Search in sources :

Example 1 with GenericUDAFBloomFilterEvaluator

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator in project hive by apache.

the class DynamicPartitionPruningOptimization method generateSemiJoinOperatorPlan.

// Generates plan for min/max when dynamic partition pruning is ruled out.
private boolean generateSemiJoinOperatorPlan(DynamicListContext ctx, ParseContext parseContext, TableScanOperator ts, String keyBaseAlias) throws SemanticException {
    // we will put a fork in the plan at the source of the reduce sink
    Operator<? extends OperatorDesc> parentOfRS = ctx.generator.getParentOperators().get(0);
    // we need the expr that generated the key of the reduce sink
    ExprNodeDesc key = ctx.generator.getConf().getKeyCols().get(ctx.desc.getKeyIndex());
    String internalColName = null;
    ExprNodeDesc exprNodeDesc = key;
    // Find the ExprNodeColumnDesc
    while (!(exprNodeDesc instanceof ExprNodeColumnDesc) && (exprNodeDesc.getChildren() != null)) {
        exprNodeDesc = exprNodeDesc.getChildren().get(0);
    }
    if (!(exprNodeDesc instanceof ExprNodeColumnDesc)) {
        // Bail out
        return false;
    }
    internalColName = ((ExprNodeColumnDesc) exprNodeDesc).getColumn();
    if (parentOfRS instanceof SelectOperator) {
        // Make sure the semijoin branch is not on partition column.
        ExprNodeDesc expr = parentOfRS.getColumnExprMap().get(internalColName);
        while (!(expr instanceof ExprNodeColumnDesc) && (expr.getChildren() != null)) {
            expr = expr.getChildren().get(0);
        }
        if (!(expr instanceof ExprNodeColumnDesc)) {
            // Bail out
            return false;
        }
        ExprNodeColumnDesc colExpr = (ExprNodeColumnDesc) expr;
        String colName = ExprNodeDescUtils.extractColName(colExpr);
        // Fetch the TableScan Operator.
        Operator<?> op = parentOfRS.getParentOperators().get(0);
        while (op != null && !(op instanceof TableScanOperator)) {
            op = op.getParentOperators().get(0);
        }
        assert op != null;
        Table table = ((TableScanOperator) op).getConf().getTableMetadata();
        if (table.isPartitionKey(colName)) {
            // The column is partition column, skip the optimization.
            return false;
        }
    }
    List<ExprNodeDesc> keyExprs = new ArrayList<ExprNodeDesc>();
    keyExprs.add(key);
    // group by requires "ArrayList", don't ask.
    ArrayList<String> outputNames = new ArrayList<String>();
    outputNames.add(HiveConf.getColumnInternalName(0));
    // project the relevant key column
    SelectDesc select = new SelectDesc(keyExprs, outputNames);
    // Create the new RowSchema for the projected column
    ColumnInfo columnInfo = parentOfRS.getSchema().getColumnInfo(internalColName);
    ArrayList<ColumnInfo> signature = new ArrayList<ColumnInfo>();
    signature.add(columnInfo);
    RowSchema rowSchema = new RowSchema(signature);
    // Create the column expr map
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    ExprNodeDesc exprNode = null;
    if (parentOfRS.getColumnExprMap() != null) {
        exprNode = parentOfRS.getColumnExprMap().get(internalColName).clone();
    } else {
        exprNode = new ExprNodeColumnDesc(columnInfo);
    }
    if (exprNode instanceof ExprNodeColumnDesc) {
        ExprNodeColumnDesc encd = (ExprNodeColumnDesc) exprNode;
        encd.setColumn(internalColName);
    }
    colExprMap.put(internalColName, exprNode);
    // Create the Select Operator
    SelectOperator selectOp = (SelectOperator) OperatorFactory.getAndMakeChild(select, rowSchema, colExprMap, parentOfRS);
    // do a group by to aggregate min,max and bloom filter.
    float groupByMemoryUsage = HiveConf.getFloatVar(parseContext.getConf(), HiveConf.ConfVars.HIVEMAPAGGRHASHMEMORY);
    float memoryThreshold = HiveConf.getFloatVar(parseContext.getConf(), HiveConf.ConfVars.HIVEMAPAGGRMEMORYTHRESHOLD);
    ArrayList<ExprNodeDesc> groupByExprs = new ArrayList<ExprNodeDesc>();
    // Add min/max and bloom filter aggregations
    List<ObjectInspector> aggFnOIs = new ArrayList<ObjectInspector>();
    aggFnOIs.add(key.getWritableObjectInspector());
    ArrayList<ExprNodeDesc> params = new ArrayList<ExprNodeDesc>();
    params.add(new ExprNodeColumnDesc(key.getTypeInfo(), outputNames.get(0), "", false));
    ArrayList<AggregationDesc> aggs = new ArrayList<AggregationDesc>();
