postgresql高級應用之行轉(zhuǎn)列&匯總求和的實現(xiàn)思路

更新時間：2021年05月08日 09:52:37 作者：funnyZpC

這篇文章主要介紹了postgresql高級應用之行轉(zhuǎn)列&匯總求和的實現(xiàn)思路,本文給大家介紹的非常詳細，對大家的學習或工作具有一定的參考借鑒價值，需要的朋友可以參考下

前言

節(jié)前公司業(yè)務方需要做一個統(tǒng)計報表，這個報表用于統(tǒng)計當月估計幾個明星品的銷售情況，而我們的數(shù)據(jù)是按行存儲的就是日期|產(chǎn)品|渠道|銷售額這樣，說是也奇了怪了，我們買的報(guan)表(yuan)系(la)統(tǒng)(ji) 竟然不能容易地實現(xiàn)。。。，于是我看了看，然后想了想，發(fā)現(xiàn)是可以通過sql算出這樣一個報表(多虧了postgresql的高階函數(shù)😂)，然后直接將數(shù)據(jù)輸出到報表系統(tǒng) 完事兒~ ，以下我將sql關鍵部分描述下，至於對前端展示有興趣的同學可留言，可考慮作一節(jié)講講哈😄~

報表

首先，業(yè)務需要的報表長這樣子的，看起來似乎還OK哈~

接下來我先給出我的測試腳本（均測試&無bug）~

表結(jié)構(gòu)

drop table if EXISTS  report1 ;
CREATE TABLE "report1" (
  "id" numeric(22) NOT NULL,
  "date" date NOT NULL,
  "product" varchar(100),
  "channel" varchar(100),
  "amount" numeric(20,4)
);

表注釋

字段	描述
id	主鍵
date	日期
product	產(chǎn)品
channel	渠道
amount	銷售額

表數(shù)據(jù)

INSERT INTO "report1"("id", "date", "product", "channel", "amount") VALUES ('2105051726328010100000', '2021-05-04', '產(chǎn)品1', '京東', '8899.0000');
INSERT INTO "report1"("id", "date", "product", "channel", "amount") VALUES ('2105051726328010100001', '2021-05-04', '產(chǎn)品2', '京東', '99.0000');
INSERT INTO "report1"("id", "date", "product", "channel", "amount") VALUES ('2105051727068010100010', '2021-05-04', '產(chǎn)品1', '天貓', '230.0000');
INSERT INTO "report1"("id", "date", "product", "channel", "amount") VALUES ('2105051727068010100011', '2021-05-04', '產(chǎn)品2', '天貓', '9.9000');
INSERT INTO "report1"("id", "date", "product", "channel", "amount") VALUES ('2105051727068010100011', '2021-05-04', '產(chǎn)品3', '線下門店', '10.1000');
INSERT INTO "report1"("id", "date", "product", "channel", "amount") VALUES ('2105051727068010100000', '2021-05-04', '產(chǎn)品1', '其它', '10');
INSERT INTO "report1"("id", "date", "product", "channel", "amount") VALUES ('2105051727068010100099', '2021-05-04', '產(chǎn)品2', '其它', '20000');
INSERT INTO "report1"("id", "date", "product", "channel", "amount") VALUES ('2105051727068010100033', '2021-05-01', '產(chǎn)品1', '其它', '20000');
INSERT INTO "report1"("id", "date", "product", "channel", "amount") VALUES ('2105051727068010100044', '2021-05-01', '產(chǎn)品3', '線下門店', '12345');

思考

如果你看到這裏請稍稍思考下，一開篇我説過我們的數(shù)據(jù)是按 日期|產(chǎn)品|渠道|銷售額 這樣按行存儲的，以上截圖大家一看就懂，然後再看看開篇的報表截圖，我想大家可以同我一樣可以分析出以下幾點：

報表縱向看大致分三部分

一部分是前一日產(chǎn)品銷售明細
然後一部分是前一日產(chǎn)品渠道產(chǎn)品合計
最後一部分是按渠道做的月統(tǒng)計

報表橫向看大致分兩部分

一部分是前一日的數(shù)據(jù)
另一部分則是月份匯總數(shù)據(jù)

最後一部分則是所有渠道的產(chǎn)品合計、日合計、月合計

好了，問題來了，如何做呢，我是這麼想的：首先要很清楚的是你的sql大致分兩大部分(兩個子查詢)

一部分是前一日的數(shù)據(jù)另一部分則是月份匯總數(shù)據(jù)

最後需要將兩部分數(shù)據(jù)做聯(lián)表查詢，這樣太贊了，似乎完成了報表的80%，至於最後一行的求總，這裏先賣個關子哈~

第一部分數(shù)據(jù)(前一日的數(shù)據(jù))

我想我們立馬能做的第一部分sql恐怕就是行專列吧(似乎這是最容易實現(xiàn)的😄)

select
  channel,
  sum(case product when '產(chǎn)品1' then amount end) as c1,
  sum(case product when '產(chǎn)品2' then amount end) as c2,
  sum(case product when '產(chǎn)品3' then amount end) as c3
from report1
group by channel ;

sql似乎沒什麼問題，但是我們少了一列，對那就是按渠道日合計，當然如果您對postgresql窗口函數(shù)熟悉的話，這裏實現(xiàn)的方式估計你已經(jīng)猜到了(窗口over函數(shù)),上sql...

