ClickHouse/src/Formats/ProtobufSerializer.cpp

3583 lines
160 KiB
C++
Raw Normal View History

#include <Formats/ProtobufSerializer.h>
#if USE_PROTOBUF
# include <Columns/ColumnAggregateFunction.h>
# include <Columns/ColumnArray.h>
# include <Columns/ColumnDecimal.h>
# include <Columns/ColumnLowCardinality.h>
# include <Columns/ColumnMap.h>
# include <Columns/ColumnNullable.h>
# include <Columns/ColumnFixedString.h>
# include <Columns/ColumnString.h>
# include <Columns/ColumnTuple.h>
# include <Columns/ColumnVector.h>
# include <Common/PODArray.h>
# include <Common/quoteString.h>
# include <Core/DecimalComparison.h>
# include <DataTypes/DataTypeAggregateFunction.h>
# include <DataTypes/DataTypeArray.h>
# include <DataTypes/DataTypesDecimal.h>
# include <DataTypes/DataTypeDateTime64.h>
# include <DataTypes/DataTypeEnum.h>
# include <DataTypes/DataTypeFixedString.h>
# include <DataTypes/DataTypeLowCardinality.h>
# include <DataTypes/DataTypeMap.h>
# include <DataTypes/DataTypeNullable.h>
# include <DataTypes/DataTypeTuple.h>
# include <DataTypes/DataTypeString.h>
2021-03-09 14:46:52 +00:00
# include <DataTypes/Serializations/SerializationDecimal.h>
# include <DataTypes/Serializations/SerializationFixedString.h>
# include <Formats/ProtobufReader.h>
# include <Formats/ProtobufWriter.h>
# include <Formats/RowInputMissingColumnsFiller.h>
# include <IO/Operators.h>
# include <IO/ReadBufferFromString.h>
# include <IO/ReadHelpers.h>
# include <IO/WriteBufferFromString.h>
# include <IO/WriteHelpers.h>
2021-10-02 07:13:14 +00:00
# include <base/range.h>
2022-01-30 19:49:48 +00:00
# include <base/sort.h>
# include <google/protobuf/descriptor.h>
# include <google/protobuf/descriptor.pb.h>
# include <boost/algorithm/string.hpp>
# include <boost/container/flat_map.hpp>
# include <boost/container/flat_set.hpp>
# include <boost/numeric/conversion/cast.hpp>
# include <boost/range/algorithm.hpp>
# include <boost/range/algorithm_ext/erase.hpp>
2022-04-27 15:05:45 +00:00
# include <Common/logger_useful.h>
namespace DB
{
namespace ErrorCodes
{
extern const int NO_COLUMNS_SERIALIZED_TO_PROTOBUF_FIELDS;
extern const int MULTIPLE_COLUMNS_SERIALIZED_TO_SAME_PROTOBUF_FIELD;
extern const int NO_COLUMN_SERIALIZED_TO_REQUIRED_PROTOBUF_FIELD;
extern const int DATA_TYPE_INCOMPATIBLE_WITH_PROTOBUF_FIELD;
extern const int PROTOBUF_FIELD_NOT_REPEATED;
extern const int PROTOBUF_BAD_CAST;
extern const int LOGICAL_ERROR;
extern const int BAD_ARGUMENTS;
}
namespace
{
using FieldDescriptor = google::protobuf::FieldDescriptor;
using MessageDescriptor = google::protobuf::Descriptor;
using FieldTypeId = google::protobuf::FieldDescriptor::Type;
/// Compares column's name with protobuf field's name.
/// This comparison is case-insensitive and ignores the difference between '.' and '_'
struct ColumnNameWithProtobufFieldNameComparator
{
static bool equals(char c1, char c2)
{
return convertChar(c1) == convertChar(c2);
}
static bool equals(std::string_view s1, std::string_view s2)
{
return (s1.length() == s2.length())
&& std::equal(s1.begin(), s1.end(), s2.begin(), [](char c1, char c2) { return convertChar(c1) == convertChar(c2); });
}
static bool less(std::string_view s1, std::string_view s2)
{
return std::lexicographical_compare(s1.begin(), s1.end(), s2.begin(), s2.end(), [](char c1, char c2) { return convertChar(c1) < convertChar(c2); });
}
static bool startsWith(std::string_view s1, std::string_view s2)
{
return (s1.length() >= s2.length()) && equals(s1.substr(0, s2.length()), s2);
}
static char convertChar(char c)
{
c = tolower(c);
if (c == '.')
c = '_';
return c;
}
};
2022-04-07 14:29:56 +00:00
bool isGoogleWrapperMessage(const MessageDescriptor & message_descriptor)
{
auto message_type = message_descriptor.well_known_type();
return (message_type >= google::protobuf::Descriptor::WELLKNOWNTYPE_DOUBLEVALUE)
&& (message_type <= google::protobuf::Descriptor::WELLKNOWNTYPE_BOOLVALUE);
}
bool isGoogleWrapperField(const FieldDescriptor & field_descriptor)
{
const auto * message_descriptor = field_descriptor.message_type();
if (message_descriptor == nullptr)
return false;
return isGoogleWrapperMessage(*message_descriptor);
}
bool isGoogleWrapperField(const FieldDescriptor * field_descriptor)
{
if (field_descriptor == nullptr)
return false;
return isGoogleWrapperField(*field_descriptor);
}
2022-04-07 14:29:56 +00:00
std::string_view googleWrapperColumnName(const FieldDescriptor & field_descriptor)
{
assert(isGoogleWrapperField(field_descriptor));
return field_descriptor.message_type()->field(0)->name();
}
// Should we omit null values (zero for numbers / empty string for strings) while storing them.
bool shouldSkipZeroOrEmpty(const FieldDescriptor & field_descriptor, bool google_wrappers_special_treatment = false)
{
if (!field_descriptor.is_optional())
return false;
if (field_descriptor.containing_type()->options().map_entry())
return false;
if (google_wrappers_special_treatment && isGoogleWrapperField(field_descriptor))
2022-04-07 14:29:56 +00:00
return false;
return field_descriptor.message_type() || (field_descriptor.file()->syntax() == google::protobuf::FileDescriptor::SYNTAX_PROTO3);
}
// Should we pack repeated values while storing them.
bool shouldPackRepeated(const FieldDescriptor & field_descriptor)
{
if (!field_descriptor.is_repeated())
return false;
switch (field_descriptor.type())
{
case FieldTypeId::TYPE_INT32:
case FieldTypeId::TYPE_UINT32:
case FieldTypeId::TYPE_SINT32:
case FieldTypeId::TYPE_INT64:
case FieldTypeId::TYPE_UINT64:
case FieldTypeId::TYPE_SINT64:
case FieldTypeId::TYPE_FIXED32:
case FieldTypeId::TYPE_SFIXED32:
case FieldTypeId::TYPE_FIXED64:
case FieldTypeId::TYPE_SFIXED64:
case FieldTypeId::TYPE_FLOAT:
case FieldTypeId::TYPE_DOUBLE:
case FieldTypeId::TYPE_BOOL:
case FieldTypeId::TYPE_ENUM:
break;
default:
return false;
}
if (field_descriptor.options().has_packed())
return field_descriptor.options().packed();
return field_descriptor.file()->syntax() == google::protobuf::FileDescriptor::SYNTAX_PROTO3;
}
WriteBuffer & writeIndent(WriteBuffer & out, size_t size) { return out << String(size * 4, ' '); }
[[noreturn]] void wrongNumberOfColumns(size_t number_of_columns, const String & expected)
{
throw Exception(ErrorCodes::LOGICAL_ERROR, "Wrong number of columns: expected {}, specified {}", expected, number_of_columns);
}
struct ProtobufReaderOrWriter
{
ProtobufReaderOrWriter(ProtobufReader & reader_) : reader(&reader_) {} // NOLINT(google-explicit-constructor)
ProtobufReaderOrWriter(ProtobufWriter & writer_) : writer(&writer_) {} // NOLINT(google-explicit-constructor)
ProtobufReader * const reader = nullptr;
ProtobufWriter * const writer = nullptr;
};
/// Base class for all serializers which serialize a single value.
class ProtobufSerializerSingleValue : public ProtobufSerializer
{
protected:
ProtobufSerializerSingleValue(
std::string_view column_name_,
const FieldDescriptor & field_descriptor_,
const ProtobufReaderOrWriter & reader_or_writer_)
: column_name(column_name_)
, field_descriptor(field_descriptor_)
, field_typeid(field_descriptor_.type())
, field_tag(field_descriptor.number())
, reader(reader_or_writer_.reader)
, writer(reader_or_writer_.writer)
, skip_zero_or_empty(shouldSkipZeroOrEmpty(field_descriptor))
{
}
void setColumns(const ColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
column = columns[0];
}
void setColumns(const MutableColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
column = columns[0]->getPtr();
}
template <typename NumberType>
void writeInt(NumberType value)
{
auto casted = castNumber<Int64>(value);
if (casted || !skip_zero_or_empty)
writer->writeInt(field_tag, casted);
}
template <typename NumberType>
void writeSInt(NumberType value)
{
auto casted = castNumber<Int64>(value);
if (casted || !skip_zero_or_empty)
writer->writeSInt(field_tag, casted);
}
template <typename NumberType>
void writeUInt(NumberType value)
{
auto casted = castNumber<UInt64>(value);
if (casted || !skip_zero_or_empty)
writer->writeUInt(field_tag, casted);
}
template <typename FieldType, typename NumberType>
void writeFixed(NumberType value)
{
auto casted = castNumber<FieldType>(value);
if (casted || !skip_zero_or_empty)
writer->writeFixed(field_tag, casted);
}
Int64 readInt() { return reader->readInt(); }
Int64 readSInt() { return reader->readSInt(); }
UInt64 readUInt() { return reader->readUInt(); }
template <typename FieldType>
FieldType readFixed()
{
return reader->readFixed<FieldType>();
}
void writeStr(std::string_view str)
{
if (!str.empty() || !skip_zero_or_empty)
writer->writeString(field_tag, str);
}
void readStr(String & str) { reader->readString(str); }
void readStrAndAppend(PaddedPODArray<UInt8> & str) { reader->readStringAndAppend(str); }
template <typename DestType>
DestType parseFromStr(std::string_view str) const
{
try
{
DestType result;
ReadBufferFromMemory buf(str.data(), str.length());
readText(result, buf);
return result;
}
catch (...)
{
cannotConvertValue(str, "String", TypeName<DestType>);
}
}
template <typename DestType, typename SrcType>
DestType castNumber(SrcType value) const
{
if constexpr (std::is_same_v<DestType, SrcType>)
return value;
DestType result;
try
{
/// TODO: use accurate::convertNumeric() maybe?
result = boost::numeric_cast<DestType>(value);
}
catch (boost::numeric::bad_numeric_cast &)
{
cannotConvertValue(toString(value), TypeName<SrcType>, TypeName<DestType>);
}
return result;
}
[[noreturn]] void incompatibleColumnType(std::string_view column_type) const
{
throw Exception(
ErrorCodes::DATA_TYPE_INCOMPATIBLE_WITH_PROTOBUF_FIELD,
"The column {} ({}) cannot be serialized to the field {} ({}) due to their types are not compatible",
quoteString(column_name),
column_type,
quoteString(field_descriptor.full_name()),
field_descriptor.type_name());
}
[[noreturn]] void cannotConvertValue(std::string_view src_value, std::string_view src_type_name, std::string_view dest_type_name) const
{
throw Exception(
"Could not convert value '" + String{src_value} + "' from type " + String{src_type_name} + " to type "
+ String{dest_type_name} + " while " + (reader ? "reading" : "writing") + " field "
+ quoteString(field_descriptor.name()) + " " + (reader ? "for inserting into" : "extracted from") + " column "
+ quoteString(column_name),
ErrorCodes::PROTOBUF_BAD_CAST);
}
const String column_name;
const FieldDescriptor & field_descriptor;
const FieldTypeId field_typeid;
const int field_tag;
ProtobufReader * const reader;
ProtobufWriter * const writer;
ColumnPtr column;
private:
const bool skip_zero_or_empty;
};
/// Serializes any ColumnVector<NumberType> to a field of any type except TYPE_MESSAGE, TYPE_GROUP.
/// NumberType must be one of the following types: Int8, UInt8, Int16, UInt16, Int32, UInt32, Int64, UInt64,
/// Int128, UInt128, Int256, UInt256, Float32, Float64.
/// And the field's type cannot be TYPE_ENUM if NumberType is Float32 or Float64.
template <typename NumberType>
class ProtobufSerializerNumber : public ProtobufSerializerSingleValue
{
public:
using ColumnType = ColumnVector<NumberType>;
ProtobufSerializerNumber(std::string_view column_name_, const FieldDescriptor & field_descriptor_, const ProtobufReaderOrWriter & reader_or_writer_)
: ProtobufSerializerSingleValue(column_name_, field_descriptor_, reader_or_writer_)
{
setFunctions();
}
void writeRow(size_t row_num) override
{
const auto & column_vector = assert_cast<const ColumnType &>(*column);
write_function(column_vector.getElement(row_num));
}
void readRow(size_t row_num) override
{
NumberType value = read_function();
auto & column_vector = assert_cast<ColumnType &>(column->assumeMutableRef());
if (row_num < column_vector.size())
column_vector.getElement(row_num) = value;
else
column_vector.insertValue(value);
}
void insertDefaults(size_t row_num) override
{
auto & column_vector = assert_cast<ColumnType &>(column->assumeMutableRef());
if (row_num < column_vector.size())
return;
column_vector.insertValue(getDefaultNumber());
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerNumber<" << TypeName<NumberType> << ">: column " << quoteString(column_name)
<< " -> field " << quoteString(field_descriptor.full_name()) << " (" << field_descriptor.type_name()
<< ")\n";
}
private:
void setFunctions()
{
switch (field_typeid)
{
case FieldTypeId::TYPE_INT32:
{
write_function = [this](NumberType value) { writeInt(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readInt()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_int32()); };
break;
}
case FieldTypeId::TYPE_SINT32:
{
write_function = [this](NumberType value) { writeSInt(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readSInt()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_int32()); };
break;
}
case FieldTypeId::TYPE_UINT32:
{
write_function = [this](NumberType value) { writeUInt(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readUInt()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_uint32()); };
break;
}
case FieldTypeId::TYPE_INT64:
{
write_function = [this](NumberType value) { writeInt(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readInt()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_int64()); };
break;
}
case FieldTypeId::TYPE_SINT64:
{
write_function = [this](NumberType value) { writeSInt(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readSInt()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_int64()); };
break;
}
case FieldTypeId::TYPE_UINT64:
{
write_function = [this](NumberType value) { writeUInt(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readUInt()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_uint64()); };
break;
}
case FieldTypeId::TYPE_FIXED32:
{
write_function = [this](NumberType value) { writeFixed<UInt32>(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readFixed<UInt32>()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_uint32()); };
break;
}
case FieldTypeId::TYPE_SFIXED32:
{
write_function = [this](NumberType value) { writeFixed<Int32>(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readFixed<Int32>()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_int32()); };
break;
}
case FieldTypeId::TYPE_FIXED64:
{
write_function = [this](NumberType value) { writeFixed<UInt64>(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readFixed<UInt64>()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_uint64()); };
break;
}
case FieldTypeId::TYPE_SFIXED64:
{
write_function = [this](NumberType value) { writeFixed<Int64>(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readFixed<Int64>()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_int64()); };
break;
}
case FieldTypeId::TYPE_FLOAT:
{
write_function = [this](NumberType value) { writeFixed<Float32>(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readFixed<Float32>()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_float()); };
break;
}
case FieldTypeId::TYPE_DOUBLE:
{
write_function = [this](NumberType value) { writeFixed<Float64>(value); };
read_function = [this]() -> NumberType { return castNumber<NumberType>(readFixed<Float64>()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_double()); };
break;
}
case FieldTypeId::TYPE_BOOL:
{
write_function = [this](NumberType value)
{
if (value == 0)
writeUInt(0);
else if (value == 1)
writeUInt(1);
else
cannotConvertValue(toString(value), TypeName<NumberType>, field_descriptor.type_name());
};
read_function = [this]() -> NumberType
{
UInt64 u64 = readUInt();
if (u64 < 2)
return static_cast<NumberType>(u64);
else
cannotConvertValue(toString(u64), field_descriptor.type_name(), TypeName<NumberType>);
};
default_function = [this]() -> NumberType { return static_cast<NumberType>(field_descriptor.default_value_bool()); };
break;
}
case FieldTypeId::TYPE_STRING:
case FieldTypeId::TYPE_BYTES:
{
write_function = [this](NumberType value)
{
WriteBufferFromString buf{text_buffer};
writeText(value, buf);
buf.finalize();
writeStr(text_buffer);
};
read_function = [this]() -> NumberType
{
readStr(text_buffer);
return parseFromStr<NumberType>(text_buffer);
};
default_function = [this]() -> NumberType { return parseFromStr<NumberType>(field_descriptor.default_value_string()); };
break;
}
case FieldTypeId::TYPE_ENUM:
{
if (std::is_floating_point_v<NumberType>)
incompatibleColumnType(TypeName<NumberType>);
write_function = [this](NumberType value)
{
int number = castNumber<int>(value);
checkProtobufEnumValue(number);
writeInt(number);
};
read_function = [this]() -> NumberType { return castNumber<NumberType>(readInt()); };
default_function = [this]() -> NumberType { return castNumber<NumberType>(field_descriptor.default_value_enum()->number()); };
break;
}
default:
incompatibleColumnType(TypeName<NumberType>);
}
}
NumberType getDefaultNumber()
{
if (!default_number)
default_number = default_function();
return *default_number;
}
void checkProtobufEnumValue(int value) const
{
const auto * enum_value_descriptor = field_descriptor.enum_type()->FindValueByNumber(value);
if (!enum_value_descriptor)
cannotConvertValue(toString(value), TypeName<NumberType>, field_descriptor.type_name());
}
protected:
std::function<void(NumberType)> write_function;
std::function<NumberType()> read_function;
std::function<NumberType()> default_function;
String text_buffer;
private:
std::optional<NumberType> default_number;
};
/// Serializes ColumnString or ColumnFixedString to a field of any type except TYPE_MESSAGE, TYPE_GROUP.
