

		// 获取表记录集
		JavaPairRDD<Integer, Integer> pairs = rdd
				.mapToPair(new PairFunction<Tuple2<WritableComparable, SerializableWritable>, Integer, Integer>() {
					private static final long serialVersionUID = 1L;

					@SuppressWarnings("unchecked")
					@Override
					public Tuple2<Integer, Integer> call(
							Tuple2<WritableComparable, SerializableWritable> value)
							throws Exception {
						HCatRecord record = (HCatRecord) value._2.value();
						return new Tuple2((Integer) record.get(1), 1);
					}
				});

		JavaPairRDD<Integer, Integer> counts = pairs
				.reduceByKey(new Function2<Integer, Integer, Integer>() {
					private static final long serialVersionUID = 1L;

					@Override
					public Integer call(Integer i1, Integer i2) {
						return i1 + i2;
					}
				});

		JavaRDD<SerializableWritable<HCatRecord>> messageRDD = counts
				.map(new Function<Tuple2<Integer, Integer>, SerializableWritable<HCatRecord>>() {
					private static final long serialVersionUID = 1L;

					@Override
					public SerializableWritable<HCatRecord> call(
							Tuple2<Integer, Integer> arg0) throws Exception {
						HCatRecord record = new DefaultHCatRecord(2);
						record.set(0, arg0._1);
						record.set(1, arg0._2);
						return new SerializableWritable<HCatRecord>(record);
					}
				});
		// 返回处理后的数据
		return messageRDD;
	}

	/**
	 * 将处理后的数据存到输出表中
	 * 
	 * @param rdd
	 * @param dbName
	 * @param tblName
	 */
	@SuppressWarnings("rawtypes")
	public static void storeToTable(
			JavaRDD<SerializableWritable<HCatRecord>> rdd, String dbName,
			String tblName) {
		Job outputJob = null;
		try {
			outputJob = Job.getInstance();
			outputJob.setJobName("SparkExample");
			outputJob.setOutputFormatClass(SerHCatOutputFormat.class);
			outputJob.setOutputKeyClass(WritableComparable.class);
			outputJob.setOutputValueClass(SerializableWritable.class);
			SerHCatOutputFormat.setOutput(outputJob,
					OutputJobInfo.create(dbName, tblName, null));
			HCatSchema schema = SerHCatOutputFormat
					.getTableSchemaWithPart(outputJob.getConfiguration());
			SerHCatOutputFormat.setSchema(outputJob, schema);
		} catch (IOException e) {
			e.printStackTrace();
		}

		// 将RDD存储到目标表中
		rdd.mapToPair(
				new PairFunction<SerializableWritable<HCatRecord>, WritableComparable, SerializableWritable<HCatRecord>>() {
					private static final long serialVersionUID = -4658431554556766962L;

					public Tuple2<WritableComparable, SerializableWritable<HCatRecord>> call(
							SerializableWritable<HCatRecord> record)
							throws Exception {
						return new Tuple2<WritableComparable, SerializableWritable<HCatRecord>>(
								NullWritable.get(), record);
					}
				}).saveAsNewAPIHadoopDataset(outputJob.getConfiguration());

	}
	

}
 

输入表数据：

hive> select * from test_in; 
OK
120
220
321
420
521
620
721
819
919
1021

输出表数据：

hive> select * from test_out;
OK
192
214
204