    try {
        AggregationDesc min = new AggregationDesc("min", FunctionRegistry.getGenericUDAFEvaluator("min", aggFnOIs, false, false), params, false, Mode.PARTIAL1);
        AggregationDesc max = new AggregationDesc("max", FunctionRegistry.getGenericUDAFEvaluator("max", aggFnOIs, false, false), params, false, Mode.PARTIAL1);
        AggregationDesc bloomFilter = new AggregationDesc("bloom_filter", FunctionRegistry.getGenericUDAFEvaluator("bloom_filter", aggFnOIs, false, false), params, false, Mode.PARTIAL1);
        GenericUDAFBloomFilterEvaluator bloomFilterEval = (GenericUDAFBloomFilterEvaluator) bloomFilter.getGenericUDAFEvaluator();
        bloomFilterEval.setSourceOperator(selectOp);
        bloomFilterEval.setMaxEntries(parseContext.getConf().getLongVar(ConfVars.TEZ_MAX_BLOOM_FILTER_ENTRIES));
        bloomFilter.setGenericUDAFWritableEvaluator(bloomFilterEval);
        aggs.add(min);
        aggs.add(max);
        aggs.add(bloomFilter);
    } catch (SemanticException e) {
        LOG.error("Error creating min/max aggregations on key", e);
        throw new IllegalStateException("Error creating min/max aggregations on key", e);
    }
    // Create the Group by Operator
    ArrayList<String> gbOutputNames = new ArrayList<String>();
    gbOutputNames.add(SemanticAnalyzer.getColumnInternalName(0));
    gbOutputNames.add(SemanticAnalyzer.getColumnInternalName(1));
    gbOutputNames.add(SemanticAnalyzer.getColumnInternalName(2));
    GroupByDesc groupBy = new GroupByDesc(GroupByDesc.Mode.HASH, gbOutputNames, new ArrayList<ExprNodeDesc>(), aggs, false, groupByMemoryUsage, memoryThreshold, null, false, 0, false);
    ArrayList<ColumnInfo> groupbyColInfos = new ArrayList<ColumnInfo>();
    groupbyColInfos.add(new ColumnInfo(gbOutputNames.get(0), key.getTypeInfo(), "", false));
    groupbyColInfos.add(new ColumnInfo(gbOutputNames.get(1), key.getTypeInfo(), "", false));
    groupbyColInfos.add(new ColumnInfo(gbOutputNames.get(2), key.getTypeInfo(), "", false));
    GroupByOperator groupByOp = (GroupByOperator) OperatorFactory.getAndMakeChild(groupBy, new RowSchema(groupbyColInfos), selectOp);
    groupByOp.setColumnExprMap(new HashMap<String, ExprNodeDesc>());
    // Get the column names of the aggregations for reduce sink
    int colPos = 0;
    ArrayList<ExprNodeDesc> rsValueCols = new ArrayList<ExprNodeDesc>();
    for (int i = 0; i < aggs.size() - 1; i++) {
        ExprNodeColumnDesc colExpr = new ExprNodeColumnDesc(key.getTypeInfo(), gbOutputNames.get(colPos++), "", false);
        rsValueCols.add(colExpr);
    }
    // Bloom Filter uses binary
    ExprNodeColumnDesc colExpr = new ExprNodeColumnDesc(TypeInfoFactory.binaryTypeInfo, gbOutputNames.get(colPos++), "", false);
    rsValueCols.add(colExpr);
    // Create the reduce sink operator
    ReduceSinkDesc rsDesc = PlanUtils.getReduceSinkDesc(new ArrayList<ExprNodeDesc>(), rsValueCols, gbOutputNames, false, -1, 0, 1, Operation.NOT_ACID);
    ReduceSinkOperator rsOp = (ReduceSinkOperator) OperatorFactory.getAndMakeChild(rsDesc, new RowSchema(groupByOp.getSchema()), groupByOp);
    Map<String, ExprNodeDesc> columnExprMap = new HashMap<String, ExprNodeDesc>();
    rsOp.setColumnExprMap(columnExprMap);
    // Create the final Group By Operator
    ArrayList<AggregationDesc> aggsFinal = new ArrayList<AggregationDesc>();
    try {
        List<ObjectInspector> minFinalFnOIs = new ArrayList<ObjectInspector>();
        List<ObjectInspector> maxFinalFnOIs = new ArrayList<ObjectInspector>();
        List<ObjectInspector> bloomFilterFinalFnOIs = new ArrayList<ObjectInspector>();
        ArrayList<ExprNodeDesc> minFinalParams = new ArrayList<ExprNodeDesc>();
        ArrayList<ExprNodeDesc> maxFinalParams = new ArrayList<ExprNodeDesc>();
        ArrayList<ExprNodeDesc> bloomFilterFinalParams = new ArrayList<ExprNodeDesc>();
        // Use the expressions from Reduce Sink.
        minFinalFnOIs.add(rsValueCols.get(0).getWritableObjectInspector());
        maxFinalFnOIs.add(rsValueCols.get(1).getWritableObjectInspector());
        bloomFilterFinalFnOIs.add(rsValueCols.get(2).getWritableObjectInspector());
        // Coming from a ReduceSink the aggregations would be in the form VALUE._col0, VALUE._col1
        minFinalParams.add(new ExprNodeColumnDesc(rsValueCols.get(0).getTypeInfo(), Utilities.ReduceField.VALUE + "." + gbOutputNames.get(0), "", false));
