Merge

dbms/include/DB/Columns/IColumn.h

@@ -176,6 +176,7 @@ public:
 
 	/** Размножить все значения столько раз, сколько прописано в offsets.
 	  * (i-е значение размножается в offsets[i] - offsets[i - 1] значений.)
+	  * Необходимо для реализации операции ARRAY JOIN.
 	  */
 	typedef UInt64 Offset_t;
 	typedef PODArray<Offset_t> Offsets_t;

dbms/include/DB/Common/HashTable/ClearableHashMap.h

@@ -1,51 +1,17 @@
 #pragma once
 
 #include <DB/Common/HashTable/HashMap.h>
-
-
-/** Хеш-таблица, позволяющая очищать таблицу за O(1).
-  * Еще более простая, чем HashMap: Key и Mapped должны быть POD-типами.
-  *
-  * Вместо этого класса можно было бы просто использовать в HashMap в качестве ключа пару <версия, ключ>,
-  * но тогда таблица накапливала бы все ключи, которые в нее когда-либо складывали, и неоправданно росла.
-  * Этот класс идет на шаг дальше и считает ключи со старой версией пустыми местами в хеш-таблице.
-  */
-
-struct ClearableHashMapState
-{
-	UInt32 version = 1;
-
-	/// Сериализация, в бинарном и текстовом виде.
-	void write(DB::WriteBuffer & wb) const 		{ DB::writeBinary(version, wb); }
-	void writeText(DB::WriteBuffer & wb) const 	{ DB::writeText(version, wb); }
-
-	/// Десериализация, в бинарном и текстовом виде.
-	void read(DB::ReadBuffer & rb) 				{ DB::readBinary(version, rb); }
-	void readText(DB::ReadBuffer & rb) 			{ DB::readText(version, rb); }
-};
+#include <DB/Common/HashTable/ClearableHashSet.h>
 
 
 template <typename Key, typename Mapped, typename Hash>
-struct ClearableHashMapCell : public HashMapCell<Key, Mapped, Hash, ClearableHashMapState>
+struct ClearableHashMapCell : public ClearableHashTableCell<Key, HashMapCell<Key, Mapped, Hash, ClearableHashSetState>>
 {
-	typedef ClearableHashMapState State;
-	typedef HashMapCell<Key, Mapped, Hash, ClearableHashMapState> Base;
-	typedef typename Base::value_type value_type;
+	using Base = ClearableHashTableCell<Key, HashMapCell<Key, Mapped, Hash, ClearableHashSetState>>;
+	using Base::Base;
 
-	UInt32 version;
-
-	ClearableHashMapCell() {}
-	ClearableHashMapCell(const Key & key_, const State & state) : Base(key_, state), version(state.version) {}
-	ClearableHashMapCell(const value_type & value_, const State & state) : Base(value_, state), version(state.version) {}
-
-	bool isZero(const State & state) const { return version != state.version; }
-	static bool isZero(const Key & key, const State & state) { return false; }
-
-	/// Установить значение ключа в ноль.
-	void setZero() { version = 0; }
-
-	/// Нужно ли хранить нулевой ключ отдельно (то есть, могут ли в хэш-таблицу вставить нулевой ключ).
-	static constexpr bool need_zero_value_storage = false;
+	ClearableHashMapCell(const typename Base::value_type & value_, const typename Base::State & state)
+		: Base::BaseCell(value_, state), Base::version(state.version) {}
 };
 
 

dbms/include/DB/Common/HashTable/ClearableHashSet.h

@@ -0,0 +1,70 @@
+#pragma once
+
+#include <type_traits>
+#include <DB/Common/HashTable/HashSet.h>
+
+
+/** Хеш-таблица, позволяющая очищать таблицу за O(1).
+  * Еще более простая, чем HashSet: Key и Mapped должны быть POD-типами.
+  *
+  * Вместо этого класса можно было бы просто использовать в HashSet в качестве ключа пару <версия, ключ>,
+  * но тогда таблица накапливала бы все ключи, которые в нее когда-либо складывали, и неоправданно росла.
+  * Этот класс идет на шаг дальше и считает ключи со старой версией пустыми местами в хеш-таблице.
+  */
+
+
+struct ClearableHashSetState
+{
+	UInt32 version = 1;
+
+	/// Сериализация, в бинарном и текстовом виде.
+	void write(DB::WriteBuffer & wb) const 		{ DB::writeBinary(version, wb); }
+	void writeText(DB::WriteBuffer & wb) const 	{ DB::writeText(version, wb); }
+
+	/// Десериализация, в бинарном и текстовом виде.
+	void read(DB::ReadBuffer & rb) 				{ DB::readBinary(version, rb); }
+	void readText(DB::ReadBuffer & rb) 			{ DB::readText(version, rb); }
+};
+
+
+template <typename Key, typename BaseCell>
+struct ClearableHashTableCell : public BaseCell
+{
+	typedef ClearableHashSetState State;
+	typedef typename BaseCell::value_type value_type;
+
+	UInt32 version;
+
+	bool isZero(const State & state) const { return version != state.version; }
+	static bool isZero(const Key & key, const State & state) { return false; }
+
+	/// Установить значение ключа в ноль.
+	void setZero() { version = 0; }
+
+	/// Нужно ли хранить нулевой ключ отдельно (то есть, могут ли в хэш-таблицу вставить нулевой ключ).
+	static constexpr bool need_zero_value_storage = false;
+
+	ClearableHashTableCell() {}
+	ClearableHashTableCell(const Key & key_, const State & state) : BaseCell(key_, state), version(state.version) {}
+};
+
+
+template
+<
+	typename Key,
+	typename Hash = DefaultHash<Key>,
+	typename Grower = HashTableGrower<>,
+	typename Allocator = HashTableAllocator
+>
+class ClearableHashSet : public HashTable<Key, ClearableHashTableCell<Key, HashTableCell<Key, Hash, ClearableHashSetState>>, Hash, Grower, Allocator>
+{
+public:
+	typedef Key key_type;
+	typedef typename ClearableHashSet::cell_type::value_type value_type;
+
+	void clear()
+	{
+		++this->version;
+		this->m_size = 0;
+	}
+};

dbms/include/DB/Common/HashTable/HashMap.h

@@ -102,9 +102,7 @@ struct HashMapCellWithSavedHash : public HashMapCell<Key, TMapped, Hash, TState>
 
 	size_t saved_hash;
 
-	HashMapCellWithSavedHash() : Base() {}
-	HashMapCellWithSavedHash(const Key & key_, const typename Base::State & state) : Base(key_, state) {}
-	HashMapCellWithSavedHash(const typename Base::value_type & value_, const typename Base::State & state) : Base(value_, state) {}
+	using Base::Base;
 
 	bool keyEquals(const Key & key_) const { return this->value.first == key_; }
 	bool keyEquals(const Key & key_, size_t hash_) const { return saved_hash == hash_ && this->value.first == key_; }

dbms/include/DB/DataStreams/FormatFactory.h

@@ -14,10 +14,10 @@ class FormatFactory
 {
 public:
 	BlockInputStreamPtr getInput(const String & name, ReadBuffer & buf,
-		Block & sample, size_t max_block_size) const;
+		const Block & sample, size_t max_block_size) const;
 
 	BlockOutputStreamPtr getOutput(const String & name, WriteBuffer & buf,
-		Block & sample) const;
+		const Block & sample) const;
 };
 
 }

dbms/include/DB/DataStreams/ParallelAggregatingBlockInputStream.h

@@ -21,17 +21,21 @@ class ParallelAggregatingBlockInputStream : public IProfilingBlockInputStream
 public:
 	/** Столбцы из key_names и аргументы агрегатных функций, уже должны быть вычислены.
 	  */
-	ParallelAggregatingBlockInputStream(BlockInputStreams inputs, const Names & key_names,
-		const AggregateDescriptions & aggregates,	bool overflow_row_, bool final_, size_t max_threads_,
+	ParallelAggregatingBlockInputStream(
+		BlockInputStreams inputs, BlockInputStreamPtr additional_input_at_end,
+		const Names & key_names, const AggregateDescriptions & aggregates,
+		bool overflow_row_, bool final_, size_t max_threads_,
 		size_t max_rows_to_group_by_, OverflowMode group_by_overflow_mode_,
 		Compiler * compiler_, UInt32 min_count_to_compile_, size_t group_by_two_level_threshold_)
 		: aggregator(key_names, aggregates, overflow_row_, max_rows_to_group_by_, group_by_overflow_mode_,
 			compiler_, min_count_to_compile_, group_by_two_level_threshold_),
 		final(final_), max_threads(std::min(inputs.size(), max_threads_)),
 		keys_size(aggregator.getNumberOfKeys()), aggregates_size(aggregator.getNumberOfAggregates()),
-		handler(*this), processor(inputs, max_threads, handler)
+		handler(*this), processor(inputs, additional_input_at_end, max_threads, handler)
 	{
-		children.insert(children.end(), inputs.begin(), inputs.end());
+		children = inputs;
+		if (additional_input_at_end)
+			children.push_back(additional_input_at_end);
 	}
 
 	String getName() const override { return "ParallelAggregating"; }

dbms/include/DB/DataStreams/ParallelInputsProcessor.h

@@ -45,8 +45,16 @@ template <typename Handler>
 class ParallelInputsProcessor
 {
 public:
-	ParallelInputsProcessor(BlockInputStreams inputs_, size_t max_threads_, Handler & handler_)
-		: inputs(inputs_), max_threads(std::min(inputs_.size(), max_threads_)), handler(handler_)
+	/** additional_input_at_end - если не nullptr,
+	  *  то из этого источника начинают доставаться блоки лишь после того, как все остальные источники обработаны.
+	  * Это делается в основном потоке.
+	  *
+	  * Предназначено для реализации FULL и RIGHT JOIN
+	  * - где нужно сначала параллельно сделать JOIN, при этом отмечая, какие ключи не найдены,
+	  *   и только после завершения этой работы, создать блоки из ненайденных ключей.
+	  */
+	ParallelInputsProcessor(BlockInputStreams inputs_, BlockInputStreamPtr additional_input_at_end_, size_t max_threads_, Handler & handler_)
+		: inputs(inputs_), additional_input_at_end(additional_input_at_end_), max_threads(std::min(inputs_.size(), max_threads_)), handler(handler_)
 	{
 		for (size_t i = 0; i < inputs_.size(); ++i)
 			available_inputs.emplace(inputs_[i], i);
@@ -150,6 +158,25 @@ private:
 		/// Последний поток при выходе сообщает, что данных больше нет.
 		if (0 == --active_threads)
 		{
+			/// И ещё обрабатывает дополнительный источник, если такой есть.
+			if (additional_input_at_end)
+			{
+				try
+				{
+					while (Block block = additional_input_at_end->read())
+						handler.onBlock(block, thread_num);
+				}
+				catch (...)
+				{
+					exception = cloneCurrentException();
+				}
+
+				if (exception)
+				{
+					handler.onException(exception, thread_num);
+				}
+			}
+
 			handler.onFinish();
 		}
 	}
@@ -206,6 +233,7 @@ private:
 	}
 
 	BlockInputStreams inputs;
+	BlockInputStreamPtr additional_input_at_end;
 	unsigned max_threads;
 
 	Handler & handler;

dbms/include/DB/DataStreams/RemoteBlockInputStream.h

@@ -84,6 +84,16 @@ public:
 		if (!is_cancelled.compare_exchange_strong(old_val, true, std::memory_order_seq_cst, std::memory_order_relaxed))
 			return;
 
+		{
+			std::lock_guard<std::mutex> lock(external_tables_mutex);
+
+			/// Останавливаем отправку внешних данных.
+			for (auto & vec : external_tables_data)
+				for (auto & elem : vec)
+					if (IProfilingBlockInputStream * stream = dynamic_cast<IProfilingBlockInputStream *>(elem.first.get()))
+						stream->cancel();
+		}
+
 		if (!isQueryPending() || hasThrownException())
 			return;
 
@@ -107,27 +117,30 @@ protected:
 	{
 		size_t count = parallel_replicas->size();
 
-		std::vector<ExternalTablesData> instances;
-		instances.reserve(count);
-
-		for (size_t i = 0; i < count; ++i)
 		{
-			ExternalTablesData res;
-			for (const auto & table : external_tables)
+			std::lock_guard<std::mutex> lock(external_tables_mutex);
+
+			external_tables_data.reserve(count);
+
+			for (size_t i = 0; i < count; ++i)
 			{
-				StoragePtr cur = table.second;
-				QueryProcessingStage::Enum stage = QueryProcessingStage::Complete;
-				DB::BlockInputStreams input = cur->read(cur->getColumnNamesList(), ASTPtr(), context, settings,
-					stage, DEFAULT_BLOCK_SIZE, 1);
-				if (input.size() == 0)
-					res.push_back(std::make_pair(new OneBlockInputStream(cur->getSampleBlock()), table.first));
-				else
-					res.push_back(std::make_pair(input[0], table.first));
+				ExternalTablesData res;
+				for (const auto & table : external_tables)
+				{
+					StoragePtr cur = table.second;
+					QueryProcessingStage::Enum stage = QueryProcessingStage::Complete;
+					DB::BlockInputStreams input = cur->read(cur->getColumnNamesList(), ASTPtr(), context, settings,
+						stage, DEFAULT_BLOCK_SIZE, 1);
+					if (input.size() == 0)
+						res.push_back(std::make_pair(new OneBlockInputStream(cur->getSampleBlock()), table.first));
+					else
+						res.push_back(std::make_pair(input[0], table.first));
+				}
+				external_tables_data.push_back(std::move(res));
 			}
-			instances.push_back(std::move(res));
 		}
 
-		parallel_replicas->sendExternalTablesData(instances);
+		parallel_replicas->sendExternalTablesData(external_tables_data);
 	}
 
 	Block readImpl() override
@@ -302,6 +315,10 @@ private:
 	QueryProcessingStage::Enum stage;
 	Context context;
 
+	/// Потоки для чтения из временных таблиц - для последующей отправки данных на удалённые серверы для GLOBAL-подзапросов.
+	std::vector<ExternalTablesData> external_tables_data;
+	std::mutex external_tables_mutex;
+
 	/// Установили соединения с репликами, но ещё не отправили запрос.
 	std::atomic<bool> established { false };
 

dbms/include/DB/DataStreams/UnionBlockInputStream.h

@@ -28,12 +28,14 @@ using Poco::SharedPtr;
 class UnionBlockInputStream : public IProfilingBlockInputStream
 {
 public:
-	UnionBlockInputStream(BlockInputStreams inputs, size_t max_threads) :
+	UnionBlockInputStream(BlockInputStreams inputs, BlockInputStreamPtr additional_input_at_end, size_t max_threads) :
 		output_queue(std::min(inputs.size(), max_threads)),
 		handler(*this),
-		processor(inputs, max_threads, handler)
+		processor(inputs, additional_input_at_end, max_threads, handler)
 	{
 		children = inputs;
+		if (additional_input_at_end)
+			children.push_back(additional_input_at_end);
 	}
 
 	String getName() const override { return "Union"; }

dbms/include/DB/Functions/FunctionsArray.h

@@ -33,14 +33,21 @@ namespace DB
   * has(arr, x) - есть ли в массиве элемент x.
   * indexOf(arr, x) - возвращает индекс элемента x (начиная с 1), если он есть в массиве, или 0, если его нет.
   * arrayEnumerate(arr) - возаращает массив [1,2,3,..., length(arr)]
-  * arrayEnumerateUniq(arr) - возаращает массив,  параллельный данному, где для каждого элемента указано,
-  * 						  какой он по счету среди элементов с таким значением.
-  * 						  Например: arrayEnumerateUniq([10, 20, 10, 30]) = [1,  1,  2,  1]
+  *
+  * arrayUniq(arr) - считает количество разных элементов в массиве,
+  * arrayUniq(arr1, arr2, ...) - считает количество разных кортежей из элементов на соответствующих позициях в нескольких массивах.
+  *
+  * arrayEnumerateUniq(arr)
+  *  - возаращает массив,  параллельный данному, где для каждого элемента указано,
+  *  какой он по счету среди элементов с таким значением.
+  *  Например: arrayEnumerateUniq([10, 20, 10, 30]) = [1,  1,  2,  1]
+  * arrayEnumerateUniq(arr1, arr2...)
+  *  - для кортежей из элементов на соответствующих позициях в нескольких массивах.
+  *
+  * emptyArrayToSingle(arr) - заменить пустые массивы на массивы из одного элемента со значением "по-умолчанию".
   */
 
 
-
-
 class FunctionArray : public IFunction
 {
 public:
@@ -995,11 +1002,11 @@ public:
 		{
 			const ColumnArray::Offsets_t & offsets = array->getOffsets();
 
-			ColumnVector<UInt32> * res_nested = new ColumnVector<UInt32>;
+			ColumnUInt32 * res_nested = new ColumnUInt32;
 			ColumnArray * res_array = new ColumnArray(res_nested, array->getOffsetsColumn());
 			block.getByPosition(result).column = res_array;
 
-			ColumnVector<UInt32>::Container_t & res_values = res_nested->getData();
+			ColumnUInt32::Container_t & res_values = res_nested->getData();
 			res_values.resize(array->getData().size());
 			size_t prev_off = 0;
 			for (size_t i = 0; i < offsets.size(); ++i)
@@ -1034,6 +1041,239 @@ public:
 	}
 };
 
