I have a fairly simple data model with about 10 related tables and low data volumes (the largest table is 2.5K rows). I have a Q&A visualization that performs well in Power BI Desktop, with answers being rendered almost instantaneously.
The central problems is that when we publish the Q&A visualization to the Power BI Service, performance is poor. The same questions that Desktop answers almost instanteously may take 30 seconds to several minutes to render, and often time out with an error saying "We couldn't answer your question due to a network issue. Check your network connection and try again." I believe the error message is a red herring as I doubt the issue is truly network-related. I have asked a question, received the error, republished the Q&A visualization with all tables turned off in the Field synonyms set-up screen other than the ones this question requires, and gotten a result with the same question. I have also monitored CPU usage on my Azure capacity and haven't seen anything above about 10% utilization. I have also checked the data model to ensure there are no weird cartesian joins or anything that would cause unexpected bloat in the data.
Anecdotally, it seems like the performance started fine when we first published this visualization, and has degraded over time to the point of being unusable. I know this isn't due to large swings in data volumes as this data model is based on demo data that is static except for the dates which are rolled forward once per week to make the data seem current.
So, my questions:
- Is there a way to troubleshoot performance in the Power BI Service since the performance issues don't show up in the performance analyzer in Desktop?
- Do you have any hints that you can share to make the Q&A more performant in the Power BI Service?
- Is there a way of "resetting" the machine learning model in the Power BI Service to see if that is the cause of the performance degradation over time?
Thanks in advance.