template <bool is_fixed_string>
class ProtobufSerializerString : public ProtobufSerializerSingleValue
{
public:
using ColumnType = std::conditional_t<is_fixed_string, ColumnFixedString, ColumnString>;
ProtobufSerializerString(
std::string_view column_name_,
2021-02-19 13:24:57 +00:00
const std::shared_ptr<const DataTypeFixedString> & fixed_string_data_type_,
const google::protobuf::FieldDescriptor & field_descriptor_,
const ProtobufReaderOrWriter & reader_or_writer_)
: ProtobufSerializerSingleValue(column_name_, field_descriptor_, reader_or_writer_)
2021-02-19 13:24:57 +00:00
, fixed_string_data_type(fixed_string_data_type_)
, n(fixed_string_data_type->getN())
{
static_assert(is_fixed_string, "This constructor for FixedString only");
setFunctions();
prepareEnumMapping();
}
ProtobufSerializerString(
std::string_view column_name_,
const google::protobuf::FieldDescriptor & field_descriptor_,
const ProtobufReaderOrWriter & reader_or_writer_)
: ProtobufSerializerSingleValue(column_name_, field_descriptor_, reader_or_writer_)
{
static_assert(!is_fixed_string, "This constructor for String only");
setFunctions();
prepareEnumMapping();
}
void writeRow(size_t row_num) override
{
const auto & column_string = assert_cast<const ColumnType &>(*column);
write_function(std::string_view{column_string.getDataAt(row_num)});
}
void readRow(size_t row_num) override
{
auto & column_string = assert_cast<ColumnType &>(column->assumeMutableRef());
const size_t old_size = column_string.size();
typename ColumnType::Chars & data = column_string.getChars();
const size_t old_data_size = data.size();
if (row_num < old_size)
{
text_buffer.clear();
read_function(text_buffer);
}
else
{
try
{
read_function(data);
}
catch (...)
{
data.resize_assume_reserved(old_data_size);
throw;
}
}
if constexpr (is_fixed_string)
{
if (row_num < old_size)
{
2021-03-09 14:46:52 +00:00
SerializationFixedString::alignStringLength(n, text_buffer, 0);
memcpy(data.data() + row_num * n, text_buffer.data(), n);
}
else
2021-03-09 14:46:52 +00:00
SerializationFixedString::alignStringLength(n, data, old_data_size);
}
else
{
if (row_num < old_size)
{
if (row_num != old_size - 1)
throw Exception("Cannot replace a string in the middle of ColumnString", ErrorCodes::LOGICAL_ERROR);
column_string.popBack(1);
}
try
{
data.push_back(0 /* terminating zero */);
column_string.getOffsets().push_back(data.size());
}
catch (...)
{
data.resize_assume_reserved(old_data_size);
column_string.getOffsets().resize_assume_reserved(old_size);
throw;
}
}
}
void insertDefaults(size_t row_num) override
{
auto & column_string = assert_cast<ColumnType &>(column->assumeMutableRef());
const size_t old_size = column_string.size();
if (row_num < old_size)
return;
const auto & default_str = getDefaultString();
typename ColumnType::Chars & data = column_string.getChars();
const size_t old_data_size = data.size();
try
{
data.insert(default_str.data(), default_str.data() + default_str.size());
}
catch (...)
{
data.resize_assume_reserved(old_data_size);
throw;
}
if constexpr (!is_fixed_string)
{
try
{
data.push_back(0 /* terminating zero */);
column_string.getOffsets().push_back(data.size());
}
catch (...)
{
data.resize_assume_reserved(old_data_size);
column_string.getOffsets().resize_assume_reserved(old_size);
throw;
}
}
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerString<" << (is_fixed_string ? "fixed" : "") << ">: column "
<< quoteString(column_name) << " -> field " << quoteString(field_descriptor.full_name()) << " ("
<< field_descriptor.type_name() << ")\n";
}
private:
void setFunctions()
{
switch (field_typeid)
{
case FieldTypeId::TYPE_INT32:
{
write_function = [this](std::string_view str) { writeInt(parseFromStr<Int32>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readInt(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_int32()); };
break;
}
case FieldTypeId::TYPE_SINT32:
{
write_function = [this](std::string_view str) { writeSInt(parseFromStr<Int32>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readSInt(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_int32()); };
break;
}
case FieldTypeId::TYPE_UINT32:
{
write_function = [this](std::string_view str) { writeUInt(parseFromStr<UInt32>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readUInt(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_uint32()); };
break;
}
case FieldTypeId::TYPE_INT64:
{
write_function = [this](std::string_view str) { writeInt(parseFromStr<Int64>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readInt(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_int64()); };
break;
}
case FieldTypeId::TYPE_SINT64:
{
write_function = [this](std::string_view str) { writeSInt(parseFromStr<Int64>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readSInt(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_int64()); };
break;
}
case FieldTypeId::TYPE_UINT64:
{
write_function = [this](std::string_view str) { writeUInt(parseFromStr<UInt64>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readUInt(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_uint64()); };
break;
}
case FieldTypeId::TYPE_FIXED32:
{
write_function = [this](std::string_view str) { writeFixed<UInt32>(parseFromStr<UInt32>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readFixed<UInt32>(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_uint32()); };
break;
}
case FieldTypeId::TYPE_SFIXED32:
{
write_function = [this](std::string_view str) { writeFixed<Int32>(parseFromStr<Int32>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readFixed<Int32>(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_int32()); };
break;
}
case FieldTypeId::TYPE_FIXED64:
{
write_function = [this](std::string_view str) { writeFixed<UInt64>(parseFromStr<UInt64>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readFixed<UInt64>(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_uint64()); };
break;
}
case FieldTypeId::TYPE_SFIXED64:
{
write_function = [this](std::string_view str) { writeFixed<Int64>(parseFromStr<Int64>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readFixed<Int64>(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_int64()); };
break;
}
case FieldTypeId::TYPE_FLOAT:
{
write_function = [this](std::string_view str) { writeFixed<Float32>(parseFromStr<Float32>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readFixed<Float32>(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_float()); };
break;
}
case FieldTypeId::TYPE_DOUBLE:
{
write_function = [this](std::string_view str) { writeFixed<Float64>(parseFromStr<Float64>(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { toStringAppend(readFixed<Float64>(), str); };
default_function = [this]() -> String { return toString(field_descriptor.default_value_double()); };
break;
}
case FieldTypeId::TYPE_BOOL:
{
write_function = [this](std::string_view str)
{
if (str == "true")
writeUInt(1);
else if (str == "false")
writeUInt(0);
else
cannotConvertValue(str, "String", field_descriptor.type_name());
};
read_function = [this](PaddedPODArray<UInt8> & str)
{
UInt64 u64 = readUInt();
if (u64 < 2)
{
std::string_view ref(u64 ? "true" : "false");
str.insert(ref.data(), ref.data() + ref.length());
}
else
cannotConvertValue(toString(u64), field_descriptor.type_name(), "String");
};
default_function = [this]() -> String
{
return field_descriptor.default_value_bool() ? "true" : "false";
};
break;
}
case FieldTypeId::TYPE_STRING:
case FieldTypeId::TYPE_BYTES:
{
write_function = [this](std::string_view str) { writeStr(str); };
read_function = [this](PaddedPODArray<UInt8> & str) { readStrAndAppend(str); };
default_function = [this]() -> String { return field_descriptor.default_value_string(); };
break;
}
case FieldTypeId::TYPE_ENUM:
{
write_function = [this](std::string_view str) { writeInt(stringToProtobufEnumValue(str)); };
read_function = [this](PaddedPODArray<UInt8> & str) { protobufEnumValueToStringAppend(readInt(), str); };
default_function = [this]() -> String { return field_descriptor.default_value_enum()->name(); };
break;
}
default:
this->incompatibleColumnType(is_fixed_string ? "FixedString" : "String");
}
}
const PaddedPODArray<UInt8> & getDefaultString()
{
if (!default_string)
{
PaddedPODArray<UInt8> arr;
auto str = default_function();
arr.insert(str.data(), str.data() + str.size());
if constexpr (is_fixed_string)
2021-03-09 14:46:52 +00:00
SerializationFixedString::alignStringLength(n, arr, 0);
default_string = std::move(arr);
}
return *default_string;
}
template <typename NumberType>
void toStringAppend(NumberType value, PaddedPODArray<UInt8> & str)
{
WriteBufferFromVector buf{str, AppendModeTag{}};
writeText(value, buf);
}
void prepareEnumMapping()
{
if ((field_typeid == google::protobuf::FieldDescriptor::TYPE_ENUM) && writer)
{
const auto & enum_descriptor = *field_descriptor.enum_type();
for (int i = 0; i != enum_descriptor.value_count(); ++i)
{
const auto & enum_value_descriptor = *enum_descriptor.value(i);
string_to_protobuf_enum_value_map.emplace(enum_value_descriptor.name(), enum_value_descriptor.number());
}
}
}
int stringToProtobufEnumValue(std::string_view str) const
{
auto it = string_to_protobuf_enum_value_map.find(str);
if (it == string_to_protobuf_enum_value_map.end())
cannotConvertValue(str, "String", field_descriptor.type_name());
return it->second;
}
std::string_view protobufEnumValueToString(int value) const
{
const auto * enum_value_descriptor = field_descriptor.enum_type()->FindValueByNumber(value);
if (!enum_value_descriptor)
cannotConvertValue(toString(value), field_descriptor.type_name(), "String");
return enum_value_descriptor->name();
}
void protobufEnumValueToStringAppend(int value, PaddedPODArray<UInt8> & str) const
{
auto name = protobufEnumValueToString(value);
str.insert(name.data(), name.data() + name.length());
}
2021-02-19 13:24:57 +00:00
const std::shared_ptr<const DataTypeFixedString> fixed_string_data_type;
const size_t n = 0;
std::function<void(std::string_view)> write_function;
std::function<void(PaddedPODArray<UInt8> &)> read_function;
std::function<String()> default_function;
std::unordered_map<std::string_view, int> string_to_protobuf_enum_value_map;
PaddedPODArray<UInt8> text_buffer;
std::optional<PaddedPODArray<UInt8>> default_string;
};
/// Serializes ColumnVector<NumberType> containing enum values to a field of any type
/// except TYPE_MESSAGE, TYPE_GROUP, TYPE_FLOAT, TYPE_DOUBLE, TYPE_BOOL.
/// NumberType can be either Int8 or Int16.
template <typename NumberType>
class ProtobufSerializerEnum : public ProtobufSerializerNumber<NumberType>
{
public:
using ColumnType = ColumnVector<NumberType>;
using EnumDataType = DataTypeEnum<NumberType>;
using BaseClass = ProtobufSerializerNumber<NumberType>;
ProtobufSerializerEnum(
std::string_view column_name_,
const std::shared_ptr<const EnumDataType> & enum_data_type_,
const FieldDescriptor & field_descriptor_,
const ProtobufReaderOrWriter & reader_or_writer_)
: BaseClass(column_name_, field_descriptor_, reader_or_writer_), enum_data_type(enum_data_type_)
{
assert(enum_data_type);
setFunctions();
prepareEnumMapping();
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerEnum<" << TypeName<NumberType> << ">: column " << quoteString(this->column_name)
<< " -> field " << quoteString(this->field_descriptor.full_name()) << " ("
<< this->field_descriptor.type_name() << ")\n";
}
private:
void setFunctions()
{
switch (this->field_typeid)
{
case FieldTypeId::TYPE_INT32:
case FieldTypeId::TYPE_SINT32:
case FieldTypeId::TYPE_UINT32:
case FieldTypeId::TYPE_INT64:
case FieldTypeId::TYPE_SINT64:
case FieldTypeId::TYPE_UINT64:
case FieldTypeId::TYPE_FIXED32:
case FieldTypeId::TYPE_SFIXED32:
case FieldTypeId::TYPE_FIXED64:
case FieldTypeId::TYPE_SFIXED64:
{
auto base_read_function = this->read_function;
this->read_function = [this, base_read_function]() -> NumberType
{
NumberType value = base_read_function();
checkEnumDataTypeValue(value);
return value;
};
auto base_default_function = this->default_function;
this->default_function = [this, base_default_function]() -> NumberType
{
auto value = base_default_function();
checkEnumDataTypeValue(value);
return value;
};
break;
}
case FieldTypeId::TYPE_STRING:
case FieldTypeId::TYPE_BYTES:
{
this->write_function = [this](NumberType value)
{
writeStr(enumDataTypeValueToString(value));
};
this->read_function = [this]() -> NumberType
{
readStr(this->text_buffer);
return stringToEnumDataTypeValue(this->text_buffer);
};
this->default_function = [this]() -> NumberType
{
return stringToEnumDataTypeValue(this->field_descriptor.default_value_string());
};
break;
}
case FieldTypeId::TYPE_ENUM:
{
this->write_function = [this](NumberType value) { writeInt(enumDataTypeValueToProtobufEnumValue(value)); };
this->read_function = [this]() -> NumberType { return protobufEnumValueToEnumDataTypeValue(readInt()); };
this->default_function = [this]() -> NumberType { return protobufEnumValueToEnumDataTypeValue(this->field_descriptor.default_value_enum()->number()); };
break;
}
default:
this->incompatibleColumnType(enum_data_type->getName());
}
}
void checkEnumDataTypeValue(NumberType value)
{
enum_data_type->findByValue(value); /// Throws an exception if the value isn't defined in the DataTypeEnum.
