Examples with SortGroupByOperatorDescriptor org.apache.hyracks.dataflow.std.group.sort.SortGroupByOperatorDescriptor used on opensource projects

Example 1 with SortGroupByOperatorDescriptor

use of org.apache.hyracks.dataflow.std.group.sort.SortGroupByOperatorDescriptor in project asterixdb by apache.

the class SortGroupByPOperator method contributeRuntimeOperator.

@Override
public void contributeRuntimeOperator(IHyracksJobBuilder builder, JobGenContext context, ILogicalOperator op, IOperatorSchema opSchema, IOperatorSchema[] inputSchemas, IOperatorSchema outerPlanSchema) throws AlgebricksException {
    List<LogicalVariable> gbyCols = getGbyColumns();
    int[] keys = JobGenHelper.variablesToFieldIndexes(gbyCols, inputSchemas[0]);
    GroupByOperator gby = (GroupByOperator) op;
    int numFds = gby.getDecorList().size();
    int[] fdColumns = new int[numFds];
    int j = 0;
    for (Pair<LogicalVariable, Mutable<ILogicalExpression>> p : gby.getDecorList()) {
        ILogicalExpression expr = p.second.getValue();
        if (expr.getExpressionTag() != LogicalExpressionTag.VARIABLE) {
            throw new AlgebricksException("Sort group-by expects variable references.");
        }
        VariableReferenceExpression v = (VariableReferenceExpression) expr;
        LogicalVariable decor = v.getVariableReference();
        fdColumns[j++] = inputSchemas[0].findVariable(decor);
    }
    if (gby.getNestedPlans().size() != 1) {
        throw new AlgebricksException("Sort group-by currently works only for one nested plan with one root containing" + "an aggregate and a nested-tuple-source.");
    }
    ILogicalPlan p0 = gby.getNestedPlans().get(0);
    if (p0.getRoots().size() != 1) {
        throw new AlgebricksException("Sort group-by currently works only for one nested plan with one root containing" + "an aggregate and a nested-tuple-source.");
    }
    Mutable<ILogicalOperator> r0 = p0.getRoots().get(0);
    AggregateOperator aggOp = (AggregateOperator) r0.getValue();
    IPartialAggregationTypeComputer partialAggregationTypeComputer = context.getPartialAggregationTypeComputer();
    List<Object> intermediateTypes = new ArrayList<Object>();
    int n = aggOp.getExpressions().size();
    IAggregateEvaluatorFactory[] aff = new IAggregateEvaluatorFactory[n];
    int i = 0;
    IExpressionRuntimeProvider expressionRuntimeProvider = context.getExpressionRuntimeProvider();
    IVariableTypeEnvironment aggOpInputEnv = context.getTypeEnvironment(aggOp.getInputs().get(0).getValue());
    IVariableTypeEnvironment outputEnv = context.getTypeEnvironment(op);
    for (Mutable<ILogicalExpression> exprRef : aggOp.getExpressions()) {
        AggregateFunctionCallExpression aggFun = (AggregateFunctionCallExpression) exprRef.getValue();
        aff[i++] = expressionRuntimeProvider.createAggregateFunctionFactory(aggFun, aggOpInputEnv, inputSchemas, context);
        intermediateTypes.add(partialAggregationTypeComputer.getType(aggFun, aggOpInputEnv, context.getMetadataProvider()));
    }
    int[] keyAndDecFields = new int[keys.length + fdColumns.length];
    for (i = 0; i < keys.length; ++i) {
        keyAndDecFields[i] = keys[i];
    }
    for (i = 0; i < fdColumns.length; i++) {
        keyAndDecFields[keys.length + i] = fdColumns[i];
    }
    List<LogicalVariable> keyAndDecVariables = new ArrayList<LogicalVariable>();
    for (Pair<LogicalVariable, Mutable<ILogicalExpression>> p : gby.getGroupByList()) {
        keyAndDecVariables.add(p.first);
    }
    for (Pair<LogicalVariable, Mutable<ILogicalExpression>> p : gby.getDecorList()) {
