2017-06-05 09:02:05 +00:00
|
|
|
#include <stack>
|
2017-06-06 17:18:32 +00:00
|
|
|
#include <ext/map.h>
|
|
|
|
#include <ext/range.h>
|
2017-05-12 21:23:12 +00:00
|
|
|
#include <Poco/Net/IPAddress.h>
|
|
|
|
#include <Poco/ByteOrder.h>
|
|
|
|
#include <Dictionaries/TrieDictionary.h>
|
2017-06-05 09:02:05 +00:00
|
|
|
#include <Columns/ColumnVector.h>
|
|
|
|
#include <Columns/ColumnFixedString.h>
|
|
|
|
#include <Dictionaries/DictionaryBlockInputStream.h>
|
|
|
|
#include <DataTypes/DataTypeFixedString.h>
|
2017-06-13 05:14:24 +00:00
|
|
|
#include <DataTypes/DataTypeString.h>
|
2017-06-05 09:02:05 +00:00
|
|
|
#include <IO/WriteIntText.h>
|
2017-06-13 05:14:24 +00:00
|
|
|
#include <Common/formatIPv6.h>
|
2017-05-12 21:23:12 +00:00
|
|
|
#include <iostream>
|
2017-06-23 20:22:35 +00:00
|
|
|
#include <btrie.h>
|
2017-05-12 21:23:12 +00:00
|
|
|
|
2017-06-13 04:45:30 +00:00
|
|
|
|
2017-05-12 21:23:12 +00:00
|
|
|
namespace DB
|
|
|
|
{
|
|
|
|
|
|
|
|
namespace ErrorCodes
|
|
|
|
{
|
|
|
|
extern const int TYPE_MISMATCH;
|
|
|
|
extern const int ARGUMENT_OUT_OF_BOUND;
|
|
|
|
extern const int BAD_ARGUMENTS;
|
|
|
|
extern const int DICTIONARY_IS_EMPTY;
|
2017-09-01 17:21:03 +00:00
|
|
|
extern const int NOT_IMPLEMENTED;
|
2017-05-12 21:23:12 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
TrieDictionary::TrieDictionary(
|
|
|
|
const std::string & name, const DictionaryStructure & dict_struct, DictionarySourcePtr source_ptr,
|
|
|
|
const DictionaryLifetime dict_lifetime, bool require_nonempty)
|
|
|
|
: name{name}, dict_struct(dict_struct), source_ptr{std::move(source_ptr)}, dict_lifetime(dict_lifetime),
|
2017-06-05 09:02:05 +00:00
|
|
|
require_nonempty(require_nonempty), logger(&Poco::Logger::get("TrieDictionary"))
|
2017-05-12 21:23:12 +00:00
|
|
|
{
|
|
|
|
createAttributes();
|
|
|
|
trie = btrie_create();
|
|
|
|
|
|
|
|
try
|
|
|
|
{
|
|
|
|
loadData();
|
|
|
|
calculateBytesAllocated();
|
|
|
|
}
|
|
|
|
catch (...)
|
|
|
|
{
|
|
|
|
creation_exception = std::current_exception();
|
|
|
|
}
|
|
|
|
|
|
|
|
creation_time = std::chrono::system_clock::now();
|
|
|
|
}
|
|
|
|
|
|
|
|
TrieDictionary::TrieDictionary(const TrieDictionary & other)
|
|
|
|
: TrieDictionary{other.name, other.dict_struct, other.source_ptr->clone(), other.dict_lifetime, other.require_nonempty}
|
|
|
|
{
|
|
|
|
}
|
|
|
|
|
|
|
|
TrieDictionary::~TrieDictionary()
|
|
|
|
{
|
|
|
|
btrie_destroy(trie);
|
|
|
|
}
|
|
|
|
|
|
|
|
#define DECLARE(TYPE)\
|
|
|
|
void TrieDictionary::get##TYPE(\
|
2017-05-25 19:52:05 +00:00
|
|
|
const std::string & attribute_name, const Columns & key_columns, const DataTypes & key_types,\
|
2017-05-12 21:23:12 +00:00
|
|
|
PaddedPODArray<TYPE> & out) const\
|
|
|
|
{\
|
|
|
|
validateKeyTypes(key_types);\
|
|
|
|
\
|
|
|
|
const auto & attribute = getAttribute(attribute_name);\
|
|
|
|
if (!isAttributeTypeConvertibleTo(attribute.type, AttributeUnderlyingType::TYPE))\
|
|
|
|
throw Exception{\
|
|
|
|
name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type),\
|
|
|
|
ErrorCodes::TYPE_MISMATCH};\
|
|
|
|
\
|
|
|
|
const auto null_value = std::get<TYPE>(attribute.null_values);\
|
|
|
|
\
|
|
|
|
getItemsNumber<TYPE>(attribute, key_columns,\
|
2017-07-21 06:35:58 +00:00
|
|
|
[&] (const size_t row, const auto value) { out[row] = value; },\
|
|
|
|
[&] (const size_t) { return null_value; });\
|
2017-05-12 21:23:12 +00:00
|
|
|
}
|
|
|
|
DECLARE(UInt8)
|
|
|
|
DECLARE(UInt16)
|
|
|
|
DECLARE(UInt32)
|
|
|
|
DECLARE(UInt64)
|
2017-11-14 00:08:54 +00:00
|
|
|
DECLARE(UInt128)
|
2017-05-12 21:23:12 +00:00
|
|
|
DECLARE(Int8)
|
|
|
|
DECLARE(Int16)
|
|
|
|
DECLARE(Int32)
|
|
|
|
DECLARE(Int64)
|
|
|
|
DECLARE(Float32)
|
|
|
