Merge

dbms/benchmark/memsql/benchmark.sh

@@ -11,7 +11,7 @@ cat "$QUERIES_FILE" | sed "s/{table}/${TABLE}/g" | while read query; do
 	echo -n "["
 	for i in $(seq 1 $TRIES); do
 
-		RES=$(mysql -u root -h 127.0.0.1 -P 3306 --database=test -t -vvv -e "$query" 2>&1 | grep 'in set' | grep -oP '\d+\.\d+')
+		RES=$(mysql -u root -h 127.0.0.1 -P 3306 --database=test -t -vvv -e "$query" 2>&1 | grep ' set ' | grep -oP '\d+\.\d+')
 
 		[[ "$?" == "0" ]] && echo -n "$RES" || echo -n "null"
 		[[ "$i" != $TRIES ]] && echo -n ", "

dbms/benchmark/memsql/instructions.txt

@@ -1,3 +1,5 @@
+Note: column store in MemSQL was introduced in Feb 2014.
+
 http://www.memsql.com/download/
 http://docs.memsql.com/docs/latest/setup/setup_onprem.html
 wget http://download.memsql.com/8d9f4c4d99a547baa40ba097b171bd15/memsql-3.2.x86_64.deb

dbms/benchmark/memsql/queries.sql

@@ -39,5 +39,5 @@ SELECT Title, count(*) AS PageViews FROM hits_10m WHERE CounterID = 34 AND Event
 SELECT URL, count(*) AS PageViews FROM hits_10m WHERE CounterID = 34 AND EventDate >= DATE('2013-07-01') AND EventDate <= DATE('2013-07-31') AND NOT Refresh AND IsLink AND NOT IsDownload GROUP BY URL ORDER BY PageViews DESC LIMIT 1000;
 SELECT TraficSourceID, SearchEngineID, AdvEngineID, CASE WHEN SearchEngineID = 0 AND AdvEngineID = 0 THEN  Referer ELSE '' END AS Src, URL AS Dst, count(*) AS PageViews FROM hits_10m WHERE CounterID = 34 AND EventDate >= DATE('2013-07-01') AND EventDate <= DATE('2013-07-31') AND NOT Refresh GROUP BY TraficSourceID, SearchEngineID, AdvEngineID, Src, Dst ORDER BY PageViews DESC LIMIT 1000; 
 SELECT URLHash, EventDate, count(*) AS PageViews FROM hits_10m WHERE CounterID = 34 AND EventDate >= DATE('2013-07-01') AND EventDate <= DATE('2013-07-31') AND NOT Refresh AND TraficSourceID IN (-1, 6) AND RefererHash =  6202628419148573758  GROUP BY URLHash, EventDate ORDER BY PageViews DESC LIMIT 100000; 
-SELECT WindowClientWidth, WindowClientHeight, count(*) AS PageViews FROM hits_10m WHERE CounterID = 34 AND EventDate >= DATE('2013-07-01') AND EventDate <= DATE('2013-07-31') AND NOT Refresh AND NOT DontCountHits AND URLHash =  6202628419148573758 GROUP BY WindowClientWidth, WindowClientHeight ORDER BY PageViews DESC LIMIT 10000; 
+SELECT WindowClientWidth, WindowClientHeight, count(*) AS PageViews FROM hits_10m WHERE CounterID = 34 AND EventDate >= DATE('2013-07-01') AND EventDate <= DATE('2013-07-31') AND NOT Refresh AND NOT DontCountHits AND URLHash = 6202628419148573758 GROUP BY WindowClientWidth, WindowClientHeight ORDER BY PageViews DESC LIMIT 10000; 
 SELECT EventTime - INTERVAL SECOND(EventTime) SECOND AS Minute, count(*) AS PageViews FROM hits_10m WHERE CounterID = 34 AND EventDate >= DATE('2013-07-01') AND EventDate <= DATE('2013-07-02') AND NOT Refresh AND NOT DontCountHits GROUP BY Minute ORDER BY Minute;

dbms/include/DB/Storages/IStorage.h

@@ -110,7 +110,11 @@ public:
 	  */
 	TableFullWriteLockPtr lockForAlter()
 	{
-		return std::make_pair(lockDataForAlter(), lockStructureForAlter());
+		/// Порядок вычисления важен.
+		auto data_lock = lockDataForAlter();
+		auto structure_lock = lockStructureForAlter();
+
+		return {std::move(data_lock), std::move(structure_lock)};
 	}
 