+
+/// Считает количество разных элементов в массиве, или количество разных кортежей из элементов на соответствующих позициях в нескольких массивах.
+/// NOTE Реализация частично совпадает с arrayEnumerateUniq.
+class FunctionArrayUniq : public IFunction
+{
+public:
+	static constexpr auto name = "arrayUniq";
+	static IFunction * create(const Context & context) { return new FunctionArrayUniq; }
+
+	/// Получить имя функции.
+	String getName() const
+	{
+		return name;
+	}
+
+	/// Получить типы результата по типам аргументов. Если функция неприменима для данных аргументов - кинуть исключение.
+	DataTypePtr getReturnType(const DataTypes & arguments) const
+	{
+		if (arguments.size() == 0)
+			throw Exception("Number of arguments for function " + getName() + " doesn't match: passed "
+				+ toString(arguments.size()) + ", should be at least 1.",
+				ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
+		for (size_t i = 0; i < arguments.size(); ++i)
+		{
+			const DataTypeArray * array_type = typeid_cast<const DataTypeArray *>(&*arguments[i]);
+			if (!array_type)
+				throw Exception("All arguments for function " + getName() + " must be arrays; argument " + toString(i + 1) + " isn't.",
+					ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+		}
+
+		return new DataTypeUInt32;
+	}
+
+	/// Выполнить функцию над блоком.
+	void execute(Block & block, const ColumnNumbers & arguments, size_t result)
+	{
+		if (arguments.size() == 1 && executeConst(block, arguments, result))
+			return;
+
+		Columns array_columns(arguments.size());
+		const ColumnArray::Offsets_t * offsets = nullptr;
+		ConstColumnPlainPtrs data_columns(arguments.size());
+
+		for (size_t i = 0; i < arguments.size(); ++i)
+		{
+			ColumnPtr array_ptr = block.getByPosition(arguments[i]).column;
+			const ColumnArray * array = typeid_cast<const ColumnArray *>(&*array_ptr);
+			if (!array)
+			{
+				const ColumnConstArray * const_array = typeid_cast<const ColumnConstArray *>(&*block.getByPosition(arguments[i]).column);
+				if (!const_array)
+					throw Exception("Illegal column " + block.getByPosition(arguments[i]).column->getName()
+						+ " of " + toString(i + 1) + "-th argument of function " + getName(),
+						ErrorCodes::ILLEGAL_COLUMN);
+				array_ptr = const_array->convertToFullColumn();
+				array = typeid_cast<const ColumnArray *>(&*array_ptr);
+			}
+			array_columns[i] = array_ptr;
+			const ColumnArray::Offsets_t & offsets_i = array->getOffsets();
+			if (!i)
+				offsets = &offsets_i;
+			else if (offsets_i != *offsets)
+				throw Exception("Lengths of all arrays passsed to " + getName() + " must be equal.",
+					ErrorCodes::SIZES_OF_ARRAYS_DOESNT_MATCH);
+			data_columns[i] = &array->getData();
+		}
+
+		const ColumnArray * first_array = typeid_cast<const ColumnArray *>(&*array_columns[0]);
+		ColumnUInt32 * res = new ColumnUInt32;
+		block.getByPosition(result).column = res;
+
+		ColumnUInt32::Container_t & res_values = res->getData();
+		res_values.resize(offsets->size());
+
+		if (arguments.size() == 1)
+		{
+			if (!(	executeNumber<UInt8>	(first_array, res_values)
+				||	executeNumber<UInt16>	(first_array, res_values)
+				||	executeNumber<UInt32>	(first_array, res_values)
+				||	executeNumber<UInt64>	(first_array, res_values)
+				||	executeNumber<Int8>		(first_array, res_values)
+				||	executeNumber<Int16>	(first_array, res_values)
+				||	executeNumber<Int32>	(first_array, res_values)
+				||	executeNumber<Int64>	(first_array, res_values)
+				||	executeNumber<Float32>	(first_array, res_values)
+				||	executeNumber<Float64>	(first_array, res_values)
+				||	executeString			(first_array, res_values)))
+				throw Exception("Illegal column " + block.getByPosition(arguments[0]).column->getName()
+						+ " of first argument of function " + getName(),
+					ErrorCodes::ILLEGAL_COLUMN);
+		}
+		else
+		{
+			if (!execute128bit(*offsets, data_columns, res_values))
+				executeHashed(*offsets, data_columns, res_values);
+		}
+	}
+
+private:
+	/// Изначально выделить кусок памяти для 512 элементов.
+	static constexpr size_t INITIAL_SIZE_DEGREE = 9;
+
+	template <typename T>
+	bool executeNumber(const ColumnArray * array, ColumnUInt32::Container_t & res_values)
+	{
+		const ColumnVector<T> * nested = typeid_cast<const ColumnVector<T> *>(&array->getData());
+		if (!nested)
+			return false;
+		const ColumnArray::Offsets_t & offsets = array->getOffsets();
+		const typename ColumnVector<T>::Container_t & values = nested->getData();
+
+		typedef ClearableHashSet<T, DefaultHash<T>, HashTableGrower<INITIAL_SIZE_DEGREE>,
+			HashTableAllocatorWithStackMemory<(1 << INITIAL_SIZE_DEGREE) * sizeof(T)> > Set;
+
+		Set set;
+		size_t prev_off = 0;
+		for (size_t i = 0; i < offsets.size(); ++i)
+		{
+			set.clear();
+			size_t off = offsets[i];
+			for (size_t j = prev_off; j < off; ++j)
+				set.insert(values[j]);
+
+			res_values[i] = set.size();
+			prev_off = off;
+		}
+		return true;
+	}
+
+	bool executeString(const ColumnArray * array, ColumnUInt32::Container_t & res_values)
+	{
+		const ColumnString * nested = typeid_cast<const ColumnString *>(&array->getData());
+		if (!nested)
+			return false;
+		const ColumnArray::Offsets_t & offsets = array->getOffsets();
+
+		typedef ClearableHashSet<StringRef, StringRefHash, HashTableGrower<INITIAL_SIZE_DEGREE>,
+			HashTableAllocatorWithStackMemory<(1 << INITIAL_SIZE_DEGREE) * sizeof(StringRef)> > Set;
+
+		Set set;
+		size_t prev_off = 0;
+		for (size_t i = 0; i < offsets.size(); ++i)
+		{
+			set.clear();
+			size_t off = offsets[i];
+			for (size_t j = prev_off; j < off; ++j)
+				set.insert(nested->getDataAt(j));
+
+			res_values[i] = set.size();
+			prev_off = off;
+		}
+		return true;
+	}
+
+	bool executeConst(Block & block, const ColumnNumbers & arguments, size_t result)
+	{
+		const ColumnConstArray * array = typeid_cast<const ColumnConstArray *>(&*block.getByPosition(arguments[0]).column);
+		if (!array)
+			return false;
+		const Array & values = array->getData();
+
+		std::set<Field> set;
+		for (size_t i = 0; i < values.size(); ++i)
+			set.insert(values[i]);
+
+		block.getByPosition(result).column = new ColumnConstUInt32(array->size(), set.size());
+		return true;
+	}
+
+	bool execute128bit(
+		const ColumnArray::Offsets_t & offsets,
+		const ConstColumnPlainPtrs & columns,
+		ColumnUInt32::Container_t & res_values)
+	{
+		size_t count = columns.size();
+		size_t keys_bytes = 0;
+		Sizes key_sizes(count);
+		for (size_t j = 0; j < count; ++j)
+		{
+			if (!columns[j]->isFixed())
+				return false;
+			key_sizes[j] = columns[j]->sizeOfField();
+			keys_bytes += key_sizes[j];
+		}
+		if (keys_bytes > 16)
+			return false;
+
+		typedef ClearableHashSet<UInt128, UInt128HashCRC32, HashTableGrower<INITIAL_SIZE_DEGREE>,
+			HashTableAllocatorWithStackMemory<(1 << INITIAL_SIZE_DEGREE) * sizeof(UInt128)> > Set;
+
+		Set set;
+		size_t prev_off = 0;
+		for (size_t i = 0; i < offsets.size(); ++i)
+		{
+			set.clear();
+			size_t off = offsets[i];
+			for (size_t j = prev_off; j < off; ++j)
+				set.insert(packFixed<UInt128>(j, count, columns, key_sizes));
+
+			res_values[i] = set.size();
+			prev_off = off;
+		}
+
+		return true;
+	}
+
+	void executeHashed(
+		const ColumnArray::Offsets_t & offsets,
+		const ConstColumnPlainPtrs & columns,
+		ColumnUInt32::Container_t & res_values)
+	{
+		size_t count = columns.size();
+
+		typedef ClearableHashSet<UInt128, UInt128TrivialHash, HashTableGrower<INITIAL_SIZE_DEGREE>,
+			HashTableAllocatorWithStackMemory<(1 << INITIAL_SIZE_DEGREE) * sizeof(UInt128)> > Set;
+
+		Set set;
+		size_t prev_off = 0;
+		for (size_t i = 0; i < offsets.size(); ++i)
+		{
+			set.clear();
+			size_t off = offsets[i];
+			for (size_t j = prev_off; j < off; ++j)
+				set.insert(hash128(j, count, columns));
+
+			res_values[i] = set.size();
+			prev_off = off;
+		}
+	}
+};
+
+
 class FunctionArrayEnumerateUniq : public IFunction
 {
 public:
@@ -1100,11 +1340,11 @@ public:
 		}
 
 		const ColumnArray * first_array = typeid_cast<const ColumnArray *>(&*array_columns[0]);
-		ColumnVector<UInt32> * res_nested = new ColumnVector<UInt32>;
+		ColumnUInt32 * res_nested = new ColumnUInt32;
 		ColumnArray * res_array = new ColumnArray(res_nested, first_array->getOffsetsColumn());
 		block.getByPosition(result).column = res_array;
 
-		ColumnVector<UInt32>::Container_t & res_values = res_nested->getData();
+		ColumnUInt32::Container_t & res_values = res_nested->getData();
 		if (!offsets->empty())
 			res_values.resize(offsets->back());
 
@@ -1137,7 +1377,7 @@ private:
 	static constexpr size_t INITIAL_SIZE_DEGREE = 9;
 
 	template <typename T>
-	bool executeNumber(const ColumnArray * array, ColumnVector<UInt32>::Container_t & res_values)
+	bool executeNumber(const ColumnArray * array, ColumnUInt32::Container_t & res_values)
 	{
 		const ColumnVector<T> * nested = typeid_cast<const ColumnVector<T> *>(&array->getData());
 		if (!nested)
@@ -1163,7 +1403,7 @@ private:
 		return true;
 	}
 
-	bool executeString(const ColumnArray * array, ColumnVector<UInt32>::Container_t & res_values)
+	bool executeString(const ColumnArray * array, ColumnUInt32::Container_t & res_values)
 	{
 		const ColumnString * nested = typeid_cast<const ColumnString *>(&array->getData());
 		if (!nested)
@@ -1211,7 +1451,7 @@ private:
 	bool execute128bit(
 		const ColumnArray::Offsets_t & offsets,
 		const ConstColumnPlainPtrs & columns,
-		ColumnVector<UInt32>::Container_t & res_values)
+		ColumnUInt32::Container_t & res_values)
 	{
 		size_t count = columns.size();
 		size_t keys_bytes = 0;
@@ -1248,7 +1488,7 @@ private:
 	void executeHashed(
 		const ColumnArray::Offsets_t & offsets,
 		const ConstColumnPlainPtrs & columns,
-		ColumnVector<UInt32>::Container_t & res_values)
+		ColumnUInt32::Container_t & res_values)
 	{
 		size_t count = columns.size();
 
@@ -1457,13 +1697,263 @@ private:
 };
 
 
+class FunctionEmptyArrayToSingle : public IFunction
+{
+public:
+	static constexpr auto name = "emptyArrayToSingle";
+	static IFunction * create(const Context & context) { return new FunctionEmptyArrayToSingle; }
+
+	/// Получить имя функции.
+	String getName() const
+	{
+		return name;
+	}
+
+	/// Получить типы результата по типам аргументов. Если функция неприменима для данных аргументов - кинуть исключение.
+	DataTypePtr getReturnType(const DataTypes & arguments) const
+	{
+		if (arguments.size() != 1)
+			throw Exception("Number of arguments for function " + getName() + " doesn't match: passed "
+				+ toString(arguments.size()) + ", should be 1.",
+				ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
+
+		const DataTypeArray * array_type = typeid_cast<const DataTypeArray *>(arguments[0].get());
+		if (!array_type)
+			throw Exception("Argument for function " + getName() + " must be array.",
+				ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
+
+		return arguments[0]->clone();
+	}
+
+	/// Выполнить функцию над блоком.
+	void execute(Block & block, const ColumnNumbers & arguments, size_t result)
+	{
+		if (executeConst(block, arguments, result))
+			return;
+
+		const ColumnArray * array = typeid_cast<const ColumnArray *>(block.getByPosition(arguments[0]).column.get());
+		if (!array)
+			throw Exception("Illegal column " + block.getByPosition(arguments[0]).column->getName() + " of first argument of function " + getName(),
+				ErrorCodes::ILLEGAL_COLUMN);
+
+		ColumnPtr res_ptr = array->cloneEmpty();
+		block.getByPosition(result).column = res_ptr;
+		ColumnArray & res = static_cast<ColumnArray &>(*res_ptr);
+
+		const IColumn & src_data = array->getData();
+		const ColumnArray::Offsets_t & src_offsets = array->getOffsets();
+		IColumn & res_data = res.getData();
+		ColumnArray::Offsets_t & res_offsets = res.getOffsets();
+
+		if (!(	executeNumber<UInt8>	(src_data, src_offsets, res_data, res_offsets)
+			||	executeNumber<UInt16>	(src_data, src_offsets, res_data, res_offsets)
+			||	executeNumber<UInt32>	(src_data, src_offsets, res_data, res_offsets)
+			||	executeNumber<UInt64>	(src_data, src_offsets, res_data, res_offsets)
+			||	executeNumber<Int8>		(src_data, src_offsets, res_data, res_offsets)
+			||	executeNumber<Int16>	(src_data, src_offsets, res_data, res_offsets)
+			||	executeNumber<Int32>	(src_data, src_offsets, res_data, res_offsets)
+			||	executeNumber<Int64>	(src_data, src_offsets, res_data, res_offsets)
+			||	executeNumber<Float32>	(src_data, src_offsets, res_data, res_offsets)
+			||	executeNumber<Float64>	(src_data, src_offsets, res_data, res_offsets)
+			||	executeString			(src_data, src_offsets, res_data, res_offsets)
+			||	executeFixedString		(src_data, src_offsets, res_data, res_offsets)))
+			throw Exception("Illegal column " + block.getByPosition(arguments[0]).column->getName()
+				+ " of first argument of function " + getName(),
+				ErrorCodes::ILLEGAL_COLUMN);
+	}
+
+private:
+	bool executeConst(Block & block, const ColumnNumbers & arguments, size_t result)
+	{
+		if (const ColumnConstArray * const_array = typeid_cast<const ColumnConstArray *>(block.getByPosition(arguments[0]).column.get()))
+		{
+			if (const_array->getData().empty())
+			{
+				auto nested_type = typeid_cast<const DataTypeArray &>(*block.getByPosition(arguments[0]).type).getNestedType();
+
+				block.getByPosition(result).column = new ColumnConstArray(
+					block.rowsInFirstColumn(),
+					{nested_type->getDefault()},
+					nested_type->clone());
+			}
+			else
+				block.getByPosition(result).column = block.getByPosition(arguments[0]).column;
+
+			return true;
+		}
+		else
+			return false;
+	}
+
+	template <typename T>
+	bool executeNumber(
+		const IColumn & src_data, const ColumnArray::Offsets_t & src_offsets,
+		IColumn & res_data_col, ColumnArray::Offsets_t & res_offsets)
+	{
+		if (const ColumnVector<T> * src_data_concrete = typeid_cast<const ColumnVector<T> *>(&src_data))
+		{
+			const PODArray<T> & src_data = src_data_concrete->getData();
+			PODArray<T> & res_data = typeid_cast<ColumnVector<T> &>(res_data_col).getData();
+			size_t size = src_offsets.size();
+			res_offsets.resize(size);
+			res_data.reserve(src_data.size());
+
+			ColumnArray::Offset_t src_prev_offset = 0;
+			ColumnArray::Offset_t res_prev_offset = 0;
+
+			for (size_t i = 0; i < size; ++i)
+			{
+				if (src_offsets[i] != src_prev_offset)
+				{
+					size_t size_to_write = src_offsets[i] - src_prev_offset;
+					size_t prev_res_data_size = res_data.size();
+					res_data.resize(prev_res_data_size + size_to_write);
+					memcpy(&res_data[prev_res_data_size], &src_data[src_prev_offset], size_to_write * sizeof(T));
+					res_prev_offset += size_to_write;
+					res_offsets[i] = res_prev_offset;
+				}
+				else
+				{
+					res_data.push_back(T());
+					++res_prev_offset;
+					res_offsets[i] = res_prev_offset;
+				}
+
+				src_prev_offset = src_offsets[i];
+			}
+
+			return true;
+		}
+		else
+			return false;
+	}
+
+	bool executeFixedString(
+		const IColumn & src_data, const ColumnArray::Offsets_t & src_offsets,
+		IColumn & res_data_col, ColumnArray::Offsets_t & res_offsets)
+	{
+		if (const ColumnFixedString * src_data_concrete = typeid_cast<const ColumnFixedString *>(&src_data))
+		{
+			const size_t n = src_data_concrete->getN();
+			const ColumnFixedString::Chars_t & src_data = src_data_concrete->getChars();
+			ColumnFixedString::Chars_t & res_data = typeid_cast<ColumnFixedString &>(res_data_col).getChars();
+			size_t size = src_offsets.size();
+			res_offsets.resize(size);
+			res_data.reserve(src_data.size());
+
+			ColumnArray::Offset_t src_prev_offset = 0;
+			ColumnArray::Offset_t res_prev_offset = 0;
+
+			for (size_t i = 0; i < size; ++i)
+			{
+				if (src_offsets[i] != src_prev_offset)
+				{
+					size_t size_to_write = src_offsets[i] - src_prev_offset;
+					size_t prev_res_data_size = res_data.size();
+					res_data.resize(prev_res_data_size + size_to_write * n);
+					memcpy(&res_data[prev_res_data_size], &src_data[src_prev_offset], size_to_write * n);
+					res_prev_offset += size_to_write;
+					res_offsets[i] = res_prev_offset;
+				}
+				else
+				{
+					size_t prev_res_data_size = res_data.size();
+					res_data.resize(prev_res_data_size + n);
+					memset(&res_data[prev_res_data_size], 0, n);
+					++res_prev_offset;
+					res_offsets[i] = res_prev_offset;
+				}
+
+				src_prev_offset = src_offsets[i];
+			}
+
+			return true;
+		}
+		else
+			return false;
+	}
+
+	bool executeString(
+		const IColumn & src_data, const ColumnArray::Offsets_t & src_array_offsets,
+		IColumn & res_data_col, ColumnArray::Offsets_t & res_array_offsets)
+	{
+		if (const ColumnString * src_data_concrete = typeid_cast<const ColumnString *>(&src_data))
+		{
+			const ColumnString::Offsets_t & src_string_offsets = src_data_concrete->getOffsets();
+			ColumnString::Offsets_t & res_string_offsets = typeid_cast<ColumnString &>(res_data_col).getOffsets();
+
+			const ColumnString::Chars_t & src_data = src_data_concrete->getChars();
+			ColumnString::Chars_t & res_data = typeid_cast<ColumnString &>(res_data_col).getChars();
+
+			size_t size = src_array_offsets.size();
+			res_array_offsets.resize(size);
+			res_string_offsets.reserve(src_string_offsets.size());
+			res_data.reserve(src_data.size());
+
+			ColumnArray::Offset_t src_array_prev_offset = 0;
+			ColumnArray::Offset_t res_array_prev_offset = 0;
+
+			ColumnString::Offset_t src_string_prev_offset = 0;
+			ColumnString::Offset_t res_string_prev_offset = 0;
+
+			for (size_t i = 0; i < size; ++i)
+			{
+				if (src_array_offsets[i] != src_array_prev_offset)
+				{
+					size_t array_size = src_array_offsets[i] - src_array_prev_offset;
+
+					size_t bytes_to_copy = 0;
+					size_t from_string_prev_offset_local = src_string_prev_offset;
+					for (size_t j = 0; j < array_size; ++j)
+					{
+						size_t string_size = src_string_offsets[src_array_prev_offset + j] - from_string_prev_offset_local;
+
+						res_string_prev_offset += string_size;
+						res_string_offsets.push_back(res_string_prev_offset);
+
+						from_string_prev_offset_local += string_size;
+						bytes_to_copy += string_size;
+					}
+
+					size_t res_data_old_size = res_data.size();
+					res_data.resize(res_data_old_size + bytes_to_copy);
+					memcpy(&res_data[res_data_old_size], &src_data[src_string_prev_offset], bytes_to_copy);
+
+					res_array_prev_offset += array_size;
+					res_array_offsets[i] = res_array_prev_offset;
+				}
+				else
+				{
+					res_data.push_back(0);	/// Пустая строка, включая ноль на конце.
+
+					++res_string_prev_offset;
+					res_string_offsets.push_back(res_string_prev_offset);
+
+					++res_array_prev_offset;
+					res_array_offsets[i] = res_array_prev_offset;
+				}
+
+				src_array_prev_offset = src_array_offsets[i];
+
+				if (src_array_prev_offset)
+					src_string_prev_offset = src_string_offsets[src_array_prev_offset - 1];
+			}
+
+			return true;
+		}
+		else
+			return false;
+	}
+};
+
+
 struct NameHas			{ static constexpr auto name = "has"; };
 struct NameIndexOf		{ static constexpr auto name = "indexOf"; };
 struct NameCountEqual	{ static constexpr auto name = "countEqual"; };
 
-typedef FunctionArrayIndex<IndexToOne, 		NameHas>	FunctionHas;
+typedef FunctionArrayIndex<IndexToOne, 		NameHas>		FunctionHas;
 typedef FunctionArrayIndex<IndexIdentity, 	NameIndexOf>	FunctionIndexOf;
-typedef FunctionArrayIndex<IndexCount, 	NameCountEqual>	FunctionCountEqual;
+typedef FunctionArrayIndex<IndexCount, 	NameCountEqual>		FunctionCountEqual;
 
 using FunctionEmptyArrayUInt8 = FunctionEmptyArray<DataTypeUInt8>;
 using FunctionEmptyArrayUInt16 = FunctionEmptyArray<DataTypeUInt16>;

dbms/include/DB/Functions/FunctionsURL.h

@@ -404,31 +404,36 @@ struct ExtractURLParameterImpl
 		{
 			size_t cur_offset = offsets[i];
 
+			const char * str = reinterpret_cast<const char *>(&data[prev_offset]);
+
 			const char * pos = nullptr;
-
-			do
+			const char * begin = strchr(str, '?');
+			if (begin != nullptr)
 			{
-				const char * str = reinterpret_cast<const char *>(&data[prev_offset]);
-
-				const char * begin = strchr(str, '?');
-				if (begin == nullptr)
-					break;
-
-				pos = strstr(begin + 1, param_str);
-				if (pos == nullptr)
-					break;
-				if (pos != begin + 1 && *(pos - 1) != ';' && *(pos - 1) != '&')
+				pos = begin + 1;
+				while (true)
 				{
-					pos = nullptr;
-					break;
-				}
+					pos = strstr(pos, param_str);
 