|
DECLARE(Float64)
|
|
|
|
#undef DECLARE
|
|
|
|
|
|
|
|
void TrieDictionary::getString(
|
2017-05-25 19:52:05 +00:00
|
|
|
const std::string & attribute_name, const Columns & key_columns, const DataTypes & key_types,
|
2017-05-12 21:23:12 +00:00
|
|
|
ColumnString * out) const
|
|
|
|
{
|
|
|
|
validateKeyTypes(key_types);
|
|
|
|
|
|
|
|
const auto & attribute = getAttribute(attribute_name);
|
|
|
|
if (!isAttributeTypeConvertibleTo(attribute.type, AttributeUnderlyingType::String))
|
|
|
|
throw Exception{
|
|
|
|
name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type),
|
|
|
|
ErrorCodes::TYPE_MISMATCH};
|
|
|
|
|
|
|
|
const auto & null_value = StringRef{std::get<String>(attribute.null_values)};
|
|
|
|
|
|
|
|
getItemsImpl<StringRef, StringRef>(attribute, key_columns,
|
2017-12-01 20:21:35 +00:00
|
|
|
[&] (const size_t, const StringRef value) { out->insertData(value.data, value.size); },
|
2017-07-21 06:35:58 +00:00
|
|
|
[&] (const size_t) { return null_value; });
|
2017-05-12 21:23:12 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
#define DECLARE(TYPE)\
|
|
|
|
void TrieDictionary::get##TYPE(\
|
2017-05-25 19:52:05 +00:00
|
|
|
const std::string & attribute_name, const Columns & key_columns, const DataTypes & key_types,\
|
2017-05-12 21:23:12 +00:00
|
|
|
const PaddedPODArray<TYPE> & def, PaddedPODArray<TYPE> & out) const\
|
|
|
|
{\
|
|
|
|
validateKeyTypes(key_types);\
|
|
|
|
\
|
|
|
|
const auto & attribute = getAttribute(attribute_name);\
|
|
|
|
if (!isAttributeTypeConvertibleTo(attribute.type, AttributeUnderlyingType::TYPE))\
|
|
|
|
throw Exception{\
|
|
|
|
name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type),\
|
|
|
|
ErrorCodes::TYPE_MISMATCH};\
|
|
|
|
\
|
|
|
|
getItemsNumber<TYPE>(attribute, key_columns,\
|
2017-07-21 06:35:58 +00:00
|
|
|
[&] (const size_t row, const auto value) { out[row] = value; },\
|
|
|
|
[&] (const size_t row) { return def[row]; });\
|
2017-05-12 21:23:12 +00:00
|
|
|
}
|
|
|
|
DECLARE(UInt8)
|
|
|
|
DECLARE(UInt16)
|
|
|
|
DECLARE(UInt32)
|
|
|
|
DECLARE(UInt64)
|
2017-11-14 00:08:54 +00:00
|
|
|
DECLARE(UInt128)
|
2017-05-12 21:23:12 +00:00
|
|
|
DECLARE(Int8)
|
|
|
|
DECLARE(Int16)
|
|
|
|
DECLARE(Int32)
|
|
|
|
DECLARE(Int64)
|
|
|
|
DECLARE(Float32)
|
|
|
|
DECLARE(Float64)
|
|
|
|
#undef DECLARE
|
|
|
|
|
|
|
|
void TrieDictionary::getString(
|
2017-05-25 19:52:05 +00:00
|
|
|
const std::string & attribute_name, const Columns & key_columns, const DataTypes & key_types,
|
2017-05-12 21:23:12 +00:00
|
|
|
const ColumnString * const def, ColumnString * const out) const
|
|
|
|
{
|
|
|
|
validateKeyTypes(key_types);
|
|
|
|
|
|
|
|
const auto & attribute = getAttribute(attribute_name);
|
|
|
|
if (!isAttributeTypeConvertibleTo(attribute.type, AttributeUnderlyingType::String))
|
|
|
|
throw Exception{
|
|
|
|
name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type),
|
|
|
|
ErrorCodes::TYPE_MISMATCH};
|
|
|
|
|
|
|
|
getItemsImpl<StringRef, StringRef>(attribute, key_columns,
|
2017-12-01 20:21:35 +00:00
|
|
|
[&] (const size_t, const StringRef value) { out->insertData(value.data, value.size); },
|
2017-07-21 06:35:58 +00:00
|
|
|
[&] (const size_t row) { return def->getDataAt(row); });
|
2017-05-12 21:23:12 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
#define DECLARE(TYPE)\
|
|
|
|
void TrieDictionary::get##TYPE(\
|
2017-05-25 19:52:05 +00:00
|
|
|
const std::string & attribute_name, const Columns & key_columns, const DataTypes & key_types,\
|
2017-05-12 21:23:12 +00:00
|
|
|
const TYPE def, PaddedPODArray<TYPE> & out) const\
|
|
|
|
{\
|
|
|
|
validateKeyTypes(key_types);\
|
|
|
|
\
|
|
|
|
const auto & attribute = getAttribute(attribute_name);\
|
|
|
|
