Пример #1
0
  private void handleSampling(DriverContext context, MapWork mWork, JobConf job, HiveConf conf)
      throws Exception {
    assert mWork.getAliasToWork().keySet().size() == 1;

    String alias = mWork.getAliases().get(0);
    Operator<?> topOp = mWork.getAliasToWork().get(alias);
    PartitionDesc partDesc = mWork.getAliasToPartnInfo().get(alias);

    ArrayList<String> paths = mWork.getPaths();
    ArrayList<PartitionDesc> parts = mWork.getPartitionDescs();

    List<Path> inputPaths = new ArrayList<Path>(paths.size());
    for (String path : paths) {
      inputPaths.add(new Path(path));
    }

    Path tmpPath = context.getCtx().getExternalTmpPath(inputPaths.get(0));
    Path partitionFile = new Path(tmpPath, ".partitions");
    ShimLoader.getHadoopShims().setTotalOrderPartitionFile(job, partitionFile);
    PartitionKeySampler sampler = new PartitionKeySampler();

    if (mWork.getSamplingType() == MapWork.SAMPLING_ON_PREV_MR) {
      console.printInfo("Use sampling data created in previous MR");
      // merges sampling data from previous MR and make partition keys for total sort
      for (Path path : inputPaths) {
        FileSystem fs = path.getFileSystem(job);
        for (FileStatus status : fs.globStatus(new Path(path, ".sampling*"))) {
          sampler.addSampleFile(status.getPath(), job);
        }
      }
    } else if (mWork.getSamplingType() == MapWork.SAMPLING_ON_START) {
      console.printInfo("Creating sampling data..");
      assert topOp instanceof TableScanOperator;
      TableScanOperator ts = (TableScanOperator) topOp;

      FetchWork fetchWork;
      if (!partDesc.isPartitioned()) {
        assert paths.size() == 1;
        fetchWork = new FetchWork(inputPaths.get(0), partDesc.getTableDesc());
      } else {
        fetchWork = new FetchWork(inputPaths, parts, partDesc.getTableDesc());
      }
      fetchWork.setSource(ts);

      // random sampling
      FetchOperator fetcher = PartitionKeySampler.createSampler(fetchWork, conf, job, ts);
      try {
        ts.initialize(conf, new ObjectInspector[] {fetcher.getOutputObjectInspector()});
        OperatorUtils.setChildrenCollector(ts.getChildOperators(), sampler);
        while (fetcher.pushRow()) {}
      } finally {
        fetcher.clearFetchContext();
      }
    } else {
      throw new IllegalArgumentException("Invalid sampling type " + mWork.getSamplingType());
    }
    sampler.writePartitionKeys(partitionFile, conf, job);
  }
Пример #2
0
  public int executeInChildVM(DriverContext driverContext) {
    // execute in child jvm
    try {
      // generate the cmd line to run in the child jvm
      Context ctx = driverContext.getCtx();
      String hiveJar = conf.getJar();

      String hadoopExec = conf.getVar(HiveConf.ConfVars.HADOOPBIN);
      conf.setVar(
          ConfVars.HIVEADDEDJARS, Utilities.getResourceFiles(conf, SessionState.ResourceType.JAR));
      // write out the plan to a local file
      Path planPath = new Path(ctx.getLocalTmpPath(), "plan.xml");
      MapredLocalWork plan = getWork();
      LOG.info("Generating plan file " + planPath.toString());

      OutputStream out = null;
      try {
        out = FileSystem.getLocal(conf).create(planPath);
        SerializationUtilities.serializePlan(plan, out);
        out.close();
        out = null;
      } finally {
        IOUtils.closeQuietly(out);
      }

      String isSilent = "true".equalsIgnoreCase(System.getProperty("test.silent")) ? "-nolog" : "";

      String jarCmd;

      jarCmd = hiveJar + " " + ExecDriver.class.getName();
      String hiveConfArgs = ExecDriver.generateCmdLine(conf, ctx);
      String cmdLine =
          hadoopExec
              + " jar "
              + jarCmd
              + " -localtask -plan "
              + planPath.toString()
              + " "
              + isSilent
              + " "
              + hiveConfArgs;

      String workDir = (new File(".")).getCanonicalPath();
      String files = Utilities.getResourceFiles(conf, SessionState.ResourceType.FILE);

      if (!files.isEmpty()) {
        cmdLine = cmdLine + " -files " + files;

        workDir = ctx.getLocalTmpPath().toUri().getPath();

        if (!(new File(workDir)).mkdir()) {
          throw new IOException("Cannot create tmp working dir: " + workDir);
        }

        for (String f : StringUtils.split(files, ',')) {
          Path p = new Path(f);
          String target = p.toUri().getPath();
          String link = workDir + Path.SEPARATOR + p.getName();
          if (FileUtil.symLink(target, link) != 0) {
            throw new IOException("Cannot link to added file: " + target + " from: " + link);
          }
        }
      }

