Examples with AbstractSingleActivityOperatorDescriptor org.apache.hyracks.dataflow.std.base.AbstractSingleActivityOperatorDescriptor used on opensource projects

Example 1 with AbstractSingleActivityOperatorDescriptor

use of org.apache.hyracks.dataflow.std.base.AbstractSingleActivityOperatorDescriptor in project asterixdb by apache.

the class SecondaryBTreeOperationsHelper method buildLoadingJobSpec.

@Override
public JobSpecification buildLoadingJobSpec() throws AlgebricksException {
    JobSpecification spec = RuntimeUtils.createJobSpecification(metadataProvider.getApplicationContext());
    boolean isEnforcingKeyTypes = index.isEnforcingKeyFileds();
    int[] fieldPermutation = createFieldPermutationForBulkLoadOp(index.getKeyFieldNames().size());
    IIndexDataflowHelperFactory dataflowHelperFactory = new IndexDataflowHelperFactory(metadataProvider.getStorageComponentProvider().getStorageManager(), secondaryFileSplitProvider);
    if (dataset.getDatasetType() == DatasetType.EXTERNAL) {
        /*
             * In case of external data,
             * this method is used to build loading jobs for both initial load on index creation
             * and transaction load on dataset referesh
             */
        // Create external indexing scan operator
        ExternalScanOperatorDescriptor primaryScanOp = createExternalIndexingOp(spec);
        // Assign op.
        AbstractOperatorDescriptor sourceOp = primaryScanOp;
        if (isEnforcingKeyTypes && !enforcedItemType.equals(itemType)) {
            sourceOp = createCastOp(spec, dataset.getDatasetType());
            spec.connect(new OneToOneConnectorDescriptor(spec), primaryScanOp, 0, sourceOp, 0);
        }
        AlgebricksMetaOperatorDescriptor asterixAssignOp = createExternalAssignOp(spec, index.getKeyFieldNames().size(), secondaryRecDesc);
        // If any of the secondary fields are nullable, then add a select op that filters nulls.
        AlgebricksMetaOperatorDescriptor selectOp = null;
        if (anySecondaryKeyIsNullable || isEnforcingKeyTypes) {
            selectOp = createFilterNullsSelectOp(spec, index.getKeyFieldNames().size(), secondaryRecDesc);
        }
        // Sort by secondary keys.
        ExternalSortOperatorDescriptor sortOp = createSortOp(spec, secondaryComparatorFactories, secondaryRecDesc);
        // Create secondary BTree bulk load op.
        AbstractSingleActivityOperatorDescriptor secondaryBulkLoadOp;
        IOperatorDescriptor root;
        if (externalFiles != null) {
            // Transaction load
            secondaryBulkLoadOp = createExternalIndexBulkModifyOp(spec, fieldPermutation, dataflowHelperFactory, GlobalConfig.DEFAULT_TREE_FILL_FACTOR);
        } else {
            // Initial load
            secondaryBulkLoadOp = createExternalIndexBulkLoadOp(spec, fieldPermutation, dataflowHelperFactory, GlobalConfig.DEFAULT_TREE_FILL_FACTOR);
        }
        AlgebricksMetaOperatorDescriptor metaOp = new AlgebricksMetaOperatorDescriptor(spec, 1, 0, new IPushRuntimeFactory[] { new SinkRuntimeFactory() }, new RecordDescriptor[] { secondaryRecDesc });
        spec.connect(new OneToOneConnectorDescriptor(spec), secondaryBulkLoadOp, 0, metaOp, 0);
        root = metaOp;
        spec.connect(new OneToOneConnectorDescriptor(spec), sourceOp, 0, asterixAssignOp, 0);
        if (anySecondaryKeyIsNullable || isEnforcingKeyTypes) {
            spec.connect(new OneToOneConnectorDescriptor(spec), asterixAssignOp, 0, selectOp, 0);
