八、MapReduce–job提交源码分析

一、源码分析

1、提交job的入口

通过 job.waitForCompletion(true)完成job的提交以及运行，下面从这个方法入手分析源码。

//-----------------job.java
public boolean waitForCompletion(boolean verbose) throws IOException, InterruptedException, ClassNotFoundException {
    //如果job的状态为未运行，则提交任务
    if (this.state == Job.JobState.DEFINE) {
        this.submit();
    }

    if (verbose) {
        //监控并打印运行信息
        this.monitorAndPrintJob();
    } else {
        int completionPollIntervalMillis = getCompletionPollInterval(this.cluster.getConf());

        while(!this.isComplete()) {
            try {
                Thread.sleep((long)completionPollIntervalMillis);
            } catch (InterruptedException var4) {
            }
        }
    }

    return this.isSuccessful();
}

2、this.submit() 提交job

//-----------------job.java
public void submit() throws IOException, InterruptedException, ClassNotFoundException {
    //确定job状态为未运行
    this.ensureState(Job.JobState.DEFINE);
    //使用新api
    this.setUseNewAPI();
    //主要就是初始化cluster对象中的client,用于和集群连接通信。分为yarn client和local client
    this.connect();
    //通过cluster对象获取job提交器，将存储job信息的文件系统以及client作为参数
    final JobSubmitter submitter = this.getJobSubmitter(this.cluster.getFileSystem(), this.cluster.getClient());
    //提交job，并运行
    this.status = (JobStatus)this.ugi.doAs(new PrivilegedExceptionAction<JobStatus>() {
        public JobStatus run() throws IOException, InterruptedException, ClassNotFoundException {
            //这里是提交job，运行，返回状态
            return submitter.submitJobInternal(Job.this, Job.this.cluster);
        }
    });
    this.state = Job.JobState.RUNNING;
    LOG.info("The url to track the job: " + this.getTrackingURL());
}

上面这里涉及到三个重要过程方法：
this.connect() 主要初始化了提交job的client
this.getJobSubmitter() 给job封装了很多api
submitter.submitJobInternal(Job.this, Job.this.cluster) 提交job，并运行
下面详细看看这三个方法具体做了啥

3、this.connect()初始化client

//-----------------job.java
private synchronized void connect() throws IOException, InterruptedException, ClassNotFoundException {
    //创建cluster连接对象,用于连接集群,提供了很多api
    if (this.cluster == null) {
        this.cluster = (Cluster)this.ugi.doAs(new PrivilegedExceptionAction<Cluster>() {
            public Cluster run() throws IOException, InterruptedException, ClassNotFoundException {
                return new Cluster(Job.this.getConfiguration());
            }
        });
    }

}

这代码最重要的就是创建了一个 Cluster对象，下面看看这个类的构造方法。

//----------------------------Cluster.java
public Cluster(Configuration conf) throws IOException {
    this((InetSocketAddress)null, conf);
}

public Cluster(InetSocketAddress jobTrackAddr, Configuration conf) throws IOException {
    this.fs = null;
    this.sysDir = null;
    //job工作目录
    this.stagingAreaDir = null;
    this.jobHistoryDir = null;
    //客户端和server通信协议提供者
    this.providerList = null;
    //将job的配置conf保存
    this.conf = conf;
    //获取当前用户
    this.ugi = UserGroupInformation.getCurrentUser();
    //对job提交器client进行初始化
    this.initialize(jobTrackAddr, conf);
}

//这里就是初始化client的方法了，主要就是获得 this.client
private void initialize(InetSocketAddress jobTrackAddr, Configuration conf) throws IOException {
    this.initProviderList();
    Iterator i$ = this.providerList.iterator();

    while(i$.hasNext()) {
        /*
        provider这里也有分 YarnClientProtocolProvider 以及LocalClientProtocolProvider
        即本地和yarn两种provider
        */
        ClientProtocolProvider provider = (ClientProtocolProvider)i$.next();
        LOG.debug("Trying ClientProtocolProvider : " + provider.getClass().getName());
        ClientProtocol clientProtocol = null;

        try {
            /*判断jobTrackAddr是否为空，也就是以远程集群还是本地的方式运行job.
              远程集群的话，就创建yarn 提交器,:YARNRunner，通过YarnClientProtocolProvider创建
              本地的话，就创建本地local 提交器:LocalRunner，通过 LocalClientProtocolProvider创建