      // Inherit Java system variables
      String hadoopOpts;
      StringBuilder sb = new StringBuilder();
      Properties p = System.getProperties();
      for (String element : HIVE_SYS_PROP) {
        if (p.containsKey(element)) {
          sb.append(" -D" + element + "=" + p.getProperty(element));
        }
      }
      hadoopOpts = sb.toString();
      // Inherit the environment variables
      String[] env;
      Map<String, String> variables = new HashMap<String, String>(System.getenv());
      // The user can specify the hadoop memory

      // if ("local".equals(conf.getVar(HiveConf.ConfVars.HADOOPJT))) {
      // if we are running in local mode - then the amount of memory used
      // by the child jvm can no longer default to the memory used by the
      // parent jvm
      // int hadoopMem = conf.getIntVar(HiveConf.ConfVars.HIVEHADOOPMAXMEM);
      int hadoopMem = conf.getIntVar(HiveConf.ConfVars.HIVEHADOOPMAXMEM);
      if (hadoopMem == 0) {
        // remove env var that would default child jvm to use parent's memory
        // as default. child jvm would use default memory for a hadoop client
        variables.remove(HADOOP_MEM_KEY);
      } else {
        // user specified the memory for local mode hadoop run
        console.printInfo(" set heap size\t" + hadoopMem + "MB");
        variables.put(HADOOP_MEM_KEY, String.valueOf(hadoopMem));
      }
      // } else {
      // nothing to do - we are not running in local mode - only submitting
      // the job via a child process. in this case it's appropriate that the
      // child jvm use the same memory as the parent jvm

      // }

      // Set HADOOP_USER_NAME env variable for child process, so that
      // it also runs with hadoop permissions for the user the job is running as
      // This will be used by hadoop only in unsecure(/non kerberos) mode
      String endUserName = Utils.getUGI().getShortUserName();
      LOG.debug("setting HADOOP_USER_NAME\t" + endUserName);
      variables.put("HADOOP_USER_NAME", endUserName);

      if (variables.containsKey(HADOOP_OPTS_KEY)) {
        variables.put(HADOOP_OPTS_KEY, variables.get(HADOOP_OPTS_KEY) + hadoopOpts);
      } else {
        variables.put(HADOOP_OPTS_KEY, hadoopOpts);
      }

      // For Windows OS, we need to pass HIVE_HADOOP_CLASSPATH Java parameter while starting
      // Hiveserver2 using "-hiveconf hive.hadoop.classpath=%HIVE_LIB%". This is to combine path(s).
      if (HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_HADOOP_CLASSPATH) != null) {
        if (variables.containsKey("HADOOP_CLASSPATH")) {
          variables.put(
              "HADOOP_CLASSPATH",
              variables.get("HADOOP_CLASSPATH")
                  + ";"
                  + HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_HADOOP_CLASSPATH));
        } else {
          variables.put(
              "HADOOP_CLASSPATH", HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_HADOOP_CLASSPATH));
        }
      }

      if (variables.containsKey(MapRedTask.HIVE_DEBUG_RECURSIVE)) {
        MapRedTask.configureDebugVariablesForChildJVM(variables);
      }

      if (UserGroupInformation.isSecurityEnabled() && UserGroupInformation.isLoginKeytabBased()) {
        // If kerberos security is enabled, and HS2 doAs is enabled,
        // then additional params need to be set so that the command is run as
        // intended user
        secureDoAs = new SecureCmdDoAs(conf);
        secureDoAs.addEnv(variables);
      }

      // If HIVE_LOCAL_TASK_CHILD_OPTS is set, child VM environment setting
      // HADOOP_CLIENT_OPTS will be replaced with HIVE_LOCAL_TASK_CHILD_OPTS.
      // HADOOP_OPTS is updated too since HADOOP_CLIENT_OPTS is appended
      // to HADOOP_OPTS in most cases. This way, the local task JVM can
      // have different settings from those of HiveServer2.
      if (variables.containsKey(HIVE_LOCAL_TASK_CHILD_OPTS_KEY)) {
        String childOpts = variables.get(HIVE_LOCAL_TASK_CHILD_OPTS_KEY);
        if (childOpts == null) {
          childOpts = "";
        }
        String clientOpts = variables.put(HADOOP_CLIENT_OPTS, childOpts);
        String tmp = variables.get(HADOOP_OPTS_KEY);
        if (tmp != null && !StringUtils.isBlank(clientOpts)) {
          tmp = tmp.replace(clientOpts, childOpts);
          variables.put(HADOOP_OPTS_KEY, tmp);
        }
      }

      env = new String[variables.size()];
      int pos = 0;
      for (Map.Entry<String, String> entry : variables.entrySet()) {
        String name = entry.getKey();
        String value = entry.getValue();
        env[pos++] = name + "=" + value;
        LOG.debug("Setting env: " + env[pos - 1]);
      }

