我有一个复杂的查询，它在查询窗口中运行2秒钟，但作为存储过程大约5分钟。为什么以存储过程的形式运行需要花费更长的时间？
这是我的查询的样子。
它采用一组特定的记录（由@id和标识@createdDate）和特定的时间范围（从开始的1年@startDate），并返回已发送信件的摘要列表以及由于这些信件而产生的估计付款。
CREATE PROCEDURE MyStoredProcedure
@id int,
@createdDate varchar(20),
@startDate varchar(20)
AS
SET NOCOUNT ON
-- Get the number of records * .7
-- Only want to return records containing letters that were sent on 70% or more of the records
DECLARE @limit int
SET @limit = IsNull((SELECT Count(*) FROM RecordsTable WITH (NOLOCK) WHERE ForeignKeyId = @id AND Created = @createdDate), 0) * .07
SELECT DateSent as [Date]
, LetterCode as [Letter Code]
, Count(*) as [Letters Sent]
, SUM(CASE WHEN IsNull(P.DatePaid, '1/1/1753') BETWEEN DateSent AND DateAdd(day, 30, DateSent) THEN IsNull(P.TotalPaid, 0) ELSE 0 END) as [Amount Paid]
INTO #tmpTable
FROM (
-- Letters Table. Filter for specific letters
SELECT DateAdd(day, datediff(day, 0, LR.DateProcessed), 0) as [DateSent] -- Drop time from datetime
, LR.LetterCode -- Letter Id
, M.RecordId -- Record Id
FROM LetterRequest as LR WITH (NOLOCK)
INNER JOIN RecordsTable as M WITH (NOLOCK) ON LR.RecordId = M.RecordId
WHERE ForeignKeyId = @id AND Received = @createdDate
AND LR.Deleted = 0 AND IsNull(LR.ErrorDescription, '') = ''
AND LR.DateProcessed BETWEEN @startDate AND DateAdd(year, 1, @startDate)
AND LR.LetterCode IN ('a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o')
) as T
LEFT OUTER JOIN (
-- Payment Table. Payments that bounce are entered as a negative payment and are accounted for
SELECT PH.RecordId, PH.DatePaid, PH.TotalPaid
FROM PaymentHistory as PH WITH (NOLOCK)
INNER JOIN RecordsTable as M WITH (NOLOCK) ON PH.RecordId = M.RecordId
LEFT OUTER JOIN PaymentHistory as PR WITH (NOLOCK) ON PR.ReverseOfUId = PH.UID
WHERE PH.SomeString LIKE 'P_'
AND PR.UID is NULL
AND PH.DatePaid BETWEEN @startDate AND DateAdd(day, 30, DateAdd(year, 1, @startDate))
AND M.ForeignKeyId = @id AND M.Created = @createdDate
) as P ON T.RecordId = P.RecordId
GROUP BY DateSent, LetterCode
--HAVING Count(*) > @limit
ORDER BY DateSent, LetterCode
SELECT *
FROM #tmpTable
WHERE [Letters Sent] > @limit
DROP TABLE #tmpTable最终结果如下所示：
日期字母代码字母已发送金额已付 2012年1月1日1245 12345.67 2012年1月1日b 2301 1234.56 2012年1月1日c 1312 7894.45 2012年1月1日1455 2345.65 2012年1月1日c 3611 3213.21
我在找出速度下降的位置时遇到了问题，因为一切在查询编辑器中运行都非常快。只有当我将查询移动到存储过程时，它开始需要很长时间才能运行。
我确定它与生成的查询执行计划有关，但是我对SQL的了解还不足以识别可能导致问题的原因。
应该注意的是，查询中使用的所有表都有数百万条记录。
有人可以向我解释为什么作为存储过程运行需要比在查询编辑器中花费更多的时间，并且可以帮助我确定查询的哪一部分作为存储过程运行时可能导致性能问题？
RECOMPILE提示，因为我真的不想在每次运行时都重新编译该查询，并且您链接的文章中提到将参数复制到本地变量等效于usingOPTIMIZE FOR UNKNOWN，这似乎仅在2008年及以后。我想现在我将坚持将参数复制到局部变量，这使我的查询执行时间降低到1-2秒。