12.20.3 MySQL Handling of GROUP BY

SQL-92 and earlier does not permit queries for which the select list, HAVING condition, or ORDER BY list refer to nonaggregated columns that are not named in the GROUP BY clause. For example, this query is illegal in standard SQL-92 because the nonaggregated name column in the select list does not appear in the GROUP BY:

For the query to be legal in SQL-92, the name column must be omitted from the select list or named in the GROUP BY clause.

SQL:1999 and later permits such nonaggregates per optional feature T301 if they are functionally dependent on GROUP BY columns: If such a relationship exists between name and custid, the query is legal. This would be the case, for example, were custid a primary key of customers.

MySQL implements detection of functional dependence. If the ONLY_FULL_GROUP_BY SQL mode is enabled (which it is by default), MySQL rejects queries for which the select list, HAVING condition, or ORDER BY list refer to nonaggregated columns that are neither named in the GROUP BY clause nor are functionally dependent on them.

If ONLY_FULL_GROUP_BY is disabled, a MySQL extension to the standard SQL use of GROUP BY permits the select list, HAVING condition, or ORDER BY list to refer to nonaggregated columns even if the columns are not functionally dependent on GROUP BY columns. This causes MySQL to accept the preceding query. In this case, the server is free to choose any value from each group, so unless they are the same, the values chosen are nondeterministic, which is probably not what you want. Furthermore, the selection of values from each group cannot be influenced by adding an ORDER BY clause. Result set sorting occurs after values have been chosen, and ORDER BY does not affect which value within each group the server chooses. Disabling ONLY_FULL_GROUP_BY is useful primarily when you know that, due to some property of the data, all values in each nonaggregated column not named in the GROUP BY are the same for each group.

You can achieve the same effect without disabling ONLY_FULL_GROUP_BY by using ANY_VALUE() to refer to the nonaggregated column.

The following discussion demonstrates functional dependence, the error message MySQL produces when functional dependence is absent, and ways of causing MySQL to accept a query in the absence of functional dependence.

This query might be invalid with ONLY_FULL_GROUP_BY enabled because the nonaggregated address column in the select list is not named in the GROUP BY clause:

The query is valid if name is a primary key of t or is a unique NOT NULL column. In such cases, MySQL recognizes that the selected column is functionally dependent on a grouping column. For example, if name is a primary key, its value determines the value of address because each group has only one value of the primary key and thus only one row. As a result, there is no randomness in the choice of address value in a group and no need to reject the query.

The query is invalid if name is not a primary key of t or a unique NOT NULL column. In this case, no functional dependency can be inferred and an error occurs:

If you know that, for a given data set, each name value in fact uniquely determines the address value, address is effectively functionally dependent on name. To tell MySQL to accept the query, you can use the ANY_VALUE() function:

Alternatively, disable ONLY_FULL_GROUP_BY.

The preceding example is quite simple, however. In particular, it is unlikely you would group on a single primary key column because every group would contain only one row. For addtional examples demonstrating functional dependence in more complex queries, see Section 12.20.4, “Detection of Functional Dependence”.

If a query has aggregate functions and no GROUP BY clause, it cannot have nonaggregated columns in the select list, HAVING condition, or ORDER BY list with ONLY_FULL_GROUP_BY enabled:

Without GROUP BY, there is a single group and it is nondeterministic which name value to choose for the group. Here, too, ANY_VALUE() can be used, if it is immaterial which name value MySQL chooses:

ONLY_FULL_GROUP_BY also affects handling of queries that use DISTINCT and ORDER BY. Consider the case of a table t with three columns c1, c2, and c3 that contains these rows:

c1 c2 c3
1  2  A
3  4  B
1  2  C

Suppose that we execute the following query, expecting the results to be ordered by c3:

To order the result, duplicates must be eliminated first. But to do so, should we keep the first row or the third? This arbitrary choice influences the retained value of c3, which in turn influences ordering and makes it arbitrary as well. To prevent this problem, a query that has DISTINCT and ORDER BY is rejected as invalid if any ORDER BY expression does not satisfy at least one of these conditions:

Another MySQL extension to standard SQL permits references in the HAVING clause to aliased expressions in the select list. For example, the following query returns name values that occur only once in table orders:

The MySQL extension permits the use of an alias in the HAVING clause for the aggregated column:

Standard SQL permits only column expressions in GROUP BY clauses, so a statement such as this is invalid because FLOOR(value/100) is a noncolumn expression:

MySQL extends standard SQL to permit noncolumn expressions in GROUP BY clauses and considers the preceding statement valid.

Standard SQL also does not permit aliases in GROUP BY clauses. MySQL extends standard SQL to permit aliases, so another way to write the query is as follows:

The alias val is considered a column expression in the GROUP BY clause.

In the presence of a noncolumn expression in the GROUP BY clause, MySQL recognizes equality between that expression and expressions in the select list. This means that with ONLY_FULL_GROUP_BY SQL mode enabled, the query containing GROUP BY id, FLOOR(value/100) is valid because that same FLOOR() expression occurs in the select list. However, MySQL does not try to recognize functional dependence on GROUP BY noncolumn expressions, so the following query is invalid with ONLY_FULL_GROUP_BY enabled, even though the third selected expression is a simple formula of the id column and the FLOOR() expression in the GROUP BY clause:

A workaround is to use a derived table: