我有这个查询:
SELECT SQL_NO_CACHE
COUNT(*) AS `numrows`
FROM
(`citations`)
LEFT JOIN
`projects` ON `projects`.`project_id` = `citations`.`project_id`
LEFT JOIN
`users` ON `users`.`user_id` = `projects`.`user_id`
WHERE
`users`.`role` = '0'
AND `citations`.`created` BETWEEN 1360213200 AND 1360299599
AND `citations`.`in_card` = '0'
AND `citations`.`citation_id` NOT IN (SELECT
user_stats_citations.citation_id
FROM
user_stats_citations,
user_stats FORCE INDEX (user_stats_type_index)
WHERE
user_stats_citations.user_stat_id = user_stats.id
AND user_stats.type IN (69 , 70, 71, 75, 76));
我在用户表上有这些索引:
users 0 PRIMARY 1 user_id A 42836 (NULL) (NULL) BTREE
users 1 users_industry_id_index 1 industry_id A 118 (NULL) (NULL) YES BTREE
users 1 users_sponsor_index 1 sponsor A 12 (NULL) (NULL) YES BTREE
这是 EXPLAIN EXTENDED 的输出
id select_type table type possible_keys key key_len ref rows filtered Extra
1 PRIMARY users ALL PRIMARY \N \N \N 42836 100.00 Using where
1 PRIMARY projects ref PRIMARY\,projects_user_id_index projects_user_id_index 4 citelighter.users.user_id 1 100.00 Using where; Using index
1 PRIMARY citations ref citations_project_id_index citations_project_id_index 4 citelighter.projects.project_id 4 100.00 Using index condition; Using where
2 SUBQUERY user_stats range user_stats_type_index user_stats_type_index 2 \N 410768 100.00 Using where; Using index
2 SUBQUERY user_stats_citations ref user_stats_citations_index_user_stat_id\,user_stats_citations_index_citation_id user_stats_citations_index_user_stat_id 8 citelighter.user_stats.id 1 100.00 \N
我尝试在用户 LEFT JOIN 上添加 FORCE INDEX,但未使用索引。你能帮我解决这个问题吗,因为这个查询在我的本地大约需要 10 秒,在生产环境上大约需要 1 秒。
最佳答案
我注意到的第一件事是,where 子句中的谓词:WHERE users.role = '0'
将您的 LEFT JOIN
转换为 INNER JOIN
,所以您也可以将它们设为内部连接。
其次,MySQL 在优化相关子查询方面存在问题,并且在派生表方面也表现不佳。例如在 this simple query :
SELECT *
FROM (SELECT * FROM T) T
JOIN (SELECT * FROM T) T2 ON T.ID = T2.ID;
尽管 ID 是 T
上的主键,但主键不用于连接,因为它无法从派生表中级联出来。同样,有时当你写:
SELECT *
FROM T
WHERE Afield NOT IN (SELECT Afield FROM T WHERE AnotherField = 1);
MySQL 不一定具体化子查询并使用它,它通常会将查询重写为:
SELECT *
FROM T
WHERE NOT EXISTS (SELECT 1
FROM T T2
WHERE T.Afield = T2.Afield
AND T2.AnotherField = 1);
子查询是针对外部查询中的每一行执行的,因此,如果外部查询中有大量行,则为每一行执行子查询会变得非常昂贵。解决方案是尽可能避免子查询。在您的情况下,您可以将查询重写为:
SELECT SQL_NO_CACHE
COUNT(*) AS `numrows`
FROM `citations`
INNER JOIN `projects`
ON `projects`.`project_id` = `citations`.`project_id`
INNER JOIN `users`
ON `users`.`user_id` = `projects`.`user_id`
LEFT JOIN (user_stats_citations
INNER JOIN user_stats
ON user_stats_citations.user_stat_id = user_stats.id
AND user_stats.type IN (69 , 70, 71, 75, 76))
ON user_stats_citations.citation_id = `citations`.`citation_id`
WHERE `users`.`role` = '0'
AND `citations`.`created` BETWEEN 1360213200 AND 1360299599
AND `citations`.`in_card` = '0'
AND user_stats_citations.citation_id IS NULL;
如果没有子查询,就没有派生表,也没有子查询的逐行执行。这应该会缩短执行时间。
关于MySQL 不使用主索引,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/22219820/