12.21.2 Window Function Concepts and Syntax

This section describes how to use window functions. Examples use the same sales information data set as found in the discussion of the GROUPING() function in Section 12.20.2, “GROUP BY Modifiers”:

A window function performs an aggregate-like operation on a set of query rows. However, whereas an aggregate operation groups query rows into a single result row, a window function produces a result for each query row:

For example, using the sales information table, these two queries perform aggregate operations that produce a single global sum for all rows taken as a group, and sums grouped per country:

By contrast, window operations do not collapse groups of query rows to a single output row. Instead, they produce a result for each row. Like the preceding queries, the following query uses SUM(), but this time as a window function:

Each window operation in the query is signified by inclusion of an OVER clause that specifies how to partition query rows into groups for processing by the window function:

Window functions are permitted only in the select list and ORDER BY clause. Query result rows are determined from the FROM clause, after WHERE, GROUP BY, and HAVING processing, and windowing execution occurs before ORDER BY, LIMIT, and SELECT DISTINCT.

The OVER clause is permitted for many aggregate functions, which therefore can be used as window or nonwindow functions, depending on whether the OVER clause is present or absent:

For details about each aggregate function, see Section 12.20.1, “Aggregate (GROUP BY) Function Descriptions”.

MySQL also supports nonaggregate functions that are used only as window functions. For these, the OVER clause is mandatory:

For details about each nonaggregate function, see Section 12.21.1, “Window Function Descriptions”.

As an example of one of those nonaggregate window functions, this query uses ROW_NUMBER(), which produces the row number of each row within its partition. In this case, rows are numbered per country. By default, partition rows are unordered and row numbering is nondeterministic. To sort partition rows, include an ORDER BY clause within the window definition. The query uses unordered and ordered partitions (the row_num1 and row_num2 columns) to illustrate the difference between omitting and including ORDER BY:

As mentioned previously, to use a window function (or treat an aggregate function as a window function), include an OVER clause following the function call. The OVER clause has two forms:

Both forms define how the window function should process query rows. They differ in whether the window is defined directly in the OVER clause, or supplied by a reference to a named window defined elsewhere in the query:

For OVER (window_spec) syntax, the window specification has several parts, all optional:

If OVER() is empty, the window consists of all query rows and the window function computes a result using all rows. Otherwise, the clauses present within the parentheses determine which query rows are used to compute the function result and how they are partitioned and ordered: