在带有/不带时区的日期或时间戳的查询中处理 generate_series()

Question

我有一个查询要根据按date 和employee_id 分组的日期系列生成报告。日期应基于特定时区，在本例中为“Asia/Kuala_Lumpur”。但这可能会根据用户所在时区的位置而改变。

SELECT 
     d::date AT TIME ZONE 'Asia/Kuala_Lumpur' AS created_date,  
     e.id,  
     e.name,
     e.division_id,
     ARRAY_AGG(
       a.id
     ) as rows,        
     MIN(a.created_at) FILTER (WHERE a.activity_type = 1) as min_time_in,
     MAX(a.created_at) FILTER (WHERE a.activity_type = 2) as max_time_out,
     ARRAY_AGG(
       CASE
           WHEN a.activity_type = 1
           THEN a.created_at
           ELSE NULL
       END
     ) as check_ins,
     ARRAY_AGG(
       CASE
           WHEN a.activity_type = 2
           THEN a.created_at
           ELSE NULL
       END
     ) as check_outs        
FROM    (SELECT MIN(created_at), MAX(created_at) FROM attendance) AS r(startdate,enddate)
  , generate_series(
        startdate::timestamp, 
        enddate::timestamp, 
        interval '1 day') g(d)
    CROSS JOIN  employee e
    LEFT JOIN   attendance a ON a.created_at::date = d::date AND e.id = a.employee_id
    where d::date = date '2020-11-20' and division_id = 1
GROUP BY 
    created_date
  , e.id
  , e.name
  , e.division_id
ORDER BY 
    created_date
  , e.id;

表attendance的定义和样本数据：

CREATE TABLE attendance (
    id int,
    employee_id int,
    activity_type int,
    created_at timestamp with time zone NOT NULL
);

INSERT INTO attendance VALUES
( 1, 1, 1,'2020-11-18 07:10:25 +00:00'),
( 2, 2, 1,'2020-11-18 07:30:25 +00:00'),
( 3, 3, 1,'2020-11-18 07:50:25 +00:00'),
( 4, 2, 2,'2020-11-18 19:10:25 +00:00'),
( 5, 3, 2,'2020-11-18 19:22:38 +00:00'),
( 6, 1, 2,'2020-11-18 20:01:05 +00:00'),
( 7, 1, 1,'2020-11-19 07:11:23 +00:00'),
( 8, 1, 2,'2020-11-19 16:21:53 +00:00'), <-- Asia/Kuala_Lumpur +8 should be in 20.11 (refer to the check_outs field in the results output)
( 9, 1, 1,'2020-11-19 19:11:23 +00:00'), <-- Asia/Kuala_Lumpur +8 should be in 20.11 (refer to the check_ins field in the results output)
(10, 1, 2,'2020-11-19 20:21:53 +00:00'), <-- Asia/Kuala_Lumpur +8 should be in 20.11 (refer to the check_outs field in the results output)
(11, 1, 1,'2020-11-20 07:41:38 +00:00'),
(12, 1, 2,'2020-11-20 08:52:01 +00:00');

这是一个fiddle 进行测试。

查询不包含时区 Asia/Kuala_Lumpur +8 的输出中的第 8-10 行，但它应该包含。结果显示“行”字段11,12。

如何修复查询，以便它根据给定时区的日期生成报告？ （意味着我可以将Asia/Kuala_Lumpur 更改为America/New_York 等）

我被告知要做这样的事情：

where created_at >= timestamp '2020-11-20' AT TIME ZONE 'Asia/Kuala_Lumpur'
and   created_at <  timestamp '2020-11-20' AT TIME ZONE 'Asia/Kuala_Lumpur' + interval '1 day'

但我不确定如何应用它。在this fiddle 中似乎无法正常工作。它应该包括第 8、9、10、11、12 行，但只显示第 8、9、10 行。

Answer 1

数据库设计

考虑对您的设置进行一些修改：

CREATE TABLE employee (
  id           int PRIMARY KEY  -- !
, name         text             -- do NOT use char(n) !
, division_id  int
);

CREATE  TABLE attendance (
  id             int PRIMARY KEY  --!
, employee_id    int NOT NULL REFERENCES employee -- FK!
, activity_type  int
, created_at     timestamptz NOT NULL
);

定义 PK 可以更轻松地聚合行，因为 PK 涵盖了 GROUP BY 子句中的整行。见：

• Why can't I exclude dependent columns from `GROUP BY` when I aggregate by a key?

