优化MySQL查询以避免“使用临时”和“使用filesort”
我知道这里有一千个类似的问题,但没有一个像我一样处理复杂的查询(并且我的MySQL技巧并不重要。真正了解如何适应它们)优化MySQL查询以避免“使用临时”和“使用filesort”
这就是:
explain select
`ev`.`EventID` AS `EventID`
,`ev`.`EventName` AS `EventName`
,concat(`ev`.`EventDate`,' ',`ev`.`StartTime`) AS `EventDT`
,`ev`.`NumberTicketsAvailable` AS `TotalTickets`
,`ev`.`Soldout` AS `Soldout`
,count((case when (`ec`.`CartStatus` = 'InCart') then 1 else NULL end)) AS `InCartCount`
,count((case when (`ec`.`CartStatus` = 'InPayment') then 1 else NULL end)) AS `InPaymentCount`
,count((case when (`ec`.`CartStatus` = 'Paid') then 1 else NULL end)) AS `PaidCount`
,count((case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 604800 second) > now())) then 1 else NULL end)) AS `PaidOverWeek`
,count((case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 432000 second) > now())) then 1 else NULL end)) AS `PaidOverFiveDays`
,count((case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 86400 second) > now())) then 1 else NULL end)) AS `PaidOverDay`
,count((case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 43200 second) > now())) then 1 else NULL end)) AS `PaidOverHalfDay`
,count((case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 21600 second) > now())) then 1 else NULL end)) AS `PaidOverQuarterDay`
,count((case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 10800 second) > now())) then 1 else NULL end)) AS `PaidOverThreeHours`
,count((case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 3600 second) > now())) then 1 else NULL end)) AS `PaidOverHour`
from (`Events` `ev`
left join (`Events_EventCart_Rel` `eecr`
left join `EventCart` `ec`
on((`eecr`.`EventCartID` = `ec`.`EventCartID`)))
on((`ev`.`EventID` = `eecr`.`EventID`)))
where (`eecr`.`Active` = 1 AND `eecr`.`Deleted` = 'No')
group by
`ev`.`EventID`
,`ev`.`EventName`
,`ev`.`EventDate`
,`ev`.`StartTime`
,`ev`.`NumberTicketsAvailable`
,`ev`.`Soldout`;
这样做的结果是这样的:
+-id-+-select_type-+-table-+--type--+--------possible_keys--------+----key----+-key_len-+----------ref----------+--rows--+---------------------------Extra---------------------------+
| 1| SIMPLE | eecr | index | EventID,EventID_2,EventID_3 | EventID_3 | 10 | {null} | 17609 | Using where; Using index; Using temporary; Using filesort |
| 1| SIMPLE | ev | eq_ref | PRIMARY | PRIMARY | 4 | eecr.EventID | 1 | Using where |
| 1| SIMPLE | ec | eq_ref | PRIMARY | PRIMARY | 4 | eecr.EventCartID | 1 | |
+----+-------------+-------+--------+-----------------------------+-----------+---------+-----------------------+--------+-----------------------------------------------------------+
而表定义:
CREATE TABLE IF NOT EXISTS `Events` (
`EventID` int(10) unsigned NOT NULL AUTO_INCREMENT,
`EventName` varchar(150) NOT NULL,
`StartTime` char(8) NOT NULL DEFAULT '00:00:00',
`EndTime` char(8) NOT NULL DEFAULT '00:00:00',
`EventDate` varchar(20) NOT NULL,
`NumberTicketsAvailable` smallint(6) DEFAULT NULL,
`Soldout` enum('yes','no') DEFAULT 'no',
#...
PRIMARY KEY (`EventID`),
KEY `EndTime` (`EndTime`,`EventDate`),
KEY `StartTime` (`StartTime`,`EventDate`),
KEY `EventDate` (`EventDate`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1
CREATE TABLE IF NOT EXISTS `Events_EventCart_Rel` (
`ID` int(11) NOT NULL AUTO_INCREMENT,
`EventCartID` int(11) NOT NULL,
`EventID` int(11) NOT NULL,
`DateAdded` datetime NOT NULL,
`PersonID` int(11) NOT NULL,
`SeatTypeID` int(11) NOT NULL,
`MealChoiceID` int(11) NOT NULL,
`Active` tinyint(1) NOT NULL DEFAULT '1',
`Deleted` enum('Yes','No') NOT NULL DEFAULT 'No',
`ModifiedByAdmin` enum('Yes','No') NOT NULL DEFAULT 'No',
PRIMARY KEY (`ID`),
KEY `EventID` (`EventID`,`PersonID`),
KEY `EventCartID` (`EventCartID`),
KEY `EventID_2` (`EventID`),
KEY `EventID_3` (`EventID`,`EventCartID`,`Active`,`Deleted`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1
CREATE TABLE IF NOT EXISTS `EventCart` (
`EventCartID` int(11) NOT NULL AUTO_INCREMENT,
`RegistrantsID` int(11) NOT NULL DEFAULT '0',
`DateRecordCreated` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
`DateRecordModified` timestamp NOT NULL DEFAULT '0000-00-00 00:00:00',
`CartStatus` enum('InCart','InPayment','Paid') NOT NULL DEFAULT 'InCart',
`ModifiedByAdmin` enum('yes','no') NOT NULL DEFAULT 'no',
PRIMARY KEY (`EventCartID`),
KEY `rid` (`RegistrantsID`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1
为了抢占任何疑惑看待这些列 - 是的,有很多遗留类型的东西,我没有时间修复代码。
我发现至少在MySQL中,几乎所有使用GROUP BY
的查询都会调用一个临时表。这是你的巨大性能成本。尝试研究有它的使用分析器时间花在:
编辑:我修正了以下内容SET PROFILING
(不SET PROFILES
):
SET PROFILING = On;
SELECT ...the whole query you want to profile...
