Top 5 unique rows without Distinct from 7Million rows - postgresql

Question

Postgres 12 on GCP.
Table with approx 7-Million rows and growing.
select Distinct col1, col2
from tab_a
where col3='abc'
and col4='xyz'
order by col2
limit 5;
with Distinct this query take around 2.1 to 2.8 sec
without Distinct it took 0.25 sec, but my table got duplicate data as per business requirement.
Is there anyway I can get top 5 unique records without costly Distinct clause?
I can do following but its not the robust solution:
select Distinct
col1, col2
from (
select col1, col2
from tab_a
where col3='abc'
and col4='xyz'
order by col2
limit 50
)
limit 5;
Can someone guide me a more robust solution please?
Wishes

Answer 1

SQL, especially for large tables, relies on indexes to run queries efficiently. You didn't tell us anything about your indexes, so there's some guessing in this answer.
Creating the following index should help your first query a lot. There's nothing wrong with that first query: if it meets your business requirements go ahead and use it.
CREATE INDEX CONCURRENTLY tab_a_distinct
ON tab_a USING BTREE
(col3, col4, col2, col1);
Why will this help?
BTREE indexes work as if they were sorted in the order of the items in the index.
You match for equality on col3 and col4 so postgreSQL can random-access the BTREE index to the first matching row, and then scan the index sequentially until it finds no more matching rows.
You want your output ordered by col2, so that column is next. postgreSQL will scan the index until it has the five rows you need then stop.
You want DISTINCT values of col2 and col3. postgreSQL can get them from the index.
In other words, this is a covering composite index for your query.
Read this to learn more. https://use-the-index-luke.com/