            主要是根据 mapreduce.framework.name 在conf中的值是local还是yarn来创建对应的runner
            */
            if (jobTrackAddr == null) {
                clientProtocol = provider.create(conf);
            } else {
                clientProtocol = provider.create(jobTrackAddr, conf);
            }

            if (clientProtocol != null) {
                this.clientProtocolProvider = provider;

                //可以看到这里client就是上面通过provider创建的
                this.client = clientProtocol;
                LOG.debug("Picked " + provider.getClass().getName() + " as the ClientProtocolProvider");

                //只要成功创建了client 和 provider就退出
                break;
            }

            LOG.debug("Cannot pick " + provider.getClass().getName() + " as the ClientProtocolProvider - returned null protocol");
        } catch (Exception var7) {
            LOG.info("Failed to use " + provider.getClass().getName() + " due to error: ", var7);
        }
    }

    if (null == this.clientProtocolProvider || null == this.client) {
        throw new IOException("Cannot initialize Cluster. Please check your configuration for mapreduce.framework.name and the correspond server addresses.");
    }
}

可以看到Cluster对象主要就是初始化了 clientProtocolProvider 以及 client 两个对象。
也就是provider和client，client是通过provider.create创建的。

下面可以看看ClientProtocolProvider和 ClientProtocol这两个类。这两个类都是抽象类，那么看他们对应有哪些实现子类。

ClientProtocolProvider：
    YarnClientProtocolProvider
    LocalClientProtocolProvider

ClientProtocol：
    YARNRunner
    LocalJobRunner

可以看看YarnClientProtocolProvider 以及LocalClientProtocolProvider的create方法

public class LocalClientProtocolProvider extends ClientProtocolProvider {
   .........
    public ClientProtocol create(Configuration conf) throws IOException {
        String framework = conf.get("mapreduce.framework.name", "local");
        if (!"local".equals(framework)) {
            return null;
        } else {
            conf.setInt("mapreduce.job.maps", 1);
            //创建LocalJobRunner
            return new LocalJobRunner(conf);
        }
    }
    .....................
}

//===============================================================
public class YarnClientProtocolProvider extends ClientProtocolProvider {
...................................
    public ClientProtocol create(Configuration conf) throws IOException {
        //创建 YARNRunner
        return "yarn".equals(conf.get("mapreduce.framework.name")) ? new YARNRunner(conf) : null;
    }
...........................
}

总的来说，就是provider分为YarnClientProtocolProvider 以及LocalClientProtocolProvider，分别用于创建client中的 YARNRunner 和 LocalJobRunner。表示job运行方式有本地和yarn两种。

至此，this.client以及this.provider这两个在Cluster对象中的对象初始化完成。

4、this.getJobSubmitter()封装submitter

//-------------------job.java
public JobSubmitter getJobSubmitter(FileSystem fs, ClientProtocol submitClient) throws IOException {
        return new JobSubmitter(fs, submitClient);
}

创建个 JobSubmitter对象，看看构造方法

//------------------JobSubmitter.java
JobSubmitter(FileSystem submitFs, ClientProtocol submitClient) throws IOException {
        this.submitClient = submitClient;
        this.jtFs = submitFs;
    }

看起来，没啥特别， 就是把文件系统fs以及 上面cluster中初始化的client保存起来。但是其实这个类中有很多方法后面会调用。后面讲

5、submitter.submitJobInternal()提交job

这个方法是整个job提交过程中的核心，要注意看

//------------------JobSubmitter.java
JobStatus submitJobInternal(Job job, Cluster cluster) throws ClassNotFoundException, InterruptedException, IOException {
    //检查配置的输出是否已存在，已存在会抛出异常
    this.checkSpecs(job);
    Configuration conf = job.getConfiguration();
    addMRFrameworkToDistributedCache(conf);
    //获取所有job工作总目录
    Path jobStagingArea = JobSubmissionFiles.getStagingDir(cluster, conf);
    //获取ip地址对象
    InetAddress ip = InetAddress.getLocalHost();
    //设置提交job的主机名和ip
    if (ip != null) {
        this.submitHostAddress = ip.getHostAddress();
        this.submitHostName = ip.getHostName();
        conf.set("mapreduce.job.submithostname", this.submitHostName);
        conf.set("mapreduce.job.submithostaddress", this.submitHostAddress);
    }

