Examples with Operator org.apache.hadoop.hive.ql.exec.Operator used on opensource projects

Example 51 with Operator

use of org.apache.hadoop.hive.ql.exec.Operator in project hive by apache.

the class SemanticAnalyzer method genGroupByPlanGroupByOperator1.

/**
   * Generate the GroupByOperator for the Query Block (parseInfo.getXXX(dest)).
   * The new GroupByOperator will be a child of the reduceSinkOperatorInfo.
   *
   * @param parseInfo
   * @param dest
   * @param reduceSinkOperatorInfo
   * @param mode
   *          The mode of the aggregation (MERGEPARTIAL, PARTIAL2)
   * @param genericUDAFEvaluators
   *          The mapping from Aggregation StringTree to the
   *          genericUDAFEvaluator.
   * @param groupingSets
   *          list of grouping sets
   * @param groupingSetsPresent
   *          whether grouping sets are present in this query
   * @param groupingSetsNeedAdditionalMRJob
   *          whether grouping sets are consumed by this group by
   * @return the new GroupByOperator
   */
@SuppressWarnings("nls")
private Operator genGroupByPlanGroupByOperator1(QBParseInfo parseInfo, String dest, Operator reduceSinkOperatorInfo, GroupByDesc.Mode mode, Map<String, GenericUDAFEvaluator> genericUDAFEvaluators, List<Integer> groupingSets, boolean groupingSetsPresent, boolean groupingSetsNeedAdditionalMRJob) throws SemanticException {
    ArrayList<String> outputColumnNames = new ArrayList<String>();
    RowResolver groupByInputRowResolver = opParseCtx.get(reduceSinkOperatorInfo).getRowResolver();
    RowResolver groupByOutputRowResolver = new RowResolver();
    groupByOutputRowResolver.setIsExprResolver(true);
    ArrayList<ExprNodeDesc> groupByKeys = new ArrayList<ExprNodeDesc>();
    ArrayList<AggregationDesc> aggregations = new ArrayList<AggregationDesc>();
    List<ASTNode> grpByExprs = getGroupByForClause(parseInfo, dest);
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    for (int i = 0; i < grpByExprs.size(); ++i) {
        ASTNode grpbyExpr = grpByExprs.get(i);
        ColumnInfo exprInfo = groupByInputRowResolver.getExpression(grpbyExpr);
        if (exprInfo == null) {
            throw new SemanticException(ErrorMsg.INVALID_COLUMN.getMsg(grpbyExpr));
        }
        groupByKeys.add(new ExprNodeColumnDesc(exprInfo));
        String field = getColumnInternalName(i);
        outputColumnNames.add(field);
        ColumnInfo oColInfo = new ColumnInfo(field, exprInfo.getType(), "", false);
        groupByOutputRowResolver.putExpression(grpbyExpr, oColInfo);
        addAlternateGByKeyMappings(grpbyExpr, oColInfo, reduceSinkOperatorInfo, groupByOutputRowResolver);
        colExprMap.put(field, groupByKeys.get(groupByKeys.size() - 1));
    }
    // This is only needed if a new grouping set key is being created
    int groupingSetsPosition = -1;
    // For grouping sets, add a dummy grouping key
    if (groupingSetsPresent) {
        groupingSetsPosition = groupByKeys.size();
        // This function is called for GroupBy2 to add grouping id as part of the groupby keys
        if (!groupingSetsNeedAdditionalMRJob) {
            addGroupingSetKey(groupByKeys, groupByInputRowResolver, groupByOutputRowResolver, outputColumnNames, colExprMap);
        } else {
            // The grouping set has not yet been processed. Create a new grouping key
            // Consider the query: select a,b, count(1) from T group by a,b with cube;
            // where it is being executed in 2 map-reduce jobs
            // The plan for 1st MR is TableScan -> GroupBy1 -> ReduceSink -> GroupBy2 -> FileSink
            // GroupBy1/ReduceSink worked as if grouping sets were not present
            // This function is called for GroupBy2 to create new rows for grouping sets
            // For each input row (a,b), 4 rows are created for the example above:
            // (a,b), (a,null), (null, b), (null, null)
            createNewGroupingKey(groupByKeys, outputColumnNames, groupByOutputRowResolver, colExprMap);
        }
    }
    HashMap<String, ASTNode> aggregationTrees = parseInfo.getAggregationExprsForClause(dest);
    // get the last colName for the reduce KEY
    // it represents the column name corresponding to distinct aggr, if any
    String lastKeyColName = null;
    List<ExprNodeDesc> reduceValues = null;
    if (reduceSinkOperatorInfo.getConf() instanceof ReduceSinkDesc) {
        List<String> inputKeyCols = ((ReduceSinkDesc) reduceSinkOperatorInfo.getConf()).getOutputKeyColumnNames();
        if (inputKeyCols.size() > 0) {