if (!isAttributeTypeConvertibleTo(attribute.type, AttributeUnderlyingType::TYPE))\
|
|
|
|
throw Exception{\
|
|
|
|
name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type),\
|
|
|
|
ErrorCodes::TYPE_MISMATCH};\
|
|
|
|
\
|
|
|
|
getItemsNumber<TYPE>(attribute, key_columns,\
|
2017-07-21 06:35:58 +00:00
|
|
|
[&] (const size_t row, const auto value) { out[row] = value; },\
|
|
|
|
[&] (const size_t) { return def; });\
|
2017-05-12 21:23:12 +00:00
|
|
|
}
|
|
|
|
DECLARE(UInt8)
|
|
|
|
DECLARE(UInt16)
|
|
|
|
DECLARE(UInt32)
|
|
|
|
DECLARE(UInt64)
|
2017-11-14 00:08:54 +00:00
|
|
|
DECLARE(UInt128)
|
2017-05-12 21:23:12 +00:00
|
|
|
DECLARE(Int8)
|
|
|
|
DECLARE(Int16)
|
|
|
|
DECLARE(Int32)
|
|
|
|
DECLARE(Int64)
|
|
|
|
DECLARE(Float32)
|
|
|
|
DECLARE(Float64)
|
|
|
|
#undef DECLARE
|
|
|
|
|
|
|
|
void TrieDictionary::getString(
|
2017-05-25 19:52:05 +00:00
|
|
|
const std::string & attribute_name, const Columns & key_columns, const DataTypes & key_types,
|
2017-05-12 21:23:12 +00:00
|
|
|
const String & def, ColumnString * const out) const
|
|
|
|
{
|
|
|
|
validateKeyTypes(key_types);
|
|
|
|
|
|
|
|
const auto & attribute = getAttribute(attribute_name);
|
|
|
|
if (!isAttributeTypeConvertibleTo(attribute.type, AttributeUnderlyingType::String))
|
|
|
|
throw Exception{
|
|
|
|
name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type),
|
|
|
|
ErrorCodes::TYPE_MISMATCH};
|
|
|
|
|
|
|
|
getItemsImpl<StringRef, StringRef>(attribute, key_columns,
|
2017-12-01 20:21:35 +00:00
|
|
|
[&] (const size_t, const StringRef value) { out->insertData(value.data, value.size); },
|
2017-07-21 06:35:58 +00:00
|
|
|
[&] (const size_t) { return StringRef{def}; });
|
2017-05-12 21:23:12 +00:00
|
|
|
}
|
|
|
|
|
2017-05-25 19:52:05 +00:00
|
|
|
void TrieDictionary::has(const Columns & key_columns, const DataTypes & key_types, PaddedPODArray<UInt8> & out) const
|
2017-05-12 21:23:12 +00:00
|
|
|
{
|
|
|
|
validateKeyTypes(key_types);
|
|
|
|
|
|
|
|
const auto & attribute = attributes.front();
|
|
|
|
|
|
|
|
switch (attribute.type)
|
|
|
|
{
|
|
|
|
case AttributeUnderlyingType::UInt8: has<UInt8>(attribute, key_columns, out); break;
|
|
|
|
case AttributeUnderlyingType::UInt16: has<UInt16>(attribute, key_columns, out); break;
|
|
|
|
case AttributeUnderlyingType::UInt32: has<UInt32>(attribute, key_columns, out); break;
|
|
|
|
case AttributeUnderlyingType::UInt64: has<UInt64>(attribute, key_columns, out); break;
|
2017-11-14 00:08:54 +00:00
|
|
|
case AttributeUnderlyingType::UInt128: has<UInt128>(attribute, key_columns, out); break;
|
2017-05-12 21:23:12 +00:00
|
|
|
case AttributeUnderlyingType::Int8: has<Int8>(attribute, key_columns, out); break;
|
|
|
|
case AttributeUnderlyingType::Int16: has<Int16>(attribute, key_columns, out); break;
|
|
|
|
case AttributeUnderlyingType::Int32: has<Int32>(attribute, key_columns, out); break;
|
|
|
|
case AttributeUnderlyingType::Int64: has<Int64>(attribute, key_columns, out); break;
|
|
|
|
case AttributeUnderlyingType::Float32: has<Float32>(attribute, key_columns, out); break;
|
|
|
|
case AttributeUnderlyingType::Float64: has<Float64>(attribute, key_columns, out); break;
|
|
|
|
case AttributeUnderlyingType::String: has<StringRef>(attribute, key_columns, out); break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void TrieDictionary::createAttributes()
|
|
|
|
{
|
|
|
|
const auto size = dict_struct.attributes.size();
|
|
|
|
attributes.reserve(size);
|
|
|
|
|
|
|
|
for (const auto & attribute : dict_struct.attributes)
|
|
|
|
{
|
|
|
|
attribute_index_by_name.emplace(attribute.name, attributes.size());
|
|
|
|
attributes.push_back(createAttributeWithType(attribute.underlying_type, attribute.null_value));
