use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderOmittedValues method testSkipRows.
/**
* Test that omitting the call to saveRow() effectively discards
* the row. Note that the vectors still contain values in the
* discarded position; just the various pointers are unset. If
* the batch ends before the discarded values are overwritten, the
* discarded values just exist at the end of the vector. Since vectors
* start with garbage contents, the discarded values are simply a different
* kind of garbage. But, if the client writes a new row, then the new
* row overwrites the discarded row. This works because we only change
* the tail part of a vector; never the internals.
*/
@Test
public void testSkipRows() {
TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).addNullable("b", MinorType.VARCHAR).buildSchema();
ResultSetLoaderImpl.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();
int rowNumber = 0;
for (int i = 0; i < 14; i++) {
rootWriter.start();
rowNumber++;
rootWriter.scalar(0).setInt(rowNumber);
if (i % 3 == 0) {
rootWriter.scalar(1).setNull();
} else {
rootWriter.scalar(1).setString("b-" + rowNumber);
}
if (i % 2 == 0) {
rootWriter.save();
}
}
RowSet result = fixture.wrap(rsLoader.harvest());
// result.print();
SingleRowSet expected = fixture.rowSetBuilder(result.batchSchema()).addRow(1, null).addRow(3, "b-3").addRow(5, "b-5").addRow(7, null).addRow(9, "b-9").addRow(11, "b-11").addRow(13, null).build();
// expected.print();
RowSetUtilities.verify(expected, result);
rsLoader.close();
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderOmittedValues method testOmittedValuesAtEndWithOverflow.
/**
* Test "holes" at the end of a batch when batch overflows. Completed
* batch must be finalized correctly, new batch initialized correct,
* for the missing values.
*/
@Test
public void testOmittedValuesAtEndWithOverflow() {
TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).add("b", MinorType.VARCHAR).addNullable("c", MinorType.VARCHAR).addNullable("d", MinorType.VARCHAR).buildSchema();
ResultSetLoaderImpl.ResultSetOptions options = new ResultSetOptionBuilder().rowCountLimit(ValueVector.MAX_ROW_COUNT).readerSchema(schema).build();
ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
RowSetLoader rootWriter = rsLoader.writer();
// Fill the batch. Column d has some values. Column c is worst case: no values.
rsLoader.startBatch();
byte[] value = new byte[533];
Arrays.fill(value, (byte) 'X');
int rowNumber = 0;
while (!rootWriter.isFull()) {
rootWriter.start();
rowNumber++;
rootWriter.scalar(0).setInt(rowNumber);
rootWriter.scalar(1).setBytes(value, value.length);
if (rowNumber < 10_000) {
rootWriter.scalar(3).setString("d-" + rowNumber);
}
rootWriter.save();
assertEquals(rowNumber, rsLoader.totalRowCount());
}
// Harvest and verify
RowSet result = fixture.wrap(rsLoader.harvest());
assertEquals(rowNumber - 1, result.rowCount());
RowSetReader reader = result.reader();
int rowIndex = 0;
while (reader.next()) {
int expectedRowNumber = 1 + rowIndex;
assertEquals(expectedRowNumber, reader.scalar(0).getInt());
assertTrue(reader.scalar(2).isNull());
if (expectedRowNumber < 10_000) {
assertEquals("d-" + expectedRowNumber, reader.scalar(3).getString());
} else {
assertTrue(reader.scalar(3).isNull());
}
rowIndex++;
}
// Start count for this batch is one less than current
// count, because of the overflow row.
int startRowNumber = rowNumber;
// Write a few more rows to the next batch
rsLoader.startBatch();
for (int i = 0; i < 10; i++) {
rootWriter.start();
rowNumber++;
rootWriter.scalar(0).setInt(rowNumber);
rootWriter.scalar(1).setBytes(value, value.length);
if (i > 5) {
rootWriter.scalar(3).setString("d-" + rowNumber);
}
rootWriter.save();
assertEquals(rowNumber, rsLoader.totalRowCount());
}
// Verify that holes were preserved.
