Erlo

项目上线后,谈一下感触比较深的一点:查询优化

时间:2019-07-11 10:04   阅读:92次   来源:博客园页面报错

点赞

前言

  开心一刻

    儿子有道题不会做,喊我过去教他。我推了推一旁的老公:我头疼,你去吧。老公不动,我:零花钱涨一千。话音刚落,老公就屁颠屁颠跑去儿子房间。进去不到几分钟,一声怒吼伴随着儿子的哭声传来的瞬间,老公从儿子房间出来,边走边说:“朽木不可雕也。”儿子从房间探出半个身子,一脸委屈:“爸爸也不会做,他说给我一块钱,让我明天早点去学校抄同学的。还让我不要告诉你,我不肯,他就吼我。”

问题背景

  前段时间,被紧急调到一个新项目,支撑新项目的开发。跌跌撞撞之下,项目也正常上线了,期间收获颇多,无论是业务上的,还是业务之外的。业务上的就不多说了,不具通用性,意义不大,有一点业务之外的东东给我的感触比较深,特记录下来,与大家分享下 : 查询优化

查询优化

  完整示例工程:data-init,包括数据库表的 ddl 和 dml,以及数据批量的生成

  相关表

    涉及的表不多,一共三张:额度表、记录表 、 存款表

    额度表 t_custmor_credit

CREATE TABLE t_customer_credit (
  id INT(11) UNSIGNED NOT NULL AUTO_INCREMENT COMMENT '自增主键',
  login_name VARCHAR(50) NOT NULL COMMENT '名称',
    credit_type TINYINT(1) NOT NULL COMMENT '额度类型,1:自由资金,2:冻结资金,3:优惠',
    amount DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '额度值',
    create_by VARCHAR(50) NOT NULL COMMENT '创建者',
    create_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '创建时间',
    update_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP COMMENT '创建时间',
    update_by VARCHAR(50) NOT NULL COMMENT '修改者',
  PRIMARY KEY (id)
);

View Code

      记录每个顾客的当前额度,额度一共分三种:自由资金、冻结资金和优惠,也就是说每个顾客会有 3 条记录来表示他的各个额度。表中数据如下

    额度记录 t_custmor_credit_record

CREATE TABLE t_customer_credit_record (
  id INT(11) UNSIGNED NOT NULL AUTO_INCREMENT COMMENT '自增主键',
  login_name VARCHAR(50) NOT NULL COMMENT '名称',
    credit_type TINYINT(1) NOT NULL COMMENT '额度类型,参考t_custmor_credit的credit_type',
    bill_no VARCHAR(50) NOT NULL COMMENT '订单号',
    amount_before DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '前额度值',
    amount_change DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '额度变化值',
    amount_after DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '后额度值',
    create_by VARCHAR(50) NOT NULL COMMENT '创建者',
    create_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '创建时间',
    remark VARCHAR(500) NOT NULL DEFAULT '' COMMENT '备注',
  PRIMARY KEY (id)
);

View Code

      记录顾客额度的每一次变化,只要有额度变化(不管是哪个额度进行了变化),都会新增3条记录,每个类型的额度都会新增一条记录。另外,该表只会有数据的插入,不会有数据的删、改。表中数据如下

    存款表 t_custmor_deposit

CREATE TABLE t_customer_deposit (
  id INT(11) UNSIGNED NOT NULL AUTO_INCREMENT COMMENT '自增主键',
  login_name VARCHAR(50) NOT NULL COMMENT '名称',
    bill_no VARCHAR(50) NOT NULL COMMENT '订单号',
    amount DECIMAL(22,6) NOT NULL DEFAULT '0.00000' COMMENT '存款金额',
    deposit_state TINYINT(1) NOT NULL COMMENT '存款状态: 1成功,2失败,3未知',
    channal TINYINT(2) NOT NULL COMMENT '存款渠道: 1:银联,2支付宝,3微信',
    create_by VARCHAR(50) NOT NULL COMMENT '创建者',
    create_time DATETIME NOT NULL DEFAULT CURRENT_TIMESTAMP COMMENT '创建时间',
    remark VARCHAR(500) NOT NULL DEFAULT '' COMMENT '备注',
  PRIMARY KEY (id)
);

View Code

      记录顾客的每一次存款,该表只会有数据的插入,不会有数据的删、改。表中数据如下

  产品需求

    分页展示如下列表(具体的过滤条件就不列了,我们就当是初始状态,还未输入过滤条件)

     实现比较简单,t_custmor_credit_record 左关联 t_custmor_deposit 就好,但是我们的额度记录表与需求列表有些许的出入,需要做一下简单的行转列。

