#pragma once #include #include #include #include #include #include #include #include #include namespace DB { const auto initial_array_size = 1024; const auto max_array_size = 500000; class FlatDictionary final : public IDictionary { public: FlatDictionary(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(); } FlatDictionary(const FlatDictionary & other) : FlatDictionary{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 "Flat"; } 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 { getItems(*hierarchical_attribute, ids, out); } #define DECLARE_MULTIPLE_GETTER(TYPE)\ void get##TYPE(const std::string & attribute_name, const PODArray & ids, PODArray & out) const override\ {\ 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, out);\ } DECLARE_MULTIPLE_GETTER(UInt8) DECLARE_MULTIPLE_GETTER(UInt16) DECLARE_MULTIPLE_GETTER(UInt32) DECLARE_MULTIPLE_GETTER(UInt64) DECLARE_MULTIPLE_GETTER(Int8) DECLARE_MULTIPLE_GETTER(Int16) DECLARE_MULTIPLE_GETTER(Int32) DECLARE_MULTIPLE_GETTER(Int64) DECLARE_MULTIPLE_GETTER(Float32) DECLARE_MULTIPLE_GETTER(Float64) #undef DECLARE_MULTIPLE_GETTER void getString(const std::string & attribute_name, const PODArray & ids, ColumnString * out) const override { 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 & attr = *std::get>>(attribute.arrays); const auto & null_value = StringRef{std::get(attribute.null_values)}; for (const auto i : ext::range(0, ids.size())) { const auto id = ids[i]; const auto string_ref = id < attr.size() ? attr[id] : null_value; out->insertData(string_ref.data, string_ref.size); } query_count.fetch_add(ids.size(), std::memory_order_relaxed); } #define DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(TYPE)\ void get##TYPE(\ const std::string & attribute_name, const PODArray & ids, const PODArray & def,\ PODArray & out) const override\ {\ 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, def, out);\ } DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(UInt8) DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(UInt16) DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(UInt32) DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(UInt64) DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(Int8) DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(Int16) DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(Int32) DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(Int64) DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(Float32) DECLARE_MULTIPLE_GETTER_WITH_DEFAULT(Float64) #undef DECLARE_MULTIPLE_GETTER_WITH_DEFAULT void getString( const std::string & attribute_name, const PODArray & ids, const ColumnString * const def, ColumnString * const out) const override { 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 & attr = *std::get>>(attribute.arrays); for (const auto i : ext::range(0, ids.size())) { const auto id = ids[i]; /// @todo getDataAt or with terminating zero? const auto string_ref = id < attr.size() ? attr[id] : def->getDataAt(i); out->insertData(string_ref.data, string_ref.size); } query_count.fetch_add(ids.size(), std::memory_order_relaxed); } private: struct attribute_t final { AttributeUnderlyingType type; std::tuple null_values; std::tuple>, std::unique_ptr>, std::unique_ptr>, std::unique_ptr>, std::unique_ptr>, std::unique_ptr>, std::unique_ptr>, std::unique_ptr>, std::unique_ptr>, std::unique_ptr>, std::unique_ptr>> arrays; 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 & array_ref = std::get>>(attribute.arrays); bytes_allocated += sizeof(PODArray) + array_ref->storage_size(); bucket_count = array_ref->capacity(); } 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) { const auto & null_value_ref = std::get(attribute.null_values) = null_value.get::Type>(); std::get>>(attribute.arrays) = std::make_unique>(initial_array_size, null_value_ref); } 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: { const auto & null_value_ref = std::get(attr.null_values) = null_value.get(); std::get>>(attr.arrays) = std::make_unique>(initial_array_size, StringRef{null_value_ref}); attr.string_arena = std::make_unique(); break; } } return attr; } template void getItems(const attribute_t & attribute, const PODArray & ids, PODArray & out) const { const auto & attr = *std::get>>(attribute.arrays); const auto null_value = std::get(attribute.null_values); for (const auto i : ext::range(0, ids.size())) { const auto id = ids[i]; out[i] = id < attr.size() ? attr[id] : null_value; } query_count.fetch_add(ids.size(), std::memory_order_relaxed); } template void getItems(const attribute_t & attribute, const PODArray & ids, const PODArray & def, PODArray & out) const { const auto & attr = *std::get>>(attribute.arrays); for (const auto i : ext::range(0, ids.size())) { const auto id = ids[i]; out[i] = id < attr.size() ? attr[id] : def[i]; } query_count.fetch_add(ids.size(), std::memory_order_relaxed); } template void setAttributeValueImpl(attribute_t & attribute, const id_t id, const T value) { auto & array = *std::get>>(attribute.arrays); if (id >= array.size()) array.resize_fill(id + 1, std::get(attribute.null_values)); array[id] = value; } void setAttributeValue(attribute_t & attribute, const id_t id, const Field & value) { if (id >= max_array_size) throw Exception{ name + ": identifier should be less than " + toString(max_array_size), ErrorCodes::ARGUMENT_OUT_OF_BOUND }; 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 & array = *std::get>>(attribute.arrays); if (id >= array.size()) array.resize_fill(id + 1, StringRef{std::get(attribute.null_values)}); const auto & string = value.get(); const auto string_in_arena = attribute.string_arena->insert(string.data(), string.size()); array[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]; } 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; }; }