SQL join table and limiting condition matching result - sql

Question

I have a table of rates and transactions from which I want to find out conversion rate based on the latest updated currency rate (related to transaction timestamp)
Table - rates
('2018-04-01 00:00:00', 'EUR', 'RUB', '1.71'),
('2018-04-01 01:00:05', 'EUR', 'RUB', '1.82'),
('2018-04-01 00:00:00', 'USD', 'RUB', '0.71'),
('2018-04-01 00:00:05', 'USD', 'RUB', '0.82'),
('2018-04-01 00:01:00', 'USD', 'RUB', '0.92'),
('2018-04-01 01:02:00', 'USD', 'RUB', '0.62'),
Table - transactions
('2018-04-01 00:00:00', 1, 'EUR', 2.45),
('2018-04-01 01:00:00', 1, 'EUR', 8.45),
('2018-04-01 01:30:00', 1, 'USD', 3.5),
My attempt to limit those additional data
select * from transactions tr1
left outer join rates ex1
on tr1.ts >= ex1.ts
and tr1.currency = ex1.from_currency;
The result I'm getting contains all of the exchange rate update that has happened previously
2
2018-04-01 00:00:00 1 EUR 2.45 2018-04-01 00:00:00 EUR RUB 1.71 (correct)
2018-04-01 01:00:00 1 EUR 8.45 2018-04-01 00:00:00 EUR RUB 1.71 (correct)
2018-04-01 01:30:00 1 USD 3.5 2018-04-01 01:02:00 USD RUB 0.62 (only this should remain)
2018-04-01 01:30:00 1 USD 3.5 2018-04-01 00:01:00 USD RUB 0.92
2018-04-01 01:30:00 1 USD 3.5 2018-04-01 00:00:05 USD RUB 0.82
2018-04-01 01:30:00 1 USD 3.5 2018-04-01 00:00:00 USD RUB 0.71
I tried to define my own statement (my previous query):
where ex1.ts = (select max(ex2.ts) from rates ex2
where ex2.from_currency=ex1.from_currency
and ex2.to_currency=ex1.from_currency);
But that doesn't return anything...

Answer 1

Postgres has the very handy distinct on for getting one row per group. This should do what you want:
select t.*, r.*
from transactions t left join
(select distinct on (from_currency, to_currency) r.*
from rates r
order by from_currency, to_currency, ts desc
) r
on r.from_currency = t.currency and
r.to_currency = 'RUB';
EDIT:
If you want the latest date for each transaction, then use a lateral join:
select t.*, r.*
from transactions t left join lateral
(select r.*
from rates r
where r.from_currency = t.currency and
r.ts <= t.ts
order by r.ts desc
limit 1
) r
on 1=1;

Answer 2

You can solve this using Window/Analytic functions. In the partition, we order by ex1.ts in descending order, so that we get the rate closest to the transaction timestamp ts. tr1.ts >= ex.ts condition in the Join ensures that we are only getting the exchange rates on or before the transaction time.
select dt.*
from
(
select tr1.*, ex1.*,
row_number() over (partition by ex1.from_currency order by ex1.ts desc) as rn
from transactions tr1
left outer join exchange_rates ex1
on tr1.ts >= ex1.ts
and tr1.currency = ex1.from_currency
) as dt
where dt.rn = 1