For string columns, indexes can be created that use only the leading part of column values, using col_name(length) syntax to specify an index prefix length:

If a specified index prefix exceeds the maximum column data type size, CREATE INDEX handles the index as follows:

The statement shown here creates an index using the first 10 characters of the name column (assuming that name has a nonbinary string type):

CREATE INDEX part_of_name ON customer (name(10));

If names in the column usually differ in the first 10 characters, lookups performed using this index should not be much slower than using an index created from the entire name column. Also, using column prefixes for indexes can make the index file much smaller, which could save a lot of disk space and might also speed up INSERT operations.

A “normal” index indexes column values or prefixes of column values. For example, in the following table, the index entry for a given t1 row includes the full col1 value and a prefix of the col2 value consisting of its first 10 characters:

CREATE TABLE t1 (
  col1 VARCHAR(10),
  col2 VARCHAR(20),
  INDEX (col1, col2(10))
);

Functional key parts that index expression values canalso be used in place of column or column prefix values. Use of functional key parts enables indexing of values not stored directly in the table. Examples:

CREATE TABLE t1 (col1 INT, col2 INT, INDEX func_index ((ABS(col1))));
CREATE INDEX idx1 ON t1 ((col1 + col2));
CREATE INDEX idx2 ON t1 ((col1 + col2), (col1 - col2), col1);
ALTER TABLE t1 ADD INDEX ((col1 * 40) DESC);

An index with multiple key parts can mix nonfunctional and functional key parts.

ASC and DESC are supported for functional key parts.

Functional key parts must adhere to the following rules. An error occurs if a key part definition contains disallowed constructs.

For CREATE TABLE ... LIKE, the destination table preserves functional key parts from the original table.

Functional indexes are implemented as hidden virtual generated columns, which has these implications:

UNIQUE is supported for indexes that include functional key parts. However, primary keys cannot include functional key parts. A primary key requires the generated column to be stored, but functional key parts are implemented as virtual generated columns, not stored generated columns.

SPATIAL and FULLTEXT indexes cannot have functional key parts.

If a table contains no primary key, InnoDB automatically promotes the first UNIQUE NOT NULL index to the primary key. This is not supported for UNIQUE NOT NULL indexes that have functional key parts.

Nonfunctional indexes raise a warning if there are duplicate indexes. Indexes that contain functional key parts do not have this feature.

To remove a column that is referenced by a functional key part, the index must be removed first. Otherwise, an error occurs.

Although nonfunctional key parts support a prefix length specification, this is not possible for functional key parts. The solution is to use SUBSTRING() (or CAST(), as described later in this section). For a functional key part containing the SUBSTRING() function to be used in a query, the WHERE clause must contain SUBSTRING() with the same arguments. In the following example, only the second SELECT is able to use the index because that is the only query in which the arguments to SUBSTRING() match the index specification:

CREATE TABLE tbl (
  col1 LONGTEXT,
  INDEX idx1 ((SUBSTRING(col1, 1, 10)))
);
SELECT * FROM tbl WHERE SUBSTRING(col1, 1, 9) = '123456789';
SELECT * FROM tbl WHERE SUBSTRING(col1, 1, 10) = '1234567890';

Functional key parts enable indexing of values that cannot be indexed otherwise, such as JSON values. However, this must be done correctly to achieve the desired effect. For example, this syntax does not work:

CREATE TABLE employees (
  data JSON,
  INDEX ((data->>'$.name'))
);

The syntax fails because:

To index the JSON column, you could try using the CAST() function as follows:

CREATE TABLE employees (
  data JSON,
  INDEX ((CAST(data->>'$.name' AS CHAR(30))))
);

The hidden generated column is assigned the VARCHAR(30) data type, which can be indexed. But this approach produces a new issue when trying to use the index:

As a result, there is a collation mismatch between the indexed expression in the preceding table definition and the WHERE clause expression in the following query:

SELECT * FROM employees WHERE data->>'$.name' = 'James';

The index is not used because the expressions in the query and the index differ. To support this kind of scenario for functional key parts, the optimizer automatically strips CAST() when looking for an index to use, but only if the collation of the indexed expression matches that of the query expression. For an index with a functional key part to be used, either of the following two solutions work (although they differ somewhat in effect):