            lastKeyColName = inputKeyCols.get(inputKeyCols.size() - 1);
        }
        reduceValues = ((ReduceSinkDesc) reduceSinkOperatorInfo.getConf()).getValueCols();
    }
    int numDistinctUDFs = 0;
    boolean containsDistinctAggr = false;
    for (Map.Entry<String, ASTNode> entry : aggregationTrees.entrySet()) {
        ASTNode value = entry.getValue();
        String aggName = unescapeIdentifier(value.getChild(0).getText());
        ArrayList<ExprNodeDesc> aggParameters = new ArrayList<ExprNodeDesc>();
        boolean isDistinct = (value.getType() == HiveParser.TOK_FUNCTIONDI);
        containsDistinctAggr = containsDistinctAggr || isDistinct;
        // side, so always look for the parameters: d+e
        if (isDistinct) {
            // 0 is the function name
            for (int i = 1; i < value.getChildCount(); i++) {
                ASTNode paraExpr = (ASTNode) value.getChild(i);
                ColumnInfo paraExprInfo = groupByInputRowResolver.getExpression(paraExpr);
                if (paraExprInfo == null) {
                    throw new SemanticException(ErrorMsg.INVALID_COLUMN.getMsg(paraExpr));
                }
                String paraExpression = paraExprInfo.getInternalName();
                assert (paraExpression != null);
                if (isDistinct && lastKeyColName != null) {
                    // if aggr is distinct, the parameter is name is constructed as
                    // KEY.lastKeyColName:<tag>._colx
                    paraExpression = Utilities.ReduceField.KEY.name() + "." + lastKeyColName + ":" + numDistinctUDFs + "." + getColumnInternalName(i - 1);
                }
                ExprNodeDesc expr = new ExprNodeColumnDesc(paraExprInfo.getType(), paraExpression, paraExprInfo.getTabAlias(), paraExprInfo.getIsVirtualCol());
                ExprNodeDesc reduceValue = isConstantParameterInAggregationParameters(paraExprInfo.getInternalName(), reduceValues);
                if (reduceValue != null) {
                    // this parameter is a constant
                    expr = reduceValue;
                }
                aggParameters.add(expr);
            }
        } else {
            ColumnInfo paraExprInfo = groupByInputRowResolver.getExpression(value);
            if (paraExprInfo == null) {
                throw new SemanticException(ErrorMsg.INVALID_COLUMN.getMsg(value));
            }
            String paraExpression = paraExprInfo.getInternalName();
            assert (paraExpression != null);
            aggParameters.add(new ExprNodeColumnDesc(paraExprInfo.getType(), paraExpression, paraExprInfo.getTabAlias(), paraExprInfo.getIsVirtualCol()));
        }
        if (isDistinct) {
            numDistinctUDFs++;
        }
        Mode amode = groupByDescModeToUDAFMode(mode, isDistinct);
        GenericUDAFEvaluator genericUDAFEvaluator = null;
        genericUDAFEvaluator = genericUDAFEvaluators.get(entry.getKey());
        assert (genericUDAFEvaluator != null);
        GenericUDAFInfo udaf = getGenericUDAFInfo(genericUDAFEvaluator, amode, aggParameters);
        aggregations.add(new AggregationDesc(aggName.toLowerCase(), udaf.genericUDAFEvaluator, udaf.convertedParameters, (mode != GroupByDesc.Mode.FINAL && isDistinct), amode));
        String field = getColumnInternalName(groupByKeys.size() + aggregations.size() - 1);
        outputColumnNames.add(field);
        groupByOutputRowResolver.putExpression(value, new ColumnInfo(field, udaf.returnType, "", false));
    }
    float groupByMemoryUsage = HiveConf.getFloatVar(conf, HiveConf.ConfVars.HIVEMAPAGGRHASHMEMORY);
    float memoryThreshold = HiveConf.getFloatVar(conf, HiveConf.ConfVars.HIVEMAPAGGRMEMORYTHRESHOLD);
    // Nothing special needs to be done for grouping sets if
    // this is the final group by operator, and multiple rows corresponding to the
    // grouping sets have been generated upstream.
    // However, if an addition MR job has been created to handle grouping sets,
    // additional rows corresponding to grouping sets need to be created here.
    Operator op = putOpInsertMap(OperatorFactory.getAndMakeChild(new GroupByDesc(mode, outputColumnNames, groupByKeys, aggregations, groupByMemoryUsage, memoryThreshold, groupingSets, groupingSetsPresent && groupingSetsNeedAdditionalMRJob, groupingSetsPosition, containsDistinctAggr), new RowSchema(groupByOutputRowResolver.getColumnInfos()), reduceSinkOperatorInfo), groupByOutputRowResolver);
    op.setColumnExprMap(colExprMap);
    return op;
}

