ClickHouse/src/Dictionaries/DictionaryHelpers.h

720 lines
25 KiB
C++
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

#pragma once
#include <Common/HashTable/HashMap.h>
#include <Columns/IColumn.h>
#include <Columns/ColumnDecimal.h>
#include <Columns/ColumnString.h>
#include <Columns/ColumnVector.h>
#include <Columns/ColumnArray.h>
#include <Columns/ColumnsNumber.h>
#include <Columns/ColumnNullable.h>
#include <Columns/ColumnSparse.h>
#include <DataTypes/DataTypesDecimal.h>
#include <DataTypes/DataTypeArray.h>
#include <DataTypes/DataTypeNullable.h>
#include <Core/Block.h>
#include <Dictionaries/IDictionary.h>
#include <Dictionaries/DictionaryStructure.h>
#include <Processors/Executors/PullingPipelineExecutor.h>
#include <QueryPipeline/QueryPipelineBuilder.h>
namespace DB
{
namespace ErrorCodes
{
extern const int TYPE_MISMATCH;
extern const int BAD_ARGUMENTS;
}
class Arena;
/** Simple helper for getting default.
* Initialized with default value and default values column.
* If default values column is not null default value is taken from column.
* If default value is null default value is taken from initializer.
*/
class DefaultValueProvider final
{
public:
explicit DefaultValueProvider(Field default_value_, ColumnPtr default_values_column_ = nullptr)
: default_value(std::move(default_value_))
, default_values_column(default_values_column_)
{
}
inline bool isConstant() const { return default_values_column == nullptr; }
Field getDefaultValue(size_t row) const
{
if (default_values_column)
return (*default_values_column)[row];
return default_value;
}
private:
Field default_value;
ColumnPtr default_values_column;
};
/** Support class for dictionary storages.
The main idea is that during fetch we create all columns, but fill only columns that client requested.
We need to create other columns during fetch, because in case of serialized storage we can skip
unnecessary columns serialized in cache with skipSerializedInArena method.
When result is fetched from the storage client of storage can filterOnlyNecessaryColumns
and get only columns that match attributes_names_to_fetch.
*/
class DictionaryStorageFetchRequest
{
public:
DictionaryStorageFetchRequest(const DictionaryStructure & structure,
const Strings & attributes_to_fetch_names,
const DataTypes & attributes_to_fetch_types,
const Columns * const attributes_to_fetch_default_values_columns = nullptr)
: attributes_to_fetch_filter(structure.attributes.size(), false)
{
size_t attributes_to_fetch_size = attributes_to_fetch_names.size();
assert(attributes_to_fetch_size == attributes_to_fetch_types.size());
bool has_default = attributes_to_fetch_default_values_columns;
assert(!has_default || attributes_to_fetch_size == attributes_to_fetch_default_values_columns->size());
for (size_t i = 0; i < attributes_to_fetch_size; ++i)
attributes_to_fetch_name_to_index.emplace(attributes_to_fetch_names[i], i);
if (attributes_to_fetch_name_to_index.size() != attributes_to_fetch_name_to_index.size())
throw Exception(ErrorCodes::BAD_ARGUMENTS, "Attribute names to fetch should be unique");
size_t attributes_size = structure.attributes.size();
dictionary_attributes_names_and_types.reserve(attributes_size);
attributes_default_value_providers.reserve(attributes_size);
for (size_t attribute_index = 0; attribute_index < attributes_size; ++attribute_index)
{
const auto & dictionary_attribute = structure.attributes[attribute_index];
dictionary_attributes_names_and_types.emplace_back(dictionary_attribute.name, dictionary_attribute.type);
auto attribute_to_fetch_index_it = attributes_to_fetch_name_to_index.find(dictionary_attribute.name);
if (attribute_to_fetch_index_it == attributes_to_fetch_name_to_index.end())
{
attributes_default_value_providers.emplace_back(dictionary_attribute.null_value);
continue;
}
attributes_to_fetch_filter[attribute_index] = true;
size_t attributes_to_fetch_index = attribute_to_fetch_index_it->second;
const auto & attribute_to_fetch_result_type = attributes_to_fetch_types[attributes_to_fetch_index];
if (!attribute_to_fetch_result_type->equals(*dictionary_attribute.type))
throw Exception(ErrorCodes::TYPE_MISMATCH,
"Attribute {} type does not match, expected {}, found {}",
dictionary_attribute.name,
attribute_to_fetch_result_type->getName(),
dictionary_attribute.type->getName());
if (has_default)
{
const auto & attribute_to_fetch_default_value_column =