        maxFinalParams.add(new ExprNodeColumnDesc(rsValueCols.get(1).getTypeInfo(), Utilities.ReduceField.VALUE + "." + gbOutputNames.get(1), "", false));
        bloomFilterFinalParams.add(new ExprNodeColumnDesc(rsValueCols.get(2).getTypeInfo(), Utilities.ReduceField.VALUE + "." + gbOutputNames.get(2), "", false));
        AggregationDesc min = new AggregationDesc("min", FunctionRegistry.getGenericUDAFEvaluator("min", minFinalFnOIs, false, false), minFinalParams, false, Mode.FINAL);
        AggregationDesc max = new AggregationDesc("max", FunctionRegistry.getGenericUDAFEvaluator("max", maxFinalFnOIs, false, false), maxFinalParams, false, Mode.FINAL);
        AggregationDesc bloomFilter = new AggregationDesc("bloom_filter", FunctionRegistry.getGenericUDAFEvaluator("bloom_filter", bloomFilterFinalFnOIs, false, false), bloomFilterFinalParams, false, Mode.FINAL);
        GenericUDAFBloomFilterEvaluator bloomFilterEval = (GenericUDAFBloomFilterEvaluator) bloomFilter.getGenericUDAFEvaluator();
        bloomFilterEval.setSourceOperator(selectOp);
        bloomFilterEval.setMaxEntries(parseContext.getConf().getLongVar(ConfVars.TEZ_MAX_BLOOM_FILTER_ENTRIES));
        bloomFilter.setGenericUDAFWritableEvaluator(bloomFilterEval);
        aggsFinal.add(min);
        aggsFinal.add(max);
        aggsFinal.add(bloomFilter);
    } catch (SemanticException e) {
        LOG.error("Error creating min/max aggregations on key", e);
        throw new IllegalStateException("Error creating min/max aggregations on key", e);
    }
    GroupByDesc groupByDescFinal = new GroupByDesc(GroupByDesc.Mode.FINAL, gbOutputNames, new ArrayList<ExprNodeDesc>(), aggsFinal, false, groupByMemoryUsage, memoryThreshold, null, false, 0, false);
    GroupByOperator groupByOpFinal = (GroupByOperator) OperatorFactory.getAndMakeChild(groupByDescFinal, new RowSchema(rsOp.getSchema()), rsOp);
    groupByOpFinal.setColumnExprMap(new HashMap<String, ExprNodeDesc>());
    // for explain purpose
    if (parseContext.getContext().getExplainConfig() != null && parseContext.getContext().getExplainConfig().isFormatted()) {
        List<String> outputOperators = new ArrayList<>();
        outputOperators.add(groupByOpFinal.getOperatorId());
        rsOp.getConf().setOutputOperators(outputOperators);
    }
    // Create the final Reduce Sink Operator
    ReduceSinkDesc rsDescFinal = PlanUtils.getReduceSinkDesc(new ArrayList<ExprNodeDesc>(), rsValueCols, gbOutputNames, false, -1, 0, 1, Operation.NOT_ACID);
    ReduceSinkOperator rsOpFinal = (ReduceSinkOperator) OperatorFactory.getAndMakeChild(rsDescFinal, new RowSchema(groupByOpFinal.getSchema()), groupByOpFinal);
    rsOpFinal.setColumnExprMap(columnExprMap);
    LOG.debug("DynamicMinMaxPushdown: Saving RS to TS mapping: " + rsOpFinal + ": " + ts);
    parseContext.getRsOpToTsOpMap().put(rsOpFinal, ts);
    // for explain purpose
    if (parseContext.getContext().getExplainConfig() != null && parseContext.getContext().getExplainConfig().isFormatted()) {
        List<String> outputOperators = new ArrayList<>();
        outputOperators.add(ts.getOperatorId());
        rsOpFinal.getConf().setOutputOperators(outputOperators);
    }
    // Save the info that is required at query time to resolve dynamic/runtime values.
    RuntimeValuesInfo runtimeValuesInfo = new RuntimeValuesInfo();
    TableDesc rsFinalTableDesc = PlanUtils.getReduceValueTableDesc(PlanUtils.getFieldSchemasFromColumnList(rsValueCols, "_col"));
    List<String> dynamicValueIDs = new ArrayList<String>();
    dynamicValueIDs.add(keyBaseAlias + "_min");
    dynamicValueIDs.add(keyBaseAlias + "_max");
    dynamicValueIDs.add(keyBaseAlias + "_bloom_filter");
    runtimeValuesInfo.setTableDesc(rsFinalTableDesc);
    runtimeValuesInfo.setDynamicValueIDs(dynamicValueIDs);
    runtimeValuesInfo.setColExprs(rsValueCols);
    parseContext.getRsToRuntimeValuesInfoMap().put(rsOpFinal, runtimeValuesInfo);
    return true;
}
Also used : HashMap(java.util.HashMap) LinkedHashMap(java.util.LinkedHashMap) ArrayList(java.util.ArrayList) RuntimeValuesInfo(org.apache.hadoop.hive.ql.parse.RuntimeValuesInfo) GenericUDAFBloomFilterEvaluator(org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator) SemanticException(org.apache.hadoop.hive.ql.parse.SemanticException) ObjectInspector(org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector) Table(org.apache.hadoop.hive.ql.metadata.Table)

