private TypeProtos.DataMode getDataMode(ColumnDescriptor column) {
if (column.getMaxRepetitionLevel() > 0) {
return DataMode.REPEATED;
} else if (column.getMaxDefinitionLevel() == 0) {
return TypeProtos.DataMode.REQUIRED;
} else {
return TypeProtos.DataMode.OPTIONAL;
}
}
/**
* Returns data type length for a given {@see ColumnDescriptor} and it's corresponding {@see
* SchemaElement}. Neither is enough information alone as the max repetition level (indicating if
* it is an array type) is in the ColumnDescriptor and the length of a fixed width field is stored
* at the schema level.
*
* @return the length if fixed width, else -1
*/
private int getDataTypeLength(ColumnDescriptor column, SchemaElement se) {
if (column.getType() != PrimitiveType.PrimitiveTypeName.BINARY) {
if (column.getMaxRepetitionLevel() > 0) {
return -1;
}
if (column.getType() == PrimitiveType.PrimitiveTypeName.FIXED_LEN_BYTE_ARRAY) {
return se.getType_length() * 8;
} else {
return getTypeLengthInBits(column.getType());
}
} else {
return -1;
}
}
private void readPageV2(DataPageV2 page) throws IOException {
this.pageValueCount = page.getValueCount();
this.repetitionLevelColumn =
createRLEIterator(
descriptor.getMaxRepetitionLevel(), page.getRepetitionLevels(), descriptor);
int bitWidth = BytesUtils.getWidthFromMaxInt(descriptor.getMaxDefinitionLevel());
this.defColumn = new VectorizedRleValuesReader(bitWidth);
this.definitionLevelColumn = new ValuesReaderIntIterator(this.defColumn);
this.defColumn.initFromBuffer(this.pageValueCount, page.getDefinitionLevels().toByteArray());
try {
initDataReader(page.getDataEncoding(), page.getData().toByteArray(), 0);
} catch (IOException e) {
throw new IOException("could not read page " + page + " in col " + descriptor, e);
}
}
@Override
public void setup(OperatorContext operatorContext, OutputMutator output)
throws ExecutionSetupException {
this.operatorContext = operatorContext;
if (!isStarQuery()) {
columnsFound = new boolean[getColumns().size()];
nullFilledVectors = new ArrayList<>();
}
columnStatuses = new ArrayList<>();
// totalRecords = footer.getBlocks().get(rowGroupIndex).getRowCount();
List<ColumnDescriptor> columns = footer.getFileMetaData().getSchema().getColumns();
allFieldsFixedLength = true;
ColumnDescriptor column;
ColumnChunkMetaData columnChunkMetaData;
int columnsToScan = 0;
mockRecordsRead = 0;
MaterializedField field;
// ParquetMetadataConverter metaConverter = new ParquetMetadataConverter();
FileMetaData fileMetaData;
logger.debug(
"Reading row group({}) with {} records in file {}.",
rowGroupIndex,
footer.getBlocks().get(rowGroupIndex).getRowCount(),
hadoopPath.toUri().getPath());
totalRecordsRead = 0;
// TODO - figure out how to deal with this better once we add nested reading, note also look
// where this map is used below
// store a map from column name to converted types if they are non-null
Map<String, SchemaElement> schemaElements =
ParquetReaderUtility.getColNameToSchemaElementMapping(footer);
// loop to add up the length of the fixed width columns and build the schema
for (int i = 0; i < columns.size(); ++i) {
column = columns.get(i);
SchemaElement se = schemaElements.get(column.getPath()[0]);
MajorType mt =
ParquetToDrillTypeConverter.toMajorType(
column.getType(),
se.getType_length(),
getDataMode(column),
se,
fragmentContext.getOptions());
field = MaterializedField.create(toFieldName(column.getPath()), mt);
if (!fieldSelected(field)) {
continue;
}
columnsToScan++;
int dataTypeLength = getDataTypeLength(column, se);
if (dataTypeLength == -1) {
allFieldsFixedLength = false;
} else {
bitWidthAllFixedFields += dataTypeLength;