 	/** Не дает изменять данные в таблице. (Более того, не дает посмотреть на структуру таблицы с намерением изменить данные).

dbms/src/Interpreters/InterpreterRenameQuery.cpp

@@ -26,6 +26,37 @@ InterpreterRenameQuery::InterpreterRenameQuery(ASTPtr query_ptr_, Context & cont
 }
 
 
+struct RenameDescription
+{
+	RenameDescription(const ASTRenameQuery::Element & elem, const String & path, const String & current_database) :
+		from_database_name(elem.from.database.empty() ? current_database : elem.from.database),
+		from_database_name_escaped(escapeForFileName(from_database_name)),
+		from_table_name(elem.from.table),
+		from_table_name_escaped(escapeForFileName(from_table_name)),
+		from_metadata_path(path + "metadata/" + from_database_name_escaped + "/"
+			+ (!from_table_name.empty() ?  from_table_name_escaped + ".sql" : "")),
+		to_database_name(elem.to.database.empty() ? current_database : elem.to.database),
+		to_database_name_escaped(escapeForFileName(to_database_name)),
+		to_table_name(elem.to.table),
+		to_table_name_escaped(escapeForFileName(to_table_name)),
+		to_metadata_path(path + "metadata/" + to_database_name_escaped + "/"
+			+ (!to_table_name.empty() ?  to_table_name_escaped + ".sql" : ""))
+	{}
+
+	String from_database_name;
+	String from_database_name_escaped;
+	String from_table_name;
+	String from_table_name_escaped;
+	String from_metadata_path;
+
+	String to_database_name;
+	String to_database_name_escaped;
+	String to_table_name;
+	String to_table_name_escaped;
+	String to_metadata_path;
+};
+
+
 void InterpreterRenameQuery::execute()
 {
 	String path = context.getPath();
@@ -37,38 +68,59 @@ void InterpreterRenameQuery::execute()
 	  *  или состояние может стать неконсистентным. (Это имеет смысл исправить.)
 	  */
 
-	for (ASTRenameQuery::Elements::const_iterator it = rename.elements.begin(); it != rename.elements.end(); ++it)
+	std::vector<RenameDescription> descriptions;
+	descriptions.reserve(rename.elements.size());
+
+	/// Для того, чтобы захватывать блокировки таблиц в одном и том же порядке в разных RENAME-ах.
+	struct UniqueTableName
 	{
-		String from_database_name = it->from.database.empty() ? current_database : it->from.database;
-		String from_database_name_escaped = escapeForFileName(from_database_name);
-		String from_table_name = it->from.table;
-		String from_table_name_escaped = escapeForFileName(from_table_name);
-		String from_metadata_path = path + "metadata/" + from_database_name_escaped + "/" + (!from_table_name.empty() ?  from_table_name_escaped + ".sql" : "");
+		String database_name;
+		String table_name;
 
-		String to_database_name = it->to.database.empty() ? current_database : it->to.database;
-		String to_database_name_escaped = escapeForFileName(to_database_name);
-		String to_table_name = it->to.table;
-		String to_table_name_escaped = escapeForFileName(to_table_name);
-		String to_metadata_path = path + "metadata/" + to_database_name_escaped + "/" + (!to_table_name.empty() ?  to_table_name_escaped + ".sql" : "");
+		UniqueTableName(const String & database_name_, const String & table_name_)
+			: database_name(database_name_), table_name(table_name_) {}
 
