CTE slow performance on Left join - sql-server

Question

I need to provide a report that shows all users on a table and their scores. Not all users on said table will have a score, so in my solution I calculate the score first using a few CTE's then in a final CTE i pull a full roster and assign a default score to users with no actual score.
While the CTE's are not overly complex, they are also not simple. Separately when I run the calculation part of the CTE's for users with an actual score, it runs in less than a second. When I join to a final CTE that grabs the full roster and assigns default scores where the nulls appear (no actual score) the wheels completely fall off and it never completes.
I've experimented with switching up the indexes and refreshing them to no avail. I have noticed the join at agent_effectiveness when switched to INNER runs in one second, but I need it to be a LEFT join so it will pull in the whole roster even when no score is present.
EDIT*
Execution Plan Inner Join
Execution Plan Left Join
WITH agent_split_stats AS (
Select
racf,
agent_stats.SkillGroupSkillTargetID,
aht_target.EnterpriseName,
aht_target.target,
Sum(agent_stats.CallsHandled) as n_calls_handled,
CASE WHEN (Sum(agent_stats.TalkInTime) + Sum(agent_stats.IncomingCallsOnHoldTime) + Sum(agent_stats.WorkReadyTime)) = 0 THEN 1 ELSE
(Sum(agent_stats.TalkInTime) + Sum(agent_stats.IncomingCallsOnHoldTime) + Sum(agent_stats.WorkReadyTime)) END
AS total_handle_time
from tblAceyusAgntSklGrp as agent_stats
-- GET TARGETS
INNER JOIN tblCrosswalkWghtPhnEffTarget as aht_target
ON aht_target.SgId = agent_stats.SkillGroupSkillTargetID
AND agent_stats.DateTime BETWEEN aht_target.StartDt and aht_target.EndDt
-- GET RACF
INNER JOIN tblAgentMetricCrosswalk as xwalk
ON xwalk.SkillTargetID = agent_stats.SkillTargetID
--GET TAU DATA LIKE START DATE AND GRADUATED FLAG
INNER JOIN tblTauClassList AS T
ON T.SaRacf = racf
WHERE
--FILTERS BY A ROLLING 15 BUSINESS DAYS UNLESS THE DAYS BETWEEN CURRENT DATE AND TAU START DATE ARE <15
agent_stats.DateTime >=
CASE WHEN dbo.fn_WorkDaysAge(TauStart, GETDATE()) <15 THEN TauStart ELSE
dbo.fn_WorkDate15(TauStart)
END
And Graduated = 'No'
--WPE FILTERS TO ENSURE ACCURATE DATA
AND CallsHandled <> 0
AND Target is not null
Group By
racf, agent_stats.SkillGroupSkillTargetID, aht_target.EnterpriseName, aht_target.target
),
agent_split_stats_with_weight AS (
-- calculate weights
-- one row = one advocate + split
SELECT
agent_split_stats.*,
agent_split_stats.n_calls_handled/SUM(agent_split_stats.n_calls_handled) OVER(PARTITION BY agent_split_stats.racf) AS [weight]
FROM agent_split_stats
),
agent_split_effectiveness AS (
-- calculate the raw Effectiveness score for each eligible advocate/split
-- one row = one agent + split, with their raw Effectiveness score and the components of that
SELECT
agent_split_stats_with_weight.*,
-- these are the components of the Effectiveness score
(((agent_split_stats_with_weight.target * agent_split_stats_with_weight.n_calls_handled) / agent_split_stats_with_weight.total_handle_time)*100)*agent_split_stats_with_weight.weight AS effectiveness_sum
FROM agent_split_stats_with_weight
), -- this is where we show effectiveness per split select * from agent_split_effectiveness
agent_effectiveness AS (
-- sum all of the individual effectiveness raw scores for each agent to get each agent's raw score
SELECT
racf AS SaRacf,
ROUND(SUM(effectiveness_sum),2) AS WpeScore
FROM agent_split_effectiveness
GROUP BY racf
),
--GET FULL CLASS LIST, TAU DATES, GOALS FOR WHOLE CLASS
tau AS (
Select L.SaRacf, TauStart, Goal as WpeGoal
,CASE WHEN agent_effectiveness.WpeScore IS NULL THEN 1 ELSE WpeScore END as WpeScore
FROM tblTauClassList AS L
LEFT JOIN agent_effectiveness
ON agent_effectiveness.SaRacf = L.SaRacf
LEFT JOIN tblCrosswalkTauGoal AS G
ON G.Year = TauYear
AND G.Bucket = 'Wpe'
WHERE TermDate IS NULL
AND Graduated = 'No'
)
SELECT tau.*,
CASE WHEN dbo.fn_WorkDaysAge(TauStart, GETDATE()) > 14 --MUST BE AT LEAST 15 DAYS TO PASS
AND WpeScore >= WpeGoal THEN 'Pass'
ELSE 'Fail' END
from tau
This style of query runs fine in 3 other different calculation types (different score types). So i am unsure why its failing so badly here. Actual results should be a list of individuals, a date, a score, a goal and a score. When no score exists, a default score is provided. Additionally there is a pass/fail metric using the score/goal.