|
|
|
|
|
|
|
|
if (attribute.hierarchical)
|
|
|
|
throw Exception{
|
|
|
|
name + ": hierarchical attributes not supported for dictionary of type " + getTypeName(),
|
|
|
|
ErrorCodes::TYPE_MISMATCH};
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void TrieDictionary::loadData()
|
|
|
|
{
|
|
|
|
auto stream = source_ptr->loadAll();
|
|
|
|
stream->readPrefix();
|
|
|
|
|
|
|
|
/// created upfront to avoid excess allocations
|
2017-10-04 00:22:00 +00:00
|
|
|
const auto keys_size = dict_struct.key->size();
|
2017-05-12 21:23:12 +00:00
|
|
|
StringRefs keys(keys_size);
|
|
|
|
|
|
|
|
const auto attributes_size = attributes.size();
|
|
|
|
|
|
|
|
while (const auto block = stream->read())
|
|
|
|
{
|
|
|
|
const auto rows = block.rows();
|
|
|
|
element_count += rows;
|
|
|
|
|
2017-05-25 19:52:05 +00:00
|
|
|
const auto key_column_ptrs = ext::map<Columns>(ext::range(0, keys_size),
|
2017-07-21 06:35:58 +00:00
|
|
|
[&] (const size_t attribute_idx)
|
2017-05-25 20:09:48 +00:00
|
|
|
{
|
|
|
|
return block.safeGetByPosition(attribute_idx).column;
|
2017-05-12 21:23:12 +00:00
|
|
|
});
|
|
|
|
|
2017-05-25 19:52:05 +00:00
|
|
|
const auto attribute_column_ptrs = ext::map<Columns>(ext::range(0, attributes_size),
|
2017-07-21 06:35:58 +00:00
|
|
|
[&] (const size_t attribute_idx)
|
2017-05-25 20:09:48 +00:00
|
|
|
{
|
|
|
|
return block.safeGetByPosition(keys_size + attribute_idx).column;
|
2017-05-12 21:23:12 +00:00
|
|
|
});
|
|
|
|
|
|
|
|
for (const auto row_idx : ext::range(0, rows))
|
|
|
|
{
|
|
|
|
/// calculate key once per row
|
|
|
|
const auto key_column = key_column_ptrs.front();
|
|
|
|
|
|
|
|
for (const auto attribute_idx : ext::range(0, attributes_size))
|
|
|
|
{
|
|
|
|
const auto & attribute_column = *attribute_column_ptrs[attribute_idx];
|
|
|
|
auto & attribute = attributes[attribute_idx];
|
|
|
|
setAttributeValue(attribute, key_column->getDataAt(row_idx), attribute_column[row_idx]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
stream->readSuffix();
|
|
|
|
|
|
|
|
if (require_nonempty && 0 == element_count)
|
|
|
|
throw Exception{
|
|
|
|
name + ": dictionary source is empty and 'require_nonempty' property is set.",
|
|
|
|
ErrorCodes::DICTIONARY_IS_EMPTY};
|
|
|
|
}
|
|
|
|
|
|
|
|
template <typename T>
|
|
|
|
void TrieDictionary::addAttributeSize(const Attribute & attribute)
|
|
|
|
{
|
|
|
|
const auto & vec = *std::get<ContainerPtrType<T>>(attribute.maps);
|
|
|
|
bytes_allocated += sizeof(ContainerType<T>) + (vec.capacity() * sizeof(T));
|
|
|
|
bucket_count = vec.size();
|
|
|
|
}
|
|
|
|
|
|
|
|
void TrieDictionary::calculateBytesAllocated()
|
|
|
|
{
|
|
|
|
bytes_allocated += attributes.size() * sizeof(attributes.front());
|
|
|
|
|
|
|
|
for (const auto & attribute : attributes)
|
|
|
|
{
|
|
|
|
switch (attribute.type)
|
|
|
|
{
|
|
|
|
case AttributeUnderlyingType::UInt8: addAttributeSize<UInt8>(attribute); break;
|
|
|
|
case AttributeUnderlyingType::UInt16: addAttributeSize<UInt16>(attribute); break;
|
|
|
|
case AttributeUnderlyingType::UInt32: addAttributeSize<UInt32>(attribute); break;
|
|
|
|
case AttributeUnderlyingType::UInt64: addAttributeSize<UInt64>(attribute); break;
|
2017-11-14 00:08:54 +00:00
|
|
|
case AttributeUnderlyingType::UInt128: addAttributeSize<UInt128>(attribute); break;
|
2017-05-12 21:23:12 +00:00
|
|
|
case AttributeUnderlyingType::Int8: addAttributeSize<Int8>(attribute); break;
|
|
|
|
case AttributeUnderlyingType::Int16: addAttributeSize<Int16>(attribute); break;
|
|
|
|
case AttributeUnderlyingType::Int32: addAttributeSize<Int32>(attribute); break;
|
|
|
|
case AttributeUnderlyingType::Int64: addAttributeSize<Int64>(attribute); break;
|
|
|
|
