我们有以下 mysql 表(简化为直接进入重点)
CREATE TABLE `MONTH_RAW_EVENTS` (
`idEvent` int(11) unsigned NOT NULL,
`city` varchar(45) NOT NULL,
`country` varchar(45) NOT NULL,
`ts` datetime NOT NULL,
`idClient` varchar(45) NOT NULL,
`event_category` varchar(45) NOT NULL,
... bunch of other fields
PRIMARY KEY (`idEvent`),
KEY `idx_city` (`city`),
KEY `idx_country` (`country`),
KEY `idClient` (`idClient`),
) ENGINE=InnoDB;
CREATE TABLE `compilation_table` (
`idClient` int(11) unsigned DEFAULT NULL,
`city` varchar(200) DEFAULT NULL,
`month` int(2) DEFAULT NULL,
`year` int(4) DEFAULT NULL,
`events_profile` int(10) unsigned NOT NULL DEFAULT '0',
`events_others` int(10) unsigned NOT NULL DEFAULT '0',
`events_total` int(10) unsigned NOT NULL DEFAULT '0',
KEY `idx_month` (`month`),
KEY `idx_year` (`year`),
KEY `idx_idClient` (`idClient`),
KEY `idx_city` (`city`)
) ENGINE=InnoDB;
MONTH_RAW_EVENTS
包含近 2000 万行用户在网站中执行的操作,大小接近 4GBcompilation_table
每个月都有一个客户/城市摘要,我们用它来实时显示网站上的统计数据我们每月处理一次统计信息(从第一个表到第二个),并且我们正在尝试优化执行此类操作的查询(因为到目前为止,我们正在处理需要很长时间的 PHP 中的所有内容)
这是我们提出的查询,它在使用小数据子集时似乎可以完成工作,
处理全套数据需要6个多小时的问题
INSERT INTO compilation_table (idClient,city,month,year,events_profile,events_others)
SELECT IFNULL(OTHERS.idClient,AP.idClient) as idClient,
IF(IFNULL(OTHERS.city,AP.city)='','Others',IFNULL(OTHERS.city,AP.city)) as city,
01,2014,
IFNULL(AP.cnt,0) as events_profile,
IFNULL(OTHERS.cnt,0) as events_others
FROM
(
SELECT idClient,CONCAT(city,', ',country) as city,count(*) as cnt
FROM `MONTH_RAW_EVENTS` WHERE `ts`>'2014-01-01 00:00:00' AND `ts`<='2014-01-31 23:59:59'
AND `event_category`!='CLIENT PROFILE'
GROUP BY idClient,city
) as OTHERS
LEFT JOIN
(
SELECT idClient,CONCAT(city,', ',country) as city,count(*) as cnt
FROM `MONTH_RAW_EVENTS` WHERE `ts`>'2014-01-01 00:00:00' AND `ts`<='2014-01-31 23:59:59'
AND `event_category`='CLIENT PROFILE'
GROUP BY idClient,city
) as CLIPROFILE
ON CLIPROFILE.city=OTHERS.city and CLIPROFILE.idClient=OTHERS.idClient
UNION
SELECT IFNULL(OTHERS.idClient,CLIPROFILE.idClient) as idClient,
IF(IFNULL(OTHERS.city,CLIPROFILE.city)='','Others',IFNULL(OTHERS.city,CLIPROFILE.city)) as city,
01,2014,
IFNULL(CLIPROFILE.cnt,0) as events_profile,
IFNULL(OTHERS.cnt,0) as events_others
FROM
(
SELECT idClient,CONCAT(city,', ',country) as city,count(*) as cnt
FROM `MONTH_RAW_EVENTS` WHERE `ts`>'2014-01-01 00:00:00' AND `ts`<='2014-01-31 23:59:59'
AND `event_category`!='CLIENT PROFILE'
GROUP BY idClient,city
) as OTHERS
RIGHT JOIN
(
SELECT idClient,CONCAT(city,', ',country) as city,count(*) as cnt
FROM `MONTH_RAW_EVENTS` WHERE `ts`>'2014-01-01 00:00:00' AND `ts`<='2014-01-31 23:59:59'
AND `event_category`='CLIENT PROFILE'
GROUP BY idClient,city
) as CLIPROFILE
ON CLIPROFILE.city=OTHERS.city and CLIPROFILE.idClient=OTHERS.idClient
我们要做的是在 Mysql 中进行 FULL Outer Join,因此查询的基本架构如下: the one proposed here
我们如何优化查询?我们一直在尝试不同的索引,左右切换,但 8 小时后仍然没有完成运行,
MySQL服务器是一台Percona MySQL 5.5专用机,2cpu,2GB内存,SSD盘,
我们使用 Percona 工具优化了此类服务器的配置,
任何帮助将非常感激,
谢谢
最佳答案
你正在做一个 UNION 导致 DISTINCT 处理。
通常最好将完全连接重写为左连接加上右连接的不匹配行(如果它是正确的 1:n 连接)
OTHERS LEFT JOIN CLIPROFILE
ON CLIPROFILE.city=OTHERS.city and CLIPROFILE.idClient=OTHERS.idClient
union all
OTHERS RIGHT JOIN CLIPROFILE
ON CLIPROFILE.city=OTHERS.city and CLIPROFILE.idClient=OTHERS.idClient
WHERE OTHERS.idClient IS NULL
此外,您可能会在加入临时表之前在临时表中实现派生表的结果,因此计算只完成一次(我不知道 MySQL 的优化器是否足够智能以自动执行此操作)。
此外,将城市/国家作为单独的列进行分组和加入并在外部步骤中将 CONCAT(city,', ',country) 作为城市可能会更有效。
关于mysql - 针对海量数据优化MySQL Full外连接,我们在Stack Overflow上找到一个类似的问题:https://stackoverflow.com/questions/22284887/