SQL Window Functions

From WikiOD

Setting up a flag if other rows have a common property[edit | edit source]

Let's say I have this data:

Table items

id name tag
1 example unique_tag
2 foo simple
42 bar simple
3 baz hello
51 quux world

I'd like to get all those lines and know if a tag is used by other lines

SELECT id, name, tag, COUNT(*) OVER (PARTITION BY tag) > 1 AS flag FROM items

The result will be:

id name tag flag
1 example unique_tag false
2 foo simple true
42 bar simple true
3 baz hello false
51 quux world false

In case your database doesn't have OVER and PARTITION you can use this to produce the same result:

SELECT id, name, tag, (SELECT COUNT(tag) FROM items B WHERE tag = A.tag) > 1 AS flag FROM items A

Getting a running total[edit | edit source]

Given this data:

date amount
2016-03-12 200
2016-03-11 -50
2016-03-14 100
2016-03-15 100
2016-03-10 -250
SELECT date, amount, SUM(amount) OVER (ORDER BY date ASC) AS running
FROM operations
ORDER BY date ASC

will give you

date amount running
2016-03-10 -250 -250
2016-03-11 -50 -300
2016-03-12 200 -100
2016-03-14 100 0
2016-03-15 100 -100

Finding "out-of-sequence" records using the LAG() function[edit | edit source]

Given these sample data:

ID STATUS STATUS_TIME STATUS_BY
1 ONE 2016-09-28-19.47.52.501398 USER_1
3 ONE 2016-09-28-19.47.52.501511 USER_2
1 THREE 2016-09-28-19.47.52.501517 USER_3
3 TWO 2016-09-28-19.47.52.501521 USER_2
3 THREE 2016-09-28-19.47.52.501524 USER_4

Items identified by ID values must move from STATUS 'ONE' to 'TWO' to 'THREE' in sequence, without skipping statuses. The problem is to find users (STATUS_BY) values who violate the rule and move from 'ONE' immediately to 'THREE'.

The LAG() analytical function helps to solve the problem by returning for each row the value in the preceding row:

SELECT * FROM (
 SELECT 
  t.*, 
  LAG(status) OVER (PARTITION BY id ORDER BY status_time) AS prev_status 
  FROM test t
) t1 WHERE status = 'THREE' AND prev_status != 'TWO'

In case your database doesn't have LAG() you can use this to produce the same result:

SELECT A.id, A.status, B.status as prev_status, A.status_time, B.status_time as prev_status_time
FROM Data A, Data B
WHERE A.id = B.id
AND   B.status_time = (SELECT MAX(status_time) FROM Data where status_time < A.status_time and id = A.id)
AND   A.status = 'THREE' AND NOT B.status = 'TWO'

Adding the total rows selected to every row[edit | edit source]

SELECT your_columns, COUNT(*) OVER() as Ttl_Rows FROM your_data_set
id name Ttl_Rows
1 example 5
2 foo 5
3 bar 5
4 baz 5
5 quux 5

Instead of using two queries to get a count then the line, you can use an aggregate as a window function and use the full result set as the window.

This can be used as a base for further calculation without the complexity of extra self joins.

Getting the N most recent rows over multiple grouping[edit | edit source]

Given this data

User_ID Completion_Date
1 2016-07-20
1 2016-07-21
2 2016-07-20
2 2016-07-21
2 2016-07-22
;with CTE as
(SELECT *, 
        ROW_NUMBER() OVER (PARTITION BY User_ID 
                           ORDER BY Completion_Date DESC) Row_Num
FROM    Data)
SELECT * FORM CTE WHERE Row_Num <= n

Using n=1, you'll get the one most recent row per user_id:

User_ID Completion_Date Row_Num
1 2016-07-21 1
2 2016-07-22 1

This article is an extract of the original Stack Overflow Documentation created by contributors and released under CC BY-SA 3.0. This website is not affiliated with Stack Overflow