        keyAndDecVariables.add(GroupByOperator.getDecorVariable(p));
    }
    for (LogicalVariable var : keyAndDecVariables) {
        aggOpInputEnv.setVarType(var, outputEnv.getVarType(var));
    }
    compileSubplans(inputSchemas[0], gby, opSchema, context);
    IOperatorDescriptorRegistry spec = builder.getJobSpec();
    IBinaryComparatorFactory[] compFactories = new IBinaryComparatorFactory[gbyCols.size()];
    IBinaryComparatorFactoryProvider bcfProvider = context.getBinaryComparatorFactoryProvider();
    i = 0;
    for (LogicalVariable v : gbyCols) {
        Object type = aggOpInputEnv.getVarType(v);
        if (orderColumns[i].getOrder() == OrderKind.ASC) {
            compFactories[i] = bcfProvider.getBinaryComparatorFactory(type, true);
        } else {
            compFactories[i] = bcfProvider.getBinaryComparatorFactory(type, false);
        }
        i++;
    }
    RecordDescriptor recordDescriptor = JobGenHelper.mkRecordDescriptor(context.getTypeEnvironment(op), opSchema, context);
    IAggregateEvaluatorFactory[] merges = new IAggregateEvaluatorFactory[n];
    List<LogicalVariable> usedVars = new ArrayList<LogicalVariable>();
    IOperatorSchema[] localInputSchemas = new IOperatorSchema[1];
    localInputSchemas[0] = new OperatorSchemaImpl();
    for (i = 0; i < n; i++) {
        AggregateFunctionCallExpression aggFun = (AggregateFunctionCallExpression) aggOp.getMergeExpressions().get(i).getValue();
        aggFun.getUsedVariables(usedVars);
    }
    i = 0;
    for (Object type : intermediateTypes) {
        aggOpInputEnv.setVarType(usedVars.get(i++), type);
    }
    for (LogicalVariable keyVar : keyAndDecVariables) {
        localInputSchemas[0].addVariable(keyVar);
    }
    for (LogicalVariable usedVar : usedVars) {
        localInputSchemas[0].addVariable(usedVar);
    }
    for (i = 0; i < n; i++) {
        AggregateFunctionCallExpression mergeFun = (AggregateFunctionCallExpression) aggOp.getMergeExpressions().get(i).getValue();
        merges[i] = expressionRuntimeProvider.createAggregateFunctionFactory(mergeFun, aggOpInputEnv, localInputSchemas, context);
    }
    RecordDescriptor partialAggRecordDescriptor = JobGenHelper.mkRecordDescriptor(context.getTypeEnvironment(op), localInputSchemas[0], context);
    IAggregatorDescriptorFactory aggregatorFactory = new SimpleAlgebricksAccumulatingAggregatorFactory(aff, keyAndDecFields);
    IAggregatorDescriptorFactory mergeFactory = new SimpleAlgebricksAccumulatingAggregatorFactory(merges, keyAndDecFields);
    INormalizedKeyComputerFactory normalizedKeyFactory = null;
    INormalizedKeyComputerFactoryProvider nkcfProvider = context.getNormalizedKeyComputerFactoryProvider();
    if (nkcfProvider == null) {
        normalizedKeyFactory = null;
    }
    Object type = aggOpInputEnv.getVarType(gbyCols.get(0));
    normalizedKeyFactory = orderColumns[0].getOrder() == OrderKind.ASC ? nkcfProvider.getNormalizedKeyComputerFactory(type, true) : nkcfProvider.getNormalizedKeyComputerFactory(type, false);
    SortGroupByOperatorDescriptor gbyOpDesc = new SortGroupByOperatorDescriptor(spec, frameLimit, keys, keyAndDecFields, normalizedKeyFactory, compFactories, aggregatorFactory, mergeFactory, partialAggRecordDescriptor, recordDescriptor, false);
    contributeOpDesc(builder, gby, gbyOpDesc);
    ILogicalOperator src = op.getInputs().get(0).getValue();
    builder.contributeGraphEdge(src, 0, op, 0);
}

Example 2 with SortGroupByOperatorDescriptor

use of org.apache.hyracks.dataflow.std.group.sort.SortGroupByOperatorDescriptor in project asterixdb by apache.

