Examples of org.apache.hadoop.hive.ql.plan.JoinDesc

org.apache.hadoop.hive.ql.plan.JoinDesc
Join operator Descriptor implementation.

    for (int i = 0; i < join.getJoinCond().length; i++) {
      JoinCond condn = join.getJoinCond()[i];
      joinCondns[i] = new JoinCondDesc(condn);
    }


    JoinDesc desc = new JoinDesc(exprMap, outputColumnNames,
        join.getNoOuterJoin(), joinCondns, filterMap);
    desc.setReversedExprs(reversedExprs);
    JoinOperator joinOp = (JoinOperator) OperatorFactory.getAndMakeChild(desc,
        new RowSchema(outputRS.getColumnInfos()), rightOps);
    joinOp.setColumnExprMap(colExprMap);
    joinOp.setPosToAliasMap(posToAliasMap);


    if (join.getNullSafes() != null) {
      boolean[] nullsafes = new boolean[join.getNullSafes().size()];
      for (int i = 0; i < nullsafes.length; i++) {
        nullsafes[i] = join.getNullSafes().get(i);
      }
      desc.setNullSafes(nullsafes);
    }
    return putOpInsertMap(joinOp, outputRS);
  }

View Full Code Here

      return;
    }


    String baseTmpDir = parseCtx.getContext().getMRTmpFileURI();


    JoinDesc joinDescriptor = joinOp.getConf();
    Map<Byte, List<ExprNodeDesc>> joinValues = joinDescriptor.getExprs();
    int numAliases = joinValues.size();


    Map<Byte, String> bigKeysDirMap = new HashMap<Byte, String>();
    Map<Byte, Map<Byte, String>> smallKeysDirMap = new HashMap<Byte, Map<Byte, String>>();
    Map<Byte, String> skewJoinJobResultsDir = new HashMap<Byte, String>();
    Byte[] tags = joinDescriptor.getTagOrder();
    for (int i = 0; i < numAliases; i++) {
      Byte alias = tags[i];
      String bigKeysDir = getBigKeysDir(baseTmpDir, alias);
      bigKeysDirMap.put(alias, bigKeysDir);
      Map<Byte, String> smallKeysMap = new HashMap<Byte, String>();
      smallKeysDirMap.put(alias, smallKeysMap);
      for (Byte src2 : tags) {
        if (!src2.equals(alias)) {
          smallKeysMap.put(src2, getSmallKeysDir(baseTmpDir, alias, src2));
        }
      }
      skewJoinJobResultsDir.put(alias, getBigKeysSkewJoinResultDir(baseTmpDir,
          alias));
    }


    joinDescriptor.setHandleSkewJoin(true);
    joinDescriptor.setBigKeysDirMap(bigKeysDirMap);
    joinDescriptor.setSmallKeysDirMap(smallKeysDirMap);
    joinDescriptor.setSkewKeyDefinition(HiveConf.getIntVar(parseCtx.getConf(),
        HiveConf.ConfVars.HIVESKEWJOINKEY));


    HashMap<String, Task<? extends Serializable>> bigKeysDirToTaskMap =
      new HashMap<String, Task<? extends Serializable>>();
    List<Serializable> listWorks = new ArrayList<Serializable>();
    List<Task<? extends Serializable>> listTasks = new ArrayList<Task<? extends Serializable>>();
    MapredWork currPlan = (MapredWork) currTask.getWork();


    TableDesc keyTblDesc = (TableDesc) currPlan.getKeyDesc().clone();
    List<String> joinKeys = Utilities
        .getColumnNames(keyTblDesc.getProperties());
    List<String> joinKeyTypes = Utilities.getColumnTypes(keyTblDesc
        .getProperties());