            spec.connect(new OneToOneConnectorDescriptor(spec), selectOp, 0, sortOp, 0);
        } else {
            spec.connect(new OneToOneConnectorDescriptor(spec), asterixAssignOp, 0, sortOp, 0);
        }
        spec.connect(new OneToOneConnectorDescriptor(spec), sortOp, 0, secondaryBulkLoadOp, 0);
        spec.addRoot(root);
        spec.setConnectorPolicyAssignmentPolicy(new ConnectorPolicyAssignmentPolicy());
        return spec;
    } else {
        // Create dummy key provider for feeding the primary index scan.
        IOperatorDescriptor keyProviderOp = DatasetUtil.createDummyKeyProviderOp(spec, dataset, metadataProvider);
        JobId jobId = IndexUtil.bindJobEventListener(spec, metadataProvider);
        // Create primary index scan op.
        IOperatorDescriptor primaryScanOp = DatasetUtil.createPrimaryIndexScanOp(spec, metadataProvider, dataset, jobId);
        // Assign op.
        IOperatorDescriptor sourceOp = primaryScanOp;
        if (isEnforcingKeyTypes && !enforcedItemType.equals(itemType)) {
            sourceOp = createCastOp(spec, dataset.getDatasetType());
            spec.connect(new OneToOneConnectorDescriptor(spec), primaryScanOp, 0, sourceOp, 0);
        }
        AlgebricksMetaOperatorDescriptor asterixAssignOp = createAssignOp(spec, index.getKeyFieldNames().size(), secondaryRecDesc);
        // If any of the secondary fields are nullable, then add a select op that filters nulls.
        AlgebricksMetaOperatorDescriptor selectOp = null;
        if (anySecondaryKeyIsNullable || isEnforcingKeyTypes) {
            selectOp = createFilterNullsSelectOp(spec, index.getKeyFieldNames().size(), secondaryRecDesc);
        }
        // Sort by secondary keys.
        ExternalSortOperatorDescriptor sortOp = createSortOp(spec, secondaryComparatorFactories, secondaryRecDesc);
        // Create secondary BTree bulk load op.
        TreeIndexBulkLoadOperatorDescriptor secondaryBulkLoadOp = createTreeIndexBulkLoadOp(spec, fieldPermutation, dataflowHelperFactory, GlobalConfig.DEFAULT_TREE_FILL_FACTOR);
        AlgebricksMetaOperatorDescriptor metaOp = new AlgebricksMetaOperatorDescriptor(spec, 1, 0, new IPushRuntimeFactory[] { new SinkRuntimeFactory() }, new RecordDescriptor[] { secondaryRecDesc });
        // Connect the operators.
        spec.connect(new OneToOneConnectorDescriptor(spec), keyProviderOp, 0, primaryScanOp, 0);
        spec.connect(new OneToOneConnectorDescriptor(spec), sourceOp, 0, asterixAssignOp, 0);
        if (anySecondaryKeyIsNullable || isEnforcingKeyTypes) {
            spec.connect(new OneToOneConnectorDescriptor(spec), asterixAssignOp, 0, selectOp, 0);
            spec.connect(new OneToOneConnectorDescriptor(spec), selectOp, 0, sortOp, 0);
        } else {
            spec.connect(new OneToOneConnectorDescriptor(spec), asterixAssignOp, 0, sortOp, 0);
        }
        spec.connect(new OneToOneConnectorDescriptor(spec), sortOp, 0, secondaryBulkLoadOp, 0);
        spec.connect(new OneToOneConnectorDescriptor(spec), secondaryBulkLoadOp, 0, metaOp, 0);
        spec.addRoot(metaOp);
        spec.setConnectorPolicyAssignmentPolicy(new ConnectorPolicyAssignmentPolicy());
        return spec;
    }
}

Example 2 with AbstractSingleActivityOperatorDescriptor

use of org.apache.hyracks.dataflow.std.base.AbstractSingleActivityOperatorDescriptor in project asterixdb by apache.

the class SecondaryRTreeOperationsHelper method buildLoadingJobSpec.

