Search in sources :

Example 21 with RepeatedValueVector

use of org.apache.drill.exec.vector.complex.RepeatedValueVector in project drill by apache.

the class TestRecordBatchSizer method testEmptyBatchRepeatedFixedWidth.

@Test
public void testEmptyBatchRepeatedFixedWidth() {
    TupleMetadata schema = new SchemaBuilder().addArray("a", MinorType.BIGINT).addArray("b", MinorType.FLOAT8).buildSchema();
    RowSetBuilder builder = fixture.rowSetBuilder(schema);
    RowSet rows = builder.build();
    // Run the record batch sizer on the resulting batch.
    RecordBatchSizer sizer = new RecordBatchSizer(rows.container());
    assertEquals(2, sizer.columns().size());
    /**
     * stdDataSize:8*5, stdNetSize:8*5+4, dataSizePerEntry:0, netSizePerEntry:0,
     * totalDataSize:0, totalNetSize:0, valueCount:0,
     * elementCount:0, cardinality:0, isVariableWidth:false
     */
    verifyColumnValues(sizer.columns().get("a"), 40, 44, 0, 0, 0, 0, 0, 0, 0, false);
    verifyColumnValues(sizer.columns().get("b"), 40, 44, 0, 0, 0, 0, 0, 0, 0, false);
    // Verify memory allocation is done correctly based on std size for empty batch.
    SingleRowSet empty = fixture.rowSet(schema);
    VectorAccessible accessible = empty.vectorAccessible();
    UInt4Vector offsetVector;
    ValueVector dataVector;
    for (VectorWrapper<?> vw : accessible) {
        ValueVector v = vw.getValueVector();
        RecordBatchSizer.ColumnSize colSize = sizer.getColumn(v.getField().getName());
        // Allocates to nearest power of two
        colSize.allocateVector(v, testRowCount);
        offsetVector = ((RepeatedValueVector) v).getOffsetVector();
        assertEquals((Integer.highestOneBit(testRowCount) << 1), offsetVector.getValueCapacity());
        dataVector = ((RepeatedValueVector) v).getDataVector();
        assertEquals(Integer.highestOneBit(testRowCount * STD_REPETITION_FACTOR << 1), dataVector.getValueCapacity());
        v.clear();
        // Allocates the same as value passed since it is already power of two.
        // -1 is done for adjustment needed for offset vector.
        colSize.allocateVector(v, testRowCountPowerTwo - 1);
        offsetVector = ((RepeatedValueVector) v).getOffsetVector();
        assertEquals(testRowCountPowerTwo, offsetVector.getValueCapacity());
        dataVector = ((RepeatedValueVector) v).getDataVector();
        assertEquals(Integer.highestOneBit((testRowCountPowerTwo - 1) * STD_REPETITION_FACTOR) << 1, dataVector.getValueCapacity());
        v.clear();
        // Allocate for max rows.
        colSize.allocateVector(v, ValueVector.MAX_ROW_COUNT - 1);
        offsetVector = ((RepeatedValueVector) v).getOffsetVector();
        assertEquals(ValueVector.MAX_ROW_COUNT, offsetVector.getValueCapacity());
        dataVector = ((RepeatedValueVector) v).getDataVector();
        assertEquals(Integer.highestOneBit(((ValueVector.MAX_ROW_COUNT - 1) * STD_REPETITION_FACTOR << 1)), dataVector.getValueCapacity());
        v.clear();
        // Allocate for 0 rows. should atleast do allocation for 1 row.
        colSize.allocateVector(v, 0);
        offsetVector = ((RepeatedValueVector) v).getOffsetVector();
        assertEquals(ValueVector.MIN_ROW_COUNT + 1, offsetVector.getValueCapacity());
        dataVector = ((RepeatedValueVector) v).getDataVector();
        assertEquals(ValueVector.MIN_ROW_COUNT, dataVector.getValueCapacity());
        v.clear();
    }
    empty.clear();
    rows.clear();
}
Also used : ValueVector(org.apache.drill.exec.vector.ValueVector) RepeatedValueVector(org.apache.drill.exec.vector.complex.RepeatedValueVector) RowSetBuilder(org.apache.drill.exec.physical.rowSet.RowSetBuilder) SingleRowSet(org.apache.drill.exec.physical.rowSet.RowSet.SingleRowSet) ColumnSize(org.apache.drill.exec.record.RecordBatchSizer.ColumnSize) TupleMetadata(org.apache.drill.exec.record.metadata.TupleMetadata) SchemaBuilder(org.apache.drill.exec.record.metadata.SchemaBuilder) SingleRowSet(org.apache.drill.exec.physical.rowSet.RowSet.SingleRowSet) RowSet(org.apache.drill.exec.physical.rowSet.RowSet) UInt4Vector(org.apache.drill.exec.vector.UInt4Vector) SubOperatorTest(org.apache.drill.test.SubOperatorTest) Test(org.junit.Test)