}
std::string_view enumDataTypeValueToString(NumberType value) const { return std::string_view{enum_data_type->getNameForValue(value)}; }
NumberType stringToEnumDataTypeValue(const String & str) const { return enum_data_type->getValue(str); }
void prepareEnumMapping()
{
if (this->field_typeid != FieldTypeId::TYPE_ENUM)
return;
const auto & enum_descriptor = *this->field_descriptor.enum_type();
/// We have two mappings:
/// enum_data_type: "string->NumberType" and protobuf_enum: string->int".
/// And here we want to make from those two mapping a new mapping "NumberType->int" (if we're writing protobuf data),
/// or "int->NumberType" (if we're reading protobuf data).
auto add_to_mapping = [&](NumberType enum_data_type_value, int protobuf_enum_value)
{
if (this->writer)
enum_data_type_value_to_protobuf_enum_value_map.emplace(enum_data_type_value, protobuf_enum_value);
else
protobuf_enum_value_to_enum_data_type_value_map.emplace(protobuf_enum_value, enum_data_type_value);
};
auto iless = [](std::string_view s1, std::string_view s2) { return ColumnNameWithProtobufFieldNameComparator::less(s1, s2); };
boost::container::flat_map<std::string_view, int, decltype(iless)> string_to_protobuf_enum_value_map;
typename decltype(string_to_protobuf_enum_value_map)::sequence_type string_to_protobuf_enum_value_seq;
2021-06-15 19:55:21 +00:00
for (int i : collections::range(enum_descriptor.value_count()))
string_to_protobuf_enum_value_seq.emplace_back(enum_descriptor.value(i)->name(), enum_descriptor.value(i)->number());
string_to_protobuf_enum_value_map.adopt_sequence(std::move(string_to_protobuf_enum_value_seq));
std::vector<NumberType> not_found_by_name_values;
not_found_by_name_values.reserve(enum_data_type->getValues().size());
/// Find mapping between enum_data_type and protobuf_enum by name (case insensitively),
/// i.e. we add to the mapping
/// NumberType(enum_data_type) -> "NAME"(enum_data_type) ->
/// -> "NAME"(protobuf_enum, same name) -> int(protobuf_enum)
for (const auto & [name, value] : enum_data_type->getValues())
{
auto it = string_to_protobuf_enum_value_map.find(name);
if (it != string_to_protobuf_enum_value_map.end())
add_to_mapping(value, it->second);
else
not_found_by_name_values.push_back(value);
}
if (!not_found_by_name_values.empty())
{
/// Find mapping between two enum_data_type and protobuf_enum by value.
/// If the same value has different names in enum_data_type and protobuf_enum
/// we can still add it to our mapping, i.e. we add to the mapping
/// NumberType(enum_data_type) -> int(protobuf_enum, same value)
for (NumberType value : not_found_by_name_values)
{
if (enum_descriptor.FindValueByNumber(value))
add_to_mapping(value, value);
}
}
size_t num_mapped_values = this->writer ? enum_data_type_value_to_protobuf_enum_value_map.size()
: protobuf_enum_value_to_enum_data_type_value_map.size();
if (!num_mapped_values && !enum_data_type->getValues().empty() && enum_descriptor.value_count())
{
throw Exception(
"Couldn't find mapping between data type " + enum_data_type->getName() + " and the enum " + quoteString(enum_descriptor.full_name())
+ " in the protobuf schema",
ErrorCodes::DATA_TYPE_INCOMPATIBLE_WITH_PROTOBUF_FIELD);
}
}
int enumDataTypeValueToProtobufEnumValue(NumberType value) const
{
auto it = enum_data_type_value_to_protobuf_enum_value_map.find(value);
if (it == enum_data_type_value_to_protobuf_enum_value_map.end())
cannotConvertValue(toString(value), enum_data_type->getName(), this->field_descriptor.type_name());
return it->second;
}
NumberType protobufEnumValueToEnumDataTypeValue(int value) const
{
auto it = protobuf_enum_value_to_enum_data_type_value_map.find(value);
if (it == protobuf_enum_value_to_enum_data_type_value_map.end())
cannotConvertValue(toString(value), this->field_descriptor.type_name(), enum_data_type->getName());
return it->second;
}
Int64 readInt() { return ProtobufSerializerSingleValue::readInt(); }
void writeInt(Int64 value) { ProtobufSerializerSingleValue::writeInt(value); }
void writeStr(std::string_view str) { ProtobufSerializerSingleValue::writeStr(str); }
void readStr(String & str) { ProtobufSerializerSingleValue::readStr(str); }
[[noreturn]] void cannotConvertValue(std::string_view src_value, std::string_view src_type_name, std::string_view dest_type_name) const { ProtobufSerializerSingleValue::cannotConvertValue(src_value, src_type_name, dest_type_name); }
const std::shared_ptr<const EnumDataType> enum_data_type;
std::unordered_map<NumberType, int> enum_data_type_value_to_protobuf_enum_value_map;
std::unordered_map<int, NumberType> protobuf_enum_value_to_enum_data_type_value_map;
};
/// Serializes a ColumnDecimal<DecimalType> to any field except TYPE_MESSAGE, TYPE_GROUP, TYPE_ENUM.
/// DecimalType must be one of the following types: Decimal32, Decimal64, Decimal128, Decimal256, DateTime64.
template <typename DecimalType>
class ProtobufSerializerDecimal : public ProtobufSerializerSingleValue
{
public:
using ColumnType = ColumnDecimal<DecimalType>;
ProtobufSerializerDecimal(
std::string_view column_name_,
const DataTypeDecimalBase<DecimalType> & decimal_data_type_,
const FieldDescriptor & field_descriptor_,
const ProtobufReaderOrWriter & reader_or_writer_)
: ProtobufSerializerSingleValue(column_name_, field_descriptor_, reader_or_writer_)
, precision(decimal_data_type_.getPrecision())
, scale(decimal_data_type_.getScale())
{
setFunctions();
}
void writeRow(size_t row_num) override
{
const auto & column_decimal = assert_cast<const ColumnType &>(*column);
write_function(column_decimal.getElement(row_num));
}
void readRow(size_t row_num) override
{
DecimalType decimal = read_function();
auto & column_decimal = assert_cast<ColumnType &>(column->assumeMutableRef());
if (row_num < column_decimal.size())
column_decimal.getElement(row_num) = decimal;
else
column_decimal.insertValue(decimal);
}
void insertDefaults(size_t row_num) override
{
auto & column_decimal = assert_cast<ColumnType &>(column->assumeMutableRef());
if (row_num < column_decimal.size())
return;
column_decimal.insertValue(getDefaultDecimal());
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerDecimal<" << TypeName<DecimalType> << ">: column " << quoteString(column_name)
<< " -> field " << quoteString(field_descriptor.full_name()) << " (" << field_descriptor.type_name()
<< ")\n";
}
private:
void setFunctions()
{
switch (field_typeid)
{
case FieldTypeId::TYPE_INT32:
{
write_function = [this](const DecimalType & decimal) { writeInt(decimalToNumber<Int32>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readInt()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_int32()); };
break;
}
case FieldTypeId::TYPE_SINT32:
{
write_function = [this](const DecimalType & decimal) { writeSInt(decimalToNumber<Int32>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readSInt()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_int32()); };
break;
}
case FieldTypeId::TYPE_UINT32:
{
write_function = [this](const DecimalType & decimal) { writeUInt(decimalToNumber<UInt32>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readUInt()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_uint32()); };
break;
}
case FieldTypeId::TYPE_INT64:
{
write_function = [this](const DecimalType & decimal) { writeInt(decimalToNumber<Int64>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readInt()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_int64()); };
break;
}
case FieldTypeId::TYPE_SINT64:
{
write_function = [this](const DecimalType & decimal) { writeSInt(decimalToNumber<Int64>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readSInt()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_int64()); };
break;
}
case FieldTypeId::TYPE_UINT64:
{
write_function = [this](const DecimalType & decimal) { writeUInt(decimalToNumber<UInt64>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readUInt()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_uint64()); };
break;
}
case FieldTypeId::TYPE_FIXED32:
{
write_function = [this](const DecimalType & decimal) { writeFixed<UInt32>(decimalToNumber<UInt32>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readFixed<UInt32>()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_uint32()); };
break;
}
case FieldTypeId::TYPE_SFIXED32:
{
write_function = [this](const DecimalType & decimal) { writeFixed<Int32>(decimalToNumber<Int32>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readFixed<Int32>()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_int32()); };
break;
}
case FieldTypeId::TYPE_FIXED64:
{
write_function = [this](const DecimalType & decimal) { writeFixed<UInt64>(decimalToNumber<UInt64>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readFixed<UInt64>()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_uint64()); };
break;
}
case FieldTypeId::TYPE_SFIXED64:
{
write_function = [this](const DecimalType & decimal) { writeFixed<Int64>(decimalToNumber<Int64>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readFixed<Int64>()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_int64()); };
break;
}
case FieldTypeId::TYPE_FLOAT:
{
write_function = [this](const DecimalType & decimal) { writeFixed<Float32>(decimalToNumber<Float32>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readFixed<Float32>()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_float()); };
break;
}
case FieldTypeId::TYPE_DOUBLE:
{
write_function = [this](const DecimalType & decimal) { writeFixed<Float64>(decimalToNumber<Float64>(decimal)); };
read_function = [this]() -> DecimalType { return numberToDecimal(readFixed<Float64>()); };
default_function = [this]() -> DecimalType { return numberToDecimal(field_descriptor.default_value_double()); };
break;
}
case FieldTypeId::TYPE_BOOL:
{
if (std::is_same_v<DecimalType, DateTime64>)
incompatibleColumnType(TypeName<DecimalType>);
else
{
write_function = [this](const DecimalType & decimal)
{
if (decimal.value == 0)
writeInt(0);
else if (DecimalComparison<DecimalType, int, EqualsOp>::compare(decimal, 1, scale, 0))
writeInt(1);
else
{
WriteBufferFromOwnString buf;
2021-08-16 08:03:23 +00:00
writeText(decimal, scale, buf, false);
cannotConvertValue(buf.str(), TypeName<DecimalType>, field_descriptor.type_name());
}
};
read_function = [this]() -> DecimalType
{
UInt64 u64 = readUInt();
if (u64 < 2)
return numberToDecimal(static_cast<UInt64>(u64 != 0));
else
cannotConvertValue(toString(u64), field_descriptor.type_name(), TypeName<DecimalType>);
};
default_function = [this]() -> DecimalType
{
return numberToDecimal(static_cast<Int64>(field_descriptor.default_value_bool()));
};
}
break;
}
case FieldTypeId::TYPE_STRING:
case FieldTypeId::TYPE_BYTES:
{
write_function = [this](const DecimalType & decimal)
{
decimalToString(decimal, text_buffer);
writeStr(text_buffer);
};
read_function = [this]() -> DecimalType
{
readStr(text_buffer);
return stringToDecimal(text_buffer);
};
default_function = [this]() -> DecimalType { return stringToDecimal(field_descriptor.default_value_string()); };
break;
}
default:
incompatibleColumnType(TypeName<DecimalType>);
}
}
DecimalType getDefaultDecimal()
{
if (!default_decimal)
default_decimal = default_function();
return *default_decimal;
}
template <typename NumberType>
DecimalType numberToDecimal(NumberType value) const
{
return convertToDecimal<DataTypeNumber<NumberType>, DataTypeDecimal<DecimalType>>(value, scale);
}
template <typename NumberType>
NumberType decimalToNumber(const DecimalType & decimal) const
{
return DecimalUtils::convertTo<NumberType>(decimal, scale);
}
void decimalToString(const DecimalType & decimal, String & str) const
{
WriteBufferFromString buf{str};
if constexpr (std::is_same_v<DecimalType, DateTime64>)
2021-08-16 08:03:23 +00:00
writeDateTimeText(decimal, scale, buf);
else
2021-08-16 08:03:23 +00:00
writeText(decimal, scale, buf, false);
}
DecimalType stringToDecimal(const String & str) const
{
ReadBufferFromString buf(str);
DecimalType decimal{0};
if constexpr (std::is_same_v<DecimalType, DateTime64>)
readDateTime64Text(decimal, scale, buf);
else
2021-03-09 14:46:52 +00:00
SerializationDecimal<DecimalType>::readText(decimal, buf, precision, scale);
return decimal;
}
const UInt32 precision;
const UInt32 scale;
std::function<void(const DecimalType &)> write_function;
std::function<DecimalType()> read_function;
std::function<DecimalType()> default_function;
std::optional<DecimalType> default_decimal;
String text_buffer;
};
using ProtobufSerializerDateTime64 = ProtobufSerializerDecimal<DateTime64>;
/// Serializes a ColumnVector<UInt16> containing dates to a field of any type except TYPE_MESSAGE, TYPE_GROUP, TYPE_BOOL, TYPE_ENUM.
class ProtobufSerializerDate : public ProtobufSerializerNumber<UInt16>
{
public:
ProtobufSerializerDate(
std::string_view column_name_,
const FieldDescriptor & field_descriptor_,
const ProtobufReaderOrWriter & reader_or_writer_)
: ProtobufSerializerNumber<UInt16>(column_name_, field_descriptor_, reader_or_writer_)
{
setFunctions();
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerDate: column " << quoteString(column_name) << " -> field "
<< quoteString(field_descriptor.full_name()) << " (" << field_descriptor.type_name() << ")\n";
}
private:
void setFunctions()
{
switch (field_typeid)
{
case FieldTypeId::TYPE_INT32:
case FieldTypeId::TYPE_SINT32:
case FieldTypeId::TYPE_UINT32:
case FieldTypeId::TYPE_INT64:
case FieldTypeId::TYPE_SINT64:
case FieldTypeId::TYPE_UINT64:
case FieldTypeId::TYPE_FIXED32:
case FieldTypeId::TYPE_SFIXED32:
case FieldTypeId::TYPE_FIXED64:
case FieldTypeId::TYPE_SFIXED64:
case FieldTypeId::TYPE_FLOAT:
case FieldTypeId::TYPE_DOUBLE:
break; /// already set in ProtobufSerializerNumber<UInt16>::setFunctions().