@Override
public JobSpecification buildLoadingJobSpec() throws AsterixException, AlgebricksException {
    /***************************************************
         * [ About PointMBR Optimization ]
         * Instead of storing a MBR(4 doubles) for a point(2 doubles) in RTree leaf node,
         * PointMBR concept is introduced.
         * PointMBR is a way to store a point as 2 doubles in RTree leaf node.
         * This reduces RTree index size roughly in half.
         * In order to fully benefit from the PointMBR concept, besides RTree,
         * external sort operator during bulk-loading (from either data loading or index creation)
         * must deal with point as 2 doubles instead of 4 doubles. Otherwise, external sort will suffer from twice as
         * many doubles as it actually requires. For this purpose,
         * PointMBR specific optimization logic is added as follows:
         * 1) CreateMBR function in assign operator generates 2 doubles, instead of 4 doubles.
         * 2) External sort operator sorts points represented with 2 doubles.
         * 3) Bulk-loading in RTree takes 4 doubles by reading 2 doubles twice and then,
         * do the same work as non-point MBR cases.
         ***************************************************/
    JobSpecification spec = RuntimeUtils.createJobSpecification(metadataProvider.getApplicationContext());
    int[] fieldPermutation = createFieldPermutationForBulkLoadOp(numNestedSecondaryKeyFields);
    int numNestedSecondaryKeFieldsConsideringPointMBR = isPointMBR ? numNestedSecondaryKeyFields / 2 : numNestedSecondaryKeyFields;
    RecordDescriptor secondaryRecDescConsideringPointMBR = isPointMBR ? secondaryRecDescForPointMBR : secondaryRecDesc;
    boolean isEnforcingKeyTypes = index.isEnforcingKeyFileds();
    IIndexDataflowHelperFactory indexDataflowHelperFactory = new IndexDataflowHelperFactory(metadataProvider.getStorageComponentProvider().getStorageManager(), secondaryFileSplitProvider);
    if (dataset.getDatasetType() == DatasetType.INTERNAL) {
        // Create dummy key provider for feeding the primary index scan.
        IOperatorDescriptor keyProviderOp = DatasetUtil.createDummyKeyProviderOp(spec, dataset, metadataProvider);
        JobId jobId = IndexUtil.bindJobEventListener(spec, metadataProvider);
        // Create primary index scan op.
        IOperatorDescriptor primaryScanOp = DatasetUtil.createPrimaryIndexScanOp(spec, metadataProvider, dataset, jobId);
        // Assign op.
        IOperatorDescriptor sourceOp = primaryScanOp;
        if (isEnforcingKeyTypes && !enforcedItemType.equals(itemType)) {
            sourceOp = createCastOp(spec, dataset.getDatasetType());
            spec.connect(new OneToOneConnectorDescriptor(spec), primaryScanOp, 0, sourceOp, 0);
        }
        AlgebricksMetaOperatorDescriptor asterixAssignOp = createAssignOp(spec, numNestedSecondaryKeFieldsConsideringPointMBR, secondaryRecDescConsideringPointMBR);
        // If any of the secondary fields are nullable, then add a select op that filters nulls.
        AlgebricksMetaOperatorDescriptor selectOp = null;
        if (anySecondaryKeyIsNullable || isEnforcingKeyTypes) {
            selectOp = createFilterNullsSelectOp(spec, numNestedSecondaryKeFieldsConsideringPointMBR, secondaryRecDescConsideringPointMBR);
        }
        // Sort by secondary keys.
        ExternalSortOperatorDescriptor sortOp = createSortOp(spec, new IBinaryComparatorFactory[] { MetadataProvider.proposeLinearizer(keyType, secondaryComparatorFactories.length) }, isPointMBR ? secondaryRecDescForPointMBR : secondaryRecDesc);
        // Create secondary RTree bulk load op.
        TreeIndexBulkLoadOperatorDescriptor secondaryBulkLoadOp = createTreeIndexBulkLoadOp(spec, fieldPermutation, indexDataflowHelperFactory, GlobalConfig.DEFAULT_TREE_FILL_FACTOR);
        AlgebricksMetaOperatorDescriptor metaOp = new AlgebricksMetaOperatorDescriptor(spec, 1, 0, new IPushRuntimeFactory[] { new SinkRuntimeFactory() }, new RecordDescriptor[] {});
        // Connect the operators.
