If InnoDB prints an operating system error in a file operation, usually the problem is one of the following:

If something goes wrong when InnoDB attempts to initialize its tablespace or its log files, you should delete all files created by InnoDB. This means all ibdata files and all ib_logfiles. In case you created some InnoDB tables, delete the corresponding .frm files for these tables (and any .ibd files if you are using multiple tablespaces) from the MySQL database directories as well. Then you can try the InnoDB database creation again. It is best to start the MySQL server from a command prompt so that you see what is happening.

By default, each client that connects to the MySQL server begins with autocommit mode enabled, which automatically commits every SQL statement you run. To use multiple-statement transactions, you can switch autocommit off with the SQL statement SET AUTOCOMMIT = 0 and use COMMIT and ROLLBACK to commit or roll back your transaction. If you want to leave autocommit on, you can enclose your transactions between START TRANSACTION and COMMIT or ROLLBACK. The following example shows two transactions. The first is committed; the second is rolled back.

shell> mysql test
Welcome to the MySQL monitor.  Commands end with ; or \g.
Your MySQL connection id is 5 to server version: 3.23.50-log
Type 'help;' or '\h' for help. Type '\c' to clear the buffer.
mysql> CREATE TABLE CUSTOMER (A INT, B CHAR (20), INDEX (A))
    -> ENGINE=InnoDB;
Query OK, 0 rows affected (0.00 sec)
mysql> BEGIN;
Query OK, 0 rows affected (0.00 sec)
mysql> INSERT INTO CUSTOMER VALUES (10, 'Heikki');
Query OK, 1 row affected (0.00 sec)
mysql> COMMIT;
Query OK, 0 rows affected (0.00 sec)
mysql> SET AUTOCOMMIT=0;
Query OK, 0 rows affected (0.00 sec)
mysql> INSERT INTO CUSTOMER VALUES (15, 'John');
Query OK, 1 row affected (0.00 sec)
mysql> ROLLBACK;
Query OK, 0 rows affected (0.00 sec)
mysql> SELECT * FROM CUSTOMER;
+------+--------+
| A    | B      |
+------+--------+
|   10 | Heikki |
+------+--------+
1 row in set (0.00 sec)
mysql>

In APIs like PHP, Perl DBI/DBD, JDBC, ODBC, or the standard C call interface of MySQL, you can send transaction control statements such as COMMIT to the MySQL server as strings just like any other SQL statements such as SELECT or INSERT. Some APIs also offer separate special transaction commit and rollback functions or methods.

Important: You should not convert MySQL system tables in the mysql database (such as user or host) to the InnoDB type. The system tables must always be of the MyISAM type.

If you want all your (non-system) tables to be created as InnoDB tables, you can simply add the line default-table-type=innodb to the [mysqld] section of your my.cnf or my.ini file.

InnoDB does not have a special optimization for separate index creation the way the MyISAM storage engine does. Therefore, it does not pay to export and import the table and create indexes afterward. The fastest way to alter a table to InnoDB is to do the inserts directly to an InnoDB table. That is, use ALTER TABLE ... ENGINE=INNODB, or create an empty InnoDB table with identical definitions and insert the rows with INSERT INTO ... SELECT * FROM ....

If you have UNIQUE constraints on secondary keys, you can speed up a table import by turning off the uniqueness checks temporarily during the import session: SET UNIQUE_CHECKS=0;. For big tables, this saves a lot of disk I/O because InnoDB can then use its insert buffer to write secondary index records as a batch.

To get better control over the insertion process, it might be good to insert big tables in pieces:

INSERT INTO newtable SELECT * FROM oldtable
   WHERE yourkey > something AND yourkey <= somethingelse;

After all records have been inserted, you can rename the tables.

During the conversion of big tables, you should increase the size of the InnoDB buffer pool to reduce disk I/O. Do not use more than 80% of the physical memory, though. You can also increase the sizes of the InnoDB log files and the log files.

Make sure that you do not fill up the tablespace: InnoDB tables require a lot more disk space than MyISAM tables. If an ALTER TABLE runs out of space, it starts a rollback, and that can take hours if it is disk-bound. For inserts, InnoDB uses the insert buffer to merge secondary index records to indexes in batches. That saves a lot of disk I/O. In rollback, no such mechanism is used, and the rollback can take 30 times longer than the insertion.

In the case of a runaway rollback, if you do not have valuable data in your database, it may be advisable to kill the database process rather than wait for millions of disk I/O operations to complete. For the complete procedure, see Section 15.8.1, “Forcing Recovery”.

If you specify an AUTO_INCREMENT column for a table, the InnoDB table handle in the data dictionary contains a special counter called the auto-increment counter that is used in assigning new values for the column. The auto-increment counter is stored only in main memory, not on disk.

InnoDB uses the following algorithm to initialize the auto-increment counter for a table T that contains an AUTO_INCREMENT column named ai_col: After a server startup, when a user first does an insert to a table T, InnoDB executes the equivalent of this statement:

SELECT MAX(ai_col) FROM T FOR UPDATE;

The value retrieved by the statement is incremented by one and assigned to the column and the auto-increment counter of the table. If the table is empty, the value 1 is assigned. If the auto-increment counter is not initialized and the user invokes a SHOW TABLE STATUS statement that displays output for the table T, the counter is initialized (but not incremented) and stored for use by later inserts. Note that in this initialization we do a normal exclusive-locking read on the table and the lock lasts to the end of the transaction.

InnoDB follows the same procedure for initializing the auto-increment counter for a freshly created table.

Note that if the user specifies NULL or 0 for the AUTO_INCREMENT column in an INSERT, InnoDB treats the row as if the value had not been specified and generates a new value for it.

After the auto-increment counter has been initialized, if a user inserts a row that explicitly specifies the column value, and the value is bigger than the current counter value, the counter is set to the specified column value. If the user does not explicitly specify a value, InnoDB increments the counter by one and assigns the new value to the column.

When accessing the auto-increment counter, InnoDB uses a special table level AUTO-INC lock that it keeps to the end of the current SQL statement, not to the end of the transaction. The special lock release strategy was introduced to improve concurrency for inserts into a table containing an AUTO_INCREMENT column. Two transactions cannot have the AUTO-INC lock on the same table simultaneously.

Note that you may see gaps in the sequence of values assigned to the AUTO_INCREMENT column if you roll back transactions that have gotten numbers from the counter.

The behavior of the auto-increment mechanism is not defined if a user assigns a negative value to the column or if the value becomes bigger than the maximum integer that can be stored in the specified integer type.

Beginning with MySQL 5.0.3, InnoDB supports the AUTO_INCREMENT = n table option in CREATE TABLE and ALTER TABLE statements, to set the initial counter value or alter the current counter value. The effect of this option is canceled by a server restart, for reasons discussed earlier in this section.

InnoDB also supports foreign key constraints. The syntax for a foreign key constraint definition in InnoDB looks like this:

[CONSTRAINT symbol] FOREIGN KEY [id] (index_col_name, ...)
    REFERENCES tbl_name (index_col_name, ...)
    [ON DELETE {RESTRICT | CASCADE | SET NULL | NO ACTION}]
    [ON UPDATE {RESTRICT | CASCADE | SET NULL | NO ACTION}]

Foreign keys definitions are subject to the following conditions:

InnoDB rejects any INSERT or UPDATE operation that attempts to create a foreign key value in a child table without a matching candidate key value in the parent table. The action InnoDB takes for any UPDATE or DELETE operation that attempts to update or delete a candidate key value in the parent table that has some matching rows in the child table is dependent on the referential action specified using ON UPDATE and ON DETETE subclauses of the FOREIGN KEY clause. When the user attempts to delete or update a row from a parent table, and there are one or more matching rows in the child table, InnoDB supports five options regarding the action to be taken:

InnoDB supports the same options when the candidate key in the parent table is updated. With CASCADE, the foreign key column(s) in the child table are set to new value(s) of the candidate key in the parent table. In the same way, the updates cascade if updated column(s) in the child table reference foreign keys in another table.