-				pos += param_len;
-			} while (false);
+					if (pos == nullptr)
+						break;
+
+					if (pos[-1] != '?' && pos[-1] != '&')
+					{
+						pos += param_len;
+						continue;
+					}
+					else
+					{
+						pos += param_len;
+						break;
+					}
+				}
+			}
 
 			if (pos != nullptr)
 			{
-				const char * end = strpbrk(pos, "&;#");
+				const char * end = strpbrk(pos, "&#");
 				if (end == nullptr)
 					end = pos + strlen(pos);
 

dbms/include/DB/Interpreters/ExpressionActions.h

@@ -66,6 +66,7 @@ public:
 
 	/// Для ARRAY_JOIN
 	NameSet array_joined_columns;
+	bool array_join_is_left;
 
 	/// Для JOIN
 	const Join * join = nullptr;
@@ -122,13 +123,14 @@ public:
 		return a;
 	}
 
-	static ExpressionAction arrayJoin(const NameSet & array_joined_columns)
+	static ExpressionAction arrayJoin(const NameSet & array_joined_columns, bool array_join_is_left)
 	{
 		if (array_joined_columns.empty())
 			throw Exception("No arrays to join", ErrorCodes::LOGICAL_ERROR);
 		ExpressionAction a;
 		a.type = ARRAY_JOIN;
 		a.array_joined_columns = array_joined_columns;
+		a.array_join_is_left = array_join_is_left;
 		return a;
 	}
 

dbms/include/DB/Interpreters/ExpressionAnalyzer.h

@@ -226,6 +226,10 @@ private:
 	/// Превратить перечисление значений или подзапрос в ASTSet. node - функция in или notIn.
 	void makeSet(ASTFunction * node, const Block & sample_block);
 
+	/// Замена скалярных подзапросов на значения-константы.
+	void executeScalarSubqueries();
+	void executeScalarSubqueriesImpl(ASTPtr & ast);
+
 	/// Находит глобальные подзапросы в секциях GLOBAL IN/JOIN. Заполняет external_tables.
 	void initGlobalSubqueriesAndExternalTables();
 	void initGlobalSubqueries(ASTPtr & ast);

dbms/include/DB/Interpreters/InterpreterCreateQuery.h

@@ -24,8 +24,7 @@ public:
 	  */
 	BlockIO execute() override
 	{
-		executeImpl(false);
-		return {};
+		return executeImpl(false);
 	}
 
 	/** assume_metadata_exists - не проверять наличие файла с метаданными и не создавать его
@@ -45,7 +44,7 @@ public:
 		const ColumnDefaults & column_defaults);
 
 private:
-	void executeImpl(bool assume_metadata_exists);
+	BlockIO executeImpl(bool assume_metadata_exists);
 
 	/// AST в список столбцов с типами. Столбцы типа Nested развернуты в список настоящих столбцов.
 	using ColumnsAndDefaults = std::pair<NamesAndTypesList, ColumnDefaults>;

dbms/include/DB/Interpreters/InterpreterSelectQuery.h

@@ -109,7 +109,7 @@ private:
 	// Переименовать столбцы каждого запроса цепочки UNION ALL в такие же имена, как в первом запросе.
 	void renameColumns();
 
-	/** Из какой таблицы читать. При JOIN, возвращается "левая" таблицы.
+	/** Из какой таблицы читать. При JOIN, возвращается "левая" таблица.
 	 */
 	void getDatabaseAndTableNames(String & database_name, String & table_name);
 
@@ -120,22 +120,43 @@ private:
 	/// Разные стадии выполнения запроса.
 
 	/// Вынимает данные из таблицы. Возвращает стадию, до которой запрос был обработан в Storage.
-	QueryProcessingStage::Enum executeFetchColumns(BlockInputStreams & streams);
+	QueryProcessingStage::Enum executeFetchColumns();
+
+	void executeWhere(ExpressionActionsPtr expression);
+	void executeAggregation(ExpressionActionsPtr expression, bool overflow_row, bool final);
+	void executeMergeAggregated(bool overflow_row, bool final);
+	void executeTotalsAndHaving(bool has_having, ExpressionActionsPtr expression, bool overflow_row);
+	void executeHaving(ExpressionActionsPtr expression);
+	void executeExpression(ExpressionActionsPtr expression);
+	void executeOrder();
+	void executeMergeSorted();
+	void executePreLimit();
+	void executeUnion();
+	void executeLimit();
+	void executeProjection(ExpressionActionsPtr expression);
+	void executeDistinct(bool before_order, Names columns);
+	void executeSubqueriesInSetsAndJoins(std::unordered_map<String, SubqueryForSet> & subqueries_for_sets);
+
+	template <typename Transform>
+	void transformStreams(Transform && transform)
+	{
+		for (auto & stream : streams)
+			transform(stream);
+
+		if (stream_with_non_joined_data)
+			transform(stream_with_non_joined_data);
+	}
+
+	bool hasNoData() const
+	{
+		return streams.empty() && !stream_with_non_joined_data;
+	}
+
+	bool hasMoreThanOneStream() const
+	{
+		return streams.size() + (stream_with_non_joined_data ? 1 : 0) > 1;
+	}
 
-	void executeWhere(                   BlockInputStreams & streams, ExpressionActionsPtr expression);
-	void executeAggregation(             BlockInputStreams & streams, ExpressionActionsPtr expression, bool overflow_row, bool final);
-	void executeMergeAggregated(         BlockInputStreams & streams, bool overflow_row, bool final);
-	void executeTotalsAndHaving(         BlockInputStreams & streams, bool has_having, ExpressionActionsPtr expression, bool overflow_row);
-	void executeHaving(                  BlockInputStreams & streams, ExpressionActionsPtr expression);
-	void executeExpression(              BlockInputStreams & streams, ExpressionActionsPtr expression);
-	void executeOrder(                   BlockInputStreams & streams);
-	void executeMergeSorted(             BlockInputStreams & streams);
-	void executePreLimit(                BlockInputStreams & streams);
-	void executeUnion(                   BlockInputStreams & streams);
-	void executeLimit(                   BlockInputStreams & streams);
-	void executeProjection(              BlockInputStreams & streams, ExpressionActionsPtr expression);
-	void executeDistinct(                BlockInputStreams & streams, bool before_order, Names columns);
-	void executeSubqueriesInSetsAndJoins(BlockInputStreams & streams, std::unordered_map<String, SubqueryForSet> & subqueries_for_sets);
 
 	void ignoreWithTotals();
 
@@ -156,9 +177,21 @@ private:
 	QueryProcessingStage::Enum to_stage;
 	size_t subquery_depth;
 	std::unique_ptr<ExpressionAnalyzer> query_analyzer;
-	BlockInputStreams streams;
 	NamesAndTypesList table_column_names;
 
+	/** Потоки данных.
+	  * Исходные потоки данных получаются в функции executeFetchColumns.
+	  * Затем они преобразуются (оборачиваются в другие потоки) с помощью функций execute*,
+	  *  чтобы получить целый конвейер выполнения запроса.
+	  */
+	BlockInputStreams streams;
+
+	/** При выполнении FULL или RIGHT JOIN, здесь будет поток данных, из которого можно прочитать "неприсоединённые" строки.
+	  * Он имеет особое значение, так как чтение из него должно осуществляться после чтения из основных потоков.
+	  * Он подклеивается к основным потокам в UnionBlockInputStream или ParallelAggregatingBlockInputStream.
+	  */
+	BlockInputStreamPtr stream_with_non_joined_data;
+
 	/// Являемся ли мы первым запросом SELECT цепочки UNION ALL?
 	bool is_first_select_inside_union_all;
 

dbms/include/DB/Interpreters/Join.h

@@ -18,8 +18,9 @@ namespace DB
 
 /** Структура данных для реализации JOIN-а.
   * По сути, хэш-таблица: ключи -> строки присоединяемой таблицы.
+  * Исключение - CROSS JOIN, где вместо хэш-таблицы просто набор блоков без ключей.
   *
-  * JOIN-ы бывают восьми типов: ANY/ALL x LEFT/INNER/RIGHT/FULL.
+  * JOIN-ы бывают девяти типов: ANY/ALL × LEFT/INNER/RIGHT/FULL, а также CROSS.
   *
   * Если указано ANY - выбрать из "правой" таблицы только одну, первую попавшуюся строку, даже если там более одной соответствующей строки.
   * Если указано ALL - обычный вариант JOIN-а, при котором строки могут размножаться по числу соответствующих строк "правой" таблицы.
@@ -213,6 +214,7 @@ private:
 		KEY_64,
 		KEY_STRING,
 		HASHED,
+		CROSS,
 	};
 
 	Type type = Type::EMPTY;
@@ -249,6 +251,8 @@ private:
 	template <ASTJoin::Kind KIND, ASTJoin::Strictness STRICTNESS, typename Maps>
 	void joinBlockImpl(Block & block, const Maps & maps) const;
 
+	void joinBlockImplCross(Block & block) const;
+
 	/// Проверить не превышены ли допустимые размеры множества
 	bool checkSizeLimits() const;
 

dbms/include/DB/Interpreters/evaluateMissingDefaults.h

@@ -41,9 +41,14 @@ inline void evaluateMissingDefaults(Block & block,
 	/// evaluate default values for defaulted columns
 	ExpressionAnalyzer{default_expr_list, context, {}, required_columns}.getActions(true)->execute(copy_block);
 
-	/// move evaluated columns to the original block
+	/// move evaluated columns to the original block, materializing them at the same time
 	for (auto & column_name_type : copy_block.getColumns())
+	{
+		if (column_name_type.column->isConst())
+			column_name_type.column = static_cast<const IColumnConst &>(*column_name_type.column).convertToFullColumn();
+
 		block.insert(std::move(column_name_type));
+	}
 }
 
 }

dbms/include/DB/Parsers/ASTJoin.h

@@ -32,7 +32,8 @@ public:
 		Inner,	/// Оставить только записи, для которых в "правой" таблице есть соответствующая.
 		Left,	/// Если в "правой" таблице нет соответствующих записей, заполнить столбцы значениями "по-умолчанию".
 		Right,
-		Full
+		Full,
+		Cross	/// Прямое произведение. strictness и using_expr_list не используются.
 	};
 
 	Locality locality = Local;
@@ -61,7 +62,8 @@ public:
 				kind == Inner ? "Inner"
 				: (kind == Left ? "Left"
 				: (kind == Right ? "Right"
-				: "Full")), wb);
+				: (kind == Full ? "Full"
+				: "Cross"))), wb);
 
 			writeString("Join", wb);
 		}

dbms/include/DB/Parsers/ASTSelectQuery.h

@@ -50,6 +50,7 @@ public:
 	ASTPtr select_expression_list;
 	ASTPtr database;
 	ASTPtr table;	/// Идентификатор, табличная функция или подзапрос (рекурсивно ASTSelectQuery)
+	bool array_join_is_left = false;	/// LEFT ARRAY JOIN
 	ASTPtr array_join_expression_list;	/// ARRAY JOIN
 	ASTPtr join;						/// Обычный (не ARRAY) JOIN.
 	bool final = false;

dbms/include/DB/Parsers/ASTSubquery.h

@@ -2,7 +2,7 @@
 
 #include <DB/DataTypes/IDataType.h>
 
-#include <DB/Parsers/IAST.h>
+#include <DB/Parsers/ASTWithAlias.h>
 
 
 namespace DB
@@ -11,12 +11,12 @@ namespace DB
 
 /** Подзарос SELECT
   */
-class ASTSubquery : public IAST
+class ASTSubquery : public ASTWithAlias
 {
 public:
 	ASTSubquery() = default;
-	ASTSubquery(const StringRange range_) : IAST(range_) {}
-	
+	ASTSubquery(const StringRange range_) : ASTWithAlias(range_) {}
+
 	/** Получить текст, который идентифицирует этот элемент. */
 	String getID() const override { return "Subquery"; }
 

dbms/include/DB/Storages/MergeTree/MergeTreeData.h

@@ -779,6 +779,13 @@ public:
 		return it == std::end(column_sizes) ? 0 : it->second;
 	}
 
+	using ColumnSizes = std::unordered_map<std::string, size_t>;
+	ColumnSizes getColumnSizes() const
+	{
+		Poco::ScopedLock<Poco::FastMutex> lock{data_parts_mutex};
+		return column_sizes;
+	}
+
 	/// Для ATTACH/DETACH/DROP PARTITION.
 	static String getMonthName(const Field & partition);
 	static DayNum_t getMonthDayNum(const Field & partition);
@@ -810,7 +817,7 @@ private:
 
 	NamesAndTypesListPtr columns;
 	/// Актуальные размеры столбцов в сжатом виде
-	std::unordered_map<std::string, size_t> column_sizes;
+	ColumnSizes column_sizes;
 
 	BrokenPartCallback broken_part_callback;
 

dbms/include/DB/Storages/MergeTree/MergeTreeReader.h

@@ -247,7 +247,7 @@ private:
 						++right;
 				}
 
-				/// Если правее засечек нет, просто используем DEFAULT_BUFFER_SIZE
+				/// Если правее засечек нет, просто используем max_read_buffer_size
 				if (right >= (*marks).size() || (right + 1 == (*marks).size() &&
 					(*marks)[right].offset_in_compressed_file == (*marks)[all_mark_ranges[i].end].offset_in_compressed_file))
 				{

dbms/src/AggregateFunctions/AggregateFunctionFactory.cpp

@@ -652,7 +652,7 @@ AggregateFunctionPtr AggregateFunctionFactory::get(const String & name, const Da
 		AggregateFunctionPtr nested = get(String(name.data(), name.size() - strlen("State")), argument_types, recursion_level + 1);
 		return new AggregateFunctionState(nested);
 	}
-	else if (recursion_level == 0 && name.size() > strlen("Merge") && !(strcmp(name.data() + name.size() - strlen("Merge"), "Merge")))
+	else if (recursion_level <= 1 && name.size() > strlen("Merge") && !(strcmp(name.data() + name.size() - strlen("Merge"), "Merge")))
 	{
 		/// Для агрегатных функций вида aggMerge, где agg - имя другой агрегатной функции.
 		if (argument_types.size() != 1)
@@ -668,7 +668,7 @@ AggregateFunctionPtr AggregateFunctionFactory::get(const String & name, const Da
 
 		return new AggregateFunctionMerge(nested);
 	}
-	else if (recursion_level <= 1 && name.size() >= 3 && name[name.size() - 2] == 'I' && name[name.size() - 1] == 'f')
+	else if (recursion_level <= 2 && name.size() >= 3 && name[name.size() - 2] == 'I' && name[name.size() - 1] == 'f')
 	{
 		if (argument_types.empty())
 			throw Exception{
@@ -682,7 +682,7 @@ AggregateFunctionPtr AggregateFunctionFactory::get(const String & name, const Da
 		AggregateFunctionPtr nested = get(String(name.data(), name.size() - 2), nested_dt, recursion_level + 1);
 		return new AggregateFunctionIf(nested);
 	}
-	else if (recursion_level <= 2 && name.size() > strlen("Array") && !(strcmp(name.data() + name.size() - strlen("Array"), "Array")))
+	else if (recursion_level <= 3 && name.size() > strlen("Array") && !(strcmp(name.data() + name.size() - strlen("Array"), "Array")))
 	{
 		/// Для агрегатных функций вида aggArray, где agg - имя другой агрегатной функции.
 		size_t num_agruments = argument_types.size();
@@ -695,7 +695,7 @@ AggregateFunctionPtr AggregateFunctionFactory::get(const String & name, const Da
 			else
 				throw Exception("Illegal type " + argument_types[i]->getName() + " of argument #" + toString(i + 1) + " for aggregate function " + name + ". Must be array.", ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
 		}
-		AggregateFunctionPtr nested = get(String(name.data(), name.size() - strlen("Array")), nested_arguments, recursion_level + 2); /// + 2, чтобы ни один другой модификатор не мог идти перед Array
+		AggregateFunctionPtr nested = get(String(name.data(), name.size() - strlen("Array")), nested_arguments, recursion_level + 3); /// + 3, чтобы ни один другой модификатор не мог идти перед Array
 		return new AggregateFunctionArray(nested);
 	}
 	else
@@ -765,14 +765,14 @@ bool AggregateFunctionFactory::isAggregateFunctionName(const String & name, int
 	if (recursion_level <= 0 && name.size() > strlen("State") && !(strcmp(name.data() + name.size() - strlen("State"), "State")))
 		return isAggregateFunctionName(String(name.data(), name.size() - strlen("State")), recursion_level + 1);
 	/// Для агрегатных функций вида aggMerge, где agg - имя другой агрегатной функции.
-	if (recursion_level <= 0 && name.size() > strlen("Merge") && !(strcmp(name.data() + name.size() - strlen("Merge"), "Merge")))
+	if (recursion_level <= 1 && name.size() > strlen("Merge") && !(strcmp(name.data() + name.size() - strlen("Merge"), "Merge")))
 		return isAggregateFunctionName(String(name.data(), name.size() - strlen("Merge")), recursion_level + 1);
 	/// Для агрегатных функций вида aggIf, где agg - имя другой агрегатной функции.
-	if (recursion_level <= 1 && name.size() >= 3 && name[name.size() - 2] == 'I' && name[name.size() - 1] == 'f')
+	if (recursion_level <= 2 && name.size() >= 3 && name[name.size() - 2] == 'I' && name[name.size() - 1] == 'f')
 		return isAggregateFunctionName(String(name.data(), name.size() - 2), recursion_level + 1);
 	/// Для агрегатных функций вида aggArray, где agg - имя другой агрегатной функции.
-	if (recursion_level <= 2 && name.size() > strlen("Array") && !(strcmp(name.data() + name.size() - strlen("Array"), "Array")))
-		return isAggregateFunctionName(String(name.data(), name.size() - strlen("Array")), recursion_level + 2); /// + 2, чтобы ни один другой модификатор не мог идти перед Array
+	if (recursion_level <= 3 && name.size() > strlen("Array") && !(strcmp(name.data() + name.size() - strlen("Array"), "Array")))
+		return isAggregateFunctionName(String(name.data(), name.size() - strlen("Array")), recursion_level + 3); /// + 3, чтобы ни один другой модификатор не мог идти перед Array
 
 	return false;
 }

dbms/src/Client/Client.cpp

@@ -369,6 +369,19 @@ private:
 	}
 
 
+	/** Проверка для случая, когда в терминал вставляется многострочный запрос из буфера обмена.
+	  * Позволяет не начинать выполнение одной строчки запроса, пока весь запрос не будет вставлен.
+	  */
+	static bool hasDataInSTDIN()
+	{
+		timeval timeout = { 0, 0 };
+		fd_set fds;
+		FD_ZERO(&fds);
+		FD_SET(STDIN_FILENO, &fds);
+		return select(1, &fds, 0, 0, &timeout) == 1;
+	}
+
+
 	void loop()
 	{
 		String query;
@@ -395,7 +408,7 @@ private:
 
 			query += line;
 
-			if (!ends_with_backslash && (ends_with_semicolon || has_vertical_output_suffix || !config().has("multiline")))
+			if (!ends_with_backslash && (ends_with_semicolon || has_vertical_output_suffix || (!config().has("multiline") && !hasDataInSTDIN())))
 			{
 				if (query != prev_query)
 				{
@@ -464,6 +477,12 @@ private:
 			copyData(in, out);
 		}
 
+		process(line);
+	}
+
+
+	bool process(const String & line)
+	{
 		if (config().has("multiquery"))
 		{
 			/// Несколько запросов, разделенных ';'.
@@ -494,17 +513,20 @@ private:
 				while (isWhitespace(*begin) || *begin == ';')
 					++begin;
 
-				process(query, ast);
+				if (!processSingleQuery(query, ast))
+					return false;
 			}
+
+			return true;
 		}
 		else
 		{
-			process(line);
+			return processSingleQuery(line);
 		}
 	}
 
 
-	bool process(const String & line, ASTPtr parsed_query_ = nullptr)
+	bool processSingleQuery(const String & line, ASTPtr parsed_query_ = nullptr)
 	{
 		if (exit_strings.end() != exit_strings.find(line))
 			return false;
@@ -838,15 +860,8 @@ private:
 	}
 
 
-	void onData(Block & block)
+	void initBlockOutputStream(const Block & block)
 	{
-		if (written_progress_chars)
-			clearProgress();
-
-		if (!block)
-			return;
-
-		processed_rows += block.rows();
 		if (!block_std_out)
 		{
 			String current_format = format;
@@ -869,8 +884,21 @@ private:
 			block_std_out = context.getFormatFactory().getOutput(current_format, std_out, block);
 			block_std_out->writePrefix();
 		}
+	}
 
-		/// Загаловочный блок с нулем строк использовался для инициализации block_std_out,
+
+	void onData(Block & block)
+	{
+		if (written_progress_chars)
+			clearProgress();
+
+		if (!block)
+			return;
+
+		processed_rows += block.rows();
+		initBlockOutputStream(block);
+
+		/// Заголовочный блок с нулем строк использовался для инициализации block_std_out,
 		/// выводить его не нужно
 		if (block.rows() != 0)
 		{
@@ -885,11 +913,13 @@ private:
 
 	void onTotals(Block & block)
 	{
+		initBlockOutputStream(block);
 		block_std_out->setTotals(block);
 	}
 
 	void onExtremes(Block & block)
 	{
+		initBlockOutputStream(block);
 		block_std_out->setExtremes(block);
 	}
 

dbms/src/Core/Block.cpp

@@ -302,7 +302,13 @@ std::string Block::dumpStructure() const
 	{
 		if (it != data.begin())
 			res << ", ";
-		res << it->name << ' ' << it->type->getName() << ' ' << it->column->getName() << ' ' << it->column->size();
+
+		res << it->name << ' ' << it->type->getName();
+
+		if (it->column)
+			res << ' ' << it->column->getName() << ' ' << it->column->size();
+		else
+			res << "nullptr";
 	}
 	return res.str();
 }

dbms/src/DataStreams/FormatFactory.cpp

@@ -26,7 +26,7 @@ namespace DB
 {
 
 BlockInputStreamPtr FormatFactory::getInput(const String & name, ReadBuffer & buf,
-	Block & sample, size_t max_block_size) const
+	const Block & sample, size_t max_block_size) const
 {
 	if (name == "Native")
 		return new NativeBlockInputStream(buf);
@@ -48,7 +48,7 @@ BlockInputStreamPtr FormatFactory::getInput(const String & name, ReadBuffer & bu
 
 
 BlockOutputStreamPtr FormatFactory::getOutput(const String & name, WriteBuffer & buf,
-	Block & sample) const
+	const Block & sample) const
 {
 	if (name == "Native")
 		return new NativeBlockOutputStream(buf);

dbms/src/DataStreams/tests/union_stream.cpp

@@ -38,7 +38,7 @@ int main(int argc, char ** argv)
 		streams.emplace_back(new DB::LimitBlockInputStream(table->read(column_names, 0, DB::Context{}, DB::Settings(), stage2, 1)[0], 30, 2000));
 		streams.emplace_back(new DB::LimitBlockInputStream(table->read(column_names, 0, DB::Context{}, DB::Settings(), stage3, 1)[0], 30, 100));
 
-		DB::UnionBlockInputStream union_stream(streams, 2);
+		DB::UnionBlockInputStream union_stream(streams, nullptr, 2);
 
 		DB::FormatFactory format_factory;
 		DB::WriteBufferFromFileDescriptor wb(STDERR_FILENO);

dbms/src/DataStreams/tests/union_stream2.cpp

@@ -44,7 +44,7 @@ int main(int argc, char ** argv)
 		for (size_t i = 0, size = streams.size(); i < size; ++i)
 			streams[i] = new DB::AsynchronousBlockInputStream(streams[i]);
 		
-		DB::BlockInputStreamPtr stream = new DB::UnionBlockInputStream(streams, settings.max_threads);
+		DB::BlockInputStreamPtr stream = new DB::UnionBlockInputStream(streams, nullptr, settings.max_threads);
 		stream = new DB::LimitBlockInputStream(stream, 10);
 
 		DB::FormatFactory format_factory;

dbms/src/Functions/FunctionsArray.cpp

@@ -13,6 +13,7 @@ void registerFunctionsArray(FunctionFactory & factory)
 	factory.registerFunction<FunctionCountEqual>();
 	factory.registerFunction<FunctionArrayEnumerate>();
 	factory.registerFunction<FunctionArrayEnumerateUniq>();
+	factory.registerFunction<FunctionArrayUniq>();
 	factory.registerFunction<FunctionEmptyArrayUInt8>();
 	factory.registerFunction<FunctionEmptyArrayUInt16>();
 	factory.registerFunction<FunctionEmptyArrayUInt32>();
@@ -26,6 +27,7 @@ void registerFunctionsArray(FunctionFactory & factory)
 	factory.registerFunction<FunctionEmptyArrayDate>();
 	factory.registerFunction<FunctionEmptyArrayDateTime>();
 	factory.registerFunction<FunctionEmptyArrayString>();
+	factory.registerFunction<FunctionEmptyArrayToSingle>();
 	factory.registerFunction<FunctionRange>();
 }
 

dbms/src/Interpreters/Context.cpp

@@ -5,6 +5,7 @@
 #include <Poco/SharedPtr.h>
 #include <Poco/Mutex.h>
 #include <Poco/File.h>
+#include <Poco/UUIDGenerator.h>
 
 #include <Yandex/logger_useful.h>
 
@@ -96,6 +97,8 @@ struct ContextShared
 	/// Создаются при создании Distributed таблиц, так как нужно дождаться пока будут выставлены Settings
 	Poco::SharedPtr<Clusters> clusters;
 
+	Poco::UUIDGenerator uuid_generator;
+
 	bool shutdown_called = false;
 
 
@@ -587,8 +590,12 @@ void Context::setCurrentDatabase(const String & name)
 
 void Context::setCurrentQueryId(const String & query_id)
 {
+	String query_id_to_set = query_id;
+	if (query_id_to_set.empty())	/// Если пользователь не передал свой query_id, то генерируем его самостоятельно.
+		query_id_to_set = shared->uuid_generator.createRandom().toString();
+
 	Poco::ScopedLock<Poco::Mutex> lock(shared->mutex);
-	current_query_id = query_id;
+	current_query_id = query_id_to_set;
 }
 
 

dbms/src/Interpreters/ExpressionActions.cpp

@@ -6,6 +6,7 @@
 #include <DB/DataTypes/DataTypeNested.h>
 #include <DB/DataTypes/DataTypeArray.h>
 #include <DB/Functions/IFunction.h>
+#include <DB/Functions/FunctionsArray.h>
 #include <set>
 
 
@@ -268,6 +269,24 @@ void ExpressionAction::execute(Block & block) const
 			if (!any_array)
 				throw Exception("ARRAY JOIN of not array: " + *array_joined_columns.begin(), ErrorCodes::TYPE_MISMATCH);
 
+			/// Если LEFT ARRAY JOIN, то создаём столбцы, в которых пустые массивы заменены на массивы с одним элементом - значением по-умолчанию.
+			std::map<String, ColumnPtr> non_empty_array_columns;
+			if (array_join_is_left)
+			{
+				for (const auto & name : array_joined_columns)
+				{
+					auto src_col = block.getByName(name);
+
+					Block tmp_block{src_col, {{}, src_col.type, {}}};
+
+					FunctionEmptyArrayToSingle().execute(tmp_block, {0}, 1);
+					non_empty_array_columns[name] = tmp_block.getByPosition(1).column;
+				}
+
+				any_array_ptr = non_empty_array_columns.begin()->second;
+				any_array = typeid_cast<const ColumnArray *>(&*any_array_ptr);
+			}
+
 			size_t columns = block.columns();
 			for (size_t i = 0; i < columns; ++i)
 			{
@@ -278,7 +297,8 @@ void ExpressionAction::execute(Block & block) const
 					if (!typeid_cast<const DataTypeArray *>(&*current.type))
 						throw Exception("ARRAY JOIN of not array: " + current.name, ErrorCodes::TYPE_MISMATCH);
 
-					ColumnPtr array_ptr = current.column;
+					ColumnPtr array_ptr = array_join_is_left ? non_empty_array_columns[current.name] : current.column;
+
 					if (array_ptr->isConst())
 						array_ptr = dynamic_cast<const IColumnConst &>(*array_ptr).convertToFullColumn();
 
@@ -379,7 +399,7 @@ std::string ExpressionAction::toString() const
 			break;
 
 		case ARRAY_JOIN:
-			ss << "ARRAY JOIN ";
+			ss << (array_join_is_left ? "LEFT " : "") << "ARRAY JOIN ";
 			for (NameSet::const_iterator it = array_joined_columns.begin(); it != array_joined_columns.end(); ++it)
 			{
 				if (it != array_joined_columns.begin())
@@ -761,7 +781,7 @@ std::string ExpressionActions::getID() const
 			ss << actions[i].result_name;
 		if (actions[i].type == ExpressionAction::ARRAY_JOIN)
 		{
-			ss << "{";
+			ss << (actions[i].array_join_is_left ? "LEFT ARRAY JOIN" : "ARRAY JOIN") << "{";
 			for (NameSet::const_iterator it = actions[i].array_joined_columns.begin();
 				 it != actions[i].array_joined_columns.end(); ++it)
 			{

dbms/src/Interpreters/ExpressionAnalyzer.cpp

@@ -87,6 +87,18 @@ const std::unordered_set<String> possibly_injective_function_names
 	"dictGetDateTime"
 };
 
+static bool functionIsInOperator(const String & name)
+{
+	return name == "in" || name == "notIn";
+}
+
+static bool functionIsInOrGlobalInOperator(const String & name)
+{
+	return name == "in" || name == "notIn" || name == "globalIn" || name == "globalNotIn";
+}
+
+
+
 void ExpressionAnalyzer::init()
 {
 	select_query = typeid_cast<ASTSelectQuery *>(&*ast);
@@ -95,6 +107,7 @@ void ExpressionAnalyzer::init()
 	LogicalExpressionsOptimizer logical_expressions_optimizer(select_query, settings);
 	logical_expressions_optimizer.optimizeDisjunctiveEqualityChains();
 
+	/// Добавляет в множество известных алиасов те, которые объявлены в структуре таблицы (ALIAS-столбцы).
 	addStorageAliases();
 
 	/// Создаёт словарь aliases: alias -> ASTPtr
@@ -103,6 +116,9 @@ void ExpressionAnalyzer::init()
 	/// Common subexpression elimination. Rewrite rules.
 	normalizeTree();
 
+	/// Выполнение скалярных подзапросов - замена их на значения-константы.
+	executeScalarSubqueries();
+
 	/// GROUP BY injective function elimination.
 	optimizeGroupBy();
 
@@ -146,7 +162,10 @@ void ExpressionAnalyzer::analyzeAggregation()
 
 	if (select_query && select_query->join)
 	{
-		getRootActions(typeid_cast<ASTJoin &>(*select_query->join).using_expr_list, true, false, temp_actions);
+		auto join = typeid_cast<ASTJoin &>(*select_query->join);
+		if (join.using_expr_list)
+			getRootActions(join.using_expr_list, true, false, temp_actions);
+
 		addJoinAction(temp_actions, true);
 	}
 
@@ -385,7 +404,7 @@ void ExpressionAnalyzer::normalizeTreeImpl(
 		}
 
 		/// Может быть указано IN t, где t - таблица, что равносильно IN (SELECT * FROM t).
-		if (func_node->name == "in" || func_node->name == "notIn" || func_node->name == "globalIn" || func_node->name == "globalNotIn")
+		if (functionIsInOrGlobalInOperator(func_node->name))
 			if (ASTIdentifier * right = typeid_cast<ASTIdentifier *>(&*func_node->arguments->children.at(1)))
 				right->kind = ASTIdentifier::Table;
 
@@ -525,6 +544,145 @@ void ExpressionAnalyzer::normalizeTreeImpl(
 	finished_asts[initial_ast] = ast;
 }
 
+
+void ExpressionAnalyzer::executeScalarSubqueries()
+{
+	if (!select_query)
+		executeScalarSubqueriesImpl(ast);
+	else
+	{
+		for (auto & child : ast->children)
+		{
+			/// Не опускаемся в FROM и JOIN.
+			if (child.get() != select_query->table.get() && child.get() != select_query->join.get())
+				executeScalarSubqueriesImpl(child);
+		}
+	}
+}
+
+
+static ASTPtr addTypeConversion(ASTLiteral * ast_, const String & type_name)
+{
+	if (0 == type_name.compare(0, strlen("Array"), "Array"))
+		return ast_;	/// Преобразование типов для массивов пока не поддерживаем.
+
+	auto ast = std::unique_ptr<ASTLiteral>(ast_);
+	ASTFunction * func = new ASTFunction(ast->range);
+	ASTPtr res = func;
+	func->alias = ast->alias;
+	ast->alias.clear();
+	func->kind = ASTFunction::FUNCTION;
+	func->name = "to" + type_name;
+	ASTExpressionList * exp_list = new ASTExpressionList(ast->range);
+	func->arguments = exp_list;
+	func->children.push_back(func->arguments);
+	exp_list->children.push_back(ast.release());
+	return res;
+}
+
+
+void ExpressionAnalyzer::executeScalarSubqueriesImpl(ASTPtr & ast)
+{
+	/** Заменяем подзапросы, возвращающие ровно одну строку
+	  * ("скалярные" подзапросы) на соответствующие константы.
+	  *
+	  * Если подзапрос возвращает более одного столбца, то он заменяется на кортеж констант.
+	  *
+	  * Особенности:
+	  *
+	  * Замена происходит во время анализа запроса, а не во время основной стадии выполнения.
+	  * Это значит, что не будет работать индикатор прогресса во время выполнения этих запросов,
+	  *  а также такие запросы нельзя будет прервать.
+	  *
+	  * Зато результат запросов может быть использован для индекса в таблице.
+	  *
+	  * Скалярные подзапросы выполняются на сервере-инициаторе запроса.
+	  * На удалённые серверы запрос отправляется с уже подставленными константами.
+	  */
+
+	if (ASTSubquery * subquery = typeid_cast<ASTSubquery *>(ast.get()))
+	{
+		Context subquery_context = context;
+		Settings subquery_settings = context.getSettings();
+		subquery_settings.limits.max_result_rows = 1;
+		subquery_settings.extremes = 0;
+		subquery_context.setSettings(subquery_settings);
+
+		ASTPtr query = subquery->children.at(0);
+		BlockIO res = InterpreterSelectQuery(query, subquery_context, QueryProcessingStage::Complete, subquery_depth + 1).execute();
+
+		Block block;
+		try
+		{
+			block = res.in->read();
+
+			if (!block)
+				throw Exception("Scalar subquery returned empty result", ErrorCodes::INCORRECT_RESULT_OF_SCALAR_SUBQUERY);
+
+			if (block.rows() != 1 || res.in->read())
+				throw Exception("Scalar subquery returned more than one row", ErrorCodes::INCORRECT_RESULT_OF_SCALAR_SUBQUERY);
+		}
+		catch (const Exception & e)
+		{
+			if (e.code() == ErrorCodes::TOO_MUCH_ROWS)
+				throw Exception("Scalar subquery returned more than one row", ErrorCodes::INCORRECT_RESULT_OF_SCALAR_SUBQUERY);
+			else
+				throw;
+		}
+
+		size_t columns = block.columns();
+		if (columns == 1)
+		{
+			ASTLiteral * lit = new ASTLiteral(ast->range, (*block.getByPosition(0).column)[0]);
+			lit->alias = subquery->alias;
+			ast = addTypeConversion(lit, block.getByPosition(0).type->getName());
+		}
+		else
+		{
+			ASTFunction * tuple = new ASTFunction(ast->range);
+			tuple->alias = subquery->alias;
+			ast = tuple;
+			tuple->kind = ASTFunction::FUNCTION;
+			tuple->name = "tuple";
+			ASTExpressionList * exp_list = new ASTExpressionList(ast->range);
+			tuple->arguments = exp_list;
+			tuple->children.push_back(tuple->arguments);
+
+			exp_list->children.resize(columns);
+			for (size_t i = 0; i < columns; ++i)
+			{
+				exp_list->children[i] = addTypeConversion(
+					new ASTLiteral(ast->range, (*block.getByPosition(i).column)[0]),
+					block.getByPosition(i).type->getName());
+			}
+		}
+	}
+	else
+	{
+		/** Не опускаемся в подзапросы в аргументах IN.
+		  * Но если аргумент - не подзапрос, то глубже внутри него могут быть подзапросы, и в них надо опускаться.
+		  */
+		ASTFunction * func = typeid_cast<ASTFunction *>(ast.get());
+		if (func && func->kind == ASTFunction::FUNCTION
+			&& functionIsInOrGlobalInOperator(func->name))
+		{
+			for (auto & child : ast->children)
+			{
+				if (child.get() != func->arguments)
+					executeScalarSubqueriesImpl(child);
+				else
+					for (size_t i = 0, size = func->arguments->children.size(); i < size; ++i)
+						if (i != 1 || !typeid_cast<ASTSubquery *>(func->arguments->children[i].get()))
+							executeScalarSubqueriesImpl(func->arguments->children[i]);
+			}
+		}
+		else
+			for (auto & child : ast->children)
+				executeScalarSubqueriesImpl(child);
+	}
+}
+
+
 void ExpressionAnalyzer::optimizeGroupBy()
 {
 	if (!(select_query && select_query->group_expression_list))
@@ -654,7 +812,7 @@ void ExpressionAnalyzer::makeSetsForIndexImpl(ASTPtr & node, const Block & sampl
 		makeSetsForIndexImpl(child, sample_block);
 
 	ASTFunction * func = typeid_cast<ASTFunction *>(node.get());
-	if (func && func->kind == ASTFunction::FUNCTION && (func->name == "in" || func->name == "notIn"))
+	if (func && func->kind == ASTFunction::FUNCTION && functionIsInOperator(func->name))
 	{
 		IAST & args = *func->arguments;
 		ASTPtr & arg = args.children.at(1);
@@ -690,7 +848,8 @@ static SharedPtr<InterpreterSelectQuery> interpretSubquery(
 	  * Так как результат этого поздапроса - ещё не результат всего запроса.
 	  * Вместо этого работают ограничения
 	  *  max_rows_in_set, max_bytes_in_set, set_overflow_mode,
-	  *  max_rows_in_join, max_bytes_in_join, join_overflow_mode.
+	  *  max_rows_in_join, max_bytes_in_join, join_overflow_mode,
+	  *  которые проверяются отдельно (в объектах Set, Join).
 	  */
 	Context subquery_context = context;
 	Settings subquery_settings = context.getSettings();
@@ -1210,7 +1369,7 @@ void ExpressionAnalyzer::getActionsImpl(ASTPtr ast, bool no_subqueries, bool onl
 				actions_stack.addAction(ExpressionAction::copyColumn(arg->getColumnName(), result_name));
 				NameSet joined_columns;
 				joined_columns.insert(result_name);
-				actions_stack.addAction(ExpressionAction::arrayJoin(joined_columns));
+				actions_stack.addAction(ExpressionAction::arrayJoin(joined_columns, false));
 			}
 
 			return;
@@ -1218,7 +1377,7 @@ void ExpressionAnalyzer::getActionsImpl(ASTPtr ast, bool no_subqueries, bool onl
 
 		if (node->kind == ASTFunction::FUNCTION)
 		{
-			if (node->name == "in" || node->name == "notIn" || node->name == "globalIn" || node->name == "globalNotIn")
+			if (functionIsInOrGlobalInOperator(node->name))
 			{
 				if (!no_subqueries)
 				{
@@ -1507,7 +1666,7 @@ void ExpressionAnalyzer::addMultipleArrayJoinAction(ExpressionActionsPtr & actio
 		result_columns.insert(result_source.first);
 	}
 
-	actions->add(ExpressionAction::arrayJoin(result_columns));
+	actions->add(ExpressionAction::arrayJoin(result_columns, select_query->array_join_is_left));
 }
 
 bool ExpressionAnalyzer::appendArrayJoin(ExpressionActionsChain & chain, bool only_types)
@@ -1548,7 +1707,8 @@ bool ExpressionAnalyzer::appendJoin(ExpressionActionsChain & chain, bool only_ty
 	ExpressionActionsChain::Step & step = chain.steps.back();
 
 	ASTJoin & ast_join = typeid_cast<ASTJoin &>(*select_query->join);
-	getRootActions(ast_join.using_expr_list, only_types, false, step.actions);
+	if (ast_join.using_expr_list)
+		getRootActions(ast_join.using_expr_list, only_types, false, step.actions);
 
 	/// Не поддерживается два JOIN-а с одинаковым подзапросом, но разными USING-ами.
 	String join_id = ast_join.table->getColumnName();
@@ -1888,7 +2048,7 @@ void ExpressionAnalyzer::collectUsedColumns()
 	}
 
 /*	for (const auto & name_type : columns_added_by_join)
-		std::cerr << "JOINed column (required, not key): " << name_type.first << std::endl;
+		std::cerr << "JOINed column (required, not key): " << name_type.name << std::endl;
 	std::cerr << std::endl;*/
 
 	/// Вставляем в список требуемых столбцов столбцы, нужные для вычисления ARRAY JOIN.
@@ -1968,14 +2128,17 @@ void ExpressionAnalyzer::collectJoinedColumns(NameSet & joined_columns, NamesAnd
 		nested_result_sample = InterpreterSelectQuery::getSampleBlock(subquery, context);
 	}
 
-	auto & keys = typeid_cast<ASTExpressionList &>(*node.using_expr_list);
-	for (const auto & key : keys.children)
+	if (node.using_expr_list)
 	{
-		if (!join_key_names_left_set.insert(key->getColumnName()).second)
-			throw Exception("Duplicate column in USING list", ErrorCodes::DUPLICATE_COLUMN);
+		auto & keys = typeid_cast<ASTExpressionList &>(*node.using_expr_list);
+		for (const auto & key : keys.children)
+		{
+			if (!join_key_names_left_set.insert(key->getColumnName()).second)
+				throw Exception("Duplicate column in USING list", ErrorCodes::DUPLICATE_COLUMN);
 
-		if (!join_key_names_right_set.insert(key->getAliasOrColumnName()).second)
-			throw Exception("Duplicate column in USING list", ErrorCodes::DUPLICATE_COLUMN);
+			if (!join_key_names_right_set.insert(key->getAliasOrColumnName()).second)
+				throw Exception("Duplicate column in USING list", ErrorCodes::DUPLICATE_COLUMN);
+		}
 	}
 
 	for (const auto i : ext::range(0, nested_result_sample.columns()))

dbms/src/Interpreters/InterpreterCreateQuery.cpp

@@ -7,7 +7,7 @@
 #include <DB/IO/WriteHelpers.h>
 
 #include <DB/DataStreams/MaterializingBlockInputStream.h>
-#include <DB/DataStreams/copyData.h>
+#include <DB/DataStreams/NullAndDoCopyBlockInputStream.h>
 
 #include <DB/Parsers/ASTCreateQuery.h>
 #include <DB/Parsers/ASTNameTypePair.h>
@@ -42,7 +42,7 @@ InterpreterCreateQuery::InterpreterCreateQuery(ASTPtr query_ptr_, Context & cont
 }
 
 
-void InterpreterCreateQuery::executeImpl(bool assume_metadata_exists)
+BlockIO InterpreterCreateQuery::executeImpl(bool assume_metadata_exists)
 {
 	String path = context.getPath();
 	String current_database = context.getCurrentDatabase();
@@ -81,7 +81,7 @@ void InterpreterCreateQuery::executeImpl(bool assume_metadata_exists)
 		if (!create.if_not_exists || !context.isDatabaseExist(database_name))
 			context.addDatabase(database_name);
 
-		return;
+		return {};
 	}
 
 	SharedPtr<InterpreterSelectQuery> interpreter_select;
@@ -118,7 +118,7 @@ void InterpreterCreateQuery::executeImpl(bool assume_metadata_exists)
 			if (context.isTableExist(database_name, table_name))
 			{
 				if (create.if_not_exists)
-					return;
+					return {};
 				else
 					throw Exception("Table " + database_name + "." + table_name + " already exists.", ErrorCodes::TABLE_ALREADY_EXISTS);
 			}
@@ -251,9 +251,16 @@ void InterpreterCreateQuery::executeImpl(bool assume_metadata_exists)
 	/// Если запрос CREATE SELECT, то вставим в таблицу данные
 	if (create.select && storage_name != "View" && (storage_name != "MaterializedView" || create.is_populate))
 	{
-		BlockInputStreamPtr from = new MaterializingBlockInputStream(interpreter_select->execute().in);
-		copyData(*from, *res->write(query_ptr));
+		BlockIO io;
+		io.in_sample = select_sample;
+		io.in = new NullAndDoCopyBlockInputStream(
+			new MaterializingBlockInputStream(interpreter_select->execute().in),
+			res->write(query_ptr));
+
+		return io;
 	}
+
+	return {};
 }
 
 InterpreterCreateQuery::ColumnsAndDefaults InterpreterCreateQuery::parseColumns(ASTPtr expression_list)

dbms/src/Interpreters/InterpreterInsertQuery.cpp

@@ -100,6 +100,7 @@ BlockIO InterpreterInsertQuery::execute()
 		InterpreterSelectQuery interpreter_select{query.select, context};
 		BlockInputStreamPtr in{interpreter_select.execute().in};
 		res.in = new NullAndDoCopyBlockInputStream{in, out};
+		res.in_sample = interpreter_select.getSampleBlock();
 	}
 
 	return res;

dbms/src/Interpreters/InterpreterSelectQuery.cpp

@@ -16,6 +16,7 @@
 #include <DB/DataStreams/CreatingSetsBlockInputStream.h>
 #include <DB/DataStreams/MaterializingBlockInputStream.h>
 #include <DB/DataStreams/FormatFactory.h>
+#include <DB/DataStreams/ConcatBlockInputStream.h>
 
 #include <DB/Parsers/ASTSelectQuery.h>
 #include <DB/Parsers/ASTIdentifier.h>
@@ -316,7 +317,7 @@ BlockIO InterpreterSelectQuery::execute()
 {
 	(void) executeWithoutUnion();
 
-	if (streams.empty())
+	if (hasNoData())
 	{
 		BlockIO res;
 		res.in = new NullBlockInputStream;
@@ -324,14 +325,11 @@ BlockIO InterpreterSelectQuery::execute()
 		return res;
 	}
 
-	executeUnion(streams);
+	executeUnion();
 
 	/// Ограничения на результат, квота на результат, а также колбек для прогресса.
 	if (IProfilingBlockInputStream * stream = dynamic_cast<IProfilingBlockInputStream *>(&*streams[0]))
 	{
-		stream->setProgressCallback(context.getProgressCallback());
-		stream->setProcessListElement(context.getProcessListElement());
-
 		/// Ограничения действуют только на конечный результат.
 		if (to_stage == QueryProcessingStage::Complete)
 		{
@@ -365,8 +363,10 @@ const BlockInputStreams & InterpreterSelectQuery::executeWithoutUnion()
 			streams.insert(streams.end(), others.begin(), others.end());
 		}
 
-		for (auto & stream : streams)
+		transformStreams([&](auto & stream)
+		{
 			stream = new MaterializingBlockInputStream(stream);
+		});
 	}
 	else
 		executeSingleQuery();
@@ -391,18 +391,19 @@ void InterpreterSelectQuery::executeSingleQuery()
 	union_within_single_query = false;
 
 	/** Вынем данные из Storage. from_stage - до какой стадии запрос был выполнен в Storage. */
-	QueryProcessingStage::Enum from_stage = executeFetchColumns(streams);
+	QueryProcessingStage::Enum from_stage = executeFetchColumns();
 
 	LOG_TRACE(log, QueryProcessingStage::toString(from_stage) << " -> " << QueryProcessingStage::toString(to_stage));
 
 	if (to_stage > QueryProcessingStage::FetchColumns)
 	{
+		bool has_join		= false;
 		bool has_where      = false;
 		bool need_aggregate = false;
 		bool has_having     = false;
 		bool has_order_by   = false;
 
-		ExpressionActionsPtr array_join;
+		ExpressionActionsPtr before_join;
 		ExpressionActionsPtr before_where;
 		ExpressionActionsPtr before_aggregation;
 		ExpressionActionsPtr before_having;
@@ -430,7 +431,17 @@ void InterpreterSelectQuery::executeSingleQuery()
 			need_aggregate = query_analyzer->hasAggregation();
 
 			query_analyzer->appendArrayJoin(chain, !first_stage);
-			query_analyzer->appendJoin(chain, !first_stage);
+
+			if (query_analyzer->appendJoin(chain, !first_stage))
+			{
+				has_join = true;
+				before_join = chain.getLastActions();
+				chain.addStep();
+
+				auto join = typeid_cast<const ASTJoin &>(*query.join);
+				if (join.kind == ASTJoin::Full || join.kind == ASTJoin::Right)
+					stream_with_non_joined_data = before_join->createStreamWithNonJoinedDataIfFullOrRightJoin(settings.max_block_size);
+			}
 
 			if (query_analyzer->appendWhere(chain, !first_stage))
 			{
@@ -475,7 +486,7 @@ void InterpreterSelectQuery::executeSingleQuery()
 		 *  чтобы запрос был проанализирован, и в нём могли бы быть обнаружены ошибки (например, несоответствия типов).
 		 *  Иначе мог бы вернуться пустой результат на некорректный запрос.
 		 */
-		if (streams.empty())
+		if (hasNoData())
 			return;
 
 		/// Перед выполнением HAVING уберем из блока лишние столбцы (в основном, ключи агрегации).
@@ -500,15 +511,19 @@ void InterpreterSelectQuery::executeSingleQuery()
 
 		if (first_stage)
 		{
+			if (has_join)
+				for (auto & stream : streams)	/// Применяем ко всем источникам кроме stream_with_non_joined_data.
+					stream = new ExpressionBlockInputStream(stream, before_join);
+
 			if (has_where)
-				executeWhere(streams, before_where);
+				executeWhere(before_where);
 
 			if (need_aggregate)
-				executeAggregation(streams, before_aggregation, aggregate_overflow_row, aggregate_final);
+				executeAggregation(before_aggregation, aggregate_overflow_row, aggregate_final);
 			else
 			{
-				executeExpression(streams, before_order_and_select);
-				executeDistinct(streams, true, selected_columns);
+				executeExpression(before_order_and_select);
+				executeDistinct(true, selected_columns);
 			}
 
 			/** При распределённой обработке запроса,
@@ -520,13 +535,13 @@ void InterpreterSelectQuery::executeSingleQuery()
 				&& !need_aggregate && !has_having)
 			{
 				if (has_order_by)
-					executeOrder(streams);
+					executeOrder();
 
 				if (has_order_by && query.limit_length)
-					executeDistinct(streams, false, selected_columns);
+					executeDistinct(false, selected_columns);
 
 				if (query.limit_length)
-					executePreLimit(streams);
+					executePreLimit();
 			}
 		}
 
@@ -538,21 +553,21 @@ void InterpreterSelectQuery::executeSingleQuery()
 			{
 				/// Если нужно объединить агрегированные результаты с нескольких серверов
 				if (!first_stage)
-					executeMergeAggregated(streams, aggregate_overflow_row, aggregate_final);
+					executeMergeAggregated(aggregate_overflow_row, aggregate_final);
 
 				if (!aggregate_final)
-					executeTotalsAndHaving(streams, has_having, before_having, aggregate_overflow_row);
+					executeTotalsAndHaving(has_having, before_having, aggregate_overflow_row);
 				else if (has_having)
-					executeHaving(streams, before_having);
+					executeHaving(before_having);
 
-				executeExpression(streams, before_order_and_select);
-				executeDistinct(streams, true, selected_columns);
+				executeExpression(before_order_and_select);
+				executeDistinct(true, selected_columns);
 
-				need_second_distinct_pass = query.distinct && (streams.size() > 1);
+				need_second_distinct_pass = query.distinct && hasMoreThanOneStream();
 			}
 			else if (query.group_by_with_totals && !aggregate_final)
 			{
-				executeTotalsAndHaving(streams, false, nullptr, aggregate_overflow_row);
+				executeTotalsAndHaving(false, nullptr, aggregate_overflow_row);
 			}
 
 			if (has_order_by)
@@ -562,49 +577,53 @@ void InterpreterSelectQuery::executeSingleQuery()
 				  *  - поэтому, делаем merge сортированных потоков с удалённых серверов.
 				  */
 				if (!first_stage && !need_aggregate && !(query.group_by_with_totals && !aggregate_final))
-					executeMergeSorted(streams);
+					executeMergeSorted();
 				else	/// Иначе просто сортировка.
-					executeOrder(streams);
+					executeOrder();
 			}
 
-			executeProjection(streams, final_projection);
+			executeProjection(final_projection);
 
 			/// На этой стадии можно считать минимумы и максимумы, если надо.
 			if (settings.extremes)
-				for (auto & stream : streams)
+			{
+				transformStreams([&](auto & stream)
+				{
 					if (IProfilingBlockInputStream * p_stream = dynamic_cast<IProfilingBlockInputStream *>(&*stream))
 						p_stream->enableExtremes();
+				});
+			}
 
 			/** Оптимизация - если источников несколько и есть LIMIT, то сначала применим предварительный LIMIT,
 				* ограничивающий число записей в каждом до offset + limit.
 				*/
-			if (query.limit_length && streams.size() > 1 && !query.distinct)
-				executePreLimit(streams);
+			if (query.limit_length && hasMoreThanOneStream() && !query.distinct)
+				executePreLimit();
 
 			if (need_second_distinct_pass)
 				union_within_single_query = true;
 
-			if (union_within_single_query)
-				executeUnion(streams);
+			if (union_within_single_query || stream_with_non_joined_data)
+				executeUnion();
 
 			if (streams.size() == 1)
 			{
 				/// Если было более одного источника - то нужно выполнить DISTINCT ещё раз после их слияния.
 				if (need_second_distinct_pass)
-					executeDistinct(streams, false, Names());
+					executeDistinct(false, Names());
 
-				executeLimit(streams);
+				executeLimit();
 			}
 		}
 	}
 
 	/** Если данных нет. */
-	if (streams.empty())
+	if (hasNoData())
 		return;
 
 	SubqueriesForSets subqueries_for_sets = query_analyzer->getSubqueriesForSets();
 	if (!subqueries_for_sets.empty())
-		executeSubqueriesInSetsAndJoins(streams, subqueries_for_sets);
+		executeSubqueriesInSetsAndJoins(subqueries_for_sets);
 }
 
 
@@ -620,9 +639,9 @@ static void getLimitLengthAndOffset(ASTSelectQuery & query, size_t & length, siz
 	}
 }
 
-QueryProcessingStage::Enum InterpreterSelectQuery::executeFetchColumns(BlockInputStreams & streams)
+QueryProcessingStage::Enum InterpreterSelectQuery::executeFetchColumns()
 {
-	if (!streams.empty())
+	if (!hasNoData())
 		return QueryProcessingStage::FetchColumns;
 
 	/// Интерпретатор подзапроса, если подзапрос
@@ -741,8 +760,10 @@ QueryProcessingStage::Enum InterpreterSelectQuery::executeFetchColumns(BlockInpu
 			context, settings_for_storage, from_stage,
 			settings.max_block_size, max_streams);
 
-		for (auto & stream : streams)
+		transformStreams([&](auto & stream)
+		{
 			stream->addTableLock(table_lock);
+		});
 	}
 	else
 	{
@@ -768,38 +789,36 @@ QueryProcessingStage::Enum InterpreterSelectQuery::executeFetchColumns(BlockInpu
 
 		QuotaForIntervals & quota = context.getQuota();
 
-		for (auto & stream : streams)
+		transformStreams([&](auto & stream)
 		{
 			if (IProfilingBlockInputStream * p_stream = dynamic_cast<IProfilingBlockInputStream *>(&*stream))
 			{
 				p_stream->setLimits(limits);
 				p_stream->setQuota(quota);
 			}
-		}
+		});
 	}
 
 	return from_stage;
 }
 
 
-void InterpreterSelectQuery::executeWhere(BlockInputStreams & streams, ExpressionActionsPtr expression)
+void InterpreterSelectQuery::executeWhere(ExpressionActionsPtr expression)
 {
-	for (auto & stream : streams)
+	transformStreams([&](auto & stream)
 	{
 		stream = new ExpressionBlockInputStream(stream, expression);
 		stream = new FilterBlockInputStream(stream, query.where_expression->getColumnName());
-	}
+	});
 }
 
 
-void InterpreterSelectQuery::executeAggregation(BlockInputStreams & streams, ExpressionActionsPtr expression, bool overflow_row, bool final)
+void InterpreterSelectQuery::executeAggregation(ExpressionActionsPtr expression, bool overflow_row, bool final)
 {
-	for (auto & stream : streams)
+	transformStreams([&](auto & stream)
 	{
 		stream = new ExpressionBlockInputStream(stream, expression);
-	}
-
-	BlockInputStreamPtr & stream = streams[0];
+	});
 
 	Names key_names;
 	AggregateDescriptions aggregates;
@@ -808,23 +827,37 @@ void InterpreterSelectQuery::executeAggregation(BlockInputStreams & streams, Exp
 	/// Если источников несколько, то выполняем параллельную агрегацию
 	if (streams.size() > 1)
 	{
-		stream = new ParallelAggregatingBlockInputStream(streams, key_names, aggregates, overflow_row, final,
+		streams[0] = new ParallelAggregatingBlockInputStream(streams, stream_with_non_joined_data, key_names, aggregates, overflow_row, final,
 			settings.max_threads, settings.limits.max_rows_to_group_by, settings.limits.group_by_overflow_mode,
 			settings.compile ? &context.getCompiler() : nullptr, settings.min_count_to_compile, settings.group_by_two_level_threshold);
 
+		stream_with_non_joined_data = nullptr;
 		streams.resize(1);
 	}
 	else
-		stream = new AggregatingBlockInputStream(stream, key_names, aggregates, overflow_row, final,
+	{
+		BlockInputStreams inputs;
+		if (!streams.empty())
+			inputs.push_back(streams[0]);
+		else
+			streams.resize(1);
+
+		if (stream_with_non_joined_data)
+			inputs.push_back(stream_with_non_joined_data);
+
+		streams[0] = new AggregatingBlockInputStream(new ConcatBlockInputStream(inputs), key_names, aggregates, overflow_row, final,
 			settings.limits.max_rows_to_group_by, settings.limits.group_by_overflow_mode,
 			settings.compile ? &context.getCompiler() : nullptr, settings.min_count_to_compile, 0);
+
+		stream_with_non_joined_data = nullptr;
+	}
 }
 
 
-void InterpreterSelectQuery::executeMergeAggregated(BlockInputStreams & streams, bool overflow_row, bool final)
+void InterpreterSelectQuery::executeMergeAggregated(bool overflow_row, bool final)
 {
 	/// Склеим несколько источников в один
-	executeUnion(streams);
+	executeUnion();
 
 	/// Теперь объединим агрегированные блоки
 	Names key_names;
@@ -834,20 +867,19 @@ void InterpreterSelectQuery::executeMergeAggregated(BlockInputStreams & streams,
 }
 
 
-void InterpreterSelectQuery::executeHaving(BlockInputStreams & streams, ExpressionActionsPtr expression)
+void InterpreterSelectQuery::executeHaving(ExpressionActionsPtr expression)
 {
-	for (auto & stream : streams)
+	transformStreams([&](auto & stream)
 	{
 		stream = new ExpressionBlockInputStream(stream, expression);
 		stream = new FilterBlockInputStream(stream, query.having_expression->getColumnName());
-	}
+	});
 }
 
 
-void InterpreterSelectQuery::executeTotalsAndHaving(BlockInputStreams & streams, bool has_having,
-	ExpressionActionsPtr expression, bool overflow_row)
+void InterpreterSelectQuery::executeTotalsAndHaving(bool has_having, ExpressionActionsPtr expression, bool overflow_row)
 {
-	executeUnion(streams);
+	executeUnion();
 
 	Names key_names;
 	AggregateDescriptions aggregates;
@@ -858,12 +890,12 @@ void InterpreterSelectQuery::executeTotalsAndHaving(BlockInputStreams & streams,
 }
 
 
-void InterpreterSelectQuery::executeExpression(BlockInputStreams & streams, ExpressionActionsPtr expression)
+void InterpreterSelectQuery::executeExpression(ExpressionActionsPtr expression)
 {
-	for (auto & stream : streams)
+	transformStreams([&](auto & stream)
 	{
 		stream = new ExpressionBlockInputStream(stream, expression);
-	}
+	});
 }
 
