Examples with TreeIndexBulkLoadOperatorDescriptor org.apache.hyracks.storage.am.common.dataflow.TreeIndexBulkLoadOperatorDescriptor used on opensource projects

Example 6 with TreeIndexBulkLoadOperatorDescriptor

use of org.apache.hyracks.storage.am.common.dataflow.TreeIndexBulkLoadOperatorDescriptor in project asterixdb by apache.

the class AbstractBTreeOperatorTest method loadPrimaryIndex.

protected void loadPrimaryIndex() throws Exception {
    JobSpecification spec = new JobSpecification();
    FileSplit[] ordersSplits = new FileSplit[] { new ManagedFileSplit(NC1_ID, "data" + File.separator + "tpch0.001" + File.separator + "orders-part1.tbl") };
    IFileSplitProvider ordersSplitProvider = new ConstantFileSplitProvider(ordersSplits);
    RecordDescriptor ordersDesc = inputRecordDesc;
    FileScanOperatorDescriptor ordScanner = new FileScanOperatorDescriptor(spec, ordersSplitProvider, new DelimitedDataTupleParserFactory(inputParserFactories, '|'), ordersDesc);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, ordScanner, NC1_ID);
    ExternalSortOperatorDescriptor sorter = new ExternalSortOperatorDescriptor(spec, 1000, new int[] { 0 }, new IBinaryComparatorFactory[] { PointableBinaryComparatorFactory.of(UTF8StringPointable.FACTORY) }, ordersDesc);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, sorter, NC1_ID);
    int[] fieldPermutation = { 0, 1, 2, 4, 5, 7 };
    TreeIndexBulkLoadOperatorDescriptor primaryBtreeBulkLoad = new TreeIndexBulkLoadOperatorDescriptor(spec, primaryRecDesc, fieldPermutation, 0.7f, true, 1000L, true, primaryHelperFactory);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, primaryBtreeBulkLoad, NC1_ID);
    NullSinkOperatorDescriptor nsOpDesc = new NullSinkOperatorDescriptor(spec);
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, nsOpDesc, NC1_ID);
    spec.connect(new OneToOneConnectorDescriptor(spec), ordScanner, 0, sorter, 0);
    spec.connect(new OneToOneConnectorDescriptor(spec), sorter, 0, primaryBtreeBulkLoad, 0);
    spec.connect(new OneToOneConnectorDescriptor(spec), primaryBtreeBulkLoad, 0, nsOpDesc, 0);
    spec.addRoot(nsOpDesc);
    runTest(spec);
}

Example 7 with TreeIndexBulkLoadOperatorDescriptor

use of org.apache.hyracks.storage.am.common.dataflow.TreeIndexBulkLoadOperatorDescriptor in project asterixdb by apache.

the class PrimaryIndexBulkLoadExample method createJob.