Be aware that although the optimizer supports automatically stripping CAST() with indexed generated columns, the following approach does not work because it produces a different result with and without an index (Bug#27337092):

mysql> CREATE TABLE employees (
         data JSON,
         generated_col VARCHAR(30) AS (CAST(data->>'$.name' AS CHAR(30)))
       );
Query OK, 0 rows affected, 1 warning (0.03 sec)

mysql> INSERT INTO employees (data)
       VALUES ('{"name": "james"}'), ('{"name": "James"}');
Query OK, 2 rows affected, 1 warning (0.01 sec)
Records: 2  Duplicates: 0  Warnings: 1

mysql> SELECT * FROM employees WHERE data->>'$.name' = 'James';
+-------------------+---------------+
| data              | generated_col |
+-------------------+---------------+
| {"name": "James"} | James         |
+-------------------+---------------+
1 row in set (0.00 sec)

mysql> ALTER TABLE employees ADD INDEX idx (generated_col);
Query OK, 0 rows affected, 1 warning (0.03 sec)
Records: 0  Duplicates: 0  Warnings: 1

mysql> SELECT * FROM employees WHERE data->>'$.name' = 'James';
+-------------------+---------------+
| data              | generated_col |
+-------------------+---------------+
| {"name": "james"} | james         |
| {"name": "James"} | James         |
+-------------------+---------------+
2 rows in set (0.01 sec)

A UNIQUE index creates a constraint such that all values in the index must be distinct. An error occurs if you try to add a new row with a key value that matches an existing row. If you specify a prefix value for a column in a UNIQUE index, the column values must be unique within the prefix length. A UNIQUE index permits multiple NULL values for columns that can contain NULL.

If a table has a PRIMARY KEY or UNIQUE NOT NULL index that consists of a single column that has an integer type, you can use _rowid to refer to the indexed column in SELECT statements, as follows:

FULLTEXT indexes are supported only for InnoDB and MyISAM tables and can include only CHAR, VARCHAR, and TEXT columns. Indexing always happens over the entire column; column prefix indexing is not supported and any prefix length is ignored if specified. See Section 14.9, “Full-Text Search Functions”, for details of operation.

InnoDB supports multi-valued indexes. A multi-valued index is a secondary index defined on a column that stores an array of values. A “normal” index has one index record for each data record (1:1). A multi-valued index can have multiple index records for a single data record (N:1). Multi-valued indexes are intended for indexing JSON arrays. For example, a multi-valued index defined on the array of zip codes in the following JSON document creates an index record for each zip code, with each index record referencing the same data record.

{
    "user":"Bob",
    "user_id":31,
    "zipcode":[94477,94536]
}

Creating multi-valued Indexes

You can create a multi-valued index in a CREATE TABLE, ALTER TABLE, or CREATE INDEX statement. This requires using CAST(... AS ... ARRAY) in the index definition, which casts same-typed scalar values in a JSON array to an SQL data type array. A virtual column is then generated transparently with the values in the SQL data type array; finally, a functional index (also referred to as a virtual index) is created on the virtual column. It is the functional index defined on the virtual column of values from the SQL data type array that forms the multi-valued index.

The examples in the following list show the three different ways in which a multi-valued index zips can be created on an array $.zipcode on a JSON column custinfo in a table named customers. In each case, the JSON array is cast to an SQL data type array of UNSIGNED integer values.

A multi-valued index can also be defined as part of a composite index. This example shows a composite index that includes two single-valued parts (for the id and modified columns), and one multi-valued part (for the custinfo column):

CREATE TABLE customers (
    id BIGINT NOT NULL AUTO_INCREMENT PRIMARY KEY,
    modified DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
    custinfo JSON
    );

ALTER TABLE customers ADD INDEX comp(id, modified,
    (CAST(custinfo->'$.zipcode' AS UNSIGNED ARRAY)) );

Only one multi-valued key part can be used in a composite index. The multi-valued key part may be used in any order relative to the other parts of the key. In other words, the ALTER TABLE statement just shown could have used comp(id, (CAST(custinfo->'$.zipcode' AS UNSIGNED ARRAY), modified)) (or any other ordering) and still have been valid.

Using multi-valued Indexes

The optimizer uses a multi-valued index to fetch records when the following functions are specified in a WHERE clause:

We can demonstrate this by creating and populating the customers table using the following CREATE TABLE and INSERT statements:

mysql> CREATE TABLE customers (
    ->     id BIGINT NOT NULL AUTO_INCREMENT PRIMARY KEY,
    ->     modified DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
    ->     custinfo JSON
    ->     );
Query OK, 0 rows affected (0.51 sec)

mysql> INSERT INTO customers VALUES
    ->     (NULL, NOW(), '{"user":"Jack","user_id":37,"zipcode":[94582,94536]}'),
    ->     (NULL, NOW(), '{"user":"Jill","user_id":22,"zipcode":[94568,94507,94582]}'),
    ->     (NULL, NOW(), '{"user":"Bob","user_id":31,"zipcode":[94477,94507]}'),
    ->     (NULL, NOW(), '{"user":"Mary","user_id":72,"zipcode":[94536]}'),
    ->     (NULL, NOW(), '{"user":"Ted","user_id":56,"zipcode":[94507,94582]}');
Query OK, 5 rows affected (0.07 sec)
Records: 5  Duplicates: 0  Warnings: 0

First we execute three queries on the customers table, one each using MEMBER OF(), JSON_CONTAINS(), and JSON_OVERLAPS(), with the result from each query shown here:

mysql> SELECT * FROM customers
    ->     WHERE 94507 MEMBER OF(custinfo->'$.zipcode');
+----+---------------------+-------------------------------------------------------------------+
| id | modified            | custinfo                                                          |
+----+---------------------+-------------------------------------------------------------------+
|  2 | 2019-06-29 22:23:12 | {"user": "Jill", "user_id": 22, "zipcode": [94568, 94507, 94582]} |
|  3 | 2019-06-29 22:23:12 | {"user": "Bob", "user_id": 31, "zipcode": [94477, 94507]}         |
|  5 | 2019-06-29 22:23:12 | {"user": "Ted", "user_id": 56, "zipcode": [94507, 94582]}         |
+----+---------------------+-------------------------------------------------------------------+
3 rows in set (0.00 sec)

mysql> SELECT * FROM customers
    ->     WHERE JSON_CONTAINS(custinfo->'$.zipcode', CAST('[94507,94582]' AS JSON));
+----+---------------------+-------------------------------------------------------------------+
| id | modified            | custinfo                                                          |
+----+---------------------+-------------------------------------------------------------------+
|  2 | 2019-06-29 22:23:12 | {"user": "Jill", "user_id": 22, "zipcode": [94568, 94507, 94582]} |
|  5 | 2019-06-29 22:23:12 | {"user": "Ted", "user_id": 56, "zipcode": [94507, 94582]}         |
+----+---------------------+-------------------------------------------------------------------+
2 rows in set (0.00 sec)

mysql> SELECT * FROM customers
    ->     WHERE JSON_OVERLAPS(custinfo->'$.zipcode', CAST('[94507,94582]' AS JSON));
+----+---------------------+-------------------------------------------------------------------+
| id | modified            | custinfo                                                          |
+----+---------------------+-------------------------------------------------------------------+
|  1 | 2019-06-29 22:23:12 | {"user": "Jack", "user_id": 37, "zipcode": [94582, 94536]}        |
|  2 | 2019-06-29 22:23:12 | {"user": "Jill", "user_id": 22, "zipcode": [94568, 94507, 94582]} |
|  3 | 2019-06-29 22:23:12 | {"user": "Bob", "user_id": 31, "zipcode": [94477, 94507]}         |
|  5 | 2019-06-29 22:23:12 | {"user": "Ted", "user_id": 56, "zipcode": [94507, 94582]}         |
+----+---------------------+-------------------------------------------------------------------+
4 rows in set (0.00 sec)

Next, we run EXPLAIN on each of the previous three queries:

mysql> EXPLAIN SELECT * FROM customers
    ->     WHERE 94507 MEMBER OF(custinfo->'$.zipcode');
+----+-------------+-----------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table     | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-----------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | customers | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    5 |   100.00 | Using where |
+----+-------------+-----------+------------+------+---------------+------+---------+------+------+----------+-------------+
1 row in set, 1 warning (0.00 sec)

mysql> EXPLAIN SELECT * FROM customers
    ->     WHERE JSON_CONTAINS(custinfo->'$.zipcode', CAST('[94507,94582]' AS JSON));
+----+-------------+-----------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table     | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-----------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | customers | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    5 |   100.00 | Using where |
+----+-------------+-----------+------------+------+---------------+------+---------+------+------+----------+-------------+
1 row in set, 1 warning (0.00 sec)

mysql> EXPLAIN SELECT * FROM customers
    ->     WHERE JSON_OVERLAPS(custinfo->'$.zipcode', CAST('[94507,94582]' AS JSON));
+----+-------------+-----------+------------+------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table     | partitions | type | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-----------+------------+------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | customers | NULL       | ALL  | NULL          | NULL | NULL    | NULL |    5 |   100.00 | Using where |
+----+-------------+-----------+------------+------+---------------+------+---------+------+------+----------+-------------+
1 row in set, 1 warning (0.01 sec)