    Map<Byte, TableDesc> tableDescList = new HashMap<Byte, TableDesc>();
    Map<Byte, List<ExprNodeDesc>> newJoinValues = new HashMap<Byte, List<ExprNodeDesc>>();
    Map<Byte, List<ExprNodeDesc>> newJoinKeys = new HashMap<Byte, List<ExprNodeDesc>>();
    // used for create mapJoinDesc, should be in order
    List<TableDesc> newJoinValueTblDesc = new ArrayList<TableDesc>();


    for (Byte tag : tags) {
      newJoinValueTblDesc.add(null);
    }


    for (int i = 0; i < numAliases; i++) {
      Byte alias = tags[i];
      List<ExprNodeDesc> valueCols = joinValues.get(alias);
      String colNames = "";
      String colTypes = "";
      int columnSize = valueCols.size();
      List<ExprNodeDesc> newValueExpr = new ArrayList<ExprNodeDesc>();
      List<ExprNodeDesc> newKeyExpr = new ArrayList<ExprNodeDesc>();


      boolean first = true;
      for (int k = 0; k < columnSize; k++) {
        TypeInfo type = valueCols.get(k).getTypeInfo();
        String newColName = i + "_VALUE_" + k; // any name, it does not matter.
        newValueExpr
            .add(new ExprNodeColumnDesc(type, newColName, "" + i, false));
        if (!first) {
          colNames = colNames + ",";
          colTypes = colTypes + ",";
        }
        first = false;
        colNames = colNames + newColName;
        colTypes = colTypes + valueCols.get(k).getTypeString();
      }


      // we are putting join keys at last part of the spilled table
      for (int k = 0; k < joinKeys.size(); k++) {
        if (!first) {
          colNames = colNames + ",";
          colTypes = colTypes + ",";
        }
        first = false;
        colNames = colNames + joinKeys.get(k);
        colTypes = colTypes + joinKeyTypes.get(k);
        newKeyExpr.add(new ExprNodeColumnDesc(TypeInfoFactory
            .getPrimitiveTypeInfo(joinKeyTypes.get(k)), joinKeys.get(k),
            "" + i, false));
      }


      newJoinValues.put(alias, newValueExpr);
      newJoinKeys.put(alias, newKeyExpr);
      tableDescList.put(alias, Utilities.getTableDesc(colNames, colTypes));


      // construct value table Desc
      String valueColNames = "";
      String valueColTypes = "";
      first = true;
      for (int k = 0; k < columnSize; k++) {
        String newColName = i + "_VALUE_" + k; // any name, it does not matter.
        if (!first) {
          valueColNames = valueColNames + ",";
          valueColTypes = valueColTypes + ",";
        }
        valueColNames = valueColNames + newColName;
        valueColTypes = valueColTypes + valueCols.get(k).getTypeString();
        first = false;
      }
      newJoinValueTblDesc.set(Byte.valueOf((byte) i), Utilities.getTableDesc(
          valueColNames, valueColTypes));
    }


    joinDescriptor.setSkewKeysValuesTables(tableDescList);
    joinDescriptor.setKeyTableDesc(keyTblDesc);


    for (int i = 0; i < numAliases - 1; i++) {
      Byte src = tags[i];
      MapredWork newPlan = PlanUtils.getMapRedWork();


      // This code has been only added for testing
      boolean mapperCannotSpanPartns =
        parseCtx.getConf().getBoolVar(
          HiveConf.ConfVars.HIVE_MAPPER_CANNOT_SPAN_MULTIPLE_PARTITIONS);
      newPlan.setMapperCannotSpanPartns(mapperCannotSpanPartns);


      MapredWork clonePlan = null;
      try {
        String xmlPlan = currPlan.toXML();
        StringBuilder sb = new StringBuilder(xmlPlan);
        ByteArrayInputStream bis;
        bis = new ByteArrayInputStream(sb.toString().getBytes("UTF-8"));
        clonePlan = Utilities.deserializeMapRedWork(bis, parseCtx.getConf());
      } catch (UnsupportedEncodingException e) {
        throw new SemanticException(e);
      }


      Operator<? extends Serializable>[] parentOps = new TableScanOperator[tags.length];
      for (int k = 0; k < tags.length; k++) {
        Operator<? extends Serializable> ts = OperatorFactory.get(
            TableScanDesc.class, (RowSchema) null);
        ((TableScanOperator)ts).setTableDesc(tableDescList.get((byte)k));
        parentOps[k] = ts;
      }
      Operator<? extends Serializable> tblScan_op = parentOps[i];


