PostgreSQL -``令人震惊的ISTINCT ON''和``ROUP BY''语法

Question

我意识到数据库查询返回了意外的结果，这是由于我不正确地使用了“ DISTINCT ON”和“ GROUP BY”

我希望有人能使我对此有所了解。实际的查询非常复杂，因此我将其简化：

我有一个由object_id和时间戳组成的表/内部查询：

CREATE TABLE test_select ( object_id INT , event_timestamp timestamp );
COPY test_select (object_id , event_timestamp) FROM stdin (DELIMITER '|');
1           | 2013-01-27 21:01:20
1           | 2012-06-28 14:36:26
1           | 2013-02-21 04:16:48
2           | 2012-06-27 19:53:05
2           | 2013-02-03 17:35:58
3           | 2012-06-14 20:17:00
3           | 2013-02-15 19:03:34
4           | 2012-06-13 13:59:47
4           | 2013-02-23 06:31:16
5           | 2012-07-03 01:45:56
5           | 2012-06-11 21:33:26
\.

我正在尝试选择一个不同的ID，按反向计时的时间戳排序/去重复

所以结果应该是[4，1，3，2，5]

我认为这可以满足我的需求（似乎）：

SELECT object_id  
FROM test_select 
GROUP BY object_id 
ORDER BY max(event_timestamp) DESC
;

出于测试/审核的目的，有时我想包括timestamp字段。我似乎无法弄清楚如何在该查询中包含另一个字段。

任何人都可以指出以上我的sql中明显的问题，或有关如何包括审核信息的建议吗？

Answer 1

为了能够选择所有列，而不仅是object_id和MAX(event_timestamp)，您可以使用DISTINCT ON

SELECT DISTINCT ON (object_id) 
    object_id, event_timestamp    ---, more columns
FROM test_select 
ORDER BY object_id, event_timestamp DESC ;

如果希望结果按event_timestamp DESC而不是按进行排序object_id，则需要将其包括在派生表或CTE中：

SELECT *
FROM 
  ( SELECT DISTINCT ON (object_id) 
        object_id, event_timestamp    ---, more columns
    FROM test_select 
    ORDER BY object_id, event_timestamp DESC 
  ) AS t
ORDER BY event_timestamp DESC ;

另外，您可以使用窗口功能，例如ROW_NUMBER()：

WITH cte AS
  ( SELECT ROW_NUMBER() OVER (PARTITION BY object_id 
                              ORDER BY event_timestamp DESC) 
             AS rn, 
           object_id, event_timestamp    ---, more columns
    FROM test_select 
  )
SELECT object_id, event_timestamp    ---, more columns
FROM cte
WHERE rn = 1
ORDER BY event_timestamp DESC ;

或MAX()与OVER：

WITH cte AS
  ( SELECT MAX(event_timestamp) OVER (PARTITION BY object_id) 
             AS max_event_timestamp, 
           object_id, event_timestamp    ---, more columns
    FROM test_select 
  )
SELECT object_id, event_timestamp    ---, more columns
FROM cte
WHERE event_timestamp = max_event_timestamp
ORDER BY event_timestamp DESC ;