use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderRepeatedList method test2DLateSchema.
@Test
public void test2DLateSchema() {
final TupleMetadata schema = new SchemaBuilder().add("id", MinorType.INT).addRepeatedList("list2").addArray(MinorType.VARCHAR).resumeSchema().buildSchema();
final ResultSetLoaderImpl.ResultSetOptions options = new ResultSetOptionBuilder().build();
final ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
final RowSetLoader writer = rsLoader.writer();
// Add columns dynamically
writer.addColumn(schema.metadata(0));
writer.addColumn(schema.metadata(1).cloneEmpty());
// Yes, this is ugly. The whole repeated array idea is awkward.
// The only place it is used at present is in JSON where the
// awkwardness is mixed in with JSON complexity.
// Consider improving this API in the future.
((RepeatedListWriter) writer.array(1)).defineElement(schema.metadata(1).childSchema());
do2DTest(schema, rsLoader);
rsLoader.close();
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderRepeatedList method do2DTest.
private void do2DTest(TupleMetadata schema, ResultSetLoader rsLoader) {
final RowSetLoader writer = rsLoader.writer();
// Sanity check of writer structure
assertEquals(2, writer.size());
final ObjectWriter listObj = writer.column("list2");
assertEquals(ObjectType.ARRAY, listObj.type());
final ArrayWriter listWriter = listObj.array();
assertEquals(ObjectType.ARRAY, listWriter.entryType());
final ArrayWriter innerWriter = listWriter.array();
assertEquals(ObjectType.SCALAR, innerWriter.entryType());
final ScalarWriter strWriter = innerWriter.scalar();
assertEquals(ValueType.STRING, strWriter.valueType());
// Sanity test of schema
final TupleMetadata rowSchema = writer.tupleSchema();
assertEquals(2, rowSchema.size());
final ColumnMetadata listSchema = rowSchema.metadata(1);
assertEquals(MinorType.LIST, listSchema.type());
assertEquals(DataMode.REPEATED, listSchema.mode());
assertTrue(listSchema instanceof RepeatedListColumnMetadata);
assertEquals(StructureType.MULTI_ARRAY, listSchema.structureType());
assertNotNull(listSchema.childSchema());
final ColumnMetadata elementSchema = listSchema.childSchema();
assertEquals(listSchema.name(), elementSchema.name());
assertEquals(MinorType.VARCHAR, elementSchema.type());
assertEquals(DataMode.REPEATED, elementSchema.mode());
// Write values
rsLoader.startBatch();
writer.addRow(1, objArray(strArray("a", "b"), strArray("c", "d"))).addRow(2, objArray(strArray("e"), strArray(), strArray("f", "g", "h"))).addRow(3, objArray()).addRow(4, objArray(strArray(), strArray("i"), strArray()));
// Verify the values.
// (Relies on the row set level repeated list tests having passed.)
final RowSet expected = fixture.rowSetBuilder(schema).addRow(1, objArray(strArray("a", "b"), strArray("c", "d"))).addRow(2, objArray(strArray("e"), strArray(), strArray("f", "g", "h"))).addRow(3, objArray()).addRow(4, objArray(strArray(), strArray("i"), strArray())).build();
RowSetUtilities.verify(expected, fixture.wrap(rsLoader.harvest()));
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderUnions method testSimpleList.
/**
* Test for the case of a list defined to contain exactly one type.
* Relies on the row set tests to verify that the single type model
* works for lists. Here we test that the ResultSetLoader put the
* pieces together correctly.
*/
@Test
public void testSimpleList() {
// Schema with a list declared with one type, not expandable
final TupleMetadata schema = new SchemaBuilder().add("id", MinorType.INT).addList("list").addType(MinorType.VARCHAR).resumeSchema().buildSchema();
schema.metadata("list").variantSchema().becomeSimple();
final ResultSetLoaderImpl.ResultSetOptions options = new ResultSetOptionBuilder().readerSchema(schema).build();
final ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
final RowSetLoader writer = rsLoader.writer();
// Sanity check: should be an array of Varchar because we said the
// types within the list is not expandable.
final ArrayWriter arrWriter = writer.array("list");
assertEquals(ObjectType.SCALAR, arrWriter.entryType());
final ScalarWriter strWriter = arrWriter.scalar();
assertEquals(ValueType.STRING, strWriter.valueType());
// Can write a batch as if this was a repeated Varchar, except
// that any value can also be null.
rsLoader.startBatch();
writer.addRow(1, strArray("fred", "barney")).addRow(2, null).addRow(3, strArray("wilma", "betty", "pebbles"));
// Verify
final SingleRowSet expected = fixture.rowSetBuilder(schema).addRow(1, strArray("fred", "barney")).addRow(2, null).addRow(3, strArray("wilma", "betty", "pebbles")).build();
RowSetUtilities.verify(expected, fixture.wrap(rsLoader.harvest()));
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderUnions method testUnionOverflow.
@Test
public void testUnionOverflow() {
final TupleMetadata schema = new SchemaBuilder().add("id", MinorType.INT).addUnion("u").addType(MinorType.INT).addType(MinorType.VARCHAR).resumeSchema().buildSchema();
final ResultSetLoaderImpl.ResultSetOptions options = new ResultSetOptionBuilder().rowCountLimit(ValueVector.MAX_ROW_COUNT).readerSchema(schema).build();
final ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator(), options);
final RowSetLoader writer = rsLoader.writer();
// Fill the batch with enough data to cause overflow.
// Fill even rows with a Varchar, odd rows with an int.
// Data must be large enough to cause overflow before 32K rows
// (the half that get strings.
