Calculate the time difference in the same row dynamically - sql

Question

Is there any way to calculate the time difference in SQL between rows within the same column based on the 'DOWN' and 'UP' values like this:
There are 3 scenarios(that I'm aware of):
Yellow, Orange and Green: there is a state_id 2(down) and after that a state_id 5(up), so the time difference needs to be calculated between the two rows;
Blue: there are multiple state_id 2(down) and after that one state_id 5(up), so the time difference needs to be calculated between the first row and last row;
Red: there is only a state_id 2(down) because it is still down with any update, so the time difference needs to be calculated till the end of the month.
I hope you can help me out.

Answer 1

Was first considering to use LAG for this.
But using a cummulative SUM, and the window version of MIN works also for more than 2 DOWN's:
-- test reference data
declare #State table (id int, state varchar(4));
insert into #State (id, state) values
(2,'DOWN'),
(5,'UP')
-- test data, using a table variable
declare #AlertState table (alert_id int identity(1,1), host_id int, state_time datetime, state_id int);
insert into #AlertState (host_id, state_time, state_id) values
(119, GetDate()-0.32, 2),
(119, GetDate()-0.31, 5),
(119, GetDate()-0.24, 2),
(119, GetDate()-0.23, 2),
(119, GetDate()-0.22, 2),
(119, GetDate()-0.21, 5),
(119, GetDate()-0.15, 5),
(119, GetDate()-0.11, 2);
-- The query
select alert_id, host_id, state_time, state_id,
diff_min = (
case
when state_id = 5 then
datediff(minute, min(state_time) over (partition by host_id, stategroup), state_time)
when state_id = 2 and stategroup is null then
datediff(minute, state_time, cast(EOMONTH(GetDate()) as datetime)+1)
end),
s.state
from (
select alert_id, host_id, state_time, state_id,
sum(case state_id when 5 then 1 end) over (partition by host_id order by state_time desc) as stategroup
from #AlertState
where state_id in (2,5)
) q
left join #State s on s.id = q.state_id
order by state_time, alert_id;

Answer 2

The way I did this before is
Select a.state_time as downtime,
(
select min(inner.state_time) from tablename downentry where
inner.state_time > outer.state_time and downentry.state='UP'
) as uptime
from tablename upentry
where state = 'DOWN'
Then you need to find the datediff between them, and if uptime is null, the datediff between downtime and 'endofmonth'
It's potentially quite poor performing, so I always wrote the answer out to a data warehouse, but think it gives the results you're asking for.

Answer 3

SQL2012+
You could try following solution:
SELECT y.group_id, host_id = MIN(host_id), start_time = MIN(state_time), end_time = MAX(state_time), diff_minute = DATEDIFF(MINUTE, MIN(state_time), MAX(state_time))
FROM (
SELECT *, group_id = SUM(x.new_group_start) OVER(ORDER BY x.host_id, x.state_time)
FROM (
SELECT *, new_group_start = IIF(a.state_id = 'DOWN' AND ISNULL(LAG(a.state_id) OVER(ORDER BY a.host_id, a.state_time), 'UP') = 'UP', 1, 0)
FROM #Alerts a
) x
) y
GROUP BY y.group_id
ORDER BY y.group_id
Demo