Example 2 with GenericUDAFBloomFilterEvaluator

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator in project hive by apache.

the class VectorUDAFBloomFilter method init.

@Override
public void init(AggregationDesc desc) throws HiveException {
    GenericUDAFBloomFilterEvaluator udafBloomFilter = (GenericUDAFBloomFilterEvaluator) desc.getGenericUDAFEvaluator();
    expectedEntries = udafBloomFilter.getExpectedEntries();
}
Also used : GenericUDAFBloomFilterEvaluator(org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator)

Example 3 with GenericUDAFBloomFilterEvaluator

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator in project hive by apache.

the class VectorUDAFBloomFilterMerge method init.

@Override
public void init(AggregationDesc desc) throws HiveException {
    GenericUDAFBloomFilterEvaluator udafBloomFilter = (GenericUDAFBloomFilterEvaluator) desc.getGenericUDAFEvaluator();
    expectedEntries = udafBloomFilter.getExpectedEntries();
}
Also used : GenericUDAFBloomFilterEvaluator(org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator)

Example 4 with GenericUDAFBloomFilterEvaluator

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator in project hive by apache.

the class VectorUDAFBloomFilter method init.

private void init() {
    GenericUDAFBloomFilterEvaluator udafBloomFilter = (GenericUDAFBloomFilterEvaluator) vecAggrDesc.getEvaluator();
    expectedEntries = udafBloomFilter.getExpectedEntries();
    bitSetSize = -1;
    byteStream = new ByteArrayOutputStream();
    // Instantiate the ValueProcessor based on the input type
    ColumnVector.Type colVectorType;
    try {
        colVectorType = inputExpression.getOutputColumnVectorType();
    } catch (HiveException e) {
        throw new RuntimeException(e);
    }
    switch(colVectorType) {
        case LONG:
        case DECIMAL_64:
            valueProcessor = new ValueProcessorLong();
            break;
        case DOUBLE:
            valueProcessor = new ValueProcessorDouble();
            break;
        case DECIMAL:
            valueProcessor = new ValueProcessorDecimal();
            break;
        case BYTES:
            valueProcessor = new ValueProcessorBytes();
            break;
        case TIMESTAMP:
            valueProcessor = new ValueProcessorTimestamp();
            break;
        default:
            throw new IllegalStateException("Unsupported column vector type " + colVectorType);
    }
}
Also used : HiveException(org.apache.hadoop.hive.ql.metadata.HiveException) GenericUDAFBloomFilterEvaluator(org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator) Type(org.apache.hadoop.hive.ql.exec.vector.ColumnVector.Type) ByteArrayOutputStream(java.io.ByteArrayOutputStream) DecimalColumnVector(org.apache.hadoop.hive.ql.exec.vector.DecimalColumnVector) BytesColumnVector(org.apache.hadoop.hive.ql.exec.vector.BytesColumnVector) LongColumnVector(org.apache.hadoop.hive.ql.exec.vector.LongColumnVector) ColumnVector(org.apache.hadoop.hive.ql.exec.vector.ColumnVector) TimestampColumnVector(org.apache.hadoop.hive.ql.exec.vector.TimestampColumnVector) DoubleColumnVector(org.apache.hadoop.hive.ql.exec.vector.DoubleColumnVector)

