Sensitivity label metadata and semantic model partial deployment

[dagen6]

Hi Michael,

I am consolidating a couple of questions into a single post, I hope that's ok 🙂

1. Is there any function within the sempy.fabric or sempy_labs package - available also to non-Fabric admins - that would return the sensitivity label ("Internal", "Confidential", etc.) of objects in a workspace? I tried looking for it but can't seem to find it.

2. The sempy_labs.deploy_semantic_model() function is really great. I'm just wondering, would there be a way to use it to only do a partial deployment of the model? Meaning, we deploy the model, but only selected columns and measures? Some context here: in my org there is the need to share semantic models with stakeholders, but some of the attributes are sensitive and cannot be shared.

3. [Somehow related to point 2.] Is there any function that would hide columns or measures in a semantic model? The programmatic equivalent of hitting the "hide" eye icon on an attribute in the Power BI Desktop model view. Some context: in my org we are after a way to only share columns and measures that are not used in visuals - it would be great to have an automatic and programmatic way to do that, that would not require clicking on objects and hiding them one by one.

Thank you very much in advance!

[m-kovalsky]

1. The only functions which only work for admins are the ones in the admin package (and several of the capacity migration functions leverage these admin functions). All other functions are available to non-admins.
2. Seems along the lines of master model. One relatively simple thing I could do is to add an optional 'perspective' parameter. You would create a perspective with the tables/measures/columns/hierarchies you want to be in the deployed model. Then you would enter the name of that perspective in the 'perspective' parameter of deploy_semantic_model. The function would deploy the model to the new location and remove all objects not in the given perspective. How does that sound?
3. To hide a column/measure you can use TOM. These settings can be set to True or False. Same pattern applies to tables, columns, measures, hierarchies.

from sempy_labs.tom import connect_semantic_model(dataset=dataset, workspace=workspace, readonly=False) as tom:
    #tom.model.Tables["MyTable"].IsHidden=True # hide a table
    #tom.model.Tables["MyTable"].Columns["MyColumn"] = True # hide a column
    #tom.model.Tables["MyTable"].Measures["MyMeasure"] = True # hide a measure
    #tom.model.Tables["MyTable"].Hierarchies["MyHierarchy"] = True # hide a hierarchy

[dagen6]

Hi Michael,

Thank you so much for your answer 🙂 Please see my comments below.

1. Understood. I found the function sempy_labs.admin.scan_workspaces() that seems to return what I'm after (objects sensitivity label). As you confirmed, it will work only if you are an admin. I also tried list_datasets() and list_reports() in sempy.fabric, but they don't return sensitivity label info - so I guess the only solution for non-admins here is to ask admins to share the complete info they see when they run sempy_labs.admin.scan_workspaces().

2. That sounds great, and I believe that other users would also benefit a lot from such an extension!

3. Great, thanks for showing the way! The only missing piece in the puzzle, then, would be how to identify which columns are used on visuals and which not, something like the output Measure Killer shows - but that should already be covered by another existing function, right? Could you just point me to it?

[m-kovalsky]

1. Functions in the 'admin' subpackage are wrappers for Admin APIs. They naturally provide more info than the non-admin APIs. Semantic Link's functions are all for non-admins. They haven't incorporated admin API wrapper functions (which is why I made them in semantic link labs).
2. Check this out. Need to test more but it seems pretty neat. You would just have to make sure that you check for dependencies (making sure that the necessary columns exist for supporting various measures, relationships, hierarchies, roles etc.
3. See this and this. The second function is the same as the first only it checks all the downstream reports from a given semantic model within that workspace.

[dagen6]

Thanks a ton Michael.

As for 2., I will have a look at the updated code and be back with a feedback 🙂

When it comes to 3., awesome, I will explore the functions you indicated.

[dagen6]

Hey Michael, I had a look at the code and it looks good to me, thanks!
Please let me know once this enhancement is finalized, I will test the new version of deploy_semantic_model() 🙂

[m-kovalsky]

So the code in the PR works, I’ve tested it. The next step is to solve for dependencies. Whatever model remains must be valid. So if a column not in there perspective but a measure/role etc which uses that column is in the remaining model, the column must be added into the model. This will be a bit of work.

[m-kovalsky]

See the updates to deploy_semantic_model in 0.9.4.