case FieldTypeId::TYPE_STRING:
case FieldTypeId::TYPE_BYTES:
{
write_function = [this](UInt16 value)
{
dateToString(static_cast<DayNum>(value), text_buffer);
writeStr(text_buffer);
};
read_function = [this]() -> UInt16
{
readStr(text_buffer);
return stringToDate(text_buffer);
};
default_function = [this]() -> UInt16 { return stringToDate(field_descriptor.default_value_string()); };
break;
}
default:
incompatibleColumnType("Date");
}
}
static void dateToString(DayNum date, String & str)
{
WriteBufferFromString buf{str};
writeText(date, buf);
}
static DayNum stringToDate(const String & str)
{
DayNum date;
ReadBufferFromString buf{str};
readDateText(date, buf);
return date;
}
};
/// Serializes a ColumnVector<UInt32> containing datetimes to a field of any type except TYPE_MESSAGE, TYPE_GROUP, TYPE_BOOL, TYPE_ENUM.
class ProtobufSerializerDateTime : public ProtobufSerializerNumber<UInt32>
{
public:
ProtobufSerializerDateTime(
std::string_view column_name_,
const DataTypeDateTime & type,
const FieldDescriptor & field_descriptor_,
const ProtobufReaderOrWriter & reader_or_writer_)
: ProtobufSerializerNumber<UInt32>(column_name_, field_descriptor_, reader_or_writer_),
date_lut(type.getTimeZone())
{
setFunctions();
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerDateTime: column " << quoteString(column_name) << " -> field "
<< quoteString(field_descriptor.full_name()) << " (" << field_descriptor.type_name() << ")\n";
}
protected:
const DateLUTImpl & date_lut;
void setFunctions()
{
switch (field_typeid)
{
case FieldTypeId::TYPE_INT32:
case FieldTypeId::TYPE_SINT32:
case FieldTypeId::TYPE_UINT32:
case FieldTypeId::TYPE_INT64:
case FieldTypeId::TYPE_SINT64:
case FieldTypeId::TYPE_UINT64:
case FieldTypeId::TYPE_FIXED32:
case FieldTypeId::TYPE_SFIXED32:
case FieldTypeId::TYPE_FIXED64:
case FieldTypeId::TYPE_SFIXED64:
case FieldTypeId::TYPE_FLOAT:
case FieldTypeId::TYPE_DOUBLE:
break; /// already set in ProtobufSerializerNumber<UInt32>::setFunctions().
case FieldTypeId::TYPE_STRING:
case FieldTypeId::TYPE_BYTES:
{
write_function = [this](UInt32 value)
{
dateTimeToString(value, text_buffer, date_lut);
writeStr(text_buffer);
};
read_function = [this]() -> UInt32
{
readStr(text_buffer);
return stringToDateTime(text_buffer, date_lut);
};
default_function = [this]() -> UInt32 { return stringToDateTime(field_descriptor.default_value_string(), date_lut); };
break;
}
default:
incompatibleColumnType("DateTime");
}
}
static void dateTimeToString(time_t tm, String & str, const DateLUTImpl & lut)
{
WriteBufferFromString buf{str};
writeDateTimeText(tm, buf, lut);
}
static time_t stringToDateTime(const String & str, const DateLUTImpl & lut)
{
ReadBufferFromString buf{str};
time_t tm = 0;
readDateTimeText(tm, buf, lut);
2021-03-15 20:40:33 +00:00
if (tm < 0)
tm = 0;
return tm;
}
};
2021-05-03 23:16:45 +00:00
/// Serializes a ColumnVector<UUID> containing UUIDs to a field of type TYPE_STRING or TYPE_BYTES.
class ProtobufSerializerUUID : public ProtobufSerializerSingleValue
{
public:
ProtobufSerializerUUID(
std::string_view column_name_,
const google::protobuf::FieldDescriptor & field_descriptor_,
const ProtobufReaderOrWriter & reader_or_writer_)
: ProtobufSerializerSingleValue(column_name_, field_descriptor_, reader_or_writer_)
{
setFunctions();
}
2021-05-03 23:16:45 +00:00
void writeRow(size_t row_num) override
{
const auto & column_vector = assert_cast<const ColumnVector<UUID> &>(*column);
write_function(column_vector.getElement(row_num));
}
void readRow(size_t row_num) override
{
UUID value = read_function();
auto & column_vector = assert_cast<ColumnVector<UUID> &>(column->assumeMutableRef());
if (row_num < column_vector.size())
column_vector.getElement(row_num) = value;
else
column_vector.insertValue(value);
}
void insertDefaults(size_t row_num) override
{
auto & column_vector = assert_cast<ColumnVector<UUID> &>(column->assumeMutableRef());
if (row_num < column_vector.size())
return;
column_vector.insertDefault();
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerUUID: column " << quoteString(column_name) << " -> field "
<< quoteString(field_descriptor.full_name()) << " (" << field_descriptor.type_name() << ")\n";
}
private:
void setFunctions()
{
if ((field_typeid != FieldTypeId::TYPE_STRING) && (field_typeid != FieldTypeId::TYPE_BYTES))
incompatibleColumnType("UUID");
2021-05-03 23:16:45 +00:00
write_function = [this](UUID value)
{
2021-05-03 23:16:45 +00:00
uuidToString(value, text_buffer);
writeStr(text_buffer);
};
2021-05-03 23:16:45 +00:00
read_function = [this]() -> UUID
{
readStr(text_buffer);
2021-05-07 18:16:27 +00:00
return parse<UUID>(text_buffer);
};
2021-05-07 18:16:27 +00:00
default_function = [this]() -> UUID { return parse<UUID>(field_descriptor.default_value_string()); };
}
static void uuidToString(const UUID & uuid, String & str)
{
WriteBufferFromString buf{str};
writeText(uuid, buf);
}
2021-05-03 23:16:45 +00:00
std::function<void(UUID)> write_function;
std::function<UUID()> read_function;
std::function<UUID()> default_function;
String text_buffer;
};
using ProtobufSerializerInterval = ProtobufSerializerNumber<Int64>;
/// Serializes a ColumnAggregateFunction to a field of type TYPE_STRING or TYPE_BYTES.
class ProtobufSerializerAggregateFunction : public ProtobufSerializerSingleValue
{
public:
ProtobufSerializerAggregateFunction(
std::string_view column_name_,
const std::shared_ptr<const DataTypeAggregateFunction> & aggregate_function_data_type_,
const google::protobuf::FieldDescriptor & field_descriptor_,
const ProtobufReaderOrWriter & reader_or_writer_)
: ProtobufSerializerSingleValue(column_name_, field_descriptor_, reader_or_writer_)
, aggregate_function_data_type(aggregate_function_data_type_)
, aggregate_function(aggregate_function_data_type->getFunction())
{
if ((field_typeid != FieldTypeId::TYPE_STRING) && (field_typeid != FieldTypeId::TYPE_BYTES))
incompatibleColumnType(aggregate_function_data_type->getName());
}
void writeRow(size_t row_num) override
{
const auto & column_af = assert_cast<const ColumnAggregateFunction &>(*column);
dataToString(column_af.getData()[row_num], text_buffer);
writeStr(text_buffer);
}
void readRow(size_t row_num) override
{
auto & column_af = assert_cast<ColumnAggregateFunction &>(column->assumeMutableRef());
Arena & arena = column_af.createOrGetArena();
AggregateDataPtr data;
readStr(text_buffer);
data = stringToData(text_buffer, arena);
if (row_num < column_af.size())
{
auto * old_data = std::exchange(column_af.getData()[row_num], data);
aggregate_function->destroy(old_data);
}
else
column_af.getData().push_back(data);
}
void insertDefaults(size_t row_num) override
{
auto & column_af = assert_cast<ColumnAggregateFunction &>(column->assumeMutableRef());
if (row_num < column_af.size())
return;
Arena & arena = column_af.createOrGetArena();
AggregateDataPtr data = stringToData(field_descriptor.default_value_string(), arena);
column_af.getData().push_back(data);
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerAggregateFunction: column " << quoteString(column_name) << " -> field "
<< quoteString(field_descriptor.full_name()) << " (" << field_descriptor.type_name() << ")\n";
}
private:
void dataToString(ConstAggregateDataPtr data, String & str) const
{
WriteBufferFromString buf{str};
aggregate_function->serialize(data, buf);
}
AggregateDataPtr stringToData(const String & str, Arena & arena) const
{
size_t size_of_state = aggregate_function->sizeOfData();
AggregateDataPtr data = arena.alignedAlloc(size_of_state, aggregate_function->alignOfData());
try
{
aggregate_function->create(data);
ReadBufferFromMemory buf(str.data(), str.length());
2021-05-30 13:57:30 +00:00
aggregate_function->deserialize(data, buf, std::nullopt, &arena);
return data;
}
catch (...)
{
aggregate_function->destroy(data);
throw;
}
}
const std::shared_ptr<const DataTypeAggregateFunction> aggregate_function_data_type;
const AggregateFunctionPtr aggregate_function;
String text_buffer;
};
/// Serializes a ColumnNullable.
class ProtobufSerializerNullable : public ProtobufSerializer
{
public:
explicit ProtobufSerializerNullable(std::unique_ptr<ProtobufSerializer> nested_serializer_)
: nested_serializer(std::move(nested_serializer_))
{
}
void setColumns(const ColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
column = columns[0];
const auto & column_nullable = assert_cast<const ColumnNullable &>(*column);
ColumnPtr nested_column = column_nullable.getNestedColumnPtr();
nested_serializer->setColumns(&nested_column, 1);
}
void setColumns(const MutableColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
ColumnPtr column0 = columns[0]->getPtr();
setColumns(&column0, 1);
}
void writeRow(size_t row_num) override
{
const auto & column_nullable = assert_cast<const ColumnNullable &>(*column);
const auto & null_map = column_nullable.getNullMapData();
if (!null_map[row_num])
nested_serializer->writeRow(row_num);
}
void readRow(size_t row_num) override
{
auto & column_nullable = assert_cast<ColumnNullable &>(column->assumeMutableRef());
auto & nested_column = column_nullable.getNestedColumn();
auto & null_map = column_nullable.getNullMapData();
size_t old_size = null_map.size();
nested_serializer->readRow(row_num);
if (row_num < old_size)
{
null_map[row_num] = false;
}
else
{
size_t new_size = nested_column.size();
if (new_size != old_size + 1)
throw Exception("Size of ColumnNullable is unexpected", ErrorCodes::LOGICAL_ERROR);
try
{
null_map.push_back(false);
}
catch (...)
{
nested_column.popBack(1);
throw;
}
}
}
void insertDefaults(size_t row_num) override
{
auto & column_nullable = assert_cast<ColumnNullable &>(column->assumeMutableRef());
if (row_num < column_nullable.size())
return;
column_nullable.insertDefault();
}
void insertNestedDefaults(size_t row_num)
{
auto & column_nullable = assert_cast<ColumnNullable &>(column->assumeMutableRef());
if (row_num < column_nullable.size())
return;
column_nullable.getNestedColumn().insertDefault();
column_nullable.getNullMapData().push_back(0);
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerNullable ->\n";
nested_serializer->describeTree(out, indent + 1);
}
private:
const std::unique_ptr<ProtobufSerializer> nested_serializer;
ColumnPtr column;
};
/// Serializes a ColumnMap.
class ProtobufSerializerMap : public ProtobufSerializer
{
public:
explicit ProtobufSerializerMap(std::unique_ptr<ProtobufSerializer> nested_serializer_)
: nested_serializer(std::move(nested_serializer_))
{
}
void setColumns(const ColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
const auto & column_map = assert_cast<const ColumnMap &>(*columns[0]);
ColumnPtr nested_column = column_map.getNestedColumnPtr();
nested_serializer->setColumns(&nested_column, 1);
}
void setColumns(const MutableColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
ColumnPtr column0 = columns[0]->getPtr();
setColumns(&column0, 1);
}
void writeRow(size_t row_num) override { nested_serializer->writeRow(row_num); }
void readRow(size_t row_num) override { nested_serializer->readRow(row_num); }
void insertDefaults(size_t row_num) override { nested_serializer->insertDefaults(row_num); }
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerMap ->\n";
nested_serializer->describeTree(out, indent + 1);
}
private:
const std::unique_ptr<ProtobufSerializer> nested_serializer;
};
/// Serializes a ColumnLowCardinality.
class ProtobufSerializerLowCardinality : public ProtobufSerializer
{
public:
explicit ProtobufSerializerLowCardinality(std::unique_ptr<ProtobufSerializer> nested_serializer_)
: nested_serializer(std::move(nested_serializer_))
{
}
void setColumns(const ColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
column = columns[0];
const auto & column_lc = assert_cast<const ColumnLowCardinality &>(*column);
ColumnPtr nested_column = column_lc.getDictionary().getNestedColumn();
nested_serializer->setColumns(&nested_column, 1);
read_value_column_set = false;
}
void setColumns(const MutableColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
ColumnPtr column0 = columns[0]->getPtr();
setColumns(&column0, 1);
}
void writeRow(size_t row_num) override
{
const auto & column_lc = assert_cast<const ColumnLowCardinality &>(*column);
size_t unique_row_number = column_lc.getIndexes().getUInt(row_num);
nested_serializer->writeRow(unique_row_number);
}
void readRow(size_t row_num) override
{
auto & column_lc = assert_cast<ColumnLowCardinality &>(column->assumeMutableRef());
if (!read_value_column_set)
{
if (!read_value_column)
{
ColumnPtr nested_column = column_lc.getDictionary().getNestedColumn();
read_value_column = nested_column->cloneEmpty();
}
nested_serializer->setColumns(&read_value_column, 1);
read_value_column_set = true;
}
read_value_column->popBack(read_value_column->size());
nested_serializer->readRow(0);
if (row_num < column_lc.size())
{
if (row_num != column_lc.size() - 1)
throw Exception("Cannot replace an element in the middle of ColumnLowCardinality", ErrorCodes::LOGICAL_ERROR);
column_lc.popBack(1);
}
column_lc.insertFromFullColumn(*read_value_column, 0);
}
void insertDefaults(size_t row_num) override
{
auto & column_lc = assert_cast<ColumnLowCardinality &>(column->assumeMutableRef());
if (row_num < column_lc.size())
return;
if (!default_value_column)
{
ColumnPtr nested_column = column_lc.getDictionary().getNestedColumn();
default_value_column = nested_column->cloneEmpty();
nested_serializer->setColumns(&default_value_column, 1);
nested_serializer->insertDefaults(0);
read_value_column_set = false;
}
column_lc.insertFromFullColumn(*default_value_column, 0);
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerLowCardinality ->\n";
nested_serializer->describeTree(out, indent + 1);
}
private:
const std::unique_ptr<ProtobufSerializer> nested_serializer;
ColumnPtr column;
MutableColumnPtr read_value_column;
bool read_value_column_set = false;
MutableColumnPtr default_value_column;
};
/// Serializes a ColumnArray to a repeated field.
class ProtobufSerializerArray : public ProtobufSerializer
{
public:
explicit ProtobufSerializerArray(std::unique_ptr<ProtobufSerializer> element_serializer_)
: element_serializer(std::move(element_serializer_))
{
}
void setColumns(const ColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
column = columns[0];
const auto & column_array = assert_cast<const ColumnArray &>(*column);
ColumnPtr data_column = column_array.getDataPtr();
element_serializer->setColumns(&data_column, 1);
}
void setColumns(const MutableColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
ColumnPtr column0 = columns[0]->getPtr();
setColumns(&column0, 1);
}
void writeRow(size_t row_num) override
{
const auto & column_array = assert_cast<const ColumnArray &>(*column);
const auto & offsets = column_array.getOffsets();
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(offsets[row_num - 1], offsets[row_num]))
element_serializer->writeRow(i);
}
void readRow(size_t row_num) override
{
auto & column_array = assert_cast<ColumnArray &>(column->assumeMutableRef());
auto & offsets = column_array.getOffsets();
size_t old_size = offsets.size();
if (row_num + 1 < old_size)
throw Exception("Cannot replace an element in the middle of ColumnArray", ErrorCodes::LOGICAL_ERROR);
auto data_column = column_array.getDataPtr();
size_t old_data_size = data_column->size();
try
{
element_serializer->readRow(old_data_size);
size_t data_size = data_column->size();
if (data_size != old_data_size + 1)
throw Exception("Size of ColumnArray is unexpected", ErrorCodes::LOGICAL_ERROR);
if (row_num < old_size)
offsets.back() = data_size;
else
offsets.push_back(data_size);
}
catch (...)