Example 52 with Operator

use of org.apache.hadoop.hive.ql.exec.Operator in project hive by apache.

the class SemanticAnalyzer method genGroupByPlan2MR.

/**
   * Generate a Group-By plan using a 2 map-reduce jobs (5 operators will be
   * inserted):
   *
   * ReduceSink ( keys = (K1_EXP, K2_EXP, DISTINCT_EXP), values = (A1_EXP,
   * A2_EXP) ) NOTE: If DISTINCT_EXP is null, partition by rand() SortGroupBy
   * (keys = (KEY.0,KEY.1), aggregations = (count_distinct(KEY.2), sum(VALUE.0),
   * count(VALUE.1))) ReduceSink ( keys = (0,1), values=(2,3,4)) SortGroupBy
   * (keys = (KEY.0,KEY.1), aggregations = (sum(VALUE.0), sum(VALUE.1),
   * sum(VALUE.2))) Select (final selects).
   *
   * @param dest
   * @param qb
   * @param input
   * @return
   * @throws SemanticException
   *
   *           Generate a Group-By plan using a 2 map-reduce jobs. Spray by the
   *           grouping key and distinct key (or a random number, if no distinct
   *           is present) in hope of getting a uniform distribution, and
   *           compute partial aggregates grouped by the reduction key (grouping
   *           key + distinct key). Evaluate partial aggregates first, and spray
   *           by the grouping key to compute actual aggregates in the second
   *           phase. The aggregation evaluation functions are as follows:
   *           Partitioning Key: random() if no DISTINCT grouping + distinct key
   *           if DISTINCT
   *
   *           Sorting Key: grouping key if no DISTINCT grouping + distinct key
   *           if DISTINCT
   *
   *           Reducer: iterate/terminatePartial (mode = PARTIAL1)
   *
   *           STAGE 2
   *
   *           Partitioning Key: grouping key
   *
   *           Sorting Key: grouping key if no DISTINCT grouping + distinct key
   *           if DISTINCT
   *
   *           Reducer: merge/terminate (mode = FINAL)
   */
@SuppressWarnings("nls")
private Operator genGroupByPlan2MR(String dest, QB qb, Operator input) throws SemanticException {
    QBParseInfo parseInfo = qb.getParseInfo();
    ObjectPair<List<ASTNode>, List<Integer>> grpByExprsGroupingSets = getGroupByGroupingSetsForClause(parseInfo, dest);
    List<ASTNode> grpByExprs = grpByExprsGroupingSets.getFirst();
    List<Integer> groupingSets = grpByExprsGroupingSets.getSecond();
    // HIVE-3508 has been filed for this
    if (!groupingSets.isEmpty()) {
        throw new SemanticException(ErrorMsg.HIVE_GROUPING_SETS_AGGR_NOMAPAGGR.getMsg());
    }
    // ////// 1. Generate ReduceSinkOperator
    // There is a special case when we want the rows to be randomly distributed
    // to
    // reducers for load balancing problem. That happens when there is no
    // DISTINCT
    // operator. We set the numPartitionColumns to -1 for this purpose. This is
    // captured by WritableComparableHiveObject.hashCode() function.
    ReduceSinkOperator reduceSinkOperatorInfo = genGroupByPlanReduceSinkOperator(qb, dest, input, grpByExprs, (parseInfo.getDistinctFuncExprsForClause(dest).isEmpty() ? -1 : Integer.MAX_VALUE), false, -1, false, false);
    // ////// 2. Generate GroupbyOperator
    Map<String, GenericUDAFEvaluator> genericUDAFEvaluators = new LinkedHashMap<String, GenericUDAFEvaluator>();
    GroupByOperator groupByOperatorInfo = (GroupByOperator) genGroupByPlanGroupByOperator(parseInfo, dest, reduceSinkOperatorInfo, reduceSinkOperatorInfo, GroupByDesc.Mode.PARTIAL1, genericUDAFEvaluators);
    int numReducers = -1;
    if (grpByExprs.isEmpty()) {
        numReducers = 1;
    }
    // ////// 3. Generate ReduceSinkOperator2
    Operator reduceSinkOperatorInfo2 = genGroupByPlanReduceSinkOperator2MR(parseInfo, dest, groupByOperatorInfo, grpByExprs.size(), numReducers, false);
    // ////// 4. Generate GroupbyOperator2
    Operator groupByOperatorInfo2 = genGroupByPlanGroupByOperator2MR(parseInfo, dest, reduceSinkOperatorInfo2, GroupByDesc.Mode.FINAL, genericUDAFEvaluators, false);
    return groupByOperatorInfo2;
}