      ArrayList<String> aliases = new ArrayList<String>();
      String alias = src.toString();
      aliases.add(alias);
      String bigKeyDirPath = bigKeysDirMap.get(src);
      newPlan.getPathToAliases().put(bigKeyDirPath, aliases);








      newPlan.getAliasToWork().put(alias, tblScan_op);
      PartitionDesc part = new PartitionDesc(tableDescList.get(src), null);




      newPlan.getPathToPartitionInfo().put(bigKeyDirPath, part);
      newPlan.getAliasToPartnInfo().put(alias, part);


      Operator<? extends Serializable> reducer = clonePlan.getReducer();
      assert reducer instanceof JoinOperator;
      JoinOperator cloneJoinOp = (JoinOperator) reducer;


      String dumpFilePrefix = "mapfile"+PlanUtils.getCountForMapJoinDumpFilePrefix();
      MapJoinDesc mapJoinDescriptor = new MapJoinDesc(newJoinKeys, keyTblDesc,
          newJoinValues, newJoinValueTblDesc, newJoinValueTblDesc,joinDescriptor
          .getOutputColumnNames(), i, joinDescriptor.getConds(),
          joinDescriptor.getFilters(), joinDescriptor.getNoOuterJoin(), dumpFilePrefix);
      mapJoinDescriptor.setTagOrder(tags);
      mapJoinDescriptor.setHandleSkewJoin(false);


      MapredLocalWork localPlan = new MapredLocalWork(
          new LinkedHashMap<String, Operator<? extends Serializable>>(),

View Full Code Here

    for (int i = 0; i < join.getJoinCond().length; i++) {
      JoinCond condn = join.getJoinCond()[i];
      joinCondns[i] = new JoinCondDesc(condn);
    }


    JoinDesc desc = new JoinDesc(exprMap, outputColumnNames,
        join.getNoOuterJoin(), joinCondns, filterMap);
    desc.setReversedExprs(reversedExprs);
    desc.setFilterMap(join.getFilterMap());


    JoinOperator joinOp = (JoinOperator) OperatorFactory.getAndMakeChild(desc,
        new RowSchema(outputRS.getColumnInfos()), rightOps);
    joinOp.setColumnExprMap(colExprMap);
    joinOp.setPosToAliasMap(posToAliasMap);


    if (join.getNullSafes() != null) {
      boolean[] nullsafes = new boolean[join.getNullSafes().size()];
      for (int i = 0; i < nullsafes.length; i++) {
        nullsafes[i] = join.getNullSafes().get(i);
      }
      desc.setNullSafes(nullsafes);
    }
    return putOpInsertMap(joinOp, outputRS);
  }

View Full Code Here

    // get parseCtx for this Join Operator
    ParseContext parseCtx = physicalContext.getParseContext();
    QBJoinTree joinTree = parseCtx.getJoinContext().get(joinOp);


    // start to generate multiple map join tasks
    JoinDesc joinDesc = joinOp.getConf();


    if (aliasToSize == null) {
      aliasToSize = new HashMap<String, Long>();
    }


    try {
      long aliasTotalKnownInputSize =
          getTotalKnownInputSize(context, currWork, pathToAliases, aliasToSize);


      Set<Integer> bigTableCandidates = MapJoinProcessor.getBigTableCandidates(joinDesc
          .getConds());


      // no table could be the big table; there is no need to convert
      if (bigTableCandidates.isEmpty()) {
        return null;

View Full Code Here

      throw new SemanticException(e.getMessage());
    }


    BigTableSelectorForAutoSMJ bigTableMatcher =
      (BigTableSelectorForAutoSMJ) ReflectionUtils.newInstance(bigTableMatcherClass, null);
    JoinDesc joinDesc = joinOp.getConf();
    JoinCondDesc[] joinCondns = joinDesc.getConds();
    Set<Integer> joinCandidates = MapJoinProcessor.getBigTableCandidates(joinCondns);
    if (joinCandidates.isEmpty()) {
      // This is a full outer join. This can never be a map-join
      // of any type. So return false.
      return false;

