Question

假设我在SQL Server中有一些orderID。每个orderID可能有多个timestamps，可以为空。找到明确的orderID

的快捷方法是什么？

所有都有非空timestamps
昨天的最后一次timestamps

示例：

今天是2018-02-08，数据如下，结果应该是OrderID 11。

OrderID     | Timestamp
------------+-------------------------
   11       | 2018-02-07 10:08:52.740    
   11       | 2018-02-06 10:08:52.740    
   22       | 2018-02-03 10:08:52.740    
   22       | 2018-02-04 10:08:52.740    
   33       | 2018-02-07 10:08:52.740    
   33       | NULL

PS：这张表有大约10亿条记录。每个orderID有大约3到4个时间戳。

Answer 1

这是group by和having：

select orderid
from t
group by orderid
having count(timestamp) = count(*) and  -- no NULLs
       max(timestamp) >= dateadd(day, -1, cast(getdate() as date)) and
       max(timestamp) < cast(getdate() as date);

如果将某些逻辑移动到where子句，可能会稍快一些。这是有效的，因为您只关心NULL时间戳和昨天的时间戳：

select orderid
from t
where timestamp is null or
      (timestamp >= dateadd(day, -1, cast(getdate() as date)) and
       timestamp < cast(getdate() as date)
      )
group by orderid
having count(timestamp) = count(*) and  -- no NULLs
       max(timestamp) is not null;

在聚合之前过滤可以加快查询速度。

Answer 2

虽然直接的解决方案是使用GROUP BY / HAVING / MIN / MAX，但是当您处理十亿行时需要使用WHERE：

查找行时间戳比今天更新 - 1
对它们进行分组并找到最大值低于今天的行（这可确保最近一行是昨天而不是今天）
对于生成的订单ID，请检查其COUNT all = COUNT不是空时间戳

WITH cte AS (
    SELECT OrderID
    FROM testdata
    WHERE Timestamp >= CAST(CURRENT_TIMESTAMP - 1 AS DATE)
    GROUP BY OrderID
    HAVING MAX(Timestamp) < CAST(CURRENT_TIMESTAMP AS DATE)
)
SELECT testdata.OrderID
FROM cte
INNER JOIN testdata ON cte.OrderID = testdata.OrderID
GROUP BY testdata.OrderID
HAVING COUNT(*) = COUNT(Timestamp)

Answer 3

您也可以尝试使用此查询：

;WITH CTE AS (
    SELECT OrderID, 
           FIRST_VALUE(Timestamp) OVER (PARTITION BY OrderID
                                        ORDER BY CASE 
                                                    WHEN Timestamp IS NULL THEN 0
                                                    ELSE 1
                                                 END, Timestamp DESC) AS first_timestamp
    FROM mytable
)
SELECT DISTINCT OrderID
FROM CTE 
WHERE  first_timestamp >= DATEADD(DAY, -1, CAST(GETDATE() as DATE)) AND
       first_timestamp < CAST(GETDATE() AS DATE);

Demo here

修改

假设以后没有日期，您可以使用GROUP BY对查询进行替换，请使用NOT EXISTS的以下查询：

SELECT DISTINCT OrderID FROM mytable AS t WHERE Timestamp >= DATEADD(DAY, -1, CAST(GETDATE() AS DATE)) AND -- Exclude OrderID slices that contain at least one `NULL` Timestamp NOT EXISTS (SELECT 1 FROM mytable AS x WHERE x.OrderID = t.OrderID AND x.Timestamp IS NULL) AND -- Exclude OrderID slices with today's date, or any other future date, -- as last Timestamp NOT EXISTS (SELECT 1 FROM mytable AS x WHERE x.OrderID = t.OrderID AND Timestamp >= DATEADD(DAY, 0, CAST(GETDATE() AS DATE)))

Demo here

Answer 4

Select distinct(orderId) from Table 
where 
(day(timestamp)=day(DATEADD(day, -1, getdate()))
and month(timestamp)=month(DATEADD(day, -1, getdate()))
and year(timestamp)=year(DATEADD(day, -1, getdate()))
) and orderid not in (select distinct(orderid) where timestamp is null)

这将返回昨天发生的所有订单ID 编辑只有那些没有单一空值的

我添加了

and orderid not in (select distinct(orderid) where timestamp is null)

这将选择所有具有空值的orderid。因此我将它们排除在外

在这种情况下，选择不同的orderID的快捷方式是什么？

4 个答案: