Limit Dimension Data based on Fact Table in Power Query

[Syndicate_Admin]

There are 2 tables:

1. Dimension Table D - EMPNO (25,000,000 rows) sourced from Oracle DB

2. Fact Table F - Excel source

join on D.EMPNO=F.EMPNO using Merge queries in Power Query.

The performance is extremely slow using Import/DirectQuery mode. I tried to limit the rows in Dimension table by following the below steps:

1. Buffered List to extract unique list of EMPNO in F - EmpList 

Emplist = let
Source = List.Distinct(F[EMPNO]),
#"Buffered List" = List.Buffer(Source)
in
#"Buffered List"

2. Filter Dimension D  i.e. Table.SelectRows(#"Changed Type",each List.Contains(EmpList, [EMPNO]))

It works but it is extremely slow when "Enable Load" is enabled for Dimension D. Although the Step (2) returns few rows(129 rows), when I apply the changes in Power Query, it loads all the rows for Dimension D.

I was hoping to get some guidance if there is anything wrong or can be done differently.

[Syndicate_Admin]

Hi @BA_Pete 

The report works fine based on the established relationship.

But the filter the Dim table based on Buffered list and save the changes steps takes a long time.

[Syndicate_Admin]

Hi @LD2022 ,

Ok. My guess is that you're applying a query step to your dimension table that is breaking query folding or causing unnecesary processing. Using the method that we have, Power Query should be streaming the F[EMPNO] values to your SQL server in a native query, and should be really fast. Can you right-click the final step in your dimension query and check whether the 'Native Query' option is lit up or greyed-out please?

If it's greyed-out, select your dimension table as we have it set up now, go to the Home tab, select Advanced Editor and copy all the code in there. Then paste the whole lot into a code window here using this button:

Before you close the code window, please just overtype any sensitive connection string values (file paths, server paths etc.) with 'XXX', but keep the code structure itself, as this may be important.

I should be able to see if there's anything breaking query folding or causing unnecessary processing time that we can work around.

Pete

[Syndicate_Admin]

Hi @BA_Pete 

Below are the steps I have done:

DIM_EMPLOYEE=

let
    Source = Oracle.Database("XXX", [HierarchicalNavigation=true, Query="SELECT * FROM XX.DIM_EMPLOYEE"])
    
in
    Source

Create a reference table as below:

Filtered_Emp=

Filtered_Emp= let
    Source = DIM_EMPLOYEE,
    #"Filtered Rows" = Table.SelectRows(Source, each List.Contains(List.Buffer(EmpList),[CUSTOMER_NUMBER]))
in
    #"Filtered Rows"

The native query is greyed out.

I have disabled load for everything except the F and  Filtered_Emp. When I hit apply in Power Query, it tries to load all the 2 million records athough the Filtered_Emp has 200 records only.

[Syndicate_Admin]

Ok. Your current query setup won't allow query folding.

Can you try connecting without using an explicit SQL statement please?

Connect to the DB again, and leave the SQL statement blank:

Just hit OK, then select your dim_employee table from the table list and import.

Assuming this goes as planned, select the [EMPNO] column in your new dimension table and filter just one value. This should be fast to implement and sets us up a new step with the correct structure.

Edit that new step in the formula bar, so it uses our streaming list filter, something like this:

= Table.SelectRows(
    previousStepName,
    each List.Contains( List.Buffer( factTableName[EMPNO] ), [EMPNO] )
)

This setup should allow PQ to stream the [EMPNO] values from your Excel source into the SQL native query sent to the SQL source for your dim_employee table.

Pete 

[Syndicate_Admin]

Hi @BA_Pete 

I followed the steps that you suggested but still no change in the performance.

let
    Source = Oracle.Database("XXXX", [HierarchicalNavigation=true]),
    XX = Source{[Schema="XX"]}[Data],
    DIM_EMPLOYEE = XX{[Name="DIM_EMPLOYEE"]}[Data],
    #"Filtered Rows" = Table.SelectRows(DIM_EMPLOYEE, each List.Contains(List.Buffer(Fact[EMPNO]),[EMPNO]))
in
    #"Filtered Rows"

[Syndicate_Admin]

Hi @LD2022 ,

If you right-click your #"Filtered Rows" step, is 'Native Query' now selectable, or is it still greyed-out?

If it's still greyed out, try changing one of the column data types to the same type, but not a text column. For example, change a date type column to date type, or a decimal type column to decimal. This is purely to force PQ to re-evaluate the query and hopefully force it to generate a native query to source.

If native query IS selectable, then you're probably at the limit of optimisation. The Oracle Data Access Client (ODAC) requirement isn't the best in this regard.

Pete

Pete

[Syndicate_Admin]

Hi @BA_Pete 

I tried to perform the suggested steps from scratch and when I am applying the filter to limit the rows in the dimension, it says"This stepsresults in a query that is not supported in DirectQuery mode".

[Syndicate_Admin]

Hi @LD2022 ,

You shouldn't need to use Direct Query mode any more.

I believe you only wanted to use DQ due to the large number of rows in your dimension table. If we can get this method to fold to the source, it will work very quickly in Import mode as you will only actually be importing the number of rows that correlate to to values in your fact table (about 126 IIRC).

Go to your Model view in PBI Desktop, select your DQ dimension table, go to the Properties pane on the right, expand 'Advanced' at the bottom of this pane and choose Import. Then go and check your dimension query in PQ to see if the last step has 'Native Query' lit up.

Pete

[Syndicate_Admin]

Hi @BA_Pete 

It worked 🙂 But I also noticed that the moment I add any steps to the Dimension i.e. rename column/change datatype, the Native query option is greyed out and it started loading the data again. I will probably add a note in PQ not to add any steps in future else it might stop working.

Thank you so much @BA_Pete for the solution.

[Syndicate_Admin]

Hi @BA_Pete 

It worked 🙂 But I also noticed that the moment I add any steps to the Dimension i.e. rename column/change datatype, the Native query option is greyed out and it started loading the data again. I will probably add a note in PQ not to add any steps in future else it might stop working.

Thank you so much @BA_Pete for the solution.