    //通过client向集群申请运行job，获取到对应的jobid.这个submitclient是前面cluster初始化完成的
    JobID jobId = this.submitClient.getNewJobID();
    job.setJobID(jobId);
    //创建存储job相关资源数据的目录对象.存储job配置文件、切片信息文件、程序jar包等
    Path submitJobDir = new Path(jobStagingArea, jobId.toString());
    JobStatus status = null;

    JobStatus var24;
    try {
        conf.set("mapreduce.job.user.name", UserGroupInformation.getCurrentUser().getShortUserName());
        conf.set("hadoop.http.filter.initializers", "org.apache.hadoop.yarn.server.webproxy.amfilter.AmFilterInitializer");
        conf.set("mapreduce.job.dir", submitJobDir.toString());
        LOG.debug("Configuring job " + jobId + " with " + submitJobDir + " as the submit dir");
        //获取访问namenode中特定目录授权
        TokenCache.obtainTokensForNamenodes(job.getCredentials(), new Path[]{submitJobDir}, conf);
        this.populateTokenCache(conf, job.getCredentials());
        //验证token相关
        if (TokenCache.getShuffleSecretKey(job.getCredentials()) == null) {
            KeyGenerator keyGen;
            try {
                keyGen = KeyGenerator.getInstance("HmacSHA1");
                keyGen.init(64);
            } catch (NoSuchAlgorithmException var19) {
                throw new IOException("Error generating shuffle secret key", var19);
            }

            SecretKey shuffleKey = keyGen.generateKey();
            TokenCache.setShuffleSecretKey(shuffleKey.getEncoded(), job.getCredentials());
        }

        if (CryptoUtils.isEncryptedSpillEnabled(conf)) {
            conf.setInt("mapreduce.am.max-attempts", 1);
            LOG.warn("Max job attempts set to 1 since encrypted intermediatedata spill is enabled");
        }

        //复制job的临时文件，以及运行的jar包到submitJobDir下
        this.copyAndConfigureFiles(job, submitJobDir);
        //获取存储job配置信息文件路径，一般命名为：submitJobDir/job.xml
        Path submitJobFile = JobSubmissionFiles.getJobConfPath(submitJobDir);
        LOG.debug("Creating splits at " + this.jtFs.makeQualified(submitJobDir));
        //将切片信息存储到submitJobDir下,并返回切片数目。会调用 InputFormat.getSplits()来获取规划的切片信息
        //切片信息会写入到 submitJobDir/job.split，切片信息条目的元信息写入到 submitJobDir/job.splitmetainfo
        int maps = this.writeSplits(job, submitJobDir);
        conf.setInt("mapreduce.job.maps", maps);
        LOG.info("number of splits:" + maps);
        //传输队列名称    
        String queue = conf.get("mapreduce.job.queuename", "default");
        //submitClient其实就是cluster的client
        AccessControlList acl = this.submitClient.getQueueAdmins(queue);
        conf.set(QueueManager.toFullPropertyName(queue, QueueACL.ADMINISTER_JOBS.getAclName()), acl.getAclString());
        TokenCache.cleanUpTokenReferral(conf);
        if (conf.getBoolean("mapreduce.job.token.tracking.ids.enabled", false)) {
            ArrayList<String> trackingIds = new ArrayList();
            Iterator i$ = job.getCredentials().getAllTokens().iterator();

            while(i$.hasNext()) {
                Token<? extends TokenIdentifier> t = (Token)i$.next();
                trackingIds.add(t.decodeIdentifier().getTrackingId());
            }

            conf.setStrings("mapreduce.job.token.tracking.ids", (String[])trackingIds.toArray(new String[trackingIds.size()]));
        }

        ReservationId reservationId = job.getReservationId();
        if (reservationId != null) {
            conf.set("mapreduce.job.reservation.id", reservationId.toString());
        }