Example 53 with Operator

use of org.apache.hadoop.hive.ql.exec.Operator in project hive by apache.

the class SemanticAnalyzer method genLimitPlan.

@SuppressWarnings("nls")
private Operator genLimitPlan(String dest, QB qb, Operator input, int offset, int limit) throws SemanticException {
    // A map-only job can be optimized - instead of converting it to a
    // map-reduce job, we can have another map
    // job to do the same to avoid the cost of sorting in the map-reduce phase.
    // A better approach would be to
    // write into a local file and then have a map-only job.
    // Add the limit operator to get the value fields
    RowResolver inputRR = opParseCtx.get(input).getRowResolver();
    LimitDesc limitDesc = new LimitDesc(offset, limit);
    globalLimitCtx.setLastReduceLimitDesc(limitDesc);
    Operator limitMap = putOpInsertMap(OperatorFactory.getAndMakeChild(limitDesc, new RowSchema(inputRR.getColumnInfos()), input), inputRR);
    if (LOG.isDebugEnabled()) {
        LOG.debug("Created LimitOperator Plan for clause: " + dest + " row schema: " + inputRR.toString());
    }
    return limitMap;
}

Example 54 with Operator

use of org.apache.hadoop.hive.ql.exec.Operator in project hive by apache.

the class SemanticAnalyzer method genGroupByPlanMapGroupByOperator.