      LOG.info("Executing: " + cmdLine);

      // Run ExecDriver in another JVM
      executor = Runtime.getRuntime().exec(cmdLine, env, new File(workDir));

      CachingPrintStream errPrintStream = new CachingPrintStream(System.err);

      StreamPrinter outPrinter = new StreamPrinter(executor.getInputStream(), null, System.out);
      StreamPrinter errPrinter = new StreamPrinter(executor.getErrorStream(), null, errPrintStream);

      outPrinter.start();
      errPrinter.start();

      int exitVal = jobExecHelper.progressLocal(executor, getId());

      // wait for stream threads to finish
      outPrinter.join();
      errPrinter.join();

      if (exitVal != 0) {
        LOG.error("Execution failed with exit status: " + exitVal);
        if (SessionState.get() != null) {
          SessionState.get().addLocalMapRedErrors(getId(), errPrintStream.getOutput());
        }
      } else {
        LOG.info("Execution completed successfully");
      }

      return exitVal;
    } catch (Exception e) {
      LOG.error("Exception: " + e, e);
      return (1);
    } finally {
      if (secureDoAs != null) {
        secureDoAs.close();
      }
    }
  }
Пример #3
0
  /** Execute a query plan using Hadoop. */
  @SuppressWarnings({"deprecation", "unchecked"})
  @Override
  public int execute(DriverContext driverContext) {

    IOPrepareCache ioPrepareCache = IOPrepareCache.get();
    ioPrepareCache.clear();

    boolean success = true;

    Context ctx = driverContext.getCtx();
    boolean ctxCreated = false;
    Path emptyScratchDir;

    MapWork mWork = work.getMapWork();
    ReduceWork rWork = work.getReduceWork();

    try {
      if (ctx == null) {
        ctx = new Context(job);
        ctxCreated = true;
      }

      emptyScratchDir = ctx.getMRTmpPath();
      FileSystem fs = emptyScratchDir.getFileSystem(job);
      fs.mkdirs(emptyScratchDir);
    } catch (IOException e) {
      e.printStackTrace();
      console.printError(
          "Error launching map-reduce job",
          "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
      return 5;
    }

    HiveFileFormatUtils.prepareJobOutput(job);
    // See the javadoc on HiveOutputFormatImpl and HadoopShims.prepareJobOutput()
    job.setOutputFormat(HiveOutputFormatImpl.class);

    job.setMapperClass(ExecMapper.class);

    job.setMapOutputKeyClass(HiveKey.class);
    job.setMapOutputValueClass(BytesWritable.class);

    try {
      String partitioner = HiveConf.getVar(job, ConfVars.HIVEPARTITIONER);
      job.setPartitionerClass(JavaUtils.loadClass(partitioner));
    } catch (ClassNotFoundException e) {
      throw new RuntimeException(e.getMessage(), e);
    }

    if (mWork.getNumMapTasks() != null) {
      job.setNumMapTasks(mWork.getNumMapTasks().intValue());
    }

    if (mWork.getMaxSplitSize() != null) {
      HiveConf.setLongVar(
          job, HiveConf.ConfVars.MAPREDMAXSPLITSIZE, mWork.getMaxSplitSize().longValue());
    }

    if (mWork.getMinSplitSize() != null) {
      HiveConf.setLongVar(
          job, HiveConf.ConfVars.MAPREDMINSPLITSIZE, mWork.getMinSplitSize().longValue());
    }

    if (mWork.getMinSplitSizePerNode() != null) {
      HiveConf.setLongVar(
          job,
          HiveConf.ConfVars.MAPREDMINSPLITSIZEPERNODE,
          mWork.getMinSplitSizePerNode().longValue());
    }

    if (mWork.getMinSplitSizePerRack() != null) {
      HiveConf.setLongVar(
          job,
          HiveConf.ConfVars.MAPREDMINSPLITSIZEPERRACK,
          mWork.getMinSplitSizePerRack().longValue());
    }

    job.setNumReduceTasks(rWork != null ? rWork.getNumReduceTasks().intValue() : 0);
    job.setReducerClass(ExecReducer.class);