Example 22 with RepeatedValueVector

use of org.apache.drill.exec.vector.complex.RepeatedValueVector in project drill by apache.

the class TestOutputBatchSize method testSizerRepeatedList.

@Test
public void testSizerRepeatedList() throws Exception {
    List<String> inputJsonBatches = Lists.newArrayList();
    StringBuilder batchString = new StringBuilder();
    StringBuilder newString = new StringBuilder();
    newString.append("[ [1,2,3,4], [5,6,7,8] ]");
    numRows = 9;
    batchString.append("[");
    for (int i = 0; i < numRows; i++) {
        batchString.append("{\"c\" : " + newString);
        batchString.append("},");
    }
    batchString.append("{\"c\" : " + newString);
    batchString.append("}");
    batchString.append("]");
    inputJsonBatches.add(batchString.toString());
    // Create a dummy scanBatch to figure out the size.
    RecordBatch scanBatch = new ScanBatch(new MockPhysicalOperator(), fragContext, getReaderListForJsonBatches(inputJsonBatches, fragContext));
    VectorAccessible va = new BatchIterator(scanBatch).iterator().next();
    RecordBatchSizer sizer = new RecordBatchSizer(va);
    assertEquals(1, sizer.columns().size());
    RecordBatchSizer.ColumnSize column = sizer.columns().get("c");
    assertNotNull(column);
    /**
     * stdDataSize:8*5*5, stdNetSize:8*5*5 + 4*5 + 4*5 + 4,
     * dataSizePerEntry:8*8, netSizePerEntry:8*8 + 4*2 + 4,
     * totalDataSize:8*8*10, totalNetSize:netSizePerEntry*10, valueCount:10,
     * elementCount:10, estElementCountPerArray:1, isVariableWidth:false
     */
    assertEquals(200, column.getStdDataSizePerEntry());
    assertEquals(244, column.getStdNetSizePerEntry());
    assertEquals(64, column.getDataSizePerEntry());
    assertEquals(76, column.getNetSizePerEntry());
    assertEquals(640, column.getTotalDataSize());
    assertEquals(760, column.getTotalNetSize());
    assertEquals(10, column.getValueCount());
    assertEquals(20, column.getElementCount());
    assertEquals(2, column.getCardinality(), 0.01);
    assertEquals(false, column.isVariableWidth());
    final int testRowCount = 1000;
    final int testRowCountPowerTwo = 2048;
    for (VectorWrapper<?> vw : va) {
        ValueVector v = vw.getValueVector();
        v.clear();
        RecordBatchSizer.ColumnSize colSize = sizer.getColumn(v.getField().getName());
        // Allocates to nearest power of two
        colSize.allocateVector(v, testRowCount);
        // offset vector of delegate vector i.e. outer array should have row count number of values.
        UInt4Vector offsetVector = ((RepeatedListVector) v).getOffsetVector();
        assertEquals((Integer.highestOneBit(testRowCount) << 1), offsetVector.getValueCapacity());
        // Get inner vector of delegate vector.
        ValueVector vector = ((RepeatedValueVector) v).getDataVector();
        // Data vector of inner vector should
        // have 2 (outer array cardinality) * 4 (inner array cardinality) * row count number of values.
        ValueVector dataVector = ((RepeatedValueVector) vector).getDataVector();
        assertEquals(Integer.highestOneBit((testRowCount * 8) << 1), dataVector.getValueCapacity());
        // offset vector of inner vector should have
        // 2 (outer array cardinality) * row count number of values.
        offsetVector = ((RepeatedValueVector) vector).getOffsetVector();
        assertEquals((Integer.highestOneBit(testRowCount * 2) << 1), offsetVector.getValueCapacity());
        v.clear();
        // Allocates the same as value passed since it is already power of two.
        // -1 is done for adjustment needed for offset vector.