{
if (data_column->size() > old_data_size)
data_column->assumeMutableRef().popBack(data_column->size() - old_data_size);
if (offsets.size() > old_size)
column_array.getOffsetsColumn().popBack(offsets.size() - old_size);
throw;
}
}
void insertDefaults(size_t row_num) override
{
auto & column_array = assert_cast<ColumnArray &>(column->assumeMutableRef());
if (row_num < column_array.size())
return;
column_array.insertDefault();
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerArray ->\n";
element_serializer->describeTree(out, indent + 1);
}
private:
const std::unique_ptr<ProtobufSerializer> element_serializer;
ColumnPtr column;
};
/// Serializes a ColumnTuple as a repeated field (just like we serialize arrays).
class ProtobufSerializerTupleAsArray : public ProtobufSerializer
{
public:
ProtobufSerializerTupleAsArray(
std::string_view column_name_,
const std::shared_ptr<const DataTypeTuple> & tuple_data_type_,
const FieldDescriptor & field_descriptor_,
std::vector<std::unique_ptr<ProtobufSerializer>> element_serializers_)
: column_name(column_name_)
, tuple_data_type(tuple_data_type_)
, tuple_size(tuple_data_type->getElements().size())
, field_descriptor(field_descriptor_)
, element_serializers(std::move(element_serializers_))
{
assert(tuple_size);
assert(tuple_size == element_serializers.size());
}
void setColumns(const ColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
column = columns[0];
const auto & column_tuple = assert_cast<const ColumnTuple &>(*column);
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(tuple_size))
{
auto element_column = column_tuple.getColumnPtr(i);
element_serializers[i]->setColumns(&element_column, 1);
}
current_element_index = 0;
}
void setColumns(const MutableColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
ColumnPtr column0 = columns[0]->getPtr();
setColumns(&column0, 1);
}
void writeRow(size_t row_num) override
{
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(tuple_size))
element_serializers[i]->writeRow(row_num);
}
void readRow(size_t row_num) override
{
auto & column_tuple = assert_cast<ColumnTuple &>(column->assumeMutableRef());
size_t old_size = column_tuple.size();
if (row_num >= old_size)
current_element_index = 0;
insertDefaults(row_num);
if (current_element_index >= tuple_size)
{
throw Exception(
ErrorCodes::PROTOBUF_BAD_CAST,
"Column {}: More than {} elements was read from the field {} to fit in the data type {}",
quoteString(column_name),
tuple_size,
quoteString(field_descriptor.full_name()),
tuple_data_type->getName());
}
element_serializers[current_element_index]->readRow(row_num);
++current_element_index;
}
void insertDefaults(size_t row_num) override
{
auto & column_tuple = assert_cast<ColumnTuple &>(column->assumeMutableRef());
size_t old_size = column_tuple.size();
if (row_num > old_size)
return;
try
{
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(tuple_size))
element_serializers[i]->insertDefaults(row_num);
}
catch (...)
{
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(tuple_size))
{
auto element_column = column_tuple.getColumnPtr(i)->assumeMutable();
if (element_column->size() > old_size)
element_column->popBack(element_column->size() - old_size);
}
throw;
}
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerTupleAsArray: column " << quoteString(column_name) << " ("
<< tuple_data_type->getName() << ") -> field " << quoteString(field_descriptor.full_name()) << " ("
<< field_descriptor.type_name() << ") ->\n";
for (const auto & element_serializer : element_serializers)
element_serializer->describeTree(out, indent + 1);
}
private:
const String column_name;
const std::shared_ptr<const DataTypeTuple> tuple_data_type;
const size_t tuple_size;
const FieldDescriptor & field_descriptor;
const std::vector<std::unique_ptr<ProtobufSerializer>> element_serializers;
ColumnPtr column;
size_t current_element_index = 0;
};
/// Serializes a message (root or nested) in the protobuf schema.
class ProtobufSerializerMessage : public ProtobufSerializer
{
public:
struct FieldDesc
{
std::vector<size_t> column_indices;
const FieldDescriptor * field_descriptor;
std::unique_ptr<ProtobufSerializer> field_serializer;
};
ProtobufSerializerMessage(
std::vector<FieldDesc> && field_descs_,
const FieldDescriptor * parent_field_descriptor_,
bool with_length_delimiter_,
bool google_wrappers_special_treatment_,
std::unique_ptr<RowInputMissingColumnsFiller> missing_columns_filler_,
const ProtobufReaderOrWriter & reader_or_writer_)
: parent_field_descriptor(parent_field_descriptor_)
, with_length_delimiter(with_length_delimiter_)
, google_wrappers_special_treatment(google_wrappers_special_treatment_)
, missing_columns_filler(std::move(missing_columns_filler_))
, should_skip_if_empty(parent_field_descriptor
? shouldSkipZeroOrEmpty(*parent_field_descriptor, google_wrappers_special_treatment_) : false)
, reader(reader_or_writer_.reader)
, writer(reader_or_writer_.writer)
{
field_infos.reserve(field_descs_.size());
for (auto & desc : field_descs_)
field_infos.emplace_back(std::move(desc.column_indices), *desc.field_descriptor, std::move(desc.field_serializer));
2022-01-30 19:49:48 +00:00
::sort(field_infos.begin(), field_infos.end(),
[](const FieldInfo & lhs, const FieldInfo & rhs) { return lhs.field_tag < rhs.field_tag; });
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(field_infos.size()))
field_index_by_field_tag.emplace(field_infos[i].field_tag, i);
}
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
void setHasEnvelopeAsParent()
{
has_envelope_as_parent = true;
}
void setColumns(const ColumnPtr * columns_, size_t num_columns_) override
{
if (!num_columns_)
wrongNumberOfColumns(num_columns_, ">0");
std::vector<ColumnPtr> field_columns;
for (const FieldInfo & info : field_infos)
{
field_columns.clear();
field_columns.reserve(info.column_indices.size());
for (size_t column_index : info.column_indices)
{
if (column_index >= num_columns_)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Wrong column index {}, expected column indices <{}", column_index, num_columns_);
field_columns.emplace_back(columns_[column_index]);
}
info.field_serializer->setColumns(field_columns.data(), field_columns.size());
}
if (reader || (google_wrappers_special_treatment && isGoogleWrapperField(parent_field_descriptor)))
{
mutable_columns.resize(num_columns_);
for (size_t i : collections::range(num_columns_))
mutable_columns[i] = columns_[i]->assumeMutable();
std::vector<UInt8> column_is_missing;
column_is_missing.resize(num_columns_, true);
for (const FieldInfo & info : field_infos)
for (size_t i : info.column_indices)
column_is_missing[i] = false;
has_missing_columns = (std::find(column_is_missing.begin(), column_is_missing.end(), true) != column_is_missing.end());
}
}
void setColumns(const MutableColumnPtr * columns_, size_t num_columns_) override
{
Columns cols;
cols.reserve(num_columns_);
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(num_columns_))
cols.push_back(columns_[i]->getPtr());
setColumns(cols.data(), cols.size());
}
void writeRow(size_t row_num) override
{
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
if (parent_field_descriptor || has_envelope_as_parent)
writer->startNestedMessage();
else
writer->startMessage();
for (const FieldInfo & info : field_infos)
{
if (info.should_pack_repeated)
writer->startRepeatedPack();
info.field_serializer->writeRow(row_num);
if (info.should_pack_repeated)
writer->endRepeatedPack(info.field_tag, true);
}
if (parent_field_descriptor)
{
bool is_group = (parent_field_descriptor->type() == FieldTypeId::TYPE_GROUP);
writer->endNestedMessage(parent_field_descriptor->number(), is_group,
2022-04-27 08:12:03 +00:00
should_skip_if_empty || (google_wrappers_special_treatment && isNullGoogleWrapper(row_num)));
}
else if (has_envelope_as_parent)
{
writer->endNestedMessage(1, false, should_skip_if_empty);
}
else
writer->endMessage(with_length_delimiter);
}
void readRow(size_t row_num) override
{
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
if (parent_field_descriptor || has_envelope_as_parent)
reader->startNestedMessage();
else
reader->startMessage(with_length_delimiter);
if (!field_infos.empty())
{
last_field_index = 0;
last_field_tag = field_infos[0].field_tag;
size_t old_size = mutable_columns.empty() ? 0 : mutable_columns[0]->size();
try
{
int field_tag;
while (reader->readFieldNumber(field_tag))
{
size_t field_index = findFieldIndexByFieldTag(field_tag);
if (field_index == static_cast<size_t>(-1))
continue;
auto * field_serializer = field_infos[field_index].field_serializer.get();
field_serializer->readRow(row_num);
field_infos[field_index].field_read = true;
}
for (auto & info : field_infos)
{
if (info.field_read)
info.field_read = false;
else
{
2022-04-27 08:12:03 +00:00
if (google_wrappers_special_treatment && isNullableGoogleWrapper())
{
2022-04-27 08:12:03 +00:00
auto * nullable_ser = dynamic_cast<ProtobufSerializerNullable*>(info.field_serializer.get());
nullable_ser->insertNestedDefaults(row_num);
}
else
{
info.field_serializer->insertDefaults(row_num);
}
}
}
}
catch (...)
{
for (auto & column : mutable_columns)
{
if (column->size() > old_size)
column->popBack(column->size() - old_size);
}
throw;
}
}
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
if (parent_field_descriptor || has_envelope_as_parent)
reader->endNestedMessage();
else
reader->endMessage(false);
addDefaultsToMissingColumns(row_num);
}
void insertDefaults(size_t row_num) override
{
for (const FieldInfo & info : field_infos)
info.field_serializer->insertDefaults(row_num);
addDefaultsToMissingColumns(row_num);
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
size_t num_columns = 0;
for (const auto & field_info : field_infos)
num_columns += field_info.column_indices.size();
writeIndent(out, indent) << "ProtobufSerializerMessage: " << num_columns << " columns ->";
if (parent_field_descriptor)
out << " field " << quoteString(parent_field_descriptor->full_name()) << " (" << parent_field_descriptor->type_name() << ")";
2021-12-14 16:08:08 +00:00
for (const auto & field_info : field_infos)
{
out << "\n";
writeIndent(out, indent + 1) << "Columns #";
for (size_t j = 0; j != field_info.column_indices.size(); ++j)
{
if (j)
out << ", ";
out << field_info.column_indices[j];
}
out << " ->\n";
field_info.field_serializer->describeTree(out, indent + 2);
}
}
private:
size_t findFieldIndexByFieldTag(int field_tag)
{
while (true)
{
if (field_tag == last_field_tag)
return last_field_index;
if (field_tag < last_field_tag)
break;
if (++last_field_index >= field_infos.size())
break;
last_field_tag = field_infos[last_field_index].field_tag;
}
last_field_tag = field_tag;
auto it = field_index_by_field_tag.find(field_tag);
if (it == field_index_by_field_tag.end())
last_field_index = static_cast<size_t>(-1);
else
last_field_index = it->second;
return last_field_index;
}
void addDefaultsToMissingColumns(size_t row_num)
{
if (has_missing_columns)
missing_columns_filler->addDefaults(mutable_columns, row_num);
}
2022-04-27 08:12:03 +00:00
bool isNullGoogleWrapper(size_t row_num)
{
return isGoogleWrapperField(parent_field_descriptor) && mutable_columns[0].get()->isNullAt(row_num);
}
2022-04-27 08:12:03 +00:00
bool isNullableGoogleWrapper()
2022-04-14 10:20:15 +00:00
{
return isGoogleWrapperField(parent_field_descriptor) && mutable_columns[0].get()->isNullable();
}
struct FieldInfo
{
FieldInfo(
std::vector<size_t> && column_indices_,
const FieldDescriptor & field_descriptor_,
std::unique_ptr<ProtobufSerializer> field_serializer_)
: column_indices(std::move(column_indices_))
, field_descriptor(&field_descriptor_)
, field_tag(field_descriptor_.number())
, should_pack_repeated(shouldPackRepeated(field_descriptor_))
, field_serializer(std::move(field_serializer_))
{
}
std::vector<size_t> column_indices;
const FieldDescriptor * field_descriptor;
int field_tag;
bool should_pack_repeated;
std::unique_ptr<ProtobufSerializer> field_serializer;
bool field_read = false;
};
const FieldDescriptor * const parent_field_descriptor;
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
bool has_envelope_as_parent = false;
const bool with_length_delimiter;
const bool google_wrappers_special_treatment;
const std::unique_ptr<RowInputMissingColumnsFiller> missing_columns_filler;
const bool should_skip_if_empty;
ProtobufReader * const reader;
ProtobufWriter * const writer;
std::vector<FieldInfo> field_infos;
std::unordered_map<int, size_t> field_index_by_field_tag;
MutableColumns mutable_columns;
bool has_missing_columns = false;
int last_field_tag = 0;
size_t last_field_index = static_cast<size_t>(-1);
};
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
/// Serializes a top-level envelope message in the protobuf schema.