Note that InnoDB supports foreign key references within a table and in these cases child table really means dependent records within the table.

InnoDB needs indexes on foreign keys and referenced keys so that foreign key checks can be fast and not require a table scan. In MySQL 5.0, the index on the foreign key is created automatically. This is in contrast to some older versions, in which indexes had to be created explicitly or the creation of foreign key constraints would fail.

Corresponding columns in the foreign key and the referenced key must have similar internal data types inside InnoDB so that they can be compared without a type conversion. The size and sign of integer types must be the same. The length of string types need not be the same. If you specify a SET NULL action, make sure that you have not declared the columns in the child table as NOT NULL.

If MySQL reports an error number 1005 from a CREATE TABLE statement, and the error message string refers to errno 150, this means that the table creation failed because a foreign key constraint was not correctly formed. Similarly, if an ALTER TABLE fails and it refers to errno 150, that means a foreign key definition would be incorrectly formed for the altered table. In MySQL 5.0, you can use SHOW INNODB STATUS to display a detailed explanation of the most recent InnoDB foreign key error in the server.

Note: InnoDB does not check foreign key constraints on those foreign key or referenced key values that contain a NULL column.

Deviation from SQL standards: If in the parent table there are several rows that have the same referenced key value, then InnoDB acts in foreign key checks as if the other parent rows with the same key value do not exist. For example, if you have defined a RESTRICT type constraint, and there is a child row with several parent rows, InnoDB does not allow the deletion of any of those parent rows.

InnoDB performs cascading operations through a depth-first algorithm, based on records in the indexes corresponding to the foreign key constraints.

Deviation from SQL standards: If ON UPDATE CASCADE or ON UPDATE SET NULL recurses to update the same table it has previously updated during the cascade, it acts like RESTRICT. This means that you cannot use self-referential ON UPDATE CASCADE or ON UPDATE SET NULL operations. This is to prevent infinite loops resulting from cascaded updates. A self-referential ON DELETE SET NULL, on the other hand, is possible, as is a self-referential ON DELETE CASCADE. In MySQL 5.0, cascading operations may not be nested more than 15 levels deep.

Deviation from SQL standards: Like MySQL in general, in an SQL statement that inserts, deletes, or updates many rows, InnoDB checks UNIQUE and FOREIGN KEY constraints row-by-row. According to the SQL standard, the default behavior should be deferred checking, that is, constraints are only checked after the entire SQL statement has been processed. Until InnoDB implements deferred constraint checking, some things will be impossible, such as deleting a record that refers to itself via a foreign key.

Note: Currently, triggers are not activated by cascaded foreign key actions.

A simple example that relates parent and child tables through a single-column foreign key:

CREATE TABLE parent(id INT NOT NULL,
                    PRIMARY KEY (id)
) TYPE=INNODB;
CREATE TABLE child(id INT, parent_id INT,
                   INDEX par_ind (parent_id),
                   FOREIGN KEY (parent_id) REFERENCES parent(id)
                     ON DELETE CASCADE
) TYPE=INNODB;

A more complex example in which a product_order table has foreign keys for two other tables. One foreign key references a two-column index in the product table. The other references a single-column index in the customer table:

CREATE TABLE product (category INT NOT NULL, id INT NOT NULL,
                      price DECIMAL,
                      PRIMARY KEY(category, id)) TYPE=INNODB;
CREATE TABLE customer (id INT NOT NULL,
                      PRIMARY KEY (id)) TYPE=INNODB;
CREATE TABLE product_order (no INT NOT NULL AUTO_INCREMENT,
                      product_category INT NOT NULL,
                      product_id INT NOT NULL,
                      customer_id INT NOT NULL,
                      PRIMARY KEY(no),
                      INDEX (product_category, product_id),
                      FOREIGN KEY (product_category, product_id)
                        REFERENCES product(category, id)
                        ON UPDATE CASCADE ON DELETE RESTRICT,
                      INDEX (customer_id),
                      FOREIGN KEY (customer_id)
                        REFERENCES customer(id)) TYPE=INNODB;

InnoDB allows you to add a new foreign key constraint to a table by using ALTER TABLE:

ALTER TABLE yourtablename
    ADD [CONSTRAINT symbol] FOREIGN KEY [id] (index_col_name, ...)
    REFERENCES tbl_name (index_col_name, ...)
    [ON DELETE {RESTRICT | CASCADE | SET NULL | NO ACTION}]
    [ON UPDATE {RESTRICT | CASCADE | SET NULL | NO ACTION}]

Remember to create the required indexes first. You can also add a self-referential foreign key constraint to a table using ALTER TABLE.

InnoDB also supports the use of ALTER TABLE to drop foreign keys:

ALTER TABLE yourtablename DROP FOREIGN KEY fk_symbol;

If the FOREIGN KEY clause included a CONSTRAINT name when you created the foreign key, you can refer to that name to drop the foreign key. Otherwise, the fk_symbol value is internally generated by InnoDB when the foreign key is created. To find out the symbol when you want to drop a foreign key, use the SHOW CREATE TABLE statement. An example:

mysql> SHOW CREATE TABLE ibtest11c\G
*************************** 1. row ***************************
       Table: ibtest11c
Create Table: CREATE TABLE `ibtest11c` (
  `A` int(11) NOT NULL auto_increment,
  `D` int(11) NOT NULL default '0',
  `B` varchar(200) NOT NULL default '',
  `C` varchar(175) default NULL,
  PRIMARY KEY  (`A`,`D`,`B`),
  KEY `B` (`B`,`C`),
  KEY `C` (`C`),
  CONSTRAINT `0_38775` FOREIGN KEY (`A`, `D`)
REFERENCES `ibtest11a` (`A`, `D`)
ON DELETE CASCADE ON UPDATE CASCADE,
  CONSTRAINT `0_38776` FOREIGN KEY (`B`, `C`)
REFERENCES `ibtest11a` (`B`, `C`)
ON DELETE CASCADE ON UPDATE CASCADE
) ENGINE=INNODB CHARSET=latin1
1 row in set (0.01 sec)

mysql> ALTER TABLE ibtest11c DROP FOREIGN KEY 0_38775;

The InnoDB parser allows you to use backticks around table and column names in a FOREIGN KEY ... REFERENCES ... clause. The InnoDB parser also takes into account the setting of the lower_case_table_names system variable.

InnoDB returns a table's foreign key definitions as part of the output of the SHOW CREATE TABLE statement:

SHOW CREATE TABLE tbl_name;

From this version, mysqldump also produces correct definitions of tables to the dump file, and does not forget about the foreign keys.

You can display the foreign key constraints for a table like this:

SHOW TABLE STATUS FROM db_name LIKE 'tbl_name';

The foreign key constraints are listed in the Comment column of the output.

When performing foreign key checks, InnoDB sets shared row-level locks on child or parent records it has to look at. InnoDB checks foreign key constraints immediately; the check is not deferred to transaction commit.