        colSize.allocateVector(v, testRowCountPowerTwo - 1);
        // offset vector of delegate vector i.e. outer array should have row count number of values.
        offsetVector = ((RepeatedListVector) v).getOffsetVector();
        assertEquals(testRowCountPowerTwo, offsetVector.getValueCapacity());
        // Get inner vector of delegate vector.
        vector = ((RepeatedValueVector) v).getDataVector();
        // Data vector of inner vector should
        // have 2 (outer array cardinality) * 4 (inner array cardinality) * row count number of values.
        dataVector = ((RepeatedValueVector) vector).getDataVector();
        assertEquals(testRowCountPowerTwo * 8, dataVector.getValueCapacity());
        // offset vector of inner vector should have
        // 2 (outer array cardinality) * row count number of values.
        offsetVector = ((RepeatedValueVector) vector).getOffsetVector();
        assertEquals(testRowCountPowerTwo * 2, offsetVector.getValueCapacity());
        v.clear();
        // MAX ROW COUNT
        colSize.allocateVector(v, ValueVector.MAX_ROW_COUNT - 1);
        // offset vector of delegate vector i.e. outer array should have row count number of values.
        offsetVector = ((RepeatedListVector) v).getOffsetVector();
        assertEquals(ValueVector.MAX_ROW_COUNT, offsetVector.getValueCapacity());
        // Get inner vector of delegate vector.
        vector = ((RepeatedValueVector) v).getDataVector();
        // Data vector of inner vector should
        // have 2 (outer array cardinality) * 4 (inner array cardinality) * row count number of values.
        dataVector = ((RepeatedValueVector) vector).getDataVector();
        assertEquals(ValueVector.MAX_ROW_COUNT * 8, dataVector.getValueCapacity());
        // offset vector of inner vector should have
        // 2 (outer array cardinality) * row count number of values.
        offsetVector = ((RepeatedValueVector) vector).getOffsetVector();
        assertEquals(ValueVector.MAX_ROW_COUNT * 2, offsetVector.getValueCapacity());
        v.clear();
        // MIN ROW COUNT
        colSize.allocateVector(v, 0);
        // offset vector of delegate vector i.e. outer array should have 1 value.
        offsetVector = ((RepeatedListVector) v).getOffsetVector();
        assertEquals(ValueVector.MIN_ROW_COUNT, offsetVector.getValueCapacity());
        // Get inner vector of delegate vector.
        vector = ((RepeatedValueVector) v).getDataVector();
        // Data vector of inner vector should have 1 value
        dataVector = ((RepeatedValueVector) vector).getDataVector();
        assertEquals(ValueVector.MIN_ROW_COUNT, dataVector.getValueCapacity());
        // offset vector of inner vector should have
        // 2 (outer array cardinality) * 1.
        offsetVector = ((RepeatedValueVector) vector).getOffsetVector();
        assertEquals(ValueVector.MIN_ROW_COUNT * 2, offsetVector.getValueCapacity());
        v.clear();
    }
}
Also used : VectorAccessible(org.apache.drill.exec.record.VectorAccessible) RecordBatch(org.apache.drill.exec.record.RecordBatch) RepeatedValueVector(org.apache.drill.exec.vector.complex.RepeatedValueVector) UInt4Vector(org.apache.drill.exec.vector.UInt4Vector) RepeatedValueVector(org.apache.drill.exec.vector.complex.RepeatedValueVector) ValueVector(org.apache.drill.exec.vector.ValueVector) RecordBatchSizer(org.apache.drill.exec.record.RecordBatchSizer) RepeatedListVector(org.apache.drill.exec.vector.complex.RepeatedListVector) ScanBatch(org.apache.drill.exec.physical.impl.ScanBatch) Test(org.junit.Test)

