SQL Server 2014 - Use previous value when date not present - sql

Question

I asked a similar question yesterday but I was not very good in my description of what I wanted. This will be far clearer.
Lead/Lag is not getting me what I need. Its close, but not enough.
Using SQL Server 2014 for client, actual server built on SQL 2012.
Here is my code:
Creating Team Table
CREATE TABLE ##TeamTable
([UserID] varchar(50), [CurrentTeam] varchar(5), [ChangeDate] datetime)
;
INSERT INTO ##TeamTable
([UserID], [CurrentTeam], [ChangeDate])
VALUES
('User1', 'Team1', '6/1/2016'),
('User1', 'Team2', '9/1/2016'),
('User1', 'Team3', '12/1/2016'),
('User2', 'Team1', '4/1/2016'),
('User2', 'Team2', '10/1/2016'),
('User2', 'Team3', '11/1/2016');
Now to create data table I need to join to
CREATE TABLE ##DataTable
([UserID] varchar(50), Month_sk datetime, Media varchar(50), NCO int)
INSERT INTO ##DataTable
([UserID] , Month_sk , Media , NCO )
VALUES
('User1', '2016-06-01 00:00:00', 'Fax', 100),
('User1', '2016-06-01 00:00:00', 'Voice', 120),
('User1', '2016-07-01 00:00:00', 'Voice', 90),
('User1', '2016-07-01 00:00:00', 'Email', 100),
('User1', '2016-08-01 00:00:00', 'Voice', 150),
('User1', '2016-08-01 00:00:00', 'Email', 100),
('User1', '2016-09-01 00:00:00', 'Voice', 100),
('User1', '2016-09-01 00:00:00', 'Email', 120),
('User1', '2016-10-01 00:00:00', 'Voice', 90),
('User1', '2016-10-01 00:00:00', 'Email', 100),
('User1', '2016-11-01 00:00:00', 'Voice', 150),
('User1', '2016-11-01 00:00:00', 'Email', 100),
('User1', '2016-12-01 00:00:00', 'Voice', 150),
('User1', '2016-12-01 00:00:00', 'Email', 100),
('User2', '2016-04-01 00:00:00', 'Fax', 100),
('User2', '2016-04-01 00:00:00', 'Voice', 120),
('User2', '2016-05-01 00:00:00', 'Fax', 100),
('User2', '2016-05-01 00:00:00', 'Voice', 120),
('User2', '2016-06-01 00:00:00', 'Fax', 100),
('User2', '2016-06-01 00:00:00', 'Voice', 120),
('User2', '2016-07-01 00:00:00', 'Voice', 90),
('User2', '2016-07-01 00:00:00', 'Email', 100),
('User2', '2016-08-01 00:00:00', 'Voice', 150),
('User2', '2016-08-01 00:00:00', 'Email', 100),
('User2', '2016-09-01 00:00:00', 'Voice', 100),
('User2', '2016-09-01 00:00:00', 'Email', 120),
('User2', '2016-10-01 00:00:00', 'Voice', 90),
('User2', '2016-10-01 00:00:00', 'Email', 100),
('User2', '2016-11-01 00:00:00', 'Voice', 150),
('User2', '2016-11-01 00:00:00', 'Email', 100),
('User2', '2016-12-01 00:00:00', 'Voice', 150),
('User2', '2016-12-01 00:00:00', 'Email', 100);
Here is a basic Select to show whats going on:
SELECT b.UserID
,b.Media
,b.NCO
,Month_sk
,CurrentTeam
FROM ##DataTable b
LEFT OUTER JOIN ##TeamTable a on b.UserID = a.UserID and b.Month_sk = a.ChangeDate
order by UserID, Month_sk, media
This gives me a result set that looks like this:
What I need is for where I have nulls, that it would be pulling in the previous team name that's not null. So in User1 case, those 4 nulls for months of July and August would say Team1 since that was the team he was last on. Same for the nulls after Team2, those should say Team2.
Lead/Lag is close or I'm not using it right. Hopefully with all this code, this makes someone's jobs way easier.
UPDATE:
Lag/Lead gives same results. Still need the nulls to fill in
SELECT b.UserID
,b.Media
,b.NCO
,Month_sk
,CurrentTeam
,LAG(CurrentTeam,1, currentteam) OVER(PARTITION BY a.userid, changedate ORDER BY ChangeDate) as Lag
FROM ##DataTable b
LEFT OUTER JOIN ##TeamTable a on b.UserID = a.UserID and b.Month_sk = a.ChangeDate
order by UserID, Month_sk, media