select
  channel,
  day_sum,
  sum(case product when '產(chǎn)品1' then amount end) as c1,
  sum(case product when '產(chǎn)品2' then amount end) as c2,
  sum(case product when '產(chǎn)品3' then amount end) as c3
from
  ( select *,sum(amount) over (partition by channel) as day_sum from report1  where date=to_date('2021-05-04','yyyy-MM-dd') ) as t1
group by t1.channel ,t1.day_sum;

哈哈，上圖的day_sum估計大家很熟悉了吧，哈哈哈~
看來已經(jīng)成功地完成了日數(shù)據(jù)部分，這裏可能的難點可能就兩點

一是使用聚合函數(shù)(sum)+分組(group by)做行專列(當然postgresql也有其他很好用的行專列擴展，這裏就不介紹啦~)另一個是使用窗口函數(shù)(over)對明細提前做 按渠道的窗口匯總，這樣渠道日合計(行）的數(shù)據(jù)就有啦~

想想是不是很容易😂，接下來我們看看第二部分數(shù)據(jù)怎麼獲取~

第二部分數(shù)據(jù)(月份匯總數(shù)據(jù))

月份匯總的數(shù)據(jù)看似簡單的可怕，如果您熟練掌握postgresql中的日期處理的話估計分分鐘就能搞定，這裏就不耍大刀了，直接放出sql，哈哈哈😄

select 
  channel,sum(amount) as month_sum from report1 
where 
  date>=date(date_trunc('month',to_date('2021-05-04','yyyy-MM-dd'))) and date < date(date_trunc('month',to_date('2021-05-04','yyyy-MM-dd')) + '1 month') 
group by  
  channel

報表數(shù)據(jù)最終求解

現(xiàn)在，我們將求解的兩部分數(shù)據(jù)按渠道channel字段做inner join合并以上兩部分數(shù)據(jù)，合并后的數(shù)據(jù)大致是這樣子的

這個是sql

select
    ttt.channel,
    sum(ttt.day_sum) as day_sum,
    sum(ttt.month_sum) as month_sum,
    sum(ttt.c1) as c1,
    sum(ttt.c2) as c2,
    sum(ttt.c3) as c3
from (
        select tt1.*,tt2.month_sum from
        (
        select
        channel,
      day_sum,
        sum(case product when '產(chǎn)品1' then amount end) as c1,
        sum(case product when '產(chǎn)品2' then amount end) as c2,
        sum(case product when '產(chǎn)品3' then amount end) as c3
        from
        ( select *,sum(amount) over (partition by channel) as day_sum from report1  where date=to_date('2021-05-04','yyyy-MM-dd') ) as t1
        group by t1.channel ,t1.day_sum
        ) as tt1 left join
        (
          select channel,sum(amount) as month_sum from report1 where date>=date(date_trunc('month',to_date('2021-05-04','yyyy-MM-dd'))) and date < date(date_trunc('month',to_date('2021-05-04','yyyy-MM-dd')) + '1 month') group by  channel
        ) as tt2 on tt1.channel = tt2.channel
    ) ttt
GROUP BY ttt.channel
order by channel asc

看，匯總的數(shù)據(jù)已經(jīng)有了,已經(jīng)可以算作是最終結(jié)果了（如果你需要報表系統(tǒng)來計算匯總行數(shù)據(jù)的話），當然，我們的報表系統(tǒng)過於繁瑣(不是不能做，而是太麻煩)，需要你將做好的菜喂給它吃，這時，該怎麼辦呢。。。，哈哈哈我們似乎忘記了很久不用的rollup函數(shù)(一開始我也沒發(fā)現(xiàn)有這麼個函數(shù)哈哈)，試試看吧

select
    ttt.channel,
    sum(ttt.day_sum) as day_sum,
    sum(ttt.month_sum) as month_sum,
    sum(ttt.c1) as c1,
    sum(ttt.c2) as c2,
    sum(ttt.c3) as c3
    from (
        select tt1.*,tt2.month_sum from
        (
        select
        channel,
      day_sum,
        sum(case product when '產(chǎn)品1' then amount end) as c1,
        sum(case product when '產(chǎn)品2' then amount end) as c2,
        sum(case product when '產(chǎn)品3' then amount end) as c3
        from
        ( select *,sum(amount) over (partition by channel) as day_sum from report1  where date=to_date('2021-05-04','yyyy-MM-dd') ) as t1
        group by t1.channel ,t1.day_sum
        ) as tt1 left join
        (
          select channel,sum(amount) as month_sum from report1 where date>=date(date_trunc('month',to_date('2021-05-04','yyyy-MM-dd'))) and date < date(date_trunc('month',to_date('2021-05-04','yyyy-MM-dd')) + '1 month') group by  channel
        ) as tt2 on tt1.channel = tt2.channel
    ) ttt
    group by rollup(ttt.channel)
    order by channel asc