View Full Code Here

    JoinOperator op, QBJoinTree joinTree, int mapJoinPos, boolean noCheckOuterJoin,
    boolean validateMapJoinTree)
    throws SemanticException {


    // outer join cannot be performed on a table which is being cached
    JoinDesc desc = op.getConf();
    JoinCondDesc[] condns = desc.getConds();
    Byte[] tagOrder = desc.getTagOrder();


    if (!noCheckOuterJoin) {
      if (checkMapJoin(mapJoinPos, condns) < 0) {
        throw new SemanticException(ErrorMsg.NO_OUTER_MAPJOIN.getMsg());
      }

View Full Code Here

  public static MapJoinOperator convertJoinOpMapJoinOp(HiveConf hconf,
      LinkedHashMap<Operator<? extends OperatorDesc>, OpParseContext> opParseCtxMap,
      JoinOperator op, QBJoinTree joinTree, int mapJoinPos, boolean noCheckOuterJoin)
      throws SemanticException {


    JoinDesc desc = op.getConf();
    JoinCondDesc[] condns = desc.getConds();
    Byte[] tagOrder = desc.getTagOrder();


    // outer join cannot be performed on a table which is being cached
    if (!noCheckOuterJoin) {
      if (checkMapJoin(mapJoinPos, condns) < 0) {
        throw new SemanticException(ErrorMsg.NO_OUTER_MAPJOIN.getMsg());
      }
    }


    Map<Byte, List<ExprNodeDesc>> keyExprMap = new HashMap<Byte, List<ExprNodeDesc>>();


    // Walk over all the sources (which are guaranteed to be reduce sink
    // operators).
    // The join outputs a concatenation of all the inputs.
    QBJoinTree leftSrc = joinTree.getJoinSrc();
    List<Operator<? extends OperatorDesc>> oldReduceSinkParentOps =
        new ArrayList<Operator<? extends OperatorDesc>>();
    if (leftSrc != null) {
      // assert mapJoinPos == 0;
      Operator<? extends OperatorDesc> parentOp = op.getParentOperators().get(0);
      assert parentOp.getParentOperators().size() == 1;
      oldReduceSinkParentOps.add(parentOp);
    }




    byte pos = 0;
    for (String src : joinTree.getBaseSrc()) {
      if (src != null) {
        Operator<? extends OperatorDesc> parentOp = op.getParentOperators().get(pos);
        assert parentOp.getParentOperators().size() == 1;
        oldReduceSinkParentOps.add(parentOp);
      }
      pos++;
    }


    // get the join keys from old parent ReduceSink operators
    for (pos = 0; pos < op.getParentOperators().size(); pos++) {
      ReduceSinkOperator parent = (ReduceSinkOperator) oldReduceSinkParentOps.get(pos);
      ReduceSinkDesc rsconf = parent.getConf();
      List<ExprNodeDesc> keys = rsconf.getKeyCols();
      keyExprMap.put(pos, keys);
    }


    List<ExprNodeDesc> keyCols = keyExprMap.get(Byte.valueOf((byte) 0));
    StringBuilder keyOrder = new StringBuilder();
    for (int i = 0; i < keyCols.size(); i++) {
      keyOrder.append("+");
    }


    Map<String, ExprNodeDesc> colExprMap = op.getColumnExprMap();
    List<ColumnInfo> schema = new ArrayList<ColumnInfo>(op.getSchema().getSignature());
    Map<Byte, List<ExprNodeDesc>> valueExprs = op.getConf().getExprs();
    Map<Byte, List<ExprNodeDesc>> newValueExprs = new HashMap<Byte, List<ExprNodeDesc>>();
    for (Map.Entry<Byte, List<ExprNodeDesc>> entry : valueExprs.entrySet()) {
      byte tag = entry.getKey();
      Operator<?> terminal = oldReduceSinkParentOps.get(tag);


      List<ExprNodeDesc> values = entry.getValue();
      List<ExprNodeDesc> newValues = ExprNodeDescUtils.backtrack(values, op, terminal);
      newValueExprs.put(tag, newValues);
      for (int i = 0; i < schema.size(); i++) {
        ColumnInfo column = schema.get(i);
        if (column == null) {
          continue;
        }
        ExprNodeDesc expr = colExprMap.get(column.getInternalName());
        int index = ExprNodeDescUtils.indexOf(expr, values);
        if (index >= 0) {
          colExprMap.put(column.getInternalName(), newValues.get(index));
          schema.set(i, null);
        }
      }
    }