Example 23 with RepeatedValueVector

use of org.apache.drill.exec.vector.complex.RepeatedValueVector in project drill by apache.

the class TestRecordBatchSizer method testSizerRepeatedFixedWidth.

@Test
public void testSizerRepeatedFixedWidth() {
    TupleMetadata schema = new SchemaBuilder().addArray("a", MinorType.BIGINT).addArray("b", MinorType.FLOAT8).buildSchema();
    RowSetBuilder builder = fixture.rowSetBuilder(schema);
    for (long i = 0; i < 10; i++) {
        builder.addRow(new long[] { 1, 2, 3, 4, 5 }, new double[] { i * 0.1, i * 0.1, i * 0.1, i * 0.2, i * 0.3 });
    }
    RowSet rows = builder.build();
    // Run the record batch sizer on the resulting batch.
    RecordBatchSizer sizer = new RecordBatchSizer(rows.container());
    assertEquals(2, sizer.columns().size());
    /**
     * stdDataSize:8*5, stdNetSize:8*5+4, dataSizePerEntry:5*8, netSizePerEntry:5*8+4,
     * totalDataSize:5*8*5, totalNetSize:5*8*10+5*8, valueCount:10,
     * elementCount:50, cardinality:5, isVariableWidth:false
     */
    verifyColumnValues(sizer.columns().get("a"), 40, 44, 40, 44, 400, 440, 10, 50, 5, false);
    verifyColumnValues(sizer.columns().get("b"), 40, 44, 40, 44, 400, 440, 10, 50, 5, false);
    SingleRowSet empty = fixture.rowSet(schema);
    VectorAccessible accessible = empty.vectorAccessible();
    UInt4Vector offsetVector;
    ValueVector dataVector;
    for (VectorWrapper<?> vw : accessible) {
        ValueVector v = vw.getValueVector();
        RecordBatchSizer.ColumnSize colSize = sizer.getColumn(v.getField().getName());
        // Allocates to nearest power of two
        colSize.allocateVector(v, testRowCount);
        offsetVector = ((RepeatedValueVector) v).getOffsetVector();
        assertEquals((Integer.highestOneBit(testRowCount) << 1), offsetVector.getValueCapacity());
        dataVector = ((RepeatedValueVector) v).getDataVector();
        assertEquals(Integer.highestOneBit((testRowCount * 5) << 1), dataVector.getValueCapacity());
        v.clear();
        // Allocates the same as value passed since it is already power of two.
        // -1 is done for adjustment needed for offset vector.
        colSize.allocateVector(v, testRowCountPowerTwo - 1);
        offsetVector = ((RepeatedValueVector) v).getOffsetVector();
        assertEquals(testRowCountPowerTwo, offsetVector.getValueCapacity());
        dataVector = ((RepeatedValueVector) v).getDataVector();
        assertEquals(Integer.highestOneBit((testRowCountPowerTwo - 1) * 5) << 1, dataVector.getValueCapacity());
        v.clear();
        // Allocate for max rows.
        colSize.allocateVector(v, ValueVector.MAX_ROW_COUNT - 1);
        offsetVector = ((RepeatedValueVector) v).getOffsetVector();
        assertEquals(ValueVector.MAX_ROW_COUNT, offsetVector.getValueCapacity());
        dataVector = ((RepeatedValueVector) v).getDataVector();
        assertEquals(Integer.highestOneBit(((ValueVector.MAX_ROW_COUNT - 1) * 5) << 1), dataVector.getValueCapacity());
        v.clear();
        // Allocate for 0 rows. should atleast do allocation for 1 row.
        colSize.allocateVector(v, 0);
        offsetVector = ((RepeatedValueVector) v).getOffsetVector();
        assertEquals(ValueVector.MIN_ROW_COUNT + 1, offsetVector.getValueCapacity());
        dataVector = ((RepeatedValueVector) v).getDataVector();
        assertEquals(ValueVector.MIN_ROW_COUNT, dataVector.getValueCapacity());
        v.clear();
    }
    empty.clear();
    rows.clear();
}
Also used : ValueVector(org.apache.drill.exec.vector.ValueVector) RepeatedValueVector(org.apache.drill.exec.vector.complex.RepeatedValueVector) RowSetBuilder(org.apache.drill.exec.physical.rowSet.RowSetBuilder) SingleRowSet(org.apache.drill.exec.physical.rowSet.RowSet.SingleRowSet) ColumnSize(org.apache.drill.exec.record.RecordBatchSizer.ColumnSize) TupleMetadata(org.apache.drill.exec.record.metadata.TupleMetadata) SchemaBuilder(org.apache.drill.exec.record.metadata.SchemaBuilder) SingleRowSet(org.apache.drill.exec.physical.rowSet.RowSet.SingleRowSet) RowSet(org.apache.drill.exec.physical.rowSet.RowSet) UInt4Vector(org.apache.drill.exec.vector.UInt4Vector) SubOperatorTest(org.apache.drill.test.SubOperatorTest) Test(org.junit.Test)

