如何在同一输出行中对多个子查询进行分组

Question

我正在尝试从两个已经存在的表中提取交易详情：

1. transactions，包含收到的总金额，

2. bills，交易中收到的每张钞票都有一行，并包含钞票的面额。

两者都使用共同的session id 进行索引。 [更正： 只有transactions 表在session id 上被索引。]

我已加入表格并进行了子查询，以计算每笔交易中每张钞票面额的数量（有多少 10、20 等）。我想为每笔交易获取一条记录，所有计数都在同一行。

我做到了这个查询：

SELECT
t.session,
to_char(t.amount::numeric, '"$"9990D99') AS "USD",
(select count(b.denom) where b.denom = '50' ) AS "50",
(select count(b.denom) where b.denom = '20') AS "20",
(select count(b.denom) where b.denom = '10') AS "10",
(select count(b.denom) where b.denom = '5') AS "5",
(select count(b.denom) where b.denom = '1') AS "1"

FROM transactions AS t JOIN bills AS b USING (session)
GROUP BY
t.session, t.amount, b.denom
ORDER BY 
t.session,
b.denom ASC;

...它正确地给了我钞票的数量，但每个面额都有一行：

   session    |    USD    | 50 | 20 | 10 | 5 | 1
--------------+-----------+----+----+----+---+----
 c64af32f1815 | $  135.00 |    |    |    | 1 |
 c64af32f1815 | $  135.00 |    |    |  1 |   |
 c64af32f1815 | $  135.00 |    |  6 |    |   |
 643e096b6542 | $  175.00 |    |    |    |   | 10
 643e096b6542 | $  175.00 |    |    |    | 1 |
 643e096b6542 | $  175.00 |    |  8 |    |   |
 ce7d2c647eff | $  200.00 |  4 |    |    |   |

我想要的是这样，每笔交易一行：

   session    |    USD    | 50 | 20 | 10 | 5 | 1
--------------+-----------+----+----+----+---+----
 c64af32f1815 | $  135.00 |    |  6 |  1 | 1 |
 643e096b6542 | $  175.00 |    |  8 |    | 1 | 10
 ce7d2c647eff | $  200.00 |  4 |    |    |   |

我需要了解什么才能修复此查询？

修改后的查询（遵循@erwin 的建议以避免子查询）：

SELECT
t.session,
to_char(t.amount::numeric, '"$"9990D99') AS "USD",
COUNT(NULLIF(b.denom = '100', FALSE)) AS "100",
COUNT(NULLIF(b.denom = '50', FALSE)) AS "50",
COUNT(NULLIF(b.denom = '20', FALSE)) AS "20",
COUNT(NULLIF(b.denom = '10', FALSE)) AS "10",
COUNT(NULLIF(b.denom = '5', FALSE)) AS "5",
COUNT(NULLIF(b.denom = '1', FALSE)) AS "1"

FROM transactions AS t JOIN bills AS b USING (session)
GROUP BY
t.session, t.amount, b.denom
ORDER BY 
t.session,
b.denom ASC;

此查询仍会为每个聚合（计数）函数调用生成一行输出。

Answer 1

不要使用相关子查询。那是低效的。
并且不要在GROUP BY 子句中包含b.denom。这是你的主要错误。

Postgres 9.4+

在 Postgres 9.4 或更高版本中使用专用聚合 FILTER 功能：

SELECT t.session
     , to_char(t.amount::numeric, '"$"9990D99') AS "USD"
     , count(*) FILTER (WHERE b.denom = '50')   AS "50"  -- !
     , count(*) FILTER (WHERE b.denom = '20')   AS "20"  -- !
     , ...
FROM   ...
GROUP  BY t.session, t.amount  -- !
ORDER  BY ...

解释和更多链接：

• How can I simplify this game statistics query?

