Query to select same event code with at least one hour interval - sql

Question

I have a sample table
CREATE TABLE [dbo].[wt](
[id] [int] NULL,
[dt] [datetime] NULL,
[txt] [nvarchar](50) NULL
) ON [PRIMARY]
GO
INSERT INTO [dbo].[wt]
([id]
,[dt]
,[txt])
VALUES
(1, '2017-01-01 00:01:00.000', 't1'),
(2, '2017-01-01 00:03:00.000', 't1'),
(3, '2017-01-01 00:02:00.000', 't1'),
(4, '2017-01-01 01:04:00.000', 't1'),
(5, '2017-01-01 02:10:00.000', 't1'),
(6, '2017-01-01 00:01:00.000', 't1'),
(7, '2017-01-01 01:05:00.000', 't1'),
(8, '2017-01-01 02:10:00.000', 't2'),
(9, '2017-01-01 00:03:00.000', 't2'),
(10,'2017-01-01 01:04:00.000', 't2'),
(11,'2017-01-01 00:52:00.000', 't1')
I would like to have a list of txt code and dt date grouped by txt code where interval beetwen txt occurrence is at least one hour and nothing in-between.
To clarify when t1 first occures at '2017-01-01 00:01:00.000'
then next occurrence I am looking for is after at least one hour
which will be '2017-01-01 01:04:00.000'
third occurrence I am looking for is after at least one hour from '2017-01-01 01:04:00.000' and so on.
After some searching I found something like this
;with a as (
select txt, dt,
rn = row_number() over (partition by txt order by dt asc)
from [wt]),
b as (
select txt, dt, dt as dt2, rn, null tm, 0 recurrence
from a
where rn = 1
union all
select a.txt, a.dt, a.dt,
a.rn, datediff(MINUTE,a.dt,b.dt) tm,
case when dateadd(MINUTE,-60,a.dt) < b.dt then recurrence + 1 else 0 end
from b join a
on b.rn = a.rn - 1 and b.txt = a.txt
)
select txt, dt, rn, tm, recurrence
from b
where recurrence = 0
order by txt, dt
but this wasn't good because the interval isn't counted from first occurrence but from last, so I got
txt dt rn tm recurrence
t1 2017-01-01 00:01:00.000 1 NULL 0
t1 2017-01-01 02:10:00.000 8 -65 0
t2 2017-01-01 00:03:00.000 1 NULL 0
t2 2017-01-01 01:04:00.000 2 -61 0
t2 2017-01-01 02:10:00.000 3 -66 0
I think I found a workaround because in this case I could group record within same hour but I am not happy with that solution.
select txt, min(dt) dt
into #ttwt
from [wt]
group by txt, substring(convert(varchar,dt,120),1,14)+'00:00.000'
;with a as (
select txt, dt,
rn = row_number() over (partition by txt order by dt asc)
from #ttwt),
b as (
select txt, dt, dt as dt2, rn, null tm, 0 recurrence
from a
where rn = 1
union all
select a.txt, a.dt, a.dt,
a.rn, datediff(MINUTE,a.dt,b.dt) tm,
case when dateadd(MINUTE,-60,a.dt) < b.dt then recurrence + 1 else 0 end
from b join a
on b.rn = a.rn - 1 and b.txt = a.txt
)
select txt, dt, rn, tm, recurrence
from b
where recurrence = 0
order by txt, dt
drop table #ttwt
txt dt rn tm recurrence
t1 2017-01-01 00:01:00.000 1 NULL 0
t1 2017-01-01 01:04:00.000 2 -63 0
t1 2017-01-01 02:10:00.000 3 -66 0
t2 2017-01-01 00:03:00.000 1 NULL 0
t2 2017-01-01 01:04:00.000 2 -61 0
t2 2017-01-01 02:10:00.000 3 -66 0
Any suggestions to improve the script so it will let the interval be any entered value in minutes would be appreciated.

