void IDataType::updateAvgValueSizeHint(const IColumn & column, double & avg_value_size_hint)
{
/// Update the average value size hint if amount of read rows isn't too small
size_t column_size = column.size();
if (column_size > 10)
{
double current_avg_value_size = static_cast<double>(column.byteSize()) / column_size;
/// Heuristic is chosen so that avg_value_size_hint increases rapidly but decreases slowly.
if (current_avg_value_size > avg_value_size_hint)
avg_value_size_hint = std::min(1024., current_avg_value_size); /// avoid overestimation
else if (current_avg_value_size * 2 < avg_value_size_hint)
avg_value_size_hint = (current_avg_value_size + avg_value_size_hint * 3) / 4;
}
}
void NativeBlockInputStream::readData(const IDataType & type, IColumn & column, ReadBuffer & istr, size_t rows)
{
/** Для массивов требуется сначала десериализовать смещения, а потом значения.
*/
if (const DataTypeArray * type_arr = typeid_cast<const DataTypeArray *>(&type))
{
IColumn & offsets_column = *typeid_cast<ColumnArray &>(column).getOffsetsColumn();
type_arr->getOffsetsType()->deserializeBinary(offsets_column, istr, rows, 0);
if (offsets_column.size() != rows)
throw Exception("Cannot read all data in NativeBlockInputStream.", ErrorCodes::CANNOT_READ_ALL_DATA);
if (rows)
readData(
*type_arr->getNestedType(),
typeid_cast<ColumnArray &>(column).getData(),
istr,
typeid_cast<const ColumnArray &>(column).getOffsets()[rows - 1]);
}
else
type.deserializeBinary(column, istr, rows, 0); /// TODO Использовать avg_value_size_hint.
if (column.size() != rows)
throw Exception("Cannot read all data in NativeBlockInputStream.", ErrorCodes::CANNOT_READ_ALL_DATA);
}
void IMergedBlockOutputStream::writeData(
const String & name,
const IDataType & type,
const IColumn & column,
OffsetColumns & offset_columns,
bool skip_offsets)
{
size_t size = column.size();
size_t prev_mark = 0;
while (prev_mark < size)
{
size_t limit = 0;
/// If there is `index_offset`, then the first mark goes not immediately, but after this number of rows.
if (prev_mark == 0 && index_offset != 0)
limit = index_offset;
else
{
limit = storage.index_granularity;
/// Write marks.
type.enumerateStreams([&] (const IDataType::SubstreamPath & substream_path)
{
bool is_offsets = !substream_path.empty() && substream_path.back().type == IDataType::Substream::ArraySizes;
if (is_offsets && skip_offsets)
return;
String stream_name = IDataType::getFileNameForStream(name, substream_path);
/// Don't write offsets more than one time for Nested type.
if (is_offsets && offset_columns.count(stream_name))
return;
ColumnStream & stream = *column_streams[stream_name];
/// There could already be enough data to compress into the new block.
if (stream.compressed.offset() >= min_compress_block_size)
stream.compressed.next();
writeIntBinary(stream.plain_hashing.count(), stream.marks);
writeIntBinary(stream.compressed.offset(), stream.marks);
}, {});
}
IDataType::OutputStreamGetter stream_getter = [&] (const IDataType::SubstreamPath & substream_path) -> WriteBuffer *
{
bool is_offsets = !substream_path.empty() && substream_path.back().type == IDataType::Substream::ArraySizes;
if (is_offsets && skip_offsets)
return nullptr;
String stream_name = IDataType::getFileNameForStream(name, substream_path);
/// Don't write offsets more than one time for Nested type.
if (is_offsets && offset_columns.count(stream_name))
return nullptr;
return &column_streams[stream_name]->compressed;
};
type.serializeBinaryBulkWithMultipleStreams(column, stream_getter, prev_mark, limit, true, {});
/// So that instead of the marks pointing to the end of the compressed block, there were marks pointing to the beginning of the next one.
type.enumerateStreams([&] (const IDataType::SubstreamPath & substream_path)
{
bool is_offsets = !substream_path.empty() && substream_path.back().type == IDataType::Substream::ArraySizes;
if (is_offsets && skip_offsets)
return;
String stream_name = IDataType::getFileNameForStream(name, substream_path);
/// Don't write offsets more than one time for Nested type.
if (is_offsets && offset_columns.count(stream_name))
return;
column_streams[stream_name]->compressed.nextIfAtEnd();
}, {});
prev_mark += limit;
}
/// Memoize offsets for Nested types, that are already written. They will not be written again for next columns of Nested structure.
type.enumerateStreams([&] (const IDataType::SubstreamPath & substream_path)
{
bool is_offsets = !substream_path.empty() && substream_path.back().type == IDataType::Substream::ArraySizes;
if (is_offsets)
{
String stream_name = IDataType::getFileNameForStream(name, substream_path);
offset_columns.insert(stream_name);
}
}, {});
}
void IDataType::insertDefaultInto(IColumn & column) const
{
column.insertDefault();
}
