Updated on 2024-09-03 GMT+08:00

Window Functions

Regular aggregate functions return a single value calculated from values in a row, or group all rows into a single output row. Window functions perform a calculation across a set of rows and return a value for each row.

A window function call represents the application of an aggregate-like function over some portion of the rows selected by a query. Therefore, aggregate functions (Aggregate Functions) can also be used as window functions. A window function can scan all rows and display the raw data and aggregation analysis results at the same time.

Precautions

  • Column-store tables support only the window functions rank (expression) and row_number (expression) and the aggregate functions sum, count, avg, min, and max. Row-store tables do not have such restrictions.
  • A single query can contain one or more window function expressions.
  • Window functions can appear only in output columns. If you want to use the values of a window function for condition filtering, you need to nest the window function in the subquery and use the aliases of the window function expression at the outer layer for condition filtering. Example:
    SELECT classid, id, score FROM(SELECT *, avg(score) OVER(PARTITION BY classid) as avg_score FROM score) WHERE score >= avg_score;
  • In the query block where the window function is located, the GROUP BY expression can be used for grouping and deduplication. In this case, the PARTITION BY clause in the window function must be a subset of the GROUP BY expression to ensure that the window function performs calculation on the deduplication result. The expression of the ORDER BY clause must be a subset of the GROUP BY expression or an aggregate function of an aggregate operation. Example:
    SELECT classid,rank() OVER(PARTITION BY classid ORDER BY sum(score)) as avg_score FROM score GROUP BY classid, id;

Syntax

A window function uses the OVER clause to define a window. The OVER clause is used for grouping data and sorting the elements in a group. Window functions are used for generating sequence numbers for the values in the group.

1
2
3
4
function_name ([expression [, expression ... ]]) OVER ( window_definition ) 
function_name ([expression [, expression ... ]]) OVER window_name
function_name ( * ) OVER ( window_definition ) 
function_name ( * ) OVER window_name

window_definition is defined as follows:

1
2
3
4
[ existing_window_name ] 
[ PARTITION BY expression [, ...] ]
[ ORDER BY expression [ ASC | DESC | USING operator ] [ NULLS { FIRST | LAST } ] [, ...] ]
[ frame_clause ]

The PARTITION BY option specifies that rows with the same PARTITION BY expression value are grouped.

The ORDER BY option is used to control the order in which the window function processes rows. ORDER BY must be followed by a column name. If ORDER BY is followed by a number, the number is processed as a constant and does not sort the target column.

frame_clause is defined as follows:

1
2
[ RANGE | ROWS ] frame_start 
[ RANGE | ROWS ] BETWEEN frame_start AND frame_end

When you need to specify a window to calculate the results of all rows in a group, you need to specify the start row and end row of the window range. The window range supports the RANGE and ROWS modes. The ROWS mode specifies the window by the physical unit (row), and the RANGE mode specifies the window as the logical offset.

In RANGE and ROWS, you can use BETWEEN frame_start AND frame_end to specify the window's first and last rows. If only frame_start is specified, frame_end is CURRENT ROW by default.

The values of frame_start and frame_end are as follows:

  • CURRENT ROW: The current row is used as the window frame's start or end point.
  • N PRECEDING: The window frame starts from the nth row to the current row.
  • UNBOUNDED PRECEDING: The window frame starts at the first row of the partition.
  • N FOLLOWING: The window frame starts from the current row to the nth row.
  • UNBOUNDED FOLLOWING: The window frame ends with the last row of the partition.

frame_start cannot be UNBOUNDED FOLLOWING, frame_end cannot be UNBOUNDED PRECEDING, and frame_end cannot be earlier than frame_start. For example, RANGE BETWEEN CURRENT ROW AND N PRECEDING is not allowed.

For clusters of version 8.3.0.100 or later, the LAST_VALUE function supports the IGNORE NULLS syntax. This syntax returns the last value in a non-null window. If all values are NULL, NULL is returned. The format is as follows:

1
LAST_VALUE (expression [IGNORE NULLS]) OVER (window_definition) 

Currently, only ROWS between CURRENT ROW and UNBOUNDED FOLLOWING and ROWS BETWEEN UNBOUNDED PRECEDING and CURRENT ROW are supported.

RANK()

Description: Generates non-consecutive sequence numbers for the values in each group. The same values have the same rank value but with sequence numbers.

Return type: bigint

Example:

In the score(id, classid, score) table, the rows are student ID, class ID, and exam score.

Use the RANK function to sort student scores.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
CREATE TABLE score(id int,classid int,score int);
INSERT INTO score VALUES(1,1,95),(2,2,95),(3,2,85),(4,1,70),(5,2,88),(6,1,70);

SELECT id, classid, score,RANK() OVER(ORDER BY score DESC) FROM score;
 id | classid | score | rank
----+---------+-------+------
  1 |       1 |    95 |    1
  2 |       2 |    95 |    1
  6 |       1 |    70 |    5
  4 |       1 |    70 |    5
  5 |       2 |    88 |    3
  3 |       2 |    85 |    4
(6 rows)

ROW_NUMBER()

Description: Generates consecutive sequence numbers for the values in each group. The same values have different sequence numbers.

Return type: bigint

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id, classid, score,ROW_NUMBER() OVER(ORDER BY score DESC) FROM score ORDER BY score DESC;
 id | classid | score | row_number
----+---------+-------+------------
  1 |       1 |    95 |          1
  2 |       2 |    95 |          2
  5 |       2 |    88 |          3
  3 |       2 |    85 |          4
  6 |       1 |    70 |          5
  4 |       1 |    70 |          6
(6 rows)

DENSE_RANK()

Description: Generates consecutive sequence numbers for the values in each group. The same values have the same rank value number and the same sequence number.