    // construct valueTableDescs and valueFilteredTableDescs
    List<TableDesc> valueTableDescs = new ArrayList<TableDesc>();
    List<TableDesc> valueFiltedTableDescs = new ArrayList<TableDesc>();
    int[][] filterMap = desc.getFilterMap();
    for (pos = 0; pos < op.getParentOperators().size(); pos++) {
      List<ExprNodeDesc> valueCols = newValueExprs.get(Byte.valueOf((byte) pos));
      int length = valueCols.size();
      List<ExprNodeDesc> valueFilteredCols = new ArrayList<ExprNodeDesc>(length);
      // deep copy expr node desc
      for (int i = 0; i < length; i++) {
        valueFilteredCols.add(valueCols.get(i).clone());
      }
      if (filterMap != null && filterMap[pos] != null && pos != mapJoinPos) {
        ExprNodeColumnDesc isFilterDesc = new ExprNodeColumnDesc(TypeInfoFactory
            .getPrimitiveTypeInfo(serdeConstants.SMALLINT_TYPE_NAME), "filter", "filter", false);
        valueFilteredCols.add(isFilterDesc);
      }




      keyOrder = new StringBuilder();
      for (int i = 0; i < valueCols.size(); i++) {
        keyOrder.append("+");
      }


      TableDesc valueTableDesc = PlanUtils.getMapJoinValueTableDesc(PlanUtils
          .getFieldSchemasFromColumnList(valueCols, "mapjoinvalue"));
      TableDesc valueFilteredTableDesc = PlanUtils.getMapJoinValueTableDesc(PlanUtils
          .getFieldSchemasFromColumnList(valueFilteredCols, "mapjoinvalue"));


      valueTableDescs.add(valueTableDesc);
      valueFiltedTableDescs.add(valueFilteredTableDesc);
    }


    Map<Byte, List<ExprNodeDesc>> filters = desc.getFilters();
    Map<Byte, List<ExprNodeDesc>> newFilters = new HashMap<Byte, List<ExprNodeDesc>>();
    for (Map.Entry<Byte, List<ExprNodeDesc>> entry : filters.entrySet()) {
      byte srcTag = entry.getKey();
      List<ExprNodeDesc> filter = entry.getValue();


      Operator<?> terminal = op.getParentOperators().get(srcTag);
      newFilters.put(srcTag, ExprNodeDescUtils.backtrack(filter, op, terminal));
    }
    desc.setFilters(filters = newFilters);


    // create dumpfile prefix needed to create descriptor
    String dumpFilePrefix = "";
    if( joinTree.getMapAliases() != null ) {
      for(String mapAlias : joinTree.getMapAliases()) {
        dumpFilePrefix = dumpFilePrefix + mapAlias;
      }
      dumpFilePrefix = dumpFilePrefix+"-"+PlanUtils.getCountForMapJoinDumpFilePrefix();
    } else {
      dumpFilePrefix = "mapfile"+PlanUtils.getCountForMapJoinDumpFilePrefix();
    }


    List<String> outputColumnNames = op.getConf().getOutputColumnNames();
    TableDesc keyTableDesc = PlanUtils.getMapJoinKeyTableDesc(hconf,
        PlanUtils.getFieldSchemasFromColumnList(keyCols, MAPJOINKEY_FIELDPREFIX));
    JoinCondDesc[] joinCondns = op.getConf().getConds();
    MapJoinDesc mapJoinDescriptor = new MapJoinDesc(keyExprMap, keyTableDesc, newValueExprs,
        valueTableDescs, valueFiltedTableDescs, outputColumnNames, mapJoinPos, joinCondns,
        filters, op.getConf().getNoOuterJoin(), dumpFilePrefix);
    mapJoinDescriptor.setStatistics(op.getConf().getStatistics());
    mapJoinDescriptor.setTagOrder(tagOrder);
    mapJoinDescriptor.setNullSafes(desc.getNullSafes());
    mapJoinDescriptor.setFilterMap(desc.getFilterMap());


    // reduce sink row resolver used to generate map join op
    RowResolver outputRS = opParseCtxMap.get(op).getRowResolver();


    MapJoinOperator mapJoinOp = (MapJoinOperator) OperatorFactory.getAndMakeChild(

View Full Code Here

      assert(!stack.isEmpty());


      // LineageCtx
      LineageCtx lCtx = (LineageCtx) procCtx;
      JoinOperator op = (JoinOperator)nd;
      JoinDesc jd = op.getConf();


      // The input operator to the join is always a reduce sink operator
      ReduceSinkOperator inpOp = (ReduceSinkOperator)getParent(stack);
      ReduceSinkDesc rd = inpOp.getConf();
      int tag = rd.getTag();


      // Iterate over the outputs of the join operator and merge the
      // dependencies of the columns that corresponding to the tag.
      int cnt = 0;
      List<ExprNodeDesc> exprs = jd.getExprs().get((byte)tag);
      for(ColumnInfo ci : op.getSchema().getSignature()) {
        if (jd.getReversedExprs().get(ci.getInternalName()) != tag) {
          continue;
        }


        // Otherwise look up the expression corresponding to this ci
        ExprNodeDesc expr = exprs.get(cnt++);

View Full Code Here

        LOG.info("Cannot guess if CommonJoinOperator will optimize " +
            joinOp.getName() + " " + joinOp.getIdentifier());
        continue;
      }


      JoinDesc joinDesc = joinOp.getConf();
      Byte[] order = joinDesc.getTagOrder();
      int numAliases = order.length;
      Set<Integer> bigTableCandidates =
          MapJoinProcessor.getBigTableCandidates(joinDesc.getConds());
      if (bigTableCandidates.isEmpty()) {
        continue;
      }


      long ThresholdOfSmallTblSizeSum = HiveConf.getLongVar(pCtx.getConf(),

View Full Code Here

    Task<? extends Serializable> child =
        children != null && children.size() == 1 ? children.get(0) : null;


    Path baseTmpDir = parseCtx.getContext().getMRTmpPath();


    JoinDesc joinDescriptor = joinOp.getConf();
    Map<Byte, List<ExprNodeDesc>> joinValues = joinDescriptor.getExprs();
    int numAliases = joinValues.size();


    Map<Byte, Path> bigKeysDirMap = new HashMap<Byte, Path>();
    Map<Byte, Map<Byte, Path>> smallKeysDirMap = new HashMap<Byte, Map<Byte, Path>>();
    Map<Byte, Path> skewJoinJobResultsDir = new HashMap<Byte, Path>();
    Byte[] tags = joinDescriptor.getTagOrder();
    for (int i = 0; i < numAliases; i++) {
      Byte alias = tags[i];
      bigKeysDirMap.put(alias, getBigKeysDir(baseTmpDir, alias));
      Map<Byte, Path> smallKeysMap = new HashMap<Byte, Path>();
      smallKeysDirMap.put(alias, smallKeysMap);
      for (Byte src2 : tags) {
        if (!src2.equals(alias)) {
          smallKeysMap.put(src2, getSmallKeysDir(baseTmpDir, alias, src2));
        }
      }
      skewJoinJobResultsDir.put(alias, getBigKeysSkewJoinResultDir(baseTmpDir,
          alias));
    }


    joinDescriptor.setHandleSkewJoin(true);
    joinDescriptor.setBigKeysDirMap(bigKeysDirMap);
    joinDescriptor.setSmallKeysDirMap(smallKeysDirMap);
    joinDescriptor.setSkewKeyDefinition(HiveConf.getIntVar(parseCtx.getConf(),
        HiveConf.ConfVars.HIVESKEWJOINKEY));


    HashMap<Path, Task<? extends Serializable>> bigKeysDirToTaskMap =
      new HashMap<Path, Task<? extends Serializable>>();
    List<Serializable> listWorks = new ArrayList<Serializable>();
    List<Task<? extends Serializable>> listTasks = new ArrayList<Task<? extends Serializable>>();
    MapredWork currPlan = (MapredWork) currTask.getWork();