/**
   * Generate the map-side GroupByOperator for the Query Block
   * (qb.getParseInfo().getXXX(dest)). The new GroupByOperator will be a child
   * of the inputOperatorInfo.
   *
   * @param mode
   *          The mode of the aggregation (HASH)
   * @param genericUDAFEvaluators
   *          If not null, this function will store the mapping from Aggregation
   *          StringTree to the genericUDAFEvaluator in this parameter, so it
   *          can be used in the next-stage GroupBy aggregations.
   * @return the new GroupByOperator
   */
@SuppressWarnings("nls")
private Operator genGroupByPlanMapGroupByOperator(QB qb, String dest, List<ASTNode> grpByExprs, Operator inputOperatorInfo, GroupByDesc.Mode mode, Map<String, GenericUDAFEvaluator> genericUDAFEvaluators, List<Integer> groupingSetKeys, boolean groupingSetsPresent) throws SemanticException {
    RowResolver groupByInputRowResolver = opParseCtx.get(inputOperatorInfo).getRowResolver();
    QBParseInfo parseInfo = qb.getParseInfo();
    RowResolver groupByOutputRowResolver = new RowResolver();
    groupByOutputRowResolver.setIsExprResolver(true);
    ArrayList<ExprNodeDesc> groupByKeys = new ArrayList<ExprNodeDesc>();
    ArrayList<String> outputColumnNames = new ArrayList<String>();
    ArrayList<AggregationDesc> aggregations = new ArrayList<AggregationDesc>();
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    for (int i = 0; i < grpByExprs.size(); ++i) {
        ASTNode grpbyExpr = grpByExprs.get(i);
        ExprNodeDesc grpByExprNode = genExprNodeDesc(grpbyExpr, groupByInputRowResolver);
        if ((grpByExprNode instanceof ExprNodeColumnDesc) && ExprNodeDescUtils.indexOf(grpByExprNode, groupByKeys) >= 0) {
            // Skip duplicated grouping keys, it happens when define column alias.
            grpByExprs.remove(i--);
            continue;
        }
        groupByKeys.add(grpByExprNode);
        String field = getColumnInternalName(i);
        outputColumnNames.add(field);
        groupByOutputRowResolver.putExpression(grpbyExpr, new ColumnInfo(field, grpByExprNode.getTypeInfo(), "", false));
        colExprMap.put(field, groupByKeys.get(groupByKeys.size() - 1));
    }
    // The grouping set key is present after the grouping keys, before the distinct keys
    int groupingSetsPosition = -1;
    // for the grouping set (corresponding to the rollup).
    if (groupingSetsPresent) {
        groupingSetsPosition = groupByKeys.size();
        createNewGroupingKey(groupByKeys, outputColumnNames, groupByOutputRowResolver, colExprMap);
    }
    // If there is a distinctFuncExp, add all parameters to the reduceKeys.
    if (!parseInfo.getDistinctFuncExprsForClause(dest).isEmpty()) {
        List<ASTNode> list = parseInfo.getDistinctFuncExprsForClause(dest);
        for (ASTNode value : list) {
            // 0 is function name
            for (int i = 1; i < value.getChildCount(); i++) {
                ASTNode parameter = (ASTNode) value.getChild(i);
                if (groupByOutputRowResolver.getExpression(parameter) == null) {