    // set input format information if necessary
    setInputAttributes(job);

    // Turn on speculative execution for reducers
    boolean useSpeculativeExecReducers =
        HiveConf.getBoolVar(job, HiveConf.ConfVars.HIVESPECULATIVEEXECREDUCERS);
    HiveConf.setBoolVar(
        job, HiveConf.ConfVars.HADOOPSPECULATIVEEXECREDUCERS, useSpeculativeExecReducers);

    String inpFormat = HiveConf.getVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT);

    if (mWork.isUseBucketizedHiveInputFormat()) {
      inpFormat = BucketizedHiveInputFormat.class.getName();
    }

    LOG.info("Using " + inpFormat);

    try {
      job.setInputFormat(JavaUtils.loadClass(inpFormat));
    } catch (ClassNotFoundException e) {
      throw new RuntimeException(e.getMessage(), e);
    }

    // No-Op - we don't really write anything here ..
    job.setOutputKeyClass(Text.class);
    job.setOutputValueClass(Text.class);

    // Transfer HIVEAUXJARS and HIVEADDEDJARS to "tmpjars" so hadoop understands
    // it
    String auxJars = HiveConf.getVar(job, HiveConf.ConfVars.HIVEAUXJARS);
    String addedJars = HiveConf.getVar(job, HiveConf.ConfVars.HIVEADDEDJARS);
    if (StringUtils.isNotBlank(auxJars) || StringUtils.isNotBlank(addedJars)) {
      String allJars =
          StringUtils.isNotBlank(auxJars)
              ? (StringUtils.isNotBlank(addedJars) ? addedJars + "," + auxJars : auxJars)
              : addedJars;
      LOG.info("adding libjars: " + allJars);
      initializeFiles("tmpjars", allJars);
    }

    // Transfer HIVEADDEDFILES to "tmpfiles" so hadoop understands it
    String addedFiles = HiveConf.getVar(job, HiveConf.ConfVars.HIVEADDEDFILES);
    if (StringUtils.isNotBlank(addedFiles)) {
      initializeFiles("tmpfiles", addedFiles);
    }
    int returnVal = 0;
    boolean noName = StringUtils.isEmpty(HiveConf.getVar(job, HiveConf.ConfVars.HADOOPJOBNAME));

    if (noName) {
      // This is for a special case to ensure unit tests pass
      HiveConf.setVar(job, HiveConf.ConfVars.HADOOPJOBNAME, "JOB" + Utilities.randGen.nextInt());
    }
    String addedArchives = HiveConf.getVar(job, HiveConf.ConfVars.HIVEADDEDARCHIVES);
    // Transfer HIVEADDEDARCHIVES to "tmparchives" so hadoop understands it
    if (StringUtils.isNotBlank(addedArchives)) {
      initializeFiles("tmparchives", addedArchives);
    }

    try {
      MapredLocalWork localwork = mWork.getMapRedLocalWork();
      if (localwork != null && localwork.hasStagedAlias()) {
        if (!ShimLoader.getHadoopShims().isLocalMode(job)) {
          Path localPath = localwork.getTmpPath();
          Path hdfsPath = mWork.getTmpHDFSPath();

          FileSystem hdfs = hdfsPath.getFileSystem(job);
          FileSystem localFS = localPath.getFileSystem(job);
          FileStatus[] hashtableFiles = localFS.listStatus(localPath);
          int fileNumber = hashtableFiles.length;
          String[] fileNames = new String[fileNumber];

          for (int i = 0; i < fileNumber; i++) {
            fileNames[i] = hashtableFiles[i].getPath().getName();
          }

          // package and compress all the hashtable files to an archive file
          String stageId = this.getId();
          String archiveFileName = Utilities.generateTarFileName(stageId);
          localwork.setStageID(stageId);

          CompressionUtils.tar(localPath.toUri().getPath(), fileNames, archiveFileName);
          Path archivePath = Utilities.generateTarPath(localPath, stageId);
          LOG.info("Archive " + hashtableFiles.length + " hash table files to " + archivePath);

          // upload archive file to hdfs
          Path hdfsFilePath = Utilities.generateTarPath(hdfsPath, stageId);
          short replication = (short) job.getInt("mapred.submit.replication", 10);
          hdfs.copyFromLocalFile(archivePath, hdfsFilePath);
          hdfs.setReplication(hdfsFilePath, replication);
          LOG.info("Upload 1 archive file  from" + archivePath + " to: " + hdfsFilePath);

          // add the archive file to distributed cache
          DistributedCache.createSymlink(job);
          DistributedCache.addCacheArchive(hdfsFilePath.toUri(), job);
          LOG.info(
              "Add 1 archive file to distributed cache. Archive file: " + hdfsFilePath.toUri());
        }
      }
      work.configureJobConf(job);
      List<Path> inputPaths = Utilities.getInputPaths(job, mWork, emptyScratchDir, ctx, false);
      Utilities.setInputPaths(job, inputPaths);

      Utilities.setMapRedWork(job, work, ctx.getMRTmpPath());

      if (mWork.getSamplingType() > 0 && rWork != null && job.getNumReduceTasks() > 1) {
        try {
          handleSampling(ctx, mWork, job);
          job.setPartitionerClass(HiveTotalOrderPartitioner.class);
        } catch (IllegalStateException e) {
          console.printInfo("Not enough sampling data.. Rolling back to single reducer task");
          rWork.setNumReduceTasks(1);
          job.setNumReduceTasks(1);
        } catch (Exception e) {
          LOG.error("Sampling error", e);
          console.printError(
              e.toString(), "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
          rWork.setNumReduceTasks(1);
          job.setNumReduceTasks(1);
        }
      }

      // remove the pwd from conf file so that job tracker doesn't show this
      // logs
      String pwd = HiveConf.getVar(job, HiveConf.ConfVars.METASTOREPWD);
      if (pwd != null) {
        HiveConf.setVar(job, HiveConf.ConfVars.METASTOREPWD, "HIVE");
      }
      JobClient jc = new JobClient(job);
      // make this client wait if job tracker is not behaving well.
      Throttle.checkJobTracker(job, LOG);

      if (mWork.isGatheringStats() || (rWork != null && rWork.isGatheringStats())) {
        // initialize stats publishing table
        StatsPublisher statsPublisher;
        StatsFactory factory = StatsFactory.newFactory(job);
        if (factory != null) {
          statsPublisher = factory.getStatsPublisher();
          List<String> statsTmpDir = Utilities.getStatsTmpDirs(mWork, job);
          if (rWork != null) {
            statsTmpDir.addAll(Utilities.getStatsTmpDirs(rWork, job));
          }
          StatsCollectionContext sc = new StatsCollectionContext(job);
          sc.setStatsTmpDirs(statsTmpDir);
          if (!statsPublisher.init(sc)) { // creating stats table if not exists
            if (HiveConf.getBoolVar(job, HiveConf.ConfVars.HIVE_STATS_RELIABLE)) {
              throw new HiveException(
                  ErrorMsg.STATSPUBLISHER_INITIALIZATION_ERROR.getErrorCodedMsg());
            }
          }
        }
      }

      Utilities.createTmpDirs(job, mWork);
      Utilities.createTmpDirs(job, rWork);

      SessionState ss = SessionState.get();
      if (HiveConf.getVar(job, HiveConf.ConfVars.HIVE_EXECUTION_ENGINE).equals("tez")
          && ss != null) {
        TezSessionState session = ss.getTezSession();
        TezSessionPoolManager.getInstance().close(session, true);
      }

      // Finally SUBMIT the JOB!
      rj = jc.submitJob(job);
      // replace it back
      if (pwd != null) {
        HiveConf.setVar(job, HiveConf.ConfVars.METASTOREPWD, pwd);
      }

      returnVal = jobExecHelper.progress(rj, jc, ctx.getHiveTxnManager());
      success = (returnVal == 0);
    } catch (Exception e) {
      e.printStackTrace();
      String mesg = " with exception '" + Utilities.getNameMessage(e) + "'";
      if (rj != null) {
        mesg = "Ended Job = " + rj.getJobID() + mesg;
      } else {
        mesg = "Job Submission failed" + mesg;
      }

      // Has to use full name to make sure it does not conflict with
      // org.apache.commons.lang.StringUtils
      console.printError(mesg, "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));

      success = false;
      returnVal = 1;
    } finally {
      Utilities.clearWork(job);
      try {
        if (ctxCreated) {
          ctx.clear();
        }

        if (rj != null) {
          if (returnVal != 0) {
            rj.killJob();
          }
          jobID = rj.getID().toString();
        }
      } catch (Exception e) {
        LOG.warn("Failed while cleaning up ", e);
      } finally {
        HadoopJobExecHelper.runningJobs.remove(rj);
      }
    }

