#pragma once #include #include #include #include #include #include #include #include #include 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; } class HashedDictionary final : public IDictionary { public: HashedDictionary(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), require_nonempty(require_nonempty) { createAttributes(); try { loadData(); calculateBytesAllocated(); } catch (...) { creation_exception = std::current_exception(); } creation_time = std::chrono::system_clock::now(); } HashedDictionary(const HashedDictionary & other) : HashedDictionary{other.name, other.dict_struct, other.source_ptr->clone(), other.dict_lifetime, other.require_nonempty} {} std::exception_ptr getCreationException() const override { return creation_exception; } std::string getName() const override { return name; } std::string getTypeName() const override { return "Hashed"; } std::size_t getBytesAllocated() const override { return bytes_allocated; } std::size_t getQueryCount() const override { return query_count.load(std::memory_order_relaxed); } double getHitRate() const override { return 1.0; } std::size_t getElementCount() const override { return element_count; } double getLoadFactor() const override { return static_cast(element_count) / bucket_count; } bool isCached() const override { return false; } DictionaryPtr clone() const override { return std::make_unique(*this); } const IDictionarySource * getSource() const override { return source_ptr.get(); } const DictionaryLifetime & getLifetime() const override { return dict_lifetime; } const DictionaryStructure & getStructure() const override { return dict_struct; } std::chrono::time_point getCreationTime() const override { return creation_time; } bool isInjective(const std::string & attribute_name) const override { return dict_struct.attributes[&getAttribute(attribute_name) - attributes.data()].injective; } bool hasHierarchy() const override { return hierarchical_attribute; } void toParent(const PODArray & ids, PODArray & out) const override { const auto null_value = std::get(hierarchical_attribute->null_values); getItems(*hierarchical_attribute, ids, [&] (const std::size_t row, const UInt64 value) { out[row] = value; }, [&] (const std::size_t) { return null_value; }); } #define DECLARE(TYPE)\ void get##TYPE(const std::string & attribute_name, const PODArray & ids, PODArray & out) const\ {\ const auto & attribute = getAttribute(attribute_name);\ if (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(attribute.null_values);\ \ getItems(attribute, ids,\ [&] (const std::size_t row, const auto value) { out[row] = value; },\ [&] (const std::size_t) { return null_value; });\ } DECLARE(UInt8) DECLARE(UInt16) DECLARE(UInt32) DECLARE(UInt64) DECLARE(Int8) DECLARE(Int16) DECLARE(Int32) DECLARE(Int64) DECLARE(Float32) DECLARE(Float64) #undef DECLARE void getString(const std::string & attribute_name, const PODArray & ids, ColumnString * out) const { const auto & attribute = getAttribute(attribute_name); if (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(attribute.null_values)}; getItems(attribute, ids, [&] (const std::size_t row, const StringRef value) { out->insertData(value.data, value.size); }, [&] (const std::size_t) { return null_value; }); } #define DECLARE(TYPE)\ void get##TYPE(\ const std::string & attribute_name, const PODArray & ids, const PODArray & def,\ PODArray & out) const\ {\ const auto & attribute = getAttribute(attribute_name);\ if (attribute.type != AttributeUnderlyingType::TYPE)\ throw Exception{\ name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type),\ ErrorCodes::TYPE_MISMATCH\ };\ \ getItems(attribute, ids,\ [&] (const std::size_t row, const auto value) { out[row] = value; },\ [&] (const std::size_t row) { return def[row]; });\ } DECLARE(UInt8) DECLARE(UInt16) DECLARE(UInt32) DECLARE(UInt64) DECLARE(Int8) DECLARE(Int16) DECLARE(Int32) DECLARE(Int64) DECLARE(Float32) DECLARE(Float64) #undef DECLARE void getString( const std::string & attribute_name, const PODArray & ids, const ColumnString * const def, ColumnString * const out) const { const auto & attribute = getAttribute(attribute_name); if (attribute.type != AttributeUnderlyingType::String) throw Exception{ name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type), ErrorCodes::TYPE_MISMATCH }; getItems(attribute, ids, [&] (const