MySQL Tip: Look at used_key_parts to Figure Out Which Part of multiple-column Key is Used

In the previous post for this ongoing “EXPLAIN FORMAT=JSON is Cool!” series, we discussed covered indexes and how the used_columns array can help to choose them wisely. There is one more type of multiple-column indexes: composite indexes. Composite indexes are just indexes on multiple columns. Covered indexes are a subgroup of the larger set “composite indexes.” In this post, we’ll discuss how “used_key_parts” can help show which part of a multiple column key is being used.

You should prioritize using composite indexes when you have queries that search on both a set of multiple columns and a single column. For example, if you run queries like:

SELECT first_name, last_name FROM employees WHERE first_name='Steve';
SELECT first_name, last_name FROM employees WHERE first_name='Steve' and last_name like 'V%';
SELECT first_name, last_name FROM employees WHERE first_name='Steve' and last_name like 'V%' and hire_date > '1990-01-01';

It would be better to have a single index on the first_name, last_name and hire_date columns rather than three indexes on first_name, a composite on (first_name, last_name) and a composite on (first_name, last_name, hire_date). But what is the best method for testing the effectiveness of the new index?

Once again, the answer is EXPLAIN FORMAT=JSON.

To illustrate this idea, let’s add a composite index on (first_name, last_name, hire_date) to the table “employees” from the standard employees database:

mysql> alter table employees add index comp (first_name, last_name, hire_date);
Query OK, 0 rows affected (9.32 sec)
Records: 0  Duplicates: 0  Warnings: 0

Now, lets check if this index is used to resolve our queries:

mysql> explain SELECT first_name, last_name FROM employees WHERE first_name='Steve';
+----+-------------+-----------+------------+------+---------------+------+---------+-------+------+----------+-------------+
| id | select_type | table     | partitions | type | possible_keys | key  | key_len | ref   | rows | filtered | Extra       |
+----+-------------+-----------+------------+------+---------------+------+---------+-------+------+----------+-------------+
|  1 | SIMPLE      | employees | NULL       | ref  | comp          | comp | 16      | const |  245 |   100.00 | Using index |
+----+-------------+-----------+------------+------+---------------+------+---------+-------+------+----------+-------------+
1 row in set, 1 warning (0.00 sec)
mysql> explain SELECT first_name, last_name FROM employees WHERE first_name='Steve' and last_name like 'V%';
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+--------------------------+
| id | select_type | table     | partitions | type  | possible_keys | key  | key_len | ref  | rows | filtered | Extra                    |
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+--------------------------+
|  1 | SIMPLE      | employees | NULL       | range | comp          | comp | 34      | NULL |    8 |   100.00 | Using where; Using index |
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+--------------------------+
1 row in set, 1 warning (0.00 sec)
mysql> explain SELECT first_name, last_name FROM employees WHERE first_name='Steve' and last_name like 'V%' and hire_date > '1990-01-01';
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+--------------------------+
| id | select_type | table     | partitions | type  | possible_keys | key  | key_len | ref  | rows | filtered | Extra                    |
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+--------------------------+
|  1 | SIMPLE      | employees | NULL       | range | comp          | comp | 37      | NULL |    8 |    33.33 | Using where; Using index |
+----+-------------+-----------+------------+-------+---------------+------+---------+------+------+----------+--------------------------+
1 row in set, 1 warning (0.00 sec)

It is used in all queries, and key_len is increasing–which shows that each query is using more parts of the index. But which part of index was actually used to resolve the WHERE  condition, and which was used to retrieve rows?

EXPLAIN FORMAT=JSON stores this information in the used_key_parts  member.