Example 5 with GenericUDAFBloomFilterEvaluator

use of org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator in project hive by apache.

the class DynamicPartitionPruningOptimization method generateSemiJoinOperatorPlan.

// Generates plan for min/max when dynamic partition pruning is ruled out.
private boolean generateSemiJoinOperatorPlan(DynamicListContext ctx, ParseContext parseContext, TableScanOperator ts, String keyBaseAlias, String internalColName, String colName, SemiJoinHint sjHint) throws SemanticException {
    // we will put a fork in the plan at the source of the reduce sink
    Operator<? extends OperatorDesc> parentOfRS = ctx.generator.getParentOperators().get(0);
    // we need the expr that generated the key of the reduce sink
    ExprNodeDesc key = ctx.getKeyCol();
    assert colName != null;
    // Fetch the TableScan Operator.
    Operator<?> op = parentOfRS;
    while (!(op == null || op instanceof TableScanOperator || op instanceof ReduceSinkOperator)) {
        op = op.getParentOperators().get(0);
    }
    Preconditions.checkNotNull(op);
    if (op instanceof TableScanOperator) {
        Table table = ((TableScanOperator) op).getConf().getTableMetadata();
        if (table.isPartitionKey(colName)) {
            // The column is partition column, skip the optimization.
            return false;
        }
    }
    // Check if there already exists a semijoin branch
    GroupByOperator gb = parseContext.getColExprToGBMap().get(key);
    if (gb != null) {
        // Already an existing semijoin branch, reuse it
        createFinalRsForSemiJoinOp(parseContext, ts, gb, key, keyBaseAlias, ctx.parent.getChildren().get(0), sjHint != null);
        // done!
        return true;
    }
    List<ExprNodeDesc> keyExprs = new ArrayList<ExprNodeDesc>();
    keyExprs.add(key);
    // group by requires "ArrayList", don't ask.
    ArrayList<String> outputNames = new ArrayList<String>();
    // project the relevant key column
    SelectDesc select = new SelectDesc(keyExprs, outputNames);
    // Create the new RowSchema for the projected column
    ColumnInfo columnInfo = parentOfRS.getSchema().getColumnInfo(internalColName);
    columnInfo = new ColumnInfo(columnInfo);
    outputNames.add(internalColName);
    ArrayList<ColumnInfo> signature = new ArrayList<ColumnInfo>();
    signature.add(columnInfo);
    RowSchema rowSchema = new RowSchema(signature);
    // Create the column expr map
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    ExprNodeDesc exprNode = null;
    if (columnInfo == null) {
        LOG.debug("No ColumnInfo found in {} for {}", parentOfRS.getOperatorId(), internalColName);
        return false;
    }
    exprNode = new ExprNodeColumnDesc(columnInfo);
    colExprMap.put(internalColName, exprNode);
    // Create the Select Operator
    SelectOperator selectOp = (SelectOperator) OperatorFactory.getAndMakeChild(select, rowSchema, colExprMap, parentOfRS);
    // do a group by to aggregate min,max and bloom filter.
    float groupByMemoryUsage = HiveConf.getFloatVar(parseContext.getConf(), HiveConf.ConfVars.HIVEMAPAGGRHASHMEMORY);
    float memoryThreshold = HiveConf.getFloatVar(parseContext.getConf(), HiveConf.ConfVars.HIVEMAPAGGRMEMORYTHRESHOLD);
    float minReductionHashAggr = HiveConf.getFloatVar(parseContext.getConf(), ConfVars.HIVEMAPAGGRHASHMINREDUCTION);
    float minReductionHashAggrLowerBound = HiveConf.getFloatVar(parseContext.getConf(), ConfVars.HIVEMAPAGGRHASHMINREDUCTIONLOWERBOUND);
    // Add min/max and bloom filter aggregations
    List<ObjectInspector> aggFnOIs = new ArrayList<ObjectInspector>();
    aggFnOIs.add(key.getWritableObjectInspector());
    ArrayList<ExprNodeDesc> params = new ArrayList<ExprNodeDesc>();
    params.add(new ExprNodeColumnDesc(key.getTypeInfo(), outputNames.get(0), "", false));
    ArrayList<AggregationDesc> aggs = new ArrayList<AggregationDesc>();
    try {
        AggregationDesc min = new AggregationDesc("min", FunctionRegistry.getGenericUDAFEvaluator("min", aggFnOIs, false, false), params, false, Mode.PARTIAL1);
        AggregationDesc max = new AggregationDesc("max", FunctionRegistry.getGenericUDAFEvaluator("max", aggFnOIs, false, false), params, false, Mode.PARTIAL1);