result = fixture.wrap(rsLoader.harvest());
assertEquals(rowNumber, rsLoader.totalRowCount());
assertEquals(rowNumber - startRowNumber + 1, result.rowCount());
// result.print();
reader = result.reader();
rowIndex = 0;
while (reader.next()) {
int expectedRowNumber = startRowNumber + rowIndex;
assertEquals(expectedRowNumber, reader.scalar(0).getInt());
assertTrue(reader.scalar(2).isNull());
if (rowIndex > 6) {
assertEquals("d-" + expectedRowNumber, reader.scalar(3).getString());
} else {
assertTrue("Row " + rowIndex + " col d should be null", reader.scalar(3).isNull());
}
rowIndex++;
}
assertEquals(rowIndex, 11);
rsLoader.close();
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderOmittedValues method testSkipOverflowRow.
/**
* Test that discarding a row works even if that row happens to be an
* overflow row.
*/
@Test
public void testSkipOverflowRow() {
TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).addNullable("b", MinorType.VARCHAR).buildSchema();
ResultSetLoaderImpl.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).setBytes(value, value.length);
if (!rootWriter.isFull()) {
rootWriter.save();
}
count++;
}
// Discard the results.
rsLoader.harvest().zeroVectors();
// Harvest the next batch. Will be empty (because overflow row
// was discarded.)
rsLoader.startBatch();
RowSet result = fixture.wrap(rsLoader.harvest());
assertEquals(0, result.rowCount());
result.clear();
rsLoader.close();
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderDictArray method testScalarValue.
@Test
public void testScalarValue() {
TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).addDictArray("d", MinorType.VARCHAR).value(MinorType.INT).resumeSchema().buildSchema();
ResultSetLoaderImpl.ResultSetOptions options = new ResultSetOptionBuilder().readerSchema(schema).build();
ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
RowSetLoader rootWriter = rsLoader.writer();
// Write a couple of rows
rsLoader.startBatch();
rootWriter.addRow(10, objArray(map("a", 1, "b", 2, "d", 4), map("a", 2, "c", 3, "d", 1, "e", 4))).addRow(20, objArray()).addRow(30, objArray(map("a", 2, "c", 4, "d", 5, "e", 6, "f", 11), map("a", 1, "d", 6, "c", 3), map("b", 2, "a", 3)));
// Verify the batch
RowSet actual = fixture.wrap(rsLoader.harvest());
SingleRowSet expected = fixture.rowSetBuilder(schema).addRow(10, objArray(map("a", 1, "b", 2, "d", 4), map("a", 2, "c", 3, "d", 1, "e", 4))).addRow(20, objArray()).addRow(30, objArray(map("a", 2, "c", 4, "d", 5, "e", 6, "f", 11), map("a", 1, "d", 6, "c", 3), map("b", 2, "a", 3))).build();
RowSetUtilities.verify(expected, actual);
rsLoader.close();
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderDictArray method testCloseWithoutHarvest.
/**
* Test that memory is released if the loader is closed with an active
* batch (that is, before the batch is harvested.)
*/
@Test
public void testCloseWithoutHarvest() {
TupleMetadata schema = new SchemaBuilder().addDictArray("d", MinorType.INT).value(MinorType.VARCHAR).resumeSchema().buildSchema();
ResultSetLoaderImpl.ResultSetOptions options = new ResultSetOptionBuilder().readerSchema(schema).rowCountLimit(ValueVector.MAX_ROW_COUNT).build();
ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
RowSetLoader rootWriter = rsLoader.writer();
ArrayWriter arrayWriter = rootWriter.array("d");
DictWriter dictWriter = arrayWriter.dict();
rsLoader.startBatch();
for (int i = 0; i < 40; i++) {
rootWriter.start();
for (int j = 0; j < 3; j++) {
dictWriter.keyWriter().setInt(i);
dictWriter.valueWriter().scalar().setString("b-" + i);
arrayWriter.save();
}
rootWriter.save();
}
// Don't harvest the batch. Allocator will complain if the
// loader does not release memory.
rsLoader.close();
}
Aggregations