use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.
the class TestResultSetReader method testBasics.
@Test
public void testBasics() {
PullResultSetReader rsReader = new PullResultSetReaderImpl(new BatchGenerator(10, 2, 1));
// Start state
try {
rsReader.reader();
fail();
} catch (IllegalStateException e) {
// Expected
}
// Ask for schema. Does an implicit next.
assertEquals(SCHEMA1, rsReader.schema());
assertEquals(1, rsReader.schemaVersion());
// Move to the first batch.
// (Don't need to do a full reader test, that is already done
// elsewhere.)
assertTrue(rsReader.next());
assertEquals(1, rsReader.schemaVersion());
RowSetReader reader1;
{
RowSetReader reader = rsReader.reader();
reader1 = reader;
assertTrue(reader.next());
assertEquals(1, reader.scalar("id").getInt());
assertEquals("Row1", reader.scalar("name").getString());
}
// Second batch, same schema.
assertTrue(rsReader.next());
assertEquals(1, rsReader.schemaVersion());
{
RowSetReader reader = rsReader.reader();
assertSame(reader1, reader);
reader1 = reader;
assertTrue(reader.next());
assertEquals(11, reader.scalar("id").getInt());
assertEquals("Row11", reader.scalar("name").getString());
}
// Batch with new schema
assertTrue(rsReader.next());
assertEquals(2, rsReader.schemaVersion());
{
RowSetReader reader = rsReader.reader();
assertNotSame(reader1, reader);
reader1 = reader;
assertTrue(reader.next());
assertEquals(21, reader.scalar("id").getInt());
assertEquals("Row21", reader.scalar("name").getString());
assertEquals(210, reader.scalar("amount").getInt());
}
assertFalse(rsReader.next());
rsReader.close();
}
use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.
the class TestResultSetLoaderMaps method testMapOverflowWithNewColumn.
/**
* Test the case in which a new column is added during the overflow row. Unlike
* the top-level schema case, internally we must create a copy of the map, and
* move vectors across only when the result is to include the schema version
* of the target column. For overflow, the new column is added after the
* first batch; it is added in the second batch that contains the overflow
* row in which the column was added.
*/
@Test
public void testMapOverflowWithNewColumn() {
final TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).addMap("m").add("b", MinorType.INT).add("c", MinorType.VARCHAR).resumeSchema().buildSchema();
final ResultSetLoaderImpl.ResultSetOptions options = new ResultSetOptionBuilder().readerSchema(schema).rowCountLimit(ValueVector.MAX_ROW_COUNT).build();
final ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
assertEquals(4, rsLoader.schemaVersion());
final RowSetLoader rootWriter = rsLoader.writer();
// Can't use the shortcut to populate rows when doing a schema
// change.
final ScalarWriter aWriter = rootWriter.scalar("a");
final TupleWriter mWriter = rootWriter.tuple("m");
final ScalarWriter bWriter = mWriter.scalar("b");
final ScalarWriter cWriter = mWriter.scalar("c");
final byte[] value = new byte[512];
Arrays.fill(value, (byte) 'X');
int count = 0;
rsLoader.startBatch();
while (!rootWriter.isFull()) {
rootWriter.start();
aWriter.setInt(count);
bWriter.setInt(count * 10);
cWriter.setBytes(value, value.length);
if (rootWriter.isFull()) {
// Overflow just occurred. Add another column.
mWriter.addColumn(SchemaBuilder.columnSchema("d", MinorType.INT, DataMode.OPTIONAL));
mWriter.scalar("d").setInt(count * 100);
}
rootWriter.save();
count++;
}
// Result set should include the original columns, but not d.
RowSet result = fixture.wrap(rsLoader.harvest());
assertEquals(4, rsLoader.schemaVersion());
assertTrue(schema.isEquivalent(result.schema()));
final BatchSchema expectedSchema = new BatchSchema(SelectionVectorMode.NONE, schema.toFieldList());
assertTrue(expectedSchema.isEquivalent(result.batchSchema()));
// Use a reader to validate row-by-row. Too large to create an expected
// result set.
RowSetReader reader = result.reader();
TupleReader mapReader = reader.tuple("m");
int rowId = 0;
while (reader.next()) {
assertEquals(rowId, reader.scalar("a").getInt());
assertEquals(rowId * 10, mapReader.scalar("b").getInt());
assertTrue(Arrays.equals(value, mapReader.scalar("c").getBytes()));
rowId++;
}
result.clear();
// Next batch should start with the overflow row
rsLoader.startBatch();
assertEquals(1, rootWriter.rowCount());
result = fixture.wrap(rsLoader.harvest());
assertEquals(1, result.rowCount());
reader = result.reader();
mapReader = reader.tuple("m");
while (reader.next()) {
assertEquals(rowId, reader.scalar("a").getInt());
assertEquals(rowId * 10, mapReader.scalar("b").getInt());
assertTrue(Arrays.equals(value, mapReader.scalar("c").getBytes()));
assertEquals(rowId * 100, mapReader.scalar("d").getInt());
}
result.clear();
rsLoader.close();
}
use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.
the class TestResultSetLoaderOverflow method testMissingArrayValues.
/**
* Test the case that an array has "missing values" before the overflow.
