Need SQL Server Query help, Newbie - sql

Question

I have a table that has the following columns.
id,
compid(used to identify a piece of equipment),
startalarmdate,
endalarmdate
when a piece of equipment goes into alarm, I insert the compid and startalarmdate(with the current time) into the table and when the equipment comes out of alarm I fill in the null in that same row in the endalarmdate column with the current time.
so I end up with something like this in the table
417,6,Sun Oct 30 18:48:17 CDT 2011,Mon Oct 31 09:49:21 CDT 2011
422,6,Mon Oct 31 10:19:19 CDT 2011,Mon Oct 31 12:19:22 CDT 2011
427,6,Mon Oct 31 20:19:56 CDT 2011,Tue Nov 01 09:50:59 CDT 2011
429,6,Tue Nov 01 21:51:41 CDT 2011,Wed Nov 02 09:52:37 CDT 2011
432,6,Wed Nov 02 21:23:23 CDT 2011,Fri Nov 04 16:26:29 CDT 2011
I was able to build a query that gives me a total downtime in hours for each event, but what I would like to do now is build a query that gives me a total hours in downtime for each day of a month. Id like it to have the compid all the way to the left, then have each day of the month to the right of the compid in a column on the same row. Id like the days with no downtime to be null. Is it possible to do that with the way this table is setup?

Answer 1

Step 1: set up a temp table containing the desired "time blocks" that you want to total for. These blocks could be for any range of time; in your example, it would be one entry for ever day (24-hour period) in the month.
CREATE TABLE #TimeRanges
(
RangeStart datetime not null
,RangeEnd datetime not null
)
Left-outer-joining this table on your data ensures you get at least one row per time block (day), even if there were no alarms occuring that day:
SELECT
tr.RangeStart -- Use start of each time block to identify the block
,md.CompId -- With left outer join, will be 0 or more rows for each block
,sum(datediff(hh
,case
when tr.RangeStart > md.StartAlarmDate then tr.RangeStart
else md.StartAlarmDate
end
,case
when tr.RangeEnd > md.EndAlarmDate then tr.RangeEnd
else md.EndAlarmDate
end)) HoursInRange
from #TimeRanges tr
left outer join MyData md
on md.StartAlarmDate < tr.RangeEnd
and md.EndAlarmDate > tr.From
group by
tr.RangeStart
,md.CompId
(I can't test this code, some debugging may be required--but the concept is solid. I'll let you worry about rounding partial hours, and whether you want > and <, or >= and <= (things may get tricky if an alarm starts and/or ends at the exact same point in time as a block boundary).
Edit/Addenda
Here's a fairly basic way to set up the temp table used in the routine (this code, I tested):
-- Set up and initialize some variables
DECLARE
#FirstDay datetime
,#NumberOfDays int
SET #FirstDay = 'Oct 1, 2011' -- Without time, this makes it "midnight the morning of" that day
SET #NumberOfDays = 91 -- Through Dec 31
-- Creates a temporary table that will persist until it is dropped or the connection is closed
CREATE TABLE #TimeRanges
(
RangeStart datetime not null
,RangeEnd datetime not null
)
-- The order in which you add rows to the table is irrelevant. By adding from last to first, I
-- only have to fuss around with one variable, instead of two (the counter and the end-point)
WHILE #NumberOfDays >= 0
BEGIN
INSERT #TimeRanges (RangeStart, RangeEnd)
values ( dateadd(dd, #NumberOfDays, #FirstDay) -- Start of day
,dateadd(dd, #NumberOfDays + 1, #FirstDay)) -- Start of the next day
SET #NumberOfDays = #NumberOfDays - 1
END
-- Review results
SELECT *
from #TimeRanges
order by RangeStart
-- Not necessary, but doesn't hurt, especially when testing code
DROP TABLE #TimeRanges
Note that by making RangeEnd the start of the next day, you have to be careful with your greaterthans and lessthans. The details can get very finicky and fussy there, and you'll want to do a lot of edge-case testing (what if alarm starts, or ends, exactly at Dec 16 2011 00:00.000). I'd go with that, because overall it's simpler to code for than for junk like 'Dec 16, 2011 23:59.997'

