Question

我的游戏服务器上有一个巨大的瓶颈，用于以下查询，用于存储当前的排行榜。

我目前只通过cron每隔5分钟调用一次此查询，但是我希望能够优化它以便每分钟或在需要时调用它。

查询耗时30秒，目前只有约2000个用户和7000个游戏（存储在Games和TopPlayerScores中）。我担心它会变得更糟！请帮帮我溢出 - 克诺比！你唯一的希望！

SET @rank=0;
INSERT INTO Board (TopScorePKID, GamePKID, UserPKID, UniquePlayerID, PlayerName, TopPlayerScore, Position, Date)
(SELECT bad.ID AS TopScorePKID, bad.GamePKID, bad.UserPKID, bad.UniquePlayerID, bad.PlayerName, bad.TopPlayerScore, @rank:=@rank+1 AS Position, bad.Date
FROM (
    SELECT g.GamePKID, g.TopPlayerScore, l.ID,  l.UserPKID, u.UniquePlayerID, u.PlayerName, (l.Date) AS Date
    FROM Games g, TopPlayerScores l, UserDetails u
    WHERE l.GamePKID = g.GamePKID
    AND u.UserPKID = l.UserPKID
    AND u.SECRET_DETAIL = 0 
    AND g.TopPlayerScore >= (SELECT DISTINCT k.TopPlayerScore AS Highest 
        FROM Games k, TopPlayerScores t 
        WHERE t.UserPKID = l.UserPKID
        AND k.GamePKID = t.GamePKID
        ORDER BY k.TopPlayerScore DESC
        LIMIT 1) 
    GROUP BY l.UserPKID
    ORDER BY g.TopPlayerScore DESC, Date ASC) 
AS bad);

请有人帮忙！我应该把它分解成观点吗？或使用Inner Join关键字？什么是最好的方法？

非常感谢甚至看着这个烂摊子：D！

更新1.0： EXPLAIN EXTENDED结果：

id  select_type table   type    possible_keys   key key_len ref rows    filtered    Extra
1   PRIMARY   ALL NULL    NULL    NULL    NULL    1521    100.00  
2   DERIVED l   ALL NULL    NULL    NULL    NULL    6923    100.00  Using temporary; Using filesort
2   DERIVED u   eq_ref  PRIMARY PRIMARY 4   DBNAME.l.UserPKID   1   100.00  Using where
2   DERIVED k   eq_ref  PRIMARY PRIMARY 4   DBNAME.l.GamePKID   1   100.00  Using where
3   DEPENDENT SUBQUERY  t   ALL NULL    NULL    NULL    NULL    6923    100.00  Using where; Using temporary; Using filesort
3   DEPENDENT SUBQUERY  g   eq_ref  PRIMARY PRIMARY 4   DBNAME.t.GamePKID   1   100.00  Using where

更新2.0： 用于查询表的有限模式

使用游戏存储游戏分数和有关游戏的其他信息

`Games` (
  `GamePKID` int(11) NOT NULL AUTO_INCREMENT,
  `TopPlayerScore` int(11) NOT NULL,
  `OTHER_MISC_STUFF_REMOVED` int(11) NOT NULL
  PRIMARY KEY (`GamePKID`)
)

使用以下内容将用户链接到游戏并存储时间/日期

`TopPlayerScores` (
  `ID` int(11) NOT NULL AUTO_INCREMENT,
  `UserPKID` int(11) NOT NULL,
  `GamePKID` int(11) NOT NULL,
  `Date` datetime NOT NULL,
  PRIMARY KEY (`ID`)
)

用于存储每个独特的播放器

`UserDetails` (
  `UserPKID` int(11) NOT NULL AUTO_INCREMENT,
  `UniquePlayerID` char(40) NOT NULL,
  `PlayerName` char(96) NOT NULL,
  `SECRET_DETAIL` tinyint(1) NOT NULL DEFAULT '0',
  `isPlayer` tinyint(4) DEFAULT NULL,
  PRIMARY KEY (`UserPKID`)
)

Answer 1

我要注意的第一件事，虽然这不会提高性能，但是你使用的JOIN语法在20多年前被ANSI 92 expcict join语法所取代，它完全是主题，但是{{3切换到更新语法的一些非常好的理由。

要注意的第二件事是您的结果将是非确定性的。您正在选择未包含在聚合或组中的列。虽然MySQL允许这样做，但您没有像MySQL那样使用该功能。 Aaron Bertrand explains州：

MySQL扩展了GROUP BY的使用，以便选择列表可以引用未在GROUP BY子句中命名的非聚合列。这意味着前面的查询在MySQL中是合法的。您可以使用此功能通过避免不必要的列排序来获得更好的性能分组。但是，这主要适用于每个中的所有值 GROUP BY中未命名的非聚合列对于每个列都是相同的组。服务器可以自由选择每个组中的任何值，所以除非它们相同，否则所选择的值是不确定的。

但是，您所包含的某些列（g.GamePKID，g.TopPlayerScore，l.ID，l.Date）不满足条件是相同的因此，如上所述，每个组，MySQL可以自由选择它喜欢的任何值，即使你有ORDER BY g.TopPlayerScore DESC, Date ASC，这也不会影响MySQL选择的每个组的单行。

