SQL window functions

Window functions let you calculate running totals, moving averages, or other aggregate-like results without collapsing rows into groups. They perform their calculations over a “window” of rows, which you can partition and order in various ways, and return a calculated value for each row in the set.

Unlike non-window aggregate functions that combine each group into a single row, window functions preserve each row’s identity and calculate an additional value for every row in the partition.

For example, the following query uses the Home sensor sample data and returns each temperature reading with the average temperature per room over the queried time range:

SELECT
  time,
  room,
  temp,
  avg(temp) OVER (PARTITION BY room) AS avg_room_temp
FROM
  home
WHERE
  time >= '2025-03-29T08:00:00Z'
  AND time <= '2025-03-29T09:00:00Z'
ORDER BY
  room,
  time

• Window frames
• Window function syntax
  • OVER clause
  • PARTITION BY clause
  • ORDER BY clause
  • Frame clause
    • Frame units
    • Frame boundaries
  • WINDOW clause
• Aggregate functions
• Ranking Functions
  • cume_dist
  • dense_rank
  • ntile
  • percent_rank
  • rank
  • row_number
• Analytical Functions
  • first_value
  • lag
  • last_value
  • lead
  • nth_value

Window frames

As window functions operate on a row, there is a set of rows in the row’s partition that the window function uses to perform the operation. This set of rows is called the window frame. Window frame boundaries can be defined using RANGE, ROW, or GROUPS frame units, each relative to the current row–for example:

SELECT
  time,
  temp,
  avg(temp) OVER (
    ORDER BY time
    RANGE INTERVAL '3 hours' PRECEDING
  ) AS 3h_moving_avg
FROM home
WHERE room = 'Kitchen'

SELECT
  time,
  temp,
  avg(temp) OVER (
    ROWS 3 PRECEDING
  ) AS moving_avg
FROM home
WHERE room = 'Kitchen'

SELECT
  time,
  room,
  temp,
  avg(temp) OVER (
    ORDER BY room
    GROUPS 1 PRECEDING
  ) AS moving_avg
FROM home

For more information about how window frames work, see the frame clause.

If you don’t specify window frames, window functions use all rows in the current partition to perform their operation.

function([expr])
  OVER(
    [PARTITION BY expr[, …]]
    [ORDER BY expr [ ASC | DESC ][, …]]
    [ frame_clause ]
    )

OVER clause

Window functions use an OVER clause that directly follows the window function’s
name and arguments.
The OVER clause syntactically distinguishes a window
function from a non-window or aggregate function and defines how to group and order rows for the window operation.

PARTITION BY clause

The PARTITION BY clause in the OVER clause divides the rows into groups, or partitions, that share the same values of the PARTITION BY expressions. The window function operates on all the rows in the same partition as the current row.

ORDER BY clause

The ORDER BY clause inside of the OVER clause controls the order that the window function processes rows in each partition. When a window clause contains an ORDER BY clause, the window frame boundaries may be explicit or implicit, limiting a window frame size in both directions relative to the current row.

Frame clause

The frame clause defines window frame boundaries and can be one of the following:

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

• Frame units
  • RANGE
  • ROWS
  • GROUPS
• Frame boundaries
  • UNBOUNDED PRECEDING
  • offset PRECEDING
  • CURRENT ROW
  • offset FOLLOWING
  • UNBOUNDED FOLLOWING

Frame units

When defining window frames, you can use one of the following frame units:

• RANGE
• ROWS
• GROUPS

RANGE

Defines frame boundaries using rows with values for columns specified in the ORDER BY clause within a value range relative to the current row value.

The offset is the difference between the current row value and surrounding row values. RANGE supports the following offset types:

• Numeric (non-negative)
• Numeric string (non-negative)
• Interval

... OVER (
  ORDER BY wind_direction
  RANGE BETWEEN 45 PRECEDING AND 45 FOLLOWING
)

... OVER (
  ORDER BY time
  RANGE BETWEEN
    INTERVAL '3 hours' PRECEDING
    AND INTERVAL '1 hour' FOLLOWING
)

ROWS

Defines window frame boundaries using row positions relative to the current row. The offset is the difference in row position from the current row. ROWS supports the following offset types:

• Numeric (non-negative)
• Numeric string (non-negative)

... OVER (
  ROWS BETWEEN 2 PRECEDING AND 1 FOLLOWING
)

GROUPS

Defines window frame boundaries using row groups. Rows with the same values for the columns in the ORDER BY clause comprise a row group.