private static JobSpecification createJob(Options options) {
    JobSpecification spec = new JobSpecification(options.frameSize);
    String[] splitNCs = options.ncs.split(",");
    // schema of tuples to be generated: 5 fields with string, string, int,
    // int, string
    // we will use field-index 2 as primary key to fill a clustered index
    RecordDescriptor recDesc = new RecordDescriptor(new ISerializerDeserializer[] { // this field will not go into B-Tree
    new UTF8StringSerializerDeserializer(), // we will use this as payload
    new UTF8StringSerializerDeserializer(), // we will use this field as key
    IntegerSerializerDeserializer.INSTANCE, // we will use this as payload
    IntegerSerializerDeserializer.INSTANCE, // we will use this as payload
    new UTF8StringSerializerDeserializer() });
    // generate numRecords records with field 2 being unique, integer values
    // in [0, 100000], and strings with max length of 10 characters, and
    // random seed 50
    DataGenOperatorDescriptor dataGen = new DataGenOperatorDescriptor(spec, recDesc, options.numTuples, 2, 0, 100000, 10, 50);
    // run data generator on first nodecontroller given
    PartitionConstraintHelper.addAbsoluteLocationConstraint(spec, dataGen, splitNCs[0]);
    // sort the tuples as preparation for bulk load
    // fields to sort on
    int[] sortFields = { 2 };
    // comparators for sort fields
    IBinaryComparatorFactory[] comparatorFactories = new IBinaryComparatorFactory[1];
    comparatorFactories[0] = PointableBinaryComparatorFactory.of(IntegerPointable.FACTORY);
    ExternalSortOperatorDescriptor sorter = new ExternalSortOperatorDescriptor(spec, options.sbSize, sortFields, comparatorFactories, recDesc);
    JobHelper.createPartitionConstraint(spec, sorter, splitNCs);
    // tuples to be put into B-Tree shall have 4 fields
    int fieldCount = 4;
    ITypeTraits[] typeTraits = new ITypeTraits[fieldCount];
    typeTraits[0] = IntegerPointable.TYPE_TRAITS;
    typeTraits[1] = UTF8StringPointable.TYPE_TRAITS;
    typeTraits[2] = IntegerPointable.TYPE_TRAITS;
    typeTraits[3] = UTF8StringPointable.TYPE_TRAITS;
    // create providers for B-Tree
    IStorageManager storageManager = BTreeHelperStorageManager.INSTANCE;
    // the B-Tree expects its keyfields to be at the front of its input
    // tuple
    // map field 2 of input tuple
    int[] fieldPermutation = { 2, 1, 3, 4 };
    // to field 0 of B-Tree tuple,
    // etc.
    IFileSplitProvider btreeSplitProvider = JobHelper.createFileSplitProvider(splitNCs, options.btreeName);
    IIndexDataflowHelperFactory dataflowHelperFactory = new IndexDataflowHelperFactory(storageManager, btreeSplitProvider);
    TreeIndexBulkLoadOperatorDescriptor btreeBulkLoad = new TreeIndexBulkLoadOperatorDescriptor(spec, recDesc, fieldPermutation, 0.7f, false, 1000L, true, dataflowHelperFactory);
    JobHelper.createPartitionConstraint(spec, btreeBulkLoad, splitNCs);
    // distribute the records from the datagen via hashing to the bulk load
    // ops
    IBinaryHashFunctionFactory[] hashFactories = new IBinaryHashFunctionFactory[1];
    hashFactories[0] = PointableBinaryHashFunctionFactory.of(UTF8StringPointable.FACTORY);
    IConnectorDescriptor hashConn = new MToNPartitioningConnectorDescriptor(spec, new FieldHashPartitionComputerFactory(new int[] { 0 }, hashFactories));
    NullSinkOperatorDescriptor nsOpDesc = new NullSinkOperatorDescriptor(spec);
    JobHelper.createPartitionConstraint(spec, nsOpDesc, splitNCs);
    spec.connect(hashConn, dataGen, 0, sorter, 0);
    spec.connect(new OneToOneConnectorDescriptor(spec), sorter, 0, btreeBulkLoad, 0);
    spec.connect(new OneToOneConnectorDescriptor(spec), btreeBulkLoad, 0, nsOpDesc, 0);
    spec.addRoot(nsOpDesc);
    return spec;
}

Example 8 with TreeIndexBulkLoadOperatorDescriptor

use of org.apache.hyracks.storage.am.common.dataflow.TreeIndexBulkLoadOperatorDescriptor 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 9 with TreeIndexBulkLoadOperatorDescriptor

use of org.apache.hyracks.storage.am.common.dataflow.TreeIndexBulkLoadOperatorDescriptor 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 10 with TreeIndexBulkLoadOperatorDescriptor

use of org.apache.hyracks.storage.am.common.dataflow.TreeIndexBulkLoadOperatorDescriptor in project asterixdb by apache.

the class SecondaryIndexOperationsHelper method createTreeIndexBulkLoadOp.

protected TreeIndexBulkLoadOperatorDescriptor createTreeIndexBulkLoadOp(JobSpecification spec, int[] fieldPermutation, IIndexDataflowHelperFactory dataflowHelperFactory, float fillFactor) throws AlgebricksException {
    TreeIndexBulkLoadOperatorDescriptor treeIndexBulkLoadOp = new TreeIndexBulkLoadOperatorDescriptor(spec, secondaryRecDesc, fieldPermutation, fillFactor, false, numElementsHint, false, dataflowHelperFactory);
    AlgebricksPartitionConstraintHelper.setPartitionConstraintInJobSpec(spec, treeIndexBulkLoadOp, secondaryPartitionConstraint);
    return treeIndexBulkLoadOp;
}