Callum

Callum Green's Blog

Buffer() M Function in Query Editor (Power BI)

Whilst I have been aware of the Buffer() M function in the Query Editor of Power BI for a while, I had never really utilised its capabilities until now.  There are two main types of buffer functionality – Table.Buffer and List.Buffer.  To define them simply, Table.Buffer puts an entire table into memory and prevents change during evaluation, whereas List.Buffer provides a stable list, meaning it has some form of count or order.

This blog will focus on the theory behind the general Buffer() M functionality, picking a specific scenario of when it can outperform the standard Query Editor behavior.  I will also demonstrate that this is not always the most effective technique within the same scenario.  The article will not give you a hard and fast rule of when to use the Buffer() function, because it can depend on a number of factors.  These are described further below.

Note:    It is assumed you have existing knowledge of Query Folding and if not, one of my previous blogs should help greatly.

Scenario

I found inspiration from Chris Webb’s example, using the Adventure Works DW 2014 database – available here.

The requirements are:

1.       Obtain the first 10,000 rows from FactInternetSales

2.       Remove the majority of columns, retaining ONLY:

a.       SalesOrderLineNumber

b.      CustomerKey

c.       SalesAmount

3.       Rank the current row based on Sales Amount.

List.Buffer()

Assuming your database exists on a local server and is named AdventureWorksDW2014, copy the following code into the Advanced Editor in the Query Editor screen.

let

    //Connect to SQL Server

    Source = Sql.Database("localhost", "AdventureWorksDW2014"),

    //Get first 2000 rows from FactInternetSales

    dbo_FactInternetSales = Table.FirstN(

          Source{[Schema="dbo",Item="FactInternetSales"]}[Data],

          10000),

    //Remove unwanted columns

    RemoveColumns = Table.SelectColumns(

          dbo_FactInternetSales,

          {"SalesOrderLineNumber", "CustomerKey","SalesAmount"}),

    //Get sorted list of values from SalesAmount column

   RankValues = List.Sort(RemoveColumns[SalesAmount], Order.Descending),

    //Calculate ranks

    AddRankColumn = Table.AddColumn(RemoveColumns , "Rank",

          each List.PositionOf(RankValues,[SalesAmount])+1)

in

    AddRankColumn

You can visibly see the rows loading – one by one.  In total, it takes nearly 1 minute to load all off the results.

Now let’s use the List.Buffer() function in the RankValues step.

Replace:

= List.Sort(RemoveColumns[SalesAmount], Order.Descending)

With:

= List.Buffer(List.Sort(RemoveColumns[SalesAmount], Order.Descending))

clip_image002

The entire transformation (from start to finish) completes in just under 2 seconds!  This is because the List.Buffer function stores the sorted values in memory and therefore, the rank calculation is only evaluated once.  The last query (and previous steps) were being evaluated multiple times.  The M language is both functional and at times, lazy.  In order to prevent the constant re-evaluation, buffer the list into memory. 

The final query output is shown below:

clip_image004

Query Folding

We will implement the same requirements, but this time using Query Folding. 

The third step in our current transformation is called ‘Removed Columns’. This is what prevents Query Folding, as this function cannot be interpreted/translated to the native SQL Server T-SQL language.  All steps below are inadvertently not supported either. 

The way around this is to write SQL Server View (in SSMS) to import just the fields required from the underlying FactInternetSales Table.  The below query will give you the same result up to the ‘Remove Columns’ step.

CREATE VIEW dbo.VwFactInternetSalesAmount

AS

       SELECT SalesOrderNumber   

                     ,[CustomerKey]

                     ,[SalesOrderLineNumber]

                     ,[SalesAmount]

                     ,RANK() over( order by [SalesAmount] desc) AS [Rank]

       FROM   [AdventureWorksDW2014].[dbo].[FactInternetSales]

clip_image006

The final steps are to filter on the top 10,000 rows and Group the Rows together – inserting the following M syntax into the last Applied Step:

let

    //Connect to SQL Server

    Source = Sql.Database(".", "AdventureWorksDW2014"),

    // Connect to SQL Server

    dbo_FactInternetSales = Source{[Schema="dbo",Item="VwFactInternetSalesAmount"]}[Data],

    #"Sorted Rows" = Table.Sort(dbo_FactInternetSales,{{"SalesOrderNumber", Order.Ascending}}),

    #"Kept First Rows" = Table.FirstN(#"Sorted Rows",10000),

    #"Grouped Rows" = Table.Group(#"Kept First Rows", {"CustomerKey", "SalesOrderLineNumber", "Rank"}, {{"TotalSalesAmount", each List.Sum([SalesAmount]), type number}})

in

    #"Grouped Rows"

 

The query now returns instantly (under 1 second).   Right click on the last applied step and select the View Native Query option, to show the underlying SQL.

select top 10000

    [rows].[CustomerKey] as [CustomerKey],

    [rows].

Loading