case AttributeUnderlyingType::Float32: addAttributeSize<Float32>(attribute); break;
|
|
|
|
case AttributeUnderlyingType::Float64: addAttributeSize<Float64>(attribute); break;
|
|
|
|
case AttributeUnderlyingType::String:
|
|
|
|
{
|
|
|
|
addAttributeSize<StringRef>(attribute);
|
|
|
|
bytes_allocated += sizeof(Arena) + attribute.string_arena->size();
|
|
|
|
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
bytes_allocated += btrie_allocated(trie);
|
|
|
|
}
|
|
|
|
|
|
|
|
void TrieDictionary::validateKeyTypes(const DataTypes & key_types) const
|
|
|
|
{
|
|
|
|
if (key_types.size() != 1)
|
|
|
|
throw Exception{
|
|
|
|
"Expected a single IP address",
|
|
|
|
ErrorCodes::TYPE_MISMATCH};
|
|
|
|
|
|
|
|
const auto & actual_type = key_types[0]->getName();
|
|
|
|
|
|
|
|
if (actual_type != "UInt32" && actual_type != "FixedString(16)")
|
|
|
|
throw Exception{
|
|
|
|
"Key does not match, expected either UInt32 or FixedString(16)",
|
|
|
|
ErrorCodes::TYPE_MISMATCH};
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
template <typename T>
|
|
|
|
void TrieDictionary::createAttributeImpl(Attribute & attribute, const Field & null_value)
|
|
|
|
{
|
|
|
|
std::get<T>(attribute.null_values) = null_value.get<typename NearestFieldType<T>::Type>();
|
|
|
|
std::get<ContainerPtrType<T>>(attribute.maps) = std::make_unique<ContainerType<T>>();
|
|
|
|
}
|
|
|
|
|
|
|
|
TrieDictionary::Attribute TrieDictionary::createAttributeWithType(const AttributeUnderlyingType type, const Field & null_value)
|
|
|
|
{
|
2017-12-01 20:21:35 +00:00
|
|
|
Attribute attr{type, {}, {}, {}};
|
2017-05-12 21:23:12 +00:00
|
|
|
|
|
|
|
switch (type)
|
|
|
|
{
|
|
|
|
case AttributeUnderlyingType::UInt8: createAttributeImpl<UInt8>(attr, null_value); break;
|
|
|
|
case AttributeUnderlyingType::UInt16: createAttributeImpl<UInt16>(attr, null_value); break;
|
|
|
|
case AttributeUnderlyingType::UInt32: createAttributeImpl<UInt32>(attr, null_value); break;
|
|
|
|
case AttributeUnderlyingType::UInt64: createAttributeImpl<UInt64>(attr, null_value); break;
|
2017-11-14 00:08:54 +00:00
|
|
|
case AttributeUnderlyingType::UInt128: createAttributeImpl<UInt128>(attr, null_value); break;
|
2017-05-12 21:23:12 +00:00
|
|
|
case AttributeUnderlyingType::Int8: createAttributeImpl<Int8>(attr, null_value); break;
|
|
|
|
case AttributeUnderlyingType::Int16: createAttributeImpl<Int16>(attr, null_value); break;
|
|
|
|
case AttributeUnderlyingType::Int32: createAttributeImpl<Int32>(attr, null_value); break;
|
|
|
|
case AttributeUnderlyingType::Int64: createAttributeImpl<Int64>(attr, null_value); break;
|
|
|
|
case AttributeUnderlyingType::Float32: createAttributeImpl<Float32>(attr, null_value); break;
|
|
|
|
case AttributeUnderlyingType::Float64: createAttributeImpl<Float64>(attr, null_value); break;
|
|
|
|
case AttributeUnderlyingType::String:
|
|
|
|
{
|
|
|
|
std::get<String>(attr.null_values) = null_value.get<String>();
|
|
|
|
std::get<ContainerPtrType<StringRef>>(attr.maps) = std::make_unique<ContainerType<StringRef>>();
|
|
|
|
attr.string_arena = std::make_unique<Arena>();
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return attr;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
template <typename OutputType, typename ValueSetter, typename DefaultGetter>
|
|
|
|
void TrieDictionary::getItemsNumber(
|
|
|
|
const Attribute & attribute,
|
2017-05-25 19:52:05 +00:00
|
|
|
const Columns & key_columns,
|
2017-05-12 21:23:12 +00:00
|
|
|
ValueSetter && set_value,
|
|
|
|
DefaultGetter && get_default) const
|
|
|
|
{
|
|
|
|
if (false) {}
|
|
|
|
#define DISPATCH(TYPE) \
|
|
|
|
else if (attribute.type == AttributeUnderlyingType::TYPE) \
|
|
|
|
