我的例子是 MYSQL 版本是 5.6.34-日志
问题 摘要以下查询耗时40 秒，ORDER_ITEM 表
有758423条记录
和PAYMENT表
有177272条记录
和submission_entry表
有2165698条记录
作为整个表计数。
详情如下:
我有这个查询，引用[1]
我添加了SQL_NO_CACHE 用于测试重新测试时的重复测试
查询。我有优化的索引引用[2]，但不显着
改进。在此处查找表结构[3]
- 查找使用过的解释计划 [4]
[1]
SELECT SQL_NO_CACHE
`payment`.`id` AS id,
`order_item`.`order_id` AS order_id,
GROUP_CONCAT(DISTINCT (CASE WHEN submission_entry.text = '' OR submission_entry.text IS NULL
THEN ' '
ELSE submission_entry.text END) ORDER BY question.var DESC SEPARATOR 0x1D) AS buyer,
event.name AS event,
COUNT(DISTINCT CASE WHEN (`order_item`.status > 0 OR (
`order_item`.status != -1 AND `order_item`.status >= -2 AND `payment`.payment_type_id != 8 AND
payment.make_order_free = 1))
THEN `order_item`.id
ELSE NULL END) AS qty,
payment.currency AS `currency`,
(SELECT SUM(order_item.sub_total)
FROM order_item
WHERE payment_id =
payment.id) AS sub_total,
CASE WHEN payment.make_order_free = 1
THEN ROUND(payment.total + COALESCE(refunds_total, 0), 2)
ELSE ROUND(payment.total, 2) END AS 'total',
`payment_type`.`name` AS payment_type,
payment_status.name AS status,
`payment_status`.`id` AS status_id,
DATE_FORMAT(CONVERT_TZ(order_item.`created`, '+0:00', '-8:00'),
'%Y-%m-%d %H:%i') AS 'created',
`user`.`name` AS 'agent',
event.id AS event_id,
payment.checked,
DATE_FORMAT(CONVERT_TZ(payment.checked_date, '+0:00', '-8:00'),
'%Y-%m-%d %H:%i') AS checked_date,
DATE_FORMAT(CONVERT_TZ(`payment`.`complete_date`, '+0:00', '-8:00'),
'%Y-%m-%d %H:%i') AS `complete date`,
`payment`.`delivery_status` AS `delivered`
FROM `order_item`
INNER JOIN `payment`
ON payment.id = `order_item`.`payment_id` AND (payment.status > 0.0 OR payment.status = -3.0)
LEFT JOIN (SELECT
sum(`payment_refund`.total) AS `refunds_total`,
payment_refunds.payment_id AS `payment_id`
FROM payment
INNER JOIN `payment_refunds` ON payment_refunds.payment_id = payment.id
INNER JOIN `payment` AS `payment_refund`
ON `payment_refund`.id = `payment_refunds`.payment_id_refund
GROUP BY `payment_refunds`.payment_id) AS `refunds` ON `refunds`.payment_id = payment.id
# INNER JOIN event_date_product ON event_date_product.id = order_item.event_date_product_id
# INNER JOIN event_date ON event_date.id = event_date_product.event_date_id
INNER JOIN event ON event.id = order_item.event_id
INNER JOIN payment_status ON payment_status.id = payment.status
INNER JOIN payment_type ON payment_type.id = payment.payment_type_id
LEFT JOIN user ON user.id = payment.completed_by
LEFT JOIN submission_entry ON submission_entry.form_submission_id = `payment`.`form_submission_id`
LEFT JOIN question ON question.id = submission_entry.question_id AND question.var IN ('name', 'email')
WHERE 1 = '1' AND (order_item.status > 0.0 OR order_item.status = -2.0)
GROUP BY `order_item`.`order_id`
HAVING 1 = '1'
ORDER BY `order_item`.`order_id` DESC
LIMIT 10
[2]
CREATE INDEX order_id
ON order_item (order_id);
CREATE INDEX payment_id
ON order_item (payment_id);
CREATE INDEX status
ON order_item (status);
第二张表
CREATE INDEX payment_type_id
ON payment (payment_type_id);
CREATE INDEX status
ON payment (status);
[3]
CREATE TABLE order_item
(
id INT AUTO_INCREMENT
PRIMARY KEY,
order_id INT NOT NULL,
form_submission_id INT NULL,
status DOUBLE DEFAULT '0' NULL,
payment_id INT DEFAULT '0' NULL
);
第二张表
CREATE TABLE payment
(
id INT AUTO_INCREMENT,
payment_type_id INT NOT NULL,
status DOUBLE NOT NULL,
form_submission_id INT NOT NULL,
PRIMARY KEY (id, payment_type_id)
);
[4] 运行代码片段以查看 HTML 格式的 EXPLAIN 表
<!DOCTYPE html>
<html>
<head>
<title></title>
</head>
<body>
<table border="1" style="border-collapse:collapse">
<tr><th>id</th><th>select_type</th><th>table</th><th>type</th><th>possible_keys</th><th>key</th><th>key_len</th><th>ref</th><th>rows</th><th>Extra</th></tr>
