Examples with RCFileMergeDesc org.apache.hadoop.hive.ql.plan.RCFileMergeDesc used on opensource projects

Example 1 with RCFileMergeDesc

use of org.apache.hadoop.hive.ql.plan.RCFileMergeDesc in project hive by apache.

the class GenMapRedUtils method createMergeTask.

/**
   * Create a block level merge task for RCFiles or stripe level merge task for
   * ORCFiles
   *
   * @param fsInputDesc
   * @param finalName
   * @param ctx
   * @param inputFormatClass
   * @return MergeWork if table is stored as RCFile or ORCFile,
   *         null otherwise
   */
public static MapWork createMergeTask(FileSinkDesc fsInputDesc, Path finalName, boolean hasDynamicPartitions, CompilationOpContext ctx) throws SemanticException {
    Path inputDir = fsInputDesc.getFinalDirName();
    TableDesc tblDesc = fsInputDesc.getTableInfo();
    List<Path> inputDirs = new ArrayList<Path>(1);
    ArrayList<String> inputDirstr = new ArrayList<String>(1);
    // in case of dynamic partitioning and list bucketing
    if (!hasDynamicPartitions && !GenMapRedUtils.isSkewedStoredAsDirs(fsInputDesc)) {
        inputDirs.add(inputDir);
    }
    inputDirstr.add(inputDir.toString());
    // internal input format class for CombineHiveInputFormat
    final Class<? extends InputFormat> internalIFClass;
    if (tblDesc.getInputFileFormatClass().equals(RCFileInputFormat.class)) {
        internalIFClass = RCFileBlockMergeInputFormat.class;
    } else if (tblDesc.getInputFileFormatClass().equals(OrcInputFormat.class)) {
        internalIFClass = OrcFileStripeMergeInputFormat.class;
    } else {
        throw new SemanticException("createMergeTask called on a table with file" + " format other than RCFile or ORCFile");
    }
    // create the merge file work
    MergeFileWork work = new MergeFileWork(inputDirs, finalName, hasDynamicPartitions, tblDesc.getInputFileFormatClass().getName());
    LinkedHashMap<Path, ArrayList<String>> pathToAliases = new LinkedHashMap<>();
    pathToAliases.put(inputDir, inputDirstr);
    work.setMapperCannotSpanPartns(true);
    work.setPathToAliases(pathToAliases);
    PartitionDesc pDesc = new PartitionDesc(tblDesc, null);
    pDesc.setInputFileFormatClass(internalIFClass);
    work.addPathToPartitionInfo(inputDir, pDesc);
    work.setListBucketingCtx(fsInputDesc.getLbCtx());
    // create alias to work which contains the merge operator
    LinkedHashMap<String, Operator<? extends OperatorDesc>> aliasToWork = new LinkedHashMap<String, Operator<? extends OperatorDesc>>();
    Operator<? extends OperatorDesc> mergeOp = null;
    final FileMergeDesc fmd;
    if (tblDesc.getInputFileFormatClass().equals(RCFileInputFormat.class)) {
        fmd = new RCFileMergeDesc();
    } else {
        fmd = new OrcFileMergeDesc();
    }
    fmd.setDpCtx(fsInputDesc.getDynPartCtx());
    fmd.setOutputPath(finalName);
    fmd.setHasDynamicPartitions(work.hasDynamicPartitions());
    fmd.setListBucketingAlterTableConcatenate(work.isListBucketingAlterTableConcatenate());
    int lbLevel = work.getListBucketingCtx() == null ? 0 : work.getListBucketingCtx().calculateListBucketingLevel();
    fmd.setListBucketingDepth(lbLevel);
    mergeOp = OperatorFactory.get(ctx, fmd);
    aliasToWork.put(inputDir.toString(), mergeOp);
    work.setAliasToWork(aliasToWork);
    return work;
}

Example 2 with RCFileMergeDesc

use of org.apache.hadoop.hive.ql.plan.RCFileMergeDesc in project hive by apache.

the class DDLTask method mergeFiles.

/**
   * First, make sure the source table/partition is not
   * archived/indexes/non-rcfile. If either of these is true, throw an
   * exception.
   *
   * The way how it does the merge is to create a BlockMergeTask from the
   * mergeFilesDesc.
   *
   * @param db
   * @param mergeFilesDesc
   * @return
   * @throws HiveException
   */
private int mergeFiles(Hive db, AlterTablePartMergeFilesDesc mergeFilesDesc, DriverContext driverContext) throws HiveException {
    ListBucketingCtx lbCtx = mergeFilesDesc.getLbCtx();
    boolean lbatc = lbCtx == null ? false : lbCtx.isSkewedStoredAsDir();
    int lbd = lbCtx == null ? 0 : lbCtx.calculateListBucketingLevel();
    // merge work only needs input and output.
    MergeFileWork mergeWork = new MergeFileWork(mergeFilesDesc.getInputDir(), mergeFilesDesc.getOutputDir(), mergeFilesDesc.getInputFormatClass().getName());
    LinkedHashMap<Path, ArrayList<String>> pathToAliases = new LinkedHashMap<>();
    ArrayList<String> inputDirstr = new ArrayList<String>(1);
    inputDirstr.add(mergeFilesDesc.getInputDir().toString());
    pathToAliases.put(mergeFilesDesc.getInputDir().get(0), inputDirstr);
    mergeWork.setPathToAliases(pathToAliases);
    mergeWork.setListBucketingCtx(mergeFilesDesc.getLbCtx());
    mergeWork.resolveConcatenateMerge(db.getConf());
    mergeWork.setMapperCannotSpanPartns(true);
    mergeWork.setSourceTableInputFormat(mergeFilesDesc.getInputFormatClass().getName());
    final FileMergeDesc fmd;
    if (mergeFilesDesc.getInputFormatClass().equals(RCFileInputFormat.class)) {
        fmd = new RCFileMergeDesc();
    } else {
        // safe to assume else is ORC as semantic analyzer will check for RC/ORC
        fmd = new OrcFileMergeDesc();
    }
    fmd.setDpCtx(null);
    fmd.setHasDynamicPartitions(false);
    fmd.setListBucketingAlterTableConcatenate(lbatc);
    fmd.setListBucketingDepth(lbd);
    fmd.setOutputPath(mergeFilesDesc.getOutputDir());
    CompilationOpContext opContext = driverContext.getCtx().getOpContext();
    Operator<? extends OperatorDesc> mergeOp = OperatorFactory.get(opContext, fmd);
    LinkedHashMap<String, Operator<? extends OperatorDesc>> aliasToWork = new LinkedHashMap<String, Operator<? extends OperatorDesc>>();
    aliasToWork.put(mergeFilesDesc.getInputDir().toString(), mergeOp);
    mergeWork.setAliasToWork(aliasToWork);
    DriverContext driverCxt = new DriverContext();
    Task task;
    if (conf.getVar(ConfVars.HIVE_EXECUTION_ENGINE).equals("tez")) {
        TezWork tezWork = new TezWork(conf.getVar(HiveConf.ConfVars.HIVEQUERYID), conf);
        mergeWork.setName("File Merge");
        tezWork.add(mergeWork);
        task = new TezTask();
        task.setWork(tezWork);
    } else {
        task = new MergeFileTask();
        task.setWork(mergeWork);
    }
    // initialize the task and execute
    task.initialize(queryState, getQueryPlan(), driverCxt, opContext);
    subtask = task;
    int ret = task.execute(driverCxt);
    if (subtask.getException() != null) {
        setException(subtask.getException());
    }
    return ret;
}