Question

I’m quite new to Oracle and I recently ran into a problem :

I have a table with the following columns : person_id, account_login, add_date.

person_id | account_login | add_date
----------|---------------|------------
1         | user1         | 2012-07-10
1         | user2         | 2012-07-09
2         | user3         | 2012-07-05
2         | user4         | 2012-07-04

A person can have multiple account_logins. For every person, I want to fetch the oldest account_login :

person_id | account_login | add_date
----------|---------------|------------
1         | user2         | 2012-07-09
2         | user4         | 2012-07-04

So what I did was :

select
  person_id,
  MAX(account_login) KEEP (DENSE_RANK FIRST ORDER BY add_date)
from
  table
group by
  person_id;

I get exactly the result I want, but I think the syntax of my query is far from good.. Oracle forces me to use an aggregate function and I used MAX() or MIN() knowing that it’s completely senseless..

Is there a way to write the query better than this? (Oracle 11g)

Answer 1

You can use DENSE_RANK as an analytic alone, and select from the resulting query:

SELECT person_id, account_login, add_date
  FROM (SELECT table.*
             , DENSE_RANK() OVER (PARTITION BY person_id ORDER BY add_date) rnk
         FROM table)
 WHERE rnk = 1;

Not sure which version you consider more elegant. It also does not address what to do if you have two accounts created on the same date for a particular person_id. If you’ve also got a time component in your add_date then you should be okay. If ties are likely, perhaps ROW_NUMBER() would give you a more desirable outcome.