    // get the list of Dynamic partition paths
    try {
      if (rj != null) {
        if (mWork.getAliasToWork() != null) {
          for (Operator<? extends OperatorDesc> op : mWork.getAliasToWork().values()) {
            op.jobClose(job, success);
          }
        }
        if (rWork != null) {
          rWork.getReducer().jobClose(job, success);
        }
      }
    } catch (Exception e) {
      // jobClose needs to execute successfully otherwise fail task
      if (success) {
        success = false;
        returnVal = 3;
        String mesg = "Job Commit failed with exception '" + Utilities.getNameMessage(e) + "'";
        console.printError(mesg, "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
      }
    }

    return (returnVal);
  }
Пример #4
0
  @Override
  public int execute(DriverContext driverContext) {

    Context ctx = driverContext.getCtx();
    boolean ctxCreated = false;

    try {
      if (ctx == null) {
        ctx = new Context(conf);
        ctxCreated = true;
      }

      // estimate number of reducers
      setNumberOfReducers();

      // auto-determine local mode if allowed
      if (!ctx.isLocalOnlyExecutionMode() && conf.getBoolVar(HiveConf.ConfVars.LOCALMODEAUTO)) {

        if (inputSummary == null) {
          inputSummary = Utilities.getInputSummary(driverContext.getCtx(), work, null);
        }

        // set the values of totalInputFileSize and totalInputNumFiles, estimating them
        // if percentage block sampling is being used
        estimateInputSize();

        // at this point the number of reducers is precisely defined in the plan
        int numReducers = work.getNumReduceTasks();

        if (LOG.isDebugEnabled()) {
          LOG.debug(
              "Task: "
                  + getId()
                  + ", Summary: "
                  + totalInputFileSize
                  + ","
                  + totalInputNumFiles
                  + ","
                  + numReducers);
        }

        String reason =
            MapRedTask.isEligibleForLocalMode(
                conf, numReducers, totalInputFileSize, totalInputNumFiles);
        if (reason == null) {
          // clone configuration before modifying it on per-task basis
          cloneConf();
          conf.setVar(HiveConf.ConfVars.HADOOPJT, "local");
          console.printInfo("Selecting local mode for task: " + getId());
          this.setLocalMode(true);
        } else {
          console.printInfo("Cannot run job locally: " + reason);
          this.setLocalMode(false);
        }
      }

      runningViaChild =
          "local".equals(conf.getVar(HiveConf.ConfVars.HADOOPJT))
              || conf.getBoolVar(HiveConf.ConfVars.SUBMITVIACHILD);

      if (!runningViaChild) {
        // we are not running this mapred task via child jvm
        // so directly invoke ExecDriver
        return super.execute(driverContext);
      }

      // we need to edit the configuration to setup cmdline. clone it first
      cloneConf();

      // propagate input format if necessary
      super.setInputAttributes(conf);

      // enable assertion
      String hadoopExec = conf.getVar(HiveConf.ConfVars.HADOOPBIN);
      String hiveJar = conf.getJar();

      String libJarsOption;
      String addedJars = Utilities.getResourceFiles(conf, SessionState.ResourceType.JAR);
      conf.setVar(ConfVars.HIVEADDEDJARS, addedJars);
      String auxJars = conf.getAuxJars();
      // Put auxjars and addedjars together into libjars
      if (StringUtils.isEmpty(addedJars)) {
        if (StringUtils.isEmpty(auxJars)) {
          libJarsOption = " ";
        } else {
          libJarsOption = " -libjars " + auxJars + " ";
        }
      } else {
        if (StringUtils.isEmpty(auxJars)) {
          libJarsOption = " -libjars " + addedJars + " ";
        } else {
          libJarsOption = " -libjars " + addedJars + "," + auxJars + " ";
        }
      }
      // Generate the hiveConfArgs after potentially adding the jars
      String hiveConfArgs = generateCmdLine(conf);

      // write out the plan to a local file
      Path planPath = new Path(ctx.getLocalTmpFileURI(), "plan.xml");
      OutputStream out = FileSystem.getLocal(conf).create(planPath);
      MapredWork plan = getWork();
      LOG.info("Generating plan file " + planPath.toString());
      Utilities.serializeMapRedWork(plan, out);

      String isSilent = "true".equalsIgnoreCase(System.getProperty("test.silent")) ? "-nolog" : "";

      String jarCmd;
      if (ShimLoader.getHadoopShims().usesJobShell()) {
        jarCmd = libJarsOption + hiveJar + " " + ExecDriver.class.getName();
      } else {
        jarCmd = hiveJar + " " + ExecDriver.class.getName() + libJarsOption;
      }

      String cmdLine =
          hadoopExec
              + " jar "
              + jarCmd
              + " -plan "
              + planPath.toString()
              + " "
              + isSilent
              + " "
              + hiveConfArgs;

