Search in sources :

Example 6 with ResultSetLoader

use of org.apache.drill.exec.physical.rowSet.ResultSetLoader in project drill by axbaretto.

the class TestResultSetLoaderMaps method testEmptyMapAddition.

/**
 * Test adding an empty map to a loader after writing the first row.
 * Then add columns in another batch. Yes, this is a bizarre condition,
 * but we must check it anyway for robustness.
 */
@Test
public void testEmptyMapAddition() {
    TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).buildSchema();
    ResultSetLoaderImpl.ResultSetOptions options = new OptionBuilder().setSchema(schema).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    assertEquals(1, rsLoader.schemaVersion());
    RowSetLoader rootWriter = rsLoader.writer();
    // Start without the map. Add a map after the first row.
    rsLoader.startBatch();
    rootWriter.addRow(10);
    int mapIndex = rootWriter.addColumn(SchemaBuilder.columnSchema("m", MinorType.MAP, DataMode.REQUIRED));
    TupleWriter mapWriter = rootWriter.tuple(mapIndex);
    rootWriter.addRow(20, objArray()).addRow(30, objArray());
    RowSet actual = fixture.wrap(rsLoader.harvest());
    assertEquals(2, rsLoader.schemaVersion());
    assertEquals(3, actual.rowCount());
    // Validate first batch
    TupleMetadata expectedSchema = new SchemaBuilder().add("a", MinorType.INT).addMap("m").resumeSchema().buildSchema();
    SingleRowSet expected = fixture.rowSetBuilder(expectedSchema).addRow(10, objArray()).addRow(20, objArray()).addRow(30, objArray()).build();
    new RowSetComparison(expected).verifyAndClearAll(actual);
    // Now add another column to the map
    rsLoader.startBatch();
    mapWriter.addColumn(SchemaBuilder.columnSchema("a", MinorType.VARCHAR, DataMode.REQUIRED));
    rootWriter.addRow(40, objArray("fred")).addRow(50, objArray("barney"));
    actual = fixture.wrap(rsLoader.harvest());
    assertEquals(3, rsLoader.schemaVersion());
    assertEquals(2, actual.rowCount());
    // Validate first batch
    expectedSchema = new SchemaBuilder().add("a", MinorType.INT).addMap("m").add("a", MinorType.VARCHAR).resumeSchema().buildSchema();
    expected = fixture.rowSetBuilder(expectedSchema).addRow(40, objArray("fred")).addRow(50, objArray("barney")).build();
    new RowSetComparison(expected).verifyAndClearAll(actual);
    rsLoader.close();
}
Also used : SingleRowSet(org.apache.drill.test.rowSet.RowSet.SingleRowSet) RowSetComparison(org.apache.drill.test.rowSet.RowSetComparison) ResultSetLoader(org.apache.drill.exec.physical.rowSet.ResultSetLoader) TupleWriter(org.apache.drill.exec.vector.accessor.TupleWriter) TupleMetadata(org.apache.drill.exec.record.metadata.TupleMetadata) SchemaBuilder(org.apache.drill.test.rowSet.schema.SchemaBuilder) SingleRowSet(org.apache.drill.test.rowSet.RowSet.SingleRowSet) RowSet(org.apache.drill.test.rowSet.RowSet) RowSetLoader(org.apache.drill.exec.physical.rowSet.RowSetLoader) SubOperatorTest(org.apache.drill.test.SubOperatorTest) Test(org.junit.Test)

Example 7 with ResultSetLoader

use of org.apache.drill.exec.physical.rowSet.ResultSetLoader in project drill by axbaretto.

the class TestResultSetLoaderMaps method testOverwriteRow.

