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

Example 31 with GroupByDesc

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

the class GlobalLimitOptimizer method checkQbpForGlobalLimit.

/**
 * Check the limit number in all sub queries
 *
 * @return if there is one and only one limit for all subqueries, return the limit
 *         if there is no limit, return 0
 *         otherwise, return null
 */
private static LimitOperator checkQbpForGlobalLimit(TableScanOperator ts) {
    Set<Class<? extends Operator<?>>> searchedClasses = new ImmutableSet.Builder<Class<? extends Operator<?>>>().add(ReduceSinkOperator.class).add(GroupByOperator.class).add(FilterOperator.class).add(LimitOperator.class).build();
    Multimap<Class<? extends Operator<?>>, Operator<?>> ops = OperatorUtils.classifyOperators(ts, searchedClasses);
    // existsOrdering AND existsPartitioning should be false.
    for (Operator<?> op : ops.get(ReduceSinkOperator.class)) {
        ReduceSinkDesc reduceSinkConf = ((ReduceSinkOperator) op).getConf();
        if (reduceSinkConf.isOrdering() || reduceSinkConf.isPartitioning()) {
            return null;
        }
    }
    // - There cannot exist any (distinct) aggregate.
    for (Operator<?> op : ops.get(GroupByOperator.class)) {
        GroupByDesc groupByConf = ((GroupByOperator) op).getConf();
        if (groupByConf.isAggregate() || groupByConf.isDistinct()) {
            return null;
        }
    }
    // - There cannot exist any sampling predicate.
    for (Operator<?> op : ops.get(FilterOperator.class)) {
        FilterDesc filterConf = ((FilterOperator) op).getConf();
        if (filterConf.getIsSamplingPred()) {
            return null;
        }
    }
    // If there is one and only one limit starting at op, return the limit
    // If there is no limit, return 0
    // Otherwise, return null
    Collection<Operator<?>> limitOps = ops.get(LimitOperator.class);
    if (limitOps.size() == 1) {
        return (LimitOperator) limitOps.iterator().next();
    } else if (limitOps.size() == 0) {
        return null;
    }
    return null;
}

Example 32 with GroupByDesc

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

the class TestVectorGroupByOperator method buildGroupByDescType.

private static Pair<GroupByDesc, VectorGroupByDesc> buildGroupByDescType(VectorizationContext ctx, String aggregate, GenericUDAFEvaluator.Mode mode, String column, TypeInfo dataType) {
    AggregationDesc agg = buildAggregationDesc(ctx, aggregate, mode, column, dataType);
    ArrayList<AggregationDesc> aggs = new ArrayList<AggregationDesc>();
    aggs.add(agg);
    ArrayList<String> outputColumnNames = new ArrayList<String>();
    outputColumnNames.add("_col0");
    GroupByDesc desc = new GroupByDesc();
    VectorGroupByDesc vectorDesc = new VectorGroupByDesc();
    desc.setOutputColumnNames(outputColumnNames);
    desc.setAggregators(aggs);
    vectorDesc.setProcessingMode(ProcessingMode.GLOBAL);
    return new Pair<GroupByDesc, VectorGroupByDesc>(desc, vectorDesc);
}

Example 33 with GroupByDesc

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

the class TestVectorGroupByOperator method buildKeyGroupByDesc.

private static Pair<GroupByDesc, VectorGroupByDesc> buildKeyGroupByDesc(VectorizationContext ctx, String aggregate, String column, TypeInfo dataTypeInfo, String key, TypeInfo keyTypeInfo) {
    Pair<GroupByDesc, VectorGroupByDesc> pair = buildGroupByDescType(ctx, aggregate, GenericUDAFEvaluator.Mode.PARTIAL1, column, dataTypeInfo);
    GroupByDesc desc = pair.fst;
    VectorGroupByDesc vectorDesc = pair.snd;
    vectorDesc.setProcessingMode(ProcessingMode.HASH);
    ExprNodeDesc keyExp = buildColumnDesc(ctx, key, keyTypeInfo);
    ArrayList<ExprNodeDesc> keys = new ArrayList<ExprNodeDesc>();
    keys.add(keyExp);
    desc.setKeys(keys);
    desc.getOutputColumnNames().add("_col1");
    return pair;
}

Example 34 with GroupByDesc

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

the class TestVectorGroupByOperator method testAggregateStringIterable.

public void testAggregateStringIterable(String aggregateName, Iterable<VectorizedRowBatch> data, Object expected) throws HiveException {
    List<String> mapColumnNames = new ArrayList<String>();
    mapColumnNames.add("A");
    VectorizationContext ctx = new VectorizationContext("name", mapColumnNames);
    Pair<GroupByDesc, VectorGroupByDesc> pair = buildGroupByDescType(ctx, aggregateName, GenericUDAFEvaluator.Mode.PARTIAL1, "A", TypeInfoFactory.stringTypeInfo);
    GroupByDesc desc = pair.fst;
    VectorGroupByDesc vectorDesc = pair.snd;
    CompilationOpContext cCtx = new CompilationOpContext();
    Operator<? extends OperatorDesc> groupByOp = OperatorFactory.get(cCtx, desc);
    VectorGroupByOperator vgo = (VectorGroupByOperator) Vectorizer.vectorizeGroupByOperator(groupByOp, ctx, vectorDesc);
    FakeCaptureVectorToRowOutputOperator out = FakeCaptureVectorToRowOutputOperator.addCaptureOutputChild(cCtx, vgo);
    vgo.initialize(hconf, null);
    for (VectorizedRowBatch unit : data) {
        vgo.process(unit, 0);
    }
    vgo.close(false);
    List<Object> outBatchList = out.getCapturedRows();
    assertNotNull(outBatchList);
    assertEquals(1, outBatchList.size());
    Object result = outBatchList.get(0);
    Validator validator = getValidator(aggregateName);
    validator.validate("_total", expected, result);
}

Example 35 with GroupByDesc

use of org.apache.hadoop.hive.ql.plan.GroupByDesc 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<Long> 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;
}