SQL: group by without using aggregate function

Question

I have the following SQL query:

select * from my_table

then the returned results looks like:

my_id    field_1    field_2   ...
 1         ...       ...
 1         ...       ... 
 1         ...       ...
 2
 3
 3

I only want to keep one record per my_id , perhaps taking the record with minimum value in field_2

Therefore, I am expecting the following query to fail becauseI haven't put a desired aggregation function after select :

select * from my_table group by my_id order by my_id

However, the query went through and returned table have no duplicated my_id . Therefore, I am wondering is there a default method SQL is using if I didn't specify an aggregation function before group by ?

Answer 1

I am expecting the following query to fail because I haven't put a desired aggregation ..

Unfortunately, some SQL Implementations allow GROUP BY without an aggregate function ¹ .

In this case the result is "not defined / implementation defined" - see the specific implementation documentation to see if it provides any guarantees. That is, while it is still guaranteed that the my_id column is unique, values from any row could be returned for other output fields .

(Obviously, if my_id were a key / candidate key - but it is not in this case - then this doesn't make a difference as only one row could be selected..)

For increased compatibility and predictable results, use an aggregate function for every field which is not covered by the GROUP BY - even if the particular SQL/RDBMS does not enforce or "require" the use aggregates.

---

¹ While the original query is "accepted" in MySQL ( depending on ONLY_FULL_GROUP_BY ), both PostgreSQL and SQL Server would have rejected the original query "as expected".

Answer 2

You can use a correlated subquery:

select t.*
from my_table t
where t.field2 = (select min(t2.field2)
                  from my_table t2
                  where t2.my_id = t.my_id
                 );