getItemsImpl<TYPE, OutputType>(attribute, key_columns, std::forward<ValueSetter>(set_value), std::forward<DefaultGetter>(get_default));
|
|
|
|
DISPATCH(UInt8)
|
|
|
|
DISPATCH(UInt16)
|
|
|
|
DISPATCH(UInt32)
|
|
|
|
DISPATCH(UInt64)
|
2017-11-14 00:08:54 +00:00
|
|
|
DISPATCH(UInt128)
|
2017-05-12 21:23:12 +00:00
|
|
|
DISPATCH(Int8)
|
|
|
|
DISPATCH(Int16)
|
|
|
|
DISPATCH(Int32)
|
|
|
|
DISPATCH(Int64)
|
|
|
|
DISPATCH(Float32)
|
|
|
|
DISPATCH(Float64)
|
|
|
|
#undef DISPATCH
|
|
|
|
else
|
|
|
|
throw Exception("Unexpected type of attribute: " + toString(attribute.type), ErrorCodes::LOGICAL_ERROR);
|
|
|
|
}
|
|
|
|
|
|
|
|
template <typename AttributeType, typename OutputType, typename ValueSetter, typename DefaultGetter>
|
|
|
|
void TrieDictionary::getItemsImpl(
|
|
|
|
const Attribute & attribute,
|
2017-05-25 19:52:05 +00:00
|
|
|
const Columns & key_columns,
|
2017-05-12 21:23:12 +00:00
|
|
|
ValueSetter && set_value,
|
|
|
|
DefaultGetter && get_default) const
|
|
|
|
{
|
|
|
|
auto & vec = *std::get<ContainerPtrType<AttributeType>>(attribute.maps);
|
|
|
|
|
|
|
|
const auto first_column = key_columns.front();
|
|
|
|
const auto rows = first_column->size();
|
|
|
|
if (first_column->isNumeric())
|
|
|
|
{
|
|
|
|
for (const auto i : ext::range(0, rows))
|
|
|
|
{
|
|
|
|
auto addr = Int32(first_column->get64(i));
|
|
|
|
uintptr_t slot = btrie_find(trie, addr);
|
2017-11-14 00:08:54 +00:00
|
|
|
set_value(i, slot != BTRIE_NULL ? static_cast<OutputType>(vec[slot]) : get_default(i));
|
2017-05-12 21:23:12 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
for (const auto i : ext::range(0, rows))
|
|
|
|
{
|
|
|
|
auto addr = first_column->getDataAt(i);
|
|
|
|
if (addr.size != 16)
|
|
|
|
throw Exception("Expected key to be FixedString(16)", ErrorCodes::LOGICAL_ERROR);
|
2017-06-05 09:02:05 +00:00
|
|
|
|
2017-05-12 21:23:12 +00:00
|
|
|
uintptr_t slot = btrie_find_a6(trie, reinterpret_cast<const UInt8*>(addr.data));
|
2017-11-14 00:08:54 +00:00
|
|
|
set_value(i, slot != BTRIE_NULL ? static_cast<OutputType>(vec[slot]) : get_default(i));
|
2017-05-12 21:23:12 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
query_count.fetch_add(rows, std::memory_order_relaxed);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
template <typename T>
|
|
|
|
bool TrieDictionary::setAttributeValueImpl(Attribute & attribute, const StringRef key, const T value)
|
|
|
|
{
|
|
|
|
// Insert value into appropriate vector type
|
|
|
|
auto & vec = *std::get<ContainerPtrType<T>>(attribute.maps);
|
|
|
|
size_t row = vec.size();
|
|
|
|
vec.push_back(value);
|
|
|
|
|
|
|
|
// Parse IP address and subnet length from string (e.g. 2a02:6b8::3/64)
|
|
|
|
Poco::Net::IPAddress addr, mask;
|
|
|
|
std::string addr_str(key.toString());
|
|
|
|
size_t pos = addr_str.find('/');
|
|
|
|
if (pos != std::string::npos)
|
|
|
|
{
|
|
|
|
|
|
|
|
addr = Poco::Net::IPAddress(addr_str.substr(0, pos));
|
|
|
|
mask = Poco::Net::IPAddress(std::stoi(addr_str.substr(pos + 1), nullptr, 10), addr.family());
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
addr = Poco::Net::IPAddress(addr_str);
|
|
|
|
mask = Poco::Net::IPAddress(addr.length() * 8, addr.family());
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Here we might overwrite the same key with the same slot as each key can map to multiple attributes.
|
|
|
|
* However, all columns have equal number of rows so it is okay to store only row number for each key
|
|
|
|
* instead of building a trie for each column. This comes at the cost of additional lookup in attribute
|
|
|
|
* vector on lookup time to return cell from row + column. The reason for this is to save space,
|
|
|
|
* and build only single trie instead of trie for each column.