Answer 1

As #Habo mentioned, we need the actual execution plan (e.g. run the query with "include actual execution plan" turned on.) I looked over what you posted and there is nothing there that will explain the problem. The difference with the actual plan vs the estimated plan is that the actual number of rows retrieved are recorded; this is vital for troubleshooting poorly performing queries.
That said, I do see a HUGE problem with both queries. It's a problem that, once fixed will, improve both queries to less than a second. Your query is leveraging two scalar user Defined Functions (UDFs): dbo.fn_WorkDaysAge & dbo.fn_WorkDate15. Scalar UDFs ruin
everything. Not only are they slow, they force a serial execution plan which makes any query they are used in much slower.
I don't have the code for dbo.fn_WorkDaysAge or dbo.fn_WorkDate15 I have my own "WorkDays" function which is inline (code below). The syntax is a little different but the performance benefits are worth the effort. Here's the syntax difference:
-- Scalar
SELECT d.*, workDays = dbo.countWorkDays_scalar(d.StartDate,d.EndDate)
FROM <sometable> AS d;
-- Inline version
SELECT d.*, f.workDays
FROM <sometable> AS d
CROSS APPLY dbo.countWorkDays(d.StartDate,d.EndDate) AS f;
Here's a performance test I put together to show the difference between an inline version vs the scalar version:
-- SAMPLE DATA
IF OBJECT_ID('tempdb..#dates') IS NOT NULL DROP TABLE #dates;
WITH E1(x) AS (SELECT 1 FROM (VALUES(1),(1),(1),(1),(1),(1),(1),(1),(1),(1)) AS x(x)),
E3(x) AS (SELECT 1 FROM E1 a, E1 b, E1 c),
iTally AS (SELECT N=ROW_NUMBER() OVER (ORDER BY (SELECT 1)) FROM E3 a, E3 b)
SELECT TOP (100000)
StartDate = CAST(DATEADD(DAY,-ABS(CHECKSUM(NEWID())%1000),GETDATE()) AS DATE),
EndDate = CAST(DATEADD(DAY,+ABS(CHECKSUM(NEWID())%1000),GETDATE()) AS DATE)
INTO #dates
FROM iTally;
-- PERFORMANCE TESTS
PRINT CHAR(10)+'Scalar Version (always serial):'+CHAR(10)+REPLICATE('-',60);
GO
DECLARE #st DATETIME = GETDATE(), #workdays INT;
SELECT #workdays = dbo.countWorkDays_scalar(d.StartDate,d.EndDate)
FROM #dates AS d;
PRINT DATEDIFF(MS,#st,GETDATE());
GO 3
PRINT CHAR(10)+'Inline Version:'+CHAR(10)+REPLICATE('-',60);
GO
DECLARE #st DATETIME = GETDATE(), #workdays INT;
SELECT #workdays = f.workDays
FROM #dates AS d
CROSS APPLY dbo.countWorkDays(d.StartDate,d.EndDate) AS f
PRINT DATEDIFF(MS,#st,GETDATE());
GO 3
Results:
Scalar Version (always serial):
------------------------------------------------------------
Beginning execution loop
380
363
350
Batch execution completed 3 times.
Inline Version:
------------------------------------------------------------
Beginning execution loop
47
47
46
Batch execution completed 3 times.
As you can see - the inline version about 8 times faster than the scalar version. Replacing those scalar UDFs with an inline version will almost certainly speed this query up regardless of join type.
