SQL joining most recent event by criteria to missing value - sql

Question

I have a SQL table that records interactions and the changes that happen in an interaction by interactionkey, user, group, and skill. I want to find the duration of each of the actions (A,B,C) by the grouping variables. Whenever action C happens, the skillKey is left blank (not NULL) and I need it to take the value of the last Skill in that interaction by the user and group so it is grouped together. The first table is the raw SQL data for 1 interaction, and the second is how I need it to look. Edit: I'm using Microsoft SQL Server Management Studio.
Here's what I have so far but it doesn't account for the last skill the user used in the group and interactionkey so it remains blank and adds it up seperately
SELECT
[InteractionKey],
[User],
[StartTime],
[SkillKey],
[GroupKey],
SUM(CASE WHEN ActionKey = 'A' THEN ActionDuration ELSE 0 END) AS 'ActionADuration',
SUM(CASE WHEN ActionKey = 'B' THEN ActionDuration ELSE 0 END) AS 'ActionBDuration',
SUM(CASE WHEN ActionKey = 'C' THEN ActionDuration ELSE 0 END) AS 'ActionCDuration'
FROM
(SELECT
[ActionKey],
[InteractionKey],
[SkillKey],
[GroupKey],
SUM(ActionDuration) AS 'ActionDuration',
[User],
CAST(StartTime AS DATE)
FROM
[InteractionTable]
GROUP BY
InteractionKey, User, SkillKey, GroupKey, ActionKey,
CAST(StartTime AS DATE)) sub
GROUP BY
InteractionKey, User, Date, SkillKey, GroupKey
ORDER BY
InteractionKey
Raw SQL Server table:
Desired output:

Answer 1

All you need is to prepare another "table" with SkillKey already filled as you need, and then use this table in your query.
I will use CTE in my code,
and it is different for divverent versions of SQL Server.
The first one is preferable, but it's for servers starting with 2012.
If you are on lower version use the second query.
-- for ##version >= 2012
with cte as
(
select *,
case
when SkillKey <> ''
then SkillKey
else lag(SkillKey) over(partition by InteractionKey, [User], GroupKey order by [Date])
end as SkillKey
from InteractionTable
)
SELECT
[InteractionKey],
[User],
[StartTime],
[SkillKey],
[GroupKey],
SUM(CASE WHEN ActionKey = 'A' THEN ActionDuration ELSE 0 END) AS 'ActionADuration',
SUM(CASE WHEN ActionKey = 'B' THEN ActionDuration ELSE 0 END) AS 'ActionBDuration',
SUM(CASE WHEN ActionKey = 'C' THEN ActionDuration ELSE 0 END) AS 'ActionCDuration'
FROM
(SELECT
[ActionKey],
[InteractionKey],
[SkillKey],
[GroupKey],
SUM(ActionDuration) AS 'ActionDuration',
[User],
CAST(StartTime AS DATE)
FROM
cte
GROUP BY
InteractionKey, User, SkillKey, GroupKey, ActionKey,
CAST(StartTime AS DATE)) sub
GROUP BY
InteractionKey, User, Date, SkillKey, GroupKey
ORDER BY
InteractionKey
The second:
-- for ##version >= 2005
with r as
(
select *,
row_number() over(partition by InteractionKey, [User], GroupKey order by [Date]) as rn
from InteractionTable
)
,cte as
(
select r1.*,
case
when r1.SkillKey <> ''
then r1.SkillKey
else r2.SkillKey
end as SkillKey
from r r1
left join r r2
on r1.rn = r2.rn + 1
)
SELECT
[InteractionKey],
[User],
[StartTime],
[SkillKey],
[GroupKey],
SUM(CASE WHEN ActionKey = 'A' THEN ActionDuration ELSE 0 END) AS 'ActionADuration',
SUM(CASE WHEN ActionKey = 'B' THEN ActionDuration ELSE 0 END) AS 'ActionBDuration',
SUM(CASE WHEN ActionKey = 'C' THEN ActionDuration ELSE 0 END) AS 'ActionCDuration'
FROM
(SELECT
[ActionKey],
[InteractionKey],
[SkillKey],
[GroupKey],
SUM(ActionDuration) AS 'ActionDuration',
[User],
CAST(StartTime AS DATE)
FROM
cte
GROUP BY
InteractionKey, User, SkillKey, GroupKey, ActionKey,
CAST(StartTime AS DATE)) sub
GROUP BY
InteractionKey, User, Date, SkillKey, GroupKey
ORDER BY
InteractionKey