the class Groupby method createJob.

private static JobSpecification createJob(FileSplit[] inSplits, FileSplit[] outSplits, int htSize, long fileSize, int frameLimit, int frameSize, String alg, boolean outPlain) {
    JobSpecification spec = new JobSpecification(frameSize);
    IFileSplitProvider splitsProvider = new ConstantFileSplitProvider(inSplits);
    FileScanOperatorDescriptor fileScanner = new FileScanOperatorDescriptor(spec, splitsProvider, new DelimitedDataTupleParserFactory(lineitemParserFactories, '|'), lineitemDesc);
    createPartitionConstraint(spec, fileScanner, inSplits);
    // Output: each unique string with an integer count
    RecordDescriptor outDesc = new RecordDescriptor(new ISerializerDeserializer[] { IntegerSerializerDeserializer.INSTANCE, // IntegerSerializerDeserializer.INSTANCE,
    IntegerSerializerDeserializer.INSTANCE });
    // Specify the grouping key, which will be the string extracted during
    // the scan.
    int[] keys = new int[] { 0 };
    AbstractOperatorDescriptor grouper;
    if (alg.equalsIgnoreCase("hash")) {
        // external hash graph
        grouper = new ExternalGroupOperatorDescriptor(spec, htSize, fileSize, keys, frameLimit, new IBinaryComparatorFactory[] { // PointableBinaryComparatorFactory.of(IntegerPointable.FACTORY),
        PointableBinaryComparatorFactory.of(IntegerPointable.FACTORY) }, new IntegerNormalizedKeyComputerFactory(), new MultiFieldsAggregatorFactory(new IFieldAggregateDescriptorFactory[] { new CountFieldAggregatorFactory(false) }), new MultiFieldsAggregatorFactory(new IFieldAggregateDescriptorFactory[] { new IntSumFieldAggregatorFactory(keys.length, false) }), outDesc, outDesc, new HashSpillableTableFactory(new IBinaryHashFunctionFamily[] { MurmurHash3BinaryHashFunctionFamily.INSTANCE }));
        createPartitionConstraint(spec, grouper, outSplits);
    } else if (alg.equalsIgnoreCase("sort")) {
        grouper = new SortGroupByOperatorDescriptor(spec, frameLimit, keys, keys, new IntegerNormalizedKeyComputerFactory(), new IBinaryComparatorFactory[] { // PointableBinaryComparatorFactory.of(IntegerPointable.FACTORY),
        PointableBinaryComparatorFactory.of(IntegerPointable.FACTORY) }, new MultiFieldsAggregatorFactory(new IFieldAggregateDescriptorFactory[] { new CountFieldAggregatorFactory(true) }), new MultiFieldsAggregatorFactory(new IFieldAggregateDescriptorFactory[] { new IntSumFieldAggregatorFactory(keys.length, true) }), outDesc, outDesc, false);
        createPartitionConstraint(spec, grouper, outSplits);
    } else {
        System.err.println("unknow groupby alg:" + alg);
        return null;
    }
    // Connect scanner with the grouper
    IConnectorDescriptor scanGroupConnDef2 = new MToNPartitioningConnectorDescriptor(spec, new FieldHashPartitionComputerFactory(keys, new IBinaryHashFunctionFactory[] { // PointableBinaryHashFunctionFactory.of(IntegerPointable.FACTORY),
    PointableBinaryHashFunctionFactory.of(IntegerPointable.FACTORY) }));
    spec.connect(scanGroupConnDef2, fileScanner, 0, grouper, 0);
    IFileSplitProvider outSplitProvider = new ConstantFileSplitProvider(outSplits);
    AbstractSingleActivityOperatorDescriptor writer = outPlain ? new PlainFileWriterOperatorDescriptor(spec, outSplitProvider, "|") : new FrameFileWriterOperatorDescriptor(spec, outSplitProvider);
    createPartitionConstraint(spec, writer, outSplits);
    IConnectorDescriptor groupOutConn = new OneToOneConnectorDescriptor(spec);
    spec.connect(groupOutConn, grouper, 0, writer, 0);
    spec.addRoot(writer);
    return spec;
}