SQL使用任意顺序查找第一个非空值

Posted 2021-04-02

鉴于下表：

ID | Value
----------
 1 | NULL
 2 |  B
 3 |  C
 4 | NULL

我想根据不同的顺序从Value列中获取第一个非null值，例如：

SELECT FIRST_NON_NULL(Value ORDER BY ID) FROM MY_TABLE

这将返回B.

SELECT FIRST_NON_NULL(Value ORDER BY ID DESC) FROM MY_TABLE

这将返回C.

附：不必是值函数，只需要所需的SELECT语句。谢谢。

---

编辑：想看看是否有可扩展的版本。

ID | Value1 | Value2
--------------------
 1 |  NULL  |   AA
 2 |   B    |  NULL
 3 |   C    |   CC
 4 |  NULL  |  NULL

SELECT FIRST_NON_NULL(Value1, Value2 ORDER BY ID) FROM MY_TABLE

这将返回B，AA

SELECT FIRST_NON_NULL(Value1, Value2 ORDER BY ID DESC) FROM MY_TABLE

这将返回C，CC

答案

做就是了 ：

SELECT TOP 1 Value
FROM mytable
WHERE Value IS NOT NULL
ORDER BY ID

要获取最后一个非空值，请切换顺序：

ORDER BY ID DESC

要将相同的逻辑扩展为更多列，可以使用以下查询：

SELECT (SELECT TOP 1 Value1 FROM mytable 
        WHERE Value1 IS NOT NULL ORDER BY ID) AS min_Value1,
       (SELECT TOP 1 Value2 FROM mytable 
        WHERE Value2 IS NOT NULL ORDER BY ID) AS min_Value2

Demo here

另一答案

原始答案绝对是所提供示例的最佳答案。

对于那些需要在特定分区和大容量内执行此操作的人来说，这会变得有问题和密集，因为它可以转化为后端的大量单个表命中。

我有一个使用窗口函数的解决方案，解决了大数据/分区的问题。

扩展原始示例数据如下 - 让我们在数据中有两个集合，基于GroupID（1和2）：

GroupID | ID | Value1 | Value2
--------------------------
1       |  1 |  NULL  |   AA
1       |  2 |   B    |  NULL
1       |  3 |   C    |   CC
1       |  4 |  NULL  |  NULL
2       |  5 |   E    |   EE
2       |  6 |   F    |  NULL
2       |  7 |  NULL  |   GG
2       |  8 |  NULL  |  NULL

我想为每个GroupID值提供这些第一个/最后一个非空值，如下所示：

GroupID | FirstValue1 | FirstValue2 | LastValue1 | LastValue2
-------------------------------------------------------------
1       | B           | AA          | C          | CC
2       | E           | EE          | F          | GG

如果我应用前1个逻辑来扩展，并且我有很多行，那么它将返回执行大量的单个表查询。但是，如果我使用窗口函数，我可以让它评估它在内存中的数据，以便更有效地获取数据，特别是在需要分区和大量数据时。

这是解决方案：

SELECT
  GroupID,
  FirstValue1,
  FirstValue2,
  LastValue1,
  LastValue2
FROM
  (
  SELECT
    GroupID,
    ID, Value1, Value2,
    -- Sets a rank so we can reduce to 1 row per GroupID in outer query
    DENSE_RANK() OVER (
      PARTITION BY GroupID
      ORDER BY ID
    ) IDRank,
    FIRST_VALUE(Value1) OVER (
      PARTITION BY GroupID
      ORDER BY
        CASE WHEN Value1 IS NULL THEN 2 ELSE 1 END, -- Prioritize non-null Value1
        ID                                          -- And ascending ID
    ) FirstValue1,
    FIRST_VALUE(Value2) OVER (
      PARTITION BY GroupID
      ORDER BY
        CASE WHEN Value2 IS NULL THEN 2 ELSE 1 END, -- Prioritize non-null Value2
        ID                                          -- And ascending ID
    ) FirstValue2,
    FIRST_VALUE(Value1) OVER (
      PARTITION BY GroupID
      ORDER BY
        CASE WHEN Value1 IS NULL THEN 2 ELSE 1 END, -- Prioritize non-null Value1
        ID DESC                                     -- And descending ID
    ) LastValue1,
    FIRST_VALUE(Value2) OVER (
      PARTITION BY GroupID
      ORDER BY
        CASE WHEN Value2 IS NULL THEN 2 ELSE 1 END, -- Prioritize non-null Value2
        ID DESC                                     -- And descending ID
    ) LastValue2
  FROM MY_TABLE
  ) BestValues
WHERE IDRank = 1  --Ensures we get only one row per GroupID
ORDER BY GroupID

解释 - FIRST_VALUE在每个不同的GroupID内进行评估（因为我们设置了PARTITION BY GroupID）。在该窗口中，它评估该GroupID的行，具有非空值的情况，然后按ID排序（升序或降序）。由于它是一个窗口函数，因此将值放在记录级别上的原始粒度，因此我们还添加了一个DENSE_RANK来排名，并使我们能够在外部查询中将每个GroupID下降到一行。

---

p.s.，如果您想自己运行它，请在查询上方包含以下内容，以便在CTE中将所需的样本数据生成为“MY_TABLE”。

WITH MY_TABLE AS (
  SELECT 0 as GroupID, 0 AS ID, CAST(NULL AS VARCHAR(10)) as Value1, CAST(NULL AS VARCHAR(10)) as Value2 WHERE 0=1 UNION ALL --Dummy row to set types

  SELECT 1 AS GroupID, 1 AS ID, NULL AS Value1, 'AA' AS Value2 UNION ALL
  SELECT 1 AS GroupID, 2 AS ID, 'B'  AS Value1, NULL AS Value2 UNION ALL
  SELECT 1 AS GroupID, 3 AS ID, 'C'  AS Value1, 'CC' AS Value2 UNION ALL
  SELECT 1 AS GroupID, 4 AS ID, NULL AS Value1, NULL AS Value2 UNION ALL
  SELECT 2 AS GroupID, 5 AS ID, 'E'  AS Value1, 'EE' AS Value2 UNION ALL
  SELECT 2 AS GroupID, 6 AS ID, 'F'  AS Value1, NULL AS Value2 UNION ALL
  SELECT 2 AS GroupID, 7 AS ID, NULL AS Value1, NULL AS Value2 UNION ALL
  SELECT 2 AS GroupID, 8 AS ID, NULL AS Value1, 'GG' AS Value2
)

另一答案

DISTINCT语句用于仅返回不同的（不同的）值

SELECT distinct TOP 1 Value
FROM mytable
WHERE Value IS NOT NULL
ORDER BY ID desc