选择聚合 SQL 语句的子项的 SUM 和 COUNT

Question

背景

我正在开发一个界面，显示与促销活动相关的用户行为的汇总统计数据。用于存储这些操作的数据库是ClickHouse（为了大规模聚合统计）。据我所知，查询语法可以非常接近地模拟常规 MySQL。

相关表格如下所示：

page_visit_actions

• 促销_id
• 动作类型
• campaign_id
• ...

购买

• 促销_id
• 价格
• campaign_id
• ...

任务

我需要显示由

page_visit_actions

表中的任意列聚合执行的某些操作的数量。例如，这可能是按促销或按日期汇总的产品份额数量。

最少的设置

CREATE TABLE page_visit_actions (
    promotion_id VARCHAR(100),
    action_type VARCHAR(100),
    campaign_id int UNSIGNED
);

CREATE TABLE purchases (
    promotion_id VARCHAR(100),
    price FLOAT(10, 2),
    campaign_id int UNSIGNED
);


INSERT INTO page_visit_actions (promotion_id, action_type, campaign_id) VALUES ('5ac4ea14-f157-45a0-a706-765f4c624920', 'share_product', 17);
INSERT INTO page_visit_actions (promotion_id, action_type, campaign_id) VALUES ('5ac4ea14-f157-45a0-a706-765f4c624920', 'add_product_to_favorites', 17);
INSERT INTO page_visit_actions (promotion_id, action_type, campaign_id) VALUES ('7140beb4-5fc9-46b1-9790-0b580b6c83f3', 'add_product_to_favorites', 17);
INSERT INTO page_visit_actions (promotion_id, action_type, campaign_id) VALUES ('97a207e6-c9f8-4edf-8f9f-397544c790b9', 'add_product_to_favorites', 17);

INSERT INTO purchases (promotion_id, price, campaign_id) VALUES ('5ac4ea14-f157-45a0-a706-765f4c624920', 15.0, 17);
INSERT INTO purchases (promotion_id, price, campaign_id) VALUES ('5ac4ea14-f157-45a0-a706-765f4c624920', 5.0, 17);
INSERT INTO purchases (promotion_id, price, campaign_id) VALUES ('97a207e6-c9f8-4edf-8f9f-397544c790b9', 5.0, 17);

查询

我从这个简单的聚合查询开始，以获取每个促销的操作：

select 
promotion_id, 
count(*) as total, 
sum(case when action_type = 'share_product' then 1 else 0 end) as share, 
sum(case when action_type = 'add_product_to_favorites' then 1 else 0 end) as favorite, 
sum(case when action_type = 'add_product_to_basket' then 1 else 0 end) as basket 
from page_visit_actions
group by promotion_id;

查询按预期工作：

+--------------------------------------+-------+-------+----------+--------+
| promotion_id                         | total | share | favorite | basket |
+--------------------------------------+-------+-------+----------+--------+
| 5ac4ea14-f157-45a0-a706-765f4c624920 |     2 |     1 |        1 |      0 |
| 7140beb4-5fc9-46b1-9790-0b580b6c83f3 |     1 |     0 |        1 |      0 |
| 97a207e6-c9f8-4edf-8f9f-397544c790b9 |     1 |     0 |        1 |      0 |
+--------------------------------------+-------+-------+----------+--------+

问题

但是，当我被要求添加购买总额和收入以进行汇总统计时，我遇到了问题，并尝试了以下查询：

select 
page_visit_actions.promotion_id, 
count(page_visit_actions.promotion_id) as total, 
sum(case when action_type = 'share_product' then 1 else 0 end) as share, 
sum(case when action_type = 'add_product_to_favorites' then 1 else 0 end) as favorite, 
sum(case when action_type = 'add_product_to_basket' then 1 else 0 end) as basket,
count(purchases.promotion_id) as purchases,
sum(purchases.price) as revenue
from page_visit_actions
left join purchases on page_visit_actions.promotion_id=purchases.promotion_id
group by page_visit_actions.promotion_id;

我可以使用

promotion_id

和

page_visit_actions

共有的列

purchases

来连接表，但这会产生搞乱计数统计的副作用，因为多个操作可以引用同一个促销（促销是按每个用户提供，但用户可以分享和收藏促销活动，从而导致 2 个操作引用相同的促销活动），因此同一购买最终会加入多个操作。

因此，我尝试查询结果为：

+--------------------------------------+-------+-------+----------+--------+-----------+---------+
| promotion_id                         | total | share | favorite | basket | purchases | revenue |
+--------------------------------------+-------+-------+----------+--------+-----------+---------+
| 5ac4ea14-f157-45a0-a706-765f4c624920 |     4 |     2 |        2 |      0 |         4 |   40.00 |
| 7140beb4-5fc9-46b1-9790-0b580b6c83f3 |     1 |     0 |        1 |      0 |         0 |    NULL |
| 97a207e6-c9f8-4edf-8f9f-397544c790b9 |     1 |     0 |        1 |      0 |         1 |    5.00 |
+--------------------------------------+-------+-------+----------+--------+-----------+---------+

但是，正确的统计数据应该是这样的：

+--------------------------------------+-------+-------+----------+--------+-----------+---------+
| promotion_id                         | total | share | favorite | basket | purchases | revenue |
+--------------------------------------+-------+-------+----------+--------+-----------+---------+
| 5ac4ea14-f157-45a0-a706-765f4c624920 |     2 |     1 |        1 |      0 |         2 |   20.00 |
| 7140beb4-5fc9-46b1-9790-0b580b6c83f3 |     1 |     0 |        1 |      0 |         0 |    NULL |
| 97a207e6-c9f8-4edf-8f9f-397544c790b9 |     1 |     0 |        1 |      0 |         1 |    5.00 |
+--------------------------------------+-------+-------+----------+--------+-----------+---------+

问题

如何查询聚合表（page_visit_actions）相关的子项（购买）的

count()

和

sum()

，而不弄乱原来的聚合组统计数据？

Answer 1

您需要分别聚合两个表，因为它们每个都与

promotion

具有一对多关系（您尚未显示）。

将所有内容连接在一起然后再聚合是聚合查询中的一个常见错误。这会给你带来错误的结果，因为你最终会在所有子表之间得到笛卡尔连接。

select 
  pva.*,
  p.purchases,
  p.revenue
from (
    select
      pva.promotion_id,
      count(*) as count_actions,
      count(case when pva.action_type = 'share_product' then 1 end) as share,
      count(case when pva.action_type = 'add_product_to_favorites' then 1 end) as favorite,
      count(case when pva.action_type = 'add_product_to_basket' then 1 end) as basket
    from page_visit_actions pva
    group by
      pva.promotion_id
) pva
left join (
    select
      p.promotion_id,
      count(*) as purchases,
      sum(p.price) as revenue
    from purchases p
    group by
      p.promotion_id
) p on pva.promotion_id = p.promotion_id;

db<>小提琴