// 16 MB / 32 K = 512 bytes
// Make a bit bigger to overflow early.
final int strLength = 600;
final byte[] value = new byte[strLength - 6];
Arrays.fill(value, (byte) 'X');
final String strValue = new String(value, Charsets.UTF_8);
int count = 0;
rsLoader.startBatch();
while (!writer.isFull()) {
if (count % 2 == 0) {
writer.addRow(count, String.format("%s%06d", strValue, count));
} else {
writer.addRow(count, count * 10);
}
count++;
}
// Number of rows should be driven by vector size.
// Our row count should include the overflow row
final int expectedCount = ValueVector.MAX_BUFFER_SIZE / strLength * 2;
assertEquals(expectedCount + 1, count);
// Loader's row count should include only "visible" rows
assertEquals(expectedCount, writer.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());
// Verify the data.
RowSetReader reader = result.reader();
int readCount = 0;
while (reader.next()) {
assertEquals(readCount, reader.scalar(0).getInt());
if (readCount % 2 == 0) {
assertEquals(String.format("%s%06d", strValue, readCount), reader.variant(1).scalar().getString());
} else {
assertEquals(readCount * 10, reader.variant(1).scalar().getInt());
}
readCount++;
}
assertEquals(readCount, result.rowCount());
result.clear();
// Write a few more rows to verify the overflow row.
rsLoader.startBatch();
for (int i = 0; i < 1000; i++) {
if (count % 2 == 0) {
writer.addRow(count, String.format("%s%06d", strValue, count));
} else {
writer.addRow(count, count * 10);
}
count++;
}
result = fixture.wrap(rsLoader.harvest());
assertEquals(1001, result.rowCount());
final int startCount = readCount;
reader = result.reader();
while (reader.next()) {
assertEquals(readCount, reader.scalar(0).getInt());
if (readCount % 2 == 0) {
assertEquals(String.format("%s%06d", strValue, readCount), reader.variant(1).scalar().getString());
} else {
assertEquals(readCount * 10, reader.variant(1).scalar().getInt());
}
readCount++;
}
assertEquals(readCount - startCount, result.rowCount());
result.clear();
rsLoader.close();
}
use of org.apache.drill.exec.physical.resultSet.RowSetLoader in project drill by apache.
the class TestResultSetLoaderUnions method testRepeatedListOfUnion.
/**
* The repeated list type is way off in the weeds in Drill. It is not fully
* supported and the semantics are very murky as a result. It is not clear
* how such a structure fits into SQL or into an xDBC client. Still, it is
* part of Drill at present and must be supported in the result set loader.
* <p>
* This test models using the repeated list as a 2D array of UNION.
*/
@Test
public void testRepeatedListOfUnion() {
final ResultSetLoader rsLoader = new ResultSetLoaderImpl(fixture.allocator());
final RowSetLoader writer = rsLoader.writer();
// Can write a batch as if this was a repeated Varchar, except
// that any value can also be null.
rsLoader.startBatch();
writer.addColumn(MaterializedField.create("id", Types.required(MinorType.INT)));
// A union requires a structured column. The only tool to build that a present
// is the schema builder, so we use that and grab a single column.
final TupleMetadata dummySchema = new SchemaBuilder().addRepeatedList("list").addArray(MinorType.UNION).resumeSchema().buildSchema();
writer.addColumn(dummySchema.metadata(0));
// Sanity check: should be an array of array of variants.
final ArrayWriter outerArrWriter = writer.array("list");
assertEquals(ObjectType.ARRAY, outerArrWriter.entryType());
final ArrayWriter innerArrWriter = outerArrWriter.array();
assertEquals(ObjectType.VARIANT, innerArrWriter.entryType());
final VariantWriter variant = innerArrWriter.variant();
// No types, so all we can do is add a null list, or a list of nulls.
writer.addRow(1, null).addRow(2, objArray()).addRow(3, objArray(null, null)).addRow(4, objArray(variantArray(), variantArray())).addRow(5, objArray(variantArray(null, null), variantArray(null, null)));
// Add a String. Now we can create a list of strings and/or nulls.
variant.addMember(MinorType.VARCHAR);
assertTrue(variant.hasType(MinorType.VARCHAR));
writer.addRow(6, objArray(variantArray("fred", "wilma", null), variantArray("barney", "betty", null)));
// Add a map
final TupleWriter mapWriter = variant.addMember(MinorType.MAP).tuple();
mapWriter.addColumn(MetadataUtils.newScalar("first", Types.optional(MinorType.VARCHAR)));
mapWriter.addColumn(MetadataUtils.newScalar("last", Types.optional(MinorType.VARCHAR)));
// Add a map-based record
writer.addRow(7, objArray(variantArray(mapValue("fred", "flintstone"), mapValue("wilma", "flintstone")), variantArray(mapValue("barney", "rubble"), mapValue("betty", "rubble"))));
// Verify
final RowSet result = fixture.wrap(rsLoader.harvest());
final TupleMetadata schema = new SchemaBuilder().add("id", MinorType.INT).addRepeatedList("list").addList().addType(MinorType.VARCHAR).addMap().addNullable("first", MinorType.VARCHAR).addNullable("last", MinorType.VARCHAR).resumeUnion().resumeRepeatedList().resumeSchema().buildSchema();
final SingleRowSet expected = fixture.rowSetBuilder(schema).addRow(1, null).addRow(2, objArray()).addRow(3, objArray(null, null)).addRow(4, objArray(variantArray(), variantArray())).addRow(5, objArray(variantArray(null, null), variantArray(null, null))).addRow(6, objArray(variantArray("fred", "wilma", null), variantArray("barney", "betty", null))).addRow(7, objArray(variantArray(mapValue("fred", "flintstone"), mapValue("wilma", "flintstone")), variantArray(mapValue("barney", "rubble"), mapValue("betty", "rubble")))).build();
RowSetUtilities.verify(expected, result);
}
Aggregations