        spec.connect(new OneToOneConnectorDescriptor(spec), keyProviderOp, 0, primaryScanOp, 0);
        spec.connect(new OneToOneConnectorDescriptor(spec), sourceOp, 0, asterixAssignOp, 0);
        if (anySecondaryKeyIsNullable || isEnforcingKeyTypes) {
            spec.connect(new OneToOneConnectorDescriptor(spec), asterixAssignOp, 0, selectOp, 0);
            spec.connect(new OneToOneConnectorDescriptor(spec), selectOp, 0, sortOp, 0);
        } else {
            spec.connect(new OneToOneConnectorDescriptor(spec), asterixAssignOp, 0, sortOp, 0);
        }
        spec.connect(new OneToOneConnectorDescriptor(spec), sortOp, 0, secondaryBulkLoadOp, 0);
        spec.connect(new OneToOneConnectorDescriptor(spec), secondaryBulkLoadOp, 0, metaOp, 0);
        spec.addRoot(metaOp);
        spec.setConnectorPolicyAssignmentPolicy(new ConnectorPolicyAssignmentPolicy());
    } else {
        // External dataset
        /*
             * In case of external data, this method is used to build loading jobs for both
             * initial load on index creation
             * and transaction load on dataset referesh
             */
        // Create external indexing scan operator
        ExternalScanOperatorDescriptor primaryScanOp = createExternalIndexingOp(spec);
        AbstractOperatorDescriptor sourceOp = primaryScanOp;
        if (isEnforcingKeyTypes && !enforcedItemType.equals(itemType)) {
            sourceOp = createCastOp(spec, dataset.getDatasetType());
            spec.connect(new OneToOneConnectorDescriptor(spec), primaryScanOp, 0, sourceOp, 0);
        }
        // Assign op.
        AlgebricksMetaOperatorDescriptor asterixAssignOp = createExternalAssignOp(spec, numNestedSecondaryKeFieldsConsideringPointMBR, secondaryRecDescConsideringPointMBR);
        // If any of the secondary fields are nullable, then add a select op that filters nulls.
        AlgebricksMetaOperatorDescriptor selectOp = null;
        if (anySecondaryKeyIsNullable || isEnforcingKeyTypes) {
            selectOp = createFilterNullsSelectOp(spec, numNestedSecondaryKeFieldsConsideringPointMBR, secondaryRecDescConsideringPointMBR);
        }
        // Sort by secondary keys.
        ExternalSortOperatorDescriptor sortOp = createSortOp(spec, new IBinaryComparatorFactory[] { MetadataProvider.proposeLinearizer(keyType, secondaryComparatorFactories.length) }, isPointMBR ? secondaryRecDescForPointMBR : secondaryRecDesc);
        // Create secondary RTree bulk load op.
        IOperatorDescriptor root;
        AbstractSingleActivityOperatorDescriptor secondaryBulkLoadOp;
        if (externalFiles != null) {
            // Transaction load
            secondaryBulkLoadOp = createExternalIndexBulkModifyOp(spec, fieldPermutation, indexDataflowHelperFactory, GlobalConfig.DEFAULT_TREE_FILL_FACTOR);
        } else {
            // Initial load
            secondaryBulkLoadOp = createExternalIndexBulkLoadOp(spec, fieldPermutation, indexDataflowHelperFactory, GlobalConfig.DEFAULT_TREE_FILL_FACTOR);
        }
        AlgebricksMetaOperatorDescriptor metaOp = new AlgebricksMetaOperatorDescriptor(spec, 1, 0, new IPushRuntimeFactory[] { new SinkRuntimeFactory() }, new RecordDescriptor[] { secondaryRecDesc });
        spec.connect(new OneToOneConnectorDescriptor(spec), secondaryBulkLoadOp, 0, metaOp, 0);
        root = metaOp;
        spec.connect(new OneToOneConnectorDescriptor(spec), sourceOp, 0, asterixAssignOp, 0);
        if (anySecondaryKeyIsNullable || isEnforcingKeyTypes) {
            spec.connect(new OneToOneConnectorDescriptor(spec), asterixAssignOp, 0, selectOp, 0);
            spec.connect(new OneToOneConnectorDescriptor(spec), selectOp, 0, sortOp, 0);
        } else {
            spec.connect(new OneToOneConnectorDescriptor(spec), asterixAssignOp, 0, sortOp, 0);
        }
        spec.connect(new OneToOneConnectorDescriptor(spec), sortOp, 0, secondaryBulkLoadOp, 0);
        spec.addRoot(root);
        spec.setConnectorPolicyAssignmentPolicy(new ConnectorPolicyAssignmentPolicy());
    }
    return spec;
}