        AggregationDesc bloomFilter = new AggregationDesc("bloom_filter", FunctionRegistry.getGenericUDAFEvaluator("bloom_filter", aggFnOIs, false, false), params, false, Mode.PARTIAL1);
        GenericUDAFBloomFilterEvaluator bloomFilterEval = (GenericUDAFBloomFilterEvaluator) bloomFilter.getGenericUDAFEvaluator();
        bloomFilterEval.setSourceOperator(selectOp);
        if (sjHint != null && sjHint.getNumEntries() > 0) {
            LOG.debug("Setting size for " + keyBaseAlias + " to " + sjHint.getNumEntries() + " based on the hint");
            bloomFilterEval.setHintEntries(sjHint.getNumEntries());
        }
        bloomFilterEval.setMaxEntries(parseContext.getConf().getLongVar(ConfVars.TEZ_MAX_BLOOM_FILTER_ENTRIES));
        bloomFilterEval.setMinEntries(parseContext.getConf().getLongVar(ConfVars.TEZ_MIN_BLOOM_FILTER_ENTRIES));
        bloomFilterEval.setFactor(parseContext.getConf().getFloatVar(ConfVars.TEZ_BLOOM_FILTER_FACTOR));
        bloomFilter.setGenericUDAFWritableEvaluator(bloomFilterEval);
        aggs.add(min);
        aggs.add(max);
        aggs.add(bloomFilter);
    } catch (SemanticException e) {
        LOG.error("Error creating min/max aggregations on key", e);
        throw new IllegalStateException("Error creating min/max aggregations on key", e);
    }
    // Create the Group by Operator
    ArrayList<String> gbOutputNames = new ArrayList<String>();
    gbOutputNames.add(SemanticAnalyzer.getColumnInternalName(0));
    gbOutputNames.add(SemanticAnalyzer.getColumnInternalName(1));
    gbOutputNames.add(SemanticAnalyzer.getColumnInternalName(2));
    GroupByDesc groupBy = new GroupByDesc(GroupByDesc.Mode.HASH, gbOutputNames, new ArrayList<ExprNodeDesc>(), aggs, false, groupByMemoryUsage, memoryThreshold, minReductionHashAggr, minReductionHashAggrLowerBound, null, false, -1, false);
    ArrayList<ColumnInfo> groupbyColInfos = new ArrayList<ColumnInfo>();
    groupbyColInfos.add(new ColumnInfo(gbOutputNames.get(0), key.getTypeInfo(), "", false));
    groupbyColInfos.add(new ColumnInfo(gbOutputNames.get(1), key.getTypeInfo(), "", false));
    groupbyColInfos.add(new ColumnInfo(gbOutputNames.get(2), key.getTypeInfo(), "", false));
    GroupByOperator groupByOp = (GroupByOperator) OperatorFactory.getAndMakeChild(groupBy, new RowSchema(groupbyColInfos), selectOp);
    groupByOp.setColumnExprMap(new HashMap<String, ExprNodeDesc>());
    // Get the column names of the aggregations for reduce sink
    int colPos = 0;
    ArrayList<ExprNodeDesc> rsValueCols = new ArrayList<ExprNodeDesc>();
    Map<String, ExprNodeDesc> columnExprMap = new HashMap<String, ExprNodeDesc>();
    for (int i = 0; i < aggs.size() - 1; i++) {
        ExprNodeColumnDesc colExpr = new ExprNodeColumnDesc(key.getTypeInfo(), gbOutputNames.get(colPos), "", false);
        rsValueCols.add(colExpr);
        columnExprMap.put(gbOutputNames.get(colPos), colExpr);
        colPos++;
    }
    // Bloom Filter uses binary
    ExprNodeColumnDesc colExpr = new ExprNodeColumnDesc(TypeInfoFactory.binaryTypeInfo, gbOutputNames.get(colPos), "", false);
    rsValueCols.add(colExpr);
    columnExprMap.put(gbOutputNames.get(colPos), colExpr);
    colPos++;
    // Create the reduce sink operator
    ReduceSinkDesc rsDesc = PlanUtils.getReduceSinkDesc(new ArrayList<ExprNodeDesc>(), rsValueCols, gbOutputNames, false, -1, 0, 1, Operation.NOT_ACID, NullOrdering.defaultNullOrder(parseContext.getConf()));
    ReduceSinkOperator rsOp = (ReduceSinkOperator) OperatorFactory.getAndMakeChild(rsDesc, new RowSchema(groupByOp.getSchema()), groupByOp);
    rsOp.setColumnExprMap(columnExprMap);
    rsOp.getConf().setReducerTraits(EnumSet.of(ReduceSinkDesc.ReducerTraits.QUICKSTART));
    // Create the final Group By Operator
    ArrayList<AggregationDesc> aggsFinal = new ArrayList<AggregationDesc>();
    try {
        List<ObjectInspector> minFinalFnOIs = new ArrayList<ObjectInspector>();
        List<ObjectInspector> maxFinalFnOIs = new ArrayList<ObjectInspector>();
        List<ObjectInspector> bloomFilterFinalFnOIs = new ArrayList<ObjectInspector>();
        ArrayList<ExprNodeDesc> minFinalParams = new ArrayList<ExprNodeDesc>();
        ArrayList<ExprNodeDesc> maxFinalParams = new ArrayList<ExprNodeDesc>();