}
}
// rowGroupOffset =
// footer.getBlocks().get(rowGroupIndex).getColumns().get(0).getFirstDataPageOffset();
if (columnsToScan != 0 && allFieldsFixedLength) {
recordsPerBatch =
(int)
Math.min(
Math.min(
batchSize / bitWidthAllFixedFields,
footer.getBlocks().get(0).getColumns().get(0).getValueCount()),
65535);
} else {
recordsPerBatch = DEFAULT_RECORDS_TO_READ_IF_NOT_FIXED_WIDTH;
}
try {
ValueVector vector;
SchemaElement schemaElement;
final ArrayList<VarLengthColumn<? extends ValueVector>> varLengthColumns = new ArrayList<>();
// initialize all of the column read status objects
boolean fieldFixedLength;
// the column chunk meta-data is not guaranteed to be in the same order as the columns in the
// schema
// a map is constructed for fast access to the correct columnChunkMetadata to correspond
// to an element in the schema
Map<String, Integer> columnChunkMetadataPositionsInList = new HashMap<>();
BlockMetaData rowGroupMetadata = footer.getBlocks().get(rowGroupIndex);
int colChunkIndex = 0;
for (ColumnChunkMetaData colChunk : rowGroupMetadata.getColumns()) {
columnChunkMetadataPositionsInList.put(
Arrays.toString(colChunk.getPath().toArray()), colChunkIndex);
colChunkIndex++;
}
for (int i = 0; i < columns.size(); ++i) {
column = columns.get(i);
columnChunkMetaData =
rowGroupMetadata
.getColumns()
.get(columnChunkMetadataPositionsInList.get(Arrays.toString(column.getPath())));
schemaElement = schemaElements.get(column.getPath()[0]);
MajorType type =
ParquetToDrillTypeConverter.toMajorType(
column.getType(),
schemaElement.getType_length(),
getDataMode(column),
schemaElement,
fragmentContext.getOptions());
field = MaterializedField.create(toFieldName(column.getPath()), type);
// the field was not requested to be read
if (!fieldSelected(field)) {
continue;
}
fieldFixedLength = column.getType() != PrimitiveType.PrimitiveTypeName.BINARY;
vector =
output.addField(
field,
(Class<? extends ValueVector>)
TypeHelper.getValueVectorClass(type.getMinorType(), type.getMode()));
if (column.getType() != PrimitiveType.PrimitiveTypeName.BINARY) {
if (column.getMaxRepetitionLevel() > 0) {
final RepeatedValueVector repeatedVector = RepeatedValueVector.class.cast(vector);
ColumnReader<?> dataReader =
ColumnReaderFactory.createFixedColumnReader(
this,
fieldFixedLength,
column,
columnChunkMetaData,
recordsPerBatch,
repeatedVector.getDataVector(),
schemaElement);
varLengthColumns.add(
new FixedWidthRepeatedReader(
this,
dataReader,
getTypeLengthInBits(column.getType()),
-1,
column,
columnChunkMetaData,
false,
repeatedVector,
schemaElement));
} else {
columnStatuses.add(
ColumnReaderFactory.createFixedColumnReader(
this,
fieldFixedLength,
column,
columnChunkMetaData,
recordsPerBatch,
vector,
schemaElement));
}
} else {
// create a reader and add it to the appropriate list
varLengthColumns.add(
ColumnReaderFactory.getReader(
this, -1, column, columnChunkMetaData, false, vector, schemaElement));
}
}
varLengthReader = new VarLenBinaryReader(this, varLengthColumns);
if (!isStarQuery()) {
List<SchemaPath> projectedColumns = Lists.newArrayList(getColumns());
SchemaPath col;
for (int i = 0; i < columnsFound.length; i++) {
col = projectedColumns.get(i);
assert col != null;
if (!columnsFound[i] && !col.equals(STAR_COLUMN)) {
nullFilledVectors.add(
(NullableIntVector)
output.addField(
MaterializedField.create(
col.getAsUnescapedPath(), Types.optional(TypeProtos.MinorType.INT)),
(Class<? extends ValueVector>)
TypeHelper.getValueVectorClass(
TypeProtos.MinorType.INT, DataMode.OPTIONAL)));
}
}
}
} catch (Exception e) {
handleAndRaise("Failure in setting up reader", e);
}
}