/**
 * Version of the {#link TestResultSetLoaderProtocol#testOverwriteRow()} test
 * that uses nested columns.
 */
@Test
public void testOverwriteRow() {
    TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).addMap("m").add("b", MinorType.INT).add("c", MinorType.VARCHAR).resumeSchema().buildSchema();
    ResultSetLoaderImpl.ResultSetOptions options = new OptionBuilder().setSchema(schema).setRowCountLimit(ValueVector.MAX_ROW_COUNT).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    RowSetLoader rootWriter = rsLoader.writer();
    // Can't use the shortcut to populate rows when doing overwrites.
    ScalarWriter aWriter = rootWriter.scalar("a");
    TupleWriter mWriter = rootWriter.tuple("m");
    ScalarWriter bWriter = mWriter.scalar("b");
    ScalarWriter cWriter = mWriter.scalar("c");
    // Write 100,000 rows, overwriting 99% of them. This will cause vector
    // overflow and data corruption if overwrite does not work; but will happily
    // produce the correct result if everything works as it should.
    byte[] value = new byte[512];
    Arrays.fill(value, (byte) 'X');
    int count = 0;
    rsLoader.startBatch();
    while (count < 100_000) {
        rootWriter.start();
        count++;
        aWriter.setInt(count);
        bWriter.setInt(count * 10);
        cWriter.setBytes(value, value.length);
        if (count % 100 == 0) {
            rootWriter.save();
        }
    }
    // Verify using a reader.
    RowSet result = fixture.wrap(rsLoader.harvest());
    assertEquals(count / 100, result.rowCount());
    RowSetReader reader = result.reader();
    TupleReader mReader = reader.tuple("m");
    int rowId = 1;
    while (reader.next()) {
        assertEquals(rowId * 100, reader.scalar("a").getInt());
        assertEquals(rowId * 1000, mReader.scalar("b").getInt());
        assertTrue(Arrays.equals(value, mReader.scalar("c").getBytes()));
        rowId++;
    }
    result.clear();
    rsLoader.close();
}
Also used : TupleReader(org.apache.drill.exec.vector.accessor.TupleReader) SingleRowSet(org.apache.drill.test.rowSet.RowSet.SingleRowSet) RowSet(org.apache.drill.test.rowSet.RowSet) ResultSetLoader(org.apache.drill.exec.physical.rowSet.ResultSetLoader) TupleWriter(org.apache.drill.exec.vector.accessor.TupleWriter) TupleMetadata(org.apache.drill.exec.record.metadata.TupleMetadata) SchemaBuilder(org.apache.drill.test.rowSet.schema.SchemaBuilder) RowSetLoader(org.apache.drill.exec.physical.rowSet.RowSetLoader) RowSetReader(org.apache.drill.test.rowSet.RowSetReader) ScalarWriter(org.apache.drill.exec.vector.accessor.ScalarWriter) SubOperatorTest(org.apache.drill.test.SubOperatorTest) Test(org.junit.Test)

Example 8 with ResultSetLoader

use of org.apache.drill.exec.physical.rowSet.ResultSetLoader in project drill by axbaretto.

the class TestResultSetLoaderOverflow method testVectorSizeLimit.

/**
 * Test that the writer detects a vector overflow. The offending column
 * value should be moved to the next batch.
 */
@Test
public void testVectorSizeLimit() {
    TupleMetadata schema = new SchemaBuilder().add("s", MinorType.VARCHAR).buildSchema();
    ResultSetOptions options = new OptionBuilder().setRowCountLimit(ValueVector.MAX_ROW_COUNT).setSchema(schema).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    RowSetLoader rootWriter = rsLoader.writer();
    rsLoader.startBatch();
    byte[] value = new byte[512];
    Arrays.fill(value, (byte) 'X');
    int count = 0;
    while (!rootWriter.isFull()) {
        rootWriter.start();
        rootWriter.scalar(0).setBytes(value, value.length);
        rootWriter.save();
        count++;
    }
    // Number of rows should be driven by vector size.
    // Our row count should include the overflow row
    int expectedCount = ValueVector.MAX_BUFFER_SIZE / value.length;
    assertEquals(expectedCount + 1, count);
    // Loader's row count should include only "visible" rows
    assertEquals(expectedCount, rootWriter.rowCount());
    // Total count should include invisible and look-ahead rows.
    assertEquals(expectedCount + 1, rsLoader.totalRowCount());
    // Result should exclude the overflow row
    RowSet result = fixture.wrap(rsLoader.harvest());
    assertEquals(expectedCount, result.rowCount());
    result.clear();
    // Next batch should start with the overflow row
    rsLoader.startBatch();
    assertEquals(1, rootWriter.rowCount());
    assertEquals(expectedCount + 1, rsLoader.totalRowCount());
    result = fixture.wrap(rsLoader.harvest());
    assertEquals(1, result.rowCount());
    result.clear();
    rsLoader.close();
}
Also used : ResultSetLoader(org.apache.drill.exec.physical.rowSet.ResultSetLoader) TupleMetadata(org.apache.drill.exec.record.metadata.TupleMetadata) SchemaBuilder(org.apache.drill.test.rowSet.schema.SchemaBuilder) RowSet(org.apache.drill.test.rowSet.RowSet) RowSetLoader(org.apache.drill.exec.physical.rowSet.RowSetLoader) ResultSetOptions(org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions) SubOperatorTest(org.apache.drill.test.SubOperatorTest) Test(org.junit.Test)