Other problems I see include:
I see a lot of Index scans, this is a sign you need more filtering and/or better indexes.
dbo.tblCrosswalkWghtPhnEffTarget does not have any indexes which means it will always get scanned.
Functions used for performance test:
-- INLINE VERSION
----------------------------------------------------------------------------------------------
IF OBJECT_ID('dbo.countWorkDays') IS NOT NULL DROP FUNCTION dbo.countWorkDays;
GO
CREATE FUNCTION dbo.countWorkDays (#startDate DATETIME, #endDate DATETIME)
/*****************************************************************************************
[Purpose]:
Calculates the number of business days between two dates (Mon-Fri) and excluded weekends.
dates.countWorkDays does not take holidays into considerations; for this you would need a
seperate "holiday table" to perform an antijoin against.
The idea is based on the solution in this article:
https://www.sqlservercentral.com/Forums/Topic153606.aspx?PageIndex=16
[Author]:
Alan Burstein
[Compatibility]:
SQL Server 2005+
[Syntax]:
--===== Autonomous
SELECT f.workDays
FROM dates.countWorkDays(#startdate, #enddate) AS f;
--===== Against a table using APPLY
SELECT t.col1, t.col2, f.workDays
FROM dbo.someTable t
CROSS APPLY dates.countWorkDays(t.col1, t.col2) AS f;
[Parameters]:
#startDate = datetime; first date to compare
#endDate = datetime; date to compare #startDate to
[Returns]:
Inline Table Valued Function returns:
workDays = int; number of work days between #startdate and #enddate
[Dependencies]:
N/A
[Developer Notes]:
1. NULL when either input parameter is NULL,
2. This function is what is referred to as an "inline" scalar UDF." Technically it's an
inline table valued function (iTVF) but performs the same task as a scalar valued user
defined function (UDF); the difference is that it requires the APPLY table operator
to accept column values as a parameter. For more about "inline" scalar UDFs see this
article by SQL MVP Jeff Moden: http://www.sqlservercentral.com/articles/T-SQL/91724/
and for more about how to use APPLY see the this article by SQL MVP Paul White:
http://www.sqlservercentral.com/articles/APPLY/69953/.
Note the above syntax example and usage examples below to better understand how to
use the function. Although the function is slightly more complicated to use than a
scalar UDF it will yield notably better performance for many reasons. For example,
unlike a scalar UDFs or multi-line table valued functions, the inline scalar UDF does
not restrict the query optimizer's ability generate a parallel query execution plan.
3. dates.countWorkDays requires that #enddate be equal to or later than #startDate. Otherwise
a NULL is returned.
4. dates.countWorkDays is NOT deterministic. For more deterministic functions see:
https://msdn.microsoft.com/en-us/library/ms178091.aspx
[Examples]:
--===== 1. Basic Use
SELECT f.workDays
FROM dates.countWorkDays('20180608', '20180611') AS f;
---------------------------------------------------------------------------------------
[Revision History]:
Rev 00 - 20180625 - Initial Creation - Alan Burstein
*****************************************************************************************/
RETURNS TABLE WITH SCHEMABINDING AS RETURN
SELECT workDays =
-- If #startDate or #endDate are NULL then rerturn a NULL
CASE WHEN SIGN(DATEDIFF(dd, #startDate, #endDate)) > -1 THEN
(DATEDIFF(dd, #startDate, #endDate) + 1) --total days including weekends
-(DATEDIFF(wk, #startDate, #endDate) * 2) --Subtact 2 days for each full weekend
-- Subtract 1 when startDate is Sunday and Substract 1 when endDate is Sunday:
-(CASE WHEN DATENAME(dw, #startDate) = 'Sunday' THEN 1 ELSE 0 END)
-(CASE WHEN DATENAME(dw, #endDate) = 'Saturday' THEN 1 ELSE 0 END)
END;
GO
-- SCALAR VERSION
----------------------------------------------------------------------------------------------
IF OBJECT_ID('dbo.countWorkDays_scalar') IS NOT NULL DROP FUNCTION dbo.countWorkDays_scalar;
GO
CREATE FUNCTION dbo.countWorkDays_scalar (#startDate DATETIME, #endDate DATETIME)
RETURNS INT WITH SCHEMABINDING AS
BEGIN
RETURN
(
SELECT workDays =
-- If #startDate or #endDate are NULL then rerturn a NULL
CASE WHEN SIGN(DATEDIFF(dd, #startDate, #endDate)) > -1 THEN
(DATEDIFF(dd, #startDate, #endDate) + 1) --total days including weekends
-(DATEDIFF(wk, #startDate, #endDate) * 2) --Subtact 2 days for each full weekend
-- Subtract 1 when startDate is Sunday and Substract 1 when endDate is Sunday:
-(CASE WHEN DATENAME(dw, #startDate) = 'Sunday' THEN 1 ELSE 0 END)
-(CASE WHEN DATENAME(dw, #endDate) = 'Saturday' THEN 1 ELSE 0 END)
END
);
END
GO
UPDATE BASED ON OP'S QUESTION IN THE COMMENTS:
First for the inline table valued function version of each function. Note that I'm using my own tables and don't have time to make the names match your environment but I did my best to include comments in the code. Also note that if, in your function, workingday = '1' is simply pulling weekdays then you'll find my function above to be a much faster alternative to your dbo.fn_WorkDaysAge function. If workingday = '1' also filters out holidays then it won't work.