我不会使用“名称”作为列名。它不是描述性的。每隔一列可以命名为“名称”。考虑：

• Any downsides of using data type "text" for storing strings?
• How to implement a many-to-many relationship in PostgreSQL?

查询

SELECT *
FROM  (        -- complete employee/date grid for division in range
   SELECT g.d::date AS the_date, id AS employee_id, name, division_id
   FROM  (
      SELECT generate_series(MIN(created_at) AT TIME ZONE 'Asia/Kuala_Lumpur'
                           , MAX(created_at) AT TIME ZONE 'Asia/Kuala_Lumpur'
                           , interval '1 day')
      FROM   attendance
      ) g(d)
   CROSS  JOIN employee e
   WHERE  e.division_id = 1
   ) de
LEFT   JOIN (  -- checkins & checkouts per employee/date for division in range
   SELECT employee_id, ts::date AS the_date
        , array_agg(id) as rows
        , min(ts)             FILTER (WHERE activity_type = 1) AS min_check_in
        , max(ts)             FILTER (WHERE activity_type = 2) AS max_check_out
        , array_agg(ts::time) FILTER (WHERE activity_type = 1) AS check_ins
        , array_agg(ts::time) FILTER (WHERE activity_type = 2) AS check_outs
   FROM  (
      SELECT a.id, a.employee_id, a.activity_type, a.created_at AT TIME ZONE 'Asia/Kuala_Lumpur' AS ts  -- convert to timestamp
      FROM   employee   e
      JOIN   attendance a ON a.employee_id = e.id
   -- WHERE  a.created_at >= timestamp '2020-11-20' AT TIME ZONE 'Asia/Kuala_Lumpur' -- "sargable" expressions
   -- AND    a.created_at <  timestamp '2020-11-21' AT TIME ZONE 'Asia/Kuala_Lumpur' -- exclusive upper bound (includes all of 2020-11-20);
      AND    e.division_id = 1
      ORDER  BY a.employee_id, a.created_at, a.activity_type  -- optional to guarantee sorted arrays
   ) sub
   GROUP  BY 1, 2
   ) a USING (the_date, employee_id)
ORDER  BY 1, 2;

db小提琴here

请注意，我的查询会输出 Asia/Kuala_Lumpur 的本地日期和时间：

test=> SELECT timestamptz '2020-11-20 08:52:01 +0' AT TIME ZONE 'Asia/Kuala_Lumpur' AS local_ts;
      local_ts       
---------------------
 2020-11-20 16:52:01

从哪里开始？您需要了解时区和 Postgres 数据类型 timestamp with time zone (timestamptz) 与 timestamp without time zone (timestamp) 的概念。否则，将是无休止的混乱。从这里开始：

• Ignoring time zones altogether in Rails and PostgreSQL

最值得注意的是，timestamptz不存储时区：

• Time zone storage in data type "timestamp with time zone"

当简单地将timestamptz 转换为date 或timestamp 时，假定会话的当前时区设置。 不是你想要的。使用 AT TIME ZONE 构造显式提供时区以避免这种错误。在你的小提琴中，你有两个：

  ...
  , generate_series(
        startdate::timestamp AT TIME ZONE 'Asia/Kuala_Lumpur', 
        enddate::timestamp AT TIME ZONE 'Asia/Kuala_Lumpur', 
        interval '1 day') g(d)
   ...

也不做你想做的事。在（错误！）转换为timestamp 之后，AT TIME ZONE 构造将值转换回timestamptz。

此外，您的查询会生成所有用户的完整笛卡尔积，以及表 attendance 中的最大天数范围，只是为了将其减少到一天：

    where created_at >= timestamp '2020-11-20' AT TIME ZONE 'Asia/Kuala_Lumpur'
    and   created_at <  timestamp '2020-11-20' AT TIME ZONE 'Asia/Kuala_Lumpur' + interval '1 day'

WHERE 子句终于完成了它应该做的事情。但是首先生成完整的天数，然后丢弃大部分天数是没有意义的。 （看来你是同时从my other fiddle 复制的？）

我注释掉了WHERE 子句，并在我的查询中保留了generate_series() 的优化版本作为概念证明。延伸阅读：

• Generating time series between two dates in PostgreSQL