(*attributes_to_fetch_default_values_columns)[attributes_to_fetch_index];
attributes_default_value_providers.emplace_back(dictionary_attribute.null_value,
attribute_to_fetch_default_value_column);
}
}
}
DictionaryStorageFetchRequest() = default;
/// Check requested attributes size
ALWAYS_INLINE size_t attributesSize() const
{
return dictionary_attributes_names_and_types.size();
}
/// Check if attribute with attribute_name was requested to fetch
ALWAYS_INLINE bool containsAttribute(const String & attribute_name) const
{
return attributes_to_fetch_name_to_index.contains(attribute_name);
}
/// Check if attribute with attribute_index should be filled during fetch
ALWAYS_INLINE bool shouldFillResultColumnWithIndex(size_t attribute_index) const
{
return attributes_to_fetch_filter[attribute_index];
}
const DataTypePtr & dataTypeAtIndex(size_t attribute_index) const
{
return dictionary_attributes_names_and_types[attribute_index].type;
}
const DefaultValueProvider & defaultValueProviderAtIndex(size_t attribute_index) const
{
return attributes_default_value_providers[attribute_index];
}
/// Create columns for each of dictionary attributes
MutableColumns makeAttributesResultColumns() const
{
MutableColumns result;
result.reserve(dictionary_attributes_names_and_types.size());
for (const auto & name_and_type : dictionary_attributes_names_and_types)
result.emplace_back(name_and_type.type->createColumn());
return result;
}
Columns makeAttributesResultColumnsNonMutable() const
{
Columns result;
result.reserve(dictionary_attributes_names_and_types.size());
for (const auto & name_and_type : dictionary_attributes_names_and_types)
result.emplace_back(name_and_type.type->createColumn());
return result;
}
/// Filter only requested columns
Columns filterRequestedColumns(MutableColumns & fetched_mutable_columns) const
{
Columns result(attributes_to_fetch_name_to_index.size());
size_t dictionary_attributes_size = dictionary_attributes_names_and_types.size();
for (size_t attribute_index = 0; attribute_index < dictionary_attributes_size; ++attribute_index)
{
if (!shouldFillResultColumnWithIndex(attribute_index))
continue;
const auto & dictionary_attribute_name = dictionary_attributes_names_and_types[attribute_index].name;
size_t fetch_attribute_index = attributes_to_fetch_name_to_index.find(dictionary_attribute_name)->second;
result[fetch_attribute_index] = std::move(fetched_mutable_columns[attribute_index]);
}
return result;
}
private:
NamesAndTypes dictionary_attributes_names_and_types;
std::unordered_map<String, size_t> attributes_to_fetch_name_to_index;
std::vector<bool> attributes_to_fetch_filter;
std::vector<DefaultValueProvider> attributes_default_value_providers;
};
static inline void insertDefaultValuesIntoColumns( /// NOLINT
MutableColumns & columns,
const DictionaryStorageFetchRequest & fetch_request,
size_t row_index)
{
size_t columns_size = columns.size();
for (size_t column_index = 0; column_index < columns_size; ++column_index)
{
const auto & column = columns[column_index];
const auto & default_value_provider = fetch_request.defaultValueProviderAtIndex(column_index);
if (fetch_request.shouldFillResultColumnWithIndex(column_index))
column->insert(default_value_provider.getDefaultValue(row_index));
}
}
/// Deserialize column value and insert it in columns.
/// Skip unnecessary columns that were not requested from deserialization.
static inline void deserializeAndInsertIntoColumns( /// NOLINT
MutableColumns & columns,
const DictionaryStorageFetchRequest & fetch_request,
const char * place_for_serialized_columns)
{
size_t columns_size = columns.size();
for (size_t column_index = 0; column_index < columns_size; ++column_index)
{
const auto & column = columns[column_index];
if (fetch_request.shouldFillResultColumnWithIndex(column_index))
place_for_serialized_columns = column->deserializeAndInsertFromArena(place_for_serialized_columns);
else
place_for_serialized_columns = column->skipSerializedInArena(place_for_serialized_columns);
}
}
/**
* In Dictionaries implementation String attribute is stored in arena and StringRefs are pointing to it.
*/
template <typename DictionaryAttributeType>
using DictionaryValueType =
std::conditional_t<std::is_same_v<DictionaryAttributeType, String>, StringRef, DictionaryAttributeType>;
/**
* Used to create column with right type for DictionaryAttributeType.
