Capacity dashboard doesn't share the client that is making interactive queries.

[dbeavon3]

In some cases the capacity dashboard doesn't share the client that is making queries against a dataset.

The following spike happens every day. It fully uses my P1 capacity.  See image below:

I'm assuming that the user name ("Power BI Service") implies that there is another dataset or report involved.  Whatever client is connecting here, it isn't very clear to me.  The Power BI service itself is the user that is indicated here. If that is so, then the Power BI service should tell me how and why it is consuming all my "CU"s.

I will try to use a profiler trace via XMLA endpoint to dig a bit deeper.  There is no remote user name, or else I would just deny them access and see what they say about it the next day!

Any tips would be appreciated.  On theory is that perhaps this is an Entra service principal, and the "capacity dashboard" just doesn't know how to display those types of accounts.  I think that service-principal-stuff was show-horned into the product as an afterthought of sorts.

[dbeavon3]

Hi @Anonymous 

I agree with you, and I'm certain there is some sort of a scheduled refresh that causes the spike. There are lots of PBI users with access to the dataset (read/build).  Given the number of users who can connect to this, it isn't easy to know whose scheduled refresh is responsible.

Normally a scheduled refresh would show the name of the user who is connecting (via the "Analysis Services" import queries).

However in this example, there is no name associated with the queries. 

As I mentioned, I suspect it may not be a normal user who is reading this data, but an Entra service principal (app registration).  Does that make sense?  Have you ever used service principals?  Have you ever seen them appear on a capacity dashboard?   I'm guessing there isn't a well-defined behavior for service principals, so it is falling back on "Power BI Service" whenever it is displaying the activity for those users.

[Anonymous]

Hi @dbeavon3 

Check if there are any scheduled refreshes for datasets that might be causing these spikes. High-frequency refreshes or large datasets can consume significant resources. You can adjust the frequency or optimize the datasets to mitigate this.

Best Regards,

Jayleny

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