                    ExprNodeDesc distExprNode = genExprNodeDesc(parameter, groupByInputRowResolver);
                    groupByKeys.add(distExprNode);
                    String field = getColumnInternalName(groupByKeys.size() - 1);
                    outputColumnNames.add(field);
                    groupByOutputRowResolver.putExpression(parameter, new ColumnInfo(field, distExprNode.getTypeInfo(), "", false));
                    colExprMap.put(field, groupByKeys.get(groupByKeys.size() - 1));
                }
            }
        }
    }
    // For each aggregation
    HashMap<String, ASTNode> aggregationTrees = parseInfo.getAggregationExprsForClause(dest);
    assert (aggregationTrees != null);
    boolean containsDistinctAggr = false;
    for (Map.Entry<String, ASTNode> entry : aggregationTrees.entrySet()) {
        ASTNode value = entry.getValue();
        String aggName = unescapeIdentifier(value.getChild(0).getText());
        ArrayList<ExprNodeDesc> aggParameters = new ArrayList<ExprNodeDesc>();
        // 0 is the function name
        for (int i = 1; i < value.getChildCount(); i++) {
            ASTNode paraExpr = (ASTNode) value.getChild(i);
            ExprNodeDesc paraExprNode = genExprNodeDesc(paraExpr, groupByInputRowResolver);
            aggParameters.add(paraExprNode);
        }
        boolean isDistinct = value.getType() == HiveParser.TOK_FUNCTIONDI;
        containsDistinctAggr = containsDistinctAggr || isDistinct;
        boolean isAllColumns = value.getType() == HiveParser.TOK_FUNCTIONSTAR;
        Mode amode = groupByDescModeToUDAFMode(mode, isDistinct);
        GenericUDAFEvaluator genericUDAFEvaluator = getGenericUDAFEvaluator(aggName, aggParameters, value, isDistinct, isAllColumns);
        assert (genericUDAFEvaluator != null);
        GenericUDAFInfo udaf = getGenericUDAFInfo(genericUDAFEvaluator, amode, aggParameters);
        aggregations.add(new AggregationDesc(aggName.toLowerCase(), udaf.genericUDAFEvaluator, udaf.convertedParameters, isDistinct, amode));
        String field = getColumnInternalName(groupByKeys.size() + aggregations.size() - 1);
        outputColumnNames.add(field);
        if (groupByOutputRowResolver.getExpression(value) == null) {
            groupByOutputRowResolver.putExpression(value, new ColumnInfo(field, udaf.returnType, "", false));
        }
        // GroupByOperators
        if (genericUDAFEvaluators != null) {
            genericUDAFEvaluators.put(entry.getKey(), genericUDAFEvaluator);
        }
    }
    float groupByMemoryUsage = HiveConf.getFloatVar(conf, HiveConf.ConfVars.HIVEMAPAGGRHASHMEMORY);
    float memoryThreshold = HiveConf.getFloatVar(conf, HiveConf.ConfVars.HIVEMAPAGGRMEMORYTHRESHOLD);
    Operator op = putOpInsertMap(OperatorFactory.getAndMakeChild(new GroupByDesc(mode, outputColumnNames, groupByKeys, aggregations, false, groupByMemoryUsage, memoryThreshold, groupingSetKeys, groupingSetsPresent, groupingSetsPosition, containsDistinctAggr), new RowSchema(groupByOutputRowResolver.getColumnInfos()), inputOperatorInfo), groupByOutputRowResolver);
    op.setColumnExprMap(colExprMap);
    return op;
}