*/
template <typename DictionaryAttributeType>
class DictionaryAttributeColumnProvider
{
public:
using ColumnType =
std::conditional_t<std::is_same_v<DictionaryAttributeType, Array>, ColumnArray,
std::conditional_t<std::is_same_v<DictionaryAttributeType, String>, ColumnString,
ColumnVectorOrDecimal<DictionaryAttributeType>>>;
using ColumnPtr = typename ColumnType::MutablePtr;
static ColumnPtr getColumn(const DictionaryAttribute & dictionary_attribute, size_t size)
{
if constexpr (std::is_same_v<DictionaryAttributeType, Array>)
{
if (const auto * array_type = typeid_cast<const DataTypeArray *>(dictionary_attribute.type.get()))
{
auto nested_column = array_type->getNestedType()->createColumn();
return ColumnArray::create(std::move(nested_column));
}
else
{
throw Exception(ErrorCodes::TYPE_MISMATCH, "Unsupported attribute type.");
}
}
if constexpr (std::is_same_v<DictionaryAttributeType, String>)
{
return ColumnType::create();
}
else if constexpr (std::is_same_v<DictionaryAttributeType, UUID>)
{
return ColumnType::create(size);
}
else if constexpr (std::is_same_v<DictionaryAttributeType, IPv4>)
{
return ColumnType::create(size);
}
else if constexpr (std::is_same_v<DictionaryAttributeType, IPv6>)
{
return ColumnType::create(size);
}
else if constexpr (is_decimal<DictionaryAttributeType>)
{
auto nested_type = removeNullable(dictionary_attribute.type);
auto scale = getDecimalScale(*nested_type);
return ColumnType::create(size, scale);
}
else if constexpr (is_arithmetic_v<DictionaryAttributeType>)
{
return ColumnType::create(size);
}
else
throw Exception(ErrorCodes::TYPE_MISMATCH, "Unsupported attribute type.");
}
};
/**
* DictionaryDefaultValueExtractor used to simplify getting default value for IDictionary function `getColumn`.
* Provides interface for getting default value with operator[];
*
* If default_values_column is null then attribute_default_value will be used.
* If default_values_column is not null in constructor than this column values will be used as default values.
*
* For nullable dictionary attribute isNullAt method is provided.
*/
template <typename DictionaryAttributeType>
class DictionaryDefaultValueExtractor
{
using DefaultColumnType = typename DictionaryAttributeColumnProvider<DictionaryAttributeType>::ColumnType;
public:
using DefaultValueType = DictionaryValueType<DictionaryAttributeType>;
explicit DictionaryDefaultValueExtractor(
Field attribute_default_value,
ColumnPtr default_values_column_)
{
if (default_values_column_ != nullptr &&
isColumnConst(*default_values_column_))
{
attribute_default_value = (*default_values_column_)[0];
default_values_column_ = nullptr;
}
if (default_values_column_ == nullptr)
{
use_attribute_default_value = true;
if (attribute_default_value.isNull())
default_value_is_null = true;
else
default_value = static_cast<DictionaryAttributeType>(attribute_default_value.get<DictionaryAttributeType>());
}
else
{
const IColumn * default_values_column_ptr = default_values_column_.get();
if (const ColumnNullable * nullable_column = typeid_cast<const ColumnNullable *>(default_values_column_.get()))
{
default_values_column_ptr = nullable_column->getNestedColumnPtr().get();
is_null_map = &nullable_column->getNullMapColumn();
}
if (const auto * const default_col = checkAndGetColumn<DefaultColumnType>(default_values_column_ptr))
{
default_values_column = default_col;
use_attribute_default_value = false;
}
else
throw Exception(ErrorCodes::TYPE_MISMATCH, "Type of default column is not the same as dictionary attribute type.");
}
}
DefaultValueType operator[](size_t row)
{
if (use_attribute_default_value)
return static_cast<DefaultValueType>(default_value);
assert(default_values_column != nullptr);
if constexpr (std::is_same_v<DefaultColumnType, ColumnArray>)
{
Field field = (*default_values_column)[row];
return field.get<Array>();
}
else if constexpr (std::is_same_v<DefaultColumnType, ColumnString>)
return default_values_column->getDataAt(row);
else
return default_values_column->getData()[row];
}
bool isNullAt(size_t row)
{
if (default_value_is_null)
return true;
if (is_null_map)
return is_null_map->getData()[row];
return false;
}
private:
DictionaryAttributeType default_value {};
const DefaultColumnType * default_values_column = nullptr;
const ColumnUInt8 * is_null_map = nullptr;
bool use_attribute_default_value = false;
bool default_value_is_null = false;
};
template <DictionaryKeyType key_type>
class DictionaryKeysArenaHolder;
