Transpose sequential times data to derive duration - sql

Question

I have a table of actions within a session and duration (milliseconds) between each step:
+-----------------------------------------------------------------------+
| | userid | sessionid | action sequence | action | milliseconds | |
| +--------+-----------+-----------------+-------------+--------------+ |
| | 1 | 1 | 1 | event start | 0 | |
| | 1 | 1 | 2 | other | 188114 | |
| | 1 | 1 | 3 | event end | 248641 | |
| | 1 | 1 | 4 | other | 398215 | |
| | 1 | 1 | 5 | event start | 488284 | |
| | 1 | 1 | 6 | other | 528445 | |
| | 1 | 1 | 7 | other | 572711 | |
| | 1 | 1 | 8 | event end | 598123 | |
| | 1 | 2 | 1 | event start | 0 | |
| | 1 | 2 | 2 | event end | 54363 | |
| | 2 | 1 | 1 | other | 0 | |
| | 2 | 1 | 2 | other | 2345 | |
| | 2 | 1 | 1 | other | 75647 | |
| | 3 | 1 | 2 | other | 0 | |
| | 3 | 1 | 3 | event start | 34678 | |
| | 3 | 1 | 4 | other | 46784 | |
| | 3 | 1 | 5 | other | 78905 | |
| | 4 | 1 | 1 | event start | 0 | |
| | 4 | 1 | 2 | other | 7454 | |
| | 4 | 1 | 3 | other | 11245 | |
| | 4 | 1 | 4 | event end | 24567 | |
| | 4 | 1 | 5 | other | 29562 | |
| | 4 | 1 | 6 | other | 43015 | |
| +--------+-----------+-----------------+-------------+--------------+ |
I would like to capture complete events -- sessions containing both an event start and end (some may have a start but no end, an end but no start, or neither -- I don't want those), and their start and end times. Ultimately I want to track duration by transposing the sequential rows of times into columns so I can calculate a difference. The above data table would ideally be transposed into:
+--------+-----------+---------------+--------+--------+
| userid | sessionid | full event id | start | end |
+--------+-----------+---------------+--------+--------+
| 1 | 1 | 1 | 0 | 248641 |
| 1 | 1 | 2 | 488284 | 598123 |
| 1 | 2 | 1 | 0 | 54363 |
| 4 | 1 | 1 | 0 | 24567 |
+--------+-----------+---------------+--------+--------+
I attempted something like:
select a.userid, a.sessionid, a.milliseconds as start, b.milliseconds as end
from #table a
inner join #table b
on a.userid=b.userid
and a.sessionid=b.sessionid
and a.action='event start'
and b.action='event end'
However, that doesn't work since some users may have multiple event start and ends in on session (like userid 1). I am stuck on how to best transpose the times data for each event. Thanks for you help!

Answer 1

So, given your above data:
CREATE TABLE test_table (
`userid` int,
`sessionid` int,
`actionSequence` int,
`action` varchar(11),
`milliseconds` int
);
INSERT INTO test_table
(`userid`, `sessionid`, `actionSequence`, `action`, `milliseconds`)
VALUES
(1, 1, 1, 'event start', 0),
(1, 1, 2, 'other', 188114),
(1, 1, 3, 'event end', 248641),
(1, 1, 4, 'other', 398215),
(1, 1, 5, 'event start', 488284),
(1, 1, 6, 'other', 528445),
(1, 1, 7, 'other', 572711),
(1, 1, 8, 'event end', 598123),
(1, 2, 1, 'event start', 0),
(1, 2, 2, 'event end', 54363),
(2, 1, 1, 'other', 0),
(2, 1, 2, 'other', 2345),
(2, 1, 1, 'other', 75647),
(3, 1, 2, 'other', 0),
(3, 1, 3, 'event start', 34678),
(3, 1, 4, 'other', 46784),
(3, 1, 5, 'other', 78905),
(4, 1, 1, 'event start', 0),
(4, 1, 2, 'other', 7454),
(4, 1, 3, 'other', 11245),
(4, 1, 4, 'event end', 24567),
(4, 1, 5, 'other', 29562),
(4, 1, 6, 'other', 43015);
The following query should get you where you want to be (you were on the right track):
SELECT
tt1.userid,
tt1.sessionid,
tt1.actionSequence,
tt1.milliseconds AS startMS,
MIN(tt2.milliseconds) AS endMS,
MIN(tt2.milliseconds) - tt1.milliseconds AS totalMS
FROM test_table tt1
INNER JOIN test_table tt2
ON tt2.userid = tt1.userid
AND tt2.sessionid = tt1.sessionid
AND tt2.actionSequence > tt1.actionSequence
AND tt2.action = 'event end'
WHERE tt1.action = 'event start'
GROUP BY tt1.userid, tt1.sessionid, tt1.actionSequence, startMS
Giving you this result set:
userid sessionid actionSequence startMS endMS totalMS
1 1 1 0 248641 248641
1 1 5 488284 598123 109839
1 2 1 0 54363 54363
4 1 1 0 24567 24567
The GROUP BY is important, because there are two rows with action = 'event end' and sequence > 1 for sessionid = 1 and userid = 1, so (I assume) we want the one closest to the current sequence, i.e. the MIN(milliseconds). As you can see, it also allows you to go ahead and take the difference of the two columns in this result set, saving you the extra step you were planning :]
Here is a SQLFiddle of this query in action on MySQL 5.6. You did not specify an RDBMS, but I believe the language used by this query should be simple enough to work in any sql engine.