Hadoop API：遍历文件分区目录，并根据目录下的数据进行并行提交spark任务

hadoop api提供了一些遍历文件的api,通过该api可以实现遍历文件目录：

import java.io.FileNotFoundException;
import java.io.IOException;
import java.net.URI;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import java.util.concurrent.CountDownLatch;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;

public class BatchSubmitMain {
    public static void main(String[] args) throws Exception {
        String mrTableName = args[0];
        String fglibTableName = args[1];

        Configuration conf = new Configuration();
        /*
         * <property> <name>fs.defaultFS</name> <value>hdfs://hcluster</value>
         * </property>
         */
        conf.set("fs.defaultFS", "hdfs://hcluster");
        FileSystem fileSystem = FileSystem.get(conf);

        String mrFilePath = "/myuser/hivedb/" + mrTableName;
        String fglibFilePath = "/myuser/hivedb/" + fglibTableName;

        System.out.println(mrFilePath);
        List<String> mrObjectIdItems = getObjectIdItems(fileSystem, mrFilePath);

        System.out.println(fglibFilePath);
        List<String> fglibObjectIdItems = getObjectIdItems(fileSystem, fglibFilePath);

        List<String> objectIdItems = new ArrayList<>();

        for (String mrObjectId : mrObjectIdItems) {
            for (String fglibObjectId : fglibObjectIdItems) {
                if (mrObjectId == fglibObjectId) {
                    objectIdItems.add(mrObjectId);
                }
            }
        }

        String submitShPath = "/app/myaccount/service/submitsparkjob.sh";

        CountDownLatch threadSignal = new CountDownLatch(objectIdItems.size());

        for (int ii = 0; ii < objectIdItems.size(); ii++) {
            String objectId = objectIdItems.get(ii);
            Thread thread = new ImportThread(objectId, submitShPath, threadSignal);
            thread.start();
        }

        threadSignal.await();

        System.out.println(Thread.currentThread().getName() + "complete");
    }

    private static List<String> getObjectIdItems(FileSystem fileSystem, String filePath) throws FileNotFoundException, IOException {
        List<String> objectItems = new ArrayList<>();

        Path path = new Path(filePath);
        // 获取文件列表
        FileStatus[] files = fileSystem.listStatus(path);
        // 展示文件信息
        for (int i = 0; i < files.length; i++) {
            try {
                if (files[i].isDirectory()) {
                    String[] fileItems = files[i].getPath().getName().split("/");
                    String objectId = fileItems[fileItems.length - 1].replace("objectid=", "");
                    objectItems.add(objectId);
                    System.out.println(objectId);
                }
            } catch (Exception e) {
                e.printStackTrace();
            }
        }

        return objectItems;
    }

    /**
     * @param hdfs
     *            FileSystem 对象
     * @param path
     *            文件路径
     */
    public static void iteratorShowFiles(FileSystem hdfs, Path path) {
        try {
            if (hdfs == null || path == null) {
                return;
            }

            // 获取文件列表
            FileStatus[] files = hdfs.listStatus(path);

            // 展示文件信息
            for (int i = 0; i < files.length; i++) {
                try {
                    if (files[i].isDirectory()) {
                        System.out.print(">>>" + files[i].getPath() + ", dir owner:" + files[i].getOwner());
                        // 递归调用
                        iteratorShowFiles(hdfs, files[i].getPath());
                    } else if (files[i].isFile()) {
                        System.out.print(" " + files[i].getPath() + ",length:" + files[i].getLen() + ", owner:" + files[i].getOwner());
                    }
                } catch (Exception e) {
                    e.printStackTrace();
                }
            }
        } catch (Exception e) {
            e.printStackTrace();
        }
    }

}

并行执行sh的线程：

import java.util.concurrent.CountDownLatch;

public class ImportThread extends Thread {
    private final JavaShellInvoker javaShellInvoker = new JavaShellInvoker();

    private CountDownLatch countDownLatch;
    private String objectId;
    private String submitShPath;

    public ImportThread(String objectId, String submitShPath, CountDownLatch countDownLatch) {
        this.objectId = objectId;
        this.submitShPath = submitShPath;
        this.countDownLatch = countDownLatch;
    }

    @Override
    public void run() {
        System.out.println(Thread.currentThread().getName() + "start... " + this.submitShPath + " " + this.objectId.toString());// 打印开始标记

        try {
            int result = this.javaShellInvoker.executeShell("mrraster", this.submitShPath, this.objectId);
            if (result != 0) {
                System.out.println(Thread.currentThread().getName() + " result type is error");
            }
        } catch (Exception e) {
            e.printStackTrace();
            System.out.println(Thread.currentThread().getName() + "-error:" + e.getMessage());
        }

        this.countDownLatch.countDown();// 计时器减1
        System.out.println(Thread.currentThread().getName() + " complete,last " + this.countDownLatch.getCount() + " threads");// 打印结束标记
    }
}

执行sh的java代码：

import java.io.File;
import java.text.SimpleDateFormat;
import java.util.Date;

public class JavaShellInvoker {
    private static final String executeShellLogFile = "./executeShell_%s_%s.log";

    public int executeShell(String shellCommandType, String shellCommand, String args) throws Exception {
        int success = 0;

        args = (args == null) ? "" : args;

        String now = new SimpleDateFormat("yyyy-MM-dd").format(new Date());
        File logFile = new File(String.format(executeShellLogFile, shellCommandType, now));

        ProcessBuilder pb = new ProcessBuilder("sh", shellCommand, args);
        pb.redirectOutput(ProcessBuilder.Redirect.appendTo(logFile));
        pb.redirectError(ProcessBuilder.Redirect.appendTo(logFile));

        Process pid = null;

        try {
            pid = pb.start();
            success = pid.waitFor();
        } catch (Exception ex) {
            success = 2;
            System.out.println("executeShell-error:" + ex.getMessage());
            throw ex;
        } finally {
            if (pid.isAlive()) {
                success = pid.exitValue();
                pid.destroy();
            }
        }

        return success;
    }
}

submitsparkjob.sh

#!/bin/sh
source ../login.sh
spark-submit --master yarn-cluster --class MySparkJobMainClass --driver-class-path /app/myaccount/service/jars/ojdbc7.jar --jars /app/myaccount/service/jars/ojdbc7.jar --num-executors
 20 --driver-memory 6g --executor-cores 1 --executor-memory 8g MySparkJobJar.jar $1

执行BatchSubmit.jar的命令：

hadoop jar BatchSubmit.jar