Grouping by date range combined with another field? - sql

Question

In SQL Server 2008, I have something like the following:
Create table #RateHistory (RatePlan char(1), EventDate datetime)
Insert into #RateHistory (RatePlan, EventDate)
VALUES
('a','10/01/2013')
,('a','10/04/2013')
,('a','10/06/2013')
,('a','10/08/2013')
,('b','10/21/2013')
,('b','11/05/2013')
,('b','11/12/2013')
,('b','12/05/2013')
,('a','12/08/2013')
,('a','12/09/2013')
,('a','12/10/2013')
,('a','12/15/2013')
I'd like to see an output like this:
Rateplan MinDate MaxDate
-------- ----------- -----------
a 2013-10-01 2013-10-08
b 2013-10-21 2013-12-05
a 2013-12-08 2013-12-15
(originally this was a bit different, but I believe this result set makes it clearer what I actually need, which is the correct grouping)
Note that RatePlan "a" shows up twice, and that I want it to be grouped separately - once for the 10/1/2013 to 10/8/2013 data, and once for the 12/8/2013 to 12/15/2013 data. I've got the solution I need with this :
-- Get initial row numbers
;with Test as (
Select
*
,RowNumber = ROW_NUMBER() over (order by EventDate)
from #RateHistory
)
-- Get initial row numbers
, Test2 as (
SelecT
Main.RowNumber
,Main.EventDate
,Main.RatePlan
,FollowingRatePlan = Following.RatePlan
,NewGroup =
case
when Main.RatePlan <> Following.RatePlan
-- if Following RatePlan is null, that means this is the last record
or (Following.RatePlan is null )
then Main.EventDate
else null
end
from Test Main
left join Test following
on Following.RowNumber = Main.RowNumber + 1
)
, Test3 as (
select
#RateHistory.RatePlan
,#RateHistory.EventDate
,MaxDate = min(Test2.NewGroup)
from #RateHistory
join Test2
on #RateHistory .RatePlan = Test2.RatePlan
and #RateHistory .EventDate <= Test2.NewGroup
where Test2.NewGroup is not null
group by
#RateHistory.RatePlan
,#RateHistory.EventDate
)
select Rateplan, MinDate = MIN(EventDate) , MaxDate
from Test3
group by RatePlan,MaxDate
...but I'm thinking - there's GOT to be a better, more elegant way of doing this. Thoughts? If nobody has anything better, I'll just go ahead and put this in as an answer...
Thanks!

Answer 1

I can think of a solution using correlated scalar sub-queries. You tell me if it's more elegant. Or better performing.
select distinct
rh0.RatePlan,
(
select min(EventDate)
from RateHistory rh1
where rh1.RatePlan = rh0.RatePlan
and rh1.EventDate <= rh0.EventDate
and not exists
(
select * from RateHistory rh2
where rh2.RatePlan != rh0.RatePlan
and rh2.EventDate > rh1.EventDate
and rh2.EventDate < rh0.EventDate
)
) as mindate,
(
select max(EventDate)
from RateHistory rh1
where rh1.RatePlan = rh0.RatePlan
and rh1.EventDate >= rh0.EventDate
and not exists
(
select * from RateHistory rh2
where rh2.RatePlan != rh0.RatePlan
and rh2.EventDate < rh1.EventDate
and rh2.EventDate > rh0.EventDate
)
) as maxdate
from RateHistory rh0
order by mindate
Check out the SQLFiddle. BTW 2012 has some cool features that could make your version of the query more elegant.