Add time bucketing function `time_bucket(<interval>,<timestamp>)` || `date_bin(<interval>,<timestamp>)`

Use case: Right now one can only use date_trunc() to easily define time buckets. date_trunc() only supports predefine time intervals like 1 minute, 1 hour, etc. . In time-series use cases it is often necessary to define different time bucket sizes like e.g. ‘5 minutes’ or ‘20 minutes’

a workaround for this is the - error prone - integer division on the timestamp e.g.

SELECT 
(ts / ( 300 * 1000)) * 300 * 1000 AS "bucket",
avg(metric)
FROM data
GROUP BY bucket
ORDER BY bucket DESC LIMIT 10;

Feature description Have the ability to pass other time bucket sizes / intervals to date_trunc(), e.g. date_trunc('5 minutes',ts) or have a separate time_bucket(INTERVAL,TIMESTAMP) function that takes an interval e.g. '5 minutes' and a timestamp as input and returns the corresponding bucket (i.e. a timestamp that has the starting value of the bucket)

SELECT
time_bucket('5 minutes', ts) AS "bucket",
avg(metric)
FROM data
GROUP BY bucket
ORDER BY bucket DESC LIMIT 10;

---

Timescale time_bucket() function TS docs GH implementation

[proddata]

PostgreSQL 14 will implement date_bin(stride, source, origin) which is quite similar to the above mentioned time_bucket() definition. To improve compatibility with Postgres Wire Protocol, I think this should be the preferred solution. However the origin parameter should be optional, for usability reasons.

default origin is unix 0, Thursday, 01.01.1970 - should be there an option to provide custom origin for easier bucketing by weeks? with origin = unix ts 0, time_bucket(‘1 week’::INTERVAL, ‘1969-01-09T00:00:00Z’::timestamp without time zone) returns -31449600000 which is Thursday, 02.01.1970.

This doesn’t necessarily have to do with the default origin, but I would expect time_bucket('1 week',<timestamp>) to return a Monday 00:00:00.000Z. I think this could be achieved by e.g. setting the default origin to 2001-01-01 or any other year that starts with a Monday.

To follow the new date_bin function, it might be good to be able to specify a origin.

@BaurzhanSakhariev

[mfussenegger]

should the null handling be the same as date_trunc?

select DATE_TRUNC('day', null) - returns null select DATE_TRUNC(null, CURRENT_TIMESTAMP) - returns null select DATE_TRUNC(null, null) - returns null

yes