Please refer to the following schema at SQL Fiddle when reading the use case below.
In a timekeeping application I'm working with, there is a business process where individuals are entitled to a differential pay bonus if they are scheduled to work in an evening or night shift. The same bonuses are applied if a benefit paycode like Vacation or Personal time is applied.
In those circumstances, their paycode balances are applied to to times they are normally scheduled to work. Sometimes, 2+ codes may need to be applied if balances are low so you may have one benefit code, like Vacation, applied to the beginning of the work day, and then another code, like Personal Time, applied to the remainder.
It becomes more complex when you also have to accommodate for a meal break, effectively splitting 2+ benefit balances across 2+ shift segments.
A real world example, described in the SQL Fiddle above...
- I work 8 hours daily, 9pm-6am in two segments; 9pm-12am and 1am-6am.
- I take the day off on 2/5/21 and it's put on my schedule for the usual 9pm-6am slot.
- I have only 6 hours of Vacation available, so I also apply 2 hours of Personal time to get my full 8 hours of pay.
- My benefit codes would be distributed as follows (this is the result set I want)
- 9am-12pm: Vacation (all of Segment 1)
- 1pm-4pm: Vacation (part of Segment 2)
- 4pm-6pm: Personal (remainder of Segment 2)
My initial thought is that if it were just one paycode (i.e. Vacation) to apply, I can just use a windowing function to allocate a running total/declining balance to each shift segment until I ran out of paycode time.
However, because I may need to apply more than one paycode, decision logic has to be applied to switch to the new code and its available balance to resume the allocation at where the previous code ran out of time.
In the SQL Fiddle you'll see I'm leaning towards using an Allocations table to keep track of how I divvy up the paycode allocations across the segments and use subqueries to keep track of the running PayCode and Shift Segment balances.
Where I'm struggling is on the "looping mechanism" to check the balance of both the current paycode and current segment to see when I have to shift to the next segment and/or paycode.
I won't ask for an exact T-SQL solution but would like a reference to a comparable solution or guidance regarding what should work (i.e., use a CTE to refer to the prior calculated allocations). I just want to be sure I can avoid a cursor for this use case, partly to avoid the antiquity, but more importantly for scalability as several hundred employees in a 40k+ workforce could be eligible for this on any given workday.
EDIT 1: Copied DDL/DML script from SQL Fiddle
/*SCENARIO-I normally work the night shift for my 24/7/365 business.-My hours are 9pm-6am; 8 hours of work, 9pm-12am and 1am-6am time segments, with 1 hour unpaid meal break in-between-Per business policy, from 7pm-11pm, personnel get paid a little extra in an evening differential-From 11pm-7am, personnel get paid a little more with a night differential-Accurate logging of hours by the correct benefit code is essential for the correct differential code to be applied in time for payroll-I'm taking the night off on 2/5/21-I have 6 hours of Vacation left and part of a Personal Day in my accrual buckets-My manager is told by Payroll to 1) post 6 hours of Vacation starting @ 2/5/21 @ 9pm 2) post 2 hours of Personal starting @ 2/5/21 @ 9pm to fill out the 8 hours for the shift-The application is supposed to do the following on payroll daya) Apply 3 hours of Vacation between 9pm and 12am for the segment before my meal breakb) Apply the remaining three hours of Vacation between 1am and 4am (the first half of the second segment)c) Apply the two hours of Personal to the remainder of the second segment (4am-6am)d) Apply matching differential codes based on the time each benefit code was in the evening an night zoneThis scenario is to address parts a) through c)*/CREATE TABLE PayCodes ( PayCodeID int IDENTITY(1,1) NOT NULL PRIMARY KEY, PersonNum nvarchar(6), EventDate date, PaycodeName nvarchar(50), PaycodeStart datetime2(0), PaycodeDuration int, PaycodeEventRowNumber int) --this last column is to determine the order of processing within the Event Date for a Person/*For our use case and to ensure the timekeeping system doesn't treat the two codes as being part of two seperate shifts Both benefit paycodes have the same start time, 9pm on 2/5*/INSERT INTO PayCodes(PersonNum,EventDate,PaycodeName,PaycodeStart,PaycodeDuration) VALUES ('163613','2021-02-05','Vacation','2021-02-05 21:00',6*3600)INSERT INTO PayCodes(PersonNum,EventDate,PaycodeName,PaycodeStart,PaycodeDuration)VALUES ('163613','2021-02-05','Vacation 2','2021-02-05 21:00',2*3600)/* The larger duration gets the higher priority*/UPDATE PayCodesSET PaycodeEventRowNumber = RNFROM PayCodesINNER JOIN (SELECT PaycodeID, ROW_NUMBER() OVER (PARTITION BY PersonNum, EventDate ORDER BY PayCodeStart, PayCodeDuration DESC) RN FROM PayCodes) RNSON PayCodes.PayCodeID=RNS.PayCodeID/* Segments for this subroutine will normally be created by selecting all paycodes with identical starttimes for the same person computing the total duration across them Then running the total duration through the segment splitter (not in this sample) to split the duration on approved meal breaks*/SET NOCOUNT ONCREATE TABLE Segments (SegmentID int IDENTITY(1,1) NOT NULL PRIMARY KEY, PersonNum nvarchar(6), EventDate date, SegmentStart datetime2(0), SegmentEnd datetime2(0), SegmentEventRowNumber int) --this last column is to determine the order of processing within the Event Date for a Person--Before the meal breakINSERT INTO Segments(PersonNum,EventDate,SegmentStart,SegmentEnd)VALUES ('163613','2021-02-05','2021-02-05 21:00','2021-02-06 00:00')--After the meal breakINSERT INTO Segments(PersonNum,EventDate,SegmentStart,SegmentEnd)VALUES ('163613','2021-02-05','2021-02-06 01:00','2021-02-06 06:00')/* The segments are sorted in chronological order within the event date for a person- In my case, my shift is a 2/5 shift even though I crossed over into 2/6; the event date keeps the segments together*/UPDATE SegmentsSET SegmentEventRowNumber = RNFROM SegmentsINNER JOIN (SELECT SegmentID, ROW_NUMBER() OVER (PARTITION BY PersonNum,EventDate ORDER BY SegmentStart) RN FROM Segments) RNSON Segments.SegmentID=RNS.SegmentID/*I figured I needed an allocation table to keep track of how I'm distributing the benefit code time*/CREATE TABLE AllocatedPaycodes( AllocatedPaycodeID int IDENTITY(1,1) NOT NULL PRIMARY KEY, SegmentID int, PaycodeID int, DurationInSec int)/* CHALLENGE/PROBLEMHow do I write a subroutine to 1) loop through all the paycodes for a PersonNum/EventDate pairing 2) loop through all the segments for the same PersonNum/EventDate pairing, a) applying the *available* amount of paycode duration balance b) against the *available* amount of segment time 3) essentially, keeping track of prior allocations both for the same segment and for the same paycode 4) and hopefully avoid using a cursor*/