Get the count of statuses by date but only count continuous rows - sql-server

Question

I have this data:
ID Name Status Date
1 Machine1 Active 2018-01-01
2 Machine2 Fault 2018-01-01
3 Machine3 Active 2018-01-01
4 Machine1 Fault 2018-01-02
5 Machine2 Active 2018-01-02
6 Machine3 Active 2018-01-02
7 Machine2 Active 2018-01-03
8 Machine1 Fault 2018-01-03
9 Machine2 Active 2018-01-04
10 Machine1 Fault 2018-01-04
11 Machine3 Active 2018-01-06
INPUT
and i want these data in output
EXPECTED OUTPUT
Name Last Status Count
Machine1 Fault 3
Machine2 Active 3
Machine3 Active 1 Because Date is not Continuous
*Count : Last number of status in continuous history

Answer 1

I believe it is as simple as this:
WITH cte1 AS (
SELECT
Name,
Status,
DATEADD(DAY, ROW_NUMBER() OVER (PARTITION BY Name, Status ORDER BY Date DESC) - 1, Date) AS GroupingDate
FROM testdata
), cte2 AS (
SELECT
Name,
Status,
RANK() OVER (PARTITION BY Name ORDER BY GroupingDate DESC) AS GroupingNumber
FROM cte1
)
SELECT Name, Status AS LastStatus, COUNT(*) AS LastStatusCount
FROM cte2
WHERE GroupingNumber = 1
GROUP BY Name, Status
ORDER BY Name
Result and DBFiddle:
| Name | LastStatus | LastStatusCount |
|----------|------------|-----------------|
| Machine1 | Fault | 3 |
| Machine2 | Active | 3 |
| Machine3 | Active | 1 |
In order to understand how this works, look at the intermediate values generated by CTE:
| Name | Status | Date | RowNumber | GroupingDate | GroupingNumber |
|----------|--------|---------------------|-----------|---------------------|----------------|
| Machine1 | Fault | 04/01/2018 00:00:00 | 1 | 04/01/2018 00:00:00 | 1 |
| Machine1 | Fault | 03/01/2018 00:00:00 | 2 | 04/01/2018 00:00:00 | 1 |
| Machine1 | Fault | 02/01/2018 00:00:00 | 3 | 04/01/2018 00:00:00 | 1 |
| Machine1 | Active | 01/01/2018 00:00:00 | 1 | 01/01/2018 00:00:00 | 4 |
| Machine2 | Active | 04/01/2018 00:00:00 | 1 | 04/01/2018 00:00:00 | 1 |
| Machine2 | Active | 03/01/2018 00:00:00 | 2 | 04/01/2018 00:00:00 | 1 |
| Machine2 | Active | 02/01/2018 00:00:00 | 3 | 04/01/2018 00:00:00 | 1 |
| Machine2 | Fault | 01/01/2018 00:00:00 | 1 | 01/01/2018 00:00:00 | 4 |
| Machine3 | Active | 06/01/2018 00:00:00 | 1 | 06/01/2018 00:00:00 | 1 |
| Machine3 | Active | 02/01/2018 00:00:00 | 2 | 03/01/2018 00:00:00 | 2 |
| Machine3 | Active | 01/01/2018 00:00:00 | 3 | 03/01/2018 00:00:00 | 2 |
The trick here is that if two numbers are contiguous then subtracting contiguous numbers from them will result in same value. E.g. if we have 5, 6, 8, 9 then subtracting 1, 2, 3, 4 in that order will produce 4, 4, 5, 5.

Answer 2

I think this will work, though SQLFiddle is having a fit at the moment, so I can't test:
SELECT [Name], [Status], ct as [Count]
FROM (
SELECT
[name],
[status],
[date],
1 + (SUM( grp ) OVER (PARTITION BY [name], [status] ORDER BY [date] ROWS BETWEEN 1 PRECEDING AND 0 FOLLOWING ) * grp) ct,
row_number() over(partition by [name] order by [date] desc) rn
FROM
(
SELECT *, CASE WHEN LAG([Date]) OVER(PARTITION BY [name], [status] ORDER BY [date] ) = DATEADD(day, -1, [date]) THEN 1 ELSE 0 END grp
FROM t
) x
) y
WHERE
rn = 1
It first off uses LAG to look at the current row and the previous row (grouping the data into machine name and status, ordering the data by date) and if the current date is 1 day different to the previous date it records a 1 else a 0
These 1s and zeroes are summed in a running total fashion, resetting when the machine name or status changes (the partition of the sum() over() )
Also we want to consider the data just in terms of the machine name, and we only want the latest record from each machine, so we partition by the machine name, and count in order of date descending, then just pick (with the where clause) the rows that are numbered 1 for each machine
It actually makes a lot more sense if you run the queries separately, like this
Calculate the "is the current report consecutive with the previous report, for the given status and machine" 1 = yes, 0 = no:
SELECT *, CASE WHEN LAG([Date]) OVER(PARTITION BY [name], [status] ORDER BY [date] ) = DATEADD(day, -1, [date]) THEN 1 ELSE 0 END grp
FROM t
Calculate the "what is the running total of the current block of consecutive reports":
SELECT
[name],
[status],
[date],
1 + (SUM( grp ) OVER (PARTITION BY [name], [status] ORDER BY [date] ROWS BETWEEN 1 PRECEDING AND 0 FOLLOWING ) * grp) ct,
row_number() over(partition by [name] order by [date] desc) rn
FROM
(
SELECT *, CASE WHEN LAG([Date]) OVER(PARTITION BY [name], [status] ORDER BY [date] ) = DATEADD(day, -1, [date]) THEN 1 ELSE 0 END grp
FROM t
) x
Then of course, the whole thing but without the where clause, so you can see the data we're discarding:
SELECT [Name], [Status], ct as [Count]
FROM (
SELECT
[name],
[status],
[date],
1 + (SUM( grp ) OVER (PARTITION BY [name], [status] ORDER BY [date] ROWS BETWEEN 1 PRECEDING AND 0 FOLLOWING ) * grp) ct,
row_number() over(partition by [name] order by [date] desc) rn
FROM
(
SELECT *, CASE WHEN LAG([Date]) OVER(PARTITION BY [name], [status] ORDER BY [date] ) = DATEADD(day, -1, [date]) THEN 1 ELSE 0 END grp
FROM t
) x
) y
Fiddle finally woke up:
http://www.sqlfiddle.com/#!18/77dae/2