|
|
|
|
*/
|
|
|
|
if (addr.family() == Poco::Net::IPAddress::IPv4)
|
|
|
|
{
|
|
|
|
UInt32 addr_v4 = Poco::ByteOrder::toNetwork(*reinterpret_cast<const UInt32*>(addr.addr()));
|
|
|
|
UInt32 mask_v4 = Poco::ByteOrder::toNetwork(*reinterpret_cast<const UInt32*>(mask.addr()));
|
|
|
|
return btrie_insert(trie, addr_v4, mask_v4, row) == 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
const uint8_t* addr_v6 = reinterpret_cast<const uint8_t*>(addr.addr());
|
|
|
|
const uint8_t* mask_v6 = reinterpret_cast<const uint8_t*>(mask.addr());
|
|
|
|
return btrie_insert_a6(trie, addr_v6, mask_v6, row) == 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool TrieDictionary::setAttributeValue(Attribute & attribute, const StringRef key, const Field & value)
|
|
|
|
{
|
|
|
|
switch (attribute.type)
|
|
|
|
{
|
|
|
|
case AttributeUnderlyingType::UInt8: return setAttributeValueImpl<UInt8>(attribute, key, value.get<UInt64>());
|
|
|
|
case AttributeUnderlyingType::UInt16: return setAttributeValueImpl<UInt16>(attribute, key, value.get<UInt64>());
|
|
|
|
case AttributeUnderlyingType::UInt32: return setAttributeValueImpl<UInt32>(attribute, key, value.get<UInt64>());
|
|
|
|
case AttributeUnderlyingType::UInt64: return setAttributeValueImpl<UInt64>(attribute, key, value.get<UInt64>());
|
2017-11-14 00:08:54 +00:00
|
|
|
case AttributeUnderlyingType::UInt128: return setAttributeValueImpl<UInt128>(attribute, key, value.get<UInt128>());
|
2017-05-12 21:23:12 +00:00
|
|
|
case AttributeUnderlyingType::Int8: return setAttributeValueImpl<Int8>(attribute, key, value.get<Int64>());
|
|
|
|
case AttributeUnderlyingType::Int16: return setAttributeValueImpl<Int16>(attribute, key, value.get<Int64>());
|
|
|
|
case AttributeUnderlyingType::Int32: return setAttributeValueImpl<Int32>(attribute, key, value.get<Int64>());
|
|
|
|
case AttributeUnderlyingType::Int64: return setAttributeValueImpl<Int64>(attribute, key, value.get<Int64>());
|
|
|
|
case AttributeUnderlyingType::Float32: return setAttributeValueImpl<Float32>(attribute, key, value.get<Float64>());
|
|
|
|
case AttributeUnderlyingType::Float64: return setAttributeValueImpl<Float64>(attribute, key, value.get<Float64>());
|
|
|
|
case AttributeUnderlyingType::String:
|
|
|
|
{
|
|
|
|
const auto & string = value.get<String>();
|
|
|
|
const auto string_in_arena = attribute.string_arena->insert(string.data(), string.size());
|
|
|
|
setAttributeValueImpl<StringRef>(attribute, key, StringRef{string_in_arena, string.size()});
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return {};
|
|
|
|
}
|
|
|
|
|
|
|
|
const TrieDictionary::Attribute & TrieDictionary::getAttribute(const std::string & attribute_name) const
|
|
|
|
{
|
|
|
|
const auto it = attribute_index_by_name.find(attribute_name);
|
|
|
|
if (it == std::end(attribute_index_by_name))
|
|
|
|
throw Exception{
|
|
|
|
name + ": no such attribute '" + attribute_name + "'",
|
|
|
|
ErrorCodes::BAD_ARGUMENTS};
|
|
|
|
|
|
|
|
return attributes[it->second];
|
|
|
|
}
|
|
|
|
|
|
|
|
template <typename T>
|
2017-12-01 20:21:35 +00:00
|
|
|
void TrieDictionary::has(const Attribute &, const Columns & key_columns, PaddedPODArray<UInt8> & out) const
|
2017-05-12 21:23:12 +00:00
|
|
|
{
|
|
|
|
const auto first_column = key_columns.front();
|
|
|
|
const auto rows = first_column->size();
|
|
|
|
if (first_column->isNumeric())
|
|
|
|
{
|
|
|
|
for (const auto i : ext::range(0, rows))
|
|
|
|
{
|
|
|
|
auto addr = Int32(first_column->get64(i));
|
|
|
|
uintptr_t slot = btrie_find(trie, addr);
|
|
|
|
out[i] = (slot != BTRIE_NULL);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
for (const auto i : ext::range(0, rows))
|
|
|
|
{
|
|
|
|
auto addr = first_column->getDataAt(i);
|
|
|
|
if (unlikely(addr.size != 16))
|
|
|
|
throw Exception("Expected key to be FixedString(16)", ErrorCodes::LOGICAL_ERROR);
|
2017-06-05 09:02:05 +00:00
|
|
|
|
2017-05-12 21:23:12 +00:00
|
|
|
uintptr_t slot = btrie_find_a6(trie, reinterpret_cast<const UInt8*>(addr.data));
|
|
|
|
out[i] = (slot != BTRIE_NULL);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-06-05 09:02:05 +00:00
|
|
|
query_count.fetch_add(rows, std::memory_order_relaxed);
|
|
|
|
}
|
|
|
|
|
|
|
|
template <typename Getter, typename KeyType>
|
|
|
|
void TrieDictionary::trieTraverse(const btrie_t * tree, Getter && getter) const
|
|
|
|
{
|
|
|
|
KeyType key = 0;
|
|
|
|
