Dataflow version control

[DebbieE]

I am working with a team who are worried that if a dataflow changes then lots of reports break that are joined to the dataflow because people who create the datasets and data reports don't know that its been changed.

What they have come up with is having dataflow versions. So people will be working against specific dataflows.

Say Dataflow V1

Dataflow V2 

Dataflow  V3 etc,

I dont think this is a good idea after reading up on governance. the data duplication for example to me is a no go but they would like more information on why this isnt a good idea,

If anyone has any thoughts on this. That would be great

[samaguire]

This maybe a bit late, however, I came across this searching for something else. Literally just having finished reading the MS Doc for BYO ADLS Gen2 for Dataflows I can say if they want version control they should set this up. The system takes snapshots of not only the data, but the Dataflow metadata with every refresh. Configuring dataflow storage to use Azure Data Lake Gen 2 - Power BI | Microsoft Docs

[MFelix]

Hi @DebbieE ,

I use some dataflow on my projects and has you refer we do not tend to use versions of the dataflows, I don't know what type of changes you refer, but if the changes is adding columns or data within the dataflow that should not give you any issue since what can happen is that the columns added aren't picked up.

However if the changes passes trough formatting change, renaming, deleting  then your users can have an issue.

Using the lineage you can check wich datasets are connected to the dataflow and send warnings about updates, another options is to create a power automate to send an alert to users when there are refresh or updates to dataflow

https://docs.microsoft.com/en-us/power-query/dataflows/send-notification-when-dataflow-refresh-compl...

The link above is for a refresh of a dataflow but maybe can be adjusted for an update.

Hope this helps to get some guidance.

[DebbieE]

can I give them any specific reasons about version control being the wrong move. I mentioned extra complexity and data duplication and they are still arguing for it. So some good reasons not to do it is what I am after.