I have a couple of stored procedures on SQL Server 2005 that I've noticed will suddenly take a significantly long time to complete when invoked from my ASP.NET MVC app running in an IIS6 web farm of four servers. Normal, expected completion time is less than a second; unexpected anomalous completion time is 25-45 seconds. The problem doesn't seem to ever correct itself.
However, if I ALTER the stored procedure (even if I don't change anything in the procedure, except to perhaps add a space to the script created by SSMS Modify command), the completion time reverts to expected completion time.
IIS and SQL Server are running on separate boxes, both running Windows Server 2003 R2 Enterprise Edition. SQL Server is Standard Edition. All machines have dual Xeon E5450 3GHz CPUs and 4GB RAM. SQL Server is accessed using its TCP/IP protocol over gigabit ethernet (not sure what physical medium).
The problem is present from all web servers in the web farm. When I invoke the procedure from a query window in SSMS on my development machine, the procedure completes in normal time. This is strange because I was under the impression that SSMS used the same SqlClient driver as in .NET. When I point my development instance of the web app to the production database, I again get the anomalous long completion time. If my SqlCommand Timeout is too short, I get
System.Data.SqlClient.SqlException:
Timeout expired. The timeout period
elapsed prior to completion of the
operation or the server is not
responding.
Question: Why would performing ALTER on the stored procedure, without actually changing anything in it, restore the completion time to less than a second, as expected?
Edit: To clarify, when the procedure is running slow for the app, it simultaneously runs fine in SSMS with the same parameters. The only difference I can discern is login credentials (next time I notice the behavior, I'll be checking from SSMS with the same creds). The ultimate goal is to get the procs to sustainably run with expected speed without requiring occasional intervention.
Resolution: I wanted to to update this question in case others are experiencing this issue. Following the leads of the answers below, I was able to consistently reproduce this behavior. In order to test, I utilize sp_recompile and pass it one of the susceptible sprocs. I then initiate a website request from my browser that will invoke the sproc with atypical parameters. Lastly, I initiate a website request to a page that invokes the sproc with typical parameters, and observe that the request does not complete because of a SQL timeout on the sproc invocation.
To resolve this on SQL Server 2005, I've added OPTIMIZE FOR hints to my SELECT. The sprocs that were vulnerable all have the "all-in-one" pattern described in this article. This pattern is certainly not ideal but was a necessary trade-off given the timeframe for the project.
