Examples with ITupleReference org.apache.hyracks.dataflow.common.data.accessors.ITupleReference used on opensource projects

Example 56 with ITupleReference

use of org.apache.hyracks.dataflow.common.data.accessors.ITupleReference in project asterixdb by apache.

the class LSMInvertedIndex method flush.

@Override
public ILSMDiskComponent flush(ILSMIOOperation operation) throws HyracksDataException {
    LSMInvertedIndexFlushOperation flushOp = (LSMInvertedIndexFlushOperation) operation;
    // Create an inverted index instance to be bulk loaded.
    LSMInvertedIndexDiskComponent component = createDiskInvIndexComponent(componentFactory, flushOp.getTarget(), flushOp.getDeletedKeysBTreeTarget(), flushOp.getBloomFilterTarget(), true);
    // Create a scan cursor on the BTree underlying the in-memory inverted index.
    LSMInvertedIndexMemoryComponent flushingComponent = (LSMInvertedIndexMemoryComponent) flushOp.getFlushingComponent();
    RangePredicate nullPred = new RangePredicate(null, null, true, true, null, null);
    // Search the deleted keys BTree to calculate the number of elements for BloomFilter
    IIndexAccessor deletedKeysBTreeAccessor = flushingComponent.getDeletedKeysBTree().createAccessor(NoOpOperationCallback.INSTANCE, NoOpOperationCallback.INSTANCE);
    IIndexCursor btreeCountingCursor = ((BTreeAccessor) deletedKeysBTreeAccessor).createCountingSearchCursor();
    deletedKeysBTreeAccessor.search(btreeCountingCursor, nullPred);
    long numBTreeTuples = 0L;
    try {
        while (btreeCountingCursor.hasNext()) {
            btreeCountingCursor.next();
            ITupleReference countTuple = btreeCountingCursor.getTuple();
            numBTreeTuples = IntegerPointable.getInteger(countTuple.getFieldData(0), countTuple.getFieldStart(0));
        }
    } finally {
        btreeCountingCursor.close();
    }
    ILSMDiskComponentBulkLoader componentBulkLoader = createComponentBulkLoader(component, 1.0f, false, numBTreeTuples, false, false);
    // Create a scan cursor on the deleted keys BTree underlying the in-memory inverted index.
    IIndexCursor deletedKeysScanCursor = deletedKeysBTreeAccessor.createSearchCursor(false);
    deletedKeysBTreeAccessor.search(deletedKeysScanCursor, nullPred);
    try {
        while (deletedKeysScanCursor.hasNext()) {
            deletedKeysScanCursor.next();
            ((LSMInvertedIndexDiskComponentBulkLoader) componentBulkLoader).delete(deletedKeysScanCursor.getTuple());
        }
    } finally {
        deletedKeysScanCursor.close();
    }
    // Scan the in-memory inverted index
    InMemoryInvertedIndexAccessor memInvIndexAccessor = (InMemoryInvertedIndexAccessor) flushingComponent.getInvIndex().createAccessor(NoOpOperationCallback.INSTANCE, NoOpOperationCallback.INSTANCE);
    BTreeAccessor memBTreeAccessor = memInvIndexAccessor.getBTreeAccessor();
    IIndexCursor scanCursor = memBTreeAccessor.createSearchCursor(false);
    memBTreeAccessor.search(scanCursor, nullPred);
    // Bulk load the disk inverted index from the in-memory inverted index.
    try {
        while (scanCursor.hasNext()) {
            scanCursor.next();
            componentBulkLoader.add(scanCursor.getTuple());
        }
    } finally {
        scanCursor.close();
    }
    if (component.getLSMComponentFilter() != null) {
        List<ITupleReference> filterTuples = new ArrayList<>();
        filterTuples.add(flushingComponent.getLSMComponentFilter().getMinTuple());
        filterTuples.add(flushingComponent.getLSMComponentFilter().getMaxTuple());
        filterManager.updateFilter(component.getLSMComponentFilter(), filterTuples);
        filterManager.writeFilter(component.getLSMComponentFilter(), ((OnDiskInvertedIndex) component.getInvIndex()).getBTree());
    }
    flushingComponent.getMetadata().copy(component.getMetadata());
    componentBulkLoader.end();
    return component;
}