Example 9 with ResultSetLoader

use of org.apache.drill.exec.physical.rowSet.ResultSetLoader in project drill by axbaretto.

the class TestResultSetLoaderOverflow method testOversizeArray.

/**
 * Case where a single array fills up the vector to the maximum size
 * limit. Overflow won't work here; the attempt will fail with a user
 * exception.
 */
@Test
public void testOversizeArray() {
    TupleMetadata schema = new SchemaBuilder().addArray("s", MinorType.VARCHAR).buildSchema();
    ResultSetOptions options = new OptionBuilder().setRowCountLimit(ValueVector.MAX_ROW_COUNT).setSchema(schema).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    RowSetLoader rootWriter = rsLoader.writer();
    // Create a single array as the column value in the first row. When
    // this overflows, an exception is thrown since overflow is not possible.
    rsLoader.startBatch();
    byte[] value = new byte[473];
    Arrays.fill(value, (byte) 'X');
    rootWriter.start();
    ScalarWriter array = rootWriter.array(0).scalar();
    try {
        for (int i = 0; i < ValueVector.MAX_ROW_COUNT; i++) {
            array.setBytes(value, value.length);
        }
        fail();
    } catch (UserException e) {
        assertTrue(e.getMessage().contains("column value is larger than the maximum"));
    }
    rsLoader.close();
}
Also used : ResultSetLoader(org.apache.drill.exec.physical.rowSet.ResultSetLoader) TupleMetadata(org.apache.drill.exec.record.metadata.TupleMetadata) SchemaBuilder(org.apache.drill.test.rowSet.schema.SchemaBuilder) UserException(org.apache.drill.common.exceptions.UserException) RowSetLoader(org.apache.drill.exec.physical.rowSet.RowSetLoader) ScalarWriter(org.apache.drill.exec.vector.accessor.ScalarWriter) ResultSetOptions(org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions) SubOperatorTest(org.apache.drill.test.SubOperatorTest) Test(org.junit.Test)

Example 10 with ResultSetLoader

use of org.apache.drill.exec.physical.rowSet.ResultSetLoader in project drill by axbaretto.

the class TestResultSetLoaderOverflow method testSizeLimitOnArray.