To make it easier to reload dump files for tables that have foreign key relationships, mysqldump automatically includes a statement in the dump output to set FOREIGN_KEY_CHECKS to 0. This avoids problems with tables having to be reloaded in a particular order when the dump is reloaded. It is also possible to set this variable manually:

mysql> SET FOREIGN_KEY_CHECKS = 0;
mysql> SOURCE dump_file_name;
mysql> SET FOREIGN_KEY_CHECKS = 1;

This allows you to import the tables in any order if the dump file contains tables that are not correctly ordered for foreign keys. It also speeds up the import operation. Setting FOREIGN_KEY_CHECKS to 0 can also be useful for ignoring foreign key constraints during LOAD DATA and ALTER TABLE operations.

InnoDB does not allow you to drop a table that is referenced by a FOREIGN KEY constraint, unless you do SET FOREIGN_KEY_CHECKS=0. When you drop a table, the constraints that were defined in its create statement are also dropped.

If you re-create a table that was dropped, it must have a definition that conforms to the foreign key constraints referencing it. It must have the right column names and types, and it must have indexes on the referenced keys, as stated earlier. If these are not satisfied, MySQL returns error number 1005 and refers to errno 150 in the error message string.

MySQL replication works for InnoDB tables as it does for MyISAM tables. It is also possible to use replication in a way where the table type on the slave is not the same as the original table type on the master. For example, you can replicate modifications to an InnoDB table on the master to a MyISAM table on the slave.

To set up a new slave for a master, you have to make a copy of the InnoDB tablespace and the log files, as well as the .frm files of the InnoDB tables, and move the copies to the slave. For the proper procedure to do this, see Section 15.9, “Moving an InnoDB Database to Another Machine”.

If you can shut down the master or an existing slave, you can take a cold backup of the InnoDB tablespace and log files and use that to set up a slave. To make a new slave without taking down any server you can also use the non-free (commercial) InnoDB Hot Backup tool.

One minor limitation in InnoDB replication is that LOAD TABLE FROM MASTER does not work for InnoDB type tables. There are two possible workarounds:

Transactions that fail on the master do not affect replication at all. MySQL replication is based on the binary log where MySQL writes SQL statements that modify data. A slave reads the binary log of the master and executes the same SQL statements. However, statements that occur within a transaction are not written to the binary log until the transaction commits, at which point all statements in the transaction are written at once. If a statement fails, for example, because of a foreign key violation, or if a transaction is rolled back, no SQL statements are written to the binary log, and the transaction is not executed on the slave at all.

In MySQL 5.0, you can store each InnoDB table and its indexes in its own file. This feature is called “multiple tablespaces” because in effect each table has its own tablespace.

Using multiple tablespaces can be beneficial to users who want to move specific tables to separate physical disks or who wish to restore backups of single tables quickly without interrupting the use of the remaining InnoDB tables.

You can enable multiple tablespaces by adding this line to the [mysqld] section of my.cnf:

[mysqld]
innodb_file_per_table

After restarting the server, InnoDB stores each newly created table into its own file tbl_name.ibd in the database directory where the table belongs. This is similar to what the MyISAM storage engine does, but MyISAM divides the table into a data file tbl_name.MYD and the index file tbl_name.MYI. For InnoDB, the data and the indexes are stored together in the .ibd file. The tbl_name.frm file is still created as usual.

If you remove the innodb_file_per_table line from my.cnf and restart the server, InnoDB creates tables inside the shared tablespace files again.

innodb_file_per_table affects only table creation. If you start the server with this option, new tables are created using .ibd files, but you can still access tables that exist in the shared tablespace. If you remove the option, new tables are created in the shared tablespace, but you can still access any tables that were created using multiple tablespaces.

InnoDB always needs the shared tablespace. The .ibd files are not sufficient for InnoDB to operate. The shared tablespace consists of the familiar ibdata files where InnoDB puts its internal data dictionary and undo logs.

Note: You cannot freely move .ibd files between database directories as you can with MyISAM table files. This is because the table definition is stored in the InnoDB shared tablespace, and because InnoDB must preserve the consistency of transaction IDs and log sequence numbers.

Within a given MySQL installation, you can move an .ibd file and the associated table from one database to another with a RENAME TABLE statement:

RENAME TABLE old_db_name.tbl_name TO new_db_name.tbl_name;

If you have a “clean” backup of an .ibd file, you can restore it to the MySQL installation from which it originated as follows:

In this context, a “clean” .ibd file backup means:

You can make a clean backup .ibd file using the following method:

Another method for making a clean copy of an .ibd file is to use the commercial InnoDB Hot Backup tool:

If there is database page corruption, you may want to dump your tables from the database with SELECT INTO OUTFILE; usually, most of the data obtained in this way is intact. Even so, the corruption may cause SELECT * FROM tbl_name or InnoDB background operations to crash or assert, or even make InnoDB roll-forward recovery crash. However, you can use to force the InnoDB storage engine to start up while preventing background operations from running, so that you are able to dump your tables. For example, you can add the following line to the [mysqld] section of your option file before restarting the server:

[mysqld]
innodb_force_recovery = 4

The allowable non-zero values for innodb_force_recovery follow. A larger number includes all precautions of lower numbers. If you are able to dump your tables with an option value of at most 4, then you are relatively safe that only some data on corrupt individual pages is lost. A value of 6 is more dramatic, because database pages are left in an obsolete state, which in turn may introduce more corruption into B-trees and other database structures.

The database must not otherwise be used with any of these options enabled. As a safety measure, InnoDB prevents users from performing INSERT, UPDATE, or DELETE operations when innodb_force_recovery is set to a value greater than 0.

You may DROP or CREATE tables even if forced recovery is used. If you know that a given table is causing a crash on rollback, you can drop it. You can also use this to stop a runaway rollback caused by a failing mass import or ALTER TABLE. You can kill the mysqld process and set innodb_force_recovery to 3 to bring the database up without the rollback, then DROP the table that is causing the runaway rollback.

InnoDB implements a checkpoint mechanism known as “fuzzy” checkpointing. InnoDB flushes modified database pages from the buffer pool in small batches. There is no need to flush the buffer pool in one single batch, which would in practice stop processing of user SQL statements during the checkpointing process.

During crash recovery, InnoDB looks for a checkpoint label written to the log files. It knows that all modifications to the database before the label are present in the disk image of the database. Then InnoDB scans the log files forward from the checkpoint, applying the logged modifications to the database.

InnoDB writes to the log files on a rotating basis. All committed modifications that make the database pages in the buffer pool different from the images on disk must be available in the log files in case InnoDB has to do a recovery. This means that when InnoDB starts to reuse a log file, it has to make sure that the database page images on disk contain the modifications logged in the log file InnoDB is going to reuse. In other words, InnoDB must create a checkpoint and this often involves flushing of modified database pages to disk.

The preceding description explains why making your log files very large may save disk I/O in checkpointing. It often makes sense to set the total size of the log files as big as the buffer pool or even bigger. The drawback of big log files is that crash recovery can take longer because there is more logged information to apply to the database.

InnoDB implements standard row-level locking where there are two types of locks:

If a transaction A holds an exclusive (X) lock on tuple t, then a request from some distinct transaction B for a lock of either type on t cannot be granted immediately. Instead, transaction B has to wait for transaction A to release its lock on tuple t.

