#pragma once #include #include #include #include #include #include #include #include #include #include #include #include #include #include namespace ProfileEvents { extern const Event DictCacheKeysRequested; extern const Event DictCacheKeysRequestedMiss; extern const Event DictCacheKeysRequestedFound; extern const Event DictCacheKeysExpired; extern const Event DictCacheKeysNotFound; extern const Event DictCacheKeysHit; extern const Event DictCacheRequestTimeNs; extern const Event DictCacheRequests; extern const Event DictCacheLockWriteNs; extern const Event DictCacheLockReadNs; } namespace CurrentMetrics { extern const Metric DictCacheRequests; } namespace DB { class CacheDictionary final : public IDictionary { public: CacheDictionary(const std::string & name, const DictionaryStructure & dict_struct, DictionarySourcePtr source_ptr, const DictionaryLifetime dict_lifetime, const std::size_t size); CacheDictionary(const CacheDictionary & other); std::exception_ptr getCreationException() const override { return {}; } std::string getName() const override { return name; } std::string getTypeName() const override { return "Cache"; } std::size_t getBytesAllocated() const override { return bytes_allocated + (string_arena ? string_arena->size() : 0); } std::size_t getQueryCount() const override { return query_count.load(std::memory_order_relaxed); } double getHitRate() const override { return static_cast(hit_count.load(std::memory_order_acquire)) / query_count.load(std::memory_order_relaxed); } std::size_t getElementCount() const override { return element_count.load(std::memory_order_relaxed); } double getLoadFactor() const override { return static_cast(element_count.load(std::memory_order_relaxed)) / size; } bool isCached() const override { return true; } 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 PaddedPODArray & ids, PaddedPODArray & out) const override; /* void isInVectorVector(const PaddedPODArray & child_ids, const PaddedPODArray & ancestor_ids, PaddedPODArray & out) const override; void isInVectorConstant(const PaddedPODArray & child_ids, const Key ancestor_id, PaddedPODArray & out) const override; void isInConstantVector(const Key child_id, const PaddedPODArray & ancestor_ids, PaddedPODArray & out) const override; */ #define DECLARE(TYPE)\ void get##TYPE(const std::string & attribute_name, const PaddedPODArray & ids, PaddedPODArray & out) const; 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 PaddedPODArray & ids, ColumnString * out) const; #define DECLARE(TYPE)\ void get##TYPE(\ const std::string & attribute_name, const PaddedPODArray & ids, const PaddedPODArray & def,\ PaddedPODArray & out) const; 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 PaddedPODArray & ids, const ColumnString * const def, ColumnString * const out) const; #define DECLARE(TYPE)\ void get##TYPE(\ const std::string & attribute_name, const PaddedPODArray & ids, const TYPE def, PaddedPODArray & out) const; 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 PaddedPODArray & ids, const String & def, ColumnString * const out) const; void has(const PaddedPODArray & ids, PaddedPODArray & out) const override; private: template using ContainerType = Value[]; template using ContainerPtrType = std::unique_ptr>; struct CellMetadata final { using time_point_t = std::chrono::system_clock::time_point; using time_point_rep_t = time_point_t::rep; using time_point_urep_t = std::make_unsigned_t; static constexpr UInt64 EXPIRES_AT_MASK = std::numeric_limits::max(); static constexpr UInt64 IS_DEFAULT_MASK = ~EXPIRES_AT_MASK; UInt64 id; /// Stores both expiration time and `is_default` flag in the most significant bit time_point_urep_t data; /// Sets expiration time, resets `is_default` flag to false time_point_t expiresAt() const { return ext::safe_bit_cast(data & EXPIRES_AT_MASK); } void setExpiresAt(const time_point_t & t) { data = ext::safe_bit_cast(t); } bool isDefault() const { return (data & IS_DEFAULT_MASK) == IS_DEFAULT_MASK; } void setDefault() { data |= IS_DEFAULT_MASK; } }; struct Attribute final { AttributeUnderlyingType type; std::tuple< UInt8, UInt16, UInt32, UInt64, Int8, Int16, Int32, Int64, Float32, Float64, String> null_values; std::tuple< ContainerPtrType, ContainerPtrType, ContainerPtrType, ContainerPtrType, ContainerPtrType, ContainerPtrType, ContainerPtrType, ContainerPtrType, ContainerPtrType, ContainerPtrType, ContainerPtrType> arrays; }; void createAttributes(); Attribute createAttributeWithType(const AttributeUnderlyingType type, const Field & null_value); template void getItemsNumber( Attribute & attribute, const PaddedPODArray & ids, PaddedPODArray & out, DefaultGetter && get_default) const; template void getItemsNumberImpl( Attribute & attribute, const PaddedPODArray & ids, PaddedPODArray & out, DefaultGetter && get_default) const; template void getItemsString( Attribute & attribute, const PaddedPODArray & ids, ColumnString * out, DefaultGetter && get_default) const; template void update( const std::vector & requested_ids, PresentIdHandler && on_cell_updated, AbsentIdHandler && on_id_not_found) const; UInt64 getCellIdx(const Key id) const; void setDefaultAttributeValue(Attribute & attribute, const Key idx) const; void setAttributeValue(Attribute & attribute, const Key idx, const Field & value) const; Attribute & getAttribute(const std::string & attribute_name) const; struct FindResult { const size_t cell_idx; const bool valid; const bool outdated; }; FindResult findCellIdx(const Key & id, const CellMetadata::time_point_t now) const; const std::string name; const DictionaryStructure dict_struct; const DictionarySourcePtr source_ptr; const DictionaryLifetime dict_lifetime; mutable Poco::RWLock rw_lock; /// Actual size will be increased to match power of 2 const std::size_t size; /// all bits to 1 mask (size - 1) (0b1000 - 1 = 0b111) const std::size_t size_overlap_mask; /// Max tries to find cell, overlaped with mask: if size = 16 and start_cell=10: will try cells: 10,11,12,13,14,15,0,1,2,3 static constexpr std::size_t max_collision_length = 10; const UInt64 zero_cell_idx{getCellIdx(0)}; std::map attribute_index_by_name; mutable std::vector attributes; mutable std::vector cells; Attribute * hierarchical_attribute = nullptr; std::unique_ptr string_arena; mutable std::mt19937_64 rnd_engine; mutable std::size_t bytes_allocated = 0; mutable std::atomic element_count{0}; mutable std::atomic hit_count{0}; mutable std::atomic query_count{0}; const std::chrono::time_point creation_time = std::chrono::system_clock::now(); }; }