use of org.apache.drill.exec.vector.accessor.ScalarWriter 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();
}
use of org.apache.drill.exec.vector.accessor.ScalarWriter in project drill by apache.
the class TestResultSetLoaderProtocol method testOverwriteRow.
/**
* The writer protocol allows a client to write to a row any number of times
* before invoking {@code save()}. In this case, each new value simply
* overwrites the previous value. Here, we test the most basic case: a simple,
* flat tuple with no arrays. We use a very large Varchar that would, if
* overwrite were not working, cause vector overflow.
* <p>
* The ability to overwrite rows is seldom needed except in one future use
* case: writing a row, then applying a filter "in-place" to discard unwanted
* rows, without having to send the row downstream.
* <p>
* Because of this use case, specific rules apply when discarding row or
* overwriting values.
* <ul>
* <li>Values can be written once per row. Fixed-width columns actually allow
* multiple writes. But, because of the way variable-width columns work,
* multiple writes will cause undefined results.</li>
* <li>To overwrite a row, call <tt>start()</tt> without calling
* <tt>save()</tt> on the previous row. Doing so ignores data for the
* previous row and starts a new row in place of the old one.</li>
* </ul>
* Note that there is no explicit method to discard a row. Instead,
* the rule is that a row is not saved until <tt>save()</tt> is called.
*/
@Test
public void testOverwriteRow() {
TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).add("b", MinorType.VARCHAR).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();
// Can't use the shortcut to populate rows when doing overwrites.
ScalarWriter aWriter = rootWriter.scalar("a");
ScalarWriter bWriter = rootWriter.scalar("b");
// 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.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();
int rowId = 1;
while (reader.next()) {
assertEquals(rowId * 100, reader.scalar("a").getInt());
assertTrue(Arrays.equals(value, reader.scalar("b").getBytes()));
rowId++;
}
result.clear();
rsLoader.close();
}
use of org.apache.drill.exec.vector.accessor.ScalarWriter in project drill by apache.
the class TestScalarAccessors method testAppendWithArray.
/**
* Test the ability to append bytes to a VarChar column. Should work for
* Var16Char, but that type is not yet supported in Drill.
*/
@Test
public void testAppendWithArray() {
TupleMetadata schema = new SchemaBuilder().addArray("col", MinorType.VARCHAR).buildSchema();
DirectRowSet rs = DirectRowSet.fromSchema(fixture.allocator(), schema);
RowSetWriter writer = rs.writer(100);
ArrayWriter arrayWriter = writer.array("col");
ScalarWriter colWriter = arrayWriter.scalar();
byte[] first = "abc".getBytes();
byte[] second = "12345".getBytes();
for (int i = 0; i < 3; i++) {
colWriter.setBytes(first, first.length);
colWriter.appendBytes(second, second.length);
arrayWriter.save();
colWriter.setBytes(second, second.length);
colWriter.appendBytes(first, first.length);
arrayWriter.save();
colWriter.setBytes(first, first.length);
colWriter.appendBytes(second, second.length);
arrayWriter.save();
writer.save();
}
RowSet actual = writer.done();
RowSet expected = new RowSetBuilder(fixture.allocator(), schema).addSingleCol(strArray("abc12345", "12345abc", "abc12345")).addSingleCol(strArray("abc12345", "12345abc", "abc12345")).addSingleCol(strArray("abc12345", "12345abc", "abc12345")).build();
RowSetUtilities.verify(expected, actual);
}
use of org.apache.drill.exec.vector.accessor.ScalarWriter in project drill by axbaretto.
the class TestResultSetLoaderMaps method testBasics.
@Test
public void testBasics() {
TupleMetadata schema = new SchemaBuilder().add("a", MinorType.INT).addMap("m").add("c", MinorType.INT).add("d", MinorType.VARCHAR).resumeSchema().add("e", MinorType.VARCHAR).buildSchema();
ResultSetLoaderImpl.ResultSetOptions options = new OptionBuilder().setSchema(schema).build();
ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
RowSetLoader rootWriter = rsLoader.writer();
// Verify structure and schema
assertEquals(5, rsLoader.schemaVersion());
TupleMetadata actualSchema = rootWriter.schema();
assertEquals(3, actualSchema.size());
assertTrue(actualSchema.metadata(1).isMap());
assertEquals(2, actualSchema.metadata("m").mapSchema().size());
assertEquals(2, actualSchema.column("m").getChildren().size());
rsLoader.startBatch();
// Write a row the way that clients will do.
ScalarWriter aWriter = rootWriter.scalar("a");
TupleWriter mWriter = rootWriter.tuple("m");
ScalarWriter cWriter = mWriter.scalar("c");
ScalarWriter dWriter = mWriter.scalar("d");
ScalarWriter eWriter = rootWriter.scalar("e");
rootWriter.start();
aWriter.setInt(10);
cWriter.setInt(110);
dWriter.setString("fred");
eWriter.setString("pebbles");
rootWriter.save();
try {
mWriter.addColumn(SchemaBuilder.columnSchema("c", MinorType.INT, DataMode.OPTIONAL));
fail();
} catch (IllegalArgumentException e) {
// Expected
}
// Write another using the test-time conveniences
rootWriter.addRow(20, objArray(210, "barney"), "bam-bam");
// Harvest the batch
RowSet actual = fixture.wrap(rsLoader.harvest());
assertEquals(5, rsLoader.schemaVersion());
assertEquals(2, actual.rowCount());
// Validate data
SingleRowSet expected = fixture.rowSetBuilder(schema).addRow(10, objArray(110, "fred"), "pebbles").addRow(20, objArray(210, "barney"), "bam-bam").build();
new RowSetComparison(expected).verifyAndClearAll(actual);
rsLoader.close();
}
use of org.apache.drill.exec.vector.accessor.ScalarWriter in project drill by axbaretto.
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() {
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);
assertEquals(4, rsLoader.schemaVersion());
RowSetLoader rootWriter = rsLoader.writer();
// Can't use the shortcut to populate rows when doing a schema
// change.
ScalarWriter aWriter = rootWriter.scalar("a");
TupleWriter mWriter = rootWriter.tuple("m");
ScalarWriter bWriter = mWriter.scalar("b");
ScalarWriter cWriter = mWriter.scalar("c");
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()));
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();
}
Aggregations