Postgres - Fast way to sum over rows from last day of month - sql

Question

I want to query a table and sum a column for all of the rows from the last day of the month.
Let's use the following table as an example:
CREATE TABLE example(dt date, value int)
(The real table has many more columns and is relatively large, and the real query is more complicated)
I have the following query:
SELECT dt, SUM(value)
FROM example
WHERE dt IN (SELECT DISTINCT
date_trunc('MONTH', generate_series('2012-01-01'::date,
'2016-12-01'::date,
interval '1 day') + INTERVAL '1 MONTH - 1 day')::date)
GROUP BY dt
It runs in about ~2 seconds on my real table.
However, if I generate the full list of end-of-month days in my range and parameterise the query like so:
SELECT dt, SUM(value)
FROM example
WHERE dt IN ('2012-01-31', ...)
GROUP BY dt
It's much quicker, ~750ms.
I would prefer not to generate the dates and pass them through to the query like that, is there a way I can do this entirely in SQL and make it as fast as the latter version?

Answer 1

The sub-select is needlessly complicated. It can be simplified to:
SELECT dt, SUM(value)
FROM example
WHERE dt IN (SELECT d::date
from generate_series('2012-01-01'::date, '2016-12-01'::date, interval '1 month') dates (d)
GROUP BY dt; --<< the group by is necessary
Maybe that speeds up the query.
You can also try to put the date generation into a CTE:
with dates (d) as (
SELECT t::date
from generate_series('2012-01-01'::date, '2016-12-01'::date, interval '1 month') t
)
SELECT dt, SUM(value)
FROM example
WHERE dt IN ( select d from dates)
GROUP BY dt;
Sometimes doing a JOIN is also more efficient:
with dates (d) as (
SELECT t::date
from generate_series('2012-01-01'::date, '2016-12-01'::date, interval '1 month') t
)
SELECT dt, SUM(value)
FROM example
JOIN dates on example.dt = dates.d
GROUP BY dt;

Answer 2

The performance problem in your query comes from the fact that you are generating a daily series. Change it to monthly, remove the distinct and add a group by
select dt, sum(value)
from
example
inner join (
select date_trunc('month', dt) + interval '1 month - 1 day' as dt
from generate_series('2012-01-01'::date, '2016-12-01', '1 month') gs (dt)
) d using (dt)
group by dt