当多次连接子查询时，PostgreSQL 子查询 COUNT 失败

Question

我有 2 个表：

Table class:

id serial4 PRIMARY KEY
name varchar(64)
code varchar(64)

Table class_event，我在其中存储与 classes 相关的事件，例如“开始”和“结束”。

id serial4
class_id int4 NOT NULL  // ->  FK to the class table
event_type varchar(1) NOT NULL  // -> 's' for started, 'e' for ended.

我需要查询每个 class 开始和结束的次数。这有效：

select
    c.code,
    c.name,
    count(started.id) "started"
from "class" c
left join (select id, class_id, event_type from "class_event" where event_type = 's') started 
    on started.klass_id = c.id
group by c.code, c.name
order by started desc;

但是当我做完全相同的事情来获得结束的数量时 classes 它显示不正确的数量：

select
    c.code,
    c.name,
    count(started.id) "started",
    count(ended.id) "ended"
from "class" c
left join (select id, class_id, event_type from "class_event" where event_type = 's') started 
    on started.klass_id = c.id
left join (select id, class_id, event_type from "class_event" where event_type = 'e') ended 
    on ended.klass_id = c.id
group by c.code, c.name
order by started desc;

此外，查询的执行时间要长得多。有什么我想念的吗？

Answer 1

你可以尝试使用条件聚合函数

select
    c.code,
    c.name,
    count(CASE WHEN event_type = 's' THEN ended.id END) "started",
    count(CASE WHEN event_type = 'e' THEN ended.id END) "ended"
from "class" c
left join "class_event" started 
    on started.class_id = c.id
group by c.code, c.name
order by started desc;

Answer 2

Is there anything I'm missing?

是的，多个联接乘以行。这与此处讨论的问题完全相同：

Two SQL LEFT JOINS produce incorrect result

当您查询整个 table 时，先聚合然后再加入通常更干净、更快速。参见：

Query with LEFT JOIN not returning rows for count of 0

这也原则上避免了原来的问题，即使是多重连接——我们不需要。

SELECT * 
FROM   class c
LEFT   JOIN (
   SELECT class_id AS id
        , count(*) FILTER (WHERE event_type = 's') AS started
        , count(*) FILTER (WHERE event_type = 'e') AS ended
   FROM   class_event
   GROUP  BY 1
   ) e  USING (id)
ORDER  BY e.started DESC NULLS LAST;

NULLS LAST 因为可以想象类中的一些在 table class_event 中没有相关行（还），并且结果 NULL 值肯定不应该排在最前面。参见：

Sort by column ASC, but NULL values first?

关于聚合 FILTER 子句：

Aggregate columns with additional (distinct) filters
For absolute performance, is SUM faster or COUNT?

旁白：

对于一手满满的允许值，我会考虑数据类型 "char" 而不是 event_type 的 varchar(1)。参见：

Any downsides of using data type "text" for storing strings?

当多次连接子查询时，PostgreSQL 子查询 COUNT 失败

PostgreSQL subquery COUNT fails when the subquery is joined more than once

sql

postgresql

left-join

aggregate