use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class JsonMessageReader method parseAndWrite.
private void parseAndWrite(ConsumerRecord<?, ?> record, byte[] recordArray) {
stream.setValue(new ByteArrayInputStream(recordArray));
if (kafkaJsonLoader == null) {
JsonLoaderOptions jsonLoaderOptions = new JsonLoaderOptions();
jsonLoaderOptions.allTextMode = readOptions.isAllTextMode();
jsonLoaderOptions.readNumbersAsDouble = readOptions.isReadNumbersAsDouble();
jsonLoaderOptions.skipMalformedRecords = readOptions.isSkipInvalidRecords();
jsonLoaderOptions.allowNanInf = readOptions.isAllowNanInf();
jsonLoaderOptions.enableEscapeAnyChar = readOptions.isAllowEscapeAnyChar();
jsonLoaderOptions.skipMalformedDocument = readOptions.isSkipInvalidRecords();
kafkaJsonLoader = (KafkaJsonLoader) new KafkaJsonLoader.KafkaJsonLoaderBuilder().resultSetLoader(resultSetLoader).standardOptions(negotiator.queryOptions()).options(jsonLoaderOptions).errorContext(negotiator.parentErrorContext()).fromStream(() -> stream).build();
}
RowSetLoader rowWriter = resultSetLoader.writer();
rowWriter.start();
if (kafkaJsonLoader.parser().next()) {
writeValue(rowWriter, MetaDataField.KAFKA_TOPIC, record.topic());
writeValue(rowWriter, MetaDataField.KAFKA_PARTITION_ID, record.partition());
writeValue(rowWriter, MetaDataField.KAFKA_OFFSET, record.offset());
writeValue(rowWriter, MetaDataField.KAFKA_TIMESTAMP, record.timestamp());
writeValue(rowWriter, MetaDataField.KAFKA_MSG_KEY, record.key() != null ? record.key().toString() : null);
rowWriter.save();
}
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderTorture method doTortureTest.
private void doTortureTest() {
TupleMetadata schema = new SchemaBuilder().add("n0", MinorType.INT).addMap("m1").addNullable("n1", MinorType.INT).addMapArray("m2").addNullable("n2", MinorType.INT).addNullable("s2", MinorType.VARCHAR).addMap("m3").addNullable("n3", MinorType.INT).addArray("s3", MinorType.VARCHAR).resumeMap().resumeMap().resumeSchema().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();
TestSetup setup = new TestSetup();
BatchWriter batchWriter = new BatchWriter(setup, rootWriter);
int totalRowCount = 0;
ReadState readState = new ReadState();
for (int batchCount = 0; batchCount < 10; batchCount++) {
rsLoader.startBatch();
batchWriter.writeBatch();
// Now the hard part. Verify the above batch.
RowSet result = fixture.wrap(rsLoader.harvest());
// result.print();
// Should have overflowed
int savedCount = batchWriter.rowCount();
assertEquals(savedCount, result.rowCount());
totalRowCount += savedCount;
assertEquals(totalRowCount, rsLoader.totalRowCount());
assertEquals(batchCount + 1, rsLoader.batchCount());
BatchReader reader = new BatchReader(setup, result.reader(), readState);
reader.verify();
result.clear();
}
// Last row overflow row
{
rsLoader.startBatch();
// Use this to visualize a string buffer. There is also a method
// to visualize offset vectors. These two are the most pesky vectors
// to get right.
// VectorPrinter.printStrings((VarCharVector) ((NullableVarCharVector) ((AbstractScalarWriter) batchWriter.s2Writer).vector()).getValuesVector(), 0, 8);
RowSet result = fixture.wrap(rsLoader.harvest());
// Use this here, or earlier, when things go amiss and you need
// to see what the actual results might be.
// result.print();
totalRowCount++;
assertEquals(totalRowCount, rsLoader.totalRowCount());
BatchReader reader = new BatchReader(setup, result.reader(), readState);
reader.verify();
result.clear();
}
rsLoader.close();
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader 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.resultSet.RowSetLoader 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.resultSet.RowSetLoader 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();
}
Aggregations