MariaDB MAX 排序失败

Question

编辑：我已经修改了问题和示例，因为上一个示例仅包含一个城市。

我正在寻找每个城市一行，其中 max_wind 或 max_gust 值最高。

这是一个基本查询 (http://sqlfiddle.com/#!9/42f63e/2)：

        SELECT *
          FROM high_wind 
         WHERE timestamp > NOW() - INTERVAL 72 HOUR;                                                                              
+------+-----------+----------+----------+---------------------+
| hid  | city      | max_wind | max_gust | timestamp           |
+------+-----------+----------+----------+---------------------+
| 4784 | Vancouver | 63       | 90       | 2021-01-13 08:26:00 |
| 4785 | Vancouver | 54       | 71       | 2021-01-13 08:40:00 |
| 4786 | Calgary   | 52       | 71       | 2021-01-13 15:35:00 |
| 4787 | Calgary   | 57       | 79       | 2021-01-13 17:53:00 |
| 4788 | Calgary   | 67       | 86       | 2021-01-13 18:14:00 |
| 4789 | Calgary   | 70       | 84       | 2021-01-13 18:40:00 |
| 4790 | Calgary   | 51       | 86       | 2021-01-13 19:00:00 |
| 4791 | Calgary   | 55       | 75       | 2021-01-13 20:00:00 |
| 4792 | Saskatoon | 66       | 81       | 2021-01-14 00:18:00 |
| 4793 | Saskatoon | 53       | 73       | 2021-01-14 01:26:00 |
| 4794 | Saskatoon | 59       | 77       | 2021-01-14 01:44:00 |
| 4795 | Saskatoon | 72       | 91       | 2021-01-14 02:00:00 |
| 4796 | Saskatoon | 77       | 103      | 2021-01-14 03:00:00 |
| 4797 | Saskatoon | 52       | 65       | 2021-01-14 05:00:00 |
| 4798 | Saskatoon | 57       | 68       | 2021-01-14 06:13:00 |
| 4799 | Saskatoon | 50       | 64       | 2021-01-14 09:00:00 |
| 4800 | Saskatoon | 51       | 61       | 2021-01-14 15:00:00 |
| 4801 | Resolute  | 51       |          | 2021-01-15 04:00:00 |
+------+-----------+----------+----------+---------------------+
18 rows in set (0.004 sec)

根据上述数据样本，这将是最终的预期结果：

+------+-----------+----------+----------+---------------------+
| hid  | city      | max_wind | max_gust | timestamp           |
+------+-----------+----------+----------+---------------------+
| 4784 | Vancouver | 63       | 90       | 2021-01-13 08:26:00 |
| 4788 | Calgary   | 67       | 86       | 2021-01-13 18:14:00 |
| 4796 | Saskatoon | 77       | 103      | 2021-01-14 03:00:00 |
| 4801 | Resolute  | 51       |          | 2021-01-15 04:00:00 |
+------+-----------+----------+----------+---------------------+

我以为我找到了使用以下 INNER JOIN 查询的解决方案，但结果好坏参半。在这里它正确地产生了两个结果，但错过了另外两个。

        SELECT *
          FROM high_wind a 
          JOIN 
             ( SELECT city
                    , MAX(max_wind) max_wind
                   , MAX(max_gust) max_gust 
              FROM high_wind
             WHERE timestamp >= NOW() - INTERVAL 72 HOUR 
             GROUP 
                BY city
             ) b 
            ON a.city = b.city 
           AND a.max_wind = b.max_wind 
           AND a.max_gust = b.max_gust 
         WHERE timestamp >= NOW() - INTERVAL 72 HOUR;          
                    
+------+-----------+----------+----------+---------------------+
| hid  | city      | max_wind | max_gust | timestamp           |
+------+-----------+----------+----------+---------------------+
| 4784 | Vancouver | 63       | 90       | 2021-01-13 08:26:00 |
| 4801 | Resolute  | 51       |          | 2021-01-15 04:00:00 |
+------+-----------+----------+----------+---------------------+
2 rows in set (0.006 sec)

Answer 1

SELECT hw.*
    FROM (
        SELECT MAX(wind) AS max_wind,
               MAX(gust) AS max_gust
            FROM high_wind
            WHERE ...
         ) AS mx
    JOIN high_wind AS hw  ON (   hw.wind = mx.max_wind
                              OR hw.gust = mx.max_gust )
    WHERE ...
    ORDER BY ...

如果最大值是针对同一城市和时间的，您可能会得到 1 行。

如果不同城市或时间的 MAX(..) 值相同，则可能超过 2 行。

如果您在 high_wind 中有一百万行，我们可以讨论索引以帮助提高性能。除非您添加 INDEX(wind) 和 INDEX(gust)，否则我的查询将需要 2 次全表扫描。但如果您需要WHERE 中的某些内容（例如省份限制或日期范围），这些索引就不够用了。

对于简单的WHERE timestamp > ...，添加

INDEX(timestamp)
INDEX(wind, timestamp)
INDEX(gust, timestamp)

更多

不要将VARCHAR 用于数字（或日期）值：

`max_wind` varchar(20) DEFAULT NULL,

这样，“102”小于“99”！