Window Functions¶

Window functions perform calculations across rows of the query result. They run after the HAVING clause but before the ORDER BY clause. Invoking a window function requires special syntax using the OVER clause to specify the window as follows:

function(args) OVER (
    [PARTITION BY expression]
    [ORDER BY expression [ASC|DESC]]
    [frame]
)

A frame is one of:

{RANGE|ROWS|GROUPS} frame_start {RANGE|ROWS|GROUPS} BETWEEN frame_start AND frame_end

frame_start and frame_end can be any of:

UNBOUNDED PRECEDING
expression PRECEDING
CURRENT ROW
expression FOLLOWING
UNBOUNDED FOLLOWING

The window definition has 3 components:

The PARTITION BY clause separates the input rows into different partitions. This is analogous to how the GROUP BY clause separates rows into different groups for aggregate functions. If PARTITION BY is not specified, the entire input is treated as a single partition.
The ORDER BY clause determines the order in which input rows will be processed by the window function. If ORDER BY is not specified, the ordering is undefined. Note that the ORDER BY clause within window functions does not support ordinals. You need to use actual expressions
The frame clause specifies the sliding window of rows to be processed by the function for a given input row. A frame can be ROWS type, RANGE type or GROUPS type, and it runs from frame_start to frame_end. If frame_end is not specified, a default value of CURRENT ROW is used.

In ROWS mode, CURRENT ROW refers specifically to the current row. In RANGE and GROUPS mode, CURRENT ROW refers to any peer row of the current row for the purpose of the ORDER BY. If no ORDER BY is specified, all rows are considered peers of the current row. In RANGE and GROUPS mode a frame start of CURRENT ROW refers to the first peer row of the current row, while a frame end of CURRENT ROW refers to the last peer row of the current row.

In ROWS mode, frame starts and ends of expression PRECEDING or expression FOLLOWING define the start or end of the frame as the specified number of rows before or after the current row. The expression must be of type INTEGER.

In RANGE mode, frame starts and ends of expression PRECEDING or expression FOLLOWING define the start or end of the frame as the value difference of the sort key from the current row. The sort key must either be the same type of expression or can be coerced to the same type as expression.

In GROUPS mode, frame starts and ends of expression PRECEDING or expression FOLLOWING define the start or end of the frame as the number of groups from the current row. A group includes all rows with the same value on the sort key. The type of expression must be INTEGER or BIGINT.

If no frame is specified, a default frame of RANGE UNBOUNDED PRECEDING is used.

Examples¶

The following query ranks orders for each clerk by price:

SELECT orderkey, clerk, totalprice,
       rank() OVER (PARTITION BY clerk
                    ORDER BY totalprice DESC) AS rnk
FROM orders
ORDER BY clerk, rnk

The following queries demonstrate the difference between ROWS, RANGE and GROUPS in frame definition:

SELECT
    ARRAY_AGG(v) OVER (
        ORDER BY k ASC ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING
    )
FROM (
    VALUES (1, 'a'), (1, 'b'), (3, 'c'), (3, 'd'), (5, 'e')
) t(k, v); -- ['a', 'b'], ['a', 'b', 'c'], ['b', 'c', 'd'], ['c', 'd', 'e'], ['d', 'e']

SELECT
    ARRAY_AGG(v) OVER (
        ORDER BY k ASC RANGE BETWEEN 1 PRECEDING AND 1 FOLLOWING
    )
FROM (
    VALUES (1, 'a'), (1, 'b'), (3, 'c'), (3, 'd'), (5, 'e')
) t(k, v); -- ['a', 'b'], ['a', 'b'], ['c', 'd'], ['c', 'd'], ['e']

SELECT
    ARRAY_AGG(v) OVER (
        ORDER BY k ASC GROUPS BETWEEN 1 PRECEDING AND 1 FOLLOWING
    )
FROM (
    VALUES (1, 'a'), (1, 'b'), (3, 'c'), (3, 'd'), (5, 'e')
) t(k, v); -- ['a', 'b', 'c', 'd'], ['a', 'b', 'c', 'd'], ['a', 'b', 'c', 'd', 'e'], ['a', 'b', 'c', 'd', 'e'], ['c', 'd', 'e']

Aggregate Functions¶

All Aggregate Functions can be used as window functions by adding the OVER clause. The aggregate function is computed for each row over the rows within the current row’s window frame.

For example, the following query produces a rolling sum of order prices by day for each clerk:

SELECT clerk, orderdate, orderkey, totalprice,
       sum(totalprice) OVER (PARTITION BY clerk
                             ORDER BY orderdate) AS rolling_sum
FROM orders
ORDER BY clerk, orderdate, orderkey

Ranking Functions¶

cume_dist() -> double()¶: Returns the cumulative distribution of a value in a group of values. The result is the number of rows preceding or peer with the row in the window ordering of the window partition divided by the total number of rows in the window partition. Thus, any tie values in the ordering will evaluate to the same distribution value.

dense_rank() -> bigint()¶: Returns the rank of a value in a group of values. This is similar to rank(), except that tie values do not produce gaps in the sequence.

ntile(n) -> bigint()¶

Divides the rows for each window partition into n buckets ranging from 1 to at most n. Bucket values will differ by at most 1. If the number of rows in the partition does not divide evenly into the number of buckets, then the remainder values are distributed one per bucket, starting with the first bucket.

For example, with 6 rows and 4 buckets, the bucket values would be as follows: 1 1 2 2 3 4

percent_rank() -> double()¶: Returns the percentage ranking of a value in group of values. The result is (r - 1) / (n - 1) where r is the rank() of the row and n is the total number of rows in the window partition.

rank() -> bigint()¶: Returns the rank of a value in a group of values. The rank is one plus the number of rows preceding the row that are not peer with the row. Thus, tie values in the ordering will produce gaps in the sequence. The ranking is performed for each window partition.

row_number() -> bigint()¶: Returns a unique, sequential number for each row, starting with one, according to the ordering of rows within the window partition.

Value Functions¶

Value functions provide an option to specify how null values should be treated when evaluating the function. Nulls can either be ignored (IGNORE NULLS) or respected (RESPECT NULLS). By default, null values are respected. If IGNORE NULLS is specified, all rows where the value expression is null are excluded from the calculation. If IGNORE NULLS is specified and the value expression is null for all rows, the default_value is returned, or if it is not specified, null is returned.

first_value(x) -> [same as input]()¶: Returns the first value of the window.

last_value(x) -> [same as input]()¶: Returns the last value of the window.

nth_value(x, offset) -> [same as input]()¶: Returns the value at the specified offset from beginning the window. Offsets start at 1. The offset can be any scalar expression. If the offset is null or greater than the number of values in the window, null is returned. It is an error for the offset to be zero or negative.

lead(x[, offset [, default_value]]) -> [same as input]()¶: Returns the value at offset rows after the current row in the window partition. Offsets start at 0, which is the current row. The offset can be any scalar expression. The default offset is 1. If the offset is null, null is returned. If the offset refers to a row that is not within the partition, the default_value is returned, or if it is not specified null is returned.

lag(x[, offset [, default_value]]) -> [same as input]()¶: Returns the value at offset rows before the current row in the window partition. Offsets start at 0, which is the current row. The offset can be any scalar expression. The default offset is 1. If the offset is null, null is returned. If the offset refers to a row that is not within the partition, the default_value is returned, or if it is not specified null is returned.