-		/// Заблокировать таблицу нужно при незаблокированном контексте.
-		StoragePtr table = context.getTable(from_database_name, from_table_name);
-		auto table_lock = table->lockForAlter();
+		bool operator< (const UniqueTableName & rhs) const
+		{
+			return std::tie(database_name, table_name) < std::tie(rhs.database_name, rhs.table_name);
+		}
+	};
 
-		/** Все таблицы переименовываются под глобальной блокировкой. */
-		Poco::ScopedLock<Poco::Mutex> lock(context.getMutex());
+	std::set<UniqueTableName> unique_tables;
 
-		context.assertTableDoesntExist(to_database_name, to_table_name);
+	for (const auto & elem : rename.elements)
+	{
+		descriptions.emplace_back(elem, path, current_database);
+		unique_tables.emplace(descriptions.back().from_database_name, descriptions.back().from_table_name);
+	}
+
+	std::vector<IStorage::TableFullWriteLockPtr> locks;
+	locks.reserve(unique_tables.size());
+
+	for (const auto & names : unique_tables)
+		if (auto table = context.tryGetTable(names.database_name, names.table_name))
+			locks.emplace_back(table->lockForAlter());
+
+	/** Все таблицы заблокированы. Теперь можно блокировать Context. Порядок важен, чтобы избежать deadlock-ов.
+	  * Это обеспечивает атомарность всех указанных RENAME с точки зрения пользователя СУБД,
+	  *  но лишь в случаях, когда в процессе переименования не было исключений и сервер не падал.
+	  */
+
+	Poco::ScopedLock<Poco::Mutex> lock(context.getMutex());
+
+	for (const auto & elem : descriptions)
+	{
+		context.assertTableDoesntExist(elem.to_database_name, elem.to_table_name);
 
 		/// Уведомляем таблицу о том, что она переименовывается. Если таблица не поддерживает переименование - кинется исключение.
-
-		table->rename(path + "data/" + to_database_name_escaped + "/", to_database_name, to_table_name);
+		StoragePtr table = context.getTable(elem.from_database_name, elem.from_table_name);
+		table->rename(path + "data/" + elem.to_database_name_escaped + "/", elem.to_database_name, elem.to_table_name);
 
 		/// Пишем новый файл с метаданными.
 		{
 			String create_query;
 			{
-				ReadBufferFromFile in(from_metadata_path, 1024);
+				ReadBufferFromFile in(elem.from_metadata_path, 1024);
 				WriteBufferFromString out(create_query);
 				copyData(in, out);
 			}
@@ -82,22 +134,22 @@ void InterpreterRenameQuery::execute()
 
 			/// Распарсенный запрос должен заканчиваться на конец входных данных или на точку с запятой.
 			if (!parse_res || (pos != end && *pos != ';'))
-				throw Exception(getSyntaxErrorMessage(parse_res, create_query.data(), end, pos, expected, "in file " + from_metadata_path),
+				throw Exception(getSyntaxErrorMessage(parse_res, create_query.data(), end, pos, expected, "in file " + elem.from_metadata_path),
 					ErrorCodes::SYNTAX_ERROR);
 
-			typeid_cast<ASTCreateQuery &>(*ast).table = to_table_name;
+			typeid_cast<ASTCreateQuery &>(*ast).table = elem.to_table_name;
 
-			Poco::FileOutputStream ostr(to_metadata_path);
+			Poco::FileOutputStream ostr(elem.to_metadata_path);
 			formatAST(*ast, ostr, 0, false);
 			ostr << '\n';
 		}
 
 		/// Переименовываем таблицу в контексте.
-		context.addTable(to_database_name, to_table_name,
-			context.detachTable(from_database_name, from_table_name));
+		context.addTable(elem.to_database_name, elem.to_table_name,
+			context.detachTable(elem.from_database_name, elem.from_table_name));
 
 		/// Удаляем старый файл с метаданными.
-		Poco::File(from_metadata_path).remove();
+		Poco::File(elem.from_metadata_path).remove();
 	}
 }