SQL and Temporal data - sql

Question

Given a table of appointments, like this:
User Start End
UserA 2016-01-15 12:00:00 2016-01-15 14:00:00
UserA 2016-01-15 15:00:00 2016-01-15 17:00:00
UserB 2016-01-15 13:00:00 2016-01-15 15:00:00
UserB 2016-01-15 13:32:00 2016-01-15 15:00:00
UserB 2016-01-15 15:30:00 2016-01-15 15:30:00
UserB 2016-01-15 15:45:00 2016-01-15 16:00:00
UserB 2016-01-15 17:30:00 2016-01-15 18:00:00
I want to create a list of distinct time intervals in which the same amount of people have an appointment:
Start End Count
2016-01-15 12:00:00 2016-01-15 13:00:00 1
2016-01-15 13:00:00 2016-01-15 14:00:00 2
2016-01-15 14:00:00 2016-01-15 15:45:00 1
2016-01-15 15:45:00 2016-01-15 16:00:00 2
2016-01-15 16:00:00 2016-01-15 17:00:00 1
2016-01-15 17:00:00 2016-01-15 17:30:00 0
2016-01-15 17:30:00 2016-01-15 18:00:00 1
How would I do this in SQL, preferably SQL Server 2008?
EDIT: To clarify: Manually, the result is obtained by making one row for each user, marking the blocked time, and then summing up the count of rows that have a mark:
Time 12 13 14 15 16 17
UserA xxxxxxxx xxxxxxxx
UserB xxxxxxxx x xx
Count 1 2 1 21 0 1
That result set would start at the minimum time available, end at the maximum time available, and while the ASCII art has only a 15min resolution, I would require at least resolution to the minute. I guess you can leave the rows with "0" out of the result, if this is easier for you.

Answer 1

There's got to be an easier way than this, but at least you can probably follow each step individually:
declare #t table ([User] varchar(19) not null,Start datetime2 not null,[End] datetime2 not null)
insert into #t([User], Start, [End]) values
('UserA','2016-01-15T12:00:00','2016-01-15T14:00:00'),
('UserA','2016-01-15T15:00:00','2016-01-15T17:00:00'),
('UserB','2016-01-15T13:00:00','2016-01-15T15:00:00'),
('UserB','2016-01-15T13:32:00','2016-01-15T15:00:00'),
('UserB','2016-01-15T15:30:00','2016-01-15T15:30:00'),
('UserB','2016-01-15T15:45:00','2016-01-15T16:00:00'),
('UserB','2016-01-15T17:30:00','2016-01-15T18:00:00')
;With Times as (
select Start as Point from #t
union
select [End] from #t
), Ordered as (
select Point,ROW_NUMBER() OVER (ORDER BY Point) as rn
from Times
), Periods as (
select
o1.Point as Start,
o2.Point as [End]
from
Ordered o1
inner join
Ordered o2
on
o1.rn = o2.rn - 1
), UserCounts as (
select p.Start,p.[End],COUNT(distinct [User]) as Cnt,ROW_NUMBER() OVER (Order BY p.[Start]) as rn
from
Periods p
left join
#t t
on
p.Start < t.[End] and
t.Start < p.[End]
group by
p.Start,p.[End]
), Consolidated as (
select uc.*
from
UserCounts uc
left join
UserCounts uc_anti
on
uc.rn = uc_anti.rn + 1 and
uc.Cnt = uc_anti.Cnt
where
uc_anti.Cnt is null
union all
select c.Start,uc.[End],c.Cnt,uc.rn
from
Consolidated c
inner join
UserCounts uc
on
c.Cnt = uc.Cnt and
c.[End] = uc.Start
)
select
Start,MAX([End]) as [End],Cnt
from
Consolidated
group by
Start,Cnt
order by Start
CTEs are - Times - since any given start or end stamp can start or end a period in the final results, we just get them all in one column - so the Ordered can number them, and so that Periods can then re-assembly them into each smallest possible period.
UserCounts then goes back to the original data and finds out how many Users where overlapped by each calculated period.
Consolidated is the trickiest CTE to follow, but it's basically merging periods that abut each other where the user count is equal.
Results:
Start End Cnt
--------------------------- --------------------------- -----------
2016-01-15 12:00:00.0000000 2016-01-15 13:00:00.0000000 1
2016-01-15 13:00:00.0000000 2016-01-15 14:00:00.0000000 2
2016-01-15 14:00:00.0000000 2016-01-15 15:45:00.0000000 1
2016-01-15 15:45:00.0000000 2016-01-15 16:00:00.0000000 2
2016-01-15 16:00:00.0000000 2016-01-15 17:00:00.0000000 1
2016-01-15 17:00:00.0000000 2016-01-15 17:30:00.0000000 0
2016-01-15 17:30:00.0000000 2016-01-15 18:00:00.0000000 1
(And I even got the zero row I was unsure I'd be able to conjure into existence)

Answer 2

This kind of query is much easier to write if you have a calendar table. But in this example I've built one on the fly using a recursive CTE. The CTE returns the appointment blocks, which we can then join to the appointment data. I couldn't determine the interval pattern in your sample data, so I've shown the results in blocks of one hour. You could modify this section, or define your own within a second table.
Sample Data
/* Table variables make sharing data easier
*/
DECLARE #Sample TABLE
(
[User] VARCHAR(50),
[Start] DATETIME,
[End] DATETIME
)
;
INSERT INTO #Sample
(
[User],
[Start],
[End]
)
VALUES
('UserA', '2016-01-15 12:00:00', '2016-01-15 14:00:00'),
('UserA', '2016-01-15 15:00:00', '2016-01-15 17:00:00'),
('UserB', '2016-01-15 13:00:00', '2016-01-15 15:00:00'),
('UserB', '2016-01-15 13:32:00', '2016-01-15 15:00:00'),
('UserB', '2016-01-15 15:30:00', '2016-01-15 15:30:00'),
('UserB', '2016-01-15 15:45:00', '2016-01-15 16:00:00'),
('UserB', '2016-01-15 17:30:00', '2016-01-15 18:00:00')
;
I've used two variables to limit the returned results to just those appointments that fall within the given start and end point.
/* Set an start and end point for the next query
*/
DECLARE #Start DATETIME = '2016-01-15 12:00:00';
DECLARE #End DATETIME = '2016-01-15 18:00:00';
WITH Calendar AS
(
/* Anchor returns start of first appointment
*/
SELECT
#Start AS [Start],
DATEADD(SECOND, -1, DATEADD(HOUR, 1, #Start)) AS [End]
UNION ALL
/* Recursion, keep adding new records until end of last appointment
*/
SELECT
DATEADD(HOUR, 1, [Start]) AS [Start],
DATEADD(HOUR, 1, [End]) AS [End]
FROM
Calendar
WHERE
[End] <= #End
)
SELECT
c [Start],
c [End],
COUNT(DISTINCT s [User]) AS [Count]
FROM
Calendar AS c
LEFT OUTER JOIN #Sample AS s ON s [Start] BETWEEN c [Start] AND c [End]
OR s [End] BETWEEN c [Start] AND c [End]
GROUP BY
c [Start],
c [End]
;
Because an appointment can exceed one hour it may contribute to more than one row. This explains why 7 sample rows leads to a returned total of 9.