  需求实现

    我们先来看看最初的SQL查询,这可能是很多人最容易想到的

SELECT MIN(tcd.channal) channal, MAX(tccr.id) mId,tccr.login_name,tccr.bill_no,tccr.create_time,
    IF(credit_type=1,amount_before,0) AS freeBefore,
    IF(credit_type=1,amount_change,0) AS freeChange,
    IF(credit_type=1,amount_after,0) AS freeAfter,
    IF(credit_type=2,amount_before,0) AS freezeBefore,
    IF(credit_type=2,amount_change,0) AS freezeChange,
    IF(credit_type=2,amount_after,0) AS freezeAfter,
    IF(credit_type=3,amount_before,0) AS promotionBefore,
    IF(credit_type=3,amount_change,0) AS promotionChange,
    IF(credit_type=3,amount_after,0) AS promotionAfter
FROM t_customer_credit_record tccr
LEFT JOIN t_customer_deposit tcd ON tccr.bill_no = tcd.bill_no
GROUP BY tccr.bill_no,tccr.login_name,tccr.create_time
ORDER BY mId desc
LIMIT 0, 10;

View Code

    数据量少的时候,也许能在我们接受的时间内查出我们需要的结果,一旦数据量多了,这个SQL就跑不动了;我们先看下 60w 数据的情况下,我们只进行 t_custmor_credit_record 单表查询

SELECT MAX(id) mId,login_name,bill_no,create_time,
    IF(credit_type=1,amount_before,0) AS freeBefore,
    IF(credit_type=1,amount_change,0) AS freeChange,
    IF(credit_type=1,amount_after,0) AS freeAfter,
    IF(credit_type=2,amount_before,0) AS freezeBefore,
    IF(credit_type=2,amount_change,0) AS freezeChange,
    IF(credit_type=2,amount_after,0) AS freezeAfter,
    IF(credit_type=3,amount_before,0) AS promotionBefore,
    IF(credit_type=3,amount_change,0) AS promotionChange,
    IF(credit_type=3,amount_after,0) AS promotionAfter
FROM t_customer_credit_record
GROUP BY bill_no,login_name,create_time
ORDER BY mId desc
LIMIT 0, 10;

View Code

    效果如下

    花了近 8 秒,这还只是单表,如果执行上面的联表SQL,那时间又得增加不少(我试验的结果是直接卡住了,看不到查询结果);

  优化处理

    加索引

      查询慢的时候,我们最容易想到的优化方式往往就是加索引;上述SQL执行的时候,t_custmor_credit_record 和 t_custmor_deposit都没有建索引(主键索引除外),那么我们就加索引呗。我的项目中加的是唯一索引,做了唯一约束,那我这里也加唯一索引

ALTER TABLE t_customer_credit_record ADD UNIQUE uk_unique (bill_no,login_name,create_time,credit_type);
ALTER TABLE t_customer_deposit ADD UNIQUE uk_billno (bill_no);

      此时我们看下SQL执行效果

      我们发现,t_custmor_credit_record 单表查询的效率几乎没变,将近 8 秒,但 t_custmor_credit_record 与 t_custmor_deposit 联表的查询却在 11 秒内有结果了。加了索引为什么还这么慢了? 难道没走索引?

      我们是不是发现了什么? IF函数对联表查询是否走索引有影响,也对单表的查询速度有影响。上图中的 t_custmor_credit_record 单表查询,有IF函数,查询时间近 8 秒,没有IF函数,查询时间 2 秒左右;t_custmor_credit_record 与 t_custmor_deposit 联表查,有IF函数,t_custmor_credit_record 走的是全表查,查询时间近 11 秒,没有IF函数,t_custmor_credit_record 走的是索引,查询时间 3 秒不到。那么我们有没有什么办法拿掉这个IF函数呢?

    使用 CASE...WHEN....THEN 代替 IF

SELECT MAX(id) mId,login_name,bill_no,create_time,
    CASE credit_type WHEN 1 THEN amount_before ELSE 0 END AS freeBefore,
    CASE credit_type WHEN 1 THEN amount_change ELSE 0 END AS freeChange,
    CASE credit_type WHEN 1 THEN amount_after ELSE 0 END AS freeAfter,
    CASE credit_type WHEN 2 THEN amount_before ELSE 0 END AS freeChange,
    CASE credit_type WHEN 2 THEN amount_change ELSE 0 END AS freeChange,
    CASE credit_type WHEN 2 THEN amount_after ELSE 0 END AS freeChange,
    CASE credit_type WHEN 3 THEN amount_before ELSE 0 END AS promotionBefore,
    CASE credit_type WHEN 3 THEN amount_change ELSE 0 END AS promotionChange,
    CASE credit_type WHEN 3 THEN amount_after ELSE 0 END AS promotionAfter
FROM t_customer_credit_record
GROUP BY bill_no,login_name,create_time
ORDER BY mId desc
LIMIT 0, 10;