If transaction A holds a shared (S) lock on tuple t, then

Additionally, InnoDB supports multiple granularity locking which allows coexistence of record locks and locks on entire tables. To make locking at multiple granularity levels practical, additional types of locks, called intention locks are used. Intention locks are table locks in InnoDB. The idea behind intention locks is for a transaction to indicate which type of lock (shared or exclusive) it will require later for a row in that table. There are two types of intention locks used in InnoDB (assume that transaction T has requested a lock of the indicated type on table R):

The intention locking protocol is as follows:

These rules can be conveniently summarized by means of a lock type compatibility matrix:

A lock is granted to a requesting transaction if it is compatible with existing locks. A lock is not granted to a requesting transaction if it conflicts with existing locks. A transaction waits until the conflicting existing lock is released. If a lock request conflicts with an existing lock and cannot be granted because it would cause deadlock, an error occurs.

Thus, intention locks do not block anything except full table requests (for example, LOCK TABLES ... WRITE). The main purpose of IX and IS locks is to show that someone is locking a row, or going to lock a row in the table.

The following example illustrates how an error can occur when a lock request would cause a deadlock. The example involves two clients, A and B.

First, client A creates a table containing one row, and then begins a transaction. Within the transaction, A obtains an S lock on the row by selecting it in share mode:

mysql> CREATE TABLE t (i INT) ENGINE = InnoDB;
Query OK, 0 rows affected (1.07 sec)

mysql> INSERT INTO t (i) VALUES(1);
Query OK, 1 row affected (0.09 sec)

mysql> START TRANSACTION;
Query OK, 0 rows affected (0.00 sec)

mysql> SELECT * FROM t WHERE i = 1 LOCK IN SHARE MODE;
+------+
| i    |
+------+
|    1 |
+------+
1 row in set (0.10 sec)

Next, client B begins a transaction and attempts to delete the row from the table:

mysql> START TRANSACTION;
Query OK, 0 rows affected (0.00 sec)

mysql> DELETE FROM t WHERE i = 1;

The delete operation requires an X lock. The lock cannot be granted because it is incompatible with the S lock that client A holds, so the request goes on the queue of lock requests for the row and client B blocks.

Finally, client A also attempts to delete the row from the table:

mysql> DELETE FROM t WHERE i = 1;
ERROR 1213 (40001): Deadlock found when trying to get lock;
try restarting transaction

Deadlock occurs here because client A needs an X lock to delete the row. However, that lock request cannot be granted because client B is already has a request for an X lock and is waiting for client A to release its S lock. Nor can the S lock held by A be upgraded to an X lock because of the prior request by B for an X lock. As a result, InnoDB generates an error for client A and releases its locks. At that point, the lock request for client B can be granted and B deletes the row from the table.

In InnoDB, all user activity occurs inside a transaction. If the autocommit mode is enabled, each SQL statement forms a single transaction on its own. MySQL always starts a new connection with autocommit enabled.

If the autocommit mode is switched off with SET AUTOCOMMIT = 0, then we can consider that a user always has a transaction open. A SQL COMMIT or ROLLBACK statement ends the current transaction and a new one starts. Both statements release all InnoDB locks that were set during the current transaction. A COMMIT means that the changes made in the current transaction are made permanent and become visible to other users. A ROLLBACK statement, on the other hand, cancels all modifications made by the current transaction.

If the connection has autocommit enabled, the user can still perform a multiple-statement transaction by starting it with an explicit START TRANSACTION or BEGIN statement and ending it with COMMIT or ROLLBACK.

In terms of the SQL:1992 transaction isolation levels, the InnoDB default is REPEATABLE READ. In MySQL 5.0, InnoDB offers all four transaction isolation levels described by the SQL standard. You can set the default isolation level for all connections by using the --transaction-isolation option on the command line or in option files. For example, you can set the option in the [mysqld] section of my.cnf like this:

[mysqld]
transaction-isolation = {READ-UNCOMMITTED | READ-COMMITTED
                         | REPEATABLE-READ | SERIALIZABLE}

A user can change the isolation level of a single session or all new incoming connections with the SET TRANSACTION statement. Its syntax is as follows:

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

Note that there are hyphens in the level names for the --transaction-isolation option, but not for the SET TRANSACTION statement.

The default behavior is to set the isolation level for the next (not started) transaction. If you use the GLOBAL keyword, the statement sets the default transaction level globally for all new connections created from that point on (but not existing connections). You need the SUPER privilege to do this. Using the SESSION keyword sets the default transaction level for all future transactions performed on the current connection.

Any client is free to change the session isolation level (even in the middle of a transaction), or the isolation level for the next transaction.

You can query the global and session transaction isolation levels with these statements:

SELECT @@global.tx_isolation;
SELECT @@tx_isolation;

In row-level locking, InnoDB uses next-key locking. That means that besides index records, InnoDB can also lock the “gap” preceding an index record to block insertions by other users immediately before the index record. A next-key lock refers to a lock that locks an index record and the gap before it. A gap lock refers to a lock that only locks a gap before some index record.

A detailed description of each isolation level in InnoDB follows:

A consistent read means that InnoDB uses its multi-versioning to present to a query a snapshot of the database at a point in time. The query see the changes made by exactly those transactions that committed before that point of time, and no changes made by later or uncommitted transactions. The exception to this rule is that the query sees the changes made by the transaction itself that issues the query.

If you are running with the default REPEATABLE READ isolation level, then all consistent reads within the same transaction read the snapshot established by the first such read in that transaction. You can get a fresher snapshot for your queries by committing the current transaction and after that issuing new queries.

Consistent read is the default mode in which InnoDB processes SELECT statements in READ COMMITTED and REPEATABLE READ isolation levels. A consistent read does not set any locks on the tables it accesses, and therefore other users are free to modify those tables at the same time a consistent read is being performed on the table.

Note that consistent read does not work over DROP TABLE and over ALTER TABLE. Consistent read does not work over DROP TABLE because MySQL can't use table that has been dropped and InnoDB destroys the table. Consistent read does not work over ALTER TABLE because it is executed inside of the transaction which creates a new table and inserts rows from the old table to the new table. Now, when you reissue the consistent read, it will not see any rows in the new table, because they were inserted in a transaction that is not visible in the snapshot that the consistent read reads.

In some circumstances, a consistent read is not convenient. For example, you might want to add a new row into your table child, and make sure that the child has a parent in table parent. The following example shows how to implement referential integrity in your application code.

Suppose that you use a consistent read to read the table parent and indeed see the parent of the child in the table. Can you safely add the child row to table child? No, because it may happen that meanwhile some other user deletes the parent row from the table parent, without you being aware of it.

The solution is to perform the SELECT in a locking mode using LOCK IN SHARE MODE:

SELECT * FROM parent WHERE NAME = 'Jones' LOCK IN SHARE MODE;

Performing a read in share mode means that we read the latest available data, and set a shared mode lock on the rows we read. A shared mode lock prevents others from updating or deleting the row we have read. Also, if the latest data belongs to a yet uncommitted transaction of another client connection, we wait until that transaction commits. After we see that the preceding query returns the parent 'Jones', we can safely add the child record to the child table and commit our transaction.

Let us look at another example: We have an integer counter field in a table child_codes that we use to assign a unique identifier to each child added to table child. Obviously, using a consistent read or a shared mode read to read the present value of the counter is not a good idea, since two users of the database may then see the same value for the counter, and a duplicate-key error occurs if two users attempt to add children with the same identifier to the table.

