12.21.1 Window Function Descriptions

CUME_DIST() over_clause

Returns the cumulative distribution of a value within a group of values; that is, the percentage of partition values less than or equal to the value in the current row. This represents the number of rows preceding or peer with the current row in the window ordering of the window partition divided by the total number of rows in the window partition. Return values range from 0 to 1.

This function should be used with ORDER BY to sort partition rows into the desired order. Without ORDER BY, all rows are peers and have value N/N = 1, where N is the partition size.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”.

The following query shows, for the set of values in the val column, the CUME_DIST() value for each row, as well as the percentage rank value returned by the similar PERCENT_RANK() function. For reference, the query also displays row numbers using ROW_NUMBER():

DENSE_RANK() over_clause

Returns the rank of the current row within its partition, without gaps. Peers are considered ties and receive the same rank. This function assigns consecutive ranks to peer groups; the result is that groups of size greater than one do not produce noncontiguous rank numbers. For an example, see the RANK() function description.

This function should be used with ORDER BY to sort partition rows into the desired order. Without ORDER BY, all rows are peers.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”.

FIRST_VALUE(expr) [null_treatment] over_clause

Returns the value of expr from the first row of the window frame.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”. null_treatment is as described in the section introduction.

The following query demonstrates FIRST_VALUE(), LAST_VALUE(), and two instances of NTH_VALUE():

Each function uses the rows in the current frame, which, per the window definition shown, extends from the first partition row to the current row. For the NTH_VALUE() calls, the current frame does not always include the requested row; in such cases, the return value is NULL.

LAG(expr [, N[, default]]) [null_treatment] over_clause

Returns the value of expr from the row that lags (precedes) the current row by N rows within its partition. If there is no such row, the return value is default. For example, if N is 3, the return value is default for the first two rows. If N or default are missing, the defaults are 1 and NULL, respectively.

N must be a literal nonnegative integer. If N is 0, expr is evaluated for the current row.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”. null_treatment is as described in the section introduction.

LAG() (and the similar LEAD() function) are often used to compute differences between rows. The following query shows a set of time-ordered observations and, for each one, the LAG() and LEAD() values from the adjoining rows, as well as the differences between the current and adjoining rows:

In the example, the LAG() and LEAD() calls use the default N and default values of 1 and NULL, respectively.

The first row shows what happens when there is no previous row for LAG(): The function returns the default value (in this case, NULL). The last row shows the same thing when there is no next row for LEAD().

LAG() and LEAD() also serve to compute sums rather than differences. Consider this data set, which contains the first few numbers of the Fibonacci series:

The following query shows the LAG() and LEAD() values for the rows adjacent to the current row. It also uses those functions to add to the current row value the values from the preceding and following rows. The effect is to generate the next number in the Fibonacci series, and the next number after that:

One way to generate the initial set of Fibonacci numbers is to use a recursive common table expression. For an example, see Fibonacci Series Generation.

LAST_VALUE(expr) [null_treatment] over_clause

Returns the value of expr from the last row of the window frame.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”. null_treatment is as described in the section introduction.

For an example, see the FIRST_VALUE() function description.

LEAD(expr [, N[, default]]) [null_treatment] over_clause

Returns the value of expr from the row that leads (follows) the current row by N rows within its partition. If there is no such row, the return value is default. For example, if N is 3, the return value is default for the last two rows. If N or default are missing, the defaults are 1 and NULL, respectively.

N must be a literal nonnegative integer. If N is 0, expr is evaluated for the current row.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”. null_treatment is as described in the section introduction.

For an example, see the LAG() function description.

NTH_VALUE(expr, N) [from_first_last] [null_treatment] over_clause

Returns the value of expr from the N-th row of the window frame. If there is no such row, the return value is NULL.

N must be a literal positive integer.

from_first_last is part of the SQL standard, but the MySQL implementation permits only FROM FIRST (which is also the default). This means that calculations begin at the first row of the window. FROM LAST is parsed, but produces an error. To obtain the same effect as FROM LAST (begin calculations at the last row of the window), use ORDER BY to sort in reverse order.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”. null_treatment is as described in the section introduction.

For an example, see the FIRST_VALUE() function description.

NTILE(N) over_clause

Divides a partition into N groups (buckets), assigns each row in the partition its bucket number, and returns the bucket number of the current row within its partition. For example, if N is 4, NTILE() divides rows into four buckets. If N is 100, NTILE() divides rows into 100 buckets.

N must be a literal positive integer. Bucket number return values range from 1 to N.

This function should be used with ORDER BY to sort partition rows into the desired order.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”.

The following query shows, for the set of values in the val column, the percentile values resulting from dividing the rows into two or four groups. For reference, the query also displays row numbers using ROW_NUMBER():

PERCENT_RANK() over_clause

Returns the percentage of partition values less than the value in the current row, excluding the highest value. Return values range from 0 to 1 and represent the row relative rank, calculated as the result of this formula, where rank is the row rank and rows is the number of partition rows:

(rank - 1) / (rows - 1)

This function should be used with ORDER BY to sort partition rows into the desired order. Without ORDER BY, all rows are peers.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”.

For an example, see the CUME_DIST() function description.

RANK() over_clause

Returns the rank of the current row within its partition, with gaps. Peers are considered ties and receive the same rank. This function does not assign consecutive ranks to peer groups if groups of size greater than one exist; the result is noncontiguous rank numbers.

This function should be used with ORDER BY to sort partition rows into the desired order. Without ORDER BY, all rows are peers.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”.

The following query shows the difference between RANK(), which produces ranks with gaps, and DENSE_RANK(), which produces ranks without gaps. The query shows rank values for each member of a set of values in the val column, which contains some duplicates. RANK() assigns peers (the duplicates) the same rank value, and the next greater value has a rank higher by the number of peers minus one. DENSE_RANK() also assigns peers the same rank value, but the next higher value has a rank one greater. For reference, the query also displays row numbers using ROW_NUMBER():

ROW_NUMBER() over_clause

Returns the number of the current row within its partition. Rows numbers range from 1 to the number of partition rows.

ORDER BY affects the order in which rows are numbered. Without ORDER BY, row numbering is nondeterministic.

ROW_NUMBER() assigns peers different row numbers. To assign peers the same value, use RANK() or DENSE_RANK(). For an example, see the RANK() function description.

over_clause is as described in Section 12.21.2, “Window Function Concepts and Syntax”.