    TableDesc keyTblDesc = (TableDesc) currPlan.getReduceWork().getKeyDesc().clone();
    List<String> joinKeys = Utilities
        .getColumnNames(keyTblDesc.getProperties());
    List<String> joinKeyTypes = Utilities.getColumnTypes(keyTblDesc
        .getProperties());


    Map<Byte, TableDesc> tableDescList = new HashMap<Byte, TableDesc>();
    Map<Byte, RowSchema> rowSchemaList = new HashMap<Byte, RowSchema>();
    Map<Byte, List<ExprNodeDesc>> newJoinValues = new HashMap<Byte, List<ExprNodeDesc>>();
    Map<Byte, List<ExprNodeDesc>> newJoinKeys = new HashMap<Byte, List<ExprNodeDesc>>();
    // used for create mapJoinDesc, should be in order
    List<TableDesc> newJoinValueTblDesc = new ArrayList<TableDesc>();


    for (Byte tag : tags) {
      newJoinValueTblDesc.add(null);
    }


    for (int i = 0; i < numAliases; i++) {
      Byte alias = tags[i];
      List<ExprNodeDesc> valueCols = joinValues.get(alias);
      String colNames = "";
      String colTypes = "";
      int columnSize = valueCols.size();
      List<ExprNodeDesc> newValueExpr = new ArrayList<ExprNodeDesc>();
      List<ExprNodeDesc> newKeyExpr = new ArrayList<ExprNodeDesc>();
      ArrayList<ColumnInfo> columnInfos = new ArrayList<ColumnInfo>();


      boolean first = true;
      for (int k = 0; k < columnSize; k++) {
        TypeInfo type = valueCols.get(k).getTypeInfo();
        String newColName = i + "_VALUE_" + k; // any name, it does not matter.
        ColumnInfo columnInfo = new ColumnInfo(newColName, type, alias.toString(), false);
        columnInfos.add(columnInfo);
        newValueExpr.add(new ExprNodeColumnDesc(
            columnInfo.getType(), columnInfo.getInternalName(),
            columnInfo.getTabAlias(), false));
        if (!first) {
          colNames = colNames + ",";
          colTypes = colTypes + ",";
        }
        first = false;
        colNames = colNames + newColName;
        colTypes = colTypes + valueCols.get(k).getTypeString();
      }


      // we are putting join keys at last part of the spilled table
      for (int k = 0; k < joinKeys.size(); k++) {
        if (!first) {
          colNames = colNames + ",";
          colTypes = colTypes + ",";
        }
        first = false;
        colNames = colNames + joinKeys.get(k);
        colTypes = colTypes + joinKeyTypes.get(k);
        ColumnInfo columnInfo = new ColumnInfo(joinKeys.get(k), TypeInfoFactory
            .getPrimitiveTypeInfo(joinKeyTypes.get(k)), alias.toString(), false);
        columnInfos.add(columnInfo);
        newKeyExpr.add(new ExprNodeColumnDesc(
            columnInfo.getType(), columnInfo.getInternalName(),
            columnInfo.getTabAlias(), false));
      }


      newJoinValues.put(alias, newValueExpr);
      newJoinKeys.put(alias, newKeyExpr);
      tableDescList.put(alias, Utilities.getTableDesc(colNames, colTypes));
      rowSchemaList.put(alias, new RowSchema(columnInfos));


      // construct value table Desc
      String valueColNames = "";
      String valueColTypes = "";
      first = true;
      for (int k = 0; k < columnSize; k++) {
        String newColName = i + "_VALUE_" + k; // any name, it does not matter.
        if (!first) {
          valueColNames = valueColNames + ",";
          valueColTypes = valueColTypes + ",";
        }
        valueColNames = valueColNames + newColName;
        valueColTypes = valueColTypes + valueCols.get(k).getTypeString();
        first = false;
      }
      newJoinValueTblDesc.set(Byte.valueOf((byte) i), Utilities.getTableDesc(
          valueColNames, valueColTypes));
    }


    joinDescriptor.setSkewKeysValuesTables(tableDescList);
    joinDescriptor.setKeyTableDesc(keyTblDesc);


    for (int i = 0; i < numAliases - 1; i++) {
      Byte src = tags[i];
      MapWork newPlan = PlanUtils.getMapRedWork().getMapWork();