std::size_t row, const StringRef value) { out->insertData(value.data, value.size); }, [&] (const std::size_t row) { return def->getDataAt(row); }); } #define DECLARE(TYPE)\ void get##TYPE(\ const std::string & attribute_name, const PODArray & ids, const TYPE & def, PODArray & out) const\ {\ const auto & attribute = getAttribute(attribute_name);\ if (attribute.type != AttributeUnderlyingType::TYPE)\ throw Exception{\ name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type),\ ErrorCodes::TYPE_MISMATCH\ };\ \ getItems(attribute, ids,\ [&] (const std::size_t row, const auto value) { out[row] = value; },\ [&] (const std::size_t) { return def; });\ } DECLARE(UInt8) DECLARE(UInt16) DECLARE(UInt32) DECLARE(UInt64) DECLARE(Int8) DECLARE(Int16) DECLARE(Int32) DECLARE(Int64) DECLARE(Float32) DECLARE(Float64) #undef DECLARE void getString( const std::string & attribute_name, const PODArray & ids, const String & def, ColumnString * const out) const { const auto & attribute = getAttribute(attribute_name); if (attribute.type != AttributeUnderlyingType::String) throw Exception{ name + ": type mismatch: attribute " + attribute_name + " has type " + toString(attribute.type), ErrorCodes::TYPE_MISMATCH }; getItems(attribute, ids, [&] (const std::size_t row, const StringRef value) { out->insertData(value.data, value.size); }, [&] (const std::size_t) { return StringRef{def}; }); } void has(const PODArray & ids, PODArray & out) const override { const auto & attribute = attributes.front(); switch (attribute.type) { case AttributeUnderlyingType::UInt8: has(attribute, ids, out); break; case AttributeUnderlyingType::UInt16: has(attribute, ids, out); break; case AttributeUnderlyingType::UInt32: has(attribute, ids, out); break; case AttributeUnderlyingType::UInt64: has(attribute, ids, out); break; case AttributeUnderlyingType::Int8: has(attribute, ids, out); break; case AttributeUnderlyingType::Int16: has(attribute, ids, out); break; case AttributeUnderlyingType::Int32: has(attribute, ids, out); break; case AttributeUnderlyingType::Int64: has(attribute, ids, out); break; case AttributeUnderlyingType::Float32: has(attribute, ids, out); break; case AttributeUnderlyingType::Float64: has(attribute, ids, out); break; case AttributeUnderlyingType::String: has(attribute, ids, out); break; } } private: template using CollectionType = HashMap; template using CollectionPtrType = std::unique_ptr>; struct attribute_t final { AttributeUnderlyingType type; std::tuple< UInt8, UInt16, UInt32, UInt64, Int8, Int16, Int32, Int64, Float32, Float64, String> null_values; std::tuple< CollectionPtrType, CollectionPtrType, CollectionPtrType, CollectionPtrType, CollectionPtrType, CollectionPtrType, CollectionPtrType, CollectionPtrType, CollectionPtrType, CollectionPtrType, CollectionPtrType> maps; std::unique_ptr string_arena; }; void 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) { hierarchical_attribute = &attributes.back(); if (hierarchical_attribute->type != AttributeUnderlyingType::UInt64) throw Exception{ name + ": hierarchical attribute must be UInt64.", ErrorCodes::TYPE_MISMATCH }; } } } void loadData() { auto stream = source_ptr->loadAll(); stream->readPrefix(); while (const auto block = stream->read()) { const auto & id_column = *block.getByPosition(0).column; element_count += id_column.size(); for (const auto attribute_idx : ext::range(0, attributes.size())) { const auto & attribute_column = *block.getByPosition(attribute_idx + 1).column; auto & attribute = attributes[attribute_idx]; for (const auto row_idx : ext::range(0, id_column.size())) setAttributeValue(attribute, id_column[row_idx].get(), 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 void addAttributeSize(const attribute_t & attribute) { const auto & map_ref = std::get>(attribute.maps); bytes_allocated += sizeof(CollectionType) + map_ref->getBufferSizeInBytes(); bucket_count = map_ref->getBufferSizeInCells(); } void calculateBytesAllocated() { bytes_allocated += attributes.size() * sizeof(attributes.front()); for (const auto & attribute : attributes) { switch (attribute.type) { case AttributeUnderlyingType::UInt8: addAttributeSize(attribute); break; case AttributeUnderlyingType::UInt16: addAttributeSize(attribute); break; case AttributeUnderlyingType::UInt32: addAttributeSize(attribute); break; case