        //将job的configuration信息写入到 submitJobDir/job.xml
        this.writeConf(conf, submitJobFile);
        this.printTokens(jobId, job.getCredentials());
        //通过client提交job，包括job资源目录，验证信息.
        //这里要看使用的client是YARNRunner还是LocalRunner
        //最后返回提交job的状态
        status = this.submitClient.submitJob(jobId, submitJobDir.toString(), job.getCredentials());
        if (status == null) {
            throw new IOException("Could not launch job");
        }

        var24 = status;
    } finally {
        //如果提交任务失败，则删除jobdir
        if (status == null) {
            LOG.info("Cleaning up the staging area " + submitJobDir);
            if (this.jtFs != null && submitJobDir != null) {
                this.jtFs.delete(submitJobDir, true);
            }
        }

    }

    return var24;
}

总结一下上面的主要流程：
（1）Path jobStagingArea = JobSubmissionFiles.getStagingDir(cluster, conf);
获取job总的工作目录

（2）JobID jobId = this.submitClient.getNewJobID();
job.setJobID(jobId);
通过处理client向集群申请jobid，并保持到job的配置信息中。

（3）Path submitJobDir = new Path(jobStagingArea, jobId.toString());
获取当前job的工作目录，以及jobid命名

（4）this.copyAndConfigureFiles(job, submitJobDir);
复制job的临时文件，运行的jar包到submitJobDir下

（5）Path submitJobFile = JobSubmissionFiles.getJobConfPath(submitJobDir);
获取job配置信息文件的路径。命名为：submitJobDir/job.xml

（6）int maps = this.writeSplits(job, submitJobDir);
将切片信息存储到submitJobDir下,并返回切片数目。会调用 InputFormat.getSplits()来获取规划的切片信息。切片信息会写入到 submitJobDir/job.split，切片信息条目的元信息写入到 submitJobDir/job.splitmetainfo。

（7）this.writeConf(conf, submitJobFile);
将job配置信息写入到 submitJobDir/job.xml 中

（8）status = this.submitClient.submitJob(jobId, submitJobDir.toString(), job.getCredentials());
正式提交job，获取job的提交状态

下面挑比较复杂的看看这些的具体实现。

重点在于job任务的资源的生成，如切片文件的生成。

=================================================================

（1）this.copyAndConfigureFiles(job, submitJobDir);

复制job的临时文件，运行的jar包到submitJobDir下

//------------------JobSubmitter.java
private void copyAndConfigureFiles(Job job, Path jobSubmitDir) throws IOException {
        JobResourceUploader rUploader = new JobResourceUploader(this.jtFs);
        rUploader.uploadFiles(job, jobSubmitDir);
        job.getWorkingDirectory();
    }

//----------------------JobResourceUploader.java
public void uploadFiles(Job job, Path submitJobDir) throws IOException {
......................
        String files = conf.get("tmpfiles");
        String libjars = conf.get("tmpjars");
        String archives = conf.get("tmparchives");
        String jobJar = job.getJar();
..................代码长，就截取一点，这些就是要复制到job目录的文件类型
}

可以看到主要复制jar包以及相关的文件到job工作目录下。

（2）Path submitJobFile = JobSubmissionFiles.getJobConfPath(submitJobDir);

获取job配置信息文件的路径。命名为：submitJobDir/job.xml

//-----------------------------JobSubmissionFiles.java
public static Path getJobConfPath(Path jobSubmitDir) {
        return new Path(jobSubmitDir, "job.xml");
}

（3）int maps = this.writeSplits(job, submitJobDir);

将切片信息存储到submitJobDir下,并返回切片数目。会调用 InputFormat.getSplits()来获取规划的切片信息。切片信息会写入到 submitJobDir/job.split，切片信息条目的元信息写入到 submitJobDir/job.splitmetainfo。返回的是切片数目