      // This code has been only added for testing
      boolean mapperCannotSpanPartns =
        parseCtx.getConf().getBoolVar(
          HiveConf.ConfVars.HIVE_MAPPER_CANNOT_SPAN_MULTIPLE_PARTITIONS);
      newPlan.setMapperCannotSpanPartns(mapperCannotSpanPartns);


      MapredWork clonePlan = Utilities.clonePlan(currPlan);


      Operator<? extends OperatorDesc>[] parentOps = new TableScanOperator[tags.length];
      for (int k = 0; k < tags.length; k++) {
        Operator<? extends OperatorDesc> ts =
            GenMapRedUtils.createTemporaryTableScanOperator(rowSchemaList.get((byte)k));
        ((TableScanOperator)ts).setTableDesc(tableDescList.get((byte)k));
        parentOps[k] = ts;
      }
      Operator<? extends OperatorDesc> tblScan_op = parentOps[i];


      ArrayList<String> aliases = new ArrayList<String>();
      String alias = src.toString();
      aliases.add(alias);
      Path bigKeyDirPath = bigKeysDirMap.get(src);
      newPlan.getPathToAliases().put(bigKeyDirPath.toString(), aliases);


      newPlan.getAliasToWork().put(alias, tblScan_op);
      PartitionDesc part = new PartitionDesc(tableDescList.get(src), null);


      newPlan.getPathToPartitionInfo().put(bigKeyDirPath.toString(), part);
      newPlan.getAliasToPartnInfo().put(alias, part);


      Operator<? extends OperatorDesc> reducer = clonePlan.getReduceWork().getReducer();
      assert reducer instanceof JoinOperator;
      JoinOperator cloneJoinOp = (JoinOperator) reducer;


      String dumpFilePrefix = "mapfile"+PlanUtils.getCountForMapJoinDumpFilePrefix();
      MapJoinDesc mapJoinDescriptor = new MapJoinDesc(newJoinKeys, keyTblDesc,
          newJoinValues, newJoinValueTblDesc, newJoinValueTblDesc,joinDescriptor
          .getOutputColumnNames(), i, joinDescriptor.getConds(),
          joinDescriptor.getFilters(), joinDescriptor.getNoOuterJoin(), dumpFilePrefix);
      mapJoinDescriptor.setTagOrder(tags);
      mapJoinDescriptor.setHandleSkewJoin(false);
      mapJoinDescriptor.setNullSafes(joinDescriptor.getNullSafes());


      MapredLocalWork localPlan = new MapredLocalWork(
          new LinkedHashMap<String, Operator<? extends OperatorDesc>>(),
          new LinkedHashMap<String, FetchWork>());
      Map<Byte, Path> smallTblDirs = smallKeysDirMap.get(src);

View Full Code Here

0 1 2 3 4 5 6 7

TOP

Related Classes of org.apache.hadoop.hive.ql.plan.JoinDesc

org.apache.hadoop.hive.ql.exec.SkewJoinHandler

org.apache.hadoop.hive.ql.optimizer.AbstractSMBJoinProc

org.apache.hadoop.hive.ql.optimizer.ConstantPropagateProcFactory$ConstantPropagateJoinProc

org.apache.hadoop.hive.ql.optimizer.ConvertJoinMapJoin

org.apache.hadoop.hive.ql.optimizer.correlation.CorrelationOptimizer

org.apache.hadoop.hive.ql.optimizer.lineage.OpProcFactory$JoinLineage

org.apache.hadoop.hive.ql.optimizer.MapJoinProcessor

org.apache.hadoop.hive.ql.optimizer.physical.BucketingSortingOpProcFactory$JoinInferrer

org.apache.hadoop.hive.ql.optimizer.physical.CommonJoinResolver$CommonJoinTaskDispatcher

org.apache.hadoop.hive.ql.optimizer.physical.CommonJoinTaskDispatcher

All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.