MySQL Transactions, Part III - BDB Tables, Table locking and Savepoints - Page 2

September 21, 2004

Table locking for all table types

InnoDB and BDB tables both make use of row level locking, whereby only the row/s being manipulated are locked. This means that other rows can still be manipulated, reducing the risk of contention, but is also a less optimal process than table level locking if most of the queries are SELECT's. MyISAM tables do not have row-locking, which is why they are not a great choice if the application has a high ration of INSERT's or UPDATE's to SELECT's. However, all table types can make use of table-locking, using the LOCK TABLE statement.

Since this is the only kind of locking available to MyISAM tables, let's create a MyISAM table to test on:

mysql> CREATE TABLE myisam_table(f INT);
Query OK, 0 rows affected (0.01 sec)

There are two main kinds of locks - read locks and write locks. A read lock only allows reads on the table, not writes, while a write lock only allows that connection to read and write - all other connections are blocked. Let's see these in action:

Connection 1:

mysql> LOCK TABLE myisam_table READ;
Query OK, 0 rows affected (0.00 sec)

Connection 2:

mysql> INSERT INTO myisam_table VALUES(1);

The second connection hangs, waiting for the table to be unlocked.

Connection 1:

mysql> UNLOCK TABLES;
Query OK, 0 rows affected (0.00 sec)

Connection 2:

mysql> INSERT INTO myisam_table VALUES(1);
Query OK, 1 row affected (1 min 16.34 sec)

However, a read from the second connection is processed immediately:

Connection 1:

mysql> LOCK TABLE myisam_table READ;
Query OK, 0 rows affected (0.00 sec)

Connection 2:

mysql> SELECT * FROM myisam_table;
+------+
| f    |
+------+
|    1 |
+------+
1 row in set (0.00 sec)

Connection 1:

mysql> UNLOCK TABLES;
Query OK, 0 rows affected (0.00 sec)

Let's try the same statements with a write lock.

Connection 1:

mysql> LOCK TABLE myisam_table WRITE;
Query OK, 0 rows affected (0.01 sec)

Connection 2:

mysql> SELECT * FROM myisam_table;

The connection hangs, waiting for the lock to be released.

Connection 1:

mysql> UNLOCK TABLE;
Query OK, 0 rows affected (0.01 sec)

Connection 2:

mysql> SELECT * FROM myisam_table;
+------+
| f    |
+------+
|    1 |
+------+
1 row in set (6.22 sec)

If you have been thinking about this a bit, you may wonder what happens if the connection that creates a READ lock attempts to INSERT a record. Since this is forbidden on a read lock even to the originating connection, and if the connection had to wait, there would be deadlock since it would then be unable to release the lock, MySQL returns an immediate error.

mysql> LOCK TABLE myisam_table READ;
Query OK, 0 rows affected (0.00 sec)

mysql> INSERT INTO myisam_table VALUES(2);
ERROR 1099: Table 'myisam_table' was locked with a READ lock and can't be updated

mysql> UNLOCK TABLE;
Query OK, 0 rows affected (0.00 sec)

Locking priority

Write locks have a higher priority than read locks. A write lock will always be obtained ahead of any prior waiting read locks. Let's demonstrate this:

Connection 1:

mysql> LOCK TABLE myisam_table WRITE;
Query OK, 0 rows affected (0.00 sec)

Connection 2:

mysql> LOCK TABLE myisam_table READ;

The connection hangs, waiting for the earlier lock to be released.

Connection 3:

mysql> LOCK TABLE myisam_table WRITE;

Now there are two locks waiting, a read lock, and a write lock requested after the read lock. When we release the first lock, the lock from the third connection, since it is a higher precedence write lock, is obtained.

Connection 1:

mysql> UNLOCK TABLE;
Query OK, 0 rows affected (0.00 sec)

Connection 2 is still waiting, but connection 3 is ready for action again:

mysql> LOCK TABLE myisam_table WRITE;
Query OK, 0 rows affected (5.52 sec)

mysql> UNLOCK TABLE;
Query OK, 0 rows affected (0.00 sec) 

Only now is the second connection's lock obtained. It does not take much imagination to understand why table locking can lead to performance implications if there are many write locks.

Sometimes you genuinely want a write lock to have a lower priority than a read lock. An example we use at Independent Online is the article tracker. Every article read is tracked, but the INSERT statement is a lower priority than the read, which of course impacts a reader. Using the same MyISAM table and the same set of locks as before except that the third connection is a low priority lock, let's examine what happens.

Connection 1:

mysql> LOCK TABLE myisam_table WRITE;
Query OK, 0 rows affected (0.00 sec)

Connection 2:

mysql> LOCK TABLE myisam_table READ;

The connection hangs, waiting for the earlier lock to be released.

Connection 3:

mysql> LOCK TABLE myisam_table LOW_PRIORITY WRITE;

Connection 1:

mysql> UNLOCK TABLE;
Query OK, 0 rows affected (0.00 sec)

This time, Connection 2 is obtained, while the write lock in connection 3 is still waiting.

Connection 2:

mysql> LOCK TABLE myisam_table READ;
Query OK, 0 rows affected (3 min 8.29 sec)

mysql> UNLOCK TABLE;
Query OK, 0 rows affected (0.00 sec) 

Connection 3:

mysql> LOCK TABLE myisam_table LOW_PRIORITY WRITE;
Query OK, 0 rows affected (2 min 12.23 sec)

mysql> UNLOCK TABLE;
Query OK, 0 rows affected (0.00 sec)

Savepoints

Finally, for this month, we will go back to InnoDB tables, and examine a recent addition to MySQL - savepoints. These allow you to rollback part of a transaction, rather than the all or nothing approach usually found with transactions. Savepoints only work with InnoDB tables - for those who have not been following the series, you can recreate the table as it is with the following statements (though the records are not really necessary):

mysql> CREATE TABLE t (f INT) TYPE = InnoDB;
Query OK, 0 rows affected (0.03 sec)

mysql> INSERT INTO t values (1),(2),(3),(4),(55),(6),(7),(88);
Query OK, 8 rows affected (0.00 sec)
Records: 8  Duplicates: 0  Warnings: 0

Inside the same transaction, we will insert two new records, one before the savepoint, and one after:

mysql> INSERT INTO t VALUES(9);
Query OK, 1 row affected (0.05 sec)

mysql> SAVEPOINT x;
Query OK, 0 rows affected (0.06 sec)

mysql> INSERT INTO t VALUES(10);
Query OK, 1 row affected (0.00 sec)

mysql> ROLLBACK TO SAVEPOINT x;
Query OK, 0 rows affected (0.06 sec)

mysql> SELECT * FROM t;
+------+
| f    |
+------+
|    1 |
|    2 |
|    3 |
|    4 |
|   55 |
|    6 |
|    7 |
|   88 |
|    9 |
+------+
9 rows in set (0.00 sec)

The first insert has been performed - effectively a savepoint could also be termed COMMIT UNTIL, so anything before the savepoint is now committed. The second insert, after the savepoint, is rolled back.

That concludes the series on transactions. I hope that you have gotten a taste for how MySQL handles them, and some of its quirks, but as always, there is no better way to learn than diving in and making all the mistakes yourself. Good luck.

» See All Articles by Columnist Ian Gilfillan








The Network for Technology Professionals

Search:

About Internet.com

Legal Notices, Licensing, Permissions, Privacy Policy.
Advertise | Newsletters | E-mail Offers