Quickly querying with effective dates on rows themselves with dates - sql

Question

Suppose I have a table called measurement. This table's purpose to measure a numeric "value" (which itself is calculated from other data) for a "series_id" at a particular "date".
Now let's add effective dating to this table with "effective_start" (inclusive) and "effective_end" (inclusive) fields.
DDL:
CREATE TABLE public.measurement
(
date date NOT NULL,
effective_end date NOT NULL,
effective_start date NOT NULL,
series_id character varying(255) COLLATE pg_catalog."default" NOT NULL,
value numeric,
CONSTRAINT measurement_pkey PRIMARY KEY (date, effective_end, effective_start, series_id)
)
My challenge is to now quickly, and with SQL only (I have Java code and a partial query that solves this), construct a query that results the following:
For all series, at a particular date in time (query parameter), return back the measurement that is the most recent (maximum "date") that was effective at the particular date in time being queried.
My current "all-SQL" solution is a view, combined with a query over the view:
DDL for the view:
CREATE OR REPLACE VIEW public.known_at AS
SELECT o.date,
o.effective_end,
o.effective_start,
o.series_id,
o.value
FROM measurement o
JOIN ( SELECT o_1.series_id,
min(o_1.effective_start) AS effective_start,
o_1.date
FROM measurement o_1
GROUP BY o_1.series_id, o_1.date) x ON o.series_id::text = x.series_id::text AND o.effective_start = x.effective_start AND o.date = x.date
JOIN ( SELECT o_1.series_id,
o_1.effective_start,
max(o_1.date) AS date
FROM measurement o_1
GROUP BY o_1.series_id, o_1.effective_start) y ON x.series_id::text = y.series_id::text AND x.effective_start = y.effective_start AND x.date = y.date
WHERE o.date <= o.effective_start
ORDER BY o.date DESC, o.series_id DESC;
Query:
select k.* from known_at k
inner join (
select
k.series_id,
max(k.date) as date
from known_at k
-- the passed in date here is a parameter as described above
where k.date <= '2020-03-26'
group by k.series_id) as mx
on k.series_id = mx.series_id and k.date = mx.date
order by k.series_id;
Unfortunately, the combination view and query is slow (~400ms) despite btree indices on series_id, date, effective_end, and effective_start. How can I do better?

Answer 1

I think this query should give you the results you want, though without having your dataset it's hard to say what its performance would be like. For this query, I'd recommend a multi-column index on (effective_start, effective_end, series_id, date DESC).
SELECT DISTINCT ON (series_id) *
FROM measurement
WHERE effective_start <= '2020-03-26' -- the passed-in date
AND effective_end >= '2020-03-26' -- the passed-in date
ORDER BY series_id, date DESC;
Explanation: The query filters for rows that include the passed-in date within the effective period, then for each series_id in the filtered rows, the row with the max date is taken.
Also, you may want to consider using a daterange type for the effective dates. Range types come with some useful range operators.