Return type: bigint

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id, classid, score,DENSE_RANK() OVER(ORDER BY score DESC) FROM score;
 id | classid | score | dense_rank
----+---------+-------+------------
  1 |       1 |    95 |          1
  2 |       2 |    95 |          1
  5 |       2 |    88 |          2
  3 |       2 |    85 |          3
  6 |       1 |    70 |          4
  4 |       1 |    70 |          4
(6 rows)

PERCENT_RANK()

Description: Generates corresponding sequence numbers for the values in each group. That is, the function calculates the value according to the formula Sequence number = (Rank – 1)/(Total rows – 1). Rank is the corresponding sequence number generated based on the RANK function for the value and Total rows is the total number of elements in a group.

Return type: double precision

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id, classid, score,PERCENT_RANK() OVER(ORDER BY score DESC) FROM score;
 id | classid | score | percent_rank
----+---------+-------+--------------
  1 |       1 |    95 |            0
  2 |       2 |    95 |            0
  3 |       2 |    85 |           .6
  4 |       1 |    70 |           .8
  5 |       2 |    88 |           .4
  6 |       1 |    70 |           .8
(6 rows)

CUME_DIST()

Description: Generates accumulative distribution sequence numbers for the values in each group. That is, the function calculates the value according to the following formula: Sequence number = Number of rows preceding or peer with current row/Total rows.

Return type: double precision

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id,classid,score,CUME_DIST() OVER(ORDER BY score DESC) FROM score;
 id | classid | score |    cume_dist
----+---------+-------+------------------
  1 |       1 |    95 | .333333333333333
  2 |       2 |    95 | .333333333333333
  5 |       2 |    88 |               .5
  3 |       2 |    85 | .666666666666667
  4 |       1 |    70 |                1
  6 |       1 |    70 |                1
(6 rows)

NTILE(num_buckets integer)

Description: Equally allocates sequential data sets to the buckets whose quantity is specified by num_buckets according to num_buckets integer and allocates the bucket number to each row. Divide the partition as equally as possible.

Return type: integer

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id,classid,score,NTILE(3) OVER(ORDER BY score DESC) FROM score;
 id | classid | score | ntile
----+---------+-------+-------
  1 |       1 |    95 |     1
  2 |       2 |    95 |     1
  5 |       2 |    88 |     2
  3 |       2 |    85 |     2
  4 |       1 |    70 |     3
  6 |       1 |    70 |     3
(6 rows)

LAG(value any [, offset integer [, default any ]])

Description: Generates lag values for the corresponding values in each group. That is, the value of the row obtained by moving forward the row corresponding to the current value by offset (integer) is the sequence number. If the row does not exist after the moving, the result value is the default value. If omitted, offset defaults to 1 and default to null.

Return type: same as the parameter type

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id,classid,score,LAG(id,3) OVER(ORDER BY score DESC) FROM score;
 id | classid | score | lag
----+---------+-------+-----
  1 |       1 |    95 |
  2 |       2 |    95 |
  5 |       2 |    88 |
  3 |       2 |    85 |   1
  4 |       1 |    70 |   2
  6 |       1 |    70 |   5
(6 rows)

LEAD(value any [, offset integer [, default any ]])

Description: Generates leading values for the corresponding values in each group. That is, the value of the row obtained by moving backward the row corresponding to the current value by offset (integer) is the sequence number. If the number of rows after the moving exceeds the total number for the current group, the result value is the default value. If omitted, offset defaults to 1 and default to null.

Return type: same as the parameter type

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id,classid,score,LEAD(id,3) OVER(ORDER BY score DESC) FROM score;
 id | classid | score | lead
----+---------+-------+------
  1 |       1 |    95 |    3
  2 |       2 |    95 |    4
  5 |       2 |    88 |    6
  3 |       2 |    85 |
  4 |       1 |    70 |
  6 |       1 |    70 |
(6 rows)

FIRST_VALUE(value any)

Description: Returns the first value of each group.

Return type: same as the parameter type

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id,classid,score,FIRST_VALUE(id) OVER(ORDER BY score DESC) FROM score;
 id | classid | score | first_value
----+---------+-------+-------------
  1 |       1 |    95 |           1
  2 |       2 |    95 |           1
  5 |       2 |    88 |           1
  3 |       2 |    85 |           1
  4 |       1 |    70 |           1
  6 |       1 |    70 |           1
(6 rows)

LAST_VALUE(value any)

Description: Returns the last value of each group.

Return type: same as the parameter type

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id,classid,score,LAST_VALUE(id) OVER(ORDER BY score DESC) FROM score;
 id | classid | score | last_value
----+---------+-------+------------
  1 |       1 |    95 |          2
  2 |       2 |    95 |          2
  5 |       2 |    88 |          5
  3 |       2 |    85 |          3
  4 |       1 |    70 |          6
  6 |       1 |    70 |          6
(6 rows)

NTH_VALUE(value any, nth integer)

Description: The nth row for a group is the returned value. If the row does not exist, NULL is returned by default.

Return type: same as the parameter type

Example:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
SELECT id,classid,score,NTH_VALUE(id,3) OVER(ORDER BY score DESC) FROM score;
 id | classid | score | nth_value
----+---------+-------+-----------
  1 |       1 |    95 |
  2 |       2 |    95 |
  5 |       2 |    88 |         5
  3 |       2 |    85 |         5
  4 |       1 |    70 |         5
  6 |       1 |    70 |         5
(6 rows)