PQ in Azure Data Factory - Mashup Engine or Transposition

[Ben-Dev]

Hello,

The documentation for Azure Data Factory's 'Execute Power Query activity' states:
> To achieve scale with your Power Query activity, Azure Data Factory translates your M script into a data flow script so that you can execute your Power Query at scale using the Azure Data Factory data flow Spark environment.

Does this mean that, outside of ADF's Query Editor, there's no PQ mashup engine involved? In other words, that the saved M script is transposed into a Spark-powered data flow script and run entirely by Spark without a mashup engine executing?

Thanks!

[v-stephen-msft]

Hi @Ben-Dev ,

The engine makes it easy for data movement and transformation. You could refer to the following article which has a good introduction.

What Is Power Query? Introduction to Data Mash-up Engine of Power BI - RADACAD

Best Regards,

Stephen Tao

If this post helps, then please consider Accept it as the solution to help the other members find it more quickly.

[Ben-Dev]

Hi @v-stephen-msft,

Thanks for this info!

Actually, my question is specifically about how Power Query works when used from ADF, not how it works in general.

I understand that "normal" Power Query (e.g. the PQ that ships with Power BI and Excel) uses the PQ mashup engine to evaluate M expressions and produce the requested output. 

However, the quote from the above-mentioned article stating that ADF's PQ activity "translates your M script into a data flow script" makes me think that maybe the mashup engine is not involved at all when PQ is run from ADF—so, when using PQ from ADF, instead of my M expression being run by a mashup engine, it is never touched by a mashup engine but instead the M expression is simply translated into an ADF data flow (Spark) script and run by Spark.

Is that correct?

/CC: @Ehren 

[Ehren · Accepted solution]

so, when using PQ from ADF, instead of my M expression being run by a mashup engine, it is never touched by a mashup engine but instead the M expression is simply translated into an ADF data flow (Spark) script and run by Spark

That's correct.

[Ben-Dev]

Thank you!