      String workDir = (new File(".")).getCanonicalPath();
      String files = Utilities.getResourceFiles(conf, SessionState.ResourceType.FILE);
      if (!files.isEmpty()) {
        cmdLine = cmdLine + " -files " + files;

        workDir = (new Path(ctx.getLocalTmpFileURI())).toUri().getPath();

        if (!(new File(workDir)).mkdir()) {
          throw new IOException("Cannot create tmp working dir: " + workDir);
        }

        for (String f : StringUtils.split(files, ',')) {
          Path p = new Path(f);
          String target = p.toUri().getPath();
          String link = workDir + Path.SEPARATOR + p.getName();
          if (FileUtil.symLink(target, link) != 0) {
            throw new IOException("Cannot link to added file: " + target + " from: " + link);
          }
        }
      }

      LOG.info("Executing: " + cmdLine);
      Process executor = null;

      // Inherit Java system variables
      String hadoopOpts;
      StringBuilder sb = new StringBuilder();
      Properties p = System.getProperties();
      for (String element : HIVE_SYS_PROP) {
        if (p.containsKey(element)) {
          sb.append(" -D" + element + "=" + p.getProperty(element));
        }
      }
      hadoopOpts = sb.toString();
      // Inherit the environment variables
      String[] env;
      Map<String, String> variables = new HashMap(System.getenv());
      // The user can specify the hadoop memory

      if ("local".equals(conf.getVar(HiveConf.ConfVars.HADOOPJT))) {
        // if we are running in local mode - then the amount of memory used
        // by the child jvm can no longer default to the memory used by the
        // parent jvm
        int hadoopMem = conf.getIntVar(HiveConf.ConfVars.HIVEHADOOPMAXMEM);
        if (hadoopMem == 0) {
          // remove env var that would default child jvm to use parent's memory
          // as default. child jvm would use default memory for a hadoop client
          variables.remove(HADOOP_MEM_KEY);
        } else {
          // user specified the memory for local mode hadoop run
          variables.put(HADOOP_MEM_KEY, String.valueOf(hadoopMem));
        }
      } else {
        // nothing to do - we are not running in local mode - only submitting
        // the job via a child process. in this case it's appropriate that the
        // child jvm use the same memory as the parent jvm
      }

      if (variables.containsKey(HADOOP_OPTS_KEY)) {
        variables.put(HADOOP_OPTS_KEY, variables.get(HADOOP_OPTS_KEY) + hadoopOpts);
      } else {
        variables.put(HADOOP_OPTS_KEY, hadoopOpts);
      }

      if (variables.containsKey(HIVE_DEBUG_RECURSIVE)) {
        configureDebugVariablesForChildJVM(variables);
      }

      env = new String[variables.size()];
      int pos = 0;
      for (Map.Entry<String, String> entry : variables.entrySet()) {
        String name = entry.getKey();
        String value = entry.getValue();
        env[pos++] = name + "=" + value;
      }
      // Run ExecDriver in another JVM
      executor = Runtime.getRuntime().exec(cmdLine, env, new File(workDir));

      StreamPrinter outPrinter =
          new StreamPrinter(
              executor.getInputStream(), null, SessionState.getConsole().getChildOutStream());
      StreamPrinter errPrinter =
          new StreamPrinter(
              executor.getErrorStream(), null, SessionState.getConsole().getChildErrStream());

      outPrinter.start();
      errPrinter.start();

      int exitVal = jobExecHelper.progressLocal(executor, getId());

      if (exitVal != 0) {
        LOG.error("Execution failed with exit status: " + exitVal);
      } else {
        LOG.info("Execution completed successfully");
      }

      return exitVal;
    } catch (Exception e) {
      e.printStackTrace();
      LOG.error("Exception: " + e.getMessage());
      return (1);
    } finally {
      try {
        // creating the context can create a bunch of files. So make
        // sure to clear it out
        if (ctxCreated) {
          ctx.clear();
        }

      } catch (Exception e) {
        LOG.error("Exception: " + e.getMessage());
      }
    }
  }
Пример #5
0
  @Override
  /** start a new map-reduce job to do the merge, almost the same as ExecDriver. */
  public int execute(DriverContext driverContext) {
    HiveConf.setVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT, CombineHiveInputFormat.class.getName());
    success = true;
    ShimLoader.getHadoopShims().setNullOutputFormat(job);
    job.setMapperClass(work.getMapperClass());

    Context ctx = driverContext.getCtx();
    boolean ctxCreated = false;
    try {
      if (ctx == null) {
        ctx = new Context(job);
        ctxCreated = true;
      }
    } catch (IOException e) {
      e.printStackTrace();
      console.printError(
          "Error launching map-reduce job",
          "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
      return 5;
    }

    job.setMapOutputKeyClass(NullWritable.class);
    job.setMapOutputValueClass(NullWritable.class);
    if (work.getNumMapTasks() != null) {
      job.setNumMapTasks(work.getNumMapTasks());
    }

