SQL Window functions (OVER, PARTITION_BY, ...)

Introduction

When you run an SQL Query you select rows, but what if you want to have a summary per multiple rows, for example you want to get the top basketball for each country, in this case we don't only group by country, but we want also to get the top player for each of the country.  This means we want to group by country and then select the first player.  In standard SQL we do this with joining with same table, but we could also use partition by and windowing functions.

For each row the window function is computed across the rows that fall into the same partition as the current row. 

Window functions are permitted only in the SELECT list and the ORDER BY clause of the query

They are forbidden elsewhere, such as in GROUP BY, HAVING and WHERE clauses. This is because they logically execute after the processing of those clauses

Over, Partition By

So in order to do a window we need this input:

- How do we want to group the data which windows do we want to have?

so 

def createWindowOnRows(ByWhat): DataWithWindows

In order to specify the ByWhat part you use the OVER and also the PARTITION BY by simply specif

OVER (PARTITION BY depname)

Now that we know that we can partition the rows that we get back by department name, all we need to do is to combine it with a simple standard SQL query. 

The SQL query that we are going to combine that windowing function is a standard SQL query that contains aggregation functions but do not contain in our simple case a group by because we already do the group by with the partition by. 

So, our basic structure of simple query that would use over and partition by in order to window the data to get aggregated data as output is: 

SELECT x,y,count(z), OVER (PARTITION BY x) FROM MyTable

Note the following:

1. We have a standard SQL SELECT statement
2. One of the fields in our case in SELECT was in window we said OVER (PARTITION BY x)
3. One of the fields in our case had an aggregation we had count(z)
4. One of the fields (we had a total of 3) in select did not have any aggregation nor any window therefore we get an actual row for each such value in our case that was employee number, so we did get a row for each such employee number it was not aggregated but for each such employee number we did get the average salary for his department.

So what we see is that the window function 

performed a calculation across a set of rows that are somehow related to the current row.  The rows retain their separate identity.

It's as if the current row has looked around into multiple rows in the window and added response from the window into the current row.

You can treat the OVER as a function just as any other function like COUNT only in this case the function is a window function where in the input of this function you give it the way you want to partition your data your windowing.

Rank()

The window function will always contain an OVER clause, this is the main function, the main that we call. The partition by is how we ask SQL to divide the rows to window them.  

With rank you control the order of the rows that fall into the window add a new column with name rank according to the order by

SELECT depname, empno, salary, rank() OVER (PARTITION BY depname ORDER BY salary DESC) FROM empsalary;

Note that it would produce a new value, the new value in the select is a new column named rank,  this new column would have the order that we specify in the ORDER BY salary DESC that resides inside the OVER function.

You can have the same values in the rank column you would have actually the same values across departments because the rank is computed per each window.

References - https://www.postgresql.org/docs/9.1/tutorial-window.html