13.4. MySQL Transactional and Locking Statements

START TRANSACTION | BEGIN [WORK]
COMMIT [WORK] [AND [NO] CHAIN] [[NO] RELEASE]
ROLLBACK [WORK] [AND [NO] CHAIN] [[NO] RELEASE]
SET AUTOCOMMIT = {0 | 1}

The START TRANSACTION and BEGIN statement begin a new transaction. COMMIT commits the current transaction, making its changes permanent. ROLLBACK rolls back the current transaction, canceling its changes. The SET AUTOCOMMIT statement disables or enables the default autocommit mode for the current connection.

Beginning with MySQL 5.0.3, the optional WORK keyword is supported for COMMIT and RELEASE, as are the CHAIN and RELEASE clauses. CHAIN and RELEASE can be used for additional control over transaction completion. The value of the completion_type system variable determines the default completion behavior. See Section 5.2.2, “Server System Variables”.

The AND CHAIN clause causes a new transaction to begin as soon as the current one ends, and the new transaction has the same isolation level as the just-terminated transaction. The RELEASE clause causes the server to disconnect the current client connection after terminating the current transaction. Including the NO keyword suppresses CHAIN or RELEASE completion, which can be useful if the completion_type system variable is set to cause chaining or release completion by default.

By default, MySQL runs with autocommit mode enabled. This means that as soon as you execute a statement that updates (modifies) a table, MySQL stores the update on disk.

If you are using a transaction-safe storage engine (such as InnoDB, BDB, or NDB Cluster), you can disable autocommit mode with the following statement:

SET AUTOCOMMIT=0;

After disabling autocommit mode by setting the AUTOCOMMIT variable to zero, you must use COMMIT to store your changes to disk or ROLLBACK if you want to ignore the changes you have made since the beginning of your transaction.

To disable autocommit mode for a single series of statements, use the START TRANSACTION statement:

START TRANSACTION;
SELECT @A:=SUM(salary) FROM table1 WHERE type=1;
UPDATE table2 SET summary=@A WHERE type=1;
COMMIT;

With START TRANSACTION, autocommit remains disabled until you end the transaction with COMMIT or ROLLBACK. The autocommit mode then reverts to its previous state.

BEGIN and BEGIN WORK are supported as aliases of START TRANSACTION for initiating a transaction. START TRANSACTION is standard SQL syntax and is the recommended way to start an ad-hoc transaction.

The BEGIN statement differs from the use of the BEGIN keyword that starts a BEGIN ... END compound statement. The latter does not begin a transaction. See Section 17.2.5, “BEGIN ... END Compound Statement Syntax”.

You can also begin a transaction like this:

START TRANSACTION WITH CONSISTENT SNAPSHOT;

The WITH CONSISTENT SNAPSHOT clause starts a consistent read for storage engines that are capable of it. Currently, this applies only to InnoDB. The effect is the same as issuing a START TRANSACTION followed by a SELECT from any InnoDB table. See Section 14.2.10.4, “Consistent Non-Locking Read”.

The WITH CONSISTENT SNAPSHOT clause does not change the current transaction isolation level, so it provides a consistent snapshot only if the current isolation level is one that allows consistent read (REPEATABLE READ or SERIALIZABLE).

Beginning a transaction causes an implicit UNLOCK TABLES to be performed.

For best results, transactions should be performed using only tables managed by a single transactional storage engine. Otherwise, the following problems can occur:

Each transaction is stored in the binary log in one chunk, upon COMMIT. Transactions that are rolled back are not logged. (Exception: Modifications to non-transactional tables cannot be rolled back. If a transaction that is rolled back includes modifications to non-transactional tables, the entire transaction is logged with a ROLLBACK statement at the end to ensure that the modifications to those tables are replicated.) See Section 5.12.3, “The Binary Log”.

You can change the isolation level for transactions with SET TRANSACTION ISOLATION LEVEL. See Section 13.4.6, “SET TRANSACTION Syntax”.

Rolling back can be a slow operation that may occur without the user having explicitly asked for it (for example, when an error occurs). Because of this, SHOW PROCESSLIST displays Rolling back in the State column for the connection during implicit and explicit (ROLLBACK SQL statement) rollbacks.

Some statements cannot be rolled back. In general, these include data definition language (DDL) statements, such as those that create or drop databases, those that create, drop, or alter tables or stored routines.

You should design your transactions not to include such statements. If you issue a statement early in a transaction that cannot be rolled back, and then another statement later fails, the full effect of the transaction cannot be rolled back in such cases by issuing a ROLLBACK statement.

