I have table that holds records with tasks, status and time when triggered:
Table tblwork:
+-------------+------------+---------------------+-----+
| task | status | stime | id |
+-------------+------------+---------------------+-----+
| A | 1 | 2018-03-07 20:00:00 | 1 |
| A | 2 | 2018-03-07 20:30:00 | 2 |
| A | 1 | 2018-03-07 21:00:00 | 3 |
| A | 3 | 2018-03-07 21:30:00 | 4 |
| B | 1 | 2018-03-07 22:30:00 | 5 |
| B | 3 | 2018-03-07 23:30:00 | 6 |
+-------------+------------+---------------------+-----+
Status 1 means start, 2 - pause, 3 - end
Then I need to calculate how much time is spent for each task excluding pause (status = 2). This is how I do it:
SELECT t1.id, t1.task,
SUM(timestampdiff(second,IFNULL(
(SELECT MAX(t2.stime) FROM tblwork t2 WHERE t2.task='B' AND t2.stime< t1.stime) ,t1.stime),t1.stime)) myTimeDiffSeconds
FROM tblwork t1
WHERE t1.task='B' and (t1.status = 1 or t1.status = 3);
Now I want to get table for all tasks
SELECT t1.id, t1.task,
SUM(timestampdiff(second,IFNULL(
(SELECT MAX(t2.stime) FROM tblwork t2 WHERE t2.stime< t1.stime) ,t1.stime),t1.stime)) myTimeDiffSeconds
FROM tblwork t1
WHERE (t1.status = 1 or t1.status = 3) GROUP BY t1.taks
I get this result:
+-------------+------------+---------------------+
| task | id | mytimedifference |
+-------------+------------+---------------------+
| A | 1 | 3600 |
| B | 3 | 2421217 |
+-------------+------------+---------------------+
Calculation for A is correct B is wrong, it should be 3600 second but i don't understand why.
Assuming there is always a start for each pause and end, wouldn't something like this be more direct?
I'm not quite sure offhand how big the values that come out of TO_SECONDS() are for current timestamps; but if they are an issue when being summed, if could be changed to
You can detect "abnormal" data by adding the following to the select expression list
Note: any "unclosed" intervals will be marked as abnormal; without much more complicated and expensive start and end checking for intervals to differentiate "open" from "invalid", it's probably the best that can be done. The sum used for additonal "integrityCheck" equaling -1 might hint at an open ended interval, but could also indicate an erroneous double-start.