Probably bad index, full table scan

问题

Can you help me with index my tables?

Problem is that i indexed my tables, but i still have "full table scan" in my explain

this is my (working) query, but on big tables it could be slow, and i dont know how to change this

EXPLAIN select * from  stats_clicked s
join visitor v on s.visitor_id=v.id

ps. index3 - I dont wan't many times values (1,5) when visitor=1 refresh page with id=5

CREATE TABLE `visitor` (
   `id` int(11) NOT NULL AUTO_INCREMENT,
   `visited_time` int(11) DEFAULT NULL,
   PRIMARY KEY (`id`)
 ) ENGINE=InnoDB AUTO_INCREMENT=1 DEFAULT CHARSET=utf8;

CREATE TABLE `stats_clicked` (
   `id` int(11) NOT NULL AUTO_INCREMENT,
   `visitor_id` int(11) NOT NULL,
   `page_clicked_id` int(11) DEFAULT NULL,
   PRIMARY KEY (`id`),
   UNIQUE KEY `index3` (`visitor_id`,`page_clicked_id`),
   KEY `index1` (`visitor_id`)
 ) ENGINE=InnoDB AUTO_INCREMENT=1 DEFAULT CHARSET=utf8;


insert into visitor (`visited_time`) values
(1467122944),(1467122944),(1467122944),
(1467122944),(1467122944),(1467122944),
(1467122944),(1467122944),(1467122944),
(1467122944),(1467122944),(1467122944),
(1467122944),(1467122944),(1467122944);

insert into `stats_clicked` ( `visitor_id`,`page_clicked_id`) values
(1,47),(2,24),(3,83),(3,8),(3,85),(3,88),(4,57),
(5,2),(6,1),(7,28),(8,83),(9,11),(9,16),(9,1),(10,17),
(11,70),(12,73),(13,97),(14,57),(15,30),(15,2),(15,22);

回答1:

If I perform what you did above, I get

EXPLAIN select * from  stats_clicked s 
join visitor v on s.visitor_id=v.id 
+----+-------------+-------+------+---------------+--------+---------+--------------------+------+-------------+
| id | select_type | table | type | possible_keys | key    | key_len | ref                | rows | Extra       |
+----+-------------+-------+------+---------------+--------+---------+--------------------+------+-------------+
|  1 | SIMPLE      | v     | ALL  | PRIMARY       | NULL   | NULL    | NULL               |   15 | NULL        |
|  1 | SIMPLE      | s     | ref  | index3,index1 | index3 | 4       | so_gibberish2.v.id |    1 | Using index |
+----+-------------+-------+------+---------------+--------+---------+--------------------+------+-------------+

However if I truncate then do the following load of a lot of data (ending up with over 100K rows):

truncate table visitor;

insert into visitor (`visited_time`) values
(1467122944),(1467122944),(1467122944),
(1467122944),(1467122944),(1467122944),
(1467122944),(1467122944),(1467122944),
(1467122944),(1467122944),(1467122944),
(1467122944),(1467122944),(1467122944);

insert into visitor (`visited_time`) values
(1467122945),(1467122945),(1467122945),
(1467122945),(1467122945),(1467122945),
(1467122945),(1467122945),(1467122945),
(1467122945),(1467122945),(1467122945),
(1467122945),(1467122945),(1467122945),


insert into visitor (`visited_time`) values
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946),
(1467122946),(1467122946),(1467122946);

insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;
insert visitor(`visited_time`) select `visited_time` from visitor;

select count(*) from visitor;
-- 104448 rows

This results in NOT a table scan:

EXPLAIN select * from  stats_clicked s 
join visitor v on s.visitor_id=v.id; 

+----+-------------+-------+--------+---------------+---------+---------+----------------------------+------+-------------+
| id | select_type | table | type   | possible_keys | key     | key_len | ref                        | rows | Extra       |
+----+-------------+-------+--------+---------------+---------+---------+----------------------------+------+-------------+
|  1 | SIMPLE      | s     | index  | index3,index1 | index3  | 9       | NULL                       |   22 | Using index |
|  1 | SIMPLE      | v     | eq_ref | PRIMARY       | PRIMARY | 4       | so_gibberish2.s.visitor_id |    1 | NULL        |
+----+-------------+-------+--------+---------------+---------+---------+----------------------------+------+-------------+

The reason is listed in the Manual Page How MySQL Uses Indexes:

Indexes are less important for queries on small tables, or big tables where report queries process most or all of the rows. When a query needs to access most of the rows, reading sequentially is faster than working through an index. Sequential reads minimize disk seeks, even if not all the rows are needed for the query.

The reason being listed above. In your question's example, you had too few rows for making the index use worth it. So the db engine chose its allegedly (and probably) faster way of not using the index on your small table.

回答2:

stats_clicked does not need id. In fact, its existence slows down all likely queries with that table. More details.

来源：https://stackoverflow.com/questions/38102137/probably-bad-index-full-table-scan

标签

mysql

indexing

full-table-scan