最近在做新闻评论数据的网络爬虫，需要将爬去到的评论数据存储在MySQL数据库中，提供评论的查询和展示功能。后来随着数据量的上升，MySQL查询性能急剧下降，有时一个查询要等几分种才有结果，无奈之下只好尝试下数据库分区。

开始评论表的结构如下:

CREATE TABLE `comment` (
  `appID` int(11) NOT NULL,
  `commentID` varchar(50) NOT NULL,
  `uname` varchar(100) DEFAULT NULL,
  `content` text,
  `cdate` datetime DEFAULT NULL,
  `ucity` varchar(50) DEFAULT NULL,
  `upnum` int(11) DEFAULT NULL,
  `downnum` int(11) DEFAULT NULL,
  `newsID` varchar(50) DEFAULT NULL,
  `isDeleted` tinyint(1) DEFAULT '0',
  PRIMARY KEY (`commentID`,`appID`),
  KEY `abc` (`cdate`) USING BTREE,
  KEY `bcd` (`commentID`) USING BTREE
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4;

开始以为这样搞，就万事大吉，后来随着数据的增多，达到5千万级别，发现查询越来越慢，因为评论的查询大部分查询的是评论的内容content,使用like ‘%expression%’做的，当然也可以使用lucene或者solr来创建倒排索引进行查询，本文目标不在此。经过调研发现，评论表的数据具有很高的时效性，也可以对评论表按时间进行分区，来加快查询速度。

创建分区的步骤如下：

1.增加时间cdate为主键

SQL语句如下：

ALTER TABLE `comment`
MODIFY COLUMN `cdate`  datetime NOT NULL AFTER `content`,
DROP PRIMARY KEY,
ADD PRIMARY KEY (`commentID`, `appID`, `cdate`);

2.执行分区SQL语句，这里直接创建两年内的分区。自动维护比较麻烦，需要创建存储过程和事件调度器（可以简单理解为时间触发器，按时间周期触发某种操作，这里特指，执行存储过程创建新的分区操作），其中存储过程比较难写，时间计算逻辑复杂。

分区SQL语句：

alter table comment_1 PARTITION BY RANGE(TO_DAYS (cdate))
(
PARTITION p20150115 VALUES LESS THAN (TO_DAYS('2015-01-15')),
PARTITION p20150131 VALUES LESS THAN (TO_DAYS('2015-01-31')),
PARTITION p20150215 VALUES LESS THAN (TO_DAYS('2015-02-15')),
PARTITION p20150228 VALUES LESS THAN (TO_DAYS('2015-02-28')),
PARTITION p20150315 VALUES LESS THAN (TO_DAYS('2015-03-15')),
PARTITION p20150331 VALUES LESS THAN (TO_DAYS('2015-03-31')),
PARTITION p20150415 VALUES LESS THAN (TO_DAYS('2015-04-15')),
PARTITION p20150430 VALUES LESS THAN (TO_DAYS('2015-04-30')),
PARTITION p20150515 VALUES LESS THAN (TO_DAYS('2015-05-15')),
PARTITION p20150531 VALUES LESS THAN (TO_DAYS('2015-05-31')),
PARTITION p20150615 VALUES LESS THAN (TO_DAYS('2015-06-15')),
PARTITION p20150630 VALUES LESS THAN (TO_DAYS('2015-06-30')),
PARTITION p20150715 VALUES LESS THAN (TO_DAYS('2015-07-15')),
PARTITION p20150731 VALUES LESS THAN (TO_DAYS('2015-07-31')),
PARTITION p20150815 VALUES LESS THAN (TO_DAYS('2015-08-15')),
PARTITION p20150831 VALUES LESS THAN (TO_DAYS('2015-08-31')),
PARTITION p20150915 VALUES LESS THAN (TO_DAYS('2015-09-15')),
PARTITION p20150930 VALUES LESS THAN (TO_DAYS('2015-09-30')),
PARTITION p20151015 VALUES LESS THAN (TO_DAYS('2015-10-15')),
PARTITION p20151031 VALUES LESS THAN (TO_DAYS('2015-10-31')),
PARTITION p20151115 VALUES LESS THAN (TO_DAYS('2015-11-15')),
PARTITION p20151130 VALUES LESS THAN (TO_DAYS('2015-11-30')),
PARTITION p20151215 VALUES LESS THAN (TO_DAYS('2015-12-15')),
PARTITION p20151231 VALUES LESS THAN (TO_DAYS('2015-12-31')),
PARTITION p20160115 VALUES LESS THAN (TO_DAYS('2016-01-15')),
PARTITION p20160131 VALUES LESS THAN (TO_DAYS('2016-01-31')),
PARTITION p20160215 VALUES LESS THAN (TO_DAYS('2016-02-15')),
PARTITION p20160228 VALUES LESS THAN (TO_DAYS('2016-02-28')),
PARTITION p20160315 VALUES LESS THAN (TO_DAYS('2016-03-15')),
PARTITION p20160331 VALUES LESS THAN (TO_DAYS('2016-03-31')),
PARTITION p20160415 VALUES LESS THAN (TO_DAYS('2016-04-15')),
PARTITION p20160430 VALUES LESS THAN (TO_DAYS('2016-04-30')),
PARTITION p20160515 VALUES LESS THAN (TO_DAYS('2016-05-15')),
PARTITION p20160531 VALUES LESS THAN (TO_DAYS('2016-05-31')),
PARTITION p20160615 VALUES LESS THAN (TO_DAYS('2016-06-15')),
PARTITION p20160630 VALUES LESS THAN (TO_DAYS('2016-06-30')),
PARTITION p20160715 VALUES LESS THAN (TO_DAYS('2016-07-15')),
PARTITION p20160731 VALUES LESS THAN (TO_DAYS('2016-07-31')),
PARTITION p20160815 VALUES LESS THAN (TO_DAYS('2016-08-15')),
PARTITION p20160831 VALUES LESS THAN (TO_DAYS('2016-08-31')),
PARTITION p20160915 VALUES LESS THAN (TO_DAYS('2016-09-15')),
PARTITION p20160930 VALUES LESS THAN (TO_DAYS('2016-09-30')),
PARTITION p20161015 VALUES LESS THAN (TO_DAYS('2016-10-15')),
PARTITION p20161031 VALUES LESS THAN (TO_DAYS('2016-10-31')),
PARTITION p20161115 VALUES LESS THAN (TO_DAYS('2016-11-15')),
PARTITION p20161130 VALUES LESS THAN (TO_DAYS('2016-11-30')),
PARTITION p20161215 VALUES LESS THAN (TO_DAYS('2016-12-15')),
PARTITION p20161231 VALUES LESS THAN (TO_DAYS('2016-12-31'))
);

创建分区耗时：

在122万数据集上新建分区耗时接近36分钟（ps:本地磁盘为机械硬盘，性能较差，曾一度认为分区失败，不过最后还是创建分区成功）

查看分区结果：

查看方式：执行SQL语句：

  select
  partition_name part,  
  partition_expression expr,  
  partition_description descr,  
  table_rows  
  from information_schema.partitions  where 
  table_schema = schema()  
  and table_name='comment';

程序查询验证：

执行SQL语句：

select * from comment where content like '%南京%' and cdate > '2015-11-01 00:00:00'

查询结果：

折线图如下

结论

1.分区查询时间平均比未分区查询快上3~4倍。

2.对已有大量数据的数据表添加分区比较耗时。