Example 57 with ITupleReference

use of org.apache.hyracks.dataflow.common.data.accessors.ITupleReference in project asterixdb by apache.

the class InMemoryInvertedIndex method delete.

public void delete(ITupleReference tuple, BTreeAccessor btreeAccessor, IIndexOperationContext ictx) throws HyracksDataException {
    InMemoryInvertedIndexOpContext ctx = (InMemoryInvertedIndexOpContext) ictx;
    ctx.getTupleIter().reset(tuple);
    while (ctx.getTupleIter().hasNext()) {
        ctx.getTupleIter().next();
        ITupleReference deleteTuple = ctx.getTupleIter().getTuple();
        try {
            btreeAccessor.delete(deleteTuple);
        } catch (HyracksDataException e) {
            if (e.getErrorCode() != ErrorCode.UPDATE_OR_DELETE_NON_EXISTENT_KEY) {
                // Ignore this exception, since a document may have duplicate tokens.
                throw e;
            }
        }
    }
}

Example 58 with ITupleReference

use of org.apache.hyracks.dataflow.common.data.accessors.ITupleReference in project asterixdb by apache.

the class InMemoryInvertedListCursor method printInvList.

@SuppressWarnings("rawtypes")
@Override
public String printInvList(ISerializerDeserializer[] serdes) throws HyracksDataException {
    StringBuilder strBuilder = new StringBuilder();
    try {
        while (btreeCursor.hasNext()) {
            btreeCursor.next();
            ITupleReference tuple = btreeCursor.getTuple();
            ByteArrayInputStream inStream = new ByteArrayInputStream(tuple.getFieldData(1), tuple.getFieldStart(1), tuple.getFieldLength(1));
            DataInput dataIn = new DataInputStream(inStream);
            Object o = serdes[0].deserialize(dataIn);
            strBuilder.append(o.toString() + " ");
        }
    } finally {
        btreeCursor.close();
        btreeCursor.reset();
    }
    btreeAccessor.search(btreeCursor, btreePred);
    return strBuilder.toString();
}

Example 59 with ITupleReference

use of org.apache.hyracks.dataflow.common.data.accessors.ITupleReference in project asterixdb by apache.

the class OnDiskInvertedIndex method validate.

@Override
public void validate() throws HyracksDataException {
    btree.validate();
    // Scan the btree and validate the order of elements in each inverted-list.
    IIndexAccessor btreeAccessor = btree.createAccessor(NoOpOperationCallback.INSTANCE, NoOpOperationCallback.INSTANCE);
    IIndexCursor btreeCursor = btreeAccessor.createSearchCursor(false);
    MultiComparator btreeCmp = MultiComparator.create(btree.getComparatorFactories());
    RangePredicate rangePred = new RangePredicate(null, null, true, true, btreeCmp, btreeCmp);
    int[] fieldPermutation = new int[tokenTypeTraits.length];
    for (int i = 0; i < tokenTypeTraits.length; i++) {
        fieldPermutation[i] = i;
    }
    PermutingTupleReference tokenTuple = new PermutingTupleReference(fieldPermutation);
    IInvertedIndexAccessor invIndexAccessor = (IInvertedIndexAccessor) createAccessor(NoOpOperationCallback.INSTANCE, NoOpOperationCallback.INSTANCE);
    IInvertedListCursor invListCursor = invIndexAccessor.createInvertedListCursor();
    MultiComparator invListCmp = MultiComparator.create(invListCmpFactories);
    try {
        // Search key for finding an inverted-list in the actual index.
        ArrayTupleBuilder prevBuilder = new ArrayTupleBuilder(invListTypeTraits.length);
        ArrayTupleReference prevTuple = new ArrayTupleReference();
        btreeAccessor.search(btreeCursor, rangePred);
        while (btreeCursor.hasNext()) {
            btreeCursor.next();
            tokenTuple.reset(btreeCursor.getTuple());
            // Validate inverted list by checking that the elements are totally ordered.
            invIndexAccessor.openInvertedListCursor(invListCursor, tokenTuple);
            invListCursor.pinPages();
            try {
                if (invListCursor.hasNext()) {
                    invListCursor.next();
                    ITupleReference invListElement = invListCursor.getTuple();
                    // Initialize prev tuple.
                    TupleUtils.copyTuple(prevBuilder, invListElement, invListElement.getFieldCount());
                    prevTuple.reset(prevBuilder.getFieldEndOffsets(), prevBuilder.getByteArray());
                }
                while (invListCursor.hasNext()) {
                    invListCursor.next();
                    ITupleReference invListElement = invListCursor.getTuple();
                    // Compare with previous element.
                    if (invListCmp.compare(invListElement, prevTuple) <= 0) {
                        throw new HyracksDataException("Index validation failed.");
                    }
                    // Set new prevTuple.
                    TupleUtils.copyTuple(prevBuilder, invListElement, invListElement.getFieldCount());
                    prevTuple.reset(prevBuilder.getFieldEndOffsets(), prevBuilder.getByteArray());
                }
            } finally {
                invListCursor.unpinPages();
            }
        }
    } finally {
        btreeCursor.close();
    }
}

