#pragma once #include #include #include #include #include #include #include #include #include namespace DB { class RangeHashedDictionary final : public IDictionaryBase { public: RangeHashedDictionary( 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(); } RangeHashedDictionary(const RangeHashedDictionary & other) : RangeHashedDictionary{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 "RangeHashed"; } 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; } #define DECLARE_MULTIPLE_GETTER(TYPE)\ void get##TYPE(\ const std::string & attribute_name, const PODArray & ids, const PODArray & dates,\ PODArray & out) const\ {\ const auto & attribute = getAttributeWithType(attribute_name, AttributeUnderlyingType::TYPE);\ getItems(attribute, ids, dates, 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, const PODArray & dates, ColumnString * out) const { const auto & attribute = getAttributeWithType(attribute_name, AttributeUnderlyingType::String); const auto & attr = *std::get>(attribute.maps); const auto & null_value = std::get(attribute.null_values); for (const auto i : ext::range(0, ids.size())) { const auto it = attr.find(ids[i]); if (it != std::end(attr)) { const auto date = dates[i]; const auto & ranges_and_values = it->second; const auto val_it = std::find_if(std::begin(ranges_and_values), std::end(ranges_and_values), [date] (const value_t & v) { return v.range.contains(date); }); const auto string_ref = val_it != std::end(ranges_and_values) ? val_it->value : null_value; out->insertData(string_ref.data, string_ref.size); } else out->insertData(null_value.data(), null_value.size()); } query_count.fetch_add(ids.size(), std::memory_order_relaxed); } private: struct range_t : std::pair { using std::pair::pair; bool contains(const UInt16 date) const { const auto & left = first; const auto & right = second; if (left <= date && date <= right) return true; const auto has_left_bound = 0 < left && left <= DATE_LUT_MAX_DAY_NUM; const auto has_right_bound = 0 < right && right <= DATE_LUT_MAX_DAY_NUM; if ((!has_left_bound || left <= date) && (!has_right_bound || date <= right)) return true; return false; } }; template struct value_t final { range_t range; T value; }; template using values_t = std::vector>; template using collection_t = HashMap>; template using ptr_t = std::unique_ptr>; struct attribute_t final { public: AttributeUnderlyingType type; std::tuple null_values; std::tuple, ptr_t, ptr_t, ptr_t, ptr_t, ptr_t, ptr_t, ptr_t, ptr_t, ptr_t, ptr_t> 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) throw Exception{ name + ": hierarchical attributes not supported by " + getName() + " dictionary.", ErrorCodes::BAD_ARGUMENTS }; } } void loadData() { auto stream = source_ptr->loadAll(); stream->readPrefix(); while (const auto block = stream->read()) { const auto & id_column = *block.getByPosition(0).column; const auto & min_range_column = *block.getByPosition(1).column; const auto & max_range_column = *block.getByPosition(2).column; element_count += id_column.size(); for (const auto attribute_idx : ext::range(0, attributes.size())) { const auto & attribute_column = *block.getByPosition(attribute_idx + 3).column; auto & attribute = attributes[attribute_idx]; for (const auto row_idx : ext::range(0, id_column.size())) setAttributeValue(attribute, id_column[row_idx].get(), range_t(min_range_column[row_idx].get(), max_range_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(collection_t) + 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: { const auto & null_value_ref = 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, const PODArray & dates, PODArray & out) const { const auto & attr = *std::get>(attribute.maps); const auto null_value = std::get(attribute.null_values); for (const auto i : ext::range(0, ids.size())) { const auto it = attr.find(ids[i]); if (it != std::end(attr)) { const auto date = dates[i]; const auto & ranges_and_values = it->second; const auto val_it = std::find_if(std::begin(ranges_and_values), std::end(ranges_and_values), [date] (const value_t & v) { return v.range.contains(date); }); out[i] = val_it != std::end(ranges_and_values) ? val_it->value : null_value; } else out[i] = null_value; } query_count.fetch_add(ids.size(), std::memory_order_relaxed); } template void setAttributeValueImpl(attribute_t & attribute, const id_t id, const range_t & range, const T value) { auto & map = *std::get>(attribute.maps); const auto it = map.find(id); if (it != map.end()) { auto & values = it->second; const auto insert_it = std::lower_bound(std::begin(values), std::end(values), range, [] (const value_t & lhs, const range_t & range) { return lhs.range < range; }); values.insert(insert_it, value_t{ range, value }); } else map.insert({ id, values_t{ value_t{ range, value } } }); } void setAttributeValue(attribute_t & attribute, const id_t id, const range_t & range, const Field & value) { switch (attribute.type) { case AttributeUnderlyingType::UInt8: setAttributeValueImpl(attribute, id, range, value.get()); break; case AttributeUnderlyingType::UInt16: setAttributeValueImpl(attribute, id, range, value.get()); break; case AttributeUnderlyingType::UInt32: setAttributeValueImpl(attribute, id, range, value.get()); break; case AttributeUnderlyingType::UInt64: setAttributeValueImpl(attribute, id, range, value.get()); break; case AttributeUnderlyingType::Int8: setAttributeValueImpl(attribute, id, range, value.get()); break; case AttributeUnderlyingType::Int16: setAttributeValueImpl(attribute, id, range, value.get()); break; case AttributeUnderlyingType::Int32: setAttributeValueImpl(attribute, id, range, value.get()); break; case AttributeUnderlyingType::Int64: setAttributeValueImpl(attribute, id, range, value.get()); break; case AttributeUnderlyingType::Float32: setAttributeValueImpl(attribute, id, range, value.get()); break; case AttributeUnderlyingType::Float64: setAttributeValueImpl(attribute, id, range, 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()); const StringRef string_ref{string_in_arena, string.size()}; const auto it = map.find(id); if (it != map.end()) { auto & values = it->second; const auto insert_it = std::lower_bound(std::begin(values), std::end(values), range, [] (const value_t & lhs, const range_t & range) { return lhs.range < range; }); values.insert(insert_it, value_t{ range, string_ref }); } else map.insert({ id, values_t{ value_t{ range, string_ref } } }); 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 attribute_t & getAttributeWithType(const std::string & name, const AttributeUnderlyingType type) const { const auto & attribute = getAttribute(name); if (attribute.type != type) throw Exception{ name + ": type mismatch: attribute " + name + " has type " + toString(attribute.type), ErrorCodes::TYPE_MISMATCH }; return attribute; } 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; 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; }; }