    // zero reducers
    job.setNumReduceTasks(0);

    if (work.getMinSplitSize() != null) {
      HiveConf.setLongVar(
          job, HiveConf.ConfVars.MAPREDMINSPLITSIZE, work.getMinSplitSize().longValue());
    }

    if (work.getInputformat() != null) {
      HiveConf.setVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT, work.getInputformat());
    }

    String inpFormat = HiveConf.getVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT);
    if ((inpFormat == null) || (!StringUtils.isNotBlank(inpFormat))) {
      inpFormat = ShimLoader.getHadoopShims().getInputFormatClassName();
    }

    LOG.info("Using " + inpFormat);

    try {
      job.setInputFormat((Class<? extends InputFormat>) (Class.forName(inpFormat)));
    } catch (ClassNotFoundException e) {
      throw new RuntimeException(e.getMessage());
    }

    String outputPath = this.work.getOutputDir();
    Path tempOutPath = Utilities.toTempPath(new Path(outputPath));
    try {
      FileSystem fs = tempOutPath.getFileSystem(job);
      if (!fs.exists(tempOutPath)) {
        fs.mkdirs(tempOutPath);
      }
    } catch (IOException e) {
      console.printError("Can't make path " + outputPath + " : " + e.getMessage());
      return 6;
    }

    RCFileBlockMergeOutputFormat.setMergeOutputPath(job, new Path(outputPath));

    job.setOutputKeyClass(NullWritable.class);
    job.setOutputValueClass(NullWritable.class);

    HiveConf.setBoolVar(
        job,
        HiveConf.ConfVars.HIVEMERGECURRENTJOBHASDYNAMICPARTITIONS,
        work.hasDynamicPartitions());

    int returnVal = 0;
    RunningJob rj = null;
    boolean noName = StringUtils.isEmpty(HiveConf.getVar(job, HiveConf.ConfVars.HADOOPJOBNAME));

    String jobName = null;
    if (noName && this.getQueryPlan() != null) {
      int maxlen = conf.getIntVar(HiveConf.ConfVars.HIVEJOBNAMELENGTH);
      jobName = Utilities.abbreviate(this.getQueryPlan().getQueryStr(), maxlen - 6);
    }

    if (noName) {
      // This is for a special case to ensure unit tests pass
      HiveConf.setVar(
          job,
          HiveConf.ConfVars.HADOOPJOBNAME,
          jobName != null ? jobName : "JOB" + Utilities.randGen.nextInt());
    }

    try {
      addInputPaths(job, work);

      Utilities.setMapRedWork(job, work, ctx.getMRTmpFileURI());

      // remove the pwd from conf file so that job tracker doesn't show this
      // logs
      String pwd = HiveConf.getVar(job, HiveConf.ConfVars.METASTOREPWD);
      if (pwd != null) {
        HiveConf.setVar(job, HiveConf.ConfVars.METASTOREPWD, "HIVE");
      }
      JobClient jc = new JobClient(job);

      String addedJars = Utilities.getResourceFiles(job, SessionState.ResourceType.JAR);
      if (!addedJars.isEmpty()) {
        job.set("tmpjars", addedJars);
      }

      // make this client wait if job trcker is not behaving well.
      Throttle.checkJobTracker(job, LOG);

      // Finally SUBMIT the JOB!
      rj = jc.submitJob(job);

      returnVal = jobExecHelper.progress(rj, jc);
      success = (returnVal == 0);

    } catch (Exception e) {
      e.printStackTrace();
      String mesg = " with exception '" + Utilities.getNameMessage(e) + "'";
      if (rj != null) {
        mesg = "Ended Job = " + rj.getJobID() + mesg;
      } else {
        mesg = "Job Submission failed" + mesg;
      }

      // Has to use full name to make sure it does not conflict with
      // org.apache.commons.lang.StringUtils
      console.printError(mesg, "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));

      success = false;
      returnVal = 1;
    } finally {
      try {
        if (ctxCreated) {
          ctx.clear();
        }
        if (rj != null) {
          if (returnVal != 0) {
            rj.killJob();
          }
          HadoopJobExecHelper.runningJobKillURIs.remove(rj.getJobID());
          jobID = rj.getID().toString();
        }
        RCFileMergeMapper.jobClose(outputPath, success, job, console);
      } catch (Exception e) {
      }
    }

    return (returnVal);
  }