Postgres 9.3-

对于旧版本（Postgres 9.3 或更早版本），有多种（不太优雅的）替代方案：

SELECT t.session
     , to_char(t.amount::numeric, '"$"9990D99') AS "USD"  -- why cast to numeric?
     , count(b.denom = '100' OR NULL) AS "100"  -- bad column name
     , count(b.denom =  '50' OR NULL) AS "50"
     , count(b.denom =  '20' OR NULL) AS "20"
     , count(b.denom =  '10' OR NULL) AS "10"
     , count(b.denom =   '5' OR NULL) AS "5"
     , count(b.denom =   '1' OR NULL) AS "1"
FROM   transactions t
JOIN   bills b USING (session)
GROUP  BY t.session, t.amount
ORDER  BY t.session;

清理后：

SELECT t.session, to_char(t.amount, '"$"9990D99') AS usd
     , d100, d50, d20, d10, d5, d1
FROM   transactions t
LEFT   JOIN (
   SELECT session
        , nullif(count(denom = 100 OR NULL), 0) AS d100
        , nullif(count(denom =  50 OR NULL), 0) AS d50
        , nullif(count(denom =  20 OR NULL), 0) AS d20
        , nullif(count(denom =  10 OR NULL), 0) AS d10
        , nullif(count(denom =   5 OR NULL), 0) AS d5
        , nullif(count(denom =   1 OR NULL), 0) AS d1
   FROM   bills
   GROUP  BY 1
   ) b USING (session)
ORDER  BY session;

发生了什么变化？

• 如果可以避免，不要使用非法标识符，那么你也不需要双引号。

• 假设integer 代表bills.denom，所以我们也不需要在常量周围加上单引号。

• 您似乎希望在没有找到账单的列中出现NULL。将结果包装在NULLIF()

• 由于您检索整个表，因此在您加入之前进行聚合会更快。

SQL Fiddle.

更多技巧和解释：

• For absolute performance, is SUM faster or COUNT?

更快的选择：crosstab()

为了获得最佳性能，请使用带有crosstab() 的实际交叉表。您需要安装附加模块tablefunc。 如果您不熟悉，请先阅读本文：

• PostgreSQL Crosstab Query

SELECT t.session, to_char(t.amount, '"$"9990D99') AS usd
     , d100, d50, d20, d10, d5, d1
FROM   transactions t
LEFT   JOIN crosstab(
     $$SELECT session, denom, count(*)::int
       FROM   bills
       GROUP  BY 1,2
       ORDER  BY 1$$
   , $$SELECT unnest ('{100,50,20,10,5,1}'::int[])$$
   ) AS ct (session text, d100 int, d50 int, d20 int, d10 int, d5 int, d1 int)
     USING (session);

Answer 2

你可以添加查询全局吗：

SELECT
session,
"USD",
sum("50") AS "50",
sum("20") AS "20",
sum("10") AS "10",
sum("5") AS "5",
sum("1") AS "1"

from (SELECT
t.session,
to_char(t.amount::numeric, '"$"9990D99') AS "USD",
(select count(b.denom) where b.denom = '50' ) AS "50",
(select count(b.denom) where b.denom = '20') AS "20",
(select count(b.denom) where b.denom = '10') AS "10",
(select count(b.denom) where b.denom = '5') AS "5",
(select count(b.denom) where b.denom = '1') AS "1"

FROM transactions AS t JOIN bills AS b USING (session)
GROUP BY
t.session, t.amount, b.denom
ORDER BY 
t.session,
b.denom ASC)youre_query
GROUP BY session,"USD"

Answer 3

我想你快到了。您需要在查询后进行另一个分组。例如：

WITH q1 as (SELECT
    t.session,
    to_char(t.amount::numeric, '"$"9990D99') AS "USD",
    (select count(b.denom) where b.denom = '50' ) AS "50",
    (select count(b.denom) where b.denom = '20') AS "20",
    (select count(b.denom) where b.denom = '10') AS "10",
    (select count(b.denom) where b.denom = '5') AS "5",
    (select count(b.denom) where b.denom = '1') AS "1"

    FROM transactions AS t JOIN bills AS b USING (session)
    GROUP BY t.session, t.amount, b.denom)

SELECT session, "USD", SUM("50") AS "50", SUM("20") AS "20", SUM("10") AS "10",
       SUM("5") AS "5", SUM("1") AS "1"
FROM q1
GROUP BY session, "USD"