第三，MySQL docs这可能会影响性能。如果您可以将这些更改为JOIN，您应该会看到性能提升。

考虑到这一切，我会重写您的查询：

SET @rank=0;
INSERT INTO Board (TopScorePKID, GamePKID, UserPKID, UniquePlayerID, PlayerName, TopPlayerScore, Position, Date)
SELECT  bad.ID AS TopScorePKID, 
        bad.GamePKID, 
        bad.UserPKID, 
        bad.UniquePlayerID, 
        bad.PlayerName, 
        bad.TopPlayerScore, 
        @rank:=@rank+1 AS Position, 
        bad.Date
FROM    (   SELECT  g.GamePKID, 
                    g.TopPlayerScore, 
                    l.ID,  
                    l.UserPKID, 
                    u.UniquePlayerID, 
                    u.PlayerName, 
                    l.Date
            FROM    Games g
                    INNER JOIN TopPlayerScores l
                        ON l.GamePKID = g.GamePKID
                    INNER JOIN UserDetails u
                        ON u.UserPKID = l.UserPKID
                    INNER JOIN
                    (   SELECT  TopPlayerScores.UserPKID, MAX(games.TopPlayerScore) AS MaxPlayerScore
                        FROM    TopPlayerScores
                                INNER JOIN Games
                                    ON Games.GamePKID = TopPlayerScores.GamePKID
                        GROUP BY TopPlayerScores.UserPKID
                    ) MaxScore
                        ON MaxScore.UserPKID = l.UserPKID
                        AND MaxScore.MaxPlayerScore = g.TopPlayerScore
            WHERE   u.SECRET_DETAIL = 0 
        ) AS bad
ORDER BY bad.TopPlayerScore DESC, bad.Date ASC;

<强> MySQL has limitations with correlated subqueries

子查询MaxScore应该具有将结果限制为每个玩家一行（只有他们的最高得分）的效果，尽管可能需要额外的逻辑来处理关系（例如，玩家在更多场景中具有相同的得分）比一场比赛）。在不知道确切要求的情况下，我无法纠正这一点。

修改

为了消除玩家在2个或更多游戏中获得相同最高分的重复项，并使其真正具有确定性，您需要添加更多子查询：

SET @rank=0; SELECT bad.ID AS TopScorePKID, bad.GamePKID, bad.UserPKID, bad.UniquePlayerID, bad.PlayerName, bad.TopPlayerScore, @rank:=@rank+1 AS Position, bad.Date FROM ( SELECT Games.GamePKID, Games.TopPlayerScore, TopPlayerScores.ID, TopPlayerScores.UserPKID, UserDetails.UniquePlayerID, UserDetails.PlayerName, TopPlayerScores.Date FROM Games INNER JOIN TopPlayerScores ON TopPlayerScores.GamePKID = Games.GamePKID INNER JOIN UserDetails ON UserDetails.UserPKID = TopPlayerScores.UserPKID INNER JOIN ( SELECT TopPlayerScores.UserPKID, MAX(games.TopPlayerScore) AS TopPlayerScore FROM TopPlayerScores INNER JOIN Games ON Games.GamePKID = TopPlayerScores.GamePKID GROUP BY TopPlayerScores.UserPKID ) MaxScore ON MaxScore.UserPKID = TopPlayerScores.UserPKID AND MaxScore.TopPlayerScore = Games.TopPlayerScore INNER JOIN ( SELECT TopPlayerScores.UserPKID, games.TopPlayerScore, MAX(Date) AS Date FROM TopPlayerScores INNER JOIN Games ON Games.GamePKID = TopPlayerScores.GamePKID GROUP BY TopPlayerScores.UserPKID, games.TopPlayerScore ) MaxScoreDate ON MaxScoreDate.UserPKID = TopPlayerScores.UserPKID AND MaxScoreDate.TopPlayerScore = Games.TopPlayerScore AND MaxScoreDate.Date = TopPlayerScores.Date WHERE UserDetails.SECRET_DETAIL = 0 ) AS bad ORDER BY bad.TopPlayerScore DESC, bad.Date ASC;

<强> Example on SQL Fiddle

N.B。如果/当MySQL引入诸如ROW_NUMBER()之类的分析函数，或者如果你切换到已经支持它们的DBMS，这个查询将变得更加简单，所以为了防止这些事情发生，这里有一个使用ROW_NUMBER的解决方案（）`的

SELECT bad.ID AS TopScorePKID, bad.GamePKID, bad.UserPKID, bad.UniquePlayerID, bad.PlayerName, bad.TopPlayerScore, ROW_NUMBER() OVER(ORDER BY TopPlayerScore DESC) AS Position, bad.Date FROM ( SELECT Games.GamePKID, Games.TopPlayerScore, TopPlayerScores.ID, TopPlayerScores.UserPKID, UserDetails.UniquePlayerID, UserDetails.PlayerName, TopPlayerScores.Date, ROW_NUMBER(PARTITION BY UserDetails.UserPKID ORDER BY Games.TopPlayerScore DESC, TopPlayerScores.Date DESC) AS RN FROM Games INNER JOIN TopPlayerScores ON TopPlayerScores.GamePKID = Games.GamePKID INNER JOIN UserDetails ON UserDetails.UserPKID = TopPlayerScores.UserPKID WHERE UserDetails.SECRET_DETAIL = 0 ) AS bad WHERE bad.RN = 1 ORDER BY bad.TopPlayerScore DESC, bad.Date ASC;

<强> Example on SQL Fiddle

如何优化MySQL游戏排行榜 - 大规模子查询问题

1 个答案: