参数值不同导致查询慢
我在 SQL Server 2008 中有一个存储过程。它基本上构建一个字符串，然后使用 EXEC() 运行查询：
SELECT * FROM [dbo].[StaffRequestExtInfo] WITH(nolock,readuncommitted)
WHERE [NoteDt] < @EndDt
AND [NoteTypeCode] = @RequestTypeO
AND ([FNoteDt] >= @StartDt AND [FNoteDt] <= @EndDt)
AND [FStaffID] = @StaffID
AND [FNoteTypeCode]<>@RequestTypeC
ORDER BY [LocName] ASC,[NoteID] ASC,[CNoteDt] ASC
除了 @RequestTypeO 和 @RequestTypeF 之外的所有内容都作为存储过程参数传入。另外两个是从参数构建到局部变量中的。通常，查询运行时间不到一秒。然而，对于 @StaffID 的一个特定值，执行计划是不同的，并且速度慢了大约 30 倍。在这两种情况下，返回的数据量通常是相同的，但执行时间会增加。
我尝试重新编译存储过程。我还尝试将@StaffID“复制”到本地@LocalStaffID。这两种方法都没有任何区别。
有什么想法吗？
更新：尝试使用以下方法删除特定计划：
DECLARE @ph VARBINARY(64), @pt VARCHAR(128), @sql VARCHAR(1024)
DECLARE cur CURSOR FAST_FORWARD FOR
SELECT p.plan_handle
FROM sys.[dm_exec_cached_plans] p
CROSS APPLY sys.dm_exec_sql_text(p.plan_handle) t
WHERE t.text LIKE N'%cms_selectStaffRequests%'
OPEN cur
FETCH NEXT FROM cur INTO @ph
WHILE @@FETCH_STATUS = 0
BEGIN
SELECT @pt = master.dbo.fn_varbintohexstr(@ph)
PRINT 'DBCC FREEPROCCACHE(' + @pt + ')'
SET @sql = 'DBCC FREEPROCCACHE(' + @pt + ')'
EXEC(@sql)
FETCH NEXT FROM cur INTO @ph
END
CLOSE cur
DEALLOCATE cur
要么删除了错误的计划，要么最终重新创建了相同的计划，但没有效果。
I have an sproc in SQL Server 2008. It basically builds a string, and then runs the query using EXEC():
SELECT * FROM [dbo].[StaffRequestExtInfo] WITH(nolock,readuncommitted)
WHERE [NoteDt] < @EndDt
AND [NoteTypeCode] = @RequestTypeO
AND ([FNoteDt] >= @StartDt AND [FNoteDt] <= @EndDt)
AND [FStaffID] = @StaffID
AND [FNoteTypeCode]<>@RequestTypeC
ORDER BY [LocName] ASC,[NoteID] ASC,[CNoteDt] ASC
All but @RequestTypeO and @RequestTypeF are passed in as sproc parameters. The other two are built from a parameter into local variables. Normally, the query runs under one second. However, for one particular value of @StaffID, the execution plan is different and about 30x slower. In either case, the amount of data returned is generally the same, but execution time goes way up.
I tried to recompile the sproc. I also tried to "copy" @StaffID into a local @LocalStaffID. Neither approach made any difference.
Any ideas?
UPDATE: Tried to drop specific plans using:
DECLARE @ph VARBINARY(64), @pt VARCHAR(128), @sql VARCHAR(1024)
DECLARE cur CURSOR FAST_FORWARD FOR
SELECT p.plan_handle
FROM sys.[dm_exec_cached_plans] p
CROSS APPLY sys.dm_exec_sql_text(p.plan_handle) t
WHERE t.text LIKE N'%cms_selectStaffRequests%'
OPEN cur
FETCH NEXT FROM cur INTO @ph
WHILE @@FETCH_STATUS = 0
BEGIN
SELECT @pt = master.dbo.fn_varbintohexstr(@ph)
PRINT 'DBCC FREEPROCCACHE(' + @pt + ')'
SET @sql = 'DBCC FREEPROCCACHE(' + @pt + ')'
EXEC(@sql)
FETCH NEXT FROM cur INTO @ph
END
CLOSE cur
DEALLOCATE cur
Either the wrong plans were dropped, or the same plans ended up being recreated, but it had no effect.
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（2）
检查 FStaffID 列中值的分布/频率/基数，并检查您的索引。可能您有一名员工承担 50% 的工作（可能是 DBA:)，这可能会改变优化器选择使用哪些索引以及读取数据的方式。
或者，动态代码生成的执行计划可能会被保存并重新使用，从而导致查询性能不佳（如 HLGEM 所说）。我不太了解细节，但是 SQL 2008 在执行此操作时比它的前辈有更多的方式让您感到困惑。
Check the distribution/frequency/cardinality of the values in column FStaffID, and review your indexes. It may be that you have one staff member doing 50% of the work (probably the DBA :) and that may change how the optimizer chooses which indexes to use and how the data is read.
Alternatively, the execution plan generated by the dynamic code may be being saved and re-used, resulting in a poorly performing query (like HLGEM says). I'm not up on the details, but SQL 2008 has more ways to confuse you while doing this than its predecessors.
在查询中的主基表上执行
UPDATE STATISTICS ...WITH FULLSCAN会导致“慢”值与慢计划不相关。Doing an
UPDATE STATISTICS ... WITH FULLSCANon the main base table in the query resulted in the "slow" value not being associated with a slow plan.