Duplicate VS Reference in Power Query in Power BI and Fabric Dataflows

When you work with tables and queries in Power Query and Power BI or dataflow in Microsoft Fabric, you can copy them through Duplicate or Reference actions. It has always been a question in my sessions and courses: what is the difference between these actions. The explanation is simple but very important to understand. Because when you know the difference, you will use it properly. In this video, I’ll explain what a Reference is and the difference between that and a Duplicate.
read my article about it here:
radacad.com/re...

[@officesuperhero4651]

Another great video. I learn so much from you.

[@RADACAD]

Thanks :) Glad to help

[@sheasafarnezhad4178]

Thank you so much, I have a question. I have a huge fact table getting data from the OData feed daily snapshot. - When I duplicate this table, does Power BI act like this is a new query and load the data twice? Or it understand that it is the same data and shortens the refresh time?

[@RADACAD]

Duplicate will create a new query and load it twice. if you want it to be loaded once, and then re-use it, I suggest using something like the Shortcut feature in Fabric Lakehouse.

[@kkravch]

Thank you. If it is possible, can you show how to reference a specific step in the query, not the final one. In your example if I want to reference the first query before you filtered it. I remember someone did it, but I don't remember who and how... Thanks again.

[@mrbartuss1]

You would have to use metadata. I think it was one of the Chandeep's tricks kzbin.info/www/bejne/ZoTNqZKtnseNiZI

[@RADACAD]

For that you can right click on the step and use "EXTRACT PREVIOUS" this will create a query up until previous step, and this one as a reference to that. I might create a short video about that too :)

[@mrbartuss1]

You can also use metadata. It was one of the Chandeep's tricks in his recent video with Chandoo

[@kkravch]

@@mrbartuss1 Thank you.

[@kkravch]

@@RADACAD Thank you. If you are going to do it, can you please touch on the performance side of running "extra" parallel queries, especially with mid or large dataset. Maybe vs the converting the table to record to refer the metadata way (Chandoo's). Thanks again.