SHOW PROFILES;
SHOW PROFILE FOR QUERY 1;
详情请参阅http://dev.mysql.com/doc/refman/5.1/en/show-profiles.html。
没有太多的事情可以解决这个问题。它有时最好的性能的原因,以消除GROUP BY
和聚合函数:
select
`ev`.`EventID` AS `EventID`
,`ev`.`EventName` AS `EventName`
,concat(`ev`.`EventDate`,' ',`ev`.`StartTime`) AS `EventDT`
,`ev`.`NumberTicketsAvailable` AS `TotalTickets`
,`ev`.`Soldout` AS `Soldout`
,case when (`ec`.`CartStatus` = 'InCart') then 1 else 0 end AS `InCartCounter`
,case when (`ec`.`CartStatus` = 'InPayment') then 1 else 0 end AS `InPaymentCounter`
,case when (`ec`.`CartStatus` = 'Paid') then 1 else 0 end AS `PaidCounter`
,case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 604800 second) > now())) then 1 else 0 end AS `PaidOverWeekCounter`
,case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 432000 second) > now())) then 1 else 0 end AS `PaidOverFiveDaysCounter`
,case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 86400 second) > now())) then 1 else 0 end AS `PaidOverDayCounter`
,case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 43200 second) > now())) then 1 else 0 end AS `PaidOverHalfDayCounter`
,case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 21600 second) > now())) then 1 else 0 end AS `PaidOverQuarterDayCounter`
,case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 10800 second) > now())) then 1 else 0 end AS `PaidOverThreeHoursCounter`
,case when ((`ec`.`CartStatus` = 'Paid') and ((`ec`.`DateRecordModified` + interval 3600 second) > now())) then 1 else 0 end AS `PaidOverHourCounter`
from `Events` `ev`
inner join `Events_EventCart_Rel` `eecr`
on `ev`.`EventID` = `eecr`.`EventID`
inner join `EventCart` `ec`
on `eecr`.`EventCartID` = `ec`.`EventCartID`
where `eecr`.`Active` = 1 and `eecr`.`Deleted` = 'No'
在应用程序代码
然后,取所有行,并遍历它们,计算总计数,当您去。例如,在PHP:
$stmt = $pdo->query($sql);
$events = array();
$counters = array("InCartCounter", "InPaymentCounter", "PaidCounter",
"PaidOverWeekCounter", "PaidOverFiveDaysCounter", "PaidOverDayCounter",
"PaidOverHalfDayCounter", "PaidOverQuarterDayCounter",
"PaidOverThreeHoursCounter", "PaidOverHourCounter");
while ($row = $stmt->fetch())
{
if (!isset($events[$row["EventID"]])) {
$events[$row["EventID"]] = $row;
} else {
foreach ($counters as $key) {
$events[$row["EventID"]][$key] += $row[$key];
}
}
}
它看起来像一个大量的代码和麻烦做SQL 应该能够更有效地做一些事情,但在MySQL和GROUP BY
写更多的情况下,应用程序代码往往是更好。
PS:在示例SQL查询中,我将您的连接更改为内部连接。我不认为你需要外部连接。
不幸的是,它看起来像我们的MySQL服务器没有安装/启用Profiler(这是正确的版本,但设置PROFILER抛出一个错误。)当我们的系统管理员有时间打开它时,我将不得不重新访问。 – 2010-09-08 20:51:06
我的歉意,我的语法错了。它是'SET PROFILING = On'。我会在上面编辑我的答案以便正确。 – 2010-09-08 22:58:11
解释计划对我来说听起来很合理。有什么困难? – 2010-09-07 15:55:41
这需要约2.5秒。无论如何都不可怕,但我喜欢贪婪。这是针对“实时”报告的,我希望尽可能减少执行时间。 – 2010-09-07 16:03:21