MySQL上的0.07s查询在MariaDB上需要11.68s?

时间:2018-08-25 07:01:57

标签: mysql sql mariadb query-optimization explain

涉及不同的硬件(MySQL在我的笔记本电脑上,MariaDB在服务器上),但通常差异最多是2倍而不是166倍!

这些表在每个实例上包含相同的数据(_cache_card中的18,000行和card_legality中的157,000行)。

查询

SELECT * FROM _cache_card AS c 
WHERE c.id IN (SELECT card_id FROM card_legality WHERE format_id = 35);

说明

MariaDB:

+------+--------------+---------------+------+---------------------------------+-----------+---------+-------+-------+-------------------------------------------------+
| id   | select_type  | table         | type | possible_keys                   | key       | key_len | ref   | rows  | Extra                                           |
+------+--------------+---------------+------+---------------------------------+-----------+---------+-------+-------+-------------------------------------------------+
|    1 | PRIMARY      | <subquery2>   | ALL  | distinct_key                    | NULL      | NULL    | NULL  |  9414 |                                                 |
|    1 | PRIMARY      | c             | ALL  | NULL                            | NULL      | NULL    | NULL  | 18567 | Using where; Using join buffer (flat, BNL join) |
|    2 | MATERIALIZED | card_legality | ref  | format_id,idx_card_id_format_id | format_id | 4       | const |  9414 |                                                 |
+------+--------------+---------------+------+---------------------------------+-----------+---------+-------+-------+-------------------------------------------------+

MySQL:

+----+--------------+---------------+------------+--------+---------------------------------+------------+---------+------------+-------+----------+-------------+
| id | select_type  | table         | partitions | type   | possible_keys                   | key        | key_len | ref        | rows  | filtered | Extra       |
+----+--------------+---------------+------------+--------+---------------------------------+------------+---------+------------+-------+----------+-------------+
|  1 | SIMPLE       | c             | NULL       | ALL    | NULL                            | NULL       | NULL    | NULL       | 18055 |   100.00 | Using where |
|  1 | SIMPLE       | <subquery2>   | NULL       | eq_ref | <auto_key>                      | <auto_key> | 4       | cards.c.id |     1 |   100.00 | NULL        |
|  2 | MATERIALIZED | card_legality | NULL       | ref    | format_id,idx_card_id_format_id | format_id  | 4       | const      | 37828 |   100.00 | NULL        |
+----+--------------+---------------+------------+--------+---------------------------------+------------+---------+------------+-------+----------+-------------+

创建表

两者:

CREATE TABLE `card_legality` (
  `id` int(11) NOT NULL AUTO_INCREMENT,
  `card_id` int(11) NOT NULL,
  `format_id` int(11) NOT NULL,
  `legality` varchar(190) COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  PRIMARY KEY (`id`),
  KEY `format_id` (`format_id`),
  KEY `idx_card_id_format_id` (`card_id`,`format_id`,`legality`),
  CONSTRAINT `card_legality_ibfk_1` FOREIGN KEY (`card_id`) REFERENCES `card` (`id`),
  CONSTRAINT `card_legality_ibfk_2` FOREIGN KEY (`format_id`) REFERENCES `format` (`id`)
) ENGINE=InnoDB AUTO_INCREMENT=1190863 DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci

(这里的输出是字符对字符相同。)

MariaDB:

CREATE TABLE `_cache_card` (
  `id` int(11) NOT NULL DEFAULT 0,
  `layout` varchar(190) CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci NOT NULL,
  `face_id` int(11) NOT NULL DEFAULT 0,
  `name` longtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `mana_cost` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `cmc` double DEFAULT NULL,
  `power` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `toughness` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `loyalty` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `type` longtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `text` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `search_text` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `image_name` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `hand` mediumtext DEFAULT NULL,
  `life` mediumtext DEFAULT NULL,
  `starter` mediumtext DEFAULT NULL,
  `position` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `name_ascii` longtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `card_id` mediumtext DEFAULT NULL,
  `names` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `legalities` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  `pd_legal` int(1) DEFAULT NULL,
  `bugs` mediumtext CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci DEFAULT NULL,
  KEY `idx_name_name` (`name`(142))
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4

MySQL:

CREATE TABLE `_cache_card` (
  `id` int(11) NOT NULL DEFAULT '0',
  `layout` varchar(190) COLLATE utf8mb4_unicode_ci NOT NULL,
  `face_id` int(11) NOT NULL DEFAULT '0',
  `name` longtext COLLATE utf8mb4_unicode_ci,
  `mana_cost` mediumtext COLLATE utf8mb4_unicode_ci,
  `cmc` double DEFAULT NULL,
  `power` mediumtext COLLATE utf8mb4_unicode_ci,
  `toughness` mediumtext COLLATE utf8mb4_unicode_ci,
  `loyalty` mediumtext COLLATE utf8mb4_unicode_ci,
  `type` longtext COLLATE utf8mb4_unicode_ci,
  `text` mediumtext COLLATE utf8mb4_unicode_ci,
  `search_text` mediumtext COLLATE utf8mb4_unicode_ci,
  `image_name` mediumtext COLLATE utf8mb4_unicode_ci,
  `hand` mediumtext CHARACTER SET utf8mb4,
  `life` mediumtext CHARACTER SET utf8mb4,
  `starter` mediumtext CHARACTER SET utf8mb4,
  `position` mediumtext COLLATE utf8mb4_unicode_ci,
  `name_ascii` longtext COLLATE utf8mb4_unicode_ci,
  `card_id` mediumtext CHARACTER SET utf8mb4,
  `names` mediumtext COLLATE utf8mb4_unicode_ci,
  `legalities` mediumtext COLLATE utf8mb4_unicode_ci,
  `pd_legal` int(1) DEFAULT NULL,
  `bugs` mediumtext COLLATE utf8mb4_unicode_ci,
  KEY `idx_name_name` (`name`(142))
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci

其他详细信息

MariaDB:

version=10.2.16-MariaDB
optimizer_switch=index_merge=on,index_merge_union=on,index_merge_sort_union=on,index_merge_intersection=on,index_merge_sort_intersection=off,engine_condition_pushdown=off,index_condition_pushdown=on,derived_merge=on,derived_with_keys=on,firstmatch=on,loosescan=on,materialization=on,in_to_exists=on,semijoin=on,partial_match_rowid_merge=on,partial_match_table_scan=on,subquery_cache=on,mrr=off,mrr_cost_based=off,mrr_sort_keys=off,outer_join_with_cache=on,semijoin_with_cache=on,join_cache_incremental=on,join_cache_hashed=on,join_cache_bka=on,optimize_join_buffer_size=off,table_elimination=on,extended_keys=on,exists_to_in=on,orderby_uses_equalities=on,condition_pushdown_for_derived=on

MySQL:

version=5.7.17
optimizer_switch=index_merge=on,index_merge_union=on,index_merge_sort_union=on,index_merge_intersection=on,engine_condition_pushdown=on,index_condition_pushdown=on,mrr=on,mrr_cost_based=on,block_nested_loop=on,batched_key_access=off,materialization=on,semijoin=on,loosescan=on,firstmatch=on,duplicateweedout=on,subquery_materialization_cost_based=on,use_index_extensions=on,condition_fanout_filter=on,derived_merge=on

在card_legality(format_id,card_id)上添加索引

这使事情变得更慢。

 CREATE INDEX idx_format_id_card_id ON card_legality(format_id,card_id); 

现在超过15秒。

EXPLAIN(在MariaDB上)说:

+------+--------------+---------------+------+---------------------------------------------+-----------------------+---------+-------+-------+-------------------------------------------------+
| id   | select_type  | table         | type | possible_keys                               | key                   | key_len | ref   | rows  | Extra                                           |
+------+--------------+---------------+------+---------------------------------------------+-----------------------+---------+-------+-------+-------------------------------------------------+
|    1 | PRIMARY      | <subquery2>   | ALL  | distinct_key                                | NULL                  | NULL    | NULL  | 16942 |                                                 |
|    1 | PRIMARY      | c             | ALL  | NULL                                        | NULL                  | NULL    | NULL  | 17653 | Using where; Using join buffer (flat, BNL join) |
|    2 | MATERIALIZED | card_legality | ref  | idx_card_id_format_id,idx_format_id_card_id | idx_format_id_card_id | 4       | const | 16942 | Using index                                     |
+------+--------------+---------------+------+---------------------------------------------+-----------------------+---------+-------+-------+-------------------------------------------------+

4 个答案:

答案 0 :(得分:1)

我要添加索引:

myuser

DBFiddle Demo


$ mysql -u myuser -p <password on prompt>

答案 1 :(得分:1)

从优化的角度来看,

IN可能很棘手。我想知道这是否可以在两个系统上更好地工作:

SELECT c.* 
FROM _cache_card  c 
WHERE EXISTS (SELECT 1
              FROM card_legality cl
              WHERE cl.card_id = c.id AND format_id = 35
             );

答案 2 :(得分:1)

目前还不清楚为什么MariaDB当前使用这种次优执行计划。可能是在假设您的数据分配有问题(尽管我不确定在哪种情况下这将是最佳计划)。使用optimize table card_legality, _cache_card;修复统计信息可能会有所帮助。

如果不是,并且由于我们在注释中确定(card_id,format_id)是唯一的,那么我将尝试添加以下索引

CREATE UNIQUE INDEX uidx_card_legality ON card_legality(format_id, card_id)

并使用

SELECT c.* 
FROM _cache_card AS c
JOIN card_legality l FORCE INDEX (uidx_card_legality)
ON l.card_id = c.id AND l.format_id = 35;

这基本上是MySQL当前执行查询的方式(在动态创建该索引的同时),尽管您似乎已经用Lukasz的答案尝试了该索引,但是它没有用。

您应该删除force index(这是绝对要确保MariaDB没有回旋余地来做其他事情),并检查MariaDB / MySQL是否仍在使用它。还要测试其他format_id值会发生什么,因为35可能是一个异常值(例如,您可能只有少数几种格式的条目),为此优化执行计划可能会减慢查询所有其他值。当然,请确保您正在比较相似的结果集,就像MariaDB的格式35的10k条目,而MySQL没有,则不公平。

答案 3 :(得分:1)

之所以出现差异,是因为MySQL 5.6和MariaDB 10.0出现了-他们分别开发了几个改进的优化程序。您遇到了一个涉及IN的构造,其中一个进行了重大改进,而另一个未(可能尚未)进行了改进。

只要IN ( SELECT ... )可行,就避免使用JOIN

EXISTS( SELECT 1 ... )是另一个可以尝试的构造。

索引:

PRIMARY KEY on every table !
card_legality:  INDEX(format_id, card_id) -- in this order
_cache_card:  (id)  -- This seems like a serious omission !

某些会影响性能的事情:在较小的*TEXT就足够时使用VARCHAR

计时时,运行查询两次。第一个将内容复制到RAM(buffer_pool);第二个是比较现实的。

多少内存?每个innodb_buffer_pool_size的值是什么?