        ArrayList<ExprNodeDesc> bloomFilterFinalParams = new ArrayList<ExprNodeDesc>();
        // Use the expressions from Reduce Sink.
        minFinalFnOIs.add(rsValueCols.get(0).getWritableObjectInspector());
        maxFinalFnOIs.add(rsValueCols.get(1).getWritableObjectInspector());
        bloomFilterFinalFnOIs.add(rsValueCols.get(2).getWritableObjectInspector());
        // Coming from a ReduceSink the aggregations would be in the form VALUE._col0, VALUE._col1
        minFinalParams.add(new ExprNodeColumnDesc(rsValueCols.get(0).getTypeInfo(), Utilities.ReduceField.VALUE + "." + gbOutputNames.get(0), "", false));
        maxFinalParams.add(new ExprNodeColumnDesc(rsValueCols.get(1).getTypeInfo(), Utilities.ReduceField.VALUE + "." + gbOutputNames.get(1), "", false));
        bloomFilterFinalParams.add(new ExprNodeColumnDesc(rsValueCols.get(2).getTypeInfo(), Utilities.ReduceField.VALUE + "." + gbOutputNames.get(2), "", false));
        AggregationDesc min = new AggregationDesc("min", FunctionRegistry.getGenericUDAFEvaluator("min", minFinalFnOIs, false, false), minFinalParams, false, Mode.FINAL);
        AggregationDesc max = new AggregationDesc("max", FunctionRegistry.getGenericUDAFEvaluator("max", maxFinalFnOIs, false, false), maxFinalParams, false, Mode.FINAL);
        AggregationDesc bloomFilter = new AggregationDesc("bloom_filter", FunctionRegistry.getGenericUDAFEvaluator("bloom_filter", bloomFilterFinalFnOIs, false, false), bloomFilterFinalParams, false, Mode.FINAL);
        GenericUDAFBloomFilterEvaluator bloomFilterEval = (GenericUDAFBloomFilterEvaluator) bloomFilter.getGenericUDAFEvaluator();
        bloomFilterEval.setSourceOperator(selectOp);
        if (sjHint != null && sjHint.getNumEntries() > 0) {
            bloomFilterEval.setHintEntries(sjHint.getNumEntries());
        }
        bloomFilterEval.setMaxEntries(parseContext.getConf().getLongVar(ConfVars.TEZ_MAX_BLOOM_FILTER_ENTRIES));
        bloomFilterEval.setMinEntries(parseContext.getConf().getLongVar(ConfVars.TEZ_MIN_BLOOM_FILTER_ENTRIES));
        bloomFilterEval.setFactor(parseContext.getConf().getFloatVar(ConfVars.TEZ_BLOOM_FILTER_FACTOR));
        bloomFilter.setGenericUDAFWritableEvaluator(bloomFilterEval);
        aggsFinal.add(min);
        aggsFinal.add(max);
        aggsFinal.add(bloomFilter);
    } catch (SemanticException e) {
        LOG.error("Error creating min/max aggregations on key", e);
        throw new IllegalStateException("Error creating min/max aggregations on key", e);
    }
    GroupByDesc groupByDescFinal = new GroupByDesc(GroupByDesc.Mode.FINAL, gbOutputNames, new ArrayList<ExprNodeDesc>(), aggsFinal, false, groupByMemoryUsage, memoryThreshold, minReductionHashAggr, minReductionHashAggrLowerBound, null, false, 0, false);
    GroupByOperator groupByOpFinal = (GroupByOperator) OperatorFactory.getAndMakeChild(groupByDescFinal, new RowSchema(rsOp.getSchema()), rsOp);
    groupByOpFinal.setColumnExprMap(new HashMap<String, ExprNodeDesc>());
    createFinalRsForSemiJoinOp(parseContext, ts, groupByOpFinal, key, keyBaseAlias, ctx.parent.getChildren().get(0), sjHint != null);
    return true;
}
Also used : TableScanOperator(org.apache.hadoop.hive.ql.exec.TableScanOperator) HashMap(java.util.HashMap) ArrayList(java.util.ArrayList) ColumnInfo(org.apache.hadoop.hive.ql.exec.ColumnInfo) SelectOperator(org.apache.hadoop.hive.ql.exec.SelectOperator) GenericUDAFBloomFilterEvaluator(org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator) ExprNodeColumnDesc(org.apache.hadoop.hive.ql.plan.ExprNodeColumnDesc) ExprNodeDesc(org.apache.hadoop.hive.ql.plan.ExprNodeDesc) SelectDesc(org.apache.hadoop.hive.ql.plan.SelectDesc) ReduceSinkDesc(org.apache.hadoop.hive.ql.plan.ReduceSinkDesc) GroupByDesc(org.apache.hadoop.hive.ql.plan.GroupByDesc) SemanticException(org.apache.hadoop.hive.ql.parse.SemanticException) RowSchema(org.apache.hadoop.hive.ql.exec.RowSchema) ObjectInspector(org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector) Table(org.apache.hadoop.hive.ql.metadata.Table) GroupByOperator(org.apache.hadoop.hive.ql.exec.GroupByOperator) SemiJoinHint(org.apache.hadoop.hive.ql.parse.SemiJoinHint) ReduceSinkOperator(org.apache.hadoop.hive.ql.exec.ReduceSinkOperator) AggregationDesc(org.apache.hadoop.hive.ql.plan.AggregationDesc)