Each of the following statements (and any synonyms for them) implicitly end a transaction, as if you had done a COMMIT before executing the statement:

Transactions cannot be nested. This is a consequence of the implicit COMMIT performed for any current transaction when you issue a START TRANSACTION statement or one of its synonyms.

Statements that cause implicit cannot be used in an XA transaction while the transaction is in an ACTIVE state.

SAVEPOINT identifier
ROLLBACK [WORK] TO SAVEPOINT identifier
RELEASE SAVEPOINT identifier

InnoDB supports the SQL statements SAVEPOINT and ROLLBACK TO SAVEPOINT. Starting from MySQL 5.0.3, RELEASE SAVEPOINT and the optional WORK keyword for ROLLBACK are supported as well.

The SAVEPOINT statement sets a named transaction savepoint with a name of identifier. If the current transaction has a savepoint with the same name, the old savepoint is deleted and a new one is set.

The ROLLBACK TO SAVEPOINT statement rolls back a transaction to the named savepoint. Modifications that the current transaction made to rows after the savepoint was set are undone in the rollback, but InnoDB does not release the row locks that were stored in memory after the savepoint. (Note that for a new inserted row, the lock information is carried by the transaction ID stored in the row; the lock is not separately stored in memory. In this case, the row lock is released in the undo.) Savepoints that were set at a later time than the named savepoint are deleted.

If the ROLLBACK TO SAVEPOINT statement returns the following error, it means that no savepoint with the specified name exists:

ERROR 1181: Got error 153 during ROLLBACK

The RELEASE SAVEPOINT statement removes the named savepoint from the set of savepoints of the current transaction. No commit or rollback occurs. It is an error if the savepoint does not exist.

All savepoints of the current transaction are deleted if you execute a COMMIT, or a ROLLBACK that does not name a savepoint.

Beginning with MySQL 5.0.17, a new savepoint level is created when a stored function is invoked or a trigger is activated. The savepoints on previous levels become unavailable and thus do not conflict with savepoints on the new level. When the function or trigger terminates, any savepoints it created are released and the previous savepoint level is restored.

LOCK TABLES
    tbl_name [AS alias] {READ [LOCAL] | [LOW_PRIORITY] WRITE}
    [, tbl_name [AS alias] {READ [LOCAL] | [LOW_PRIORITY] WRITE}] ...
UNLOCK TABLES

LOCK TABLES locks base tables (but not views) for the current thread. If any of the tables are locked by other threads, it blocks until all locks can be acquired. UNLOCK TABLES releases any locks held by the current thread. All tables that are locked by the current thread are implicitly unlocked when the thread issues another LOCK TABLES, or when the connection to the server is closed.

A table lock protects only against inappropriate reads or writes by other clients. The client holding the lock, even a read lock, can perform table-level operations such as DROP TABLE.

Note the following regarding the use of LOCK TABLES with transactional tables:

To use LOCK TABLES, you must have the LOCK TABLES privilege and the SELECT privilege for the involved tables.

The main reasons to use LOCK TABLES are to emulate transactions or to get more speed when updating tables. This is explained in more detail later.

If a thread obtains a READ lock on a table, that thread (and all other threads) can only read from the table. If a thread obtains a WRITE lock on a table, only the thread holding the lock can write to the table. Other threads are blocked from reading or writing the table until the lock has been released.

The difference between READ LOCAL and READ is that READ LOCAL allows non-conflicting INSERT statements (concurrent inserts) to execute while the lock is held. However, this cannot be used if you are going to manipulate the database files outside MySQL while you hold the lock. For InnoDB tables, READ LOCAL is the same as READ as of MySQL 5.0.13. (Before that, READ LOCAL essentially does nothing: It does not lock the table at all, so for InnoDB tables, the use of READ LOCAL is deprecated because a plain consistent-read SELECT does the same thing, and no locks are needed.)