Example 24 with RepeatedValueVector

use of org.apache.drill.exec.vector.complex.RepeatedValueVector in project drill by apache.

the class HyperReaderBuilder method buildDict.

private AbstractObjectReader buildDict(VectorAccessor va, ColumnMetadata metadata) {
    boolean isArray = metadata.isArray();
    ValueVector vector = va.vector();
    VectorAccessor dictAccessor;
    if (isArray) {
        ValueVector dictVector = ((RepeatedValueVector) vector).getDataVector();
        dictAccessor = new VectorAccessors.SingleVectorAccessor(dictVector);
    } else {
        dictAccessor = va;
    }
    List<AbstractObjectReader> readers = buildMapMembers(dictAccessor, metadata.tupleSchema());
    AbstractObjectReader reader = DictReaderImpl.build(metadata, dictAccessor, readers);
    if (!isArray) {
        return reader;
    }
    return ArrayReaderImpl.buildTuple(metadata, va, reader);
}
Also used : RepeatedValueVector(org.apache.drill.exec.vector.complex.RepeatedValueVector) ValueVector(org.apache.drill.exec.vector.ValueVector) AbstractObjectReader(org.apache.drill.exec.vector.accessor.reader.AbstractObjectReader) RepeatedValueVector(org.apache.drill.exec.vector.complex.RepeatedValueVector) BaseHyperVectorAccessor(org.apache.drill.exec.vector.accessor.reader.VectorAccessors.BaseHyperVectorAccessor) VectorAccessor(org.apache.drill.exec.vector.accessor.reader.VectorAccessor) VectorAccessors(org.apache.drill.exec.vector.accessor.reader.VectorAccessors)

Example 25 with RepeatedValueVector

use of org.apache.drill.exec.vector.complex.RepeatedValueVector in project drill by apache.

the class ColumnBuilder method buildPrimitive.