For the first two queries, the following result is shown:

mysql> explain format=json SELECT first_name, last_name FROM employees WHERE first_name='Steve'G
*************************** 1. row ***************************
EXPLAIN: {
  "query_block": {
    "select_id": 1,
    "cost_info": {
      "query_cost": "51.22"
    },
    "table": {
      "table_name": "employees",
      "access_type": "ref",
      "possible_keys": [
        "comp"
      ],
      "key": "comp",
      "used_key_parts": [
        "first_name"
      ],
      "key_length": "16",
      "ref": [
        "const"
      ],
      "rows_examined_per_scan": 245,
      "rows_produced_per_join": 245,
      "filtered": "100.00",
      "using_index": true,
      "cost_info": {
        "read_cost": "2.22",
        "eval_cost": "49.00",
        "prefix_cost": "51.22",
        "data_read_per_join": "11K"
      },
      "used_columns": [
        "first_name",
        "last_name"
      ]
    }
  }
}
1 row in set, 1 warning (0.00 sec)

Only the first_name field of the index was used for the query with the WHERE first_name='Steve' condition.

mysql> explain format=json SELECT first_name, last_name FROM employees WHERE first_name='Steve' and last_name like 'V%'G
*************************** 1. row ***************************
EXPLAIN: {
  "query_block": {
    "select_id": 1,
    "cost_info": {
      "query_cost": "4.24"
    },
    "table": {
      "table_name": "employees",
      "access_type": "range",
      "possible_keys": [
        "comp"
      ],
      "key": "comp",
      "used_key_parts": [
        "first_name",
        "last_name"
      ],
      "key_length": "34",
      "rows_examined_per_scan": 8,
      "rows_produced_per_join": 8,
      "filtered": "100.00",
      "using_index": true,
      "cost_info": {
        "read_cost": "2.65",
        "eval_cost": "1.60",
        "prefix_cost": "4.25",
        "data_read_per_join": "384"
      },
      "used_columns": [
        "first_name",
        "last_name"
      ],
      "attached_condition": "((`employees`.`employees`.`first_name` = 'Steve') and (`employees`.`employees`.`last_name` like 'V%'))"
    }
  }
}
1 row in set, 1 warning (0.00 sec)

And two fields, first_name and last_name, were used for the second query.

     "key": "comp",
      "used_key_parts": [
        "first_name",
        "last_name"
      ],

But, surprisingly the same result happens for the last query, although it queries column hire_date  too:

mysql> explain format=json SELECT first_name, last_name FROM employees WHERE first_name='Steve' and last_name like 'V%' and hire_date > '1990-01-01'G
*************************** 1. row ***************************
EXPLAIN: {
  "query_block": {
    "select_id": 1,
    "cost_info": {
      "query_cost": "4.24"
    },
    "table": {
      "table_name": "employees",
      "access_type": "range",
      "possible_keys": [
        "comp"
      ],
      "key": "comp",
      "used_key_parts": [
        "first_name",
        "last_name"
      ],
      "key_length": "37",
      "rows_examined_per_scan": 8,
      "rows_produced_per_join": 2,
      "filtered": "33.33",
      "using_index": true,
      "cost_info": {
        "read_cost": "3.71",
        "eval_cost": "0.53",
        "prefix_cost": "4.25",
        "data_read_per_join": "127"
      },
      "used_columns": [
        "first_name",
        "last_name",
        "hire_date"
      ],
      "attached_condition": "((`employees`.`employees`.`first_name` = 'Steve') and (`employees`.`employees`.`last_name` like 'V%') and (`employees`.`employees`.`hire_date` > '1990-01-01'))"
    }
  }
}
1 row in set, 1 warning (0.00 sec)

This is most likely because there are too many values in the hire_date column that satisfy the conditions, so it is easier to retrieve a data set using part of the index and then check the condition for the hire_date column.

This means that since we don’t retrieve hire_date, we can drop it from the index. We might be a bit leary as to what table rows will be accessed to perform final comparison with hire_date column, but in this case it’s fine:

