Finding A Time When A Value Changed - sql-server

Question

I am still learning many new things about SQL such as PARTITION BY and CTEs. I am currently working on a query which I have cobbled together from a similar question I found online. However, I can not seem to get it to work as intended.
The problem is as follows -- I have been tasked to show rank promotions in an organization from the begining of 2022 to today. I am working with 2 primary tables, an EMPLOYEES table and a PERIODS table. This periods table captures a snapshot of any given employee each month - including their rank at the time. Each of these months is also assigned a PeriodID (e.g. Jan 2022 = PeriodID 131). Our EMPLOYEE table holds the employees current rank. These ranks are stored as an int (e.g. 1,2,3 with 1 being lowest rank). It is possible for an employee to rank up more than once in any given month.
I have simplified the used query as much as I can for the sake of this problem. Query follows as:
;WITH x AS
(
SELECT
e.EmployeeID, p.PeriodID, p.RankID,
rn = ROW_NUMBER() OVER (PARTITION BY e.EmployeeID ORDER BY p.PeriodID DESC)
FROM employees e
LEFT JOIN periods p on p.EmployeeID= e.EmployeeID
WHERE p.PeriodID <= 131 AND p.PeriodID >=118 --This is the time range mentioned above
),
rest AS (SELECT * FROM x WHERE rn > 1)
SELECT
main.EmployeeID,
PeriodID = MIN(
CASE
WHEN main.CurrentRankID = Rest.RankID
THEN rest.PeriodID ELSE main.PeriodID
END),
main.RankID, rest.RankID
FROM x AS main LEFT OUTER JOIN rest ON main.EmployeeID = rest.EmployeeID
AND rest.rn >1
LEFT JOIN periods p on p.EmployeeID = e.EmployeeID
WHERE main.rn = 1
AND NOT EXISTS
(
SELECT 1 FROM rest AS rest2
WHERE EmployeeID = rest.EmployeeID
AND rn < rest.rn
AND main.RankID <> rest.RankID
)
and p.PeriodID <= 131 AND p.PeriodID >=118
GROUP BY main.EmployeeID, main.PeriodID, main.RankID, rest.RankID
As mentioned before, this query was borrowed from a similar question and modified for my own use. I imagine the bones of the query is good and maybe I have messed up a variable somewhere but I can not seem to locate the problem line. The end goal is for the query to result in a table showing the EmployeeID, PeriodID, the rank they are being promoted from, and the rank they are being promoted to in the month the promotion was earned. Similar to the below.
EmployeeID
PeriodID
PerviousRankID
NewRank
123
131
1
2
123
133
2
3
Instead, my query is spitting out repeating previous/current ranks and the PeriodIDs seem to be static (such as what is shown below).
EmployeeID
PeriodID
PerviousRankID
NewRank
123
131
1
1
123
131
1
1
I am hoping someone with a greater knowledge base on these functions is able to quickly notice my mistake.

Answer 1

If we assume some example DML/DDL (it's really helpful to provide this with your question):
DECLARE #Employees TABLE (EmployeeID INT IDENTITY, Name VARCHAR(20), RankID INT);
DECLARE #Periods TABLE (PeriodID INT, EmployeeID INT, RankID INT);
INSERT INTO #Employees (Name, RankID) VALUES ('Jonathan', 10),('Christopher', 10),('James', 10),('Jean-Luc', 8);
INSERT INTO #Periods (PeriodID, EmployeeID, RankID) VALUES
(1,1,1),(2,1,1),(3,1,1),(4,1,8 ),(5,1,10),(6,1,10),
(1,2,1),(2,2,1),(3,2,1),(4,2,8 ),(5,2,8 ),(6,2,10),
(1,3,1),(2,3,1),(3,3,7),(4,3,10),(5,3,10),(6,3,10),
(1,4,1),(2,4,1),(3,4,1),(4,4,8 ),(5,4,9 ),(6,4,9 )
Then we can accomplish what I think you're looking for using a OUTER APPLY then aggregates the values based on the current-row values:
SELECT e.EmployeeID, e.Name, e.RankID AS CurrentRank, ap.PeriodID AS ThisPeriod, p.PeriodID AS LastRankChangePeriodID, p.RankID AS LastRankChangedFrom, ap.RankID - p.RankID AS LastRankChanged
FROM #Employees e
LEFT OUTER JOIN #Periods ap
ON e.EmployeeID = ap.EmployeeID
OUTER APPLY (
SELECT EmployeeID, MAX(PeriodID) AS PeriodID
FROM #Periods
WHERE EmployeeID = e.EmployeeID
AND RankID <> ap.RankID
AND PeriodID < ap.PeriodID
GROUP BY EmployeeID
) a
LEFT OUTER JOIN #Periods p
ON a.EmployeeID = p.EmployeeID
AND a.PeriodID = p.PeriodID
ORDER BY e.EmployeeID, ap.PeriodID DESC
Using the correlated subquery we get a view of the data which we can filter using the current-row values, and we aggregate that to return the period we're looking for (where it's before this period, and it's not the same rank). Then it's just a join back to the Periods table to get the values.
You used an LEFT JOIN, so I've preserved that using an OUTER APPLY. If you wanted to filter using it, it would be a CROSS APPLY instead.
EmployeeID
Name
CurrentRank
ThisPeriod
LastRankChangePeriodID
LastRankChangedFrom
LastRankChanged
1
Jonathan
10
6
4
8
2
1
Jonathan
10
5
4
8
2
1
Jonathan
10
4
3
1
7
1
Jonathan
10
3
1
Jonathan
10
2
1
Jonathan
10
1
2
Christopher
10
6
5
8
2
2
Christopher
10
5
3
1
7
2
Christopher
10
4
3
1
7
2
Christopher
10
3
2
Christopher
10
2
2
Christopher
10
1
3
James
10
6
3
7
3
3
James
10
5
3
7
3
3
James
10
4
3
7
3
3
James
10
3
2
1
6
3
James
10
2
3
James
10
1
4
Jean-Luc
8
6
5
9
-1
4
Jean-Luc
8
5
4
8
1
4
Jean-Luc
8
4
3
1
7
4
Jean-Luc
8
3
4
Jean-Luc
8
2
4
Jean-Luc
8
1
Now we can see what the previous change looked like for each period. Currently Jonathan is has RankID 10. Last time that was different was in PeriodID 4 when it was 8. The same was true for PeriodID 5. In PeriodID 4 he had RankID 8, and prior to that he had RankID 1. Before that his Rank hadn't changed.
Jean-Luc was actually demoted as his last change. I don't know if this is possible within your model.