<tr><td>1</td><td>PRIMARY</td><td>payment_status</td><td>range</td><td>PRIMARY</td><td>PRIMARY</td><td>8</td><td>NULL</td><td>4</td><td>Using where; Using temporary; Using filesort</td></tr>
<tr><td>1</td><td>PRIMARY</td><td>payment</td><td>ref</td><td>PRIMARY,payment_type_id,status</td><td>status</td><td>8</td><td>exp_live_18092017.payment_status.id</td><td>17357</td><td></td></tr>
<tr><td>1</td><td>PRIMARY</td><td>payment_type</td><td>eq_ref</td><td>PRIMARY</td><td>PRIMARY</td><td>4</td><td>exp_live_18092017.payment.payment_type_id</td><td>1</td><td></td></tr>
<tr><td>1</td><td>PRIMARY</td><td>user</td><td>eq_ref</td><td>PRIMARY</td><td>PRIMARY</td><td>4</td><td>exp_live_18092017.payment.completed_by</td><td>1</td><td></td></tr>
<tr><td>1</td><td>PRIMARY</td><td>submission_entry</td><td>ref</td><td>form_submission_id,idx_submission_entry_1</td><td>form_submission_id</td><td>4</td><td>exp_live_18092017.payment.form_submission_id</td><td>2</td><td></td></tr>
<tr><td>1</td><td>PRIMARY</td><td>question</td><td>eq_ref</td><td>PRIMARY,var</td><td>PRIMARY</td><td>4</td><td>exp_live_18092017.submission_entry.question_id</td><td>1</td><td>Using where</td></tr>
<tr><td>1</td><td>PRIMARY</td><td>order_item</td><td>ref</td><td>status,payment_id</td><td>payment_id</td><td>5</td><td>exp_live_18092017.payment.id</td><td>3</td><td>Using where</td></tr>
<tr><td>1</td><td>PRIMARY</td><td>event</td><td>eq_ref</td><td>PRIMARY</td><td>PRIMARY</td><td>4</td><td>exp_live_18092017.order_item.event_id</td><td>1</td><td></td></tr>
<tr><td>1</td><td>PRIMARY</td><td><derived3></td><td>ref</td><td>key0</td><td>key0</td><td>5</td><td>exp_live_18092017.payment.id</td><td>10</td><td>Using where</td></tr>
<tr><td>3</td><td>DERIVED</td><td>payment_refunds</td><td>index</td><td>payment_id,payment_id_refund</td><td>payment_id</td><td>4</td><td>NULL</td><td>1110</td><td></td></tr>
<tr><td>3</td><td>DERIVED</td><td>payment</td><td>ref</td><td>PRIMARY</td><td>PRIMARY</td><td>4</td><td>exp_live_18092017.payment_refunds.payment_id</td><td>1</td><td>Using index</td></tr>
<tr><td>3</td><td>DERIVED</td><td>payment_refund</td><td>ref</td><td>PRIMARY</td><td>PRIMARY</td><td>4</td><td>exp_live_18092017.payment_refunds.payment_id_refund</td><td>1</td><td></td></tr>
<tr><td>2</td><td>DEPENDENT SUBQUERY</td><td>order_item</td><td>ref</td><td>payment_id</td><td>payment_id</td><td>5</td><td>func</td><td>3</td><td></td></tr></table>
</body>
</html>预期结果
必须小于5秒而不是40秒
重要 更新
1)回复评论1:那两个表根本就没有外键
更新 1: 在本地上，原始查询需要40 秒 如果我删除仅以下内容，它将变为25 秒 节省15 秒
GROUP_CONCAT(DISTINCT (CASE WHEN submission_entry.text = '' OR submission_entry.text IS NULL
THEN ' '
ELSE submission_entry.text END) ORDER BY question.var DESC SEPARATOR 0x1D) AS buyer
如果我在 40 秒 左右的同一时间移除仅 没有保存!
COUNT(DISTINCT CASE WHEN (`order_item`.status > 0 OR (
`order_item`.status != -1 AND `order_item`.status >= -2 AND `payment`.payment_type_id != 8 AND
payment.make_order_free = 1))
THEN `order_item`.id
ELSE NULL END) AS qty,
如果我移除仅，它需要大约36 秒 节省4 秒
(SELECT SUM(order_item.sub_total)
FROM order_item
WHERE payment_id =
payment.id) AS sub_total,
CASE WHEN payment.make_order_free = 1
THEN ROUND(payment.total + COALESCE(refunds_total, 0), 2)
ELSE ROUND(payment.total, 2) END AS 'total',
最佳答案
删除HAVING 1=1；优化器可能不够聪明，无法忽略它。请提供EXPLAIN SELECT(不是在 html 中)以查看优化器正在做什么。
在这种情况下使用复合 PK 似乎是错误的:PRIMARY KEY (id, payment_type_id)。请证明这一点。
请解释status的含义或DOUBLE的必要性:status DOUBLE
要找出查询为何如此缓慢的原因需要一些努力。让我们从抛出规范化部分开始，例如日期和事件名称和货币。那就是将查询缩减到足以找到所需的行，但不是每行的详细信息。如果它仍然很慢，让我们调试它。如果它“快”了，然后再添加其他内容，一项一项地找出导致性能问题的原因。
只是id是每个表的PRIMARY KEY吗？还是有更多异常(exception)情况(如payment)？
为 question.var 指定一个值似乎是“错误的”，但随后使用 LEFT 来暗示它是可选的。请将所有 LEFT JOINs 更改为 INNER JOINs 除非我在这个问题上弄错了。
是否有任何表(可能是 submission_entry 和 event_date_product)是“多对多”映射表？如果是这样，请按照提示进行操作 here以获得一些性能提升。
当您回来时，请为每个表提供SHOW CREATE TABLE。
关于mysql - group by 和 group concat ，不使用主 pk 优化 mysql 查询，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/46510329/