Here, LOCK IN SHARE MODE is not a good solution because if two users read the counter at the same time, at least one of them ends up in deadlock when attempting to update the counter.

In this case, there are two good ways to implement the reading and incrementing of the counter: (1) update the counter first by incrementing it by 1 and only after that read it, or (2) read the counter first with a lock mode FOR UPDATE, and increment after that. The latter approach can be implemented as follows:

SELECT counter_field FROM child_codes FOR UPDATE;
UPDATE child_codes SET counter_field = counter_field + 1;

A SELECT ... FOR UPDATE reads the latest available data, setting exclusive locks on each row it reads. Thus, it sets the same locks a searched SQL UPDATE would set on the rows.

Please note that the above is merely an example of how SELECT ... FOR UPDATE works. In MySQL, the specific task of generating a unique identifier actually can be accomplished using only a single access to the table:

UPDATE child_codes SET counter_field = LAST_INSERT_ID(counter_field + 1);
SELECT LAST_INSERT_ID();

The SELECT statement merely retrieves the identifier information (specific to the current connection). It does not access any table.

In row-level locking, InnoDB uses an algorithm called next-key locking. InnoDB performs the row-level locking in such a way that when it searches or scans an index of a table, it sets shared or exclusive locks on the index records it encounters. Thus, the row-level locks are actually index record locks.

The locks InnoDB sets on index records also affect the “gap” before that index record. If a user has a shared or exclusive lock on record R in an index, another user cannot insert a new index record immediately before R in the index order. This locking of gaps is done to prevent the so-called “phantom problem”. Suppose that you want to read and lock all children from the child table having an identifier value greater than 100, with the intention of updating some column in the selected rows later:

SELECT * FROM child WHERE id > 100 FOR UPDATE;

Suppose that there is an index on the id column. The query scans that index starting from the first record where id is bigger than 100. If the locks set on the index records would not lock out inserts made in the gaps, a new row might meanwhile be inserted to the table. If you execute the same SELECT within the same transaction, you would see a new row in the result set returned by the query. This is contrary the isolation principle of transactions: A transaction should be able to run so that the data it has read does not change during the transaction. If we regard a set of rows as a data item, the new “phantom” child would violate this isolation principle.

When InnoDB scans an index, it can also lock the gap after the last record in the index. Just that happens in the previous example: The locks set by InnoDB prevent any insert to the table where id would be bigger than 100.

You can use next-key locking to implement a uniqueness check in your application: If you read your data in share mode and do not see a duplicate for a row you are going to insert, then you can safely insert your row and know that the next-key lock set on the successor of your row during the read prevents anyone meanwhile inserting a duplicate for your row. Thus, the next-key locking allows you to “lock” the non-existence of something in your table.

Suppose that you are running in the default REPEATABLE READ isolation level. When you issue a consistent read, that is, an ordinary SELECT statement, InnoDB gives your transaction a timepoint according to which your query sees the database. If another transaction deletes a row and commits after your timepoint was assigned, you do not see the row as having been deleted. Inserts and updates are treated similarly.

You can advance your timepoint by committing your transaction and then doing another SELECT.

This is called multi-versioned concurrency control .

               User A                 User B

           SET AUTOCOMMIT=0;      SET AUTOCOMMIT=0;
time
|          SELECT * FROM t;
|          empty set
|                                 INSERT INTO t VALUES (1, 2);
|
v          SELECT * FROM t;
           empty set
                                  COMMIT;

           SELECT * FROM t;
           empty set

           COMMIT;

           SELECT * FROM t;
           ---------------------
           |    1    |    2    |
           ---------------------
           1 row in set

In this example, user A sees the row inserted by B only when B has committed the insert and A has committed as well, so that the timepoint is advanced past the commit of B.

If you want to see the “freshest” state of the database, you should use either the READ COMMITTED isolation level or a locking read:

SELECT * FROM t LOCK IN SHARE MODE;

A locking read, an UPDATE, or a DELETE generally set record locks on every index record that is scanned in the processing of the SQL query. It does not matter if there are WHERE conditions in the query that would exclude the row from the result set of the query. InnoDB does not remember the exact WHERE condition, but only knows which index ranges were scanned. The record locks are normally next-key locks that also block inserts to the “gap” immediately before the record.

If the locks to be set are exclusive, then InnoDB always retrieves also the clustered index record and sets a lock on it.

If you do not have indexes suitable for your query and MySQL has to scan the whole table to process the query, every row of the table becomes locked, which in turn blocks all inserts by other users to the table. It is important to create good indexes so that your queries do not unnecessarily need to scan many rows.

MySQL begins each client connection with autocommit mode enabled by default. When autocommit is enabled, MySQL does a commit after each SQL statement if that statement did not return an error.

If you have the autocommit mode off and close a connection without calling an explicit commit of your transaction, then MySQL rolls back your transaction.

If an SQL statement returns an error, the commit/rollback behavior depends on the error. See Section 15.15, “InnoDB Error Handling”.

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.

InnoDB automatically detects a deadlock of transactions and rolls back a transaction or transactions to prevent the deadlock. InnoDB tries to pick small transactions to roll back, the size of a transaction being determined by the number of rows inserted, updated, or deleted.

InnoDB is aware of table locks if innodb_table_locks=1 (1 is the default), and the MySQL layer above it knows about row-level locks. Otherwise, InnoDB cannot detect deadlocks where a table lock set by a MySQL LOCK TABLES statement or a lock set by a storage engine other than InnoDB is involved. You must resolve these situations by setting the value of the innodb_lock_wait_timeout system variable.

When InnoDB performs a complete rollback of a transaction, all the locks of the transaction are released. However, if just a single SQL statement is rolled back as a result of an error, some of the locks set by the SQL statement may be preserved. This is because InnoDB stores row locks in a format such it cannot know afterward which lock was set by which SQL statement.

Deadlocks are a classic problem in transactional databases, but they are not dangerous unless they are so frequent that you cannot run certain transactions at all. Normally, you must write your applications so that they are always prepared to re-issue a transaction if it gets rolled back because of a deadlock.

InnoDB uses automatic row-level locking. You can get deadlocks even in the case of transactions that just insert or delete a single row. That is because these operations are not really “atomic”; they automatically set locks on the (possibly several) index records of the row inserted or deleted.

You can cope with deadlocks and reduce the likelihood of their occurrence with the following techniques:

InnoDB includes InnoDB Monitors that print information about the InnoDB internal state. You can use the SQL statement SHOW INNODB STATUS to fetch the output of the standard InnoDB Monitor to your SQL client. This information is useful in performance tuning. (If you are using the mysql interactive SQL client, the output is more readable if you replace the usual semicolon statement terminator with \G.) For a discussion of InnoDB lock modes see Section 15.10.1, “InnoDB Lock Modes”.

mysql> SHOW INNODB STATUS\G

Another way to use InnoDB Monitors is to let them continuously write data to the standard output of the server mysqld. In this case, no output is sent to clients. When switched on, InnoDB Monitors print data about every 15 seconds. Server output usually is directed to the .err log in the MySQL data directory. This data is useful in performance tuning. On Windows, you must start the server from a command prompt in a console window with the --console option if you want to direct the output to the window rather than to the error log.