The offset is the difference in row group position relative to the current row group. GROUPS supports the following offset types:

• Numeric (non-negative)
• Numeric string (non-negative)

... OVER (
  ORDER BY country, city
  GROUPS ...
)

... OVER (
  ORDER BY country, city
  GROUPS 2 PRECEDING
)

Frame boundaries

Frame boundaries (frame_start and frame_end) define the boundaries of each frame that the window function operates on.

• UNBOUNDED PRECEDING
• offset PRECEDING
• CURRENT_ROW](#current-row)
• offset> FOLLOWING
• UNBOUNDED FOLLOWING

UNBOUNDED PRECEDING

Starts at the first row of the partition and ends at the current row.

UNBOUNDED PRECEDING

offset PRECEDING

Starts at offset frame units before the current row and ends at the current row. For example, 3 PRECEDING includes 3 rows before the current row.

<offset> PRECEDING

CURRENT ROW

Both starts and ends at the current row when used as a boundary.

CURRENT ROW

offset FOLLOWING

Starts at the current row and ends at offset frame units after the current row. For example, 3 FOLLOWING includes 3 rows after the current row.

<offset> FOLLOWING

UNBOUNDED FOLLOWING

Starts at the current row and ends at the last row of the partition.

offset FOLLOWING

Use a specified offset of frame units after the current row as a frame boundary.

offset FOLLOWING

UNBOUNDED FOLLOWING

Use the current row to the end of the current partition the frame boundary.

UNBOUNDED FOLLOWING

WINDOW clause

Use the WINDOW clause to define a reusable alias for a window specification. This is useful when multiple window functions in your query share the same window definition.

Instead of repeating the same OVER clause for each function, define the window once and reference it by alias–for example:

SELECT
  sum(net_gain) OVER w,
  avg(net_net) OVER w
FROM
  finance
WINDOW w AS ( PARTITION BY ticker ORDER BY time DESC);

Aggregate functions

All aggregate functions can be used as window functions.

Ranking Functions

• cume_dist
• dense_rank
• ntile
• percent_rank
• rank
• row_number

cume_dist

Returns the cumulative distribution of a value within a group of values. The returned value is greater than 0 and less than or equal to 1 and represents the relative rank of the value in the set of values. The ORDER BY clause in the OVER clause is used to correctly calculate the cumulative distribution of the current row value.

cume_dist()

SELECT
  time,
  room,
  temp,
  cume_dist() OVER (
    PARTITION BY room
    ORDER BY temp 
  ) AS cume_dist
FROM home
WHERE
  time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T12:00:00Z'

dense_rank

Returns the rank of the current row in its partition. Ranking is consecutive; assigns duplicate values the same rank number and the rank sequence continues with the next distinct value (unlike rank()).

The ORDER BY clause in the OVER clause determines ranking order.

dense_rank()

SELECT
  time,
  room,
  temp,
  dense_rank() OVER (
    PARTITION BY room
    ORDER BY temp 
  ) AS dense_rank
FROM home
WHERE
  time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T12:00:00Z'

SELECT
  id,
  rank() OVER(ORDER BY id),
  dense_rank() OVER(ORDER BY id),
  row_number() OVER(ORDER BY id)
FROM my_table;

ntile

Distributes the rows in an ordered partition into the specified number of groups. Each group is numbered, starting at one. For each row, ntile returns the group number to which the row belongs. Group numbers range from 1 to the expression value, dividing the partition as equally as possible. The ORDER BY clause in the OVER clause determines ranking order.

ntile(expression)

Arguments

• expression: An integer. The number of groups to split the partition into.

SELECT
  time,
  temp,
  ntile(4) OVER (
    ORDER BY time 
  ) AS ntile
FROM home
WHERE
  room = 'Kitchen'
  AND time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T15:00:00Z'

percent_rank

Returns the percentage rank of the current row within its partition. The returned value is between 0 and 1, computed as:

(rank - 1) / (total_rows - 1)

The ORDER BY clause in the OVER clause determines the ranking order.

percent_rank()

SELECT
  time,
  room,
  temp,
  percent_rank() OVER (
    PARTITION BY room
    ORDER BY temp 
  ) AS percent_rank
FROM home
WHERE
  time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T11:00:00Z'

rank

Returns the rank of the current row in its partition. For duplicate values, rank assigns them the same rank number, skips subsequent ranks (unlike dense_rank()), and then continues ranking with the next distinct value.