mysql> flush status;
Query OK, 0 rows affected (0.06 sec)
mysql> SELECT first_name, last_name FROM employees WHERE first_name='Steve' and last_name like 'V%' and hire_date > '1990-01-01';
+------------+-----------+
| first_name | last_name |
+------------+-----------+
| Steve      | Vecchi    |
| Steve      | Veldwijk  |
| Steve      | Vickson   |
+------------+-----------+
3 rows in set (0.00 sec)
mysql> show status like 'Handler%';
+----------------------------+-------+
| Variable_name              | Value |
+----------------------------+-------+
| Handler_commit             | 1     |
| Handler_delete             | 0     |
| Handler_discover           | 0     |
| Handler_external_lock      | 2     |
| Handler_mrr_init           | 0     |
| Handler_prepare            | 0     |
| Handler_read_first         | 0     |
| Handler_read_key           | 1     |
| Handler_read_last          | 0     |
| Handler_read_next          | 8     |
| Handler_read_prev          | 0     |
| Handler_read_rnd           | 0     |
| Handler_read_rnd_next      | 0     |
| Handler_rollback           | 0     |
| Handler_savepoint          | 0     |
| Handler_savepoint_rollback | 0     |
| Handler_update             | 0     |
| Handler_write              | 0     |
+----------------------------+-------+
18 rows in set (0.00 sec)
mysql> alter table employees drop index comp;
Query OK, 0 rows affected (0.27 sec)
Records: 0  Duplicates: 0  Warnings: 0
mysql> alter table employees add index comp (first_name, last_name);
Query OK, 0 rows affected (7.57 sec)
Records: 0  Duplicates: 0  Warnings: 0
mysql> flush status;
Query OK, 0 rows affected (0.03 sec)
mysql> SELECT first_name, last_name FROM employees WHERE first_name='Steve' and last_name like 'V%' and hire_date > '1990-01-01';
+------------+-----------+
| first_name | last_name |
+------------+-----------+
| Steve      | Vecchi    |
| Steve      | Veldwijk  |
| Steve      | Vickson   |
+------------+-----------+
3 rows in set (0.00 sec)
mysql> show status like 'Handler%';
+----------------------------+-------+
| Variable_name              | Value |
+----------------------------+-------+
| Handler_commit             | 1     |
| Handler_delete             | 0     |
| Handler_discover           | 0     |
| Handler_external_lock      | 2     |
| Handler_mrr_init           | 0     |
| Handler_prepare            | 0     |
| Handler_read_first         | 0     |
| Handler_read_key           | 1     |
| Handler_read_last          | 0     |
| Handler_read_next          | 8     |
| Handler_read_prev          | 0     |
| Handler_read_rnd           | 0     |
| Handler_read_rnd_next      | 0     |
| Handler_rollback           | 0     |
| Handler_savepoint          | 0     |
| Handler_savepoint_rollback | 0     |
| Handler_update             | 0     |
| Handler_write              | 0     |
+----------------------------+-------+
18 rows in set (0.00 sec)

As you see, the  Handler_*  variables are same for both indexes. The reason for this is that in this case the optimizer can use index condition pushdown optimization:

mysql> explain format=json SELECT first_name, last_name FROM employees WHERE first_name='Steve' and last_name like 'V%' and hire_date > '1990-01-01'G
*************************** 1. row ***************************
EXPLAIN: {
  "query_block": {
    "select_id": 1,
    "cost_info": {
      "query_cost": "12.21"
    },
    "table": {
      "table_name": "employees",
      "access_type": "range",
      "possible_keys": [
        "comp"
      ],
      "key": "comp",
      "used_key_parts": [
        "first_name",
        "last_name"
      ],
      "key_length": "34",
      "rows_examined_per_scan": 8,
      "rows_produced_per_join": 2,
      "filtered": "33.33",
      "index_condition": "((`employees`.`employees`.`first_name` = 'Steve') and (`employees`.`employees`.`last_name` like 'V%'))",
      "cost_info": {
        "read_cost": "11.68",
        "eval_cost": "0.53",
        "prefix_cost": "12.21",
        "data_read_per_join": "127"
      },
      "used_columns": [
        "first_name",
        "last_name",
        "hire_date"
      ],
      "attached_condition": "(`employees`.`employees`.`hire_date` > '1990-01-01')"
    }
  }
}
1 row in set, 1 warning (0.00 sec)

In the output above we don’t have a member:

"using_index": true,

But there is information about index condition:

"index_condition": "((`employees`.`employees`.`first_name` = 'Steve') and (`employees`.`employees`.`last_name` like 'V%'))",

However,  query_cost  is higher in this case: 12.21 against 4.24 for the composite index of the three fields.

Conclusion: The used_key_parts field of the EXPLAIN FORMAT=JSON output can help us to identify how effective our composite indexes are.