Monitor output includes information of the following types:

To cause the standard InnoDB Monitor to write to the standard output of mysqld, use the following SQL statement:

CREATE TABLE innodb_monitor(a INT) ENGINE=INNODB;

The monitor can be stopped by issuing the following statement:

DROP TABLE innodb_monitor;

The CREATE TABLE syntax is just a way to pass a command to the InnoDB engine through MySQL's SQL parser: The only things that matter are the table name innodb_monitor and that it be an InnoDB table. The structure of the table is not relevant at all for the InnoDB Monitor. If you shut down the server when the monitor is running, and you want to start the monitor again, you must drop the table before you can issue a new CREATE TABLE statement to start the monitor. This syntax may change in a future release.

You can use innodb_lock_monitor in a similar fashion. This is the same as innodb_monitor, except that it also provides a great deal of lock information. A separate innodb_tablespace_monitor prints a list of created file segments existing in the tablespace and validates the tablespace allocation data structures. In addition, there is innodb_table_monitor with which you can print the contents of the InnoDB internal data dictionary.

A sample of InnoDB Monitor output:

mysql> SHOW INNODB STATUS\G
*************************** 1. row ***************************
Status:
=====================================
030709 13:00:59 INNODB MONITOR OUTPUT
=====================================
Per second averages calculated from the last 18 seconds
----------
SEMAPHORES
----------
OS WAIT ARRAY INFO: reservation count 413452, signal count 378357
--Thread 32782 has waited at btr0sea.c line 1477 for 0.00 seconds the semaphore:
X-lock on RW-latch at 41a28668 created in file btr0sea.c line 135
a writer (thread id 32782) has reserved it in mode wait exclusive
number of readers 1, waiters flag 1
Last time read locked in file btr0sea.c line 731
Last time write locked in file btr0sea.c line 1347
Mutex spin waits 0, rounds 0, OS waits 0
RW-shared spins 108462, OS waits 37964; RW-excl spins 681824, OS waits 375485
------------------------
LATEST FOREIGN KEY ERROR
------------------------
030709 13:00:59 Transaction:
TRANSACTION 0 290328284, ACTIVE 0 sec, process no 3195, OS thread id 34831 inser
ting
15 lock struct(s), heap size 2496, undo log entries 9
MySQL thread id 25, query id 4668733 localhost heikki update
insert into ibtest11a (D, B, C) values (5, 'khDk' ,'khDk')
Foreign key constraint fails for table test/ibtest11a:
,
  CONSTRAINT `0_219242` FOREIGN KEY (`A`, `D`) REFERENCES `ibtest11b` (`A`, `D`)
 ON DELETE CASCADE ON UPDATE CASCADE
Trying to add in child table, in index PRIMARY tuple:
 0: len 4; hex 80000101; asc ....;; 1: len 4; hex 80000005; asc ....;; 2: len 4;
 hex 6b68446b; asc khDk;; 3: len 6; hex 0000114e0edc; asc ...N..;; 4: len 7; hex
 00000000c3e0a7; asc .......;; 5: len 4; hex 6b68446b; asc khDk;;
But in parent table test/ibtest11b, in index PRIMARY,
the closest match we can find is record:
RECORD: info bits 0 0: len 4; hex 8000015b; asc ...[;; 1: len 4; hex 80000005; a
sc ....;; 2: len 3; hex 6b6864; asc khd;; 3: len 6; hex 0000111ef3eb; asc ......
;; 4: len 7; hex 800001001e0084; asc .......;; 5: len 3; hex 6b6864; asc khd;;
------------------------
LATEST DETECTED DEADLOCK
------------------------
030709 12:59:58
*** (1) TRANSACTION:
TRANSACTION 0 290252780, ACTIVE 1 sec, process no 3185, OS thread id 30733 inser
ting
LOCK WAIT 3 lock struct(s), heap size 320, undo log entries 146
MySQL thread id 21, query id 4553379 localhost heikki update
INSERT INTO alex1 VALUES(86, 86, 794,'aA35818','bb','c79166','d4766t','e187358f'
,'g84586','h794',date_format('2001-04-03 12:54:22','%Y-%m-%d %H:%i'),7
*** (1) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 0 page no 48310 n bits 568 table test/alex1 index symbole
trx id 0 290252780 lock mode S waiting
Record lock, heap no 324 RECORD: info bits 0 0: len 7; hex 61613335383138; asc a
a35818;; 1:
*** (2) TRANSACTION:
TRANSACTION 0 290251546, ACTIVE 2 sec, process no 3190, OS thread id 32782 inser
ting
130 lock struct(s), heap size 11584, undo log entries 437
MySQL thread id 23, query id 4554396 localhost heikki update
REPLACE INTO alex1 VALUES(NULL, 32, NULL,'aa3572','','c3572','d6012t','', NULL,'
h396', NULL, NULL, 7.31,7.31,7.31,200)
*** (2) HOLDS THE LOCK(S):
RECORD LOCKS space id 0 page no 48310 n bits 568 table test/alex1 index symbole
trx id 0 290251546 lock_mode X locks rec but not gap
Record lock, heap no 324 RECORD: info bits 0 0: len 7; hex 61613335383138; asc a
a35818;; 1:
*** (2) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 0 page no 48310 n bits 568 table test/alex1 index symbole
trx id 0 290251546 lock_mode X locks gap before rec insert intention waiting
Record lock, heap no 82 RECORD: info bits 0 0: len 7; hex 61613335373230; asc aa
35720;; 1:
*** WE ROLL BACK TRANSACTION (1)
------------
TRANSACTIONS
------------
Trx id counter 0 290328385
Purge done for trx's n:o < 0 290315608 undo n:o < 0 17
Total number of lock structs in row lock hash table 70
LIST OF TRANSACTIONS FOR EACH SESSION:
---TRANSACTION 0 0, not started, process no 3491, OS thread id 42002
MySQL thread id 32, query id 4668737 localhost heikki
show innodb status
---TRANSACTION 0 290328384, ACTIVE 0 sec, process no 3205, OS thread id 38929 in
serting
1 lock struct(s), heap size 320
MySQL thread id 29, query id 4668736 localhost heikki update
insert into speedc values (1519229,1, 'hgjhjgghggjgjgjgjgjggjgjgjgjgjgggjgjgjlhh
gghggggghhjhghgggggghjhghghghghghhhhghghghjhhjghjghjkghjghjghjghjfhjfh
---TRANSACTION 0 290328383, ACTIVE 0 sec, process no 3180, OS thread id 28684 co
mmitting
1 lock struct(s), heap size 320, undo log entries 1
MySQL thread id 19, query id 4668734 localhost heikki update
insert into speedcm values (1603393,1, 'hgjhjgghggjgjgjgjgjggjgjgjgjgjgggjgjgjlh
hgghggggghhjhghgggggghjhghghghghghhhhghghghjhhjghjghjkghjghjghjghjfhjf
---TRANSACTION 0 290328327, ACTIVE 0 sec, process no 3200, OS thread id 36880 st
arting index read
LOCK WAIT 2 lock struct(s), heap size 320
MySQL thread id 27, query id 4668644 localhost heikki Searching rows for update
update ibtest11a set B = 'kHdkkkk' where A = 89572
------- TRX HAS BEEN WAITING 0 SEC FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 0 page no 65556 n bits 232 table test/ibtest11a index PRIM
ARY trx id 0 290328327 lock_mode X waiting
Record lock, heap no 1 RECORD: info bits 0 0: len 9; hex 73757072656d756d00; asc
 supremum.;;
------------------
---TRANSACTION 0 290328284, ACTIVE 0 sec, process no 3195, OS thread id 34831 ro
llback of SQL statement
ROLLING BACK 14 lock struct(s), heap size 2496, undo log entries 9
MySQL thread id 25, query id 4668733 localhost heikki update
insert into ibtest11a (D, B, C) values (5, 'khDk' ,'khDk')
---TRANSACTION 0 290327208, ACTIVE 1 sec, process no 3190, OS thread id 32782
58 lock struct(s), heap size 5504, undo log entries 159
MySQL thread id 23, query id 4668732 localhost heikki update
REPLACE INTO alex1 VALUES(86, 46, 538,'aa95666','bb','c95666','d9486t','e200498f
','g86814','h538',date_format('2001-04-03 12:54:22','%Y-%m-%d %H:%i'),
---TRANSACTION 0 290323325, ACTIVE 3 sec, process no 3185, OS thread id 30733 in
serting
4 lock struct(s), heap size 1024, undo log entries 165
MySQL thread id 21, query id 4668735 localhost heikki update
INSERT INTO alex1 VALUES(NULL, 49, NULL,'aa42837','','c56319','d1719t','', NULL,
'h321', NULL, NULL, 7.31,7.31,7.31,200)
--------
FILE I/O
--------
I/O thread 0 state: waiting for i/o request (insert buffer thread)
I/O thread 1 state: waiting for i/o request (log thread)
I/O thread 2 state: waiting for i/o request (read thread)
I/O thread 3 state: waiting for i/o request (write thread)
Pending normal aio reads: 0, aio writes: 0,
 ibuf aio reads: 0, log i/o's: 0, sync i/o's: 0
Pending flushes (fsync) log: 0; buffer pool: 0
151671 OS file reads, 94747 OS file writes, 8750 OS fsyncs
25.44 reads/s, 18494 avg bytes/read, 17.55 writes/s, 2.33 fsyncs/s
-------------------------------------
INSERT BUFFER AND ADAPTIVE HASH INDEX
-------------------------------------
Ibuf for space 0: size 1, free list len 19, seg size 21,
85004 inserts, 85004 merged recs, 26669 merges
Hash table size 207619, used cells 14461, node heap has 16 buffer(s)
1877.67 hash searches/s, 5121.10 non-hash searches/s
---
LOG
---
Log sequence number 18 1212842764
Log flushed up to   18 1212665295
Last checkpoint at  18 1135877290
0 pending log writes, 0 pending chkp writes
4341 log i/o's done, 1.22 log i/o's/second
----------------------
BUFFER POOL AND MEMORY
----------------------
Total memory allocated 84966343; in additional pool allocated 1402624
Buffer pool size   3200
Free buffers       110
Database pages     3074
Modified db pages  2674
Pending reads 0
Pending writes: LRU 0, flush list 0, single page 0
Pages read 171380, created 51968, written 194688
28.72 reads/s, 20.72 creates/s, 47.55 writes/s
Buffer pool hit rate 999 / 1000
--------------
ROW OPERATIONS
--------------
0 queries inside InnoDB, 0 queries in queue
Main thread process no. 3004, id 7176, state: purging
Number of rows inserted 3738558, updated 127415, deleted 33707, read 755779
1586.13 inserts/s, 50.89 updates/s, 28.44 deletes/s, 107.88 reads/s
----------------------------
END OF INNODB MONITOR OUTPUT
============================
1 row in set (0.05 sec)

