How to Setup Dataset Scale-Out

[tk6]

Good Morning,

My team recently came across this article about the new PBI feature for Dataset Scale-Out by @kayu  (https://powerbi.microsoft.com/en-us/blog/announcing-the-public-preview-of-power-bi-dataset-scale-out...). However, in trying to get this configured and running a small trial run on a small dataset, we can't seem to properly configure/setup the Dataset Scale-Out feature.

We've mainly been using this article (https://learn.microsoft.com/en-us/power-bi/enterprise/service-premium-scale-out) to help with the pre-requisites. We are testing using a ~2GB dataset and manually refreshing in the PBI service on an A1 SKU (3 GB memory). The expected result is that this refresh would be successful if the Dataset Scale-Out feature was active, but it is failing due to a memory issue.

We have confirmed that the following pieces have been setup properly:

• Premium Scale-Out tenant setting is enabled
• Dataset is large format
• XMLA read/write operations enabled

We believe our issue is coming from the template XMLA request (https://powerbiblogscdn.azureedge.net/wp-content/uploads/2023/01/XMLA-command.txt). We have tried updating the WorkspaceName with the workspace name, the workspace connect, and a workspace name that does not exist and each time has returned the same results within the SSMS v19.1 console:

<return xmlns="urn:schemas-microsoft-com:xml-analysis">
  <results xmlns="http://schemas.microsoft.com/analysisservices/2003/xmla-multipleresults">
    <root xmlns="urn:schemas-microsoft-com:xml-analysis:empty" />
  </results>
</return>

Is there an expected output from the XMLA request if it successfully runs? Is there anywhere in the PBI service to validate that a workspace has the DataSet Scale-Out feature active?

Any help would be very much appreciated. Thank you so much for your time and help!

[tk6]

@kayu , thank you so much for your response! Our team will slot this work for an upcoming sprint (we originally intended that this was going to be a quicker POC). I will be sure to get a response and mark as solution once we can get this working. Thank you so much!

[kayu]

Hi tk6, it looks like the XMLA request succeeded. At least that's what an empty XMLA response like the one you posted above implies. It might be worth checking the syncStatus API, but if your refresh is still failing, the replicas might still be on the same version. Note that refresh isolation does not automatically lower the memory requirements of your refresh. You would have to change the refresh methodology. My recommendation, start with a small evaluation dataset first. If you place it in the same workspace and configure it to use the large storage format, scale-out should be enabled by default because you have it enabled on the workspace. Then refresh this small dataset. Check the syncStatus. Replicas should be out of sync. Then manually sync the replicas. Then try things like Process Clear. Then read replica should be unaffected because you aren't syncing yet. Then do a Process Full and then a sync. Everything should be back in good shape. Once you are confident, try this approach on the larger dataset to see if it helps you lower the memory footprint of your dataset refresh on the read/write replica. 
Hope this helps,
Cheers,

Kay