//------------------------JobSubmitter.java
private int writeSplits(JobContext job, Path jobSubmitDir) throws IOException, InterruptedException, ClassNotFoundException {
    JobConf jConf = (JobConf)job.getConfiguration();
    int maps;
    if (jConf.getUseNewMapper()) {
        maps = this.writeNewSplits(job, jobSubmitDir);
    } else {
        maps = this.writeOldSplits(jConf, jobSubmitDir);
    }

    return maps;
}

没什么特别的，主要就是区分新旧api，我们看 this.writeNewSplits

//------------------------JobSubmitter.java
private <T extends InputSplit> int writeNewSplits(JobContext job, Path jobSubmitDir) throws IOException, InterruptedException, ClassNotFoundException {
    Configuration conf = job.getConfiguration();
    //反射获取指定的inputformat对象，默认TextInputFormat
    InputFormat<?, ?> input = (InputFormat)ReflectionUtils.newInstance(job.getInputFormatClass(), conf);
    //通过inputformat的getSplits() 生成获取规划切片信息
    List<InputSplit> splits = input.getSplits(job);
    T[] array = (InputSplit[])((InputSplit[])splits.toArray(new InputSplit[splits.size()]));
    Arrays.sort(array, new JobSubmitter.SplitComparator());
    //创建切片文件原始数据文件，以及元数据文件
    JobSplitWriter.createSplitFiles(jobSubmitDir, conf, jobSubmitDir.getFileSystem(conf), array);
    return array.length;
}

获取 inputformat对象，通过inputformat的getSplits() 获取规划切片信息，然后JobSplitWriter.createSplitFiles()创建切片信息文件。下面最后这个方法

//------------------JobSplitWriter.createSplitFiles
public static <T extends InputSplit> void createSplitFiles(Path jobSubmitDir, Configuration conf, FileSystem fs, T[] splits) throws IOException, InterruptedException {
    //创建切片输出流，文件命名为 jobSubmitDir/job.split
    FSDataOutputStream out = createFile(fs, JobSubmissionFiles.getJobSplitFile(jobSubmitDir), conf);
    //将数组中的每个切片元信息进行序列化,并将切片信息写入到jobSubmitDir/job.split中
    //返回的是每个切片条目的元信息，比如每条切片信息在 job.split中的起始位置，长度等
    SplitMetaInfo[] info = writeNewSplits(conf, splits, out);
    out.close();
    //将切片信息文件的元信息写入到文件 jobSubmitDir/job.splitmetainfo 中
    writeJobSplitMetaInfo(fs, JobSubmissionFiles.getJobSplitMetaFile(jobSubmitDir), new FsPermission(JobSubmissionFiles.JOB_FILE_PERMISSION), 1, info);
}

这里主要生成两个主要文件
jobSubmitDir/job.split：切片信息文件，记录每个切片的信息，比如路径，block位置，偏移量等
jobSubmitDir/job.splitmetainfo：切片信息文件中每个信息条目的索引位置，如每条切片信息在 job.split中的起始位置，长度等

下面看看这两个文件的生成
首先是jobSubmitDir/job.split

private static <T extends InputSplit> SplitMetaInfo[] writeNewSplits(Configuration conf, T[] array, FSDataOutputStream out) throws IOException, InterruptedException {
    SplitMetaInfo[] info = new SplitMetaInfo[array.length];
    if (array.length != 0) {
        SerializationFactory factory = new SerializationFactory(conf);
        int i = 0;
        int maxBlockLocations = conf.getInt("mapreduce.job.max.split.locations", 10);
        long offset = out.getPos();
        InputSplit[] arr$ = array;
        int len$ = array.length;

        //循环将切片信息中每一条切片信息写入到文件中，并生成每条切片信息的元信息
        for(int i$ = 0; i$ < len$; ++i$) {
            T split = arr$[i$];
            long prevCount = out.getPos();
            Text.writeString(out, split.getClass().getName());
            Serializer<T> serializer = factory.getSerializer(split.getClass());
            serializer.open(out);
            //将切片信息对象序列化存储到文件中
            serializer.serialize(split);
            long currCount = out.getPos();
            String[] locations = split.getLocations();
            if (locations.length > maxBlockLocations) {
                LOG.warn("Max block location exceeded for split: " + split + " splitsize: " + locations.length + " maxsize: " + maxBlockLocations);
                locations = (String[])Arrays.copyOf(locations, maxBlockLocations);
            }

