Calculating a rolling count over a specified window

user_3393505 0
Viewed 463

The sample data may help explain what I'm trying to do more than explaining it, so I'll begin with that.

Here's the data I'm currently working with:

+-------------------------+--------------+
|        CallStart        | CallDuration |
+-------------------------+--------------+
| 2017-09-15 09:15:15.313 | 00:01:28     |
| 2017-09-15 09:15:15.317 | 00:01:45     |
| 2017-09-15 09:16:45.603 | 00:01:31     |
| 2017-09-15 09:17:00.637 | 00:01:24     |
| 2017-09-15 09:18:20.853 | 00:01:42     |
| 2017-09-15 09:18:25.870 | 00:01:24     |
| 2017-09-15 11:27:05.117 | 00:00:59     |
| 2017-09-15 11:31:16.053 | 00:01:18     |
| 2017-09-15 11:34:41.627 | 00:01:00     |
| 2017-09-15 12:16:45.413 | 00:01:01     |
| 2017-09-15 12:18:15.820 | 00:01:05     |
| 2017-09-15 12:30:43.607 | 00:01:04     |
| 2017-09-15 12:31:48.817 | 00:00:55     |
| 2017-09-15 12:35:14.563 | 00:00:59     |
| 2017-09-15 12:42:10.947 | 00:00:43     |
| 2017-09-15 12:56:28.807 | 00:01:14     |
| 2017-09-15 13:05:10.643 | 00:00:37     |
| 2017-09-15 13:20:08.400 | 00:00:37     |
| 2017-09-15 14:30:12.607 | 00:00:59     |
| 2017-09-15 14:31:22.807 | 00:00:49     |
| 2017-09-15 15:19:47.240 | 00:01:07     |
| 2017-09-15 16:04:47.753 | 00:00:55     |
| 2017-09-15 16:58:08.080 | 00:00:55     |
| 2017-09-15 17:05:04.557 | 00:00:50     |
| 2017-09-15 17:20:42.753 | 00:00:58     |
| 2017-09-15 17:28:09.140 | 00:01:05     |
| 2017-09-15 17:39:46.690 | 00:00:38     |
| 2017-09-15 17:40:21.957 | 00:01:05     |
| 2017-09-15 17:43:47.570 | 00:01:08     |
| 2017-09-15 17:47:23.390 | 00:01:05     |
| 2017-09-15 17:47:28.410 | 00:00:56     |
| 2017-09-15 17:51:59.380 | 00:01:04     |
+-------------------------+--------------+

I am trying to get a rolling COUNT(*) for the number of occurrences in this data over a 15-minute timeframe. The expected result for this data would be the following:

+-------------------------+--------------+------------------+
|        CallStart        | CallDuration | DropsIn15Minutes |
+-------------------------+--------------+------------------+
| 2017-09-15 09:15:15.313 | 00:01:28     |                1 |
| 2017-09-15 09:15:15.317 | 00:01:45     |                2 |
| 2017-09-15 09:16:45.603 | 00:01:31     |                3 |
| 2017-09-15 09:17:00.637 | 00:01:24     |                4 |
| 2017-09-15 09:18:20.853 | 00:01:42     |                5 |
| 2017-09-15 09:18:25.870 | 00:01:24     |                6 |
| 2017-09-15 11:27:05.117 | 00:00:59     |                1 |
| 2017-09-15 11:31:16.053 | 00:01:18     |                2 |
| 2017-09-15 11:34:41.627 | 00:01:00     |                3 |
| 2017-09-15 12:16:45.413 | 00:01:01     |                1 |
| 2017-09-15 12:18:15.820 | 00:01:05     |                2 |
| 2017-09-15 12:30:43.607 | 00:01:04     |                3 |
| 2017-09-15 12:31:48.817 | 00:00:55     |                3 |
| 2017-09-15 12:35:14.563 | 00:00:59     |                3 |
| 2017-09-15 12:42:10.947 | 00:00:43     |                4 |
| 2017-09-15 12:56:28.807 | 00:01:14     |                2 |
| 2017-09-15 13:05:10.643 | 00:00:37     |                2 |
| 2017-09-15 13:20:08.400 | 00:00:37     |                2 |
| 2017-09-15 14:30:12.607 | 00:00:59     |                1 |
| 2017-09-15 14:31:22.807 | 00:00:49     |                2 |
| 2017-09-15 15:19:47.240 | 00:01:07     |                1 |
| 2017-09-15 16:04:47.753 | 00:00:55     |                1 |
| 2017-09-15 16:58:08.080 | 00:00:55     |                1 |
| 2017-09-15 17:05:04.557 | 00:00:50     |                2 |
| 2017-09-15 17:20:42.753 | 00:00:58     |                1 |
| 2017-09-15 17:28:09.140 | 00:01:05     |                2 |
| 2017-09-15 17:39:46.690 | 00:00:38     |                2 |
| 2017-09-15 17:40:21.957 | 00:01:05     |                3 |
| 2017-09-15 17:43:47.570 | 00:01:08     |                3 |
| 2017-09-15 17:47:23.390 | 00:01:05     |                4 |
| 2017-09-15 17:47:28.410 | 00:00:56     |                5 |
| 2017-09-15 17:51:59.380 | 00:01:04     |                6 |
+-------------------------+--------------+------------------+

Sample data:

Create Table #Calls 
(
    CallStart DateTime,
    CallDuration Time(0)
);
Insert Into #Calls
Values (N'2017-09-15T09:15:15.313', N'00:01:28'),
    (N'2017-09-15T09:15:15.317', N'00:01:45'),
    (N'2017-09-15T09:16:45.603', N'00:01:31'),
    (N'2017-09-15T09:17:00.637', N'00:01:24'),
    (N'2017-09-15T09:18:20.853', N'00:01:42'),
    (N'2017-09-15T09:18:25.87', N'00:01:24'),
    (N'2017-09-15T11:27:05.117', N'00:00:59'),
    (N'2017-09-15T11:31:16.053', N'00:01:18'),
    (N'2017-09-15T11:34:41.627', N'00:01:00'),
    (N'2017-09-15T12:16:45.413', N'00:01:01'),
    (N'2017-09-15T12:18:15.82', N'00:01:05'),
    (N'2017-09-15T12:30:43.607', N'00:01:04'),
    (N'2017-09-15T12:31:48.817', N'00:00:55'),
    (N'2017-09-15T12:35:14.563', N'00:00:59'),
    (N'2017-09-15T12:42:10.947', N'00:00:43'),
    (N'2017-09-15T12:56:28.807', N'00:01:14'),
    (N'2017-09-15T13:05:10.643', N'00:00:37'),
    (N'2017-09-15T13:20:08.4', N'00:00:37'),
    (N'2017-09-15T14:30:12.607', N'00:00:59'),
    (N'2017-09-15T14:31:22.807', N'00:00:49'),
    (N'2017-09-15T15:19:47.24', N'00:01:07'),
    (N'2017-09-15T16:04:47.753', N'00:00:55'),
    (N'2017-09-15T16:58:08.08', N'00:00:55'),
    (N'2017-09-15T17:05:04.557', N'00:00:50'),
    (N'2017-09-15T17:20:42.753', N'00:00:58'),
    (N'2017-09-15T17:28:09.14', N'00:01:05'),
    (N'2017-09-15T17:39:46.69', N'00:00:38'),
    (N'2017-09-15T17:40:21.957', N'00:01:05'),
    (N'2017-09-15T17:43:47.57', N'00:01:08'),
    (N'2017-09-15T17:47:23.39', N'00:01:05'),
    (N'2017-09-15T17:47:28.41', N'00:00:56'),
    (N'2017-09-15T17:51:59.38', N'00:01:04');

I can somewhat get this to work via the following:

Select  CallStart,
        CallDuration,
        DropsIn15Minutes = 
        (
            Select  Count(*) 
            From    #Calls C2 
            Where   C2.CallStart Between DateAdd(Minute, -15, C1.CallStart) 
                                 And     C1.CallStart
        )
From    #Calls  C1

But, I would like to avoid the sub-query, in favor of a COUNT(*) OVER () (or any other solution if possible) solution.

