mysql - 如何重写SQL查询以在MariaDB和MySQL上实现相同的性能

Question

如今，您经常读到“ MariaDB是MySQL的替代品”。但实际上，有时会存在巨大差异。例如。这个：

有一个使用MariaDB查询的应用程序运行良好。现在必须使用MySQL，而且令人惊讶的是性能不足。

如何在MySQL上达到相同的性能？


我是否缺少任何设置？使用mariadb，它可以快速运行。
我应该修改查询吗？
那么最好的查询是什么？


该查询确定当前值和一年前的值，但是每天没有值，因此将搜索距一年前最近的日期。

这是查询。下面是创建一些测试数据的代码。

SELECT c.id, c.nname as n, a.wert as x,
(SELECT z.wert FROM tbl_history z
 WHERE z.indize_id = a.indize_id
 AND z.hdate > DATE_SUB(a.hdate, INTERVAL 1 YEAR)
 ORDER BY z.hdate LIMIT 1
) as y
FROM tbl_history a
LEFT JOIN tbl_indize c ON a.indize_id = c.id
WHERE a.hdate = 
(SELECT b.hdate FROM tbl_history b
 WHERE b.indize_id = a.indize_id ORDER BY b.hdate DESC LIMIT 1)
AND a.indize_id = 1;

执行时间


MariaDB：0.01秒
的MySQL：12.23秒


这是一些虚拟数据

CREATE TABLE tbl_indize (
  id int(11) NOT NULL AUTO_INCREMENT,
  nname varchar(145) NOT NULL,
  PRIMARY KEY (id)
);

CREATE TABLE tbl_history (
  id int(11) NOT NULL AUTO_INCREMENT,
  indize_id int(11) NOT NULL,
  hdate date NOT NULL,
  wert decimal(18,6) DEFAULT NULL,
  PRIMARY KEY (id),
  UNIQUE KEY idx_id_date (indize_id,hdate)
);

INSERT INTO tbl_indize (nname) values ('test');

INSERT INTO tbl_history (indize_id, hdate, wert) values (1, '1970-01-01', RAND());
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 4096 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 2048 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 1024 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 512 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 256 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 128 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 64 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 32 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 16 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 8 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 4 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 2 DAY), SQRT(wert * RAND()) FROM tbl_history;
INSERT INTO tbl_history (indize_id, hdate, wert) SELECT indize_id, date_add(hdate, INTERVAL 1 DAY), SQRT(wert * RAND()) FROM tbl_history;
DELETE FROM tbl_history WHERE DAYOFWEEK(hdate) = 1;
DELETE FROM tbl_history WHERE DAYOFWEEK(hdate) = 2;

这是解释

玛丽亚数据库

+------+--------------------+-------+-------+---------------+-------------+---------+---------------+------+------------------------------------+
| id   | select_type        | table | type  | possible_keys | key         | key_len | ref           | rows | Extra                              |
+------+--------------------+-------+-------+---------------+-------------+---------+---------------+------+------------------------------------+
|    1 | PRIMARY            | a     | ALL   | idx_id_date   | NULL        | NULL    | NULL          | 5877 | Using where                        |
|    1 | PRIMARY            | c     | const | PRIMARY       | PRIMARY     | 4       | const         |    1 |                                    |
|    3 | DEPENDENT SUBQUERY | b     | ref   | idx_id_date   | idx_id_date | 4       | t.a.indize_id | 2938 | Using where; Using index           |
|    2 | DEPENDENT SUBQUERY | z     | ref   | idx_id_date   | idx_id_date | 4       | t.a.indize_id | 2938 | Using index condition; Using where |
+------+--------------------+-------+-------+---------------+-------------+---------+---------------+------+------------------------------------+

的MySQL

+----+--------------------+-------+------------+-------+---------------+-------------+---------+---------------+------+----------+------------------------------------------+
| id | select_type        | table | partitions | type  | possible_keys | key         | key_len | ref           | rows | filtered | Extra                                    |
+----+--------------------+-------+------------+-------+---------------+-------------+---------+---------------+------+----------+------------------------------------------+
|  1 | PRIMARY            | a     | NULL       | ALL   | idx_id_date   | NULL        | NULL    | NULL          | 5877 |    99.57 | Using where                              |
|  1 | PRIMARY            | c     | NULL       | const | PRIMARY       | PRIMARY     | 4       | const         |    1 |   100.00 | NULL                                     |
|  3 | DEPENDENT SUBQUERY | b     | NULL       | ref   | idx_id_date   | idx_id_date | 4       | t.a.indize_id |    1 |   100.00 | Using where; Using index; Using filesort |
|  2 | DEPENDENT SUBQUERY | z     | NULL       | ref   | idx_id_date   | idx_id_date | 4       | t.a.indize_id |    1 |    33.33 | Using index condition; Using filesort    |
+----+--------------------+-------+------------+-------+---------------+-------------+---------+---------------+------+----------+------------------------------------------+

Answer 1

玩了之后，我发现了一个改进：

SELECT c.id, c.nname as n, a.wert as x,
(SELECT z.wert FROM tbl_history z
 WHERE z.indize_id = a.indize_id
 AND z.hdate > DATE_SUB(a.hdate, INTERVAL 1 YEAR)
 ORDER BY z.hdate LIMIT 1
) as y
FROM tbl_history a
LEFT JOIN tbl_indize c ON a.indize_id = c.id
LEFT JOIN (SELECT MAX(hdate) as d, indize_id FROM tbl_history GROUP BY indize_id) x
ON x.indize_id = a.indize_id
WHERE a.hdate = x.d
AND a.indize_id = 1;

现在两个数据库的速度都可以。


MariaDB：0.00秒
MySQL：0.00秒


好吧-非常兼容^^

基本食谱

将where子句中的子选择替换为join-selects。