/**
 * Build a primitive column. Check if the column is projected. If not,
 * allocate a dummy writer for the column. If projected, then allocate
 * a vector, a writer, and the column state which binds the two together
 * and manages the column.
 *
 * @param parent schema of the new primitive column
 * @param columnSchema implied projection type for the column
 * @return column state for the new column
 */
private ColumnState buildPrimitive(ContainerState parent, ColumnMetadata columnSchema) {
    final ValueVector vector;
    if (parent.projection().projection(columnSchema).isProjected || allowCreation(parent)) {
        // Create the vector for the column.
        vector = parent.vectorCache().vectorFor(columnSchema.schema());
        // from that requested. Update the schema to match.
        if (parent.vectorCache().isPermissive() && !vector.getField().isEquivalent(columnSchema.schema())) {
            columnSchema = ((PrimitiveColumnMetadata) columnSchema).mergeWith(vector.getField());
        }
    } else {
        // Column is not projected. No materialized backing for the column.
        vector = null;
    }
    // Create the writer.
    final AbstractObjectWriter colWriter = ColumnWriterFactory.buildColumnWriter(columnSchema, vector);
    // Build the vector state which manages the vector.
    VectorState vectorState;
    if (vector == null) {
        vectorState = new NullVectorState();
    } else if (columnSchema.isArray()) {
        vectorState = new RepeatedVectorState(colWriter.array(), (RepeatedValueVector) vector);
    } else if (columnSchema.isNullable()) {
        vectorState = new NullableVectorState(colWriter, (NullableVector) vector);
    } else {
        vectorState = SimpleVectorState.vectorState(columnSchema, colWriter.events(), vector);
    }
    // Create the column state which binds the vector and writer together.
    return new PrimitiveColumnState(parent.loader(), colWriter, vectorState);
}
Also used : RepeatedValueVector(org.apache.drill.exec.vector.complex.RepeatedValueVector) ValueVector(org.apache.drill.exec.vector.ValueVector) NullableVector(org.apache.drill.exec.vector.NullableVector) PrimitiveColumnState(org.apache.drill.exec.physical.resultSet.impl.ColumnState.PrimitiveColumnState) AbstractObjectWriter(org.apache.drill.exec.vector.accessor.writer.AbstractObjectWriter) MapVectorState(org.apache.drill.exec.physical.resultSet.impl.TupleState.MapVectorState) RepeatedListVectorState(org.apache.drill.exec.physical.resultSet.impl.RepeatedListState.RepeatedListVectorState) UnionVectorState(org.apache.drill.exec.physical.resultSet.impl.UnionState.UnionVectorState) OffsetVectorState(org.apache.drill.exec.physical.resultSet.impl.SingleVectorState.OffsetVectorState) ListVectorState(org.apache.drill.exec.physical.resultSet.impl.ListState.ListVectorState) SimpleVectorState(org.apache.drill.exec.physical.resultSet.impl.SingleVectorState.SimpleVectorState)

Aggregations

RepeatedValueVector (org.apache.drill.exec.vector.complex.RepeatedValueVector)26 ValueVector (org.apache.drill.exec.vector.ValueVector)22 UInt4Vector (org.apache.drill.exec.vector.UInt4Vector)14 Test (org.junit.Test)13 ColumnSize (org.apache.drill.exec.record.RecordBatchSizer.ColumnSize)9 RepeatedMapVector (org.apache.drill.exec.vector.complex.RepeatedMapVector)9 SubOperatorTest (org.apache.drill.test.SubOperatorTest)9 SingleRowSet (org.apache.drill.exec.physical.rowSet.RowSet.SingleRowSet)7 SchemaBuilder (org.apache.drill.exec.record.metadata.SchemaBuilder)7 TupleMetadata (org.apache.drill.exec.record.metadata.TupleMetadata)7 RowSet (org.apache.drill.exec.physical.rowSet.RowSet)6 RowSetBuilder (org.apache.drill.exec.physical.rowSet.RowSetBuilder)6 ScanBatch (org.apache.drill.exec.physical.impl.ScanBatch)4 RecordBatch (org.apache.drill.exec.record.RecordBatch)4 RecordBatchSizer (org.apache.drill.exec.record.RecordBatchSizer)4 TypedFieldId (org.apache.drill.exec.record.TypedFieldId)4 VectorAccessible (org.apache.drill.exec.record.VectorAccessible)4 RepeatedListVector (org.apache.drill.exec.vector.complex.RepeatedListVector)4 MaterializedField (org.apache.drill.exec.record.MaterializedField)3 TransferPair (org.apache.drill.exec.record.TransferPair)3