SELECT MIN(tcd.channal) channal, MAX(tccr.id) mId,tccr.login_name,tccr.bill_no,tccr.create_time,
    CASE credit_type WHEN 1 THEN amount_before ELSE 0 END AS freeBefore,
    CASE credit_type WHEN 1 THEN amount_change ELSE 0 END AS freeChange,
    CASE credit_type WHEN 1 THEN amount_after ELSE 0 END AS freeAfter,
    CASE credit_type WHEN 2 THEN amount_before ELSE 0 END AS freeChange,
    CASE credit_type WHEN 2 THEN amount_change ELSE 0 END AS freeChange,
    CASE credit_type WHEN 2 THEN amount_after ELSE 0 END AS freeChange,
    CASE credit_type WHEN 3 THEN amount_before ELSE 0 END AS promotionBefore,
    CASE credit_type WHEN 3 THEN amount_change ELSE 0 END AS promotionChange,
    CASE credit_type WHEN 3 THEN amount_after ELSE 0 END AS promotionAfter
FROM t_customer_credit_record tccr
LEFT JOIN t_customer_deposit tcd ON tccr.bill_no = tcd.bill_no
GROUP BY tccr.bill_no,tccr.login_name,tccr.create_time
ORDER BY mId desc
LIMIT 0, 10;

View Code

      我们可以看到,执行时间与 IF 所差无几,执行计划也是与 IF 的一致,这也就反映出不是 IF的问题,应该是 GROUP BY 的问题。我们用 GROUP BY 结合 IF(或 CASE...WHEN....THEN),就是为了将 3 条额度记录合并成一条、行转列之后输出我们想要的结果,那有没有不用 GROUP BY、又能实现我们需求的方式了?

    自联代替 GROUP BY

      我们再仔细琢磨下这个需求,咋一看,确实需要行转列,那么就需要用到 GROUP BY,那么效率也就低了,这似乎是无解了? 真的非要行转列吗,假设我们将额度记录拆分成 3 张表:一张表只存自由资金的额度变化、一张表只存冻结资金的额度变化、一张表只存优惠的额度变化,这样是不是只需要联表查而不要用 GROUP BY 来进行行转列了? 有小伙伴有可能会问:t_custmor_credit_record 表已经定了,数据都跑了不少了,再将其进行拆分,既要改表(同时还要迁移数据),还要改代码,工程量会很大! 我们换个角度来看 t_custmor_credit_record ,目前它是 3 中额度记录的一个总和表,我们能不能从它的身上做文章,变化出我们想要的那 3 张表,然后进行联表查询呢? 肯定可以的,类似如下

-- 自由资金额度记录表
SELECT * FROM t_customer_credit_record WHERE credit_type = 1;
-- 冻结资金额度记录表
SELECT * FROM t_customer_credit_record WHERE credit_type = 2;
-- 优惠额度记录表
SELECT * FROM t_customer_credit_record WHERE credit_type = 3;

      接下来的 SQL 怎么写,我想大家都知道了吧,自联就行了,写法有很多种,常见的写法有如下 4 种

-- 不用group by,做法1, 个人比较推荐, 但此种方式不支持存款表的过滤条件
SELECT d.channal,a.amount_before AS freeBefore,a.amount_change AS freeChange, a.amount_after freeAfter,
    b.amount_before AS freezeBefore,b.amount_change AS freezeChange, b.amount_after freezeAfter,
    c.amount_before AS promotionBefore,c.amount_change AS promotionChange, c.amount_after promotionAfter
FROM (
    SELECT * FROM t_customer_credit_record WHERE credit_type = 1 ORDER BY id DESC LIMIT 0, 10
) a
LEFT JOIN t_customer_credit_record b ON a.bill_no = b.bill_no AND b.credit_type = 2
LEFT JOIN t_customer_credit_record c ON a.bill_no = c.bill_no AND c.credit_type = 3
LEFT JOIN t_customer_deposit d ON a.bill_no = d.bill_no;

-- 不用group by,做法2, 此种方式支持存款表的过滤条件
SELECT a.channal,a.amount_before AS freeBefore,a.amount_change AS freeChange, a.amount_after freeAfter,
    b.amount_before AS freezeBefore,b.amount_change AS freezeChange, b.amount_after freezeAfter,
    c.amount_before AS promotionBefore,c.amount_change 下一篇:python数据库-MySQL单表查询基本操作(50)                

评论留言

还没有评论留言,赶紧来抢楼吧~~

Erlo大厅()

* 这里是“吐槽厅”,所有人可看,只保留当天信息。

  • Erlo.vip2019-09-19 10:59:18Hello、欢迎使用吐槽厅,这里是个吐槽的地方。
  • 返回顶部