Question

I have a simple table with a unit_id oid, time timestamp, diag bytea. The primary key is a combination of both time and unit_id.

The idea behind this query is to get the latest row (largest timestamp) for each unique unit_id. However the rows for each unit_id with the latest time are not always returned.

I really want to group by just the unit_id, but postgres makes me use diag also, since I am selecting that.

SELECT DISTINCT ON(unit_id) max(time) as time, diag, unit_id 
FROM diagnostics.unit_diag_history  
GROUP BY unit_id, diag

Answer 1

Any time you start thinking that you want a localized GROUP BY you should start thinking about window functions instead.

I think you’re after something like this:

select unit_id, time, diag
from (
    select unit_id, time, diag,
           rank() over (partition by unit_id order by time desc) as rank
    from diagnostics.unit_diag_history
) as dt
where rank = 1

You might want to add something to the ORDER BY to consistently break ties as well but that wouldn’t alter the overall technique.