 
@@ -898,12 +930,12 @@ static size_t getLimitForSorting(ASTSelectQuery & query)
 }
 
 
-void InterpreterSelectQuery::executeOrder(BlockInputStreams & streams)
+void InterpreterSelectQuery::executeOrder()
 {
 	SortDescription order_descr = getSortDescription(query);
 	size_t limit = getLimitForSorting(query);
 
-	for (auto & stream : streams)
+	transformStreams([&](auto & stream)
 	{
 		IProfilingBlockInputStream * sorting_stream = new PartialSortingBlockInputStream(stream, order_descr, limit);
 
@@ -916,53 +948,51 @@ void InterpreterSelectQuery::executeOrder(BlockInputStreams & streams)
 		sorting_stream->setLimits(limits);
 
 		stream = sorting_stream;
-	}
-
-	BlockInputStreamPtr & stream = streams[0];
+	});
 
 	/// Если потоков несколько, то объединяем их в один
-	executeUnion(streams);
+	executeUnion();
 
 	/// Сливаем сортированные блоки.
-	stream = new MergeSortingBlockInputStream(
-		stream, order_descr, settings.max_block_size, limit,
+	streams[0] = new MergeSortingBlockInputStream(
+		streams[0], order_descr, settings.max_block_size, limit,
 		settings.limits.max_bytes_before_external_sort, context.getTemporaryPath());
 }
 
 
-void InterpreterSelectQuery::executeMergeSorted(BlockInputStreams & streams)
+void InterpreterSelectQuery::executeMergeSorted()
 {
 	SortDescription order_descr = getSortDescription(query);
 	size_t limit = getLimitForSorting(query);
 
-	BlockInputStreamPtr & stream = streams[0];
-
 	/// Если потоков несколько, то объединяем их в один
-	if (streams.size() > 1)
+	if (hasMoreThanOneStream())
 	{
 		/** MergingSortedBlockInputStream читает источники последовательно.
 		  * Чтобы данные на удалённых серверах готовились параллельно, оборачиваем в AsynchronousBlockInputStream.
 		  */
-		for (auto & stream : streams)
+		transformStreams([&](auto & stream)
+		{
 			stream = new AsynchronousBlockInputStream(stream);
+		});
 
 		/// Сливаем сортированные источники в один сортированный источник.
-		stream = new MergingSortedBlockInputStream(streams, order_descr, settings.max_block_size, limit);
+		streams[0] = new MergingSortedBlockInputStream(streams, order_descr, settings.max_block_size, limit);
 		streams.resize(1);
 	}
 }
 
 
-void InterpreterSelectQuery::executeProjection(BlockInputStreams & streams, ExpressionActionsPtr expression)
+void InterpreterSelectQuery::executeProjection(ExpressionActionsPtr expression)
 {
-	for (auto & stream : streams)
+	transformStreams([&](auto & stream)
 	{
 		stream = new ExpressionBlockInputStream(stream, expression);
-	}
+	});
 }
 
 
-void InterpreterSelectQuery::executeDistinct(BlockInputStreams & streams, bool before_order, Names columns)
+void InterpreterSelectQuery::executeDistinct(bool before_order, Names columns)
 {
 	if (query.distinct)
 	{
@@ -976,31 +1006,38 @@ void InterpreterSelectQuery::executeDistinct(BlockInputStreams & streams, bool b
 		if (!query.order_expression_list || !before_order)
 			limit_for_distinct = limit_length + limit_offset;
 
-		for (auto & stream : streams)
+		transformStreams([&](auto & stream)
 		{
 			stream = new DistinctBlockInputStream(stream, settings.limits, limit_for_distinct, columns);
-		}
+		});
 
-		if (streams.size() > 1)
+		if (hasMoreThanOneStream())
 			union_within_single_query = true;
 	}
 }
 
 
-void InterpreterSelectQuery::executeUnion(BlockInputStreams & streams)
+void InterpreterSelectQuery::executeUnion()
 {
 	/// Если до сих пор есть несколько потоков, то объединяем их в один
-	if (streams.size() > 1)
+	if (hasMoreThanOneStream())
 	{
-		streams[0] = new UnionBlockInputStream(streams, settings.max_threads);
+		streams[0] = new UnionBlockInputStream(streams, stream_with_non_joined_data, settings.max_threads);
+		stream_with_non_joined_data = nullptr;
 		streams.resize(1);
 		union_within_single_query = false;
 	}
+	else if (stream_with_non_joined_data)
+	{
+		streams.push_back(stream_with_non_joined_data);
+		stream_with_non_joined_data = nullptr;
+		union_within_single_query = false;
+	}
 }
 
 
 /// Предварительный LIMIT - применяется в каждом источнике, если источников несколько, до их объединения.
-void InterpreterSelectQuery::executePreLimit(BlockInputStreams & streams)
+void InterpreterSelectQuery::executePreLimit()
 {
 	size_t limit_length = 0;
 	size_t limit_offset = 0;
@@ -1009,18 +1046,18 @@ void InterpreterSelectQuery::executePreLimit(BlockInputStreams & streams)
 	/// Если есть LIMIT
 	if (query.limit_length)
 	{
-		for (auto & stream : streams)
+		transformStreams([&](auto & stream)
 		{
 			stream = new LimitBlockInputStream(stream, limit_length + limit_offset, 0);
-		}
+		});
 
-		if (streams.size() > 1)
+		if (hasMoreThanOneStream())
 			union_within_single_query = true;
 	}
 }
 
 
-void InterpreterSelectQuery::executeLimit(BlockInputStreams & streams)
+void InterpreterSelectQuery::executeLimit()
 {
 	size_t limit_length = 0;
 	size_t limit_offset = 0;
@@ -1029,20 +1066,22 @@ void InterpreterSelectQuery::executeLimit(BlockInputStreams & streams)
 	/// Если есть LIMIT
 	if (query.limit_length)
 	{
-		BlockInputStreamPtr & stream = streams[0];
-		stream = new LimitBlockInputStream(stream, limit_length, limit_offset);
+		transformStreams([&](auto & stream)
+		{
+			stream = new LimitBlockInputStream(stream, limit_length, limit_offset);
+		});
 	}
 }
 
 
-void InterpreterSelectQuery::executeSubqueriesInSetsAndJoins(BlockInputStreams & streams, SubqueriesForSets & subqueries_for_sets)
+void InterpreterSelectQuery::executeSubqueriesInSetsAndJoins(SubqueriesForSets & subqueries_for_sets)
 {
 	/// Если запрос не распределённый, то удалим создание временных таблиц из подзапросов (предназначавшихся для отправки на удалённые серверы).
 	if (!(storage && storage->isRemote()))
 		for (auto & elem : subqueries_for_sets)
 			elem.second.table.reset();
 
-	executeUnion(streams);
+	executeUnion();
 	streams[0] = new CreatingSetsBlockInputStream(streams[0], subqueries_for_sets, settings.limits);
 }
 

dbms/src/Interpreters/Join.cpp

@@ -4,6 +4,10 @@
 #include <DB/Parsers/ASTJoin.h>
 #include <DB/Interpreters/Join.h>
 #include <DB/DataStreams/IProfilingBlockInputStream.h>
+#include <DB/Core/ColumnNumbers.h>
+
+/*#include <DB/DataStreams/TabSeparatedBlockOutputStream.h>
+ *#include <DB/IO/WriteBufferFromFileDescriptor.h>*/
 
 
 namespace DB
@@ -18,6 +22,9 @@ Join::Type Join::chooseMethod(const ConstColumnPlainPtrs & key_columns, bool & k
 	size_t keys_bytes = 0;
 	key_sizes.resize(keys_size);
 
+	if (keys_size == 0)
+		return Type::CROSS;
+
 	for (size_t j = 0; j < keys_size; ++j)
 	{
 		if (!key_columns[j]->isFixed())
@@ -57,6 +64,7 @@ static void initImpl(Maps & maps, Join::Type type)
 		case Join::Type::KEY_64:		maps.key64		.reset(new typename Maps::MapUInt64); 	break;
 		case Join::Type::KEY_STRING:	maps.key_string	.reset(new typename Maps::MapString); 	break;
 		case Join::Type::HASHED:		maps.hashed		.reset(new typename Maps::MapHashed);	break;
+		case Join::Type::CROSS:																	break;
 
 		default:
 			throw Exception("Unknown JOIN keys variant.", ErrorCodes::UNKNOWN_AGGREGATED_DATA_VARIANT);
@@ -101,6 +109,9 @@ void Join::init(Type type_)
 {
 	type = type_;
 
+	if (kind == ASTJoin::Cross)
+		return;
+
 	if (!getFullness(kind))
 	{
 		if (strictness == ASTJoin::Any)
@@ -120,21 +131,41 @@ void Join::init(Type type_)
 size_t Join::getTotalRowCount() const
 {
 	size_t res = 0;
-	res += getTotalRowCountImpl(maps_any);
-	res += getTotalRowCountImpl(maps_all);
-	res += getTotalRowCountImpl(maps_any_full);
-	res += getTotalRowCountImpl(maps_all_full);
+
+	if (type == Type::CROSS)
+	{
+		for (const auto & block : blocks)
+			res += block.rowsInFirstColumn();
+	}
+	else
+	{
+		res += getTotalRowCountImpl(maps_any);
+		res += getTotalRowCountImpl(maps_all);
+		res += getTotalRowCountImpl(maps_any_full);
+		res += getTotalRowCountImpl(maps_all_full);
+	}
+
 	return res;
 }
 
 size_t Join::getTotalByteCount() const
 {
 	size_t res = 0;
-	res += getTotalByteCountImpl(maps_any);
-	res += getTotalByteCountImpl(maps_all);
-	res += getTotalByteCountImpl(maps_any_full);
-	res += getTotalByteCountImpl(maps_all_full);
-	res += pool.size();
+
+	if (type == Type::CROSS)
+	{
+		for (const auto & block : blocks)
+			res += block.bytes();
+	}
+	else
+	{
+		res += getTotalByteCountImpl(maps_any);
+		res += getTotalByteCountImpl(maps_all);
+		res += getTotalByteCountImpl(maps_any_full);
+		res += getTotalByteCountImpl(maps_all_full);
+		res += pool.size();
+	}
+
 	return res;
 }
 
@@ -254,7 +285,11 @@ template <> struct Inserter<ASTJoin::All, Join::MapsAllFull::MapString> : Insert
 template <ASTJoin::Strictness STRICTNESS, typename Maps>
 void Join::insertFromBlockImpl(Maps & maps, size_t rows, const ConstColumnPlainPtrs & key_columns, size_t keys_size, Block * stored_block)
 {
-	if (type == Type::KEY_64)
+	if (type == Type::CROSS)
+	{
+		/// Ничего не делаем. Уже сохранили блок, и этого достаточно.
+	}
+	else if (type == Type::KEY_64)
 	{
 		typedef typename Maps::MapUInt64 Map;
 		Map & res = *maps.key64;
@@ -377,9 +412,9 @@ bool Join::insertFromBlock(const Block & block)
 	{
 		key_columns[i] = block.getByName(key_names_right[i]).column;
 
-		if (key_columns[i]->isConst())
+		if (auto * col_const = dynamic_cast<const IColumnConst *>(key_columns[i]))
 		{
-			materialized_columns.emplace_back(dynamic_cast<const IColumnConst &>(*key_columns[i]).convertToFullColumn());
+			materialized_columns.emplace_back(col_const->convertToFullColumn());
 			key_columns[i] = materialized_columns.back();
 		}
 	}
@@ -389,31 +424,51 @@ bool Join::insertFromBlock(const Block & block)
 	blocks.push_back(block);
 	Block * stored_block = &blocks.back();
 
-	/// Удаляем из stored_block ключевые столбцы, так как они не нужны.
-	for (const auto & name : key_names_right)
-		stored_block->erase(stored_block->getPositionByName(name));
+	if (getFullness(kind))
+	{
+		/// Переносим ключевые столбцы в начало блока.
+		size_t key_num = 0;
+		for (const auto & name : key_names_right)
+		{
+			size_t pos = stored_block->getPositionByName(name);
+			ColumnWithTypeAndName col = stored_block->getByPosition(pos);
+			stored_block->erase(pos);
+			stored_block->insert(key_num, col);
+			++key_num;
+		}
+	}
+	else
+	{
+		/// Удаляем из stored_block ключевые столбцы, так как они не нужны.
+		for (const auto & name : key_names_right)
+			stored_block->erase(stored_block->getPositionByName(name));
+	}
 
 	/// Редкий случай, когда соединяемые столбцы являются константами. Чтобы не поддерживать отдельный код, материализуем их.
 	for (size_t i = 0, size = stored_block->columns(); i < size; ++i)
 	{
 		ColumnPtr col = stored_block->getByPosition(i).column;
-		if (col->isConst())
-			stored_block->getByPosition(i).column = dynamic_cast<IColumnConst &>(*col).convertToFullColumn();
+		if (auto * col_const = dynamic_cast<const IColumnConst *>(col.get()))
+			stored_block->getByPosition(i).column = col_const->convertToFullColumn();
 	}
 
-	if (!getFullness(kind))
+	if (kind != ASTJoin::Cross)
 	{
-		if (strictness == ASTJoin::Any)
-			insertFromBlockImpl<ASTJoin::Any>(maps_any, rows, key_columns, keys_size, stored_block);
+		/// Заполняем нужную хэш-таблицу.
+		if (!getFullness(kind))
+		{
+			if (strictness == ASTJoin::Any)
+				insertFromBlockImpl<ASTJoin::Any>(maps_any, rows, key_columns, keys_size, stored_block);
+			else
+				insertFromBlockImpl<ASTJoin::All>(maps_all, rows, key_columns, keys_size, stored_block);
+		}
 		else
-			insertFromBlockImpl<ASTJoin::All>(maps_all, rows, key_columns, keys_size, stored_block);
-	}
-	else
-	{
-		if (strictness == ASTJoin::Any)
-			insertFromBlockImpl<ASTJoin::Any>(maps_any_full, rows, key_columns, keys_size, stored_block);
-		else
-			insertFromBlockImpl<ASTJoin::All>(maps_all_full, rows, key_columns, keys_size, stored_block);
+		{
+			if (strictness == ASTJoin::Any)
+				insertFromBlockImpl<ASTJoin::Any>(maps_any_full, rows, key_columns, keys_size, stored_block);
+			else
+				insertFromBlockImpl<ASTJoin::All>(maps_all_full, rows, key_columns, keys_size, stored_block);
+		}
 	}
 
 	if (!checkSizeLimits())
@@ -443,7 +498,8 @@ template <typename Map>
 struct Adder<ASTJoin::Left, ASTJoin::Any, Map>
 {
 	static void add(const Map & map, const typename Map::key_type & key, size_t num_columns_to_add, ColumnPlainPtrs & added_columns,
-		size_t i, IColumn::Filter * filter, IColumn::Offset_t & current_offset, IColumn::Offsets_t * offsets)
+		size_t i, IColumn::Filter * filter, IColumn::Offset_t & current_offset, IColumn::Offsets_t * offsets,
+		size_t num_columns_to_skip)
 	{
 		typename Map::const_iterator it = map.find(key);
 
@@ -451,7 +507,7 @@ struct Adder<ASTJoin::Left, ASTJoin::Any, Map>
 		{
 			it->second.setUsed();
 			for (size_t j = 0; j < num_columns_to_add; ++j)
-				added_columns[j]->insertFrom(*it->second.block->unsafeGetByPosition(j).column.get(), it->second.row_num);
+				added_columns[j]->insertFrom(*it->second.block->unsafeGetByPosition(num_columns_to_skip + j).column.get(), it->second.row_num);
 		}
 		else
 		{
@@ -465,7 +521,8 @@ template <typename Map>
 struct Adder<ASTJoin::Inner, ASTJoin::Any, Map>
 {
 	static void add(const Map & map, const typename Map::key_type & key, size_t num_columns_to_add, ColumnPlainPtrs & added_columns,
-					size_t i, IColumn::Filter * filter, IColumn::Offset_t & current_offset, IColumn::Offsets_t * offsets)
+		size_t i, IColumn::Filter * filter, IColumn::Offset_t & current_offset, IColumn::Offsets_t * offsets,
+		size_t num_columns_to_skip)
 	{
 		typename Map::const_iterator it = map.find(key);
 
@@ -475,7 +532,7 @@ struct Adder<ASTJoin::Inner, ASTJoin::Any, Map>
 
 			it->second.setUsed();
 			for (size_t j = 0; j < num_columns_to_add; ++j)
-				added_columns[j]->insertFrom(*it->second.block->unsafeGetByPosition(j).column.get(), it->second.row_num);
+				added_columns[j]->insertFrom(*it->second.block->unsafeGetByPosition(num_columns_to_skip + j).column.get(), it->second.row_num);
 		}
 		else
 			(*filter)[i] = 0;
@@ -486,7 +543,8 @@ template <ASTJoin::Kind KIND, typename Map>
 struct Adder<KIND, ASTJoin::All, Map>
 {
 	static void add(const Map & map, const typename Map::key_type & key, size_t num_columns_to_add, ColumnPlainPtrs & added_columns,
-					size_t i, IColumn::Filter * filter, IColumn::Offset_t & current_offset, IColumn::Offsets_t * offsets)
+		size_t i, IColumn::Filter * filter, IColumn::Offset_t & current_offset, IColumn::Offsets_t * offsets,
+		size_t num_columns_to_skip)
 	{
 		typename Map::const_iterator it = map.find(key);
 
@@ -497,7 +555,7 @@ struct Adder<KIND, ASTJoin::All, Map>
 			for (auto current = &static_cast<const typename Map::mapped_type::Base_t &>(it->second); current != nullptr; current = current->next)
 			{
 				for (size_t j = 0; j < num_columns_to_add; ++j)
-					added_columns[j]->insertFrom(*current->block->unsafeGetByPosition(j).column.get(), current->row_num);
+					added_columns[j]->insertFrom(*current->block->unsafeGetByPosition(num_columns_to_skip + j).column.get(), current->row_num);
 
 				++rows_joined;
 			}
@@ -538,19 +596,34 @@ void Join::joinBlockImpl(Block & block, const Maps & maps) const
 	{
 		key_columns[i] = block.getByName(key_names_left[i]).column;
 
-		if (key_columns[i]->isConst())
+		if (auto * col_const = dynamic_cast<const IColumnConst *>(key_columns[i]))
 		{
-			materialized_columns.emplace_back(dynamic_cast<const IColumnConst &>(*key_columns[i]).convertToFullColumn());
+			materialized_columns.emplace_back(col_const->convertToFullColumn());
 			key_columns[i] = materialized_columns.back();
 		}
 	}
 
+	size_t existing_columns = block.columns();
+
+	/** Если используется FULL или RIGHT JOIN, то столбцы из "левой" части надо материализовать.
+	  * Потому что, если они константы, то в "неприсоединённых" строчках, у них могут быть другие значения
+	  *  - значения по-умолчанию, которые могут отличаться от значений этих констант.
+	  */
+	if (getFullness(kind))
+	{
+		for (size_t i = 0; i < existing_columns; ++i)
+		{
+			auto & col = block.getByPosition(i).column;
+
+			if (auto * col_const = dynamic_cast<IColumnConst *>(col.get()))
+				col = col_const->convertToFullColumn();
+		}
+	}
+
 	/// Добавляем в блок новые столбцы.
 	size_t num_columns_to_add = sample_block_with_columns_to_add.columns();
 	ColumnPlainPtrs added_columns(num_columns_to_add);
 
-	size_t existing_columns = block.columns();
-
 	for (size_t i = 0; i < num_columns_to_add; ++i)
 	{
 		const ColumnWithTypeAndName & src_column = sample_block_with_columns_to_add.getByPosition(i);
@@ -575,6 +648,16 @@ void Join::joinBlockImpl(Block & block, const Maps & maps) const
 	if (strictness == ASTJoin::All)
 		offsets_to_replicate.reset(new IColumn::Offsets_t(rows));
 