template <>
class DictionaryKeysArenaHolder<DictionaryKeyType::Simple>
{
public:
static Arena * getComplexKeyArena() { return nullptr; }
};
template <>
class DictionaryKeysArenaHolder<DictionaryKeyType::Complex>
{
public:
Arena * getComplexKeyArena() { return &complex_key_arena; }
private:
Arena complex_key_arena;
};
template <DictionaryKeyType key_type>
class DictionaryKeysExtractor
{
public:
using KeyType = std::conditional_t<key_type == DictionaryKeyType::Simple, UInt64, StringRef>;
explicit DictionaryKeysExtractor(const Columns & key_columns_, Arena * complex_key_arena_)
: key_columns(key_columns_)
, complex_key_arena(complex_key_arena_)
{
assert(!key_columns.empty());
if constexpr (key_type == DictionaryKeyType::Simple)
{
key_columns[0] = recursiveRemoveSparse(key_columns[0]->convertToFullColumnIfConst());
const auto * vector_col = checkAndGetColumn<ColumnVector<UInt64>>(key_columns[0].get());
if (!vector_col)
throw Exception(ErrorCodes::TYPE_MISMATCH, "Column type mismatch for simple key expected UInt64");
}
keys_size = key_columns.front()->size();
}
inline size_t getKeysSize() const
{
return keys_size;
}
inline size_t getCurrentKeyIndex() const
{
return current_key_index;
}
inline KeyType extractCurrentKey()
{
assert(current_key_index < keys_size);
if constexpr (key_type == DictionaryKeyType::Simple)
{
const auto & column_vector = static_cast<const ColumnVector<UInt64> &>(*key_columns[0]);
const auto & data = column_vector.getData();
auto key = data[current_key_index];
++current_key_index;
return key;
}
else
{
size_t allocated_size_for_columns = 0;
const char * block_start = nullptr;
for (const auto & column : key_columns)
{
StringRef serialized_data = column->serializeValueIntoArena(current_key_index, *complex_key_arena, block_start);
allocated_size_for_columns += serialized_data.size;
}
++current_key_index;
current_complex_key = StringRef{block_start, allocated_size_for_columns};
return current_complex_key;
}
}
void rollbackCurrentKey() const
{
if constexpr (key_type == DictionaryKeyType::Complex)
complex_key_arena->rollback(current_complex_key.size);
}
PaddedPODArray<KeyType> extractAllKeys()
{
PaddedPODArray<KeyType> result;
result.reserve(keys_size - current_key_index);
for (; current_key_index < keys_size;)
{
auto value = extractCurrentKey();
result.emplace_back(value);
}
return result;
}
void reset()
{
current_key_index = 0;
}
private:
Columns key_columns;
size_t keys_size = 0;
size_t current_key_index = 0;
KeyType current_complex_key {};
Arena * complex_key_arena;
};
/// Deserialize columns from keys array using dictionary structure
MutableColumns deserializeColumnsFromKeys(
const DictionaryStructure & dictionary_structure,
const PaddedPODArray<StringRef> & keys,
size_t start,
size_t end);
/// Deserialize columns with type and name from keys array using dictionary structure
ColumnsWithTypeAndName deserializeColumnsWithTypeAndNameFromKeys(
const DictionaryStructure & dictionary_structure,
const PaddedPODArray<StringRef> & keys,
size_t start,
size_t end);
/** Merge block with blocks from stream. If there are duplicate keys in block they are filtered out.
* In result block_to_update will be merged with blocks from stream.
* Note: readPrefix readImpl readSuffix will be called on stream object during function execution.
*/
template <DictionaryKeyType dictionary_key_type>
void mergeBlockWithPipe(
size_t key_columns_size,
Block & block_to_update,
QueryPipeline pipeline)
{
using KeyType = std::conditional_t<dictionary_key_type == DictionaryKeyType::Simple, UInt64, StringRef>;
Columns saved_block_key_columns;
saved_block_key_columns.reserve(key_columns_size);
/// Split into keys columns and attribute columns
for (size_t i = 0; i < key_columns_size; ++i)
saved_block_key_columns.emplace_back(block_to_update.safeGetByPosition(i).column);
DictionaryKeysArenaHolder<dictionary_key_type> arena_holder;
DictionaryKeysExtractor<dictionary_key_type> saved_keys_extractor(saved_block_key_columns, arena_holder.getComplexKeyArena());
auto saved_keys_extracted_from_block = saved_keys_extractor.extractAllKeys();
/**
* We create filter with our block to update size, because we want to filter out values that have duplicate keys
* if they appear in blocks that we fetch from stream.
* But first we try to filter out duplicate keys from existing block.
* For example if we have block with keys 1, 2, 2, 2, 3, 3
* Our filter will have [1, 0, 0, 1, 0, 1] after first stage.
