Si realizo lo que hiciste anteriormente, obtengo
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 |
+----+-------------+-------+------+---------------+--------+---------+--------------------+------+-------------+
Sin embargo, si trunco, hago la siguiente carga de una gran cantidad de datos (terminando con más de 100 000 filas):
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
Esto resulta en NO un escaneo de tabla:
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 |
+----+-------------+-------+--------+---------------+---------+---------+----------------------------+------+-------------+
El motivo se indica en la página del manual Cómo usa MySQL los índices :
La razón se enumera arriba. En el ejemplo de su pregunta, tenía muy pocas filas para que el uso del índice valiera la pena. Entonces, el motor de db eligió su supuesta (y probablemente) forma más rápida de no usar el índice en su tabla pequeña.