*/
@Test
public void testMissingArrayValues() {
TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).add("b", MinorType.VARCHAR).addArray("c", MinorType.INT).buildSchema();
ResultSetOptions options = new ResultSetOptionBuilder().rowCountLimit(ValueVector.MAX_ROW_COUNT).readerSchema(schema).build();
ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
RowSetLoader rootWriter = rsLoader.writer();
byte[] value = new byte[512];
Arrays.fill(value, (byte) 'X');
int blankAfter = ValueVector.MAX_BUFFER_SIZE / 512 * 2 / 3;
ScalarWriter cWriter = rootWriter.array("c").scalar();
rsLoader.startBatch();
int rowId = 0;
while (rootWriter.start()) {
rootWriter.scalar("a").setInt(rowId);
rootWriter.scalar("b").setBytes(value, value.length);
if (rowId < blankAfter) {
for (int i = 0; i < 3; i++) {
cWriter.setInt(rowId * 3 + i);
}
}
rootWriter.save();
rowId++;
}
VectorContainer container = rsLoader.harvest();
BatchValidator.validate(container);
RowSet result = fixture.wrap(container);
assertEquals(rowId - 1, result.rowCount());
RowSetReader reader = result.reader();
ArrayReader cArray = reader.array("c");
ScalarReader cReader = cArray.scalar();
while (reader.next()) {
assertEquals(reader.offset(), reader.scalar("a").getInt());
assertArrayEquals(value, reader.scalar("b").getBytes());
if (reader.offset() < blankAfter) {
assertEquals(3, cArray.size());
for (int i = 0; i < 3; i++) {
assertTrue(cArray.next());
assertEquals(reader.offset() * 3 + i, cReader.getInt());
}
} else {
assertEquals(0, cArray.size());
}
}
result.clear();
rsLoader.close();
}
use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.
the class TestResultSetLoaderOverflow method testOverflowWithNullables.
@Test
public void testOverflowWithNullables() {
TupleMetadata schema = new SchemaBuilder().add("n", MinorType.INT).addNullable("a", MinorType.VARCHAR).addNullable("b", MinorType.VARCHAR).addNullable("c", MinorType.VARCHAR).buildSchema();
ResultSetOptions options = new ResultSetOptionBuilder().rowCountLimit(ValueVector.MAX_ROW_COUNT).readerSchema(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).setInt(count);
rootWriter.scalar(1).setNull();
rootWriter.scalar(2).setBytes(value, value.length);
rootWriter.scalar(3).setNull();
rootWriter.save();
count++;
}
// Result should exclude the overflow row
{
VectorContainer container = rsLoader.harvest();
BatchValidator.validate(container);
RowSet result = fixture.wrap(container);
assertEquals(count - 1, result.rowCount());
RowSetReader reader = result.reader();
while (reader.next()) {
assertEquals(reader.offset(), reader.scalar(0).getInt());
assertTrue(reader.scalar(1).isNull());
assertArrayEquals(value, reader.scalar(2).getBytes());
assertTrue(reader.scalar(3).isNull());
}
result.clear();
}
// Next batch should start with the overflow row
rsLoader.startBatch();
{
VectorContainer container = rsLoader.harvest();
BatchValidator.validate(container);
RowSet result = fixture.wrap(container);
RowSetReader reader = result.reader();
assertEquals(1, result.rowCount());
assertTrue(reader.next());
assertEquals(count - 1, reader.scalar(0).getInt());
assertTrue(reader.scalar(1).isNull());
assertArrayEquals(value, reader.scalar(2).getBytes());
assertTrue(reader.scalar(3).isNull());
result.clear();
}
rsLoader.close();
}
use of org.apache.drill.exec.physical.rowSet.RowSetReader in project drill by apache.
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 ResultSetOptionBuilder().rowCountLimit(ValueVector.MAX_ROW_COUNT).readerSchema(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 valuesPerArray = 13;
int count = 0;
{
int rowSize = 0;
int totalSize = 0;
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.
VectorContainer container = rsLoader.harvest();
BatchValidator.validate(container);
RowSet result = fixture.wrap(container);
assertEquals(expectedCount, result.rowCount());
RowSetReader reader = result.reader();
reader.setPosition(expectedCount - 1);
ArrayReader arrayReader = reader.array(0);
ScalarReader strReader = arrayReader.scalar();
assertEquals(valuesPerArray, arrayReader.size());
for (int i = 0; i < valuesPerArray; i++) {
assertTrue(arrayReader.next());
String cellValue = strValue + (count - 1) + "." + i;
assertEquals(cellValue, strReader.getString());
}
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();
assertEquals(1, rootWriter.rowCount());
assertEquals(count, rsLoader.totalRowCount());
VectorContainer container = rsLoader.harvest();
BatchValidator.validate(container);
RowSet result = fixture.wrap(container);
assertEquals(1, result.rowCount());
RowSetReader reader = result.reader();
reader.next();
ArrayReader arrayReader = reader.array(0);
ScalarReader strReader = arrayReader.scalar();
assertEquals(valuesPerArray, arrayReader.size());
for (int i = 0; i < valuesPerArray; i++) {
assertTrue(arrayReader.next());
String cellValue = strValue + count + "." + i;
assertEquals(cellValue, strReader.getString());
}
result.clear();
}
rsLoader.close();
}
Aggregations