const KeyType high_bit = ~((~key) >> 1);
|
|
|
|
|
|
|
|
btrie_node_t * node;
|
|
|
|
node = tree->root;
|
|
|
|
|
|
|
|
std::stack<btrie_node_t *> stack;
|
|
|
|
while (node)
|
|
|
|
{
|
|
|
|
stack.push(node);
|
|
|
|
node = node->left;
|
|
|
|
}
|
|
|
|
|
|
|
|
auto getBit = [&high_bit](size_t size) { return size ? (high_bit >> (size - 1)) : 0; };
|
|
|
|
|
|
|
|
while (!stack.empty())
|
|
|
|
{
|
|
|
|
node = stack.top();
|
|
|
|
stack.pop();
|
|
|
|
|
|
|
|
if (node && node->value != BTRIE_NULL)
|
|
|
|
getter(key, stack.size());
|
|
|
|
|
|
|
|
if (node && node->right)
|
|
|
|
{
|
|
|
|
stack.push(NULL);
|
|
|
|
key |= getBit(stack.size());
|
|
|
|
stack.push(node->right);
|
|
|
|
while (stack.top()->left)
|
|
|
|
stack.push(stack.top()->left);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
key &= ~getBit(stack.size());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
Columns TrieDictionary::getKeyColumns() const
|
|
|
|
{
|
2017-12-14 01:43:19 +00:00
|
|
|
auto ip_column = ColumnFixedString::create(IPV6_BINARY_LENGTH);
|
|
|
|
auto mask_column = ColumnVector<UInt8>::create();
|
2017-06-05 09:02:05 +00:00
|
|
|
|
2017-09-01 17:21:03 +00:00
|
|
|
#if defined(__SIZEOF_INT128__)
|
2018-01-10 00:04:08 +00:00
|
|
|
auto getter = [& ip_column, & mask_column](__uint128_t ip, size_t mask)
|
|
|
|
{
|
2017-06-05 09:02:05 +00:00
|
|
|
UInt64 * ip_array = reinterpret_cast<UInt64 *>(&ip);
|
|
|
|
ip_array[0] = Poco::ByteOrder::fromNetwork(ip_array[0]);
|
|
|
|
ip_array[1] = Poco::ByteOrder::fromNetwork(ip_array[1]);
|
|
|
|
std::swap(ip_array[0], ip_array[1]);
|
2017-06-13 05:14:24 +00:00
|
|
|
ip_column->insertData(reinterpret_cast<const char *>(ip_array), IPV6_BINARY_LENGTH);
|
2017-06-05 09:02:05 +00:00
|
|
|
mask_column->insert(static_cast<UInt8>(mask));
|
|
|
|
};
|
|
|
|
|
|
|
|
trieTraverse<decltype(getter), __uint128_t>(trie, std::move(getter));
|
2017-09-01 17:21:03 +00:00
|
|
|
#else
|
|
|
|
throw Exception("TrieDictionary::getKeyColumns is not implemented for 32bit arch", ErrorCodes::NOT_IMPLEMENTED);
|
|
|
|
#endif
|
2017-12-15 03:19:14 +00:00
|
|
|
return {std::move(ip_column), std::move(mask_column)};
|
2017-06-05 09:02:05 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
BlockInputStreamPtr TrieDictionary::getBlockInputStream(const Names & column_names, size_t max_block_size) const
|
|
|
|
{
|
|
|
|
using BlockInputStreamType = DictionaryBlockInputStream<TrieDictionary, UInt64>;
|
|
|
|
|
2017-12-15 03:19:14 +00:00
|
|
|
auto getKeys = [](const Columns & columns, const std::vector<DictionaryAttribute> & attributes)
|
2017-06-05 09:02:05 +00:00
|
|
|
{
|
|
|
|
const auto & attr = attributes.front();
|
|
|
|
return ColumnsWithTypeAndName({ColumnWithTypeAndName(columns.front(),
|
2017-06-13 05:14:24 +00:00
|
|
|
std::make_shared<DataTypeFixedString>(IPV6_BINARY_LENGTH), attr.name)});
|
2017-06-05 09:02:05 +00:00
|
|
|
};
|
2017-12-15 03:19:14 +00:00
|
|
|
auto getView = [](const Columns & columns, const std::vector<DictionaryAttribute> & attributes)
|
2017-06-05 09:02:05 +00:00
|
|
|
{
|
2017-12-14 01:43:19 +00:00
|
|
|
auto column = ColumnString::create();
|
2017-12-15 03:19:14 +00:00
|
|
|
const auto & ip_column = static_cast<const ColumnFixedString &>(*columns.front());
|
|
|
|
const auto & mask_column = static_cast<const ColumnVector<UInt8> &>(*columns.back());
|
2017-06-05 09:02:05 +00:00
|
|
|
char buffer[48];
|
2017-12-15 03:19:14 +00:00
|
|
|
for (size_t row : ext::range(0, ip_column.size()))
|
2017-06-05 09:02:05 +00:00
|
|
|
{
|
2017-12-15 03:19:14 +00:00
|
|
|
UInt8 mask = mask_column.getElement(row);
|
2017-06-05 09:02:05 +00:00
|
|
|
char * ptr = buffer;
|
2017-12-15 03:19:14 +00:00
|
|
|
formatIPv6(reinterpret_cast<const unsigned char *>(ip_column.getDataAt(row).data), ptr);
|
2017-06-05 09:02:05 +00:00
|
|
|
*(ptr - 1) = '/';
|
|
|
|
auto size = detail::writeUIntText(mask, ptr);
|
|
|
|
column->insertData(buffer, size + (ptr - buffer));
|
|
|
|
}
|
2017-12-15 03:19:14 +00:00
|
|
|
return ColumnsWithTypeAndName{ColumnWithTypeAndName(std::move(column), std::make_shared<DataTypeString>(), attributes.front().name)};
|
2017-06-05 09:02:05 +00:00
|
|
|
};
|
|
|
|
return std::make_shared<BlockInputStreamType>(shared_from_this(), max_block_size, getKeyColumns(), column_names,
|
|
|
|
std::move(getKeys), std::move(getView));
|
|
|
|
}
|
2017-05-12 21:23:12 +00:00
|
|
|
|
|
|
|
}
|