“复制到 tmp 表”非常慢

Question

这是我的查询示例：

SELECT
    nickname, 
    CASE class_id
      WHEN 1 THEN 'Druid'
      WHEN 2 THEN 'Necromancer'
      WHEN 3 THEN 'Mage'
      WHEN 4 THEN 'Priest'
      WHEN 5 THEN 'Warrior'
      WHEN 6 THEN 'Stalker'
      WHEN 7 THEN 'Paladin'
      WHEN 8 THEN 'Psionic'
    END class_name,
    ROUND(AVG(level),2) level,
    ROUND(AVG(tabard_id),2) tabard,
    CASE rank_id
      WHEN 1 THEN 'Leader'
      WHEN 2 THEN 'Officer'
      WHEN 3 THEN 'Veteran'
      WHEN 4 THEN 'HonoryMember'
      WHEN 5 THEN 'OrdinaryMember'
      WHEN 6 THEN 'Alt'
      WHEN 7 THEN 'Apprentice'
      WHEN 8 THEN 'Penalty'
    END rank_name,
    ROUND(AVG(loyality),2) loyality,
    ROUND((MAX(authority)-MIN(authority))/AVG(tabard_id)) authority_effective,
    MAX(authority)-MIN(authority) authority_delta,
    MIN(authority) authority_begin,
    MAX(authority) authority_end
FROM users
    LEFT JOIN level_history ON level_history.users_id = users.id
    LEFT JOIN tabard_history ON tabard_history.users_id = users.id
    LEFT JOIN rank_history ON rank_history.users_id = users.id
    LEFT JOIN loyality_history ON loyality_history.users_id = users.id
    LEFT JOIN authority_history ON authority_history.users_id = users.id
    LEFT JOIN guilds_has_users ON guilds_has_users.users_id = users.id
    LEFT JOIN report ON report.id = authority_history.report_id
      AND report.id = level_history.report_id
      AND report.id = loyality_history.report_id
      AND report.id = rank_history.report_id
      AND report.id = tabard_history.report_id
WHERE report.date BETWEEN '2011-10-24 00:00:00' AND '2011-10-30 23:59:59'
  AND guilds_has_users.active = 1
GROUP BY users.id;

解释那个选择：

id  select_type   table               type    possible_keys                                            key                          key_len   ref                                           rows    Extra
1   SIMPLE        guilds_has_users    ref     fk_guilds_has_users_users1,active_IDX                    active_IDX                   1         const                                         139     Using where; Using temporary; Using filesort
1   SIMPLE        users               eq_ref  PRIMARY                                                  PRIMARY                      4         z92985_orlandino.guilds_has_users.users_id    1    
1   SIMPLE        level_history       ref     fk_level_history_users1,fk_level_history_report1,u...    fk_level_history_users1      4         z92985_orlandino.guilds_has_users.users_id    1       Using where
1   SIMPLE        report              eq_ref  PRIMARY,date_IDX,id_date_IDX                             PRIMARY                      4         z92985_orlandino.level_history.report_id      1       Using where
1   SIMPLE        tabard_history      ref     fk_tabard_history_users1,fk_tabard_history_report1...    fk_tabard_history_users1     4         z92985_orlandino.level_history.users_id       1       Using where
1   SIMPLE        rank_history        ref     fk_rank_history_users1,fk_rank_history_report1,use...    fk_rank_history_users1       4         z92985_orlandino.users.id                     1       Using where
1   SIMPLE        loyality_history    ref     fk_loyality_history_users1,fk_loyality_history_rep...    fk_loyality_history_users1   4         z92985_orlandino.rank_history.users_id        1       Using where
1   SIMPLE        authority_history   ref     fk_authority_history_users1,fk_authority_history_r...    fk_authority_history_users1  4         z92985_orlandino.level_history.users_id       1       Using where

该选择的分析告诉我：

(139 total, Query took 4.4918 sec)
Copying to tmp table 4.488318

以及一些关于 MySQL 变量的信息：

SHOW VARIABLES LIKE '%buffer%';

Variable_name              Value
bulk_insert_buffer_size    8388608
join_buffer_size           131072
key_buffer_size            12884901888
myisam_sort_buffer_size    8388608
net_buffer_length          16384
preload_buffer_size        32768
read_buffer_size           131072
read_rnd_buffer_size       25165824
sort_buffer_size           2097144
sql_buffer_result          OFF

为什么复制到 tmp 表这么慢？如何提高我的查询速度？

PS：我无法配置 MySQL，因为我的托管服务提供商不允许。

Answer 1

您可能需要尝试在会话中设置某些变量

• tmp_table_size
• max_heap_table_size

这些特定值对于您的数据库连接来说可能太小而无法有效地完成查询。这些可以设置如下：

• 要查看这些设置当前具有哪些值，请执行以下操作：
  • SHOW VARIABLES LIKE 'max_heap_table_size';
  • SHOW VARIABLES LIKE 'tmp_table_size';
• 要将 max_heap_table_size 设置为 64M，请执行以下操作：
  • SET max_heap_table_size = 1024 * 1024 * 64;
• 要将 tmp_table_size 设置为 32M，请执行以下操作：
  • SET tmp_table_size = 1024 * 1024 * 32;

请参阅有关临时表使用的MySQL 文档

如果您无法在自己的会话中设置这些值，请联系您的托管服务提供商以在 my.cnf 中动态设置它们。

试一试 ！！！

Answer 2

您是否可以将查询减少到仅生成输出绝对必要的那些表，或者将查询拆分为多个单独的查询以提取信息的不同部分？

您可能会发现，对数据运行三个单独的查询比运行一个庞大的查询更快——尤其是当您的数据库开始增长到数万行时。

我还从我自己的工作中注意到LEFT JOIN查询不一定是最有效的，所以只在绝对必要时才使用它们......

希望有帮助:)