Update time only if not a minute passed - postgresql

Question

I have created the next table.
-- TABLE user_time
user_id integer PRIMARY KEY,
prev_time TIMESTAMP WITHOUT TIME ZONE DEFAULT NOW(),
total_time INTERVAL DEFAULT interval '0 second'
I have to add an interval value to the total_time, e.g.
total_time = total_time + NOW() - prev_time
only if not a minute passed since prev_time (so, less than 1 minute passed) in a single query.
The next construction is about what I want but it's wrong:
UPDATE user_time SET total_time = total_time +
(
SELECT NOW() - prev_time incinterval,
CASE
WHEN incinterval < interval '1 minute' THEN incinterval
ELSE interval '0 second'
END
FROM user_time WHERE user_id=6
)
Firstly the SELECT is wrong, PostgreSQL does not recognize incinterval in the CASE construction. Secondly there is the first extra column in SELECT which creates a pseudo name.
Do you have an idea how to correct the query OR
Is it the common practice to increment total time with the condition and store it to a database with a single query?

Answer 1

You can not expect Postgres to respect your alias like this inside a SELECT. Here's the way to go:
UPDATE user_time SET total_time = total_time +
(
SELECT
CASE
WHEN NOW() - prev_time < interval '1 minute'
THEN NOW() - prev_time
ELSE interval '0 second'
END
FROM user_time WHERE user_id=6
)
Or to use your alias:
UPDATE user_time SET total_time = total_time +
(
SELECT
CASE
WHEN incinterval < interval '1 minute'
THEN incinterval
ELSE interval '0 second'
END
FROM(
SELECT
NOW() - prev_time incinterval
FROM user_time WHERE user_id=6
) foo
)
Edit after comment:
Simply add , prev_time = NOW() after the last parenthesis in any option you choose from those above.

Answer 2

The deadlock issue can be resolved by doing the analysis post hoc. Just log the queries and "roll up" clusters that meet your interval spec. This allows for variable-length intervals, which can be very useful for sites that provide dense content.
The "post hoc" process can/should run periodically in overlapping ranges to catch longer intervals.
There is an relevant example of a generalizable temporal clustering query using analytic functions from about 8 years ago on asktom.oracle.com.