You may like this solution also…

== Solution #1

You could build a 60 rows table with two columns : “minute” and “nearest”. Those 60 rows are pre-computed. Make “minute” the PK.

By joining, you can get the nearest increment, and then you can compute back the rounded timestamp.

create table report.nearest
(
“minute” integer,
nearest time,
constraint nearest_pk primary key (“minute”)
) ;

–
– To be done once, when application is deployed
–

insert into report.nearest values (0, time ’00:00′) ;
insert into report.nearest values (1, time ’00:00′) ;
…
insert into report.nearest values (14, time ’00:00′) ;
insert into report.nearest values (15, time ’00:15′) ;
insert into report.nearest values (16, time ’00:16′) ;
…

create table report.billing
(
tstamp timestamp,
whatever integer
) ;

insert into report.billing values(timestamp ’2001-02-16 20:14:12′, 1) ;
insert into report.billing values(timestamp ’2001-02-16 20:15:34′, 2) ;

select
B.tstamp,
N.nearest,
date_trunc(‘hour’, B.tstamp) + N.nearest as rounded
from
report.billing B,
report.nearest N
where
extract(minute from B.tstamp) = N.”minute” ;

Results are :

TSTAMP NEAREST ROUNDED
——————- ——– ——————-
2001-02-16 20:14:12 00:00:00 2001-02-16 20:00:00
2001-02-16 20:15:34 00:15:00 2001-02-16 20:15:00

The underlying principle is simple: you can tradeoff storage for computing power (and way back). Here, storage is table report.nearest and computing power is your round_timestamp function.

Btw, in database terminology, I guess “nearest” is a dimension table and “billing” is the fact table.

I don’t know if this solution “is the most elegant” as you state but it is “full SQL”: usually, this is to be prefered because database engines are better at “crunching” sets of data and optmizing SQL requests rather than running procedural code.

== Solution #2

If you manage lots of billing records within a batch process (I mean 100K or millions rows maybe), you may consider to compute more data ahead of the batch time window.

Maybe extract(minute from B.tstamp) could be precomputed in a new column when the billing record is created ? The effect on whole performance of the application could be good … or worse, it depends on the environment and must be validated.

Note: this denormalizes the data model to get better performance and should be documented.

Hope this helps.