Mysql, slow avg query - mysql

Question

I have a single table book_log Mysql 5.7
+------------+----------+------+-----+---------+----------------+
| Field | Type | Null | Key | Default | Extra |
+------------+----------+------+-----+---------+----------------+
| id | int(11) | NO | PRI | NULL | auto_increment |
| book_id | int(11) | YES | MUL | NULL | |
| type | int(11) | NO | MUL | NULL | |
| value | int(11) | YES | | NULL | |
| created_at | datetime | NO | | NULL | |
+------------+----------+------+-----+---------+----------------+
Book table makes connection with series (One series can have many books)
Create table info :
book_log | CREATE TABLE `book_log` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`book_id` int(11) DEFAULT NULL,
`type` int(11) NOT NULL,
`value` int(11) DEFAULT NULL,
`created_at` datetime NOT NULL,
PRIMARY KEY (`id`),
KEY `IDX_7E42115316A2B381` (`book_id`),
KEY `IDX_TYPE` (`type`),
KEY `IDX_ME` (`book_id`,`type`) USING BTREE,
CONSTRAINT `FK_7E42115316A2B381` FOREIGN KEY (`book_id`) REFERENCES `book` (`id`)
) ENGINE=InnoDB AUTO_INCREMENT=1158962 DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci |
book | CREATE TABLE `book` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`series_id` int(11) DEFAULT NULL,
`language_id` int(11) DEFAULT NULL,
`position` int(11) NOT NULL,
`dir` varchar(255) COLLATE utf8mb4_unicode_ci NOT NULL,
`title` varchar(255) COLLATE utf8mb4_unicode_ci NOT NULL,
`created_at` datetime NOT NULL,
`updated_at` datetime NOT NULL,
PRIMARY KEY (`id`),
KEY `IDX_CBE5A3315278319C` (`series_id`),
KEY `IDX_CBE5A33182F1BAF4` (`language_id`),
CONSTRAINT `FK_CBE5A3315278319C` FOREIGN KEY (`series_id`) REFERENCES `series` (`id`),
CONSTRAINT `FK_CBE5A33182F1BAF4` FOREIGN KEY (`language_id`) REFERENCES `language` (`id`)
) ENGINE=InnoDB AUTO_INCREMENT=55022 DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci |
I make the avg value for a given series
select AVG(value)
from book_log
join book b on book_log.book_id = b.id
where type = 20 and b.series_id = ?;
Explain :
+----+-------------+----------+------------+------+------------------------------+----------------------+---------+----------------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+----------+------------+------+------------------------------+----------------------+---------+----------------+------+----------+-------------+
| 1 | SIMPLE | b | NULL | ref | PRIMARY,IDX_CBE5A3315278319C | IDX_CBE5A3315278319C | 5 | const | 212 | 100.00 | Using index |
| 1 | SIMPLE | book_log | NULL | ref | IDX_7E42115316A2B381,IDX_ME | IDX_7E42115316A2B381 | 5 | bdd.b.id | 33 | 100.00 | NULL |
+----+-------------+----------+------------+------+------------------------------+----------------------+---------+----------------+------+----------+-------------+
Or
select AVG(value)
from book_log
where type = 20 AND book_id IN (
select id from book where series_id = ?
);
Explain :
+----+-------------+----------+------------+------+--------------------------------------+----------------------+---------+-------------------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+----------+------------+------+--------------------------------------+----------------------+---------+-------------------+------+----------+-------------+
| 1 | SIMPLE | book | NULL | ref | PRIMARY,IDX_CBE5A3315278319C | IDX_CBE5A3315278319C | 5 | const | 212 | 100.00 | Using index |
| 1 | SIMPLE | book_log | NULL | ref | IDX_7E42115316A2B381,IDX_TYPE,IDX_ME | IDX_7E42115316A2B381 | 5 | bdd.book.id | 33 | 3.72 | Using where |
+----+-------------+----------+------------+------+--------------------------------------+----------------------+---------+-------------------+------+----------+-------------+
I have 10 973 results for these query, 42ms for a count(*) but more than 1 sec for the avg query.
I don't understand why is it so long.
Any idea ?
Thx.

Answer 1

You can expect to COUNT(*) be as fast or faster than SUM(somecol) or AVG(othercolumn). Why? The database server is, by the rules of SQL, to apply any optimization that yields the correct anwer. COUNT(*) has some serious optimization to it.
But the aggregate functions that do arithmetic ; they must instead examine every record. So, slower.
You can create a purpose-built index for your query.
It is this:
ALTER TABLE book_log
ADD INDEX type_id_val
(type, book_id, val)
I chose these columns for the index because your query searches for a particuler type in the index. Upon finding the first row of the chosen type, MySQL can range-scan through just the index and not the table. So, faster. It's called a covering index.