AttributeUnderlyingType::UInt64: addAttributeSize(attribute); break; case AttributeUnderlyingType::Int8: addAttributeSize(attribute); break; case AttributeUnderlyingType::Int16: addAttributeSize(attribute); break; case AttributeUnderlyingType::Int32: addAttributeSize(attribute); break; case AttributeUnderlyingType::Int64: addAttributeSize(attribute); break; case AttributeUnderlyingType::Float32: addAttributeSize(attribute); break; case AttributeUnderlyingType::Float64: addAttributeSize(attribute); break; case AttributeUnderlyingType::String: { addAttributeSize(attribute); bytes_allocated += sizeof(Arena) + attribute.string_arena->size(); break; } } } } template void createAttributeImpl(attribute_t & attribute, const Field & null_value) { std::get(attribute.null_values) = null_value.get::Type>(); std::get>(attribute.maps) = std::make_unique>(); } attribute_t createAttributeWithType(const AttributeUnderlyingType type, const Field & null_value) { attribute_t attr{type}; switch (type) { case AttributeUnderlyingType::UInt8: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::UInt16: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::UInt32: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::UInt64: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::Int8: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::Int16: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::Int32: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::Int64: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::Float32: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::Float64: createAttributeImpl(attr, null_value); break; case AttributeUnderlyingType::String: { std::get(attr.null_values) = null_value.get(); std::get>(attr.maps) = std::make_unique>(); attr.string_arena = std::make_unique(); break; } } return attr; } template void getItems( const attribute_t & attribute, const PODArray & ids, ValueSetter && set_value, DefaultGetter && get_default) const { const auto & attr = *std::get>(attribute.maps); const auto rows = ext::size(ids); for (const auto i : ext::range(0, rows)) { const auto it = attr.find(ids[i]); set_value(i, it != attr.end() ? it->second : get_default(i)); } query_count.fetch_add(rows, std::memory_order_relaxed); } template void setAttributeValueImpl(attribute_t & attribute, const id_t id, const T value) { auto & map = *std::get>(attribute.maps); map.insert({ id, value }); } void setAttributeValue(attribute_t & attribute, const id_t id, const Field & value) { switch (attribute.type) { case AttributeUnderlyingType::UInt8: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::UInt16: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::UInt32: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::UInt64: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::Int8: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::Int16: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::Int32: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::Int64: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::Float32: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::Float64: setAttributeValueImpl(attribute, id, value.get()); break; case AttributeUnderlyingType::String: { auto & map = *std::get>(attribute.maps); const auto & string = value.get(); const auto string_in_arena = attribute.string_arena->insert(string.data(), string.size()); map.insert({ id, StringRef{string_in_arena, string.size()} }); break; } } } const attribute_t & 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 void has(const attribute_t & attribute, const PODArray & ids, PODArray & out) const { const auto & attr = *std::get>(attribute.maps); const auto rows = ext::size(ids); for (const auto i : ext::range(0, rows)) out[i] = attr.find(ids[i]) != std::end(attr); query_count.fetch_add(rows, std::memory_order_relaxed); } const std::string name; const DictionaryStructure dict_struct; const DictionarySourcePtr source_ptr; const DictionaryLifetime dict_lifetime; const bool require_nonempty; std::map attribute_index_by_name; std::vector attributes; const attribute_t * hierarchical_attribute = nullptr; std::size_t bytes_allocated = 0; std::size_t element_count = 0; std::size_t bucket_count = 0; mutable std::atomic query_count{0}; std::chrono::time_point creation_time; std::exception_ptr creation_exception; }; }