Examples with ResultSetOptions org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions used on opensource projects

Example 1 with ResultSetOptions

use of org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions in project drill by axbaretto.

the class TestResultSetLoaderLimits method testDynamicLimit.

/**
 * Test that the row limit can change between batches.
 */
@Test
public void testDynamicLimit() {
    // Start with a small limit.
    ResultSetOptions options = new OptionBuilder().setRowCountLimit(TEST_ROW_LIMIT).build();
    ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
    assertEquals(TEST_ROW_LIMIT, rsLoader.targetRowCount());
    RowSetLoader rootWriter = rsLoader.writer();
    rootWriter.addColumn(SchemaBuilder.columnSchema("s", MinorType.VARCHAR, DataMode.REQUIRED));
    rsLoader.startBatch();
    int count = fillToLimit(rootWriter);
    assertEquals(TEST_ROW_LIMIT, count);
    assertEquals(count, rootWriter.rowCount());
    rsLoader.harvest().clear();
    // Reset the batch size larger and fill a second batch
    int newLimit = 8000;
    rsLoader.setTargetRowCount(newLimit);
    rsLoader.startBatch();
    count = fillToLimit(rootWriter);
    assertEquals(newLimit, count);
    assertEquals(count, rootWriter.rowCount());
    rsLoader.harvest().clear();
    // Put the limit back to a lower number.
    newLimit = 1000;
    rsLoader.setTargetRowCount(newLimit);
    rsLoader.startBatch();
    count = fillToLimit(rootWriter);
    assertEquals(newLimit, count);
    assertEquals(count, rootWriter.rowCount());
    rsLoader.harvest().clear();
    rsLoader.close();
}

Example 2 with ResultSetOptions

use of org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions 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();
}

Example 3 with ResultSetOptions

use of org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions 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();
}

Example 4 with ResultSetOptions

use of org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions 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();
}

Example 5 with ResultSetOptions

use of org.apache.drill.exec.physical.rowSet.impl.ResultSetLoaderImpl.ResultSetOptions in project drill by axbaretto.

the class TestResultSetLoaderOverflow method testCloseWithOverflow.

/**
 * Load a batch to overflow. Then, close the loader with the overflow
 * batch unharvested. The Loader should release the memory allocated
 * to the unused overflow vectors.
 */
@Test
public void testCloseWithOverflow() {
    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++;
    }
    assertTrue(count < ValueVector.MAX_ROW_COUNT);
    // Harvest the full batch
    RowSet result = fixture.wrap(rsLoader.harvest());
    result.clear();
    // Close without harvesting the overflow batch.
    rsLoader.close();
}