2022-03-17 10:35:23 +00:00
/// "Envelope" means that the contained subtree of serializers is enclosed in a message just once,
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
/// i.e. only when the first and the last row read/write trigger a read/write of the msg header.
class ProtobufSerializerEnvelope : public ProtobufSerializer
{
public:
ProtobufSerializerEnvelope(
std::unique_ptr<ProtobufSerializerMessage>&& serializer_,
const ProtobufReaderOrWriter & reader_or_writer_)
: serializer(std::move(serializer_))
, reader(reader_or_writer_.reader)
, writer(reader_or_writer_.writer)
{
// The inner serializer has a backreference of type protobuf::FieldDescriptor * to it's parent
// serializer. If it is unset, it considers itself the top-level message, otherwise a nested
// message and accordingly it makes start/endMessage() vs. startEndNestedMessage() calls into
// Protobuf(Writer|Reader). There is no field descriptor because Envelopes merely forward calls
// but don't contain data to be serialized. We must still force the inner serializer to act
// as nested message.
serializer->setHasEnvelopeAsParent();
}
void setColumns(const ColumnPtr * columns_, size_t num_columns_) override
{
serializer->setColumns(columns_, num_columns_);
}
void setColumns(const MutableColumnPtr * columns_, size_t num_columns_) override
{
serializer->setColumns(columns_, num_columns_);
}
void writeRow(size_t row_num) override
{
if (first_call_of_write_row)
{
writer->startMessage();
first_call_of_write_row = false;
}
serializer->writeRow(row_num);
}
void finalizeWrite() override
{
writer->endMessage(/*with_length_delimiter = */ true);
}
void readRow(size_t row_num) override
{
if (first_call_of_read_row)
{
reader->startMessage(/*with_length_delimiter = */ true);
first_call_of_read_row = false;
}
int field_tag;
2022-03-17 10:35:23 +00:00
[[maybe_unused]] bool ret = reader->readFieldNumber(field_tag);
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
assert(ret);
serializer->readRow(row_num);
}
void insertDefaults(size_t row_num) override
{
serializer->insertDefaults(row_num);
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerEnvelope ->\n";
serializer->describeTree(out, indent + 1);
}
std::unique_ptr<ProtobufSerializerMessage> serializer;
ProtobufReader * const reader;
ProtobufWriter * const writer;
bool first_call_of_write_row = true;
bool first_call_of_read_row = true;
};
/// Serializes a tuple with explicit names as a nested message.
class ProtobufSerializerTupleAsNestedMessage : public ProtobufSerializer
{
public:
explicit ProtobufSerializerTupleAsNestedMessage(std::unique_ptr<ProtobufSerializerMessage> message_serializer_)
: message_serializer(std::move(message_serializer_))
{
}
void setColumns(const ColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
const auto & column_tuple = assert_cast<const ColumnTuple &>(*columns[0]);
size_t tuple_size = column_tuple.tupleSize();
assert(tuple_size);
Columns element_columns;
element_columns.reserve(tuple_size);
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(tuple_size))
element_columns.emplace_back(column_tuple.getColumnPtr(i));
message_serializer->setColumns(element_columns.data(), element_columns.size());
}
void setColumns(const MutableColumnPtr * columns, [[maybe_unused]] size_t num_columns) override
{
if (num_columns != 1)
wrongNumberOfColumns(num_columns, "1");
ColumnPtr column0 = columns[0]->getPtr();
setColumns(&column0, 1);
}
void writeRow(size_t row_num) override { message_serializer->writeRow(row_num); }
void readRow(size_t row_num) override { message_serializer->readRow(row_num); }
void insertDefaults(size_t row_num) override { message_serializer->insertDefaults(row_num); }
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerTupleAsNestedMessage ->\n";
message_serializer->describeTree(out, indent + 1);
}
private:
const std::unique_ptr<ProtobufSerializerMessage> message_serializer;
};
/// Serializes a flattened Nested data type (an array of tuples with explicit names)
/// as a repeated nested message.
class ProtobufSerializerFlattenedNestedAsArrayOfNestedMessages : public ProtobufSerializer
{
public:
explicit ProtobufSerializerFlattenedNestedAsArrayOfNestedMessages(
const std::vector<std::string_view> & column_names_,
const FieldDescriptor * parent_field_descriptor_,
std::unique_ptr<ProtobufSerializerMessage> message_serializer_,
const std::function<String(size_t)> & get_root_desc_function_)
: parent_field_descriptor(parent_field_descriptor_)
, message_serializer(std::move(message_serializer_))
, get_root_desc_function(get_root_desc_function_)
{
column_names.reserve(column_names_.size());
for (const auto & column_name : column_names_)
column_names.emplace_back(column_name);
}
void setColumns(const ColumnPtr * columns, size_t num_columns) override
{
if (!num_columns)
wrongNumberOfColumns(num_columns, ">0");
data_columns.clear();
data_columns.reserve(num_columns);
offset_columns.clear();
offset_columns.reserve(num_columns);
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(num_columns))
{
const auto & column_array = assert_cast<const ColumnArray &>(*columns[i]);
data_columns.emplace_back(column_array.getDataPtr());
auto offset_column = column_array.getOffsetsPtr();
if (std::binary_search(offset_columns.begin(), offset_columns.end(), offset_column))
continue;
/// Keep `offset_columns` sorted.
offset_columns.insert(std::upper_bound(offset_columns.begin(), offset_columns.end(), offset_column), offset_column);
/// All the columns listed in `offset_columns` should have equal offsets.
if (i >= 1)
{
const auto & column_array0 = assert_cast<const ColumnArray &>(*columns[0]);
if (!column_array0.hasEqualOffsets(column_array))
{
throw Exception(ErrorCodes::PROTOBUF_BAD_CAST,
"Column #{} {} and column #{} {} are supposed to have equal offsets according to the following serialization tree:\n{}",
0, quoteString(column_names[0]), i, quoteString(column_names[i]), get_root_desc_function(0));
}
}
}
message_serializer->setColumns(data_columns.data(), data_columns.size());
}
void setColumns(const MutableColumnPtr * columns, size_t num_columns) override
{
Columns cols;
cols.reserve(num_columns);
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(num_columns))
cols.push_back(columns[i]->getPtr());
setColumns(cols.data(), cols.size());
}
void writeRow(size_t row_num) override
{
const auto & offset_column0 = assert_cast<const ColumnArray::ColumnOffsets &>(*offset_columns[0]);
size_t start_offset = offset_column0.getElement(row_num - 1);
size_t end_offset = offset_column0.getElement(row_num);
2021-06-15 19:55:21 +00:00
for (size_t i : collections::range(start_offset, end_offset))
message_serializer->writeRow(i);
}
void readRow(size_t row_num) override
{
size_t old_size = offset_columns[0]->size();
if (row_num + 1 < old_size)
throw Exception("Cannot replace an element in the middle of ColumnArray", ErrorCodes::LOGICAL_ERROR);
size_t old_data_size = data_columns[0]->size();
try
{
message_serializer->readRow(old_data_size);
size_t data_size = data_columns[0]->size();
if (data_size != old_data_size + 1)
throw Exception("Unexpected number of elements of ColumnArray has been read", ErrorCodes::LOGICAL_ERROR);
if (row_num < old_size)
{
for (auto & offset_column : offset_columns)
assert_cast<ColumnArray::ColumnOffsets &>(offset_column->assumeMutableRef()).getData().back() = data_size;
}
else
{
for (auto & offset_column : offset_columns)
assert_cast<ColumnArray::ColumnOffsets &>(offset_column->assumeMutableRef()).getData().push_back(data_size);
}
}
catch (...)
{
for (auto & data_column : data_columns)
{
if (data_column->size() > old_data_size)
data_column->assumeMutableRef().popBack(data_column->size() - old_data_size);
}
for (auto & offset_column : offset_columns)
{
if (offset_column->size() > old_size)
offset_column->assumeMutableRef().popBack(offset_column->size() - old_size);
}
throw;
}
}
void insertDefaults(size_t row_num) override
{
size_t old_size = offset_columns[0]->size();
if (row_num < old_size)
return;
try
{
size_t data_size = data_columns[0]->size();
for (auto & offset_column : offset_columns)
assert_cast<ColumnArray::ColumnOffsets &>(offset_column->assumeMutableRef()).getData().push_back(data_size);
}
catch (...)
{
for (auto & offset_column : offset_columns)
{
if (offset_column->size() > old_size)
offset_column->assumeMutableRef().popBack(offset_column->size() - old_size);
}
throw;
}
}
void describeTree(WriteBuffer & out, size_t indent) const override
{
writeIndent(out, indent) << "ProtobufSerializerFlattenedNestedAsArrayOfNestedMessages: columns ";
for (size_t i = 0; i != column_names.size(); ++i)
{
if (i)
out << ", ";
out << "#" << i << " " << quoteString(column_names[i]);
}
out << " ->";
if (parent_field_descriptor)
out << " field " << quoteString(parent_field_descriptor->full_name()) << " (" << parent_field_descriptor->type_name() << ") ->\n";
message_serializer->describeTree(out, indent + 1);
}
private:
Strings column_names;
const FieldDescriptor * parent_field_descriptor;
const std::unique_ptr<ProtobufSerializerMessage> message_serializer;
const std::function<String(size_t)> get_root_desc_function;
Columns data_columns;
Columns offset_columns;
};
/// Produces a tree of ProtobufSerializers which serializes a row as a protobuf message.
class ProtobufSerializerBuilder
{
public:
explicit ProtobufSerializerBuilder(const ProtobufReaderOrWriter & reader_or_writer_) : reader_or_writer(reader_or_writer_) {}
std::unique_ptr<ProtobufSerializer> buildMessageSerializer(
const Strings & column_names,
const DataTypes & data_types,
std::vector<size_t> & missing_column_indices,
const MessageDescriptor & message_descriptor,
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
bool with_length_delimiter,
bool with_envelope,
bool google_wrappers_special_treatment)
{
root_serializer_ptr = std::make_shared<ProtobufSerializer *>();
get_root_desc_function = [root_serializer_ptr = root_serializer_ptr](size_t indent) -> String
{
WriteBufferFromOwnString buf;
(*root_serializer_ptr)->describeTree(buf, indent);
return buf.str();
};
std::vector<size_t> used_column_indices;
auto message_serializer = buildMessageSerializerImpl(
/* num_columns = */ column_names.size(),
column_names.data(),
data_types.data(),
message_descriptor,
with_length_delimiter,
google_wrappers_special_treatment,
/* parent_field_descriptor = */ nullptr,
used_column_indices,
/* columns_are_reordered_outside = */ false,
/* check_nested_while_filling_missing_columns = */ true);
if (!message_serializer)
{
throw Exception(
"Not found matches between the names of the columns {" + boost::algorithm::join(column_names, ", ")
+ "} and the fields {" + boost::algorithm::join(getFieldNames(message_descriptor), ", ") + "} of the message "
+ quoteString(message_descriptor.full_name()) + " in the protobuf schema",
ErrorCodes::NO_COLUMNS_SERIALIZED_TO_PROTOBUF_FIELDS);
}
missing_column_indices.clear();
missing_column_indices.reserve(column_names.size() - used_column_indices.size());
auto used_column_indices_sorted = std::move(used_column_indices);
2022-01-30 19:49:48 +00:00
::sort(used_column_indices_sorted.begin(), used_column_indices_sorted.end());
boost::range::set_difference(collections::range(column_names.size()), used_column_indices_sorted,
std::back_inserter(missing_column_indices));
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
if (!with_envelope)
{
*root_serializer_ptr = message_serializer.get();
#if 0
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
LOG_INFO(&Poco::Logger::get("ProtobufSerializer"), "Serialization tree:\n{}", get_root_desc_function(0));
#endif
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
return message_serializer;
}
else
{
auto envelope_serializer = std::make_unique<ProtobufSerializerEnvelope>(std::move(message_serializer), reader_or_writer);
*root_serializer_ptr = envelope_serializer.get();
#if 0
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
LOG_INFO(&Poco::Logger::get("ProtobufSerializer"), "Serialization tree:\n{}", get_root_desc_function(0));
#endif
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
return envelope_serializer;
}
}
private:
/// Collects all field names from the message (used only to format error messages).
static Strings getFieldNames(const MessageDescriptor & message_descriptor)
{
Strings field_names;
field_names.reserve(message_descriptor.field_count());
2021-06-15 19:55:21 +00:00
for (int i : collections::range(message_descriptor.field_count()))
field_names.emplace_back(message_descriptor.field(i)->name());
return field_names;
}
static bool columnNameEqualsToFieldName(std::string_view column_name, const FieldDescriptor & field_descriptor)
{
std::string_view suffix;
return columnNameStartsWithFieldName(column_name, field_descriptor, suffix) && suffix.empty();
}
/// Checks if a passed column's name starts with a specified field's name.
/// The function also assigns `suffix` to the rest part of the column's name
/// which doesn't match to the field's name.
/// The function requires that rest part of the column's name to be started with a dot '.' or underline '_',
/// but doesn't include those '.' or '_' characters into `suffix`.
static bool columnNameStartsWithFieldName(std::string_view column_name, const FieldDescriptor & field_descriptor, std::string_view & suffix)
{
size_t matching_length = 0;
const MessageDescriptor & containing_type = *field_descriptor.containing_type();
if (containing_type.options().map_entry())
{
/// Special case. Elements of the data type Map are named as "keys" and "values",
/// but they're internally named as "key" and "value" in protobuf schema.
if (field_descriptor.number() == 1)
{
if (ColumnNameWithProtobufFieldNameComparator::startsWith(column_name, "keys"))
matching_length = strlen("keys");
else if (ColumnNameWithProtobufFieldNameComparator::startsWith(column_name, "key"))
matching_length = strlen("key");
}
else if (field_descriptor.number() == 2)
{
if (ColumnNameWithProtobufFieldNameComparator::startsWith(column_name, "values"))
matching_length = strlen("values");
else if (ColumnNameWithProtobufFieldNameComparator::startsWith(column_name, "value"))
matching_length = strlen("value");
}
}
if (!matching_length && ColumnNameWithProtobufFieldNameComparator::startsWith(column_name, field_descriptor.name()))
{
matching_length = field_descriptor.name().length();
}
if (column_name.length() == matching_length)
return true;
if ((column_name.length() < matching_length + 2) || !field_descriptor.message_type())
return false;
char first_char_after_matching = column_name[matching_length];
if (!ColumnNameWithProtobufFieldNameComparator::equals(first_char_after_matching, '.'))
return false;
suffix = column_name.substr(matching_length + 1);
return true;
}
/// Finds fields in the protobuf message which can be considered as matching
/// for a specified column's name. The found fields can be nested messages,
/// for that case suffixes are also returned.
/// This is only the first filter, buildMessageSerializerImpl() does other checks after calling this function.
static bool findFieldsByColumnName(
std::string_view column_name,
const MessageDescriptor & message_descriptor,
std::vector<std::pair<const FieldDescriptor *, std::string_view /* suffix */>> & out_field_descriptors_with_suffixes,
bool google_wrappers_special_treatment)
{
out_field_descriptors_with_suffixes.clear();
/// Find all fields which have the same name as column's name (case-insensitively); i.e. we're checking
/// field_name == column_name.
2021-06-15 19:55:21 +00:00
for (int i : collections::range(message_descriptor.field_count()))
{
const auto & field_descriptor = *message_descriptor.field(i);
if (columnNameEqualsToFieldName(column_name, field_descriptor))
{
std::string_view suffix =
google_wrappers_special_treatment && isGoogleWrapperField(field_descriptor)
? googleWrapperColumnName(field_descriptor)
: "";
out_field_descriptors_with_suffixes.emplace_back(&field_descriptor, suffix);
break;
}
}
if (!out_field_descriptors_with_suffixes.empty())
return true; /// We have an exact match, no need to compare prefixes.
/// Find all fields which name is used as prefix in column's name; i.e. we're checking
/// column_name == field_name + '.' + nested_message_field_name
2021-06-15 19:55:21 +00:00
for (int i : collections::range(message_descriptor.field_count()))
{
const auto & field_descriptor = *message_descriptor.field(i);
std::string_view suffix;
if (columnNameStartsWithFieldName(column_name, field_descriptor, suffix))
{
out_field_descriptors_with_suffixes.emplace_back(&field_descriptor, suffix);
}
}
/// Shorter suffixes first.