Answer 1

(Moving update notes to end)
I think the easiest solution (conceptually) is to join against all months up to month_sk and then filter to get only the last match. This "feels" potentially inefficient, so you'd want to test it with realistic data volume and if there's a problem then look for something better. (But "something better" may involve changes to the physical data model...)
So:
select userid, media, nco, month_sk, currentteam
from (SELECT b.UserID
, b.Media
, b.NCO
, Month_sk
, CurrentTeam
, rank() over(partition by b.userID
order by a.changeDate desc) n
FROM ##DataTable b
INNER JOIN ##TeamTable a
on b.UserID = a.UserID
and b.Month_sk >= a.ChangeDate
) x
where n = 1
order by UserID, Month_sk, media
Note that in previous versions I used row_number() over() instead of rank() over()... and you can do that, but if you do then you have to include in the partitioning key any data from the b table that could cause a duplication of a row from the a table during the join. Using rank ensures that all such duplicates share their rank as they ought to.
UPDATE - After I initially wrote this, I deleted it because I thought I'd misread your question; but as I was writing a replacement realized I may have had it right in the first place. So here it is, with a caveat:
This assumes that the only reason you get the NULL value is the outer join. If ever the "right hand" table has a row and just a value for a column therein is NULL, then getting the previous value for that column would require further work with subqueries or analytic funcitons. But even then lead/lag may not work, since they are position based. (I think something with LAST_VALUE might be more suitable, but will leave the details of that unless it's needed.)
UPDATE 2 - based on your description of the data model in below comments, I'm changing the query to show an inner join as it sounds like that will work (once you broaden the join criteria) and should be more efficient.
UPDATE 3 - I did misread your sample data and got the partitioning expression for calculating n wrong. Should be fixed assuming the values from the b table are unique. If not it's still fixable but requires more trickery...

Answer 2

You can do this with an APPLY and a sub query like this.
SELECT
userid,
media,
nco,
month_sk,
currentteam
FROM
##DataTable td
OUTER APPLY (
SELECT TOP (1)
CurrentTeam,
ChangeDate
FROM
##TeamTable tt
WHERE
tt.UserID = td.UserID
and tt.ChangeDate <= td.Month_sk
ORDER BY
tt.ChangeDate desc
) dataTableWithTeam
ORDER BY
td.UserID,
td.Month_sk,
td.media

Answer 3

In this version, I first identify the appropriate "linking" month in the CTE, and then use that as a lookup in the final join. (It got much easier once I realized Media and NCO played no real part in the join.)
WITH cteDateLookup
as (
-- Get the ChangeDate for this User/Month
SELECT
b.UserID
,b.Month_sk
,max(a.ChangeDate) ChangeDate
from ##DataTable b
left outer join ##TeamTable a
on b.UserID = a.UserID
and b.Month_sk >= a.ChangeDate
group by
b.UserID
,b.Month_sk
)
-- Use the cte as a "lookup" for the appropriate date
SELECT
b.UserID
,b.Media
,b.NCO
,b.Month_sk
,a.CurrentTeam
from ##DataTable b
left outer join cteDateLookup cte
on cte.UserId = b.UserId
and b.Month_sk = cte.Month_sk
left outer join ##TeamTable a
on a.UserId = cte.UserId
and a.ChangeDate = cte.ChangeDate
order by
b.UserID
,b.Month_sk
,b.media