* We also update saved_key_to_index hash map for keys to point to their latest indexes.
* For example if in blocks from stream we will get keys [4, 2, 3]
* Our filter will be [1, 0, 0, 0, 0, 0].
* After reading all blocks from stream we filter our duplicate keys from block_to_update and insert loaded columns.
*/
IColumn::Filter filter(saved_keys_extracted_from_block.size(), true);
HashMap<KeyType, size_t> saved_key_to_index;
saved_key_to_index.reserve(saved_keys_extracted_from_block.size());
size_t indexes_to_remove_count = 0;
for (size_t i = 0; i < saved_keys_extracted_from_block.size(); ++i)
{
auto saved_key = saved_keys_extracted_from_block[i];
auto [it, was_inserted] = saved_key_to_index.insert(makePairNoInit(saved_key, i));
if (!was_inserted)
{
size_t index_to_remove = it->getMapped();
filter[index_to_remove] = false;
it->getMapped() = i;
++indexes_to_remove_count;
}
}
auto result_fetched_columns = block_to_update.cloneEmptyColumns();
PullingPipelineExecutor executor(pipeline);
Block block;
while (executor.pull(block))
{
convertToFullIfSparse(block);
Columns block_key_columns;
block_key_columns.reserve(key_columns_size);
/// Split into keys columns and attribute columns
for (size_t i = 0; i < key_columns_size; ++i)
block_key_columns.emplace_back(block.safeGetByPosition(i).column);
DictionaryKeysExtractor<dictionary_key_type> update_keys_extractor(block_key_columns, arena_holder.getComplexKeyArena());
PaddedPODArray<KeyType> update_keys = update_keys_extractor.extractAllKeys();
for (auto update_key : update_keys)
{
const auto * it = saved_key_to_index.find(update_key);
if (it != nullptr)
{
size_t index_to_filter = it->getMapped();
filter[index_to_filter] = false;
++indexes_to_remove_count;
}
}
size_t rows = block.rows();
for (size_t column_index = 0; column_index < block.columns(); ++column_index)
{
const auto update_column = block.safeGetByPosition(column_index).column;
MutableColumnPtr & result_fetched_column = result_fetched_columns[column_index];
result_fetched_column->insertRangeFrom(*update_column, 0, rows);
}
}
size_t result_fetched_rows = result_fetched_columns.front()->size();
size_t filter_hint = filter.size() - indexes_to_remove_count;
for (size_t column_index = 0; column_index < block_to_update.columns(); ++column_index)
{
auto & column = block_to_update.getByPosition(column_index).column;
column = column->filter(filter, filter_hint);
MutableColumnPtr mutable_column = column->assumeMutable();
const IColumn & fetched_column = *result_fetched_columns[column_index];
mutable_column->insertRangeFrom(fetched_column, 0, result_fetched_rows);
}
}
/**
* Returns ColumnVector data as PaddedPodArray.
* If column is constant parameter backup_storage is used to store values.
*/
/// TODO: Remove
template <typename T>
static const PaddedPODArray<T> & getColumnVectorData(
const IDictionary * dictionary,
const ColumnPtr column,
PaddedPODArray<T> & backup_storage)
{
bool is_const_column = isColumnConst(*column);
auto full_column = recursiveRemoveSparse(column->convertToFullColumnIfConst());
auto vector_col = checkAndGetColumn<ColumnVector<T>>(full_column.get());
if (!vector_col)
{
throw Exception(ErrorCodes::TYPE_MISMATCH,
"{}: type mismatch: column has wrong type expected {}",
dictionary->getDictionaryID().getNameForLogs(),
TypeName<T>);
}
if (is_const_column)
{
// With type conversion and const columns we need to use backup storage here
auto & data = vector_col->getData();
backup_storage.assign(data);
return backup_storage;
}
else
{
return vector_col->getData();
}
}
template <typename T>
static ColumnPtr getColumnFromPODArray(const PaddedPODArray<T> & array)
{
auto column_vector = ColumnVector<T>::create();
column_vector->getData().reserve(array.size());
column_vector->getData().insert(array.begin(), array.end());
return column_vector;
}
template <typename T>
static ColumnPtr getColumnFromPODArray(PaddedPODArray<T> && array)
{
auto column_vector = ColumnVector<T>::create();
column_vector->getData() = std::move(array);
return column_vector;
}
template <typename T>
static ColumnPtr getColumnFromPODArray(const PaddedPODArray<T> & array, size_t start, size_t length)
{
auto column_vector = ColumnVector<T>::create();
column_vector->getData().reserve(length);
column_vector->getData().insert(array.begin() + start, array.begin() + start + length);
return column_vector;
}
}