2022-01-30 19:49:48 +00:00
::sort(out_field_descriptors_with_suffixes.begin(), out_field_descriptors_with_suffixes.end(),
[](const std::pair<const FieldDescriptor *, std::string_view /* suffix */> & f1,
const std::pair<const FieldDescriptor *, std::string_view /* suffix */> & f2)
{
return f1.second.length() < f2.second.length();
});
return !out_field_descriptors_with_suffixes.empty();
}
/// Removes TypeIndex::Array from the specified vector of data types,
/// and also removes corresponding elements from two other vectors.
template <typename T1, typename T2>
static void removeNonArrayElements(DataTypes & data_types, std::vector<T1> & elements1, std::vector<T2> & elements2)
{
size_t initial_size = data_types.size();
assert(initial_size == elements1.size() && initial_size == elements2.size());
data_types.reserve(initial_size * 2);
elements1.reserve(initial_size * 2);
elements2.reserve(initial_size * 2);
2021-06-16 17:31:59 +00:00
for (size_t i : collections::range(initial_size))
{
if (data_types[i]->getTypeId() == TypeIndex::Array)
{
data_types.push_back(std::move(data_types[i]));
elements1.push_back(std::move(elements1[i]));
elements2.push_back(std::move(elements2[i]));
}
}
data_types.erase(data_types.begin(), data_types.begin() + initial_size);
elements1.erase(elements1.begin(), elements1.begin() + initial_size);
elements2.erase(elements2.begin(), elements2.begin() + initial_size);
}
/// Treats specified column indices as indices in another vector of column indices.
/// Useful for handling of nested messages.
static void transformColumnIndices(std::vector<size_t> & column_indices, const std::vector<size_t> & outer_indices)
{
for (size_t & idx : column_indices)
idx = outer_indices[idx];
}
/// Builds a serializer for a protobuf message (root or nested).
///
/// Some of the passed columns might be skipped, the function sets `used_column_indices` to
/// the list of those columns which match any fields in the protobuf message.
///
/// Normally `columns_are_reordered_outside` should be false - if it's false it means that
/// the used column indices will be passed to ProtobufSerializerMessage, which will write/read
/// only those columns and set the rest of columns by default.
/// Set `columns_are_reordered_outside` to true if you're going to reorder columns
/// according to `used_column_indices` returned and pass to
/// ProtobufSerializerMessage::setColumns() only the columns which are actually used.
std::unique_ptr<ProtobufSerializerMessage> buildMessageSerializerImpl(
size_t num_columns,
const String * column_names,
const DataTypePtr * data_types,
const MessageDescriptor & message_descriptor,
bool with_length_delimiter,
bool google_wrappers_special_treatment,
const FieldDescriptor * parent_field_descriptor,
std::vector<size_t> & used_column_indices,
bool columns_are_reordered_outside,
bool check_nested_while_filling_missing_columns)
{
std::vector<std::string_view> column_names_sv;
column_names_sv.reserve(num_columns);
for (size_t i = 0; i != num_columns; ++i)
column_names_sv.emplace_back(column_names[i]);
return buildMessageSerializerImpl(
num_columns,
column_names_sv.data(),
data_types,
message_descriptor,
with_length_delimiter,
google_wrappers_special_treatment,
parent_field_descriptor,
used_column_indices,
columns_are_reordered_outside,
check_nested_while_filling_missing_columns);
}
std::unique_ptr<ProtobufSerializerMessage> buildMessageSerializerImpl(
size_t num_columns,
const std::string_view * column_names,
const DataTypePtr * data_types,
const MessageDescriptor & message_descriptor,
bool with_length_delimiter,
bool google_wrappers_special_treatment,
const FieldDescriptor * parent_field_descriptor,
std::vector<size_t> & used_column_indices,
bool columns_are_reordered_outside,
bool check_nested_while_filling_missing_columns)
{
std::vector<ProtobufSerializerMessage::FieldDesc> field_descs;
boost::container::flat_map<const FieldDescriptor *, std::string_view> field_descriptors_in_use;
used_column_indices.clear();
used_column_indices.reserve(num_columns);
boost::container::flat_set<size_t> used_column_indices_sorted;
used_column_indices_sorted.reserve(num_columns);
size_t sequential_column_index = 0;
auto add_field_serializer = [&](std::string_view column_name_,
std::vector<size_t> && column_indices_,
const FieldDescriptor & field_descriptor_,
std::unique_ptr<ProtobufSerializer> field_serializer_)
{
auto it = field_descriptors_in_use.find(&field_descriptor_);
if (it != field_descriptors_in_use.end())
{
throw Exception(
"Multiple columns (" + backQuote(StringRef{it->second}) + ", "
+ backQuote(StringRef{column_name_}) + ") cannot be serialized to a single protobuf field "
+ quoteString(field_descriptor_.full_name()),
ErrorCodes::MULTIPLE_COLUMNS_SERIALIZED_TO_SAME_PROTOBUF_FIELD);
}
used_column_indices.insert(used_column_indices.end(), column_indices_.begin(), column_indices_.end());
used_column_indices_sorted.insert(column_indices_.begin(), column_indices_.end());
auto column_indices_to_pass_to_message_serializer = std::move(column_indices_);
if (columns_are_reordered_outside)
{
for (auto & index : column_indices_to_pass_to_message_serializer)
index = sequential_column_index++;
}
field_descs.push_back({std::move(column_indices_to_pass_to_message_serializer), &field_descriptor_, std::move(field_serializer_)});
field_descriptors_in_use.emplace(&field_descriptor_, column_name_);
};
std::vector<std::pair<const FieldDescriptor *, std::string_view>> field_descriptors_with_suffixes;
/// We're going through all the passed columns.
2021-06-16 17:31:59 +00:00
for (size_t column_idx : collections::range(num_columns))
{
if (used_column_indices_sorted.count(column_idx))
continue;
const auto & column_name = column_names[column_idx];
const auto & data_type = data_types[column_idx];
if (!findFieldsByColumnName(column_name, message_descriptor, field_descriptors_with_suffixes, google_wrappers_special_treatment))
continue;
if ((field_descriptors_with_suffixes.size() == 1) && field_descriptors_with_suffixes[0].second.empty())
{
/// Simple case: one column is serialized as one field.
const auto & field_descriptor = *field_descriptors_with_suffixes[0].first;
auto field_serializer = buildFieldSerializer(column_name, data_type,
field_descriptor, field_descriptor.is_repeated(), google_wrappers_special_treatment);
if (field_serializer)
{
add_field_serializer(column_name, {column_idx}, field_descriptor, std::move(field_serializer));
continue;
}
}
for (const auto & [field_descriptor, suffix] : field_descriptors_with_suffixes)
{
if (!suffix.empty())
{
/// Complex case: one or more columns are serialized as a nested message.
std::vector<size_t> nested_column_indices;
std::vector<std::string_view> nested_column_names;
nested_column_indices.reserve(num_columns - used_column_indices.size());
nested_column_names.reserve(num_columns - used_column_indices.size());
nested_column_indices.push_back(column_idx);
nested_column_names.push_back(suffix);
2021-06-15 19:55:21 +00:00
for (size_t j : collections::range(column_idx + 1, num_columns))
{
if (used_column_indices_sorted.count(j))
continue;
std::string_view other_suffix;
if (!columnNameStartsWithFieldName(column_names[j], *field_descriptor, other_suffix))
continue;
nested_column_indices.push_back(j);
nested_column_names.push_back(other_suffix);
}
DataTypes nested_data_types;
nested_data_types.reserve(nested_column_indices.size());
for (size_t j : nested_column_indices)
nested_data_types.push_back(data_types[j]);
/// Now we have up to `nested_message_column_names.size()` columns
/// which can be serialized as a nested message.
/// We will try to serialize those columns as one nested message,
/// then, if failed, as an array of nested messages (on condition if those columns are array).
bool has_fallback_to_array_of_nested_messages = false;
if (field_descriptor->is_repeated())
{
bool has_arrays
= boost::range::find_if(
nested_data_types, [](const DataTypePtr & dt) { return (dt->getTypeId() == TypeIndex::Array); })
!= nested_data_types.end();
if (has_arrays)
has_fallback_to_array_of_nested_messages = true;
}
/// Try to serialize those columns as one nested message.
try
{
std::vector<size_t> used_column_indices_in_nested;
auto nested_message_serializer = buildMessageSerializerImpl(
nested_column_names.size(),
nested_column_names.data(),
nested_data_types.data(),
*field_descriptor->message_type(),
/* with_length_delimiter = */ false,
google_wrappers_special_treatment,
field_descriptor,
used_column_indices_in_nested,
/* columns_are_reordered_outside = */ true,
/* check_nested_while_filling_missing_columns = */ false);
/// `columns_are_reordered_outside` is true because column indices are
/// going to be transformed and then written to the outer message,
/// see add_field_serializer() below.
if (nested_message_serializer)
{
transformColumnIndices(used_column_indices_in_nested, nested_column_indices);
add_field_serializer(
column_name,
std::move(used_column_indices_in_nested),
*field_descriptor,
std::move(nested_message_serializer));
break;
}
}
catch (Exception & e)
{
if ((e.code() != ErrorCodes::PROTOBUF_FIELD_NOT_REPEATED) || !has_fallback_to_array_of_nested_messages)
throw;
}
if (has_fallback_to_array_of_nested_messages)
{
/// Try to serialize those columns as an array of nested messages.
removeNonArrayElements(nested_data_types, nested_column_names, nested_column_indices);
for (DataTypePtr & dt : nested_data_types)
dt = assert_cast<const DataTypeArray &>(*dt).getNestedType();
std::vector<size_t> used_column_indices_in_nested;
auto nested_message_serializer = buildMessageSerializerImpl(
nested_column_names.size(),
nested_column_names.data(),
nested_data_types.data(),
*field_descriptor->message_type(),
/* with_length_delimiter = */ false,
google_wrappers_special_treatment,
field_descriptor,
used_column_indices_in_nested,
/* columns_are_reordered_outside = */ true,
/* check_nested_while_filling_missing_columns = */ false);
/// `columns_are_reordered_outside` is true because column indices are
/// going to be transformed and then written to the outer message,
/// see add_field_serializer() below.
if (nested_message_serializer)
{
std::vector<std::string_view> column_names_used;
2021-12-14 16:08:08 +00:00
column_names_used.reserve(used_column_indices_in_nested.size());
for (size_t i : used_column_indices_in_nested)
column_names_used.emplace_back(nested_column_names[i]);
auto field_serializer = std::make_unique<ProtobufSerializerFlattenedNestedAsArrayOfNestedMessages>(
std::move(column_names_used), field_descriptor, std::move(nested_message_serializer), get_root_desc_function);
transformColumnIndices(used_column_indices_in_nested, nested_column_indices);
add_field_serializer(column_name, std::move(used_column_indices_in_nested), *field_descriptor, std::move(field_serializer));
break;
}
}
}
}
}
/// Check that we've found matching columns for all the required fields.
if ((message_descriptor.file()->syntax() == google::protobuf::FileDescriptor::SYNTAX_PROTO2)
&& reader_or_writer.writer)
{
2021-06-15 19:55:21 +00:00
for (int i : collections::range(message_descriptor.field_count()))
{
const auto & field_descriptor = *message_descriptor.field(i);
if (field_descriptor.is_required() && !field_descriptors_in_use.count(&field_descriptor))
throw Exception(
"Field " + quoteString(field_descriptor.full_name()) + " is required to be set",
ErrorCodes::NO_COLUMN_SERIALIZED_TO_REQUIRED_PROTOBUF_FIELD);
}
}
if (field_descs.empty())
return nullptr;
std::unique_ptr<RowInputMissingColumnsFiller> missing_columns_filler;
if (reader_or_writer.reader)
{
if (check_nested_while_filling_missing_columns)
missing_columns_filler = std::make_unique<RowInputMissingColumnsFiller>(num_columns, column_names, data_types);
else
missing_columns_filler = std::make_unique<RowInputMissingColumnsFiller>();
}
return std::make_unique<ProtobufSerializerMessage>(
std::move(field_descs), parent_field_descriptor, with_length_delimiter, google_wrappers_special_treatment,
std::move(missing_columns_filler), reader_or_writer);
}
/// Builds a serializer for one-to-one match:
/// one column is serialized as one field in the protobuf message.
std::unique_ptr<ProtobufSerializer> buildFieldSerializer(
std::string_view column_name,
const DataTypePtr & data_type,
const FieldDescriptor & field_descriptor,
bool allow_repeat,
bool google_wrappers_special_treatment)
{
auto data_type_id = data_type->getTypeId();
switch (data_type_id)
{
case TypeIndex::UInt8: return std::make_unique<ProtobufSerializerNumber<UInt8>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::UInt16: return std::make_unique<ProtobufSerializerNumber<UInt16>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::UInt32: return std::make_unique<ProtobufSerializerNumber<UInt32>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::UInt64: return std::make_unique<ProtobufSerializerNumber<UInt64>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::UInt128: return std::make_unique<ProtobufSerializerNumber<UInt128>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::UInt256: return std::make_unique<ProtobufSerializerNumber<UInt256>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Int8: return std::make_unique<ProtobufSerializerNumber<Int8>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Int16: return std::make_unique<ProtobufSerializerNumber<Int16>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Int32: return std::make_unique<ProtobufSerializerNumber<Int32>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Int64: return std::make_unique<ProtobufSerializerNumber<Int64>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Int128: return std::make_unique<ProtobufSerializerNumber<Int128>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Int256: return std::make_unique<ProtobufSerializerNumber<Int256>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Float32: return std::make_unique<ProtobufSerializerNumber<Float32>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Float64: return std::make_unique<ProtobufSerializerNumber<Float64>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Date: return std::make_unique<ProtobufSerializerDate>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::DateTime: return std::make_unique<ProtobufSerializerDateTime>(column_name, assert_cast<const DataTypeDateTime &>(*data_type), field_descriptor, reader_or_writer);
case TypeIndex::DateTime64: return std::make_unique<ProtobufSerializerDateTime64>(column_name, assert_cast<const DataTypeDateTime64 &>(*data_type), field_descriptor, reader_or_writer);
case TypeIndex::String: return std::make_unique<ProtobufSerializerString<false>>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::FixedString: return std::make_unique<ProtobufSerializerString<true>>(column_name, typeid_cast<std::shared_ptr<const DataTypeFixedString>>(data_type), field_descriptor, reader_or_writer);
case TypeIndex::Enum8: return std::make_unique<ProtobufSerializerEnum<Int8>>(column_name, typeid_cast<std::shared_ptr<const DataTypeEnum8>>(data_type), field_descriptor, reader_or_writer);
case TypeIndex::Enum16: return std::make_unique<ProtobufSerializerEnum<Int16>>(column_name, typeid_cast<std::shared_ptr<const DataTypeEnum16>>(data_type), field_descriptor, reader_or_writer);
case TypeIndex::Decimal32: return std::make_unique<ProtobufSerializerDecimal<Decimal32>>(column_name, assert_cast<const DataTypeDecimal<Decimal32> &>(*data_type), field_descriptor, reader_or_writer);
case TypeIndex::Decimal64: return std::make_unique<ProtobufSerializerDecimal<Decimal64>>(column_name, assert_cast<const DataTypeDecimal<Decimal64> &>(*data_type), field_descriptor, reader_or_writer);
case TypeIndex::Decimal128: return std::make_unique<ProtobufSerializerDecimal<Decimal128>>(column_name, assert_cast<const DataTypeDecimal<Decimal128> &>(*data_type), field_descriptor, reader_or_writer);
case TypeIndex::Decimal256: return std::make_unique<ProtobufSerializerDecimal<Decimal256>>(column_name, assert_cast<const DataTypeDecimal<Decimal256> &>(*data_type), field_descriptor, reader_or_writer);
case TypeIndex::UUID: return std::make_unique<ProtobufSerializerUUID>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::Interval: return std::make_unique<ProtobufSerializerInterval>(column_name, field_descriptor, reader_or_writer);
case TypeIndex::AggregateFunction: return std::make_unique<ProtobufSerializerAggregateFunction>(column_name, typeid_cast<std::shared_ptr<const DataTypeAggregateFunction>>(data_type), field_descriptor, reader_or_writer);
case TypeIndex::Nullable:
{
const auto & nullable_data_type = assert_cast<const DataTypeNullable &>(*data_type);
auto nested_serializer = buildFieldSerializer(column_name, nullable_data_type.getNestedType(),
field_descriptor, allow_repeat, google_wrappers_special_treatment);
if (!nested_serializer)
return nullptr;
return std::make_unique<ProtobufSerializerNullable>(std::move(nested_serializer));
}
case TypeIndex::LowCardinality:
{
const auto & low_cardinality_data_type = assert_cast<const DataTypeLowCardinality &>(*data_type);
auto nested_serializer
= buildFieldSerializer(column_name, low_cardinality_data_type.getDictionaryType(),
field_descriptor, allow_repeat, google_wrappers_special_treatment);
if (!nested_serializer)
return nullptr;
return std::make_unique<ProtobufSerializerLowCardinality>(std::move(nested_serializer));
}
case TypeIndex::Map:
{
const auto & map_data_type = assert_cast<const DataTypeMap &>(*data_type);
auto nested_serializer = buildFieldSerializer(column_name, map_data_type.getNestedType(),
field_descriptor, allow_repeat, google_wrappers_special_treatment);
if (!nested_serializer)
return nullptr;
return std::make_unique<ProtobufSerializerMap>(std::move(nested_serializer));
}
case TypeIndex::Array:
{
/// Array is serialized as a repeated field.
const auto & array_data_type = assert_cast<const DataTypeArray &>(*data_type);
if (!allow_repeat)
2021-02-24 15:56:32 +00:00
throwFieldNotRepeated(field_descriptor, column_name);
auto nested_serializer = buildFieldSerializer(column_name, array_data_type.getNestedType(), field_descriptor,
/* allow_repeat = */ false, // We do our repeating now, so for nested type we forget about the repeating.
google_wrappers_special_treatment);
if (!nested_serializer)
return nullptr;
return std::make_unique<ProtobufSerializerArray>(std::move(nested_serializer));
}
case TypeIndex::Tuple:
{
/// Tuple is serialized in one of two ways:
/// 1) If the tuple has explicit names then it can be serialized as a nested message.
