SQL Server: Difference between PARTITION BY and GROUP BY

Question

I've been using GROUP BY for all types of aggregate queries over the years. Recently, I've been reverse-engineering some code that uses PARTITION BY to perform aggregations. In reading through all the documentation I can find about PARTITION BY, it sounds a lot like GROUP BY, maybe with a little extra functionality added in? Are they two versions of the same general functionality, or are they something different entirely?

Answer 1

partition by doesn't actually roll up the data. It allows you to reset something on a per group basis. For example, you can get an ordinal column within a group by partitioning on the grouping field and using rownum() over the rows within that group. This gives you something that behaves a bit like an identity column that resets at the beginning of each group.

Answer 2

They're used in different places. group by modifies the entire query, like:

select customerId, count(*) as orderCount
from Orders
group by customerId

But partition by just works on a window function, like row_number:

select row_number() over (order by orderId, partition by customerId)
    as OrderNumberForThisCustomer
from Orders

A group by normally reduces the number of rows returned by rolling them up and calculating averages or sums for each row. partition by does not affect the number of rows returned, but it changes how a window function's result is calculated.

Answer 3

PARTITION BY Divides the result set into partitions. The window function is applied to each partition separately and computation restarts for each partition.

Found at this link: OVER Clause

Answer 4

It provides rolled-up data without rolling up

i.e. Suppose I want to return the relative position of sales region

Using PARTITION BY, I can return the sales amount for a given region and the MAX amount across all sales regions in the same row.

This does mean you will have repeating data, but it may suit the end consumer in the sense that data has been aggregated but no data has been lost - as would be the case with GROUP BY.