Time spent between log entries in a table - sql

Question

I have a table which tracks the user actions in a web-site. A simplified version is as follows
user_id | action_time | module_name
--------+-------------------------+------------
1 | 2014-03-02 11:13:08.775 | home
1 | 2014-03-02 11:13:08.345 | user
1 | 2014-03-02 11:13:08.428 | discussions
How much time did a user spend on each screen? So take the least action_time for a user, get the next one, find the difference.
I think this calls for a recursive query, but not able to get my head around it. One thing - I wouldn't know when to stop. After some "module" the user could have just closed the browser, without bothering to logout. So "closure" is a bit tricky.

Answer 1

This can be surprisingly simple with the window function lead()
SELECT *
, lead(action_time) OVER (PARTITION BY user_id ORDER BY action_time)
- action_time AS time_spent
FROM tbl;
That's all.
time_spent is NULL for the last action of a user, where no other action follows - which seems perfectly adequate.

Answer 2

this example of how to make a 'range aggregate' using windowing functions and a lot of nested subqueries. I just adapted it to partition and group by user_id, and it seems to do what you want:
SELECT user_id, min(login_time) as login_time, max(logout_time) as logout_time
FROM (
SELECT user_id, login_time, logout_time,
max(new_start) OVER (PARTITION BY user_id ORDER BY login_time, logout_time) AS left_edge
FROM (
SELECT user_id, login_time, logout_time,
CASE
WHEN login_time <= max(lag_logout_time) OVER (
PARTITION BY user_id ORDER BY login_time, logout_time
) THEN NULL
ELSE login_time
END AS new_start
FROM (
SELECT
user_id,
login_time,
logout_time,
lag(logout_time) OVER (PARTITION BY user_id ORDER BY login_time, logout_time) AS lag_logout_time
FROM app_log
) AS s1
) AS s2
) AS s3
GROUP BY user_id, left_edge
ORDER BY user_id, min(login_time)
Results in:
user_id | login_time | logout_time
---------+---------------------+---------------------
1 | 2014-01-01 08:00:00 | 2014-01-01 10:49:00
1 | 2014-01-01 10:55:00 | 2014-01-01 11:00:00
2 | 2014-01-01 09:00:00 | 2014-01-01 11:49:00
2 | 2014-01-01 11:55:00 | 2014-01-01 12:00:00
(4 rows)
It works by first detecting the beginning of each new range (partitioned by user_id), then extending and grouping by the detected ranges. I found I had to read that article very carefully to understand it!
The article suggests it can be simplified with Postgresql>=9.0 by removing the innermost subquery and changing the window range, but I could not get that to work.