Some notes on the output:

InnoDB sends diagnostic output to stderr or to files rather than to stdout or fixed-size memory buffers, in order to avoid potential buffer overflows. As a side effect, the output of SHOW INNODB STATUS is written to a status file every fifteen seconds. The name of the file is innodb_status.pid, where pid is the server process ID. This file is created in the MySQL data directory. InnoDB removes the file for a normal shutdown. If abnormal shutdowns have occurred, instances of these status files may be present and must be removed manually. Before removing them, you might want to examine them to see if they contain useful information to the cause of abnormal shutdowns. IN MySQL 5.0, innodb_status.pid file is created only if the configuration option innodb_status_file=1 is set.

All indexes in InnoDB are B-trees where the index records are stored in the leaf pages of the tree. The default size of an index page is 16KB. When new records are inserted, InnoDB tries to leave 1/16 of the page free for future insertions and updates of the index records.

If index records are inserted in a sequential order (ascending or descending), the resulting index pages are about 15/16 full. If records are inserted in a random order, the pages are from 1/2 to 15/16 full. If the fillfactor of an index page drops below 1/2, InnoDB tries to contract the index tree to free the page.

It is a common situation in a database application that the primary key is a unique identifier and new rows are inserted in the ascending order of the primary key. Thus, the insertions to the clustered index do not require random reads from a disk.

On the other hand, secondary indexes are usually non-unique, and insertions into secondary indexes happen in a relatively random order. This would cause a lot of random disk I/O operations without a special mechanism used in InnoDB.

If an index record should be inserted to a non-unique secondary index, InnoDB checks whether the secondary index page is in the buffer pool. If that is the case, InnoDB does the insertion directly to the index page. If the index page is not found in the buffer pool, InnoDB inserts the record to a special insert buffer structure. The insert buffer is kept so small that it fits entirely in the buffer pool, and insertions can be done very fast.

Periodically, the insert buffer is merged into the secondary index trees in the database. Often it is possible to merge several insertions to the same page of the index tree, saving disk I/O operations. It has been measured that the insert buffer can speed up insertions into a table up to 15 times.

The insert buffer merging may continue to happen after the inserting transaction has been committed. In fact, it may continue to happen after a server shutdown and restart (see Section 15.8.1, “Forcing Recovery”).

The insert buffer merging may take many hours, when many secondary indexes must be updated, and many rows have been inserted. During this time, disk I/O will be increased, which can cause significant slowdown on disk-bound queries. Another significant background I/O operation is the purge thread (see Section 15.12, “Implementation of Multi-Versioning”).

If a table fits almost entirely in main memory, the fastest way to perform queries on it is to use hash indexes. InnoDB has an automatic mechanism that monitors index searches made to the indexes defined for a table. If InnoDB notices that queries could benefit from building a hash index, it does so automatically.

Note that the hash index is always built based on an existing B-tree index on the table. InnoDB can build a hash index on a prefix of any length of the key defined for the B-tree, depending on the pattern of searches that InnoDB observes for the B-tree index. A hash index can be partial: It is not required that the whole B-tree index is cached in the buffer pool. InnoDB builds hash indexes on demand for those pages of the index that are often accessed.

In a sense, InnoDB tailors itself through the adaptive hash index mechanism to ample main memory, coming closer to the architecture of main memory databases.

Records in InnoDB tables have the following characteristics:

InnoDB uses simulated asynchronous disk I/O: InnoDB creates a number of threads to take care of I/O operations, such as read-ahead.

There are two read-ahead heuristics in InnoDB:

InnoDB uses a novel file flush technique called doublewrite . It adds safety to recovery following an operating system crash or a power outage, and improves performance on most varieties of Unix by reducing the need for fsync() operations.

Doublewrite means that before writing pages to a data file, InnoDB first writes them to a contiguous tablespace area called the doublewrite buffer. Only after the write and the flush to the doublewrite buffer has completed does InnoDB write the pages to their proper positions in the data file. If the operating system crashes in the middle of a page write, InnoDB can later find a good copy of the page from the doublewrite buffer during recovery.

In MySQL 5.0, you can also use raw disk partitions as tablespace data files. By using a raw disk, you can perform non-buffered I/O on Windows and on some Unix systems without filesystem overhead, which may improve performance.

