13.2.8 REPLACE Syntax

REPLACE [LOW_PRIORITY | DELAYED]
    [INTO] tbl_name [(col_name,...)]
    {VALUES | VALUE} ({expr | DEFAULT},...),(...),...

Or:

REPLACE [LOW_PRIORITY | DELAYED]
    [INTO] tbl_name
    SET col_name={expr | DEFAULT}, ...

Or:

REPLACE [LOW_PRIORITY | DELAYED]
    [INTO] tbl_name [(col_name,...)]
    SELECT ...

REPLACE works exactly like INSERT, except that if an old row in the table has the same value as a new row for a PRIMARY KEY or a UNIQUE index, the old row is deleted before the new row is inserted. See Section 13.2.5, “INSERT Syntax”.

REPLACE is a MySQL extension to the SQL standard. It either inserts, or deletes and inserts. For another MySQL extension to standard SQL—that either inserts or updates—see Section 13.2.5.3, “INSERT ... ON DUPLICATE KEY UPDATE Syntax”.

Note that unless the table has a PRIMARY KEY or UNIQUE index, using a REPLACE statement makes no sense. It becomes equivalent to INSERT, because there is no index to be used to determine whether a new row duplicates another.

Values for all columns are taken from the values specified in the REPLACE statement. Any missing columns are set to their default values, just as happens for INSERT. You cannot refer to values from the current row and use them in the new row. If you use an assignment such as SET col_name = col_name + 1, the reference to the column name on the right hand side is treated as DEFAULT(col_name), so the assignment is equivalent to SET col_name = DEFAULT(col_name) + 1.

To use REPLACE, you must have both the INSERT and DELETE privileges for the table.

The REPLACE statement returns a count to indicate the number of rows affected. This is the sum of the rows deleted and inserted. If the count is 1 for a single-row REPLACE, a row was inserted and no rows were deleted. If the count is greater than 1, one or more old rows were deleted before the new row was inserted. It is possible for a single row to replace more than one old row if the table contains multiple unique indexes and the new row duplicates values for different old rows in different unique indexes.

The affected-rows count makes it easy to determine whether REPLACE only added a row or whether it also replaced any rows: Check whether the count is 1 (added) or greater (replaced).

If you are using the C API, the affected-rows count can be obtained using the mysql_affected_rows() function.

Currently, you cannot replace into a table and select from the same table in a subquery.

MySQL uses the following algorithm for REPLACE (and LOAD DATA ... REPLACE):

  1. Try to insert the new row into the table

  2. While the insertion fails because a duplicate-key error occurs for a primary key or unique index:

    1. Delete from the table the conflicting row that has the duplicate key value

    2. Try again to insert the new row into the table

It is possible that in the case of a duplicate-key error, a storage engine may perform the REPLACE as an update rather than a delete plus insert, but the semantics are the same. There are no user-visible effects other than a possible difference in how the storage engine increments Handler_xxx status variables.

Because the results of REPLACE ... SELECT statements depend on the ordering of rows from the SELECT and this order cannot always be guaranteed, it is possible when logging these statements for the master and the slave to diverge. For this reason, in MySQL 5.5.18 and later, REPLACE ... SELECT statements are flagged as unsafe for statement-based replication. With this change, such statements produce a warning in the log when using the STATEMENT binary logging mode, and are logged using the row-based format when using MIXED mode. See also Section 16.1.2.1, “Advantages and Disadvantages of Statement-Based and Row-Based Replication”.

A REPLACEstatement that acts on a partitioned table using a storage engine such as MyISAM that employs table-level locks locks all partitions of the table. This does not occur with tables using storage engines such as InnoDB that employ row-level locking. This issue is resolved in MySQL 5.6. See Section 18.5.4, “Partitioning and Table-Level Locking”, for more information.