None of the three queries just shown are able to use any keys. To solve this problem, we can add a multi-valued index on the zipcode array in the JSON column (custinfo), like this:

mysql> ALTER TABLE customers
    ->     ADD INDEX zips( (CAST(custinfo->'$.zipcode' AS UNSIGNED ARRAY)) );
Query OK, 0 rows affected (0.47 sec)
Records: 0  Duplicates: 0  Warnings: 0

When we run the previous EXPLAIN statements again, we can now observe that the queries can (and do) use the index zips that was just created:

mysql> EXPLAIN SELECT * FROM customers
    ->     WHERE 94507 MEMBER OF(custinfo->'$.zipcode');
+----+-------------+-----------+------------+------+---------------+------+---------+-------+------+----------+-------------+
| id | select_type | table     | partitions | type | possible_keys | key  | key_len | ref   | rows | filtered | Extra       |
+----+-------------+-----------+------------+------+---------------+------+---------+-------+------+----------+-------------+
|  1 | SIMPLE      | customers | NULL       | ref  | zips          | zips | 9       | const |    1 |   100.00 | Using where |
+----+-------------+-----------+------------+------+---------------+------+---------+-------+------+----------+-------------+
1 row in set, 1 warning (0.00 sec)

mysql> EXPLAIN SELECT * FROM customers
    ->     WHERE JSON_CONTAINS(custinfo->'$.zipcode', CAST('[94507,94582]' AS JSON));
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table     | partitions | type  | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | customers | NULL       | range | zips          | zips | 9       | NULL |    6 |   100.00 | Using where |
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+-------------+
1 row in set, 1 warning (0.00 sec)

mysql> EXPLAIN SELECT * FROM customers
    ->     WHERE JSON_OVERLAPS(custinfo->'$.zipcode', CAST('[94507,94582]' AS JSON));
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+-------------+
| id | select_type | table     | partitions | type  | possible_keys | key  | key_len | ref  | rows | filtered | Extra       |
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+-------------+
|  1 | SIMPLE      | customers | NULL       | range | zips          | zips | 9       | NULL |    6 |   100.00 | Using where |
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+-------------+
1 row in set, 1 warning (0.01 sec)

A multi-valued index can be defined as a unique key. If defined as a unique key, attempting to insert a value already present in the multi-valued index returns a duplicate key error. If duplicate values are already present, attempting to add a unique multi-valued index fails, as shown here:

mysql> ALTER TABLE customers DROP INDEX zips;
Query OK, 0 rows affected (0.55 sec)
Records: 0  Duplicates: 0  Warnings: 0

mysql> ALTER TABLE customers
    ->     ADD UNIQUE INDEX zips((CAST(custinfo->'$.zipcode' AS UNSIGNED ARRAY)));
ERROR 1062 (23000): Duplicate entry '[94507, ' for key 'customers.zips'
mysql> ALTER TABLE customers
    ->     ADD INDEX zips((CAST(custinfo->'$.zipcode' AS UNSIGNED ARRAY)));
Query OK, 0 rows affected (0.36 sec)
Records: 0  Duplicates: 0  Warnings: 0

Characteristics of Multi-Valued Indexes

Multi-valued indexes have the additional characteristics listed here:

Limitations and Restrictions on Multi-valued Indexes

Multi-valued indexes are subject to the limitations and restrictions listed here:

The MyISAM, InnoDB, NDB, and ARCHIVE storage engines support spatial columns such as POINT and GEOMETRY. (Section 13.4, “Spatial Data Types”, describes the spatial data types.) However, support for spatial column indexing varies among engines. Spatial and nonspatial indexes on spatial columns are available according to the following rules.

Spatial indexes on spatial columns have these characteristics:

Nonspatial indexes on spatial columns (created with INDEX, UNIQUE, or PRIMARY KEY) have these characteristics:

Following the key part list, index options can be given. An index_option value can be any of the following:

ALGORITHM and LOCK clauses may be given to influence the table copying method and level of concurrency for reading and writing the table while its indexes are being modified. They have the same meaning as for the ALTER TABLE statement. For more information, see Section 15.1.10, “ALTER TABLE Statement”

NDB Cluster supports online operations using the same ALGORITHM=INPLACE syntax used with the standard MySQL Server. See Section 25.6.12, “Online Operations with ALTER TABLE in NDB Cluster”, for more information.