Answer 2

As mentionned by #paulbailey, you want to use the DATEDIFF function to get the amount of downtime.
To extract the dates and downtime period (I'm adding a bit more columns that you might need)..
SELECT compid,
YEAR(startalarmdate) AS [Year],
MONTH(startalarmdate) AS [Month],
DAY(startalarmdate) AS [Day],
DATEDIFF(ss, startalarmdate, endalarmdate) AS DowntimeInSeconds --You will need to convert thid later to the format you wish to use
FROM YourTable
/* WHERE CLAUSE - Most probably a date range */
Now this gives you the downtime in seconds for each days that had a downtime.
To get the amount of downtime per day is easy as grouping by day and SUMing up the downtimes (again adding more columns that you might need)..
SELECT compid,
[Year],
[Month],
[Day],
SUM(DowntimeInSeconds) AS TotalDowntimeInSeconds
FROM (SELECT compid,
YEAR(startalarmdate) AS [Year],
MONTH(startalarmdate) AS [Month],
DAY(startalarmdate) AS [Day],
DATEDIFF(ss, startalarmdate, endalarmdate) AS DowntimeInSeconds --You will need to convert thid later to the format you wish to use
FROM YourTable
/* WHERE CLAUSE - Most probably a date range */) AS GetDowntimes
GROUP BY compid, [Year], [Month], [Day]
ORDER BY [Year], [Month], [Day], compid
And I believe this should help you get where you want to.
Edit:
To have the days that have no downtime included in this result, you need to first have a list of ALL days present in a month. You take this list and you LEFT OUTER JOIN the result from the above query (you will have to remove the ORDER BY first).

Answer 3

The case statement in Philip Kelley's answer does not work, although the main principal of filling a temp table with dates and left joining stands true. For my version I've used the same variable to start - an input date and the number of days to report on.
DECLARE #StartDate DATETIME, #Days INT
SELECT #StartDate = GETDATE(),
#Days = 5
-- REMOVE ANY TIME FROM THE STARTDATE
SET #StartDate = DATEADD(DAY, 0, DATEDIFF(DAY, 0, #StartDate))
-- CREATE THE TEMP TABLE OF DATES USING THE SAME METHODOLOGY
DECLARE #Dates TABLE (AlarmDate SMALLDATETIME NOT NULL PRIMARY KEY)
WHILE (#Days > 0)
BEGIN
INSERT #Dates VALUES (DATEADD(DAY, #Days, #StartDate))
SET #Days = #Days - 1
END
-- NOW SELECT THE DATA
SELECT AlarmDate,
CompID,
CONVERT(DECIMAL(10, 2), ISNULL(SUM(DownTime), 0) / 3600.0) [DownTime]
FROM #Dates
LEFT JOIN
( SELECT DATEADD(DAY, 0, DATEDIFF(DAY, 0, StartAlarmDate)) [StartAlarmDate],
CompID,
DATEDIFF(SECOND, StartAlarmDate, CASE WHEN EndAlarmDate >= DATEADD(DAY, 1, DATEDIFF(DAY, 0, StartAlarmDate)) THEN DATEADD(DAY, 1, DATEDIFF(DAY, 0, StartAlarmDate)) ELSE EndAlarmDate END) [DownTime]
FROM [yourTable]
UNION ALL
SELECT DATEADD(DAY, 0, DATEDIFF(DAY, 0, EndAlarmDate)) [Date],
CompID,
DATEDIFF(SECOND, DATEADD(DAY, 1, DATEDIFF(DAY, 0, StartAlarmDate)), EndAlarmDate) [DownTime]
FROM [yourTable]
WHERE EndAlarmDate >= DATEADD(DAY, 1, DATEDIFF(DAY, 0, StartAlarmDate))
) data
ON StartAlarmDate = AlarmDate
GROUP BY AlarmDate, CompID
I have used seconds for the date diff and divided by 3600.0 after the seconds have been summed up as 60 rows each with a difference of a minute would sum to 0 when using hours for a datediff.