CREATE FUNCTION dbo.fn_WorkDaysAge_itvf
(
#first_date DATETIME,
#second_date DATETIME
)
RETURNS TABLE AS RETURN
SELECT WorkDays = COUNT(*)
FROM dbo.dimdate -- DateDimension
WHERE DateValue -- [date]
BETWEEN #first_date AND #second_date
AND IsWeekend = 0 --workingday = '1'
GO
CREATE FUNCTION dbo.fn_WorkDate15_itvf
(
#TauStartDate DATETIME
)
RETURNS TABLE AS RETURN
WITH DATES AS
(
SELECT
ROW_NUMBER() OVER(Order By DateValue Desc) as RowNum, DateValue
FROM dbo.dimdate -- DateDimension
WHERE DateValue BETWEEN #TauStartDate AND --GETDATE() testing below
CASE WHEN GETDATE() < #TauStartDate + 200 THEN GETDATE() ELSE #TauStartDate + 200 END
AND IsWeekend = 0 --workingday = '1'
)
--Get the 15th businessday from the current date
SELECT DateValue
FROM DATES
WHERE RowNum = 16;
GO
Now, to replace your scalar UDFs with the inline table valued functions, you would do this (note my comments):
WITH agent_split_stats AS (
Select
racf,
agent_stats.SkillGroupSkillTargetID,
aht_target.EnterpriseName,
aht_target.target,
Sum(agent_stats.CallsHandled) as n_calls_handled,
CASE WHEN (Sum(agent_stats.TalkInTime) + Sum(agent_stats.IncomingCallsOnHoldTime) + Sum(agent_stats.WorkReadyTime)) = 0 THEN 1 ELSE
(Sum(agent_stats.TalkInTime) + Sum(agent_stats.IncomingCallsOnHoldTime) + Sum(agent_stats.WorkReadyTime)) END
AS total_handle_time
from tblAceyusAgntSklGrp as agent_stats
INNER JOIN tblCrosswalkWghtPhnEffTarget as aht_target
ON aht_target.SgId = agent_stats.SkillGroupSkillTargetID
AND agent_stats.DateTime BETWEEN aht_target.StartDt and aht_target.EndDt
INNER JOIN tblAgentMetricCrosswalk as xwalk
ON xwalk.SkillTargetID = agent_stats.SkillTargetID
INNER JOIN tblTauClassList AS T
ON T.SaRacf = racf
-- INLINE FUNCTIONS HERE:
CROSS APPLY dbo.fn_WorkDaysAge_itvf(TauStart, GETDATE()) AS wd
CROSS APPLY dbo.fn_WorkDate15_itvf(TauStart) AS w15
-- NEW WHERE CLAUSE:
WHERE agent_stats.DateTime >=
CASE WHEN wd.workdays < 15 THEN TauStart ELSE w15.workdays END
And Graduated = 'No'
AND CallsHandled <> 0
AND Target is not null
Group By
racf, agent_stats.SkillGroupSkillTargetID, aht_target.EnterpriseName, aht_target.target
),
agent_split_stats_with_weight AS (
SELECT
agent_split_stats.*,
agent_split_stats.n_calls_handled/SUM(agent_split_stats.n_calls_handled) OVER(PARTITION BY agent_split_stats.racf) AS [weight]
FROM agent_split_stats
),
agent_split_effectiveness AS
(
SELECT
agent_split_stats_with_weight.*,
(((agent_split_stats_with_weight.target * agent_split_stats_with_weight.n_calls_handled) /
agent_split_stats_with_weight.total_handle_time)*100)*
agent_split_stats_with_weight.weight AS effectiveness_sum
FROM agent_split_stats_with_weight
),
agent_effectiveness AS
(
SELECT
racf AS SaRacf,
ROUND(SUM(effectiveness_sum),2) AS WpeScore
FROM agent_split_effectiveness
