GROUP BY clause order omitting results in Oracle 11g query

Question

I have a simple query that appears to give the desired result:

select op.opr, op.last, op.dept, count(*) as counter 
from DWRVWR.BCA_M_OPRIDS1 op
where op.opr = '21B'
group by op.opr, op.last ,op.dept;

My original query returns no results. The only difference was the order of the group by clause:

select op.opr, op.last, op.dept, count(*) as counter 
from DWRVWR.BCA_M_OPRIDS1 op
where op.opr = '21B'
group by op.opr, op.dept, op.last;

In actuality, this was part of a much larger, more complicated query, but I narrowed down the problem to this. All documentation I was able to find states that the order of the group by clause doesn't matter. I really want to understand why I am getting different results, as I would have to review all of my queries that use the group by clause, if there is a potential issue. I'm using SQL Developer, if it matters.

Also, if the order of the group by clause did not matter and every field not used in an aggregate function is required to be listed in the group by clause, wouldn't the group by clause simply be redundant and seemingly unnecessary?

Answer 1

All documentation I was able to find states that the order of the group by clause doesn't matter

That's not entirely true, it depends.

The grouping functionality is not impacted by the order of columns in the GROUP BY clause. It will produce the same group set regardless of the order. Perhaps that's what those documentation that you found were referring to. However the order does matter for other aspects.

Before Oracle 10g, the GROUP BY performed implicitly an ORDER BY , so the order of the columns in the GROUP BY clause did matter. The group sets are the same, but only ordered differently. Starting with Oracle10g, if you want the result set to be in any specific order, then you must add an ORDER BY clause. Other databases have similar history.

Another case where the order matters is if you have indexes on the table. Multi-column indexes are only used if the columns exactly match the columns specified in the GROUP BY or ORDER BY clauses. So if you change the order, your query will not use the index and will perform differently. The result is the same, but the performance is not.

Also the order of the columns in the GROUP BY clause becomes important if you use some features like ROLLUP . This time the results themselves will not be the same.

It is recommended to follow the best practice of listing the fields in the GROUP BY clause in the order of the hierarchy. This makes the query more readable and more easily maintainable.

Also, if the order of the group by clause did not matter and every field not used in an aggregate function is required to be listed in the group by clause, wouldn't the group by clause simply be redundant and seemingly unnecessary?

No, the GROUP BY clause is mandatory in the standard SQL and in Oracle. There is only one exception in which you can omit the GROUP BY clause, if you want the aggregate functions to apply to the entire result set. In this case, your SELECT list must consist only of aggregate expressions.