Aggregations

GenericUDAFBloomFilterEvaluator (org.apache.hadoop.hive.ql.udf.generic.GenericUDAFBloomFilter.GenericUDAFBloomFilterEvaluator)8 ArrayList (java.util.ArrayList)2 HashMap (java.util.HashMap)2 GroupByOperator (org.apache.hadoop.hive.ql.exec.GroupByOperator)2 ReduceSinkOperator (org.apache.hadoop.hive.ql.exec.ReduceSinkOperator)2 Table (org.apache.hadoop.hive.ql.metadata.Table)2 SemanticException (org.apache.hadoop.hive.ql.parse.SemanticException)2 AggregationDesc (org.apache.hadoop.hive.ql.plan.AggregationDesc)2 ExprNodeDesc (org.apache.hadoop.hive.ql.plan.ExprNodeDesc)2 ObjectInspector (org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector)2 ByteArrayOutputStream (java.io.ByteArrayOutputStream)1 LinkedHashMap (java.util.LinkedHashMap)1 ColumnInfo (org.apache.hadoop.hive.ql.exec.ColumnInfo)1 RowSchema (org.apache.hadoop.hive.ql.exec.RowSchema)1 SelectOperator (org.apache.hadoop.hive.ql.exec.SelectOperator)1 TableScanOperator (org.apache.hadoop.hive.ql.exec.TableScanOperator)1 BytesColumnVector (org.apache.hadoop.hive.ql.exec.vector.BytesColumnVector)1 ColumnVector (org.apache.hadoop.hive.ql.exec.vector.ColumnVector)1 Type (org.apache.hadoop.hive.ql.exec.vector.ColumnVector.Type)1 DecimalColumnVector (org.apache.hadoop.hive.ql.exec.vector.DecimalColumnVector)1