15.7.2.1 Transaction Isolation Levels

REPEATABLE READ

This is the default isolation level for InnoDB. Consistent reads within the same transaction read the snapshot established by the first read. This means that if you issue several plain (nonlocking) SELECT statements within the same transaction, these SELECT statements are consistent also with respect to each other. See Section 15.7.2.3, “Consistent Nonlocking Reads”.

For locking reads (SELECT with FOR UPDATE or FOR SHARE), UPDATE, and DELETE statements, locking depends on whether the statement uses a unique index with a unique search condition, or a range-type search condition.

READ COMMITTED

Each consistent read, even within the same transaction, sets and reads its own fresh snapshot. For information about consistent reads, see Section 15.7.2.3, “Consistent Nonlocking Reads”.

For locking reads (SELECT with FOR UPDATE or FOR SHARE), UPDATE statements, and DELETE statements, InnoDB locks only index records, not the gaps before them, and thus permits the free insertion of new records next to locked records. Gap locking is only used for foreign-key constraint checking and duplicate-key checking.

Because gap locking is disabled, phantom problems may occur, as other sessions can insert new rows into the gaps. For information about phantoms, see Section 15.7.4, “Phantom Rows”.

Only row-based binary logging is supported with the READ COMMITTED isolation level. If you use READ COMMITTED with binlog_format=MIXED, the server automatically uses row-based logging.

Using READ COMMITTED has additional effects:

Consider the following example, beginning with this table:

In this case, the table has no indexes, so searches and index scans use the hidden clustered index for record locking (see Section 15.6.2.1, “Clustered and Secondary Indexes”) rather than indexed columns.

Suppose that one session performs an UPDATE using these statements:

Suppose also that a second session performs an UPDATE by executing these statements following those of the first session:

As InnoDB executes each UPDATE, it first acquires an exclusive lock for each row, and then determines whether to modify it. If InnoDB does not modify the row, it releases the lock. Otherwise, InnoDB retains the lock until the end of the transaction. This affects transaction processing as follows.

When using the default REPEATABLE READ isolation level, the first UPDATE acquires an x-lock on each row that it reads and does not release any of them:

x-lock(1,2); retain x-lock
x-lock(2,3); update(2,3) to (2,5); retain x-lock
x-lock(3,2); retain x-lock
x-lock(4,3); update(4,3) to (4,5); retain x-lock
x-lock(5,2); retain x-lock

The second UPDATE blocks as soon as it tries to acquire any locks (because first update has retained locks on all rows), and does not proceed until the first UPDATE commits or rolls back:

x-lock(1,2); block and wait for first UPDATE to commit or roll back

If READ COMMITTED is used instead, the first UPDATE acquires an x-lock on each row that it reads and releases those for rows that it does not modify:

x-lock(1,2); unlock(1,2)
x-lock(2,3); update(2,3) to (2,5); retain x-lock
x-lock(3,2); unlock(3,2)
x-lock(4,3); update(4,3) to (4,5); retain x-lock
x-lock(5,2); unlock(5,2)

For the second UPDATE, InnoDB does a “semi-consistent” read, returning the latest committed version of each row that it reads to MySQL so that MySQL can determine whether the row matches the WHERE condition of the UPDATE:

x-lock(1,2); update(1,2) to (1,4); retain x-lock
x-lock(2,3); unlock(2,3)
x-lock(3,2); update(3,2) to (3,4); retain x-lock
x-lock(4,3); unlock(4,3)
x-lock(5,2); update(5,2) to (5,4); retain x-lock

However, if the WHERE condition includes an indexed column, and InnoDB uses the index, only the indexed column is considered when taking and retaining record locks. In the following example, the first UPDATE takes and retains an x-lock on each row where b = 2. The second UPDATE blocks when it tries to acquire x-locks on the same records, as it also uses the index defined on column b.

The effects of using the READ COMMITTED isolation level are the same as enabling the deprecated innodb_locks_unsafe_for_binlog configuration option, with these exceptions:

READ COMMITTED therefore offers finer and more flexible control than innodb_locks_unsafe_for_binlog.

READ UNCOMMITTED

SELECT statements are performed in a nonlocking fashion, but a possible earlier version of a row might be used. Thus, using this isolation level, such reads are not consistent. This is also called a dirty read. Otherwise, this isolation level works like READ COMMITTED.

SERIALIZABLE

This level is like REPEATABLE READ, but InnoDB implicitly converts all plain SELECT statements to SELECT ... FOR SHARE if autocommit is disabled. If autocommit is enabled, the SELECT is its own transaction. It therefore is known to be read only and can be serialized if performed as a consistent (nonlocking) read and need not block for other transactions. (To force a plain SELECT to block if other transactions have modified the selected rows, disable autocommit.)