Answer 1

If I have understood correctly I think the following does what you need.
CREATE TABLE #T (id INT , rn INT, txt VARCHAR(10), dt DATETIME, lagDiff INT, runningDiff INT)
INSERT INTO #T (id, rn, txt, dt, lagDiff, runningDiff)
SELECT id
, ROW_NUMBER() OVER( PARTITION BY txt ORDER BY dt, id) -1 rn
, txt
, dt
, DATEDIFF(MINUTE, COALESCE(LAG(dt) OVER( PARTITION BY txt ORDER BY dt, id), dt), dt) Diff
, DATEDIFF(MINUTE, COALESCE(FIRST_VALUE(dt) OVER( PARTITION BY txt ORDER BY dt, id), dt), dt) RunningDiff
FROM wt
; WITH CTE AS (
SELECT *, 1 AS Level
FROM #T
WHERE rn = 0
UNION ALL
SELECT T.*, CTE.Level + 1
FROM #T T
INNER JOIN CTE ON CTE.txt = T.txt AND CTE.rn < T.rn AND T.runningDiff - 60 > CTE.runningDiff
WHERE T.rn > 0
)
, X AS (
SELECT txt
, Level
, MIN(rn) rn
FROM CTE
GROUP BY txt, Level
)
SELECT #T.*
FROM X
INNER JOIN #T ON #T.txt = X.txt AND #T.rn = X.rn
Output
+----+----+-----+-------------------------+---------+-------------+
| id | rn | txt | dt | lagDiff | runningDiff |
+----+----+-----+-------------------------+---------+-------------+
| 1 | 0 | t1 | 2017-01-01 00:01:00.000 | 0 | 0 |
| 4 | 5 | t1 | 2017-01-01 01:04:00.000 | 12 | 63 |
| 5 | 7 | t1 | 2017-01-01 02:10:00.000 | 65 | 129 |
| 9 | 0 | t2 | 2017-01-01 00:03:00.000 | 0 | 0 |
| 10 | 1 | t2 | 2017-01-01 01:04:00.000 | 61 | 61 |
| 8 | 2 | t2 | 2017-01-01 02:10:00.000 | 66 | 127 |
+----+----+-----+-------------------------+---------+-------------+

Answer 2

I kind of like a method that is a bubble sort. The problem I have found when doing recursive operations is they work great for small sets(think less than 5 or 10k), then behave horrid when you get larger. For this reason I like a cursor approach were you are essentially saying: "Are you larger than a criteria? Yes, No. Insert or Ignore, Delete, move on." This way you are evaluating over every item once and once only, not every variation of a theme of recursion.
DECLARE #Temp TABLE
(
id INT
, dt DATETIME
, txt VARCHAR(8)
, rwn INT
)
DECLARE #Holder TABLE
(
id INT
, dt DATETIME
, txt VARCHAR(8)
, Dif int
)
INSERT INTO #Temp
SELECT *, row_number() over (partition by txt order by dt, id) AS rn
From wt
WHILE EXISTS (SELECT 1 FROM #Temp)
BEGIN
DECLARE
#CurId INT
, #CurDt DATETIME
, #Curtxt VARCHAR(8)
, #LastDate DATETIME
;
SELECT TOP 1 #CurId = Id, #CurDt = Dt, #Curtxt = txt FROM #Temp ORDER BY txt, rwn
--If there is not entry you need a single entry
IF NOT EXISTS (SELECT TOP 1 * FROM #Holder)
BEGIN
INSERT INTO #Holder VALUES (#CurId, #CurDt, #curtxt, null)
END
ELSE
--if you reset the grouping you need to reset and begin anew
IF (SELECT rwn FROM #Temp WHERE Id = #CurId) = 1
BEGIN
INSERT INTO #Holder VALUES (#CurId, #CurDt, #curtxt, null)
END
--if you are going along check the logic for the difference of what the last was compared to the current
ELSE
BEGIN
SELECT TOP 1 #LastDate = dt FROM #Holder ORDER BY id desc
IF DATEDIFF(HOUR, #LastDate, #CurDt) >= 1
BEGIN
INSERT INTO #Holder VALUES (#CurId, #CurDt, #curtxt, DATEDIFF(MINUTE, #LastDate, #CurDt))
END
END
--Delete the running values and loop again
DELETE #Temp WHERE Id = #CurId
END
Select *
From #Holder