+	/** Для LEFT/INNER JOIN, сохранённые блоки не содержат ключи.
+	  * Для FULL/RIGHT JOIN, сохранённые блоки содержат ключи;
+	  *  но они не будут использоваться на этой стадии соединения (а будут в AdderNonJoined), и их нужно пропустить.
+	  */
+	size_t num_columns_to_skip = 0;
+	if (getFullness(kind))
+		num_columns_to_skip = keys_size;
+
+//	std::cerr << num_columns_to_skip << "\n" << block.dumpStructure() << "\n" << blocks.front().dumpStructure() << "\n";
+
 	if (type == Type::KEY_64)
 	{
 		typedef typename Maps::MapUInt64 Map;
@@ -586,7 +669,8 @@ void Join::joinBlockImpl(Block & block, const Maps & maps) const
 		{
 			/// Строим ключ
 			UInt64 key = column.get64(i);
-			Adder<KIND, STRICTNESS, Map>::add(map, key, num_columns_to_add, added_columns, i, filter.get(), current_offset, offsets_to_replicate.get());
+			Adder<KIND, STRICTNESS, Map>::add(
+				map, key, num_columns_to_add, added_columns, i, filter.get(), current_offset, offsets_to_replicate.get(), num_columns_to_skip);
 		}
 	}
 	else if (type == Type::KEY_STRING)
@@ -605,7 +689,8 @@ void Join::joinBlockImpl(Block & block, const Maps & maps) const
 			{
 				/// Строим ключ
 				StringRef key(&data[i == 0 ? 0 : offsets[i - 1]], (i == 0 ? offsets[i] : (offsets[i] - offsets[i - 1])) - 1);
-				Adder<KIND, STRICTNESS, Map>::add(map, key, num_columns_to_add, added_columns, i, filter.get(), current_offset, offsets_to_replicate.get());
+				Adder<KIND, STRICTNESS, Map>::add(
+					map, key, num_columns_to_add, added_columns, i, filter.get(), current_offset, offsets_to_replicate.get(), num_columns_to_skip);
 			}
 		}
 		else if (const ColumnFixedString * column_string = typeid_cast<const ColumnFixedString *>(&column))
@@ -618,7 +703,8 @@ void Join::joinBlockImpl(Block & block, const Maps & maps) const
 			{
 				/// Строим ключ
 				StringRef key(&data[i * n], n);
-				Adder<KIND, STRICTNESS, Map>::add(map, key, num_columns_to_add, added_columns, i, filter.get(), current_offset, offsets_to_replicate.get());
+				Adder<KIND, STRICTNESS, Map>::add(
+					map, key, num_columns_to_add, added_columns, i, filter.get(), current_offset, offsets_to_replicate.get(), num_columns_to_skip);
 			}
 		}
 		else
@@ -636,7 +722,8 @@ void Join::joinBlockImpl(Block & block, const Maps & maps) const
 				? packFixed<UInt128>(i, keys_size, key_columns, key_sizes)
 				: hash128(i, keys_size, key_columns);
 
-			Adder<KIND, STRICTNESS, Map>::add(map, key, num_columns_to_add, added_columns, i, filter.get(), current_offset, offsets_to_replicate.get());
+			Adder<KIND, STRICTNESS, Map>::add(
+				map, key, num_columns_to_add, added_columns, i, filter.get(), current_offset, offsets_to_replicate.get(), num_columns_to_skip);
 		}
 	}
 	else
@@ -654,6 +741,60 @@ void Join::joinBlockImpl(Block & block, const Maps & maps) const
 }
 
 
+void Join::joinBlockImplCross(Block & block) const
+{
+	Block res = block.cloneEmpty();
+
+	/// Добавляем в блок новые столбцы.
+	size_t num_existing_columns = res.columns();
+	size_t num_columns_to_add = sample_block_with_columns_to_add.columns();
+
+	ColumnPlainPtrs src_left_columns(num_existing_columns);
+	ColumnPlainPtrs dst_left_columns(num_existing_columns);
+	ColumnPlainPtrs dst_right_columns(num_columns_to_add);
+
+	for (size_t i = 0; i < num_existing_columns; ++i)
+	{
+		src_left_columns[i] = block.unsafeGetByPosition(i).column;
+		dst_left_columns[i] = res.unsafeGetByPosition(i).column;
+	}
+
+	for (size_t i = 0; i < num_columns_to_add; ++i)
+	{
+		const ColumnWithTypeAndName & src_column = sample_block_with_columns_to_add.unsafeGetByPosition(i);
+		ColumnWithTypeAndName new_column = src_column.cloneEmpty();
+		res.insert(new_column);
+		dst_right_columns[i] = new_column.column;
+	}
+
+	size_t rows_left = block.rowsInFirstColumn();
+
+	/// NOTE Было бы оптимальнее использовать reserve, а также методы replicate для размножения значений левого блока.
+
+	for (size_t i = 0; i < rows_left; ++i)
+	{
+		for (const Block & block_right : blocks)
+		{
+			size_t rows_right = block_right.rowsInFirstColumn();
+
+			for (size_t col_num = 0; col_num < num_existing_columns; ++col_num)
+				for (size_t j = 0; j < rows_right; ++j)
+					dst_left_columns[col_num]->insertFrom(*src_left_columns[col_num], i);
+
+			for (size_t col_num = 0; col_num < num_columns_to_add; ++col_num)
+			{
+				const IColumn * column_right = block_right.unsafeGetByPosition(col_num).column;
+
+				for (size_t j = 0; j < rows_right; ++j)
+					dst_right_columns[col_num]->insertFrom(*column_right, j);
+			}
+		}
+	}
+
+	block = res;
+}
+
+
 void Join::checkTypesOfKeys(const Block & block_left, const Block & block_right) const
 {
 	size_t keys_size = key_names_left.size();
@@ -689,6 +830,10 @@ void Join::joinBlock(Block & block) const
 		joinBlockImpl<ASTJoin::Left, ASTJoin::All>(block, maps_all_full);
 	else if (kind == ASTJoin::Right && strictness == ASTJoin::All)
 		joinBlockImpl<ASTJoin::Inner, ASTJoin::All>(block, maps_all_full);
+	else if (kind == ASTJoin::Cross)
+		joinBlockImplCross(block);
+	else
+		throw Exception("Logical error: unknown combination of JOIN", ErrorCodes::LOGICAL_ERROR);
 }
 
 
@@ -759,9 +904,45 @@ struct AdderNonJoined<ASTJoin::All, Mapped>
 class NonJoinedBlockInputStream : public IProfilingBlockInputStream
 {
 public:
-	NonJoinedBlockInputStream(const Join & parent_, Block & left_sample_block_, size_t max_block_size_)
-		: parent(parent_), left_sample_block(left_sample_block_), max_block_size(max_block_size_)
+	NonJoinedBlockInputStream(const Join & parent_, Block & left_sample_block, size_t max_block_size_)
+		: parent(parent_), max_block_size(max_block_size_)
 	{
+		/** left_sample_block содержит ключи и "левые" столбцы.
+		  * result_sample_block - ключи, "левые" столбцы и "правые" столбцы.
+		  */
+
+		size_t num_keys = parent.key_names_left.size();
+		size_t num_columns_left = left_sample_block.columns() - num_keys;
+		size_t num_columns_right = parent.sample_block_with_columns_to_add.columns();
+
+		result_sample_block = left_sample_block;
+
+		/// Добавляем в блок новые столбцы.
+		for (size_t i = 0; i < num_columns_right; ++i)
+		{
+			const ColumnWithTypeAndName & src_column = parent.sample_block_with_columns_to_add.getByPosition(i);
+			ColumnWithTypeAndName new_column = src_column.cloneEmpty();
+			result_sample_block.insert(new_column);
+		}
+
+		column_numbers_left.reserve(num_columns_left);
+		column_numbers_keys_and_right.reserve(num_keys + num_columns_right);
+
+		for (size_t i = 0; i < num_keys + num_columns_left; ++i)
+		{
+			const String & name = left_sample_block.getByPosition(i).name;
+
+			if (parent.key_names_left.end() == std::find(parent.key_names_left.begin(), parent.key_names_left.end(), name))
+				column_numbers_left.push_back(i);
+			else
+				column_numbers_keys_and_right.push_back(i);
+		}
+
+		for (size_t i = 0; i < num_columns_right; ++i)
+			column_numbers_keys_and_right.push_back(num_keys + num_columns_left + i);
+
+		columns_left.resize(num_columns_left);
+		columns_keys_and_right.resize(num_keys + num_columns_right);
 	}
 
 	String getName() const override { return "NonJoined"; }
@@ -790,56 +971,59 @@ protected:
 
 private:
 	const Join & parent;
-	Block left_sample_block;
 	size_t max_block_size;
 
+	Block result_sample_block;
+	ColumnNumbers column_numbers_left;
+	ColumnNumbers column_numbers_keys_and_right;
+	ColumnPlainPtrs columns_left;
+	ColumnPlainPtrs columns_keys_and_right;
+
 	std::unique_ptr<void, std::function<void(void *)>> position;	/// type erasure
 
 
 	template <ASTJoin::Strictness STRICTNESS, typename Maps>
 	Block createBlock(const Maps & maps)
 	{
-		Block block = left_sample_block.cloneEmpty();
+		Block block = result_sample_block.cloneEmpty();
 
-		size_t num_columns_left = left_sample_block.columns();
-		ColumnPlainPtrs columns_left(num_columns_left);
+		size_t num_columns_left = column_numbers_left.size();
+		size_t num_columns_right = column_numbers_keys_and_right.size();
 
 		for (size_t i = 0; i < num_columns_left; ++i)
 		{
-			auto & column_with_name_and_type = block.getByPosition(i);
+			auto & column_with_name_and_type = block.getByPosition(column_numbers_left[i]);
 			column_with_name_and_type.column = column_with_name_and_type.type->createColumn();
 			columns_left[i] = column_with_name_and_type.column.get();
 		}
 
-		/// Добавляем в блок новые столбцы.
-		size_t num_columns_right = parent.sample_block_with_columns_to_add.columns();
-		ColumnPlainPtrs columns_right(num_columns_right);
-
 		for (size_t i = 0; i < num_columns_right; ++i)
 		{
-			const ColumnWithTypeAndName & src_column = parent.sample_block_with_columns_to_add.getByPosition(i);
-			ColumnWithTypeAndName new_column = src_column.cloneEmpty();
-			block.insert(new_column);
-			columns_right[i] = new_column.column;
-			columns_right[i]->reserve(src_column.column->size());
+			auto & column_with_name_and_type = block.getByPosition(column_numbers_keys_and_right[i]);
+			column_with_name_and_type.column = column_with_name_and_type.type->createColumn();
+			columns_keys_and_right[i] = column_with_name_and_type.column.get();
+			columns_keys_and_right[i]->reserve(column_with_name_and_type.column->size());
 		}
 
 		size_t rows_added = 0;
 		if (parent.type == Join::Type::KEY_64)
-			rows_added = fillColumns<STRICTNESS>(*maps.key64, num_columns_left, columns_left, num_columns_right, columns_right);
+			rows_added = fillColumns<STRICTNESS>(*maps.key64, num_columns_left, columns_left, num_columns_right, columns_keys_and_right);
 		else if (parent.type == Join::Type::KEY_STRING)
-			rows_added = fillColumns<STRICTNESS>(*maps.key_string, num_columns_left, columns_left, num_columns_right, columns_right);
+			rows_added = fillColumns<STRICTNESS>(*maps.key_string, num_columns_left, columns_left, num_columns_right, columns_keys_and_right);
 		else if (parent.type == Join::Type::HASHED)
-			rows_added = fillColumns<STRICTNESS>(*maps.hashed, num_columns_left, columns_left, num_columns_right, columns_right);
+			rows_added = fillColumns<STRICTNESS>(*maps.hashed, num_columns_left, columns_left, num_columns_right, columns_keys_and_right);
 		else
 			throw Exception("Unknown JOIN variant.", ErrorCodes::UNKNOWN_SET_DATA_VARIANT);
 
-		std::cerr << "rows added: " << rows_added << "\n";
+//		std::cerr << "rows added: " << rows_added << "\n";
 
 		if (!rows_added)
 			return Block();
 
-		std::cerr << block.dumpStructure() << "\n";
+/*		std::cerr << block.dumpStructure() << "\n";
+		WriteBufferFromFileDescriptor wb(STDERR_FILENO);
+		TabSeparatedBlockOutputStream out(wb);
+		out.write(block);*/
 
 		return block;
 	}
@@ -862,7 +1046,7 @@ private:
 
 		for (; it != end; ++it)
 		{
-			std::cerr << it->second.getUsed() << "\n";
+//			std::cerr << it->second.getUsed() << "\n";
 
 			if (it->second.getUsed())
 				continue;

dbms/src/Interpreters/executeQuery.cpp

@@ -166,6 +166,15 @@ static std::tuple<ASTPtr, BlockIO> executeQueryImpl(
 		/// Держим элемент списка процессов до конца обработки запроса.
 		res.process_list_entry = process_list_entry;
 
+		if (res.in)
+		{
+			if (IProfilingBlockInputStream * stream = dynamic_cast<IProfilingBlockInputStream *>(res.in.get()))
+			{
+				stream->setProgressCallback(context.getProgressCallback());
+				stream->setProcessListElement(context.getProcessListElement());
+			}
+		}
+
 		quota.addQuery(current_time);
 
 		/// Всё, что связано с логом запросов.

dbms/src/Parsers/ExpressionElementParsers.cpp

@@ -556,12 +556,8 @@ bool ParserWithOptionalAlias::parseImpl(Pos & pos, Pos end, ASTPtr & node, Pos &
 	{
 		String alias_name = typeid_cast<ASTIdentifier &>(*alias_node).name;
 
-		if (ASTFunction * func = typeid_cast<ASTFunction *>(&*node))
-			func->alias = alias_name;
-		else if (ASTIdentifier * ident = typeid_cast<ASTIdentifier *>(&*node))
-			ident->alias = alias_name;
-		else if (ASTLiteral * lit = typeid_cast<ASTLiteral *>(&*node))
-			lit->alias = alias_name;
+		if (ASTWithAlias * ast_with_alias = dynamic_cast<ASTWithAlias *>(node.get()))
+			ast_with_alias->alias = alias_name;
 		else
 		{
 			expected = "alias cannot be here";

dbms/src/Parsers/ParserJoin.cpp

@@ -24,12 +24,13 @@ bool ParserJoin::parseImpl(Pos & pos, Pos end, ASTPtr & node, Pos & max_parsed_p
 	ParserString s_left("LEFT", true, true);
 	ParserString s_right("RIGHT", true, true);
 	ParserString s_full("FULL", true, true);
+	ParserString s_cross("CROSS", true, true);
 	ParserString s_outer("OUTER", true, true);
 	ParserString s_join("JOIN", true, true);
 	ParserString s_using("USING", true, true);
 
 	ParserNotEmptyExpressionList exp_list;
-	ParserSubquery subquery;
+	ParserWithOptionalAlias subquery(ParserPtr(new ParserSubquery));
 	ParserIdentifier identifier;
 
 	ws.ignore(pos, end);
@@ -41,15 +42,13 @@ bool ParserJoin::parseImpl(Pos & pos, Pos end, ASTPtr & node, Pos & max_parsed_p
 
 	ws.ignore(pos, end);
 
+	bool has_strictness = true;
 	if (s_any.ignore(pos, end))
 		join->strictness = ASTJoin::Any;
 	else if (s_all.ignore(pos, end))
 		join->strictness = ASTJoin::All;
 	else
-	{
-		expected = "ANY|ALL";
-		return false;
-	}
+		has_strictness = false;
 
 	ws.ignore(pos, end);
 
@@ -61,16 +60,24 @@ bool ParserJoin::parseImpl(Pos & pos, Pos end, ASTPtr & node, Pos & max_parsed_p
 		join->kind = ASTJoin::Right;
 	else if (s_full.ignore(pos, end))
 		join->kind = ASTJoin::Full;
+	else if (s_cross.ignore(pos, end))
+		join->kind = ASTJoin::Cross;
 	else
 	{
-		expected = "INNER|LEFT|RIGHT|FULL";
+		expected = "INNER|LEFT|RIGHT|FULL|CROSS";
 		return false;
 	}
 
+	if (!has_strictness && join->kind != ASTJoin::Cross)
+		throw Exception("You must specify ANY or ALL for JOIN, before INNER or LEFT or RIGHT or FULL.", ErrorCodes::SYNTAX_ERROR);
+
+	if (has_strictness && join->kind == ASTJoin::Cross)
+		throw Exception("You must not specify ANY or ALL for CROSS JOIN.", ErrorCodes::SYNTAX_ERROR);
+
 	ws.ignore(pos, end);
 
-	/// Для всех JOIN-ов кроме INNER может присутствовать не обязательное слово "OUTER".
-	if (join->kind != ASTJoin::Inner && s_outer.ignore(pos, end))
+	/// Для всех JOIN-ов кроме INNER и CROSS может присутствовать не обязательное слово "OUTER".
+	if (join->kind != ASTJoin::Inner && join->kind != ASTJoin::Cross && s_outer.ignore(pos, end))
 		ws.ignore(pos, end);
 
 	if (!s_join.ignore(pos, end, max_parsed_pos, expected))
@@ -84,22 +91,23 @@ bool ParserJoin::parseImpl(Pos & pos, Pos end, ASTPtr & node, Pos & max_parsed_p
 
 	ws.ignore(pos, end);
 
-	/// Может быть указан алиас. На данный момент, он ничего не значит и не используется.
-	ParserAlias().ignore(pos, end);
-	ws.ignore(pos, end);
+	if (join->kind != ASTJoin::Cross)
+	{
+		if (!s_using.ignore(pos, end, max_parsed_pos, expected))
+			return false;
 
-	if (!s_using.ignore(pos, end, max_parsed_pos, expected))
-		return false;
+		ws.ignore(pos, end);
 
-	ws.ignore(pos, end);
+		if (!exp_list.parse(pos, end, join->using_expr_list, max_parsed_pos, expected))
+			return false;
 
-	if (!exp_list.parse(pos, end, join->using_expr_list, max_parsed_pos, expected))
-		return false;
-
-	ws.ignore(pos, end);
+		ws.ignore(pos, end);
+	}
 
 	join->children.push_back(join->table);
-	join->children.push_back(join->using_expr_list);
+
+	if (join->using_expr_list)
+		join->children.push_back(join->using_expr_list);
 
 	return true;
 }

dbms/src/Parsers/ParserSelectQuery.cpp

@@ -23,6 +23,7 @@ bool ParserSelectQuery::parseImpl(Pos & pos, Pos end, ASTPtr & node, Pos & max_p
 	ParserString s_select("SELECT", true, true);
 	ParserString s_distinct("DISTINCT", true, true);
 	ParserString s_from("FROM", true, true);
+	ParserString s_left("LEFT", true, true);
 	ParserString s_array("ARRAY", true, true);
 	ParserString s_join("JOIN", true, true);
 	ParserString s_using("USING", true, true);
@@ -166,8 +167,22 @@ bool ParserSelectQuery::parseImpl(Pos & pos, Pos end, ASTPtr & node, Pos & max_p
 	if (!parse_final_and_sample())
 		return false;
 
-	/// ARRAY JOIN expr list
-	if (s_array.ignore(pos, end, max_parsed_pos, expected))
+	/// [LEFT] ARRAY JOIN expr list
+	Pos saved_pos = pos;
+	bool has_array_join = false;
+	if (s_left.ignore(pos, end, max_parsed_pos, expected) && ws.ignore(pos, end) && s_array.ignore(pos, end, max_parsed_pos, expected))
+	{
+		select_query->array_join_is_left = true;
+		has_array_join = true;
+	}
+	else
+	{
+		pos = saved_pos;
+		if (s_array.ignore(pos, end, max_parsed_pos, expected))
+			has_array_join = true;
+	}
+
+	if (has_array_join)
 	{
 		ws.ignore(pos, end);
 
@@ -182,7 +197,7 @@ bool ParserSelectQuery::parseImpl(Pos & pos, Pos end, ASTPtr & node, Pos & max_p
 		ws.ignore(pos, end);
 	}
 
-	/// [GLOBAL] ANY|ALL INNER|LEFT JOIN (subquery) USING tuple
+	/// [GLOBAL] [ANY|ALL] INNER|LEFT|RIGHT|FULL|CROSS [OUTER] JOIN (subquery)|table_name USING tuple
 	join.parse(pos, end, select_query->join, max_parsed_pos, expected);
 
 	if (!parse_final_and_sample())

dbms/src/Parsers/formatAST.cpp

@@ -63,12 +63,24 @@ String backQuoteIfNeed(const String & x)
 }
 
 
-String hightlight(const String & keyword, const String & color_sequence, const bool hilite)
+static String hightlight(const String & keyword, const String & color_sequence, const bool hilite)
 {
 	return hilite ? color_sequence + keyword + hilite_none : keyword;
 }
 