When you create a new data file, you must put the keyword newraw immediately after the data file size in innodb_data_file_path. The partition must be at least as large as the size that you specify. Note that 1MB in InnoDB is 1024 * 1024 bytes, whereas 1MB usually means 1,000,000 bytes in disk specifications.

[mysqld]
innodb_data_home_dir=
innodb_data_file_path=/dev/hdd1:3Gnewraw;/dev/hdd2:2Gnewraw

The next time you start the server, InnoDB notices the newraw keyword and initializes the new partition. However, do not create or change any InnoDB tables yet. Otherwise, when you next restart the server, InnoDB reinitializes the partition and your changes are lost. (Starting from 3.23.44, as a safety measure InnoDB prevents users from modifying data when any partition with newraw is specified.)

After InnoDB has initialized the new partition, stop the server, change newraw in the data file specification to raw:

[mysqld]
innodb_data_home_dir=
innodb_data_file_path=/dev/hdd1:5Graw;/dev/hdd2:2Graw

Then restart the server and InnoDB allows changes to be made.

On Windows, you can allocate a disk partition as a data file like this:

[mysqld]
innodb_data_home_dir=
innodb_data_file_path=//./D::10Gnewraw

The //./ corresponds to the Windows syntax of \\.\ for accessing physical drives.

When you use raw disk partitions, be sure that they have permissions that allow read and write access by the account used for running the MySQL server.

The data files you define in the configuration file form the tablespace of InnoDB. The files are simply concatenated to form the tablespace. There is no striping in use. Currently you cannot define where in the tablespace your tables are allocated. However, in a newly created tablespace, InnoDB allocates space starting from the first data file.

The tablespace consists of database pages with a default size of 16KB. The pages are grouped into extents of 64 consecutive pages. The “files” inside a tablespace are called segments in InnoDB. The term “rollback segment” is somewhat confusing because it actually contains many tablespace segments.

Two segments are allocated for each index in InnoDB. One is for non-leaf nodes of the B-tree, the other is for the leaf nodes. The idea here is to achieve better sequentiality for the leaf nodes, which contain the data.

When a segment grows inside the tablespace, InnoDB allocates the first 32 pages to it individually. After that InnoDB starts to allocate whole extents to the segment. InnoDB can add to a large segment up to 4 extents at a time to ensure good sequentiality of data.

Some pages in the tablespace contain bitmaps of other pages, and therefore a few extents in an InnoDB tablespace cannot be allocated to segments as a whole, but only as individual pages.

When you ask for available free space in the tablespace by issuing a SHOW TABLE STATUS, InnoDB reports the extents that are definitely free in the tablespace. InnoDB always reserves some extents for clean-up and other internal purposes; these reserved extents are not included in the free space.

When you delete data from a table, InnoDB contracts the corresponding B-tree indexes. It depends on the pattern of deletes whether that frees individual pages or extents to the tablespace, so that the freed space becomes available for other users. Dropping a table or deleting all rows from it is guaranteed to release the space to other users, but remember that deleted rows are physically removed only in an (automatic) purge operation after they are no longer needed for transaction rollbacks or consistent reads.

If there are random insertions into or deletions from the indexes of a table, the indexes may become fragmented. Fragmentation means that the physical ordering of the index pages on the disk is not close to the index ordering of the records on the pages, or that there are many unused pages in the 64-page blocks that were allocated to the index.

A symptom of fragmentation is that a table takes more space than it “should” take. How much that is exactly, is difficult to determine. All InnoDB data and indexes are stored in B-trees, and their fill factor may vary from 50% to 100%. Another symptom of fragmentation is that a table scan such as:

SELECT COUNT(*) FROM t WHERE a_non_indexed_column <> 12345;

takes more time than it “should” take. (In the query above we are “fooling” the SQL optimizer into scanning the clustered index, rather than a secondary index.) Most disks can read 10 to 50 MB/s, which can be used to estimate how fast a table scan should run.

It can speed up index scans if you periodically perform a “null” ALTER TABLE operation:

ALTER TABLE tbl_name ENGINE=INNODB

That causes MySQL to rebuild the table. Another way to perform a defragmention operation is to use mysqldump to dump the table to a text file, drop the table, and reload it from the dump file.

If the insertions to an index are always ascending and records are deleted only from the end, the InnoDB filespace management algorithm guarantees that fragmentation in the index does not occur.

The following is a non-exhaustive list of common InnoDB-specific errors that you may encounter, with information about why each occurs and how to resolve the problem.

To print the meaning of an operating system error number, use the perror program that comes with the MySQL distribution.

The following table provides a list of some common Linux system error codes. For a more complete list, see Linux source code.

The following table provides a list of some common Windows system error codes. For a complete list see the Microsoft website.

A specific issue with tables is that the MySQL server keeps data dictionary information in .frm files it stores in the database directories, while InnoDB also stores the information into its own data dictionary inside the tablespace files. If you move .frm files around, or if the server crashes in the middle of a data dictionary operation, the .frm files may end up out of sync with InnoDB's internal data dictionary.

A symptom of an out-of-sync data dictionary is that a CREATE TABLE statement fails. If this occurs, you should look in the server's error log. If the log says that the table already exists inside the InnoDB internal data dictionary, you have an orphaned table inside the InnoDB tablespace files that has no corresponding .frm file. The error message looks like this:

InnoDB: Error: table test/parent already exists in InnoDB internal
InnoDB: data dictionary. Have you deleted the .frm file
InnoDB: and not used DROP TABLE? Have you used DROP DATABASE
InnoDB: for InnoDB tables in MySQL version <= 3.23.43?
InnoDB: See the Restrictions section of the InnoDB manual.
InnoDB: You can drop the orphaned table inside InnoDB by
InnoDB: creating an InnoDB table with the same name in another
InnoDB: database and moving the .frm file to the current database.
InnoDB: Then MySQL thinks the table exists, and DROP TABLE will
InnoDB: succeed.

You can drop the orphaned table by following the instructions given in the error message. If you are still unable to use DROP TABLE successfully, the problem may be due to name completion in the mysql client. To work around this, start the mysql client with the --disable-auto-rehash option and try DROP TABLE again. (With name completion on, mysql tries to construct a list of table names, which doesn't work when a problem such as just described exists.)

Another symptom of an out-of-sync data dictionary is that MySQL prints an error that it cannot open a .InnoDB file:

ERROR 1016: Can't open file: 'child2.InnoDB'. (errno: 1)

In the error log you can find a message like this:

InnoDB: Cannot find table test/child2 from the internal data dictionary
InnoDB: of InnoDB though the .frm file for the table exists. Maybe you
InnoDB: have deleted and recreated InnoDB data files but have forgotten
InnoDB: to delete the corresponding .frm files of InnoDB tables?

This means that there is an orphaned .frm file without a corresponding table inside InnoDB. You can drop the orphaned .frm file by deleting it manually.

If MySQL crashes in the middle of an ALTER TABLE operation, you may end up with an orphaned temporary table inside the InnoDB tablespace. Using innodb_table_monitor you can see listed a table whose name is #sql-.... In MySQL 5.0, you can perform SQL statements on tables whose name contains the character ‘#’ if you enclose the name in backticks. Thus, you can drop such an orphaned table like any other orphaned table using the method described above. Note that to copy or rename a file in the Unix shell, you need to put the file name in double quotes if the file name contains ‘#’.