如果爲空，則按順序檢索不同的值

Question

我有一個表，其中包含不同類型的度量，使用id_measure_type來標識它們。我需要做一個SELECT，在該SELECT中我檢索所有期間的總和措施，並且每個單元（id_unit）butif一些措施是空的檢索另一個等等。

措施按日期，單位和小時計。這是我的表內容摘要：

 
id_unit  dateBid  hour id_measure_type  measure 
252   05/22/2013 11  6     500 
252   05/22/2013 11  4     250 
252   05/22/2013 11  1     300 
107   05/22/2013 11  4     773 
107   05/22/2013 11  1     500 
24   05/22/2013 11  6     0 
24   05/22/2013 11  4     549 
24   05/22/2013 11  1     150 

我需要一個選擇，使一個SUM所有輸入數據範圍和順序得到「最佳」測量類型：首先是id_measure_type = 6，如果其空或0則id_measure_type = 4，並且由相同的條件，則id_measure_type = 1，並且如果沒有則0

此選擇是正確的，只要有充分的措施爲6型：

SELECT id_unit, SUM(measure) AS measure 
FROM UNIT_MEASURES 
WHERE dateBid BETWEEN '03/23/2013' AND '03/24/2013' AND id_unit IN (325, 326) 
AND id_measure_type = 6 GROUP BY id_unit 

輸入是日期和單位的範圍。有一種方法可以在一個單一的選擇中做到這一點？

編輯：
我也有一個calendar表包含因此它可以用來做每一個日期和時間與它連接（如有必要，檢索每一個小時）。

編輯：
的值永遠不會成爲NULL，當「空或0」我的意思是0或失蹤的小時值。我需要每個可能的小時都在「總和」中，從「最好」的可能類型的度量中。

Answer 1

Don answer作品完美，但因爲是在表中的大量數據，它持續超過3分鐘做選擇。

我做我自己的解決方案，至極較大，醜陋，但悅目更快（其持續大約1秒）：

SELECT tUND.id_unit, ROUND(SUM(
    CASE WHEN ISNULL(t1.measure, 0) > 0 THEN t1.measure ELSE 
    CASE WHEN ISNULL(t2.measure, 0) > 0 THEN t2.measure ELSE 
    CASE WHEN ISNULL(t3.measure, 0) > 0 THEN t3.measure ELSE 0 
    END END END END 
/1000),3) AS myMeasure 
FROM calendar 
LEFT OUTER JOIN UNITS_TABLE tUND ON tUND.id_unit IN (252, 107) 
LEFT OUTER JOIN MEASURES_TABLE t1 ON t1.id_measure_type = 6 
    AND t1.dateBid = dt AND t1.hour = h AND t1.id_uf = tUND.id_uf 
LEFT OUTER JOIN MEASURES_TABLE t2 ON t2.id_unit = t1.id_unit 
    AND t2.dateBid = t1.dateBid AND t2.id_measure_type = 1 
    AND t2.id_unit = tUND.id_unit 
LEFT OUTER JOIN MEASURES_TABLE t3 ON tCNT.id_unit = tUFI.id_unit 
    AND t3.dateBid = t1.dateBid AND t3.id_measure_type = 4 
    AND t3.id_unit = tUND.id_unit 
WHERE dt BETWEEN '03/23/2013' AND '03/24/2013' 
GROUP BY tUND.id_unit 

Answer 2

我不知道如何你的一些列的是相關的，因爲您提供的每個樣本行具有相同的價值觀，並沒有measure是0，但我想你想要的是沿着這些線路：

;With Priorities as (
    select *,ROW_NUMBER() OVER (
     PARTITION BY id_unit,dateBid,hour 
     ORDER BY CASE 
      WHEN COALESCE(measure,0) > 0 THEN id_measure_type 
      else -1 END desc) as priority 
    from @t 
) 
select id_unit,SUM(measure) 
from Priorities 
where priority = 1 
group by id_unit 

假設沒有id_measure_type小於0，也沒有measure小於0 - 如果這些都是要求，這些可以適應。

（我添加了一個COALESCE，在你的「空或0」的要求實際上是在談論NULL S和0的假設）

Answer 3

不能肯定我理解錯的話，但我認爲這應該工作（其他城市你的表）

設置爲我的測試：

DECLARE @Table TABLE ([id_unit] INT, [dateBid] DATE, [hour] INT, [id_measure_type] INT, [measure] INT); 

INSERT INTO @Table 
    SELECT * 
     FROM (
     VALUES (252, GETDATE(), 11, 6, 500) 
      , (252, GETDATE(), 11, 4, 250) 
      , (252, GETDATE(), 11, 1, 300) 
      , (107, GETDATE(), 11, 4, 773) 
      , (107, GETDATE(), 11, 1, 500) 
     ) [Values]([id_unit], [dateBid], [hour], [id_measure_type], [measure]); 

實際查詢：

WITH [Filter] AS (
    SELECT * 
     , DENSE_RANK() OVER(PARTITION BY [id_unit] ORDER BY [id_measure_type] DESC) [Rank] 
     FROM @Table 
    WHERE [measure] > 0 
) 
SELECT [id_unit], SUM([measure]) 
    FROM [Filter] 
WHERE [Rank] = 1 
    AND [id_unit] IN (252, 107) 
    AND [dateBid] BETWEEN CAST(GETDATE()-1 AS DATE) AND CAST(GETDATE() AS DATE) 
GROUP BY [id_unit]; 

查看輸出的總結之前（）：

WITH [Filter] AS (
    SELECT * 
     , DENSE_RANK() OVER(PARTITION BY [id_unit] ORDER BY [id_measure_type] DESC) [Rank] 
     FROM @Table 
    WHERE [measure] > 0 
) 
SELECT * 
    FROM [Filter] 
WHERE [Rank] = 1 
    AND [id_unit] IN (252, 107) 
    AND [dateBid] BETWEEN CAST(GETDATE()-1 AS DATE) AND CAST(GETDATE() AS DATE); 

EDIT（每小時）： 我不知道，如果你在表中有更多的數據，上述樣品的總和將只（據我所知）只有SUM（）無論如何，但它仍然有效，所以我會離開它。

WITH [Filter] AS (
    SELECT * 
     , DENSE_RANK() OVER(PARTITION BY [id_unit], [Hour] ORDER BY [id_measure_type] DESC) [Rank] 
     FROM @Table 
    WHERE [measure] > 0 
) 
SELECT [id_unit], [Hour], SUM([measure]) 
    FROM [Filter] 
WHERE [Rank] = 1 
    AND [id_unit] IN (252, 107) 
    AND [dateBid] BETWEEN CAST(GETDATE()-1 AS DATE) AND CAST(GETDATE() AS DATE) 
GROUP BY [id_unit], [Hour];