Example 60 with ITupleReference

use of org.apache.hyracks.dataflow.common.data.accessors.ITupleReference in project asterixdb by apache.

the class PartitionedOnDiskInvertedIndex method openInvertedListPartitionCursors.

@Override
public boolean openInvertedListPartitionCursors(IInvertedIndexSearcher searcher, IIndexOperationContext ictx, short numTokensLowerBound, short numTokensUpperBound, InvertedListPartitions invListPartitions, List<IInvertedListCursor> cursorsOrderedByTokens) throws HyracksDataException {
    PartitionedTOccurrenceSearcher partSearcher = (PartitionedTOccurrenceSearcher) searcher;
    OnDiskInvertedIndexOpContext ctx = (OnDiskInvertedIndexOpContext) ictx;
    ITupleReference lowSearchKey = null;
    ITupleReference highSearchKey = null;
    partSearcher.setNumTokensBoundsInSearchKeys(numTokensLowerBound, numTokensUpperBound);
    if (numTokensLowerBound < 0) {
        ctx.getBtreePred().setLowKeyComparator(ctx.getPrefixSearchCmp());
        lowSearchKey = partSearcher.getPrefixSearchKey();
    } else {
        ctx.getBtreePred().setLowKeyComparator(ctx.getSearchCmp());
        lowSearchKey = partSearcher.getFullLowSearchKey();
    }
    if (numTokensUpperBound < 0) {
        ctx.getBtreePred().setHighKeyComparator(ctx.getPrefixSearchCmp());
        highSearchKey = partSearcher.getPrefixSearchKey();
    } else {
        ctx.getBtreePred().setHighKeyComparator(ctx.getSearchCmp());
        highSearchKey = partSearcher.getFullHighSearchKey();
    }
    ctx.getBtreePred().setLowKey(lowSearchKey, true);
    ctx.getBtreePred().setHighKey(highSearchKey, true);
    ctx.getBtreeAccessor().search(ctx.getBtreeCursor(), ctx.getBtreePred());
    boolean tokenExists = false;
    try {
        while (ctx.getBtreeCursor().hasNext()) {
            ctx.getBtreeCursor().next();
            ITupleReference btreeTuple = ctx.getBtreeCursor().getTuple();
            short numTokens = ShortPointable.getShort(btreeTuple.getFieldData(PARTITIONING_NUM_TOKENS_FIELD), btreeTuple.getFieldStart(PARTITIONING_NUM_TOKENS_FIELD));
            IInvertedListCursor invListCursor = partSearcher.getCachedInvertedListCursor();
            resetInvertedListCursor(btreeTuple, invListCursor);
            cursorsOrderedByTokens.add(invListCursor);
            invListPartitions.addInvertedListCursor(invListCursor, numTokens);
            tokenExists = true;
        }
    } finally {
        ctx.getBtreeCursor().close();
        ctx.getBtreeCursor().reset();
    }
    return tokenExists;
}