The ORDER BY clause in the OVER clause determines ranking order.

rank()

SELECT
  time,
  room,
  temp,
  rank() OVER (
    PARTITION BY room
    ORDER BY temp 
  ) AS rank
FROM home
WHERE
  time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T11:00:00Z'

SELECT
  id,
  rank() OVER(ORDER BY id),
  dense_rank() OVER(ORDER BY id),
  row_number() OVER(ORDER BY id)
FROM my_table;

row_number

Returns the position of the current row in its partition, counting from 1. The ORDER BY clause in the OVER clause determines row order.

row_number()

SELECT
  time,
  room,
  temp,
  row_number() OVER (
    PARTITION BY room
    ORDER BY temp 
  ) AS row_number
FROM home
WHERE
  time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T11:00:00Z'

SELECT
  id,
  rank() OVER(ORDER BY id),
  dense_rank() OVER(ORDER BY id),
  row_number() OVER(ORDER BY id)
FROM my_table;

Analytical Functions

• first_value
• lag
• last_value
• lead
• nth_value

first_value

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

first_value(expression)

Arguments

• expression: Expression to operate on. Can be a constant, column, or function, and any combination of arithmetic operators.

Related functions

last_value

SELECT
  time,
  room,
  temp,
  first_value(temp) OVER (
    PARTITION BY room
    ORDER BY time 
  ) AS first_value
FROM home
WHERE  
  time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T11:00:00Z'
ORDER BY room, time

lag

Returns the value from the row that is at the specified offset before the current row in the partition. If the offset row is outside the partition, the function returns the specified default.

lag(expression, offset, default)

Arguments

• expression: Expression to operate on. Can be a constant, column, or function, and any combination of arithmetic or string operators.
• offset: How many rows before the current row to retrieve the value of expression from. Default is 1.
• default: The default value to return if the offset is in the partition. Must be of the same type as expression.

Related functions

lead

SELECT
  time,
  room,
  temp,
  lag(temp, 1, 0) OVER (
    PARTITION BY room
    ORDER BY time
  ) AS previous_value
FROM home
WHERE
  time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T11:00:00Z'
ORDER BY room, time

last_value

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

last_value(expression)

Arguments

• expression: Expression to operate on. Can be a constant, column, or function, and any combination of arithmetic operators.

Related functions

first_value

SELECT
  time,
  room,
  temp,
  last_value(temp) OVER (
    PARTITION BY room
    ORDER BY time
    ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING
  ) AS last_value
FROM home
WHERE  
  time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T11:00:00Z'
ORDER BY room, time

lead

Returns the value from the row that is at the specified offset after the current row in the partition. If the offset row is outside the partition, the function returns the specified default.

lead(expression, offset, default)

Arguments

• expression: Expression to operate on. Can be a constant, column, or function, and any combination of arithmetic or string operators.
• offset: How many rows before the current row to retrieve the value of expression from. Default is 1.
• default: The default value to return if the offset is in the partition. Must be of the same type as expression.

Related functions

lag

SELECT
  time,
  room,
  temp,
  lead(temp, 1, 0) OVER (
    PARTITION BY room
    ORDER BY time
  ) AS next_value
FROM home
WHERE
  time >= '2025-03-29T08:00:00Z'
  AND time < '2025-03-29T11:00:00Z'
ORDER BY room, time

nth_value

Returns the value from the row that is the nth row of the window frame (counting from 1). If the nth row doesn’t exist, the function returns null.

nth_value(expression, n)

Arguments

• expression: The expression to operator on. Can be a constant, column, or function, and any combination of arithmetic or string operators.
• n: Specifies the row number in the current frame and partition to reference.

SELECT
  time,
  room,
  temp,
  nth_value(temp, 2) OVER (
    PARTITION BY room
  ) AS "2nd_temp"
FROM home
WHERE  
  time >= '2025-02-10T08:00:00Z'
  AND time < '2025-02-10T11:00:00Z'