SQL Server Query Sudden Run Time spike for some parameters - sql-server

Question

Using Microsoft SQL Server 2019;
I was investigating a Stored Procedure that has suddenly started taking 4 hours to complete for certain parameters passed.
Same SP for these parameters was taking only few minutes before and there is no change in the SP recently.
When I checked, NOT just the SP, the SELECT query inside the SP itself is running long.
When I did some research , I read about parameter sniffing issue.
So I first added OPTION(RECOMPILE) to the last SELECT statement and it started running within 1min for first 2 runs.
But then it started to take 4 hours from 3rd run.
So I changed again I used a variable to hold a value and used it in WHERE condition and it gives me result in 1 min.
Plan 1 of Final SELECT statement - Taking 4 hours to complete - uses a Key Lookup which seems to me as trouble maker
Note : WHERE condition is WHERE PR.PRSId = 1
https://www.brentozar.com/pastetheplan/?id=H1dAPJ5kj
Plan 2 of Final SELECT statement - Taking 1 min to complete
Note : WHERE condition is WHERE PR.PRSId = #PRSId
https://www.brentozar.com/pastetheplan/?id=H1VoI15Ji
From my understanding Parameter sniffing happens on SPs. But this case the SELECT query itself is having issue.
Can you explain on this behavior? Thanks
Full query set below
DECLARE #SDVId INT = 3
IF OBJECT_ID('tempdb..#tmp_DSCF') IS NOT NULL
DROP TABLE #tmp_DSCF;
SELECT
SDSDDS.DataSourceId
,C.CategoryId
,F.[Code]
,PF.PFId
INTO
#tmp_DSCF
FROM
dbo.tbl_SDV AS SDV WITH (NOLOCK)
INNER JOIN dbo.tbl_SDSDDataSource AS SDSDDS WITH (NOLOCK) ON SDSDDS.SDVId = SDV.SDVId
INNER JOIN dbo.tbl_SDC AS SDC WITH (NOLOCK) ON SDC.SDVId = SDV.SDVId
INNER JOIN dbo.tbl_Category AS C WITH (NOLOCK) ON C.CategoryId = SDC.CategoryId
INNER JOIN [dbo].[fnt_SPAF](#SDVId) AS F ON (F.Category = C.Description OR F.Category = '*')
INNER JOIN dbo.tbl_PF AS PF WITH (NOLOCK) ON PF.Description = F.Filter AND PF.CategoryId = C.CategoryId
WHERE
SDV.SDVId = #SDVId;
CREATE CLUSTERED INDEX IX_tmp_DSCF ON #tmp_DSCF(PFId, CategoryId, DataSourceId);
SELECT
PR.PRId
,DSCF.Code AS AttributeName
,PFV.PFVId AS AttributeKey
,PFV.Value AS AttributeValue
FROM
#tmp_DSCF AS DSCF
INNER JOIN [dbo].[tbl_PR] AS PR WITH (NOLOCK) ON PR.DataSourceId = DSCF.DataSourceId AND PR.CategoryId = DSCF.CategoryId
INNER JOIN [dbo].[fnt_PPAtDate](GETDATE()) AS F ON F.ProductId = PR.ProductId AND F.PFId = DSCF.PFId
INNER JOIN [dbo].[tbl_PFV] AS PFV WITH (NOLOCK) ON PFV.PFId = F.PFId AND PFV.PFVId = F.PFVId
WHERE
PR.PRSId = 1
AND PR.ProductId IS NOT NULL

Answer 1

As some comments have mentioned, the server hangs on to the execution plan for a SQL statement in a SP and reuses it. Working out an execution plan for a query takes time. Some SPs are used heavily, rapid-fire, in transactional workloads so saving that time is good.
In your case, when the parameters changed the old execution plan became not-so-useful. You found you can fix this by marking the query for "recompilation" (making a new execution plan based on the current parameters). You can also do it by marking the SP for recompilation.
For a data-extraction query like yours the recompilation time is, duh, trivial compared to the query execution time. So there's no harm to be had for marking either the query or the SP for recompilation.