Hive QL - Limiting number of rows per each item

Eric Philmore picture Eric Philmore · Aug 1, 2012 · Viewed 22.4k times · Source

If I have multiple items listed in a where clause How would one go about limiting the results to N for each item in the list?

EX:

select a_id,b,c, count(*), as sumrequests
from table_name
where
a_id in (1,2,3)
group by a_id,b,c
limit 10000

Answer

Carter Shanklin picture Carter Shanklin · Feb 6, 2014

Sounds like your question is to get the top N per a_id. You can do this with a window function, introduced in Hive 11. Something like:

SELECT a_id, b, c, count(*) as sumrequests
FROM (
    SELECT a_id, b, c, row_number() over (Partition BY a_id) as row
    FROM table_name
    ) rs
WHERE row <= 10000
AND a_id in (1, 2, 3)
GROUP BY a_id, b, c;

This will output up to 10,000 randomly-chosen rows per a_id. You can partition it further if you're looking to group by more than just a_id. You can also use order by in the window functions, there are a lot of examples out there to show additional options.