            //生成每条切片信息的元信息
            info[i++] = new SplitMetaInfo(locations, offset, split.getLength());
            offset += currCount - prevCount;
        }
    }

    return info;
}

主要就是将split中的切片信息条目对象序列化写入到文件中，并生成jobSubmitDir/job.splitmetainfo中要写入的信息，也就是切片文件的索引信息
接着看看 writeJobSplitMetaInfo()

private static void writeJobSplitMetaInfo(FileSystem fs, Path filename, FsPermission p, int splitMetaInfoVersion, SplitMetaInfo[] allSplitMetaInfo) throws IOException {
    //写入切片信息条目的元信息，创建一个输出流
    FSDataOutputStream out = FileSystem.create(fs, filename, p);
    out.write(JobSplit.META_SPLIT_FILE_HEADER);
    WritableUtils.writeVInt(out, splitMetaInfoVersion);
    WritableUtils.writeVInt(out, allSplitMetaInfo.length);
    SplitMetaInfo[] arr$ = allSplitMetaInfo;
    int len$ = allSplitMetaInfo.length;

    //逐条写入
    for(int i$ = 0; i$ < len$; ++i$) {
        SplitMetaInfo splitMetaInfo = arr$[i$];
        splitMetaInfo.write(out);
    }

    out.close();
}

这里其实很明显了，就是将切片文件索引信息写入到 jobSubmitDir/job.splitmetainfo

（4）status = this.submitClient.submitJob(jobId, submitJobDir.toString(), job.getCredentials());

正式提交job，获取job的提交状态

public JobStatus submitJob(JobID jobId, String jobSubmitDir, Credentials ts) throws IOException, InterruptedException {
    this.addHistoryToken(ts);
    //这里就是将job配置，以及job资源的hdfs目录路径传入
    ApplicationSubmissionContext appContext = this.createApplicationSubmissionContext(this.conf, jobSubmitDir, ts);

    try {
        //提交job，返回的appid
        ApplicationId applicationId = this.resMgrDelegate.submitApplication(appContext);
        //根据appid创建appMaster
        ApplicationReport appMaster = this.resMgrDelegate.getApplicationReport(applicationId);
        String diagnostics = appMaster == null ? "application report is null" : appMaster.getDiagnostics();
        if (appMaster != null && appMaster.getYarnApplicationState() != YarnApplicationState.FAILED && appMaster.getYarnApplicationState() != YarnApplicationState.KILLED) {
            return this.clientCache.getClient(jobId).getJobStatus(jobId);
        } else {
            throw new IOException("Failed to run job : " + diagnostics);
        }
    } catch (YarnException var8) {
        throw new IOException(var8);
    }
}

这里主要就是提交job，创建appMaster。最后获取job状态。

二、总结

一个job提交流程主要如下：
1、和MapReduce集群建立连接 this.connect()
这里面最重要就是创建了 client，有 YARNRunner和LocalJobRunner两种方式。后续用来和server端通信、提交job等。

2、正式提交job ，submitter.submitJobInternal(Job.this, cluster)
（1）Path jobStagingArea = JobSubmissionFiles.getStagingDir(cluster, conf);
获取job总的工作目录

（2）JobID jobId = this.submitClient.getNewJobID();
job.setJobID(jobId);
通过处理client向集群申请jobid，并保持到job的配置信息中。

（3）Path submitJobDir = new Path(jobStagingArea, jobId.toString());
获取当前job的工作目录，以及jobid命名

（4）this.copyAndConfigureFiles(job, submitJobDir);
复制job的临时文件，运行的jar包到submitJobDir下

（5）Path submitJobFile = JobSubmissionFiles.getJobConfPath(submitJobDir);
获取job配置信息文件的路径。命名为：submitJobDir/job.xml

（6）int maps = this.writeSplits(job, submitJobDir);
将切片信息存储到submitJobDir下,并返回切片数目。会调用 InputFormat.getSplits()来获取规划的切片信息。切片信息会写入到 submitJobDir/job.split，切片信息条目的元信息写入到 submitJobDir/job.splitmetainfo。

（7）this.writeConf(conf, submitJobFile);
将job配置信息写入到 submitJobDir/job.xml 中

（8）status = this.submitClient.submitJob(jobId, submitJobDir.toString(), job.getCredentials());
正式提交job，获取job的提交状态