优化查询或建议LINQ等效

Question

我有一个表，包含列date_trans，time_trans，价格。选择查询后，我想添加一个新的列“计数”，它将被计算为价格列的连续相等值，并且具有连续相等价格的前一行将从最终结果中移除。看到预期的输出：

date_trans time_trans price **Count**  
2011-02-22 09:39:59 58.02 1 
2011-02-22 09:40:03 58.1 *ROW WILL BE REMOVED 
2011-02-22 09:40:07 58.1 *ROW WILL BE REMOVED 
2011-02-22 09:40:08 58.1 3 
2011-02-22 09:40:10 58.15 1 
2011-02-22 09:40:10 58.1 *ROW WILL BE REMOVED 
2011-02-22 09:40:14 58.1 2 
2011-02-22 09:40:24 58.15 1 
2011-02-22 09:40:24 58.18 *ROW WILL BE REMOVED 
2011-02-22 09:40:24 58.18 *ROW WILL BE REMOVED 
2011-02-22 09:40:24 58.18 3 
2011-02-22 09:40:24 58.15 1 

请提出一个SQL查询或LINQ表达式从表

目前选择的，我能做到这一点是选择查询，并通过所有选定行循环，但选择数以百万计的时候行需要数小时。

我当前的代码：

string query = @"SELECT date_trans, time_trans, price 
          FROM tbl_data 
         WHERE date_trans BETWEEN '2011-02-22' AND '2011-10-21' 
         AND time_trans BETWEEN '09:30:00' AND '16:00:00'"; 

      DataTable dt = oUtil.GetDataTable(query); 

      DataColumn col = new DataColumn("Count", typeof(int)); 
      dt.Columns.Add(col); 

      int priceCount = 1; 
      for (int count = 0; count < dt.Rows.Count; count++) 
      { 
       double price = Convert.ToDouble(dt.Rows[count]["price"]); 
       double priceNext = (count == dt.Rows.Count - 1) ? 0 : Convert.ToDouble(dt.Rows[count + 1]["price"]); 
       if (price == priceNext) 
       { 
        priceCount++; 
        dt.Rows.RemoveAt(count); 
        count--; 
       } 
       else 
       { 
        dt.Rows[count]["Count"] = priceCount; 
        priceCount = 1; 
       } 
      } 

Answer 1

这是一个有趣的一个。我认为你需要会是这样的：

SELECT MAX(date_trans), MAX(time_trans), MAX(price), COUNT(*) 
FROM 
    (SELECT *, ROW_NUMBER() OVER(PARTITION BY price ORDER BY date_trans, time_trans) - ROW_NUMBER() OVER(ORDER BY date_trans, time_trans) AS grp 
    FROM transactions) grps 
GROUP BY grp 

找到了解决办法在这里：http://www.sqlmag.com/article/sql-server/solution-to-the-t-sql-puzzle-grouping-consecutive-rows-with-a-common-element

UPDATE

分组列需要还包括“价格”，否则组可能不独一无二。还有一件事是，日期和时间列应该合并到日期时间列中，以便最大日期时间值在从一天结束开始到下一开始结束的组中是正确的。 这是更正后的查询。

SELECT MAX(CAST(date_trans AS DATETIME) + CAST(time_trans AS DATETIME)) , MAX(price), COUNT(*) 
FROM 
    (SELECT *, 
     CAST(ROW_NUMBER() OVER(PARTITION BY price ORDER BY date_trans, time_trans) - ROW_NUMBER() OVER(ORDER BY date_trans, time_trans) AS NVARCHAR(255)) + '-' + CAST(price AS NVARCHAR(255)) AS grp 
    FROM transactions 
    ORDER BY date_trans, time_trans) grps 
GROUP BY grp 

查询可能更适合'grp'列作为字节数组或bigint而不是nvarchar。你还提到了你可能想要在小组内汇总的'卷'栏。