GROUP BY racf
),
tau AS
(
SELECT L.SaRacf, TauStart, Goal as WpeGoal
,CASE WHEN agent_effectiveness.WpeScore IS NULL THEN 1 ELSE WpeScore END as WpeScore
FROM tblTauClassList AS L
LEFT JOIN agent_effectiveness
ON agent_effectiveness.SaRacf = L.SaRacf
LEFT JOIN tblCrosswalkTauGoal AS G
ON G.Year = TauYear
AND G.Bucket = 'Wpe'
WHERE TermDate IS NULL
AND Graduated = 'No'
)
SELECT tau.*,
-- NEW CASE STATEMENT HERE:
CASE WHEN wd.workdays > 14 AND WpeScore >= WpeGoal THEN 'Pass' ELSE 'Fail' END
from tau
-- INLINE FUNCTIONS HERE:
CROSS APPLY dbo.fn_WorkDaysAge_itvf(TauStart, GETDATE()) AS wd
CROSS APPLY dbo.fn_WorkDate15_itvf(TauStart) AS w15;
Note that I can't test this right now but it should be correct (or close)

Answer 2

UPDATE
I accepted Alan's answer, i ended up doing the following. Posting examples hoping the formatting helps someone, it slowed me down a bit...or maybe I am just slow heh heh.
1. Changed my Scalar UDF to InlineTVF
SCALAR Function 1-
ALTER FUNCTION [dbo].[fn_WorkDaysAge]
(
-- Add the parameters for the function here
#first_date DATETIME,
#second_date DATETIME
)
RETURNS int
AS
BEGIN
-- Declare the return variable here
DECLARE #WorkDays int
-- Add the T-SQL statements to compute the return value here
SELECT #WorkDays = COUNT(*)
FROM DateDimension
WHERE Date BETWEEN #first_date AND #second_date
AND workingday = '1'
-- Return the result of the function
RETURN #WorkDays
END
iTVF function 1-
ALTER FUNCTION [dbo].[fn_iTVF_WorkDaysAge]
(
-- Add the parameters for the function here
#FirstDate as Date,
#SecondDate as Date
)
RETURNS TABLE AS RETURN
SELECT WorkDays = COUNT(*)
FROM DateDimension
WHERE Date BETWEEN #FirstDate AND #SecondDate
AND workingday = '1'
I then updated my next function the same way. I added the CROSS APPLY (something ive personally not used, im still a newbie) as indicated below and replaced the UDFs with the field names in my case statement.
Old Code
INNER JOIN tblTauClassList AS T
ON T.SaRacf = racf
WHERE
--FILTERS BY A ROLLING 15 BUSINESS DAYS UNLESS THE DAYS BETWEEN CURRENT DATE AND TAU START DATE ARE <15
agent_stats.DateTime >=
CASE WHEN dbo.fn_WorkDaysAge(TauStart, GETDATE()) <15 THEN TauStart ELSE
dbo.fn_WorkDate15(TauStart)
END
New Code
INNER JOIN tblTauClassList AS T
ON T.SaRacf = racf
--iTVFs
CROSS APPLY dbo.fn_iTVF_WorkDaysAge(TauStart, GETDATE()) as age
CROSS APPLY dbo.fn_iTVF_WorkDate_15(TauStart) as roll
WHERE
--FILTERS BY A ROLLING 15 BUSINESS DAYS UNLESS THE DAYS BETWEEN CURRENT DATE AND TAU START DATE ARE <15
agent_stats.DateTime >=
CASE WHEN age.WorkDays <15 THEN TauStart ELSE
roll.Date
END
New code runs in 3-4 seconds. I will go back and index the appropriate tables per your recommendation and probably gain more efficiency there.
Cannot thank you enough!