/// 2) Any tuple can be serialized as a repeated field, just like Array.
const auto & tuple_data_type = assert_cast<const DataTypeTuple &>(*data_type);
size_t size_of_tuple = tuple_data_type.getElements().size();
if (tuple_data_type.haveExplicitNames() && field_descriptor.message_type())
{
/// Try to serialize as a nested message.
std::vector<size_t> used_column_indices;
auto message_serializer = buildMessageSerializerImpl(
size_of_tuple,
tuple_data_type.getElementNames().data(),
tuple_data_type.getElements().data(),
*field_descriptor.message_type(),
/* with_length_delimiter = */ false,
2022-04-14 10:20:15 +00:00
google_wrappers_special_treatment,
&field_descriptor,
used_column_indices,
/* columns_are_reordered_outside = */ false,
/* check_nested_while_filling_missing_columns = */ false);
if (!message_serializer)
{
throw Exception(
"Not found matches between the names of the tuple's elements {"
+ boost::algorithm::join(tuple_data_type.getElementNames(), ", ") + "} and the fields {"
+ boost::algorithm::join(getFieldNames(*field_descriptor.message_type()), ", ") + "} of the message "
+ quoteString(field_descriptor.message_type()->full_name()) + " in the protobuf schema",
ErrorCodes::NO_COLUMNS_SERIALIZED_TO_PROTOBUF_FIELDS);
}
return std::make_unique<ProtobufSerializerTupleAsNestedMessage>(std::move(message_serializer));
}
/// Serialize as a repeated field.
if (!allow_repeat && (size_of_tuple > 1))
2021-02-24 15:56:32 +00:00
throwFieldNotRepeated(field_descriptor, column_name);
std::vector<std::unique_ptr<ProtobufSerializer>> nested_serializers;
for (const auto & nested_data_type : tuple_data_type.getElements())
{
auto nested_serializer = buildFieldSerializer(column_name, nested_data_type, field_descriptor,
/* allow_repeat = */ false, // We do our repeating now, so for nested type we forget about the repeating.
google_wrappers_special_treatment);
if (!nested_serializer)
break;
nested_serializers.push_back(std::move(nested_serializer));
}
if (nested_serializers.size() != size_of_tuple)
return nullptr;
return std::make_unique<ProtobufSerializerTupleAsArray>(
column_name,
typeid_cast<std::shared_ptr<const DataTypeTuple>>(data_type),
field_descriptor,
std::move(nested_serializers));
}
default:
throw Exception("Unknown data type: " + data_type->getName(), ErrorCodes::LOGICAL_ERROR);
}
}
2021-02-24 15:56:32 +00:00
[[noreturn]] static void throwFieldNotRepeated(const FieldDescriptor & field_descriptor, std::string_view column_name)
2021-02-24 15:56:32 +00:00
{
if (!field_descriptor.is_repeated())
throw Exception(
"The field " + quoteString(field_descriptor.full_name())
+ " must be repeated in the protobuf schema to match the column " + backQuote(StringRef{column_name}),
ErrorCodes::PROTOBUF_FIELD_NOT_REPEATED);
throw Exception(
"The field " + quoteString(field_descriptor.full_name())
+ " is repeated but the level of repeatedness is not enough to serialize a multidimensional array from the column "
+ backQuote(StringRef{column_name}) + ". It's recommended to make the parent field repeated as well.",
ErrorCodes::PROTOBUF_FIELD_NOT_REPEATED);
}
const ProtobufReaderOrWriter reader_or_writer;
std::function<String(size_t)> get_root_desc_function;
std::shared_ptr<ProtobufSerializer *> root_serializer_ptr;
};
template <typename Type>
DataTypePtr getEnumDataType(const google::protobuf::EnumDescriptor * enum_descriptor)
{
std::vector<std::pair<String, Type>> values;
for (int i = 0; i != enum_descriptor->value_count(); ++i)
{
const auto * enum_value_descriptor = enum_descriptor->value(i);
values.emplace_back(enum_value_descriptor->name(), enum_value_descriptor->number());
}
return std::make_shared<DataTypeEnum<Type>>(std::move(values));
}
std::optional<NameAndTypePair> getNameAndDataTypeFromField(const google::protobuf::FieldDescriptor * field_descriptor, bool skip_unsupported_fields, bool allow_repeat = true)
{
2021-12-16 18:48:38 +00:00
if (allow_repeat && field_descriptor->is_map())
{
auto name_and_type = getNameAndDataTypeFromField(field_descriptor, skip_unsupported_fields, false);
if (!name_and_type)
return std::nullopt;
const auto * tuple_type = assert_cast<const DataTypeTuple *>(name_and_type->type.get());
return NameAndTypePair{name_and_type->name, std::make_shared<DataTypeMap>(tuple_type->getElements())};
2021-12-16 18:48:38 +00:00
}
if (allow_repeat && field_descriptor->is_repeated())
{
auto name_and_type = getNameAndDataTypeFromField(field_descriptor, skip_unsupported_fields, false);
if (!name_and_type)
return std::nullopt;
return NameAndTypePair{name_and_type->name, std::make_shared<DataTypeArray>(name_and_type->type)};
}
switch (field_descriptor->type())
{
case FieldTypeId::TYPE_SFIXED32:
case FieldTypeId::TYPE_SINT32:
case FieldTypeId::TYPE_INT32:
return NameAndTypePair{field_descriptor->name(), std::make_shared<DataTypeInt32>()};
case FieldTypeId::TYPE_SFIXED64:
case FieldTypeId::TYPE_SINT64:
case FieldTypeId::TYPE_INT64:
return NameAndTypePair{field_descriptor->name(), std::make_shared<DataTypeInt64>()};
case FieldTypeId::TYPE_BOOL:
return NameAndTypePair{field_descriptor->name(), std::make_shared<DataTypeUInt8>()};
case FieldTypeId::TYPE_FLOAT:
return NameAndTypePair{field_descriptor->name(), std::make_shared<DataTypeFloat32>()};
case FieldTypeId::TYPE_DOUBLE:
return NameAndTypePair{field_descriptor->name(), std::make_shared<DataTypeFloat64>()};
case FieldTypeId::TYPE_UINT32:
case FieldTypeId::TYPE_FIXED32:
return NameAndTypePair{field_descriptor->name(), std::make_shared<DataTypeUInt32>()};
case FieldTypeId::TYPE_UINT64:
case FieldTypeId::TYPE_FIXED64:
return NameAndTypePair{field_descriptor->name(), std::make_shared<DataTypeUInt64>()};
case FieldTypeId::TYPE_BYTES:
case FieldTypeId::TYPE_STRING:
return NameAndTypePair{field_descriptor->name(), std::make_shared<DataTypeString>()};
case FieldTypeId::TYPE_ENUM:
{
const auto * enum_descriptor = field_descriptor->enum_type();
if (enum_descriptor->value_count() == 0)
{
if (skip_unsupported_fields)
return std::nullopt;
throw Exception("Empty enum field", ErrorCodes::BAD_ARGUMENTS);
}
int max_abs = std::abs(enum_descriptor->value(0)->number());
for (int i = 1; i != enum_descriptor->value_count(); ++i)
{
if (std::abs(enum_descriptor->value(i)->number()) > max_abs)
max_abs = std::abs(enum_descriptor->value(i)->number());
}
if (max_abs < 128)
return NameAndTypePair{field_descriptor->name(), getEnumDataType<Int8>(enum_descriptor)};
else if (max_abs < 32768)
return NameAndTypePair{field_descriptor->name(), getEnumDataType<Int16>(enum_descriptor)};
else
{
if (skip_unsupported_fields)
return std::nullopt;
throw Exception("ClickHouse supports only 8-bit and 16-bit enums", ErrorCodes::BAD_ARGUMENTS);
}
}
case FieldTypeId::TYPE_GROUP:
case FieldTypeId::TYPE_MESSAGE:
{
const auto * message_descriptor = field_descriptor->message_type();
if (message_descriptor->field_count() == 0)
{
if (skip_unsupported_fields)
return std::nullopt;
throw Exception("Empty messages are not supported", ErrorCodes::BAD_ARGUMENTS);
}
else if (message_descriptor->field_count() == 1)
{
const auto * nested_field_descriptor = message_descriptor->field(0);
auto nested_name_and_type = getNameAndDataTypeFromField(nested_field_descriptor, skip_unsupported_fields);
if (!nested_name_and_type)
return std::nullopt;
return NameAndTypePair{field_descriptor->name() + "_" + nested_name_and_type->name, nested_name_and_type->type};
}
else
{
DataTypes nested_types;
Strings nested_names;
for (int i = 0; i != message_descriptor->field_count(); ++i)
{
auto nested_name_and_type = getNameAndDataTypeFromField(message_descriptor->field(i), skip_unsupported_fields);
if (!nested_name_and_type)
continue;
nested_types.push_back(nested_name_and_type->type);
nested_names.push_back(nested_name_and_type->name);
}
if (nested_types.empty())
return std::nullopt;
return NameAndTypePair{field_descriptor->name(), std::make_shared<DataTypeTuple>(std::move(nested_types), std::move(nested_names))};
}
}
}
__builtin_unreachable();
}
}
std::unique_ptr<ProtobufSerializer> ProtobufSerializer::create(
const Strings & column_names,
const DataTypes & data_types,
std::vector<size_t> & missing_column_indices,
const google::protobuf::Descriptor & message_descriptor,
bool with_length_delimiter,
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
bool with_envelope,
bool flatten_google_wrappers,
ProtobufReader & reader)
{
return ProtobufSerializerBuilder(reader).buildMessageSerializer(column_names, data_types, missing_column_indices, message_descriptor, with_length_delimiter, with_envelope, flatten_google_wrappers);
}
std::unique_ptr<ProtobufSerializer> ProtobufSerializer::create(
const Strings & column_names,
const DataTypes & data_types,
const google::protobuf::Descriptor & message_descriptor,
bool with_length_delimiter,
Implement ProtobufList - fixes ClickHouse#16436 Introduce IO format "ProtobufList" with protobuf schema // schemafile.proto message Envelope { message MessageType { uint32 colA = 1; string colB = 2; } repeated MessageType mt = 1; } where "Envelope" is a hard-coded/expected top-level message and "MessageType" is a message with user-provided name containing the table fields to export/import, e.g. SELECT * FROM db1.tab1 FORMAT ProtobufList SETTINGS format_schema = 'schemafile:MessageType' As a result, the new format wraps a list of messages (one per row) into a single, containing message. Compare that to the schema of the existing IO formats "Protobuf" and "ProtobufSingle": message MessageType { uint32 colA = 1; string colB = 2; } The new format does not save space compared to the existing formats, but it is conceptually a bit more beautiful and also more convenenient. Implementation details: - Created new files ProtobufList(Input|Output)Format which use the existing ProtobufSerializer mechanism. The goal was to reuse as much code as possible and avoid copypasta. - I was torn between inheriting from I(Input|Output)Format vs. IRow(Input|Output)Format for ProtobufList(Input|Output)Format. The former is chunk-based which can be better for performance. Since the ProtobufSerializer mechanism is row-based but data is generally passed around in chunks, I decided for the latter to leverage the existing chunk <--> row mapping code in IRow(InputOutput)Format. - A new ProtobufSerializer called ProtobufSerializerEnvelope was introduced (--> ProtobufSerializer.cpp). It represents the top-level message which encloses the list of inner nested messages, i.e. the rows. - With the new format, parsing the schema file and matching the fields in the schema file to table column works like for the old formats. The only difference is that parsing starts one level below the "Envelope" (--> ProtobufSchema.cpp). This is more natural than forcing customers to have table columns start with "Envelope". - Creation of the ProtobufSerializer tree also works like before. What is different is that we finally add a ProtobufSerializerEnvelope as new root of the tree. It's only purpose is to write/read the top-level message for the first/last row to write/read. Caveats: - The low-level serialization code in ProtobufWriter uses an internal buffer which is flushed to the output file only in endMessage(). In the existing "Protobuf" format, this happens once per row, in the new format this happens only at the end of the serialization since row-level messages now call start/endNestedMessage(). As a future TODO to, the buffer should be flushed also in start/endNestedMessage() to reduce memory consumption.
2022-03-13 19:24:46 +00:00
bool with_envelope,
bool defaults_for_nullable_google_wrappers,
ProtobufWriter & writer)
{
std::vector<size_t> missing_column_indices;
return ProtobufSerializerBuilder(writer).buildMessageSerializer(column_names, data_types, missing_column_indices, message_descriptor, with_length_delimiter, with_envelope, defaults_for_nullable_google_wrappers);
}
NamesAndTypesList protobufSchemaToCHSchema(const google::protobuf::Descriptor * message_descriptor, bool skip_unsupported_fields)
{
NamesAndTypesList schema;
for (int i = 0; i != message_descriptor->field_count(); ++i)
{
if (auto name_and_type = getNameAndDataTypeFromField(message_descriptor->field(i), skip_unsupported_fields))
schema.push_back(*name_and_type);
}
if (schema.empty())
2022-07-20 14:33:46 +00:00
throw Exception(ErrorCodes::BAD_ARGUMENTS, "Cannot convert Protobuf schema to ClickHouse table schema, all fields have unsupported types");
return schema;
}
}
#endif