/**
 * Test a row with a single array column which overflows. Verifies
 * that all the fiddly bits about offset vectors and so on works
 * correctly. Run this test (the simplest case) if you change anything
 * about the array handling code.
 */
@Test
public void testSizeLimitOnArray() {
    TupleMetadata schema = new SchemaBuilder().addArray("s", MinorType.VARCHAR).buildSchema();
    ResultSetOptions options = new OptionBuilder().setRowCountLimit(ValueVector.MAX_ROW_COUNT).setSchema(schema).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    RowSetLoader rootWriter = rsLoader.writer();
    // Fill batch with rows of with a single array, three values each. Tack on
    // a suffix to each so we can be sure the proper data is written and moved
    // to the overflow batch.
    rsLoader.startBatch();
    byte[] value = new byte[473];
    Arrays.fill(value, (byte) 'X');
    String strValue = new String(value, Charsets.UTF_8);
    int count = 0;
    int rowSize = 0;
    int totalSize = 0;
    int valuesPerArray = 13;
    while (rootWriter.start()) {
        totalSize += rowSize;
        rowSize = 0;
        ScalarWriter array = rootWriter.array(0).scalar();
        for (int i = 0; i < valuesPerArray; i++) {
            String cellValue = strValue + (count + 1) + "." + i;
            array.setString(cellValue);
            rowSize += cellValue.length();
        }
        rootWriter.save();
        count++;
    }
    // Row count should include the overflow row.
    int expectedCount = count - 1;
    // Size without overflow row should fit in the vector, size
    // with overflow should not.
    assertTrue(totalSize <= ValueVector.MAX_BUFFER_SIZE);
    assertTrue(totalSize + rowSize > ValueVector.MAX_BUFFER_SIZE);
    // Result should exclude the overflow row. Last row
    // should hold the last full array.
    RowSet result = fixture.wrap(rsLoader.harvest());
    assertEquals(expectedCount, result.rowCount());
    RowSetReader reader = result.reader();
    reader.set(expectedCount - 1);
    ScalarElementReader arrayReader = reader.column(0).elements();
    assertEquals(valuesPerArray, arrayReader.size());
    for (int i = 0; i < valuesPerArray; i++) {
        String cellValue = strValue + (count - 1) + "." + i;
        assertEquals(cellValue, arrayReader.getString(i));
    }
    result.clear();
    // Next batch should start with the overflow row.
    // The only row in this next batch should be the whole
    // array being written at the time of overflow.
    rsLoader.startBatch();
    // VectorPrinter.printStrings((VarCharVector) ((VarCharColumnWriter) rootWriter.array(0).scalar()).vector(), 0, 5);
    // ((ResultSetLoaderImpl) rsLoader).dump(new HierarchicalPrinter());
    assertEquals(1, rootWriter.rowCount());
    assertEquals(expectedCount + 1, rsLoader.totalRowCount());
    result = fixture.wrap(rsLoader.harvest());
    // VectorPrinter.printStrings((VarCharVector) ((VarCharColumnWriter) rootWriter.array(0).scalar()).vector(), 0, 5);
    assertEquals(1, result.rowCount());
    reader = result.reader();
    reader.next();
    arrayReader = reader.column(0).elements();
    assertEquals(valuesPerArray, arrayReader.size());
    for (int i = 0; i < valuesPerArray; i++) {
        String cellValue = strValue + (count) + "." + i;
        assertEquals(cellValue, arrayReader.getString(i));
    }
    result.clear();
    rsLoader.close();
}
Also used : ScalarElementReader(org.apache.drill.exec.vector.accessor.ScalarElementReader) RowSet(org.apache.drill.test.rowSet.RowSet) ResultSetOptions(org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions) ResultSetLoader(org.apache.drill.exec.physical.rowSet.ResultSetLoader) TupleMetadata(org.apache.drill.exec.record.metadata.TupleMetadata) SchemaBuilder(org.apache.drill.test.rowSet.schema.SchemaBuilder) RowSetLoader(org.apache.drill.exec.physical.rowSet.RowSetLoader) RowSetReader(org.apache.drill.test.rowSet.RowSetReader) ScalarWriter(org.apache.drill.exec.vector.accessor.ScalarWriter) SubOperatorTest(org.apache.drill.test.SubOperatorTest) Test(org.junit.Test)

Aggregations

ResultSetLoader (org.apache.drill.exec.physical.rowSet.ResultSetLoader)45 RowSetLoader (org.apache.drill.exec.physical.rowSet.RowSetLoader)44 SubOperatorTest (org.apache.drill.test.SubOperatorTest)44 Test (org.junit.Test)44 TupleMetadata (org.apache.drill.exec.record.metadata.TupleMetadata)38 SchemaBuilder (org.apache.drill.test.rowSet.schema.SchemaBuilder)38 RowSet (org.apache.drill.test.rowSet.RowSet)34 SingleRowSet (org.apache.drill.test.rowSet.RowSet.SingleRowSet)28 RowSetComparison (org.apache.drill.test.rowSet.RowSetComparison)17 ResultSetOptions (org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions)16 TupleWriter (org.apache.drill.exec.vector.accessor.TupleWriter)14 ScalarWriter (org.apache.drill.exec.vector.accessor.ScalarWriter)13 RowSetReader (org.apache.drill.test.rowSet.RowSetReader)12 BatchSchema (org.apache.drill.exec.record.BatchSchema)6 SchemaPath (org.apache.drill.common.expression.SchemaPath)5 TupleReader (org.apache.drill.exec.vector.accessor.TupleReader)5 ArrayWriter (org.apache.drill.exec.vector.accessor.ArrayWriter)4 ScalarElementReader (org.apache.drill.exec.vector.accessor.ScalarElementReader)4 MaterializedField (org.apache.drill.exec.record.MaterializedField)3 ArrayReader (org.apache.drill.exec.vector.accessor.ArrayReader)3