驯服一个怪物MySQL查询

Question

因此，通过一些SO用户的帮助，我最终得到了一个逻辑上正确的MySQL查询，用于我正在处理的任务：检索反向按时间顺序排列的ID列表允许用户使用的新闻项目，某些类型的分组项目过滤到该组的单个代表。 （Phew！）

还有一个明显的问题，就是根据CakePHP的数据库调用调试打印输出，这个查询非常笨重且速度很慢，达到了145000毫秒，哎。

是否有一个明智的办法来驯服野兽像这样的，或者我应该承认，我贪多，我可以在这里咀嚼，寻找一个不太沉重的方法，将实现更多或更少类似的结果？所有建议表示赞赏

SELECT DISTINCT Uid.id, Uid.type 
    FROM (SELECT uids.id id, uids_uids.parent_id parent_id, uids.created date, 
       uids.type type 
      FROM uids 
      JOIN uids_uids ON uids_uids.uid_id = uids.id 
      JOIN aros_uids ON uids.id = aros_uids.uid_id 
      JOIN uids_uids ParentUids ON uids_uids.parent_id = ParentUids.uid_id 
      WHERE uids.type IN ('Document','Photo','Release','PreRelease', 
           'ArtworkResource','Event') 
      AND (uids.start_date IS NULL OR uids.start_date <= NOW()) 
      AND (uids.end_date IS NULL OR uids.end_date <= NOW()) 
      AND aros_uids.aro_id IN (3,2,86,1448) 
     ) Uid 
    JOIN (SELECT uids_uids.parent_id parent_id, MAX(uids.created) maxdate 
      FROM uids JOIN uids_uids 
      ON uids_uids.uid_id = uids.id 
      GROUP BY uids_uids.parent_id, uids.type) T2 
    ON Uid.parent_id = T2.parent_id AND Uid.date = T2.maxdate 
    ORDER BY Uid.date DESC 
    LIMIT 100 

ETA：

好了，在第一遍，我转身的子查询到意见，所以现在查询看起来像一个稍微更易于管理

SELECT DISTINCT Uid.id, Uid.type 
    FROM UidView Uid 
    JOIN UidView2 T2 
    ON Uid.parent_id = T2.parent_id AND Uid.date = T2.maxdate 
    WHERE Uid.aro_id IN (3,2,86,1448) 
    ORDER BY Uid.date DESC 
    LIMIT 100 

这一定帮助，将Cake的估计查询时间从六位数降低到2500位左右。绝对是一个好开始！

Answer 1

这里是我想尝试：

以每个派生的查询和独立运行的EXPLAIN针对每个。如评论所示，检查任何缺少索引并在需要时添加的行。发布您的EXPLAIN结果以获得任何帮助。所以

EXPLAIN SELECT uids.id id, uids_uids.parent_id parent_id, uids.created date, .... 
EXPLAIN SELECT uids_uids.parent_id parent_id, MAX(uids.created) maxdate .... 

如果添加索引不帮助或帮助不大，然后把每个子查询到一个临时表第一，并应用指标到它：这些表

CREATE TABLE temp_uid 
SELECT uids.id id, uids_uids.parent_id parent_id, uids.created date, 
      uids.type type 
     FROM uids 
     JOIN uids_uids ON uids_uids.uid_id = uids.id 
     JOIN aros_uids ON uids.id = aros_uids.uid_id 
     JOIN uids_uids ParentUids ON uids_uids.parent_id = ParentUids.uid_id 
     WHERE uids.type IN ('Document','Photo','Release','PreRelease', 
          'ArtworkResource','Event') 
     AND (uids.start_date IS NULL OR uids.start_date <= NOW()) 
     AND (uids.end_date IS NULL OR uids.end_date <= NOW()) 
     AND aros_uids.aro_id IN (3,2,86,1448); 

CREATE TABLE temp_t2 
SELECT uids_uids.parent_id parent_id, MAX(uids.created) maxdate 
     FROM uids JOIN uids_uids 
     ON uids_uids.uid_id = uids.id 
     GROUP BY uids_uids.parent_id, uids.type; 

而且JOIN：

SELECT DISTINCT Uid.id, Uid.type 
FROM temp_uid AS Uid 
JOIN temp_t2 AS T2 ON Uid.parent_id = T2.parent_id AND Uid.date = T2.maxdate 
ORDER BY Uid.date DESC 
LIMIT 100; 

正如我所提到的，您可能需要添加索引，并且可能会添加到临时表中的这些列中：

ALTER TABLE temp_uid ADD INDEX parentDateIdx (parent_id, Uid.date); 
ALTER TABLE temp_t2 ADD INDEX parentMaxDateIdx (parent_id, maxdate); 

如果您需要刷新临时表，只需截断它们并对它们执行INSERT INTO temp_uid...SELECT和INSERT INTO temp_t2...SELECT而不是CREATE...SELECT。存储过程对此非常有用。

顺便说一句，执行CREATE TABLE temp_t2...SELECT，我为每个临时表执行的方式可能无法创建最佳表结构，因此最好修改后续创建或从头开始自己完成。