When you use LOCK TABLES, you must lock all tables that you are going to use in your queries. Because LOCK TABLES will not lock views, if the operation that you are performing uses any views, you must also lock all of the base tables on which those views depend. While the locks obtained with a LOCK TABLES statement are in effect, you cannot access any tables that were not locked by the statement. Also, you cannot use a locked table multiple times in a single query. Use aliases instead, in which case you must obtain a lock for each alias separately.

mysql> LOCK TABLE t WRITE, t AS t1 WRITE;
mysql> INSERT INTO t SELECT * FROM t;
ERROR 1100: Table 't' was not locked with LOCK TABLES
mysql> INSERT INTO t SELECT * FROM t AS t1;

If your queries refer to a table by means of an alias, you must lock the table using that same alias. It does not work to lock the table without specifying the alias:

mysql> LOCK TABLE t READ;
mysql> SELECT * FROM t AS myalias;
ERROR 1100: Table 'myalias' was not locked with LOCK TABLES

Conversely, if you lock a table using an alias, you must refer to it in your queries using that alias:

mysql> LOCK TABLE t AS myalias READ;
mysql> SELECT * FROM t;
ERROR 1100: Table 't' was not locked with LOCK TABLES
mysql> SELECT * FROM t AS myalias;

WRITE locks normally have higher priority than READ locks to ensure that updates are processed as soon as possible. This means that if one thread obtains a READ lock and then another thread requests a WRITE lock, subsequent READ lock requests wait until the WRITE thread has gotten the lock and released it. You can use LOW_PRIORITY WRITE locks to allow other threads to obtain READ locks while the thread is waiting for the WRITE lock. You should use LOW_PRIORITY WRITE locks only if you are sure that eventually there will be a time when no threads have a READ lock.

LOCK TABLES works as follows:

This policy ensures that table locking is deadlock free. There are, however, other things you need to be aware of about this policy: If you are using a LOW_PRIORITY WRITE lock for a table, it means only that MySQL waits for this particular lock until there are no threads that want a READ lock. When the thread has gotten the WRITE lock and is waiting to get the lock for the next table in the lock table list, all other threads wait for the WRITE lock to be released. If this becomes a serious problem with your application, you should consider converting some of your tables to transaction-safe tables.

You can safely use KILL to terminate a thread that is waiting for a table lock. See Section 13.5.5.3, “KILL Syntax”.

Note that you should not lock any tables that you are using with INSERT DELAYED because in that case the INSERT is performed by a separate thread.

Normally, you do not need to lock tables, because all single UPDATE statements are atomic; no other thread can interfere with any other currently executing SQL statement. However, there are a few cases when locking tables may provide an advantage:

You can avoid using LOCK TABLES in many cases by using relative updates (UPDATE customer SET value=value+new_value) or the LAST_INSERT_ID() function. See Section 1.9.5.3, “Transactions and Atomic Operations”.

You can also avoid locking tables in some cases by using the user-level advisory lock functions GET_LOCK() and RELEASE_LOCK(). These locks are saved in a hash table in the server and implemented with pthread_mutex_lock() and pthread_mutex_unlock() for high speed. See Section 12.9.4, “Miscellaneous Functions”.

See Section 7.3.1, “Locking Methods”, for more information on locking policy.

You can lock all tables in all databases with read locks with the FLUSH TABLES WITH READ LOCK statement. See Section 13.5.5.2, “FLUSH Syntax”. This is a very convenient way to get backups if you have a filesystem such as Veritas that can take snapshots in time.

Note: If you use ALTER TABLE on a locked table, it may become unlocked. See Section A.7.1, “Problems with ALTER TABLE”.

SET [GLOBAL | SESSION] TRANSACTION ISOLATION LEVEL
{ READ UNCOMMITTED | READ COMMITTED | REPEATABLE READ | SERIALIZABLE }

XA {START|BEGIN} xid [JOIN|RESUME]

XA END xid [SUSPEND [FOR MIGRATE]]

XA PREPARE xid

XA COMMIT xid [ONE PHASE]

XA ROLLBACK xid

XA RECOVER

xid: gtrid [, bqual [, formatID ]]

mysql> XA RECOVER;
+----------+--------------+--------------+--------+
| formatID | gtrid_length | bqual_length | data   |
+----------+--------------+--------------+--------+
|        7 |            3 |            3 | abcdef |
+----------+--------------+--------------+--------+

mysql> XA START 'xatest';
Query OK, 0 rows affected (0.00 sec)

mysql> INSERT INTO mytable (i) VALUES(10);
Query OK, 1 row affected (0.04 sec)

mysql> XA END 'xatest';
Query OK, 0 rows affected (0.00 sec)

mysql> XA PREPARE 'xatest';
Query OK, 0 rows affected (0.00 sec)

mysql> XA COMMIT 'xatest';
Query OK, 0 rows affected (0.00 sec)

ERROR 1399 (XAE07): XAER_RMFAIL: The command cannot be executed
when global transaction is in the ACTIVE state