Example 3 with AbstractSingleActivityOperatorDescriptor

use of org.apache.hyracks.dataflow.std.base.AbstractSingleActivityOperatorDescriptor in project asterixdb by apache.

the class AggregationTest method multiKeyMinMaxStringPreClusterGroupTest.

@Test
public void multiKeyMinMaxStringPreClusterGroupTest() throws Exception {
    JobSpecification spec = new JobSpecification();
    FileScanOperatorDescriptor csvScanner = new FileScanOperatorDescriptor(spec, splitProvider, tupleParserFactory, desc);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, csvScanner, NC2_ID);
    RecordDescriptor outputRec = new RecordDescriptor(new ISerializerDeserializer[] { new UTF8StringSerializerDeserializer(), new UTF8StringSerializerDeserializer(), IntegerSerializerDeserializer.INSTANCE, new UTF8StringSerializerDeserializer() });
    int[] keyFields = new int[] { 8, 0 };
    PreclusteredGroupOperatorDescriptor grouper = new PreclusteredGroupOperatorDescriptor(spec, keyFields, new IBinaryComparatorFactory[] { PointableBinaryComparatorFactory.of(UTF8StringPointable.FACTORY), PointableBinaryComparatorFactory.of(UTF8StringPointable.FACTORY) }, new MultiFieldsAggregatorFactory(new IFieldAggregateDescriptorFactory[] { new IntSumFieldAggregatorFactory(1, true), new MinMaxStringFieldAggregatorFactory(15, true, false) }), outputRec);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, grouper, NC2_ID, NC1_ID);
    IConnectorDescriptor conn1 = new MToNPartitioningConnectorDescriptor(spec, new FieldHashPartitionComputerFactory(keyFields, new IBinaryHashFunctionFactory[] { PointableBinaryHashFunctionFactory.of(UTF8StringPointable.FACTORY), PointableBinaryHashFunctionFactory.of(UTF8StringPointable.FACTORY) }));
    spec.connect(conn1, csvScanner, 0, grouper, 0);
    AbstractSingleActivityOperatorDescriptor printer = getPrinter(spec, "multiKeyMinMaxStringPreClusterGroupTest");
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, printer, NC2_ID, NC1_ID);
    IConnectorDescriptor conn2 = new OneToOneConnectorDescriptor(spec);
    spec.connect(conn2, grouper, 0, printer, 0);
    spec.addRoot(printer);
    runTest(spec);
}

Example 4 with AbstractSingleActivityOperatorDescriptor

use of org.apache.hyracks.dataflow.std.base.AbstractSingleActivityOperatorDescriptor in project asterixdb by apache.

the class AggregationTest method singleKeyAvgPreClusterGroupTest.

