使用 ORDER BY id 时 MySQL 查询慢

Question

我有一个非常慢的查询，其中第一部分由 gem 创建（https://github.com/CanCanCommunity/cancancan，它创建选择和内部查询），我在其中添加了 ORDER BY 和 LIMIT 用于基于游标的分页。

SELECT `spree_products`.*
FROM `spree_products`
WHERE `spree_products`.`id` IN
    (SELECT `spree_products`.`id`
     FROM `spree_products`
     LEFT OUTER JOIN `spree_vendors` ON `spree_vendors`.`id` = `spree_products`.`vendor_id`
     WHERE `spree_vendors`.`active` = TRUE)
ORDER BY `spree_products`.`id` ASC
LIMIT 50;

=> 50 rows in set (1 min 3.48 sec)

这是表格：

CREATE TABLE `spree_products` (
  `id` int(11) NOT NULL AUTO_INCREMENT,
  `available_on` datetime DEFAULT NULL,
  `permalink` varchar(255) DEFAULT NULL,
  `created_at` datetime NOT NULL,
  `updated_at` datetime NOT NULL,
  `count_on_hand` int(11) DEFAULT NULL,
  `vendor_id` int(11) DEFAULT NULL,
  PRIMARY KEY (`id`),
  KEY `index_spree_products_on_vendor_id` (`vendor_id`)
) ENGINE=InnoDB AUTO_INCREMENT=37209248 DEFAULT CHARSET=utf8mb4

CREATE TABLE `spree_vendors` (
  `id` int(11) NOT NULL AUTO_INCREMENT,
  `name` varchar(255) DEFAULT NULL,
  `active` tinyint(1) DEFAULT '0',
  `created_at` datetime NOT NULL,
  `updated_at` datetime NOT NULL,
  PRIMARY KEY (`id`)
) ENGINE=InnoDB AUTO_INCREMENT=4413 DEFAULT CHARSET=utf8mb4

（我删除了不必要的字段以保持整洁）

上面查询中的EXPLAIN 返回：

+----+-------------+----------------+------------+--------+-------------------------------------------+-----------------------------------+---------+--------------------------------+------+----------+----------------------------------------------+
| id | select_type | table          | partitions | type   | possible_keys                             | key                               | key_len | ref                            | rows | filtered | Extra                                        |
+----+-------------+----------------+------------+--------+-------------------------------------------+-----------------------------------+---------+--------------------------------+------+----------+----------------------------------------------+
|  1 | SIMPLE      | spree_vendors  | NULL       | ALL    | PRIMARY                                   | NULL                              | NULL    | NULL                           | 3465 |    10.00 | Using where; Using temporary; Using filesort |
|  1 | SIMPLE      | spree_products | NULL       | ref    | PRIMARY,index_spree_products_on_vendor_id | index_spree_products_on_vendor_id | 5       | _hubert_test.spree_vendors.id  | 8613 |   100.00 | Using index                                  |
|  1 | SIMPLE      | spree_products | NULL       | eq_ref | PRIMARY                                   | PRIMARY                           | 4       | _hubert_test.spree_products.id |    1 |   100.00 | NULL                                         |
+----+-------------+----------------+------------+--------+-------------------------------------------+-----------------------------------+---------+--------------------------------+------+----------+----------------------------------------------+

当我删除 ORDER BY 时，查询速度很快：

SELECT `spree_products`.*
FROM `spree_products`
WHERE `spree_products`.`id` IN
    (SELECT `spree_products`.`id`
     FROM `spree_products`
     LEFT OUTER JOIN `spree_vendors` ON `spree_vendors`.`id` = `spree_products`.`vendor_id`
     WHERE `spree_vendors`.`active` = TRUE)
LIMIT 50;

=> 50 rows in set (0.00 sec)

当我从外部查询中保留ORDER BY 部分，但从子查询中删除WHERE 部分时，查询也很快：

SELECT `spree_products`.*
FROM `spree_products`
WHERE `spree_products`.`id` IN
    (SELECT `spree_products`.`id`
     FROM `spree_products`
     LEFT OUTER JOIN `spree_vendors` ON `spree_vendors`.`id` = `spree_products`.`vendor_id`)
ORDER BY `spree_products`.`id` ASC
LIMIT 50;

我尝试将复合索引添加到 spree_vendors.id / spree_vendors.active，但这没有帮助。

关于如何优化此查询的任何想法？

更新 1：

JOIN 的变体也很慢。 DISTINCT 由 gem 添加以防止重复记录，以防您未选择所有列：

SELECT DISTINCT `spree_products`.*
FROM `spree_products`
LEFT OUTER JOIN `spree_vendors` ON `spree_vendors`.`id` = `spree_products`.`vendor_id`
WHERE `spree_vendors`.`active` = TRUE
ORDER BY `spree_products`.`id` ASC
LIMIT 50;

=> 50 rows in set (1 min 43.13 sec)

没有DISTINCT，查询速度很快。

更新 2

有人指出，在子查询中使用LEFT OUTER JOIN 会返回整个表。但是当使用INNER JOIN 时它仍然很慢：

SELECT `spree_products`.*
FROM `spree_products`
WHERE `spree_products`.`id` IN
    (SELECT `spree_products`.`id`
     FROM `spree_products`
     INNER JOIN `spree_vendors` ON `spree_vendors`.`id` = `spree_products`.`vendor_id`
     WHERE `spree_vendors`.`active` = TRUE)
ORDER BY `spree_products`.`id` ASC
LIMIT 50;

=> 50 rows in set (1 min 3.98 sec)

Answer 1

鉴于id 必须是PRIMARY，您的查询在功能上必须与此相同：

SELECT [DISTINCT] p.*
  FROM spree_products p
  JOIN spree_vendors v
    ON v.id = p.vendor_id
 WHERE v.active = 1
 ORDER 
    BY p.id ASC
 LIMIT 50;

这将受益于 p.vendor_id 上的索引，也许还有 v.active。