Is this possible to do? Or is the sub-query the right solution for this?

sql-server sql
3 Answers

Two other options:

using cross apply():

select  
    CallStart
  , CallDuration
  , DropsIn15Minutes
from calls c
  outer apply (
    select DropsIn15Minutes = count(*)
    from calls i
    where i.callstart >= dateadd(minute,-15,c.CallStart)
      and i.callstart <= c.CallStart
    ) x
order by c.CallStart

using an inner join:

select  
    c.CallStart
  , c.CallDuration
  , DropsIn15Minutes = count(i.CallStart)
from calls c
  inner join calls i
    on i.callstart >= dateadd(minute,-15,c.CallStart)
   and i.callstart <= c.CallStart
group by c.CallStart, c.CallDuration
order by c.CallStart

rextester demo: http://rextester.com/PFCE43712

compare execution plans for all 3 (including yours): dbfiddle.uk demo

user_23334990

One way - which may perform better than a nested loops join on a range if the table is big - would be to first create a numbers table...

CREATE TABLE dbo.Numbers
(
N INT PRIMARY KEY
);

    WITH E1(N) AS 
    (
        SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL 
        SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL 
        SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1 UNION ALL SELECT 1
    )                                       -- 1*10^1 or 10 rows
    , E2(N) AS (SELECT 1 FROM E1 a, E1 b)   -- 1*10^2 or 100 rows
    , E4(N) AS (SELECT 1 FROM E2 a, E2 b)   -- 1*10^4 or 10,000 rows
    , E8(N) AS (SELECT 1 FROM E4 a, E4 b)   -- 1*10^8 or 100,000,000 rows
INSERT INTO dbo.Numbers
    SELECT TOP (60*60*24) -1 + ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) AS N FROM E8;

And then use something along the following lines.

WITH Calls
     AS (SELECT *,
                --pre-truncate all call starts to second precision
                CallStart_sec = DATEADD(SECOND, DATEDIFF(SECOND, '20000101', CallStart), '20000101')
         FROM   #Calls),
     PreAgg
     AS (SELECT CallStart_sec,
                COUNT(*) AS Cnt
         FROM   Calls
         GROUP  BY CallStart_sec),
     Dates(D)
     --Todo - something else other than hardcoding the dates
     AS (SELECT CAST('2017-09-15' AS DATETIME2)),
     RT
     AS (SELECT *,
                Cume = SUM(Cnt) OVER (ORDER BY DATEADD(SECOND, N.N, Dates.D) 
                               ROWS BETWEEN 900 PRECEDING AND CURRENT ROW)
         FROM   Dates
                INNER JOIN dbo.Numbers N
                  ON N.N BETWEEN 0 AND 86399
                LEFT JOIN PreAgg P
                  ON P.CallStart_sec = DATEADD(SECOND, N.N, Dates.D))
SELECT C.CallStart_sec AS CallStart,
       CallDuration,
       DropsIn15Minutes = Cume
FROM   Calls C
       JOIN RT
         ON RT.CallStart_sec = C.CallStart_sec
user_732260 
Select  CallStart,
        CallDuration,
COUNT(*) OVER (PARTITION BY trunc(CallStart,'mi') - 
         numtodsinterval(mod(to_char(CallStart,'mi'),15),'minute'))
as DropsIn15Minutes 
From    #Calls  C1

Lets try this out and see how it works.

Ted

user_85599720
Related