@Test
public void singleKeyAvgPreClusterGroupTest() throws Exception {
    JobSpecification spec = new JobSpecification();
    FileScanOperatorDescriptor csvScanner = new FileScanOperatorDescriptor(spec, splitProvider, tupleParserFactory, desc);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, csvScanner, NC2_ID);
    RecordDescriptor outputRec = new RecordDescriptor(new ISerializerDeserializer[] { new UTF8StringSerializerDeserializer(), IntegerSerializerDeserializer.INSTANCE, IntegerSerializerDeserializer.INSTANCE, FloatSerializerDeserializer.INSTANCE });
    int[] keyFields = new int[] { 0 };
    PreclusteredGroupOperatorDescriptor grouper = new PreclusteredGroupOperatorDescriptor(spec, keyFields, new IBinaryComparatorFactory[] { PointableBinaryComparatorFactory.of(UTF8StringPointable.FACTORY) }, new MultiFieldsAggregatorFactory(new IFieldAggregateDescriptorFactory[] { new IntSumFieldAggregatorFactory(1, true), new CountFieldAggregatorFactory(true), new AvgFieldGroupAggregatorFactory(1, true) }), outputRec);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, grouper, NC2_ID, NC1_ID);
    IConnectorDescriptor conn1 = new MToNPartitioningConnectorDescriptor(spec, new FieldHashPartitionComputerFactory(keyFields, new IBinaryHashFunctionFactory[] { PointableBinaryHashFunctionFactory.of(UTF8StringPointable.FACTORY) }));
    spec.connect(conn1, csvScanner, 0, grouper, 0);
    AbstractSingleActivityOperatorDescriptor printer = getPrinter(spec, "singleKeyAvgInmemGroupTest");
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, printer, NC2_ID, NC1_ID);
    IConnectorDescriptor conn2 = new OneToOneConnectorDescriptor(spec);
    spec.connect(conn2, grouper, 0, printer, 0);
    spec.addRoot(printer);
    runTest(spec);
}

Example 5 with AbstractSingleActivityOperatorDescriptor

use of org.apache.hyracks.dataflow.std.base.AbstractSingleActivityOperatorDescriptor in project asterixdb by apache.

the class AggregationTest method singleKeySumPreClusterGroupTest.

@Test
public void singleKeySumPreClusterGroupTest() throws Exception {
    JobSpecification spec = new JobSpecification();
    FileScanOperatorDescriptor csvScanner = new FileScanOperatorDescriptor(spec, splitProvider, tupleParserFactory, desc);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, csvScanner, NC2_ID);
    RecordDescriptor outputRec = new RecordDescriptor(new ISerializerDeserializer[] { new UTF8StringSerializerDeserializer(), IntegerSerializerDeserializer.INSTANCE, IntegerSerializerDeserializer.INSTANCE, FloatSerializerDeserializer.INSTANCE });
    int[] keyFields = new int[] { 0 };
    PreclusteredGroupOperatorDescriptor grouper = new PreclusteredGroupOperatorDescriptor(spec, keyFields, new IBinaryComparatorFactory[] { PointableBinaryComparatorFactory.of(UTF8StringPointable.FACTORY) }, new MultiFieldsAggregatorFactory(new IFieldAggregateDescriptorFactory[] { new IntSumFieldAggregatorFactory(1, true), new IntSumFieldAggregatorFactory(3, true), new FloatSumFieldAggregatorFactory(5, true) }), outputRec);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, grouper, NC2_ID, NC1_ID);
    IConnectorDescriptor conn1 = new MToNPartitioningConnectorDescriptor(spec, new FieldHashPartitionComputerFactory(keyFields, new IBinaryHashFunctionFactory[] { PointableBinaryHashFunctionFactory.of(UTF8StringPointable.FACTORY) }));
    spec.connect(conn1, csvScanner, 0, grouper, 0);
    AbstractSingleActivityOperatorDescriptor printer = getPrinter(spec, "singleKeySumInmemGroupTest");
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, printer, NC2_ID, NC1_ID);
    IConnectorDescriptor conn2 = new OneToOneConnectorDescriptor(spec);
    spec.connect(conn2, grouper, 0, printer, 0);
    spec.addRoot(printer);
    runTest(spec);
}