Example 55 with Operator

use of org.apache.hadoop.hive.ql.exec.Operator in project hive by apache.

the class SemanticAnalyzer method genBodyPlan.

@SuppressWarnings("nls")
private Operator genBodyPlan(QB qb, Operator input, Map<String, Operator> aliasToOpInfo) throws SemanticException {
    QBParseInfo qbp = qb.getParseInfo();
    TreeSet<String> ks = new TreeSet<String>(qbp.getClauseNames());
    Map<String, Operator<? extends OperatorDesc>> inputs = createInputForDests(qb, input, ks);
    Operator curr = input;
    List<List<String>> commonGroupByDestGroups = null;
    // expressions, otherwise treat all the expressions as a single group
    if (conf.getBoolVar(HiveConf.ConfVars.HIVEMULTIGROUPBYSINGLEREDUCER)) {
        try {
            commonGroupByDestGroups = getCommonGroupByDestGroups(qb, inputs);
        } catch (SemanticException e) {
            LOG.error("Failed to group clauses by common spray keys.", e);
        }
    }
    if (commonGroupByDestGroups == null) {
        commonGroupByDestGroups = new ArrayList<List<String>>();
        commonGroupByDestGroups.add(new ArrayList<String>(ks));
    }
    if (!commonGroupByDestGroups.isEmpty()) {
        // Iterate over each group of subqueries with the same group by/distinct keys
        for (List<String> commonGroupByDestGroup : commonGroupByDestGroups) {
            if (commonGroupByDestGroup.isEmpty()) {
                continue;
            }
            String firstDest = commonGroupByDestGroup.get(0);
            input = inputs.get(firstDest);
            // The conf variable used to control combining group bys into a single reducer is false
            if (commonGroupByDestGroup.size() == 1 || (qbp.getAggregationExprsForClause(firstDest).size() == 0 && getGroupByForClause(qbp, firstDest).size() == 0) || conf.getBoolVar(HiveConf.ConfVars.HIVEGROUPBYSKEW) || !conf.getBoolVar(HiveConf.ConfVars.HIVEMULTIGROUPBYSINGLEREDUCER)) {
                // Go over all the destination tables
                for (String dest : commonGroupByDestGroup) {
                    curr = inputs.get(dest);
                    if (qbp.getWhrForClause(dest) != null) {
                        ASTNode whereExpr = qb.getParseInfo().getWhrForClause(dest);
                        curr = genFilterPlan((ASTNode) whereExpr.getChild(0), qb, curr, aliasToOpInfo, false, false);
                    }
                    // Preserve operator before the GBY - we'll use it to resolve '*'
                    Operator<?> gbySource = curr;
                    if ((qbp.getAggregationExprsForClause(dest).size() != 0 || getGroupByForClause(qbp, dest).size() > 0) && (qbp.getSelForClause(dest).getToken().getType() != HiveParser.TOK_SELECTDI || qbp.getWindowingExprsForClause(dest) == null)) {
                        // multiple distincts is not supported with skew in data
                        if (conf.getBoolVar(HiveConf.ConfVars.HIVEGROUPBYSKEW) && qbp.getDistinctFuncExprsForClause(dest).size() > 1) {
                            throw new SemanticException(ErrorMsg.UNSUPPORTED_MULTIPLE_DISTINCTS.getMsg());
                        }
                        // insert a select operator here used by the ColumnPruner to reduce
                        // the data to shuffle
                        curr = genSelectAllDesc(curr);
                        // Check and transform group by *. This will only happen for select distinct *.
                        // Here the "genSelectPlan" is being leveraged.
                        // The main benefits are (1) remove virtual columns that should
                        // not be included in the group by; (2) add the fully qualified column names to unParseTranslator
                        // so that view is supported. The drawback is that an additional SEL op is added. If it is
                        // not necessary, it will be removed by NonBlockingOpDeDupProc Optimizer because it will match
                        // SEL%SEL% rule.
                        ASTNode selExprList = qbp.getSelForClause(dest);
                        if (selExprList.getToken().getType() == HiveParser.TOK_SELECTDI && selExprList.getChildCount() == 1 && selExprList.getChild(0).getChildCount() == 1) {
                            ASTNode node = (ASTNode) selExprList.getChild(0).getChild(0);
                            if (node.getToken().getType() == HiveParser.TOK_ALLCOLREF) {
                                curr = genSelectPlan(dest, qb, curr, curr);
                                RowResolver rr = opParseCtx.get(curr).getRowResolver();
                                qbp.setSelExprForClause(dest, SemanticAnalyzer.genSelectDIAST(rr));
                            }
                        }
                        if (conf.getBoolVar(HiveConf.ConfVars.HIVEMAPSIDEAGGREGATE)) {
                            if (!conf.getBoolVar(HiveConf.ConfVars.HIVEGROUPBYSKEW)) {
                                curr = genGroupByPlanMapAggrNoSkew(dest, qb, curr);
                            } else {
                                curr = genGroupByPlanMapAggr2MR(dest, qb, curr);
                            }
                        } else if (conf.getBoolVar(HiveConf.ConfVars.HIVEGROUPBYSKEW)) {
                            curr = genGroupByPlan2MR(dest, qb, curr);
                        } else {
                            curr = genGroupByPlan1MR(dest, qb, curr);
                        }
                    }
                    if (LOG.isDebugEnabled()) {
                        LOG.debug("RR before GB " + opParseCtx.get(gbySource).getRowResolver() + " after GB " + opParseCtx.get(curr).getRowResolver());
                    }
                    curr = genPostGroupByBodyPlan(curr, dest, qb, aliasToOpInfo, gbySource);
                }
            } else {
                curr = genGroupByPlan1ReduceMultiGBY(commonGroupByDestGroup, qb, input, aliasToOpInfo);
            }
        }
    }
    if (LOG.isDebugEnabled()) {
        LOG.debug("Created Body Plan for Query Block " + qb.getId());
    }
    return curr;
}