 
+static void writeAlias(const String & name, std::ostream & s, bool hilite, bool one_line)
+{
+	s << (hilite ? hilite_keyword : "") << " AS " << (hilite ? hilite_alias : "");
+
+	WriteBufferFromOStream wb(s, 32);
+	writeProbablyBackQuotedString(name, wb);
+	wb.next();
+
+	s << (hilite ? hilite_none : "");
+}
+
+
 void formatAST(const ASTExpressionList 		& ast, std::ostream & s, size_t indent, bool hilite, bool one_line, bool need_parens)
 {
 	for (ASTs::const_iterator it = ast.children.begin(); it != ast.children.end(); ++it)
@@ -151,7 +163,9 @@ void formatAST(const ASTSelectQuery 		& ast, std::ostream & s, size_t indent, bo
 
 	if (ast.array_join_expression_list)
 	{
-		s << (hilite ? hilite_keyword : "") << nl_or_ws << indent_str << "ARRAY JOIN " << (hilite ? hilite_none : "");
+		s << (hilite ? hilite_keyword : "") << nl_or_ws << indent_str
+			<< (ast.array_join_is_left ? "LEFT " : "") << "ARRAY JOIN " << (hilite ? hilite_none : "");
+
 		one_line
 			? formatAST(*ast.array_join_expression_list, s, indent, hilite, one_line)
 			: formatExpressionListMultiline(typeid_cast<const ASTExpressionList &>(*ast.array_join_expression_list), s, indent, hilite);
@@ -245,12 +259,23 @@ void formatAST(const ASTSelectQuery 		& ast, std::ostream & s, size_t indent, bo
 
 void formatAST(const ASTSubquery 			& ast, std::ostream & s, size_t indent, bool hilite, bool one_line, bool need_parens)
 {
+	/// Если есть алиас, то требуются скобки вокруг всего выражения, включая алиас. Потому что запись вида 0 AS x + 0 синтаксически некорректна.
+	if (need_parens && !ast.alias.empty())
+		s << '(';
+
 	std::string indent_str = one_line ? "" : std::string(4 * indent, ' ');
 	std::string nl_or_nothing = one_line ? "" : "\n";
 
 	s << nl_or_nothing << indent_str << "(" << nl_or_nothing;
 	formatAST(*ast.children[0], s, indent + 1, hilite, one_line);
 	s << nl_or_nothing << indent_str << ")";
+
+	if (!ast.alias.empty())
+	{
+		writeAlias(ast.alias, s, hilite, one_line);
+		if (need_parens)
+			s << ')';
+	}
 }
 
 void formatAST(const ASTCreateQuery 		& ast, std::ostream & s, size_t indent, bool hilite, bool one_line, bool need_parens)
@@ -461,17 +486,6 @@ void formatAST(const ASTInsertQuery 		& ast, std::ostream & s, size_t indent, bo
 	}
 }
 
-static void writeAlias(const String & name, std::ostream & s, bool hilite, bool one_line)
-{
-	s << (hilite ? hilite_keyword : "") << " AS " << (hilite ? hilite_alias : "");
-
-	WriteBufferFromOStream wb(s, 32);
-	writeProbablyBackQuotedString(name, wb);
-	wb.next();
-
-	s << (hilite ? hilite_none : "");
-}
-
 void formatAST(const ASTFunction 			& ast, std::ostream & s, size_t indent, bool hilite, bool one_line, bool need_parens)
 {
 	/// Если есть алиас, то требуются скобки вокруг всего выражения, включая алиас. Потому что запись вида 0 AS x + 0 синтаксически некорректна.
@@ -829,21 +843,30 @@ void formatAST(const ASTSet & ast, std::ostream & s, size_t indent, bool hilite,
 
 void formatAST(const ASTJoin & ast, std::ostream & s, size_t indent, bool hilite, bool one_line, bool need_parens)
 {
-	s << (hilite ? hilite_keyword : "")
-		<< (ast.locality == ASTJoin::Global ? "GLOBAL " : "")
-		<< (ast.strictness == ASTJoin::Any ? "ANY " : "ALL ")
-		<< (ast.kind == ASTJoin::Inner ? "INNER "
-			: (ast.kind == ASTJoin::Left ? "LEFT "
-			: (ast.kind == ASTJoin::Right ? "RIGHT "
-			: "FULL OUTER ")))
-		<< "JOIN "
+	s << (hilite ? hilite_keyword : "");
+
+	if (ast.locality == ASTJoin::Global)
+		s << "GLOBAL ";
+
+	if (ast.kind != ASTJoin::Cross)
+		s << (ast.strictness == ASTJoin::Any ? "ANY " : "ALL ");
+
+	s << (ast.kind == ASTJoin::Inner ? "INNER "
+		: (ast.kind == ASTJoin::Left ? "LEFT "
+		: (ast.kind == ASTJoin::Right ? "RIGHT "
+		: (ast.kind == ASTJoin::Cross ? "CROSS "
+		: "FULL OUTER "))));
+
+	s << "JOIN "
 		<< (hilite ? hilite_none : "");
 
 	formatAST(*ast.table, s, indent, hilite, one_line, need_parens);
 
-	s << (hilite ? hilite_keyword : "") << " USING " << (hilite ? hilite_none : "");
-
-	formatAST(*ast.using_expr_list, s, indent, hilite, one_line, need_parens);
+	if (ast.kind != ASTJoin::Cross)
+	{
+		s << (hilite ? hilite_keyword : "") << " USING " << (hilite ? hilite_none : "");
+		formatAST(*ast.using_expr_list, s, indent, hilite, one_line, need_parens);
+	}
 }
 
 void formatAST(const ASTCheckQuery & ast, std::ostream & s, size_t indent, bool hilite, bool one_line, bool need_parens)

dbms/src/Storages/StorageSystemColumns.cpp

@@ -1,6 +1,10 @@
 #include <DB/Storages/StorageSystemColumns.h>
+#include <DB/Storages/MergeTree/MergeTreeData.h>
+#include <DB/Storages/StorageMergeTree.h>
+#include <DB/Storages/StorageReplicatedMergeTree.h>
 #include <DB/Columns/ColumnString.h>
 #include <DB/DataTypes/DataTypeString.h>
+#include <DB/DataTypes/DataTypesNumberFixed.h>
 #include <DB/DataStreams/OneBlockInputStream.h>
 #include <DB/Common/VirtualColumnUtils.h>
 
@@ -15,7 +19,8 @@ StorageSystemColumns::StorageSystemColumns(const std::string & name_)
 		{ "name",               new DataTypeString },
 		{ "type",               new DataTypeString },
 		{ "default_type",       new DataTypeString },
-		{ "default_expression", new DataTypeString }
+		{ "default_expression", new DataTypeString },
+		{ "bytes",				new DataTypeUInt64 },
 	}
 {
 }
@@ -103,6 +108,7 @@ BlockInputStreams StorageSystemColumns::read(
 	ColumnPtr type_column = new ColumnString;
 	ColumnPtr default_type_column = new ColumnString;
 	ColumnPtr default_expression_column = new ColumnString;
+	ColumnPtr bytes_column = new ColumnUInt64;
 
 	size_t rows = filtered_database_column->size();
 	for (size_t i = 0; i < rows; ++i)
@@ -112,6 +118,7 @@ BlockInputStreams StorageSystemColumns::read(
 
 		NamesAndTypesList columns;
 		ColumnDefaults column_defaults;
+		std::unordered_map<String, size_t> column_sizes;
 
 		{
 			StoragePtr storage = storages.at(std::make_pair(database_name, table_name));
@@ -120,6 +127,26 @@ BlockInputStreams StorageSystemColumns::read(
 			columns = storage->getColumnsList();
 			columns.insert(std::end(columns), std::begin(storage->alias_columns), std::end(storage->alias_columns));
 			column_defaults = storage->column_defaults;
+
+			/** Данные о размерах столбцов для таблиц семейства MergeTree.
+			  * NOTE: В дальнейшем можно сделать интерфейс, позволяющий получить размеры столбцов у IStorage.
+			  */
+			if (auto storage_concrete = dynamic_cast<StorageMergeTree *>(storage.get()))
+			{
+				column_sizes = storage_concrete->getData().getColumnSizes();
+			}
+			else if (auto storage_concrete = dynamic_cast<StorageReplicatedMergeTree *>(storage.get()))
+			{
+				column_sizes = storage_concrete->getData().getColumnSizes();
+
+				auto unreplicated_data = storage_concrete->getUnreplicatedData();
+				if (unreplicated_data)
+				{
+					auto unreplicated_column_sizes = unreplicated_data->getColumnSizes();
+					for (const auto & name_size : unreplicated_column_sizes)
+						column_sizes[name_size.first] += name_size.second;
+				}
+			}
 		}
 
 		for (const auto & column : columns)
@@ -129,16 +156,26 @@ BlockInputStreams StorageSystemColumns::read(
 			name_column->insert(column.name);
 			type_column->insert(column.type->getName());
 
-			const auto it = column_defaults.find(column.name);
-			if (it == std::end(column_defaults))
 			{
-				default_type_column->insertDefault();
-				default_expression_column->insertDefault();
+				const auto it = column_defaults.find(column.name);
+				if (it == std::end(column_defaults))
+				{
+					default_type_column->insertDefault();
+					default_expression_column->insertDefault();
+				}
+				else
+				{
+					default_type_column->insert(toString(it->second.type));
+					default_expression_column->insert(queryToString(it->second.expression));
+				}
 			}
-			else
+
 			{
-				default_type_column->insert(toString(it->second.type));
-				default_expression_column->insert(queryToString(it->second.expression));
+				const auto it = column_sizes.find(column.name);
+				if (it == std::end(column_sizes))
+					bytes_column->insertDefault();
+				else
+					bytes_column->insert(it->second);
 			}
 		}
 	}
@@ -151,6 +188,7 @@ BlockInputStreams StorageSystemColumns::read(
 	block.insert(ColumnWithTypeAndName(type_column, new DataTypeString, "type"));
 	block.insert(ColumnWithTypeAndName(default_type_column, new DataTypeString, "default_type"));
 	block.insert(ColumnWithTypeAndName(default_expression_column, new DataTypeString, "default_expression"));
+	block.insert(ColumnWithTypeAndName(bytes_column, new DataTypeUInt64, "bytes"));
 
 	return BlockInputStreams{ 1, new OneBlockInputStream(block) };
 }

dbms/tests/queries/0_stateless/00201_array_uniq.reference

@@ -0,0 +1 @@
+10	10	10	1297	1297	1299	1299

dbms/tests/queries/0_stateless/00201_array_uniq.sql

@@ -0,0 +1 @@
+SELECT uniqExact(x), length(groupUniqArray(x)), arrayUniq(groupArray(x)), uniqExact(y), arrayUniq(groupArray(y)), uniqExact(concat(toString(x), '_', y)), arrayUniq(groupArray(x), groupArray(y)) FROM (SELECT round(log(intHash32(number))) AS x, toString(round(cbrt(intHash32(number)))) AS y FROM system.numbers LIMIT 10000);

dbms/tests/queries/0_stateless/00202_cross_join.reference

@@ -0,0 +1,15 @@
+0	0
+0	1
+0	2
+0	3
+0	4
+1	0
+1	1
+1	2
+1	3
+1	4
+2	0
+2	1
+2	2
+2	3
+2	4

dbms/tests/queries/0_stateless/00202_cross_join.sql

@@ -0,0 +1 @@
+SELECT x, y FROM (SELECT number AS x FROM system.numbers LIMIT 3) CROSS JOIN (SELECT number AS y FROM system.numbers LIMIT 5);

dbms/tests/queries/0_stateless/00203_full_join.reference

@@ -0,0 +1,40 @@
+1	Hello	[]
+2	Hello	[]
+3	Hello	[0,1,2]
+4		[0,1,2,3]
+5		[0,1,2,3,4]
+1	Hello
+2	Hello
+3	Hello
+4	
+5	
+1	[]
+2	[]
+3	[0,1,2]
+4	[0,1,2,3]
+5	[0,1,2,3,4]
+Hello	[]
+Hello	[]
+Hello	[0,1,2]
+	[0,1,2,3]
+	[0,1,2,3,4]
+1
+2
+3
+4
+5
+3	Hello	[0,1,2]
+4		[0,1,2,3]
+5		[0,1,2,3,4]
+3	Hello
+4	
+5	
+3	[0,1,2]
+4	[0,1,2,3]
+5	[0,1,2,3,4]
+Hello	[0,1,2]
+	[0,1,2,3]
+	[0,1,2,3,4]
+3
+4
+5

dbms/tests/queries/0_stateless/00203_full_join.sql

@@ -0,0 +1,11 @@
+SELECT k, x, y FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY FULL JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;
+SELECT k, x FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY FULL JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;
+SELECT k, y FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY FULL JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;
+SELECT x, y FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY FULL JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;
+SELECT k FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY FULL JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;
+
+SELECT k, x, y FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY RIGHT JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;
+SELECT k, x FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY RIGHT JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;
+SELECT k, y FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY RIGHT JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;
+SELECT x, y FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY RIGHT JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;
+SELECT k FROM (SELECT arrayJoin([1, 2, 3]) AS k, 'Hello' AS x) ANY RIGHT JOIN (SELECT range(k) AS y, arrayJoin([3, 4, 5]) AS k) USING k WHERE k < 10 ORDER BY k;

dbms/tests/queries/0_stateless/00204_extract_url_parameter.reference

@@ -0,0 +1 @@
+111

dbms/tests/queries/0_stateless/00204_extract_url_parameter.sql

@@ -0,0 +1 @@
+SELECT extractURLParameter('http://test.com/?testq=aaa&q=111', 'q');

dbms/tests/queries/0_stateless/00205_scalar_subqueries.reference

@@ -0,0 +1,5 @@
+1	1
+1
+1	1
+('2015-01-02','Hello')
+('2015-01-02','Hello')	('2015-01-02','Hello')	1	1

dbms/tests/queries/0_stateless/00205_scalar_subqueries.sql

@@ -0,0 +1,5 @@
+SELECT (SELECT (SELECT (SELECT (SELECT (SELECT count() FROM (SELECT * FROM system.numbers LIMIT 10)))))) = (SELECT 10), ((SELECT 1, 'Hello', [1, 2]).3)[1];
+SELECT toUInt64((SELECT 9)) IN (SELECT number FROM system.numbers LIMIT 10);
+SELECT (SELECT toDate('2015-01-02')) = toDate('2015-01-02'), 'Hello' = (SELECT 'Hello');
+SELECT (SELECT toDate('2015-01-02'), 'Hello');
+SELECT (SELECT toDate('2015-01-02'), 'Hello') AS x, x, identity((SELECT 1)), identity((SELECT 1) AS y);

dbms/tests/queries/0_stateless/00206_empty_array_to_single.reference

@@ -0,0 +1,14 @@
+[1,2]
+[0]
+[4,5,6]
+['']	['0000-00-00']	['0000-00-00 00:00:00']
+[0]	['']	['0000-00-00 00:00:00']	['0000-00-00']
+[0]	['0']	['2015-01-01 00:00:00']	['2015-01-01']
+[0,1]	['']	['2015-01-01 00:00:00','2015-01-01 00:00:01']	['2015-01-01','2015-01-02']
+[0]	['0']	['2015-01-01 00:00:00','2015-01-01 00:00:01','2015-01-01 00:00:02']	['2015-01-01','2015-01-02','2015-01-03']
+[0]	['']	['2015-01-01 00:00:00','2015-01-01 00:00:01','2015-01-01 00:00:02','2015-01-01 00:00:03']	['0000-00-00']
+[0,1]	['0']	['0000-00-00 00:00:00']	['2015-01-01']
+[0]	['']	['2015-01-01 00:00:00']	['2015-01-01','2015-01-02']
+[0]	['0']	['2015-01-01 00:00:00','2015-01-01 00:00:01']	['2015-01-01','2015-01-02','2015-01-03']
+[0,1]	['']	['2015-01-01 00:00:00','2015-01-01 00:00:01','2015-01-01 00:00:02']	['0000-00-00']
+[0]	['0']	['2015-01-01 00:00:00','2015-01-01 00:00:01','2015-01-01 00:00:02','2015-01-01 00:00:03']	['2015-01-01']

dbms/tests/queries/0_stateless/00206_empty_array_to_single.sql

@@ -0,0 +1,8 @@
+SELECT emptyArrayToSingle(arrayFilter(x -> x != 99, arrayJoin([[1, 2], [99], [4, 5, 6]])));
+SELECT emptyArrayToSingle(emptyArrayString()), emptyArrayToSingle(emptyArrayDate()), emptyArrayToSingle(emptyArrayDateTime());
+
+SELECT 
+    emptyArrayToSingle(range(number % 3)), 
+    emptyArrayToSingle(arrayMap(x -> toString(x), range(number % 2))), 
+    emptyArrayToSingle(arrayMap(x -> toDateTime('2015-01-01 00:00:00') + x, range(number % 5))), 
+    emptyArrayToSingle(arrayMap(x -> toDate('2015-01-01') + x, range(number % 4))) FROM system.numbers LIMIT 10;

dbms/tests/queries/0_stateless/00207_left_array_join.reference

@@ -0,0 +1,23 @@
+0
+1
+2
+2
+3
+4
+5
+5
+6
+7
+0	[]	0
+1	[0]	0
+2	[0,1]	0
+2	[0,1]	1
+3	[]	0
+4	[0]	0
+5	[0,1]	0
+5	[0,1]	1
+6	[]	0
+7	[0]	0
+8	[0,1]	0
+8	[0,1]	1
+9	[]	0

dbms/tests/queries/0_stateless/00207_left_array_join.sql

@@ -0,0 +1,2 @@
+SELECT number FROM system.numbers LEFT ARRAY JOIN range(number % 3) AS arr LIMIT 10;
+SELECT number, arr, x FROM (SELECT number, range(number % 3) AS arr FROM system.numbers LIMIT 10) LEFT ARRAY JOIN arr AS x;

dbms/tests/queries/0_stateless/00208_agg_state_merge.reference

@@ -0,0 +1,7 @@
+0	15	15
+1	14	14
+2	14	14
+3	15	15
+4	9	9
+5	9	9
+6	9	9

dbms/tests/queries/0_stateless/00208_agg_state_merge.sql

@@ -0,0 +1 @@
+SELECT k % 7 AS k2, finalizeAggregation(uniqMergeState(state)), uniqMerge(state) FROM (SELECT k, uniqState(x) AS state FROM (SELECT number % 11 AS k, intDiv(number, 7)  AS x FROM system.numbers LIMIT 100) GROUP BY k) GROUP BY k2 ORDER BY k2;

dbms/tests/queries/0_stateless/00209_insert_select_extremes.reference

@@ -0,0 +1,11 @@
+
+1
+1
+
+0
+
+0
+
+1
+1
+4	1	1

dbms/tests/queries/0_stateless/00209_insert_select_extremes.sql

@@ -0,0 +1,11 @@
+DROP TABLE IF EXISTS test.test;
+CREATE TABLE test.test (x UInt8) ENGINE = Log;
+
+INSERT INTO test.test SELECT 1 AS x;
+INSERT INTO test.test SELECT 1 AS x SETTINGS extremes = 1;
+INSERT INTO test.test SELECT 1 AS x GROUP BY 1 WITH TOTALS;
+INSERT INTO test.test SELECT 1 AS x GROUP BY 1 WITH TOTALS SETTINGS extremes = 1;
+
+SELECT count(), min(x), max(x) FROM test.test;
+
+DROP TABLE test.test;

dbms/tests/queries/0_stateless/00210_insert_select_extremes_http.reference

@@ -0,0 +1,3 @@
+
+1
+1

dbms/tests/queries/0_stateless/00210_insert_select_extremes_http.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+
+curl -sS http://localhost:8123/?extremes=1 -d @- <<< "DROP TABLE IF EXISTS test.test"
+curl -sS http://localhost:8123/?extremes=1 -d @- <<< "CREATE TABLE test.test (x UInt8) ENGINE = Log"
+curl -sS http://localhost:8123/?extremes=1 -d @- <<< "INSERT INTO test.test SELECT 1 AS x"
+curl -sS http://localhost:8123/?extremes=1 -d @- <<< "DROP TABLE test.test"