Question

I have a table that looks like this:

pk client value date
1  001     564  2012/5/1
2  002     245  2012/6/1
3  003     445  2012/6/6
4  001     845  2012/7/1
5  002     567  2012/8/1
6  001     123  2012/9/1

I know this can be solved with the greatest n per group and a self join, but I’m having trouble figuring it out.

Basically this is what I want for my output

client min(value) max(value) date_for_min(value) date_for_max(value)
001    123        845        2012/9/1            2012/7/1
002    245        567        2012/6/1            2012/8/1
003    445        445        2012/6/6            2012/6/6

The tricky part is getting only one row for each client with the min / max values and then the other columns that go along with those min / max values. Any ideas?

Answer 1

In the case there are multiple rows for some min or max value (for same client), I have given you the earliest date that they appear:

select t1.client, t1.MinValue, t1.MaxValue, min(t2.date) as date_for_min_value, min(t3.date) as date_for_max_value
from (
    select client, min(value) as MinValue, max(value) as MaxValue
    from MyTable
    group by client
) t1
inner join MyTable t2 on t1.client = t2.client and t1.MinValue = t2.Value
inner join MyTable t3 on t1.client = t3.client and t1.MaxValue = t3.Value
group by t1.client, t1.MinValue, t1.MaxValue