MySQL SELECT 与子查询和 GROUP BY 的性能

Question

我正在为一个旧应用程序（MySQL 5.6）构建一个报告工具。它使用单个reports 表，所有相关列都有一个索引，并且该表大约有 120'000 行。

我的查询如下，其中 GROUP BY 列和 WHERE 子句可能会根据用户配置报告的方式而改变：

SELECT
    nationality AS groupValue,
    ( SELECT COUNT( request_id ) FROM reports WHERE create_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND nationality = groupValue ) AS totalRequests,
    ( SELECT SUM(effective_amount) FROM reports WHERE pay_off_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND nationality = groupValue ) AS totalSum,
    ( SELECT COUNT( request_id ) FROM reports WHERE pay_off_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND current_status = 5 AND nationality = groupValue ) AS totalPaid,
    ( SELECT COUNT( request_id ) FROM reports WHERE failed_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND current_status = 3 AND nationality = groupValue ) AS totalRefused,
    ( SELECT COUNT( request_id ) FROM reports WHERE failed_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND current_status = 4 AND nationality = groupValue ) AS totalRenounced,
    ( SELECT COUNT( request_id ) FROM reports WHERE failed_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND current_status = 2 AND nationality = groupValue ) AS totalInProgress 
FROM
    reports 
GROUP BY
    nationality;

GROUP BY 列有大约 40 个不同的值，性能很好，但如果我选择一个有大约 200 个不同值的列（例如国籍，见下文），性能会下降到未知时间（我放弃了等待）。

我的感觉是，这可能与某些二次复杂性有关，因为 GROUP BY 在主查询和子查询中都使用，我需要这样做，因为 WHERE 子句根据它们各自的 SELECT 值而变化。但是我的SQL fu 太久忘记了……

我的问题：

1. 是什么导致性能下降？
2. 如何改进此查询？

更新一：

这是reports 表定义（简化）：

CREATE TABLE `reports` (
  `request_id` int(11) NOT NULL,
  `employee_id` int(11) NOT NULL,
  `create_date` datetime NOT NULL,
  `failed_date` datetime DEFAULT NULL,
  `pay_off_date` datetime DEFAULT NULL,
  `nationality` varchar(30) DEFAULT NULL,
  `effective_amount` decimal(13,2) DEFAULT NULL,
  `current_status` tinyint(2) DEFAULT NULL,
  PRIMARY KEY (`request_id`,`customer_id`,`employee_id`),
  UNIQUE KEY `request_id` (`request_id`),
  KEY `fk_request_id` (`request_id`),
  KEY `fk_employee_id` (`employee_id`),
  KEY `create_date_index` (`create_date`),
  KEY `failed_date_index` (`failed_date`),
  KEY `pay_off_date_index` (`pay_off_date`),
  KEY `nationality_index` (`nationality`(2))
) ENGINE=InnoDB DEFAULT CHARSET=latin1;

这是一个执行良好的查询：

id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
1	PRIMARY	reports	NULL	ALL	NULL	NULL	NULL	NULL	118923	100.00	Using temporary; Using filesort
7	DEPENDENT SUBQUERY	reports	NULL	ref	failed_date_index,nationality_index	nationality_index	5	func	720	0.12	Using where
6	DEPENDENT SUBQUERY	reports	NULL	ref	failed_date_index,nationality_index	nationality_index	5	func	720	0.12	Using where
5	DEPENDENT SUBQUERY	reports	NULL	ref	failed_date_index,nationality_index	nationality_index	5	func	720	0.12	Using where
4	DEPENDENT SUBQUERY	reports	NULL	ALL	pay_off_date_index,nationality_index	NULL	NULL	NULL	118923	1.00	Range checked for each record (index map: 0x8080)
3	DEPENDENT SUBQUERY	reports	NULL	ALL	pay_off_date_index,nationality_index	NULL	NULL	NULL	118923	10.00	Range checked for each record (index map: 0x8080)
2	DEPENDENT SUBQUERY	reports	NULL	ref	create_date_index,nationality_index	nationality_index	5	func	720	2.10	Using where

这是带有国籍的慢查询：

id	select_type	table	partitions	type	possible_keys	key	key_len	ref	rows	filtered	Extra
1	PRIMARY	reports	NULL	ALL	NULL	NULL	NULL	NULL	118923	100.00	Using temporary; Using filesort
7	DEPENDENT SUBQUERY	reports	NULL	ref	failed_date_index,nationality_index	nationality_index	5	func	720	0.12	Using where
6	DEPENDENT SUBQUERY	reports	NULL	ref	failed_date_index,nationality_index	nationality_index	5	func	720	0.12	Using where
5	DEPENDENT SUBQUERY	reports	NULL	ref	failed_date_index,nationality_index	nationality_index	5	func	720	0.12	Using where
4	DEPENDENT SUBQUERY	reports	NULL	ALL	pay_off_date_index,nationality_index	NULL	NULL	NULL	118923	1.00	Range checked for each record (index map: 0x8080)
3	DEPENDENT SUBQUERY	reports	NULL	ALL	pay_off_date_index,nationality_index	NULL	NULL	NULL	118923	10.00	Range checked for each record (index map: 0x8080)
2	DEPENDENT SUBQUERY	reports	NULL	ref	create_date_index,nationality_index	nationality_index	5	func	720	2.10	Using where

更新二

正如下面评论者所指出的，这不是 GROUP BY 的犹太教用法。我不确定它到底是如何工作的，但它确实需要每个子查询的 WHERE 子句中的 nationality = groupValue 表达式，这似乎导致子查询中的聚合在主查询中为 GROUP BY 分组（欢迎解释！）。

我不建议以这种方式使用子查询。就我而言，查询通常有效，但是一旦您分组的不同值的数量变高，它们似乎也会在性能方面下降（或进入无限循环？）。

相反，请使用答案中完美展示的 CASE。

Answer 1

您可以编写查询，使用条件聚合并避免 6 个相关子查询：

SELECT
    nationality AS groupValue,
    COUNT(CASE WHEN create_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' THEN request_id END) AS totalRequests,
    SUM(CASE WHEN pay_off_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' THEN effective_amount END) AS totalSum,
    COUNT(CASE WHEN pay_off_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND current_status = 5 THEN request_id END) AS totalPaid,
    COUNT(CASE WHEN failed_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND current_status = 3 THEN request_id END) AS totalRefused,
    COUNT(CASE WHEN failed_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND current_status = 4 THEN request_id END) AS totalRenounced,
    COUNT(CASE WHEN failed_date BETWEEN '2020-10-01 00:00:00' AND '2021-12-31 23:59:59' AND current_status = 2 THEN request_id END) AS totalInProgress 
FROM reports 
GROUP BY nationality;